Embed Size (px)
Citation preview
138 CS13 Abstracts
IP1
Control and Optimization of Subsurface Flow
Controlling the flow of fluids (e.g. water, oil, natural gasor CO2) in subsurface porous media is a technical pro-cess with many mathematical challenges. The underlyingphysics can be described with coupled nearly-elliptic andnearly-hyperbolic nonlinear partial differential equations,which require the aid of large-scale numerical simulation.The strongly heterogeneous nature of subsurface rock leadsto strong spatial variations in the coefficients. Moreover,the limited accessibility of the underground leads to verylarge uncertainties in those coefficients and severely lim-its the amount of control over the dynamic variables. Inthis talk I will address related system-theoretical aspects,reduced-order modeling techniques, and adjoint-based op-timization methods.
Jan Dirk JansenTU Delft, The Netherlands’[email protected]’
IP2
Analyzing and Generating BIG Networks
Graphs and networks are used to model interactions in avariety of contexts, and there is a growing need to accu-rately measure and model large-scale networks. We con-sider especially the role of triangles, useful for measuringsocial cohesion. This talk will focus on two topics: (1) Ac-curately estimating the number of triangles and clusteringcoefficients for BIG networks, and (2) Generating BIG ar-tificial networks that capture the degree distribution andclustering coefficients of observed data. This is joint workwith Ali Pinar, Todd Plantenga, and C. Seshadhri fromSandia National Labs and Christine Task from Purdue Uni-versity.
Tamara G. KoldaSandia National [email protected]
IP3
Certified Reduced Models and Their Applications
Models of reduced computational complexity are used ex-tensively throughout science and engineering to facilitatemodeling of complex systems for control, design, multi-scale analysis, uncertainty quantification etc. We shall dis-cuss ongoing efforts to develop reduced methods endowedwith rigorous error estimators to certify models, henceendowing it with predictive value. We outline the basicideas behind certified models and discuss computationalefficiency and efficient model construction. The perfor-mance of the certified reduced models will be illustratedthrough several examples and, time permitting, we con-clude by discussing some ideas aiming to enable the devel-opment of certified reduced models for high-dimensionalparametrized problems.
Jan S. HesthavenBrown UniversityDivision of Applied [email protected]
IP4
Modeling Cardiac Function and Dysfunction
Simulating cardiac electrophysiological function is one of
the most striking examples of a successful integrative multi-scale modeling approach applied to a living system directlyrelevant to human disease. This presentation showcasesspecific examples of the state-of-the-art in cardiac integra-tive modeling, including 1) improving ventricular ablationprocedure by using MRI reconstructed heart geometry andstructure to investigate the reentrant circuits formed in thepresence of an infarct scar; 2) developing a new out-of-the box high-frequency defibrillation methodology; 3) un-derstanding the contributions of non-myocytes to cardiacfunction and dysfunction, and others.
Natalia A. TrayanovaJohn Hopkins UniversityInstitute for Computational [email protected]
IP5
Quantum Mechanics Without Wavefunctions
In principle, predictions of the electronic structure ofatoms, molecules, and materials requires solving the many-body Schrdinger wave equation (SWE), whose eigenvaluesand eigenfunctions delineate the distribution of electronsin energy and space, respectively. In fact, the SWE cannotbe solved exactly for more than one electron but excellentapproximations are available. However, such methods scaleextremely poorly with system size and generally are not ap-plicable to large numbers of atoms or to condensed matter.Alternatively, one can solve directly for the electron den-sity distribution rather than the many-body wavefunction,within orbital-free density functional theory (OFDFT). Byso doing, the problem greatly simplifies to 3 degrees offreedom rather 3N (N being the number of electrons). Thestate of the art of OFDFT, in terms of algorithms, physicalmodels, and applications, will be discussed.
Emily A. CarterPrinceton [email protected]
IP6
Automated Astrophysics in the Big Data Era
Telescope projects are now routinely obtaining massive dig-ital movies of the dynamic nights sky. But given the grow-ing data volumes, coupled with the need to respond to tran-sient events quickly with appropriate followup resources,it is no longer possible for people to be involved in thereal-time loop. I discuss the development of a robotic tele-scopes, autonomous follow-up networks, and a machine-learning framework that act as a scalable, deterministichuman surrogate for discovery and classification in astro-nomical imaging.
Joshua S. BloomUniversity of California, [email protected]
IP7
PDE-Based Simulation Beyond Petascale
We explore fundamental computational complexity con-siderations that will drive algorithmic design choices forPDE-based simulation codes scaling to petascale and be-yond. We argue that high-order methods using implicitor semi-implicit solvers are essential to efficient simulationof multiscale problems. These methods can be realizedat per-point-costs equivalent low-order methods. We fur-
CS13 Abstracts 139
ther show that multilevel solvers having bounded iterationcounts can scale to billion-way concurrency. We analyzethe scalability of (low- or high-order) domain decomposi-tion approaches to predict parallel performance on exascalearchitectures. These predictions shed light on what exas-cale CFD computation might enable and provide insightto design requirements of exascale algorithms, codes, andarchitectures.
Paul F. FischerArgonne National [email protected]
IP8
Challenges for Algorithms and Software at Ex-treme Scale
Extreme scale systems face many challenges. The end offrequency scaling forces the use of extreme amounts of con-currency. Power constraints are forcing a reconsiderationof the processor architecture, eliminating features that pro-vide small performance benefit relative to the power con-sumed and making use of specialized processing elementssuch as GPUs. Future systems will need to combine theseand other approaches to approach Exascale performance.This talk discusses how algorithms and software need tochange to make effective use of extreme scale systems.
William D. GroppUniversity of Illinois at Urbana-ChampaignDept of Computer [email protected]
IP9
Consistent Modelling of Interface Conditions forMulti-Physics Applications
In many multi-physics applications, information transportplays an important role for the efficiency and stability ofthe applied numerical algorithms. The global accuracy isquite often dominated by local effects at the interfaces,and local singularities can pollute the numerical solution.Here we discuss several issues such as hierarchical limitingtechniques, local energy corrections and optimal estimatesfor the flux variables. The coupling is controlled by pairsof balance equations providing a very flexible framework.Different examples illustrate the abstract concepts, specialfocus is on surface based coupling techniques and highlynon-linear dimension-reduced systems.
Barbara WohlmuthM2, Centre for Mathematical Sciences,Technische Universitat Munchen, [email protected]
CP1
Planning of MR-Guided Laser Induced ThermalTherapy Using UQ Methods
Magnetic resonance-guided laser induced heating of can-cer metastases in brain is a minimally invasive alternativeto surgery. MR thermal imaging provides thermal dosefeedback to the surgeon during treatment. A generalizedpolynomial chaos implementation of the stochastic bioheatequation was critically evaluated for aiding in the planningand effective therapy delivery. High performance imple-mentations are needed clinically. Statistical comparisonsof the measured and predicted temperature field will be
presented in several in-vivo datasets.
David Fuentes, Samuel Fahrenholtz, John Hazle, JasonStaffordMD Anderson Cancer [email protected], [email protected], [email protected], [email protected]
CP1
Renewables Integration in Power System Opera-tions Modeling
The requirements for a smart, green, efficient and reliablegrid are reshaping the US power grid. The integrationof renewable resources in the power system operations in-troduces further complexity to utilities profit optimizationproblem. The Renewable Integration Model (RIM) a com-puter based system - offers insight on power systems man-agement, and answers questions on the integration of re-newable energies. The RIM’s design, solution method, andinitial results are studied in the following.
Cristina MarinoviciPacific Northwest National [email protected]
Harold KirkhamPacific Northwest National LaboratoryRichland [email protected]
Leif [email protected]
Kevin GlassPacific Northwest National LaboratoryRichland [email protected]
CP1
Heel Effect Adaptive Gain Correction of Flat PanelX-Ray Detectors
Anode heel effect causes large-scale variations in the X-raybeam intensity. Together with the localized gain variabilityof the detector, they contribute to the non-uniformities ofdigital radiographs. For mobile applications, conventionalflat field gain correction techniques degrade substantiallywithout recalibration after each source-to-image distance(SID) reset. We present a novel method to adaptively com-pensate for the heel effect with flat field techniques andminimize the effort for recalibration. The method is basedon a mathematical-physical model to estimate the heel ef-fect ratio at two different SIDs. The gain factor calibratedat the standard SID is applied to any setting with heel ef-fect compensation. The model parameters are determinedby fitting the model to directly measurable quantities. Theeffect of the proposed method is demonstrated on flat fieldimages acquired at variable SIDs.
Jue WangUnion [email protected]
Yongjian YuVarian Medical [email protected]
140 CS13 Abstracts
CP1
A Parallel Two-Level Newton-Krylov-SchwarzMethod For Three-Dimensional Blood Flow Sim-ulations
We develop a parallel scalable domain decompositionmethod for numerical simulations of blood flows in three-dimensional compliant arteries. The corresponding fluid-structure interaction is modeled by a fully coupled systemof linear elastic equation and the incompressible Navier-Stokes equations in an ALE framework. The resultingmonolithic system is discretized with a fully-implicit fi-nite element method on an unstructured moving meshand solved by the Newton-Krylov method with a two-levelSchwarz preconditioner. The investigation focuses on theefficiency and parallel scalability of the proposed two-levelalgorithm.
Yuqi WuDepartment of Applied Mathematics, University [email protected]
Xiao-Chuan CaiUniversity of Colorado, BoulderDept. of Computer [email protected]
CP2
Nonlinear Scale Interactions and Energy Pathwaysin the Ocean
We refine a novel mathematical framework to analyze non-linear scale-coupling and energy pathways in scale and inspace from high-resolution ocean simulations. We test thevalidity of the traditional paradigm for such pathways atvarious locations such as in western boundary currents,near the equator, and in the deep ocean. We investigatethe contribution of various nonlinear mechanisms to energytransfer across scales such as baroclinic and barotropic in-stabilities, barotropization, and Rossby wave generation.
Hussein Aluie, Matthew HechtLos Alamos National [email protected], [email protected]
Geoffrey Vallis, Kirk BryanPrinceton/[email protected], [email protected]
Robert EckeCenter for Nonlinear StudiesLos Alamos National [email protected]
Mathew [email protected]
Beth WingateLos Alamos National LaboratoryComputer And Computational Sciences [email protected]
CP2
A Non-Hydrostatic Spectral-Element Model of the
Atmosphere
We are developing a non-hydrostatic version of theHOMME or CAM-SE spectral-element dynamical core em-ployed by NCAR and the climate modeling community. Wewill discuss the design choices of this model including ver-tical coordinates and time stepping techniques methods,and demonstrate its performance on several standard hy-drostatic and non-hydrostatic tests.
David M. HallNational Center for Atmospheric Research1850 Table Mesa Drive, Boulder CO [email protected]
CP2
Non-Dissipative Space Time Hp-DiscontinuousGalerkin Method for the Time-Dependent MaxwellEquations
A space-time finite element method for the time-dependentMaxwell equations is presented. The method allows for lo-cal hp-refinement in space and time by employing a space-time Galerkin approach and is thus well suited for hp-adaptivity. Inspired by the so called cG methods for ODEs,nonequal test and trial spaces are employed. This allowsfor obtaining a non-dissipative method. For an efficient im-plementation, a hierarchical tensor product basis in spaceand time is proposed.
Martin LilienthalGraduate School of Computational EngineeringTU [email protected]
CP2
Discontinuous Collocation Method, Convergenceand Applications
Collocation is a successful technique to simulate PDE inone spatial dimension. Recently, we have introduced dis-continuous collocation method and applied it to PDE inseveral spatial variables. The technique employs triangularor tetrahedral partitions and piecewise polynomial interpo-lations. We proved that the method is L-infinity conver-gent. In the current work we prove L2 convergence undermore general hypothesis. Our proof is similar to Bialeckisproof for rectangular partitions. We provide improved ex-amples.
John A. LoustauHunter Collete CUNYDept of Math and [email protected]
Ariel LindorffHunter College CUNYDept of Math [email protected]
CP2
Automated Refinements for Discontinuous Collo-cation Method
Discontinuous collocation method is based on piecewisepolynomial interpolation over a triangular domain parti-tion. The method is discontinuous as the interpolationneed not be continuous at element boundaries. If we ex-pect the solution to be continuous then the size of the jump
CS13 Abstracts 141
discontinuities can be used to measure the error. In thiswork we use these values to guide a local refinement pro-cess which may be repeated until the implied error is withinprescribed bounds.
Amy WangHunter College CUNY - Department of Math and [email protected]
John A. LoustauHunter Collete CUNYDept of Math and [email protected]
CP3
Wave Dynamics of Two-Phase Flow Models
Hyperbolic conservation laws with relaxation terms can beused to model two-phase flow in chemical non-equilibrium.Relaxation terms will introduce dispersion in the wave dy-namics of the hyperbolic system. We use a relaxationparameter ξ = kε to study the transition between theequilibrium model and the frozen (non-equilibrium) model.Herein, we demonstrate the existence of a critical point forwhich the wave dynamics change from being equilibrium-like to being frozen-like.
Peder Aursand, Susanne SolemNorwegian University of Science and [email protected], [email protected]
Tore FlattenSINTEF Energy [email protected]
CP3
Dispersive Wave Propagation in Solids with Mi-crostructure
The accuracy of dispersive wave models for solids withmicrostructure is analyzed through a series of numericalsimulations. We compare numerical results obtained usingvarious descriptions of the internal structure of a mate-rial: a micromorphic model for the microstructure, regu-lar laminate structures, laminates with substructures, etc.,for a large range of material parameters and wavelengths.Numerical computations are performed by means of thefinite-volume numerical scheme, which belongs to the classof wave-propagation algorithms.
Arkadi BerezovskiInstitute of Cybernetics at Tallinn University [email protected]
Mihhail BerezovskiDepartment of Mathematical Sciences [email protected]
CP3
Efficient Boundary ElementMethods for MolecularElectrostatics Using Python and Gpus
Many biomolecular processes are studied using electrostat-ics and continuum models. One such model for proteins inionic solutions results in a coupled system of Poisson and
linearized Poisson-Boltzmann equations. Using an integralformulation, this problem is solved with boundary elementmethods, BEM. We present a treecode-accelerated BEMthat can solve biologically relevant problems using GPUhardware but with a friendly Python interface. We willpresent validation results and demonstrations with multi-ple proteins in solution.
Christopher CooperBoston [email protected]
Jaydeep BardhanDepartment of Electrical and Computer EngineeringNortheastern [email protected]
Lorena A. BarbaDepartment of Mechanical EngineeringBoston [email protected]
CP3
Enrichment by Exotic NURBS Geometrical Map-pings in Isogeometric Analysis for fracture Me-chanics
The mapping techniques are concerned with constructionsof nonconventional exotic geometrical mappings such thatthe pushforward functions of B-spline functions defined onthe parameter space into physical domain by the mappingsgenerate singular functions that resemble the given pointsingularities. In this paper, for highly accurate stress anal-ysis of elastic structures with cracks or corners, we mixNURBS basis functions and singular basis functions con-structed by the design mapping and the exotic geometricmapping, respectively.
Hae-Soo OhThe Department of Mathematics and StatisticsUniversity of North Carolina at [email protected]
Hyunju KimUniversity of North Carolina at CharlotteThe Department of Mathematics and [email protected]
Jae Woo JeongDepartment of MathematicsMiami University, Hamilton, OH [email protected]
CP3
Computational Analysis of Non-Standard Wave Structures in Hydrodynamic andMagneto-Hydrodynamic Systems
We consider the Euler equations of hydrodynamics and thesystem of magneto-hydrodynamic equations both closedwith a non-convex equation of state. We study the wavestructure for both models and compute the approximatesolution of several initial value problems. We analyze theformation of complex composite waves in both models. Weinvestigate the influence of thermodynamical and magneticfield magnitudes in the formation of composite waves.
Susana SernaDepartment of Mathematics
142 CS13 Abstracts
Universitat Autonoma de [email protected]
CP3
Elimination of Oscillating Singularities at theCrack-Tips of An Interface Crack with a Help ofa Curvature-Dependent Surface Tension
A new model of fracture mechanics incorporating acurvature-dependent surface tension acting on the bound-aries of a crack is considered. The model is studied onthe example of a single straight interface crack betweentwo elastically dissimilar semi-planes. Linear elasticity isassumed for the behavior of the material of the plate ina bulk. A non-linear boundary condition with a consid-eration for a curvature-dependent surface tension is givenon the crack boundary. It is well known from linear elas-tic fracture mechanics (LEFM) that oscillating singulari-ties exist at the crack tips and lead to non-physical inter-penetration and wrinkling of the crack boundaries. Us-ing the methods of complex analysis, such as Dirichlet-to-Neumann mappings, the problem is reduced to a systemof six singular integro-differential equations. It is provedthat the introduction of the curvature-dependent surfacetension eliminates both classical power singularities of theorder 1/2 at the tips of the crack and oscillating singular-ities, thus resolving the classical contradictions of LEFM.Numerical computations are presented.
Anna ZemlyanovaDepartment of MathematicsTexas A&M [email protected]
CP4
A Coupled Poroelasticity Numerical Model byMixed Finite Elements
The numerical treatment of coupled poroelasticity is diffi-cult because of the instabilities affecting the pressure solu-tion. A coupled poroelasticity model based on Mixed Fi-nite Elements has been developed so as to alleviate the nu-merical oscillations at the interface between different ma-terials. Both a monolithic and sequential scheme are im-plemented, the former implying the use of ad-hoc precon-ditioned Krylov methods, the latter the separate solutionof flow and deformation with an outer iteration to achieveconvergence.
Massimiliano FerronatoUniversity of [email protected]
Nicola CastellettoDMMMSAUniversity of [email protected]
Carlo JannaDMMMSA - University of [email protected]
CP4
Efficient Time-Stepping for Long-Time Simulationsof Parabolic Reaction-Diffusion Equations
The flow of calcium ions in a heart cell is modeled by a sys-
tem of time-dependent parabolic reaction-diffusion equa-tions. The spatial discretization of this system by finitedifference, finite element, and finite volume methods in amethod of lines approach results in systems of ODEs thatnecessarily get stiffer and larger with increasing spatialmesh resolution. We will compare several state-of-the-artlinear multi-step and Runge-Kutta methods, all specificallydesigned for problems of this type.
Xuan HuangDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
Philipp BirkenDepartment 10, Mathematics and Natural SciencesUniversity of Kassel, [email protected]
Matthias K. GobbertUniversity of Maryland, Baltimore CountyDepartment of Mathematics and [email protected]
CP4
Adaptive Hp Finite Elements Using Cuda
Adaptive hp finite elements lead to exponential conver-gence to the exact solution for many PDEs. High polyno-mial order methods also offer the possibility of higher arith-metic intensity (ratio of computation performed to mem-ory IO) on modern many-core architectures. We present anew high-performance implementation tuned for adaptivehp finite elements that hides the complexity of performingquadrature for multiple polynomial orders, arbitrary ref-erence to physical mapping, and anisotropic and variablematerial properties. We also present performance resultsusing NVIDIA CUDA.
Chetan Jhurani, Paul MullowneyTech-X [email protected], [email protected]
CP4
A Robust Non-Negative Numerical Frameworkfor Diffusion-Controlled Bimolecular-Reactive Sys-tems
In this talk, we will present a novel non-negative computa-tional framework for diffusive-reactive systems in heteroge-neous anisotropic rigid porous media. The governing equa-tions for the concentration of reactants and product will bewritten in terms of tensorial diffusion-reaction equations.We shall restrict ourselves to fast irreversible bimolecularreactions in which Damkohler number Da >> 1. We em-ploy a linear transformation to rewrite the governing equa-tions in terms of invariants, which are unaffected by thereaction. This results in two uncoupled tensorial diffusionequations in terms of these invariants, which are solvedusing a novel non-negative solver for tensorial diffusion-type equations. The concentrations of the reactants andproduct are then calculated from invariants using algebraicmanipulations. Several representative numerical exampleswill be presented to illustrate the robustness, convergence,and the performance of the proposed computational frame-work. We will also compare the proposed formulation withother popular formulations. In particular, we will showthat the standard single-field formulation does not pro-duce reliable solutions, and the reason can be attributed
CS13 Abstracts 143
to the fact that the single-field formulation does not guar-antee non-negative solutions. We will also show that theclipping procedure (which produces non-negative solutionsbut is considered as a variational crime) over predicts theplume length.
Maruti K. Mudunuru, Kalyana NakshatralaUniversity Of Houston - Main [email protected], [email protected]
Albert J. ValocchiUniversity of Illinois at [email protected]
CP4
Complexity-Friendly Algebraic Multigrid Precon-ditioners Based on Sparsified Coarsening
Traditional algebraic multigrid (AMG) methods useGalerkin coarsening where the sparsity of the multigridhierarchy is dictated by its transfer operators. In manyscenarios, the coarse operators tend to be much denserthan the fine operator as the coarsening progresses. Suchbehavior is problematic in parallel computations where itimposes expensive communication overhead. We presenta new technique for controlling the sparsity of the AMGoperators. Our algorithm sparsifies the Galerking opera-tors while preserving the energy of the algebraically smootherror modes. Numerical experiments for large-scale Graph-Laplacian problems demonstrate the potential of this ap-proach.
Eran TreisterComputer ScienceTechnion - Israel Institute of [email protected]
Irad YavnehComputer Science Department, [email protected]
Raanan FattalComputer Science and Engineering,The Hebrew Unisersity of [email protected]
CP4
Localized Method of Approximate Particular Solu-tions for Solving Diffusion-Reaction Equa Tions inTwo-Dimensional Space
An localized meshless method, the localized method of ap-proximate particular solutions(LMAPS), was developed re-cently to solve elliptic partial differential equations. In thispaper, the method is extended to solve diffusion-reactionequations in two-dimensional space, using positive definiteradial basis function multiquadrics and non-positive def-inite radial basis function thin-plate splines. The com-parison of method with different radial basis functions aretested on three examples. The numerical experiment sug-gested that the localized method with multiquadric is muchmore stable than the method with thin-plate splines.
Guangming YaoClarkson [email protected]
CP5
Spectral Methods on GPUs with Applications inFluid Dynamics and Materials Science
GPGPU recently has drawn a big attention from the CSEcommunity. This talk exploits how spectral methods canbe benefited from GPGPU. Our focus is the design and im-plementation of spectral-collocation and spectral-Galerkinmethods for elliptic systems. And we look at applicationsof the Navier-Stokes equation in incompressible fluids, theCahn-Hilliard Equation in phase-field models, and nonlin-ear PDEs from the recently proposed phase-field-crystalmodel.
Feng ChenBrown Universityfeng chen [email protected]
CP5
Comparative Study of Various Low Mach NumberPreconditioners Applied to Steady and UnsteadyInviscid Flows
The performance of many existing compressible codes de-grades as the Mach number of the flow tends to zero andleads to inaccurate computed results. To alleviate theproblem, Roe-type based schemes have been developed forall speed flows, such as the preconditioned Roe and LM-Roe schemes. Most of these schemes tend to computesteady state solutions, but many fail to address unsteadysolutions. This paper attempts to evaluate various precon-ditioners in terms of development, performance, accuracyand limitations in both steady and unsteady simulationsat various Mach numbers by implementing them in a 3DEuler Solver using Roe Scheme on Unstructured meshes.While developing 3D Unstructured Euler Solver, math-ematical models of each preconditioning is developed byusing proper Non-Dimensionalization, Primitive variablesand Characteristics based Boundary Conditions. Precon-ditioners proposed by W.R. Briley, Weiss-Smith, Turkel,Eriksson, Choi and Merkel, and LM-Roe are studied bycomputing flow over a NACA0012 airfoil and Cylinder atvarious Mach Numbers.
Ashish GuptaUniversity of Tennessee at [email protected]
CP5
Rankine-Hugoniot-Riemann Solver for Multi-Dimensional Balance Laws with Source Terms
We developed a Rankine-Hugoniot-Riemann (RHR) solverto solve systems of multi-dimensional balance laws withsource terms. The solver incorporates the source termsand the cross fluxes as a singular term in each finite volumecell, yielding a jump in the solution satisfying a Rankine-Hugoniot condition. The Riemann problems at the cellboundaries can then be solved using a conventional Rie-mann solver. The solver is shown to be of second order forphysically relevant equations.
Halvor LundDepartment of Energy and Process EngineeringNorwegian University of Science and [email protected]
Florian MullerInstitute of Fluid Dynamics
144 CS13 Abstracts
Patrick JennyInstitute of Fluid DynamicsETH [email protected]
CP5
Implementation of Multigrid Method for theNavier-Stokes Equations in Cuda
This contribution is concerned with implementation of theultigrid method with Vanka type smoother on GPU. Themethod is used to solve problem of 2D flow over an ur-ban canopy governed by Navier-Stokes equations and dis-cretized by means of the mixed finite element method.GPU version of the algorithm achieved speed-up 5 com-pared to parallel code based on OpenMP and 26 comparedto the sequential code.
Tomas OberhuberFaculty of Nuclear Sciences and Physical EngineeringCzech Technical University in [email protected]
Vladimir Klement, Vitezslav Zabka, Petr BauerFaculty of Nuclear Sciences and Physical EngineeringCzech [email protected], [email protected],[email protected]
CP5
Numerical Simulation of Two-Phase Incompress-ible Flows with Complex Interfaces
We consider a flow problem with two different immiscibleincompressible newtonian phases (fluid-fluid or fluid-gas).A standard model for this consists of the Navier-Stokesequations with viscosity and density coefficients that arediscontinuous across the interface and with a localized sur-face functional that models interface force. An example isthe Boussinesq-Scriven model for viscous interfaces. Wepresent finite element techniques for the the numericaltreatment of such flow problems. Topics that are brieflyaddressed are the XFEM and space-time DG method. Re-sults of numerical methods will be presented.
[1] S. Groß, A. Reusken, Numerical Methods for Two-phaseIncompressible Flows, Springer 2011.
Arnold ReuskenNumerical MathematicsRWTH Aachen University, Aachen, [email protected]
CP5
Mpp Limiter for Finite Difference Rk-WenoScheme with Applications in Vlasov Simulationsand Advection in Incompressible Flows
In this talk, we will give a maximum principle preservinglimiter for WENO scheme with high order explicit Runge-Kutta time discretization with applications in Vlasov andadvection of incompressible flow problems. We directlydealing with the numerical fluxes.Our approach is veryeasy, and can be applied to finite difference scheme, sothat it could be straightforwardly extended to high dimen-sional problems. Numerical experiments demonstrate the
high accuracy and high efficiency of our method.
Tao Xiong, Jingmei QiuUniversity of [email protected], [email protected]
Zhengfu XuMichigan Technological UniversityDept. of Mathematical [email protected]
CP6
O(N) Parallel Algorithm for Computing SelectedElements of the Inverse of a Gram Matrix in Elec-tronic Structure Calculations
Recently, there has been increased interest in developingO(N) algorithms for Density Functional First-Principlesmolecular dynamics at exascale. However, the energy func-tional formulation for general non-orthogonal orbitals re-quires the knowledge of selected elements of the inverse ofan associated global Gram matrix. We present a scalableparallel algorithm, based on domain decomposition and ap-proximate inverse techniques, for computing selected ele-ments of the inverse of the Gram matrix in electronic struc-ture calculations.
Daniel Osei-Kuffuor, Sebastien HamelLawrence Livermore National [email protected], [email protected]
Jean-Luc FattebertLawrence Livermore National [email protected]
CP6
An Efficient State-Space Based Method for DirectSimulation of Particle-Laden Turbulent Flows
We present a new numerical method for the coupled simula-tion of particulate poly-disperse flow using the probabilisticdescription (Boltzmann like), within turbulent flows. Themethod is stated in a purely deterministic Eulerian frame-work using the particle density function of the velocity-position state space of the particles while the carrier phaseobeys the incompressible Navier-Stokes equations. Appli-cation of the method will also be demonstrated by simu-lating a particle-laden turbulent flow using a parallel MPI-based solver implementation.
Carlos Pantano, Reetesh RanjanMechanical Science and EngineeringUniversity of Illinois at [email protected], [email protected]
CP6
On the Equivalence of P3M and NFFT-Based FastSummation
The P3M method combines the Ewald sum with the fastFourier transform in order to calculate long range Coulombinteractions for the classical N-body problem approxi-mately with O(N log N ) arithmetic operations. Duringthis talk, we show that P3M belongs to a class of fast sum-mation algorithms that are based on nonequispaced fastFourier transforms (NFFT). This connection yields newtheoretical and practical insights. Furthermore, we presentperformance results of our massively parallel implementa-
CS13 Abstracts 145
tion.
Michael PippigChemnitz University of TechnologyDepartment of [email protected]
CP6
Fast Evaluating Matern Covariance Kernel by aCartesianTreecode
Evaluating sums of multivariate Matern kernels is a com-mon computational task in statistical and machine learn-ing community. The quadratic computational complexityof the summation is a signicant barrier to practical ap-plications. We develop a Cartesian treecode algorithmto efficiently estimate sums of the Matern Kernel. Themethod uses a farfield Taylor expansion in Cartesian coor-dinates to compute particle-cluster interactions. The Tay-lor coefficients are obtained by recurrence relations whichallows effifficient computation of high order approxima-tions. In the serial code, for a given order of accuracy, thetreecode CPU time scales as O(N log N) and the memoryusage scales as O(N), where N is the number of particles.Parallel code also gives promising scale.
Lei WangArgonne National [email protected]
CP6
Uncertainty Quantification of Molecular Systems
We propose a method to quantify the uncertainties in themolecular dynamics (MD) system to enhance the efficiency,we employ �1 minimization to recover the coefficients ingeneral polynomial chaos (gPC) expansion given the priorknowledge that the coefficients are “sparse”. We imple-ment this method to quantify the dominant terms of gPCcoefficients for the MD measurement (e.g.,viscosity, diffu-sivity, etc.) This method thoroughly exploits informationfrom limited simulations, hence it is very efficient.
Huan Lei, Xiu Yang, George KarnidakisBrown UniversityHuan [email protected], xiu [email protected],george [email protected]
CP6
A Faster Fft in the Mid-West
We describe in this paper an FFT library that was buildwhile paying special attention to locality. The libraryachieves significantly better performance than FFTW, forlong vectors. Unlike FFTW or Spiral, the recursive decom-position of the FFT is not created by a library generator;it is created by macro expansion that has a few selectableparameters. This provides an interface that can be moreeasily modified by users
Alexander YeeUniversity of Illinois [email protected]
Marc SnirUniversity of Illinois at [email protected]
CP7
Efficient Computation of Eigenpairs for LargeScale-free Graphs
Large-scale network and data analysis is an area of greatimportance. The extreme eigenvalues and eigenvectors of-fer useful insights, but are difficult (expensive) to computefor large problems. We present a computational frameworkfor large spectral graph computations based on the Trilinostoolkit. We study the effectiveness of several eigensolversand preconditioners. Preliminary results indicate graph-based preconditioners can be highly effective.
Erik G. BomanSandia National Labs, NMScalable Algorithms [email protected]
Karen D. Devine, Richard B. LehoucqSandia National [email protected], [email protected]
Nicole SlattengrenSandia National [email protected]
Kevin DeweeseUC Santa Barbarana
CP7
Versatile Batch QR Factorization on GPUs
This talk describes a versatile, GPU-based BatchQR li-brary that performs the QR factorization of an array ofmoderately sized, dense matrices of variable dimensions.It addresses a growing need for batched matrix computa-tions on the GPU as opposed to single, large matrix fac-torizations. The library exploits data as well as task-levelparallelism by decomposing each dense QR into sets of in-dependent tasks, which are performed entirely on the GPU.
Sharanyan Chetlur, Nuri YeralanUniversity of [email protected], [email protected]
CP7
Utilizing Slater Matrix Properties to Design BetterPreconditioners for Quantum Monte Carlo Meth-ods
Quantum Monte Carlo (QMC) methods are often used tosolve problems involving many body systems. In imple-menting QMC methods, sequences of Slater matrices areconstructed. Solving linear systems for each matrix is amajor part of the cost. Currently, preconditioned linearsolves are utilized, reducing this cost from O(n3) to O(n2)power sweep (n updates). We analyze the properties ofSlater matrices, attempting to further reduce the cost. Di-rect solvers will also be examined.
Arielle K. Grim McnallyVirginia TechEric de [email protected]
Eric De SturlerVirginia Tech
146 CS13 Abstracts
CP7
Augmenting and Shifting in a Restarted LanczosBidiagonalization Method
The Lanczos bidiagonalization method can be used to solveSVD and least squares problems. Generally, the methodrequires restarting. Two restarted methods for SVD prob-lems are implicit shifting and augmenting the subspaceafter each restart. In this talk, we examine the Krylovsubspaces that result from the bidiagonalization processapplied in SVD and least squares schemes using harmonicRitz values as shifts and harmonic Ritz vectors as augment-ing vectors. Computed examples and results are presented.
Daniel J. Richmond, James BaglamaUniversity of Rhode [email protected], [email protected]
CP7
Multi-Preconditioning Gmres
Standard Krylov subspace methods for the solution of non-symmetric linear systems only allow the user to choose asingle preconditioner, although in many situations theremay be a number of possibilities. Here we describe multi-preconditioned GMRES, which allows the use of more thanone preconditioner. These multiple preconditioners canbe combined in an optimal manner. We give some the-oretical results, propose a practical algorithm, and presentnumerical results from problems in domain decompositionand PDE-constrained optimization. These numerical ex-periments illustrate the applicability and potential of themulti-preconditioned approach.
Daniel B. SzyldTemple UniversityDepartment of [email protected]
Chen GreifDepartment of Computer ScienceThe University of British [email protected]
Tyrone ReesRutherford Appleton Lab, [email protected]
CP7
Multifrontal Sparse QR Factorization on GPU
GPU devices are growing in popularity as heterogeneousarchitectures become prevalent in scientific computing. Wediscuss a multifrontal sparse QR factorization method im-plemented on GPU accelerators. Our method exploitstask-based and data-based parallelism across the elimina-tion tree with frontal matrices of differing sizes, within con-tribution block assembly, and within the numeric factor-ization. Finally, we show performance results for problemsfound in the UF Sparse Matrix collection.
Nuri YeralanUniversity of [email protected]
CP8
A Parallel Method for Computing Visibility Re-gions and Its Application to Dynamic CoverageControl
We present a parallel algorithm for computing visibilityregions in complex terrain and explore its application tothe problem of dynamic coverage control with a networkof near-ground mobile sensor agents. The highly scalablemethod makes it tractable to compute the visibility regionof each sensor and thus account for possible terrain ob-structions. We show the scalability of this algorithm onGraphics Processing Units and provide coverage controlsimulations employing our novel sensor model.
Miles L. DetrixheUniversity of California Santa [email protected]
CP8
State of the Art Hifoo: H∞ Controller Optimiza-tion for Large and Sparse Systems
We present a new state of the art version of HIFOO,a MATLAB package for optimizing H∞ and H2 con-trollers for linear dynamical systems, supporting simulta-neous multiple plant stabilization. Previous versions of HI-FOO have generally been limited to smaller scale problems,due to the high asymptotic cost of computing the H∞ normof the transfer matrix. However, new sparse methods forcomputing the H∞ norm allow HIFOO to extend efficientlyto large and sparse systems.
Tim MitchellCourant Institute of Mathematical SciencesNew York [email protected]
Michael L. OvertonNew York UniversityCourant Instit. of Math. [email protected]
CP8
Hybrid Functions Approach for Optimal ControlProblems
In this work, we present a new direct computationalmethod to solve optimal control problems. The approach isbased on reducing the optimal control problems into a set ofalgebraic equations by first expanding the required solutionas a hybrid function with unknown coefficient. These hy-brid functions, which consist of block-pulse functions andBernoulli polynomials are first introduced. Numerical ex-amples are included to demonstrate the applicability andthe accuracy of the proposed method and a comparison ismade with the existing results.
Mohsen RazzaghiMississippi State [email protected]
CP8
Multigrid Solution of a Distributed Optimal Con-trol Problem Constrained by a Semilinear EllipticPde
We study a multigrid solution strategy for a distributed op-
CS13 Abstracts 147
timal control problem constrained by a semilinear ellipticPDE. Working in the discretize-then-optimize framework,we solve the reduced optimal control problem using New-tons method. Further, adjoint methods are used to com-pute matrix-vector multiplications for the reduced Hessian.In this work we introduce and analyze a matrix-free multi-grid preconditioner for the reduced Hessian which provesto be of optimal order with respect the discretization.
Jyoti SaraswatDepartment of Mathematics & StatisticsUniversity of Maryland Baltimore [email protected]
Andrei DraganescuDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
CP8
Multigrid Preconditioners for Optimal ControlProblems in Fluid Flow
We construct multigrid preconditioners to accelerate thesolution process of optimal control problems constrained bythe Stokes/Navier-Stokes equations. Our approach for theStokes control problem is to eliminate the state and adjointvariables from the optimality system and to construct effi-cient multigrid preconditioners for the Schur-complementof the block associated with these variables. Similar pre-conditioners are constructed for the reduced Hessian in theNewton-PCG method for the optimal control of the sta-tionary Navier-Stokes equations.
Ana Maria SoaneUniversity Of Maryland, Baltimore CountyDepartment of Mathematics and [email protected]
Andrei DraganescuDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
CP8
Thermoelastic Shape Optimization
Shape optimization is often devoted to systems with onephysical discipline. This talk discusses specific problems,when dealing with shape optimization problems in multi-physics systems. In particular, the coupling of the heatequation with the elasticity equation is studied as it is thecase in standard heating devices made from iron. Numeri-cal approaches to the optimization of the iron body of hotplates under the coupled multi-physics problem are pre-sented together with computational results.
Heinz ZornUniversity of [email protected]
Volker H. SchulzUniversity of TrierDepartment of [email protected]
CP9
Uncertainty Quantification for Subsurface Flow
Models Using Bayesian Nested Sampling Algo-rithm
We investigate an efficient sampling algorithm known asnested sampling, which can simultaneously perform thetasks of parameter estimation, sampling from the posteriordistribution for uncertainty quantification, and estimatingthe Bayesian evidence for model comparison. Nested sam-pling has the advantage of generality of application, andcomputational feasibility. In this talk, we report the firstsuccessful application of nested sampling for calibrationand model selection of several nonlinear subsurface flowproblems.
Ahmed H. ElSheikhDepartment of Earth Science and EngineeringImperial College London, South Kensington [email protected]
Mary F. WheelerCenter for Subsurface ModelingUniversity of Texas at [email protected]
Ibrahim HoteitKing Abdullah University of Science and Technology(KAUST)[email protected]
CP9
Variable Dimensional Bayesian Elastic Wave-FieldInversion of a 2D Heterogeneous Media
This paper introduces a methodology to infer the spatialvariation of the elastic characteristics of a 2D heteroge-neous earth model via Bayesian approach, given the probedmedium’s response to interrogating SH waves measured atthe surface. A reduced dimension, self regularized treat-ment of the inverse problem using partition modeling isintroduced, where the velocity field is discretized by a vari-able number of disjoint regions defined by Voronoi tessel-lations. A provided synthetic study indicates the strengthof the technique.
Saba S. EsmailzadehZachry Department of Civil Engineering,Texas A&M University, College Station, TX, USA [email protected]
Zenon Medina-CetinaZachry Department of Civil EngineeringTexas A&M [email protected]
Jun Won KangDepartment of Civil EngineeringNew Mexico State University, NM 88003, [email protected]
Loukas F. KallivokasThe University of Texas at [email protected]
CP9
Optimization in the Presence of Uncertainty: Ap-plications in Designing Random Heterogeneous
148 CS13 Abstracts
Media
The present paper discusses a sampling framework that en-ables the optimization/design of complex systems charac-terized by high-dimensional uncertainties and design vari-ables. We are especially concerned with problems relat-ing to random heterogeneous materials where uncertaintiesarise from the stochastic variability of their properties. Inparticular, we reformulate topology optimization problemsto account for the design of truly random composites. Themethodological advances proposed in this paper allows theanalyst to identify several local maxima that provide im-portant information with regards to the robustness of thedesign. We further propose a principled manner of intro-ducing information from approximate models that can ul-timately lead to further reductions in computational cost.
Phaedon S. KoutsourelakisHeriot Watt [email protected]
CP9
Adaptive Total Variation Regularization in ImageProcessing
We consider an adaptive version of the well-known totalvariation (TV) based regularization model in image pro-cessing. Adaptive TV based schemes try to avoid the stair-casing artifacts associated with TV regularization basedenergy minimization models. An algorithm based on amodification of the split Bregman technique proposed by[Goldstein and Osher, The split Bregman algorithm for L1
regularized problems, SIAM J. Imaging Sci. 2009], can beused for solving the adaptive case. Convergence analysis ofsuch an alternating scheme is proved using the Fenchel du-ality and a recent result of [Svaiter, On weak convergenceof the Douglas-Rachford method, SIAM J. Control Op-tim. 2011] on the weak convergence of Douglas-Rachfordsplitting method. We will provide detailed comparative ex-perimental results using the modified split Bregman, dualminimization and additive operator splitting for the gradi-ent descent scheme for the TV diffusion equation to high-light the efficiency of adaptive TV based schemes for imagedenoising, decomposition and segmentation problems.
Surya Prasath, Kannappan PalaniappanUniversity of [email protected], [email protected]
CP9
Transient Hydraulic Tomography Using the Geo-statistical Approach
Transient Hydraulic Tomography is a well known methodfor characterizing aquifers that consists of measuring pres-sure (head) responses to estimate important parameters(eg. conductivity, specific storage). We discuss variouscomputational issues regarding their large-scale implemen-tation using the Geostatical approach and techniques forcomputing the uncertainty associated with the estimate.
Arvind SaibabaStanford [email protected]
Peter K. KitanidisDept. of Civil and Environmental EngineeringStanford University
CP9
Quantification of the Loss of Information inSource Attribution Problems in Global Atmo-spheric Chemical Transport Models
It is of crucial importance to be able to identify the locationof atmospheric pollution sources in our planet in order toevaluate global emissions policies like the Kyoto protocol.It is shown that the ability to successfully use the adjointmethod to reconstruct the location and magnitude of aninstantaneous source in global chemical transport modelsis compromised in relevant global atmospheric time scales.
Mauricio SantillanaHarvard University, School of Engineering and [email protected]
CP10
Computing Solutions of Dynamic SustainabilityGames
We introduce a dynamic Nash game among firms harvest-ing a renewable resource and propose a differential varia-tional inequality (DVI) framework for modeling and solv-ing such game. We consider the application of the proposedframework to a sustainable halibut fishery. The problemis modeled using the DVI, which in turn converted to afixed-point problem that allows computationally efficientalgorithms. A numerical example is presented to show theuse of our proposed framework.
Sung ChungNew Mexico Institute of Mining and TechnologyNew Mexico [email protected]
CP10
How to Deal with Dynamic Contact Angles in Mov-ing Contact-Line Simulations
We present an efficient algorithm within phase field frame-work on how to impose dynamic contact-angle boundaryconditions for wall-bounded liquid-gas flows. Our strategyconsists of two components: (i) we ignore the boundaryconditions and re-formulate the 4-th order Cahn-Hilliardequation into two nominally de-coupled Helmholtz typeequations; (ii) we treat the dynamic contat-angle bound-ary conditions in such a manner that the two Helmholtztype equations are truly de-coupled. We will also present astrategy on how to treat the variable-density Navier-Stokesequations such that only constant and time-independentcoefficient matrices will result from the discretization, eventhough variable density and variable viscosity are involvedin the governing equations. The overall method can dealwith both dynamic and static contact angles for movingcontact line problems involving large density ratios, andthe computations for all flow variables have been com-pletely de-coupled. Simulations of air-water two-phaseflows involving solid walls of different hydrophobicities willbe presented.
Suchuan DongPurdue [email protected]
CS13 Abstracts 149
CP10
On An Implicit Time Integration Method for WavePropagations
The Bathe method is known to be very effective for fi-nite element solutions of linear and nonlinear problems instructural dynamics. In this presentation, we present dis-persion properties of the scheme and show that its desiredstability and accuracy characteristics for structural dynam-ics are also valuable for the solution of wave propagationproblems. A dispersion analysis using the CFL number ispresented and simple examples are solved to illustrate thecapabilities of the scheme for wave propagations.
Gunwoo Noh, Seounghyun HamDept. of Mechanical Engineering, [email protected], [email protected]
CP10
Numerical Simulation of the Damping Behavior ofParticle-Filled Hollow Spheres
For many industrial applications, light-weight materialsplay an important role. A new material developed atthe Fraunhofer Institute for Manufacturing Technology andAdvanced Materials uses particle filled hollow spheres em-bedded into a sandwich structure to dampen vibrationthrough friction. We present a numerical simulation ofsuch a sphere using a discrete element model. For thispurpose, techniques from molecular dynamics are adaptedand extended correspondingly. First results illustrating thedamping behavior are shown.
Tobias SteinleUniversitat [email protected]
Jadran VrabecUniversity of [email protected]
Andrea WaltherUniversitat [email protected]
CP10
Hybrid Parallel Transistor-Level Full-Chip CircuitSimulation
The emergence of multicore and manycore processors hasprovided an opportunity for application-specific simulationtools to take advantage of potential intra-node parallelism.CAD applications, being fundamental to the electronic de-sign automation industry, need to harness the availablehardware resources to be able to perform full-system sim-ulation for modern technology nodes. We will present ahybrid (MPI+threads) approach for performing paralleltransistor-level transient circuit simulation that achievesscalable performance for some challenging large-scale inte-grated circuits.
Heidi K. Thornquist, Siva Rajamanickam, Mike HerouxSandia National [email protected], [email protected],[email protected]
Erik G. BomanSandia National Labs, NMScalable Algorithms Dept.
CP10
Numerical Optimization of Instrument Settings inReference Material Certification
For metrological purposes, the National Institute of Stan-dards and Technology (NIST) provides over 1200 StandardReference Materials (SRM) of the highest quality. Devel-oping SRMs to meet todays technological needs demandsever-increasing measurement accuracy and precision. Tra-ditional methods of hand tuning chemical spectrometers byexperienced users are no longer adequate. We have devel-oped a new method for optimizing instrument settings thatdoes not reply on having an expert instrument user. Wedemonstrate how a noisy objective function is optimizedthrough a novel gradient approximation which employs ana priori estimate of noise levels.
William E. WallaceNational Institute of Standards and [email protected]
Anthony KearsleyMathematical and Computational Sciences DivisionU.S.A. National Institute of Standards and [email protected]
CP11
Multi-Scale Modelling of Droplets Deformation
Deformation and fragmentation of droplets due to growthof hydrodynamical instabilities are studied using a finitevolume-volume of fluid (FV-VOF) numerical technique forfalling droplets and droplets in a stream. Problems aresolved in both two and three dimensions. Results werecompared with stability analyses that provided based onthe linearized Navier-Stokes equations. Reasonable agree-ment is observed between the numerical and analytical so-lutions for the most amplified wave numbers, growth rateetc.
Maziyar JalaalDepartment of Mechanical Engineering, The University ofBritish Columbia, Vancouver, Canada.m [email protected]
Kian MehravaranSchool of Engineering, The University of BritishColumbia,Kelowna, [email protected]
CP11
Domain Decomposition Preconditioning for Varia-tional Monte Carlo on Insulators
The main cost of the Variational Monte Carlo(VMC) meth-ods is in constructing a sequence of Slater matrices andcomputing the ratios of determinants for successive Slatermatrices. The scaling of constructing Slater matrices forinsulators has been improved so that its cost is now linearwith the number of particles. To improve computing theratios of determinants, we employ preconditioned Krylovsubspace iteration methods. The main work here is us-ing domain decomposition preconditioner to solve the ra-tios of determinants iteratively. Iterative Krylov methodslike GMRES are performed. Since there is one supposed
150 CS13 Abstracts
moving particle in every attempt of the VMC methods,two-subdomain decomposition is used most of the times.One subdomain is the local domain around the movingparticle, the other includes the remainning particles andorbitals. To make the iteration robust and stable, an ef-fective reordering of Slater matrices and incomplete LUdecomposition are used .
Ming Livirginia polytechnic institute and state [email protected]
Eric De SturlerVirginia [email protected]
Arielle Grim-McNallyvirginia polytechnic institute and state [email protected]
CP11
Coarse-Grained Stochastic Particle-BasedReaction-Diffusion Simulation Algorithm
The stochastic, diffusive motion of molecules influences therate of molecular encounters and hence many biochemi-cal processes in living cells. Tiny time steps are typicallyrequired to accurately resolve the probabilistic encounterevents in a simulation of cellular signaling, creating thedilemma that only simplistic models can be studied at cell-biologically relevant timescales. Particular solutions of thediffusion equation, Green’s functions, allow for larger timesteps and more efficient simulations by detecting otherwiseunnoticed encounter events. However, previous implemen-tations of Green’s functions based approaches were not gen-erally applicable to reaction-diffusion problems. We gen-eralized the method to reflect the physiologically relevantcase of reversible, partially diffusion-controlled reactions.In my talk, I will discuss how the approach can be appliedin computational cell biology, taking advantage of an effi-cient approximation technique that allows us to include themathematically challenging case of 2D reaction-diffusionsystems on membranes as well. The resulting algorithm iseasily implementable, flexible and efficient and provides acoarse-grained but nevertheless detailed, stochastic repre-sentation of biochemical processes.
Thorsten Prustel, Martin Meier-SchellersheimLaboratory of Systems Biology (LSB)NIAID, [email protected], [email protected]
CP11
A Three-Dimensional Domain DecompositionMethod for Large-Scale DFT Electronic StructureCalculations
We present a 3D domain decomposition method for atomsand grids. A modified recursive bisection method is devel-oped based on inertia tensor moment to reorder the atomsalong a principal axis for atom decomposition. We definefour data structures for grid decomposition and propose amethod for solving the Poisson equation using FFT withcommunications minimized. Benchmark results show thatthe parallel efficiency at 131,072 cores is 67.7% comparedto the baseline of 16,384 cores.
Truong Vinh Truong DuyResearch Center for Simulation Science
Japan Advanced Institute of Science and [email protected]
Taisuke OzakiJapan Advanced Institute of Science & [email protected]
CP11
Multiscale Smooth Dissipative Particle Simulationof Non-Isothermal Flows
Smooth Dissipative Particle Dynamics (SDPD) is a multi-scale mesh-free method that provides a bridge between thecontinuum and molecular scales. SDPD is thermodynam-ically consistent, involves realistic transport coefficients,and includes fluctuation terms. The SDPD is implementedin our work for arbitrary 3D geometries with a methodol-ogy to model solid wall boundary conditions. The entropyequation is implemented with a velocity-entropy Verlet in-tegration algorithm. Flows with heat transfer are simu-lated for verification of the SDPD. We present also theself-diffusion coefficient derived from SDPD simulations forgases and liquids. Results show the scale dependence onSDPD particle size.
Jun Yang, Nikolaos Gatsonis, Raffaele PotamiWorcester Polytechnic [email protected], [email protected], [email protected]
CP11
Domain Decomposition Based Jacobi-Davidson Al-gorithm for Quantum Dot Simulation
A domain decomposition based Jacobi-Davidson algorithmis proposed to find the interior eigenpairs of large sparsepolynomial eigenvalue problems arising from pyramidalquantum dot simulation. We apply domain decomposi-tion method both to improve the convergence of correc-tion equation that is the most expensive part in Jacobi-Davidson algorithm and to obtain initial guesses with richinformation on target eigenpairs. The numerical resultsshow that our algorithm is effective with good scalabilityon supercomputer.
Tao ZhaoDepartment of Computer ScienceUniversity of Colorado [email protected]
Feng-Nan HwangNational Central [email protected]
Xiao-Chuan CaiDepartment of Computer ScienceUniversity of Colorado [email protected]
CP12
Mining Periodic Patterns in Digital Footprint’Event’ Data via Empirical Mode Decomposition(EMD)
Digital trace data can often be interpreted as a temporal se-ries of events (such as the times at which emails are sent orposts on on-line social networks made). We present a noveltechnique for the categorisation of individuals according tothe temporal characteristics inherent in personal ’event’
CS13 Abstracts 151
data. Our method, exemplified on synthetic and real datasets, is underpinned by projecting individuals’ data onto abasis of pseudo-periodic intrinsic mode functions obtainedfrom EMD.
Duncan S. Barrack, Chienmin ChuangHorizon Digital Economy Research InstituteUniversity of [email protected],[email protected]
Keith Hopcraft, Simon PrestonSchool of Mathematical SciencesUniversity of [email protected],[email protected]
James Goulding, Gavin SmithHorizon Digital Economy Research InstituteUniversity of [email protected],[email protected]
CP12
Recent Advances in Moment-Matching Model Or-der Reduction for Maxwell’s Equations
We present some new results for the application of moment-matching methods in the field of model order reduction forMaxwell’s equations. In detail, a different interpretation ofexisting heuristic error estimations in combination with anadaptive expansion point selection will be presented. Fur-thermore, we will show how to deal with the discrete, di-vergence conditions during the computation of the reducedorder model.
Andre BodendiekInstitut Computational MathematicsTechnical University [email protected]
CP12
A Reduced Basis Kalman Filter for ParametrizedParabolic Partial Differential Equations
Realizing a Kalman filter for solution estimations ofstochastically forced partial differential equations (SF-PDEs) is often infeasible due to the high computationalcost. Hence, we propose a reduced basis (RB) Kalman fil-ter for parametrized linear SFPDEs giving solution estima-tions in real-time. Ingredients of our contribution are theRB-construction via KL-expansion and Duhamel’s princi-ple, certification by derivation of statistical informationabout error bounds for stochastically forced RB-modelsand application to a stochastically disturbed parametrizedheat equation.
Markus Dihlmann, Bernard HaasdonkUniversity of [email protected],[email protected]
Anthony T. PateraMassachusetts Institute of TechnologyDepartment of Mechanical [email protected]
CP12
Multi-Fidelity Modeling of Solar Irradiance
We describe our approach to enhancing the production-grade solar irradiance model, based on NASA/NOAAsatellite imagery, by combining it with the real-time datafrom PV installations and weather sensors monitored byLocus Energy. The implications of leveraging ground datafor thousands of monitored PV sites with the current short-term solar irradiance forecasting model, based on opticalflow computations, is also discussed. The multi-fidelityforecasting model was calibrated and validated againstSurfRad network’s solar data.
Sergey KoltakovStanford [email protected]
CP12
Efficient Reduced Order Models for Solving Prob-lems in Fluid Mechanics Using Stochastic Colloca-tion
In this work, we propose a novel computational algorithmthat employs proper orthogonal decomposition (POD) andsparse grid stochastic collocation method to approximatethe solution to stochastic partial differential equations de-scribing fluids. In particular, we will describe the stabilityand accuracy of the proposed parallel algorithm leading toa reduction of the overall computational cost. Numericalresults validating the performance of the proposed methodwill be presented for benchmark problems.
Maziar RaissiMathematical SciencesGeorge Mason [email protected]
Padmanabhan SeshaiyerGeorge Mason [email protected]
CP12
Efficient New Greedy Algorithms for Reduced Ba-sis Methods
We propose two new algorithms to improve greedy sam-pling of high-dimensional functions. While the techniqueshave a substantial degree of generality, we frame the dis-cussion in the context of reduced basis methods for high-dimensional parametrized functions. The first algorithm,based on a saturation assumption of the error in the greedyalgorithm, is shown to result in a significant reduction ofthe workload over the standard greedy algorithm. In afurther improved approach, this is combined with an al-gorithm in which the train set for the greedy approach isadaptively sparsefied and enriched. A safety check step isadded at the end of the algorithm to certify the quality ofthe sampling.
Shun Zhang, Jan HesthavenDivision of Applied MathematicsBrown UniversityShun [email protected], jan [email protected]
Benjamin StammUniversity of California, Berkeley, [email protected]
152 CS13 Abstracts
MS1
Discontinuous Galerkin Methods in ConvectionDominated Application Models
We present discontinuous Galerkin (DG) methods arisingin convection dominated application models. The firstmodel is the DG Advanced Circulation model for hurri-cane storm surge, where recent additions to the code in-clude layered dynamic sediment, local time-stepping al-gorithms, and non-hydrostatic pressure formalisms. Thesecond model we address, is a reduced magnetohydrody-namic code for solving convective scrape-off layer dynamicsin tokamak edge plasmas.
Craig MichoskiUT [email protected]
Clint DawsonInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
Kyle T. MandliUniversity of Texas at [email protected]
Francois WaelbroeckUT [email protected]
MS1
Numerical Simulation of Cylindrical SolitaryWaves in Periodic Media
We study the behavior of nonlinear waves in a two-dimensional medium with material properties that varyperiodically in space. The system considered has no dis-persive terms; however, the material heterogeneity leads toreflections and effective dispersion. This dispersion com-bined with the nonlinearity of the system breaks the initialprofile into solitary waves. Interaction of two solitary wavesis studied. The results are typical of solitary waves; the twopulses retain their identity after the interaction.
Manuel Quezada De [email protected]
MS1
Numerical Inverse Scattering: Uniformly AccurateResolution of Dispersion
The Riemann–Hilbert formulations of the Korteweg-deVries and Nonlinear Schrodinger equations have provedto be numerically valuable. Borrowing ideas from themethod of nonlinear steepest descent, the resulting numer-ical schemes are seen to be asymptotically reliable. Wederive sufficient conditions for a uniformly accurate nu-merical method. We resolve highly-oscillatory solutions ofdispersive equations on unbounded domains with uniformaccuracy.
Thomas D. TrogdonDepartment of Applied MathematicsUniversity of [email protected]
Sheehan OlverThe University of [email protected]
MS1
Grid Resolution Requirements in NonhydrostaticInternal Wave Modeling
Abstract not available at time of publication.
Sean VitousekEnvironmental Fluid Mechanics [email protected]
MS2
Kalman Smoothing Approach for 4D Imaging
We present a formulation for modeling of large scale im-ages that change dynamically in time. We use an optimiza-tion formulation of Kalman smoothing, combining a slowlyvarying dynamic prior with PDE constrained optimizationused in 3D imaging. We discuss computational aspects andshow how matrix free algorithms can be used to solve theproblem.
Aleksandr Aravkinthe University of British [email protected]
Eldad HaberDepartment of MathematicsThe University of British [email protected]
MS2
4D Biomedical Imaging and Compressed Sensing
In state-of-the-art biomedicine, 4D time-evolving processesarise at different scales in space and time, e.g. researchersstudy cell migration in high-resolution/-speed microscopyor motion of organs (heart, lung,) in tomography. Theseproblems offer interesting challenges for modeling and non-linear large-scale PDE optimization. The goal of this talkis to present recent modeling and numerics for 4D imagingsubject to transport equations. Moreover, we will focuson the design question of compressed sensing in space andtime.
Christoph BruneDepartment of MathematicsUniversity of California Los [email protected]
MS2
Electromagnetic Imaging of Subsurface Flow
Abstract not available at time of publication.
Eldad HaberDepartment of MathematicsThe University of British [email protected]
MS2
Reliability of 4D Image Based Simulations of Blood
CS13 Abstracts 153
Flow in Arteries
Integration of 4D images and computational fluid dynam-ics is an effective approach for the numerical simulation ofblood in moving arteries. The reliability of this approachdepends on the quality of the images and the registrationprocess tracking the wall motion in time. Here we discussthe accuracy of the hemodynamics simulations in a patientspecific aorta for different sampling frequencies of the im-ages. The relevance of appropriate selection of boundaryconditions will be addressed.
Alessandro VenezianiMathCS, Emory University, Atlanta, [email protected]
Tiziano PasseriniDepartment of Math & CSEmory [email protected]
Marina PiccinelliDepartment of Math&CSEmory [email protected]
MS3
An Overview of Silent Data Corruption and its Ef-fects on Linear Solvers
Exascale machines are expected to challenge current HPCcomputing methods. Fault rates due to high componentcount will likely be exacerbated by decreasing componentreliability. To help algorithm developers prepare for a fu-ture that may include a constant rate of faults, I will dis-cuss current hardware trends, how they relate to existingfailure modes, and how simulations of faults affect existingcodes. I will end with a few glimmers of hope for futurearchitectures.
Sean BlanchardLos Alamos National [email protected]
Nathan DeBardelebenLos Alamos National [email protected]
MS3
Fault-tolerant Iterative Linear Solvers via SelectiveReliability
Protecting computations and data from corruption due tohardware errors costs energy. However, energy increasinglyconstrains modern computers. As processor counts con-tinue to grow, it will become too expensive to correct errorsat system levels, before they reach user code. We will showinstead that if the system provides a programming modelthat lets applications apply reliability only when and whereneeded, we can develop algorithms that compute the rightanswer despite hardware faults.
Mark HoemmenSandia National [email protected]
MS3
Fault Tolerant Qmr
As architectures are developed with smaller componentsoperating a lower voltages, soft errors (i.e., bit flips) be-come potentially worrying, and in fact may not even bedetected. Previous work has investigated the sensitivityseveral Krylov methods to soft errors. Here, we explorefault tolerance issues in QMR, compare its performancein the presence of bit flips, and introduce a fault tolerantvariant of QMR.
Victoria Howle, Ashley MeekTexas [email protected], [email protected]
Mark HoemmenSandia National [email protected]
MS3
Algorithmic Strategies for Soft-error Resilience inSparse Linear Solvers
Increasing rates of soft errors in many-core and multi-corearchitectures can adversely affect the accuracy and conver-gence of a wide variety of parallel numerical applications,such as, implicit and explicit methods for the solution ofpartial differential equations, the latter being particularlysusceptible to instabilities arising from soft errors. In thistalk we present methods to characterize the growth andpropagation of soft errors in sparse iterative kernels anddiscuss light-weight strategies to recover from the result-ing error states.
Radhika S. SaksenaPennsylvania State [email protected]
Manu ShantharamThe University of [email protected]
Padma RaghavanThe Pennsylvania State Univ.Dept of Computer Science [email protected]
MS5
Fast Implicit Maxwell Solver for Linear Wave Prop-agation in Cold Plasmas
We present a new implicit Maxwell solver, based onthe ‘method of lines transpose’ for the wave equation.The fields update satisfies a modified Helmholtz equation,which is solved by a Green’s function approach. Combin-ing an ADI procedure with an accelerated 1D method, thecomputational cost of multi-dimensional solvers is linearin the number of unknowns. A fully implicit exponentialdiscretization of the linear plasma response is employed,obtaining second order accuracy and unconditional stabil-ity.
Yaman GucluDepartment of MathematicsMichigan State [email protected]
Matthew F. Causley
154 CS13 Abstracts
New Jersey Institute of Technology (NJIT)[email protected]
Andrew J. Christlieb, Yingda ChengDepartment of MathematicsMichigan State [email protected], [email protected]
MS5
Discontinuous Galerkin Schemes for the (Gyro) Ki-netic Simulations of Plasmas
Recently developed discontinuous Galerkin (DG) algo-rithms are extended for the solution of kinetic and gyro-kinetic simulations of plasmas. The schemes are applicableto a wide variety of systems describable by a Hamiltonianevolution equation, coupled to field equations. Conserva-tion properties of the algorithms, specially for energy, mo-mentum and L2 norm are discussed. Special basis func-tions for velocity space are explored in the context of rep-resenting small deviations from a Maxwellian distributionefficiently.
Ammar HakimPlasma Physics LaboratoryPrinceton [email protected]
MS5
A High-order Unstaggered Constrained TransportMethod for the 3D Ideal MagnetohydrodynamicEquations based on the Method of Lines
We present finite volume methods for the 3D ideal mag-netohydrodynamic (MHD) equations using a constrainedtransport (CT) technique, in which an evolution equationfor the magnetic potential is solved during each time step.A divergence free magnetic field is obtained by computingthe curl of the magnetic potential. In contrast to the 2Dcase, the evolution equation for the magnetic potential isnot unique but instead depends on the choice of the gaugecondition. By using the so-called Weyl gauge, we obtain aweak hyperbolic system for the evolution of the magneticpotential that requires an appropriate discretization.
Christiane HelzelDepartment of [email protected]
James A. RossmanithIowa State UniversityDepartment of [email protected]
Bertram TaetzDepartment of MathematicsRuhr-University [email protected]
MS5
Two-Fluid Higher-Moment Modeling of Fast Mag-netic Reconnection
Higher-moment fluid models efficiently approximate kineticmodels for moderate Mach and Knudsen numbers. We seekminimal modeling requirements to resolve fast magneticreconnection with a fluid model. For 2D problems invariant
under 180-degree rotation about the out-of-plane axis, non-singular steady-state magnetic reconnection is impossiblein a conservative fluid model that lacks heat flux. To avoidthe difficulties of diffusive closures, we use higher-momenthyperbolic models and compare simulations of magneticreconnection with kinetic simulations.
Evan A. [email protected]
MS6
Parallel Multiscale Modeling of High Explosives forTransportation Accidents
The Uintah Computational Framework provides a scalablearchitecture for simulating fires and detonation of largearrays of explosives in transportation accidents. The sim-ulations utilize Kraken for these complex fluid-structureinteraction problems. Uintah combines validated reactionmodels with robust multi-material mechanics and scalingto 250k cores to provide the predictive capability neededto improve transportation safety on the nation’s highwaysand railways for the first time.
Jacqueline BeckvermitDepartment of ChemsitryUniversity of [email protected]
Qingyu MengSCI InstituteUniveristy of [email protected]
Todd HarmanDepartment of Mechanical EngineeringUniversity of [email protected]
Martin BerzinsScientific Computing and Imaging InstituteUniversity of [email protected]
Charles WightDept. of ChemistryUniversity of [email protected]
MS6
Simulations of the Universe: Toward PetascaleCosmology
Numerical simulations are now the dominant tool in the-oretical cosmology. As advances in computer power andsoftware algorithms continue into the petascale regime, itbecomes possible to simulate the entire visible universeat high mass and spatial resolutions. We will present apetascale-optimized version of the gravity/hydrodynamicscode GADGET using an efficient and low-risk incremen-tal strategy. Hybrid shared memory, load balancing andother improvements allow the new p-GADGET to run onpetascale systems.
Tiziana DiMatteoDepartment of PhysicsCarnegie Mellon [email protected]
CS13 Abstracts 155
MS6
Petascale Challenges in Astrophysics: Core-Collapse Supernovae
Understanding the explosive deaths of massive stars hasbeen a major focus in stellar astrophysics for five decades.Among many other roles, these explosions are the source ofmany of the elements essential for life. Despite ∼50 yearsof concerted effort, we still do not know how massive starsexplode. I will review the multi-physics, multi-scale, multi-dimensional challenges of core-collapse supernova modelingand describe the massively parallel, state-of-the-art effortsto unveil the mechanism of explosion.
Joshua C. DolenceDepartment of Astrophysical SciencesPrinceton [email protected]
MS6
Correlated Electronic Wave-function Calculationsfor Large Molecules at the Petascale Level
We describe how correlated electronic wave function cal-culations on large molecular systems may be carried outusing a massively parallel algorithm. The presented algo-rithm employs two levels of parallelization and allows forefficient parallization over many thousand nodes for largemolecule systems. With the presented algorithm the high-order polynomial scaling of a standard correlation wavefunctions calculations is reduced to linear, while error con-trol compared to a conventional calculation is maintained.
Kasper KristensenDepartment of ChemistryAarhus [email protected]
MS7
Random Light Polarization Dynamics in an ActiveOptical Medium
Resonant interaction of light with a randomly-prepared,lambda-configuration active optical medium is describedby exact solutions of a completely-integrable, system ofrandom partial differential equations, thus combining theopposing concepts of integrability and disorder. An opticalpulse passing through such a material will switch randomlybetween left- and right-hand circular polarizations. Exactprobability distributions of the electric-field envelope vari-ables describing the light polarization and their switchingtimes will be presented together with stochastic simula-tions.
Gregor KovacicRensselaer Polytechnic InstDept of Mathematical [email protected]
Katherine NewhallCourant Institute of Mathematical ScienceNew York [email protected]
Peter R. KramerRensselaer Polytechnic InstituteDepartment of Mathematical [email protected]
Ildar R. GabitovDepartment of Mathematics, University of [email protected]
Ethan AkinsUniversity of California, [email protected]
MS7
Thermally Induced Magnetization Reversals
Driving nanomagnets by spin-polarized currents offers ex-citing prospects in magnetoelectronics, but the actual re-sponse of the magnets to such currents remains poorlyunderstood. The underdamped dynamics require exceed-ingly long simulations of stochastic trajectories to computethermally-induced reversal times and other relevant fea-tures. Rather, we derive an averaged equation describingthe diffusion of energy on a graph that could be used toanalyze the behavior of a broad range of other nongradient,underdamped systems.
Katherine NewhallCourant Institute of Mathematical ScienceNew York [email protected]
Eric Vanden-EijndenCourant [email protected]
MS7
Limitations in Reduction of Wind Power Intermit-tency with Storage Technologies
Stochastic variations and unpredictability of wind energyare the major concerns of power industry and hinder thewide scale adoption of wind power. Compensation of shortterm variability is one of the major challenges that the in-dustry will face in the coming years. Our study focuseson statistical analysis of fluctuations of wind power on theminute to hour time scales. Using the publicly availablewind measurement data we show that the statistics of fluc-tuations is strongly non-Gaussian and highly correlatedin this time frame. Specifically we show that traditionalGaussian can underestimate the probability of rare eventsby several orders of magnitude. In the second part of ourwork we analyze the potential impact of advanced controland storage technologies in reducing the intermittency ofwind power. Using the convex optimization techniques westudy the theoretical limits on the performance of storagetechnologies. Specifically we analyze the interplay betweenthe statistics of electric power fluctuations and the char-acteristics of storage available in the system. We quantifythe trade-off between the reduction in power intermittency,storage capacity, and charging rate. In the end we presenta general approach to the intermittency mitigation prob-lem that incorporates multiple objectives and system con-straints.
Konstantin TuritsynMassachusetts Institute of [email protected]
156 CS13 Abstracts
MS7
Advances in Parallel Tempering
Parallel tempering (replica exchange Monte Carlo) is animportant tool for exploring free-energy landscapes withmany metastable minimas. For nearly-degenerate free-energy wells, the equilibration time is slow; it is diffusive inthe number of replicas the system. Here we discuss benefitsof MC schemes that violate detailed balance in acceleratingthe convergence of parallel tempering.
Marija VuceljaCourant InstituteNew York [email protected]
Jon MachtaDepartment of PhysicsUniversity of Massachusetts [email protected]
MS8
Stochastic Reduced--Order Models for Multi--Scale Simulation of Laser Weld Failure
Micro laser welds are ubiquitous in intricate engineeringsystems. These partial penetration welds result in sharp,crack-like notches at their root. Further, the geometry andconstitutive behavior of these welds are complex and sub-ject to significant variability and uncertainty. Modelingthese welds in large systems is necessary to predict reli-ability; however, traditional modeling approaches cannotadequately resolve the sources of uncertainty. To this end,a surrogate model, based on stochastic reduced order mod-els (SROMs) and detailed finite element solutions, is devel-oped to represent the laser welds in the system-level modelsfor the goal of achieving efficient and accurate estimatesof system reliability. *Sandia is a multiprogram labora-tory operated by Sandia Corporation, a Lockheed MartinCompany, for the United States Department of EnergysNational Nuclear Security Administration under contractDE-AC04-94AL85000.
John M. EmerySandia National [email protected]
Mircea GrigoriuSchool of Civil EngineeringCornell [email protected]
Richard FieldSandia National [email protected]
MS8
Multifidelity Simulation of Large Scale Mass Flow
Large scale mass flows are modeled using a number ofmodeling assumptions to make the models computationallytractable. The first set of modeling assumptions are basedon simplifications inherent in the physics (e.g. depth av-eraging) and numerical methodology. For ensemble basedhazard analysis of such flows we create even simpler sta-tistical surrogates (Gaussian process etc.). In this paperwe will characterize the impact of each such assumption
on the simulation outcome.
Abani K. PatraSUNY at BuffaloDept of Mechanical [email protected]
Bruce PitmanDept of Mathematics,SUNY at [email protected]
Dinesh KumarDept of Mechanical EngineeringSUNY at [email protected]
Ramona StefanescuUniversity at [email protected]
MS8
Continuum Scale Constitutive Laws Extractedfrom Atomistic Simulations Using Bayesian Infer-ence and Uncertainty Quantification Methods
In atomistic-to-continuum simulations, it is crucial to de-rive constitutive relationships from atomistic simulationsto compensate for the unresolved degrees of freedom. Inthis study, we extract the heat conduction law from molec-ular dynamics (MD) simulations, together with its associ-ated uncertainty. This latter is due to the intrinsic noisein MD simulations. We use Bayesian inference to buildthe constitutive law. We present two approaches based onpolynomial chaos expansions (parametric) and Gaussianprocesses(non-parametric), respectively. We then propa-gate the obtained constitutive law into a continuum scalesimulation and compare with an equivalent MD simulation.
Maher SalloumSandia National LaboratoriesScalable and Secure Systems Research [email protected]
Jeremy TempletonSandia National [email protected]
MS8
Comparing Multiple Sources of Epistemic Uncer-tainty in Geophysical Simulations
Epistemic uncertainty – uncertainty due to a lack of modelrefinement – arises through assumptions made in physicalmodels, numerical approximation, and imperfect statisticalmodels. We will compare the effects of model parameteruncertainty to those introduced by numerical approxima-tions of geophysical model output. With the objective ofmodel-based geophysical hazard mapping in mind, we pro-pose methodology which can account for these sources ofuncertainty and quickly assess their impact on resultinghazard maps.
Elaine SpillerMarquette [email protected]
CS13 Abstracts 157
MS9
Optimal Control of Partial Differential AlgebraicEquations with Application to Fuel Cell Plants
Realistic multiphysics models describing Molten Carbon-ate Fuel Cells (MCFC) incorporate several aspects fromelectrochemistry, reaction kinetics, fluid dynamics as wellas heat convection and diffusion. This gives rise to a cou-pled system of Partial Differential and Algebraic Equations(PDAE) including 21 PDE of hyperbolic and parabolictype.The ability of fast and save load changes is very importantfor stationary power plants based on MCFC. The task ofcontrolling the gas inflows of the system to perform a fastload change constitutes on optimal control problem withPDAE and control constraints. Based on the gradient in-formation provided by the solution of the adjoint PDAEsystem, we solve the problem with a quasi-Newton methodof BFGS-type. We conclude with numerical experiments.
Armin RundInstitute for Mathematics and Scientific ComputingUniversity of [email protected]
MS9
Shape Calculus in Optics
Fabrication based perturbations, in nano-optics, can be ex-pressed in terms of a PDE constraint shape optimization.The resulting minimization problem is solved with a steep-est descent algorithm based on the method of mapping.Considering the regularity requirements of shape gradi-ent, we study and compare the so called continuous andfully-discrete methods based on a finite element discretiza-tion. We investigate both theoretical and implementationaspects in the framework of a 2D transmission problem.
Sahar SargheiniDepartment Information Technology and ElectricalEngineeringETH, [email protected]
Alberto Paganiniseminar for applied mathematicsETH, [email protected]
MS9
Multidisciplinary Shape and Topology Optimiza-tion
Shape optimization is a challenging issue of high practicalinterest. Mostly, it is treated in a single disciplinary con-text. But many industrial problems like aeroelastic designor instationary thermo-elastic design depend on a multi-disciplinary framework. The coupling of multiple disci-plines in shape and topology optimization leads to addi-tional mathematical issues, which are to be solved in nu-merical approaches. Efficient implementations as well asresults in practical applications will be presented.
Roland StoffelUniversity of TrierTrier, [email protected]
Volker H. Schulz
University of TrierDepartment of [email protected]
MS9
Design Optimization for Wave Propagation Prob-lems
By using adjoint based numerical deign optimization it ispossible to design highly efficient components for profes-sional audio. This talk presents an overview of the typi-cal steps in the design optimization process. The designof loudspeaker horns and subwoofer boxes illustrates thisprocess. Here, the Helmholtz equation models the wavepropagation, shape and topology optimization algorithmsdesign the components, and, as a final step, the compo-nents are build and analyzed in an anechoic chamber.
Eddie WadbroDepartment of Computing ScienceUmea [email protected]
MS10
Adaptive Algorithms for Simulations on ExtremeScales
We have develop a fast method that can capture piece-wise smooth functions in high dimensions with high orderand low computational cost. This method can be used forboth approximation and error estimation of stochastic sim-ulations where the computations can either be guided orcome from a legacy database. The focus of this talk willbe to describe how these fast methods are adaptive to highperformance computing.
Rick ArchibaldComputational Mathematics GroupOak Ridge National [email protected]
MS10
Adaptive Strategies for Random Elliptic PartialDifferential Equations
Solutions of elliptic boundary value problems with randomoperators admit efficient approximations by polynomialson the parameter domain. I will give an overview of adap-tive strategies for constructing suitable spaces of polynomi-als and computing approximate solutions on these spaces,and discuss the merits of high-order finite elements for ran-dom partial differential equations.
Claude J. GittelsonPurdue [email protected]
MS10
Adaptive Sequential Design for Efficient Emulationusing Gaussian Processes
We provide enhancements of a sequential design approachto build Gaussian Process surrogates, where points are cho-sen according to the entropy-based mutual information cri-terion. Points are selected according to updated parame-ters of the covariance structure to boost efficiency. Weprove that this method is close to optimal. We illustratethe benefits of our improvements on several synthetic data
158 CS13 Abstracts
sets and on the investigation of parametrizations in theWhole Atmosphere Community Climate Model.
Joakim Beck, Serge GuillasUniversity College [email protected], [email protected]
MS10
High Dimensional Multiphysics Metamodeling forCombustion Engine Stability
Cycle-to-cycle variations in power output of combustionengines is a major obstacle to increasing fuel efficiency,however, the causes of such variations are not fully under-stood. We take a multiphysics computationally expensiveengine model that combines turbulence, thermodynamicsand chemistry. Using high dimensional interpolation tech-niques, we create a cheap model approximation that is usedto study the correlation between the various parameters ofengine operations and the cycle-to-cycle power variations.
Miroslav StoyanovFlorida State UniversityDepartment of Scientific [email protected]
Clayton G. Webster, Charles FinneyOak Ridge National [email protected], [email protected]
Sreekanth PannalaComputer Science and Mathematics DivisionOak Ridge National [email protected]
Stuart Daw, Robert Wagner, Kevin Edwards, JohneyGreenOak Ridge National [email protected], [email protected],[email protected], [email protected]
MS11
A Fast Algorithm for Biharmonic Equation andApplications
We present an accurate fast algorithm to solve the Bihar-monic equation within a unit disk of the complex plane.This accurate and fast algorithm is derived through ex-act analysis and properties of convolution integrals usingGreen’s function method in complex plane. The fast al-gorithm has an asymptotic operation count O(lnN) perpoint. It has been applied to several Stokes’ and Navier-Stokes flow problems.
Aditi GhoshTexas A&M [email protected]
Prabir DaripaTexas A&M UniversityDepartment of [email protected]
MS11
A Parallel Fast Summation Algorithm for Volume
Potentials
Traditional FMM based approaches for evaluating poten-tials due to continuous source distributions use large num-ber of source points, require smoothing functions, and needfrequent remeshing for time varying distributions. We in-stead use piecewise polynomials and adaptive octree torepresent source distribution and output potential. Ourmethod uses kernel independent FMM, enabling use ofwide range of kernels (Laplace, Stokes, Helmholtz etc.).The implementation scales to the largest supercomputersavailable and demonstrates excellent per core performance.
Dhairya MalhotraInstitute of Computational Engineering and SciencesThe University of Texas at [email protected]
George BirosUniversity of Texas at [email protected]
MS11
BEM++ – a new C++/Python Library forBoundary-element Calculations
BEM++ is a new general-purpose open-source library forboundary-element calculations. All standard kernels forLaplace, Helmholtz and modified Helmholtz are availableand more (Maxwell, elasticity) are being added. Acceler-ated assembly based on the adaptive cross approximation(ACA) algorithm is supported on CPUs via an interfaceto the AHMED library, and implementation of ACA onGPUs is ongoing. The library is tightly integrated withTrilinos to allow access to its iterative solver capabilities.Python wrappers allow users to rapidly develop boundary-element codes and easily visualise and postprocess resultsof their calculations. More information is available atwww.bempp.org.
Wojciech SmigajDepartment of MathematicsUniversity College [email protected]
Simon ArridgeDepartment of Computer ScienceUniversity College [email protected]
Timo BetckeUniversity College [email protected]
Joel PhillipsDepartment of MathematicsUniversity College [email protected]
Martin SchweigerDepartment of Computer ScienceUniversity College [email protected]
MS11
Fast Numerical Greens Functions with Application
CS13 Abstracts 159
to Magnetic Resonance Imaging
A common problem in a variety of electromagnetic appli-cations is to determine fields from a variety of sources inthe presence of a given scatterer. If there is an analyticrepresentation for the Greens function given the scatterer,then the standard approach is to use that Greens func-tion to simplify the field calculations. For general scat-terer geometries, the Greens function must be computednumerically and tabulated, an intractable process in threespace dimensions as the table must be six-dimensional. Inthis talk we describe combined SVD-plus-sampling methodfor constant-time Green’s function evaluation, and thendemonstrate its use in optimizing coil excitation in low-frequency Magnetic Resonance Imaging.
Amit HochmanResearch Lab. of [email protected]
Jorge Fernandez VillenaInst Engn Sistemas & Comp Invest & Desenvolviment,[email protected]
L. Miguel SilveiraUniv Tecn Lisboa, IST, P-1000029 Lisbon, [email protected]
Luca DanielM.I.T.Research Lab in [email protected]
Jacob WhiteDept. of Elec. Eng. and Comp. Sci and Res. Lab. [email protected]
MS12
Why Julia?
Julia is a language designed from the beginning to effi-ciently execute the kinds of programs users write in Mat-lab, R, Python/SciPy, and similar environments. Perfor-mance is a major motivation, but there is also ongoingwork to speed up existing languages. I will discuss whatwe might want other than performance, and what advan-tages we gain by designing a new language. Fortunately,performance does not trade off against all other desirablefeatures.
Jeff BezansonComputer [email protected]
Stephan [email protected]
Virah [email protected]
Alan EdelmanISC [email protected]
MS12
Numba, Compiling Numpy Functions from Pythonwith Llvm
Abstract not available at time of publication.
Travis OliphantContinuum [email protected]
MS12
Rllvm and RLLVMCompile
R is increasingly used for both data analysis and also devel-oping new statistical methods. There are various efforts tomake this interpreted language faster, with technical and”practical” complications. Instead, we have created bind-ings to the LLVM libraries/APIs to allow R programmerscompile their own code and parts of the R language or newsimilar DSLs. Rather than a ”single” compiler within R,others can use the bindings to develop different compilationstrategies that target run-time, memory usage and paral-lelism for different hardware, e.g GPUs, cores with sharedmemory and distributed memory clusters.
Duncan W. Temple LangStatisticsUC [email protected]
MS12
Introduction to JIT Compiling and Code Genera-tion in Scientific Computing
Domain specific languages, code generation, and just-in-time compiling are all becoming standard part of scientificcomputing. Originally the target for high level languageswas to produce more expressive ways of writing code overthe traditional languages, i.e. Fortran and C/C++. Nowthese high level codes are starting to compete, and some-times beat, low level languages using advanced compilertechniques. This introduction will overview the methodsand tradeoffs used in modern high level scientific languages.
Andy R. TerrelTexas Advanced Computing CenterUniversity of [email protected]
MS13
A Narrow-band Gradient-Augmented Level SetMethod for Incompressible Two-phase Flow
We have incorporated the gradient-augmented level setmethod (GALSM) for use in two-phase incompressible flowsimulations by interpolating velocity values and introduc-ing a re-initialization procedure. The method is conductedon a narrow band around the interface, reducing computa-tional effort, while maintaining an optimally local advec-tion scheme and providing sub-grid resolution. Ocean wavesimulation is the primary motivation, and numerous bench-marks have been conducted, including a new comparisonwith wave tank data.
Curtis Lee, John DolbowDuke [email protected], [email protected]
160 CS13 Abstracts
Peter J. MuchaUniversity of North [email protected]
MS13
A Particle-enhanced Gradient-augmented LevelSet Method
The original gradient-augmented level set methods use aprojection on an interpolant space that is based only onthe information provided on the grid. In this presentation,we show how we can incorporate information from Lagra-gian particles to improve the accuracy of the method whilepreserving the projection properties. We will address thequestions of topological changes and show applications tomultiphase Navier-Stokes.
Olivier MercierMcGill [email protected]
MS13
Model Equations for Jet-schemes
Model equations describe the behavior of numericalschemes in the asymptotic limit when the mesh-size van-ishes, and provide information about the nature of the algo-rithm error in this limit. In particular, they can be used todetect the causes of long wave instabilities (if present), anddevise ways to correct them. Because jet-schemes use anadvect–and–project approach in function space, standardmethods for obtaining model equations cannot be used.The numerical solution by a jet scheme implicitly carries(small amplitude) grid size structure (as given by the localpolynomial interpolant) — modulated on a longer scale bythe solution’s variations. In this talk we will discuss howhomogenization techniques can be used to obtain modelequations for jet schemes, and apply it to some illustrativeexamples.
Rodolfo R. RosalesMassachusetts Inst of TechDepartment of [email protected]
Benjamin SeiboldTemple [email protected]
Jean-Christophe NaveMcGill [email protected]
MS13
An Overview of Gradient-augmented Methods andJet Schemes
This talk provides an introduction to the philosophy andmethodology of jet schemes and gradient-augmented levelset methods, and an overview of how they relate to othercomputational approaches for advection and interface evo-lution problems.
Benjamin SeiboldTemple [email protected]
MS14
A Method for Non-Intrusive Model Reduction andAdjoint-based Optimization
We propose a method to enable existing model reductionmethods (POD, DEIM, etc) and adjoint-based optimiza-tion without modification to the simulation source code.We infer a twin model to mirror the underlying modelgoverning a PDE simulation. Our method features: Di-mensionality reduction of inputs by exploiting physics in-variance; Independence with the numerical scheme imple-mented in the original PDE simulation; A goal-orientedapproach to minimize the discrepancy between the twinmodel and the original simulation model.
Qiqi WangMassachusetts Institute of [email protected]
Hector KlieConocoPhillips [email protected]
MS14
Adaptive Proxies for Integrated Oil Field Produc-tion Optimization under Uncertainty
The treatment of geological parameter uncertainty in nu-merical reservoir simulation is necessary to qualify the re-sults achieved. Similarly, the back-pressure effects and dy-namics of a full transient system (from source to sink) canonly be elicited by treating the interdependent boundaryconditions between the component reservoir, surface andprocess models. However, as the resulting coupled simula-tion is then extremely costly to evaluate, adaptive proxymethods are required to effectively optimize the compositemodel under uncertainty.
Benoit Couet, Kashif Rashid, William BaileySchlumberger-Doll [email protected], [email protected], [email protected]
MS14
Local-Global Model Reduction Techniques forPorous Media Flow Simulation
The development of algorithms for modeling and simula-tion of heterogeneous porous media poses challenges re-lated to the variability of scales and requires efficient solu-tion methodologies due to their large-scale nature. This isespecially daunting in simulating unconventional reservoirsthat are embedded in fractured media. In this talk, I willdescribe a local-global model reduction method that com-bines multiscale techniques with the reduced-order meth-ods in a seamless fashion. The aim is to reduce the degreesof freedoms in the state-space by computing global reducedorder models written on a coarser space.
Eduardo GildinPetroleum Engineering DepartmentTexas A&M [email protected]
CS13 Abstracts 161
MS14
Controllability and Observability in Two-phasePorous Media Flow
Large-scale nonlinear flow simulation models are frequentlyused in oil and gas reservoir engineering to make decisionson well locations, depletion strategies, production scenar-ios, etc. The quality of these decisions is, among other as-pects, determined by the controllability and observabilityproperties of the reservoir model at hand. We use empiricalGramians to analyze the controllability and observabilityof two-phase reservoir flow. We conclude that the positionof the wells and the dynamics of the front between reservoirfluids determine the controllability and observability prop-erties of the state variables (pressures and saturations).Therefore, for fixed well positions, reduced-order modelsshould focus on modeling the fluid front(s).
Jan Dirk JansenDelft University of TechnologyDepartment of [email protected]
Jorn van DorenDelft University of Technology (TU Delft)[email protected]
Paul Van den HofTU Eindhoven and TU [email protected]
MS15
Fast and Reliable Trust-region Eigensolvers
Error resilient solvers are necessary to address the reducedreliability of future high-performance computing systems.Surrogate-based methods, including trust-region methods,are naturally suited for this context, because the analysesand mechanisms surrounding surrogate error are a start-ing point for addressing soft errors and expensive sanitychecks. We present an implementation and analysis of atrust-region-based eigenvalue solver adapted to low relia-bility computing due to soft-errors or low precision types.
Christopher G. BakerOak Ridge National [email protected]
MS15
Parallel Implementations of the Trace Minimiza-tion Scheme TraceMIN for the Sparse SymmetricEigenvalue Problem
Large scale sparse symmetric eigenvalue problems arise inmany computational science and engineering applications.Often, the large size of these problems requires the develop-ment of eigensolvers that scale well on parallel computingplatforms. We compare the effectiveness and robustnessof our eigensolver for the symmetric generalized eigenvalueproblem, the trace minimization scheme TraceMIN, againstthe well-known sparse eigensolvers in Sandia’s Trilinos li-brary. Our results show that TraceMIN is more robust andhas higher parallel scalability.
Alicia KlinvexDepartment of Computer SciencePurdue [email protected]
Faisal SaiedPurdue [email protected]
Ahmed SamehDepartment of Computer SciencePurdue [email protected]
MS15
Absolute Value Preconditioning for Symmetric In-definite Linear Systems
We introduce a novel strategy, absolute value precondi-tioning, for constructing symmetric positive definite (SPD)preconditioners for linear systems with symmetric indefi-nite matrices. We consider a model problem with a shifteddiscrete negative Laplacian, and suggest a geometric multi-grid (MG) preconditioner, where the inverse of the matrixabsolute value appears only on the coarse grid, while op-erations on finer grids are based on the Laplacian. Ournumerical tests demonstrate practical effectiveness of thenew preconditioner. [http://arxiv.org/abs/1104.4530]
Andrew KnyazevMitsubishi Electric Research [email protected]
Eugene VecharynskiUniversity of Colorado [email protected]
MS15
A MATLAB Interface for PRIMME for SolvingEigenvalue and Singular Value Problems
The software package PRIMME for solving large, sparse,Hermitian eigenvalue problems implements many state-of-the-art preconditioned eigenvalue iterative methods andembeds expert knowledge into dynamically choosing tech-niques. As stand-alone or through a SLEPc interface, itsrobustness, efficiency, and user-friendliness have made itthe first choice for hundreds of groups around the world.First, we present primme eigs(), a MATLAB interface toPRIMME’s full functionality with an interface as simpleas eigs(). Second, we present primme svds(), which solvesthe same SVD problem as svds() but with the power ofPRIMME’s methods. For smallest singular values it is pos-sible to dynamically switch between ATA and augmentedmatrix techniques.
Andreas StathopoulosCollege of William & MaryDepartment of Computer [email protected]
Lingfei WuDepartment of Computer ScienceCollege of William & [email protected]
MS16
Computing Least Squares Condition Numbers
The notion of condition number provides us with a theo-retical framework to measure the numerical sensitivity ofa problem solution to perturbations in its data. We derive
162 CS13 Abstracts
condition numbers and estimates for linear least squaresand total least squares problems using various metrics tomeasure errors. We present numerical experiments to com-pare exact values and statistical estimates. We also pro-pose performance results using new routines on top of themulticore-GPU library MAGMA.
Marc BaboulinINRIA/University [email protected]
MS16
Numerical Issues in Testing Linear Algebra Algo-rithms
We discuss a variety of numerical issues that arise in test-ing linear algebra algorithms. These include how to choosean error measure (backward, forward, normwise, or com-ponentwise) and interpret the observed errors; why tinynormwise relative errors (orders of magnitude smaller thanthe unit roundoff) sometimes appear and how to deal withthem when producing performance profiles; and the impli-cations for testing of a possible lack of bitwise reproducibil-ity of results on HPC systems.
Nicholas J. HighamUniversity of ManchesterSchool of [email protected]
Nicholas DingleSchool of Mathematics, University of [email protected]
MS16
Accuracy and Stability Issues for Randomized Al-gorithms
The advantage of randomized algorithms is speed and sim-plicity. For very large problems, they can be faster thandeterministic algorithms, and they are often simple to im-plement. We will discuss the numerical sensitivity and sta-bility of randomized algorithms, as well as the error due torandomization, and the effect of the coherence and lever-age scores of the matrix. Algorithms under considerationinclude matrix multiplication, least squares solvers, andlow-rank approximations.
Ilse IpsenNorth Carolina State UniversityDepartment of [email protected]
MS16
Towards a Reliable Performance Evaluation of Ac-curate Summation Algorithms
Recent floating point summation algorithms compute thebest accurate sum even for arbitrary ill-conditioned prob-lems. So the run-time efficiency of these algorithms be-comes the criteria to decide which is the best. Neitherthe classic flop count nor experimental timings are relevantmeasures of the actual performance of such core numeri-cal algorithms. We justify these claims and we present areliable performance evaluation of some accurate summa-tion algorithms thanks to an automatic instruction levelparallelism analysis.
Philippe Langlois, Bernard Goossens, David Parello
University of [email protected], [email protected],[email protected]
MS17
Mortar Methods for Flow in Heterogeneous PorousMedia
For an elliptic problem with a heterogeneous coefficient inmixed form, nonoverlapping mortar domain decompositionis efficient in parallel if the mortar space is small. Wedefine a new multiscale mortar space using homogenizationtheory. In the locally periodic case, we prove the methodachieves optimal order error estimates in the discretizationparameters and heterogeneity period. Numerical resultsshow that the method works well even when the coefficientis not locally periodic.
Todd ArbogastDept of Math; C1200University of Texas, [email protected]
Hailong XiaoICESThe University of Texas at [email protected]
MS17
A Multiscale Discontinuous Galerkin Method forthe Schrodinger Equation in the Simulation ofSemiconductor Devices
We develop and analyze a multiscale discontinuousGalerkin method for the stationary Schrodinger equationin the simulation of nanoscale semiconductor structures.The solution of the Schrodinger equation has a small wavelength and oscillate at much smaller space scale for highelectron energies. We incorporate the oscillatory behaviorof the solution into the multiscale basis of the discontin-uous Galerkin finite element method so that the problemcan be solved on coarser meshes.
Bo DongUniversity of Massachusetts [email protected]
Wei WangDepartment of Mathematics and StatisticsFlorida International [email protected]
Chi-Wang ShuBrown UniversityDiv of Applied [email protected]
MS17
A Multiscale Mixed Method Based on a Nonover-lapping Domain Decomposition Procedure
We use a nonoverlapping iterative domain decompositionprocedure based on the Robin interface condition to de-velop a multiscale mixed method for calculation of thevelocity field in heterogeneous porous media. Hybridizedmixed finite elements are used for the spatial discretiza-tion of the equations. We define local, multiscale mixed
CS13 Abstracts 163
basis functions to represent the discrete solutions in thesubdomains. In the numerical approximations, subspacesof the vector space spanned by these basis functions canbe appropriately chosen, which determines the balance be-tween numerical accuracy and efficiency. Several numericalexperiments are discussed to illustrate the important fea-tures of the procedure.
Alexandre FranciscoUniversidade Federal [email protected]
Victor E. GintingDepartment of MathematicsUniversity of [email protected]
Felipe PereiraCenter for Fundamentals of Subsurface FlowUniversity of [email protected]
Joyce RigeloUniversity of [email protected]
MS17
Relaxing the CFL Number of the DiscontinuousGalerkin Methods
Discontinuous Galerkin methods applied to hyperbolicproblems have a CFL number that is inversely propor-tional to the order of approximation. We propose a mod-ification to the scheme which allows for a significant timestep increase while preserving spatial accuracy. We furtherdiscuss how a CFL can be relaxed on non-uniform grids.These algorithms can be especially useful when discontinu-ities and under-resolved scales are present in the solution.
Lilia KrivodonovaUniversity of [email protected]
MS18
A Systematic Construction of SuperconvergentHDG Methods
Abstract not available at time of publication.
Bernardo CockburnSchool of MathematicsUniversity of [email protected]
Ke ShiUniversity of Minnesota, [email protected]
MS18
Overview of the Discontinuous Petrov-GalerkinMethods
Abstract not available at time of publication.
Leszek DemkowiczInstitute for Computational Engineering and Sciences(ICES)The University of Texas
MS18
Effect of Inexact Test Functions in the DPGMethod
The discontinuous Petrov Galerkin (DPG) method usesstandard trial spaces for approximating solutions of bound-ary value problems. However, its test functions are locallydefined so as to get optimal inf-sup stability. Althoughit is now clear how to apply hybridization techniques tolocalize the problem of computing these optimal test func-tions to a single mesh element, the resulting problem onone element can often be infinite dimensional. We provethat replacing this infinite dimensional problem by a prac-tical finite dimensional problem does not affect convergencerates for many examples. Thus ”inexactly” computed testspaces are enough in the DPG framework. The analysisproceeds by adapting an abstract Fortin operator to thePetrov-Galerkin context. The same operator also turnsout to be an essential tool in a posteriori error analysis ofa natural residual-based estimator provided by the DPGformalism. Recent results in this direction will also be pre-sented.
Jay GopalakrishnanPortland state [email protected]
MS18
Hybridizable Discontinuous Galerkin Methods forWave Propagation Problems at High Frequency
We present a class of hybridizable discontinuous Galerkin(HDG) methods for wave propagation problems. Themethods are fully implicit and high-order accurate in bothspace and time, yet computationally attractive owing totheir distinctive features. First, they reduce the globaldegrees of freedom to the numerical trace. Second, theyprovide optimal convergence rates for both the field vari-able and their gradient. Third, they allow us to improvethe numerical solution by performing a simple local post-processing.
Cuong Nguyen, Jaime PeraireMassachusetts Institute of [email protected], [email protected]
Bernardo CockburnSchool of MathematicsUniversity of [email protected]
MS19
An Immersed Finite Element Method for Fluid-Structure Interaction Problems and Applicationsto Hydrocephalus
We present a fully variational formulation of an immersedmethod for FSI problems based on the finite elementmethod. Our formulation does not require Dirac-δ distri-butions. The immersed solid can be viscoelastic of differen-tial type or hyperelastic, but is not otherwise restricted inthis constitutive class. Our discrete formulation is shownto have the natural stability estimates of the companioncontinuum problem. We include standard test cases and
164 CS13 Abstracts
some preliminary applications to hydrocephalus.
Luca HeltaiSISSA: International School for Advanced StudiesTrieste, [email protected]
Saswati RoyDepartment of Engineering Science and MechanicsThe Pennsylvania State [email protected]
Francesco CostanzoDepartment of Engineering Science and MechanicsPennsylvania State [email protected]
MS19
An Immersed Boundary Energy-Based Method forIncompressible Viscoelasticity
Fluid-structure interaction problems, in which the dynam-ics of a deformable structure is coupled to the dynamics ofa fluid, are prevalent in biology. For example, the heartcan be modeled as an elastic boundary that interacts withthe blood circulating through it. The immersed boundarymethod is a popular method for simulating fluid-structureproblems. The traditional immersed boundary method dis-cretizes the elastic structure using a network of springs.This makes it difficult to use material models from con-tinuum mechanics within the framework of the immersedboundary method. In this talk, I present a new immersedboundary method that uses continuum mechanics to dis-cretize the elastic structure, with a finite-element-like dis-cretization. This method is first applied to a warm-upproblem, in which a viscoelastic incompressible materialfills a two-dimensional periodic domain. Then we applythe method to a three-dimensional fluid-structure interac-tion problem.
Dharshi DevendranUniversity of [email protected]
MS19
An Approach to using Finite Element ElasticityModels with the Immersed Boundary Method
The immersed boundary (IB) method treats problems inwhich a structure is immersed in a fluid flow. The IBmethod was first introduced for the case in which the struc-ture is composed of systems of elastic fibers, but subse-quent extensions have treated increasingly general mechan-ics models. This talk will describe one version of the IBmethod that uses elasticity models approximated via stan-dard finite element methods, and will describe applicationsto cardiac and cardiovascular mechanics.
Boyce GriffithNew York University School of [email protected]
MS19
Augmenting the Immersed Boundary Method withRbfs: Applications to Modeling of Platelets inHemodynamic Flows
The Immersed Boundary (IB) Method is a widely-used nu-
merical methodology for the simulation of fluid-structureinteraction problems. Most IB simulations represent theirstructures with piecewise-linear approximations and utilizeHookean spring models to approximate structural forces.Our specific motivation is the modeling of platelets inhemodynamic flows. In this talk, we present our attemptsat using Radial Basis Functions (RBFs) as the means bywhich to represent platelets in an Immersed Boundary sim-ulation. We will report results for two-dimensional andthree-dimensional platelet flows. Furthermore, we will re-port other possible extensions/additions that can be madein the Immersed Boundary context using RBFs.
Robert KirbyUniversity of [email protected]
Varun ShankarSchool of ComputingUniversity of [email protected]
Grady B. WrightDepartment of MathematicsBoise State University, Boise [email protected]
Aaron L. FogelsonUniversity of [email protected]
MS20
H∈-Optimal Reduced Models for Structured Sys-tems
We present new necessary conditions for H∈-optimal ap-proximation of linear dynamical systems by reduced ordersystems having additional special structure such as sec-ond order systems or port-Hamiltonian systems. This givesrise to distributed interpolatory conditions similar to thosearising in weighted H∈ model reduction which suggests, inturn, natural computational strategies for resolution.
Christopher A. BeattieVirginia Polytechnic Institute and State [email protected]
Peter BennerMax-Planck-Institute forDynamics of Complex Technical [email protected]
MS20
Preserving Lagrangian Structure in NonlinearModel Reduction
Nonlinear mechanical systems described by Lagrangian dy-namics arise in structural dynamics and molecular dynam-ics. Standard nonlinear model-reduction methods (e.g.,discrete empirical interpolation) destroy Lagrangian struc-ture, and with it critical properties such as energy conser-vation and symplecticity. This talk presents an efficientnonlinear model-reduction methodology that preserves La-grangian structure. The method approximates the sys-tem’s ‘Lagrangian ingredients’: the Riemannian metric,potential-energy function, dissipation function, and exter-nal force. These approximations preserve salient properties
CS13 Abstracts 165
while ensuring computational efficiency.
Kevin T. CarlbergSandia National [email protected]
Ray S. TuminaroSandia National LaboratoriesComputational Mathematics and [email protected]
Paul BoggsSandia National [email protected]
MS20
Passivity-preserving Algorithm for Multiport Pa-rameterized Modeling in the Frequency Domain
Given a collection of frequency response datasets, sweptover design and geometrical parameters of interest, weidentify a closed-form parameterized dynamical model us-ing constrained fitting. The parameterized models are de-veloped by polynomial or rational multivariate approxima-tion between non-parameterized transfer matrices. Theuser of the generated models will be able to instantiatereduced models guaranteed to be stable and passive forany values of design parameters. Examples will includesystems with multiple ports and parameters.
Zohaib [email protected]
Luca DanielM.I.T.Research Lab in [email protected]
MS20
Hierarchical Structure-Preserving Phase Mod-elling of Oscillators
It has long been known that coupled oscillators, when syn-chronized, respond to external inputs as a single “macro’oscillator. Reliance on this phenomenon is common inmany domains, including biology (circadian rhythms, heartpacemaking systems) and nanotechnology (eg, spin-torquebased microwave sources). To our knowledge, no proof ofthis remarkable phenomenon has been available. We pro-vide such a proof, along with effective computational meth-ods for extracting quantitative characterizations of “macro’oscillators from its individual.
Jaijeet RoychowdhuryUniversity of California at [email protected]
MS21
Parallelization and Performance Issues in AdaptiveSimulation of Wave Propagation
In this presentation, we will discuss questions of computa-tional performance that occur in the adaptive simulationof wave propagation problems - in our case, parallel adap-tive mesh refinement techniques for tsunami simulation andsimulation of seismic wave propagation on unstructuredmeshes. For tsunami simulation, we will present paral-
lelization approaches (based on the Sierpinski space-fillingcurve) for efficient dynamical refinement and coarsening ofstructured, but fully adaptive triangular grids. For exam-ple, we use dynamical scheduling of subpartitions, togetherwith a split&join approach, to allow varying computationalload of partitions and even varying number of computecores. For SeisSol, a Discontinuous-Galerkin-based packagefor seismic wave propagation and dynamic rupture simula-tion, we will present first results on using code generationapproaches to speed up the basic computational kernels.We will also discuss options to improve memory access andload balancing of the adopted ADER-DG method.
Michael Bader, Alexander BreuerTechnische Universitat MunchenDepartment of [email protected], [email protected]
Alexander HeineckeTU [email protected]
Martin SchreiberTechnische Universitat MunchenDepartment of [email protected]
Christian PeltiesLMU [email protected]
MS21
Finite-Volume and Discontinuous Galerkin Algo-rithms for Multifluid Simulations of Plasmas
Plasmas display a rich variety of behavior, from kinetic tofluid, and span vast spatial and temporal scales. This talkwill give an overview of explicit finite-volume and discon-tinuous Galerkin algorithms for the solution of multi fluidplasma models. The framework of weakly nonlinear hy-perbolic equations will be developed and applied to para-sitic decay of waves from parametric resonance arising fromquadratic nonlinearities. The application of this theory tounderstanding losses in radio-frequency heating of plasmaswill be discussed.
Ammar HakimPlasma Physics LaboratoryPrinceton [email protected]
MS21
Simulation of Poroelastic Wave Propagation inCLAWPACK for Geophysical and Medical Appli-cations
We use the CLAWPACK (Conservation LAWs PACKage)finite volume method code to solve Biot’s equations for dy-namics of a porous, fluid-saturated elastic medium. Theseequations were developed to model fluid-saturated rock for-mations, but are also applicable to other porous solids, suchas in vivo bone. At low frequency Biot’s equations are asystem of hyperbolic PDEs with a relaxation source term,which may be stiff depending on the time scales associatedwith wave propagation. This talk gives a brief introductionto the class of numerical methods used, including issues as-sociated the incorporation of the stiff source term. Numer-ical results are shown Cartesian grids, comparing againstrecent discontinuous Galerkin results for geophysical prob-
166 CS13 Abstracts
lems, and on logically rectangular mapped grids capable ofmodeling moderately complex geometry similar to bones.
Grady I. LemoineDepartment of Applied MathematicsUniversity of [email protected]
MS21
Parallelization of Hybridizable Discontinu-ous Galerkin Methods for a Nonhydrostatic FreeSurface Primitive Equation Ocean Model
Higher-order finite element methods are becoming increas-ingly common in ocean modelling. Hybridizable discontin-uous Galerkin (HDG) methods are attractive because theyare highly parallelizable, and the high computational costassociated with standard DG and LDG methods can be al-leviated. Studies on parallel numerical schemes for solvingthe 2D/3D nonhydrostatic Navier–Stokes equations cou-pled to a free surface equation are presented, along withaccuracy and performance implications.
Chris Mirabito, Mattheus Ueckermann, Patrick Haley,Pierre LermusiauxMassachusetts Institute of [email protected], [email protected], [email protected],[email protected]
MS22
Optimal Low Rank Matrix Inverse Approximationfor Image Processing
In many applications, the desired solution of an inverseproblem can be well-represented using only a few vectorsof a certain basis, e.g., the singular vectors. In this work,we design an optimal low-rank matrix inverse approxima-tion by incorporating probabilistic information from train-ing data and solving an empirical Bayes risk minimizationproblem. We propose an efficient update approach for com-puting a low-rank regularized matrix, and provide numer-ical results for problems from image processing.
Julianne ChungVirginia [email protected]
Matthias ChungDepartment of MathematicsTexas State [email protected]
MS22
Windowed Regularization for Image Deblurring viaOperator Approximation
Recent work by Chung, Easley and O’Leary (SIAM J. Sci.Comput., 2011), has shown that by windowing the com-ponents in the spectral domain of the blurring operatorand employing different regularization parameters for eachwindow, superior image restorations can be obtained. Weconsider suitable surrogate representations of the blurringoperator that allow us to make this new regularization tech-nique, as well as other known methods, applicable when theSVD cannot be computed.
Misha E. KilmerTufts [email protected]
Julianne ChungVirginia [email protected]
Dianne P. O’LearyUniversity of Maryland, College ParkDepartment of Computer [email protected]
MS22
Parallel Algorithms for Iterative Image Recon-struction
Abstract not available at time of publication.
James G. NagyDepartment of Mathematics & Computer ScienceEmory [email protected]
MS22
Tracking Objects Using 3D Edge Detectors
Edge detection determines the boundary of objects in animage. In some applications a sequence of images recordsa 2D representation of a scene changing over time, giv-ing 3D data. New 3D edge detectors, particularly ones wedeveloped using shearlets and hybrid shearlet-Canny algo-rithms, identify edges of complicated objects much morereliably than standard approaches, especially under highnoise conditions. We also use edge information to deriveposition and velocity via an optimization algorithm.
Dianne P. O’LearyUniversity of Maryland, College ParkDepartment of Computer [email protected]
Glenn EasleySystem Planning [email protected]
David SchugDepartment of MathematicsUniversity of Maryland at College [email protected]
MS23
Flexible BiCGStab and Its Use in Practice
BiCGStab is one of the de facto methods of choice for solv-ing linear systems in many application domains. Motivatedby recent development of flexible preconditioners in highperformance computing, we study BiCGStab under flexiblepreconditioning. In this talk, we will present some analysisresults of the convergence behavior of flexible BiCGStab,and show its successful use in practice. In an important ap-plication, PFLOTRAN, we demonstrate that the run timeof flexible BiCGStab preconditioned by multigrid is signif-icantly improved over the currently known fastest record.
Jie ChenArgonne National [email protected]
Lois Curfman McInnesArgonne National Laboratory
CS13 Abstracts 167
Mathematics and Computer Science [email protected]
Hong ZhangArgonne National [email protected]
MS23
Krylov Solvers for Singular Hermitian, ComplexSymmetric, and Skew Hermitian Linear Systems
Most existing Krylov subspace algorithms for linear sys-tems assume non-singularity of the matrices or opera-tors. MINRES-QLP (Choi, Paige, and Saunders 2011) isa MINRES-like algorithm but designed for computing theunique pseudoinverse solution of a singular symmetric orHermitian problem using short recurrence relations in so-lution updates and stopping conditions. On a nonsingu-lar system, MINRES-QLP is more stable than MINRES(Paige and Saunders 1975). We design similarly stable al-gorithms for singular (or nonsingular) complex-symmetric,skew-symmetric, and skew-Hermitian linear systems or lin-ear least-squares problems. Our goal is to provide oneefficient implementation prototyped in Matlab for thesedifferent classes of linear systems. We present extensivenumerical experiments to demonstrate the scalability androbustness of these algorithms.
Sou-Cheng ChoiStanford [email protected]
MS23
LAMG: Fast Multilevel Linear Solver and Eigen-solver of the Graph Laplacian
Graph Laplacians arise in large-scale applications includ-ing machine learning, clustering, transportation networks,and CFD. We present Lean Algebraic Multigrid (LAMG):a linear solver of Ax = b, where A is a graph Laplacian.It combines novel methodologies: piecewise-constant inter-polation; aggregation via a novel node proximity measure;multilevel acceleration; and perturbed Galerkin coarsening.LAMG is demonstrated to scale linearly with the number ofedges for real-world graphs with up to 47,000,000 edges. Fi-nally, we present a linear-scaling LAMG-based eigensolverfor computing several of the lowest eigenpairs of A.
Oren E. LivneDepartment of Human GeneticsUniversity of [email protected]
Achi BrandtWeizmann Institute of ScienceMicrosoft, [email protected]
MS23
Flexible and Multi-Shift IDR Algorithms for Solv-ing Large Sparse Linear Systems
The Induced Dimension Reduction algorithm, IDR(s), isone of the most efficient methods for solving large sparsenonsymmetric linear systems of equations. We present twouseful extensions of IDR(s), namely a flexible variant and amulti-shift variant. The algorithms exploit the underlyingHessenberg decomposition computed by IDR(s) to gener-
ate basis vectors for the Krylov subspace. The approxi-mate solutions are computed using a quasi-minimizationapproach. Numerical examples are presented to show theeffectiveness of these IDR(s) variants compared to existingones and to other Krylov subspace methods.
Martin B. van GijzenDelft Inst. of Appl. MathematicsDelft University of [email protected]
Gerard SleijpenDepartment of MathematicsUtrecht [email protected]
Jens-Peter M. ZemkeInst. of Numerical SimulationTU [email protected]
MS24
Solution Methods for Phase Field Methods withArbitrary Numbers of Variables
A unique characteristic of some phase field models, for ex-ample typical phase field models of grain growth, is thatthey can have an arbitrary number of variables, dependingon the the number of grains being modeled. Finite ele-ment approaches typically focus on problems with a largenumber of nodes, but only between one to five variables.In this work, we present techniques for solving systems ofequations with anywhere from two to 1000 variables, focus-ing on determining the method that results in the shortestcomputation time and the least memory usage. This effortis a part of the MARMOT phase field framework.
Derek Gaston, Michael TonksIdaho National [email protected], [email protected]
MS24
Optimal Control of a Semi-discrete Cahn-Hilliard/ Navier-Stokes System
Optimal boundary control of a time-discrete Cahn-HilliardNavier- Stokes system is studied. For the Cahn-Hilliardpart of the system, a general class of free energy poten-tials is considered including the double-obstacle potential.The existence of an optimal solution to the time discretecontrol problem as well as an approximate version thereofis established. Moreover a C-stationarity system for theoriginal problem is derived and numerically realized.
Michael HintermuellerHumboldt-University of [email protected]
MS24
Time-Stepping Methods for Phase Field Models
Phase field models are flexible approaches to representcomplex materials microstructure evolution. This comes,however, at the cost of including singular terms such asinfinite potential and logarithmic components for express-ing certain phenomena, which pose significant difficultiesfor numerical approximation schemes. In this work we de-scribe time-stepping approaches that accommodate such
168 CS13 Abstracts
singularities, including ones that use variational inequali-ties in the subproblem. We present analytical and numeri-cal results to demonstrate the properties of such methods.
Jungho Lee, Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected], [email protected]
MS24
Computational Challenges Posed by Phase FieldMethods
The phase field method is a novel approach to solvingfree-boundary problems without explicitly tracking thelocation of an interface. However, the diffuse interfaceused in the method introduces a new length scale thatpresents significant computational challenges when per-forming three-dimensional simulations over realistic lengthand time scales. Even more challenging is the recently de-veloped phase field crystal method. A survey of the com-putational challenges posed by phase field methods will begiven.
Peter VoorheesNorthwestern UniversityDept. of Material Science and [email protected]
MS25
Soft Error Resilience for One-sided Dense LinearAlgebra Algorithms
Soft errors threaten computing systems by producing datacorruption that’s hard to detect and correct. Current re-search of soft-error resilience for dense linear solvers offerslimited capability on large-scale systems, and suffers fromboth soft error and round-off error. This work proposes afault tolerant algorithm that can recover the solution of adense linear system from multiple spatial and temporal softerrors. Experimental results on Kraken confirm scalabilityand negligible overhead in solution recovery.
Jack J. DongarraDepartment of Computer ScienceThe University of [email protected]
Peng DuThe University of [email protected]
Piotr LuszczekDepartment of Electrical Engineering and ComputerScienceUniversity of Tennessee, [email protected]
MS25
Quantifying the Impact of Bit Flips on NumericalMethods
The collective surface area and increasing density of com-ponents has led to an increase in the number of observed bitflips. This effort works towards rigorously quantifying theimpact of bit flips on floating point arithmetic, using bothanalytical modeling and Monte Carlo sampling, with the
eventual goal to utilize these results to provide insight intothe vulnerability of leadership-class computing systems tosilent faults, and ultimately provide a theoretical basis forfuture silent data corruption research.
James ElliottNorth Carolina State [email protected]
MS25
Application Robustification: Fortifying (Even Dis-crete) Applications Against Hardware Errors
Many applications need to solve combinatorial or dis-crete problems. Examples include sorting and optimizationproblems on graphs such as maximum flow. Most discretealgorithms were not designed to tolerate hardware faultswhich cause incorrect arithmetic or corruption of valuesin memory. We will talk about how to transform or re-lax discrete problems into forms which can be solved us-ing fault-tolerant numerical algorithms, including some ofthose discussed in this minisymposium. We will also dis-cuss algorithmic detection and correction schemes for linearalgebra problems.
Joseph Sloan, Rakesh KumarUniversity of Illinois at [email protected], [email protected]
MS25
Classifying Soft Error Vulnerabilities in Extreme-Scale Scientific Applications Using Bifit
Extreme-scale scientific applications are at a significant riskof being hit by soft errors on future supercomputers. Tobetter understand this risk, we have built an empirical faultinjection tool - BIFIT. BIFIT is designed with capabilityto inject faults at specific targets: execution point and datastructure. We apply BIFIT to three scientific applicationsand investigate their vulnerability to soft errors. We areable to identify relationships between vulnerabilities andclasses of data structures.
Jeffrey S. VetterOak Ridge National LaboratoryGeorgia Institute of [email protected]
Dong LiOak Ridge National [email protected]
Weikuan YuAuburn [email protected]
MS26
Presentations To Be Announced
Abstract not available at time of publication.
MS27
An Energy- and Charge-conserving, Implicit, Elec-trostatic Particle-in-Cell Algorithm in MappedMeshes
Recently, a fully implicit algorithm for electrostatic plasma
CS13 Abstracts 169
simulation has been proposed in 1D. The approach employsa Jacobian-free Newton-Krylov solver, which is made prac-tical by the nonlinear elimination of particle quantities infavor of field quantities. Its fully implicit character en-ables exact charge and energy conservation, lending theapproach superior accuracy properties. In this talk, wewill introduce the approach, and its extention to mappedmeshes (for mesh adaption) in a multidimensional setting.
Luis ChaconLos Alamos National [email protected]
Guangye [email protected]
Daniel BarnesCoronado [email protected]
MS27
Gyrokinetic Edge Plasma Simulation Using Con-tinuum Methods
Understanding the edge plasma region of a tokamak isimportant to maintaining the confinement of the burningplasma in the core. Because of highly non-equilibrium be-havior, a kinetic plasma description is required over a sig-nificant fraction of the edge region. We will discuss algo-rithmic advances for the 4D gyrokinetic edge plasma codeCOGENT, which is based on a fourth-order finite volumeformulation for mapped, multiblock grids. Results will bepresented for real magnetic geometries.
Jeffrey A. Hittinger, Milo DorrCenter for Applied Scientific ComputingLawrence Livermore National [email protected], [email protected]
Phillip CollelaApplied Numerical Algorithms GroupLawrence Berkeley National [email protected]
Peter MccorquodaleLawrence Berkeley National [email protected]
MS27
Positivity-Preserving Hybrid Semi-Lagrangian DGSchemes for Vlasov-Poisson
The Vlasov-Poisson equations describe the evolution ofa collisionless plasma. The large velocities of the sys-tem create a severe time-step restriction from which thedominant approach in the plasma physics communityis the particle-in-cell (PIC) method. In this work, wepresent a discontinuous-Galerkin method which utilizessemi-Lagrangian methods together with classical RKDGmethods bootstrapped through local time-stepping. Ourmethod utilizes unstructured physical grids which accom-modate complicated geometries. Our method conservesmass and is positivity preserving.
James A. RossmanithIowa State UniversityDepartment of [email protected]
David C. SealDepartment of MathematicsMichigan State [email protected]
Andrew J. ChristliebMichigan State UniverityDepartment of [email protected]
MS27
Toward Gyrokinetic Particle-in-cell Simulations ofFusion Energy Dynamics at the Extreme Scale
Abstract: The Gyrokinetic Particle-in-cell (PIC) methodhas been successfully applied in studies of low-frequencymicroturbulence in magnetic fusion plasmas. While theexcellent scaling of PIC codes on modern computing plat-forms is well established, significant challenges remain inachieving high on-chip concurrency for the new path toexascale systems. In addressing associated issues, it isnecessary to deal with the basic gather-scatter operationand the relatively low computational intensity in the PICmethod. Significant advancements have been achieved inoptimizing gather-scatter operations in the gyrokinetic PICmethod for next-generation multi-core CPU and GPU ar-chitectures. In particular, we will present on new tech-niques that improve locality, reduce memory conflict, andefficiently utilize shared memory on GPUs. Performancebenchmarks on two high-end computing platforms – theIBM BlueGene/Q (Mira) system at the Argonne Leader-ship Computing Facility (ALCF) and the Cray XK6 (Ti-tan) with the latest GPU at Oak Ridge Leadership Com-puting Facility (OLCF) will be presented.
Bei WangPrinceton Institute of Computational Science andEngineeringPrinceton [email protected]
Stephane EthierPrinceton Plasma Physics [email protected]
William TangPrinceton [email protected]
MS28
Simulating Flows in 2020:Challenges for Traditional CFD Applications andWeather/ocean/climate Predictions
We present an analysis of the needs of Computational andGeophysical Fluid Dynamics applications in the time hori-zon of Exascale Era machines. Empasis is given to the thescientific/engineering targets and to the the algorithms ex-pected to be used in relation to the stresses they are ex-pected to place on hardware and software. A discussion ofvarious architectural options is provided along with theirexpected advantages and disadvantages. We also considerthe question of uncertainty quantification and how thattransforms many problems from pure ”capability” onesto mixed ”capacity-capability” ones and discuss how thatchanges the proposed machine balance.
Constantinos Evangelinos
170 CS13 Abstracts
Computational Science CenterIBM T.J. Watson [email protected]
MS28
The Importance of Modeling Solvers: A CaseStudy Using Algebraic Multigrid
As the computers used to solve CSE problems becomelarger and more massively parallel, it is becoming increas-ingly important to develop performance models of thesolvers. In this talk, we present a performance model ofthe solve cycle of algebraic multigrid and illustrate how ithas guided our efforts to improve its scalability on emergingparallel machines. We additionally use the model to makepredictions for solving larger problems on future machines.
Hormozd GahvariUniversity of Illinois at [email protected]
William D. GroppUniversity of Illinois at Urbana-ChampaignDept of Computer [email protected]
Kirk E. JordanIBM T.J. Watson [email protected]
Martin Schulz, Ulrike Meier YangLawrence Livermore National [email protected], [email protected]
MS28
Driving to Exascale - Challenges in Systems andApplications
A new day is dawning for High Performance Computing(HPC) Systems as we see core counts continue to rise.In this talk, we will briefly describe the direction we seecomputer systems taking as we move to exascale, usingthe IBM Blue Gene/Q as an early exemplar. We describesome of the hardware and software challenges that the com-putational science community can address through algo-rithm development for exascale, overviewing this minisym-posium.
Kirk E. JordanIBM T.J. Watson [email protected]
Constantinos EvangelinosComputational Science CenterIBM T.J. Watson [email protected]
Vipin SachdevaIBM Research, [email protected]
MS28
Cloud Computing Pratices for Scientific Comput-ing Applications
Cloud computing provides on-demand access to comput-ing utilities, an abstraction of unlimited computing re-
sources, and support for on-demand scale up, scale downand scale out. Furthermore, dynamically federated infras-tructure can support heterogeneous applications require-ments. Clouds are also joining HPC systems as a viableplatform for scientific exploration and discovery. There-fore, understanding application formulations and usagemodes that are meaningful in such a hybrid infrastructure,and how application workflows can effectively utilize it, iscritical.
Moustafa AbdelBaky, Manish ParasharElectrical and Computer EngineeringRutgers [email protected], [email protected]
Kirk E. JordanIBM T.J. Watson [email protected]
MS29
Probabilistic Predictions of Micro-anomalies fromMacro-scale Response
We will present a probabilistic approach to character-ize continuum (macro-level) constitutive properties of het-erogeneous materials from a limited amount of micro-structural information as typically available in practice.This approach is particularly amenable to include the ef-fects of micro-cracks, that are not discernible by nakedeyes, into the continuum constitutive material properties.Such microlevel defects can be potentially dangerous forstructural systems due to fatigue cyclic loading that resultsin initiation of fatigue cracks. Distinct difference in proba-bilistic characteristics of macro-level responses of computersimulated model is observed depending on presence or ab-sence of micro-cracks. This finding opens up the possibil-ity of detecting micro-cracks (∼10–100 microns size) frommeasurements of macro-level experimental responses (say,of a structure of dimension ∼20 m).
Sonjoy Das, Sourish ChakravartyUniversity at [email protected], [email protected]
MS29
A Stochastic Multiscale Method for the Elastody-namic Wave Equation Arising from Fiber Compos-ites
The long-term structural degradation of composite struc-tures in fiber-reinforced materials is largely influencedby micro mechanical events. Moreover, due to the ran-dom character of the fiber locations and diameters, mate-rial properties, and fracture parameters, most mechanicalquantities must be expressed in statistical terms. We there-fore consider a multiscale problem governed by the linearstochastic elastodynamic wave equation and propose a nu-merical method for computing statistical moments of somegiven quantities of interest in regions of relatively smallsize. The method uses the homogenized global solutionto construct a good local approximation that captures themicroscale features of the real solution. We present numer-ical examples to verify the accuracy and efficiency of themethod.
Mohammad [email protected]
CS13 Abstracts 171
Raul F. TemponeMathematics, Computational Sciences & EngineeringKing Abdullah University of Science and [email protected]
MS29
Quantification of Subscale Effects in HomogenizedModels through Adaptive Information Measures
We consider the problem of computing bulk behaviorof particle systems that are far away from equilibrium.Continuum-level expressions of conservation are closed bysampling the particle dynamics in a predictor-corrector for-mulation. The continuum predictor step constrains theprobable final states of the particle ensemble. We con-centrate on the problem of minimizing prediction errorsthrough variational formulations for the particle probabil-ity density function based on adaptive information mea-sures.
Anil ShenoyUniversity of North [email protected]
Sorin MitranUniversity of North Carolina Chapel [email protected]
MS29
Building Surrogates of Very Expensive ComputerCodes: Applications to Uncertainty Quantification
To account for the epistemic uncertainty induced by thefinite number of evaluations of a computer code, we definea probability measure over the space of possible surrogates.Each sample from this measure is a candidate surrogate forthe code. By quantifying the informational content of theinput space, we devise active learning schemes that areable to enhance the quality of the surrogate for particulartasks. Non-stationarity (localized features, discontinuities)can be captured by employing binary tree models.
Ilias BilionisCenter for Applied MathematicsCornell University, [email protected]
Nicholas ZabarasCornell [email protected]
MS30
Quantifying Uncertainty using a-posteriori En-hanced Sparse Grid Approximations
Adaptive sparse grids are popular for approximating out-puts from high-dimensional simulation codes. When thesecodes solve both a forward and adjoint PDE formulation,a posteriori error estimation techniques can be used toachieve significant increases in the observed rate of con-vergence. The gains in efficiency are achieved by using theerror estimate to both guide adaptivity and enhance thefinal approximation.
John D. JakemanSandia National [email protected]
Tim WildeySandia National [email protected]
MS30
Sparse Grid Data Mining for Approximating High-dimensional Functions
Sparse grid data mining techniques can be used to con-struct mesh-based approximations of high-dimensionalfunctions from randomly positioned data, unlike tradi-tional approaches based on numerical discretizations thatuse interpolation. In this talk we employ sparse grid datamining to construct approximations of functions from un-gridded legacy data. Unlike traditional regression-basedPolynomial Chaos methods such an approach can be usedto approximate both smooth and discontinuous functions,the latter ones requiring suitable adaptive refinement.
Dirk PflugerUniversitat Stuttgart, SimTech-IPVSSimulation of Large [email protected]
MS30
Model Order Reduction for Complex DynamicalSystems in Uncertainty Quantification
We consider systems of ordinary differential equations ordifferential algebraic equations, which result from the mod-eling of technical applications. Typically, uncertainties ap-pear in some parameters of the systems. An uncertaintyquantification is based on the substitution of uncertain pa-rameters by random variables. The corresponding prob-lems become complex in case of a huge dimension of thedynamical systems and/or a large number of random pa-rameters. We investigate approaches for model order re-duction to reduce the complexity of the systems, i.e., ef-ficient numerical methods shall be achieved. On the onehand, parameterized model order reduction provides a re-duction of the computational effort in sampling schemes.On the other hand, stochastic Galerkin methods yield de-terministic systems of an even larger dimension to be re-duced. We present numerical simulations of correspondingapplications.
Roland PulchUniversity of [email protected]
E. Jan W. ter MatenTechnical University of [email protected]
MS31
An Implicit Maxwell Solver based on the Methodof Lines Transpose
We present a novel method for solving the wave equationimplicitly, to address scale separation and complex geome-tries when simulating plasma phenomenon. We apply themethod of lines transpose (MOLT ) to the wave equationto obtain a boundary integral solution. Additionally, wedevelop a fast wave solver in higher dimensions which uti-lizes an ADI splitting to extend the fast one-dimensionalalgorithm that we have developed.
Matthew F. Causley
172 CS13 Abstracts
New Jersey Institute of Technology (NJIT)[email protected]
Andrew J. ChristliebMichigan State UniverityDepartment of [email protected]
MS31
Second Kind Integral Equation Formulation for theModified Biharmonic Equation and its Applica-tions
A system of Fredholm second kind integral equations isconstructed for the modified biharmonic equation in two di-mensions with gradient boundary conditions. Such bound-ary value problem arises naturally when the incompress-ible Navier-Stokes equations in pure stream-function for-mulation are solved via a semidiscretization scheme. Theadvantages of such an approach are two fold: first, thevelocity is automatically divergence free, and second, com-plicated (nonlocal) boundary conditions for the vorticityare avoided. Our scheme can be coupled with standardfast algorithms such as FFT, fast multipole methods, orfast direct solvers to achieve optimal complexity, bring-ing accurate large-scale long-time fluid simulations withinpractical reach.
Mary-Catherine KropinskiSimon Fraser [email protected]
Shidong JiangDepartment of Mathematical SciencesNew Jersey Institute of [email protected]
Bryan D. QuaifeInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
MS31
High Volume Fraction Simulations of Two-Dimensional Vesicle Suspensions
A boundary integral equation method for simulating in-extensible vesicles in 2D viscous fluid was developed byVeerapaneni et al. I will discuss extensions that allow usto consider suspensions with a high concentration of vesi-cles. If time permits, results for simulating tracers, andcomputing pressures and tensors will be presented.
Bryan D. QuaifeInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
George BirosUniversity of Texas at [email protected]
MS31
Simulations of Fiber and Particle Suspensions Ac-celerated by a Spectrally Accurate FFT-basedEwald Method
Numerical methods for 3D simulations of fiber and particle
suspensions as based on boundary integral discretizationswill be discussed. The simulations are accelerated by aspectrally accurate FFT based Ewald method, as appliedto different fundamental solutions and periodic boundaryconditions. This new spectral Ewald method compares fa-vorably to the established so-called SPME method, espe-cially regarding memory. This reduced memory use is a keyfeature in allowing for the simulations that are presented.
Anna-Karin TornbergKTHStockholm, [email protected]
MS32
Using High Level Languages in Petascale Applica-tions with PyClaw
Development of scientific software involves tradeoffs be-tween ease of use, generality, and performance. We em-phasize the importance of Pythonic interfaces to low-levellanguages and libraries. We then describe how an opensource scientific Python stack allows us to implement Py-Claw, a general hyperbolic PDE solver with simply writtenapplication scripts that scale to large supercomputers.
Aron AhmadiaKing Abdullah University of Science and [email protected]
MS32
Using R with Open MPI and CUDA
At SuperComputing 2011, the University of Houston -Downtown received a Little Fe Cluster for teaching andresearch purposes. The software is the Bootable ClusterCD, which contains Open MPI and CUDA. We will dis-cuss the history of the cluster and the software. We willthen consider the uses of R with the cluster and presentsome empirical results to demonstrate speed up.
Erin M. HodgessUniversity of Houston - [email protected]
MS32
LOO.PY: Transformation-Based Programming forLoops
Creating peak-performance compute codes on CPUs andGPUs often requires accomodating machine granularitiessuch as vector widths, core counts, and on-chip mem-ory sizes. Transforming even mathematically simple algo-rithms to suit these restrictions is onerous and error-prone.Loo.py is a tool that automates these transformations, al-lowing practitioners to experiment quickly while helping toensure correct code. Loo.py is built for run-time code gen-eration infrastructure using PyOpenCL, which interfacesPython with many types of compute devices.
Andreas KloecknerCourant Institute of Mathematical SciencesNew York [email protected]
Tim WarburtonDepartment of Computational and Applied MathRice [email protected]
CS13 Abstracts 173
MS32
Bringing Exploratory Analytics to Big Data onLeadership Class HPC Platforms
The need for high-level languages in big data analysiscomes from two challenges. First, we need to easily pro-totype complex analyses with a syntax close to high-levelmathematical expressions and have a diverse toolbox ofknown analytics. Second, portability and optimization ofcodes on large computing platforms is increasingly difficult,calling for higher-level approaches. We begin to solve bothof these challenges with pbdR (r-pbd.org). This project el-evates R, a high-level language for data with arguably themost diverse analytics toolbox, to leadership class HPCplatforms. We do this by a tight coupling with scalableand portable HPC libraries and by building additional in-frastructure to allow high-level management of big data oncomplex architectures.
George OstrouchovComputer Science and Mathematics DivisionOak Ridge National [email protected]
Drew SchmidtUniversity of Tennessee [email protected]
Wei-Chen ChenComputer Science and Mathematics DivisionOak Ridge National [email protected]
Pragneshkumar PatelUniversity of Tennessee [email protected]
MS33
Optimal Runge-Kutta-type Smoothers for TimeDependent Problems
We consider implicit time integration schemes for time de-pendent nonlinear PDEs. The appearing nonlinear systemsare typically solved using either preconditioned JFNK orthe FAS variant of multigrid, where the latter are typicallydesigned for steady problems. This leads to suboptimalschemes. With regards to parallel computations, low stor-age preconditioners respectively multigrid smoothers areimportant. In this talk, we discuss possibilities of findingoptimal low storage Runge-Kutta type smoothers, usingtime dependent model problems.
Philipp BirkenDepartment 10, Mathematics and Natural SciencesUniversity of Kassel, [email protected]
Antony JamesonProfessor, Department of Aeronautics & AstronauticsStanford [email protected]
MS33
Efficient Exponential Integrators for Large StiffSystems of ODEs
We discuss construction, analysis and implementation ofexponential propagation iterative (EPI) methods for solv-
ing large stiff nonlinear systems of ODEs. These schemesallow integration with large time steps compared to ex-plicit methods and computational savings per time stepcompared to implicit and other exponential schemes. Wepresent high-order adaptive Runge-Kutta-type (EPIRK)integrators and compare their performance with otherstate-of-the-art methods on serial and parallel machines.
Mayya TokmanUniversity of California, MercedSchool of Natural [email protected]
John LoffeldUniversity of [email protected]
MS33
Discontinuous Galerkin Methods and ImplicitWave Propagation
We consider discontinuous Galerkin discretizations in spacefor first-order hyperbolic linear systems with heterogeneouscoefficients and full upwind flux. In time we compare stan-dard explicit schemes with sufficiently small time steps,stable implicit Runge-Kutta methods, and exponential in-tegrators, where the linear systems are solved by a parallelmultigrid method with block smoothing. Numerical exam-ples for acoustic, elastic, and electro-magnetic waves arepresented. We show that the implicit methods are moreefficient since the same accuracy in can be achieved withlarger time steps.
Christian WienersInstitute for Applied and Numerical MathematicsKarlsruhe Institute of Technology, [email protected]
MS33
Efficient Unsteady Flow Simulation using GMRES-E with Rosenbrock Time Integration Schemes
In this contribution the computational efficiency of Rosen-brock time integration schemes is compared to multi-stageDIRK and ESDIRK schemes. The computational benefitsof Rosenbrock should come from the linear implicitness andconstant system matrix for all stages within a time step.Solving one time step therefore reduces to solving the samelinear system with a different right-hand-side per stage.In solving the linear systems with GMRES, both the re-tained effectiveness of the preconditioner and the reuse ofKrylov vectors through enrichment improve computationalefficiency.
Alexander H. van ZuijlenFaculty Aerospace EngineeringDelft University of Technology, [email protected]
David BlomDelft University of [email protected]
Hester BijlFaculty Aerospace EngineeringDelft University of Technology, [email protected]
174 CS13 Abstracts
MS34
Global/Local Surrogate Based Reservoir Manage-ment Optimization
In the waterflooding optimal management problem the ob-jective is to maximize the NPV using as controls the ratesof injector and producer wells. The duration of each cyclemay also be included as design variables. As the numeri-cal simulation has high computational cost we use krigingbased surrogate models. The trust region based SequentialApproximate Optimization technique is used for local op-timization. As the problem may be multimodal we use ahybrid Genetic Algorithm strategy.
Bernardo HorowitzUniversidade Federal de Pernambuco - UFPECivil Engineering [email protected]
Silvana Afonso, Leonardo OliveiraUniversidade Federal de PernambucoCivil Engineering [email protected], [email protected]
MS34
Efficient EnKF Conditioning with Reduced-OrderModeling
In this work we propose to use reduced-order modelingtechniques to reduce the computational burden of theensemble Kalman filter for large-scale problems. Thereduced-order model proposed here can perform dimensionreduction in both state and parameter spaces. The high-dimensional state space is reduced by the proper orthogo-nal decomposition, while the high-dimensional parameterspace is reduced by the discrete cosine transform. Theefficiency and robustness of the reduced-order model aredemonstrated by an uncertainty quantification example ofsubsurface transport.
Binghuai LinMITDepartment of Civil and Environmental [email protected]
Dennis McLaughlinCivil Engineering [email protected]
MS34
Using Geological Uncertainty to Simplify History-Match Optimization
Geological realism in history-matching is fundamental toaccurate reservoir model predictions. We frame this asa Bayesian optimization problem, fitting the observableswhile conforming to prior uncertainties. Principal compo-nent analysis greatly reduces the number of optimizationvariables by describing only the intrinsic degrees of free-dom allowed by the prior. Using the SPE Brugge Chal-lenge model as a test case, we demonstrate improvementover previously published data, with higher NPV and lowerpredictive error.
Michael Prange, William BaileySchlumberger-Doll [email protected], [email protected]
Thomas Dombrowsky
Longhorn Technology [email protected]
MS34
Efficient Surrogate Surface Global Optimizationfor Estimating Carbon Sequestration Plumes withSparse Observations
Estimation of sequestered CO2 and pressure plumes is im-portant but difficult because monitoring data is very sparseand inverse optimization problem has multiple local op-tima. Each objective function evaluation requires expen-sive forward simulation of 3-D, highly nonlinear, multi-phase, multi-constituent set of PDEs. We get good currentestimates and forecasts of plumes with a surrogate responsesurface global optimization algorithm Stochastic RBF withsmall number of original model simulations. Variants of thealgorithm are presented.
Christine A. Shoemaker, Antoine EspinetCornell [email protected], [email protected]
Christine DoughtyLawrence Berkeley National [email protected]
MS35
A Stochastic Dimensional Reduction Multiscale Fi-nite ElementMethod with Applications for Subsur-face Flows in Random Porous Media
Stochastic modeling has become a widely accepted ap-proach to quantify uncertainty of flows in random porousmedia. To efficiently treat the high-dimensionality of thestochastic space and the inherent heterogeneity of porousmedia, we propose a new truncated high dimensional modelrepresentation technique and combine it with a mixed mul-tiscale finite element method. To capture the non-localspatial features of the media and the effects of impor-tant random variables, we hierarchically incorporate someglobal information individually from each of random pa-rameters. Numerical experiments are carried out for sub-surface flows in random porous media.
Lijian JiangIMA, University of [email protected]
David MoultonLos Alamos National LaboratoryApplied Mathematics and Plasma [email protected]
MS35
Geometric Integration and Analysis of GeneralMultiscale Systems
In order to accelerate computations, improve long time ac-curacy of numerical simulations, and sample statistics dis-tribution by dynamics, we develop multiscale geometric in-tegrators. These integrators employ coarse steps that donot resolve the fast timescale in the system; nevertheless,they capture the effective contribution of the fast dynam-ics. Distinct from existing approaches, an identification ofunderlying slow variables is not required, and intrinsic ge-ometric structures (e.g., symplecticity, conservation laws,
CS13 Abstracts 175
and invariant distribution) can be preserved by the numer-ical simulation.
Molei TaoCourant [email protected]
Houman OwhadiApplied [email protected]
Jerrold MarsdenCaltech(deceased)
MS35
Multiscale Discontinuous Galerkin Method for El-liptic Equations with Rapidly Oscillatory Coeffi-cients
In this talk, a special discontinuous Galerkin method isproposed for a class of second order elliptic problems withrough coefficients, whose local oscillating directions varysmoothly within the domain. The key ingredient of themethod lies in choosing special approximation space to cap-ture the multiscale solutions without having to resolve thefinest scale therein. Theoretical proof and numerical ex-amples would be presented for the second order method intwo dimensional case.
Yifan ZhangBrown Universityyifan [email protected]
Wei WangDepartment of Mathematics and StatisticsFlorida International [email protected]
Johnny GuzmanBrown Universityjohnny [email protected]
Chi-Wang ShuBrown UniversityDiv of Applied [email protected]
MS35
Numerical Homogenization: From Higher OrderPoincare Inequalities to Optimally Localized Basis
We introduce a new method for the numerical homoge-nization of divergence form elliptic equations with rough(�L∞) coefficients. Our method does not rely on conceptsof ergodicity or scale-separation but on the property thatthe solution operator is compact from �L2 to H1. The ap-proximation space is generated as an interpolation space(over a coarse mesh of resolution H) minimizing the �L2
norm of source terms; its (pre-)computation involves solv-
ing O(H−�) bi-harmonic PDEs on localized sub-domainsof size O(H(lnH)∈); its accuracy (O(H) in H1 norm) isestablished via the introduction of a new classs of higher-order Poincare inequalities. The method naturally gener-alizes to time dependent problems.
Lei Zhang
University of [email protected]
MS36
Parallel Constrained Minimization Methods
Abstract not available at time of publication.
Rolf KrauseInstitute of Computational ScienceUniversity of [email protected]
MS36
Sparse Quadrature Approach to Bayesian InverseProblems
In this talk, we present a novel, deterministic approach toinverse problems for identification of parameters in differ-ential equations from noisy measurements. Based on theparametric deterministic formulation of Bayesian inverseproblems with input parameter from infinite-dimensional,separable Banach spaces, we develop a practical compu-tational algorithm for the efficient approximation of theinfinite-dimensional integrals with respect to the posteriormeasure.
Claudia SchillingsETH ZurichSeminar for Applied [email protected]
Christoph SchwabETH [email protected]
MS36
Multigrid Optimization on GPGPU
GPGPU is a cost-effective means for the usage of a paral-lel computer with several hundred cores. In this talk, thecomputational potential of this architecture for optimiza-tion problems is investigated. In particular, a specific im-plementation of multi grid methods for large optimal con-trol problems on several stream processors are discussed,where a domain decomposition approach is employed forthe distribution on multiple GPU.
Volker H. SchulzUniversity of TrierDepartment of [email protected]
Christian WagnerDepartment of MathematicsUniversity of [email protected]
MS36
GPU Accelerated Discontinuous Galerkin Methodsfor Simulation and Optimization
Recent studies have shown advantages of the application ofstreaming processors for highly arithmetic expensive prob-lems arising from partial differential equations. This talkintroduces a high-order discontinuous Galerkin implemen-
176 CS13 Abstracts
tation on GPUs for Euler equations with focus on its par-allel performance. In particular, GPU acceleration for thecorresponding discrete adjoint is addressed and its appli-cation in optimization is discussed. The importance of ac-curate boundary resolutions is demonstrated and a meshcurvature approach dealing with complex domains is pre-sented.
Martin SiebenbornUniversity of [email protected]
Volker H. SchulzUniversity of TrierDepartment of [email protected]
MS37
A hp-nonconforming Discontinuous Galerkin Spec-tral Element Method: Analysis and Application toLarge Scale Seismic Inversions
We analyze the consistency, stability, and convergence of anhp discontinuous Galerkin spectral element method. Ouranalytical results are developed for both conforming andnon-conforming approximations on hexahedral meshes. Amortar-based non-conforming approximation is developedto treat both h and p non-conforming meshes simultane-ously. We demonstrate the scalability and accuracy of theproposed method on large-scale global seismic wave inver-sion.
Tan Bui-ThanhThe University of Texas at [email protected]
Omar GhattasUniversity of Texas at [email protected]
MS37
Output-based Space-time Adaptation for DG Sim-ulations with Moving Meshes
We address the question of how to adapt a mesh for prob-lems with deformable domains, like a wing in flappingflight, to obtain accurate outputs. We present a space-timeadaptation strategy that uses output error estimates at-tained from a set of discrete, unsteady adjoint solutions.We derive an additional adjoint for the geometric conser-vation law, and show the accuracy and efficiency of ourstrategy for Navier-Stokes problems solved within a DGframework.
Steve Kast, Krzysztof FidkowskiUniversity of [email protected], [email protected]
MS37
Discontinuous Petrov-Galerkin Finite ElementMethod for the Analysis of Plate and Shell Struc-tures
Stress recovery and numerical locking present common dif-ficulties encountered in the finite element analysis of plateand shell structures. One way to address these issues isto base the numerical discretization on a mixed variationalprinciple where the stresses are declared as independent
unknowns next to the displacements. The discontinuousPetrov-Galerkin (DPG) finite element methodology pro-vides a systematic approach to devise such a variationalformulation and ensure its stability also in the discrete (fi-nite dimensional) setting. In this talk, we discuss recentdevelopments in the analysis of plate and shell deforma-tions utilizing the DPG framework.
Antti H. NiemiAalto University, [email protected]
MS37
A High-Order Implicit-Explicit DiscontinuousGalerkin Scheme for Fluid-Structure Interaction
We present a discontinuous Galerkin scheme for fully cou-pled fluid-structure interaction problems. A high-orderDG-ALE formulation is used for the fluid, and standardCG schemes for beams and membranes are used for thesolids. Using implicit-explicit Runge-Kutta time integra-tors with explicit predictors for the coupling variables, wecan integrate each problem domain independently usingefficient parallel implicit solvers while retaining the highformal order of accuracy.
Per-Olof PerssonUniversity of California BerkeleyDept. of [email protected]
Bradley FroehleUniversity of California, BerkeleyDept. of [email protected]
MS38
ComparativeReview of Fractional-step Approaches to the Im-mersed Boundary Method
In the various immersed boundary methods based ona fractional-step approach, there are either unjustifiedboundary conditions or overlooked conservation proper-ties. Often, this is caused by mixing continuum and dis-crete equations in the analysis. With a careful, fully dis-crete derivation, we construct a unifying perspective of IBmethods that resolves these issues. We have developed anopen-source GPU code, cuIBM, that provides a commonframework for IB methods integrating these improvements.
Lorena A. BarbaDepartment of Mechanical EngineeringBoston [email protected]
Anush KrishnanBoston [email protected]
MS38
Increased Accuracy of Immersed Boundary Meth-ods Using Fourier Approximations of Delta Func-tions
In immersed boundary methods, the fluid and structurecommunicate through smoothed approximate delta func-
CS13 Abstracts 177
tions with small spatial support. We take a different ap-proach and construct highly accurate approximations tothe delta function directly in Fourier space. This methodleads to high-order accuracy away from the boundary andsignificantly smaller errors near the boundary. We presentaccuracy tests and simulation results from an applicationin cell biology in which large errors in the traditional IBmethod produce unphysical results.
Robert D. GuyMathematics DepartmentUniversity of California [email protected]
David HartenstineDepartment of MathematicsWestern Washington [email protected]
Wanda StrychalskiDepartment of MathematicsUniversity of California, [email protected]
MS38
Immersed Boundary Method for the Incompress-ible Navier-Stokes Equations Based on the LatticeGreen’s Function Method
A parallel, three-dimensional immersed boundary methodis developed to solve external, viscous incompressible flowson an infinite domain. The equations are formally dis-cretized on an infinite staggered Cartesian grid. The Lat-tice Green’s Function method is used to reduce the problemto a finite region. The fully discrete equations are obtainedby combining an integrating factor technique and a half-explicit Runge-Kutta scheme, and solved using a nestedprojection technique. Results for test problems are pre-sented.
Sebastian LiskaMechanical and Civil EngineeringCalifornia Institute of [email protected]
Tim ColoniusDivision of Engineering and Applied ScienceCalifornia Institute of [email protected]
MS38
Dynamics of an Elastic Rod with Curvature andTwist: Stokes Formulation
We develop a Lagrangian numerical algorithm for an elas-tic rod immersed in a viscous, incompressible fluid at zeroReynolds number. The elasticity of the rod is describedby a version of the Kirchhoff rod model. The coupling tothe fluid is accomplished by the use of the method of regu-larized Stokeslets for the force, and regularized rotlets anddipoles for the torque. This method will be compared tothe generalized IB method and asymptotic results.
Sarah D. OlsonWorcester Polytechnic [email protected]
Sookkyung LimUniversity of Cincinnati
Ricardo CortezTulane UniversityMathematics [email protected]
MS39
Linear Algebra Libraries with DAG Runtimes onGPUs
Nowadays many clusters integrate GPUs accelerators intheir architectures that provide a huge amount of compu-tational units rarely fully exploited. We present in thistalk how tile algorithms and DAG schedulers as PaRSECor StarPU can allow the programmer to integrate GPUsin their algorithms. We will present dense linear algebraalgorithms as Cholesky or LU factorizations that exploitdistributed architectures equipped with GPUs.
George BosilcaUniversity of Tennessee - [email protected]
Aurelien BouteillerUniversity of [email protected]
Mathieu FavergeU. of Tennessee - [email protected]
Thomas HeraultUniv. of Tennessee - [email protected]
MS39
Parallel LU Factorizations on GPUs in AORSA
We describe the performance of a parallel dense matrixsolver in AORSA, the All ORders Spectral Algorithm fu-sion application for modeling the response of plasma toradio frequency waves in a tokamak device, which takesadvantage of Graphics Processing Unit (GPU) accelerationand is compatible with ScaLAPACK LU factorization rou-tines. The left-looking out-of-core algorithm factors ma-trices that are larger than the available GPU memory andachieves 170 GFlops (double precision) per GPU-equippednode on the CrayXK6.
Judith Hill, Ed D’Azevedo, David GreenOak Ridge National [email protected], [email protected], [email protected]
MS39
A Performance Study of Solving a Large Dense Ma-trix for Radiation Heat Transfer
This talk presents a performance study of solving a largedense matrix arising from the thermal radiation problem.The radiosity exchanged among a number of radiating sur-faces depends on their view factors. The radiosity matrixis a strictly diagonally dominant matrix. But in a limitednumber of applications the radiosity matrix is also SPD.We will present the LU and LLT procedures and resultsused in solving the matrix on a GPU cluster.
Kwai L. Wong
178 CS13 Abstracts
Joint Institute for Computational ScienceUniversity of Tennessee/[email protected]
Edurado D’[email protected]
Harvy HuChinese University of Hong [email protected]
Shiquan SuUniversity of [email protected]
MS39
Mult-GPU Tridiagonalzation onShared-and-distributed-memory Systems
Tridiagonalization of a symmetric dense matrix is an im-portant computational kernel in many symmetric eigen-value problems. In this talk, we describe our extensionof one-stage tridiagonalization to use multiple GPUs onshared- and distributed-memory computers. We presentexperimental results to demonstrate that the tridiagonalreduction time of LAPACK or ScaLAPACK can be reducedsignificantly using GPUs.
Tingxing [email protected]
Jack J. DongarraUniversity of [email protected]
Stanimire TomovInnovative Computing Laboratory, Computer ScienceDeptUniversity of Tennessee, [email protected]
Ichitaro [email protected]
MS40
A Structured Quasi-Arnoldi Procedure for ModelOrder Reduction of Second-order Systems
Most Krylov subspace-based structure-preserving modelorder reduction methods for the second-order systems con-sist of two stages. The first stage is to generate basis vec-tors of the underlying Krylov subspaces. The second stageis to perform explicit subspace projections via matrix-matrix multiplications to generate the reduced-order mod-els. For very large scale systems, the second stage couldbe prohibitively expensive due to the costs of data storageand communication. In this talk, we discuss a StructuredQuasi-Arnoldi (SQA) procedure to avoid the second stage.The SQA procedure is based on a Krylov-type decom-position that allows us to derive the structure-preservingreduced-order model directly from the computed Krylov-type decomposition. Numerical examples will be presentedto illustrate accuracy and efficiency of the SQA procedure.
Yung-Ta LiFu Jen Catholic University, [email protected]
Wen-wei LiNational Chiao-Tung University, [email protected]
Zhaojun BaiDepartments of Computer Science and MathematicsUniversity of California, [email protected]
MS40
A New Approach to Model Order Reduction of theNavier-Stokes Equations
A new approach to model order reduction of the Navier-Stokes equations is proposed. Unlike traditional ap-proaches, this method does not rely on empirical turbu-lence modeling or modification of the Navier-Stokes equa-tions. It provides spatial basis functions different from theusual proper orthogonal decomposition basis function inthat, in addition to optimally representing the solution,the new basis functions also provide stable reduced-ordermodels. The proposed approach is illustrated with twotest cases: two-dimensional flow inside a square lid-drivencavity and a two-dimensional mixing layer.
Maciej Balajewicz, Earl DowellDuke [email protected], [email protected]
Bernd NoackInstitut PPRIME, [email protected]
MS40
Reduced Order Models Preserving Spectral Con-tinuity for Wave Propagation in Unbounded Do-mains
We address the main issues obstructing the application ofthe Krylov subspace based model reduction to the time-domain solution of the exterior wave problems. To avoidspurious reflections and instability, the ROM should pre-serve in some sense delicate spectral properties of the orig-inal problem, i. e., absolutely continuous spectral measuresupported on real negative semiaxis. We introduce such aROM, give its rigorous justification, discuss approaches topreconditioning and present large scale seismic examples.
Vladimir L. DruskinSchlumberger-Doll [email protected]
Leonid KnizhnermanSchlumberger, [email protected]
Olga PodgornovaSchlumberger Moscow [email protected]
Rob RemisCircuits and Systems Group
CS13 Abstracts 179
Delft University of [email protected]
Mikhail ZaslavskySchlumberger-Doll [email protected]
MS40
A Hyper-Reduction Method for Nonlinear Dy-namic Finite Element Models
A model order reduction method for nonlinear systems aris-ing from the discretization of second-order hyperbolic prob-lems using a finite element method in space and a finitedifference method in time is presented. Its main compo-nents are a Galerkin projection onto a reduced-order basisconstructed using snapshots and the proper orthogonal de-composition method, a hyper-reduction scheme that pre-serves important properties of both the physical systemand its finite element representation, and a fast computerimplementation.
Charbel FarhatStanford [email protected]
MS41
Automating Adjoints for Earth Science Applica-tions
Adjoint models are crucial ingredients in solving inverseproblems in the geosciences. However, deriving adjointmodels is typically very difficult. I present a new tech-nique for rapidly deriving adjoints of finite element mod-els via symbolic analysis and code generation (based onthe FEniCS project). I give examples of its application inocean circulation, mantle convection and ice modelling.
Patrick FarrellDepartment of Earth Science and EngineeringImperial College [email protected]
Simon W. FunkeImperial College LondonDepartment of Earth Science and [email protected]
David HamImperial [email protected]
Marie E. RognesSimula Research [email protected]
MS41
Second-order Adjoints in Seismic Tomography
We present an introductory and general review of second-order adjoints for the efficient computation of Hessian-vector products. Using the specific example of large seis-mic tomographic inverse problems, we demonstrate howsencond-order adjoints can be used to accelerate conver-gence towards an optimal Earth model, and to analyse res-olution and trade-offs in the final solution.
Andreas Fichtner
Utrecht [email protected]
Naiara KortaBarcelona Center for Subsurface [email protected]
Jeannot TrampertUtrecht [email protected]
MS41
Estimating the Rheology of the Earth’s Mantle: AnApplication of the Adjoint Method in Geodynamics
Determining rheologic parameters of the Earth with thehelp of numerical simulations of mantle flow is one of theessential tasks in geodynamics. But where standard for-ward simulations suffer from the problem of an unknowninitial condition and instantaneous models are non-uniquewith respect to rheology (Schaber et al. (2009)), the ad-joint method in geodynamics can be used to create a time-series of the temperature distribution of the Earth’s mantleover the last 40 Ma that is both sensitive to rheology andconsistent with present-day observations.
Andre HorbachDepartment of Earth and Environmental Sciences, [email protected]
Hans-Peter BungeLMU Munich, [email protected]
MS41
Calibration of Stratigraphical Models
Calibration of the model is a type of inverse problem thatestimates coefficients on the basis of observations. Deposi-tional models were chosen because of their significance inthe basin research and the available observations: seismicstudies provide the thickness of the the deposited layers,while well-log data give the information on the type of sed-iments and the history of deposition.
Lyudmyla Vynnytska, Stuart ClarkSimula Research [email protected], [email protected]
MS42
An Algorithm for Shape Detection in ComputedTomography
Computed tomography (CT) is essential to modernmedicine. Usually practitioners’ interest in CT is to iden-tify and quantify structures found in CT images. Tradi-tionally, this is done by post-processing of reconstructedimages (i.e. segmenting the images). In this talk, we pro-pose to address this problem directly; we extract the geo-metric structures directly from measured data, bypassingthe reconstruction phase. We formulate this task as a shapeoptimization problem and devise an efficient numerical al-gorithm to perform the optimization.
Gunay DoganTheiss Research, [email protected]
180 CS13 Abstracts
MS42
Image Segmentation Methods Based on CentroidalVoronoi Tessellation and Its Variants
In this talk we review some recent progresses on imagesegmentation methods based on centroidal Voronoi tessel-lation and its variants. The classic CVT model, the edge-weighted CVT (EWCVT) model, the local variation andedge-weighted CVT (LVEWCVT) model, and their imple-mentation algorithms will be discussed in detail. We willalso illustrate and compare these interesting segmentationmethods through extensive numerical examples.
Jie WangArizona State [email protected]
Lili JuUniversity of South CarolinaDepartment of [email protected]
Xiaoqiang WangFlorida State [email protected]
MS42
Image Registration for the Future: Fast, Scalable,and Highly Memory-Efficient Algorithms
Image registration is a fundamental problem in many imag-ing applications. The ever growing image sizes and newmodalities pose a significant challenge for efficient algo-rithms. In the first part of the talk, we present techniquesfor fully matrixfree implementations of multimodal para-metric registration algorithms. Besides of virtually reduc-ing memory consumption to zero, an excellent scalabilityon multi-core architectures is achieved. The second partof deals with selected computational aspects of nonlinearapproaches yielding diffeomorphic transformations.
Jan RuhaakInstitute of Mathematics and Image ComputingUniversity of [email protected]
MS42
A Mesh Warping Algorithm for Brain Biomechan-ics Boundary Evolution Tracking
Hydrocephalus is a neurological disease which causes ven-tricular dilation due to abnormalities in the cerebrospinalfluid circulation. Similarly, a brain tumor involves the ab-normal growth of brain cells. To aid neurosurgeons intreatment planning, we propose an automated geometriccomputational approach for tracking evolution of the ap-plicable brain boundaries via the level set method and amesh warping technique. Our method uses brain geome-tries taken from medical images and may incorporate otherrelevant clinical information.
Suzanne M. ShontzThe Pennsylvania State [email protected]
Corina DrapacaDepartment of Engineering Science and MechanicsPennsylvania State [email protected]
MS43
Pulling Fibers
Abstract not available at time of publication.
Linda CummingsDepartment of Mathematical SciencesNew Jersey Institute of [email protected]
MS43
An Application of Matrix Theory to the Evolutionof Coupled Modes
In order to overcome loss in optical fibers, experimental-ists are interested in employing parametric amplifiers us-ing four-wave mixing. A variety of other signal-processingfunctions also use parametric amplifiers based on four-wavemixing. Upon linearizing the nonlinear Schrodinger equa-tion typically used to model such amplifiers, one obtainsa system of ODEs for the complex amplitude. The so-lution of this system can be expressed as the product oftransfer matrices with the initial condition and its conju-gate. Physical insight into the fiber-optic system can beobtained by examining the theoretical properties of thesematrices. This presentation explores these properties.
Joseph D. FehribachWorcester Polytechnic [email protected]
David A. EdwardsUniversity of DelawareNewark, [email protected]
Richard O. MooreNew Jersey Institute of [email protected]
Colin J. McKinstrieBell Laboratories, [email protected]
MS43
Modeling Glass Temperature in a Tempering Fur-nace
Equations governing the temperature distribution in aglass plate as it progresses through a tempering furnace areformulated. Heat transfer due to natural convection, forcedconvection from an array of impinging gas jets, conductiondue to contact with supporting rollers, and radiative heat-ing from heating elements is described. Two and three di-mensional time-dependent numerical simulations allow usto explore how to control the glass temperature profile byadjusting heating conditions.
Harrison PotterDuke UniversityDepartment of [email protected]
MS43
A Homogenization Analysis of the CompressibleFlow Between a Slider and a Moving Rough Sur-
CS13 Abstracts 181
face
The compressible flow between a slider and a moving roughsurface is examined asymptotically and numerical in thelimit of very small gap height. The amplitude and wave-length of the roughness are assumed to be of the order ofthe gap height. A two-scale homogenization analysis is em-ployed to determine a nonlinear elliptic partial differentialequation governing the leading-order pressure in the gapon the scale of the slider. The equation involves coefficientfunctions which are determined numerically by averagingStokes flows on the scale of the roughness. Comments anda brief analysis is given on the reduction of the governingequation for pressure in the limit of long wavelength of thesurface roughness.
Burt S. TilleyMathematical Sciences DepartmentWorcester Polytechnic [email protected]
Donald W. SchwendemanRensselaer Polytechnic InstituteDepartment of Mathematical [email protected]
Colin PleaseSchool of MathematicsUniversity of [email protected]
Ferdinand HendriksHGST, a Western Digital [email protected]
MS44
Performance Modeling of the Eigen-K DenseEigensolver on Massively Parallel Machines
We aim for developing an efficient dense eigensolver onmassively parallel machines such as the K computer inJapan. In our consideration, performance modeling of thesolver is helpful to performance tuning. In this talk, weintroduce an approach to modeling the performance of theEigen-K solver developed by Imamura, where the parallelexecution time of the solver is estimated from the singlenode performance and the internode communication per-formance.
Takeshi FukayaKobe [email protected]
Toshiyuki ImamuraThe University of [email protected]
Yusaku YamamotoKobe [email protected]
MS44
A Hybrid CPU-GPU Generalized Eigensolver forElectronic Structure Calculations
The adoption of hybrid GPU-CPU nodes in traditionalsupercomputing platforms opens acceleration opportuni-
ties for electronic structure calculations in materials sci-ence and chemistry applications, where medium-sized gen-eralized eigenvalue problems must be solved many times.We developed a novel, architecture aware algorithm thatis state-of-the-art in HPC, significantly outperforming ex-isting libraries. We describe the algorithm and analyzeits performance impact on applications of interest whendifferent fractions of eigenvectors are needed by the hostelectronic structure code.
Azzam HaidarDepartment of Electrical Engineering and ComputerScienceUniversity of Tennessee, [email protected]
Raffaele SolcaETHEidgen{o}ssische Technische Hochschule Z{u}[email protected]
Stan TomovComputer Science DepartmentUniversity of [email protected]
Jack DongarraUniversity of [email protected]
Thomas SchulthessETHEidgenossische Technische Hochschule [email protected]
MS44
ELPA: A Highly Scalable Eigensolver for PetaflopApplications
The symmetric eigenproblem is of great interest in manyscientific disciplines (e.g., quantum chemistry, networkanalysis) and represents a severe bottleneck for several nu-merical simulations. ELPA is a new direct eigensolver,which has been consequently designed towards massivelyparallel systems. Compared to state-of-the-art librariesfor distributed memory systems, it leads to significant im-provements in both, efficiency and scalability. In this talkwe present the ELPA library and address the most recentdevelopments and results.
Thomas AuckenthalerTU Munchen Institut fur [email protected]
Alexander HeineckeTU [email protected]
Hermann LedererRechenzentrum [email protected]
Thomas K. HuckleInstitut fuer InformatikTechnische Universitaet [email protected]
182 CS13 Abstracts
MS44
Eigensolvers in Combustion Simulations
Thermoacoustic instabilities are an important concern inthe design of gas turbine combustion chambers. Theirstudy leads to the numerical solution of large sparse un-symmetric eignevalue problems. In this talk we discussvarious numerical techniques to exploit a priori informa-tion on the sought eigenmodes. We illustrate the featuresof these different techniques on small toy examples as wellas on large scale industrial calculations.
Luc GiraudINRIAToulouse, [email protected]
Pablo SalasInria - Bordeaux Sud Ouestjont Inria-CERFACS lab on [email protected]
Vasseur [email protected]
MS45
Guaranteed Stability and Passivity of Reduced Or-der Models
Some efficient model reduction methods fail to guaranteeimportant properties of the reduced system, such as sta-bility and passivity. In this contribution we will show howmethods from control theory can be combined with modelreduction methods in order to guarantee these properties.We will concentrate on the application of such methods tointerpolatory reduction approaches without deteriorationof the interpolation properties. We will apply the resultsto challenging examples from S-parameter design in elec-tronic devices.
Athanasios C. AntoulasDept. of Elec. and Comp. Eng.Rice [email protected]
MS45
Fast Solver for Large Scale Inverse Problems usingData-driven Reduced Order Models
A novel data-driven model reduction technique is devel-oped for solving large-scale inverse problems. The pro-posed technique exploits the fact that the solution of theinverse problem often concentrates on a low dimensionalmanifold. Unlike typical MCMC approaches for solvingthe inverse problem, our approach avoids repeated evalua-tion of expensive forward models by coupling the inferencealgorithm with the reduced-order model. This maintainsthe accuracy of the inference and also results in a lower-dimensional reduced model than obtained with the typicalPOD approach.
Tiangang [email protected]
Youssef M. Marzouk, Karen E. WillcoxMassachusetts Institute of [email protected], [email protected]
MS45
A Reproducing-Kernel Framework for H∈ ModelOrder Reduction
The Iterative Rational Krylov (IRKA) algorithm for modelorder reduction has recently attracted attention because ofits effectiveness in real world applications. The key ideais to construct a reduced order model that satisfies a setof necessary optimality conditions formulated as interpo-latory conditions: the reduced r-th order transfer functioninterpolates the full n-th order function and its deriva-tive(s) in the reflected (about the imaginary axis) images ofthe reduced order poles. This is formulated as a fixed pointproblem, and the interpolation nodes are generated asσ(k+1) = φ(σ(k)). Here φ(·) computes the eigenvalues of thePetrov-Galerking projection of the state matrix to rational
Krylov subspaces computed at σ(k) = (σ(k)1 , . . . , σ
(k)r ). The
most expensive part of the IRKA is computing the trans-fer function at a sequence of dynamically generated pointsin the right half-plane C+. Also, this setting is not suit-able for a data driven framework, since we are given onlythe transfer function values along the imaginary axis. Onthe other hand, if we interpret the necessary optimalityconditions as orthogonality (in H∈) of the residual to thetangent space at the reduced order model to the manifoldof the models of given reduced order, we can deploy richtheory of the geometry of linear systems, as well as repro-ducing kernel space property of H∈. All action takes placeon iR, and, with a suitable kernel, function evaluation canbe achieved by a linear functional, implemented using nu-merical quadrature. This new framework is well suited fordata driven applications. It also offers efficient implemen-tation and deeper insight in the behavior and properties ofthe numerical algorithm.
Zlatko DrmacUniversity of ZagrebDepartment of [email protected]
Christopher A. BeattieVirginia Polytechnic Institute and State [email protected]
Serkan GugercinVirginia Tech.Department of [email protected]
MS45
Model Reduction for Indoor-Air Behavior inEnergy-Efficient Building Design
We present a two-step approach for generating reduced or-der models for the indoor-air environment in control de-sign for energy efficient buildings. Using a data-drivenmodel reduction approach, we first construct an interme-diate reduced-model directly from input and output mea-surements. We then apply optimal interpolatory modelreduction techniques to this intermediate model to obtainthe final reduced model. Numerical results illustrate thatthe reduced model accurately represents the input-outputbehavior of the full-order system.
Jeff BorggaardVirginia TechDepartment of [email protected]
CS13 Abstracts 183
Eugene CliffVirginia TechInterdisciplinary Center for Applied [email protected]
Serkan GugercinVirginia Tech.Department of [email protected]
MS46
Issues in the Management and Analysis of LargeData Sets Arising in Complex Problems
Reliable fast and accurate estimation of useful quantitiesof interest from analysis of large computational and ex-perimental data sets and from modeling and simulation ofcomplex systems involving multi-physics and multi-scalepose many significant challenges that require sophisticatedmathematical and statistical tools. In this talk, we will dis-cuss some of these challenges and their solutions against thebackdrop of some complex problems arising in heteroge-neous environment such as turbulence, porous media flowsand so on. This talk will be based on an ongoing work finedetails of which will have to wait until the talk.
Prabir DaripaTexas A&M UniversityDepartment of [email protected]
MS46
Numerical Investigation of the Effective Propertiesof Tight Fractured Porous Shale Rock
We focus on the effect of micro-pores and micro-fracturesand their interactions on the overall effective mechanical,thermal and hydrological flow and transport properties oftight fractured porous shale rock under several conditionsof uncertainties. Uncertainty associated with the physi-cal and chemical aggregates of tight shale rock and theirparametrization is investigated using brute-force Monte-Carlo method.
Souheil M. [email protected]
MS46
Covariance Models Based on Local Interaction(Spartan) Functionals
We derive two-dimensional covariance functions corre-sponding to Gibbs random fields based on quadratic en-ergy functionals with local spatial interactions. We thendevelop the Karhunen-Loeve representation and explicitsolutions for simple boundary geometries. We illustratethe parametric dependence of the realizations using two-dimensional Karhunen-Loeve expansions. We develop in-verse covariance functions in d ∈ Z+-dimensions using afrequency cutoff and approximating the respective spec-tral integral. Finally, we discuss potential applications inspatial interpolation and simulation problems.
Dionissios T. HristopulosTechnical University of [email protected]
MS46
The Effect of Material Heterogeneity in ComputingLocal Deformation Effects
Model predictions of large deformation geologic structuresare based on constitutive models that are typically cali-brated using experimental data that assume material ho-mogeneity at a micro- to meso-scale. This work showscomputational representations of the kinematics of gran-ular materials using Discrete and Finite Elements as anattempt to identify and characterize the main sources ofmaterial heterogeneity as stochastic processes, and the im-pact these have on the simulation of the material’s mecha-nistic performance.
Zenon Medina-CetinaZachry Department of Civil EngineeringTexas A&M [email protected]
Ahran SongStochastic Geomechanics LaboratoryTexas A&M University, TX, [email protected]
Patrick R. NobleTexas A&M [email protected]
Tam DuongStochastic Geomechanics LaboratoryTexas A&M University, TX, [email protected]
MS47
Robust Integral Solver for 3D Acoustic Scatteringfrom Doubly-Periodic Media
We construct a new class of high-order accurate integralequation solvers for the scattering of time-harmonic scalarwaves from a doubly-periodic array of smooth obstacles,or a surface with a doubly-periodic array of bumps. Thiscombines recent QBX surface quadratures with a simplequasi-periodizing scheme based on matching on unit cellwalls. No quasi-periodic Green’s function nor lattice sumsare needed; the solver is thus robust for all scattering pa-rameters including Wood’s anomalies.
Alex BarnettDepartment of MathematicsDartmouth [email protected]
Leslie Greengard, Zydrunas GimbutasCourant InstituteNew York [email protected],[email protected]
MS47
A Fast Direct Solver for Quasi-periodic ScatteringProblems
In this talk, we present an integral equation based directsolver for the two dimensional scattering of time-harmonicplane waves from an infinite periodic array of obstacles in ahomogeneous background medium. By coupling a recentlydeveloped periodizing scheme, potential theory and ”fast”
184 CS13 Abstracts
direct linear algebra, the direct solver has O(N) complexityand is robust for all incident angles. For design problemswhere multiple incident angles are needed, the solver isextremely efficient.
Adrianna GillmanUniversity of Colorado at BoulderDepartment of Applied [email protected]
Alex BarnettDepartment of MathematicsDartmouth [email protected]
MS47
Efficient Representations for the Fundamental So-lutions of Stokes Flow
In this talk, we present a simple and efficient method forthe evaluation of the fundamental solutions for Stokes flowin a half space. We show that both the direct and imagecontributions can be decomposed using harmonic functionswith a physically meaningful interpretation of the corre-sponding components. This decomposition is easy to in-corporate with existing FMM libraries. This is joint workwith Leslie Greengard.
Zydrunas GimbutasCourant InstituteNew York [email protected]
MS47
A Direct Solver for Variable Coefficient EllipticPDEs
The talk describes a highly accurate technique for solvingelliptic PDEs with variable coefficients and smooth solu-tions. The domain is tessellated into squares, and the dif-ferential operator is discretized via high order (p=10 or20) spectral differentiation on each square. A hierarchicaldirect solver is used to solve the resulting discrete system.The method is very efficient; e.g., a Helmholtz problem ona domain of size 200x200 wavelengths is solved to ten digitsof accuracy in ten minutes on a standard laptop (using 6Mdegrees of freedom).
Gunnar MartinssonUniv. of Colorado at [email protected]
MS48
A Semi-implicit Gradient-augmented Level SetMethod
Here a semi-implicit formulation of the gradient-augmentedlevel set method is presented. Stability is enhanced bythe addition of a high-order smoothing term added to thegradient-augmented level set equations. The new approachis a hybrid Lagrangian-Eulerian method which allows forthe investigation of flows based on the curvature of theinterface and the intrinsic Laplacian of the curvature. Inthis talk the method will be outlined and sample resultswill be presented. The influence of the smoothing termon the stability and accuracy of the method will also beshown.
Ebrahim M. Kolahdouz
SUNY [email protected]
David SalacUniversity at Buffalo - [email protected]
MS48
Efficient Synchronous Update of Multiple Level SetFunctions using Jet Schemes
In this talk we will present a technique to evolve a referencemap. This map can then be used to reference more thanone level set function, and thus provide for an efficient wayto advect multiple level set functions. Time permitting, wewill present various applications for this approach.
Jean-Christophe Nave, Olivier MercierMcGill [email protected], [email protected]
Rodolfo R. RosalesMassachusetts Inst of TechDepartment of [email protected]
Benjamin SeiboldTemple [email protected]
MS48
An Augmented Fast Marching Method for LevelSet Reinitialization
Here a new gradient-augmented fast marching method forreinitialization of level set functions is presented. Thismethod will calculate the signed distance function and upto the second-order derivatives of the signed distance func-tion for arbitrary interfaces. Sample results in both two-and three-dimensions will be show that the resulting levelset and curvature field are smooth even for coarse grids.
David SalacUniversity at Buffalo - [email protected]
MS48
Jet Schemes for Hamilton-Jacobi Equations usingan Evolve-and-project Framework
Jet schemes are based on tracking characteristics and us-ing suitable Hermite interpolations to achieve high order.For Hamilton-Jacobi equations, the characteristic equa-tions are in general nonlinear, and explicit schemes for solv-ing characteristic equations can yield incorrect results. Wetherefore propose an implicit update rule that is based onsolving a constrained polynomial optimization problem ineach grid cell, and then reconstructing the solution fromHermite interpolations and evolving it in time.
Dong ZhouTemple [email protected]
MS49
Fluctuating Hydrodynamics for Dynamic Simula-tions of Coarse-Grained Implicit-Solvent Models of
CS13 Abstracts 185
Lipid Bilayer Membranes
Many coarse-grained models have been developed for ef-ficient equilibrium studies of lipid bilayer membranes bytreating implicitly interactions mediated by the solvent.However, for problems involving dynamics, such CG stud-ies require accounting for the momentum transfer throughthe missing solvent degrees of freedom. We introduce anew thermostat for such CG models and computationalmethods based on fluctuating hydrodynamics. We showour approach yields results significantly different than con-ventional Langevin dynamics. We then present a numberof simulation results both for self-assembled planar-bilayersheets and vesicles.
Paul J. AtzbergerUniversity of California-Santa [email protected]
MS49
Modeling of Thermal Fluctuations in Multicompo-nent Reacting Systems
We present a multicomponent version of the fluctuatingNavier-Stokes equations that includes detailed transportand chemical reactions. The resulting system includesstochastic flux terms for momentum, energy and speciestransport and a Langevin-type model for fluctuations inthe chemical reactions. We discuss isses in the numericalsolution of the resulting systems and illustrate the impactof fluctuations on numerical solutions Turing patterns.
John B. BellCCSELawrence Berkeley [email protected]
Alejandro GarciaSan Jose State [email protected]
Aleksandar DonevCourant Institute of Mathematical SciencesNew York [email protected]
Kaushik BalakrishnanCCSELawrence Berkeley [email protected]
MS49
Low Mach Number Fluctuating Hydrodynamics ofDiffusively Mixing Fluids
We formulate low Mach number fluctuating hydrodynamicequations appropriate for modeling diffusive mixing in mix-tures of fluids of unequal density. These equations elimi-nate the fast isoentropic fluctuations in pressure associatedwith the propagation of sound waves by replacing the equa-tion of state with a local thermodynamic constraint. Wedemonstrate that the low Mach number model preservesthe spatio-temporal spectrum of the slower diffusive fluc-tuations in the linearized setting. We develop a strictlyconservative finite-volume spatial discretization of the lowMach number fluctuating equations in both two and threedimensions. We construct several explicit Runge-Kuttatemporal integrators that strictly maintain the equation ofstate constraint. The resulting spatio-temporal discretiza-
tion is second-order accurate deterministically and main-tains fluctuation-dissipation balance in the stochastic set-ting. We apply our algorithms to model the development ofgiant concentration fluctuations in the presence of concen-tration gradients, and investigate the validity of commonsimplifications neglecting the spatial non-homogeneity ofdensity and transport properties. As a validation of the lowMach number fluctuating equations and our algorithm, weperform simulations of diffusive mixing of two fluids of dif-ferent densities in two dimensions and compare the resultsof low Mach number continuum simulations to hard-diskmolecular dynamics simulations. Excellent agreement isobserved between the particle and continuum simulationsof giant fluctuations during time-dependent diffusive mix-ing.
Aleksandar DonevCourant Institute of Mathematical SciencesNew York [email protected]
Andy NonakaLawrence Berkeley National [email protected]
John B. BellCCSELawrence Berkeley [email protected]
Alejandro GarciaSan Jose State [email protected]
Thomas FaiCourant [email protected]
MS49
Implicit and Explicit Solvent Models for the Simu-lation of a Single Polymer Chain in Solution: Lat-tice Boltzmann vs Brownian Dynamics
We present a comparative study of two computer simula-tion methods to obtain static and dynamic properties ofdilute polymer solutions. The first approach is a recentlyestablished hybrid algorithm based upon dissipative cou-pling between Molecular Dynamics and lattice Boltzmann(LB), while the second is standard Brownian Dynamics(BD) with fluctuating hydrodynamic interactions. Apply-ing these methods to the same physical system (a singlepolymer chain in a good solvent in thermal equilibrium) al-lows us to draw a detailed and quantitative comparison interms of both accuracy and efficiency. It is found that thestatic conformations of the LB model are distorted whenthe box length L is too small compared to the chain size.Furthermore, some dynamic properties of the LB model aresubject to an L−1 finite size effect, while the BD model di-rectly reproduces the asymptotic L → ∞ behavior. Apartfrom these finite size effects, it is also found that in orderto obtain the correct dynamic properties for the LB simu-lations, it is crucial to properly thermalize all the kineticmodes. Only in this case, the results are in excellent agree-ment with each other, as expected. Moreover, BrownianDynamics is found to be much more efficient than latticeBoltzmann as long as the degree of polymerization is notexcessively large.
Burkhard Duenweg
186 CS13 Abstracts
Max Planck Institute [email protected]
MS50
Computational Methods for Parametric Sensitivi-ties in the Chemical Kinetic Context
I will discuss methods for the computation of paramet-ric sensitivities for stochastically modeled biochemical sys-tems. In particular, I will discuss a new finite differencemethod that arises from a non-trivial coupling of a nominaland perturbed process. The method is easy to implementand produces an estimator with a much lower variance thanthe previous state of the art for a wide array of systems.
David F. AndersonDepartment of MathematicsUniversity of Wisconsin [email protected]
MS50
Infinite Swapping Schemes for Accelerated MonteCarlo Approximation
Abstract not available at time of publication.
Paul DupuisDivision of Applied MathematicsBrown [email protected]
MS50
Parameterisation and Multilevel Approximationsof Coarse-grained Dynamics in KMC Simulations
We present an information-theoretic approach to parame-terisation of coarse-grained dynamics defined as continuoustime Markov chains. Rates of the coarse-grained processare parameterised and optimal parameters are selected byminimisation of the relative entropy on the path space.This approach extends techniques also known as inverseMonte Carlo to models with non-equilibrium stationarystates, for example, systems driven by external parame-ters or reaction-diffusion systems in catalysis. This a jointwork with P. Plechac.
Markos A. KatsoulakisUMass, AmherstDept of [email protected]
MS50
Off-lattice KMC Simulation of HeteroepitaxialGrowth
In this talk I will review previous work on weakly off-lattice KMC simulation of quantum dots, including recentprogress on a two-scale domain decomposition approach.This method is used to study various phenomena thattake place during heteroepitaxial growth. For example,it is demonstrated that faceted quantum dots occur viathe layer-by-layer nucleation of pre-pyramids on top of acritical layer with faceting occurring by anisotropic surfacediffusion. It is also shown that the dot growth is enhancedby the depletion of the critical layer which leaves behind awetting layer. Capping simulations provide insight into themechanisms behind dot erosion and ring formation. I willthen discuss efforts to extend these methods to fully off-
lattice simulations. This is joint work with Peter Smerekaand Henry Boateng.
Tim SchulzeUniversity of TennesseeDepartment of [email protected]
MS51
An Efficient and Scalable Lanczos-based Eigen-solver for Multi-core Systems
We describe an efficient scalable symmetric iterative eigen-solver for finding few lowest eigenpairs on distributedmulti-core platforms. We show over 80% computationalefficiency by major reductions in communication overheadfor the SpMV and basis orthogonalization tasks. In par-ticular, we present strategies for hiding communication onmulti-core platforms. We demonstrate the effectiveness ofthese techniques by reporting the performance improve-ments in large-scale eigenvalue problems arising in nuclearstructure calculations.
Hasan Metin Aktulga, Chao YangLawrence Berkeley National [email protected], [email protected]
Esmond G. NgLawrence Berkeley National [email protected]
Pieter Maris, James VaryIowa State [email protected], [email protected]
MS51
Density Functional Electronic Band Structure Cal-culations with a Complex Moment Based Eigen-solver
First-principles electronic band structure calculationsbased on the density functional theory is one of the bestchoices for understanding and predicting phenomena inmaterial sciences. In electronic band structure calculationsof large systems, one needs to solve large sparse interioreigenproblems. We present an efficient approach for solv-ing these eigenproblems using a complex moment basedeigensolver. Numerical experiments on the K computerare also presented.
Yasunori FutamuraUniversity of [email protected]
Tetsuya SakuraiDepartment of Computer ScienceUniversity of [email protected]
Shinnosuke Furuya, Jun-Ichi IwataThe University of [email protected], [email protected]
MS51
Computing a Large Number of Eigenpairs onMulit-/many-core Systems
We examine two techniques for compute a relatively large
CS13 Abstracts 187
number of eigenpairs of a sparse symmetric matrix. One isbased on multiple shift-invert Lanczos. The other is basedon a projection method that makes use of a contour inte-gral representation of a projection operator. We comparethe pros and cons of both approaches and discuss a num-ber of practical issues of implementing these methods onmulti/many core systems
Chao Yang, Hasan Metin AktulgaLawrence Berkeley National [email protected], [email protected]
Christopher HaineUniversity of [email protected]
Lin LinLawrence Berkeley [email protected]
MS51
Computing Eigenspace by a Penalty Approach
Abstract not available at time of publication.
Yin ZhangRice UniversityDept of Comp and Applied [email protected]
Xin LiuAcademy of Mathematics and Systems ScienceChinese Academy of [email protected]
Zaiwen WenDepartment of MathematicsShanghai Jiaotong [email protected]
Chao YangLawrence Berkeley National [email protected]
MS52
Limited Data-Driven Uncertainty Quantification
In this talk, we will present work on limited data-drivenstochastic collocation approach to include the effect of un-certain design parameters during complex multi-physicssimulation of microsystems. The proposed framework com-prises of two key steps: firstly, probabilistic characteri-zation of the input uncertain parameters based on avail-able experimental information, and secondly, propagationof these uncertainties through the predictive model to rele-vant quantities of interest. The uncertain input parametersare modeled as independent random variables, for whichthe distributions are estimated based on available experi-mental observations, using a nonparametric diffusion mix-ing based estimator. The diffusion based estimator derivesfrom the analogy between the kernel density estimation(KDE) procedure and the heat dissipation equation, andconstructs density estimates that are smooth and asymp-totically consistent. The diffusion model allows for the in-corporation of the prior density and leads to an improveddensity estimate, in comparison to the standard KDE ap-proach, as demonstrated through several numerical exam-ples. Following the characterization step, the uncertainties
are propagated to the output dependent variables using thestochastic collocation approach, based on sparse grid inter-polation. The stochastic multiphysics framework is used tostudy uncertainties in microsystems.
Narayana R. AluruDepartment of Mechanical Science and EngineeringBeckman Institute, University of [email protected]
MS52
A Decomposition Approach to Uncertainty Analy-sis of Multidisciplinary Systems
Uncertainty analysis for complex systems can become cum-bersome and computationally intractable. This talk de-scribes an approach for decomposing and distributing theuncertainty analysis task amongst the various componentscomprising a complex system. The approach draws on mul-tidisciplinary analysis and optimization methods, densityestimation, and sequential Monte Carlo methods. The dis-tributed multidisciplinary uncertainty analysis approach isprovably convergent and is compared to a traditional all-at-once Monte Carlo uncertainty analysis approach for severalexamples.
Sergio AmaralDepartment of Aeronautics & AstronauticsMassachusetts Institute of [email protected]
Doug Allaire, Karen E. WillcoxMassachusetts Institute of [email protected], [email protected]
MS52
A Finite Element Method for Density Estimationwith Gaussian Priors
A variational problem characterizing the density estimatordefined by the maximum a posteriori method with Gaus-sian process priors is derived. It is shown that this problemis well posed and can be solved with Newton’s method. Nu-merically, the solution is approximated by a Galerkin/finiteelement method with piecewise multilinear functions onuniform grids. Error bounds for this method are givenand numerical experiments are performed for one-, two-,and three-dimensional examples.
Markus HeglandAustralian National [email protected]
MS52
Density Estimation for Large Datasets with SparseGrids
Kernel density estimation can become computationally ex-pensive for large data sets. Furthermore, its performancehighly depends on the choice of the kernel bandwidth. Oursparse-grid-based method overcomes these drawbacks tosome extent. We give details on how to estimate densityfunctions on sparse grids and show numerical experimentswith large data sets to demonstrate that our method iscompetitive with respect to accuracy and superior to con-ventional approaches with respect to computational com-plexity.
Benjamin Peherstorfer
188 CS13 Abstracts
SCCS, Department of InformaticsTechnische Universitat [email protected]
MS53
Analysis of Ship Structural Hydroelasticity by us-ing a Fully-coupled Higher-order BEM and FEM
This study considers the problem of ship hydroelasticity,which is an important technical issue in the design of ultra-large marine vessels. To analyze the ship structural hy-droelasticity, two initial boundary value problems shouldbe solved simultaneously: hydrodynamic problem for shipmotion and dynamic structural problem. In the presentstudy, a partitioned method is applied for the coupledfluid-structure interaction problem. The fluid domain sur-rounding a flexible body is solved using a Rankine panelmethod based on higher-order B-spline basis function, andthe structural domain is handled with a three-dimensionalfinite element method. The two distinct methods are fullycoupled in the time domain by using an implicit iterativescheme. The detailed numerical method with stable time-marching is described, and the validation of the developedmethod is introduced. The numerical results include a realship application.
Yonghwan KimSeoul National UniversityNaval Architecture and Ocean [email protected]
MS53
Flexible Ring Flapping in a Uniform Flow
An improved version of the immersed boundary methodfor simulating an initially circular or elliptic flexible ringpinned at one point in a uniform flow has been developed.A penalty method derived from fluid compressibility wasused to ensure the conservation of the internal volumeof the flexible ring. A new bistability phenomenon wasobserved: for certain aspect ratios, two periodically flap-ping states coexist with different amplitudes in a particularReynolds number range.
Boyoung Kim, Hyung Jin SungDepartment of Mechanical [email protected], [email protected]
MS53
Numerical Modeling of the Interaction betweenMoving Solid Structures and Two-phase FluidFlows: Application in Ocean Wave Energy Con-verters
We are developing a computational tool for interactionanalysis of two-phase fluid flows with a moving solid object.Two-step projection method with GPU acceleration solvesthe flow equations. The volume-of-fluid method tracksthe fluid interfaces while the fast-fictitious-domain methodsimulates the interactions between a moving solid objectand two-phase fluid flows. A geometrical reconstructionis employed to handle liquid-gas-solid interfaces. We willpresent results of canonical tests and preliminary resultson ocean wave energy converters.
Amirmahdi Ghasemi, Ashish PathakUniversity of Massachusetts [email protected], [email protected]
Mehdi RaessiDepartment of Mechanical EngineeringUniversity of Massachusetts [email protected]
MS53
Generalized Fictitious Methods for Fluid-structureInteractions
We present two methods for fluid-structure interaction(FSI) problems: the fictitious pressure method and thefictitious mass method. For simplified problems we obtainexpressions for the convergence rate index, which demon-strates a similarity of fictitious methods to the FSI ap-proach with Robin boundary conditions. In numericaltests, we verify the selection of optimal values for the fic-titious parameters, and develop an empirical analysis forcomplex geometries and apply it to 3D patient-specific flex-ible brain arteries.
Yue YuApplied [email protected]
Hyoungsu BaekMassachusetts Institute of TechnologyMathematics [email protected]
George E. KarniadakisBrown UniversityDivision of Applied Mathematicsgeorge [email protected]
MS54
Parallel Computing for Long-Time Simulations ofCalcium Waves in a Heart Cell
The flow of calcium ions in a heart cell is modeled bya system of time-dependent reaction-diffusion equations.The large number of calcium release sites requiring high-resolution meshes and the need for large final times requireparallel computing for effective simulations. Using Krylovsubspace methods offers the opportunity for efficient par-allel computing, with choices in parallel algorithms drivenby the special considerations of the available CPU, GPU,or hybrid CPU-GPU architecture.
Matthias K. GobbertUniversity of Maryland, Baltimore CountyDepartment of Mathematics and [email protected]
Xuan HuangDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
Stefan KopeczInstitute for MathematicsUniversity of [email protected]
Philipp BirkenDepartment 10, Mathematics and Natural SciencesUniversity of Kassel, [email protected]
CS13 Abstracts 189
Andreas MeisterUniversity of KasselInstitute of [email protected]
MS54
High End of Mesh-based PDE Simulations
Abstract not available at time of publication.
David E. [email protected]
MS54
A Memory Efficient Finite Volume Method forAdvection-Diffusion-Reaction Systems
Advection-Diffusion-Reaction Systems occur in a wide va-riety of applications. The use of matrix-free parallel meth-ods allows for simulations on high resolution meshes, sinceno memory is needed to store system matrices and the par-alellism distributes the workload among several processes.We present a method of this type based on a finite volumediscretization and demonstrate its performance simulatingcalcium flow in human heart cells.
Jonas SchaferInstitute of MathematicsUniversity of [email protected]
Xuan HuangDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
Stefan KopeczInstitute for MathematicsUniversity of [email protected]
Philipp BirkenDepartment 10, Mathematics and Natural SciencesUniversity of Kassel, [email protected]
Matthias K. GobbertUniversity of Maryland, Baltimore CountyDepartment of Mathematics and [email protected]
Andreas MeisterUniversity of KasselInstitute of [email protected]
MS54
Asynchronous Multilevel Algorithms
Iterative algorithms for solving elliptic PDE are tradition-ally formulated in an imperative style that prescribes afixed order of operations. In a parallel environment, thisimplies a rigid sequence of communication and synchro-nization steps. In the talk we will present alternative asyn-chronous strategies based on a greedy and randomized ver-sion of the Gauss-Southwell algorithm and the theory of
stable multilevel splittings.
Ulrich J. RuedeUniversity of Erlangen-NurembergDepartment of Computer Science (Simulation)[email protected]
MS55
Reduced-space inexact-Newton-Krylov for High-dimensional Optimization
Abstract not available at time of publication.
Jason E. Hickenpostdoctoral fellow, Stanford [email protected]
MS55
A Matrix-Free Augmented Lagrangian Approachto Structural Optimization
In structural optimization, it is common practice to ag-gregate the failure constraints to reduce the cost of com-puting the constraint Jacobian. We present an alternativeapproach in which the constraint Jacobian is never formed;only matrix-vector products are needed. We illustrate ourapproach using an augmented Lagrangian optimization al-gorithm and several example problems. We also show howadopting the matrix-free approach can lead to lower-massstructural designs compared to the aggregation approach.
Andrew LambeUniversity of [email protected]
Joaquim MartinsUniversity of [email protected]
Sylvain ArreckxEcole Polytechnique de [email protected]
Dominique OrbanGERAD and Dept. Mathematics and IndustrialEngineeringEcole Polytechnique de [email protected]
MS55
Adjoint-Based Equivalent Area Methods for Super-sonic Low-Boom Design
Abstract not available at time of publication.
Francisco PalaciosStanford [email protected]
MS55
Practical Experience witha Multi-Objective Model-Management FrameworkOptimization Method
Solving multi-objective optimization problems that involvecomputationally expensive functions is a normal part of
190 CS13 Abstracts
many engineering applications. Runtime issues are mag-nified in a multidisciplinary design optimization setting.Normal-Boundary Intersection (NBI) is a multi-objectiveoptimization method. Running NBI directly requires pro-hibitive computational cost. Running on surrogate approx-imations to the simulation fails to produce sufficiently ac-curate solutions. We combine the use of surrogates andsimulations in a way similar to the general surrogate man-agement framework. We conclude with a representativeaircraft design.
Joseph P. SimonisThe Boeing [email protected]
Evin CramerBoeingMath & [email protected]
Joerg M. GablonskyMathematics & Engineering AnalysisPhantom Works - The Boeing [email protected]
Laura Lurati, Paul SellersThe Boeing [email protected], [email protected]
MS56
Four Modules for Teaching CUDA in a Computa-tional Science Context
Four modules developed for the UPEP project deal withusing NVIDIA’s CUDA programming environment to per-form scientific computations on graphics cards. The firstis a straightforward introduction to CUDA in the contextof matrix multiplication. Two modules introduce dynamicprogramming and show how such problems can be solvedwithin CUDA, and the last shows how to integrate CUDAwith OpenGL in order to do high performance scientific vi-sualization. We will give an overview and demonstration.
Robert HochbergUniversity of DallasDepartment of [email protected]
MS56
Supporting Petascale Education: The Blue WatersUndergraduate Petascale Education Program
The Blue Waters project, in collaboration with the Na-tional Computational Science Institute and national HPCprograms, has launched a coordinated effort to preparecurrent and future generations of students for the grow-ing complexity of computing paradigms. To support thiseffort, the Blue Waters Undergraduate Petascale Educa-tion Program (BW-UPEP) seeks to promote understand-ing and interest in petascale computing and its applicationsamong undergraduate students and faculty by supportingthe development of high quality undergraduate curricularmaterials.
Jennifer K. HouchinsShodor Education Foundation [email protected]
MS56
Classroom Explorations of N-body GravitationalSimulations using GalaxSeeHPC
GalaxSeeHPC is the most recent release of the GalaxSee N-body solver designed for classroom study of gravitationalsystems, adding periodic boundary conditions, increasedcontrol over input, and additional force calculation meth-ods to the previous GalaxSee-MPI code. GalaxSeeHPCand its corresponding two Blue Waters Petascale Educa-tion Project modules allow students to study both the tech-niques required to scale N-body solutions to millions ofparticles as well as the new science enabled by larger N.
David A. JoinerKean UniversityAssistant [email protected]
MS56
Biofilms: Linked for Life (Understanding Biofilmsthrough Modeling and Simulation)
Most microbial organisms do not exist as individuals, butwithin communities of interconnected members. In thispresentation, we discuss development of a simulation ofsuch biofilm structural growth appropriate for modeling,simulation, or high performance computing courses. Con-sideration of cellular automaton simulations, boundaryconditions, and diffusion can empower students to developsimilar simulations for other applications. Moreover, ex-tensions of the basic model can illustrate and motivate theneed for high performance computing in computational sci-ence.
George W. ShifletWofford [email protected]
Angela B. ShifletMcCalla Professor of Math. & CS, Dir. of ComputationalSci.Wofford [email protected]
MS57
Bernstein-Bezier Techniques in High Order FiniteElement Analysis
Algorithms are presented that enable the element matri-ces for the standard finite element space, consisting ofcontinuous piecewise polynomials of degree n on simpli-cial elements in Rd, to be computed in optimal complex-ity O(n2d). The algorithms (i) take account of numericalquadrature; (ii) are applicable to non-linear problems; and,(iii) do not rely on pre-computed arrays containing valuesof one-dimensional basis functions at quadrature points (al-though these can be used if desired). The elements arebased on Bernstein-Bezier polynomials and are the first toachieve optimal complexity for the standard finite elementspaces on simplicial elements.
Mark AinsworthUniversity of StrathclydeDepartment of [email protected]
CS13 Abstracts 191
MS57
High-order Methods for Fractional DifferentialEquations
Modeling of non-classic phenomena in science, finance, bi-ology and engineering increasing utilizes fractional differ-ential equations, highlighting the need for robust, accu-rate and efficient computational models for such operators.However, despite fractional calculus being almost as oldas classic calculus, the development of computational tech-niques is less advanced. In this talk we introduce discontin-uous Galerkin and spectral penalty methods for fractionalPDEs, consider accuracy and stability, and illustrate per-formance on test problems.
Jan S. HesthavenBrown UniversityDivision of Applied [email protected]
Weihua DengDepartment of MathematicsLanzhou University, [email protected]
Qinwu XuDivision of Applied MathematicsBrown University, Providence, RIqinwu [email protected]
MS57
High-order Virtual Element Methods
The Virtual Element spaces are just like the usual FiniteElement spaces with the addition of suitable non polyno-mial functions. This is not a new idea; the novelty hereis to choose the degrees of freedom in such a way that theelementary stiffness matrix can be computed without ac-tually computing these non polynomial functions. In doingthat we can easily deal with complicated element geome-tries and higher order methods.
Alessandro RussoMilan [email protected]
MS58
Mint: A User-fiendly C-to-CUDA Code Translator
Aiming at automated source-to-source code translationfrom C to CUDA, we have developed the Mint frame-work. Users only need to annotate serial C code with a fewcompiler directives, specifying host-device data transfersplus the parallelization depth and granularity of loop nests.Mint then generates CUDA code as output, while carryingout on-chip memory optimizations that will greatly benefit3D stencil computations. Several real-world applicationshave been ported to GPU using Mint.
Xing CaiSimula Research Laboratory1325 Lysaker, [email protected]
Didem UnatLawrence Berkeley National [email protected]
Scott Baden
University of California, San [email protected]
MS58
PyOP2 - An Abstraction for Performance-portableSimulation Software
OP2 is an abstraction which expresses mesh-based simula-tion code in terms of numerical kernels applied in parallelover a mesh. This enables performance portability andfrees the developer from the increasingly complex detailsof parallel programming. Here we present PyOP2, a just-in-time compiled OP2 which delays kernel compilation andparallel strategy formation until runtime. PyOP2 presentsa high-level programmable interface. Alternatively PyOP2code for finite elements may be generated automatically bythe FEniCS system.
Carlo BertolliImperial College [email protected]
Mike GilesUniversity of [email protected]
David HamImperial [email protected]
Paul Kelly, Nicolas Loriant, Graham MarkallImperial College [email protected], [email protected],[email protected]
Lawrence MitchellUniversity of [email protected]
Giham MudaligeUniversity of [email protected]
Florian RathgeberImperial College [email protected]
MS58
Achieving High Performance and Portability inStencil Computations
Physis is an application framework for stencil computationsthat is designed to achieve both performance and produc-tivity in large-scale parallel computing systems, in partic-ular heterogeneous GPU-accelerated supercomputers. Theframework consists of an implicitly parallel domain-specificlanguage for stencil computations and its translators forvarious target parallel architectures. This talk presentsthe current status of the framework and its performancestudies using several application benchmarks.
Naoya MaruyamaRIKEN Advanced Institute for Computational [email protected]
Satoshi MatsuokaTokyo Insitute of [email protected]
192 CS13 Abstracts
MS58
Automating the Communication-computationOverlap with Bamboo
MPI has been a popular standard for implementing HPCapplications. However, codes running at scale would re-quire significant optimizations, one of which is the coderestructuring to mask communication. Such optimizationsrequire aggressive effort and complicate the software main-tenance. We present Bamboo, a translator that automatesthe optimization by transforming MPI into a data-drivenform. Experiments on up to 98304 processors demon-strate that Bamboo significantly speeds up its input, meet-ing/exceeding the performance of hand-optimized variants.
Tan Nguyen, Scott BadenUniversity of California, San [email protected], [email protected]
MS59
Assessing the Predictive Capabilities of Miniapps
Under what conditions does an application proxy repre-sent a key performance characteristic in a full application?In this talk we define a methodology for comparing appli-cations and miniapps, providing a framework for reason-ing about important performance-impacting issues. Thismethodology will be illustrated using miniapps from theMantevo project, configured to represent the runtime char-acteristics of some key mission application codes.
Richard Barrett, Paul Crozier, Douglas DoerflerSandia National [email protected], [email protected],[email protected]
Simon D. HammondScalable Computer ArchitecturesSandia National [email protected]
Michael A. Heroux, Paul Lin, Heidi K. Thornquist,Timothy G. Trucano, Courtenay T. VaughanSandia National [email protected], [email protected],[email protected], [email protected],[email protected]
MS59
Programming Model Exploration and EfficiencyModeling using Mini and Proxy Applications
Application proxies and kernels of applications that aresmall and easily run can serve as a useful test bed fora variety of new languages. As languages are developedto combat the increasingly complex architectural changes,they become a crucial part of the development effort. Inthis talk we discuss some recent results of various program-ming models applied to kernel applications.
Alice KonigesLawrence Berekely [email protected]
MS59
Programming Models using Workflows from Prox-
ies
Abstract not available at time of publication.
James SextonIBM [email protected]
MS59
Examples of Codesign Using Application Proxies
Architectures are undergoing a sea change with a tran-sition from cheap-memory-expensive-flops to cheap-flops-expensive memory/power. This requires us to rethink ourapplications to get the most out of a differently balancedmachine than we have been used to for the past 15 years.Yet, rewriting entire applications is not practical. Proxyapplications have emerged as one possible solution to thisthorny problem. In this talk we explore the meaning androle of proxy applications in co-design and their effective-ness as a vehicle for reducing many possible paths forwardto a manageable few.
Sriram SwaminarayanLos Alamos National [email protected]
MS60
Accuracy of Some Finite-difference and Finite-element Methods for Wave Propagation at a Fluid-solid Interface
The problem of modeling wave propagation in media withsolid and fluid regions has many applications in geophysics,engineering and medicine. We have performed a grid-dispersion analysis and numerical computations to com-pare the performance of the following numerical meth-ods in handling fluid-solid interfaces: the spectral-elementmethod, and several versions of the the finite-differencemethod. We conclude that a first-order velocity stress for-mulation can be used in dealing with fluid-solid layers with-out using staggered grids necessarily.
Jonas D. De BasabeThe University of Texas at AustinInstitute for Computational Engineering and [email protected]
Mrinal SenInstitute of GeophysicsUniversity of [email protected]
MS60
The Discontinuous Enrichment Method for WavePropagation
Wave propagation problems in the medium frequencyregime are computationally challenging. One avenue ofresearch pursues higher-order discretization methods thatcan deliver both accuracy and computational efficiency atsmaller mesh resolutions. The Discontinuous EnrichmentMethod (DEM) is one example which distinguishes itselffrom competing approaches in the additional informationit incorporates in the approximation method. It has showna significant promise for acoustic and structural acousticproblems and therefore is reviewed here, together with new
CS13 Abstracts 193
applications to multi-scale problems.
Charbel Farhat, Radek TezaurStanford [email protected], [email protected]
MS60
Dispersion Reducing Techniques for FDTDSchemes
Abstract not available at time of publication.
Bezalel FinkelsteinSami Shamoon College of Engineering, [email protected]
MS60
Analysis of High Order FDTD Methods forMaxwell’s Equations in Dispersive Media
We consider models for electromagnetic wave propagationin linear dispersive media which include ordinary differ-ential equations for the electric polarization coupled toMaxwell’s equations. We discretize these models using highorder finite difference methods and study the properties ofthe corresponding discrete models. In this talk we willpresent the stability, dispersion and convergence analysisfor a class of finite difference methods that are second orderaccurate in time and have arbitrary (even) order accuracyin space in one and two dimensions.
Nathan L. Gibson, Vrushali A. BokilOregon State [email protected],[email protected]
MS60
Dispersion Reduction for Acoustic Wave Equationusing m-adaptation
We present a novel discretization strategy, dubbed m-adaptation, for acoustic wave equation in time domain in2D and 3D. This strategy is based on the optimizationwithin a family of second-order accurate Mimetic FiniteDifference (MFD) discretizations. The optimized schemehas a computational complexity of second-order schemeand is shown to be fourth-order accurate in dispersionon rectangular and cubic meshes. On square meshes theanisotropy is shown to be sixth-order accurate.
Vitaliy GyryaLos Alamos National LaboratoryApplied Mathematics and Plasma Physicsvitaliy [email protected]
Konstantin LipnikovLos Alamos National [email protected]
MS61
Upwind Methods for Second-order Wave Equations
In this talk we discuss a newly developed class of high-order accurate upwind schemes for the wave equation insecond-order form. By working directly on the equa-tions in second-order form, we avoid issues of compatibil-ity that can arise when converting from second- to first-order formulations. The schemes are based on embedding
the solution to a local Riemann-type problem that usesd’Alembert’s exact solution. High-order accuracy is ob-tained using a single-step space-time procedure. The re-sult is a method that is highly efficient in both memoryand speed, and has very attractive accuracy and robust-ness characteristics. Stability and accuracy are discussedusing normal-mode theory, and the efficacy of the approachis demonstrated for a set of challenging test problems.
Jeffrey W. BanksLawrence Livermore National [email protected]
MS61
A Performance Study of a Massively Parallel WaterWave Model for Engineering Applications
We present results of a unique study of several computa-tional techniques suitable for improving performance of afully nonlinear and dispersive water wave model intendedfor use in coastal engineering applications and performancecritical real-time ship simulator software. The model is at-tractive for description of a broad range of wave phenom-ena, flow kinematics and wave-structure interactions. Aflexible-order finite difference method is used for accuratediscretization and efficient and scalable mapping to mod-ern many-core hardware.
Allan P. Engsig-KarupTechnical University of DenmarkDepartment of Informatics and Mathematical [email protected]
Stefan L. GlimbergDTU InformaticsTechnical University of [email protected]
MS61
DG Schemes for Quadrature-based Moment-closure Models of Plasma
The dynamics of collisionless plasma can be simulated us-ing kinetic or fluid models. Kinetic models are valid overmost of the spatial and temporal scales that are of physicalrelevance in many application problems; however, they arecomputationally expensive due to the high-dimensionalityof phase space. Fluid models have a more limited range ofvalidity, but are generally computationally more tractablethan kinetic models. One critical aspect of fluid models isthe question of what assumptions to make in order to closethe fluid model. In this work we study so-called quadrature-based moment closure models for collisionless plasma. Wedevelop high-order discontinuous Galerkin (DG) schemesfor these models, and in particular, we consider severalimportant issues in the discretization of these models, in-cluding hyperbolicity, positivity, and moment-realizability.The resulting schemes are tested on several standard colli-sionless plasma test cases.
James A. RossmanithIowa State UniversityDepartment of [email protected]
Yongtao ChengUniversity of Wisconsin - MadisonDepartment of [email protected]
194 CS13 Abstracts
MS61
ManyClaw: Slicing and Dicing Riemann Solvers forNext Generation Highly Parallel Architectures
Next generation computer architectures, e.g. Intel MICand GPUs, include an order of magnitude increase in in-tranode parallelism over traditional CPUs. We presentManyClaw, a project that explores the exploitation of thisintranode parallelism in hyperbolic PDE solvers via theClawpack software package. Our goal is to separate thelow level parallelism and the physical equations thus pro-viding users the capability to leverage intranode parallelismwithout explicitly writing code to take advantage of newerarchitectures.
Andy R. TerrelTexas Advanced Computing CenterUniversity of [email protected]
Kyle T. MandliUniversity of Texas at [email protected]
MS62
GPU-Accelerated Implementations of B-SplineSignal Processing Operations for FFD-Based Im-age Registration
B-spline signal processing operations are widely used in theanalysis of two and three-dimensional images. In this talk,we investigate GPU-accelerated implementations of basicB-spline signal processing operations in CUDA, includingdirect transformations, indirect transformations, compu-tation of partial derivatives, and interpolation. We thenillustrate how these operations can be used to design afree-form deformation based image registration algorithmthat enables dense control point spacing.
Nathan D. CahillSchool of Mathematical SciencesRochester Institute of [email protected]
Alex Karantza, Sonia Lopez AlarconDepartment of Computer EngineeringRochester Institute of [email protected], [email protected]
MS62
Model Mis-specification - in Search of the MissingLink
Frequently, since our ability to simulate data is limited,only a discrete, incomplete observation operator can bespecified. It will benefit a lot if we could supplement suchreduced model somehow. In this study, we developed anuclear norm stochastic optimization technique to supple-ment the model with rank restriction given a training setof models and data.
Ning HaoTufts UniversityDepartment of [email protected]
Lior HoreshIBM Research
lhoreshus.ibm.com
Misha E. KilmerTufts [email protected]
MS62
Title Not Available at Time of Publication
Abstract not available at time of publication.
Harald KoestlerUniversity of [email protected]
MS62
Ultra-low-dose Method for Lung Cancer Screening
In the United States, lung cancer is the leading cause ofcancer death. The screening CT and the follow-up CTs ex-pose the patient to ionizing radiation, which carries with itan increased risk of malignancy. To maximize the benefit ofCT lung cancer screening, we need to minimize the radia-tion dose as low as reasonably achievable. The compressivesensing theory has shown its potential in CT reconstruc-tion for dose reduction. In contrast to the popular totalvariation (TV) based functions, a redundant dictionary ismore specific for a particular application and more effec-tive in terms of a sparse representation. In this work, wedevelop a dictionary learning based ultra-low dose methodfor lung cancer screening. Very supportive results are ob-tained from preclinical sheep lung study.
Hengyong YuDepartment of RadiologyWake Forest [email protected]
MS63
Pipelining the Fast Multipole Method over a Run-time System
Fast multipole methods (FMM) usually require a carefultuning of the algorithm for both the targeted physics andthe hardware. In this talk, we propose a new approachthat achieves high performance across architectures. Weexpress the FMM algorithm as a task flow and employ aruntime system to process the tasks on the different pro-cessing units. We carefully design the task flow, the math-ematical operators, their CPU and GPU implementationsand their schedule.
Emmanuel Agullo, Berenger Bramas, Olivier [email protected], [email protected],[email protected]
Eric F. DarveStanford UniversityMechanical Engineering [email protected]
Matthias [email protected]
Toru TakahashiNagoya University
CS13 Abstracts 195
MS63
Achieving High Performance with Multiple-GPUNon-symmetric Eigenvalue Solver
The first step in the non-symmetric eigenvalue problem isthe reduction to upper Hessenberg form. Dependenciescaused by applying an orthogonal matrix on both sidesmake this reduction difficult to parallelize. We presenta multiple-GPU implementation of the Hessenberg reduc-tion that uses the GPUs’ superior memory bandwidth toachieve high performance on memory-bound operations.Our algorithm includes optimizations to overlap work onthe CPU and GPU. We also explore accelerating the com-putation of eigenvectors.
Mark GatesComputer Science DepartmentUniversity of [email protected]
MS63
An Applications Perspective on Multi-core, Mas-sive Multi-threading, and Hybrid Systems
Multi/many-core and hybrid architectures that supportmassive multi-threading have raised considerable uncer-tainty as to what programming models might be appropri-ate. We will discuss applications that have been developedwithin the Swiss High-Performance and High-ProductivityComputing platform (www.hp2c.ch) and are exploitingemerging computer architecture quite successfully. We willsee what architectural aspects are important for the variousalgorithmic motifs that appear in applications, and whatnew programming models seem to find broad acceptanceamong scientific programmers.
Thomas SchulthessSwiss National Supercomputing [email protected]
MS63
Adapting to the Heterogeneous HPC Phenomenain the Industry
The lions share of Shells global HPC capacity is con-sumed by geophysical seismic imaging and reservoir en-gineering fluid flow simulations for oil and gas reservoirs.Legacy algorithms and software must be replaced with fun-damentally different ones that scale to 1000s of possiblyheterogeneous- cores. Geophysical Reverse Time Migra-tion is an example. In this talk, we present how we’readapting our algorithms to tackle this phenomena.
Amik St-CyrComputation and Modeling,Shell International E&P, [email protected]
Chaohui ChenComputation and ModelingShell International E&P, [email protected]
Detlef HohlShell Global Solutions (US)[email protected]
MS64
Front Tracking and Spring Fabric Model forParachute FSI
We use the front tracking method on a spring system tomodel the dynamic evolution of parachute canopy. Thecanopy surface of a parachute is represented by a trian-gulated surface mesh with preset equilibrium side length.This mechanical structure is coupled with the Navier-Stokes solver through the Impulse Method. The numericalsolutions are compared with the experimental data andthere are good agreements in the terminal descent velocityand breathing frequency of the parachute.
Xiaolin LiDepartment of Applied Math and StatSUNY at Stony [email protected]
Joungdong Kim, Yan LiSUNY at Stony BrookStony Brook, NY [email protected], [email protected]
MS64
Implicit Schemes for Fluid-Fluid and Fluid-Structure Interaction Problems in an EulerianFramework
The simulation of compressible multi-fluid flows interact-ing with flexible structures is important in many areas, es-pecially in problems involving implosions and explosions.The numerical methods utilized for these problems havehistorically used explicit time integration. In this talk wedescribe an implicit time integration scheme for these sim-ulations within an Eulerian framework (FIVER). The pre-sented scheme shows speedups of up to 40x on many prob-lems, while preserving the same level of accuracy.
Alex MainStanford [email protected]
Kevin G. WangDepartment of Aeronautics and AstronauticsStanford University, Stanford, [email protected]
Charbel FarhatStanford [email protected]
MS64
Hybridization Techniques in Coupled MultiphysicsFlow Problems
Coupled problems, involving multiple scales and/orphysics, arise in many applications; e.g., in fluid-structureinteraction or the coupling of porous media with free flow.An important question in such coupled problems is thechoice of flexible, stable, physically meaningful and well-approximating interface couplings. We show that hybridiz-able methods offer some interesting properties for address-ing these points and illustrate the proposed coupling con-cepts by examples.
Christian WalugaM2 Center of MathematicsTechnical University of Munich
196 CS13 Abstracts
Barbara WohlmuthM2, Centre for Mathematical Sciences,Technische Universitat Munchen, [email protected]
MS64
Computational Methods for Multi-Material Fluid-Structure Interaction with Dynamic Fracture
This talk addresses the interaction of multi-material com-pressible fluid flows and elastic-plastic structures subjectto large deformation and fracture. We present a high-fidelity, fluid-structure coupled computational frameworkbased on an embedded boundary method for CFD and anextended finite element method (XFEM) for CSD. We il-lustrate this framework, highlight its features, and assessits performance with the three-dimensional simulation ofpipeline explosions and underwater implosions for whichtest data is available.
Kevin G. WangDepartment of Aeronautics and AstronauticsStanford University, Stanford, [email protected]
Patrick LeaNorthwestern [email protected]
Charbel FarhatStanford [email protected]
MS65
An Adaptive Patchy Method for the Numerical So-lution of the Hamilton-Jacobi-Bellman Equation
We present an adaptive numerical method to compute so-lutions to the Hamilton-Jacobi-Bellman PDE arising froman infinite-horizon optimal control problem. The methodis based on a patchy technique and builds local polyno-mial solutions using a continuation algorithm. A level setmethod is used that adaptively changes the step-size ofthe cost levels depending on the magnitude of the error in-curred by the computed solution. Numerical experimentsare presented that illustrate the accuracy of the method.
Cesar O. Aguilar, Arthur J. KrenerNaval Postgraduate SchoolDepartment of Applied [email protected], [email protected]
MS65
An Iterative POD (I-POD) based Approach toSolving the Fokker-Planck Equation with Applica-tion to Nonlinear Filtering
The Fokker-Planck-Kolmogorov Equation (FPK) is a par-tial differential equation that governs the evolution of un-certainty through a stochastically perturbed nonlinear dy-namical system. The FPK is a linear parabolic PDE, how-ever, it suffers from the curse of dimensionality owing toits high dimensional state space. We propose an itera-tive proper orthogonal decomposition (I-POD) based ap-proach to obtaining reduced order models (ROM) of theFPK equation such that it is suitable for online implemen-
tation. The I-POD is a data based technique that can beused to extract the dominant eigenmodes of large scale lin-ear systems such as those arising from the FPK equation,which can then be used to construct a suitable ROM. TheseROMs are then utilized to solve the nonlinear filtering/data assimilation problem in a computationally tractable,real-time fashion.
Suman ChakravortyTexas A&M UniversityCollege Station, [email protected]
MS65
Computational Aspects of Optimal Control forNonlocal Problems
We consider a control problem for the nonlocal Poissonequation. We study existence and uniqueness of the op-timal control and we show the convergence of the nonlo-cal solution to the classical (local) one as the horizon ap-proaches zero. Also, we introduce a finite element (FE)discretization, we derive a priori error estimates and wediscuss advantages and disadvantages in using differentFE spaces. Theoretical results are confirmed by numeri-cal computations for one-dimensional test cases.
Marta D’EliaFlorida State [email protected]
Max GunzburgerFlorida State UniversitySchool for Computational [email protected]
MS65
Robust Control via Optimal Uncertainty Quantifi-cation
This talk will cover recent results in the optimal quan-tification of uncertainties with incomplete information onprobability measures and response functions and in pres-ence of sample data (and also, possibly, with incompletelyspecified priors). We will show how these results can beapplied to robust control. Various parts of this talk arejoint work with Clint Scovel (Caltech), Tim Sullivan (Cal-tech, Warwick), Mike McKerns (Caltech) and Michael Or-tiz (Caltech).
Houman OwhadiApplied [email protected]
MS66
Model Velocity Estimation based on SeismogramRegistration
Seismograms from seismic surveys contain subsurface ve-locity and structure information. We show that optimalmapping between a measured seismogram and its corre-sponding predicted seismogram can be obtained and effec-tively used for full waveform inversion. Highly non-conveximage registration problems due to the oscillatory waveletsare solved in a multiscale manner. Such mapping enablesus to modify the measured seismogram close to the pre-dicted one so that a low frequency velocity update is com-
CS13 Abstracts 197
puted.
Hyoungsu BaekMassachusetts Institute of TechnologyMathematics [email protected]
Laurent DemanetMathematics, [email protected]
MS66
Time-domain Seismic Imaging and Inversion
Seismic reflection imaging can be formulated in the depthdomain (original Cartesian coordinates of the subsurface)or in the time domain (coordinates defined by the so-calledimage rays). The time-domain formulation admits efficientimaging and inversion algorithms thanks to the existence ofeffective approximations, which provide prior informationfor seismic velocity estimation. I will describe recent ad-vances in developing time-domain algorithms and in solv-ing the problem of time-to-depth conversion.
Sergey FomelUniversity of Texas at [email protected]
MS66
Algorithms for Seismic Imaging with MultiplyScattered Waves
The single scattering assumption is ubiquitous in seismicimaging; almost all images are formed under the Born ap-proximation. There is ample evidence, however, that asignificant amount of energy is multiply scattered withinthe Earth. When this is the case, applying algorithms thatassume single scattering results in images with artifacts.We will discuss extensions that allow for the inclusion ofmultiply-scattered waves in imaging algorithms and discusstheir performance.
Alison [email protected]
Alan RichardsonDepartment of Earth, Atmospheric and PlanetarySciencesMassachusetts Institute of Technologyalan [email protected]
MS66
Frozen Gaussian Approximation for High Fre-quency Wave Propagation
Abstract not available at time of publication.
Xu YangCourant Institute of Mathematical SciencesNew York [email protected]
MS67
Scientific Computing Projects for Undergraduates
Technology today is changing so dramatically and soquickly that there is certainly a need to educate students
in computer literacy and applications. Computer experi-ments are as valuable to the mathematics student as labexperiments are to a chemistry student. I have integratedcomputer projects in all levels of courses to help develop thestudents’ mathematical skills that are necessary to functionin this growing technological world.
James BaglamaUniversity of Rhode [email protected]
MS67
An Innovative Scientific Computation Course forUndergraduates
Over the past five years, we have developed a problem-driven Python-based course in scientific computation forour undergraduate math majors, which replaces a conven-tional Java (or C-C++) programming course requirement.The course is taught in a computer-equipped classroomwhere students can obtain in-class help with their program-ming and mathematical errors. During my talk, we will dis-cuss our rationale for this curriculum change and provideexamples of classroom activities and student’s projects.
Adam O. HausknechtUniversity of Massachusetts DartmouthNorth Dartmouth, [email protected]
MS67
Running an Undergraduate Summer Research Pro-gram in Parallel Computing: a Challenging but Re-warding Experience
Parallel computing, the future of computing, is rarelytaught at undergraduate level due to the complexity of par-allel programming and the cost and maintenance expensesof massively parallel systems. EXERCISE- Explore Emerg-ing Computing in Science and Engineering is a new Re-search Experiences for Undergraduates (REU) site hostedat Salisbury University funded by NSF. This presentationwill focus on how to overcome the challenges of runningsuch an undergraduate research program in a primarily un-dergraduate institution.
Enyue LuSalisbury UniversitySalisbury, [email protected]
MS67
The Impact of Undergraduate Research in Scien-tific Computing on Undergrads at the Universityof Massachusetts
In this talk, we will describe undergraduate mathematicalresearch that we have supervised at the University of Mas-sachusetts Amherst. This includes honor’s theses as well assummer research. We will give descriptions of the projects,which were primarily in the area of modeling tumor angio-genesis. We also discuss the impact of these projects onthe direction and the future success of the students.
Nathaniel WhitakerDepartment of Mathematics and StatisticsUniversity of Massachusetts, [email protected]
198 CS13 Abstracts
MS68
Data-driven Optimal H∈ Model Reduction
We present a combined H∈ trust-region descent and It-erative Rational Krylov Algorithm (IRKA) approach foroptimal H∈ model reduction of multiple-input/multiple-output (MIMO) linear dynamical systems. The proposedapproach is formulated in terms of transfer function evalu-ations only, without requiring access to any particular real-ization. We consider this approach in the context of data-driven model reduction applied to extreme MIMO systems(large number of inputs or outputs).
Christopher A. BeattieVirginia Polytechnic Institute and State [email protected]
Tim CampbellNaval Research LaboratoryStennis Space [email protected]
Serkan GugercinVirginia Tech.Department of [email protected]
MS68
Model Order Reduction fo Un(Steady) Aerody-namic Applications
In this talk the fast simulation of air flowing past an air-foil is addressed using POD-based model order reduction.In particular, reduced order models are used to predict(un)steady aerodynamic flows. Numerical results for ahigh-lift wing-flap configuration are presented. Such con-figurations are used in the take-off and landing phase offlight.
Heike FassbenderTU [email protected]
MS68
Index-preserving MOR for Nonlinear DAE Sys-tems
We introduce a model order reduction procedure fordifferential-algebraic equations, which is based on explic-itly splitting the DAE into the intrinsic differential equa-tion contained in the original system and on the remainingalgebraic constraints. The index 1 case will be discussedextensively, as well as extensions to higher index. We im-plement numerically this procedure and show numericalevidence of its validity, as well as advances over existingMOR techniques.
Wil SchildersTechnical University of [email protected]
Nicodemus BanagaayaTU [email protected]
Giuseppe AliUniv. of [email protected]
Caren TischendorfHumboldt University of [email protected]
MS68
Data-Driven Model Order Reduction via ConvexOptimization: Improved Bounds for Fitting StableNonlinear Models
In this talk we present a convex optimization framework foroptimizing stable nonlinear state-space models to matchrecordings from experiment or simulation. The approachconsists of a parameterization of stable nonlinear modelsand a convex upper bound on the long-term mismatch be-tween simulated response and recordings. The resultingoptimization problem is implementable as a semidefiniteprogram. We demonstrate the approach on the task ofidentifying reduced order models of examples drawn fromthe circuits domain.
Mark Tobenkin, Yan Li, Alexandre [email protected], [email protected], [email protected]
MS69
Estimation of Uncertain Parameters Through Par-allel Inversion
We compare an optimization-based method for identify-ing stochastic parameters with a sampling-based approach.Unlike Bayesian methods, the sampling-based approach weconsider is based on solving a sequence of deterministic in-verse problems combined with ideas from sparse grid collo-cation. Knowledge of the parameters at collocation pointsallows us to approximate statistical quantities of interestvia numerical quadrature. We present numerical resultsfor an elliptic PDE where the conductivity parameter isstochastic.
Jeff BorggaardVirginia TechDepartment of [email protected]
Hans-Werner Van WykDepartment of Scientific ComputingThe Florida State [email protected]
MS69
Karhunen-Loeve Expansion for Multiple Corre-lated Stochastic Processes
In this work, we propose two numerical techniques to modeland simulate multiple correlated random processes. Thetwo techniques find the appropriate expansion for eachcorrelated random process by generalizing the Karhunen-Loeve expansion to multiple processes, while attaining theentire correlated stochastic structure. The primary differ-ence between the two methods lies in whether the randomvariables used in the expansion are independent or corre-lated. Some explicit formulae and analytical results arepresented for exponentially correlated random processes.In addition, these two methodologies are compared to eachother in terms of convergence and computational efficiencyand to other methods including mixtures of probabilisticPCA. We also simulate the tumor cell model induced by
CS13 Abstracts 199
colored noise and limit cycle oscillator with our techniques.
Heyrim ChoBrown UniversityProvidence, RIheyrim [email protected]
George E. KarniadakisBrown UniversityDivision of Applied Mathematicsgeorge [email protected]
Daniele VenturiBrown Universitydaniele [email protected]
MS69
Predictive Simulations for Problems with SolutionNonuniqueness
Large eddy simulations of turbulent mixing show examplesof nonunique but apparently converged simulations. The-ory allows a zero parameter selection of a unique solution,within the context of front tracking and dynamic subgridmodels. Experimental data confirms this result. Extrapo-lation beyond the experimental range to infinte Reynoldsnumbers is shown to be mild. Results are interpretated inthe context of inertial confinement fusion simulations.
James GlimmState University of New York, Stony [email protected]
MS69
Stochastic (w*) Convergence for Turbulent Com-bustion
We test two fundamental ideas for numerical simulationof turbulent combustion: (1) Finite rate chemistry forLarge Eddy Simulations; (2) Stochastic (w*) convergencebased on probability distribution functions and mathemat-ical ideas associated with Young measures. Convergence ismeasured in terms of L1 norms for CDFs. Our verification,validation and uncertainty quantification test platform isthe combustion within the engine of a scram jet, which isa Mach 7 experimental aircraft under study at StanfordUniversity PSAAP Center.
Tulin KamanSUNY Stony [email protected]
James GlimmState University of New York, Stony [email protected]
MS70
Hermite Methods for Hyperbolic Systems: Appli-cations and Extensions
This talk discuss applications and extensions of the Her-mite methods described in the previous talk. In particular,propagation of discontinuities, hybridization with discon-tinuous Galerkin methods will be considered. Applicationsof the Hermite methods to compressible flow and electro-magnetics will also be presented.
Daniel Appelo
Department of Mathematics and StatisticsUniversity of New [email protected]
Thomas M. HagstromSouthern Methodist UniversityDepartment of [email protected]
MS70
Hermite Methods for Hyperbolic Systems: BasicTheory
Hermite methods are spectral element methods defined onstaggered computational cells of cuboids whose degrees-of-freedom are tensor-product Taylor polynomials defined ateach vertex. When they are applied to hyperbolic systems,the time step is restricted only by the CFL constraint,independent of the polynomial degree, enabling memory-efficient implementations. In this talk we outline the the-ory of Hermite discretizations and compare their resolvingpower to that of competitive schemes.
Thomas M. HagstromSouthern Methodist UniversityDepartment of [email protected]
Chang-Young JangSouthern Methodist [email protected]
Daniel AppeloUniversity of New [email protected]
Ronald ChenUniversity of [email protected]
MS70
Accurate Solution of Diffusive Problems in Im-mersed Domains
In this talk we discuss the Correction Function Method(CFM), a general framework used to devise highly accu-rate numerical schemes to discretize diffusion dominatedphenomena in immersed domains. The combination of theCFM with Gradient Augmented Level Set Methods resultsin a powerful tool that can be applied to a variety of situa-tions in which both immersed domains and high accuracyare required. Here we present results for the Poisson andthe incompressible Navier-Stokes equations.
Alexandre N. [email protected]
Jean-Christophe NaveMcGill [email protected]
Rodolfo R. RosalesMassachusetts Inst of TechDepartment of [email protected]
200 CS13 Abstracts
MS71
Efficient Simulation of Multiscale Kinetic Trans-port
We discuss a new class of approaches for simulating multi-scale kinetic problems, with particular emphasis on appli-cations related to small-scale transport. These approachesare based on an algebraic decomposition of the distributionfunction into an equilibrium part, that is described deter-ministically (analytically or numerically), and the remain-der, which is described using a particle simulation method.We show that it is possible to derive evolution equations forthe two parts from the governing kinetic equation, leadingto a decomposition that is dynamically and automaticallyadaptive, and a multiscale method that seamlessly bridgesthe two descriptions without introducing any approxima-tion. Our discussion pays particular attention to stochas-tic particle simulation methods that are typically used tosimulate kinetic phenomena; in this context, algebraic de-composition can be thought of as control-variate variance-reduction formulation, with the nearby equilibrium servingas the control. Such formulations can provide substantialcomputational benefits in a broad spectrum of applicationsbecause a number of transport processes and phenomena ofpractical interest correspond to perturbations from nearbyequilibrium distributions. In many cases, the computa-tional cost reduction is sufficiently large to enable other-wise intractable simulations. The proposed methods willbe illustrated with applications to a variety of problemsof engineering interest, such as microscale/nanoscale gasflows and microscale/nanoscale solid-state heat transfer asmediated by phonon transport.
Nicolas HadjiconstantinouMassachussets Institute of [email protected]
MS71
Modeling and Simulation of Suspensions with aLarge Number of Interacting Micro-swimmers
Microorganisms play an important role in nature. Un-derstanding aspects of their locomotion and collective be-havior is essential to understanding many biological andphysical phenomena, as well as how to best use them intechnological applications. Designing mathematical andcomputational models to help scientists in these endeav-ors is paramount. We present a new mathematical modeland simulation method to compute the collective dynam-ics of a large colony of micro-swimmers that interact witheach-other and the fluid they are suspended in and can af-fect by their locomotion. This fast computational methoduses the immersed boundary framework and enables us toefficiently simulate thousands of interacting motile parti-cles. We illustrate the method by showing examples ofcollective dynamics in large suspensions of ”pusher” and”puller” micro-swimmers. The model satisfactorily cap-tures macroscopic structures of observed in experiments ofbacterial baths. Lastly, applications of the method willbe discussed, e.g. in for bacteria suspensions or syntheticchemically-powered particles.
Enkeleida LushiImperial College Londonand Courant [email protected]
Charles S. PeskinCourant Institute of Mathematical Sciences, New YorkUnivers
MS71
Lattice-Boltzmann-Langevin Simulations of BinaryMixtures and Wetting Instabilities in Thin FluidFilms
I will describe a hybrid numerical method for the solutionof the Model H fluctuating hydrodynamic equations forbinary mixtures. The momentum conservation equationswith Landau-Lifshitz stresses are solved using the fluctuat-ing lattice Boltzmann equation while the order parameterconservation equation with Langevin fluxes is solved us-ing stochastic method of lines ensuring that fluctuation-dissipation theorem is satisfied. The method is bench-marked by comparing static and dynamic correlations andexcellent agreement is found between analytical and nu-merical results. Thermally induced capillary fluctuationsof the interface are captured accurately, indicating that themodel can be used to study nonlinear fluctuations. I willalso discuss the performance of such an method to under-stand the instabilities of confined binary mixtures and therole of dynamic wetting.
Ignacio Pagonabarraga MoraDepartament de Fsica FonamentalUniversitat de Barcelona, [email protected]
MS71
Simulation of Osmotic Swelling by the StochasticImmersed Boundary Method
We present a numerical model that employs the stochasticimmersed boundary method to study the osmotic swellingof a microscopic vesicle. The scale is so small that in-dividual solute molecules are tracked explicitly. This isan important biological problem because water movementinside cells is generally driven by osmotic forces. Thetime-dependent Stokes equations are discretised using thestochastic immersed boundary method, and we allow thefluid to slip through the vesicle wall thus making the modellipid bilayer permeable to water. The elastic energy of themembrane has been modeled by a sum of three terms: aterm proportional to area that generates surface tension,a surface neo-Hookean energy that resists shear, and aHelfich bending energy proportional to the integral of thesquare of the sum of the principal curvatures. In additonto the membrane energy, we also employ an energy of in-teraction between the membrane and the explicitly trackedsolute particles to keep the solute within the vesicle. Wefind that the vesicle swells or shrinks (depending on its ini-tial size) and eventually fluctuates about an equilibriumsize that depends on the number of solute particles con-tained within the vesicle.
Charles S. PeskinCourant Institute of Mathematical SciencesNew York [email protected]
Chen-Hung WuCourant Institute of Mathematical [email protected]
Paul J. AtzbergerUniversity of California-Santa [email protected]
CS13 Abstracts 201
MS72
On the Evaluation of the Singular Integrals of Scat-tering Theory
I will describe an efficient method for the numerical evalu-ation of the singular integrals which arise in the discretiza-tion of certain integral operators of scattering theory givenon surfaces. Standard techniques for evaluating these inte-grals, while adequate in simple cases, become prohibitivelyexpensive when applied to integral operators given on com-plicated surfaces. The scheme I will describe, by contrast,is largely insensitive to the geometry of the underlying sur-face.
James BremerUC [email protected]
MS72
Fast Volume Integral Equation Solver for LayeredMedia
Wave scattering in layered media is studied via the Green’sfunction. The Green’s function is developed by the scatter-ing matrix technique and Sommerfeld-type integral. Then,the application of the integral operator for the Helmholtzequation in layered media is accelerated with the fast mul-tipole method (FMM) and local-expansion treecode for theBessel function. The same methods are applied to the vec-tor potential Green’s function for Maxwell’s equations inlayered media.
Min Hyung ChoDepartment of MathematicsDartmouth [email protected]
Wei CaiUniversity of North Carolina, [email protected]
MS72
Quadrature by Expansion: A New Method for theEvaluation of Layer Potentials
We present a systematic, high-order approach to the com-putation of layer potentials that works for any singularity(including hypersingular kernels), based only on the as-sumption that the field induced by the integral operator issmooth when restricted to either the interior or the exte-rior of the bounding curve or surface. The scheme, denotedQBX (quadrature by expansion), is easy to implement andcompatible with the fast multipole method.
Andreas KloecknerCourant Institute of Mathematical SciencesNew York [email protected]
Alexander BarnettDepartment of MathematicsDartmouth [email protected]
Leslie Greengard, Michael O’NeilCourant Institute of Mathematical SciencesNew York [email protected], [email protected]
MS72
Fast Algorithms for Layer Heat Potentials
We will describe a new fast algorithm for evaluating layerheat potentials. A new recurrence relation is derived whichreduces the computational complexity of the heat poten-tials from quadratic (of direct evaluation) to linear in thenumber of time-steps. The key advantages over other fastmethods are its adaptivity and insensitivity to the time-step size. When combined with high-order product inte-gration rules that overcome geometrically-induced stiffness,this algorithm can be used for solving the diffusion equa-tion accurately and efficiently on moving geometries withDirichlet or Neumann data. This is joint work with LeslieGreengard and Shidong Jiang.
Shravan VeerapaneniDepartment of MathematicsUniversity of [email protected]
MS73
A First Passage Time Algorithm for Reaction-Diffusion Processes on a 2D Lattice
Abstract not available at time of publication.
Linda R. PetzoldUniversity of California, Santa [email protected]
MS73
Parallelization and Error Analysis in Lattice Ki-netic Monte Carlo
n this talk we explain an operator splitting approach to par-allel implementation of kinetic Monte Carlo simulations ofspatially distributed particle systems on a lattice. We dis-cuss the error analysis of the algorithm that gives approx-imations of the conventional, serial kinetic Monte Carlo(KMC). A key aspect of our analysis relies on emphasiz-ing a goal-oriented approach for suitably defined macro-scopic observables (e.g., density, energy, correlations, sur-face roughness), rather than focusing on strong topologyestimates for individual trajectories. One of the key im-plications of our error analysis is that it allows us to ad-dress systematically the processor communication of differ-ent parallelization strategies for KMC by comparing their(partial) asynchrony, which in turn is measured by theirrespective fractional time step for a prescribed error toler-ance. This is a joint work with G. Arampatzis and M. A.Katsoulakis.
Petr PlechacUniversity of DelawareDepartment of Mathematical [email protected]
MS73
Efficient Algorithms and Parallel Issues for KineticMonte Carlo Modeling in Materials Science
Making kinetic Monte Carlo (KMC) applications run scal-ably in parallel involves trade-offs in performance ver-sus accuracy, particularly with respect to relaxing the re-quirement that the underlying KMC algorithm be ”ex-act” versus acceptably approximate. I’ll discuss serial andparallel algorithms we’ve developed that come down onboth sides of this issue and illustrate how we’ve imple-
202 CS13 Abstracts
mented them in our parallel KMC simulator SPPARKS(http://spparks.sandia.gov). One novel feaature of SP-PARKS is that it allows users to add new KMC models bywriting functions that calculate energy changes and enu-merate possible ”events”. The code achieves parallelismby spatially decomposing the simulation domain, which as-sumes the energetics of an event depends only on nearbyinformation. This is a good assumption for many mate-rials science problems, but I’ll also highlight examples forsurface growth and nuclear fuel sintering where this is notthe case.
Steven J. PlimptonSandia National [email protected]
MS73
Simulation of Strained Epitaxial Growth with Ki-netic Monte Carlo
We present weakly-off-lattice and off-lattice kinetic MonteCarlo models for strained epitaxial growth. Both formula-tions are based on the observation that near equilibrium,the chemical potential is the thermodynamic driving forcefor film evolution. For the weakly off-lattice system, theenergy is taken from a bonding counting, ball and springmodel whereas in the off-lattice case an intermolecular po-tential is used. This is joint work with Henry Boateng andTim Schulze.
Peter SmerekaDepartment of MathematicsUniversity of [email protected]
MS74
Stable Product of Matrices and Application inQuantum Monte Carlo Simulation on GPU Accel-erated Multicore Systems
The product singular value decomposition is one way of cir-cumventing numerical instability in calculatioing the prod-uct of many matrices. However, it is prohibitively ex-pensive in terms of floating point operations, storage re-quriements and data communication. This is particularlytrue for large-scale many-body Monte Carlo simulations incomputational materials science. In this talk, we present adifferent approach and work with a graded decompositionof the product. Furthermore, we use a pre-pivoting schemefor the grade revealing to reduce the data communicationcost and efficiently exploit highly optimized primitive ma-trix kernels on hybrid CPU and GPU systems.
Zhaojun BaiDepartments of Computer Science and MathematicsUniversity of California, [email protected]
Andres TomasDepartment of Computer ScienceUniversity of California, [email protected]
Richard ScalettarDepartment of PhysicsUniversity of California, [email protected]
MS74
Numerical Behavior of Two-step Splitting IterationMethods
In this contribution we study numerical behavior of sev-eral stationary or two-step splitting iterative methods forsolving large sparse systems of linear equations. We showthat inexact solution of inner systems associated with thesplitting matrix may considerably influence the accuracy ofcomputed approximate solutions computed in finite preci-sion arithmetic. We analyze several mathematically equiv-alent implementations and find the corresponding compo-nentwise or normwise forward or backward stable imple-mentations. The theory is then illustrated on the class ofefficient two-step iteration methods such as Hermitian andskew-Hermitian splitting methods. This is a joint workwith Zhong-Zhi Bai.
Miro RozloznikAcademy of Sciences of theCzech Republic, [email protected]
MS74
Using High-precision Arithmetic in the Design of aStopping Criterion for Lanczos
The classical Lanczos method is the most memory-efficientway to compute all the eigenvalues of a large sparse sym-metric matrix. Its convergence can be slow, but it doesconverge to all the eigenvalues and it is possible to deter-mine which eigenvalues have converged. However, unlessall the eigenvalues are distinct, it is not possible to deter-mine when all of them have been found. To address thisissue, we show that multiple eigenvalues can be dispersedby adding to A a random matrix with a small norm. Byusing high-precision arithmetic, we can perturb the eigen-values by an amount that does not affect the accuracy ofdouble-precision computed eigenvalues.
Sivan A. ToledoTel Aviv [email protected]
Alexander Alperovich, Alex DruinskyTel-Aviv [email protected], [email protected]
MS74
Computational Noise, Derivatives, and Optimiza-tion
Efficient simulation of complex phenomena often results incomputational noise. Noise destroys underlying smooth-ness that otherwise could benefit optimization algorithms.We present a non-intrusive method for estimating com-putational noise and show how this noise can be used toderive finite-difference estimates with provable approxima-tion guarantees. Building upon these results, we show howstep sizes for model minimization and improvement can beselected in derivative-free optimization.
Stefan WildArgonne National [email protected]
Jorge J. MoreArgonne National LaboratoryDiv of Math & Computer Science
CS13 Abstracts 203
MS75
Towards a Fine-Grained Parallel Implementationof the Nonsymmetric QR Algorithm
We present the first step towards a fine-grained parallel im-plementation of the non-symmetric QR algorithm targetingmulti-core processors and shared memory. Our primarygoal is high performance on the full range from small tolarge problems. To reach this goal, we overlap the criticalpath with delayed updates and parallelize both the bulgechasing and the aggressive early deflation kernels. In ad-dition, different scheduling techniques and matrix storageformats are evaluated.
Lars KarlssonUmea UniversityComputing Science and [email protected]
MS75
Parallel Multishift QR and QZ Algorithms withAdvanced Deflation Strategies - Recent Progress
Key techniques used in our novel distributed memory par-allel QR and QZ algorithms include multi-window bulgechain chasing and distributed aggressive early deflation(AED), which enable level-3 chasing and delayed updateoperations as well as improved eigenvalue convergence. Re-cent progress includes a multi-level recursive approach forperforming AED in a parallel environment leading to com-munication avoiding algorithms via data redistribution.Application and test benchmarks confirm the superb per-formance of our parallel library software.
Bo T. KagstromUmea UniversityComputing Science and [email protected]
MS75
Designing Fast Eigenvalue Solvers on ManycoreSystems
Abstract not available at time of publication.
Piotr LuszczekDepartment of Electrical Engineering and ComputerScienceUniversity of Tennessee, [email protected]
MS75
The Parallel Nonsymmetric QR Algorithm withAggressive Early Deflation
We present the new parallel nonsymmetric QR algorithmin ScaLAPACK v2.0. The multiwindow bulge chain chas-ing approach ensures that most computations in the bulgechasing stage are performed in level 3 BLAS. We also de-velop multilevel aggressive early deflation algorithms whichdecrease the total amount of communications. These tech-niques make the new approach significantly outperform thepipelined QR algorithm in ScaLAPACK v1.8.0. Both per-formance models and numerical experiments demonstrate
the efficiency of our new approach.
Meiyue ShaoMATHICSEEPF [email protected]
MS76
Lessons Learned from Managing the Open SourceLibrary Deal.II
We will review our experience with managing deal.II, anopen source project which today has 600,000 lines of codeand hundreds of users around the world. This will includethe technical side, such as ensuring quality in the long term.It will also include social aspects of managing a commu-nity, with the challenges of attracting and retaining volun-teer contributors, maintaining quality in contributions ofnewcomers, and adequate documentation and training.
Wolfgang BangerthTexas A&M [email protected]
Timo HeisterTexas A&M UniversityDepartment of [email protected]
Guido KanschatDepartment of MathematicsTexas A&M [email protected]
MS76
The Development and Adoption of the TriBITSLifecycle Model in CSE Projects
We describe a proposal for a well-defined software lifecycleprocess based on modern Lean/Agile software engineeringprinciples for research-driven CSE software. What we pro-pose is appropriate for many CSE software projects thatare initially heavily focused on research but also are ex-pected to eventually produce usable high-quality capabil-ities. We describe the motivations for this work as wellas the efforts to get it adopted in CSE projects includingTrilinos and CASL.
R. A. BartlettOak Ridge National [email protected]
Michael A. Heroux, James WillenbringSandia National [email protected], [email protected]
MS76
IPython: a Tool for the Lifecycle of ComputationalIdeas
IPython is an open source environment for interactive andparallel computing that supports all stages in the lifecycleof a scientific idea: individual exploration, collaborative de-velopment, large-scale production using parallel resources,publication and education. The web-based IPython Note-book supports multiuser collaboration and allows scientiststo share their work in an open document format that is atrue “executable paper”: notebooks can be version con-
204 CS13 Abstracts
trolled, exported to HTML or PDF for publication, andused for teaching.
Fernando PerezHelen Wills Neuroscience InstituteUniversity of California, [email protected]
MS76
libMesh: Lessons in Distributed Collaborative De-sign and Development
Science is naturally suited to open source software develop-ment. Accurate, replicable publication requires completedescription of all algorithms in numerical experiments.Source code publication allows true review and reuse ofalgorithms therein, enabling greater collaboration and co-operation between researchers. We review advantages andchallenges of open source scientific applications. Efficiencyversus usability, compatibility versus innovation, and cen-tralized versus distributed development are discussed inthe context of the MOOSE [?] simulation framework andlibMesh [?] finite element library.
Roy StognerUniversity of Texas at [email protected]
MS77
An Enhanced Derivative-free Approach to EnergyApplications
It has been well documented that derivative-free algorithmsare immensely useful for the optimization of black-boxfunctions. In this talk, I will explain how their effective-ness can be augmented by the inclusion of sensitivity cal-culations. I will describe an algorithm that monitors localsensitivities throughout the optimization process and usesthis information as a stopping criteria. I will discuss theapplicability of this approach to energy relevant applica-tions. Specifically, I will discuss planning and operationsof the electrical grid and radioactive waste disposal optionsfor nuclear power plants.
Genetha GraySandia National [email protected]
MS77
Results of Design Studies Using Derivative-freeOptimization for Multi-Layered Filters
Filtration applications appear in a variety of physical set-tings; among them are industrial filtration for polymer pro-cessing, protein separation in pharmaceutical drug purifi-cation, and oil and air filtration in the automotive indus-try. Effective filters remove large amounts of debris, butcost considerations warrant filters that have long lifetimes.Thus, one must balance the need for effective filters againstthe costs of replacement; filters that trap everything wouldhave short life spans. Alternatively, one could make a fil-ter last forever by trapping nothing. Filter design can beevaluated using computational simulators and optimiza-tion tools that balance these competing objectives. In thistalk,we summarize the optimization results obtained usinga variety of different algorithms. We discuss the use of dif-ferent objective functions, provide analysis of the designsreturned by the algorithms, and give directions for future
work.
Lea JenkinsDepartment of Mathematical SciencesClemson [email protected]
MS77
Sparse Interpolatory Models for Molecular Dynam-ics
We describe a method for using interpolatory models to ac-curately and efficiently simulate molecular excitation andrelaxation. We use sparse interpolation for efficiency andlocal error estimation and control for robustness and accu-racy. The objective of the project is to design an efficientalgorithm for simulation of light-induced molecular trans-formations. The simulation seeks to follow the relaxationpath of a molecule after excitation by light. The simulatoris a predictive tool to see if light excitation and subsequentreturn to the unexcited or ground state will produce a dif-ferent configuration than the initial one. The goals of thesimulation are not only to identify the end point, but to re-port the entire path in an high-dimensional configurationspace so that one can look for nearby paths to interest-ing configurations and examine the energy landscape nearthe path to see if low energy barriers make jumping to adifferent path possible
Carl T. KelleyNorth Carolina State UnivDepartment of Mathematicstim [email protected]
David MokrauerBAE [email protected]
James NanceNorth Carolina State [email protected]
MS77
Parameter Estimation for Modeling Threat Detec-tion in the Brain using Derivative-free Methods
Several neural network architectures have been constructedmodeling the human brain’s attentive response to stim-uli. These neural network architectures are optimized to fitdata from physical experiments. Using optimization tools,parameters are fit to these data. Parameters can vary fromconnection strength in the architecture to the existenceof a connection between nodes. We examine both whichmodel architecture best fits experimental electrophysiolog-ical data and which optimization tool achieves the best fitfor each model.
Benjamin RitzDepartment of MathematicsClarkson [email protected]
MS78
Output-Based Adaptation for Hybridized Discon-tinuous Galerkin Methods
We present output-based adaptive solutions of the steadycompressible Euler and Navier-Stokes equations using a hy-
CS13 Abstracts 205
bridized discontinuous Galerkin (DG) discretization. Errorestimates are obtained using a discrete adjoint that takesinto account both element-interior and interface approxi-mations. Adaptation consists of changing the approxima-tion orders independently on elements and interfaces. Bothhybridized and lower-cost embedded DG methods are con-sidered, and results are compared to standard DG adaptiveruns.
Krzysztof Fidkowski, Johann Dahm, Peter KleinUniversity of [email protected], [email protected], [email protected]
MS78
Recent Developments in the Flux ReconstructionMethod and Extensions to Large Eddy Simulation
Theoretical studies and numerical experiments suggestthat unstructured high-order methods can provide solu-tions to otherwise intractable fluid flow problems. How-ever, existing high-order schemes are less robust and morecomplex to implement than their low-order counterparts.Such issues have limited the adoption of high-order tech-niques in both academia and industry. In this talk ourefforts to address a range of issues currently pacing theadoption of unstructured high-order schemes will be dis-cussed.
Antony JamesonProfessor, Department of Aeronautics & AstronauticsStanford [email protected]
MS78
High-Order Flux Reconstruction Schemes: Theoryand Implementation
The Flux Reconstruction (FR) approach to high-ordermethods is simple to implement and allows various high-order schemes, such as nodal discontinuous Galerkin meth-ods, and Spectral Difference methods, to be cast within asingle unifying framework. In this talk, new theoreticalaspects of FR schemes will be discussed, and efficient im-plementation strategies for novel hardware platforms willbe presented.
Peter E. VincentDepartment of Aeronautics and AstronauticsStanford [email protected]
Freddie Witherden, Antony FarringtonDepartment of AeronauticsImperial College [email protected],[email protected]
MS78
Finite Spectral ElementMethod for IncompressibleFlows
Finite spectral method is a category of pointwise or cell-wise local spectral schemes based on Fourier integral. Wecombine the finite spectral basis function with the finite el-ement method. We can not only use element discretizationin the computation domain, but also increase the exactnessin each element. We first introduce the finite spectral in-terpolation functions and compare them with the Lagrange
interpolation functions. Then, we combine the finite spec-tral interpolations functions with finite element methodand give the 2D incompressible Navier-Stokes equations interms of the stream function and the vorticity. Finally, wesolve the benchmark lid-driving cavity problem and flowaround a cylinder with unstructured mesh.
Jian-Ping WangPeking [email protected]
MS79
Exploring Co-Design in Chapel using LULESH
Chapel is an emerging parallel programming languagewhose design and development are led by Cray Inc.LULESH is an unstructured Lagrangian explicit shock hy-drodynamics code developed by Lawrence Livermore Na-tional Laboratory. In this talk, we describe a collabora-tion between Cray and LLNL to explore the expression ofLULESH within Chapel. This codesign effort has resultedin an improved version of LULESH while also providingvaluable feedback from users on Chapel’s feature set andimplementation priorities.
Bradford L. ChamberlainCray [email protected]
MS79
Programming Model Support Necessary forAdapting High Performance Code to DifferingPlatforms
Portable performance of million line multi-physics codes re-quires programming techniques and compiler support thatenable optimization without substantial code changes. Inthis talk, we describe tuning techniques used to improvethe performance of ALE hydrodynamics mini-apps. TheLULESH and LUAU mini-aps are presented as a case studyfor performance portability, architectural impact on codeperformance and vendor co-design. We present a style andabstractions for writing multi-physics codes that can beauto-tuned for multiple platforms.
Ian Karlin, Jim McGraw, Jeff Keasler, Bert StillLawrence Livermore National [email protected], [email protected], [email protected],[email protected]
MS79
Leveraging the Cloud for Materials Proxy Applica-tions
One oft-discussed approach to exascale development isMPI+X. This two-part model envisions using X to accel-erate code on local, heterogeneous node architectures andMPI to construct the overall application and communicatebetween the nodes running X. Today, X includes languagessuch as CUDA, OpenCL, Cilk+, TBB, HMPP, OpenMP,etc.-more will follow in the future. We describe the addi-tion of ‘Cloud’ technologies that enable the rapid develop-ment of dynamic multi-scale materials science proxy ap-plications. Our discussion will focus on NoSQL databases(e.g. Riak, MongoDB) and non-traditional programminglanguages (e.g. Erlang).
Christopher MitchellLos Alamos National Laboratory
206 CS13 Abstracts
MS79
High-Level Abstractions for Portable Performanceusing LULESH
Exascale machines will be a significant departure from tra-ditional architectures, for which multiphysics simulationscodes have been developed, and will require an accom-panying change in the programming model. Obtainingportable performance necessitates introducing high-levelabstractions above the architecture-specific details. We re-port on explorations into these paradigms using LULESH,a simple shock hydrodynamics proxy application represen-tative of the data structures and numerics used in largerapplications of interest.
Charles H. StillLawrence Livermore National [email protected]
Zach DevitoStanford [email protected]
MS80
Hybrid Subspace Methods for Dimensionality Re-duction in Nonlinear Multi-Physics Models
Recent developments on hybrid subspace methods have in-troduced a new general approach to performing dimension-ality reduction for nonlinear multi-physics models. Thebasic idea is to construct linear transformation to fewerdegrees of freedom replacing the original I/O streams ofthe coupled physics models. We demonstrate the applica-tion to a typical nonlinear nuclear engineering model withmany input parameters and quantities of interest, currentlyintractable with the state-of-the-art methods due to thecomputational cost required.
Hany S. Abdel-KhalikNorth Carolina State [email protected]
MS80
Decomposition Methods for Multidisciplinary Un-certainty Analysis
The focus of this talk is a decomposition approach for mul-tidisciplinary uncertainty analysis of systems governed bypartial differential equations. The main idea of our ap-proach is to perform uncertainty analysis independently onlocal components in an ”offline” phase, and then to assem-ble global uncertainty quantification with pre-computedlocal information in an ”online” phase. In this talk, weshow how this is achieved through a combination of do-main decomposition methods, reduced basis methods, andsampling importance re-sampling.
Qifeng [email protected]
Karen E. WillcoxMassachusetts Institute of [email protected]
Tiangang Cui
MS80
Stochastic Dimension Reduction of Multi PhysicsSystems through Measure Transformation
Uncertainty quantification of multiphysics systems rep-resents numerous mathematical and computational chal-lenges. Indeed, uncertainties that arise in each physics ina fully coupled system must be captured throughout thewhole system, the so-called curse of dimensionality. Wepresent techniques for mitigating the curse of dimensional-ity in network-coupled multiphysics systems by using thestructure of the network to transform uncertainty represen-tations as they pass between components. Examples fromthe simulation of nuclear power plants will be discussed.
Eric PhippsSandia National LaboratoriesOptimization and Uncertainty Quantification [email protected]
Paul ConstantineStanford [email protected]
John Red-HorseValidation and Uncertainty Quantification ProcessesSandia National [email protected]
Tim WildeySandia National [email protected]
Roger GhanemUniversity of Southern CaliforniaAerospace and Mechanical Engineering and [email protected]
Maarten ArnstUniversite de [email protected]
MS80
Gradient-based Model Reduction forHigh-dimensional Uncertainty Quantification
This talk focuses on a Reduced Order Modeling approachfor Uncertainty Quantification, where we use gradient in-formation to partition the uncertainty domain into “active’and “passive’ subspaces, where the “passive’ subspace ischaracterized by near constant value of the quantity of in-terest. We project the model onto the low dimensional“active’ subspace and solve the resulting problem usingconventional techniques. We derive rigorous error boundsfor the projection algorithm and show convergence in L1
norm.
Miro Stoyanov, Clayton G. WebsterOak Ridge National [email protected], [email protected]
MS80
A Generalized Adjoint Framework for Sensitivity
CS13 Abstracts 207
and Global Error Estimation in Burnup Calcula-tions
We develop an abstract framework for computing the ad-joint of a general set of differential-algebraic equations, andwe apply the framework to the transport/depletion equa-tions which are used to model advanced nuclear reactordesigns. The framework efficiently generates both para-metric sensitivity information, which we use for calibrationof the (up to thousands of) random or uncertain input pa-rameters, as well as estimates for numerical discretizationerrors.
Hayes StriplingTexas A&M [email protected]
Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected]
Marvin AdamsTexas A&M [email protected]
MS81
A Hybrid Adaptive Mesh Framework for WavePropagation Algorithms on a Forest of Locally Re-fined Cartesian Meshes
We describe current efforts to develop ForestClaw, a hybridAMR finite volume code based on wave propagation algo-rithms in which non-overlapping fixed-size Cartesian gridsare stored as leaves in a forest of quad- or oct-trees. Thetree-based code p4est manages the multi-block connectiv-ity and is highly scalable in realistic applications. In jointwork with researchers at the Cascade Volcanic Observatory(L. Mastin and H. Schwaiger, CVO, Vancouver, WA), wewill present results from our efforts to use ForestClaw formodeling the transport of volcanic ash in the atmosphere.
Donna CalhounBoise State [email protected]
Carsten BursteddeInstitut fuer Numerische SimulationUniversitaet [email protected]
MS81
Title Not Available at Time of Publication
Abstract not available at time of publication.
David GeorgeU.S. Geological SurveyCascades Volcano [email protected]
MS81
Optimal Strong-Stability-Preserving Runge-KuttaMethods for Discontinuous Galerkin Spatial Dis-cretizations of Hyperbolic Problems
In this talk, we present the construction of explicit strong-
stability-preserving Runge–Kutta methods with optimalstability regions for discontinuous Galerkin spatial dis-cretizations applied to hyperbolic problems.
Ethan KubatkoDepartment of Civil and Environmental EngineeringThe Ohio State [email protected]
Benjamin YeagerThe Ohio State [email protected]
MS81
High Order Accurate RKDG Methods for the Shal-low Water Equations on Unstructured TriangularMeshes
Shallow-water equations with a non-flat bottom topogra-phy have been widely used to model flows in rivers andcoastal areas. These equations have steady-state solutionsin which the flux gradients are non-zero but exactly bal-anced by the source term. Therefore extra care must bepaid to approximate the source term numerically. Anotherimportant difficulty arising in these simulations is the ap-pearance of dry areas, and standard numerical methodsmay fail in the presence of these areas. In this presenta-tion, we will talk about recently developed high-order dis-continuous Galerkin methods on unstructured triangularmeshes, which can preserve the steady-state exactly, andat the same time are positivity preserving without loss ofmass conservation. Some numerical tests are performed toverify the positivity, well-balanced property, high-order ac-curacy, and good resolution for smooth and discontinuoussolutions.
Yulong XingDepartment of MathematicsUniveristy of Tennessee / Oak Ridge National [email protected]
MS82
A New Three-Field Stabilized Finite ElementMethod for Fluid-Structure Interactions
We present some advancement towards a monolithic solu-tion procedure for the numerical solution of fluid-structureinteractions. A new three-field stabilized finite elementformulation is presented for modelling the interactions be-tween the fluid (laminar or turbulent) and the structure(rigid or elastic). We combine this method with anisotropicmesh adaptation to ensure an accurate capturing of the dis-continuities at the interface. The accuracy of the formu-lation is demonstrated by means of 2D and 3D numericalexamples.
Elie Hachem, Thierry CoupezMINES [email protected],[email protected]
Ramon CodinaUniverstitat Politecnica de Catalunya (UPC)[email protected]
MS82
Three Dimensional Optimal Transportation Mesh-free (OTM) Simulations of Human Arterial Blood
208 CS13 Abstracts
Flow
We present a monolithic Lagrangian solution for the fluid-structure interaction problems involving large deforma-tion structure and free-surface flows. In our approach,both the fluid and structure are modeled by the Opti-mal Transportation Meshfree (OTM) method. The per-formance of the proposed method is demonstrated by athree-dimensional simulation of patient-specific modelingof arterial blood flow. Especially an anisotropic hyperelas-tic constitutive model with fiber reinforcement is developedto model the three layers of the human arterial wall.
Bo Li, Stefanie HeydenCalifornia Institute of [email protected], [email protected]
Anna PandolfiPolitecnico di [email protected]
Michael OrtizCalifornia Institute of [email protected]
MS82
Deforming Composite Grids for Fluid-structure In-teraction
In this talk, we discuss recent work on overlapping griddiscretizations of compressible fluid-structure interactionproblems. Both elastic and rigid structures are considered.Deforming composite grids (DCGs) are used to address ge-ometric complexity in a highly efficient and flexible man-ner. The FSI coupling is partitioned so that the fluid andsolid solvers remain independent. Added-mass instabili-ties are addressed through the use of a newly developedinterface projection technique.
Donald W. SchwendemanRensselaer Polytechnic InstituteDepartment of Mathematical [email protected]
Jeffrey W. BanksLawrence Livermore National [email protected]
William D. HenshawCenter for Applied Scientific ComputingLawrence Livermore National [email protected]
MS82
A Semi-local Solver for h-p Discretization of theStructure Equations
We develop a semi-local method for structure solvers,which decouples the three directions of displacement, en-abling the use of an efficient low energy preconditioner forthe conjugate gradient solver. Based on spectral elementwith Jacobi modal basis, we demonstrate high parallel ef-ficiency for structure simulations on a 3D patient-specificflexible brain arteries test problem. The new solver com-bined with spectral element for the fluid discretization,improves greatly the computational efficiency of the FSIsolver.
Yue Yu
Brown Universityyue yu [email protected]
Marco BittencourtDepartamento de mecanicoUniversidade de [email protected]
George E. KarniadakisBrown UniversityDivision of Applied Mathematicsgeorge [email protected]
MS83
Computational Issues for Boundary Control Prob-lems with Actuator Dynamics
The problem of boundary control in systems governed bypartial differential equations often leads to abstract controlsystems with unbounded input operators and very weakstate spaces. Moreover, in most practical settings the inputat the boundary v(t) is typically the output of a dynamic“actuator’ so that v(t) = Hxa(t) where the actuator statexa(t) is defined by a finite dimensional system. Althoughthe inclusion of actuator dynamics is a more realistic rep-resentation of the complex system, certain abstract formu-lations of the composite can bring additional complexityto the control problem and does not provide a practicalapproach to the development of computational methods.However, if one begins with the fundamental PDE systemand treats the actuator dynamics as boundary dynamicalsystems, then it is possible to simplify both the theoreticaland computational challenges. We illustrate this approachwith a simple boundary control problem defined by a non-linear parabolic PDE.
John A. BurnsVirginia TechInterdisciplinary Center for Applied [email protected]
MS83
A Combined Controls and Computational FluidsApproach for Estimation of a Moving GaseousSource
A combined controls and computational fluids dynamicsapproach is applied to the estimation of gas concentra-tion associated with an emitting moving source. The stateestimator uses a filter gain parameterized by the sensorposition whose motion dynamics is incorporated into thespatial process. The estimator is based on a 3D adaptive,multi-grid, multi-step finite-volume method with upwindand flux limiting. The grid is adapted with local refine-ment and coarsening during the process-state estimation.
Michael A. DemetriouWorcester Polytechnic [email protected]
Nikolaos GatsonisWorcester Polytechnic [email protected]
MS83
Challenges in Computational Nonlinear ControlTheory: a Perspective from the Air Force Office
CS13 Abstracts 209
of Scientific Research
Modeling, design and control of high performance complexAir Force Systems have put increasing demand on compu-tational resources and theoretical and algorithmic devel-opment to meet the scientific challenges in dealing withthe underlying high-dimensional, nonlinear, and stochasticproblems. While there has been serious research on de-velopment of algorithms specifically designed for controlor optimization of these problems (e.g. flow control, vi-bration control, design of structures), we have yet to meetthe needs of scalability, efficiency and accuracy required tomeet the real-world application requirements. In this talk,a brief discussion of Air Force needs in computational con-trol theory will be presented and the scientific challengeswill be addressed.
Fariba FahrooAir Force Office of Scientific [email protected]
MS83
Simultaneous Actuator Placement and ControllerDesign
Many control systems of interest, for example, active noisecontrol and control of structural vibrations, are modelledby partial differential equations. Because of the distribu-tion of the system in space the location of actuators, andsensors, in these systems is a variable in the design of acontrol system. The performance of the controlled systemis strongly dependent on these locations. Thus, actuatorand sensor locations are important variables in controllerdesign and should be considered as part of controller syn-thesis. Conditions for well-posedness of the optimal ac-tuator location problem and also for convergence of op-timal locations chosen using approximations in the caseswhere the cost is linear-quadratic (or H-2) and also for theH-infinity situation have been obtained. Algorithms havebeen developed to solve these problems, but challenges forlarge-scale problems remain. The best actuator locationsdo not always agree with physical intuition, even for simpleexamples. This supports the need for further research intothis area.
Kirsten MorrisDept. of Applied MathematicsUniversity of [email protected]
MS84
Analysis and Numerical Solutions of Quasi-steadyState Poroelasticity Problems
Abstract not available at time of publication.
Yanzhao CaoDepartment of Mathematics & StatisticsAuburn [email protected]
MS84
Adaptive Sparse Reconstruction for Prior Selectionand Robust Geophysical Inversion
Abstract not available at time of publication.
Behnam JafarpourChemical and Electrical Engineering
University of Southern [email protected]
MS84
Sparse Regularized Seismic Inverse Problem
In this talk, we frist address the sparsity-promoting sesmicdata reconstruction from L1-norm to nuclear-norm min-imization. Then we apply low-rank matrix completion(MC) with a designed texture-patch pre-transformation tothree-dimensional seismic data reconstruction. An efficientL1-norm minimizing algorithm, named approximate mes-sage passing (AMP), is extended to use for our nuclear-norm minimization problem. Numerical experiments showpromising performance of the proposed method in com-parison to traditional singular value thresholding (SVT) ofMC and recent tensor completion.
Jianwei MaDepartment of MathematicsHarbin Institute of [email protected]
MS85
Thoughts on Preparation: How to Lower the Bar-rier for using Computational Tools and Learning toProgram?
A successful undergraduate research experience has thestudent making a contribution to the research group andbecoming better prepared for a graduate program in CSE.To accomplish this, the student must quickly develop hercomputational-thinking and coding skills. Ideally, cod-ing ability should be developed integrally, with computingpresent across the curriculum. In practice, this is rare. Aresearch group hosting undergraduates can offer a ”bootcamp” of coding tasks of incremental nature to developskills.
Lorena A. BarbaDepartment of Mechanical EngineeringBoston [email protected]
MS85
A Five Year Experiment on Developing an Under-graduate Research Computing Program
In 2006 we received a CSUMS grant from the NSF. Accord-ing to the solicitation, CSUMS was to enhance the educa-tion and training of math undergraduates and to betterprepare them for fields that require integrated strengthsin computation and mathematics. This talk will describewhat we did, what we got right, what we modified, andwhat we botched. It will also address the question asto whether computing helps reinforce, or complements,or interferes with the more mathematical aspects of theprojects.
Mark HolmesRensselaer Polytechnic [email protected]
MS85
Writing and Publishing a Scientific Paper with Un-dergraduate Students
I had opportunities to write journal papers with under-
210 CS13 Abstracts
graduate students. Some papers were co-authored withone undergraduate student and some with a group of stu-dents. In this talk, I will share my experience in writinga journal paper with undergraduate students, e.g. how Istarted my research with undergraduate students and whatdifficulties and joy I found while working with them.
Jae-Hun JungSUNY at [email protected]
MS85
The PRISM Interdisciplinary Program at North-eastern
The NSF-funded PRISM program at Northeastern Uni-versity is run jointly by faculty from the Mathematics,Physics, Biology, Engineering and Education Departments.Students participate in faculty-led exploration and discov-ery courses, involving a mixture of theory, hands-on activ-ities, and data analysis, as well as virtual environments inthe Action Lab. Matlab is used throughout the programand is taught at multiple levels.
Christopher KingNortheastern [email protected]
MS86
A Model Reduction Approach for PartitionedTreatment of Uncertainty in Coupled Systems
We present a stochastic model reduction framework forpartitioned treatment of the uncertainty space in domaincoupling problems. In particular, we expand the solutionof each domain in a stochastic basis that is constructedadaptively and by calling the domain solvers separately.The novelty of the proposed approach is that the prop-agation of uncertainty is achieved through a sequence ofapproximations with respect to the dimensionality of eachindividual domain and not the combined dimensionality.
Alireza DoostanDepartment of Aerospace Engineering SciencesUniversity of Colorado, [email protected]
Mohammad HadigolAerospace Engineering SciencesUniversity of Colorado, [email protected]
Hermann Matthies, Rainer NiekampTechnische Universitat [email protected], [email protected]
MS86
Stochastic Polynomial Chaos Basis Selection in theBayesian Framework
Generalised polynomial chaos (gPC) expansions allow therepresentation of the stochastic solution of a system whoseinput parameters are random variables. It is useful to se-lect a smaller set of gPC basis to reduce the computationalcost of making measurements or avoid over-fitting thatleads to inaccurate solutions especially when the numberof the available gPC basis is much bigger that the numberof the model evaluations. In some cases, the solution is
sparse at the stochastic level and can be expanded with agPC of only a few terms. E.g., under weak conditions, inelliptic stochastic partial differentiable equations (SPDE)with high-dimensional random coefficients, the solutionsadmit sparse representations with respect to gPC basis,while the deterministic solver required is expensive. Wepropose a fully Bayesian stochastic search strategy for theselection of the important gPC basis and the evaluation ofthe associated coefficients. The proposed method combinesBayesian model selection and regularised regression meth-ods. Therefore, it accomplishes both shrinkage and basisselection while it takes into account the model uncertainty.For the evaluation of the required posterior quantities wepropose an MCMC sampler. Some of the main advantages:(1) it provides interval estimates, (2) it quantifies the im-portance of each gPC base, (3) it allows the computation ofBayesian model average estimates or Bayes-optimal predic-tors. We show that the proposed method is able to detectthe significant gPC basis and recover potential sparse pat-terns in gPC expansion on a toy example and an 1D ellipticSDE with high-dimensional random diffusion coefficients.
Guang LinPacific Northwest National [email protected]
Georgios Karagiannispacific Northwest National [email protected]
MS86
Uncertainty Propagation in Finite Element Simu-lation of Particle Driven Flow
In this talk we address the initial conditions uncertaintypropagation in the variational multiscale finite elementsimulation of particle driven flow. We use sparse gridstochastic collocation methods managed by a scientificworkflow engine on a HPC environment. Particular in-terest is devoted to a quantity of interest related to thespatial pattern of sediment deposition.
Fernando A. RochinhaThe Federal University of Rio de [email protected]
Gabriel GuerraFederal University of Rio de [email protected]
Alvaro Coutinho, Jonas DiasThe Federal University of Rio de [email protected], [email protected]
Marta Mattoso, Eduardo OgasawaraFedral University of Rio de [email protected], [email protected]¿
Erb LinsFederal University of [email protected]
MS86
Noise Propagation in the Multiscale Simulation of
CS13 Abstracts 211
Coarse Fokker-Planck Equations
We present a numerical procedure to compute the solutionof a FokkerPlanck equation for which the drift and diffu-sion parameters are unknown, but can be estimated usingappropriately chosen realizations of a fine-scale, individual-based model. If the latter model is stochastic, the estima-tion procedure introduces noise on the coarse level. We in-vestigate stability conditions for this procedure and presentan analysis of the propagation of the estimation error in thenumerical solution of the FokkerPlanck equation.
Yves FrederixDept. of Computer Science, K.U. [email protected]
Giovanni SamaeyDepartment of Computer Science, K. U. [email protected]
Dirk RooseKU LeuvenDept. of Computer [email protected]
MS87
Interoperability of PETSc and Chombo
Abstract not available at time of publication.
Mark AdamsColumbia [email protected]
MS87
BoxLib: Overview and Applications
BoxLib is a publicly available software framework for build-ing parallel block-structured AMR applications. It sup-ports grid-based and particle-mesh operations on adaptivehierarchical meshes. BoxLib-based codes use both MPIand OpenMP, have demonstrated excellent scaling behav-ior on today’s largest machines, and are in active use ina number of research areas. The BoxLib distribution con-tains an extensive User’s Guide as well as straightforwardtutorials which demonstrate how to build parallel adaptiveapplication codes using BoxLib.
Ann S. AlmgrenLawrence Berkeley National [email protected]
John B. BellCCSELawrence Berkeley [email protected]
Michael LijewskiLawrence Berkeley National [email protected]
MS87
Region-Based AMR: A New AMR Paradigm inBoxLib
Region-based AMR (RAMR) is a new paradigm in BoxLibin which different regions at the same level of spatial refine-ment may have different temporal refinement. When com-
bined with optimal subcycling, which selects a timestep foreach region based on maximizing the efficiency of the over-all algorithm, the use of RAMR can result in significantcomputational savings over traditional AMR approaches.Examples will be given from the BoxLib-based Nyx codefor cosmology.
Ethan Van Andel, Ann S. AlmgrenLawrence Berkeley National [email protected], [email protected]
John B. BellCCSELawrence Berkeley [email protected]
Michael LijewskiLawrence Berkeley National [email protected]
MS88
Building Envelope Parameter Estimation throughHeat Transfer Modeling
In order to reduce energy consumption in buildings, oneneeds to understand and be capable of modeling the under-lying heat transfer mechanisms, characteristics of buildingstructures, operations and occupant energy consumptionbehaviors. Often, some of the crucial building envelopeparameters are unknown. In order to infer thermal pa-rameters associated with buildings envelope, sensor data isutilized within an inversion procedure. The forward prob-lem involves a system of differential equations capturingthe governing heat transfer model.
Raya HoreshIBM T.J. Watson Research [email protected]
Lianjun An, Young M. Lee, Young T. Chae, Rui ZhangIBM T.J. Watson Reseach [email protected], [email protected],[email protected], [email protected]
MS88
Simulating Heat Transfer and Environmental Con-ditions in Buildings Equipped with Sensor Net-works
We consider the problem of modeling air flow, heat trans-fer, and humidity in buildings, specifically in environmentswhere natural convection is dominant. Buildings are nowa-days often equipped with a network of sensors and a datamanagement system which gathers the sensor data in realtime. Such sensor measurements serve as input data for theboundary value problems for systems of partial differentialequations comprising the physical models. We will discussmodeling techniques that facilitate performing simulationsof the physical phenomena at hand and which are suit-able for coupling with measurements gathered via sensornetworks.
Vanessa Lopez-MarreroIBM T. J. Watson Research [email protected]
MS88
A Reduced-order Energy Performance Modeling
212 CS13 Abstracts
Approach for Buildings
The reduced-order building and HVAC system models withparameter estimation methods are generic and will elimi-nate the labor intensive, time consuming task of developingand calibrating specific detailed physics-based models. Inthis talk, we will present a circuit-equivalent 3R2C ther-mal network of building, which is a widely used and simplereduced order building model. The underlying physics ofthe 3R2C model will be introduced firstly, followed by thevalidation of the 3R2C model with ASHRAE Standard 140and a case study of a reduced-order 3R2C model of a realbuilding.
Zheng ONeillUnited Technologies Research [email protected]
MS89
Three-dimensional Acoustic Scattering from Ob-stacles in a Half-space with Impedance BoundaryConditions
A classical problem in acoustic and electromagnetic scat-tering concerns the evaluation of the Greens function forthe Helmholtz equation subject to impedance boundaryconditions on a half-space. We will discuss a hybrid rep-resentation of this Greens function which combines imagesand a rapidly converging Sommerfeld-like integral. Therepresentation is valid at arbitrary source and target loca-tions, and is amenable to evaluation using fast-multipolemethods. This is joint work with Leslie Greengard.
Michael O’NeilCourant InstituteNew York [email protected]
MS89
Quadrature Methods for the Sommerfeld Integraland their Applications
Abstract not available at time of publication.
Josef SifuentesNew York UniversityCourant [email protected]
MS89
Fast Fourier Transforms of Piecewise Polynomials
We will construct a fast transform, based on low-rankapproximation, which evaluates Fourier coefficients ofpiecewise-polynomial generalized functions. These gen-eralized functions are supported on d-dimensional sim-plices such as points, lines, triangles, or tetrahedra, in D-dimensional space. The transform employs a stable newdimensional recurrence and a tree-based butterfly scheme.
John A. StrainDepartment of MathematicsUniversity of California at [email protected]
MS89
High-order Solvers for Scattering Problems in Do-
mains with Geometric Singularities
Abstract not available at time of publication.
Catalin TurcDepartment of Mathematical SciencesNew Jersey Institute of [email protected]
MS90
Uncertainty in Turbulent Flows with EmpiricalSub-cooled Boiling Models
A sensitivity study is performed by varying tunable param-eters in four different implementations of Eulerian multi-phase boiling flow models. The modeling differences be-tween the codes allows for a more complete study of mul-tiphase boiling flow, but requires coupling the codes tomeaningfully interpret the results. Parameter ranges arecompiled from the literature. A few models with relativelylittle theoretical and empirical development have a signifi-cant impact on the quantities of interest.
Isaac Asher, Krzysztof FidkowskiUniversity of [email protected], [email protected]
MS90
Determination of Nitridation Reaction ParametersUsing Bayesian Inference
In this work, we present a computational study of a flowtube experiment in order to infer reaction parameters forgraphite nitridation. We construct a two-dimensional rep-resentation of the experimental setup and model the flowusing a reacting low-Mach number approximation to theNavier-Stokes equations. We employ a Bayesian approachin order to produce probability distributions that can usedto quantify uncertainty in models where surface nitridationis important.
Paul T. BaumanICESThe University of Texas at [email protected]
MS90
Uncertainty Modeling with Stochastic PDEs forTurbulent Channel Flow
Validation of and UQ for extrapolative predictions madeby RANS turbulence models are necessary to properly in-form decisions based on such predictions. Here, we explorethe use of stochastic PDEs for this purpose. In particular,multiple stochastic PDEs describing the modeling errorsobserved in the Reynolds stress are coupled with multipledeterministic turbulence models to make uncertain predic-tions of channel flow. These predictions are compared withDNS data to assess their credibility.
Todd OliverPECOS/ICES, The University of Texas at [email protected]
Robert D. MoserUniversity of Texas at [email protected]
CS13 Abstracts 213
MS90
Calibration of Stochastic non-Boltzmann KineticModels using EAST Shock Tube Data
We use Bayesian inference to calibrate the physical andstochastic model parameters using shock tube radiationmeasurements from NASA. We propose a novel formula-tion of the stochastic model based on the physical intuitionthat the discrepancy between predictions and experimentaldata is due to assumption of equilibrium population of theenergy levels. This formulation will enable us to propagatemodel uncertainty to both the observable for the parametercalibration and the quantity of interest.
Marco PanesiUniversity of Illinois at [email protected]
MS91
H-FaIMS: A Hierarchical Fast Inverse MediumSolver
We consider the inverse medium problem for the waveequation with broadband and multi-point illumination. Weuse a nonlinear least-squares formulation. If M is the num-ber of illuminations, a Hessian matrix-vector multiplicationrequires 2M wave solves. We have developed H-FaIMS, ascheme based on hierarchical decompositions that, asymp-totically, enables Hessian approximations that scale inde-pendently of the number of the sources. We present resultsfor the 3D Helmholtz problem.
George BirosUniversity of Texas at Austinbiros@ices. utexas.edu
MS91
Interferometric Waveform Inversion via Lifting andSemidefinite Relaxation
In seismic and SAR imaging, fitting cross-correlations ofwavefields rather than the wavefields themselves can resultin much improved robustness vis-a-vis model uncertainties.This approach however raises two challenges: (i) new spu-rious local minima may complicate the inversion, and (ii)one must find a good subset of cross-correlations to makethe problem well-posed. I the talk I will explain how toaddress these two problems with lifting, semidefinite relax-ation, and expander graphs.
Laurent Demanet, Vincent JugnonMathematics, [email protected], [email protected]
MS91
Bayesian Uncertainty Quantification in FWI
We adopt the Bayesian inference formulation for seismicinversion: given observational data and their uncertaintyand a prior probability distribution describing uncertaintyin the parameter field, find the posterior probability distri-bution over the parameter field. We exploiting the relationbetween the covariance matrix for the Gaussian approxi-mation of the posterior probability density function (pdf)and the inverse of the Hessian to study this posterior pdf.A low-rank representation of the misfit Hessian is used tomake our approach computationally feasible.
Tan Bui-Thanh
The University of Texas at [email protected]
Carsten BursteddeInstitut fuer Numerische SimulationUniversitaet [email protected]
Omar GhattasUniversity of Texas at [email protected]
James R. MartinUniversity of Texas at AustinInstitute for Computational Engineering and [email protected]
Georg StadlerUniversity of Texas at [email protected]
Lucas [email protected]
MS91
Optimal Design of Simultaneous Source Encoding
Many parameter estimation problems involve the collectionof an excessively large number of observations N , but it hasbeen observed that similar results can often be obtainedby considering a far smaller number K of multiple linearsuperpositions of experiments. To find the optimal weightsof these superpositions, we formulate the problem as anoptimal experimental design problem and show that theweights can be determined by using techniques from thisfield.
Kees van den DoelUniversity of British [email protected]
Eldad HaberDepartment of MathematicsThe University of British [email protected]
Lior HoreshBusiness Analytics and Mathematical SciencesIBM TJ Watson Research [email protected]
Kai RothaugeUniversity of British [email protected]
MS92
GPU Computing with QUDA
The exponential growth of floating point power in GPUsand high memory bandwidth, has given rise to an attractiveplatform upon which to deploy HPC applications. How-ever, deploying such computations on GPUs can be non-trivial because pre-existing applications cannot be recom-piled and run while maintaining high performance. Wereview the QUDA library which is a domain-specific li-brary designed to accelerate legacy lattice quantum chro-
214 CS13 Abstracts
modynamics applications through providing a library ofthe common performance-critical algorithms.
Michael ClarkHarvard-Smithsonian Center for [email protected]
MS92
VexCL: Vector Expression Template Library forOpenCL
VexCL is modern C++ library created for ease of OpenCLdevelopement. VexCL strives to reduce amount of boiler-plate code needed to develop OpenCL applications. The li-brary provides convenient and intuitive notation for vectorarithmetic, reduction, and sparse matrix-vector multiplica-tion. Multi-device and even multi-platform computationsare supported. This talk is a brief introduction to VexCLinterface.
Denis DemidovLobachevsky Institute of Mathematics and MechanicsKazan Federal [email protected]
MS92
Developing Numerical Algorithms on Heteroge-neous Architectures with High Productivity inMind
Porting existing or developing new scientific applicationson today’s heterogeneous architectures can be a very chal-lenging and time-consuming process. The necessity of ab-stracting the underlying hardware from the numerical de-velopers becomes a crucial approach to effectively use theavailable processing units. This separation of concerns fur-ther allows the mathematician and the computer scientistto respectively concentrate on what they are good at. Thistalk will describe how some of the linear algebra communityare tackling complex GPU-based systems when it comes toimplementing high performance numerical libraries.
Emmanuel [email protected]
Jack DongarraUniversity of [email protected]
Hatem LtaiefKAUST Supercomputing LaboratoryThuwal, [email protected]
Stanimire TomovInnovative Computing Laboratory, Computer ScienceDeptUniversity of Tennessee, [email protected]
MS92
ViennaCL: GPU-accelerated Linear Algebra at theConvenience of the C++ Boost Libraries
In order to provide simple access to the vast computingresources in graphics processing units (GPUs) for generalpurpose scientific computing, the open source linear alge-
bra library ViennaCL is presented. ViennaCL is writtenin C++ and used like existing CPU-based linear algebralibraries, thus it can be integrated into existing code easily.Moreover, the generic implementations of algorithms suchas iterative solvers allow for code reuse beyond device andlibrary boundaries.
Karl RuppInstitute for Analysis and Scientific Computing, TU WienInstitute for Microelectronics, TU [email protected]
MS93
A Multi-Scale Model for Capillary Driven Contact-Line Dynamics
We present a multi-scale method to simulate the flow oftwo immiscible incompressible fluids in contact with solids.The macro model is a level set method. The contact lineis tracked explicitly and moves according to a slip veloc-ity that depends on the wall contact angle of the interfacewith the solid. The relation between wall contact angleand slip velocity is determined in a micro model based onthe phase field method. The phase field method seeks foran equilibrium slip velocity in a box around the contactpoint, prescribed a static contact angle at the solid andthe wall contact angle in the far field. The dimensions ofthe box are chosen such that physical diffusion processesaround the contact point are fully represented. We presentnumerical results for capillary-driven flows which demon-strate the convergence of results in the macro model andcompare the behavior with other approaches in contact linedynamics.
Gunilla Kreiss, Martin KronbichlerDivision of Scientific ComputingUppsala [email protected], [email protected]
MS93
A Model for Simulating the Wrinkling and Buck-ling Dynamics of a Multicomponent Vesicle
Abstract not available at time of publication.
John LowengrubDepartment of MathematicsUniversity of California at [email protected]
MS93
High-resolution Solver for the Poisson-Nernst-Planck Equations and its Applications
In this talk we present a high-resolution finite-volumemethod for solving the Poisson-Nernst-Planck equationson adaptive grids and for complicated geometries. We willhighlight the importance of local charge conservation, atthe coarse-fine grid interface, on the overall accuracy ofthe solver. Next, we utilize the solver to study the charg-ing dynamics of super-capacitors at high voltages wherenonlinear effects can lead to new charging mechanism pre-viously unknown. Finally we will discuss possible futuredirections.
Mohammad MirzadehDept. Mechanical [email protected]
CS13 Abstracts 215
MS93
Mechanical Simulation of Mammalian Acini
Acini are small groups of cells that form hollow compart-ments and serve multiple biological functions in differentorgans. They are a common source of various types of can-cer, and recent work suggests that mechanical interactionsof the cells with their local environment may play a role incancer development. This talk will describe two mechanicalsimulation studies of acini that make use of new compu-tational techniques for large-strain nonlinear elasticity andmultiple deforming boundaries.
Chris RycroftDepartment of MathematicsLawrence Berkeley National [email protected]
MS94
Avoiding Communication in Parallel Bidiagonaliza-tion of Band Matrices
Successive band reduction is a technique for reducing asymmetric band matrix to tridiagonal form for the sym-metric eigenproblem. We have shown that a careful refor-mulation of the technique can asymptotically reduce thecommunication costs (i.e., data movement) on a sequentialmachine, compared to standard algorithms. In this talk,we will present the application of this approach to the re-duction of a non-symmetric band matrix to bidiagonal form(for the SVD) in the distributed-memory parallel setting.
Grey Ballard, Nicholas KnightUC [email protected], [email protected]
James W. DemmelUniversity of CaliforniaDivision of Computer [email protected]
MS94
Improved Accuracy for MR3-based Eigensolvers
A number of algorithms exist for the dense Hermitianeigenproblem. In many cases, MRRR is the fastest one,although it does not deliver the same accuracy as Di-vide&Conquer or the QR algorithm. We demonstrate howthe use of mixed precisions in MRRR-based eigensolversleads to an improved orthogonality, even surpassing theaccuracy of DC and QR. Our approach comes with limitedperformance penalty, and increases both robustness andscalability.
Paolo BientinesiAICES, RWTH [email protected]
MS94
Spectral Divide-and-conquer Algorithms for Gen-eralized Eigenvalue Problems
Spectral divide-and-conquer (SDC) algorithms solve eigen-value problems by computing an invariant subspace for asubset of the spectrum to decouple the problem into twosmaller subproblems. Recently Nakatsukasa and Highamintroduced a stable and efficient SDC algorithm for thesymmetric eigenvalue problem and the SVD. We propose
an SDC algorithm for generalized eigenvalue problems thatminimizes communication (its main computational kernelsare QR factorization and matrix-multiplication) and hasoperation count within a small constant factor of that forthe standard QZ algorithm.
Yuji NakatsukasaDepartment of MathematicsUniversity of [email protected]
MS94
Restructuring the Symmetric QR Algorithm forPerformance
Abstract not available at time of publication.
Robert A. van de GeijnThe University of Texas at AustinDepartment of Computer [email protected]
MS95
Model Reduction for Parameter Estimation inComputational Hemodynamics
We present a variational Data Assimilation procedure tothe estimation of the Young modulus of an artery. Theapplication of this approach to real problems is compu-tationally intensive. We present a Proper OrthogonalDecomposition-based strategy for the reduction of the com-putational costs of the inverse problem associated with theparameter estimation. We address the role of surrogatemodelling (such as 1D Euler equations) and surrogate so-lutions in this context.
Luca BertagnaDept. Math & CSEmory [email protected]
Alessandro VenezianiMathCS, Emory University, Atlanta, [email protected]
MS95
Energy-stable Galerkin Reduced Order Models forPrediction and Control of Fluid Systems
The focus of this talk is the construction of POD/GalerkinROMs for real-time prediction and control of fluid systems.An energy stability analysis reveals that the inner productemployed in the Galerkin projection step of the model re-duction dictates the ROMs stability. For linearized com-pressible flow, a symmetry transform leads to a stable for-mulation for the inner product. Stability of the proposedROM is demonstrated on several model problems. Exten-sions involving flow control are described.
Irina Kalashnikova, Srinivasan Arunajatesan, Bart G.Van Bloemen WaandersSandia National [email protected], [email protected],[email protected]
MS95
Computation of Periodic Steady States with the
216 CS13 Abstracts
Harmonic Balance Reduced Basis Method
In many applications, the flow solution reaches a periodicsteady state. To compute the PSS, the harmonic balancemethod expands the solution and the operator of the prob-lem as Fourier series. These expansions are truncated andthe problem reduces to solving a set of coupled nonlinearequations. The harmonic balance method is coupled withthe reduced basis method for spatial reduction. Error esti-mates and stability issues for the complex-valued reducedbasis systems are considered.
Toni LassilaMathematics Institute of Computational Science and [email protected]
Gianluigi RozzaSISSA, International School for Advanced StudiesTrieste, [email protected]
MS95
Reduced Basis Methods for Coupled Transport-reaction Problems
We consider a parameter-dependent multiphysics prob-lem which consists of horizontal transport and coupledbivariate convection-reaction. Such problems occur e.g.in catalysts. The coupling is nonlinear through Robin-type boundary conditions. Parameters may include in-flow as well as chemical properties of the reaction. Thetruth discretization leads to a nonlinear generalized saddle-point problem. A Reduced Basis Method is developed anda-posteriori error estimates are developed. In addition,we consider other features that are often present in real-world CFD-problems such as parameter functions, long-time horizons, time-periodicity or stochastic influences.The talk is based upon joint work with Masayuki Yano,Anthony T. Patera (MIT), Dominik Lechler, Antonia Mey-erhofer, Kristina Steih, Bernhard Wieland and Oliver Zeeb(all Ulm).
Karsten UrbanInstitute of Numerical Mathematics, University of [email protected]
MS96
High Performance Computational Models ofCoastal and Hydraulic Processes in an InteractivePython Environment
The Proteus toolkit is a software package for research onmodels for coastal and hydraulic processes and improve-ments in numerics. The models considered include multi-phase flow, shallow water flow, turbulent free surface flow,and various flow-driven processes. We will discuss the ob-jectives of Proteus and recent evolution of the toolkit’s de-sign as well as present examples of how it has been usedused to construct computational models for the US ArmyCorps of Engineers.
Chris KeesU.S. Army Engineer Research and Development CenterCoastal and Hydraulics [email protected]
Matthew FarthingUS Army Engineer Research and Development Center
MS96
Developing Open Source Software: LessonsLearned from Clawpack
The Clawpack (Conservation Laws Package) open sourcesoftware project has grown substantially since 1994 andhas recently branched into several related projects, withdevelopers scattered at many institutions. I will give abrief overview of the current development process, with afocus on some aspects that may be of most interest to otherresearchers who are contemplating sharing their own codein this manner, such as choice of licenses, version controlsystems, public repositories and hosting, and the use ofvirtualization to facilitate use of the software.
Randall J. LeVequeApplied MathematicsUniversity of Washington (Seattle)[email protected]; [email protected]; rjl@washingt
MS96
Feel++, a Library and Language in C++ forGalerkin Methods, from Rapid Prototyping toLarge Scale Multi-physics Applications
Abstract not available at time of publication.
Christophe Prud’hommeUniversity of [email protected]
MS96
The waLBerla/PE Parallel Multiphysics Frame-work
WaLBerla is a large scale software framework for multi-physics-applications based on kinetic methods. It is scal-able to beyond 100 000 cores. This talk will focus on how todeal with conflicting goals like high node performance andscalability on the one side, and clean software structure,maintainability, and flexibility on the other side.
Ulrich J. RuedeUniversity of Erlangen-NurembergDepartment of Computer Science (Simulation)[email protected]
Christian FeichtingerChiar for System SimulationUniversity of Erlangen-Nuremberg, [email protected]
Harald KoestlerUniversity of [email protected]
Tobias PreclikUniversity [email protected]
Florian SchornbaumUniversity of [email protected]
CS13 Abstracts 217
MS97
Optimal Decision Making in Network Security Un-der Uncertainty
With the advent of modern cyber warfare techniques, thefield of cyber security has been forced to adapt to a chang-ing climate of malicious activity. Complicated multi-stageattacks that constantly change have become the new normand have created a perpetual state of uncertainty for net-work administrators, who struggle to adapt their defensesto these new and ever-changing threats. In this work, wecreate a model that combines the methods of a PartiallyObservable Markov Decision process (POMDP) and Hid-den Markov Models (HMM) in an attempt to optimize theaction and strategies of a Network Administrator.
Michael J. FowlerDepartment of MathematicsClarkson [email protected]
MS97
Optimization to Understand Trade-offs in Agricul-tural Practices
Seawater intrusion along coastal California threatens to de-stroy freshwater resources, which is especially detrimentalto the berry industry in that region. Surface water anal-ysis and crop simulation can further aid decision makersin planting cycles. We seek to reduce the aquifer draw byanalyzing alternative farming techniques while simultane-ously meeting demands and maximizing profits. This isaccomplished using a farm model and optimization strate-gies to analyze approaches that meet a sustainable wateryield constraint.
Kathleen FowlerClarkson UniversityDepartment of [email protected]
MS97
Exploiting Expert Knowledge for EnhancedSimulation-Based Optimization
Safeguarding water supplies from contaminated sites isaided by linking models and global optimizers. Thisresearch augmented selected derivative-free optimizersto leverage non-traditional information like site-specificknowledge and practitioner rules-of-thumb. A benchmark-ing application was used in which extraction wells mustintercept pollutants at a contaminated site. A rules en-gine adjusted candidate wells to place them within plumeboundaries with a bias toward areas of rapid pollutant mi-gration. Incorporating this expert knowledge significantlyimproved optimizer performance.
L. Shawn Matott, Camden ReslinkUniversity at [email protected], [email protected]
MS97
Revealing the Difficulties in Managing CoastalAquifer Supply Problems
This research examines the difficulties encountered whenattempting to use optimization techniques to managecoastal ground water aquifers that are utilized as a sourceof fresh water. Coastal groundwater aquifers general con-
sist of fresh water underlain by salt water. When freshwater is extracted from the aquifer, this mechanism drawsthe salt water into the fresh water zone. If the salt water isdrawn into the well, then the well is contaminated and canno longer be utilized as a source of fresh water. The move-ment of the fresh water- salt water interface in response toground water extraction can be predicted using mathemat-ical models that are solved using computationally intensivealgorithms. In this research the ground water flow modelMODFLOW coupled with the principles of the Ghyben-Herzberg approach are used to model the fresh water- saltwater interface. An optimization problem is developed thatprovides a fixed supply of fresh water while minimizing thethreat of salt water contaminating the extraction wells inthe model. Derivative free methods of optimization areused to solve this optimization problem, first using geneticalgorithms, followed by a pattern search for refining thesolutions determined. The results from this optimizationexercise reveal multiple management solutions within theboundaries of acceptance for maintaining the utility of acoastal aquifer. Statistical tools are used to examine thesolutions to the optimization problems, revealing symme-tries and balancing that occurs in the solutions to coastalmanagement systems.
Karen L. RicciardiUniversity of Massachusetts [email protected]
MS98
A Method-of-lines Approach to Computing onGeneral Surfaces
The Closest Point Method is a set of mathematical princi-ples and associated numerical techniques for solving partialdifferential equations (PDEs) posed on curved surfaces orother general domains. The method works by embeddingthe surface in a higher-dimensional space and solving thePDE in that space using simple finite difference and inter-polation schemes. This presentation outlines some currentwork on formulating the algorithm as a method of linesand on solving elliptic problems.
Colin B. MacdonaldOxford [email protected]
Ingrid Von Glehn, Yujia Chen, Tom MarzUniversity of [email protected],[email protected], [email protected]
MS98
Strong Stability Preserving Methods for Time Evo-lution of Hyperbolic PDEs
In this talk we present the development of and recent re-search in strong stability preserving (SSP) methods fortime evolution of hyperbolic PDEs. We will discuss SSPRunge–Kutta and multistep methods as well as multi-stepmultistage methods, and present the barriers and boundsthat these methods have.
Sigal GottliebDepartment of MathematicsUniversity of Massachusetts [email protected]
218 CS13 Abstracts
MS98
An ODE and PDE Test Suite for Time-steppingMethods
We have created a MATLAB test suite for ODEs and PDEswhich allow users to test different time-integration methodsin a simple way. In this talk we present the details of thetest suite and its current capabilities.
Daniel L. HiggsUniversity of Massachusetts, [email protected]
MS98
Matrix-free Integrators for Large Discretized PDEs
We present a new class of Rosenbrock integrators based ona Krylov space solution of the linear systems. We developa framework for the derivation of order conditions for thenew Rosenbrock-K methods. This new class of methods re-quire only a small number of basis vectors, determined bythe order of accuracy, but independent of the ODE underconsideration. Numerical results show favorable proper-ties of Rosenbrock-K methods when compared to currentRosenbrock and Rosenbrock-W methods.
Adrian Sandu, Paul TranquilliVirginia Polytechnic Institute andState [email protected], [email protected]
MS99
Very High Order Residual Distribution Schemesfor Laminar and Turbulent Compressible Flow
We consider the numerical approximation of compressibleviscous fluid problems by residual distribution schemesthat can be considered as a non linear version of the stabi-lized finite element method. They need conformal unstruc-tured (hybrid) meshes. We first show how to approximatescalar steady viscous, and show that optimal order can bereached. Then we extend this to fluid problems. We showhow to deal with unsteady problems. Meshing issues willbe also considered: optimal order on complex geometriesrequire the use of curved meshes. In particular we foccuson the boundary layers geometrical approximation
Remi AbgrallINRIA Bordeaux Sud-OuestUniversite de [email protected]
Dante de Santis, Mario RicchiutoINRIAdante.de [email protected], [email protected]
Cecile DobrzynskiIMB, IPB, Universite de Bordeaux, FranceBacchus, INRIA Bordeaux Sud-Ouest, [email protected]
MS99
Robust Untangling of Curvilinear Meshes
We present a technique that allows to untangle high or-der/curvilinear meshes. The technique makes use of un-constrained optimization where element Jacobians are con-strained to lie in a prescribed range through moving log-
barriers. The untangling procedure starts from a possi-bly invalid curvilinear mesh and moves mesh vertices withthe objective of producing elements that all have boundedJacobians. Provable bounds on Jacobians are computedadaptively for any kind of elements, both for surface, vol-ume, hybrid or boundary layer meshes.
Christophe GeuzaineUniversity of [email protected]
MS99
H-to-P Efficiently: A Progress Report
The spectral/hp element method can be considered asbridging the gap between the traditionally low order finiteelement method on one side and spectral methods on theother side. Consequently, a major challenge which arisesin implementing the spectral/hp element methods is to de-sign algorithms that perform efficiently for both low- andhigh-order spectral/hp discretisations, as well as discreti-sations in the intermediate regime. In this talk, we explainhow the judicious use of different implementation strate-gies – combined with an understanding of the architectureon which one plans to run – can be employed to achievehigh efficiency across a wide range of polynomial orders.We examine both static and time-dependent problems, andexamine differences between using continuous Galerkin ver-sus discontinuous Galerkin discretizations.
Robert KirbyUniversity of [email protected]
Spencer SherwinImperial College [email protected]
MS99
Simulation of An Oscillating-Wing Power Genera-tor Using a High-Order CFD Method
We developed a massively parallel 3D code of compress-ible Navier-Stokes equations on moving and deforming do-mains. The code is based on an efficient high-order cor-rection procedure via reconstruction (CPR). We will re-port our recent study of unsteady turbulent flow past anoscillating-wing power generator using this code. We ap-ply a technique of active flow control to gain improvedefficiency of this power generator.
Chunlei LiangGeorge Washington [email protected]
MS100
Exploring Code Performance Issues on Many-coreDevices using the Multifluid PPM Code as a Rep-resentative CFD Application
The PPM (piecewise-Parabolic Method) code has beenused in a detailed study of code performance issues onmany-core devices, represented by Intels new MIC co-processor, with over 50 cores. The different implemen-tation strategies that were tried will be described alongwith the implications of the observed performance for rec-ommended programming techniques for many-core devices.In particular a key technique for increasing the computa-
CS13 Abstracts 219
tional intensity of the algorithm will be described.
Paul R. WoodwardLaboratory for Computational Science and EngineeringUniversity of [email protected]
Jagan Jayaraj, Pei-Hung Lin, Michael KnoxUniversity of [email protected], [email protected],[email protected]
Simon D. HammondScalable Computer ArchitecturesSandia National [email protected]
MS100
Accelerating Mini-FE, a Finite Element Proxy Ap-plication, on GPUs
The Mantevo performance project is a collection of self-contained proxy applications that illustrate the main per-formance characteristics of important algorithms. miniFEis intended to be and approximation to an unstructuredimplicit finite element or finite volume application. Thistalk will focus on GPU algorithms for assembling a finiteelement matrix. Results on both NVIDIAs Fermi and Ke-pler GPUs will be presented.
Justin LuitjensNVIDIA [email protected]
MS100
Developing a Multi-Architecture Implicit Particle-in-Cell Proxy
PlasmaApp3D is a 3D fully implicit Particle-in-Cell proxyapp developed to explore co-design issues associated withvarious levels of hardware and algorithmic abstraction onheterogeneous architectures with multiple levels of paral-lelism. The guiding philosophy behind this proxy app isthat the physics should only be implemented once, andshould be independent of the underlying hardware specificoperations and data-management. This notion of isolatingthe physics leads to a proxy app that allows for rapid test-ing of various data-management and architecture specificoptimizations. The current goal of the PlasmaApp3D de-velopement is a proxy app optimized for shared multi-coreand GPU system, although this may be later extended toinclude other architectures. Performance results as well asdesign and development methods will be presented.
Joshua Payne, Dana Knoll, Allen McPherson, WilliamTaitano, Luis ChaconLos Alamos National [email protected], [email protected], [email protected], [email protected], [email protected]
MS100
UMMA: Unstructured Mesh Micro Apps
We begin with a discussion of our target application,Chicoma. We then cover the conception and developmentof Unstructured Mesh Micro-Applications (UMMA) on avariety of architectures and finish by relating our experi-ences introducing ideas from UMMA back into our large
application.
Dean RisingerLos Alamos National [email protected]
MS101
Hybrid FOSLS/FOSLL∗
Abstract not available at time of publication.
Thomas ManteuffelUniversity of [email protected]
MS101
Least-Squares Finite ElementMethods for CoupledGeneralized Newtonian Stokes-Darcy Flow
The coupled problem for a generalized Newtonian Stokesflow in one domain and a generalized Newtonian Darcyflow in a porous medium is considered in this talk. Theflows are treated as a stress-velocity formulation for theStokes problem and a volumetric flux-hydraulic potentialformulation for the Darcy problem. The coupling is doneby using the well known Beavers-Joseph-Saffman interfacecondition. A least-squares finite element method is usedfor the numerical approximation of the solution.
Steffen MunzenmaierGottfried Wilhelm Leibniz University of [email protected]
MS101
Goal-Oriented Least-Squares Finite ElementMeth-ods
We present an approach to augment least-squares finite el-ement formulations with user-specified goals. The methodinherits the global approximation properties of the stan-dard formulation with increased resolution of the goal. Sev-eral theoretical properties such as optimality and enhancedconvergence under general assumptions are discussed andwe present an adaptive approach that results in efficient, lo-cally refined approximations that hone in on the quantity-of-interest with a range of numerical examples to supportthe approach.
Luke OlsonDepartment of Computer ScienceUniversity of Illinois at [email protected]
Jehanzeb ChaudhryDepartment of MathematicsColorado State [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
Kuo Liu, Tom ManteuffelDepartment of Applied MathematicsUniversity of Colorado at [email protected], [email protected]
220 CS13 Abstracts
Lei TangUniversity of Colorado at [email protected]
MS101
Least Squares Finite Element Methods for Non-Newtonian Fluids with Application to Blood Flow
Due primarily to the presence of red blood cells, wholeblood may exhibit significant non-Newtonian behavior, andconsequently, accurate numerical models for blood flowshould reflect the appropriate physics. This talk examinesthree categories of nonlinear constitutive models applica-ble to blood: viscoelastic fluids of Oldroyd type, shear-thinning fluids, and yield-stress behavior. In particular,we focus on the least-squares finite element approach asthe basis for an accurate and flexible discretization tech-nique for these challenging problems.
Chad WestphalWabash [email protected]
MS102
Discretization and Solvers for the Stokes Equationsof Ice Sheet Dynamics at Continental Scale
Ice sheets exhibit incompressible creeping flow with shear-thinning rheology. On a continental scale, the flow is char-acterized by localized regions of fast flow that are sepa-rated from vast slow regions by thin transition zones. Weuse a parallel, adaptive mesh, higher-order finite elementdiscretization and an inexact Newton method for the solu-tion of the nonlinear Stokes equations modeling ice sheetdynamics. Preconditioned Krylov methods for the scalablesolution of the discretized system, and effects of a highlyanisotropic discretization are discussed.
Tobin Isaac, Georg Stadler, Omar GhattasUniversity of Texas at [email protected], [email protected],[email protected]
MS102
Resolving Grounding Line Dynamics Using theBISICLES Adaptive Mesh Refinement Model
Ice sheet dynamics span a wide range of scales. Extremelyfine spatial resolution is required to resolve the dynam-ics of features like grounding lines, while such fine resolu-tion is unnecessary over large quiescent regions. BISICLESis a scalable adaptive mesh refinement (AMR) ice sheetmodel built on the Chombo framework, with a dynamicalcore based on the vertically-integrated model of Schoof andHindmarsh (2010). We will present results demonstratingthe effectiveness of our approach.
Daniel MartinLawrence Berkeley National [email protected]
Stephen CornfordUniversity of [email protected]
William LipscombLos Alamos National [email protected]
Esmond G. NgLawrence Berkeley National [email protected]
Stephen PriceLos Alamos National [email protected]
MS102
A Finite Element Implementation of Higher-orderIce-sheet Models: Mathematical and NumericalChallenges
Several models, characterized by different complexity andaccuracy, have been proposed for describing ice-sheet dy-namics. We introduce a parallel finite element implemen-tation for some of these models which constitutes partof the land ice dynamical core for MPAS library. Wepresent large-scale simulations of the Greenland ice-sheet,and explore methods for the solution of the resulting lin-ear and nonlinear systems. We also address the estimationof model parameters such as the friction coefficient at theice-bedrock interface.
Max GunzburgerFlorida State UniversitySchool for Computational [email protected]
Matt HoffmanLos Alamos National [email protected]
Wei LengLaboratory of Scientific and Engineering ComputingChinese Academy of Sciences, Beijing, [email protected]
Mauro PeregoFlorida State [email protected]
Stephen PriceLos Alamos National [email protected]
MS102
Analysis of Convergence and Performance Variabil-ity for Continental Ice Sheet Modeling at Scale
A prototype python-based verification and validation pack-age to assess the significant development work occurringin continental-scale ice sheet models is presented. A keyaspect is a performance V&V capability to quantify algo-rithms that are efficient but also sensitive and variable,performance variability of leadership class hardware, andthe impact of developments on both speed and robustness.The assessment of alternative dynamical core performancein terms of simulation value (speed versus accuracy) is dis-cussed.
Patrick H. Worley, Katherine J. EvansOak Ridge National [email protected], [email protected]
Adrianna BoghozianUniversity of [email protected]
CS13 Abstracts 221
MS103
Fourier Continuation Methods for Therapeutic Ul-trasound
HIFU is an ultrasound therapy which focuses destructiveacoustic energy on a target (e.g. a cancerous tumor), leav-ing the surrounding tissue unharmed. For HIFU simula-tion, wherein nonlinear waves propagate many times theirfundamental wavelength, the use of high-order methods isvital for accurate and efficient simulation. We present aclass of high-order solvers, based on the Fourier Contin-uation method, that are capable of accurate and efficientHIFU simulation in large, complex domains.
Nathan AlbinKansas State [email protected]
MS103
Jet Noise DNS using Arbitrary-order HermiteMethods
We discuss the application of Hermite methods to directnumerical simulations of compressible turbulence with spe-cial attention to jet noise.
Daniel AppeloDepartment of Mathematics and StatisticsUniversity of New [email protected]
Thomas M. HagstromSouthern Methodist UniversityDepartment of [email protected]
Tim ColoniusDivision of Engineering and Applied ScienceCalifornia Institute of [email protected]
Chang Young JangSouthern Methodist [email protected]
MS103
Stable High Order Finite Difference Methods forWave Propagation Problems
During the last decade, stable high order finite differ-ence methods as well as finite volume methods appliedto initial-boundary-value-problems have been developed.The stability is due to the use of so-called summation-by-parts operators (SBP), penalty techniques for imple-menting boundary and interface conditions, and the en-ergy method for proving stability. In this talk we discusssome aspects of this technique including the relation to theinitial-boundary-value-problem. By reusing the main ideasbehind the recent development, new coupling proceduresfor multi-physics applications have been developed. Wewill present the theory by analyzing simple examples andapply to complex multi-physics problems such as fluid flowproblems, elastic and electromagnetic wave propagation,fluid-structure interaction and conjugate heat transfer.
Jan Nordstromprofessor in scientific computingDepartment of Mathematics, Linkoping [email protected]
MS103
Stability of Interacting Solitary Water Waves,Standing Waves, and Breathers
We develop a high-performance shooting algorithm tocompute new families of time-periodic and quasi-periodicsolutions of the free-surface Euler equations involvingbreathers, traveling-standing waves, and collisions ofcounter-propagating and unidirectional solitary waves ofvarious types. The wave amplitudes are too large to bewell-approximated by weakly nonlinear theory, yet we oftenobserve behavior that resembles elastic collisions of solitonsin integrable model equations. A Floquet analysis showsthat many of the new solutions are stable to harmonic per-turbations.
Jon WilkeningUC Berkeley [email protected]
MS104
2D FTLE in 3D Flows: The Accuracy of usingTwo-dimensional Data for Lagrangian Analysis inThree-dimensional Fluid Flows
In experimental, three-dimensional vortex-dominatedflows, common particle image velocimetry (PIV) data isoften collected in only the plane of interest due to equip-ment constraints. For flows with significant out of planevelocities or velocity gradients, this can create large dis-crepancies in Lagrangian analyses that require accurateparticle trajectories. A Finite Time Lyapunov Exponent(FTLE) analysis is one such example, and has been shownto be very powerful at examining vortex dynamics and in-teractions in a variety of aperiodic flows. In this work,FTLE analysis of a turbulent channel simulation was con-ducted using both full three-dimensional velocity data andmodified planar data extracted from the same computa-tional domain. When the out-of-plane velocity componentis neglected the difference in FTLE fields is non-trivial. Aquantitative comparison and computation of error is pre-sented for several planes across the width of the channel todetermine the efficacy of using 2D analyses on the inher-ently 3D flows.
Melissa GreenSyracuse [email protected]
MS104
Dimensionality Reduction in Neuro-sensory Sys-tems
Abstract not available at time of publication.
J. Nathan KutzDepartment of Applied MathematicsUniversity of [email protected]
MS104
Model Reduction for Large-scale Systems usingBalanced POD and Koopman Modes
Abstract not available at time of publication.
Clancy W. RowleyPrinceton [email protected]
222 CS13 Abstracts
MS104
Recent Advances in Discrete Empirical Interpola-tion for Nonlinear Model Reduction
Abstract not available at time of publication.
Danny C. SorensenRice [email protected]
MS105
A Geometric Load-Balancing Algorithm for Multi-core Parallel Computers
Geometric partitioning is fast and effective for load-balancing dynamic applications, particularly those requir-ing geometric locality of data (particle methods, crash sim-ulations). We present, to our knowledge, the first paral-lel implementation of a multidimensional-jagged geomet-ric partitioner. Its MPI+OpenMP implementation makesit compatible with hybrid MPI+threads applications onmulticore-based parallel architectures. By keeping data inplace throughout the algorithm, we minimize data move-ment compared to recursive bisection methods. We demon-strate the algorithm’s scalability and quality relative to re-cursive bisection.
Mehmet DeveciTHe Ohio State [email protected]
Siva RajamanickamSandia National [email protected]
Umit V. CatalyurekThe Ohio State UniversityDepartment of Biomedical [email protected]
Karen D. DevineSandia National [email protected]
MS105
Parallel Anisotropic Mesh Adaptation with Spe-cific Consideration of Flow Features
Effective PDE-based simulations in a number of fields re-quire meshes with strong anisotropy and controlled meshesthat isolate specific solution features. For example, fluidsimulations require appropriate anisotropic mesh layoutsnear boundaries (e.g., boundary layers) and at interiorportions (e.g., flows with shocks or free shear layers). Inthis talk, we will present recent advances made in localmesh modification procedures for adaptively creating suchmeshes. Parallelization of these procedures will also bediscussed.
Onkar Sahni, Ryan Molecke, Mark ShephardRensselaer Polytechnic [email protected], [email protected], [email protected]
Kedar ChitaleUniveristy of Colorado [email protected]
Kenneth Jansen
University of Colorado at [email protected]
Saurabh TendulkarSimmetrix [email protected]
Mark BeallSimmetrix, [email protected]
MS105
Predictive Load Balancing Using Mesh Adjacenciesfor Mesh Adaptation
Parallel mesh adaptation on unstructured meshes requiresthat the mesh be distributed across a large number ofprocessors such that the adapted mesh fits within mem-ory. The goal of ParMA it to dynamically partition un-structured meshes directly using the existing mesh adja-cency information to account for multiple criteria. Resultswill demonstrate the ability of ParMA to rebalance largemeshes (greater than 100,000,000 regions) on large corecount machines (greater than 32,000) accounting for mul-tiple criteria.
Cameron SmithScientific Computation Research CenterRensselaer Polytechnic [email protected]
Onkar Sahni, Mark ShephardRensselaer Polytechnic [email protected], [email protected]
MS105
Parallel Mesh Generation and Adaptation on CADGeometry
Effective parallel generation and adaptivity of meshes ofCAD geometry add complexity due to the need to ensurethe mesh conforms to the CAD geometry. We will dis-cuss procedures to support unstructured meshes, with orwithout boundary layers, on massively computers includ-ing the ability to distribute the geometric model duringmesh adaptation. This leads to substantial memory us-age reductions on high processor counts by not requiring acomplete copy of the model on each process.
Saurabh TendulkarSimmetrix [email protected]
Ottmar KlaasSimmetrix, [email protected]
Rocco NastasiaSimmetrix [email protected]
Mark BeallSimmetrix, [email protected]
MS106
Towards Real-Time Power Grid Dynamics Simula-
CS13 Abstracts 223
tion using PETSc
The need for real-time power grid dynamics simulation hasbeen a primary focus of the power system community inrecent years. In this talk we present our experiences, andpresent preliminary results, to simulate power grid dynam-ics in real-time using the high performance computing li-brary PETSc. With the range of scalable linear, nonlinear,and time-stepping solvers, PETSc has the potential to bethe mathematical and computing framework needed for anonline dynamics simulator.
Shrirang AbhyankarArgonne National [email protected]
Barry F. SmithArgonne National LabMCS [email protected]
Alexander FlueckIllinois Institute of [email protected]
MS106
Exploitation of Dynamic Information in Power Sys-tem Phasor Measurements
The power grid is becoming more dynamic with high pen-etration of intermittent renewable sources and respon-sive loads. It is essential to establish a dynamic oper-ation paradigm in contrast to todays operation built onstatic modeling. Recent developments in phasor technol-ogy provide such an opportunity with high-speed time-synchronized measurement data. This talk will discussthe mathematical and computational challenges and recentadvancements in extracting and utilizing dynamic informa-tion in phasor measurements.
Zhenyu HuangPacific Northwest National [email protected]
MS106
Next Generation Modeling and Simulation ofBuilding Energy and Control Systems
Building energy and control systems can be representedby systems of coupled stiff differential equations, algebraicequations and discrete equations. These equations couplemodels from multiple physical domains. We will presenthow these systems can be modeled using the equation-based, object-oriented Modelica language. We discuss nu-merical challenges and opportunities to use such modelsfor analysis that is outside the capabilities of conventionalbuilding simulation programs.
Michael [email protected]
MS106
Scalable Dynamic Optimization
We present an approach for nonlinear programming (NLP)based on the direct minimization of an exact difffferentiablepenalty function using trust-region Newton techniques.
The approach provides all the features required for scalabil-ity: it can effifficiently detect and exploit directions of nega-tive curvature, it is superlinearly convergent, and it enablesthe scalable computation of the Newton step through iter-ative linear algebra. Moreover, it presents features that aredesirable for parametric optimization problems that haveto be solved in a latency-limited environment, as is thecase for model predictive control and mixed-integer non-linear programming.
Victor ZavalaArgonne National [email protected]
Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected]
MS107
High Resolution Simulation of Two-phase Flows onQuadtree Grids
Abstract not available at time of publication.
Arthur [email protected]
MS107
High-Order Interface Tracking Methods for Com-pressible and Incompressible Two-Phase Flow
We present new high-order interface tracking methods forcompressible and incompressible two-phase flow. Bothsystems represent domains with a signed distance func-tion, and PDEs use embedded boundary finite volume dis-cretizations. For hyperbolic problems the level set is ad-vected using a velocity from Riemann problems. For ellip-tic problems, the interface velocity comes from coupled im-plicit solves for viscosity and pressure. We will demonstratethe methods using two-phase compressible Euler equations,and two-phase Navier-Stokes with surface tension.
Mehdi VahabUniversity of California DavisDepartment of Applied [email protected]
Greg MillerUniversity of CaliforniaDepartment of Applied [email protected]
MS108
A High-order Discontinuous Galerkin Method withLagrange Multipliers for Advection-diffusion Prob-lems
A high-order Discontinuous Galerkin method with La-grange Multipliers (DGLM) is presented for the solutionof advection-diffusion problems on unstructured meshes.Unlike other hybrid discretization methods for transportproblems, it operates directly on the second-order formthe advection-diffusion equation. Like the DiscontinuousEnrichment Method (DEM), it chooses the basis functionsamong the free-space solutions of the homogeneous formof the governing partial differential equation, and relies
224 CS13 Abstracts
on Lagrange multipliers for enforcing a weak continuityof the approximated solution across the element interfaceboundaries. However unlike DEM, the proposed hybriddiscontinuous Galerkin method approximates the Lagrangemultipliers in a subspace of polynomials, instead of a sub-space of traces on the element boundaries of the normalderivatives of a subset of the basis functions. For homo-geneous problems, the design of arbitrarily high-order ele-ments based on this DGLM method is supported by a de-tailed mathematical analysis. For non-homogeneous ones,the approximated solution is locally decomposed into itshomogeneous and particular parts. The homogeneous partis captured by the DGLM elements designed for homoge-nous problems. The particular part is obtained analyticallyafter the source term is projected onto an appropriate poly-nomial subspace. This decoupling between the two partsof the solution is another differentiator between DGLMand DEM with attractive computational advantages. An aposteriori error estimator for the proposed method is alsoderived and exploited to develop an automatic mesh re-finement algorithm. All theoretical results are confirmedby numerical simulations that furthermore highlight thepotential of the proposed high-order hybrid DG methodfor transport problems
Sebastien Brogniez, Charbel FarhatStanford [email protected], [email protected]
MS108
Entropy Stability and High-order Approximationof the Compressible Euler Equations
This talk will discuss questions regarding parabolic regu-larization of the Euler equations and entropy stability. Inparticular a sub-class of parabolic regularizations is identi-fied that yields a minimum entropy principle. The conse-quences of this property will be illustrated on a high-orderfinite elements method.
Jean-Luc GuermondDepartment of MathematicsTexas A&M, [email protected]
Bojan PopovDepartment of MathematicsTexas A&M [email protected]
Murtazo [email protected]
MS108
Discontinuous Galerkin Spectral Element Approx-imation for Wave Scattering from Moving Objects
Accurate computation of wave scattering from moving,perfectly reflecting objects, or embedded objects with ma-terial properties that differ from the surrounding medium,requires methods that accurately represent the boundarylocation and motion, propagate the scattered waves withlow dissipation and dispersion errors, and don’t introduceerrors or artifacts from the movement of the mesh. Wedescribe development of a discontinuous Galerkin spectralelement approximation to satisfy these requirements. Ex-
amples are taken from aeroacoustics and electromagnetics.
David A. KoprivaDepartment of MathematicsThe Florida State [email protected]
MS108
Discontinuous Galerkin Methods for VlasovMaxwell Equations
The Vlasov-Maxwell system is one of the important modelsto study collisionless magnetized plasmas. It couples theVlasov equation satisfied by the distribution function ofthe particle(s) and the Maxwell system, and has wide ap-plications such as in space and laboratory plasmas, andfusion. The challenges in simulation come from high-dimensionality, multiple scales, and conservation. We willpresent our recent progress in developing and analyzingdiscontinuous Galerkin methods for Vlasov-Maxwell equa-tions.
Fengyan LiRensselaer Polytechnic [email protected]
MS109
Stochastic Collocation Techniques for UncertaintyQuantification in Reactor Criticality Problems
This talk focuses on stochastic collocation method appliedto a radiation transport problem. We consider a one dimen-sional reactor model and we study the effects of materialcross-section uncertainty on the reactor criticallity, specifi-cally uncertainty in the properties of the nuclear fuel, con-trol rod and potential crud deposit. We use the DENOVOradiation transport simulator and we demonstrate conver-gence of the stochastic collocation technique even for verylarge ranges of the uncertainty.
Nick DexterUniversity of [email protected]
MS109
Local Sensitivity Derivative Enhanced Monte Carlo
We present a Local Sensitivity Derivative Enhanced MonteCarlo (LSDEMC) method that utilizes derivative informa-tion in Voronoi cells around sample points to constructsurrogate response functions, then uses MC integration tocorrect the bias in the surrogate-based evaluations. A newestimator for the mean is developed using this strategy,which has lower variance than a simple average. We illus-trate the use of LSDEMC in uncertainty propagation foran SPDE system and present some theoretical results.
Vikram GargMassachusetts Institute of [email protected]
Roy StognerUniversity of Texas at [email protected]
MS109
Adaptive ANOVA-based Probabilistic CollocationKalman Filter for Inverse Uncertainty Quantifica-
CS13 Abstracts 225
tion
Probabilistic Collocation Kalman Filter (PCKF) is an ap-proach solving inverse uncertainty quantification problems.PCKF resembles the Ensemble Kalman Filter (EnKF) ex-cept that it represents and propagates model uncertaintywith the Polynomial Chaos Expansion (PCE) instead ofan ensemble of model realizations. The accuracy and ef-ficiency of PCKF depends on the selection of PCE bases.We present an adaptive algorithm to build the PCE expres-sion for PCKF. The adaptivity is based on two techniques.First, we adopt the idea of adaptive functional ANOVA(Analysis of Variance) decomposition, which approximatesa high dimensional function with the summation of a set oflow dimensional functions. Thus instead of expanding theoriginal model into PCE, we implement the PCE expan-sion on much lower dimensions. Second, for those inverseproblems where the observations are assimilated sequen-tially, we propose the method of Dynamic Parameteriza-tion which controls the dimensionality of the problem (thenumber of random variables used to represent parameteruncertainty) in different Kalman Filter loops. We give twoillustrative examples to demonstrate our algorithm.
Guang LinPacific Northwest National [email protected]
Weixuan LiUniversity of Southern [email protected]
Dongxiao [email protected]
MS109
Intrusive Analysis and Uncertainty Quantificationfor Nuclear Engineering Models
We investigate uncertainty analysis of nuclear engineeringsimulation models. Given that sampling at high-resolution,over large parameter space, is computationally prohibitive,we argue for the use of compact representations of uncer-tainty and for extraction of additional information fromeach model run. We accelerate the standard response sur-face learning techniques by using gradient information, andalso calibrated lower-resolution data; goal-oriented modelreduction is an important tool. Our work has been testedon codes developed within NEAMS/SHARP.
Oleg RoderickArgonne National [email protected]
Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected]
MS110
Graph 500 Performance on a Distributed-MemoryCluster
The Graph 500 benchmark ranks high-performance com-puters based on speed of memory retrieval. We reporton our experience implementing this benchmark on thedistributed-memory cluster tara in the UMBC High Per-formance Computing Facility. Our best run to date using
64 nodes achieved a GTEPS rate that would have put taraat rank 58 on the June 2012 Graph 500 list. We havesubmitted an official benchmark run for publication in theGraph 500 list. Advisor: Matthias Gobbert, UMBC
Jordan AngelEast Tennessee State [email protected]
MS110
Comparative Transcriptome Analysis to IdentifyGenes Regulating Elastogenesis
Arterial smooth muscle cells produce significantly moreelastic fibers when overexpressing the V3 variant of theproteoglycan versican. To identify possible causes of thisincrease, microarray and qPCR analyses were used to mea-sure associated changes in gene expression. We refined theset of V3 influenced genes by analyzing their expressionin two other models of elastogenesis. Twenty-nine geneswere consistently upregulated or downregulated in all threemodels, suggesting these genes importance in elastogenesis.
Sharon GuffyMedical University of South [email protected]
MS110
ACTIVE: A Bayesian Approach to Asset Covari-ance Estimation
An improved stock return covariance estimation procedureis developed using an implicit Bayesian method known asshrinkage. We extend the work of Ledoit (2003) in covari-ance matrix shrinkage by incorporating the VIX, a marketimplied volatility index, into the estimation process andname our strategy ACTIVE. ACTIVE’s ability to forecastmarket volatility makes it more dynamic by allowing itto respond to changing financial regimes, making it a vi-able alternative to industry standard techniques. Advisors:Marcel Blais, Worcester Polytechnic University and CarlosMorales, Wellington
Daniel HelkeyEmmanuel [email protected]
Tejpal AhluwaliaNew Jersey Institute of [email protected]
Robert HarkMontana [email protected]
Nicholas MarshallClarkson [email protected]
MS111
Multigrid For Divergence-Conforming Discretiza-tions of the Stokes Equations
Recent years have seen renewed interest in the numericalsolution of the Stokes Equations. Divergence-conformingdiscretization methods are of interest because they allowfor the strong solution of the incompressibility condition.Monolithic multigrid methods offer the possibility of solv-
226 CS13 Abstracts
ing the resulting systems with better overall efficiency thanblock-factorization methods. We explore a BDM-baseddiscretization, BDM-P0, and present preliminary numeri-cal experiments using overlapping Vanka smoothers withinGMG, as well as analysis of the method.
James H. AdlerTufts [email protected]
Thomas Benson, Scott MaclachlanDepartment of MathematicsTufts [email protected], [email protected]
MS111
Modeling and Numerical Simulations of ImmiscibleMulti-Phase Fluids
We introduce a multi-phase fluid model derived via an en-ergetic variational approach. We present simulations ofseveral applications of the model, including the dynamicsof a buoyant air bubble or a falling solid in a two-phaseStokes flow. Possible slip effects at the interfaces of thefluids are also considered. Numerical simulations are pre-sented, including results of pressure Schur complement ap-proaches for solving the fluid system and multilevel solversfor solving the pressure system in each iteration of theseschemes.
James BrannickDepartment of MathematicsPennsylvania State [email protected]
MS111
Performance of Efficient AMG Based Precondition-ers for Implicit Resistive MHD
The resistive magnetohydrodynamics model describes thedynamics of charged fluids in the presence of electromag-netic fields. This model is strongly coupled, highly nonlin-ear, and characterized by physical mechanisms that spana wide-range of interacting time scales. Implicit time step-ping is an attractive choice for simulating a wide rangeof physical phenomena. Such methods require an effectivepreconditioner. We explore the performance of several can-didate AMG based preconditioners including fully-coupledAMG projection methods and approximate block factor-ization methods.
Roger PawlowskiMultiphysics Simulation Technologies Dept.Sandia National [email protected]
John ShadidSandia National LaboratoriesAlbuquerque, [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
Luis ChaconLos Alamos National [email protected]
MS111
An Investigation of Multigrid Smoothers for FullyImplicit 2D Incompressible Resistive MHD
We consider a number of different smoothers for use withina multigrid procedure applied to a resistive MHD applica-tion. The smoothers include a generalization of a Braess-Sarazin procedure, a generalization of a Vanka smoother,and the use of operator split (or approximation block fac-torizations) techniques to derive a new smoother. A num-ber of numerical results are presented illustrating the mer-its of the different smoothers in terms of scalability, setup,and cost per iteration.
Ray S. TuminaroSandia National LaboratoriesComputational Mathematics and [email protected]
Tom BensonTufts [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
Scott MaclachlanDepartment of MathematicsTufts [email protected]
MS112
State and Parameter Sensitivities and Estimationin Coupled Ocean/ice Shelf/ice stream Evolution
A challenge in cryospheric research is the quantificationof what controls coupled ice shelf/ice sheet evolution in re-sponse to sub-ice shelf cavity circulation changes and melt-ing. Here we present a modeling infrastructure for simulat-ing the evolution of ice sheet, ice shelf, and sub-ice shelf cir-culation, for inferring high-dimensional fields of state andparameter sensitivities through adjoint model components,and for gradient-based optimal estimation using availableobservations. Examples provided include idealized setupsas well as realistic simulations for Pine Island Ice Shelf,West Antarctica.
Patrick HeimbachMassachusetts Institute of [email protected]
Daniel [email protected]
Martin LoschAlfred Wegener [email protected]
MS112
Hierarchical Error Estimates for Adaptive ShallowIce Sheet and Ice Shelf Models
Mesh adaptation methods are necessary to face the sharpchanges of the ice flow regime in the narrow transitionzones between ice sheets and ice shelves. However, existing
CS13 Abstracts 227
refinement criteria are often empirical. To optimize auto-matically the number and the position of mesh nodes, weimplement a new hierarchical error estimate for the shal-low shelf approximation equation. The efficiency of theresulting re-meshing procedure is shown on the MISMIPexercices.
Guillaume JouvetFree University of BerlinInstitut fur [email protected]
Carsten GraserFreie Universitat BerlinInstitut fur [email protected]
MS112
A Trust Region Stochastic Newton MCMCMethod with Application to Inverse Problems Gov-erned by Ice Sheet Flows
We address the problem of quantifying uncertainty in thesolution of inverse problems governed by Stokes modelsof ice sheet flows within the framework of Bayesian infer-ence. The posterior probability density is explored usinga stochastic Newton MCMC sampling method that em-ploys local Gaussian approximations based on gradientsand Hessians (of the log posterior) as proposal densities.The method is applied to quantify uncertainties in the in-ference of basal boundary conditions for ice sheet models.
Noemi PetraInstitute for Computational Engineering and Sciences(ICES)The University of Texas at [email protected]
James R. MartinUniversity of Texas at AustinInstitute for Computational Engineering and [email protected]
Georg Stadler, Omar GhattasUniversity of Texas at [email protected], [email protected]
MS112
Initialization Strategy for Short Term Projectionsusing the Ice Sheet System Model
Accurate projections of ice sheet contribution to sea levelrise require numerical modeling. However, large-scale mod-eling of ice sheets remains scientifically and technicallychallenging. The Ice Sheet System Model (ISSM) is athree-dimensional thermo-mechanical model that relies onthe finite element method. It includes higher-order andfull-Stokes ice flow approximations, adaptive mesh refine-ment, and data assimilation among other capabilities. Wepresent here the initialization strategy adopted to improveshort-term simulations of ice sheet dynamics.
Helene SeroussiJet Propulsion [email protected]
Mathieu MorlighemUniversity of California - [email protected]
Eric LarourJet Propulsion [email protected]
Eric RignotUniversity of California - IrvineJet Propulsion [email protected]
Ala KhazendarJet Propulsion [email protected]
MS113
Local Time-stepping and Spectral Element Meth-ods for Wave Propagation
We consider the numerical solution of the wave equationon locally refined meshes. To overcome the severe stabilityrestriction of explicit time-stepping methods, due to possi-bly just a few small elements in the mesh, we consider leap-frog based local time-stepping schemes, combined with ahigh-order spectral element spatial discretization. Numer-ical experiments illustrate the efficiency of the proposedmethod.
Loredana GaudioUniversity of [email protected]
Marcus GroteUniversitat [email protected]
MS113
High Efficiency Algorithms for IncompressibleFlows and Moving Geometry
This talk will discuss some new high-order accurate al-gorithms for simulating incompressible flows using over-lapping grids for complex, possibly moving geometry.The approach is based on an approximate-factored com-pact scheme for the momentum equations together witha fourth-order accurate multigrid solver for the pressureequation. The scheme will be described and results will bepresented for some three-dimensional (parallel) computa-tions of flows with moving rigid-bodies.
William D. HenshawCenter for Applied Scientific ComputingLawrence Livermore National [email protected]
MS113
Debye Potentials for the Time Dependent MaxwellEquations
The explicit solution to the scattering problem of time de-pendent Maxwell equations on a sphere is derived. Thederivation of the explicit solution is based on a general-ization of Debye potentials for the time harmonic case andreduces the problem to two scalar wave problems - one withDirichlet condition and the other with Robin condition. Aspectrally accurate discretization scheme is discussed andnumerical examples are presented. The solution can beserved as a reference for checking the accuracy of other nu-merical methods. More importantly, it will provide someinsight towards better integral equation formulations for
228 CS13 Abstracts
the wave equation and time-dependent Maxwell equationsin a general domain.
Leslie GreengardCourant InstituteNew York [email protected]
Thomas HagstromDepartment of MathematicsSouthern Methodist [email protected]
Shidong JiangDepartment of Mathematical SciencesNew Jersey Institute of [email protected]
MS113
Symmetry-preserving High-order Schemes
In this talk we will show how to systematically constructnumerical schemes that preserve the geometrical structureof the underlying PDE. We will then show how to ex-tend these ideas to generate high-order versions of theseschemes.
Jean-Christophe NaveMcGill [email protected]
Alexander [email protected]
MS114
Combining Domain Decomposition with MultigridOptimization for Hierarchical Problems Arisingfrom Nanoporous Material Design
Multigrid optimization algorithms have been shown to im-prove the convergence on nanoporous material design prob-lems, owing to their multiscale nature. However, for morerealistic applications, the large problems corresponding tothe finest resolution levels represent a performance bottle-neck. Well-known domain decomposition techniques canaccelerate the solution of the PDE constraints, i.e. theinner loop of the optimization algorithm. In the presentwork, these ideas are used to introduce parallelism directlyin the multigrid framework.
Julien CortialSandia National Laboratories (Livermore)Quantitative Modeling and [email protected]
Paul BoggsSandia National [email protected]
Stephen G. NashGeorge Mason UniversitySystems Engineering & Operations Research [email protected]
Kevin LongTexas Tech University
David GayAMPL Optimization, [email protected]
R. Michael LewisCollege of William & [email protected]
MS114
Network Heuristics for InitialGuesses to Nanoporous Flow Optimization Prob-lems
Designing channel structures to optimize suitably con-strained flow in nanoporous materials leads to a large-scale,hierarchical, PDE-constrained optimization problem thatinvites solution by multigrid optimization. The problemhas many local solutions, and starting guesses strongly af-fect the solutions found by a descent algorithm. We sketchthe problem and describe a greedy heuristic, possibly im-proved by a Metropolis-Hastings algorithm, for choosingstarting guesses. We show some test results on regular andDelaunay triangulations.
David GayAMPL Optimization, [email protected]
Paul T. BoggsSandia National [email protected]
Robert H. NilsonSandia National Labs (retired)[email protected]
MS114
Preliminary Design of Nanoporous Materials ViaSemidefinite Programming
We discuss the preliminary design of nanoporous materialsto obtain initial designs for detailed optimization. The goalis a material that has specified storage capacity (say, forelectric charge) while allowing sufficiently rapid discharge.Assuming transport is diffusive, standard energy estimatesallow us to pose the problem as a semidefinite program.We discuss the formulation of the problem, its relation tothe fastest mixing Markov chain problem, and present nu-merical results.
Robert M. LewisCollege of William and [email protected]
David PhillipsUnited States Naval [email protected]
MS114
Software for Automating PDE-constrained Opti-mization
Sundance is a package in the Trilinos suite designed toprovide high-level components for the development of high-performance PDE simulators with built-in capabilities for
CS13 Abstracts 229
PDE-constrained optimization. We review the implica-tions of PDE-constrained optimization on simulator designrequirements, then survey the architecture of the Sundanceproblem specification components. These components al-low immediate extension of a forward simulator for use inan optimization context.
Kevin LongTexas Tech [email protected]
Paul BoggsSandia National [email protected]
Bart G. Van Bloemen WaandersSandia National [email protected]
MS115
”Wide or Tall” and ”Sparse Matrix Dense Matrix”Multiplications
This talk explores sparse matrix dense matrix (SMDM)multiplications, useful in block Krylov or block Lanczosmethods. SMDM computations are AU , V A, with Asparse, U with a few columns or V with a few rows. An-other interesting operation is a combined AU , V A, AUVoperation. In a block Lanczos or Krylov algorithm, ma-trix matrix multiplications with the ”tall” U and ”wide”V are also needed, as demonstrated here in computation ofsingular values of sparse matrices. All of these operationscan run significantly faster than BLAS-1 (vector vector)or BLAS-2 (matrix vector) operations, but are not alwayswell implemented in vendor BLAS. Tuning on multi-coreand NUMA architectures is discussed.
Gary W. HowellNorth Carolina State Universitygary [email protected]
Masha SosonkinaAmes Laboratory/DOEIowa State [email protected]
MS115
BLAS Specification Revisited
Certain Blas were not included in the 2002 updated Blasbecause their significance was not appreciated. Examplesinclude the simultaneous application of different Givenstransformations to independent rows and columns and par-allel matrix- matrix multiplication with different matrices.For the indefinite symmetric update operation, the 2002proposal suggested the rarely implemented A = A+XJXT
where J was a symmetric tridiagonal matrix. A simplerupdate that would cover this case would be updating thetriangular portion of A = A + XY T , as proposed by JohnLewis.
Linda [email protected]
MS115
Communication-Avoiding Oblique QR Factoriza-
tions
We study parallel QR factorizations where Q has or-thogonal columns with respect to an oblique inner prod-uct. We present stable, communication-avoiding algo-rithms that cast most of their computations to Level-3BLAS routines. We also provide performance comparisonswith communication-intensive algorithms. For sparse innerproduct, communication-avoiding algorithms show an in-crease in performance while performing more FLOPS thancommunication-intensive algorithms.
Bradley LoweryUniversity of Colorado, [email protected]
Julien LangouUniversity of Colorado [email protected]
MS115
Search Strategies for Empirical Autotuning in Lin-ear Algebra
We present a compiler that translates a linear algebra spec-ification into a parallel and cache-efficient implementation.The compiler achieves high performance through autotun-ing, trying many code forms to find the fastest code forthe architecture. The ability to specify linear algebra op-erations that are larger than those in the BLAS gives usgreater opportunities for optimization. This talk focuseson the challenges of empirical search in autotuning andcompares BTO’s results with optimized BLAS.
Thomas NelsonUniversity of Colorado at BoulderArgonne National [email protected]
Geoffrey BelterDept. of Electrical, Computer, and Energy EngineeringUniversity of Colorado at [email protected]
Elizabeth JessupUniversity of Colorado at BoulderDepartment of Computer [email protected]
Boyana NorrisArgonne National [email protected]
Jeremy SiekDepartment of Electrical and Computer EngineeringUniversity of Colorado at [email protected]
MS116
Automating DEIM using Automatic Differentia-tion
The Discrete Empirical Interpolation method(DEIM) usessnapshots to interpolate a nonlinear vector valued functionby knowing only a few selected components of the function.For DEIM to be practical, the user must be able to evalu-ate these selected components without having to evaluateall of the components of the function. We propose using
230 CS13 Abstracts
the operator overloading approach of automatic differenti-ation to generate these subfunctions. We demonstrate ourapproach on an example involving Miscible Flow.
Russell CardenComputational and Applied MathematicsRice [email protected]
Danny C. SorensenRice [email protected]
MS116
Reduced Basis Methods for Nonlinear DiffusionProblems
We present reduced basis approximations and associateda posteriori error estimation procedures for quadraticallynonlinear diffusion equations. We develop an efficient com-putational procedure for the evaluation of the approxima-tion and bound. The method is thus ideally suited formany-query or real-time applications. Numerical resultsare presented to confirm the rigor, sharpness and fast con-vergence of our approach.
Martin Grepl, Mohammad RastyRWTH Aachen [email protected], [email protected]
MS116
Combining a Data-driven Loewner Approach withH∈ Optimal Interpolation
We present an idea to combine H∈ optimal model order re-duction with a data-driven approach based on Loewner ma-trices. This method can be extended to parametric modelorder reduction by using radial basis function interpolationfor specially defined functions over the parameter domain.We give an estimate on the H∈ error of this method.
Sara GrundelMPI [email protected]
Nils HornungFraunhofer SCAI [email protected]
Peter BennerMax-Planck-Institute forDynamics of Complex Technical [email protected]
MS116
Model Reduction using Snapshot based Realiza-tions
Abstract not available at time of publication.
Dick LuchtenburgPrinceton [email protected]
MS117
A Multiresolution Adjoint Sensitivity Analysis of
Time-lapse Seismic Data
We present a method to speed up the computation of ad-joint sensitivities during the estimation of reservoir param-eters from time-lapse seismic data. The seismic data, inform of change in saturation over time, is first transformedinto wavelets and a small subset of the wavelets is retainedfor parameter estimation. The method reveals how to min-imize computational time by modifying the standard ad-joint equation to compute directly the adjoint sensitivitiesof only the wavelets retained.
Abeeb AwotundeKing Fahd University of Petroleum & [email protected]
MS117
Quantifying the Effect of Observations in 4D-VarData Assimilation
Data assimilation combines information from an imperfectnumerical model, noisy observations and error statistics toproduce a best estimate of the state of a physical system.The contribution of individual observations in reducing theerror can provide useful insight for pruning redundant mea-surements and designing sensor networks. The followingpresentation describes a novel systematic approach to ac-complish this in the context of 4D-Var data assimilation,with emphasis put on the computational challenges.
Alexandru CioacaVirginia [email protected]
Adrian SanduVirginia Polytechnic Institute andState [email protected]
MS117
Conjugate Unscented Transform based Approachfor Uncertainty Characterization and Data Assim-ilation
This talk will focus on recent development of mathematicaland algorithmic fundamentals for uncertainty characteriza-tion, forecasting, and data assimilation for nonlinear sys-tems. Recently developed Conjugate Unscented Transfor-mation (CUT) methodology will be presented to computemulti-dimensional expectation integrals efficiently. CUTmethodology offers the systematic development of non-product cubature rules for accurately evaluating multi di-mensional expectation integrals with respect to a symmet-ric pdf. By accurately characterizing the uncertainty as-sociated with both process and measurement models, thiswork offers systematic design of low-complexity data assim-ilation algorithms with significant improvement in nominalperformance and computational effort. The applicabilityand feasibility of these new ideas will be demonstrated onbenchmark problems and some real world problems such astracking resident space objects and forecasting toxic plume.
Nagavenkat AdurthiMAE DeaprtmentUniversity at [email protected]
Puneet Singla
CS13 Abstracts 231
Mechanical & Aerospace EngineeringUniversity at [email protected]
Tarunraj SinghDept of Mechanical EngineeringSUNY at [email protected]
Abani K. PatraSUNY at BuffaloDept of Mechanical [email protected]
MS117
Variational Data Assimilation of Chaotic Dynami-cal Systems over Climate Timescales
In this talk, we first demonstrate difficulties encounteredby conventional data assimilation methods, when the datais from observation of a chaotic dynamical system over cli-mate timescales. We then introduce a new method, theLeast Squares Sensitivity Analysis method, that overcomesthe difficulties of climate data assimilation. The LeastSquares Sensitivity method formulates an adjoint problemthat aims at computing derivatives of statistics of a chaoticdynamical system, which is used in our variational data as-similation. Our new method is demonstrated on the Lorenz63 system.
Qiqi WangMassachusetts Institute of [email protected]
MS118
Statistical Perspectives
Abstract not available at time of publication.
George CasellaDepartment of StatisticsUniversity of [email protected]
MS118
Spectral/hp Element and Discontinuous GalerkinMethods for Response-Excitation PDF Equations
Evolution equations of the joint response-excitation prob-ability density function (REPDF) generalize the existingPDF evolution equations and enable us to compute thePDF of the solution of stochastic systems driven by coloredrandom noise. This talk presents an efficient numericalmethod for this evolution equation of REPDF by consider-ing the response and excitation spaces separately. For theresponse space, a non-conforming adaptive discontinuousGalerkin method is used to resolve both local and discon-tinuous dynamics while a probabilistic collocation methodis used for the excitation space. We propose two funda-mentally different adaptive schemes for the response spaceusing either the local variance combined with the boundaryflux difference or using particle trajectories. The effective-ness of the proposed new algorithm is demonstrated in twoprototype applications dealing with randomly forced non-linear oscillators. The resulted PDF is compared againstthe one obtained from kernel density estimation based onMonte Carlo simulation and the solution of the effectiveFokker-Planck equation. The framework we develop here is
general and can be readily extended to first order stochas-tic PDEs subject to random boundary conditions, randominitial conditions, or random forcing terms.
Heyrim ChoBrown UniversityProvidence, RIheyrim [email protected]
Daniele VenturiBrown Universitydaniele [email protected]
George E. KarniadakisBrown UniversityDivision of Applied Mathematicsgeorge [email protected]
MS118
UQ and High-Performance Computing
Abstract not available at time of publication.
James R. MartinUniversity of Texas at AustinInstitute for Computational Engineering and [email protected]
MS118
Hierarchical Preconditioners for the StochasticGalerkin Finite Element Methods
Use of the stochastic Galerkin finite element methods leadsto large systems of linear equations. These systems are typ-ically solved iteratively. We propose a family of precondi-tioners that take an advantage of (the recursion in) thehierarchy of the global system matrices. Neither the globalmatrix, nor the preconditioner need to be formed explicitly.The ingredients include only the number of stiffness matri-ces from the truncated Karhunen-Loeve expansion and apreconditioner for the mean-value problem. The perfor-mance is illustrated by numerical experiments.
Bedrich Sousedik, Roger G. GhanemUniversity of Southern [email protected], [email protected]
ERIC T. PhippsSandia National [email protected]
MS119
Error Estimate for Nonlinear Model Reduction us-ing Discrete Empirical Interpolations
This work provides an error analysis of parametric nonlin-ear model reduction using Discrete Empirical InterpolationMethod together with standard projection-based model re-duction methods, such as Proper Orthogonal Decomposi-tion. The analysis will be given in the setting of ODE sys-tems arising from spatial discretizations of parabolic PDEsand in the setting of discretized steady-state problems. Theconditions under which the stability of the original systemis preserved and the reduction error is uniformly boundedwill be discussed. Some numerical examples will be usedto illustrate the applicability of this error analysis.
Saifon Chaturantabut
232 CS13 Abstracts
Virginia [email protected]
Danny C. SorensenRice [email protected]
MS119
Error Estimation for Nonlinear Reduced BasisMethods based on Empirical Operator Interpola-tion
Many applications based on numerical simulations of PDEsdepend on expensive computations and therefore use low-dimensional surrogate models. Here, it is important to con-trol the error of the surrogate model. If the surrogate isobtained with the reduced basis method, efficient operatorevaluations are required for nonlinear problems. These canbe obtained through the empirical operator interpolationmethod. We present error estimators for both the interpo-lated operators and reduced basis models for parametrizednonlinear evolution equations.
Martin DrohmannUniversity of MunsterInstitute for Computational and Applied [email protected]
Bernard HaasdonkUniversity of [email protected]
Mario OhlbergerUniversitat MunsterInstitut fur Numerische und Angewandte [email protected]
MS119
Certified Reduced Basis Approximation forComponent-Based Problems
We discuss the SCRBE Method [Huynh et.al., A StaticCondensation Reduced Basis Element Method, M2AN, ac-cepted 2012] for large component-based (e.g. 3D structural)problems. We combine RB model reduction for the high-fidelity component-local bubble spaces with a “port reduc-tion” procedure [Eftang et.al., Adaptive Port Reduction inStatic Condensation, Proceedings of MATHMOD 2012] toreduce the Schur system size.
The additional port reduction procedure is necessary inparticular in 3D when the number of port degrees of free-dom is large. The errors introduced by both the RB ap-proximations and port reduction procedure can be rigor-ously bounded with respect to an underlying FE approxi-mation.
Jens EftangMITDepartment of Mechanical [email protected]
MS119
Error Estimates for Some Galerkin Reduced-orderModels of the Semi-discrete Wave Equation
Reduced-order models for first-order dynamical systemshave been extensively analyzed. Even though the semi-
discrete wave equation can be formulated as a first-orderdynamical system, model-order reduction may destroy theoriginal structure of the wave equation. This talk willpresent an error analysis for Galerkin-based reduced sys-tems preserving the structure of the wave equation. Errorbounds are derived in the continuous setting and when theNewmark scheme is used. Numerical experiments illustratethe theoretical results.
David AmsallemStanford [email protected]
Ulrich HetmaniukUniversity of WashingtonDepartment of Applied [email protected]
MS120
The Effective Combination of Mesh Adaptationand Non-linear Thermo-mechanical Solution Com-ponents for the Modeling of Weld Failures
Computational studies of the transient failure of laser weldsrequire maintaining a high quality mesh for use in a fullycoupled thermo-mechanical, finite deformation, simulationof the necking and subsequent unloading of the specimen.Large deformations encountered during necking necessitatemethods to locally adapt the mesh. The method used toeffectively combine FASTMaths MeshAdapt and Sandia’sAlbany thermo-mechanical solution framework as well asits application in the parallel simulation of weld failureswill be presented.
Glen HansenSandia National [email protected]
Ryan MoleckeRensselaer Polytechnic [email protected]
James Foulk IIISandia National [email protected]
Mark ShephardRensselaer Polytechnic [email protected]
MS120
Parallel Interface Preservation in Mesquite forBubble-Shock Interaction Problems
We present a parallel material interface preservationscheme developed in the context of the Mesquite meshquality improvement toolkit. This scheme allows themovement and tracking of interior and exterior bound-aries and significantly improves the robustness of an ALEshock/bubble interaction test problem over standard tech-niques.
Brian MillerLawrence Livermore National [email protected]
MS120
Advances of an Unstructured Mesh Infrastructure
CS13 Abstracts 233
to Support Massively Parallel Adaptive Simula-tions on High Core Count Machines
The flexible distributed Mesh Database (FMDB) is FAST-Math iMesh/iMeshP compliant unstructured mesh infras-tructure to represent and manipulate mesh data for par-allel adaptive simulations. FMDB has been extended tofully support multiple parts per process, faster migration,and full interaction with ZOLTAN partitioning. RecentFMDB developments include making it architecture-awareand structural changes to the partition model to supporta partition taking advantage of node level shared memoryand threading, while maintaining message passing betweennodes.
Seegyoung SeolRensselaer Polytechnic InstituteScientific Computation Research [email protected]
Daniel [email protected]
Mark ShephardRensselaer Polytechnic [email protected]
MS120
Interoperable Solution Transfer Tool for CoupledMulti-physics Simulations
I will describe a MOAB-based tool for solution transferthat simplifies connection of physics codes and supportsparallel coupled multi-physics simulation; examples will in-clude problems in Nuclear Energy (coupling Nek5000 andUNIC) and reservoir simulation (coupled reservoir & seis-mic codes)
Timothy J. TautgesArgonne National [email protected]
MS121
Accurate Solutions to Scattering Problems in Non-smooth Planar Domains
Recursively Compressed Inverse Preconditioning (RCIP) isa method for obtaining accurate solutions to integral equa-tions stemming from elliptic PDEs in piecewise smooth do-mains. Tractable features include: corners, cusps, multiplejunctions, small edge separation, mixed boundary condi-tions, barely integrable solutions, and composed integraloperators. In a postprocessor, the solution to the PDE isrecovered in the entire domain using high-order analyticproduct rules close to the boundary. A range of numericalexamples is presented.
Johan HelsingCentre for Mathematical SciencesLund [email protected]
MS121
Fast Direct Solvers for the Lippmann-SchwingerEquation
Recent years have seen the development of a number of
fast direct solvers for integral equations that are extremelyeffective for systems involving many right-hand sides.Here, we adapt a recursive skeletonization-based solver tothe Lippmann-Schwinger equation for high-frequency wavescattering in 2D. For a domain discretized with N ele-ments, the algorithm incurs an O(N�/∈) precomputationcost, after which each incident wave can be analyzed inonly O(N log N ) operations.
Kenneth L. HoCourant Institute of Mathematical SciencesNew York [email protected]
MS121
Robust Charge-current Formulations for Electro-magnetic Scattering
In the electromagnetics literature, significant attentionhas been paid to the integral equation formulation ofscattering problems. Certain problems as low frequencybreakdown, high-density mesh breakdown, resonances, andcatastrophic cancellation arise when dealing with the mostcommon integral equations as MFIE, EFIE, CFIE... Inthis paper we present some new formulations that avoidmost of the problems above mentioned. The formulationsproposed are based on second kind integral equations andcurrent and charge representation for the sources.
Felipe VicoDepartamento de ComunicacionesUniversidad Politecnica de [email protected]
Leslie GreengardCourant Institute of Mathematical SciencesNew York [email protected]
Zydrunas GimbutasCourant InstituteNew York [email protected]
Miguel Ferrando BatallerDepartamento de ComunicacionesUniversidad Politecnica de [email protected]
MS121
High Frequency FMMs and Improved WKB Ap-proximations
WKB approximation is ubiquitous in several areas of sci-ence and engineering. It approximates solutions of sec-ond order ODEs by a simple formula; a major nuisanceis that it is only asymptotic, as opposed to an exact ex-pression.It turns out that such exact expressions can beconstructed via the classical Kummer’s equation. Amongother things, it leads to approximations of special func-tions in the Fresnel regime, with obvious applications infast multipole methods, etc.
Bogdan G. VioreanuApplied MathematicsYale [email protected]
234 CS13 Abstracts
Vladimir RokhlinYale [email protected]
MS122
Full Waveform Inversion for Diffraction Focusing
Seismic diffractions are an important part of seismic re-flection data. Diffractions represent the direct responseof small objects that are often the subject of geophysi-cal exploration: faults, cavities, channels, etc. By sepa-rating diffractions from specular reflections, it is possibleto formulate the seismic inverse problem as the problemof focusing diffraction energy in the imaging domain. Wedemonstrate the connection between this formulation andthe conventional full waveform inversion and highlight dif-ferences and similarities.
Sergey Fomel, Siwei LiUniversity of Texas at [email protected], [email protected]
MS122
Title Not Available at Time of Publication
Abstract not available at time of publication.
Loukas F. KallivokasThe University of Texas at [email protected]
MS122
On Parameterization of Full Waveform Inversion inExploration Geophysics
Over the last five years, acoustic full waveform inversionhas been applied to large surface seismic data sets to imagethe first kilometers of the earth crust. The problem is how-ever severely ill-posed, non-linear and expensive. Certainparameterizations, that may be case dependent, shouldhelp to reduce the number of iterations and some artifacts.During this talk, we will discuss some possible parameter-izations or pre-conditioning and their challenges and showsome examples.
Rene-Edouard PlessixShell Global Solutions [email protected]
MS122
Relaxation of Nonlinear Full Waveform Inversionvia Extended Modeling
Least squares data fitting has become known as ”full wave-form inversion” in computational seismology, and is thesubject of very active research in industry and academiaover the last couple of years. The objective function of fullwaveform inversion tends to be highly multimodal, espe-cially for data rich in reflected wave energy. Reformulationvia extended modeling and soft physicality constraints hasproven able to convexify this difficult optimization prob-lem, at least to some extent: for example, the Marmousibenchmark problem has recently yielded to extended in-version. This presentation will review the theoretical basisfor extended inversion, and outline some of the remainingcomputational challenges.
William Symes
Rice [email protected]
MS123
Competitive Product Differentiation in a ThreeLevel Market
Semiconductor chips are not sold directly to the end cus-tomer but to Original Equipment Manufacturers (OEM)who assemble a computer and sell to the end market. TheOEM acts as a filter for the customer demand. Select-ing the optimal number and the optimal spacing betweenproducts in price and performance is known as the verticaldifferentiation problem. The interaction between this op-timization problem and the game theoretic competition inthis three level market is analyzed.
Dieter ArmbrusterArizona State UniversityDepartment of [email protected]
Tulin InkayaArizona State [email protected]
Karl KempfIntel [email protected]
Hongmin LiArizona State [email protected]
Morgan DempseyIntel [email protected]
MS123
Interactions between Product Development andManufacturing Planning within INTEL
Abstract not available at time of publication.
Karl KempfIntel [email protected]
MS123
Process and Production Optimization for Indus-trial Applications
ABB Corporate Research tackles real world problems aris-ing in industry applying methods from mathematical opti-mization and optimal control. In this talk, we will presenttwo research projects: Energy-aware enterprise-wide pro-duction scheduling in the metals industry and a solutionfor plant-wide asset-performance management. The un-derlying problems can be solved by hybrid methods com-bining mathematical optimization with production intelli-gence and using asset performance information (diagnosis)to generate operations and maintenance actions (therapy)in steel and paper mills.
Sleman SalibaABB AG Corporate Research [email protected]
CS13 Abstracts 235
Iiro HarjunkoskiABB AG Corporate Research [email protected]
MS123
Production Planning with Nonlinear ClearingFunctions: A Review of Recent Results
Nonlinear clearing functions that represent the relationshipbetween the expected throughput of a planning resource ina planning period and the workload of the resource haveshown promising results in production planning. We re-view basic concepts and computational results, focusing ona study of a large semiconductor wafer fabrication facility.
Reha [email protected]
Lars MoenchFernuniversitat [email protected]
N. Baris KacarSAS [email protected]
MS124
Large Eddy Simulation of Pathological and MedicalDevice Hemodynamics using a Novel MultiblockImmersed Boundary Method Approach
Large eddy simulations of several flows involving patho-logical or medical device hemodynamics will be presented.A high-order finite-difference method is used to integratethe incompressible Navier-Stokes equations on a structuredCartesian grid. Complex geometries are handled using anovel multiblock immersed boundary method. Results arecompared to experimental measurements to assess accu-racy and gain insight into these flowfields.
K Anupindi, Y Delorme, D Shetty, Steven H. FrankelSchool of Mechanical EngineeringPurdue [email protected], [email protected],[email protected], [email protected]
MS124
Multiscale Modeling and Optimization in SingleVentricle Heart Disease
Single ventricle patients typically undergo three pallia-tive surgical procedures before the age of three. We willpresent a framework for multiscale blood flow simulationsand surgery optimization. We will discuss challenges ofmodeling the coupled system of blood flow and circulatorydynamics that arises in these patients, and how simulationsmay impact their clinical care. We present an efficient andmodular method to couple our custom finite element solverto a closed loop circulatory model.
Alison MarsdenDepartment of Mechanical and Aerospace EngineeringUniversity of California, San [email protected]
MS124
Modeling of Blood Flow in Normal and DiseasedLeft-ventricles
The current research focuses on the modeling and analysisof flow in normal and diseased left-ventricles; disease condi-tions include myocardial infarction, diastolic dysfunction,and hypertrophic obstructive cardiomyopathy. Anatomicalmodels of the left ventricles are derived from high resolu-tion CT as well as echo images, and the simulations employan immersed boundary flow solver. The effect of flow pat-terns on the performance of the left ventricle is explored indetail.
Vijay Vedula, Jung Hee SeoMechanical EngineeringJohns Hopkins [email protected], [email protected]
Albert Lardo, Theodore AbrahamMedical SchoolJohns Hopkins [email protected], [email protected]
Rajat MittalDepartment of Mechanical EngineeringThe Johns Hopkins [email protected]
MS124
Numerical Study of Pulse Wave Propagation Pat-terns in Stiffened Arteries
A 2-D axis-symmetric model of aortic pulse-wave-travelingis simulated using our modified Immersed Finite ElementMethod (mIFEM), a numerical method that accuratelysimulates the dynamics of fully-coupled fluid-structure in-teractions. This model is to validate pulse wave imagingmeasurement, a noninvasive and quantitative way of mea-suring arterial stiffness. In our model, pulse wave velocity(PWV) is measured by calculating the spatiotemporal vari-ation of the pulse wave-induced displacement of the arterialwall. Boundary conditions and material properties are tofollow the experiments. The results are in good agreementswith PWV measured from aortic phantoms.
Lucy Zhang, Jubiao YangDepartment of Mechanical, Aerospace and NuclearEngineeringRensselaer Polytechnic [email protected], [email protected]
Danial ShahmirzadiDepartment of Biomedical EngineeringColumbia [email protected]
Elisa KonofagouColumbia [email protected]
MS125
Data Analysis of Cascading Power System Failures
We present analyses of large-scale simulations of extremeevents in power grids, in particular (a) cascading failuresand (b) random fluctuations of renewable power injec-tions. In the case of cascading failures we use models that
236 CS13 Abstracts
capture some of the short-term grid dynamics (includingvoltage and frequency deviations) in approximate form,as well as generator ramp-up behavior and frequency re-sponse (broadly interpreted). Several interesting phenom-ena can be observed, in particular dominant modes of cas-cading, and observable features in the early stages of thecascade. In the case of fluctuating renewable injections weconsider various conditional distributions of wind power(conditional on short-term forecasts) and analyze resultingPDFs of line overloads.
Daniel BienstockColumbia University IEOR and APAM DepartmentsIEOR [email protected]
MS125
Scalable Stochastic Optimization for Power GridSystems
Stochastic optimization of complex energy systems re-sults in extremely large optimization problems that canbe solved only by means of high-performance computers.We present the latest algorithmic developments and imple-mentations for the scalable solution of continuous and in-teger stochastic programming problems with recourse. Wealso discuss the numerical results obtained on Argonne’s”Intrepid” BG/P platform.
Cosmin G. Petra, Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected], [email protected]
MS125
Economic Dispatch with Renewable Power Gener-ation
A two-stage non-linear stochastic formulation for the eco-nomic dispatch problem under renewable-generation un-certainty is investigated. Certain generation decisions aremade only in the first stage and fixed for the second stage,where the actual renewable generation is realized. The un-certainty in renewable output is captured by a finite num-ber of scenarios. Any resulting supply-demand mis-matchmust then be alleviated using high marginal-cost powersources. We present two decomposition algorithms to solvethis problem to optimality.
Dzung PhanIBM T.J. Watson Research [email protected]
MS125
Scalable, Parallel Stochastic Unit Commitment ForReliability Operations
Modern grid reliability systems include a security-constrained unit commitment (SCUC) optimization algo-rithm operating in tandem with a security-constrained eco-nomic dispatch (SCED) optimization algorithm. Indepen-dent system operators (ISOs) use these models to commitand dispatch generation resources. Increasing penetrationof intermittent renewable generation necessitates a shift tostochastic SCUC/ SCED algorithms. We will describe thestochastic programming SCUC/SCED formulations and al-gorithms we are currently developing under an effort re-
cently funded by the ARPA-e GENI program.
Jean-Paul WatsonSandia National LaboratoriesDiscrete Math and Complex [email protected]
MS126
Efficient Symmetric Positive Definite Second-orderAccurate Monolithic Solver for Fluid/solid Interac-tions
We introduce a second-order accurate method to simu-late strongly coupled (monolithic) fluid/rigid-body inter-actions. We take a fractional step approach, where theintermediate state variables of the fluid and of the solidare solved independently, before their interactions are en-forced via a projection step. The projection step producesa symmetric positive definite linear system that can be effi-ciently solved using the preconditioned conjugate gradientmethod. Numerical results indicate that the method issecond-order accurate in the maximum norm and demon-strate that its solutions agree quantitatively with experi-mental results.
Frederic G. GibouUC Santa [email protected]
Chohong MinEwha Womans University, [email protected]
MS126
An Adaptive Multi-material Moment-of-fluid Method for ComputingMulti-phase Flows
We combine the multimaterial Moment-of-Fluid (MOF)work of Ahn and Shashkov with the work of Kwatra et alfor removing the acoustic time step restriction in order tosolve multimaterial flows in which each material might becompressible or incompressible. The mass weights found inthe algorithm of Kwatra et al are computed directly fromthe multimaterial MOF reconstructed interface. Simula-tions for multimaterial flows are presented with applica-tions to combustion (atomization and spray) and microflu-idics.
Mark SussmanDepartment of MathematicsFlorida State [email protected]
Matt Jemison, Michael RoperFlorida State [email protected], [email protected]
Mikhail ShashkovLos Alamos National [email protected]
Marco ArientiSandia National Labs, Livermore [email protected]
Xiaoyi LiUTRC
CS13 Abstracts 237
MS126
A Sharp Level-set Approach for the DendriticGrowth
We developped a sharp level-set approach for the den-dritic growth in two spatial dimensions, where the inher-ent multi-scale nature of the physical problem is handleby the use of adaptive quadtree grids. The obtained nu-merical results show an excellent match with experimen-tal observations and theoritical predictions. In particularthe dendrites morphology (primary and secondary spacing)and the transition between the different regime (palanar-cellular-dendritic) were accurately captured.
Maxime TheillardDept. Mechanical EngineeringUCSBmaxime [email protected]
MS126
Computations of Mass Transfer in Bubbly Flowsusing a Hybrid Finite Volume Method and an Em-bedded Analytical Description
DNS of mass transfer from hundreds of bubbles in tur-bulent flows introduces resolution requirements that arecurrently difficult to match. Here we introduce an em-bedded analytical description for mass transfer in the thinboundary layer at bubble surfaces and couple it with afinite volume method for the rest of the flow. Compar-isons with fully resolved simulations of simple systems andexperiments shows good agreement, for relatively modestincrease in the computational effort.
Gretar TryggvasonUniversity of Notre DameDepartment of Aerospace and Mechanical [email protected]
Bahman AboulhasanzadehDepartment of Aerospace and Mechanical EngineeringUniversity of Notre [email protected]
Jiacai LuMechanical Engineering DepartmentWorcester Polytechnic [email protected]
MS127
A High-order Immersed Finite ElementMethod forPDEs with Discontinuous Coefficients
We present an immersed finite element for interface prob-lem with discontinuous coefficients. Here we allow the in-terface to cut elements and construct immersed finite ele-ment shape functions that satisfy jump conditions acrossthe interface. The immersed finite element spaces are usedwith a penalized weak formulation. We discuss the exis-tence and approximation properties of these finite elementspaces and solve several interface problems to show theperformance of the proposed method.
Slimane AdjeridDepartment of MathematicsVirginia Tech
Mohamed [email protected]
Tao LinVirginia [email protected]
MS127
Low-complexity Application of Finite Element Op-erators on Simplices via Bernstein Polynomials
Evaluation and differentiation of Bernstein-form polyno-mials on the simplex may be accomplished with low com-plexity, and so may computation of integral moments ofdata against the Bernstein basis. I will demonstrate thesefeatures via a recursive decomposition of an appropriategeneralized Vandermonde-type matrix. H(div) and H(curl)are similarly handled by first converting the exterior cal-culus bases of Arnold, Falk, and Winther to the Bernsteinbasis efficiently and then using the H1 techniques. I willsketch these techniques, and also comment on shared mem-ory parallelism of the resulting algorithms.
Robert C. KirbyTexas Tech UniversityRobert [email protected]
MS127
Error Analysis of the Discrete Wave EquationBased on a Full-Modal Decomposition: A Unify-ing Perspective for Higher-order Methods
Abstract not available at time of publication.
Pedro MoyKing Abdullah University of Science and [email protected]
Victor M. CaloApplied Mathematics and Computational ScienceKing Abdullah University of Science and [email protected]
MS127
Discontinuous Galerkin Method for HyperbolicEquations Involving δ-functions
In this talk, we apply discontinuous Galerkin (DG) meth-ods to solve hyperbolic equations involving δ-functions. Ingeneral, the numerical solutions are highly oscillatory nearthe singularities, which we refer to as the pollution region.We first analyze the size of the pollution region and the rateof convergence outside for some model equations, then ap-ply the method to pressureless Euler equations to show thegood performance of the DG method.
Yang YangBrown universityyang [email protected]
Chi-Wang ShuBrown UniversityDiv of Applied [email protected]
238 CS13 Abstracts
MS128
A Practically Painless Path to Petascale Paral-lelism: PETSc + Python
I describe the design of numerical software that is oper-ated with the convenience of MATLAB yet achieves effi-ciency near that of hand-coded Fortran and scales to thelargest supercomputers. Python is used for most of thecode while automatically-wrapped Fortran kernels are usedfor computationally intensive routines. For parallelism weuse PETSc via petsc4py. The software described here isPyClaw, a structured grid solver for systems of hyperbolicPDEs based on Clawpack.
David I. KetchesonMathematical and Computer Sciences & EngineeringKing Abdullah University of Science & [email protected]
MS128
Constructing and Configuring Nested and Hierar-chical Solvers in PETSc
We look at code architecture for constructing and also con-figuring deeply nested solver hierarchies. We are most con-cerned with
• how much code a user has to write
• how frequently code has to change
• how an implementation chooses among options
and the impact tradeoffs here have on the interface design.Our examples will be drawn from multiphysics PDE sim-ulation, and exhibit block solvers, algebraic and geometricmultigrid, and composition of both linear and nonlinearsolvers. However, this is not an overview of packages ca-pabilities, but rather a discussion about implementationstrategy and user experience.
Matthew G. KnepleyUniversity of [email protected]
Jed BrownMathematics and Computer Science DivisionArgonne National [email protected]
Barry F. SmithArgonne National LabMCS [email protected]
MS128
The FEniCS Project: Organization, Practices,Maintenance and Distribution
Abstract not available at time of publication.
Anders LoggSimula Research [email protected]
MS128
Joining Forces: Combining FEniCS and DUNE us-ing a High-level Form Language and Code Gener-
ation
Abstract not available at time of publication.
Steffen MuthingInstitute of Parallel and Distributed SystemsUniversity of [email protected]
MS129
Building An Applied and Computational Math De-gree Program from the Ground Up
Abstract not available at time of publication.
Jeffrey HumpherysBrigham Young [email protected]
MS129
Interdisciplinary Undergraduate Research in Com-putational Sciences as a Model for InstitutionalChange
Abstract not available at time of publication.
Padmanabhan SeshaiyerGeorge Mason [email protected]
MS129
Enabling Excel-lent Experiences: ComputationalScience Internships
Professors can help undergraduates excel through obtain-ing meaningful CSE research internships. Such internshipscan enhance students’ professional and personal lives andcan add new dimensions to a school’s program. Withstudents’ inexperience, active involvement by professors isusually a crucial. Using numerous ”success stories,” thistalk explores how advisors can help students obtain andbenefit from computational research experiences.
Angela B. ShifletMcCalla Professor of Math. & CS, Dir. of ComputationalSci.Wofford [email protected]
MS129
Modeling Across the Curriculum
The August, 2012 SIAM-NSF Workshop on Modelingacross the Curriculum will be described, and an introduc-tion to the report presented. The meeting proposal pre-dated but responded to the PCAST ”Engage to Excel”report’s call for one million new STEM graduates by 2020.Three themes were addressed: coordinated STEM curricu-lum content in K-12; undergraduate curricula in modelingand CSE as the heart of STEM; and readiness for collegeSTEM education. The last theme is a basis for assessmentof programs and how they address the ”math gap.”
Peter R. TurnerClarkson UniversitySchool of Arts & [email protected]
CS13 Abstracts 239
MS130
High-Order Multi-Stage Lax-Wendroff Time Inte-grators with Applications for Cahn-Hilliard
In this work, we present a novel class of high-order nu-merical integrators. These integrators utilize a combina-tion of Lax-Wendroff time-steps and Hermite interpolation.Given their low-storage requirements and high-order ac-curacy, this class of methods proves to be promising forhigh performance computing. We explore applications ofexplicit and implicit formulations on ordinary and partialdifferential equations. The implicit scheme is a low-storage,4th order, A-stable method, and the 4th order explicit for-mulation serves as an accurate error predictor for adaptivetime stepping on stiff PDEs. Results for the Cahn-Hilliardequations are investigated. The method is extends to 6thorder and higher.
Andrew ChristliebMichigan State [email protected]
David C. SealDepartment of MathematicsMichigan State [email protected]
Jaylan S. JonesMichigan State [email protected]
MS130
High Order Partitioned Time Stepping Methodsfor Stiff Problems
We discuss recent developments for multistage partitionedtime stepping methods for solving ODEs and PDEs. Weanalyze stability and consistency properties of different op-erator splitting strategies driven by stiff and nonstiff termsthat require integrators with different properties for stabil-ity and efficiency reasons. We focus on multistage additiveimplicit-explicit (IMEX) and Rosenbrock W schemes.
Emil M. ConstantinescuArgonne National LaboratoryMathematics and Computer Science [email protected]
MS130
Conditions for Positivity and SSP Property of Lin-early Implicit Methods and Applications to CFD
Positivity preservation and the diminishing of some con-vex functionals along the solution in time are features ofmany classes of initial (or initial-boundary) value problems,which ideally are maintained by their numerical discretiza-tion. These requirements can be fulfilled only under someconditions on the parameters of the discretization proce-dure. In this talk we investigate such conditions for lin-early implicit methods, in particular when applied to someproblems in CFD.
Zoltan HorvathSzechenyi Istvan [email protected]
Jed BrownMathematics and Computer Science DivisionArgonne National Laboratory
Matthew G. KnepleyUniversity of [email protected]
Tihamer A. Kocsis, Adrian NemethSzechenyi Istvan [email protected], [email protected]
MS130
A Comparison of High Order Explicit Runge-Kutta, Extrapolation and Integral Deferred Cor-rection Methods
Two popular approaches for solving ODEs with very highaccuracy are extrapolation and deferred correction meth-ods. When the base method on which they are built isa one-step method (i.e., a Runge-Kutta method), bothmethods result in a Runge-Kutta method with very manystages. This connection between the two methods is usedto study and investigate stability and accuracy propertiesof the two ODE solving schemes along with high order em-bedded Runge-Kutta pairs. These methods are comparedin practice by implementing them in an adaptive error con-trol framework and a small set of test problems, including athree-body chaotic problem, are used to derive conclusions.
Umair bin WaheedKing Abdullah University of Science and [email protected]
David I. KetchesonMathematical and Computer Sciences & EngineeringKing Abdullah University of Science & [email protected]
MS131
Development of Exact-to-Precision GeneralizedPerturbation Theory in Nuclear Engineering Cal-culations
Perturbation theory provides the most computationally ef-ficient approach for calculating responses variations result-ing from parameters variations without the need to re-execute the model. In this talk, we introduce EpGPT(Exact-to-Precision Generalized Perturbation Theory) in-tended to address the challenges of existing theory, includ-ing ability to calculate higher order variations and boundthe variational errors, handle models with many responses.
Hany S. Abdel-KhalikNorth Carolina State [email protected]
MS131
Utilizing Adjoints to Improve Propagation of Un-certainties through Surrogate Response Surfaces
We consider the use of surrogate response surfaces, some-times called emulators, to cheaply propagate probabilitydistributions between an input parameter space and ob-servable quantities of interest. We develop a frameworkfor both forward and inverse propagation and utilize ad-joints to improve the pointwise accuracy of the emulatorsand subsequently of the computed cumulative distribution
240 CS13 Abstracts
functions. An error analysis is provided for concrete ex-amples using polynomial chaos expansions to define thesurrogates.
Troy ButlerColorado State [email protected]
Clint DawsonInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
Tim WildeySandia National [email protected]
MS131
On an Adjoint Consistent Formulation for a Cou-pled Problem
Microfluidic devices are usually simulated using coupledmodels due to the multiscale and multiphysics nature of mi-croflows. In the Helmholtz slip electroosmotic model, cou-pling between physical models is enforced on the boundaryof the domain through a slip boundary condition. How-ever, analysis of this coupled formulation shows that theresulting forward problem may be ill-posed. We propose anew formulation of the coupling term to ensure that boththe forward and adjoint problems are well-posed.
Serge PrudhommeICESThe University of Texas at [email protected]
Vikram GargMassachusetts Institute of [email protected]
Kris van der ZeeMultiscale Engineering Fluid DynamicsEindhoven University of [email protected]
MS132
Fast Direct Solvers for H2 Matrices
Abstract not available at time of publication.
Eric F. DarveStanford UniversityMechanical Engineering [email protected]
Amirhossein AminfarStanford [email protected]
MS132
Fast Multipole Method as a Preconditioner
Recent efforts to view the FMM as an elliptic PDE solverhave opened the possibility to use it as a preconditionerfor even a broader range of applications. Use of FMMas a preconditioner requires optimization for low accuracy,which is a challenge on many-core co-processors and GPUs.This talk addresses various optimization techniques that
combine treecodes and FMMs to achieve a highly parallelFMM preconditioner.
Huda Ibeid, Rio YokotaKing Abdullah University of Science and [email protected], [email protected]
David E. [email protected]
MS132
A Treecode-accelerated Boundary IntegralPoisson-Boltzmann Solver
We present a boundary integral method for electrostaticsof solvated biomolecules described by the linear Poisson-Boltzmann equation. The method employs GMRES itera-tion and a Cartesian treecode which reduces the cost fromO(N2) to O(N log N), where N is the number of faces inthe triangulated molecular surface. We present results forthe Kirkwood sphere and a protein.
Robert KrasnyUniversity of MichiganDepartment of [email protected]
Weihua GengUniversity of [email protected]
MS132
Directional FMM for Maxwell’s Equations
For boundary element methods in electromagnetics, thereare a variety of fast algorithms used to accelerate thematrix-vector product with the impedance matrix. In thistalk, we propose a different approach based on the direc-tional multilevel algorithm by Engquist and Ying. Usingthe directional low-rank property of the Green’s functionin the high-frequency regime, it is shown that the algo-rithm achieves O(N log N) complexity for time-harmonicMaxwell problems.
Paul H. TsujiUniversity of Texas at AustinInstitute for Computational Engineering & [email protected]
Lexing YingUniversity of TexasDepartment of [email protected]
MS133
The DSP as a High-performance, General-purposeProcessor: First Experiences with FLAME
Digital Signal Processors (DSPs) are specialized architec-tures with low-power consumption, real-time processingsupport and sophisticated development tools. In this talk,we report our first experiences with the TMS320C6678multi-core DSP from Texas Instruments. We introducethe first complete BLAS implementation for this proces-sor. Building upon the BLAS and OpenMP, we use theflexibility of the FLAME methodology and the libflamelibrary to port higher-level linear algebra functionality to
CS13 Abstracts 241
the DSP parallel architectures.
Murtaza AliTexas [email protected]
Francisco IgualUniv. Complutense de [email protected]
MS133
Optimization of Blas Kernels: Close-to-the MetalTricks for Multicore and Manycore Architectures
Novel computer architectures often require non-traditionaloptimization techniques in order to allow even routines assimple as the traditional BLAS kernels to perform as theyshould. This talk will focus on techniques for optimizingthese kernels as well as higher-level techniques to employthem to greater advantage on large and/or new architec-tures.
John A. GunnelsT.J. Watson Research [email protected]
MS133
MKL, BLAS, and New Architectures
The Intel(R) Math Kernel Library (MKL) is well knownfor its high performance for HPC applications across manyscientific domains. We show some of the code generationand design behind some key techniques and models for ourlinear algebra routines and the performance, time, and ac-curacy benefits behind our tools and techniques. We focuson some of the many-core results from our newest Intel(R)Xeon Phi processors.
Greg HenryIntel [email protected]
MS133
BLIS: A New Framework and Interface for theBLAS
We present a modern alternative to the Basic Linear Alge-bra Subprograms (BLAS) that addresses many shortcom-ings in the original BLAS. This framework, which we callBLIS, allows an expert to (1) rapidly instantiate BLAS-likelibraries for new architectures, and (2) design entirely newoperations by leveraging existing components of the frame-work. Preliminary performance on select architectures ishighly competitive with high-performance open source andcommercial products.
Field G. Van ZeeDepartment of Computer SciencesThe University of Texas at [email protected]
MS134
A new Monte Carlo Method for Velocity-
dependent Neutrino and Photon Transport
Abstract not available at time of publication.
Ernazar AbdikamalovCalifornia Institute of [email protected]
MS134
Discontinuous Galerkin Methods for Vlasov-Maxwell Systems
Abstract not available at time of publication.
Yingda ChengDepartment of MathematicsMichigan State [email protected]
MS134
Asymptotic Preserving DG Methods for KineticEquations
A family of high order asymptotic preserving schemes areproposed for some kinetic equations. The methods arebased on the micro-macro decomposition of the problem,and they combine high order discontinuous Galerkin spa-tial discretizations and second order IMEX temporal dis-cretizations. Both theoretical and numerical studies arecarried out to demonstrate the performance of the meth-ods.
Juhi JangUniversity of California, Riverside, [email protected]
Fengyan LiRensselaer Polytechnic [email protected]
Jingmei QiuUniversity of [email protected]
MS134
Asymptotic Preserving Schemes for Kinetic Equa-tions in the High Field Regime
I will introduce numerical methods for two kineticequations—the Vlasov -Poisson-Fokker-Planck (VPFP)system and the semiconductor Boltzmann equation, withemphasis on the connections to the high field limit. Thetwo stiff terms under this scaling–the collision term and theforce term, make the classical kinetic solver expensive toimplement. For the VPFP system, the idea is to combinethe two stiff terms and treat them implicitly; while for thesemiconductor Boltzmann equation, we use the penaliza-tion idea to overcome another remarkable difficulty that noexplicit form of the local equilibrium is available.
Li WangUniversity of California, Los [email protected]
MS135
Simulating Optogenetic Control of the Heart
Optogenetics is the process of expressing light-sensitive
242 CS13 Abstracts
proteins in tissue then eliciting a precise electrical re-sponse with illumination. Initial experiments in opticalheart stimulation have yielded promising results, but clini-cal translation remains a distant goal. Developing the abil-ity to accurately simulate optogenetics in biophysically-detailed cardiac models is an essential step towards thisgoal. We present a comprehensive but flexible frameworkfor simulating cardiac optogenetics, modeling optical stim-ulation effects from molecule scale to organ level.
Patrick M. BoyleInstitute for Computational MedicineJohns Hopkins [email protected]
John C. Williams, Emilia EntchevaStony Brook [email protected], [email protected]
Natalia A. TrayanovaInstitute for Computational MedicineJohns Hopkins [email protected]
MS135
Reconstruction of Catheter Electrograms and 12-lead ECG in Heart-failure Patients using a Bido-main Reaction-diffusion Model of the Heart andTorso
To predict and improve response to cardiac resynchro-nization therapy in heart-failure patients, we are buildingpatient-specific models based on MRI data of 12 cases andsimulating cardiac activation, local electrograms, and thesurface electrocardiogram. Simulations used 2048 cores ofa Cray XE6. Activation order as well as electrogram andelectrocardiogram morphology are compared to their mea-sured counterparts. Here we report on our modeling tech-niques, challenges in data acquisition, and results in thefirst 4 cases.
Mark Potse, Dorian Krause, Rolf KrauseInstitute of Computational ScienceUniversity of [email protected], [email protected], [email protected]
Wilco KroonUniversity of LuganoLugano, [email protected]
Enrico CaianiPolitecnico di MilanoMilano, [email protected]
Frits PrinzenMaastricht UniversityMaastricht, The [email protected]
Angelo AuricchioFondazione Cardiocentro [email protected]
MS135
Highly Scalable Cardiac Modeling Codes for Petas-
cale Computing
Electrophysiology simulations of high-resolution humanheart ventricles with realistic anatomy and biophysicallydetailed cellular models are run on Sequoia at LLNL. Max-imum sustained performance is 11.8 PFlop/s (58.8% ofpeak) with exceptional strong scaling and time to solu-tion. For a heart at 0.10 mm resolution, the throughput is18.24 heartbeats per minute, over 1200 times faster thanthe previous state of the art. At 0.13 mm resolution, ourthroughput is just 12% below real-time simulation.
John J. RiceComputational Biology CenterIBM T.J. Watson Research [email protected]
Erik Draeger, James GlosliLawrence Livermore National [email protected], [email protected]
Slava Gurev, Changhoan Kim, John MagerleinIBM [email protected], [email protected],[email protected]
Art MirinLawrence Livermore National [email protected]
Matthias ReumannIBM [email protected]
David RichardsLawrence Livermore National [email protected]
MS135
Simulation of Cardiac Electrophysiology: EfficientTime Integration
Cardiac electrophysiology can be modeled by the bidomainequations, a multi-scale reaction-diffusion system of non-linear ODEs describing the ionic currents at the cellularscale coupled with a set of PDEs describing the propaga-tion of the electrical activity at the tissue scale. To pro-duce clinically useful data, billions of variables must beevolved. In this presentation, I discuss the most efficienttime-integration methods for the bidomain equations andreveal the secrets of their success.
Raymond J. SpiteriUniversity of Saskatchewan, CanadaDepartment of Computer [email protected]
MS136
Implicit Domain Decomposition Methods for At-mospheric Flows on the Cubed-sphere
We introduce a fully implicit second-order finite volumemethod for solving some transport problems on the cubed-sphere. To solve the linear system of equations at each timestep, we investigate some single-level and multilevel over-lapping Schwarz preconditioners. Even though, Schwarzpreconditioners are initially designed for elliptic problems,we show by numerical experiments that they scale quite
CS13 Abstracts 243
well for this class of purely hyperbolic problems on a ma-chine with a large number of processors.
Haijian YangHunan University,Changsha, Hunan 410082, [email protected]
Chao YangDept. of Computer ScienceUniversity of Colorado, [email protected]
Xiao-Chuan CaiUniversity of Colorado, BoulderDept. of Computer [email protected]
MS136
A New Multi-tracer-efficient Semi-LagrangianTransport Scheme for the Community AtmosphereModel (CAM)
Recently, a conservative semi-Lagrangian multi-tracertransport scheme (CSLAM) was integrated in the High-Order Method Modeling Environment (HOMME). A newdynamical core for the Community Atmosphere Model isbased on HOMME-Spectral Elements (CAM-SE) and thecubed-sphere grid. To improve computational efficiencywe replace the SE transport scheme for structured gridsby CSLAM but still use the velocity provided by the SEdiscretization. We compare this hybrid approach on thebaroclinic wave test example and discuss the effects of sev-eral coupling methods.
Christoph ErathNCARUniversity of Colorado, [email protected]
Mark A. TaylorSandia National Laboratories, Albuquerque, [email protected]
MS136
Application of the Cubed-Sphere Grid to TiltedBlack Hole Accretion Disks
Traditionally, general relativistic magnetohydrodynamicsimulations of black hole accretion disks have usedspherical-polar meshes. Such meshes are adequate for caseswhen the angular momentum axis of the disk is alignedwith the spin axis of the black hole. However, for mis-aligned (i.e. tilted) disks, the cubed-sphere grid offers anumber of advantages. In this talk I present some of ourwork using the cubed sphere grid for simulations of tiltedblack hole accretion disks.
P. Chris Fragile, Will DuPre, Julia WilsonCollege of [email protected], [email protected],[email protected]
Chris LindnerUniversity of [email protected]
MS136
Advances in Numerical Modeling of the Atmo-
sphere on the Cubed-sphere Grid
The desire to improve our understanding of the Earth sys-tem has driven numerical models to very fine grid reso-lutions. Simulations at these resolutions require massivelyparallel systems in order to be completed over a reasonabletime scale. In order to accomplish parallelism on this scale,modern numerical methods have relied on quasi-uniformgrids, in particular the cubed-sphere grid. This talk willboth review advances in modeling the atmosphere on thecubed-sphere grid and identify new avenues for future re-search on this subject.
Paul UllrichUC [email protected]
MS137
Computational Chemistry: Chemical Accuracyand Errors at Different Scales
Computational chemistry can be used to reliably predictthe properties of compounds with density functional the-ory and correlated molecular orbital theory. The use ofcomputational methods to design syntheses for new mate-rials, for example, for catalysis, solar energy capture, andnuclear fuels, is in its infancy. We will describe the com-plex issues that need to be addressed for the design of newmaterials syntheses and initial progress on understandingbasic steps in such reaction mechanisms.
David DixonUniversity of [email protected]
MS137
Chemical Kinetic Model Development usingBayesian Variable Selection
We present a novel approach for tractable Bayesian in-ference of chemical kinetic models from noisy and indi-rect system-level data. Formulating the problem as vari-able selection and making use of point-mass mixture pri-ors, our approach allows an exhaustive comparison ofall models comprised of subsets of elementary reactions.Adaptive Markov chain Monte Carlo methods are usedto efficiently explore the posterior distribution. We alsopresent a newly-developed proposal distribution that im-proves MCMC chain mixing in variable selection problemscontaining strong posterior correlations.
Nikhil Galagali, Youssef M. MarzoukMassachusetts Institute of [email protected], [email protected]
MS137
Split Step Adam Moulton Method for Stiff Stochas-tic Differnetial Equations
We present a split-step method for solving stochastic differ-ential equations (SDEs). The method is based on a secondorder split Adams-Moulton Formula for stiff ordinary dif-ferential equations and modified for use on stiff SDEs whichare stiff in both the deterministic and stochastic compo-nents. Its order of strong convergence is established andstability region is displayed. Numerical results show the ef-fectiveness of the method in the path wise approximation.
244 CS13 Abstracts
Abdul M. KhaliqMiddle Tennessee State UniversityDepartment of Mathematical [email protected]
David A. VossWestern Illinois [email protected]
MS137
Accurate Filtering of The Navier-Stokes Equation
Filtering in the data assimilation context is often consid-ered to be the reproduction of a deterministic point esti-mate of the state of a system from noisily observed dataand knowledge of the underlying system, resulting in a con-tinuous feedback control problem. This is in contrast tothe probabilistic interpretation of the state as a randomquantity, whose distribution reflecting our knowledge at aparticular time is referred to as the filtering distribution,i.e. the distribution of state given the filtration generatedby the random observations made until the given time. Wefocus on the former case in which the model is know, i.e.the perfect model scenario. In the perfect model scenariotwo ideas drive accurate filtering: (i) observe enough lowfrequency information, and (ii) model variance inflation:trust the observations. In this talk I will illustrate this forthe simplest filter, referred to as 3DVAR (perhaps moreaccurately called 2DVAR, if one then adapts 4DVAR to(2+1)DVAR to avoid the necessary confusion) applied tothe 2D Navier-Stokes equations, in the low and high fre-quency observation limits. The latter will yield an SPDEfor the estimator.
Kody LawMathematics InstituteUniversity of [email protected]
Andrew StuartMathematics Institute,University of [email protected]
Dirk BloemkerUniversity of Augsburg, [email protected]
Kostas ZygalakisUniversity of Southampton, [email protected]
MS138
An Application of Sparse Sensing to Partial Differ-ential Equations
Abstract not available at time of publication.
Ido BrightUniversity of [email protected]
MS138
Reduced Order Models of Unsteady Aerodynamic
Flow for Control
Abstract not available at time of publication.
Steven L. BruntonPrinceton [email protected]
MS138
TBD - Equation Free Modeling
Abstract not available at time of publication.
I. G. KevrekidisPrinceton [email protected]
MS138
Hybrid Reduced Order Integration Using theProper Orthogonal Decomposition and DynamicMode Decomposition
Abstract not available at time of publication.
Matthew O. WilliamsProgram in Applied and Computational MathematicsPrinceton [email protected]
MS139
Probabilistic Graphical Models: Applications toUQ
We develop a non-parametric probabilistic graphical modelframework that can provide a clear and efficient way to rep-resent correlated random variables in uncertainty quantifi-cation problems governed by stochastic PDEs. Given a setof training data, the basic objectives include structure de-sign, graph learning and inference problem. Using a prob-abilistic graphical model approach to UQ allows 1) captur-ing the correlations between random variables; 2) inferenceof any unobserved variables with appropriate error bars; 3)wide application prospects due to the non-parametric na-ture of the model.
Peng Chen, Nicholas ZabarasCornell [email protected], [email protected]
MS139
The Ensemble Kalman Filter for Inverse Problems
We present a novel non-standard perspective of the en-semble Kalman filter (EnKF) methodology that convertsit into a generic tool for solving inverse problems. We pro-vide numerical results to illustrate the efficacy of the pro-posed EnKF for solving a wide range of applications. Inparticular, we discuss (i) inversion of hydraulic head datato determine transmissivity in a groundwater model; (ii)inversion Eulerian velocity measurements to determine theinitial condition in an incompressible fluid.
Marco IglesiasCivil and Environmental [email protected]
Kody Law
CS13 Abstracts 245
Mathematics InstituteUniversity of [email protected]
Andrew StuartUniversity of [email protected]
MS139
Tensor-based Uncertainty Quantification of Exper-imental Data
A methodology is presented for inferring a functional repre-sentation of a stochastic quantity from a collection of its re-alizations. This collection comes from experiments, whereno specific sampling strategy such as sparse grids or low-discrepancy sequences can be assumed. We derive a newtensor-train approach based on orthogonal polynomials inorder to accurately approximate the random quantity withfew degrees of freedom, allowing for a stable solution evenwith scarce data. Provided examples support the notionof data-driven multilinear algebra as a potentially effectivetool for high-dimensional uncertainty quantification.
Tarek [email protected]
Lionel MathelinLIMSI - [email protected]
Youssef M. MarzoukMassachusetts Institute of [email protected]
MS139
Strategies for Quantification of Epistemic Uncer-tainty
Epistemic uncertainty refers to the uncertainty due to ourlack of knowledge. In this talk, we are concerned with para-metric type of uncertainty when its complete probabilisticinformation is not available. We describe a few numericalapproaches to construct reliable system response, withoutusing probability distribution function, and their ranges ofapplicabilities. We also present a method to compute boththe upper bound and lower bound of the systems responsesusing variational inequality of relative entropy. The resultsbounds can serve as estimates of the ”best case scenario”and ”worst case scenario”. We also discuss how to effi-ciently compute the bounds as a mere post-processing stepof the traditional UQ computation.
Xiaoxiao Chen, Jing Li, Xin Qi, Dongbin XiuPurdue [email protected], [email protected],[email protected], [email protected]
MS140
Recent Advances in Scalable Sparse FactorizationMethods
Because of their robustness, factorization-based solversand preconditioners play a significant role in develop-ing scalable solvers for large-scale, ill-conditioned andhighly-indefinite algebraic equations. We present our re-cent development on sparse factorization algorithms for
multi/many-core architectures and exploiting low-rankstructures. These include new scheduling algorithm forDAG execution to reduce processes idle time, incorporatelight-weight OpenMP threads in MPI processes to reducememory footprint, and superfast factorization algorithmsbased on hierarchically semi-separable (HSS) matrices forstructured sparse linear systems arising from discretizedPDEs.
Xiaoye Sherry LiComputational Research DivisionLawrence Berkeley National [email protected]
MS140
Multithreaded Sparse Kernels for Solution ofSparse Linear Systems
Solving sparse linear systems is one of the most expensiveparts of computational science applications. It is impor-tant to exploit the shared memory parallelism available inmodern multicore architectures to develop scalable sparselinear solvers. We describe a task-based model and imple-ment the sparse kernels using the model to improve multi-threaded performance. We focus on sparse kernels that areimportant for both iterative and direct solvers and presentperformance results on various multicore architectures.
Sivasankaran RajamanickamSandia National [email protected]
MS140
Accelerated Fixed Point Methods and Other Ad-vances in SUNDIALS
The SUNDIALS software suite supplies time integrationand nonlinear solvers used in implicit solution approachesto ODEs and PDEs. New capabilities going into this paral-lel, C language suite of codes include accelerated fixed pointnonlinear solvers and IMEX Runge-Kutta time integrationmethods. We will overview the fixed point solvers, their in-terfaces in SUNDIALS, and their application in subsurfaceand materials sciences. Lastly, we will present informationon the new time integrators soon to be added.
Carol S. WoodwardLawrence Livermore Nat’l [email protected]
MS140
Recent Advances on Reducing Communication inAMG
Algebraic Multigrid (AMG) solvers are an essential compo-nent of many large-scale scientific simulation codes. Theircontinued numerical scalability and efficient implementa-tion is critical for preparing these codes for emerging com-puter architectures. Previous investigations have shownthat the increasing communication complexity on coarsergrids combined with the effects of increasing numbers ofcores lead to severe performance bottlenecks for AMG onvarious multicore architectures. We present recent progresson several efforts to reduce communication in AMG.
Ulrike Meier YangLawrence Livermore National [email protected]
246 CS13 Abstracts
Robert FalgoutCenter for Applied Scientific ComputingLawrence Livermore National [email protected]
Jacob SchroderLawrence Livermore National [email protected]
MS141
Applications and Challenges in Large-scale GraphAnalysis
Emerging real-world graph problems include detectingcommunity structure in large social networks, improvingthe resilience of the electric power grid, and detectingand preventing disease in human populations. We discussthe opportunities and challenges in massive data-intensivecomputing for applications in social network analysis, ge-nomics, and security. The explosion of real-world graphdata poses substantial challenges for software, hardware,algorithms, and application experts.
Jason RiedyGeorgia Institute of TechnologySchool of Computational Science and [email protected]
Henning MeyerhenkeKarlsruhe Institute of TechnologyInstitute of Theoretical [email protected]
David A. BaderGeorgia Institute of [email protected]
MS141
Large Scale Graph Analytics and Randomized Al-gorithms for Applications in Cybersecurity
Abstract not available at time of publication.
John R. JohnsonPacific Northwest National [email protected]
Emilie HoganPacific Nortwest National [email protected]
Mahantesh HalappanavarPacific Northwest National [email protected]
MS141
Combinatorial and Numerical Algorithms for Net-work Analysis
We report on our recent efforts in developing combinatorialand numerical algorithms for network analysis. Algorithmsunder consideration include community detection and seedset expansion, algebraic distances and lean algebraic multi-grid. The development is driven by the rationale of com-bining ease-of-use and high performance.
Henning MeyerhenkeKarlsruhe Institute of Technology
Institute of Theoretical [email protected]
Christian StaudtKarlsruhe Institute of [email protected]
MS141
Anomaly Detection in Very Large Graphs: Model-ing and Computational Considerations
Graph theory provides an intuitive mathematical founda-tion for dealing with relational data, but there are numer-ous computational challenges in the detection of interestingbehavior within small subsets of vertices, especially as thegraphs grow larger and the behavior becomes more sub-tle. This presentation discusses computational considera-tions of a residuals-based subgraph detection framework,including the implications on inference with recent statis-tical models. We also present scaling properties, demon-strating analysis of a billion-vertex graph using commodityhardware.
Benjamin Miller, Nicholas Arcolano, Edward Rutledge,Matthew Schmidt, Nadya BlissMIT Lincoln [email protected], [email protected], [email protected],[email protected], [email protected]
MS142
Least-squares Finite Element Methods for Incom-pressible Flows with Improved Mass Conservation
We present a new locally conservative LSFEM for thevelocity-vorticity-pressure Stokes and Navier-Stokes equa-tions, which uses a piecewise divergence-free basis for thevelocity and standard C0 elements for the vorticity andthe pressure. Computational studies demonstrate that thenew formulation achieves optimal convergence rates andyields high conservation of mass. We also propose a simplediagonal preconditioner for the dV-VP formulation, whichsignificantly reduces the condition number of the LSFEMproblem.
Pavel BochevSandia National LaboratoriesComputational Math and [email protected]
Luke [email protected]
James [email protected]
MS142
Asymptotically Exact a posteriori Error Estima-tors for LS Methods
A new asymptotically exact a posteriori error estimatoris developed for first-order div least-squares(LS) finite el-ement methods. Let (uh , σh) be equal-order LS approxi-
mate solution for (u , σ = −A∇u). Then, E = ‖A−1/2(σh+A∇uh)‖0 is asymptotically exact a posteriori error estima-
tor for ‖A1/2∇(u − uh)‖0 or ‖A−1/2(σ − σh)‖ depending
CS13 Abstracts 247
on the order of approximate spaces for σ and u. For E tobe asymptotically for ‖A1/2∇(u−uh)‖0, we require higherorder approximation property for σ, and vice versa. Whenboth A∇u and σ are approximated in the same order ofaccuracy, the estimator becomes an equivalent error esti-mator for both errors. Confirming numerical results arepresented.
Jaeun KuOklahoma State [email protected]
Zhiqiang CaiPurdue UniversityDepartment of [email protected]
Varis CareyUniversity of Texas at [email protected]
Eun-Jae ParkYonsei [email protected]
MS142
FOSLL∗ For Nonlinear Partial Differential Equa-tions
For sufficiently regular, elliptic-like problems, a least-squares approach may be continuous/coercive in the H1
or H(div) norm, yielding convergence in H1 or H(div).Often, however, coarse approximations may have large L2-error relative to the H1 or H(div) seminorm. In contrast,the first-order system LL∗ (FOSLL∗) approach minimizeserror in a dual norm induced by the differential operator,yielding better control of the L2-error. We extend this gen-eral framework to nonlinear problems and establish conver-gence theory.
Chad WestphalWabash [email protected]
MS142
Superconvergence: Unclaimed Territories
Abstract not available at time of publication.
Zhimin ZhangWayne State UniversityDepartment of [email protected]
MS143
Power Flows and Stability of a Modern Distribu-tion Feeder
In this talk I briefly review the work at LANL on ODE andPDE modeling of a distribution feeder. In particular, I planto discuss effects of many inductive motors and of manydistributed generators on the feeder stability, bifurcationdiagram and potentially dangerous hysteretic response toa voltage fault.
Michael ChertkovLos Alamos National [email protected]
MS143
Graph-based Approaches for N-x ContingencyAnalysis of Electric Power Grids
The goal of N − x contingency selection is to pick a subsetof critical cases to assess their potential for causing a severecrippling of a power grid. Since the number grows expo-nentially, it can be overwhelming even for a moderately-sized system. We propose a novel method for N − x selec-tion using group-betweenness centrality and show that theamount of computation can be decoupled from the problemsize, thus making it feasible for large systems.
Mahantesh HalappanavarPacific Northwest National [email protected]
Yousu Chen, Zenyu Huang, Mark RicePacific Northwest National [email protected], [email protected],[email protected]
MS143
Random Chemistry and Dual Graphs: Two Waysto Understand Cascading Failures in Power Grids
Power systems are vulnerable to low probability, high im-pact failures. Because the probability distribution of black-out sizes follows a power law, new statistical methods areneeded to provide good information about the risk of largeblackouts. The Random Chemistry approach strategicallytests a model to quickly generate large unbiased sets ofcascading outage data. The dual graph method transformssimulation data into a graph that provides insight into howcascades propagate. Both methods transform models anddata into meaningful information.
Paul HinesUniversity of [email protected]
MS143
The Role of Grid Topology in Secure Power SystemOptimization Problems
The electric grid is considered critical infrastructure anddata concerning transmission lines and their connectedtopology are restricted. We consider the masking of powersystem optimization models for shared computing, and fordistributing equivalent power system models among re-searchers. We examine properties of masking transforma-tions that allow equivalent optimization models to be pub-lic, the analysis of which can be only be related back to theoriginal model through knowledge of the transformation.
Bernard Lesieutre, Alex Borden, Daniel Molzahn,Parmesh RamanathanUniversity of Wisconsin at [email protected], [email protected],[email protected], [email protected]
MS144
Closure Modeling for the Proper Orthogonal De-composition of Turbulent Flows
The reduced-order models (ROMs) are frequently used inthe simulation of complex flows to overcome the high com-putational cost of direct numerical simulations, especially
248 CS13 Abstracts
for three-dimensional nonlinear problems. The proper or-thogonal decomposition (POD), as one of the most com-monly used tools to generate ROMs, has been utilized inmany engineering and scientific applications. Its originalpromise of computationally efficient, yet accurate approx-imation of coherent structures in high Reynolds numberturbulent flows, however, still remains to be fulfilled. Tobalance the low computational cost required by ROMs andthe complexity of the targeted flows, appropriate closuremodeling strategies need to be employed. In this talk,we put forth several closure models for the POD-ROMsof structurally dominated turbulent flows. These models,which are considered state-of-the-art in large eddy simu-lation, are carefully derived and numerically investigated.We also discuss several approaches for an efficient and ac-curate numerical discretization of general nonlinear PODclosure models. We numerically illustrate these develop-ments in several computational settings, including a three-dimensional turbulent flow past a cylinder at Reynoldsnumber Re = 1000. A rigorous numerical analysis of thenew computational framework will also be presented.
Traian IliescuDepartment of MathematicsVirginia [email protected]
Zhu WangInstitute for Mathematics and its ApplicationsUniversity of [email protected]
MS144
Trust Region POD 4D VAR Data Assimilation ofa Parabolized NavierStokes Equations Model
A reduced order model based on Proper Orthogonal De-composition (POD) 4D VAR data assimilation for parabo-lized NavierStokes (PNS) equations is derived. Various ap-proaches of POD implementation of reduced order inverseproblem are compared including an ad-hoc POD adaptiv-ity along with a trust region POD adaptivity. Numericalresults show that trust region POD 4D VAR provides thebest results for all error metrics for reduced order inverseproblem of PNS equations.
Ionel M. NavonFlorida State UniversityDepartment of Scientific [email protected]
Juan DuIAP, Academia Sinica, Beijing, [email protected]
MS144
Model Reduction and Multi-fidelity Data in Uncer-tainty Quantification of Flow Models
We study the use of lower-fidelity training data for uncer-tainty quantification of advanced fluid mechanics models.We use POD-based dimensionality reduction to obtain ap-proximations of the full model. This imperfect data, cali-brated using stochastic-processes learning, can be used torecover the full model’s response to uncertainty. Our ap-proach can be generalized to using multiple sources of data.We performed experiments on simplified fluid flow models,
as well as on a professional fluid mechanics solver nek5000.
Oleg RoderickArgonne National [email protected]
MS144
Towards a Blackbox Approach for Model Reduc-tion via EIM
We will introduce two EIM (empirical interpolationmethod) -based methods for nonlinear model reduction,both inspired by discrete empirical interpolation method(DEIM). These new approaches provide comparable per-formance with DEIM at reduced cost and are promisingfor blackbox model reduction. Different points selectionalgorithms will be compared with DEIM and numericalexamples will be shown to demonstrate the performanceof the new algorithms. Finally, stability issues will be dis-cussed.
Xueyu ZhuDivision of Applied MathematicsBrown Universityxueyu [email protected]
Ramakanth MunipalliHypercomp. [email protected]
Hesthaven JanDivision of Applied MathematicsBrown Universityjan [email protected]
MS145
Approximation and Use of Set Valued Solutions toStochastic Inverse Problems
We discuss the use of set-valued solutions for approximatesolution of stochastic inverse problems for parameter deter-mination. The focus is on using observations on multiplequantities of interest.
Donald Estep, Troy ButlerColorado State [email protected] or [email protected], [email protected]
MS145
Adaptive Stochastic Collocation for PDE Opti-mization under Uncertainty
Many optimization problems in engineering and science aregoverned by partial differential equations (PDEs) with un-certain parameters. Although such problems can be for-mulated as optimization problems in Banach spaces andderivative based optimization methods can in principle beapplied, the numerical solution of these problems is morechallenging than the solution of deterministic PDE con-strained optimization problems. The difficulty is that thePDE solution is a random field and the numerical so-lution of the PDE requires a discretization of the PDEin space/time as well as in the random variables. Asa consequence, these optimization problems are substan-tially larger than the already large deterministic PDE con-strained optimization problems.In this talk we discuss the numerical solution of such op-
CS13 Abstracts 249
timization problems using stochastic collocation methods.We explore the structure of this method in gradient andHessian computations. A trust-region framework is usedto adapt the collocation points based on the progress ofthe algorithms and structure of the problem. Convergenceresults are presented. Numerical results demonstrate sig-nificant savings of our adaptive approach.
Matthias HeinkenschlossDepartment of Computational and Applied MathematicsRice [email protected]
Drew P. KouriMathematics and Computer Science DivisionArgonne National [email protected]
MS145
A Framework for Sequential Experimental Designfor Inverse Problems
Abstract not available at time of publication.
Luis TenorioColorado School of [email protected]
MS145
A Generalized Stochastic Collocation Approach toConstrained Optimization for Random Data Iden-tification Problems
Characterizing stochastic model inputs to physical andengineering problems relies on approximations in high-dimensional spaces, particularly in the case when the ex-perimental data or targets are affected by large amountsof uncertainty. To approximate these high-dimensionalproblems we integrate a generalized adaptive sparse gridstochastic collocation method with a SPDE-constrainedleast squares parameter identification approach. Rigor-ously derived error estimates will be used to show the ef-ficiency of the methods at predicting the behavior of thestochastic parameters.
Clayton G. WebsterOak Ridge National [email protected]
Max GunzburgerFlorida State [email protected]
Catalin S. TrencheaDepartment of MathematicsUniversity of [email protected]
MS146
Portability of a Large-scale Heart Model throughHybrid Parallelization
Portability of CS&E software between diverse parallel ar-chitectures is a challenging task and important for itsflexible and efficient usage. In this talk, we discuss ourexperiences with a large-scale heart model (monodomainand bidomain equations). The in-house code Propag wasported from its original design for shared memory systems
to contemporary massively parallel multicore architecturesby means of a hybrid OpenMP–MPI parallelization. Wealso report on the relevance of the newly introduced par-allel setup phase.
Thomas Dickopf, Dorian Krause, Mark Potse, RolfKrauseInstitute of Computational ScienceUniversity of [email protected], [email protected],[email protected], [email protected]
MS146
ONELAB: Open Numerical Engineering LABora-tory
We present the ONELAB software, a lightweight opensource toolkit to interface finite element solvers used ina variety of engineering disciplines. Based on the design ofthe open-source CAD modeler, mesh generator and post-processor Gmsh, ONELAB is targeted for use in educationand small- and medium-size businesses, which do not re-quire the power (and cannot justify the cost) of commercialfinite element software suites.
Christophe GeuzaineUniversity of [email protected]
MS146
Commercial Open-source: An Approach to Com-putational Fluid Dynamics
Business and open-source are no contradiction: while thisbasic truth is now widely adopted, many software usersstill struggle to embrace the world of commercial open-source and to separate between irrational fears and healthyprecautions. We present as an example the open-sourceproject Palabos (www.palabos.org) for computational fluiddynamics, which follows the logic of a commercial businessplan, yet is open-source and supports University-level re-search through academic collaboration. While its win-winbenefits are obvious, in practice this approach leads to nu-merous misunderstandings and fears which have a consid-erable impact on the project management. On the sideof commercial users, the most frequent misunderstandingsrelate to licensing and usage term, and to considerationsabout software quality and support guarantees. On theside of the academic community, fears about a future inter-ruption of the free-software guarantee, or about otherwiseunethical project management appear to be preponderant.We show that in order to overcome these difficulties, a pol-icy of extensive and open communication with the userbase is crucial, related to the project management as wellas the technical aspects of the software. In particular, weshow that both the programming and graphical user inter-faces need to be thought in such a way as to send signalswith which users from both academic and industrial com-munities feel comfortable.
Jonas LattUniversite de [email protected]
Bastien ChopardUniversity of Geneva, [email protected]
250 CS13 Abstracts
MS146
Building Effective Parallel Unstructured AdaptiveSimulation by In-memory Integration of ExistingSoftware Components
Unstructured adaptive meshing driven by error estimationprocedures support reliable mesh-based PDE simulations.Typically, interfacing of unstructured adaptive procedureswith existing analysis components is through files. File I/Ois currently the critical bottleneck in parallel adaptive anal-ysis, even when advanced parallel I/O methods are used.An approach to eliminate this bottleneck using in-memoryintegration of existing analysis procedures with adaptivemesh procedures is presented. Results demonstrate strongscaling to > 32,000 processors of the in-memory integra-tion.
Cameron SmithScientific Computation Research CenterRensselaer Polytechnic [email protected]
Onkar SahniRensselaer Polytechnic [email protected]
Mark S. ShephardRensselaer Polytechnic InstituteScientific Computation Research [email protected]
MS147
Implications of the Choice of SDC Nodes in theMultilevel PFASST Algorithm
The Parallel Full Approximation Scheme in Space andTime (PFASST) algorithm iteratively improves the solu-tion on each time slice by applying Spectral Deferred Cor-rection (SDC) sweeps to a hierarchy of discretizations atdifferent spatial and temporal resolutions. The number andtype of SDC nodes used on each level of refinement impactsseveral aspects of the scheme including stability, accuracy,convergence, and efficiency. We analyse various SDC nodesand their impact on several representative PDEs.
Matthew EmmettUniversity of North Carolina at Chapel HillDepartment of [email protected]
Michael MinionICME [email protected]
MS147
On the Convergence of Parallel Deferred Correc-tion Methods
We discuss novel convergence results for parallel de-ferred correction methods for solving initial-value prob-lems u′(t) = f(t, u(t)), u(0) = u0, with an emphasis onthe Parareal-SDC variant proposed by Minion & Williams(2008) and Minion (2011), and a new variant based onbarycentric rational interpolation. This is joint work withMartin J. Gander.
Stefan GuettelUniversity of [email protected]
MS147
An Optimized RIDC-DD Space-time Method forTime Dependent Partial Differential Equations
Recently, the Revisionist Integral Deferred Correction(RIDC) approach has been shown to be a relatively easyway to add small scale parallelism (in time) to the solutionof time dependent PDEs. In this talk I will show how largescale spatial parallelism can be added to RIDC using clas-sical and optimized Schwarz methods. This results in trulyparallel space-time methods for PDEs suitable for hybridOpenMP/MPI implementation. Initial scaling studies willdemonstrate the viability of the approach.
Ronald HaynesMemorial University of NewfoundlandSt. John’s, NL, [email protected]
Benjamin OngMichigan State [email protected]
MS147
Parallelization in Time: How Far into the FutureCan We See?
This talk will provide an overview of past and present de-velopments in parallelization of numerical methods for or-dinary and partial differential equations in the temporaldirection. I will highlight some recent efforts to extend theefficiency and applicability of parallel in time methods andgive some insight into the practical limitations of currentapproaches.
Michael MinionICME [email protected]
MS148
Efficient Methods for Wave Propagation in LayeredAbsorbing Media
In this talk we discuss Galerkin boundary integral formu-lations and their fast solution via Hierarchical matrices forthe problem of wave propagation in nested, layered, absorb-ing media. This is the forward problem in diffuse opticaltomography, where the different layers model for examplea head geometry. The software implementation of the re-sults presented in this talk is done via BEM++, a newlibrary for the solution of boundary integral equations viaGalerkin boundary elements.
Simon ArridgeDepartment of Computer ScienceUniversity College [email protected]
Timo BetckeUniversity College [email protected]
Martin SchweigerDepartment of Computer ScienceUniversity College [email protected]
Wojciech SmigajDepartment of Mathematics, University College London
CS13 Abstracts 251
Faculty of Physics, Adam Mickiewicz University [email protected]
MS148
An ADI Timestepping Preconditioner for theHelmholtz Equation
Passing to the time domain is a great way to avoid the pre-conditioning issues in the high-frequency Helmholtz equa-tion, though the complexity count is typically unfavorabledue the CFL condition. In the talk I will explain how thistimestepping restriction can be circumvented by a specificADI scheme, in the scope of narrowband solutions. As aresult, the Helmholtz equation can be solved in low com-plexity in rather general variable media.
Laurent DemanetMathematics, [email protected]
MS148
Sweeping Preconditioners for the Helmholtz Equa-tion
Many standard techniques for preconditioning do not workwell for high frequency wave propagation. We will describeand analyze new class of preconditioners based on sweep-ing processes and apply them to GMRES iterative solu-tions of frequency domain equations. Hierarchical matrixtechniques for compression and moving perfectly matchedlayers play important roles in the algorithm. The numberof operations scales essentially linearly in the number ofunknowns.
Bjorn EngquistDepartment of Mathematics and ICES, UT [email protected]
MS148
Paraxial Estimates and a Direct Structured Solverfor the Helmholtz Equation in the High-frequencyRegime
Abstract not available at time of publication.
Maarten de HoopCenter for Computational & Applied MathematicsPurdue [email protected]
MS149
Computational Engineering and Science Softwarefor Nanoscale Explorations
Interactive computational engineering and science softwarehas been developed for students to explore phenomena im-portant at the nanoscale. The software has been designedfor integration into physics, chemistry, materials science,and engineering courses. As an example, the software al-lows students to simulate the propagation of light througha photonic band gap material, and to design its internalnanoscale dimensions to tune the amount of light reflectedand transmitted at different frequencies.
Richard BraatzMassachusetts Institute of TechnologyDepartment of Chemical Engineering
MS149
Complexity Science and Computational Modeling
Complexity Science creates an opportunity to bring to theundergraduate curriculum topics that are useful and im-portant, accessible and engaging to students, and natu-rally connected to material from Mathematics, Physics,and Computer Science. At Olin College we have devel-oped an unusual new class, Computational Modeling, withan accompanying textbook, Think Complexity. Studentsin this class read seminal papers in Complexity Science,write Python programs to run experiments, and write casestudies reporting their results.
Allen DowneyOlin College of EngineeringGreen Tea [email protected]
MS149
Computational Laboratory Activities for MedicinalChemistry
Lab activities for a medicinal chemistry course were devel-oped to teach undergraduate biology and chemistry stu-dents the important role that computers play in the drugdiscovery process. During lab time, students learned howto use a docking program such as AutoDock Vina to con-duct a virtual screen of compounds and a protein visu-alization program such as PyMOL. They also gained anappreciation for how much a supercomputer can speed upthe virtual screening process.
Jimmy FrancoMerrimack [email protected]
MS149
Computational Quantum Mechanics in the Under-graduate Curriculum
Quantum Mechanics offers a fertile field for computation.The subject is mathematically and conceptually difficultand because Planck’s constant is small undergraduateshave poor quantum intuition. Computation can be invalu-able in helping students build intuition and opens up newclasses of problems, many of current experimental interest.I will describe a number of modules that I have developedfor teaching computational quantum mechanics and howthese materials are being used in the classroom.
Richard GassUniversity of [email protected]
MS150
Investigations of High Or-der Discontinuous Galerkin Methods for ImplicitLarge Eddy Simulations
We will present DG implicit LES simulations of canonicalturbulent flows and investigate the potential of stabilizedvery high-order approximations (polynomial degree ≥ 7) inunderresolved scenarios. We will focus on stabilization bypolynomial de-aliasing. Thus, the only remaining parame-ter in such a DG formulation is the choice of the numerical
252 CS13 Abstracts
flux functions. We will investigate the influence of differ-ent flux functions on the quality of iLES results for theTaylor-Green vortex.
Andrea D. BeckUniversity of [email protected]
Gregor GassnerInstitute for Aerodynamics and GasdynamicsUniversitaet [email protected]
MS150
Discretely Energy Stable Discontinuous GalerkinSpectral Element Methods
In this talk, we present a skew symmetric discontinu-ous Galerkin collocation spectral element (DGSEM) for-mulation for the non-linear Burgers equation. We provethat this formulation remains discretely conservative. Fur-thermore, we prove that in combination with the com-monly used Roe flux energy stability cannot be guaranteed,whereas in combination with the local Lax-Friedrichs fluxenergy stability can be proved. In combination, this yieldsa novel discretely energy stable and discretely conservativeDGSEM formulation.
Gregor GassnerInstitute for Aerodynamics and GasdynamicsUniversitaet [email protected]
MS150
A Solver for the Incompressible Navier-StokesEquations using DG, QBX, and Integral Equations
We present a new method for the solution of the in-compressible Navier-Stokes equation. Our approach com-bines the stiffly stable splitting scheme and a discontinu-ous Galerkin spatial discretization of the advection termswith a new integral-equation-based solver for the pressureand viscous stages. The latter is based on a volume gen-eralization of Quadrature by Expansion (QBX), originallya method for the computation of Nystrom discretizationsof layer potentials. We demonstrate the scheme’s perfor-mance in general geometries.
Andreas KloecknerCourant Institute of Mathematical SciencesNew York [email protected]
Tim WarburtonDepartment of Computational and Applied MathRice [email protected]
MS150
Shock Capturing Using Artificial Viscosity andMultiscale Methods
We present some advances in the application of high or-der Discontinuous Galerkin methods to problems involvingshock waves. First, an implicit scheme based on the diver-gence of the velocity as a driver for artificial viscosity willbe presented. Following, a multiscale approach in whichelements affected by shocks are discretized using a low or-
der sub-mesh will also be discussed. Both approaches willbe compared on the same cases to asses their performanceand robustness.
Ngoc Cuong Nguyen, David Moro, Jaime PeraireMassachusetts Institute of [email protected], [email protected], [email protected]
MS151
Error Control for Output Quantities of Interest inParameterized Partial Differential Equations
In this work, we propose and study some adaptive ap-proaches to control errors in numerical approximations ofdifferential equations with uncertain coefficients. Adaptiv-ity is based here on a posteriori error estimation and theapproaches rely on the ability to decompose the a poste-riori error estimate into contributions from the physicaldiscretization and the approximation in parameter space.Errors are evaluated in terms of quantities of interest usingwell established adjoint based methodologies.
Corey M. BryantICESThe University of Texas at [email protected]
Tim WildeySandia National [email protected]
Serge PrudhommeICESThe University of Texas at [email protected]
MS151
A Posteriori Analysis of Implicit-Explicit (imex)Methods
Implicit-Explicit (IMEX) schemes are an important andwidely used class of time integration methods for parabolicor hyperbolic partial differential equations. For example,explicit scheme may be used for the convection (or reac-tion) term, and an implicit one for the stiff diffusion term.Such schemes may preserve monotonicity properties inher-ent in the equation. In this talk we cast a class of IMEXschemes in a variational format and perform a posteriorianalysis to compute error in a quantity of interest for suchschemes.
Jehanzeb ChaudhryDepartment of MathematicsColorado State [email protected]
Don Estep, Simon TavenerColorado State [email protected], [email protected]
Victor E. GintingDepartment of MathematicsUniversity of [email protected]
MS151
On the Application of Adjoint Methods in Subsur-
CS13 Abstracts 253
face Flow Simulations
This presentation explores an application of the adjointmethod for subsurface characterization with a benchmarkproblem in water infiltration through soil. The idea is toemploy available sparse measurements of pressure and/orwater content to infer the subsurface permeability thathonors the measurement. The process that is designedto achieve this goal is to devise an iterative procedure inwhich for every iteration a possible permeability field isproposed, then a simulated response is obtained by solv-ing the associated boundary value problems, which is tobe compared with the measurements. The aim is to min-imize the difference between the simulated response andthe measurements. This difference (coined as a residual)is then used as a data in the adjoint equation associatedwith the original boundary value problem, which in turngives an information on the perturbation values, serving asa correction to the proposed permeability. The iterativeprocedure is run this way until a convergence is reached.Some numerical examples are presented to illustrate theframework.
Lawrence Bush, Victor E. GintingDepartment of MathematicsUniversity of [email protected], [email protected]
MS151
Adjoint Methods for Adjoint Inconsistent Formu-lations
Often, computing a numerically stable approximation toa PDE requires the formulation to be modified. Unfortu-nately, the adjoint of the modified formulation does notalways correspond to a modification of the formal adjoint.We demonstrate that using the adjoint of such a formu-lation to compute a posteriori error estimates can lead tounstable results. We show that an alternative approach ex-ists for DG and stabilized Galerkin methods that providesmore stable and accurate error estimates.
Tim WildeySandia National [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
John ShadidSandia National LaboratoriesAlbuquerque, [email protected]
MS152
Large-scale Stochastic Linear Inversion using Hier-archical Matrices
Large scale inverse problems, which frequently arise inearth-science applications, involve estimating unknownsfrom sparse data. The goal is to evaluate the best estimate,quantify the uncertainty in the solution, and obtain condi-tional realizations, which are computationally intractableusing conventional methods. In this talk, I will discuss thehierarchical matrix approach optimized for a realistic largescale stochastic inverse problem arising from a cross-well
seismic tomography application.
Sivaram AmbikasaranInstitute for Computational and MathematicalEngineeringStanford [email protected]
Judith Yue Li, Peter K KitanidisStanford [email protected], [email protected]
Eric F. DarveStanford UniversityMechanical Engineering [email protected]
MS152
Large-scale Biomolecular Electrostatics with Mas-sively Parallel FMM
A challenge in understanding biomolecular interactions isthat molecules are always in a solution (water moleculesand dissolved ions). The potential is described by aPoission-Boltzmann equation which is directly solved viaBEM and accelerated with FMM. We present the two di-electric formulation where the electrostatic field is calcu-lated using continuum dielectric media: the solvent and themolecule. Using FMM, we enable large-scale calculationswith millions of unknowns and advance towards solvingbiologically challenging problems.
Aparna ChandramowlishwaranGeorgia Institute of [email protected]
MS152
Optimizing the Black-box FMM for Smooth andOscillatory Kernels
A black-box FMM for smooth kernels was introduced in[Fong and Darve, 2009] and has been extended to oscil-latory kernels in [Messner, Schanz and Darve, 2012]. Itsmajor advantages are the easy implementation and adap-tation for new kernels. However, it requires significantlymore floating point operations compared to other FMMs.In my talk, I will present ways to tackle this drawback:(1) The exploitation of symmetries allow us to reduce thepre-computation time by a factor greater than 1000. (2)Blocking schemes increase the applicability of optimizedLevel 3 BLAS routines and lead to a great performance ofthe actual matrix-vector product.
Matthias MessnerStanford [email protected]
Berenger Bramas, Olivier [email protected], [email protected]
Eric F. DarveStanford UniversityMechanical Engineering [email protected]
MS152
Adaptive Parallel Scheduling of the Fast Multipole
254 CS13 Abstracts
Method
We introduce an adaptive scheduling method for parallelexecution of the fast multipole method (FMM) in a dy-namic computing environment. The scheduling is adaptiveto variation in algorithm, change in data as well as specificsand dynamics in the computing system. We present thegraph-theoretic analysis underlying the scheduling methodand experimental results with parallel FMM.
Bo ZhangDuke [email protected]
Nikos PitsianisDepartment of Electrical & Computer EngineeringAristotle [email protected]
Jingfang [email protected]
Xiaobai SunDepartment of Computer ScienceDuke [email protected]
MS153
Early Experience of Adaptation of ppOpen-AT: AnAuto-tuning Description Language
We will present an auto-tuning (AT) methodology withppOpen-AT:an AT description language. One of targets ofppOpen-AT is selection between CPU and GPU. In addi-tion to this selection, code optimization for CPU, such asloop fusion and loop separation, is also main function ofthe AT. The targets come from real numerical simulationcodes, such as earthquake simulation by FDM and electro-magnetism simulation by FVM. Performance evaluation isperformed by several current CPU architectures, like theSPARC64 IXfx of Fujitsu FX10.
Takahiro KatagiriThe University of [email protected]
Satoshi ItoInformation Technology CenterThe University of [email protected]
Satoshi OhshimaThe University of TokyoInformation Technology [email protected]
MS153
Parameter Auto-tuning for a Contour Integralbased Eigensolver using Stochastic Estimation ofEigenvalue Count
We consider a parallel eigensolver based on contour in-tegral. The method requires several parameters that arerelated to the number of eigenvalues around the contourpath. To improve the performance of the eigensolver, wepresent a method that selects suitable parameters in the
eigensolver by using a stochastic estimation of eigenvaluecount in a domain on the complex plane. The efficiency ofthe presented method is demonstrated by some numericalexperiments.
Yasuyuki Maeda, Yasunori FutamuraUniversity of [email protected],[email protected]
Tetsuya SakuraiDepartment of Computer ScienceUniversity of [email protected]
MS153
Evaluation of Genetic Algorithm on Initial VectorSettings for GMRES
GMRES(m) solver is very popular for sparse matrix com-putations in science and engineerings, but, performs verypoorly with some badly selected values of initial vectorsand Krylov dimensions. This paper evaluates our newlyproposed GA-based Automatic Tuning for GMRES(m) insize of populations of initial vectors. Experiments on 20largest unsymmetric matrices in Florida collection showthat the GA of initial vectors with ten populations is quiteeffective in convergence as well as predictive parallel per-formance.
Ken NaonoR&D departmentHitachi Asia Malaysia, [email protected]
Nordin ZakariaHPC service centerUniversiti Teknologi PETRONAS, [email protected]
Takao SakuraiCentral Research LaboratoryHitachi [email protected]
AnindyaJyoti PalHPC service center, Universiti Teknologi [email protected]
Nobutoshi SagawaR&D center, Hitachi Asia [email protected]
MS153
Energy-Aware Matrix Auto-tuning using GeneticAlgorithm and the Xabclib
We consider a form of matrix auto-tuning that includes notonly solver performance and stability but the energy per-formance as well. A simple power model is adopted that in-fers power consumption from runtime resource usage. Themodel is combined with conventional matrix solution per-formance measures into an aggregate objective score thatis then used by a Genetic Algorithm, deployed togetherwith the Xabclib matrix autotuning library, to search forthe optimal solver parameters.
Nordin Zakaria
CS13 Abstracts 255
HPC service centerUniversiti Teknologi PETRONAS, [email protected]
Ken NaonoR&D departmentHitachi Asia Malaysia, [email protected]
Anindya Jyoti PalHPC service centerUniversiti Teknologi [email protected]
Takao SakuraiCentral Research LaboratoryHitachi [email protected]
MS154
IMEX Schemes for Hyperbolic Systems and Ki-netic Equations with Diffusion Relaxation
We consider IMEX schemes for hyperbolic systems withstiffff relaxation in the so-called diffusion limit. In suchregime the system relaxes towards a convection-diffusionequation. The first objective is to show that traditionalpartitioned IMEX schemes will relax to an explicit schemefor the limit equation with no need of modication of theoriginal system. Of course the explicit scheme obtained inthe limit suffers from the classical parabolic stability re-striction on the time step. The main goal is to present anapproach, based on IMEX schemes, that in the diffusionlimit relaxes to an IMEX scheme for the convection-di ffu-sion equation, in which the diffusion is treated implicitly.This is achieved by a novel reformulation of the problem,and subsequent application of IMEX schemes to it. Severalnumerical examples including neutron transport equationsconfirm the theoretical analysis.
Sebastiano BoscarinoUniversity of [email protected]
MS154
Explicit Time Stepping for Radiative Transferbased on Mixed Variational Formulations
Based on a mixed variational formulation, we derive an ex-plicit time stepping scheme of leap-frog type for PN-FEMdiscretizations of radiative transfer. We discuss stabilityaspects, the appropriate choice of the time-step, and the ef-ficient implementation avoiding inversion of mass matrices.Numerical tests will be presented that illustrate the perfor-mance of the new algorithm on a few benchmark problems.
Herbert EggerNumerical Analysis and Scientific ComputingDarmstadt University of Technologyherbert.eggerematma.tum.de
MS154
A New Spherical-Harmonics Scheme for Multi-Dimensional Radiation Transport
Recent work by [McClarren & Hauck, Robust and accuratefiltered spherical harmonics expansions for radiative trans-
fer. J. Comput. Phys., 229(16):5597−5614, Aug. 2010]suggests that the filtered spherical harmonics method rep-resents an efficient, robust and accurate method for radia-tion transport, at least in the two-dimensional (2D) case.We extend their work to the three-dimensional (3D) caseand find that all of the advantages of the filtering approachidentified in 2D are present also in the 3D case. We refor-mulate the filter operation in a way that is independentof the timestep and of the spatial discretization. We alsoexplore different second- and fourth-order filters and findthat the second-order ones yield significantly better results.Overall, our findings suggest that the filtered spherical har-monics approach represents a very promising method for3D radiation transport calculations.
David RadiceMax-Planck Institut fur [email protected]
David RadiceMax-Planck Institut fuer [email protected]
Ernazar AbdikamalovCalifornia Institute of [email protected]
Luciano RezzollaMax-Planck Institut fur [email protected]
Christian [email protected]
MS154
StaRMAP – A Staggered Grid Approach for Arbi-trary Order Linear Moment Methods of RadiativeTransfer
A simple method to solve the PN equations of radiativetransfer is presented. A specific coupling between the mo-ments in the equations naturally induces a second-orderaccurate finite difference scheme on staggered grids. Whilethe scheme does not possess limiters, its simplicity givesrise to a very efficient implementation in MATLAB. Thecode, which is available for download, can solve problemswith ten million degrees of freedom in space, angle, andtime within a few seconds.
Benjamin SeiboldTemple [email protected]
Martin FrankRWTH Aachen UniversityCenter for Computational Engineering [email protected]
MS155
An Adaptive High Order Finite Element Schemefor the Monodomain Equation
Despite extensive research, obtaining accurate numericalsolutions to the cardiac electrical propagation equationsremains very time-consuming. Attempts have been madeto reduce the computational burden using adaptive algo-rithms, but success has been limited due to the high costs
256 CS13 Abstracts
associated with finite element matrix re-assembly. Wepresent an adaptive algorithm based on high-order finiteelements and controlled using an a posteriori error indica-tor which avoids this re-assembly problem and provides thepotential for significant efficiency improvements.
Christopher ArthursComputing LaboratoryUniversity of [email protected]
MS155
Efficiency Considerations for High-performanceComputing with Applications to Cardiac Electro-physiology
We discuss techniques for improving performance of sci-entific codes with a particular emphasis on solving thereaction-diffusion equations associated with cardiac elec-trophysiology. Some of these methods involve program-ming choices that decrease memory usage and/or process-ing time. Others methods are architecture-dependent, in-cluding traditional parallelization and graphics process-ing unit-based implementations, where decisions regardingmemory access patterns and problem decomposition arecritical. We discuss the implementation of these various ap-proaches and quantify the performance improvements theygenerate.
Elizabeth M. CherryRochester Institute of TechnologySchool of Mathematical [email protected]
MS155
Strongly Scalable Numerical Approaches for Mod-eling Coupled Cardiac Electro-mechanics at HighSpatial Resolution
Biophysically detailed multiscale computer models of theheart are increasingly important in advancing our under-standing of integrated cardiac function in health and dis-ease. However, such detailed multiphysics simulations arecomputationally vastly demanding. In previous studies,using up to 16k cores we demonstrated strong scalabilityfor solving the bidomain equations. More recently, we im-plemented a proper domain decomposition algebraic multi-grid bidomain solver which can be compiled for executionon both CPUs and GPUs in distributed memory environ-ments. In this study first results are presented on extend-ing this solver framework for electro-mechanically coupledmulti-physics simulations.
Gernot PlankInstitute of BiophysicsMedical University of Graz, [email protected]
MS155
Efficient Computational Techniques for ModelingElectro-mechanical Interactions in the Heart
Computer simulations stand out as a promising path to-wards increased understanding of electro-mechanical inter-actions in the heart, and in particular to fully uncover therole of mechano-electric feedback in post infarct arrhyth-mias. However, the simulations are challenging to performbecause of the complexity and strong non-linearity of therelevant mathematical models. In this talk we will ad-
dress these computational issues, and propose a set of so-lution methods based on operator splitting techniques. Thederived computational methods will then be applied to amodel of infarct injured sheep hearts, to study the effect ofMEF in the border zone surrounding the infarcted region.
Joakim SundnesSimula Research [email protected]
MS156
A Parallel and Dynamically Adaptive 3D Cubed-sphere Grid Framework for Hyperbolic Conserva-tion Laws
A parallel block-adaptive simulation framework is de-scribed for hyperbolic conservation laws on 3D cubed-sphere grids. We use a multi-dimensional finite-volumeapproach, which naturally maintains uniform second-orderaccuracy at the sector boundaries and corners of the cubed-sphere grid without requiring special interpolation or re-construction procedures. This simplifies implementation ofparallelism and adaptivity in our hierarchical multi-blockframework. Numerical tests demonstrate accuracy and ef-ficiency of the approach and show excellent parallel scala-bility on thousands of computing cores.
Hans De Sterck, Lucian IvanUniversity of WaterlooApplied [email protected], [email protected]
Scott NorthrupUniversity of TorontoInstitute for Aerospace [email protected]
Clinton P. GrothUniversity of Toronto Institute for Aerospace [email protected]
MS156
Grid Refinement in the GFDL High Resolution At-mosphere Model (HiRAM): Stretched and NestedGrid
Abstract not available at time of publication.
Lucas HarrisNOAA/[email protected]
MS156
Variable Resolution Capabilities of the Commu-nity Atmosphere Model’s Cubed-sphere Spectral-element Configuration
We will present results from high-resolution global atmo-spheric simulations using CAM-SE: The Community At-mosphere Model), running with the Spectral finite Elementdynamical core from NCAR’s High-Order Method Model-ing Environment (HOMME). CAM-SE uses fully unstruc-tured conforming quadrilateral grids, including cubed-sphere grids for global quasi-uniform resolution of theEarth’s atmosphere. For global 1/4 and 1/8 degree resolu-tions, CAM-SE runs efficiently on hundreds of thousandsof processors on modern Cray and IBM supercomputers
CS13 Abstracts 257
and obtains excellent simulation throughput.
Mark A. TaylorSandia National Laboratories, Albuquerque, [email protected]
Katherine J. EvansOak Ridge National [email protected]
Oksana GubaSandia National LaboratoriesNM, [email protected]
Peter H. [email protected]
Mike Levy, Rich [email protected], [email protected]
James OverfeltSandia National [email protected]
MS156
Utilizing Grid Refinement in the Cubed-sphereSpectral Element Option of CAM to Model Tropi-cal Cyclones
We utilize the variable-resolution Spectral Element dy-namical core of the NCAR Community Atmosphere Model(CAM-SE) on a cubed-sphere grid to forecast tropical cy-clones. This setup permits high resolution in low-latitudeocean basins where tropical cyclones are prevalent whilemaintaining continuity within the remainder of the globaldomain. We present short-term forecasts of selected stormsand compare performance to globally-uniform and limitedarea models currently utilized operationally for hurricaneprediction.
Colin M. ZarzyckiUniversity of [email protected]
Christiane JablonowskiUniversity of MichiganAnn Arbor MI [email protected]
MS157
The SORD Code for Rupture Dynamics
Simulations provide an important tool for investigatinghigh-frequency seismic energy generation. We model spon-taneous rupture within a 3D isotropic viscoelastic solid us-ing a mimetic generalized finite-difference method. Therelevant phenomena span many orders of magnitude in spa-tial scale, requiring high discretization resolution. Emerg-ing petascale computational resources are making possiblenew applications of this method, such as a physics-basedprobabilistic seismic hazard analysis (PSHA) map for theentire state of California.
Geoffrey Ely
Argonne National [email protected]
MS157
Earthquake Shaking from Rupture Dynamics andSeismic Wave Scattering
Predicting earthquake ground-motions in the frequencyband of engineering interest (0-10 Hz) represents an impor-tant challenge in computational science. Realistic simula-tions rely on fundamental earthquake source physics, de-scribing fault geometry and how frictional strength evolvesduring rupture, and seismic wave propagation through het-erogeneous Earth structure. The generation of syntheticbroadband seismograms requires state-of-the art computa-tional resources and highly scalable and efficient numericalcodes. The mini-symposium invites contributions that ad-dress this subject.
Walter [email protected]
Christian PeltiesLMU [email protected]
Martin Galis, Martin [email protected], [email protected]
Kirk E. JordanIBM T.J. Watson [email protected]
MS157
The SeisSol Code: Efficient Implementation ofthe ADER-DG Method for Large-Scale DynamicEarthquake Simulations
We will present a complete software solution for earthquakesimulations based on the arbitrary high-order derivativeDiscontinuous Galerkin (ADER-DG) method. The im-plementation achieves excellent scalability on large-scalehigh-performance computing infrastructure using unstruc-tured tetrahedral meshes. The presentation focuses on re-cent code optimizations, in particular the introduction ofmulti-threading and the efficient implementation of smallmatrix-matrix multiplications, as well as a large-scale dy-namic earthquake rupture case study.
Christian PeltiesLMU [email protected]
Alex BreuerTU [email protected]
Vipin SachdevaIBM Research, [email protected]
Walter [email protected]
Alice Gabriel
258 CS13 Abstracts
Michael BaderTechnische Universitat MunchenDepartment of [email protected]
Kirk E. JordanIBM T.J. Watson [email protected]
Martin [email protected]
MS157
Construction of Models and Meshes for Large-ScaleEarth Science Applications
The construction of domain definitions and meshes arecritical steps in the execution of earth science simula-tions. This presentation will discuss on-going efforts onthe extension of geometric model construction, meshing,and partitioning technologies, originally developed to workwith CAD representations, to meet the needs of large-scaleearth science simulations using well controlled unstruc-tured meshes. Emphasis will be placed on issues associatedwith the construction of the non-manifold geometric modelneeded by the mesh generators. high-level commands.
Mark S. ShephardRensselaer Polytechnic InstituteScientific Computation Research [email protected]
Cameron SmithRensselaer Polytechnic [email protected]
Mark BeallSimmetrix, [email protected]
MS158
ExtractingNovel Insight from Probabilistic Machine-learnedClassification Catalogs
As the size of astronomical light curve datasets grows, as-tronomers must become further removed from the inferenceworkflow. But instead of labeling a certain source as defi-nitely a member of one class, machine-learned classificationproduces class probability vectors. I discuss the calibrationof probabilistic classification, using taxonometic informa-tion, with an eye towards the ultimate goal: producingnovel scientific insight as a result the classification cata-logs.
Joshua S. BloomUniversity of California, [email protected]
MS158
Never Look Twice: Prefiltering Approaches forDealing with Pesky Amounts of Biological Se-
quence Data
Raw sequencing data is noisy and redundant, but it’s alsoannoyingly large and sometimes difficult to process effi-ciently. Rather than improving the performance of down-stream applications, we have focused on efficient prepro-cessing and prefiltering algorithms that lay a foundationfor streaming online compressive computing in sequenceanalysis. It turns out such algorithms work really well.
C. Titus BrownMichigan State [email protected]
MS158
Electrical Signals in the Human Brain: Computa-tional Challenges
Electrophysiological signals present a unique opportunityto record real-time activity from behaving human brains.While methods for recording large amounts of data haveimproved drastically, computational techniques for mod-eling and analyzing this data still leave much to be de-sired. Such improvements would have implications instudying mechanisms of neural communication and appli-cations such as brain-computer interfaces. This talk willdetail some leading questions in this area, and discuss chal-lenges facing computational cognitive neuroscientists.
Chris HoldgrafUniversity of California, [email protected]
MS158
Managing Large Datasets and Computation Work-flows with Python
Data collection and generation during the last decade hasbecome increasingly easier – so much so that almost allscientific communities need new solutions for processingand managing large datasets. In particular, there exist twomain problems: 1) the compute power of a single machineis not enough to complete tasks in a reasonable amountof time; and 2) the memory of a single machine is notlarge enough to load and process all the data. The recentprogress of Python-based tools for Big Data offers usersnew and novel methods of managing data workflows – toolswhich I believe elevate the novice and give power to thedomain expert. From data storage and management tomodeling and data analysis, these new tools and conceptscan help individuals, small teams, and large organizationsmanage the Big Data deluge.
Benjamin L. ZaitlenContinuum [email protected]
MS159
Robust and Scalable Strategies for Coupling Bio-geochemical Reaction and Solute Transport inEarth System Modeling Frameworks
The Earth fosters a diverse set of physical, chemical, andbiological processes that operate over a range of spatialand temporal scales from nanometers to kilometers and mi-croseconds to millennia. The coupling of these complex andoften uncertain processes can be challenging and impre-cise within numerical models. This presentation illustratescoupling strategies within the petascale reactive flow and
CS13 Abstracts 259
biogeochemical transport code PFLOTRAN and demon-strates the accuracy and parallel performance on real-worldEarth system problems.
Glenn HammondPacific Northwest National [email protected]
MS159
Moment-based Scale-bridgingAlgorithms for Multi-physics Kinetic Systems onEmerging Architectures
The advent of modern parallel architectures, and thepromise of exascale computing resources, brings new chal-lenges to computational science. Dependably taking ad-vantage of billion-way parallelism and distinct levels ofparallelism will stress solver algorithms. We are devel-oping a moment-based multi-physics scale-bridging solveralgorithm to help meet some of this challenge. Develop-ment of similar moment-based acceleration methods can befound in a variety of application areas which are describedby multi-physics kinetic systems. We will provide algo-rithm fundamentals, specific application results and openresearch questions. We will provide some indication of al-gorithm performance on emerging architectures.
Dana Knoll, Luis ChaconLos Alamos National [email protected], [email protected]
G. Chen, C. Newman, H. Park, J. Payne, R. [email protected], [email protected], [email protected],[email protected], [email protected]
J Willert, C.T. [email protected], [email protected]
MS159
Multiphysics Coupling Tools Applied to Large-scale Simulations of a Light Water Nuclear ReactorCore
Obtaining robust and efficient solutions for multiple-time-scale multi-physics systems is challenging. This talk willdescribe a general toolkit under development for use in cou-pling software codes on HPC platforms. We will discuss so-lution algorithms (including Picard iteration, Jacobian-freeNewton-Krylov, and Fully coupled Newton methods), ap-plication interfaces, and data layout/transfer requirementsin the context of HPC platforms. We will show results fora neutronics/CFD coupling for simulating an assembly ina light water nuclear reactor.
Roger PawlowskiMultiphysics Simulation Technologies Dept.Sandia National [email protected]
Roscoe BartlettOak Ridge National [email protected]
John ShadidSandia National LaboratoriesAlbuquerque, [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
MS159
Managing Complexity in Multi-physics Calcula-tions on Modern and Emerging HPC Architectures
Complexity in HPC software stems from two primarysources: complexity in the physics and deployment on het-erogeneous and changing architectures. . Different mod-eling choices imply different coupling between componentsin software; a reflection of the different physical processesbeing described. Trying to support large, evolving codebases on evolving architectures can lead to major softwaremaintenance challenges. We describe two abstractions thatallow these challenges to be addressed in a scalable mannerwithout imposing abstraction penalties. First, we employtask graphs to represent complex multiphysics problems,enabling the programmer to focus on writing individualpieces of software by removing challenges associated withintegrating that into the code base by scheduling computa-tion, memory management, etc. automatically. Secondly,we introduce a domain-specific language embedded in C++that facilitates rapid code development and alleviates theneed to maintain versions of code for execution on serial,multithreaded, or GPU architectures.
James C. SutherlandUniversity of [email protected]
MS160
A Semismooth Newton-CG Method for Full Wave-form Seismic Inversion with Parameter Constraints
We present a semismooth Newton-PCG method for fullwaveform seismic inversion that uses a Moreau-Yosida reg-ularization to handle box constraints on the material pa-rameters and a trust-region globalization. The matrix-free implementation relies on adjoint-based gradient andHessian-vector computations and parallelization with MPI.We analyze the proposed optimization method in a func-tion space setting. Numerical results are shown for an ap-plication to marine geophysical exploration.
Christian BoehmTechnische Universitaet [email protected]
Michael UlbrichTechnische Universitaet MuenchenChair of Mathematical [email protected]
MS160
Performance and Real Data Application of3D Frequency-domain Full-waveform Implemen-tations: Frequency-domain Direct-solver versusTime-domain-based Modeling
3D frequency-domain solutions of the wave-equation for
260 CS13 Abstracts
full waveform inversion (FWI) can be computed with sev-eral methods. We implemented two massively-parallel al-gorithms including the direct-solver and the time-domainmodeling coupled with a discrete Fourier transform for 3Dacoustic FWI. We applied the two approaches to ocean-bottom-cable data from the Valhall field. We discuss theperformances, pros and cons of both methods, which de-pend on the wave physics, the acquisition geometry andthe computing architectures.
Romain BrossierISTerreUniversity Joseph [email protected]
Vincent EtienneGEOAZURUniversity Nice - [email protected]
Guanghui HuISTerreUniversity Joseph [email protected]
Stephane [email protected]
Jean VirieuxISTerreUniversity Joseph [email protected]
MS160
Multi-level Multi-frequency Full Waveform Inver-sion: Convergence and Computation
Abstract not available at time of publication.
Maarten de HoopCenter for Computational & Applied MathematicsPurdue [email protected]
MS161
Efficient Methods for Computing and EstimatingProbability of Failure in PDE Systems
We present an algorithm for computing probability of fail-ure in systems governing by partial differential equations.Since in many practical systems, simulations of the PDEsystems can be highly computationally intensive, it is es-sential to avoid sampling the PDE many times. Our ap-proach is based on the availability of low-fidelity model ofthe system. The low-fidelity model is of low accuracy butcan be simulated fast. One example of low-fidelity model iscoarse grid computation for the PDE system. We present amethod to use a large number of samples of the low-fidelitymodel as a predictor of the failure probability, and then usea small number of samples of the original PDE system asa corrector. The final result of the failure probability canbe of high accuracy, and induces much reduced simulationcost.
Jing Li, Dongbin XiuPurdue [email protected], [email protected]
MS161
Nonintrusive Polynomial Chaos on Nested Un-structured Meshes
We present a novel method for construction of polynomialinterpolation grids on arbitrary Euclidean geometries onwhich there is a probability density. These grids are emi-nently suitable for uncertainty quantification: they are con-structed using the standard polynomial Chaos basis, theyare nested so that refinement strategies can be employed,they are applicable for high-dimensional spaces. Thesegrids do not suffer computational curse-of-dimensionality-type restrictions: the number of nodes in the grid is arbi-trary and user-specified.
Akil NarayanUniversity of Massachusetts [email protected]
MS161
QMC Lattice Methods for PDE with Random Co-efficients
This talk presents recent developments in applying Quasi-Monte Carlo methods (specifically lattice methods) to el-liptic PDE with random coefficients. A guiding exampleis the flow of liquid through a porous material, with thepermeability modeled as a high-dimensional random fieldwith log- normal covariance. In this work we apply the the-ory of lattice methods in weighted spaces to design latticerules with good convergence properties for expected valuesof functionals of the solution of the PDE.
Ian H. SloanUniversity of New South WalesSchool of [email protected]
Frances Y. KuoSchool of Mathematics and StatisticsUniversity of New South [email protected]
James NicholsUniversity of New South Wales, [email protected]
Christoph SchwabETH [email protected]
Ivan G. GrahamUniversity of [email protected]
Robert ScheichlUniversity of BathDepartment of Mathematical [email protected]
MS161
Sufficient Model Reduction on High-dimensionalStochastic Input in Uncertainty Quantification
In order to overcome the curse of dimensionality in high-dimensional stochastic systems, we combine conventionalmodel reduction techniques, such as principle component
CS13 Abstracts 261
analysis (PCA) and manifold learning, that focus on repre-senting stochastic input in a lower-dimensional space withsufficient model reduction (SMR) methods, such as slicedinverse regression and kernel dimension reduction for su-pervised learning, that seek sufficient reduction of predic-tors while preserving the information on responses. In thisway, a high-dimensional stochastic input could be reducedto the minimum number of random variables that can ac-curately predict particular responses of interest.
Nicholas Zabaras, Jiang WanCornell [email protected], [email protected]
MS162
Design Issues for Manycore Application-library In-terfaces
Although there has been much focus on algorithms andsolvers for scalable manycore systems, less effort has beenapplied to understanding the requirements and possibili-ties for the application-library interface. This prioritiza-tion was reasonable, given the critical need for new algo-rithms and the relatively large amount of time spent in thesolvers. However, eventually we need to have a scalablemanycore interface as well. In this presentation we discusssome of the issues and possibilities for application-libraryinterfaces on scalable manycore systems, focusing on dif-ferent ways to think about the roles and responsibilities ofeach component.
Michael A. HerouxSandia National [email protected]
MS162
Porting and Optimizing a Large-Scale CFD Appli-cation with CUDA and OpenACC
Computational Fluid Dynamics (CFD) applications areone of the most important applications executed with high-speed supercomputers. Especially, GPU-based supercom-puters have been showing remarkable performance of CFDapplications. However, GPU-programing is still difficult toobtain high performance, which prevents legacy applica-tions from being ported to GPU environments. We applyCUDA and OpenACC, which are programming languagefor GPU with different features, to a Large-Scale CFD ap-plication UPACS and evaluate these performance.
Tetsuya HoshinoTokyo Institute of [email protected]
Naoya MaruyamaRIKEN Advanced Institute for Computational [email protected]
Satoshi MatsuokaTokyo Insitute of [email protected]
MS162
Multithreading API in the PLASMA Library
The PLASMA numerical library relies on a variety ofmultithreading mechanisms, including static and dynamicthread scheduling. PLASMA’s superscalar scheduler,
QUARK, offers powerfull tools for parallel task composi-tion, such as support for nested parallelism and provisionsfor task aggregation. The dynamic nature of PLASMA’sinternal mode of operation, exposes its user, the applica-tion developer, to an array of new capabilities, such asasynchronous mode of operation, where library functionscalls act like non-blocking MPI communications, i.e. returnbefore work completion. This talk will discuss new op-portunities and difficulties steming from the asynchronousdesign paradigm.
Jakub KurzakUniversity of Tennessee [email protected]
Piotr LuszczekDepartment of Electrical Engineering and ComputerScienceUniversity of Tennessee, [email protected]
Jack DongarraUniversity of [email protected]
MS162
Implementation of FEM Application on GPU withStarPU
We are aiming at accelerating FEM application on multi-core CPUs with GPU. While it isn’t easy to obtain verygood performance because FEM application requires manynon-sequential memory accesses, our GPU implementationhas obtained better performance than multicore CPUs. Weare now trying to accelerate FEM with StarPU runtimesystem. In this talk, we talk about the implementationand performance of it.
Satoshi OhshimaThe University of TokyoInformation Technology [email protected]
Takahiro KatagiriInformation Technology Center, The University of [email protected]
Kengo NakajimaThe University of TokyoInformation Technology [email protected]
Samuel Thibault, Raymond NamystUniversity of Bordeaux 1, [email protected], [email protected]
MS163
Approximate Preconditioners for the UnsteadyNavier-Stokes Equations and Applications toHemodynamics Simulations
We present scalable preconditioners based on approximateversions of efficient preconditioners for the Navier–Stokesequations, namely the Pressure Convection-Diffusion pre-conditioner, Yosida, and SIMPLE. We exploit factoriza-tions of the linearized system where inverses are handledusing specific embedded preconditioners. Weak and strongscalability results illustrate this approach using bench-
262 CS13 Abstracts
marks relevant to hemodynamics simulations. All the com-putations are carried out using the open source finite ele-ment library LifeV based on Trilinos.
Gwenol GrandperrinEPFL LausanneCMCS - [email protected]
Alfio QuarteroniEcole Pol. Fed. de [email protected]
Simone DeparisEcole Politechnique Federale de [email protected]
MS163
An Incompressible Viscous Flow Finite ElementSolver for 1D-3D Coupled Fluid Models
The talk adresses an application of conforming finite ele-ment method (FEM) for a 1D–3D coupled incompressibleflow problem. Coupling conditions are introduced to ensurea suitable bound for the cumulative energy of the model.Motivated by the simulation of flow over an inferior venacava filter, we consider the coupling of a 1D graph and a3D flow domain with highly anisotropic inclusions. Thetetrahedra grid is locally refined and highly anisotropic.We study the stability and the accuracy of the discretiza-tion method and the performance of some state-of-the-artlinear algebraic solvers for such flow configurations.
Maxim A. OlshanskiiDepartment of MathematicsUniversity of [email protected]
Tatiana DobroserdovaMoscow State University [email protected]
MS163
PALADINS: Scalable Time-adaptive AlgebraicSplitting and Preconditioners for the Navier-StokesEquations
We consider a class of second and third order time-accuratealgebaric splitting schemes for the incompressible Navier-Stokes equations featuring a hierarchical structure prone totime-adaptivity. These schemes can be used both as solveror as preconditioner. We discuss some properties and tech-nical details on the implementation of these scheme, called”PALADINS” (Parallel ALgebraic ADaptive Incompress-ible Navier-Stokes). We present scalability results and 3Dapplications to computational hemodynamics.
Umberto E. VillaDept. of Mathematics and Computer ScienceEmory [email protected]
Alessandro VenezianiMathCS, Emory University, Atlanta, [email protected]
MS163
Computational Algorithms for Stability Analysis of
Incompressible Flows
Linear stability analysis of large-scale dynamical systemsrequires the computation of the rightmost eigenvalues of aseries of large generalized eigenvalue problems. Using theincompressible Navier-Stokes as an example, we show thatthis can be done efficiently using a new eigenvalue iterationconstructed using an idea of Meerbergen and Spence thatreformulates the problem into one with a Lyapunov struc-ture. This method requires solution of Lyapunov equa-tions, which in turn entail solutions of subproblems con-sisting of linearized Navier-Stokes systems. We describethe methodology and demonstrate efficient iterative solu-tion of the Navier-Stokes systems.
Howard C. ElmanUniversity of Maryland, College [email protected]
Minghao WuApplied Mathematics ProgramUniversity of [email protected]
MS164
POD-Based Reduced-Order Models for BoussinesqEquations
Model reduction promises to have a significant impact onthe design and control of energy efficient buildings. Thestandard proper-orthogonal decomposition (POD) com-bined with Galerkin projection is not sufficient to pro-duce accurate representations of the airflow inside buildingsat realistic parameter values (modeled by the Boussinesqequations). Therefore, we propose an LES-based closuremodel to reproduce the dissipation effects of the neglectedmodes in nonlinear terms appearing in the momentum andenergy equations. Three-dimensional airflow simulationswill be presented.
Jeff BorggaardVirginia TechDepartment of [email protected]
MS164
Window Proper Orthogonal Decomposition forContinuum and Atomistic Flow Simulations
Proper Orthogonal Decomposition (POD) is a spectralanalysis tool often employed for data postprocessing andpredictive modeling. Here we present a window POD(WPOD, Grinberg et al. ABME 2009) and demonstrateits utility in analysis of flow fields. We review the method-ology and demonstrate application of the WPOD for anal-ysis of transient and space-time intermittent flow regimes.We demonstrate that the time-varying POD eigenvaluespectrum provides an indication of a turbulent or laminarregime.
Leopold GrinbergBrown [email protected]
Alexander YakhotBen Gurion [email protected]
George E. Karniadakis
CS13 Abstracts 263
Brown UniversityDivision of Applied Mathematicsgeorge [email protected]
MS164
Reduced Basis Methods for Viscous Flows: Appli-cation to Inverse Problems in Haemodynamics
We review the current state of the art of the ReducedBasis method for Stokes and Navier-Stokes equations inparametrized geometries, and related a posteriori error es-timation. We present a fully decoupled Offline/Onlineprocedure for both reduced spaces construction and er-ror bounds evaluation, as well as a general framework forthe efficient solution of inverse problems. Furthermore, weshow some numerical results related to applications of in-terest in haemodynamics.
Andrea ManzoniInternational School for Advanced Studies, Trieste, [email protected]
Alfio QuarteroniEcole Pol. Fed. de [email protected]
Gianluigi RozzaSISSA, International School for Advanced StudiesTrieste, [email protected]
MS164
Space-time Error Bounds for Reduced-order Ap-proximations of Parametrized Boussinesq Equa-tions
We present a space-time certified reduced basis methodfor the multi-parameter unsteady Boussinesq equations.We combine a space-time Petrov-Galerkin variational for-mulation, discontinuous Galerkin temporal discretization,Brezzi-Rappaz-Raviart a posteriori error bounds, and hp-adaptive greedy sampling to provide rapid and certifiedcharacterization of the flow over a wide range of parame-ters as well as long integration times. The space-time sta-bility constant constitutes a single quantifiable measure offlow stability which may serve for example in uncertaintyquantification.
Masayuki YanoMassachusetts Institute of Technology, Cambridge, MA,[email protected]
Anthony T. PateraMassachusetts Institute of TechnologyDepartment of Mechanical [email protected]
Karsten UrbanInstitute of Numerical Mathematics, University of [email protected]
MS165
Stable Parareal in Time Method for First and Sec-ond Order Hyperbolic System
The parareal in time algorithm has been implemented onmany types of time dependent problems with some success.
However, some difficulty occurs for problems where the so-lution lacks regularity, either initially or during the evolu-tion, as it is the case for hyperbolic system. In this talkwe propose some way to cure it. We use the new methodto solve a linear wave equation and a nonlinear Burger’sequation, the results illustrate the stability of this variantof the parareal in time algorithm.
Xiaoying DaiInstitute of Computational MathematicsChinese Academy of Sciences, Beijing, [email protected]
Yvon MadayUniversite Pierre et Marie Curieand Brown [email protected]
MS165
Coarse Grid Correction for the Neumann-Neumann Waveform Relaxation Method
The Neumann-Neumann waveform relaxation (NNWR)method for the heat equation converges superlinearly forfinite time windows; however, the number of iterationsrequired for convergence to a fixed tolerance increasesquadratically in the number of subdomains. We showthat by adding a coarse grid correction step, the modifiedmethod converges in two iterations for 1D problems, inde-pendent of the number of subdomains. We also analyze itsconvergence rate for the 2D case.
Felix KwokUniversity of GenevaSection de [email protected]
MS165
Comparing Implementation Strategies for Pararealwith Spatial Parallelization
We explore and compare different implementation strate-gies for a combination of Parareal with spatial parallelism.One is a hybrid approach, using shared memory in timewith distributed memory in space, the other a fully MPI-based approach, using distributed memory and messagepassing in space as well as time. Shared memory in timeavoids messaging volume data and reduces the memoryfootprint of the code, but combining it with message pass-ing in space is not straightforward.
Daniel Ruprecht, Rolf KrauseInstitute of Computational ScienceUniversity of [email protected], [email protected]
MS165
A Large-Scale Space-Time Multilevel Solver for the3D Heat Equation
We present results on the coupling of a space-parallelparabolic multigrid solver with the Parallel Full Approx-imation Scheme in Space and Time (PFASST). By in-tertwining sweeps of spectral deferred corrections withParareal iterations, PFASST allows to obtain parallelizedtime propagators with arbitrary order. We compare theperformance of on- and inter-node parallelization in timeand demonstrate the capabilities of these approaches us-
264 CS13 Abstracts
ing large-scale benchmarks on a recent IBM Blue Gene/Qinstallation.
Robert SpeckJulich Supercomputing CentreForschungszentrum [email protected]
Matthias BoltenUniversity of [email protected]
Rolf KrauseInstitute of Computational ScienceUniversity of [email protected]
MS166
Solving the Non-linear Eigenvector Problem -FEAST-based Alternative to Self-consistent FieldMethods
The non-linear eigenvector problem of the form A[X]X =λBX, such as the one arising in electronic structure calcu-lations, is commonly addressed using a self-consistent fieldmethod (SCF) where a series of linear eigenvalue problemsneed to be solved iteratively until convergence. Here, wepresent an extension of the symmetric FEAST algorithmfor directly addressing the nonlinear eigenvector problem.We show that the new strategy offers a robust and efficientalternative to the traditional SCF methods.
Brendan GavinECE DepartmentUniversity of Massachusetts, [email protected]
Eric PolizziUniversity of Massachusetts, Amherst, [email protected]
MS166
FEAST for the Non-Symmetric Eigenvalue Prob-lem
The FEAST algorithm is extended to address the non-symmetric eigenvalue problem AX = λBX where A andB are real non-symmetric or complex non-Hermitian ma-trices. In order to obtain left and right eigenvectors, wepresent a dual subspace strategy for performing the con-tour integration on the Green’s function in a given regionof the complex plane. All desirable features of the originalalgorithm including robustness, scalability and parallelismare retained.
James KestynECE DepartmentUniversity of Massachusetts, [email protected]
Eric PolizziUniversity of Massachusetts, Amherst, [email protected]
MS166
Parallel Solution of Sparse Rank-deficient Linear
Systems
The talk is devoted to a highly parallel method for solv-ing rank-deficient sparse linear least-square problems onmulti-core platforms. The proposed method is based onthe truncated SVD approach. The algorithm computes anorthonormal basis of the kernel using the FEAST Eigen-value Solver first and then solves an extended system oflinear equations. Intel R© MKL PARDISO is used to solveresulting well-conditioned systems.
Sergey KuznetsovIntel [email protected]
MS166
Subspace Iteration + Approximate Spectral Pro-jection = FEAST
The FEAST algorithm for Hermitian generalized eigen-problem is based on a “density-matrix’ approach. TheFEAST software is well received as it exploits multiplelevels of parallelism that are available on modern architec-tures. This presentation provides some theoretical analy-sis of the FEAST algorithm that is hitherto missing. Thepoint of view is that FEAST is carrying out subspace it-eration with Rayleigh-Ritz procedure on an approximatespectral projector.
Ping T. TangIntel [email protected]
MS167
Tree Decompositions in Logical and ProbabilisticInference
Inference is the problem of answering questions fromknowledge and observations represented in a formal lan-guage or model. Research about efficient computational in-ference procedures applies insights and assumptions aboutlogical languages and probabilistic models. The tree-decomposition abstraction enables structured and fast in-ference by detecting an underlying tree structure in theknowledge or model. In this talk I will present results,insights, applications, and future promise of tree decompo-sition in inference and decision making.
Eyal AmirUniversity of Illinois, [email protected]
MS167
Supernodal and Multifrontal Sparse Matrix Fac-torization
For peak performance, sparse direct methods rely on su-pernodal and multifrontal techniques, which exploit thecliques that form during factorization. These cliques arerelated via the supernodal elimination tree which has thesame properties as trees used in tree decomposition meth-ods for dynamic programming problems on graphs. Thistalk presents supernodal trees, how their treewidth can becomputed efficiently, and the role they play in sparse directmethods, including multifrontal methods on the GPU.
Timothy A. DavisUniversity of FloridaComputer and Information Science and Engineering
CS13 Abstracts 265
MS167
Efficient Algorithms from Graph Structure Theory:Minors, Bidimensionality, and Decomposition
Graph minors, treewidth, and other graph structure theoryenable new approximation and fixed-parameter algorithmsfor graph optimization problems. This talk describes tworecent approaches. Bidimensionality theory provides gen-eral tools for designing fast (constructive, often subex-ponential) fixed-parameter algorithms, kernelizations, andapproximation algorithms (often PTASs), for a wide vari-ety of NP-hard graph problems for graphs excluding a fixedminor. Simplifying decompositions split such a graph intofew small-treewidth graphs via deletions or contractions,also leading to PTASs.
Erik [email protected]
MS167
Tree Decompositions: Adapting Algorithms forParallel Computation
Although many NP-hard graph optimization problemscan be solved in polynomial time on graphs of boundedtreewidth, the adoption of these techniques into main-stream scientific computation has been limited due tothe high memory requirements of required dynamic pro-gramming tables and excessive running times of sequen-tial codes. We discuss how INDDGO, an open sourceOpenMP/MPI software suite, addresses some of these chal-lenges through algorithms and implementations, and whatimpediments still remain.
Blair D. SullivanOak Ridge National [email protected]
MS168
QR Factorizations and SVDs for Tall-and-skinnyMatrices in MapReduce Architectures
The QR factorization and the SVD are two fundamentalmatrix decompositions with applications throughout sci-entific computing and data analysis. Matrices with manymore rows than columns, so-called “tall-and-skinny matri-ces,” are of particular interest. We describe how to com-pute a tall-and-skinny QR factorization on MapReduce ar-chitectures with four different algorithms, and our directTSQR method is the only fast and numerically stable op-tion. We extend the algorithms to compute the SVD withno performance changes. Advisors: James Demmel, UCBerkeley and David Gleich, Purdue University
Austin BensonUC [email protected]
MS168
Restoration and Analysis of Apollo Lunar Data
The Lunar Ejecta and Meteorites Experiment was designedto measure particles that collide with the surface of theMoon. Original analyses revealed more activity during thepassage of the terminator. Recently, there has been spec-
ulation that noise from other instruments may have influ-enced the results. The current analysis examines the datato see if there are correlations between impacts and electri-cal events. Analyses have revealed unusual patterns in allthree sensors of the instrument. Advisor: David Williams,NASA Goddard Space Flight Center
Missy GaddyWofford [email protected]
MS168
Visualization of Cardiac Simulations Using Amira
As the incidence of heart disease in many developed coun-tries continues to rise, research involving computationalmodels of the heart is becoming increasingly important.Various software can be used to visualize and run simula-tions on cardiac models, with the purpose of better under-standing, diagnosing, and treating heart problems. Thisresearch looked at software for segmenting CT and MRIimages to generate 3D cardiac meshes, which were used torun electrical and other simulations. Advisors: RodrigoWeber dos Santos, Angela Shiflet
Jonathan HansonFederal University of Juiz de Fora, [email protected]
MS168
Water Quality Monitoring of Maryland’s Tidal Wa-terways
The Maryland Department of Natural Resources moni-tors the Chesapeake Bay and its tributaries with mon-itoring stations located throughout the tidal waterways.The status of the stations is assessed using the WilcoxonSigned Rank Test. Our simulations demonstrated that log-transforming the data helped to reduce the Type I Error.We ranked the stations using a set of multiple compari-son methods, including the Benjamini-Hochberg rejectionmethod. An interactive GUI is created to facilitate thisdata analysis. Advisor: Nagaraj K. Neerchal
Rosemary LeBrown Universityrosemary [email protected]
MS169
Shock Capturing for High-Order DiscontinuousGalerkin Simulation of Transient Flow Problems
We present some recent result for shock capturing usinghigh-order discontinuous Galerkin schemes. We extend thesensor-based artificial viscosity approach to transient prob-lems, where we demonstrate how weakly coupled sensorsgive a simpler scheme, higher performance, and a more ro-bust behavior. We also show how different levels of smooth-ness in the sensor affect the solutions, for transonic andsupersonic flow problems in 2D and 3D.
Per-Olof PerssonUniversity of California BerkeleyDept. of [email protected]
MS169
H-to-P Efficiently: Achieving Scalable Perfor-
266 CS13 Abstracts
mance using Hybrid Parallelism and Matrix Co-alescence
Spectral/hp element methods utilise a high-order piecewisediscretisation which permits both geometric flexibility andexponential convergence properties to be attained simulta-neously. Some fluid flow problems have one or more ge-ometrically homogeneous coordinate directions and com-putation can be significantly reduced by using a spectraldiscretisation in these directions. In this talk we explorehow hybrid parallelisation of spectral-spectral/hp elementdiscretisations can be optimised for a given problem andparallel environment. Additionally, elemental matrix oper-ations typically achieve poor performance on tuned BLASimplementations due to their small size. We therefore ex-plore how the performance of each process can be furtherimproved through intelligent matrix coalescence.
Spencer Sherwin, Chris CantwellImperial College [email protected], [email protected]
David MoxeyDepartment of AeronauticsImperial College [email protected]
Robert KirbyUniversity of [email protected]
MS169
Recent Development of a High Order Riemann-Solver-Free Space-Time Discontinuous GalerkinMethod
In this talk, new development of a high-order Riemann-solver-free space-time discontinuous Galerkin method willbe reported. The alternate cell-face solution updatingstrategy is employed to replace the original cell-vertexscheme for the convenience of the boundary conditiontreatment. The resulting discontinuous Galerkin cell-facescheme (DG-CFS) preserves the original methods most im-portant features including (i) high-order in both space andtime; (ii) Riemann-solver-free approach and (iii) workingfor both advection and diffusion equations.
Shuangzhang TuJackson State [email protected]
MS169
Adjoint-Based Mesh Adaptation for a Class ofHigh-Order Hybridized Finite-Element Schemesfor Convection-Diffusion Problems
We present a goal-oriented mesh-adaptation methodology,building on a class of hybridized finite-element schemesfor (nonlinear) convection-diffusion problems. Using adiscrete-adjoint approach, sensitivities with respect to out-put functionals of interest are computed to drive the adap-tation. The theoretical framework is embedded in a unifiedformulation of a large class of hybridized, adjoint consis-tent schemes. Furthermore, a shock-capturing methodol-ogy is incorporated, enabling applications in high-speedcompressible flow simulation.
Michael Woopen, Aravind Balan, Georg MayAICES
RWTH [email protected],[email protected], [email protected]
Jochen SchutzIGPMRWTH [email protected]
MS170
Asymptotically Exact DG Error Estimates for Con-vection Problems on Tetrahedral Meshes
We present several superconvergence results and asymp-totically exact a posteriori estimates for discontinuousGalerkin methods applied to convection and convection-diffusion problems. We derive new superconvergence re-sults with accurate error estimates for three-dimensionalhyperbolic problems on tetrahedral meshes. On each ele-ment, the asymptotic behavior of DG errors depends on themesh orientation with respect to the problem characteris-tics. Thus, elements are classified according to the numberof inflow and outflow faces and in all cases enriched finiteelements spaces are needed to show pointwise supercon-vergence. Numerical results are presented to validate thetheory and show the behavior of our error estimates whenthe theory does not apply near shocks and discontinuities.
Slimane AdjeridDepartment of MathematicsVirginia [email protected]
Idir MechaiDepartment of Mathematics, Virginia [email protected]
MS170
A Posteriori Error Estimates for an LDG MethodApplied to Transient Convection-diffusion Prob-lems in One Space Dimension
In this talk, new a posteriori error estimates for a localdiscontinuous Galerkin (LDG) formulation applied to tran-sient convection-diffusion problems in one space dimensionare presented and analyzed. These a posteriori LDG errorestimates are computationally simple and are computedby solving a local steady problem with no boundary con-ditions on each element of the mesh. We first show thatthe leading error term on each element for the solution isproportional to a (p + 1)-degree right Radau polynomialwhile the leading error term for the solution’s derivativeis proportional to a (p + 1)-degree left Radau polynomial.These results are used to prove that, for smooth solutions,these error estimates at a fixed time converge to the truespatial errors in the L2-norm under mesh refinement. Moreprecisely, we show that our LDG error estimates converge
to the true spatial errors at O(〈√+/) rate. Finally, we
prove that the global effectivity indices in the L2-norm con-verge to unity at O(〈∞/∈) rate. Our computational resultsindicate that the observed numerical convergence rates arehigher than the theoretical rates.
Mahboub BaccouchDepartment of MathematicsUniversity of Nebraska at [email protected]
CS13 Abstracts 267
MS170
Adjoint Based Error Estimation for the Lax-Wendroff Method
Hyperbolic problems are of interest in many research ar-eas. A popular class of methods for solving them are finitedifference methods. While these methods have been shownto be quite effective, goal oriented error analysis is difficultto perform due to the fact that the method is not in vari-ational form. In this work we show equivalency between afinite element method and the Lax-Wendroff method, andpresent an adjoint based error representation formula..
James Collins, Don Estep, Simon TavenerColorado State [email protected], [email protected],[email protected]
MS170
Finite Volume Adjoint Error Estimates for WeakSolutions
Many codes exist for hyperbolic equations that employ fi-nite volume methods, which are often nonlinear. A methodis presented for adjoint-based a posteriori error calculationswith finite volume methods. We demonstrate the flexibil-ity in implementation of this post-processing technique anddemonstrate the accuracy with smooth and discontinuoussolutions.
Jeffrey M. ConnorsLawrence Livermore National LaboratoryCenter for Applied Scientific [email protected]
MS171
An IDE Integrated Cross-Platform Build Systemfor Scientific Applications
The goal of our work is to support the development ofscientific and engineering applications while keeping itsperformance portable. To this end, we need to estimateperformances of individual applications on various systemsand detect their non-portable parts based on the estima-tion. Therefore, in this work, we have developed an IDEintegrated cross-platform tool to automatically build ap-plications on remote HPC systems. By enabling the ap-plications to run on various systems, we can improve theirfunctional portabilities, and also expect to improve theirperformance portabilities.
Shoichi HirasawaTohoku Univ./JST [email protected]
Hiroyuki TakizawaTohoku University/JST [email protected]
Hiroaki KobayashiTohoku University, [email protected]
MS171
An Autotuning Framework for Adapting OpenCLKernels to Diverse Architectures
OpenCL promises to enable source code portability across
vendor architectures. However, compiling OpenCL ker-nels for high-performance execution on diverse hardwareis challenging. The widely varying forms of parallelismand memory organizations create a highly non-linear searchspace for compiler transformations. In particular, numer-ous combinations of transformations need to be evaluatedeither explicitly or implicitly in order to settle on the finalconfiguration for a hardware target. In order to addressthis challenge, we have developed an autotuning frame-work for OpenCL programs. It takes an OpenCL kernelas an input and evaluates combinations of compiler trans-formations. We will report the cost and benefit of usingthis autotuning framework in compiling a wide variety ofOpenCL kernels for different OpenCL devices.
Hee-Seok KimUniversity of Illinois at Urbana-Champaign [email protected]
Matthieu [email protected]
Wen-Mei HwuUniversity of Illinois at Urbana-Champaign [email protected]
MS171
Assessing Library Performance with TAU
Autotuning technologies have been successful in concealingthe complex decision-making process and restructuring forobtaining near optimal instances of kernels or libraries ona given platform. Yet, assessing the performance portabil-ity of kernels or libraries on existing and upcoming plat-forms may require the identification of additional parame-ters, and testing the performance characteristics of those.In this presentation we will discuss how performance anal-ysis utilities can be employed to search for complementaryopportunities for improving performance.
Osni A. MarquesLawrence Berkeley National LaboratoryBerkeley, [email protected]
Sameer ShendeDept. of Computer and Information ScienceUniversity of [email protected]
MS171
A Cost-Efficient Approach for Automatic Algo-rithm Selection of Collective Communications
As the size and the complexity of computers increase, se-lection and tuning the technologies for implementing com-munication libraries have become important issues. Thistalk introduces a method that selects a suitable algorithmof collective communications at runtime. In addition to thestatic information such as the number of processes and thesize of message, this method also examines the runtime in-formation such as relative locations of processes to choosean appropriate algorithm efficiently.
Takeshi NanriKyushu University, Japan
268 CS13 Abstracts
JST CREST, [email protected]
Hironobu SugiyamaKyushu [email protected]
Keiichiro FukazawaKyushu University, JapanJST [email protected]
MS172
Scalable Algorithms for Real-Space and Real-TimeFirst-Principle Simulations
The FEAST eigenvalue algorithm is presented beyond the”black-box” solver as a fundamental modeling frameworkfor the electronic structure problem. Within this frame-work, the domain decomposition muffin-tin strategy cannow be performed exactly for the all-electron DFT prob-lem without resorting to traditional approximations such asas linearization and pseudopotential techniques. Addition-ally, the FEAST framework is also ideally suited for per-forming real-time TDDFT calculations using a direct spec-tral decomposition of time-ordered evolution operators.
Eric PolizziUniversity of Massachusetts, Amherst, [email protected]
MS172
Polynomial Filtered Lanczos Algorithms and Spec-trum Slicing
We present a polynomial filtering technique for extractingextreme or interior eigenvalues of large sparse matrices.This general approach can be quite efficient in the situa-tion when a large number of eigenvalues is sought, as isthe case in electronic structure calculations for example.However, its competitiveness depends critically on a goodimplementation. The method presented relies on a com-bination of the Lanczos algorithm with partial reorthogo-nalization and polynomial filtering based on least-squarespolynomials. Details on implementation and a few numer-ical experiments will be presented.
Yousef SaadDepartment of Computer ScienceUniversity of [email protected]
Haw-ren FangDepartment of Computer Science & EngineeringUniversity of [email protected]
MS172
Fast Inversion Methods for NEGF-based Simula-tion of Nanoelectronics Devices with Scattering
We will present numerical methods that allow fast and ro-bust simulations of coupled electro-thermal transport inultra-scaled nanostructures. For that purpose the simula-tion capabilities of the existing quantum transport solverOMEN will be improved by adding electron-phonon andphonon-phonon scattering. We will present numericalmethods for computing all diagonal elements of both the
retarded and lesser Green’s function; As part of this re-search, the selected inversion approach will be extendedto complex and general matrices as needed in the NEGFformalism.
Olaf SchenkDepartment of Mathematics and Computer ScienceInstitute of Computational [email protected]
MS172
Acceleration Techniques for Electronic StructureCalculation
We present a number of techniques for accelerating den-sity functional theory based electronic structure calcula-tion. One of these techniques aims at reducing the com-plexity for evaluating the electron density. In particular,we compute the electron density without diagonalizing theKohn-Sham Hamiltonian. Another technique involves im-proving the convergence of the self-consistent field iterationby developing effective preconditioners.
Chao YangLawrence Berkeley National [email protected]
MS173
Towards an Ultra Efficient Kinetic Scheme for theBoltzmann Equation
In this work we present a new ultra efficient numericalmethod for solving kinetic equations. The key idea, onwhich the method relies, is to solve the collision part ona grid and then to solve exactly the transport linear partby following the characteristics backward in time. Themain difference between the method proposed and semi-Lagrangian methods is that here we do not need to recon-struct the distribution function at each time step. Thisallows to tremendously reduce the computational cost ofthe method and it permits for the first time to computesolutions of full six dimensional kinetic equations on a sin-gle processor laptop machine. Numerical examples, up tothe full three dimensional case, are presented.
Giacomo DimarcoInstitut de Mathematiques de [email protected]
MS173
Inverse Lax-Wendroff Method for Boundary Con-ditions of Boltzmann Type Models
In this talk we present a new algorithm based on Cartesianmesh for the numerical approximation of the kinetic modelson complex geometry boundary. Due to the high dimen-sional property, numerical algorithms based on unstruc-tured meshes for a complex geometry are not appropriate.Here we propose to adapt the inverse Lax-Wendroff proce-dure, which was recently introduced for conservation laws[S. Tan and C.W. Shu], to the kinetic equations. Applica-tions in 1D3D and 2D3D of this algorithm for Boltzmanntype operators (BGK, ES-BGK models) is then presentedand numerical results illustrate the accuracy properties ofthis algorithm.
Francis FilbetUniversity of [email protected]
CS13 Abstracts 269
Chang YangUniversity Claude Bernard, [email protected]
MS173
Conservative Spectral Method for Collision Oper-ators with Anisotropic Scattering
I will present the extension to anisotropic scattering mech-anisms for the conservative spectral method developed byGamba and Tharkabhushnanam, This method can admit awide variety of collisional models, and we study the resultsobtained in the grazing collisions (Landau) limit.
Jeff HaackDepartment of MathematicsUniversity of Texas at [email protected]
Irene M. GambaDepartment of Mathematics and ICESUniversity of [email protected]
MS173
Averaged Kinetic Equations on Graphs
We derive a kinetic equation for flows on directed graphslines with applications to production lines. We collectdata from a company to derive transition probabilitiesand present a kinetic model that allows for a homogeniza-tion procedure, yielding a macroscopic transport model forlarge networks on large time scales.
Michael HertyRWTH Aachen UniverstiyDepartment of [email protected]
MS174
A Mass and Momentum Conserving Discontinu-ous Galerkin Shallow-water Model on the Cubed-sphere
The momentum equations used for the spherical shallow-water (SW) model equations are either in non-conservativeform or in the vector-invariant flux-form with prognosticvariables (u, v, h)T . However, these formulations lack theformal conservation of momentum. A rigorous form of themass and momentum conserving flux-form SW equationswith prognostic variables (uh, vh, h)T on the cubed-spherehas been formulated, which leads to strong form of hy-perbolic SW system. This system has more technical ad-vantages than the usual vector-invariant form. Solving thissystem in non-orthogonal curvilinear geometry (such as thecubed-sphere) is a challenge due to tensorial representa-tion with numerous metric terms. Both set of equationsdiscretized using the discontinuous Galerkin method. Avariety of standard SW tests have been performed for arigorous comparison of conservation of integral invariantssuch as momentum, energy and potential enstrophy. Theresults will be presented in the seminar.
Lei BaoUniversity of Colorado, [email protected]
MS174
Numerical Framework for Atmospheric Modelingon Cubed Sphere by Multi-Moment Scheme
Multi-moment method defines more than one DOFs withineach element to build local high-order schemes. On cubedsphere, compact stencil is beneficial not only to efficientlyexchanging information over different patches, but also toeffectively reducing the excessive errors due to the discon-tinuous coordinate systems along patch boundaries. In thistalk, we will report a multi-moment SWE model up to fifth-order accuracy and a global AMR technique using multi-moment finite volume formulation on cubed sphere.
Chungang ChenXi’an Jiaotong [email protected]
Feng XiaoTokyo Institute of [email protected]
Xingliang Li, Xueshun ShenCenter of Numerical Weather Predication,China Meteorological [email protected], [email protected]
MS174
A Multi-dimensional Fourth-order Accurate Finite-volumeScheme on 3D Cubed-sphere Grids
A fourth-order accurate central essentially non-oscillatory(CENO) finite-volume scheme is presented for hyperbolicconservation laws on 3D cubed-sphere grids. The multi-dimensional approach uses k-exact reconstruction togetherwith a monotonicity procedure that switches between high-order and low-order reconstruction based on a smoothnessindicator. Hexahedral cells with trilinear faces are em-ployed to handle nonplanar cell faces, achieving uniformhigh-order accuracy throughout the cubed-sphere grid.The 3D CENO scheme is implemented in a parallel dy-namically adaptive simulation framework.
Lucian Ivan, Hans De Sterck, Andree SusantoUniversity of WaterlooApplied [email protected], [email protected],[email protected]
Clinton P. GrothUniversity of Toronto Institute for Aerospace [email protected]
MS174
Adaptive Fourth-Order Cubed Sphere Discretiza-tion for Non-Hydrostatic Atmosphere Simulations
We present an adaptive, conservative finite volume ap-proach for non-hydrostatic atmospheric dynamics on a 3Dcubed-sphere mapping with thin shells. Our method isfourth-order accurate in space, and uses higher-order leastsquares-based interpolation to compute stencil operationsnear block and refinement boundaries. Our discretizationis adaptive in both time and horizontal spatial directions,while the radial direction is treated implicitly (using afourth-order RK IMEX scheme) to eliminate time step con-
270 CS13 Abstracts
straints from vertical acoustic waves.
Hans JohansenLawrence Berkeley National LaboratoryComputational Research [email protected]
Phillip Colella, Peter MccorquodaleLawrence Berkeley National [email protected], [email protected]
Paul UlrichUC Davis - [email protected]
MS175
Immersed Boundary Method Simulations of RedBlood Cells
We have developed a variable-viscosity and variable-density Immersed Boundary method that makes effectiveuse of the Fast Fourier transform despite the variable co-efficients in the equations of motion. We have appliedthis method to study red blood cells, which rely on theirremarkable flexibility to squeeze through capillaries. Wedemonstrate that our computations recover the physiolog-ical equilibrium shapes. Further, we describe our simula-tions of shear flow and single-file motion within capillaries.More recent work that includes a realistic model of thespectrin cytoskeleton will also be discussed.
Thomas FaiCourant [email protected]
Boyce E. GriffithLeon H. Charney Division of CardiologyNew York University School of [email protected]
Yoichiro MoriSchool of MathematicsUniversity of [email protected]
Charles S. PeskinCourant Institute of Mathematical SciencesNew York [email protected]
MS175
A Narrow-band Gradient-augmented Level SetMethod for Incompressible Two-phase Flow
We have incorporated the gradient-augmented level setmethod (GALSM) for use in two-phase incompressible flowsimulations by interpolating velocity values and introduc-ing a re-initialization procedure. The method is conductedon a narrow band around the interface, reducing computa-tional effort, while maintaining an optimally local advec-tion scheme and providing sub-grid resolution. Ocean wavesimulation is the primary motivation, and numerous bench-marks have been conducted, including a new comparisonwith wave tank data.
Curtis Lee, John DolbowDuke [email protected], [email protected]
Peter J. MuchaUniversity of North [email protected]
MS175
Adjoint-Based Error and Sensitivity Analysis inTransport-Depletion Problems
We describe the implementation and testing of an ad-joint solver for the neutron/nuclide depletion equations.We do this in a flexible, multi-physics friendly frameworkfor computing uncertainty and sensitivity information viathe adjoint variable and describe our implementation in amassively-parallel transport solver. We then discuss per-formance trade-offs, algorithmic challenges, and scaling re-sults as we push the problem towards large-scale simula-tions.
Hayes Stripling, Marvin AdamsTexas A&M [email protected], [email protected]
Mihai AnitescuArgonne National LaboratoryMathematics and Computer Science [email protected]
MS175
Efficient, Parametrically-Robust Nonlinear ModelReduction using Local Reduced-Order Bases
The large computational cost associated with high-fidelityphysics-based simulations has limited their use in manypractical situations. Model Order Reduction (MOR) is anattempt to reduce the computational cost of such simu-lations by searching for an approximate solution in well-chosen, low-dimensional affine subspaces. Here, I presentrecent developments in nonlinear MOR theory that achieveincreased parametric robustness and additional speedup.These new techniques will be applied to fluid mechanicaltest problems.
Matthew J. ZahrUniversity of California, BerkeleyStanford [email protected]
David Amsallem, Charbel FarhatStanford [email protected], [email protected]
MS176
Bootstrapping Big Data in the Cloud
Bag of Little Bootstraps (BLB) is a recently developedand easily parallelizable algorithm which assesses the qual-ity of a statistical estimator function on a large data set.Via Selective Embedded Just-in-Time Specialization (SE-JITS), we dynamically convert such input high level esti-mator functions to low level, efficient Scala code to thendeploy BLB to the cloud. This allows domain experts toquickly code BLB applications, thus providing for scalableand swifter manipulation of big data.
Peter BirsingerUniversity of California, [email protected]
CS13 Abstracts 271
MS176
Big Data? Never Ask The Same Question Twice
The title refers to a holonic approach to the design of al-gorithms so that the same record never has to be readtwice. With examples of application and their economicimpact founded in LEGOs planning & logistic operations,this paper illustrates the relevance for real-time decision-making in an irreversible economic context, where the data-set evolves in discrete iterations. As aggregate run-timewith discretely evolving data-sets appears to be absent inthe debate on algorithmic design, the paper encouragesfurther research in this area.
Bjorn MadsenLEGO [email protected]
MS176
Spatial Analysis and Big Data: Challenges and Op-portunities
We share our experiences as developers of new spatial an-alytical methods with a heavy social science perspectiveand identify new challenges that arise from ”big data” inthese realms. Much of the current stack of spatial analysis,statistics, and econometric software has not taken full ad-vantage of new HPC environments due to a general lack ofconsensus on best practices. By engaging with the widercommunity of computational scientists we hope to identifypaths forward.
Sergio Rey, Luc AnselinGeoDa Center for Geospatial Analysis and ComputationArizona State [email protected], [email protected]
MS176
yt: Massively Parallel Astrophysical SimulationAnalysis Made Easy
Astrophysical simulations have reached the point where thechallenge of running ever-larger simulations is becomingovershadowed by the task of analyzing their data. Wepresent recent enhancements to the yt analysis toolkitaimed at the analysis of large datasets, including multi-level parallelism and in-situ analysis capable of communi-cating back to running simulations. The goal is to providea toolkit that allows the user to craft functionality that canbe easily made to work in parallel.
Britton SmithMichigan State [email protected]
Matthew TurkColumbia [email protected]
MS177
Experiences with Mini-Apps on Intel’s Xeon PhiArchitecture
Intel’s recently announced Xeon Phi architecture repre-sents a blend of features designed for high performancecomputing including increased length of vector processingunits, increased thread counts and greater core density.Since 2011 Sandia National Laboratories has been working
with Intel and several Office of Science Co-Design centersto port, analyze and optimize the performance of importantmini-applications and libraries on the Xeon Phi architec-ture. In this talk we present the experiences gained fromthese exercises and thoughts of the next steps we will taketo improve performance still further.
Simon D. HammondScalable Computer ArchitecturesSandia National [email protected]
Richard BarrettSandia National [email protected]
MS177
Introduction to Stampede and the Intel MIC Ar-chitecture
In this talk, we will introduced the Stampede supercom-puting system deployed at TACC over the course of thesummer and fall of 2012. Stampede is a large-scale Linuxcluster which is first to deploy a significant number of IntelXeon Phi co-processors using the Intel Many IntegratedCore Architecture (MIC). We will present the Stampedearchitecture, programing models, and some early resultsenabled by the Xeon Phi coprocessors.
Bill BarthTexas Advanced Computing CenterUniversity of Texas at [email protected]
MS177
A Unified Approach to Heterogeneous Architec-tures using the Uintah Framework
Uintah is a large-scale, parallel multi-physics framework tosimulate fluid-structure interaction problems on structuredAMR grids using the ICE flow solver and Material PointMethod. Uintah now scales to 256k cores on Jaguar byusing hybrid MPI/multi-threaded parallelism. A recentlydeveloped heterogeneous runtime system further enablescomputational tasks to be offloaded to accelerators. Inthis work we demonstrate and analyze the performance ofUintah using both native and offload modes on the IntelMIC co-processor.
Alan HumphreyScientific Computing and Imaging InstituteUniversity of [email protected]
Qingyu MengSCI InstituteUniveristy of [email protected]
Martin BerzinsScientific Computing and Imaging InstituteUniversity of [email protected]
MS177
MPI Communication on Stampede with MIC using
272 CS13 Abstracts
MVAPICH2: Early Experience
TACC Stampede is the first large-scale deployment ofIntel Many Integrated Core (MIC) architecture. MVA-PICH2 is one of the most widely-used open-sourceMPI li-braries on high performance computing clusters with In-finiBand. This talk will present our early experience in en-hancing MVAPICH2 for TACC Stampede with Intel MIC.An overview of the design enhancements (point-to-pointand collective communication) and their impact on perfor-mance of end applications will be presented.
Dhabaleswar K. Panda, Sreeram Potluri, DevendarBureddyThe Ohio State [email protected], [email protected],[email protected]
MS178
A Flexible and Extensible Multi-process Simula-tion Capability for the Terrestrial Arctic
The frozen soils of the Arctic contain vast amounts ofstored organic carbon which is vulnerable to release as tem-peratures warm and permafrost degrades. The critical pro-cess models required for simulating degradation and releaseinclude subsurface thermal hydrology of freezing/thawingsoils, thermal processes within ice wedges, mechanical de-formation processes, overland flow, and surface energy bal-ances including snow dynamics. The Arctic TerrestrialSimulator, based upon Amanzi, is being developed to en-able flexible experimentation in coupling these processes.
Ethan T. Coon, Gianmarco Manzini, Markus Berndt,Rao V. GarimellaLos Alamos National [email protected], [email protected],[email protected], [email protected]
David MoultonLos Alamos National LaboratoryApplied Mathematics and Plasma [email protected]
Scott PainterLos Alamos National [email protected]
MS178
Sundance: High-level Components for Automationof PDE Simulation Development
We describe Sundance, a component toolkit for automatingthe assembly of high-performance parallel PDE simulationsfrom high-level problem descriptions. We show how com-pact, high-level representations of a problem’s symbolicform and geometry can be interpreted to coordinate theaction of efficient low-level computational kernels.
Kevin Long, Robert C. KirbyTexas Tech [email protected], Robert [email protected]
MS178
Multilevel Preconditioner Components for Multi-
Physics Problems on Adaptively Refined Meshes
Newton-Krylov and nonlinear Krylov methods for multi-physics problems require effective preconditioners for effi-ciency. Many multi-physics problems discretized on struc-tured AMR grids require the solution of elliptic subcom-ponents as part of the preconditioner. Multilevel solutionmethods for elliptic problems that can exploit the naturalmultilevel hierarchy of structured AMR grids are discussed.We will describe the performance of both synchronous andasynchronous multilevel methods particularly in the con-text of nonlinear radiation diffusion problems.
Bobby Philip, Zhen WangOak Ridge National [email protected], [email protected]
Manuel Rodriguez Rodriguez, Mark [email protected], [email protected]
MS178
Cache-aligned Data Structures for UnstructuredMeshes
Abstract not available at time of publication.
Daniel SunderlandSandia National [email protected]
MS179
High-Performance Filtered Queries in AttributedSemantic Graphs
An analytic query views an attributed semantic graphthrough a filter that passes only edges of interest. In ourKnowledge Discovery Toolbox, an open-source system forhigh-perfomance parallel graph computation, the user candefine a Python filter that applies to every graph operation.We address the performance challenge of per-edge filter-ing with selective embedded specialization, automaticallytranslating filters into an efficiency language, CombBLAS.This greatly accelerates Python KDT graph analytics onclusters and multicore CPUs.
John R. GilbertDept of Computer ScienceUniversity of California, Santa [email protected]
Aydin BulucLawrence Berkeley National [email protected]
Armando FoxUniversity of California, [email protected]
Shoaib KamilMassachusetts Institute of [email protected]
Adam LugowskiUC Santa [email protected]
Leonid Oliker, Samuel Williams
CS13 Abstracts 273
Lawrence Berkeley National [email protected], [email protected]
MS179
Large-Scale Graph-Structured Machine Learning:GraphLab in the Cloud and GraphChi in your PC
Abstract not available at time of publication.
Carlos Guestrin, Joseph GonzalezCarnegie-Mellon [email protected], [email protected]
MS179
Are We There Yet? When to Stop a Markov Chainwhile Generating Random Graphs?
Markov chains are commonly used to generate random net-works through rewiring of its edges. However, provablebounds on the number of iterations required for an inde-pendent instance are impractical. Practitioners end up us-ing ad hoc bounds for experiments. In this work, we willpresent our methods for computing practical bounds onthe number of iterations, and present experimental resultsfor generating graphs with a given degree distributions andjoint degree distribution.
Ali PinarSandia National [email protected]
Jaideep RaySandia National Laboratories, Livermore, [email protected]
C. SeshadhriSandia National [email protected]
MS179
Analyzing Graph Structure in Streaming Data withSTINGER
Analyzing static snapshots of massive, graph-structureddata cannot keep pace with the growth of social networks,financial transactions, and other valuable data sources.Our software framework, STING (Spatio-Temporal In-teraction Networks and Graphs), uses a scalable, high-performance graph data structure to enable these appli-cations. STING supports fast insertions, deletions, andupdates on graphs with semantic information and skeweddegree distributions. STING achieves large speed-ups overparallel, static recomputation on both common multicoreand specialized multithreaded platforms.
Jason RiedyGeorgia Institute of TechnologySchool of Computational Science and [email protected]
David A. Bader, Robert C. MccollGeorgia Institute of [email protected], [email protected]
David EdigerGeorgia Institute of TechnologySchool of Computational Science and [email protected]
MS180
Open Access for Models in Molecular Biophysics –Progress and Challenges
Abstract not available at time of publication.
Jaydeep P. BardhanDepartment of Electrical and Computer EngineeringNortheastern Universityjaydeep [email protected]
MS180
Open Science in Molecular Simulations
Abstract not available at time of publication.
Ahmed IsmailDepartment of Mechanical EngineeringRWTH Aachen [email protected]
MS180
How to Succeed in Reproducible Research withoutReally Trying
Abstract not available at time of publication.
Geoffrey M. OxberryLawrence Livermore National [email protected]
MS180
Exorcising Numerical Ghosts from ab initio Calcu-lations of Electron Transport
The marriage of electron transport models with quantumchemistry codes has enabled the use of computation for ex-ploring electron transport. Unfortunately, ”better” basissets often unphysically increase the calculated properties.The cause of this ”ghost transmission” has proven elusive,in part due to the numerous, sometimes implicit, approx-imations invoked in the models. In this talk I diagnoseghost transmission, emphasize the importance of unit test-ing, and develop an open-source, ghost-busted model forelectron transport.
Matthew ReuterEugene P. Wigner FellowOak Ridge National [email protected]
MS181
Full Waveform Inversion Across the Scales
We present a newly-developed full waveform inversionscheme that incorporates seismic data on multiple spatialscales. Based on a decomposition of a multi-scale Earthmodel into various single-scale sub-models via 3D non-periodic homogenisation, the method allows us to simul-taneously resolve the details of both the Earths crust andmantle. We demonstrate the applicability and efficiency ofour technique in a multi-scale full waveform inversion forEurope and western Asia.
Andreas FichtnerUtrecht [email protected]
274 CS13 Abstracts
Paul CupillardIPG [email protected]
Erdinc SayginThe Australian National [email protected]
Yann CapdevilleUniversite de [email protected]
Tuncay TaymazIstanbul Technical [email protected]
Antonio VillasenorBarcelona Center for Subsurface [email protected]
Trampert JeannotUtrecht [email protected]
MS181
Dimensionality Reduction in FWI
Abstract not available at time of publication.
Felix J. HerrmannSeismic Laboratory for Imaging and ModelingThe University of British [email protected]
MS181
Projected-gradient Schemes and Sharp Interfacesin 3D Seismic Inversions
Abstract not available at time of publication.
Tarje Nissen-MeyerETH [email protected]
Loredana GaudioUniversity of [email protected]
Max RietmannUniversity of [email protected]
Olaf SchenkDepartment of Mathematics and Computer ScienceInstitute of Computational [email protected]
Piero BasiniUniversity of [email protected]
MS181
Elastic and Anelastic Structure of the EuropeanUpper Mantle based on Adjoint Tomography
Harnessing high-performance computers and numerical
methods to constrain physical properties of Earth’s interioris becoming one of the most important topics in structuralseismology. We use spectral-element and adjoint methodsto iteratively improve 3D elastic and anelastic structureof the European upper mantle. This study involves twostages: in stage one, only phase measurements are em-ployed to investigate radial anisotropic elastic structure.In stage two, both phase and amplitude measurements areapplied to simultaneously constrain elastic and anelasticstructure.
Hejun ZhuDepartment of GeosciencesPrinceton [email protected]
MS182
Error Analysis for Galerkin POD Approximationof the Nonstationary Boussinesq Equations
Abstract not available at time of publication.
S.S. RavindranUniversity of AlabamaHuntsville, [email protected]
MS182
Reduced Order Modeling of Buoyancy Driven In-compressible Flows
In this talk, we apply proper orthogonal decomposi-tion based reduced order modeling (ROM) to the time-dependent Boussinesq equations:
∂tu− νeΔu + (u · ∇)u + ∇p = −βθg,
nabla · u = 0,
partialtθ − aeΔθ + (u · ∇)θ = c−1p qv.
We discuss the function space setting and variational solu-tion of the equations and how it relates to the ROM prob-lem. We offer computational illustrations which indicatethat for buoyancy driven flows, the ability of the ROM toreproduce the flow depends upon both the space utilizedfor the velocity and that for the temperature. We aimto utilize ROM to reproduce the solution from a parallelBoussinesq solver which uses a Chorin projection method.
John P. RoopDepartment of MathematicsNorth Carolina A&[email protected]
MS182
New POD Error Expressions, Error Bounds,and Asymptotic Results for Model Reduction ofParabolic PDEs
The derivations of existing error bounds for POD-basedreduced order models of time varying partial differentialequations have relied on bounding the error between thePOD data and various POD projections of that data. Weprove that these data approximation errors can be com-puted exactly, using only the POD eigenvalues and modes.We apply our results to derive new POD model reductionerror bounds and convergence results for the two dimen-sional Navier-Stokes equations.
John Singler
CS13 Abstracts 275
Missouri S & TMathematics and [email protected]
MS182
Proper Orthogonal Decomposition Reduced-OrderModels of Complex Flows
In numerical simulations of complex flows, model reduc-tion techniques are frequently used to make the compu-tation feasible. Proper orthogonal decomposition is oneof the most commonly used methods to generate reduced-order models for turbulent flows dominated by coherentstructures. To balance the low computational cost requiredby a reduced-order model and the complexity of the tar-geted turbulent flows, appropriate closure modeling strate-gies need to be employed. In this talk, we will presentnew nonlinear closure methods for proper orthogonal de-composition reduced-order models, develop rigorous errorestimates, and design efficient algorithms for them. Appli-cations of the closure models in realistic engineering prob-lems such as energy efficient building design and controlwill also be discussed.
Zhu WangUniversity of [email protected]
MS183
An Overview of Computational Methods for theDynamics of Schrdinger Equations
The Schrodinger equation is a Partial Differential Equationthat can be met in different areas of physics and engineer-ing. Its most well-known use concerns quantum mechanics.Other fundamental applications are related to the study ofBose-Einstein condensates where the understanding of theGross-Pitaevskii equation plays a crucial role. The numeri-cal approximation of such equations is therefore fundamen-tal. In this talk, we will review various solutions to buildefficient numerical schemes, describe their properties andshow their efficiency in various conditions.
Christophe BesseUniversite des Sciences et Technologies de [email protected]
Xavier L. AntoineInstitut Elie Cartan Nancy (IECN)Universite de [email protected]
Weizhu BaoNational University of SingaporeDepartment of [email protected]
MS183
A Posteriori Error Control and Adaptivity forSchrodinger Equations
We derive optimal order a posteriori error bounds forCrank-Nicolson fully discrete approximations for linearSchrodinger equations. For the spatial discretization weuse finite element spaces that are allowed to change in time.The a posteriori error estimates are established using ap-propriate time-space reconstructions and a modified ellipticreconstruction that leads to estimators, which reflect cor-
rectly the physical properties of the continuous problem.Our theoretical results are validated by numerical experi-ments on various model problems.
Irene [email protected]
Theodoros KatsaounisAssociate Professor, Department of Applied MathematicsUniversity of [email protected]
MS183
An Asymptotic Preserving Numerical Method forthe Nonlinear Schrodinger Equation in the Semi-classical Regime
This is a joint work with Remi Carles and Christophe BesseWe present a new decomposition ”a la Grenier” for the de-focusing nonlinear Schrodinger equation in the semiclassi-cal regime. We prove the local well-posedness of the system(and its global well-posedness in dimension one) and itsconvergence to the semiclassical limit before shocks appearin the limit system. Moreover, we construct a numericalscheme which is Asymptotic-Preserving, i.e. which is con-sistent and of order two when the semiclassical parameterε is fixed, and which degenerates when ε → 0 into a con-sistent numerical scheme for the limit system.
Florian MehatsIRMAR - Universite de Rennes [email protected]
Remi CarlesCNRS & Universite Montpellier [email protected]
Christophe BesseUniversite des Sciences et Technologies de [email protected]
MS183
Recent Developments of Fast Algorithms for HighFrequency Waves in the Semi-classical Regime
I will review some recent developments on fast algorithmsfor computing high frequency waves in the semi-classicalregime.
Jianliang QianDepartment of MathematicsMichigan State [email protected]
MS184
A Greedy Strategy for Sparse Approximation ofPDEs with High-dimensional Random Inputs
This talk is concerned with the sparse approximation ofPDEs with high-dimensional random inputs where thequantity of interest (QOI) admits a sparse polynomialchaos (PC) expansion. We discuss a greedy strategy toselect the PC basis functions whose coefficients are ”large”from realizations of the QOI. We demonstrate the conver-gence of the proposed approach when the realizations ofQOI are generated via Monte Carlo sampling as well as a
276 CS13 Abstracts
more efficient sequential sampling design.
Jerrad HamptonUniversity of Colorado, [email protected]
Alireza DoostanDepartment of Aerospace Engineering SciencesUniversity of Colorado, [email protected]
MS184
Dimension Reduction in Nonlinear Statistical In-verse Problems
The Bayesian approach to inverse problems in principlerequires posterior sampling in high or infinite-dimensionalparameter spaces. However, the intrinsic dimensionalityof such problems is affected by prior information, limiteddata, and the smoothing properties of the forward oper-ator. Often only a few directions are needed to capturethe change from prior to posterior. We describe a methodfor identifying these directions through the solution of ageneralized eigenvalue problem, and extend it to nonlin-ear problems where the data misfit Hessian varies over pa-rameter space. This scheme leads to more efficient Rao-Blackwellized posterior sampling schemes.
James R. MartinUniversity of Texas at AustinInstitute for Computational Engineering and [email protected]
Tiangang Cui, Tarek [email protected], [email protected]
Omar GhattasUniversity of Texas at [email protected]
Youssef M. MarzoukMassachusetts Institute of [email protected]
MS184
High-dimensional Polynomial Chaos Basis Selec-tion with Bayesian Compressive Sensing
For a complex model with a large number of input param-eters, building Polynomial Chaos (PC) surrogate modelsis challenged by insufficient model simulation data as wellas by a prohibitively large number of spectral basis terms.Bayesian sparse learning approaches are implemented inorder to detect a sparse polynomial basis set that bestcaptures the model outputs. We enhanced the Bayesiancompressive sensing approach with adaptive basis growthand with a data-driven, piecewise-PC surrogate construc-tion.
Khachik Sargsyan, Cosmin SaftaSandia National [email protected], [email protected]
Bert J. DebusschereEnergy Transportation CenterSandia National Laboratories, Livermore [email protected]
Habib N. NajmSandia National LaboratoriesLivermore, CA, [email protected]
Dan Ricciuto, Peter ThorntonOak Ridge National [email protected], [email protected]
MS184
Mori-Zwanzig Approach to Nonlinear Stochas-tic Differential Equations with High-dimensionalParametric-type Uncertainty
We propose a new approach to obtain exact PDF equa-tions for low-dimensional nonlinear functionals of the solu-tion to high-dimensional stochastic differential equations,including SODEs and SPDEs. The new method does notsuffer from the curse of dimensionality and, in principle, itallows us to avoid the integration of the full stochastic sys-tem, and solve directly for the PDF of the low-dimensionalfunctional we are interested in.
Daniele VenturiBrown Universitydaniele [email protected]
George E. KarniadakisBrown UniversityDivision of Applied Mathematicsgeorge [email protected]
MS185
Global Modes for Controller Selection and Place-ment in Compressible Turbulent Flows
A method for estimating the optimal location and typeof flow control to use in compressible, turbulent flows isdeveloped. Through linearizing the compressible Navier-Stokes equations about an unstable equilibrium point, theforward and adjoint equations of motion are used to esti-mate the structural sensitivity of the flow using a general-ized “wavemaker’ concept. Matrix optimization is used toenhance the structural sensitivity over possible controllertypes (e.g., mass or energy sources) and locations, with op-timal solutions identifying advantageous control strategies.The algorithms and theory are applied to a high-subsonicseparating boundary layer appearing in an S-duct andcompared with more traditional optimal methods throughadjoint-based gradient information. It is found that opti-mum locations for mass and energy sources are typicallylocated upstream of those momentum sources, and use dif-ferent physical mechanisms for affecting the flow.
Daniel J. BodonyUniversity of Illinois at [email protected]
MS185
Optimal Active Separation Control on Airfoils us-ing Discrete Adjoint Approach
Blowing and suction type of active flow control techniquesare being widely used to prevent or delay the flow sep-aration in order to enhance the performance of aerody-namic configurations. Typically, this type of flow control isachieved by varying the actuation parameters such as am-plitude, frequency, position and direction of blowing and
CS13 Abstracts 277
suction. An efficient way of finding the optimal set of ac-tuation parameters is by using the adjoint based optimi-sation methods. In the present work, a discrete adjointsolver has been developed for the optimal active controlof unsteady incompressible viscous flows by using the Au-tomatic/Algorithmic Differentiation (AD) techniques. Anadvantage of this approach is that the discrete realisationsof the turbulence models are algorithmically differentiableand hence the frozen turbulence assumption is not requiredas it is often necessary for continuous adjoint methods. Theadjoint code is applied to the optimisation problem of liftmaximisation of a NACA 4412 airfoil using sinusoidal blow-ing and suction. Numerical results based on discrete ad-joints are compared with the continuous adjoint approach.
Nicolas R. Gauger, Anil NemiliDepartment of Mathematics and CCESRWTH Aachen [email protected],[email protected]
Emre Oezkaya, Stefanie GuentherRWTH Aachen [email protected],[email protected]
MS185
Shape Calculus and Unsteady PDE Control
Optimization and optimal control is considered for prob-lems where the shape is the actual unknown to be found.Exploiting the structure of such problems leads to sur-face formulations of the gradient that circumvent most ofthe problems usually arising in shape optimization, suchas computing sensitivities of the mesh deformation. Thetalk features applications of this technique in areas of wavepropagation problems and computational fluid dynamics.
Stephan SchmidtImperial College [email protected]
MS185
Towards Design and Optimization in Periodic andChaotic Unsteady Aerodynamics
This talk focus on challenges in gradient based optimiza-tion for both periodic unsteady flows and chaotic, aperi-odic unsteady flows. Checkpointing-based adjoint methodfor sensitivity analysis will first be introduced, followed bydemonstration of the sensitivity divergence problem thataeries when the objective function is a long time averagedquantity. The fundamental cause of this problem is ana-lyzed, and solutions to this problem, including the recentlydeveloped windowing method and Least Squares Sensitiv-ity method, are introduced.
Qiqi WangMassachusetts Institute of [email protected]
MS186
Sharing Thread Pools and Caches for Inter-libraryComposition and Multicore Performance
PETSc’s multicore approach involves a thread communi-cator that can run on the user’s choice of threading models
and provides weak structured synchronization primitives.This is necessary to support applications and other librariesthat were implemented using various programming mod-els such as raw pthreads, OpenMP, and TBB. For per-formance, affinity must be managed and data structuresorganized so that cores can effectively utilize bandwidthand share caches while avoiding contention.
Jed BrownMathematics and Computer Science DivisionArgonne National [email protected]
MS186
Manycore Performance Portability throughMapped Multidimensional Arrays
Performance on manycore-accelerators is dependent uponarchitecture-specific data access patterns. The perva-sive question of whether to use arrays-of-structures orstructures-of-arrays is a fundamentally wrong question.The correct question is “what data structure abstractionis required to achieve manycore performance-portability?’Our answer is “device-mapped multidimensional arrays.’The KokkosArray library transparently inserts, at compiletime, device-specific multidimensional array maps via con-ventional C++ template meta-programming. CPU/GPUperformance-portability is demonstrated through MPI +KokkosArray hybrid parallel FEM proxy-applications.
H. Carter EdwardsSandia National [email protected]
MS186
Multilevel Programming Paradigms for Smart-tuned Exascale Computational Science
Exascale ”hypercomputers” are expected to have highlyhierarchical architectures with nodes composed by ”lot of-core” processors and accelerators. The different program-ming levels (from clusters of processors loosely connectedto tightly connected ”lot-of-core” processors and/or accel-erators) will generate new difficult algorithm issues. Newlanguage and framework should be defined and evaluated.In this talk, we present multilevel programming paradigmsfor exascale computing and propose example based onYML.
Serge G. Petiton
CNRS/LIFL and [email protected]
MS186
A Hierarchical Parallel Implementation of a Con-tour Integral-based Eigensolver on Trilinos
We consider solving large-scale sparse eigenvalue problems.An eigensolver based on contour integral has been pro-posed by Sakurai and Sugiura. This solver has a hierar-chical structure and is suitable for massively parallel su-percomputers. In this solver, linear systems with multipleright-hand sides need to be solved. Trilinos includes usefulpackages for solving such kind of linear systems. In thistalk, we present an implementation of the eigensolver onTrilinos and show its performance.
Tetsuya SakuraiDepartment of Computer Science
278 CS13 Abstracts
University of [email protected]
Yasunori FutamuraUniversity of [email protected]
Lei DuUniversity of Tsukuba, [email protected]
Hiroto TadanoDepartment of Computer ScienceUniversity of [email protected]
MS187
Optimal Control of Variable Density Navier-StokesEquations
Optimal control problems for partial differential equa-tions arise in various applications,such as in engineer-ing,tomography and finance. Here we are concerned withthe Navier-Stokes equation as state equation and a sourcecontrol function to obtain a desired solution,also referredto as a velocity tracking problem. Using Tikhonov regular-ization and a Lagrange multiplier,a saddle point operatorsystem arises where we can use equal order discretizationfor the state variable,the control function and the Lagrangemultiplier. This enables elimination of the control function.The reduced system takes a particular two-by-two blockmatrix form for which we can construct a special precondi-tioner, the inverse of which involves just the inverses of twomatrices that are linear combinations of the block matri-ces in each row of the given matrix.In the N-S problem,theare a mass matrix and an advection perturbed diffusionmatrix. There is no need to solve any Schur complementsystem.The resulting eigenvalues are real and positive witha small condition number, which holds uniformly with re-spect to both discretization and method parameters.
Owe Axelsson, He Xin, Maya NeytchevaDepartment of Information TechnologyUppsala [email protected], [email protected],[email protected]
MS187
Efficient Augmented Lagrangian-type Precondi-tioning for the Oseen Problem using Grad-Div Sta-bilization
Grad-Div stabilization can be exploited in a preconditionerfor the Oseen Problem. It turns out that it behaves sim-ilar to the classical augmented Lagrangian approach, butwith the advantage of being able to easily construct thesystem matrix efficiently. This simplifies the constructionof inner preconditioners. I will discuss the difficulty of thetrade-off between solution accuracy from stabilization andsolver efficiency. Finally I will present numerical results todemonstrate the preconditioner.
Timo HeisterTexas A&M UniversityDepartment of [email protected]
MS187
Performance of SIMPLE-type Preconditioners inCFD Applications for Maritime Industry
CFD applications in maritime industry, for examplehull resistance prediction, involve high Reynolds num-ber flows modelled by the incompressible Reynolds-averaged Navier-Stokes equations. The system of equa-tions is discretized with a cell-centered finite-volumemethod with colocated variables. After linearization,various SIMPLE-type preconditioners can be appliedto solve the discrete system. In this presentation,we discuss their performance for flows with Reynoldsnumber up to 109 and cell aspect ratio up to 106.http://ta.twi.tudelft.nl/nw/users/vuik/papers/Kla12V.pdf
Chris KlaijMaritime Research Institute [email protected]
C VuikFaculty of Electrical Engineering, Mathematicsand Computer Science, Delft University of Technology,[email protected]
MS188
Solution of Parabolic Inverse Coefficient ProblemVia Reduced Order State Equations
We introduce a method for numerical solution of a one-dimensional parabolic inverse coefficient problem with timedomain data. The problem is formulated as non-linear leastsquares optimization. Unlike the traditional output leastsquares approach, the proposed method employs non-linearpreconditioning, which greatly improves the conditioningof the optimization functional and its convexity. As a re-sult, the method attains high quality reconstructions injust a few Gauss-Newton iterations. The construction ofthe non-linear preconditioner is based on ideas from modelorder reduction and rational approximation. We study dif-ferent choices of matching conditions for projection basedmodel order reduction and osculatory rational interpola-tion. Among those choices we identify those that corre-spond to a non-linear preconditioner with desired proper-ties. Performance of the proposed method is evaluated ina number of numerical experiments involving both smoothand discontinuous coefficients.
Alexander V. MamonovRice [email protected]
MS188
Stability-Corrected Extended Krylov Method forWavefield Problems in Unbounded Domains
In this talk, we present a new Extended Krylov SubspaceMethod for exterior wavefield problems. We start by show-ing that the solution of such problems can be expressedin terms of stability-corrected operator exponents. Sub-sequently, the exponents are approximated by structure-preserving and unconditionally stable reduced-order mod-els (ROMs) drawn from an extended Krylov space. Theperformance of the method is illustrated through a num-ber of examples in which we simulate electromagnetic and
CS13 Abstracts 279
acoustic wavefield propagation.
Rob RemisCircuits and Systems GroupDelft University of [email protected]
Vladimir L. Druskin, Mikhail ZaslavskySchlumberger-Doll [email protected], [email protected]
MS188
Krylov Subspace Methods for Large Scale Con-strained Sylvester Equations
Constrained Sylvester equations arise in various applica-tions, and in particular in control theory, in the designof reduced-order observers. In this talk we present a newformulation of the algebraic problem which possesses cer-tain advantages in the case of large scale data. Projectionsolvers for the resulting matrix equation will be discussed,whose approximate solution exactly satisfies the given con-straint. Numerical experiments will be reported to illus-trate the new methodology.
Stephen D. ShankTemple [email protected]
Valeria SimonciniUniversita’ di [email protected]
MS188
Inverse Problems for Large-Scale Dynamical Sys-tems in the H2-Optimal Model Reduction Frame-work
The Rational Krylov subspace (RKS) projection methodwith application to the inverse problems was considered.We derive a representation for the reduced Jacobian as theproduct of a time-dependent and a stationary part. Thenwe show that the RKS satisfying the Meier-Luenberger nec-essary H2 optimality condition completely annuls the in-fluence of approximation error on the inversion result. Wecompare our inversion against other nearly optimal RKS’sbased on Zolotarev problem and adaptive pole selectionalgorithm.
Mikhail ZaslavskySchlumberger-Doll [email protected]
Aria AbubakarSchlumberger Doll [email protected]
Vladimir L. Druskin, Tarek HabashySchlumberger-Doll [email protected], [email protected]
Valeria SimonciniUniversita’ di [email protected]
MS189
Improved Stability Estimates for the hp-Raviart-
Thomas Projection Operator on Quadrilaterals
In this talk we derive improved stability estimates for thehp-Raviart-Thomas projection operator on quadrilaterals.Such estimates may be useful per se, but also have impor-tant applications, e.g., in the inf-sup stability proofs and aposteriori error estimation for hp-DG methods. In particu-lar, we show that the stability constant of the RT-projectoras a mapping in H1 grows not faster than p3/2, where p isthe polynomial degree, being an improvement of the boundp2 reported by Schoetzau et. al in SIAM J. Numer. Anal.,40 (2003), 2171–2194.
Alexey ChernovUniversity of BonnHausdorff Center for [email protected]
Herbert EggerNumerical Analysis and Scientific ComputingDarmstadt University of [email protected]
MS189
Commuting Diagram of TNT Elements on Cubes
Abstract not available at time of publication.
Bernardo CockburnSchool of MathematicsUniversity of [email protected]
Weifeng QiuCity University of Hong [email protected]
MS189
The Discontinuous Petrov-Galerkin Method for theStokes Problem
We discuss well-posedness and convergence theory for thediscontinuous Petrov Galerkin (DPG) method applied tothe classical Stokes problem. The Stokes problem is aniconic troublemaker for standard Bubnov Galerkin meth-ods; if discretizations are not carefully designed, they mayexhibit non-convergence or locking. By contrast, DPG doesnot require us to treat the Stokes problem in any specialmanner. We illustrate and confirm our theoretical conver-gence estimates with numerical experiments.
Nathan RobertsUniversity of Texas at [email protected]
Tan Bui-ThanhThe University of Texas at [email protected]
Leszek DemkowiczInstitute for Computational Engineering and Sciences(ICES)The University of [email protected]
MS189
Biorthogonal Basis Functions in hp-Adaptive FEM
280 CS13 Abstracts
for Elliptic Obstacle Problems
The talk presents an hp-adaptive mixed finite element dis-cretiziation for a non-symmetric elliptic obstacle problemwhere the dual space is discretized via biorthogonal ba-sis functions. The resulting algebraic system only includesbox constraints and componentwise complementarity con-ditions. This special structure is exploited to apply efficientsemismooth Newton methods using a penalized Fischer-Burmeister NCP-function in each component. Adaptivityis accomplished via a posteriori error control which is alsointroduced. Several numerical experiments show the ap-plicability of the constructed biorthogonal basis functions.
Andreas SchroederHumboldt-Universitaet zu [email protected]
Lothar BanzLeibniz University [email protected]
MS190
A Portable OpenMP Runtime Library based onMCA APIs
Programming multicore embedded systems is a challenge.These systems typically consist of heterogeneous cores op-erating on different ISAs, OSes and dedicated memory sys-tems. Are there adequate software toolsets that can exploitcapabilities of such systems? We will discuss about the in-dustry standards, MCA APIs and how they could be usedwith OpenMP to provide the light-weight runtime libraryfor embedded platforms.
Sunita ChandrasekaranUniversity of [email protected]
MS190
Toward Parallel Applications for the Year of Ex-ascale: Requirements for Resilient, Scalable Pro-gramming Models
Presently we are on the threshold of mass deployment ofmultilevel parallelism across most application areas. Thereare many programming models, languages and architec-tures from which to pick, and the number of choices isgrowing. In this presentation we discuss some of the prin-ciples of parallel application development that have pro-duced todays codes, how we can address these principlesgoing forward, and what we need from programming mod-els.
Michael A. HerouxSandia National [email protected]
MS190
The GPU Revolution, What Computational Chem-istry and Battlefield Earth have in Common
This talk will look at the evolution of hardware featuresin commodity accelerators and the drivers influencing thearchitecture. The application of accelerators in HPC haslead to a wide array of software tools and underlying pro-gramming standards. We will snapshot the role NVIDIAhas played in this effort, and illustrate a few successes and
the challenges ahead.
Duncan [email protected]
MS190
Handling the Power, Performance and ReliabilityBattle in Programming Models
Energy consumption and resilience are primary issues forhigh supercomputer utilization at Exascale systems. Theseissues cannot be addressed in isolation as there are proventradeoffs between them. In this talk, I will discuss theimportance of programming models in designing fault tol-erance solutions, and present a case study with computa-tional chemistry. I will also present the methodologies forenergy efficiency and a case of co-design here. I will alsopresent food for thought for fault tolerance and energy ef-ficiency based on upcoming architectural trends.
Abhinav VishnuPacific National Northwest [email protected]
MS191
Automated Adjoints of Finite Element Discretiza-tions
A new approach for automatically deriving adjoint modelsis presented. The discretised PDE is formulated in a highlevel language that resembles the mathematical notation.Our approach differentiates this high level specification anduses code generation to implement the adjoint model (us-ing the FEniCS system). I demonstrate that this approachautomatically and robustly generates adjoint models for awide class of PDE models. Examples in the field of opti-mization and stability analysis are presented.
Simon W. FunkeImperial College LondonDepartment of Earth Science and [email protected]
Patrick FarrellDepartment of Earth Science and EngineeringImperial College [email protected]
Marie E. RognesSimula Research [email protected]
David HamImperial [email protected]
MS191
Oneshot Design Optimisation with Bounded Retar-dation
To complete an optimization run in a small multiple of thenumber of iterations required for a simulation run, it mustbe ensured that the contractivity factor of the combined‘one-shot’ scheme is only a certain fraction closer to 1 thanthat of the user supplied fixed-point solver. This prop-erty we call bounded retardation of the convergence speedof a one-shot optimization compared to just a simulation.
CS13 Abstracts 281
We provide a bound on the retardation factor in terms ofproblem characteristics, mostly related to the Hessian ofthe Lagrangian. A key question is how close the precondi-tioning matrix of the design step should be to the reducedHessian.
Andreas GriewankHU Berlin, MATHEON Research Center, [email protected]
MS191
Dynamically and Kinematically Consistent GlobalOcean-ice State and Parameter Estimation with aGeneral Circulation Model and its Adjoint
Over the last decade the consortium on ”Estimating theCirculation and Climate of the Ocean” (ECCO) has beenproducing optimal estimates of the global time-evolvingcirculation of the ocean. These estimates form the basisfor addressing various problems in climate research. Atthe heart of the effort is the state-of-the-art MIT oceangeneral circulation model (MITgcm) and its adjoint. Theproject has taken rigorous advantage of algorithmic differ-entiation (AD) to generate efficient, scalable adjoint code.Use of AD has enabled the maintenance of up-to-date ad-joint model versions in an environment of vigorous code de-velopment. As a result, today adjoint versions exist for theocean component itself, and for components simulating seaice, sub-ice shelf cavity circulation, and ocean biogeochem-ical processes. In this talk, we provide an overview of theestimation infrastructure, highlight sample applications re-lated decadal ocean climate variability and predictability,and discuss future directions.
Patrick HeimbachMassachusetts Institute of [email protected]
MS191
Nonlinear Adjoint Looping in Thermoacoustics
Thermoacoustic oscillations are currently one of the biggestproblems facing aircraft engine manufacturers, particularlybecause these oscillations seem to be triggered by very lit-tle noise. In this paper, nonlinear adjoint looping is usedto calculate the optimal starting perturbation of a sim-ple non-linear thermo-acoustic system. This shows thatthe system exploits non-normal transient growth to reacha stable limit cycle, even when starting with less energythan the corresponding unstable limit cycle.
Matthew P. JuniperUniversity of CambridgeDepartment of [email protected]
MS192
Parallel I/O Optimizations for a Massively ParallelElectromagnetic System
Checkpointing is a very effective approach for failurerestart and post processing. However, this approach couldresult in heavy I/O load and may cause an I/O bottleneckon a massively parallel systems, especially future exascalesystems with extremely high concurrency. In this talk, wepresent our application-level checkpoint approaches usingtuned collective I/O, application data aggregation I/O anda threaded data aggregation model for asynchronous I/O.We discuss some production performance improvement of a
massively parallel electromagnetic solver (NekCEM) on thedifferent supercomputer architectures including the IBMBlue Gene/Q.
Jing FuDepartment of Computer ScienceRensselaer Polytechnic [email protected]
MiSun MinArgonne National LaboratoryMathematics and Computer Science [email protected]
Robert LathamArgonne National [email protected]
Christopher CarothersComputer ScienceRensselaer Polytechnic [email protected]
MS192
Accelerating Performance of a Petascale Electro-magnetic Solver NekCEM with MPI+CUDA
Modern supercomputing architectures are increasing thenumber of cores per node and adding GPU co-processorsto increase the instruction throughput. Compute-intensiveapplications need to take advantage of higher throughputby employing both distributed and shared memory pro-gramming models. In this talk, we discuss our approachfor accelerating computational electromagnetics applica-tion code NEKCEM using MPI and CUDA programmingmodels and demonstrate its performance on the leadership-class computing systems such as Eureka/Gadzooks on theIBM BG/P and Cray Titan.
Azamat MametjanovMathematics and Computer ScienceArgonne National [email protected]
MiSun MinArgonne National LaboratoryMathematics and Computer Science [email protected]
Boyana NorrisArgonne National [email protected]
MS192
A Scalable Electromagnetic Solver for Applicationsin Nanoscale Materials
We will present recent advances in algorithmic and soft-ware development for efficient and accurate electromagnet-ics modeling that can benefit a wide range of relevant re-search communities and industries involved in the produc-tion of plasmonic devices, photovoltaic cells, electronic andstorage devices, and semiconductors. The core algorithmsare implemented into a petascale electromagnetic code thatis based on an efficient communication kernel and memory-reduced framework. Performance and scalability analysison the advanced computing architectures such as the IBMBG/Q and Cray XK6 will be demonstrated as well as some
282 CS13 Abstracts
preliminary results based on a hybrid MPI/shared-memorymodel.
MiSun MinArgonne National LaboratoryMathematics and Computer Science [email protected]
Jing FuDepartment of Computer ScienceRensselaer Polytechnic [email protected]
Azamat MametjanovMathematics and Computer ScienceArgonne National [email protected]
Ying HeDepartment of MathematicsPurdue [email protected]
Paul F. FischerArgonne National [email protected]
MS192
Investigation on using Different High-order basesfor Some Electromagnetic Simulations
Electromagnetic simulations need highly efficient numeri-cal methods, which depend on many factors, such as expan-sion bases, mesh, algorithms and parallel models. In thistalk, we will investigate the effects of using different ex-pansion bases in some electromagnetic simulations. Theirperformance will be compared in detail, which includes ac-curacy, speed, and parallel efficiency. Based on these, ad-vantages and disadvantages of using different bases will bestudied. Some simulation results will be presented usingthe suitable solvers.
Jin Xu, Xiaohe Zhufu, Ruifeng ZhaoInstitute of Software, ISCACSChinese Academy of Science, Chinaxu [email protected], [email protected],[email protected]
MiSun MinArgonne National LaboratoryMathematics and Computer Science [email protected]
MS193
Scalable Physics-based Preconditioning for 3D Ex-tended MHD
Extended MHD (XMHD) is a very challenging hyperbolicPDE system for implicit integration techniques due to theill-conditioning introduced by fast dispersive waves. Inthis talk, we will describe our physics-based precondition-ing approach for 3D XMHD. The method is based ona conceptual Schur-complement decomposition, which ex-ploits the nature of the hyperbolic couplings in XMHD toproduce a block diagonally dominant PDE system, well-conditioned for multilevel techniques. Numerical experi-
ments will demonstrate the scalability of the approach.
Luis ChaconLos Alamos National [email protected]
MS193
Block Preconditioners for Coupled Fluids Prob-lems
Many important scientific systems require solution of ex-tensions to standard incompressible flow models, whetherby incorporating additional nonlinear effects or by couplingto other processes. We consider approximate block factor-ization preconditioners for the algebraic systems resultingfrom linearization and FE discretization of these systems.In particular, we extend existing block-structured precon-ditioners (such as those of Elman, et al.), combining effec-tive preconditioners for Navier-Stokes with other elementsto obtain preconditioners for magnetohydrodynamics andcoupled fluids problems.
Victoria HowleTexas [email protected]
Robert C. Kirby, Geoffrey DillonTexas Tech UniversityRobert [email protected], [email protected]
MS193
Block-Oriented Preconditioners for the Solution ofthe Semiconductor Drift-Diffusion Equations
We apply block-oriented preconditioners to the implicitsolution of the drift-diffusion equations for semiconduc-tor device modeling. The equations are discretized by astabilized finite element method to produce the nonlinearcoupled system, then solved with a parallel preconditionedNewton-Krylov method. The subblocks are solved by al-gebraic multigrid methods. The performance of these pre-conditioners will be compared to preconditioners that solvethe fully-coupled algebraic system with fully-coupled AMGtechnique.
Paul LinSandia National [email protected]
John ShadidSandia National LaboratoriesAlbuquerque, [email protected]
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
MS193
The Fast Adaptive Composite-grid Method for a3-temperature Radiation Diffusion System withAdaptive Mesh Refinement
We describe the fast adaptive composite-grid (FAC) pre-conditioner applied to a non-equilibrium 3-temperature ra-diation diffusion problem. Multiple temporal and spatialscales make the associated initial value problem very chal-
CS13 Abstracts 283
lenging to solve. These challenges are addressed by fullyimplicit time integration and dynamic adaptive mesh re-finement. At every timestep, a large scale nonlinear systemis solved by the Jacobian-free Newton Krylov approach pre-conditioned by FAC. We will demonstrate the performancewith accuracy, efficiency and scalability results.
Zhen Wang, Bobby Philip, Manuel Rodriguez Rodriguez,Mark BerrillOak Ridge National [email protected], [email protected],[email protected], [email protected]
MS194
Phase Response Theory Reveals Roles of Centraland Sensory Inputs in Cockroach Locomotion
Abstract not available at time of publication.
Einat FuchsPrinceton [email protected]
MS194
Design of Experiments of Parametric Manifoldswith Application to Machine Vision and MaterialIdentification
Abstract not available at time of publication.
James PennMassachusetts Institute of [email protected]
MS194
On the Usefulness of Model Reduction Techniquesin the Quest to Eradicate Infectious Diseases
Abstract not available at time of publication.
Joshua ProctorIntellectual [email protected]
MS195
Analysis of Glassy Potential Energy Landscapes
Glasses are an ill-understood class of materials. Althoughmany potential topologies have been proposed for the un-derlying energy landscape, little is known about the trueform of the potential energy surface that governs glassy dy-namics. We explore techniques to sample the underlyingenergy landscape. We then use dimensionality reductiontechniques to extract patterns in these landscapes, search-ing for underlying trends or coarse-grained structure in thedetailed potential energy landscape of glasses.
Carmeline DsilvaPrinceton [email protected]
MS195
Mapping Sugars Along Catalytic Itineraries: ACase Study in Exploring Multi-dimensional Land-scapes
Accurately mapping the substrate free energy landscape is
computationally demanding and of particular importanceto understand enzyme function. We compared a suite ofmethods to examine how enzymes break down polysaccha-rides. The results highlight how the molecular and elec-tronic structure of carbohydrates change as a function ofperturbation from their solution-stable structures, lendingthem amenable to catalysis. These new insights will havewide-ranging applications from biomedical sciences to cel-lulose conversion for biofuel production.
Heather MayesNorthwestern [email protected]
MS195
Tensor Hypercontraction Theory: A Physically-Motivated Rank Reduction Method for ElectronicStructure Theory
The Tensor Hypercontraction representation is a new,physically-motivated compression scheme which reducesthe ubiquitous electron repulsion operator from a fourth-order tensor to a product of five second-order tensors. Ad-vantages of this representation include substantial formalscaling reductions, decreased memory and/or communi-cations requirements, and enhanced reliance on matrix-multiplication kernels. This talk discusses the mathemat-ics of the compression scheme, as well as ongoing initiativesto apply the representation within many areas of electronicstructure theory.
Robert M. ParrishGeorgia [email protected]
MS195
Absorption Spectra and Photoexcitation Dynamicsin Phenylacetylene Dendrimers using TDDFT
Photochemical conversion of solar energy involving lightharvesting molecules has been identified as a promis-ing pathway to practical alternative energy technologies.Design of higher efficiency artificial photosynthetic sys-tems necessitates a detailed understanding of photoexci-tation and energy transfer processes. Here, we use time-dependent density functional theory to describe the elec-tronic structure and photoexcitation in dendrimeric sys-tems. Absorption spectra and electronic relaxation mech-anisms of a wide range of dendrimers are computed andagreement with recent experiments is discussed.
Aaron [email protected]
Todd MartinezDepartment of ChemistryStanford [email protected]
MS196
The Community Earth System Model: EnablingHigh-resolution Climate Simulations
The Community Earth System Model (CESM) is a widely-used global climate model that enables climate simulationsover meaningful periods of time and at high resolutions.CESM is composed of separate component models that
284 CS13 Abstracts
simulate the ocean, atmosphere, ocean, sea-ice, and landsurface. A central coupler provides interpolation and re-gridding of the boundary conditions between the compo-nent models. We will discuss some examples of the com-putational challenges involved with climate modeling soft-ware.
Allison H. Baker, John M. DennisNational Center for Atmospheric [email protected], [email protected]
MS196
A Space-Time Domain Decomposition Method forStochastic Parabolic Problems
We consider a domain decomposition based implicit space-time approach for solving stochastic parabolic PDEs. Theequation is first discretized in space and time using astochastic Galerkin method and then decoupled into a se-quence of deterministic systems with a Karhunen-Loeveexpansion and double orthogonal polynomials. A Schwarzpreconditioned recycling GMRES method is employed tosolve the systems with similar structures. We report ex-periments obtained on a parallel computer with a largenumber of processors.
Cui CongUniversity of Colorado at [email protected]
Xiao-Chuan CaiUniversity of Colorado, BoulderDept. of Computer [email protected]
MS196
Resilience at Extreme Scale: System Level, Algo-rithmic Level or Both?
Resilience is a critical problem for extreme scale nu-merical simulations. The most credible solution is stillbased on checkpoint/restart with its high overheads orhardware cost. It has been shown recently that somealgorithmic approaches and some code characteristicscan help reducing these costs through combined system-algorithmic/application approaches. However, we are stilllooking for a right solution to this simple question: how toreduce simultaneously and significantly state saving andrecovery times?
Franck CappelloINRIA and University of Illinois at [email protected]
MS196
Exploiting Hierarchies in Algorithms, Software,and Applications
This presentation discusses exploiting algorithmic and soft-ware hierarchies in scientific computing. These hierarchiespresent opportunities to manage complexity, exploit thechanging landscape of high-performance computing, andreduce the time required to simulate physical phenomena.In this talk, we focus on the impact of hierarchy-awareKrylov methods in subsurface flow to overcome bottlenecksin global synchronization, and we discuss related issues inoptimization software.
Lois Curfman McInnes, Todd Munson
Argonne National LaboratoryMathematics and Computer Science [email protected], [email protected]
Jie ChenArgonne National [email protected]
Hong ZhangArgonne National [email protected]
MS197
Block Preconditioners for Implicit AtmosphericClimate Simulation in CAM-SE
We discuss the development of block preconditioners inan effort to reduce computational costs associated withimplicit time integration of atmospheric climate modelswithin CAM-SE. We construct a fully implicit frameworkbased on the shallow water equations and view the sub-sidiary linear system as a block matrix. Preconditionersare derived based on approximate block factorization.
Aaron LottLawrence Livermore Nat’l [email protected]
Katherine J. EvansOak Ridge National [email protected]
Carol S. WoodwardLawrence Livermore Nat’l [email protected]
MS197
Physics-based Preconditioners for Ocean Simula-tion
We examine physics-based preconditioners for free-surface,fully implicit, fully coupled time integration of the mo-mentum and continuity equations of ocean dynamics; thusreducing errors and increasing stability due to traditionaloperator splitting. The nonlinear system is solved viaJacobian-free Newton-Krylov, where we reformulate semi-implicit barotropic-baroclinic splitting as a preconditioner.Thus the desired solution is timestep converged withtimesteps on the order of the dynamical timescale. Weprovide numerical examples and compare to explicit meth-ods.
Christopher K. Newman, Dana A. KnollLos Alamos National [email protected], [email protected]
MS197
Implicit Solvers for Coupled Overland and Subsur-face Flow
Subsurface and overland flow models constitute a signifi-cant portion of fresh water resource simulations, and cou-pling these models can produce insights into efficient man-agement of these renewable resources. We consider asubsurface flow model coupled to kinematic and diffusivewave approximations of overland flow. We will discuss animplicit solution approach to these coupled models andoverview the formulation and effectiveness of a block pre-
CS13 Abstracts 285
conditioning approach.
Daniel Osei-KuffuorLawrence Livermore National [email protected]
Carol S. WoodwardLawrence Livermore Nat’l [email protected]
Reed M. MaxwellDepartment of Geology and Geologic EngineeringColorado School of [email protected]
Laura CondonColorado School of [email protected]
MS197
A Domain Decomposition based Implicit Methodfor Compressible Euler Equations in AtmosphericModeling
We discuss some multilevel domain decomposition basedfully implicit methods for solving the nonlinear system ofhyperbolic equations arising from global climate modeling.With the fully implicit approach, the time step size is nolonger limited by the stability condition, and with multi-level overlapping Schwarz preconditioners, good scalabil-ities are obtained on computers with a large number ofprocessors. Numerical results are provided to show theconservation properties and the parallel scalability of themethods.
Chao YangDept. of Computer ScienceUniversity of Colorado at [email protected]
Xiao-Chuan CaiUniversity of Colorado, BoulderDept. of Computer [email protected]
MS198
Multi-level Monte Carlo for Continuous TimeMarkov Chain Models of Intracellular BiochemicalProcesses
Multi-level Monte Carlo is a relatively new method thatcomputes expectations of stochastic processes to a desiredtolerance significantly faster than previous methods. I willdetail the basic idea of multi-level Monte Carlo and showhow to implement it in the continuous time Markov chainsetting, which is a modeling choice commonly used in thebiosciences.
David F. AndersonDepartment of MathematicsUniversity of Wisconsin [email protected]
MS198
Efficient Simulation of Mesoscopic Reaction-diffusion Kinetics via Operator Splitting
A fractional step method offers many advantages for spa-
tial stochastic simulation. It makes efficient simulation oflarge systems possible by using approximate methods toupdate the system due to diffusive transfers on the grid.It also simplifies more efficient parallel simulation since de-coupling the operators enables more parallelism over thesplitting time step. We propose a computable strategy toadaptively select that time step and illustrate the efficiencyof our approach in parallel implementations.
Andreas HellanderDepartment of Computer ScienceUniversity of California, Santa [email protected]
Brian DrawertUniversity of California Santa [email protected]
Michael LawsonUniverity of Califorina Santa [email protected]
Linda PetzoldUniversity of California, Santa [email protected]
MS198
First-Passage Kinetic Monte Carlo Methods forReaction-Drift-Diffusion Processes
Earlier versions of First-Passage Kinetic Monte Carlo, astochastic algorithm for simulating reaction-diffusion pro-cesses, rely on analytic solutions of the diffusion equationand do not allow for drift. We have developed a varia-tion of the algorithm using a discretization of the Fokker-Planck equation to incorporate drift due to arbitrary po-tentials. In this talk, we will demonstrate accuracy andconvergence of our algorithm, compare various implemen-tation approaches, and discuss applications to reaction-drift-diffusion processes in cell biology.
Ava J. MauroDepartment of Mathematics and StatisticsBoston [email protected]
MS198
Computational Analysis of Stochastic Reaction-diffusion Equations
How to choose the computational compartment or cell sizefor the stochastic simulation of a reaction-diffusion systemis still an open problem. We discuss a new criterion basedon a global measure of the sensitivity of the reaction net-work that predicts a grid size that assures that the con-centrations of all species converge to a spatially-uniformsolution. This criterion applies for all orders of reactionsand encompasses both diffusing and non-diffusing species
Hans G. OthmerUniversity of MinnesotaDepartment of [email protected]
MS199
Scalable Algorithms for Function Approximation
286 CS13 Abstracts
and Error Estimation on Arbitrary Sparse Samples
Stochastic collocation methods are an attractive choice tocharacterize uncertainty because of their non-intrusive na-ture. High dimensional stochastic spaces can be approxi-mated well for smooth functions with sparse grids. Therehas been a focus in extending this approach to non-smoothfunctions using adaptive sparse grids. We have developeda fast method that can capture piecewise smooth functionsin high dimensions with high order and low computationalcost. This method can be used for both approximationand error estimation of stochastic simulations where thecomputations can either be guided or come from a legacydatabase. We compare these methods to more traditionalstatistical approaches.
Rick ArchibaldComputational Mathematics GroupOak Ridge National [email protected]
MS199
Tensor-based Algorithms for the Optimal ModelReduction of Stochastic Problems
Tensor-based methods are receiving a growing attentionfor their use in high dimensional applications in scientificcomputing where functions of multiple parameters have tobe approximated. Here, we present algorithms that areable to directly construct an approximation of optimal ten-sor decompositions of the solution of stochastic equations,without a priori information on the solution. Optimalitycan be achieved with respect to a desired metric.
Anthony Nouy, Marie Billaud-Friess, Loic Giraldi, OlivierZahmLUNAM Universite, Ecole Centrale Nantes, CNRS, [email protected],[email protected], [email protected],[email protected]
MS199
Sliced Cross-validation for Surrogate Models
Multi-fold cross-validation is widely used to assess the ac-curacy of a surrogate model in uncertainty quantification.Despite its popularity, this method is known to have highvariability. We propose a method, called sliced cross-validation, to mitigate this drawback. It uses sliced space-filling designs to construct structured cross-validation sam-ples such that the data for each fold are space-filling. Ex-tensions of the method to situations with high-dimensionalinputs will be discussed. Numerical examples and theoret-ical results will be given to illustrate the proposed method.
Peter QianUniversity of Wisconsin - [email protected]
MS199
On the Consistency of Calibration Parameter Esti-mation in Deterministic Computer Experiments
Calibration parameters in deterministic computer exper-iments are those attributes that cannot be measured oravailable in physical experiments. Kennedy-OHagan sug-gested an approach to estimate them by using data fromphysical as well as computer experiments. We develop anasymptotic theory for kernel interpolations to study the
calibration consistency problem and show that the KOmethod leads to asymptotically inconsistent calibrationdue to overfitting. This calibration inconsistency can beremedied by modifying the original estimation procedure.(Based on joint work with Rui Tuo, Chinese Academy ofSciences)
Jeff WuGeorgia Institute of [email protected]
MS200
Differential Geometric MCMC Methods and Ap-plications
I will discuss the latest advances in Markov chain MonteCarlo methodology that exploit the natural underlying Rie-mannian geometry of many statistical models. Such algo-rithms automatically adapt to the local correlation struc-ture of the model, providing highly efficient means of per-forming Bayesian inference for inverse problems. I will pro-vide examples of Bayesian inference using these methodson a variety of challenging statistical models, including dy-namical systems described by nonlinear differential equa-tions.
Ben Calderhead, Mark GirolamiUniversity College [email protected], [email protected]
MS200
Derivation and Low-rank Computation of theBayesian Filter
We derive the non-linear Bayesian update (NLBU) anddevelop low-rank numerical algorithms for its evaluation.NLBU in a non-sampling functional approximation settingwill be derived from the variational problem associatedwith conditional expectation. Whereas the linear BU isa linear function of the prediction mismatch, here we willuse higher order polynomials. An important intermediatesubproblem which appears during the BU is the increaseof the stochastic dimension after each update. The reasonis the new random variables which come from the randommeasurement noise.
Alexander LitvinenkoTU Braunschweig, [email protected]
Hermann G. MatthiesInstitute of Scientific ComputingTechnische Universitat [email protected]
MS200
Implicit Particle Methods for Data Assimilation
Many applications in science and engineering require thatan uncertain model be updated by a stream of incompleteand noisy data. I will present a sequential Monte Carlomethod for this problem that can avoid many of the pitfallthat arise from a large problem dimension. The basic ideais to first identify regions of large probability and then focusattention on these regions. I will illustrate the theory withexamples from applications in geophysics.
Matthias MorzfeldDepartment of Mathematics
CS13 Abstracts 287
Lawrence Berkeley National [email protected]
Alexander J. ChorinUniversity of California, BerkeleyMathematics [email protected]
MS200
Bayesian Data Assimilation with Optimal Maps
We develop novel map-based schemes for sequential dataassimilation, i.e., nonlinear filtering and smoothing. Onescheme involves pushing forward a fixed reference mea-sure to each filtered state distribution, while an alternativescheme computes maps that push forward the filtering dis-tribution from one stage to the next. A key advantage ofthe map approach is that it inherently avoids issues of sam-ple impoverishment, since the posterior is explicitly repre-sented as a transformation of a reference measure, ratherthan with a particular set of samples. The computationalcomplexity of our algorithm is comparable to state-of-the-art particle filters.
Tarek [email protected]
Youssef M. MarzoukMassachusetts Institute of [email protected]
MS201
Control Volume Finite Element Method for Drift-diffusion Equations on General Unstructured Grids
We present a new Control Volume Finite Element Methodfor the drift-diffusion equations, which uses edge elementsto define an exponentially fitted elemental current. Thiscurrent obviates the need for the control volumes to betopologically dual to the finite elements and results in amethod that is stable and accurate on general unstructuredfinite element grids. Simulations of a silicon PN diode anda MOSFET device demonstrate the performance of the newscheme.
Pavel BochevSandia National LaboratoriesComputational Math and [email protected]
Kara Peterson, Xujiao GaoSandia Natl. [email protected], [email protected]
MS201
Lagrangian Hydrodynamics for Compressible Flu-ids
The entropy viscosity technique is extended to the La-grangian framework for solving the compressible Eulerequations. Various types of parabolic regularizations arediscussed and a minimum entropy principle is proved. Themethod is illustrated numerically on 2D and 3D test cases.
Jean-Luc GuermondDepartment of Mathematics
Texas A&M, [email protected]
Bojan PopovDepartment of MathematicsTexas A&M [email protected]
Vladimir TomovDepartment of Mathematics Texas A&M, [email protected]
MS201
High-order Curvilinear ALE Hydrodynamics
The Arbitrary Lagrangian-Eulerian (ALE) framework,which forms the basis of many shock hydrodynamics codes,consists of alternating Lagrange and advection phases. Inthis talk we will discuss our work on high-order extensionsof the ALE advection phase, including curvilinear mesh op-timization based on appropriately defined high-order topo-logical ”mesh Laplacian” and smoothing operators, as wellas new DG advection algorithms for conservative and ac-curate remap of high-order fields. We will report resultsfrom single-material advection tests in a parallel researchcode.
Robert AndersonLLNLCtr for Appl Scientific [email protected]
Veselin DobrevLawrence Livermore National [email protected]
Tzanio V. KolevCenter for Applied Scientific ComputingLawrence Livermore National [email protected]
Robert RiebenLawrence Livermore National [email protected]
MS201
Adaptive Material Interface Capturing Methods
A method used to capture a material interface carries withit a number of considerations that manifest themselves inthe character of the solution. The nature of a physicallyadmissible solution is key. Unfortunately, these considera-tions are often only implicitly manifested in the methodschosen to propagate the interface. We examine these con-siderations and how to bring them into the light of day.This provides a path forward toward better more physi-cally motivated methods.
William J. RiderSandia National [email protected]
MS202
Performances of Krylov Solvers for Reactor PhysicsSimulation on Petascale Architectures
The governing equation in neutronic simulation for reac-
288 CS13 Abstracts
tor physics application is the Boltzmann neutron transportequation. In order to solve this equation, one has to solvean eigenproblem. Efficient parallel eigensolvers are com-pulsory to reach high performances simulations for reactorphysics. We will present in this talk some performancesresults of Arnoldi eigensolvers applied to reactor physicsproblem on the petascale heterogeneous CURIE machineon both CPUs and GPUs configurations.
Christophe Calvin
CEA-Saclay/DEN/DANS/[email protected]
Jerome DuboisCEA-SaclayCEA/DEN/DANS/DM2S/[email protected]
MS202
Performance Evaluation of Multi-threaded Itera-tive Solver on Recent Processors
Performance evaluation of multi-threaded sparse triangu-lar solver is conducted on recent multi-core / many-coreprocessors such as Intel Sandy Bridge processor. A sparsetriangular solver is involved in IC(0) preconditioning, SORmethod, Gauss-Seidel smoother etc., and it is utilized inmany practical applications. Our multi-threaded solver isbased on block multi-color ordering, which is one of paral-lel ordering techniques. The effect of blocking of unknownsis examined in recent processors.
Takeshi Iwashita, Akihiro IdaAcademic Center for Computing and Media StudiesKyoto [email protected], [email protected]
Masatoshi KawaiGraduate School of InformaticsKyoto [email protected]
Hiroshi NakashimaACCMSKyoto [email protected]
MS202
Modeling of Epidemic Spread and Eigenvalue Com-putation
We present epidemiological modeling techniques of thespread of infectious diseases and show that below a thresh-old number of infected individuals, the spread of the epi-demic will stop. We highlight that this epidemic threshold,beyond which infections become endemic, can be repre-sented by the largest eigenvalue of the adjacency matrixrepresenting the network of individuals. We present an ef-ficient method to calculate the major eigenvector and reacha fast reaction facing the outbreak of epidemics.
Nahid EmadUniversity of Versailles, PRiSM [email protected]
Zifan LiuUniversity of [email protected]
Michel LamureUniversity of Lyon [email protected]
Sofian ben AmorVersailles [email protected]
MS202
Parallel CFD Code using ppOpen-HPC for Post-peta-scale Systems
ppOpen-HPC is an open source infrastructure for develop-ment and execution of large-scale scientific applications onvarious types of post-peta-scale systems with automatictuning. This talk provides an example of developmentof 3D parallel CFD code on ppOpen-HPC, and overviewsdata structures, API and strategy for automatic tuningof ppOpen-HPC. Target code is developed on multicoreclusters with OpenMP/MPI hybrid parallel programmingmodels, and on CPU-GPU heterogeneous environment.Performance of the developed code is also demonstrated.
Kengo NakajimaThe University of TokyoInformation Technology [email protected]
MS203
A Locally Conservative Eulerian-LagrangianMethod for a Two-Phase Flow Problem
We develop an Eulerian-Lagrangian numerical method fora system of two conservation laws in one space dimensionmodeling a simplified two-phase flow problem in a porousmedium. We approximate tracing along the tracelines byimposing local mass conservation principles for both phasesand optimizing self-consistency. Numerical results demon-strate that the method can handle problems with shocksand rarefactions on coarse spatial grids using time stepslarger than the CFL limit.
Todd ArbogastDept of Math; C1200University of Texas, [email protected]
Chieh-Sen Huangational Sun Yat-sen UniversityKaohsiung, [email protected]
Thomas F. [email protected]
MS203
Finite Element Methods for the Fully NonlinearMonge-Ampere Equation using a Local DiscreteHessian
In this talk, we will discuss a family of numerical meth-ods for the Monge-Ampere equation, a fully nonlinear sec-ond order PDE. The approach is based on the conceptof a discrete Hessian recently introduced by Aguilera &Morin (2009), Huang et al. (2010) and Lakkis & Pryer
CS13 Abstracts 289
(2011). However, the definition of our discrete Hessian isentirely local, making the resulting linear system withinthe Newton iteration much easier to solve. Replacing theHessian in the PDE by its discrete counterpart, we obtainschemes that converge even when the exact solution pos-sesses strong singularities.
Michael J. NeilanUniversity of PittsburghDepartment of [email protected]
MS203
New Phase-field Models and Energy Stable Numer-ical Schemes for Multiphase Flows with DifferentDensities
I shall present two new phase field models, one incompress-ible and the other quasi-incompressible, for multiphaseflows with different densities. I shall also present efficientand energy stable numerical schemes, as well as some nu-merical results which validate the flexibility and robustnessof these phase-field models.
Jie ShenPurdue UniversityDepartment of [email protected]
MS203
Numerical Approximation of Oldroyd-B Fluids
The Oldroyd–B equations model the flow of fluids contain-ing rod–like elastic molecules. This model couples the mo-mentum equation with an equation governing the evolu-tion of the elastic components, and numerical simulation isnotoriously difficult (the high Weisenberg problem). Thissystem of partial differential equations can be derived fromHamiltonian’s principle which reveals a subtle balance be-tween inertia, transport, and dissipation effects. This talkwill focus on the structural properties of these equationswhich provide insight into why naive numerical schemesmay fail, and the ingredients required to construct stablenumerical schemes.
Noel J. WalkingtonDepartment of Mathematical SciencesCarnegie Mellon [email protected]
MS204
Multilevel Approximate Inversion
Computing the diagonal entries of the inverse of a sparsematrix arises in several computational applications such ascovariance matrix analysis in uncertainty quantification, orwhen evaluating Green’s functions in computational nano-electronics. We discuss the approximate matrix inversionusing a multilevel incomplete factorization. Its approxi-mate inverse is represented recursively as a sum of matri-ces with increasing rank but decreasing norm. Taking thelevels backwards we can successively update the diagonalentries of the matrix inverse.
Matthias BollhoferTU [email protected]
MS204
Usage of Domain Decomposition Smoothers inMultigrid Methods
Local block solves are a building block of domain de-composition methods. Furthermore they can be used assmoothers for multigrid methods. In this context they havesome advantages over point smoother, e.g., a higher arith-metic intensity that is beneficial on nowadays computerarchitectures. We analyzed different block smoothers andwe will point out their advantages and disadvantages whenused in a multigrid setting.
Matthias BoltenUniversity of [email protected]
Karsten KahlBergische Universitat WuppertalDepartment of [email protected]
MS204
Bootstrap AMG
We present in this talk a Bootstrap approach to adaptiveAMG introducing the so-called “Least Squares Interpo-lation” (LSI) and a “Bootstrap Setup” which enables usto compute accurate LSI operators using a multigrid ap-proach in the setup to efficiently compute prototypes ofalgebraically smooth error. We demonstrate the potentialof the Bootstrap AMG approach in the application to avariety of problems, each illustrating a certain aspect ofthe method.
Karsten KahlBergische Universitat WuppertalDepartment of [email protected]
MS204
Robust Solution of Singularly Perturbed Problemsusing Multigrid Methods
In order to resolve important features of solutions to sin-gularly perturbed differential equations, specialized dis-cretization techniques are frequently used. Here, we con-sider classical finite difference schemes on meshes adaptedto resolve boundary layers in reaction-diffusion equations.We show that standard direct solvers exhibit poor scal-ing behaviour when solving the resulting linear systems.We consider, instead, standard robust multigrid precondi-tioners for these linear systems, and we propose and proveoptimality of a new block-structured preconditioning ap-proach.
Scott MaclachlanDepartment of MathematicsTufts [email protected]
Niall MaddenNational University of Ireland, [email protected]
MS205
Reproducibility and Computationally Intensive,
290 CS13 Abstracts
Data-driven Research
Since Jon Claerbout adopted and started promoting repro-ducible research practices much has changed. While theproblems for reproducibility of computational results hasgrown in conjunction with increases in computing powerand storage densities, there has also been a steady growthin awareness of these problems and strategies to addressthem. In this minisymposium, we will discuss several re-cent attempts to come to terms with reproducibility incomputational research. Topics will include education,publication, forensics and scientific integrity, as well as newtechnologies for provenance tracking and literate program-ming.
Kenneth J. MillmanUniversity of California, [email protected]
Vincent J. CareyChanning Division of Network MedicineHarvard Medical [email protected]
MS205
Disseminating Reproducible Computational Re-search: Tools, Innovations, and Best Practices
Computation is now widely recognized as central to thescientific enterprise, and numerous efforts are emerging toincorporate code and data sharing into standards of re-search dissemination. This goal is challenging from a num-ber of perspectives, including effective research practices.In this talk I discuss novel innovations and best practicesfor facilitating code and data sharing, both at the time ofpublication and during the research itself, that support theunderlying rational of reproducible research.
Victoria StoddenColumbia [email protected]
MS205
Rethinking How we Work with Documents
Reproducible research tools require capturing all of the dif-ferent avenues and lines of exploration in research. Thedocument should be a database that can be rendered indifferent ways for different audiences, allowing dynamic re-sults replacing code, enabling reader interactivity to ex-plore different approaches and what-ifs. We also want tobe able to programmatically query, update and verify thisdocument database. All of this leads us to a different struc-ture and approach for authoring documents.
Duncan W. Temple LangStatisticsUC [email protected]
MS205
Reproducible Research on the Web: From Home-work, Blogging to Open Journals
Reproducible research used to be tied to LATEX (e.g.Sweave in R) for statisticians, which has a steep learningcurve and lacks many features of the web. The underlyingidea of literate programming, however, is language agnos-
tic. In this talk, we introduce the R package knitr asa general-purpose tool for reproducible research, with anemphasis on dynamic report generation on the web withMarkdown, including reproducible homework, blog postsand online journals in statistics.
Yihui XieDepartment of StatisticsIowa State [email protected]
MS206
Numerical Solution of the Bloch-Torrey EquationApplied to the DMRI of Biological Tissue
We propose a numerical method to solve the Bloch-Torreypartial differential equation in multiple diffusion compart-ments to simulate the bulk magnetization of a sample un-der the influence of a diffusion gradient. We couple amass-conserving finite volume discretization in space witha stable time discretization using an explicit Runge-Kutta-Chebyshev method. We are able to solve the Bloch-TorreyPDE in multiple compartments for an arbitrary diffusionsequence with reasonable accuracy for moderately compli-cated geometries in computational time that is on the or-der of tens of minutes per bvalue on a laptop computer.We show simulation results for nearly isotropic as well asanistropic diffusion, for the PGSE as well cosine OGSEsequences.
Donna CalhounBoise State [email protected]
Jing-Rebecca LiINRIA [email protected]
MS206
Image based Simulations towards UnderstandingTissue Microstructure with MRI
Diffusion-weighted MRI and other quantitative MRI se-quences like multi-exponential T2 are known to report onthe microanatomical integrity of tissue. However, given themicroanatomical complexity of tissue, mathematical andcomputational models are important to aid in the interpre-tation of these experiments. We present on histology-basedsimulations of white matter mircoanatomy for comparisonwith quantitative MRI, as well as the computational tech-niques required to make these simulations feasible.
Kevin HarkinsVanderbilt [email protected]
MS206
Reduced Models of Multiple-compartment Diffu-sion MRI in the Intermediate Exchange Regime
We model the magnetization in biological tissue due toa diffusion gradient by a two compartment Bloch- Torreypartial differential equation with infinitely thin permeablemembranes. We formulate a ODE model for the magne-tization and show the simpler ODE model is a good ap-proximation to the Bloch-Torrey PDE model for a varietyof gradient shapes. Using the ODE model we determine ofthe change in the cellular volume fraction from the signal
CS13 Abstracts 291
attenuation obtained before and after cell swelling. Thismethod requires only the ADC and Kurtosis of the twosignal attenuations and the numerical solution of an ODEsystem.
Jing-Rebecca LiINRIA [email protected]
MS206
Time-dependent Diffusion: From MicrostructureClassification to Biomedical Applications
We show how a bulk diffusion measurement can distin-guish between different classes of microgeometry. Based onthe specific values of the dynamical exponent of a velocityautocorrelator measured with diffusion MRI, we identifythe relevant tissue microanatomy in muscles and in brain,and the microstructural changes in ischemic stroke. Ourframework presents a systematic way to identify the mostrelevant part of structural complexity with diffusion.
Dmitry S. NovikovNew York University School of [email protected]
Els FieremansNew York [email protected]
Jens Jensen, Joseph HelpernMedical University of South [email protected], [email protected]
MS207
A General Framework for Stable Reconstructionsfrom Non-uniform Fourier Samples
Abstract not available at time of publication.
Ben AdcockPurdue [email protected]
MS207
Finite Fourier Frame Approximation using the In-verse Polynomial Reconstruction Method
The inverse polynomial reconstruction method (IPRM)was developed to resolve the Gibbs phenomenon in thespectral reconstruction of piecewise analytic functions. Wedemonstrate that the IPRM is suitable for approximatingthe finite inverse Fourier frame operator as a projectiononto the polynomial space. The IPRM can also removethe Gibbs phenomenon from the Fourier frame approxi-mation of piecewise smooth functions. Numerical resultsshow that the IPRM is robust, stable, and accurate fornon-uniform Fourier data.
Jae-Hun Jung, Xinjuan ChenSUNY at [email protected], [email protected]
Anne GelbArizona State [email protected]
MS207
Convolutional Gridding and Frame Approximation
The technique of covolutional gridding (CG) has beenwidely used in applications with non-uniform (Fourier)data such as magnetic resonance imaging (MRI). On theother hand, its error analysis is not fully understood. Weconsider it as a frame approximation and present an er-ror analysis accordingly. Moreover, we propose a general-ized convolutional gridding (GCG) method as an improvedframe approximation.
Guohui SongClarkson [email protected]
Anne GelbArizona State [email protected]
MS207
Robust Sub-Linear Time Fourier Algorithms
We present a new deterministic algorithm for the sparseFourier transform problem, in which we seek to identifyk � N significant Fourier coefficients from a signal ofbandwidth N . Previous deterministic algorithms exhibitquadratic runtime scaling, while our algorithm scales lin-early with k in the average case. Via a multiscale approachour algorithm is extremely robust against noise.
Yang WangMichigan State [email protected]
MS208
Non-Gaussian Test Models for Prediction andState Estimation with Model Errors
A class of statistically exactly solvable non-Gaussian testmodels are introduced where a generalized Feynman-Kacformulation reduces the exact behavior of conditional sta-tistical moments to the solution of inhomogeneous Fokker-Planck equations modified by linear lower order couplingand source terms. This procedure is applied to a test modelwith hidden instabilities and combined with informationtheory to address the coarse-grained ensemble predictionin a perfect model and improving long range forecasting inimperfect models.
Nan ChenNew York [email protected]
MS208
Stability and Convergence of a Fully DiscreteFourier Pseudo-spectral Method for BoussinesqEquation
In this paper, we discuss the nonlinear stability and con-vergence of a fully discrete Fourier pseudo-spectral methodcoupled with specially designed time-stepping of second or-der for the numerical solution of the “good” Boussinesqequation. Our results improve the known results obtainedby Frutos et al. In particular, this stability condition doesnot impose a restriction on time-step that is dependent onthe spatial grid size.
Wenqiang Feng
292 CS13 Abstracts
Missouri University of Science and [email protected]
MS208
First and Second Order Schemes for Applicationsof Dynamic Density Functional Theory
In this talk I will present the first and second order (intime) unconditional energy stable schemes for nonlocalCahn-Hilliard (CH) equation, nonlocal Allen-Cahn (AC)equation and some models derived from dynamic densityfunctional theory (DDFT). I will briefly derive these non-local models and discuss the relation between nonlocal CHequation and DDFT. Also the relation between DDFT andclassical CH equation and phase field crystal equation willbe discussed. I will briefly introduce properties of schemessuch as stability and convergence. Numerical simulationsof nonlocal CH equation with anisotropic correlation func-tions will be given. Also numerical simulations for im-plications of DDFT, such as hard sphere model, will bepresented.
Zhen GuanThe University of [email protected]
MS208
Operator-splitting forConvection-reaction-diffusion Equations
For reaction-diffusion systems with both stiff reaction anddiffusion terms, implicit integration factor (IIF) methodand its high dimensional analog compact form (cIIF) serveas an efficient class of time-stepping methods. For non-linear hyperbolic equations, front tracking method is oneof the most powerful tools to dynamically track the sharpinterfaces. Meanwhile, weighted essentially non-oscillatory(WENO) methods are a class of start-of-the-art schemeswith uniform high order of accuracy in smooth regions ofthe solution. In this talk, IIF/cIIF is coupled with fronttracking or WENO by the second-order symmetric opera-tor splitting approach to solve advection-reaction-diffusionequations. In the methods, IIF/cIIF methods treat the stiffreaction-diffusion equations, and front tracking/WENOmethods handle hyperbolic equations that arise from theadvection part.
Xingfeng LiuUniversity of South [email protected]
MS209
Beyond Treewidth in Graphical Model Inference
While exact probabilistic inference in graphical modelstypically has cost exponential in the treewidth, we dis-cuss cases where this breaks down. For example, whenthe distribution contains determinism, graph triangula-tions implying higher than optimal treewidth can haveunboundedly faster inference. Also, when the nature,rather than the degree, of interaction is limited, inferencecost can become polynomial even for unboundedly largetreewidth. Examples include submodular interaction func-tions or those that indirectly utilize submodularity.
Jeff A. BilmesUniversity of Washington, [email protected]
MS209
Graph Width Metrics, Well-Quasi Ordered Setsand Fixed Parameter Tractability: History, Appli-cations and Scalable Implementations
Parameterized computation has evolved from a mere com-plexity theoretic aberration to a powerful and highly re-spected technique for solving difficult combinatorial prob-lems. Papers on fixed parameter tractability (FPT) nowappear with regularity in top computer science conferencesand journals. It has not always been an easy journey. Wewill survey FPT’s history, its current known range of ap-plications, the central role played by graph width metrics,and the significance of effective parallel implementations.
Michael A. LangstonDepartment of Electrical Engineering and ComputerScienceUniversity of [email protected]
MS209
Toward Tree-like Structure in Large InformaticsGraphs
Large informatics graphs such as large social and informa-tion networks are often thought of as having some sort oftree-like or hierarchical structure. We describe recent em-pirical and theoretical results aimed at extracting mean-ingful tree-like structure from large informatics graphs ina scalable and robust way. In particular, empirical proper-ties of cut-based methods such as tree decompositions andmetric-based methods such as delta-hyperbolicity, as wellas their similarities and differences, will be reviewed.
Michael MahoneyStanford UniversityApplied and Computational [email protected]
MS209
Chordal Graphs and Clique Trees in Sparse MatrixAlgorithms
We consider two problems in parallel sparse matrix com-putation where efficient algorithms are enabled by theclique tree. Simple greedy algorithms that eliminate cer-tain leaves of the clique tree solve these problems. Prov-ing the correctness of the algorithms provides fresh insightinto the collection of vertex separators in a chordal graph.Hence, in this context, sparse matrix algorithms provdenew results in chordal graph theory.
Alex PothenPurdue UniversityDepartment of Computer [email protected]
MS210
Fully Computable a posterior Error Estimators forStabilized Conforming Finite Element Approxima-tions
Abstract not available at time of publication.
Alejandro AllendesUniversidad Tecnica Federico Santa [email protected]
CS13 Abstracts 293
MS210
Error Estimation for VMS-stabilized AcousticWave Propagation
Abstract not available at time of publication.
Brian CarnesSandia National [email protected]
MS210
Output-based hp-adaptive Simulations of High-Reynolds Number Compressible Flows
We present a method for concurrent mesh and polynomial-order adaptation with the objective of direct minimizationof output error using a selection process for choosing theoptimal refinement option from a discrete set of choicesthat includes directional spatial resolution and approxima-tion order increment. The scheme is geared towards com-pressible viscous aerodynamic flows, in which solution fea-tures make certain refinement options more efficient com-pared to others. We present results for 2D and 3D turbu-lent flows.
Marco Ceze, Krzysztof FidkowskiUniversity of [email protected], [email protected]
MS210
Gradient-norm Error Estimation forPDE-constrained Optimization
When solving a PDE-constrained optimization problem nu-merically, we would like to know if the obtained designaccurately reflects the true optimum. One measure of ac-curacy would be to compute an error estimate for the ob-jective or the Lagrangrian; however, a more natural mea-sure of accuracy for an optimum is the discrepancy in thefirst-order optimality conditions. To this end, we showhow the adjoint-weighted residual method can be used toconstruct a posteriori error estimates for the norm of thegradient. The error estimate requires two additional ad-joint variables, but only at the beginning and end of eachoptimization cycle. Moreover, the adjoint systems can beformed and solved with limited additional infrastructure.
Jason E. Hickenpostdoctoral fellow, Stanford [email protected]
Juan J. AlonsoDepartment of Aeronautics and AstronauticsStanford [email protected]
MS211
Auto-tuning and Smart-tuning Approaches for Ef-ficient Krylov Solvers on Petascale Architectures
For decades, supercomputers have carried on deliveringmore and more computational power using more and morecomplex architecture. It reflects in the parameterization ofthe basic algorithms used by the computer codes to achievegood performances. We experiment a methodology basedon statistical approach for autotuning. The objective is tohelp the code users define the best set of parameters ona key algorithm (Matrix.Vector multiplication) depending
on the numerical problem and the computing environment.
Christophe Calvin
CEA-Saclay/DEN/DANS/[email protected]
Anthony Leroy DrummondLawrence Berkeley National [email protected]
France Boillod-CerneuxCNRS/[email protected]
Jerome [email protected]
Gisele Ndongo EboumUniversity of Paris XIII, [email protected]
MS211
Iterative Method for Sparse Linear Systems usingQuadruple Precision Operations on GPUs
The convergence of iterative methods such as a Krylov sub-space method may be affected by round-off errors and ex-tended precision operations may reduce the iterations toget the convergence. We implemented Krylov subspacemethods for sparse linear systems on an NVIDIA GPUusing quadruple precision operations and evaluated theirperformance. We used double-double arithmetics to per-form quadruple precision arithmetic operations. On theGPUs, the computation time of one iteration on quadrupleprecision is up to approximately twice as that on doubleprecision. In some cases by utilizing quadruple precisionoperations we can reduce the computational time to getthe convergence as compared with the computation usingfull double precision operations.
Daichi MukunokiGraduate School of Systems and Information EngineeringUniversity of [email protected]
Daisuke TakahashiUniversity of TsukubaGraduate School of Systems and Information [email protected]
MS211
Toward Tunable Multi-Scheme Parallelization
High performance computers will have more SIMD widthsand a larger number of cores. System scale ranges froma single processor ( 1TF) to a supercomputer ( 1EF).We need more flexible parallelization methodologies, whichmust be based on composite extension of Amdahl’s law andhierarchical extension of Gustafson’s law. Gradual paral-lelization and deparallelization, reusable parallel compo-nents, and autotuning will be a part of key concepts, anda vision on programming tools for them will be discussed.
Reiji SudaDepartment of Computer Science, The University ofTokyo
294 CS13 Abstracts
MS212
Communicating in Science and not being Afraid ofTenacious Self-promotion
Publish-or-perish is no news to anyone, but is that all thereis in science communication? Far from it, there are a myr-iad considerations: some political and even legal in na-ture (copyright, anyone?), but most are social. Twitter fortenure? Absurd as it may sound, there are advantages tousing social media, which some scientists are already ben-efitting from. At the very least, scientists should investin their web presence and identity. And while we mentionpresence, how about honing those presentation skills? Zenfor slides, Tufte for plots, and did you know that your fontselection can help the review of your proposal?
Lorena A. BarbaDepartment of Mechanical EngineeringBoston [email protected]
MS212
The Two Body Problem
The two body problem in physics can be solved by dealingwith a few equations, while solving the two body problemin academic career is much harder, maybe no correct solu-tion at all. Here, I just want to share my own experienceof two body problem–my struggle, attempts, failures andresult, as well as the experience of some dual-career cou-ples that I know. I hope this will be helpful for people insimilar situation.
Bo DongUniversity of Massachusetts [email protected]
MS212
Preparing for Tenure and Promotion
Preparing for tenure and/or promotion is arguably one ofthe most stressful times in one’s academic career. Whenpreparing your case, it is worth bearing in mind that mostacademic institutions have T&P processes which requireeach case be reviewed at different levels within the institu-tion before a final decision is made. I will give pointers formaking a strong case for tenure and/or promotion basedon my recent past experience making recommendations atthree of these levels within my home institution.
Misha E. KilmerTufts [email protected]
MS212
Educating Undergraduate Women in Mathematics
I will present my experiences at Spelman College, educat-ing undergraduate women in mathematics.
Monica StephensSpelman CollegeAtlanta, [email protected]
MS213
Spatial Stochastic Modelling of the Hes1 Pathway
Individual mouse embryonic stem cells have been foundto exhibit highly variable difffferentiation responses un-der the same environmental conditions. The noisy cyclicexpression of Hes1 and its downstream genes are knownto be responsible for this, but the mechanism underly-ing this variability in expression is not well understood.We show that the observed experimental data and diversedifffferentiation responses can be explained by a spatialstochastic model of the Hes1 gene regulatory network.
Marc SturrockUniversity of [email protected]
Andreas HellanderDepartment of Computer ScienceUniversity of California, Santa [email protected]
Anastasios MatzavinosIowa State [email protected]
Mark ChaplainDivision of MathematicsUniversity of [email protected]
MS213
Modeling of Stochastic Diffusion and Reaction Pro-cesses in Mixed Dimensions in Systems Biology
Abstract not available at time of publication.
Per LotstedtDepartment of Information TechnologyUppsala University, [email protected]
MS213
Intrinsic and Extrinsic Noise in Genetic Oscilla-tions
Abstract not available at time of publication.
Shev MacNamaraUniversity of OxfordOxford, [email protected]
MS213
Linear Algebra for Difference Equations, Networksand Master Equations
Abstract not available at time of publication.
Gilbert StrangMassachusetts Institute of [email protected]
MS214
Comparison of Uncertainty Quantification Meth-
CS13 Abstracts 295
ods for Nonlinear Parabolic Systems
We compare exiting probabilistic methods for propaga-tion of epistemic uncertainty in parabolic models such asthe nonlinear Richards’ equation with spatially distributed,uncertain input parameters. For the estimation of first andsecond order moments, we show that Monte Carlo methodsoutperform global and locally adaptive Stochastic Colloca-tion methods for input with short correlation lengths andhigh variances, due to an increase in problem dimensional-ity and the loss of regularity of state variables in probabilityspace.
David A. Barajas-SolanoDept. of Mechanical and Aerospace EngineeringUniversity of California, San [email protected]
Daniel M. TartakovskyUniversity of California, San [email protected]
MS214
An Implementation of Polynomial Chaos Ex-pansion on Discontinously Dependent ModelParamters
One of the main advantageous of polynomial chaos expan-sions and stochastic collocation method is it’s exponentialconvergence rate. However, this rate is lost because ofGibb’s phenomena if there is a discontinuity in the depen-dency structure between model parameters. In this talk anew all-purpose method for setting up an expansion withdependent model parameters will be presented, and fastconvergence property will be shown irrespectively of prob-ability structure.
Jonathan FeinbergSimula Research LaboratoryUniversity of [email protected]
MS214
Visualizing Gaussian Process Uncertainty usingSmooth Animations
Animations can visualize probability distributions: eachframe shows a random draw from the distribution, and iscorrelated with its neighbors so the motion is continuous.I focus on Gaussian distributions, including Gaussian Pro-cesses which are used in kriging. Existing work interpolatesbetween “keyframes’ (mutually independent random drawsfrom the distribution), but motion becomes kinky at thesekeyframes. My approach treats all frames on equal foot-ing, yielding smooth, natural-looking timetraces. Code isincluded.
Charles [email protected]
MS214
Assessment of Numerical Uncertainty in the So-lution of Inverse Problems, for Different Observa-tions Conditions
Abstract The Bayesian Inference method is applied to cali-brate parameters of the differential equation used to modeldiffusion processes. The forward model is estimated us-
ing both analytical and numerical solution of the ODE.The numerical uncertainty induced by time discretizationis studied through comparing results associated with theanalytical and numerical forward models. Also, effect oftime step size on numerical uncertainty and computationalcost is investigated.
Negin YousefpourTexas A&M [email protected]
Zenon Medina-CetinaZachry Department of Civil EngineeringTexas A&M [email protected]
Hans Petter LangtangenCenter for Biomedical ComputingSimula Research Laboratory and University of [email protected]
Are Magnus Bruaset, Stuart ClarkSimula Research [email protected], [email protected]
MS215
Sensitivity Analysis in Weak-Constraint 4D-Var:Theoretical Aspects and Applications
Suboptimal weighting of the information provided by mod-els and measurements poses a fundamental limitation onthe performance of atmospheric data assimilation systems(DAS). Theoretical aspects of adjoint sensitivity analy-sis are presented in the context of weak-constraint four-dimensional variational data assimilation. Evaluation ofthe model forecast sensitivity to the information vectorand to the DAS representation of the information errorstatistics, covariance parameter optimization, and a prioriperformance assessment are discussed.
Dacian N. DaescuPortland State UniversityDepartment of Mathematics and [email protected]
MS215
Efficient Implementations of the Ensemble KalmanFilter
This research presents efficient implementations of the en-semble Kalman filter based on Singular Value (SVD) andCholesky decompositions. In addition, a novel implementa-tion is derived from the Sherman Morrison formula. Thisdirect method exploits the special structure of the dataerror covariance matrix which, in practice, is often di-agonal. The complexity of the proposal is equivalent towell-known, efficient implementations of the EnKF. Theproposed method is tested using realistic atmospheric andoceanic models. In terms of accuracy, not significant dif-ference is shown in the results for the compared methods.Moreover, the elapsed time of the simulation is reducedwhen the proposed implementation of the EnKF is used.
Adrian SanduVirginia Polytechnic Institute andState [email protected]
Elias Nino-Ruiz
296 CS13 Abstracts
Virginia Polytechnic Institute and State [email protected]
MS215
Trust Region Adaptive POD/DEIM 4D-Var for aFinite-Element Shallow Water Equations Model
A proper orthogonal decomposition (POD) coupled withdiscrete empirical interpolation method (DEIM) constructsefficient reduced-order inverse problem four - dimensionalvariational (4D-Var) data assimilation for a nonlinear fi-nite element (FEM) shallow water equations(SWE) model.Different approaches of POD/DEIM trust-region 4D-Vardata assimilation problem are compared, including a dual-weighed method for snapshots selection. Using DEIM wereduce the computational complexity of the Trust-RegionPOD-4D-Var FEM-SWE model by decreasing the CPUtime required to calculate the solutions of the forward andadjoint models and indirectly by reducing the conditionnumber of Hessian of cost functional, thus accelerating con-vergence of minimization process.
Razvan StefanescuFlorida State [email protected]
Ionel M. NavonFlorida State UniversityDepartment of Scientific [email protected]
Xiao ChenLawrence Livermore National LaboratoryCenter for Applied Scientific [email protected]
MS215
A Hybrid Variational-ensemble Data AssimilationMethod
In this presentation we will discuss discus a possibility toselectively use methodological advantages from ensembleand variational data assimilation systems while avoidingsub-optimal features of the original algorithms. We willpresent a new development of static error covariance modelfor use in hybrid system that explores the structure of cir-culant matrices, while maintaining the complexity of exist-ing variational error covariance models.
Milija ZupanskiCooperative Institute for Research in the AtmosphereColorado State [email protected]
MS216
Data Fusion based on Coupled Matrix and TensorFactorizations
Data fusion enhances knowledge discovery, in particular,in complex data mining problems. The task of fusing data,however, is challenging since data are often incomplete,heterogeneous, i.e., in the form of higher-order tensors andmatrices, and have both overlapping and non-overlappingcomponents. We formulate data fusion as a coupled ma-trix and tensor factorization problem and propose an all-at-once optimization algorithm, which easily extends to cou-pled analysis of incomplete data sets. We demonstrate the
usefulness of our approach in metabolomics applications.
Evrim AcarUniversity of [email protected]
MS216
Data-Driven Analysis and Fusion of Medical Imag-ing Data
Data-driven methods such as independent component anal-ysis (ICA) have proven quite effective for the analysis offunctional magnetic resonance (fMRI) data and for discov-ering associations between fMRI and other medical imag-ing data types such as electroencephalography (EEG) andstructural MRI data. Without imposing strong modelingassumptions, these methods efficiently take advantage ofthe multivariate nature of fMRI data and are particularlyattractive for use in cognitive paradigms where detailed apriori models of brain activity are not available. This talkreviews major data-driven methods that have been suc-cessfully applied to fMRI analysis and fusion, and presentsexamples of their successful application for studying brainfunction in both healthy individuals and those sufferingfrom mental disorders such as schizophrenia.
Tulay AdaliUniversity of MarylandBaltimore [email protected]
MS216
MetaFac: Community Discovery via RelationalHypergraph Factorization
This work aims at discovering community structure in richmedia social networks, through analysis of time-varying,multi-relational data. Community structure represents thelatent social context of user actions. It has important ap-plications in information tasks such as search and recom-mendation. Social media has several unique challenges. (a)In social media, the context of user actions is constantlychanging and co-evolving; hence the social context con-tains time-evolving multi-dimensional relations. (b) Thesocial context is determined by the available system fea-tures and is unique in each social media website. In thiswork we propose MetaFac (MetaGraph Factorization), aframework that extracts community structures from vari-ous social contexts and interactions. Our work has threekey contributions: (1) metagraph, a novel relational hy-pergraph representation for modeling multi- relational andmulti-dimensional social data; (2) an efficient factoriza-tion method for community extraction on a given meta-graph; (3) an on-line method to handle time-varying rela-tions through incremental metagraph factorization. Exten-sive experiments on real-world social data collected fromthe Digg social media website suggest that our techniqueis scalable and is able to extract meaningful communitiesbased on the social media contexts. We illustrate the use-fulness of our framework through prediction tasks. Weoutperform baseline methods (including aspect model andtensor analysis) by an order of magnitude.
Jimeng SunIBM T.J. Watson Research [email protected]
CS13 Abstracts 297
MS216
Looking for Common Features Across a Collectionof Matrices Using the Higher-Order GSVD
The higher-order GSVD is a way of simultaneously reduc-ing each matrix in a collection {D1, . . . , DN} to a form thatpermits one to identify common features. Each Di has thesame number of columns. The invariant subspace associ-ated with the minimum eigenvalue of the very nasty matrix(S1 + . . . + Sn)(inv(S1) + . . . + inv(SN )) is involved where
Si = D′iDi. However, we are able to get at this subspace
safely via the minimization of a very interesting quadraticform. Everything reverts to the ordinary generalized SVDwhen N = 2.
Charles Van LoanCornell UniversityDepartment of Computer [email protected]
Orly AlterSci Comp & Imaging (SCI) Inst, Bioeng & HumanGenetics DeptsUniversity of [email protected]
Sri Priya [email protected]
Michael A. SaundersSystems Optimization Laboratory (SOL)Dept of Management Sci and Eng, [email protected]
MS217
Optimal Experimental Design under Uncertainty
We focus on large-scale optimal experimental design forstatistical inverse problems. We consider a model involv-ing recovery of initial concentration field in an advection-diffusion problem. We use a Bayesian inference frameworkwith Gaussian prior and likelihood. Due to linearity of theparameter-to-observable map, the posterior is also Gaus-sian, with mean and covariance given by solution of a de-terministic inverse problem. The goal is to optimize the lo-cation of observation points to maximize information gain.
Alen AlexanderianUniversity of Texas at [email protected]
Noemi PetraInstitute for Computational Engineering and Sciences(ICES)The University of Texas at [email protected]
Georg Stadler, Omar GhattasUniversity of Texas at [email protected], [email protected]
MS217
Joint Inversion
Inverse problems are inheritable non-unique and regular-ization is needed to obtain stable and reasonable solutions.The regularization adds information to the problem and
determines which solution, out of the infinitely many, isobtained. In this talk we discuss the case when a-prioriinformation exists in the form of either known structureor in the form of another inverse problem for a differentproperty. The challenge is to include such information inthe inversion process.
Eldad HaberDepartment of MathematicsThe University of British [email protected]
MS217
Approximate Dynamic Programming for Sequen-tial Bayesian Experimental Design
Optimal design maximizes the value of costly experimentaldata. Popular approaches for designing multiple experi-ments are suboptimal: these include open-loop approachesthat choose all experiments simultaneously, or greedy meth-ods that optimally design the next experiment withoutaccounting for the future. We instead formulate experi-mental design in a closed-loop dynamic programming (DP)framework that yields the true optimal sequence of ex-periments. We solve the DP problem using methods ofapproximate dynamic programming, coupled with polyno-mial surrogates and stochastic optimization. Preliminaryresults demonstrate the superiority of the closed-loop ap-proach.
Xun Huan, Youssef M. MarzoukMassachusetts Institute of [email protected], [email protected]
MS217
Estimating and using Information in Inverse Prob-lems
Information theory provides convenient mechanisms tomanage and unify various components of the inversion pro-cess especially when uncertainty has to be addressed. Herewe calculate information densities from adjoints to guidemesh adaptivity, regularization parameters, and sensor lo-cation identification. These aspects have historically beenhandled in an ad-hoc and incohesive manner. Our in-formation density approach not only provides a unifyingmechanism, but also demonstrates improved performanceon a convection-diffusion problem in comparison to stan-dard methods.
Bart VanbloemenwaandersSandia National LaboratoriesOptimization and Uncertainty Quantification [email protected]
Wolfgang BangerthTexas A&M [email protected]
MS218
SciHadoop: Array-based Query Processing inHadoop
Hadoop has become the de facto platform for large-scaledata analysis in commercial applications and increasinglyin scientific applications. Scientific data, typically struc-tured but stored in binary file formats, enables different as-sumptions and optimizations than unstructured data. Our
298 CS13 Abstracts
system, SciHadoop, utilizes knowledge of that structureto yield big performance gains by creating efficient inputsplits; enabling holistic combiners; minimizing and localiz-ing communications; producing early results; and produc-ing balanced, contiguous output.
Joe BuckUC Santa [email protected]
MS218
What is MapReduce and How can it Help with mySimulation Data?
Abstract not available at time of publication.
Paul ConstantineStanford UniversityMechanical [email protected]
David F. GleichPurdue [email protected]
MS218
Benchmarking MapReduce Implementations forScientific Applications
With data production increasing rapdily due to the evergrowing application needs, the MapReduce model and itsimplementations need to be further evaluated and opti-mized. Several MapReduce frameworks with various de-grees of conformance to the key tenets of the model areavailable today, each, optimized for specific features. Wewill discuss the design of a standard benchmark suite forquantifying, comparing, and contrasting the performanceof MapReduce platforms under a wide range of representa-tive use cases found in scientific applications. The aim ofthe benchmark is to allow scientists to choose the MapRe-duce implementation that best suits their application’sneeds.
Madhusudhan GovindarajuSUNY [email protected]
MS218
Dynamic Mode Decomposition with MapReduce
Dynamic mode decomposition (DMD) is a spectral analysistechnique that identifies coherent features of a fluid flow,based on a series of snapshots of the flow field (Schmid,2010). In order to treat databases exceeding 1 Tb in size,we develop an implementation of the DMD algorithm basedon MapReduce tall-and-skinny QR factorization (Constan-tine & Gleich, 2011). The method is demonstrated on athree dimensional simulation database of a screeching su-personic jet.
Joseph W. NicholsStanford [email protected]
MS219
Certified Parameter Optimization with Reduced
basis Surrogate Models
PDE-constrained parameter optimization problems witharbitrary output functionals may be quite expensive, whenusing standard PDE-solvers. Instead we use RB surrogatemodels for approximately solving the optimization prob-lem. Ingredients of our scheme comprise RB-spaces for thesolution, its sensitivity derivatives and rigorous a-posteriorierror bounds for the solution, derivatives, output func-tional and the suboptimal parameters. Experiments onan instationary convection-diffusion problem demonstratethe benefits of the approach.
Markus Dihlmann, Bernard HaasdonkUniversity of [email protected],[email protected]
MS219
A Certified Reducedbasis Approach for Parametrized Linear-quadraticOptimal Control Problems
In this talk, we discuss effective model reduction oflinear-quadratic optimal control problems governed byparametrized elliptic and parabolic partial differentialequations. To this end, we employ the reduced basismethod as a surrogate model for the solution of the op-timal control problem and develop rigorous and efficientlyevaluable a posteriori error bounds for the optimal controland the associated cost functional. Numerical results arepresented to confirm the validity of our approach.
Mark Kaercher, Martin GreplRWTH Aachen [email protected],[email protected]
MS219
Goal-oriented Inference for Nonlinear PDE-constrained Inverse Problems
High-dimensional parameter spaces pose a significant chal-lenge to existing inference methods, particularly in the con-text of limited data. When the quantity of interest dependson the parameter estimate, computational effort may bewasted resolving components of the parameter not requiredto make accurate predictions. We present a goal-orientedinference method that exploits the context of required pre-dictions. Previous work on linear problems is extended tothe nonlinear setting.
Chad E. [email protected]
Karen E. WillcoxMassachusetts Institute of [email protected]
MS219
Use of Reduced Order Models in Iterative Opti-mization Solvers
When solving nonlinear optimization problems with par-tial differential equations, information from the previousiteration is sometimes useful to accelerate the convergenceof the nonlinear optimization solver. We show how to useinformation obtained from a reduced order model in the
CS13 Abstracts 299
previous iterations in the solver of a quadratic subproblem,for example. This leads to deflated MINRES or CG algo-rithms, where we consider the convergence of these meth-ods. We give numerical results for an optimization problemwith a nonlinear parabolic partial differential equation inthe 3-dimensional case.
Ekkehard W. SachsUniversity of TrierVirginia [email protected]
Xuancan YeUniversity of [email protected]
MS220
Using Shannon Entropy to Estimate Convergenceof Cmfd-Accelerated Monte Carlo
Coarse Mesh Finite Difference (CMFD)[Smith, TANS, v44,1984] is a common moment based acceleration techniquethat can be used to solve transport equations. The statisti-cal error that arises from CMFD-accelerated Monte Carlocan mask the convergence of the CMFD correction val-ues. This work demonstrates how Shannon entropy[Uekiand Brown, ANS, 2003] can be used to determine when aCMFD iteration is near a convergence limit created by thestatistical noise of the Monte Carlo method.
Mathew ClevelandLawrence Livermore National [email protected]
Todd PalmerOregon State [email protected]
Nick GenitleLawrence Livermore National [email protected]
MS220
Monte Carlo Simulation Methods in Moment-based Scale-bridging Algorithm for Neutral Par-ticle Transport Problems
Recently, we have extended a moment-based scale-bridgingalgorithm to thermal radiative transfer problems. The al-gorithm accelerates a solution of the high-order (HO) ra-diation transport equation using a ”discretely consistent”low-order (LO) continuum system. In this talk, we dis-cuss a recent progress of the use of Monte Carlo simula-tion methods, which are specifically tailored to the scale-bridging algorithm, such as an energy conserving tally andasymptotic approximation for optically thick spatial cells.
Hyeongkae ParkLos Alamos National [email protected]
Jeff Densmore, Allan [email protected], [email protected]
Dana KnollLos Alamos National Laboratory
Rick [email protected]
MS220
Acceleration of Nuclear Reactor Neutronics Eigen-value Problems with Non-Linear Low-Order Pro-jection Operators
Nuclear reactor core analyses using continuous-energyMonte Carlo neutron transport models have been acceler-ated using low-order operator projections onto coarse spa-tial and energy mesh diffusion theory models. A disconti-nuity factor formulation of the diffusion equations (whichallow exact preservation of Monte Carlo solution on thecoarse mesh) is used to produce deterministic eigenvalueequations at each stage of the Monte Carlo solution. Con-verged eigenvectors obtained by solving the low-order dif-fusion equations are then used to scale and prolong theMonte Carlo source distributions, resulting in significantreductions in the number of inactive and active historiesneeded to compute reliable reactor spatial power distribu-tions and eigenvalues.
Kord Smith, Benoit Forget, Bryan Herman, Paul RomanoMassachusetts Institute of [email protected], [email protected], [email protected],[email protected]
MS220
A Hybrid Approach to Nonlinear Acceleration ofTransport Criticality Computations
In nuclear engineering, the computation of the dominanteigenvalue of the neutron transport equation is highly im-portant for the analysis of nuclear reactors. Recent workin the field has led to advancements in efficient determin-istic numerical techniques for computing this eigenvalue,including Nonlinear Diffusion Accelerated Power Iteration(NDA-PI). In an effort to obtain higher accuracy solutions,it is becoming more popular to use Monte Carlo simulationsto compute the eigenvalue and associated eigenvector, asthese simulations allow for a continuous treatment of space,angle and energy. We look to build hybrid solvers that com-bine the efficiency of NDA-PI with the high-accuracy solu-tions provided by otherwise expensive Monte Carlo simu-lations.
Jeffrey A. WillertNorth Carolina State [email protected]
MS221
DPG Methods for Transport and the Inviscid EulerEquations
The DPG method, introduced by Demkowicz andGopalakrishnan, is a Petrov-Galerkin method derived fromthe minimization of the operator residual associated witha variational formulation. For transport under the ultra-weak variational formulation, DPG demonstrates optimalconvergence for cases in which the upwind DG methodis suboptimal. Unfortunately, the DPG solution for sys-tems of transport equations manifests problems related tosolution regularity in the crosswind direction; we discussthese issues in detail and propose regularization solutions
300 CS13 Abstracts
through the vanishing viscosity method.
Jesse L. ChanUniversity of Texas at [email protected]
MS221
Comparison of Multigrid Performance for Stabi-lized and Algebraic Flux Correction FEM for Con-vection Dominated Transport
In this empirical study we will compare the performanceof Algebraic Multigrid (AMG) applied to stabilized andalgebraic flux corrected discretizations of convection dom-inated problems. The Algebraic Flux Corrected transportalgorithm considered in this study results in an operatorwith the M-matrix property which has theoretical prop-erties more amenable to AMG as compared to stabilizedmethods. The performance of these methods on bench-mark convection-diffusion and fluid flow problems will bedemonstrated.
Eric C. CyrScalable Algorithms DepartmentSandia National [email protected]
John N. ShadidSandia National [email protected]
Roger PawlowskiMultiphysics Simulation Technologies Dept.Sandia National [email protected]
Dmitri KuzminUniversity [email protected]
MS221
Characteristic Discontinuous Galerkin for TracerAdvection in Climate Modeling
We will describe our development of Characteristic Discon-tinuous Galerkin (CDG) for tracer advection, on arbitraryconvex polygon meshes, for use in climate modeling. CDGcan be viewed as an extension of Prather’s 1986 momentmethod to unsplit advection on general meshes and higherorders of accuracy. Methods for bounds preservation willbe discussed, and comparisons made with Runge-KuttaDG (Cockburn & Shu), semi-Lagrangian DG (Restelli,Bonaventura, & Sacco), and Hancock-Huynh DG (Lo &Van Leer).
Robert B. LowrieLos Alamos National LaboratoryLos Alamos, NM 87545 [email protected]
Todd RinglerLos Alamos National [email protected]
MS221
A Semi-Lagrangian Discontinuos Galerkin Trans-
port Scheme on the Cubed Sphere
Because of its geometric flexibility and high parallel ef-ficiency, DG (discontinues Galerkin) method is becomingincreasingly popular in atmospheric and ocean modeling.However, a major drawback of DG method is its strin-gent CFL stability restriction associated with explicit time-stepping. We adopt a dimension-splitting approach wherea regular semi-Lagrangian (SL) scheme is combined withthe DG method, permitting longer time steps. The SLDGscheme is inherently conservative and has the option toincorporate a local positivity-preserving filter for tracers.The SLDG scheme is tested for various benchmark advec-tion test-suites on the cubed-sphere and results will be pre-sented in the seminar.
Ram NairNCARInstitute for Mathematics Applied to [email protected]
Wei GuoUniversity of [email protected]
MS222
Mesoscale Investigations of the Influence of Cap-illary Heterogeneity on Multiphase Flow of Fluidsin Rocks
We are focused on understanding the influence of mesoscaleheterogeneity on multiphase flow of CO2 and brine. Thislargely ignored spatial scale provides the opportunity to de-velop a rigorous understanding of major controls on CO2migration when a continuum description of the multiphasephysics is applicable. Methods have been developed tocharacterize subcore scale capillary heterogeneity, to per-form high-resolution simulations that that accurately repli-cate experiments, and accurately predict multiphase flowin complex rocks.
Sally M. BensonEnergy Resources EngineeringStanford [email protected]
MS222
Multiscale Algorithms for Reactive Transport inPorous Media
I will present two multiscale methods including a Langevinapproach and a dimension reduction method based on acomputational closure. The purpose of these methods isto provide an accurate description of the system averageswhile retaining critical pore-scale information. The advan-tages, range of applicability and limitations of the men-tioned above multiscale methods will be discussed.
Alexandre TartakovskyPacific Northwest National [email protected]
MS222
Pore Scale Reactive Transport Modeling usingAdaptive, Finite Volume Methods with a Look to-ward Upscaling
We have developed a high performance simulation capabil-ity to model pore scale multi-component reactive transport
CS13 Abstracts 301
in geologic subsurface media, especially that obtained fromimage data. Our approach is conservative, accurate and ro-bust, based on adaptive, finite volume methods. As such,the adaptive functionality can be used to provide communi-cation between models at different spatial scales, and, thus,provide the ability to directly upscale local pore scale datato the Darcy scale through flux matching techniques.
David TrebotichLawrence Berkeley National [email protected]
MS222
Geometric Comparisons in Porous Media Simula-tion
We extract geometric descriptors that characterize flowproperties and the pore space of a material. By measur-ing distances between the descriptors, we compare multiplematerials.
Gunther H. WeberLawrence Berkeley [email protected]
Dmitriy MorozovLawrence Berkeley National [email protected]
MS223
Rational Krylov Subspace Methods for TransientElectromagnetic Geophysical Forward Modeling
In recent work we explored how the initial value problemfor the quasi-static Maxwell’s equations arising in transientelectromagnetic modeling (TEM) can be solved efficientlyin the frequency domain by solving a small projected modelfor each relevant frequency using simple shift-and-inverttype Krylov subspace projection much in the style of clas-sical model order reduction for linear time-invariant controlproblems. In this work we present more advanced rationalKrylov subspace methods employing more elaborate poleselection techniques. We also compare this frequency do-main approach with solving the problem in the time do-main using the same rational Krylov subspace methods toapproximate the action of the matrix exponential.
Oliver G. ErnstTU Bergakademie FreibergFakultaet Mathematik und [email protected]
Stefan GuttelSchool of MathematicsThe University of [email protected]
Ralph-Uwe BornerInstitute of GeophysicsTU Bergakademie [email protected]
MS223
Rational Approximations through Finite Element
Discretization
Abstract not available at time of publication.
Murthy N. GuddatiNorth Carolina State [email protected]
MS223
Spectrally Adaptive Rational Quadrature ofMarkov Functions
Rational Arnoldi is a powerful method for approximatingexpressions of the form f(A)b or bHf(A)b, where A is alarge sparse matrix, f(A) is a matrix function, and b is avector. The selection of asymptotically optimal parame-ters for this method is crucial for its fast convergence. Wepresent and investigate a novel strategy for the automatedparameter selection when the function f is of Markov type,such as the square root or the logarithm. The performanceof this approach is demonstrated by numerical examplesinvolving symmetric and nonsymmetric matrices. This isjoint work with Leonid Knizhnerman.
Stefan GuettelUniversity of [email protected]
MS223
Recursive Relations for Rational Krylov Methods
Abstract not available at time of publication.
Lothar ReichelKent State [email protected]
MS224
Reproducible Research and Omics: Thoughts fromthe IOM Review
Between 2007 and 2010, several genomic signatures wereused to guide patient therapy in clinical trials in cancer.Unfortunately, the signatures were wrong, and trials pro-ceeded despite warnings to this effect. The Institute ofMedicine (IOM) subsequently reviewed the level of evi-dence that should be required in such situations. Many rec-ommendations focus on reproducibility and data integrity,including directives to funders, journals, and regulatoryagencies. We briefly review the report and implications forreproducible research.
Keith BaggerlyDepartment of Bioinformatics and Computational BiologyUT MD Anderson Cancer [email protected]
MS224
A Portrait of One Scientist as a Graduate Student
In this talk, I will focus on the how of reproducible re-search. I will focus on specific tools and techniques I havefound invaluable in doing research in a reproducible man-ner. In particular, I will cover the following general top-ics (with specific examples in parentheses): version controland code provenance (git), code verification (test drivendevelopment, nosetests), data integrity (sha1, md5, git-annex), seed saving ( random seed retention ) distribution
302 CS13 Abstracts
of datasets (mirroring, git-annex, metalinks), light-weightanalysis capture ( ttyrec, ipython notebook)
Paul IvanovRedwood Center for Theoretical NeuroscienceUniversity of California, [email protected]
MS224
Reproducible Research in Graduate Education inthe Computational Sciences
Instilling good habits of reproducible computational re-search can be woven throughout graduate student educa-tion. Current software tools allow drafting of live doc-uments that contain theoretical derivations, generationof computational code, verifiable execution of code, andpreparation of reports. The talk presents experience inthis approach in graduate courses at UNC.
Sorin MitranUniversity of North Carolina Chapel [email protected]
MS224
Publishing Reproducible Research: Thoughts onJournal Policy
I will also discuss the role that journals can play in en-couraging reproducible research and will review the recentreproducibility policy at the journal Biostatistics.
Roger D. PengDepartment of BiostatisticsJohns Hopkins Bloomberg School of Public [email protected]
MS225
Modeling Gene Regulatory Networks throughBayesian Structure Learning
Gene regulatory networks modeled as Bayesian networksare known to provide high quality inference. However, ex-act Bayesian inference is NP-hard. In this talk, we presentscalable parallel exact and heuristic methods for Bayesianinference. The exact method achieves both work and spaceoptimality compared to the serial algorithm. The heuristicmethod operates by predicting and subsequently refiningparent sets of the genes in the network. Experimental re-sults are presented using synthetic and model pathways.
Olga NikolovaGraduate studentIowa State [email protected]
Jaroslaw Zola, Srinivas AluruIowa State [email protected], [email protected]
MS225
PASQUAL: Parallel Techniques for Next Genera-tion Genome Sequence Assembly
Short-read assembly of a genome from sequence data gath-ered from Next Generation Sequencing (NGS) platformshas become a highly challenging and fundamental problemin bioinformatics. In this talk we discuss algorithms and
compressed string graph-based data structures for sharedmemory parallel architectures that address several compu-tational challenges of short-read assembly. Experimentson up to 40 cores demonstrate that our method deliversthe fastest time-to-solution and the best trade-off betweenspeed, memory consumption, and solution quality.
Xing Liu, Pushkar PandeSchool of Computational Science and EngineeringGeorgia Institute of [email protected], [email protected]
Henning MeyerhenkeKarlsruhe Institute of TechnologyInstitute of Theoretical [email protected]
David A. BaderGeorgia Institute of [email protected]
MS225
Theory, Application and Challenges for Graph-theoretic Models in Computational Biology
With the combinatorics inherent in the field of compu-tational biology alongside a recent emergence of high-throughput instruments for data generation, the role ofgraph-theoretic modeling has been elevated to take a centerstage in biological discovery. In this overview talk, we dis-cuss the application potential and the imminent challengesof graph-theoretic analytics in modern-day computationalbiology viz. sheer volume of data, diversity in type, anda quest to reveal hidden data interconnectivity at differentlevels.
Ananth KalyanaramanAssociate ProfessorSchool of EECS, Washington State [email protected]
MS225
Graph Algorithms in Flow Cytometry
Flow cytometry (FC) measures fluorescence of proteins insingle cells, enabling us to identify cellular subpopulationsrelevant to immunology, cell signaling, and oncology. Re-cent experimental developments lead to large-scale, highdimensional FC data, necessitating novel combinatorial al-gorithms for analyzing the data. We describe algorithmsfor registering cell populations across multiple samples, ex-perimental conditions and times. The combinatorics em-ployed include edge covers of minimum weight and severalconcepts from phylogenetic combinatorics.
Ariful AzadPurdue [email protected]
Arif KhanDepartment of Computer SciencePurdue [email protected]
Bartek RajwaBindley Bioscience CenterPurdue [email protected]
CS13 Abstracts 303
Alex PothenPurdue UniversityDepartment of Computer [email protected]
MS226
Approximation of the Fokker-Planck Equation ofFENE Dumbbell Model
We propose a new weighted weak formulation for theFokker-Planck equation of FENE dumbbell model, andprove its well-posedness in weighted Sobolev spaces.We also propose simple and efficient semi-implicit time-discretization schemes and prove that they are uncondi-tionally stable. We then construct two Fourier-Jacobispectral-Galerkin algorithms which enjoy the followingproperties: (i) it is unconditionally stable, spectrally ac-curate and of optimal computational complexity; (ii) itconserves the volume naturally, and provide accurate ap-proximation to higher-order moments of the distributionfunction; and (iii) it can be easily extended to couplednon-homogeneous systems. Extension to full Navier-StokesFokker-Planck system by using sparse spectral methodswill also be discussed.
Jie ShenDept. of MathematicsPurdue [email protected]
MS226
Modeling Tissue Self-assembly in Bio-fabricationusing Kinetic Monte Carlo Simulations
We present a three-dimensional lattice model to study self-assembly and fusion of multicellular aggregate systems byusing kinetic Monte Carlo (KMC) simulations. This modelis developed to describe and predict the time evolution ofpostprinting morphological structure formation during tis-sue or organ maturation in a novel biofabrication process(or technology) known as bioprinting. In this new tech-nology, live multicellular aggregates as bio-ink are used tomake tissue or organ constructs via the layer-by-layer de-position technique in biocompatible hydrogels; the printedbio-constructs embedded in the hydrogels are then placedin bioreactors to undergo the self-assembly process to formthe desired functional tissue or organ products. Here weimplement our model with an efficient KMC algorithm tosimulate the making of a set of tissues/organs in severaldesigner’s geometries like a ring, a sheet and a tube, whichcan involve a large number of cells and various other sup-port materials like agarose constructs etc. We also studythe process of cell sorting/migration within the cellular ag-gregates formed by multiple types of cells with differentadhesivities.
Yi SunUniversity of South [email protected]
MS226
Theoretical and Computational Advances in Mod-eling Active Nematic Liquid Crystal Polymers andits Applications
Active liquid crystal polymers can be found in many man-made and natural material systems. Their features includespontaneous local molecular orientation and self-propelled
molecular motion. The F-actin in cytoplasm is an out-standing example of the active material system. The tra-ditional (passive) liquid crystal theory has been modifiedto account for the active (nonequilibrium) forcing. In thistalk, I will discuss a mathematical analysis of a class ofpolar nematic active liquid crystal models in simple shear,Poiseuille flows, and other simple geometries under variousstress balance conditions. 2-D numerical simulation withrespect to two types of boundary conditions will be dis-cussed as well. A rich set of spatial-temporal patterns incollective molecular orientation and flows will be revealedand so will be their relation to the active parameters in themodel.
Qi WangUniversity of South [email protected]
MS226
Numerical Stability for Incompressible Euler Equa-tion
Fully discrete pseudo spectral numerical schemes to 2-Dand 3-D incompressible Euler equation are presented inthis talk. To ensure the numerical stability for this inviscidequation, an artificial viscosity is added with a requirednumerical accuracy. A local in time convergence analysis isestablished for smooth solution and a global in time energystability is assured with a suitable choice of the artificialviscosity term.
Cheng WangUniversity of MassachusettsDartmouth, [email protected]
Sigal GottliebDepartment of MathematicsUniversity of Massachusetts [email protected]
MS227
Model Form Issues in Uncertainty Quantification
Abstract not available at time of publication.
Michael S. EldredSandia National LaboratoriesOptimization and Uncertainty Quantification [email protected]
MS227
Extreme-Scale Stochastic Inversion
We present a computational framework for solution ofdiscretized infinite-dimensional Bayesian inverse problems.We address several computational issues related to theappropriate choice of prior, consistent discretizations,tractable treatment of the Hessian of log posterior, andscalable parallel MCMC algorithms for sampling the pos-terior. We apply the framework to the problem of globalseismic inversion, for which we demonstrate scalability to1M earth parameters, 630M wave propagation unknowns,and 100K cores on the Jaguar supercomputer.
Tan Bui-ThanhThe University of Texas at [email protected]
304 CS13 Abstracts
Carsten BursteddeInstitut fuer Numerische SimulationUniversitaet [email protected]
Omar GhattasUniversity of Texas at [email protected]
James R. MartinUniversity of Texas at AustinInstitute for Computational Engineering and [email protected]
Georg StadlerUniversity of Texas at [email protected]
Lucas [email protected]
MS227
Bayesian Inference of Wind Drag using AXBTData
An adapative, sparse, pseudospectral sampling algorithmis applied to efficiently construct a polynomial chaos surro-gate of the response of the ocean circulation to uncertaindrag parameters. The surrogate is then exploited to in-fer uncertain drag parameters, using AXBT data collectedduring typhoon Fanapi. The analysis leads to sharp esti-mates for the saturation of the wind drag coefficient andthe corresponding wind speed; however, the data was notinformative regarding the drag behavior at higher speeds.
Ihab SrajDuke [email protected]
Mohamed IskandaraniRosenstiel School of Marine and Atmospheric SciencesUniversity of [email protected]
Ashwanth Srinivasan, Carlisle ThackerUniversity of [email protected], [email protected]
Justin WinokurDuke [email protected]
Alen AlexanderianJohns Hopkins [email protected]
Chia-Ying Lee, Shuyi ChenUniversity of [email protected], [email protected]
Omar M. KnioDuke [email protected]
MS227
Multiscale Methods for Large-Scale Bayesian In-
version
In many ill-posed inverse problems, especially those involv-ing transport in porous media, the forward model smoothsthe input parameters. In this context, we can exploit condi-tional independence of scales, identified through multiscaleforward solvers, to efficiently sample the Bayesian poste-rior. We sample a low-dimensional coarse-scale problemwith MCMC and “project’ these samples to the fine scalewith approximate iterative techniques. This approach iswell-suited to parallel computation. We will describe notonly the algorithm but its implementation via a new high-level inversion toolbox.
Matthew Parno, Youssef M. MarzoukMassachusetts Institute of [email protected], [email protected]
MS228
Iterative Methods and Spectral Approximation ofFast Rotating Gross-Pitaevskii Equations
The aim of this talk is to get fast converging pseudospec-tral methods for computing stationary solutions to theGross Pitaevskii equation with large rotations and nonlin-earites. The pseudospectral method is standard and basedon FFTs. Concerning the numerical solution of the result-ing non explicit linear system, we show that usual methodsused in the literature makes the solver diverge. We pro-pose to use Krylov solvers and build some physics basedpreconditioners that provide substantial gains in term ofcomputational times to reach the ground state with veryhigh accuracy. Numerical validations for 2D and 3D casewill support our approach.
Xavier L. AntoineInstitut Elie Cartan Nancy (IECN)Universite de [email protected]
Romain DuboscqIECN, Nancy, [email protected]
MS228
Improved Sobolev Gradient Methods for Solvingthe Stationary Gross-Pitaevskii Equation with Ro-tation
We compute vortex states of a rotating Bose-Einstein con-densate by direct minimization of the Gross-Pitaevskii en-ergy functional. We extensively compare different mini-mization algorithms with improved steepest descent meth-ods based on Sobolev gradients. In particular, we showthat mixed Newton-Sobolev gradient methods offer appeal-ing convergence properties. Advantages and drawbacks ofeach method are summarized. We present numerical se-tups using 6th order finite difference schemes and finiteelements with mesh adaptivity that were successfully usedto compute a rich variety of difficult cases with quantizedvortices.
Ionut DANAILALaboratoire de Mathematiques Raphael Salem, UniversitedeRouen, [email protected]
Parimah Kazemi
CS13 Abstracts 305
Ripon College, Ripon, Wisconsin, [email protected]
MS228
Dimension Reduction of the Nonlinear SchrdingerEquation with Coulomb Interaction underAnisotropic Potentials
We present a rigorous dimension reduction analysis fromthe 3D ( i.e. in 3 spatial dimensions) Schrodinger-Poissonsystem (SPS) to lower dimensional models, arising in thelimit of infinitely strong confinement in two or one space di-mensions, namely the Surface Adiabatic Model (SAM) andthe Surface Density Model (SDM) in 2D or Line AdiabaticModel (LAM) in 1D. In particular, we explain and demon-strate that the 2D Schrodinger-Poisson Model (SPM) is notappropriate to simulate a ”2D electron gas” of point par-ticles confined into a plane (or more general 2D manifold),whereas SDM as the correct model can be successfully ap-plied by utilizing the inversion of Square root of Laplacianto describe the Coulomb interaction. Finally, we study theground state and dynamics of SDM in different setups.
Yong ZhangWolfgang Pauli Institute, Wien, [email protected]
Weizhu BaoDept. of Mathematics and Center for ComputationalScience [email protected]
Huaiyu JianDept. of Mathematical Sciences, Tsinghua [email protected]
Norbert MauserWolfgang Pauli Institute c/o Fak. [email protected]
MS228
Numerical Methods for Rotating Dipolar BECbased on a Rotating Lagrange Coordinate
In the talk, a simple and efficient numerical method forstudying the dynamics of rotating dipolar BEC is pre-sented. Using coordinate transformation, we eliminate theangular rotational term from the Gross-Pitaevskii equation(GPE). Then we develop pseudospectral type methods tosolve the GPE. The accuracy of the method is discussed.In addition, the dynamics of quantized vortex lattice inrotating dipolar BEC are studied.
Yanzhi ZhangMissouri University of Science and [email protected]
Weizhu BaoNational University of SingaporeDepartment of [email protected]
Qinglin TangNational University of [email protected]
MS229
A New Spectral Method for Numerical Solution ofthe Unbounded Rough Surface Scattering Problem
I shall present a new spectral method for solving the un-bounded rough surface scattering problem. The methoduses a transformed field expansion to reduce the boundaryvalue problem with a complex scattering surface into a suc-cessive sequence of transmission problems of the Helmholtzequation with a plane surface. Hermite orthonormal basisfunctions are used to further simply the transmission prob-lems to fully decoupled one-dimensional two-point bound-ary value problems with piecewise constant wavenumbers,which can be solved efficiently by a Legendre-Galerkinmethod. Ample numerical results will be presented todemonstrate the new spectral method is efficient, accu-rate, and well suited to solve the scattering problem byunbounded rough surfaces.
Ying He, Peijun LiDepartment of MathematicsPurdue [email protected], [email protected]
Jie ShenPurdue UniversityDepartment of [email protected]
MS229
Spectral Element Discontinuous Galerkin LatticeBoltzmann Method for Convection Heat Transfer
Spectral Element Discontinuous Galerkin (SEDG) LatticeBoltzmann Method (LBM) employs body-fitted unstruc-tured mesh and is capable of dealing with complex ge-ometry with high-order accuracy. Explicit time march-ing, diagonal mass matrix and minimum communicationcost for numerical flux make SEDG-LBM very efficient flowsolver. We present SEDG-LBM algorithms based on two-distribution function and hybrid approaches and comparethem for convection heat transfer problems. Various typesof boundary conditions will be described in detail.
Saumil Patel, Kalu Chibueze UgaDepartment of Mechanical EngineeringThe City College of New [email protected], [email protected]
MiSun MinArgonne National LaboratoryMathematics and Computer Science [email protected]
Taehun LeeDepartment of Mechanical EngineeringThe City College of New [email protected]
MS229
Boundary Perturbation Methods for Surface Plas-mon Polaritons
Surface Plamson Polaritons (SPPs) arise in a number oftechnologically important applications including enhance-ment of signals, highly sensitive sensing, and the design ofmetamaterials. In this talk we discuss several new classesof Boundary Perturbation Methods for simulating SPPswhich are accurate, robust, and extremely fast. We will
306 CS13 Abstracts
outline the numerical methods, display simulation resultsfor SPP configurations, and discuss how these algorithmscan be effectively utilized for optimization and design ofdevices.
David P. NichollsUniversity of Illinois at [email protected]
MS229
Boundary Treaments for Second Order Wave Equa-tions by Pseudospectral and Runge-Kutta-NystrmMethods
Many wave phenomena are described by second order waveequations, for instance, general relativity, acoustics, andelectromagnetic waves. For these wave problems, the do-main may be large compared the wavelength, and the waveshave to propagate long distances. Hence, numerically solv-ing these problems requires long time integration and ac-cumulation of numerical dispersion error may affect thesimulation quality. It can be shown that in preserving lowaccumulation of dispersion error during long time integra-tion, high-order accurate methods are more efficient thanthe low-order methods. However, high-order schemes arevery sensitive to the imposition of boundary conditions,and great care must be exercised to ensure stable compu-tations of high-order schemes. In this talk we present ahigh-order scheme based on the pseudospectral Legendreapproximation in space and the Runge-Kutta-Nystrom al-gorithm in time, which can be adopted in a multi-domaincomputational framework, to solve second order wave equa-tions. The key toward to the success of constructing sucha stable scheme hinges upon properly imposing penaltyboundary conditions at every collocation equations. Weshall use one dimensional space problems to illustrate theconceptual ideas of the methods, and special attention ispaid upon analyzing the stability of the scheme subject tovarious types of boundary conditions, including Dirichlet,Neumann, Robin, and material interface boundary condi-tions. Through conducting energy estimates it is shownthat the scheme can be made stable by properly choos-ing the penalty parameters. The present one-dimensionalscheme can be generalized for multidimensional problemseven defined on curvilinear coordinates. Numerical exper-iments for model problems defined on one and two dimen-sional spaces are conducted, and we observe the expectedconvergence results.
Chun-Hao TengDepartment of Applied MathematicsNational Chiao Tung [email protected]
MS230
Excelling: Transition from School to Aerospace andDefense Industry
There are commonalities between being a graduate re-searcher and working in the defense industry - both requirewell defined scope of the problem and increasingly the workin done in collaborative teams. There are, however, alsosignificant differences. School environment is usually a veryprotective, productive, and creative one. You are taughtto raises expectations from your own self. Excellence isdefined by your capability to learn. The transition fromschool where the focus is on learning without constraintsto the defense industry, which is governed by requirementsset by the customer, can be a very turbulent one. In a
generic defense industry, a mark of excellence is how wellone learns to adapt to the hierarchy and conforms to acertain behavior pattern. The work in graduate school ex-pects a very out of box thinking and a solution over longertime scales than the industry which expects a very processoriented thinking. Since the goals are so different, transi-tioning from one environment to another can be turbulent,especially, if you are a strong and intelligent woman engi-neer/scientist.
Roochi ChopraRaytheonroochi [email protected]
MS230
A Dual Career: Experiences as a Researcher and aProgram Manager
In this talk I will present an overview of my experiences as aformer academic (and still active researcher) and a programmanager for applied and computational mathematics at theAir Force Office of Scientific Research (AFOSR), and willdiscuss the challenges and rewards of pursuing this dualcareer.
Fariba FahrooAir Force Office of Scientific [email protected]
MS230
Experiences in Industry
In this talk, I will discuss my experience as an industrialpost-doc at IMA UMN and as a researcher in industrialenvironment at IBM Watson research lab.
Raya HoreshInstitute of Mathematics and its ApplicationsUniversity of [email protected]
MS230
The Best of Both: Federally Funded Research andDevelopment Center (FFRDCs) at the Juncture ofIndustry and Academia
The work performed at Federally Funded Research and De-velopment Centers straddles academic research and indus-trial design and development. Such an environment pro-vides insight into the advancement of new technologies andanalysis along with insight into bringing these advancesinto real world designs. The panelist will focus on the bene-fits and drawbacks of working concurrently on an expansivearray of projects as well as the characteristics necessary tobe successful in such a career.
Julia MullenComputing & Communications CtrWorcester Polytechnic [email protected]
MS231
Energy Aware Performance Metrics
Energy aware performance metrics are absolutely necessaryin order to better appreciate the performance of algorithmson modern architectures. Although recent advances, suchas the FLOPS/WATT metric, gave an important push to
CS13 Abstracts 307
the right direction, we will show that deeper investigationsare needed, if we are to overcome the power barrier forreaching Exaflop. We will showcase tools that allow accu-rate, on chip power measurements that shed new light onthe energy requirements of important kernels.
Costas Bekas, Alessandro CurioniIBM Research - [email protected], [email protected]
MS231
Application-aware Energy Efficient High Perfor-mance Computing
The energy cost of running an HPC system can exceed thecost of the original hardware purchase. This has driven thecommunity to attempt to understand and minimize energycosts wherever possible. We present an automated frame-work, Green Queue, for customized application-aware Dy-namic Voltage-Frequency Scaling (DVFS) settings to re-duce the energy consumption of large scale scientific ap-plications. Green Queue supports making CPU clock fre-quency changes in response to intra-node and internode ob-servations about application behavior. Our intra-node ap-proach reduces CPU clock frequencies and therefore powerconsumption while CPUs lacks computational work due toinefficient data movement. Our inter-node approach re-duces clock frequencies for MPI ranks that lack computa-tional work. We investigated these techniques on a set oflarge scientific applications on 1024 cores of Gordon, anIntel Sandybridge based supercomputer at the San DiegoSupercomputer Center. Our optimal intra-node techniqueshowed an average measured energy savings of 10.6% anda maximum of 21.0% over regular application runs. Ouroptimal inter-node technique showed an average 17.4% anda maximum of 31.7% energy savings.
Laura CarringtonSan Diego Supercomputing [email protected]
MS231
A ’Roofline’ Model of Energy and What it Impliesfor Algorithm Design
Abstract not available at time of publication.
Richard VuducGeorgia Institute of [email protected]. edu
MS231
Power Bounds and Large Scale Computing
Energy and power are widely recognized as significant chal-lenges for future large scale systems. Current processors, inparticular the Intel Sandy Bridge family, already includemechanisms to limit power levels dynamically. However,these mechanisms apply without consideration of the powerlevels of other nodes, which may be lower and allow a ”hot”node to consume more power. This talk will discuss optionsand techniques for limiting power and energy consumptionthat better suit large-scale systems. It will also detail ini-tial experiences with Intel’s Running Average Power Limit(RAPL) on a large Linux system and discuss possible ex-tensions.
Bronis R. de SupinskiLivermore Computing
Lawrence Livermore National [email protected]
MS232
Convex Collective Matrix Factorization
In many realistic applications, multiple interlinked sourcesof data are available and they cannot be easily representedin the form of a single matrix. Collective matrix factoriza-tion has recently been introduced to improve generaliza-tion performances by jointly factorizing multiple relationsor matrices. In this paper, we extend the trace norm formatrix factorization to the collective matrix factorizationcase. This norm defined on the space of relations is usedto regularize the empirical loss, leading to a convex for-mulation of the problem. Similarly to the trace norm onmatrices, we show that the collective-matrix completionproblem admits a fast iterative singular-value thresholdingalgorithm. The collective trace norm is also characterizedas a decomposition norm, useful to find an optimal solutionthanks to an unconstrained minimization procedure. Em-pirically we show that stochastic gradient descent suits wellfor solving the convex collective factorization even for largescale problems. We also show that the proposed algorithmdirectly solving the convex problem is much faster than un-constrained gradient minimization optimizing in the spaceof low-rank matrices.
Guillaume BouchardXerox Research Centre [email protected]
MS232
Generalised Coupled Tensor Factorisation andGraphical Models
We discuss coupled tensor factorisation from a statisti-cal perspective, building on generalised linear models andprobabilistic graphical models. We express a model us-ing a factor graph where the factorisation is achieved viaa message passing algorithm. This provides a practicalapproach that enables development of application specificcustom models, Bayesian model selection as well as algo-rithms for joint factorisations where several tensors are fac-torised simultaneously. We illustrate the approach on sig-nal processing applications.
A. Taylan CemgilBogazici [email protected]
MS232
Linked Multilinear Component Analysis, MultiwayCanonical Correlation Analysis and Partial LeastSquares
We will present models and algorithms for multiwaycomponent analysis, tensor canonical correlation analy-sis (MCCA) and higher-order partial least squares (HO-PLS). We also discuss emerging models and approachesfor multi-block constrained matrix/tensor decompositionsin applications to group- and linked-multiway compo-nent analysis, feature extraction, classification and clus-tering. We also overview our recent related results:http://www.bsp.brain.riken.jp/ cia/recent.html#tensor
Andrzej CichockiRIKEN Brain Science [email protected]
308 CS13 Abstracts
MS232
Exact Line and Plane Search for Tensor Optimiza-tion by Global Minimization of Bivariate Polyno-mials and Rational Functions
Line search (LS) and plane search (PS) are an integralcomponent of many optimization algorithms. We pose in-dependent component analysis (ICA) as a data fusion prob-lem in which a PS subproblem naturally arises. In tensoroptimization problems LS and PS often amount to mini-mizing a polynomial. We introduce a scaled LS and PS andshow they are equivalent to minimizing a rational function.Lastly, we show how to compute the global minimizer ofreal and complex (scaled) LS and PS problems accuratelyand efficiently by means of a generalized eigenvalue decom-position.
Laurent SorberKU [email protected]
Ignat DomanovKU Leuven [email protected]
Marc Van BarelKatholieke Universiteit LeuvenDepartment of Computer [email protected]
Lieven De LathauwerKU Leuven [email protected]
MS233
NWChem Quantum Many-body Methods on theIntel MIC Architecture
The NWChem quantum many-body methods found in theTensor Contraction Engine module are widely used for sci-entific applications but are both computation and mem-ory intensive, hence require large parallel computers. Wepresent our initial findings concerning the porting of thesemethods to the Intel MIC architecture using multiple pro-gramming models. We apply inspector-executor schemesto eliminate the overhead of dynamic load-balancing andallow for greater data reuse, which reduces both intranodeand internode communication.
Jeff R. HammondArgonne National LaboratoryLeadership Computing [email protected]
David OzogUniversity of [email protected]
MS233
Optimizing Numerical Weather Prediction Perfor-mance and Scaling on the Intel MIC
Further increases in weather and climate simulation ca-pability will depend on exploiting all levels of applicationparallelism. What are the opportunities for exposing andexploiting parallelism in atmospheric models despite rela-tively large memory footprints and low computational in-tensities? What are the challenges and prospects for pro-
grammability? We present status and preliminary resultsof efforts to employ the Intel Many Integrated Core proces-sor in the Weather Research and Forecast (WRF) model.
John MichalakesNational Renewable Energy [email protected]
MS233
Parallelization Strategies for High-order Dis-cretized Hyperbolic PDEs
We target high-performance implementations for the solu-tion of hyperbolic PDEs on Stampede using a high-orderdiscontinuous Galerkin finite element discretization. Onekey challenge in achieving high throughput for these appli-cations is that these are heavily memory-bound; addition-ally the flop/mop ratio changes with the polynomial orderof the elemental basis. The other challenge arises from theasymmetry between the CPU and the co-processor makingload-balance difficult. We consider parallelization strate-gies addressing these challenges.
Hari SundarInstitute for Computational and Engineering SciencesUniversity of Texas at [email protected]
Jesse Kelly, Omar GhattasUniversity of Texas at [email protected], [email protected]
MS233
Evaluating Intel’s Many Integrated Core Architec-ture for Climate Science
Abstract not available at time of publication.
Henry TufoNCARUniversity of Colorado at [email protected]
MS234
Solving Sequences of Linear Systems for StochasticCollocation based Uncertainty Quantification
Using stochastic collocation for uncertainty quantificationin partial differential equations (PDEs) with random datarequires solving sequences of linear systems. We use Krylovsubspace recycling to obtain efficient solution of such se-quences. We propose a new recycle subspace selection cri-terion, describe a new recycling algorithm, and adapt theexisting recycling solvers. The underlying PDEs includean elliptic diffusion equation, Maxwell’s equations, and aheat equation. The results show speed ups of up-to 55% intime.
Kapil AhujaMax Planck Institutefor Dynamics of Complex Technical [email protected]
Michael ParksSandia National [email protected]
Eric De Sturler
CS13 Abstracts 309
Virginia [email protected]
Peter BennerMax-Planck-Institute forDynamics of Complex Technical [email protected]
MS234
On using AMG Preconditioners to Accelerate theXFEM-Monte Carlo Approach for UncertaintyQuantification in Homogenization
An AMG approach is proposed for solution of linear sys-tems arising from unit cell problems within the XFEM-Monte Carlo framework to quantify uncertainties in ho-mogenization. The XFEM linear systems can be parti-tioned to contain an invariant sub-matrix corresponding tothe standard DOFs along with a sub-matrix correspondingto the enrichment DOFs that change for each Monte Carlorealization. This property is harnessed to optimally reuseparts of the preconditioner thereby reducing the AMGsetup costs.
Badri HiriyurWeidlinger Associates Inc., [email protected]
MS234
An Algebraic Multigrid Solver for Elliptic PDEswith Random Coefficients
In this talk we present a study of performance for alge-braic multigrid (AMG) method on elliptic partial differ-ential equations (PDEs) with random coefficients arisingfrom discretization of groundwater flow problem. We willfurther discuss the effects of use of fixed coarse grids andcomputational cost savings by this approach.
Minho ParkUniversity of [email protected]
MS234
Preconditioning Stochastic Collocation Saddle-Point Systems
We discuss linear algebra issues and preconditioning forsaddle point systems arising from stochastic collocationdiscretisations of PDEs with random data. Model prob-lems include: a mixed formulation of a second-order ellipticproblem (with uncertain diffusion coefficient), the steady-state Navier-Stokes equations (with uncertain Reynoldsnumber) and a second-order elliptic PDE on a random do-main, solved via the fictitious domain method.
Catherine PowellSchool of MathematicsUniversity of [email protected]
MS235
Implicit Timesteping of the Community Atmo-sphere Model Spectral Element Code Utilize GPUAccelerators
With the commissioning of hybrid multicore HPC sys-tems, such as the Cray XK6 Titan supercomputer at the
Oak Ridge Leadership Computing Facility (OLCF), we arebridging the way to the exascale era of high performancecomputing which will extend the resolution of regional cli-mate predictions with realistic clouds and chemistry. Inorder to realize the full potential of Titan and similarhardware, we are enhancing the Community Earth SystemModel (CESM) to take advantage of the GPU accelera-tors. Here, we discuss the adaptation of the spectral ele-ment dycore in the Community Atmospheric Model(CAM)to run on hybrid architecture and the step towards implicittimesteping.
Rick ArchibaldComputational Mathematics GroupOak Ridge National [email protected]
MS235
Development of an IMEX Integration Method forSea Ice Dynamics
Most sea ice model numerical schemes involve a splittingin time between the momentum and the continuity equa-tions. It has been shown that this approach can lead toa numerical instability or failure of the momentum equa-tion implicit solver for large advective time steps. To curethis problem, an IMplicit EXplicit (IMEX) time integra-tion technique is being developed. Results show that theIMEX technique improves the accuracy of the solution andreduces the number of failures of a JFNK solver. These lessfrequent failures also lead to a decrease in the CPU timerequired.
Jean-Francois LemieuxEnvironment [email protected]
MS235
Fast Offline Ocean Tracer Transport Model for theCommunity Earth System Model
Abstract not available at time of publication.
Francois PrimeauEarth System Science Dept.University of California, [email protected]
MS235
Software and Algorithms for Implementing Scal-able Solvers in Climate Codes
We have had success in using the Trilinos software from theFASTMath SciDAC institute to implement implicit andsteady-state solvers in climate codes. Since each applica-tion has different requirements and capabilities, having atoolbox of available software and algorithms has led to flex-ibility and extensible implementations that can be tailoredto the application. We will show results from Ice Sheetsimulations (CISM) and progress in atmosphere, ocean, seaice, and tracer transport applications.
Andrew SalingerApplied Computational Methods Dept, Sandia NationalLabsSandia National [email protected]
Katherine J. Evans
310 CS13 Abstracts
Oak Ridge National [email protected]
MS236
An Efficient Resampling Algorithm for Estimationof Fisher Information Matrix using Prior Informa-tion
The Fisher information matrix (FIM) is a critical quantityin several aspects of mathematical modeling, e.g., confi-dence interval calculation, optimal input selection in ex-perimental design. Analytical determination of the FIMin a general setting may be difficult or almost impossi-ble due to intractable modeling requirements or/and in-tractable high-dimensional integration. We will present anefficient resampling algorithm for estimation of the FIM,particularly, for the cases when some elements of the FIMare analytically known a priori, and the rest of the elementsare unknown. Such an interesting structure of the FIM isreported to be observed in the context of many engineeringapplications.
Sonjoy DasUniversity at [email protected]
James C. SpallJohns Hopkins [email protected]
Roger GhanemUniversity of Southern CaliforniaAerospace and Mechanical Engineering and [email protected]
MS236
Design or Experiments for Multi-phyiscs Problems
We demonstrate the use of information theory on inver-sion for multi physics problems. In particular we evaluatethe use of different measurements on a multi-physics pro-totype that emulates glacier flow, atmospheric transportand heat. Information metrics are explored to help man-age different measurements. We have developed a flexibleand extensible computational framework based on variouscomponents from Trilinos. This enables large scale opti-mization through adjoints, which is used in reduced andfull space optimization algorithms.
Andrew [email protected]
Bart VanbloemenwaandersSandia National LaboratoriesOptimization and Uncertainty Quantification [email protected]
MS236
Data-driven Model Mis-specification Mitigation
In the context of inverse problems, often the ability to sim-ulate data is limited to the extent that only a discrete, in-complete observation operator can be specified. Such mis-specified observation operator may describe approximatedphysics, approximated numerics, linearization of non-linearprocesses, etc. In this study, we developed a nuclear norm
stochastic optimization technique to supplement the modelwith rank restriction given a training set of models anddata.
Lior HoreshBusiness Analytics and Mathematical SciencesIBM TJ Watson Research [email protected]
Ning HaoTufts UniversityDepartment of [email protected]
Misha E. KilmerTufts [email protected]
MS236
Coherence Metric for Optimal Compressive Sens-ing
The efficiency of compressive sensing depends on randomprojections of the sampled signal. However, in practicalencoders/decoders, one uses deterministic matrices, whichhave non-zero coherence between sampling matrices andthe bases used to represent the signal sparsely. This resultsin sub-optimal sampling. We investigate how the degree ofcoherence correlates with the accuracy of reconstruction(i.e., completeness of the estimated sparse respresentation)as well as the uncertainty in the estimated signal.
Sean A. MckennaSandia National [email protected]
Jaideep RaySandia National Laboratories, Livermore, [email protected]
MS237
Coarse-grained Modeling of Supported and Teth-ered Bilayers
Biomembranes compartmentalize cells and are fundamen-tal for any living organism. The fundamental buildingblock for biomembranes are lipid bilayers. Here we presenta multiscale molecular modeling investigation of biomem-branes. Especially the interplay between membranes andsupports will be discussed. Supported Lipid Bilayers arean abundant research platform for understanding the be-havior of cell membranes as they allow for additional me-chanical stability and enable characterization techniquesnot reachable otherwise. However, in computer simulationsthese systems have been studied only rarely up to now. Wepresent systematic studies on different length scales of thechanges that a support and tethering to it inflicts on aphospholipid bilayer using molecular modeling. We char-acterize the density and pressure profiles as well as thedensity imbalance induced by the support. We determinethe diffusion coefficients and characterize the influence ofcorrugation of the support. We also measure the free en-ergy of transfer of phospholipids between leaflets using thecoarse-grained Martini model.
Roland FallerDepartment of Chemical Engineering and MaterialsScienceUniversity of California, Davis
CS13 Abstracts 311
MS237
Stochastic Reaction-diffusion Simulation on Multi-ple Scales
Abstract not available at time of publication.
Mark FleggMathematical InstituteUniversity of [email protected]
MS237
Matrix Calculations in Diffusion Approximationsfor Molecular Motors
A discussion of matrix methods to bridge nano-scale molec-ular motor model, which includes both diffusive and ki-netic components, to a meso-scale stepping model. Such amethod serves as an alternative to monte carlo methods,facilitating rapid sensitivity analysis over a large parameterspace.
John FricksDept of StatisticsPennsylvania State [email protected]
MS237
Fluctuating Lipid Bilayer Membranes with Diffus-ing Protein Inclusions: Hybrid Continuum-ParticleNumerical Methods
Many proteins through their geometry and specific interac-tions with lipids induce changes in local membrane materialproperties. To study such phenomena we introduce a newhybrid continuum-particle description for the membrane-protein system that incorporates protein interactions, hy-drodynamic coupling, and thermal fluctuations. We inves-tigate how collective protein effects influence membranemechanical properties. We discuss interesting numericalaspects that are required to obtain good translation in-variance. Finally, we discuss a coarse grained model thatincorporates important hydrodynamics.
Jon Karl SigurdssonDepartment of MathematicsUniversity of California, Santa [email protected]
Paul J. AtzbergerUniversity of California-Santa [email protected]
MS238
Multi-level Communication Avoiding LU and QRFactorizations for Hierarchical Platforms
This study focuses on LU and QR factorizations of densematrices on multi-level hierarchical platforms. We first in-troduce a new platform model called HCP. The focus is seton reducing the communication requirements of the stud-ied algorithms at each level of the hierarchy. We extendlower bounds on communications to the HCP model andintroduce two multi-level LU and QR algorithms tailoredfor those platforms. Finally, we provide a detailed perfor-
mance analysis.
Mathias JacquelinLRI - INRIA Saclay, [email protected]
Amal KhabouNRIA Saclay, 4 Rue Jacques Monod [email protected]
Laura [email protected]
MS238
A Communication Optimal N-Body Algorithm forLong-Range Direct Interactions
Algorithms for long-range, direct particle interactions tra-ditionally work by dividing the particles among processorsand synchronously exchanging particles to compute inter-actions. This approach works well when computation dom-inates communication, but it breaks down at scale whencommunication dominates. We present a communication-optimal algorithm that uses extra memory to replicatethe particles. In practice, the algorithm yields modestspeedups on communication-bound problem instances andprovides significantly better strong scaling to tens of thou-sands of processors.
Michael DriscollUC [email protected]
Evangelos GeorganasEECSUC [email protected]
Penporn KoanantakoolUC [email protected]
Edgar SolomonikUniversity of California at [email protected]
Katherine YelickUC BerkeleyLawrence Berkeley National [email protected]
MS238
Algorithmic Adventures in 3D
We describe an extension of the Scalable Universal Ma-trix Multiplication Algorithms (SUMMA) from 2D to 3Dprocess grids; the underlying idea is to lower the com-munication volume through storing redundant copies ofone or more matrices. This talk focuses on element-wisematrix distributions, which lead to allgather-based algo-rithms. We begin by describing an allgather-based 2DSUMMA, describe its generalization to 3D process grids,and then discuss theoretical and experimental performancebenefits of the new algorithms.
Martin D. Schatz
312 CS13 Abstracts
Department of Computer SciencesThe University of Texas at [email protected]
MS238
Shape-morphing in LU Factorizations
We present a sequential communication-avoiding LU fac-torization algorithm. By communication avoiding, wemean that the volume of data transferred in cache missesor page faults (or explicit input-output operations) is closeto optimal, and that the size of each data transfer can beproportional to the size of the fast memory (cache). Thekey idea in the algorithm is to morph the data layout ofthe reduced matrix and the factors repeatedly throughoutthe algorithm. The goal is to use a column-major whencolumns are eliminated, but to use block layout during ma-trix multiplication and triangular solves. Our key findingis that the shape-morphing steps do not contribute signifi-cantly to the asymptotic communication complexity of thecomputation.
Grey BallardUC [email protected]
James W. DemmelUniversity of CaliforniaDivision of Computer [email protected]
Benjamin LipshitzUC [email protected]
Sivan A. ToledoTel Aviv [email protected]
Oded SchwartzUC [email protected]
MS239
Randomized Asynchronous Iterative LinearSolvers
Asynchronous methods for solving linear equations havebeen researched since Chazan and Miranker published theirpioneering paper on chaotic relaxation in 1969. While thesemethods were shown to converge, no bounds were provenon the rate of convergence, and in practice their conver-gence was sometimes shown to be very slow. In this talkwe will present new asynchronous iterative linear solvers.A key component in the new algorithms is the use of ran-domization. We show that unlike previous algorithms, therate of convergence for new algorithms can be bounded (inexpectancy), and that they can be competitive in practice.
Haim AvronBusiness Analytics & Mathematical SciencesIBM T.J. Watson Research [email protected]
Alex DruinskyTel-Aviv [email protected]
Anshul Gupta, Prabhanjan KambadurIBM T J Watson Research [email protected], [email protected]
MS239
Investigating the Convergence of Asynchronous It-erative Methods
When solving large sparse systems of linear equations inparallel with synchronous iterative techniques, it is neces-sary to exchange newly-computed vector elements betweenprocessors at every iteration. This talk describes our cur-rent work on developing and analysing asynchronous it-erative algorithms that avoid time-consuming synchronouscommunication operations and thus allow execution to pro-ceed without having to wait for new data to arrive. Wepresent theoretical results supported by practical experi-mentation.
Iain BethuneThe University of [email protected]
J. Mark BullEdinburgh Parallel Computing [email protected]
Nicholas DingleSchool of Mathematics, University of [email protected]
Nicholas J. HighamUniversity of ManchesterSchool of [email protected]
James HookSchool of MathematicsUniversity of [email protected]
MS239
Asynchronous Multilevel Methods on AdaptivelyRefined Grids
Emerging architectures and the drive towards exascale ma-chines has seen a renewed interest in asynchronous solutionmethods for PDE systems. In this talk we will consider theAFACx method which is an asynchronous version of theFast Adaptive Composite Grid (FAC) method. AFACx isa multilevel solution method most often used on structuredadaptively refined grids for the solution of elliptic problems.AFACx has been demonstrated for both scalar and systemelliptic PDEs. We will describe recent numerical work andthe performance of AFACx as a solver and as a precondi-tioner. Progress permitting we will also describe the use ofchaotic smoothers within AFACx for both multicore andGPU architectures.
Bobby Philip, Zhen WangOak Ridge National [email protected], [email protected]
Manuel Rodriguez Rodriguez, Mark [email protected], [email protected]
CS13 Abstracts 313
MS239
Asynchronous Stochastic Optimization Methods
Stochastic gradient descent (SGD) has become a popu-lar method for many machine learning tasks, particularwhen dealing with massive data sets. Some attempts toparallelize SGD require performance-destroying memorylocking and synchronization. We describe an approachin which a centrally stored decision variable is updatedasynchronously by the different cores. We present con-vergence theory and computational results. We also dis-cuss Lagrangian-based extensions that potentially allow ahigher degree of parallelism.
Christopher Re, Ben RechtComputer Sciences DepartmentUniversity of [email protected], [email protected]
Stephen WrightUniversity of WisconsinDept. of Computer [email protected]
MS240
Solution of Inverse Problems with Limited ForwardSolver Evaluations: a Fully Bayesian Framework
Solving inverse problems based on computationally de-manding forward solvers is ubiquitously difficult since oneis necessarily limited to just a few observations of the re-sponse surface. This limited information induces addi-tional uncertainties on the predicted design distributions.In this work we quantify this epistemic uncertainty by em-ploying our recently developed fully Bayesian surrogatesbased on Gaussian processes.
Ilias BilionisCenter for Applied MathematicsCornell University, [email protected]
Nicholas ZabarasCornell [email protected]
MS240
An Analysis of Infinite-dimensional Bayesian In-verse Shape Acoustic Scattering and its NumericalApproximation
We present and analyze an infinite dimensional Bayesianinference formulation, and its numerical approximation, forthe inverse problem of inferring the shape of an obstaclefrom scattered acoustic waves. Given a Gaussian priormeasure on the shape space, whose covariance operatoris the inverse of the Laplacian, the Bayesian solution ofthe inverse problem is the posterior measure given by theRadon-Nikodym derivative with respect to the Gaussianprior measure. The well-posedness of the Bayesian formu-lation in infinite dimensions is proved, including the justi-fication of the Radon-Nikodym derivative and the continu-ous dependence of the posterior measure on the observationdata via the Hellinger distance. Next, a finite dimensionalapproximation to the Bayesian posterior is proposed andthe corresponding approximation error is quantified. Theapproximation strategy involves a Nystrom scheme for ap-proximating a boundary integral formulation of the forwardHelmholtz problem and a Karhunen-Loeve truncation for
approximating the prior measure. The main result of thiswork is that the convergence rate for approximating theBayesian inverse problem is spectral, and this directly in-herits the spectral convergence rates of the approximationsof both the prior and the forward problem.
Tan Bui-ThanhThe University of Texas at [email protected]
Omar GhattasUniversity of Texas at [email protected]
MS240
Evaluation of GaussianApproximations to Bayesian Inverse Problems inSubsurface Models
We discuss numerical evaluations of Gaussian approxima-tions to the posterior distribution that arises in Bayesiandata assimilation for reservoir characterization. In partic-ular, we evaluate (i) maximum a posterior estimate, (ii)randomized maximum likelihood, (iii) EnKF (and vari-ants). For our evaluation, we use a state-of-the art MCMCto accurately characterize the posterior distribution of thelog-permeability conditioned to dynamic data. Therefore,MCMC provides a gold standard against which to comparethe performance of the Gaussian approximations.
Marco IglesiasCivil and Environmental [email protected]
Kody LawMathematics InstituteUniversity of [email protected]
Andrew StuartMathematics Institute,University of [email protected]
MS240
An Adaptive Sparse-Grid High-Order StochasticCollocation Method for Bayesian Inference withComputationally Expensive Simulations
We develop an adaptive sparse-grid high-order stochasticcollocation (aSG-hSC) method to quantify parametric andpredictive uncertainty of computationally expensive physi-cal systems with Bayesian approach. In our method, a sur-rogate system for the logarithm of the posterior probabil-ity density function (PPDF) is constructed using the localadaptive sparse-grid approximation. Moreover, we incor-porate the high-order local hierarchical polynomial basis(e.g. quadratic, cubic basis) to further improve the accu-racy of the surrogate system and reduce the computationalexpense.
Guannan ZhangOak Ridge National [email protected]
Max GunzburgerFlorida State [email protected]
314 CS13 Abstracts
Clayton G. WebsterOak Ridge National [email protected]
MS241
Methods for Long-time Lagrangian Transport onthe Sphere
We solve the advection equation in its Lagrangian formusing a collection of disjoint particles and panels coveringthe sphere, for applications of climate modeling. For shorttime scales, the advection equation is satisfied exactly byeach Lagrangian computational element. For longer simu-lations, we encounter the problem of mesh distortion. Weintroduce a remeshing scheme that uses adaptive refine-ment and interpolation of the Lagrangian parameter topreserve the range of the advected tracer without intro-ducing any numerical oscillations.
Peter A. BoslerUniversity of MichiganApplied and Interdisciplinary [email protected]
Christiane JablonowskiUniversity of MichiganAnn Arbor MI [email protected]
Robert KrasnyUniversity of MichiganDepartment of [email protected]
MS241
A Conservative Semi-Lagrangian TransportScheme on Spectral Element Cubed-sphere Grids
A conservative semi-Lagrangian scheme for the spectral-element (SE) spherical grids (SPELT) has been developed.Each SE is overlaid by finite-volume (FV) cells, where webuild a biquadratic mulit-moment reconstruction proce-dure. Conservation is guaranteed by a flux-based charac-teristic semi-Lagrangian approach. However, the velocityis specified on the host (SE) grid system. SPELT allows usto adopt arbitrary unstructured SE discretization and cantherefore be used for future (unstructured) climate simu-lations (CAM-SE). The new scheme is tested on cubed-sphere geometry using several challenging tests.
Christoph ErathNCARUniversity of Colorado, [email protected]
Ramachandran NairNational Center for Atmospheric [email protected]
Henry TufoNCARUniversity of Colorado at [email protected]
MS241
Advection Scheme in CAM-SE
CAM-SE is the Community Atmosphere Model (CAM)
configured with the spectral element dynamical core, whichuses conforming quadrilateral grids. The spectral elementmethod requires stabilization. If grids are quasi-uniform,adding a constant amount of hyper-viscosity is a very effec-tive form of stabilization. Here we focus on a tensor-basedhyper-viscosity operator for grids with high variability ofscales.
Oksana GubaSandia National LaboratoriesNM, [email protected]
Michael [email protected]
James OverfeltSandia National [email protected]
Mark A. TaylorSandia National Laboratories, Albuquerque, [email protected]
MS241
Optimization-based Remap and Transport: A Di-vide and Conquer Strategy for Feature-preservingDiscretizations
We present an optimization framework for the preserva-tion of physical features in PDE discretizations and re-late the recently introduced optimization-based remap offlux and mass to this framework. Remap is cast asa quadratic program whose optimal solution minimizesthe distance to a suitable target quantity, responsible forthe method’s accuracy, subject to a system of inequalityconstraints, which separately maintain physical features.We use optimization-based remap to demonstrate shape-adaptable, conservative and bound-preserving transport al-gorithms.
Denis RidzalSandia National [email protected]
Pavel BochevSandia National LaboratoriesComputational Math and [email protected]
Kara PetersonSandia Natl. [email protected]
MS242
Fast Multigrid Solvers Long Range Potentials
This talk will present multigrid solvers for computing elec-trostatic potentials. These parallel solvers can be usedas building blocks in coupled multiphysics applications,such as particle transport in electro-osmotic or electro-rheological flows.
Ulrich J. RuedeUniversity of Erlangen-NurembergDepartment of Computer Science (Simulation)[email protected]
CS13 Abstracts 315
Daniel RitterUniversity of [email protected]
Dominik BartuschatLehrstuhl f. SimulationUni [email protected]
MS242
A Finite Element Method for the Total VariationFlow Without Regularization
The TV flow, that is the subgradient flow of the en-ergy generated by the BV-norm, and related equations arecalled very singular diffusion equations, since in flat re-gions (|∇u| = 0) the diffusion is so strong that becomesa nonlocal effect. We propose a method for the solutionof this class of equations, which involves no regularizationand is unconditionally stable and convergent. To deal withthe fact that the underlying nonlinear problems are solvedonly approximately, we devise an a posteriori error estima-tor. Applications to materials science are currently underinvestigation.
Abner J. SalgadoDepartment of MathematicsUniversity of [email protected]
MS242
Conforming vs. Nonconforming Finite ElementMethods for Fluid and Solid Mechanics
The use of nonconforming elements to solve fluid and fluidmechanical problems has shown several advantages overthe use of conforming counterparts. Replacing conformingelements by nonconforming ones is simple and guaranteesnumerical stability in many cases. We give a brief reviewon the recent development in quadrilateral nonconformingfinite elements. We will then discuss several new interest-ing observations about these quadrilateral nonconformingelements. Error estimates and numerical results will bepresented.
Dongwoo SheenSeoul National UniversityDepartment of [email protected]
MS242
Superconvergence of Polynomial Spectral Colloca-tion Methods
Abstract not available at time of publication.
Zhimin ZhangWayne State UniversityDepartment of [email protected]
MS243
Hierarchical Matrix Preconditioners and thePreservation of Vectors
Approximate H-matrix preconditioners provide a robustand efficient way to solve large systems of linear equationsresulting from the FE discretization of elliptic pdes. Due
to ill-conditioning, the accuracy of the H-matrix approxi-mation has to be raised with n, which spoils the efficiency.We present a new approach that is based on the preserva-tion of certain vectors during the H-matrix approximation.The new technique significantly improves the quality of thepreconditioner.
Mario BebendorfUniversity of [email protected]
MS243
Block Filtering Factorizations
In this talk we describe a family of preconditioners thatare suitable for matrices arising from the discretization ofa system of PDEs on unstructured grids. To address thescalability problem of many existing preconditioners, thesepreconditioners preserve directions of interest from the in-put matrix and are able for example to filter low frequencymodes and alleviate their effect on convergence. The inputmatrix can have an arbitrary sparse structure, and can bereordered using nested dissection to allow a parallel im-plementation. We present a set of numerical experimentson matrices with anisotropies and jumps in the coefficientsthat show the efficiency of our preconditioners.
Laura [email protected]
Frederic NatafLaboratoire J.L. [email protected]
Riadh FezzaniINRIA, [email protected]
MS243
Elliptic Preconditioner for Accelerating the SelfConsistent Field Iteration of Kohn-Sham DensityFunctional Theory
Kohn-Sham density functional theory (KSDFT) is themost widely used electronic structure theory for condensedmatter systems. We present a new preconditioner for accel-erating the self consistent field (SCF) iteration for solvingthe Kohn-Sham equations. The new preconditioner, whichis called the elliptic preconditioner, is constructed by solv-ing an elliptic partial differential equation. The ellipticpreconditioner is shown to be effective for large inhomoge-neous systems at low temperature.
Lin LinLawrence Berkeley National [email protected]
Chao YangLawrence Berkeley National [email protected]
MS243
Multilevel Low-rank Approximation Precondition-ers
A new class of methods based on low-rank approxima-
316 CS13 Abstracts
tions which has some appealing features will be introduced.The methods handle indefiniteness quite well and are moreamenable to SIMD compuations, which makes them attrac-tive for GPUs. The method is easily defined for SymmetricPositive Definite model problems arising from Finite Differ-ence discretizations of PDEs. We will show how to extendto general situations using domain decomposition concepts.
Yousef SaadDepartment of Computer ScienceUniversity of [email protected]
Ruipeng LiDepartment of Computer Science & EngineeringUniversity of [email protected]
MS244
Application-specific Reduced Order Quadraturesfor Parameterized Problems
We present an algorithm to generate an application-specificreduced order quadrature (ROQ) for products of param-eterized functions. Such integrals need to be computedmany times, for example, while doing matched filtering tosearch for gravitational waves (GWs). If a reduced ba-sis (RB) or any other projection based model reductiontechnique is applied, the dimensionality of integrands isreduced dramatically, however the cost of evaluating thereduced integrals still scales as the size of the original prob-lem. Using discrete empirical interpolation (DEIM) pointsas ROQ nodes leads to a computational cost that only de-pends linearly on the number of reduced basis. Generationof a RB via a greedy procedure requires defining a trainingset, which for the product of functions can be very large.We notice that this direct approach can be impractical inmany applications, and propose instead a two-step greedytargeted towards approximation of such products. The ac-curacy and efficiency of the two-step greedy and ROQ areverified by the error estimates and operation counts pre-sented. In addition, we find that for the particular appli-cation in gravitational wave physics the two-step greedyspeeds up the computations in the offline stage by twoorder of magnitude and two order of magnitude savingsare observed in actual evaluations of integrals when ROQsare used as a downsampling strategy for equidistant sam-ples assimilated from data. While the primary focus is onquadrature rules for products of parameterized functions,our method is easily adapted for integrations of single pa-rameterized functions and some examples are considered.Joint work with: R. H. Nochetto, S. Field, M. Tiglio, F.Herrmann.
Harbir AntilThe University of [email protected]
MS244
Model Reduction for Fluid Flows Based on Non-linear Balanced Truncation
For a vast variety of fluid flows the dynamics are governedby the Navier-Stokes equations which are highly nonlin-ear. For this class of systems linear model reduction tech-niques fail. In this paper after showing existence of solu-tions to the required equations, a computational algorithmfor nonlinear balanced truncation is proposed for Galerkin
models corresponding to the Navier-Stokes equations. TheGalerkin models present a quadratic type nonlinearity thatcan be exploited in solving the Hamilton-Jacobi equations.The method is based on Taylor series expansion in whichthe computation of successively higher order terms reducesto solving consecutively higher order systems of linear al-gebraic equations.
Seddik DjouadiDepartment of Electrical Engineering and ComputerScienceUniversity of Tenneessee, [email protected]
Samir Sahyoun, Jin DongUniversity of [email protected], [email protected]
MS244
Combined Reduced Basis Approximations and Sta-tistical Approaches for Datat Assimulation
Reduced basis approximations allow to simulate the solu-tion to systems modeled by parameter dependent PDE’s.These methods allow to provide certified rapid on-line ap-proximation of the solutions. We explain in this presenta-tion how these complexity reduction methods can be usedto improve data assimilation by adding a large amount ofknowledge based of the model, constructed thanks to priorunderstanding by the specialists that have built the mod-els.
Yvon MadayUniversite Pierre et Marie Curieand Brown [email protected]
MS244
Certified Reduced Order Methods for OptimalFlow Control Problems
We propose a reduced basis framework for the numeri-cal solution of optimal control problems for parametrizedviscous incompressible flows. We mainly focus on con-trol problems for the (Navier-) Stokes equations involv-ing infinite-dimensional control functions, thus requiringa suitable reduction of the whole optimization problem,rather than of the sole state equation. We provide rigor-ous and efficiently evaluable a posteriori error estimates,as well as preliminary examples of application arising fromhaemodynamics.
Gianluigi RozzaSISSA, International School for Advanced StudiesTrieste, [email protected]
Federico NegriMathematics Institute of Computational Science andEng, [email protected]
Andrea ManzoniInternational School for Advanced Studies, Trieste, [email protected]
MS245
Monte-Carlo Simulation of Diffusion in Fractal Do-
CS13 Abstracts 317
mains
Diffusion-weighted MRI has became an important sourceof information about the dynamics in and the structure ofnatural or artificial materials (e.g., rocks, cements, humanorgans). In spite of intensive research, the relation betweenthe microstructure and the signal formed by diffusing nucleiremains poorly understood, mainly due to lack of efficientalgorithms and models for multi-scale porous media. Toovercome this limitation, we developed a fast random walk(FRW) algorithm with gradient encoding which exploitsthe multi-scale character of the medium. In this talk, wepresent an application of this algorithm to a Menger spongewhich is formed by multiple channels of broadly distributedsizes and often used as a model for soils and materials.Using this model, we investigate the role of multiple scalesonto diffusion-weighted signals.
Denis GrebenkovEcole Polytechnique, FranceLaboratoire de Physique de la Matiere [email protected]
Hang Tuan NguyenCEA Neurospin, [email protected]
MS245
Hermite Functions in Modeling Diffusion MRIData: From Applications to Fundamentals
Properties of Hermite functions make them ideally-suitedto problems of modeling diffusion-weighted (DW) magneticresonance (MR) signals and estimating propagators. We il-lustrate the utility of Hermite functions to characterize theanisotropy and diffusion-time dependence of the diffusionprocess. The ability of the model to represent different sig-nal profiles suggests that the approach could be used evenwhen no information regarding the underlying microstruc-ture is availablea desirable characteristic for imaging ap-plications.
Evren OzarslanDepartment of Radiology, Brigham and Women’s HospitalHarvard Medical [email protected]
Cheng Guan Koay, Peter J. [email protected], [email protected]
MS245
Diffusion Dynamics in Porous Media
Diffusion is known to cause multiple relaxation peaks inNMR relaxation behavior. However, in real porous mate-rials, pore size distribution also results in multiple or broadrelaxation times. We have devised a two-dimensional NMRmethod in order to distinguish these two scenarios. Usinganalytical and numerical solutions of the diffusion dynam-ics and numerical simulation we have demonstrated thiseffect and found it consistent with experimental results.
Yi-Qiao SongMassachusetts General Hospital and Harvard Medicalschool, [email protected]
Giovanna Carneiro
Massachusetts General Hospital and Harvard [email protected]
Larry Schwartz, Michael [email protected],[email protected]
MS246
Evaluating Noise in Complex Networks
As in all computations involving real-world systems, theresults of network analysis are affected by experimental,subjection and computational choices. However, networkanalysis algorithms are primarily based on graph theoryand therefore assume the inputs to be exact. In this talk,we will demonstrate how user choices and computationallimitations can affect network analysis results and discusshow concepts from numerical analysis such as conditioningand stability can be extended to evaluate the effect of noisein this domain.
Sanjukta Bhowmick, Sriram SrinivasanDepartment of Computer ScienceUniversity of Nebraska, [email protected],[email protected]
Vladimir UfimtsevUniversity of Nebraska, [email protected]
MS246
On the Resilience of Graph Clusterings
Are clusters found by popular graph clustering algorithmssignificant? One way to answer this is by measuring clus-ter resilience as follows: repeatedly perturb the input graphby adding one edge, and for each new edge recluster thevertices and calculate the distance between the originaland modified clustering. We hypothesize that the distribu-tion of distances contains information about the stabilityof clusters in the original graph and present applicationsof this technique to synthetic and real-world graphs.
Evan Fields, Tzu-Yi ChenPomona [email protected], [email protected]
MS246
Quantification of Uncertainty in Network Sum-mary Statistics
It is common in the analysis of network data to presenta network graph and then cite various summary statis-tics thereof (e.g., degree distribution, clustering coefficient,conductance, etc.). However, most real-world networks de-rive from low-level measurements that are noisy. Surpris-ingly, there has been little effort to date aimed at quantify-ing the uncertainty induced in network summary statisticsby noise in the underlying measurements. I will discusswork being done in my group towards this goal.
Eric D. KolaczykBoston UniversityDept of Mathematics & [email protected]
318 CS13 Abstracts
MS246
Impact of Graph Perturbations on Structural andDynamical Properties
Much of the research on networks arising in online socialmedia and bioinformatics involves networks that are in-ferred by sampling or indirect measurements, making themnoisy and incomplete. We study the impact of perturba-tions on structural and dynamical properties of networks.Specifically, find that threshold models are sensitive to per-turbations, which can significantly alter the fixed pointproperties. Further, the effects depend on the nature ofthe perturbation.
Abhijin AdigaVirginia [email protected]
Henning MortveitDept. of Mathematics and Virginia BioinformaticsInstituteVirginia [email protected]
Chris KuhlmanVirginia [email protected]
Anil VullikantiDept. of Computer Science, and Virginia BioinformaticsInst.Virginia [email protected]
MS247
Numerical Stability of Vortex Soliton under Opti-cal Lattice and Harmonic Potential
With only parabolic potential, 2D vortex soliton collapses(for the focusing case) when its norm exceeds certainthreshold (which is much smaller than the norm that willcause its intrinsic instability). After including an opti-cal lattice, there exists a second stability region (up toT=2000) when the lattice strength is moderate to large.However, some of these new stable solutions collapse af-ter T¿5000. We investigate whether it is a property of thesolution or simply a numerical artifact.
Qian-Yong ChenDepartment of Mathematics & StatisticsU. of Massachusetts at [email protected]
MS247
Analysis of Formal Order of Accuracy of WENOFinite Difference Scheme
In the presence of critical points, the formal order of accu-racy of WENO schemes is reduced if the sensitivity param-eter is chosen to be a small constant value or as a functionof grid spacing. We proved a much weaker sufficient condi-tion for the WENO-Z schemes by analyzing the nonlinearweights using some interesting properties of the smooth-ness indicators and illustrated with one dimensional Eulerequations.
Wai-Sun DonSan Diego State [email protected]
MS247
Unconditionally Stable Numerical Scheme for TwoPhase Models in Karst Aquifers
We present numerical methods that are unconditionallystable for certain phase field models of two phase flows inKarst aquifers.
Xiaoming [email protected]
MS247
Adaptive Multigrid, Discontinuous Galerkin Meth-ods for Cahn-Hilliard Type Equations
I will define and analyze a mixed DG, convex splittingscheme for a modified Cahn-Hilliard equation. The equa-tion represents a diffuse interface model for phase separa-tion in diblock coploymer blends and permits rather exoticsolutions compared to the classical Cahn-Hilliard equation,including solutions like those from the phase field crystalmodel. The talk will cover theoretical energy stability andconvergence results and also the practical, efficient solu-tion of the algebraic equations via an adaptive nonlinearmultigrid method. This is joint work with A. Aristotelous(SAMSI) and O. Karakashian (UTK).
Steven M. WiseMathematics DepartmentUniversity of [email protected]
MS248
Splitting Schemes for IncompressibleFluid-structure Interaction with Unfitted InterfaceFormulations
The stability and accuracy of the numerical approxima-tions of incompressible fluid-structure interaction problemsis very sensitive to the way the interface coupling condi-tions (kinematic and kinetic continuity) are treated at thediscrete level. In this talk we will provide an overview ofthe existing loosely coupled (or explicit coupling) schemeswithin a fitted mesh framework and then discuss their ex-tension to the unfitted case.
Miguel A. [email protected]
MS248
Immersed Finite Element Methods for ParabolicEquations with Moving Interface
Three Crank-Nicolson-type immersed finite element (IFE)methods are presented for solving parabolic equationswhose diffusion coefficient is discontinuous across a time de-pendent interface. Instead of the body-fitting mesh neededby the traditional finite elements for solving interface prob-lems, these IFE methods can use a structured mesh becauseIFEs can handle interface jump conditions without requir-ing the mesh to be aligned with the interface. Several dis-advantages of the body-fitting mesh for time-dependent in-terface problems will be discussed. And then a fixed struc-tured mesh for IFEs will be utilized to resolve these prob-lems. Numerical examples are provided to demonstrate
CS13 Abstracts 319
features of the three IFE methods.
Xiaoming HeDepartment of Mathematics and StatisticsMissouri University of Science and [email protected]
Tao LinVirginia [email protected]
Yanping LinDepartment of Mathematical and Statistics Science,[email protected]
Xu ZhangVirginia [email protected]
MS248
A High Order Discontinuous Galerkin NitscheMethod for Elliptic Problems with FictitiousBoundary
We present a DG method, based on the method of Nitsche,for elliptic problems with an immersed boundary represen-tation on a structured grid. In such methods small ele-ments typically occur at the boundary, leading to break-down of the discrete coercivity as well as numerical insta-bilities. In this work we propose a method that avoids usingsmall elements on the boundary by associating them to aneighboring element with a sufficiently large intersectionwith the domain.
August JohanssonUniversity of California, BerkeleyDept. of [email protected]
MS248
Analysis and Implementation of a Nitsche-basedDomain-bridging Method for Fluid Problems
We propose a Nitsche-based domain-bridging formulationfor a general class of stabilized finite element methods forthe Stokes problem. To ensure the stability of the methodand to avoid ill-conditioned linear algebra systems, a ghost-penalty is added by extending the least-squares stabiliza-tion terms to the overlap region. We explain how gen-eral overlapping domain methods can be implemented ef-ficiently by employing sophisticated algorithms and datastructures from the field of computational geometry.
Andre MassingSimula Research [email protected]
Mats G. LarsonUmea [email protected]
Anders Logg, Marie E. RognesSimula Research [email protected], [email protected]
MS248
A Cut Finite Element Method for a Stokes Inter-face Problem
We present a finite element method for the Stokes equa-tions involving two immiscible incompressible fluids withdifferent viscosities and with surface tension. The inter-face separating the two fluids does not need to align withthe mesh. We propose a Nitsche formulation which allowsfor discontinuities along the interface with optimal a pri-ori error estimates. A stabilization procedure is includedwhich ensures that the method produces a well conditionedstiffness matrix.
Peter HansboJonkoping [email protected]
Mats G. LarsonDepartment of MathematicsUmea [email protected]
Sara ZahediInformation TechnologyUppsala [email protected]
MS249
Relaxation Models for Two-phase Flows
Dynamical two-phase flow models represent flows in var-ious states of disequilibrium, where the driving forces to-wards full equilibrium are represented as relaxation terms.There is a deep connection between the entropy productionof the relaxation process and the stability and wave speedsof the resulting equilibrium model. For various explicitmodels, we will discuss the issues of hyperbolicity and thesubcharactistic condition; imposed equilibrium conditionswill always reduce the mixture speed of sound.
Tore FlattenSINTEF Energy [email protected]
Halvor LundDepartment of Energy and Process EngineeringNorwegian University of Science and [email protected]
MS249
A Mixture-energy-consistent 6-equation Two-phase Numerical Model for Cavitating Flows
We model cavitating flows by a variant of the 6-equationsingle-velocity two-phase model with stiff pressure relax-ation of Saurel–Petitpas–Berry. Our novel formulation em-ploys phasic total energy equations instead of the internalenergy equations of the classical system. This allows us todesign a simple numerical model that ensures consistencywith mixture total energy conservation and agreement ofthe relaxed pressure with the mixture equation of state.Heat and mass transfer terms are also considered.
Marica PelantiENSTA [email protected]
Keh-Ming Shyue
320 CS13 Abstracts
National Taiwan [email protected]
MS249
Liquid-gas Mixtures and Diffuse Interfaces Com-putations at All Speeds
All speed flows and in particular the low Mach numberasymptotics algorithms are addressed for the numerical ap-proximation of a mechanical equilibrium multiphase flowmodel. During the computation, the interface is consid-ered as a numerically diffused zone, captured as well as allpresent waves (shocks, expansion waves). Many applica-tions with liquid-gas interfaces involve a very wide rangeof Mach number variations. Therefore, it is important toaddress numerical methods free of restrictions regardingthe Mach number.
Richard Saurel, Sebastien Le MartelotPolytech [email protected],[email protected]
Boniface NkongaINRIA PUMAS - Universite de [email protected]
MS249
Multi-model Simulation of Compressible Two-phase Flows
The simulation of water flows in Pressurized Water Reac-tor requires the use of different models according to thelocal characteristics of the flow. These models may be for-mally linked by asymptotic limits but the dedicated codesare developed independently. We present how to couplethese different models and codes and how to optimize thelocation of the coupling interfaces, in order to use as muchas possible the simplest models without deteriorating theaccuracy.
Nicolas SeguinLaboratoire Jacques-Louis LionsUniversity Paris 6, [email protected]
MS250
Popularity versus Similarity in Growing Networks
Popularity is attractive – this is the formula underly-ing preferential attachment, a popular explanation for theemergence of scaling in growing networks. If new connec-tions are made preferentially to more popular nodes, thenthe resulting distribution of the number of connections thatnodes have follows power laws observed in many real net-works. Preferential attachment has been directly validatedfor some real networks, including the Internet. Preferentialattachment can also be a consequence of different underly-ing processes based on node fitness, ranking, optimization,random walks, or duplication. Here we show that pop-ularity is just one dimension of attractiveness. Anotherdimension is similarity. We develop a framework wherenew connections, instead of preferring popular nodes, op-timize certain trade-offs between popularity and similar-ity. The framework admits a geometric interpretation, inwhich popularity preference emerges from local optimiza-tion. As opposed to preferential attachment, the optimiza-tion framework accurately describes large-scale evolution
of technological (Internet), social (web of trust), and bi-ological (E.coli metabolic) networks, predicting the prob-ability of new links in them with a remarkable precision.The developed framework can thus be used for predict-ing new links in evolving networks, and provides a differ-ent perspective on preferential attachment as an emergentphenomenon.
Fragkiskos PapadopoulosDepartment of Electrical EngineeringCyprus University of [email protected]
Maksim KitsakCooperative Association for Internet Data AnalysisUniversity of California, San [email protected]
M. Angeles Serrano, Marian BogunaDepartament de Fisica FonamentalUniversity of [email protected], [email protected]
Dmitri KrioukovCooperative Association for Internet Data AnalysisUniversity of California, San [email protected]
MS250
Testing Model Fit with Algebraic Statistics
We address the problem of model validation or goodness-of-fit testing: to assess whether a given model can be con-sidered as a satisfactory generative model for the data athand. As argued in Diaconis-Sturmfels ’98, a stochasticsearch of the space of networks with given values of suffi-cient statistics by Markov Chain Monte Carlo algorithmsyields bonafide tests for goodness of fit. Markov basesguarantee to connect all networks with the same sufficientstatistics, thus enabling this stochastic search. They arethe only set of moves that guarantee that the random walkwill give the real distribution. We show the algebraic toolsused to derive Markov bases for several network models.
Sonja PetrovicDepartment of StatisticsPennsylvania State [email protected]
Alessandro Rinaldo, Stephen FienbergCarnegie Mellon [email protected], [email protected]
MS250
Utilizing Spectral Methods for Uncued AnomalyDetection in Large-scale, Dynamic Networks
Abstract not available at time of publication.
Matt Schmidt, Nadya BlissMIT Lincoln [email protected], [email protected]
CS13 Abstracts 321
MS250
Recent Results in Statistical Network Analysis
Abstract not available at time of publication.
Patrick WolfeUniversity College [email protected]
MS251
A Scalable Data Centric Model for Security andProvenance
Enthusiasm for Big Data is spurring innovation in newtechnologies that seek to transform information into knowl-edge. Using the foundation of Apache Accumulo we intro-duce a data centric security model that compliments theadvances in Big Data technology and provides adaptabilityfor data scientists. Our approach re-conceptualizes the re-lationship among data, users and applications, unlockingthe potential for innovation and serving as a mechanismfor the integration of disparate data sets. We address dataprovenance, integration of disparate data sets, and the in-teraction of a diverse community to realize the potentialof Big Data to combat the spread of infectious disease, cli-mate change, and other issues of scientific importance.
Oren J. [email protected]
MS251
Large Data Analysis using the Dynamic Dis-tributed Dimensional Data Model (D4M)
The growth of bioinformatics, social analysis, and networkscience is forcing computational scientists to handle un-structured data in the form of genetic sequences, text, andgraphs. Triple store databases are a key enabling tech-nology for this data and are used by many large Inter-net companies (e.g., Google Big Table and Amazon Dy-namo). Triple stores are highly scalable and run on com-modity clusters, but lack interfaces to support efficient de-velopment of the mathematically based algorithms of thetyped used by computational scientists. D4M (DynamicDistributed Dimensional Data Model) provides a parallellinear algebraic interface to triple stores. Using D4M, it ispossible to create composable analytics with significantlyless effort than using traditional approaches. The centralmathematical concept of D4M is the associative array thatcombines spreadsheets, triple stores, and sparse linear al-gebra. Associative arrays are group theoretic constructsthat use fuzzy algebra to extend linear algebra to wordsand strings. This talk describes the D4M technology, itsmathematical foundations, application, and performance.
Jeremy KepnerMIT Lincoln [email protected]
MS251
How to Achieve Scalable Complex Analytics
In this talk we distinguish simple (SQL) analytics which arepopular in business data warehouses from complex analyt-ics, which most scientists are interested in. These are usu-ally domain-specific codes which have matrix operations asinner loops. We first discuss why RDBMSs are not likelyto work well for complex analytics. We then continue with
a collection of tactics to make such complex analytics per-form well and scale to arbitrary size data sets.
Michael StonebrakerMIT [email protected]
MS251
Streaming Algorithms and Large AstronomicalData Sets
As scientific data sets get larger, random access patternsto data become increasingly untenable. As a result, se-quential streaming is the primary way for scalable dataanalysis. Also, the main statistical challenges are shifting,as uncertainties are no longer statistical but due to sys-tematic errors. We discuss various astronomical analysischallenges related to and streaming algorithms, e.g. incre-mental robust PCA and how these can be implemented aspart of database access patterns.
Alex SzalayJohns Hopkins [email protected]
MS252
Graphlets: A Scalable, Multi-scale Decompositionfor Large Social and Information Networks
Abstract not available at time of publication.
Edo AiroldiHarvard [email protected]
MS252
From Cells to Tissue: Coping with Heterogeneitywhen Modelling the Electrophysiology of the Hu-man Heart
Abstract not available at time of publication.
Kevin BurrageUniversity of Oxford, & Queensland University ofTechnology& Institute for Molecular Bioscience, [email protected]
MS252
Stochastic Simulation Service: Towards an In-tegrated Development Environment for Modelingand Simulation of Stochastic Biochemical Systems
Abstract not available at time of publication.
Linda R. PetzoldUniversity of California, Santa [email protected]
MS252
Exploiting Stiffness for Efficient Discrete Stochas-tic Biochemical Simulation
Gillespie’s stochastic simulation algorithm (SSA) has be-come an invaluable tool for simulating biochemical modelsin a way that captures the inherent randomness of thesesystems. However, the SSA is inefficient for models featur-
322 CS13 Abstracts
ing stochastic stiffness arising from the presence of multi-ple timescales. In this talk, we present some counterintu-itive implications of stochastic stiffness and describe meth-ods for exploiting stiffness to improve simulation efficiency.Techniques for automating and accelerating existing algo-rithms will also be discussed.
Kevin SanftSt. Olaf [email protected]
MS253
The Powers that be in HPC
The power consumption of supercomputers ultimately lim-its their performance. The current challenge is not whetherwe will can build an exaflop system by 2018, but whetherwe can do it in less than 20 megawatts. The SCAPE Lab-oratory at Virginia Tech has been studying the tradeoffsbetween performance and power for over a decade. We’vedeveloped an extensive tool chain for monitoring and man-aging power and performance in supercomputers. We willdiscuss the implications of our findings for exascale systemsand some research directions ripe for innovation.
Kirk CameronDepartment of Computer ScienceVirginia [email protected]
MS253
Lower Bounds on Algorithm Energy Consumption:Current Work and Future Directions
By extending communication lower bounds on algorithmsvia linear models of machine energy consumption, we havederived theoretical bounds on the minimal amount of en-ergy required to run an algorithm on a given machine. Touse these bounds for HW/SW cotuning, efficient param-eters for energy models must be calculated. We discussinitial approaches to parameter calculation and further de-velopment of energy bounds into broader classes of codes.
Andrew GearhartUC [email protected]
James W. DemmelUniversity of CaliforniaDivision of Computer [email protected]
MS253
Energy-Aware Dense and Sparse Linear Algebra
Power is becoming a crucial challenge that the high per-formance computing community will have to face to effi-ciently leverage the Exascale systems that will be avail-able at the end of this decade. In this talk we will ad-dress several aspects related to this issue on multicore andmany-core (GPU) processors. Specifically, we introducea simple model of power dissipation, and a tools to com-pose a power tracing framework. Then, using experimentaldata, we evaluate both the potential and real energy reduc-tion that can be attained for the task-parallel execution ofdense and sparse linear algebra operations on multi-coreand many-core processors, when idle periods are leveraged
by promoting CPU cores to a power-saving C-state.
Enrique S. Quintana-OrtıUniversidad Jaume [email protected]
MS253
Locality Aware Scheduling of Sparse Computationsfor Energy and Performance Efficiencies
We consider the problem of increasing the performance andenergy efficiencies of sparse matrix and graph computationson multicore processors. Such systems have complex non-uniform memory access (NUMA) cache and memory thatexhibit significant variations in data access latencies. Wedescribe a scheme for fine-grain task scheduling to coresthat takes into account the probabilities of hits in cache.We present results indicating that our scheme leads to nearideal speed-ups and large improvements in performanceand energy efficiencies compared to traditional methods.
Padma RaghavanThe Pennsylvania State Univ.Dept of Computer Science [email protected]
MS254
Testing of Management Objective Hypotheses bySolution of Constrained Inverse Problems
Often, in the environmental sciences, too much faith isplaced in model predictions when there is no scientific basisfor doing so. We assert that model-based decision-makingshould implement the scientific method, and present an ap-proach best described as model-based hypothesis testing.The method seeks to reject the hypothesized occurrenceof a future event by demonstrating that it is incompatiblewith historical measurements of system state and expertknowledge.
Rachel S. BlakersNCGRT, iCAM, Fenner School of Environment andSocietyAustralian National [email protected]
MS254
Employing Imprecise Knowledge for Model Param-eter Inference and Prediction
In Bayesian analysis, the presence of ambiguity in subjec-tive or intersubjective knowledge often makes it difficultto formulate precise prior probability distributions. Theconcept of imprecise probability provides a framework forcharacterizing such ambiguity. When imprecise probabili-ties are used to represent ambiguous prior knowledge aboutmodel parameters, there are consequences for posteriorsand model results. By extending earlier studies, we showthat, under weak regularity conditions, the Density Ra-tio Class of imprecise probability measures is invariant un-der Bayesian inference, marginalization, and propagationthrough deterministic models. These invariance propertiesare desirable because they make it possible to describe se-quential Bayesian learning and prediction using impreciseprobabilities within a unified framework. We proceed todescribe algorithms that exploit these invariance proper-ties to minimize the additional computational burden com-pared to the use of precise priors. To demonstrate the fea-
CS13 Abstracts 323
sibility of the proposed approach, the concepts and numer-ical methods developed are applied to a simple empiricalecological model.
Simon RinderknechtEawagSwiss Federal Institute of Aquatic Science and [email protected]
Carlo [email protected]
Mark BorsukThayer School of EngineeringDartmouth [email protected]
Nele SchuwirthEawagSwiss Federal Institute of Aquatic Science and [email protected]
Hans-Rudolf KuenschSeminar for [email protected]
Peter ReichertEawag: Swiss Federal Inst. of Aquatic Science [email protected]
MS254
Uncertainty Assessment by Algebraic PlausibleWorst-case Falsification of Conclusions
The minisymposium topic is introduced, using a typologyto highlight key aspects of deep uncertainty. Key tech-niques are summarised, including imprecise probabilities,measures of robustness and various approaches to falsify-ing hypotheses. Using groundwater management exam-ples, we show that algebraic techniques to falsification canefficiently identify plausible worst case scenarios, effectivelystress testing model conclusions. Accounting for uncer-tainty in this way helps surface assumptions and can sim-plify the application of other techniques.
Joseph H. Guillaume, Anthony J. JakemanNCGRT, iCAM, Fenner School of Environment andSocietyAustralian National [email protected], [email protected]
MS254
Many Objective Robust Decision Making for Com-plex Environmental Systems Undergoing Change
We present a many objective robust decision making frame-work, combining evolutionary optimization, robust de-cision making (RDM), and interactive visual analytics.Many objective evolutionary search enables the discoveryof tradeoffs between objectives. Subsequently, RDM deter-mines the robustness of the component tradeoff solutionsto deeply uncertain future conditions. Results suggest thatincluding robustness as a decision criterion can dramati-cally change the formulation of environmental managementproblems as well as the negotiated selection of candidate
alternatives.
Joseph R. KasprzykDepartment of Civil and Environmental EngineeringThe Pennsylvania State [email protected]
Shanthi NatarajRAND [email protected]
Patrick ReedThe Pennsylvania State [email protected]
Robert LempertRAND [email protected]
MS255
A Composed and Multilevel Solution Frameworkfor Nonlinear PDEs
Nonlinear PDE solvers enjoy a fairly large design spacethat is underexplored in practical application. Hierarchiesof solution techniques, such as local or multilevel non-linear solvers accelerated by nonlinear Krylov or quasi-Newton methods, provide a flexible alternative to thenested Newton-Krylov iteration. We present a framework,available in the PETSc library, for large-scale composedand multilevel nonlinear solvers. We demonstrate the ben-efits and drawbacks of these methods on problems of inter-est.
Peter R. BruneMathematics and Computer Science DivisionArgonne National [email protected]
Barry F. SmithArgonne National LabMCS [email protected]
MS255
A Multilevel Stochastic Collocation Algorithm forOptimization of PDEs with Uncertain Coefficients
Optimization of PDEs with uncertain coefficients is expen-sive because problem size depends on both spatial and sam-ple space discretizations. I analyze the use of MG/OPT toformulate a multilevel stochastic collocation algorithm byexploiting the hierarchical structure of sparse grids. For aclass of problems and methods, I present convegence analy-sis as well as explicit cost and error bounds for one V-cycleof MG/OPT. I provide numerical illustrations confirmingthese bounds.
Drew P. KouriMathematics and Computer Science DivisionArgonne National [email protected]
MS255
Using Inexact Gradients in a Multilevel Optimiza-tion Algorithm
Many optimization algorithms require gradients of the
324 CS13 Abstracts
model functions, but computing accurate gradients can beexpensive. We study the impact of inexact gradients onthe multilevel optimization algorithm MG/Opt. MG/Optrecursively uses coarse models to obtain search directionsfor fine models. However, MG/Opt requires fine-level gra-dients to define the recursion. We consider various possiblesources of error in the gradients, and demonstrate that inmany cases the effect of the errors is benign.
Stephen G. NashGeorge Mason UniversitySystems Engineering & Operations Research [email protected]
Robert Michael LewisCollege of William & MaryWilliamsburg, VA [email protected]
MS255
Parallel Software for the Optimization AlgorithmDIRECT with High-cost Function Evaluations
This talk concerns derivative-free global optimization usinga large number of extreme-scale function evaluations, illus-trated by parameter estimation for basis set selection inthe ab initio nuclear physics code MFDn (Many FermionDynamics, nuclear). A hierarchical three-tier scheme forintegrating a massively parallel DIRECT implementation,pVTdirect, with MFDn is considered. Additional pVTdi-rect modifications to support a dynamic number of pro-cesses to accommodate unpredictable runtime memory de-mands are discussed.
Masha SosonkinaAmes Laboratory/DOEIowa State [email protected]
Layne T. WatsonVirginia Polytechnic Institute and State UniversityDepartments of Computer Science and [email protected]
MS256
A(nother) Randomized Rank-revealing Decompo-sition: Generalized RURV
Fast, randomized rank-revealing decompositions haveemerged in the last decade, in particular, as a means toconstruct low-rank or sparse approximations of matrices.Various types of algorithms emerged, some involving lessrandomization (through the use of FFT-like matrices), andsome more. We present here one that uses more random-ization, but the results are guaranteed for a wider rangeof numerical ranks, and as a bonus, it is straightforwardlygeneralizable to a product of matrices and inverses. Thisturns out to be a very desirable feature, given that it wasdeveloped for a communication-minimizing non-symmetriceigenvalue algorithm using divide-and-conquer.
Ioana DumitriuUniversity of Washington, [email protected]
MS256
Beating MKL and Scalapack at Rectangular Ma-
trix Multiplication Using the BFS/DFS Approach
We present CARMA, a communication-avoiding parallelrectangular matrix multiplication algorithm, attaining sig-nificant speedups over both MKL and ScaLAPACK. Com-bining the recursive BFS/DFS approach of Ballard, Dem-mel, Holtz, Lipshitz and Schwartz (SPAA ’12) with the di-mension splitting technique of Frigo, Leiserson, Prokop andRamachandran (FOCS ’99), CARMA is communication-optimal, cache- and network-oblivious, and simple to im-plement (e.g., 60 lines of code for the shared-memory ver-sion), performing well on both shared- and distributed-memory machines.
James W. DemmelUniversity of CaliforniaDivision of Computer [email protected]
David EliahuUC [email protected]
Armando FoxUniversity of California, [email protected]
Shoaib KamilMassachusetts Institute of [email protected]
Benjamin Lipshitz, Oded Schwartz, Omer SpillingerUC [email protected], [email protected],[email protected]
MS256
Communication Avoiding ILU(0) Preconditioner(CA-ILU(0))
We present a parallel communication-avoiding ILU(0) pre-conditioner for systems Ax = b, where A is sparse. First,A is reordered using nested dissection, then the blocks andseparators are reordered to avoid communication at the ex-pense of performing redundant computations. Our specialreordering doesn’t affect the convergence rate of the ILU(0)preconditioned systems as compared to nested dissectionreordering of A, while it reduces data movement and redun-dant flops. Thus, CA-ILU(0) preconditioner should havegood performance.
Sophie MoufawadLRI - INRIA Saclay, [email protected]
Laura [email protected]
MS256
Scalable Numerical Algorithms for ElectronicStructure Calculations
This talk will focus on dense linear algebra and ten-sor computations in application to electronic calcula-tions. I will introduce 2.5D algorithms, which are de-signed to minimize communication between processors.
CS13 Abstracts 325
The algorithms employ limited data-replication to theoret-ically lower communication costs with respect to standard(ScaLAPACK/Elemental) algorithms. I will detail imple-mentations of 2.5D matrix multiplication and LU factor-ization, which employ topology-aware mapping on super-computers such as BG/P and BG/Q. 2.5D algorithms havebeen employed on BG/Q by QBox, a parallel DFT code forelectronic structure calculations. The communication min-imization and topology-aware mapping schemes will alsobe extended to tensor contractions as embodied by Cy-clops Tensor Framework. The Coupled Cluster method forsystems with high electronic correlation has already beenimplemented on top of this framework and is achievingstrong performance results on BG/Q.
Edgar SolomonikUniversity of California at [email protected]
MS257
Reduced Collocation Methods: Reduced basisMethods in the Collocation Framework
In this talk, we present a reduced basis method well-suitedfor the collocation framework. Two fundamentally differ-ent algorithms will be presented. This work provides areduced basis strategy to practitioners who prefer a col-location, rather than Galerkin, approach. Furthermore,one of these two algorithms eliminates a potentially costlyonline procedure that is needed for non-affine problemswith Galerkin approach. Numerical results demonstratethe high efficiency and accuracy of the reduced collocationmethods, which match or exceed that of the traditionalreduced basis method in the Galerkin framework.
Yanlai ChenDepartment of MathematicsUniversity of Massachusetts [email protected]
Sigal GottliebDepartment of MathematicsUniversity of Massachusetts [email protected]
MS257
A Component-based Reduced basis Method forMany-parameter Systems
We present a reduced basis method based that combinesstandard reduced basis approximations with a static con-densation formulation. This allows us to develop stan-dalone parametrized reduced basis components that canbe connected together to form large systems with manyparameters. Also, this enables an appealing ”system as-sembly” approach, which we demonstrate in an interactiveGUI. We show numerical results from applications drawnfrom structural analysis and acoustics.
David KnezevicHarvard [email protected]
MS257
Non-intrusive Reduced basis Approximations forParameter Dependent PDEs
Reduced basis methods take advantage of the small Kol-
mogorov dimension of the manifold of all solutions to someparameter dependent partial differential equation (whenthe parameter vary). These allow to propose rapid approxi-mation methods for real time on-line approximations basedon reduction of complexity. Numerical analysis based ona posteriori error estimation allow to provide additionaltools that allow to validate the computations with reducedbasis approximations. These approximations methods aread’hoc and require a preliminary off-line analysis that usesall the previously tools together a fine classical approxima-tion of e.g. finite element type. The reduced basis will bean alternative and complement to these classical methodsand will allow to mimic it. The rapid on-line implemen-tation and off-line learning process require to have a fullaccess to the original classical solver. When this is not thecase we have to imagine alternative tools that allow to cir-cumvent this lack of full access (as can be the case whenthe code is an industrial one). We shall present some recentadvances in the direction of the non intrusive implementa-tion.
Yvon MadayUniversite Pierre et Marie Curieand Brown [email protected]
MS257
Component-based Reduced basis Simulations:Conjugate Heat Transfer and Transient Problems
We present some applications of the Static CondensationReduced Basis Element (SCRBE) method: a domain de-composition with reduced basis at the intradomain levelto populate a Schur complement at the interdomain level.We present numerical results for a conjugate heat transferproblem (2D car radiator model) as well as transient heattransfer problems on 3D configurations, which demonstratethe flexibility, accuracy and computational efficiency of ourapproach.
Sylvain Vallaghe, Anthony T. PateraMassachusetts Institute of [email protected], [email protected]
MS258
A Fluid-Based Preconditioner for Fully Implicit Ki-netic Electrostatic Plasma Simulations
We introduce a fluid-based preconditioner for a re-cently proposed energy- and charge-conserving fully im-plicit, particle-in-cell electrostatic algorithm in 1D [Chen,Chacon, Barnes. J.Comp.Phys, 230, 7018 (2011)]. The ap-proach employs a linearized form of the first two momentequations, with suitable kinetic closures, to find approxi-mate updates of the self-consistent electric field from thelinearized Ampere’s law. The effectiveness of the result-ing preconditioner will be demonstrated on a challengingmultiscale ion acoustic wave problem.
Guangye [email protected]
Luis ChaconLos Alamos National [email protected]
326 CS13 Abstracts
MS258
Coupled Simulation of Continuum Material PointMethod with Molecular Dynamics NumericalStatistics
Material point method uses both Lagrangian and Eule-rian descriptions for material motion, avoids mesh distor-tion and numerical diffusion issues. When combined withMolecular Dynamics in a scale bridging algorithms, wehave the advantage of simulating whole domain with min-imal communication between molecular dynamic systemsto achieve parallel computation efficiency close to the em-barrassing parallelism. We present examples to show thatthis algorithm can be implemented in combined GPU/CPUplatform.
Duan Z. Zhang, Dana KnollLos Alamos National [email protected], [email protected]
MS258
Moment Acceleration of Fokker-Planck CollisionOperator for Electrostatic Plasma Physics Simu-lation
For time step sizes significantly larger than the colli-sion time scale, a fully implicit solution to the Fokker-Planck operator requires many fixed point iterations toconverge. The convergence of the fixed point iteration forthe Fokker-Planck collision operator will be accelerated us-ing an emerging class of moment based accelerators. Itis demonstrated that upon convergence, we obtain a fullyconsistent kinetic and moment solution.
William T. TaitanoUniversity of New MwxicoLos Alamos National [email protected]
Dana KnollLos Alamos National [email protected]
Anil K. PrinjaUniversity of New [email protected]
MS258
An Accelerated Free Surface, Z-Level Ocean Modelusing a Moment-Based Approach and Trilinos
We study a moment based method for a free-surface z-levelocean model. In this approach, the full three dimensionalmomentum and continuity equations are coupled to a setof two dimensional moment equations, obtained by verti-cal integration. This formulation allows relaxation of thetimestep size by isolating the stiff physics to the reducedsystem. We provide numerical examples to support ourstudy and compare to traditional implementations.
Geoff Womeldorff, Christopher K. Newman, Dana KnollLos Alamos National [email protected], [email protected], [email protected]
MS259
Problem Formulation for Multi-source Optimiza-tion in Complex Systems
The operation of a single complex artifact is governed bymany models that are combined at various stages of design,together with some statement of constrained optimizationto form a model used for designing the artifact a designproblem formulation. The use of models within variousformulations influences the tractability and robustness ofthe solution. We discuss the synthesis of models originatingfrom a variety of sources into problem formulations and theresulting computational properties.
Natalia AlexandrovNASA Langley Research [email protected]
MS259
Sensitivity Analysis and Information Fusion forMulti-source Optimization
A multifidelity approach to design seeks to exploit opti-mally all available information. Existing methods gener-ally calibrate or replace low-fidelity results using higherfidelity information. Here we propose an approach basedon estimation theory to fuse information from multifidelitysources. Mathematical interrogation of the uncertainty inquantities of interest is achieved via global sensitivity anal-ysis, which provides guidance for adaptation of model fi-delity. The methodology is demonstrated on a wing-sizingproblem for a high-altitude, long-endurance vehicle.
Doug Allaire, Karen E. WillcoxMassachusetts Institute of [email protected], [email protected]
MS259
On the Use of Self-organizing Maps for Data Man-agement in a Multifidelity Analysis Context
The use of Self Organizing Maps (SOM), unsupervised neu-ral networks also known as Kohonens Maps, for the man-agement of data in a variable fidelity analysis environmentis discussed. In particular the employment of SOM forscreening purposes is introduced and applied to the multi-fidelity aerodynamic analysis of an aircraft wing. Moreoverthe possibility to use those networks to address local errorsof low fidelity models for corrective purposes is finally pro-posed.
Laura MaininiPolytechnic University of [email protected]
MS259
Numerical Bouillabaisse: Combining Experiments,Simulations and Machine Learning to Guide Opti-mization
In conventional optimization one iteratively samples a sin-gle source of information concerning a function f(.), e.g.,one samples f(x) itself. However in practice, often at eachstep one must choose which of several information sourcesabout f to sample. Here I illustrate how to use semi-supervised machine learning to statistically combine sam-ples from multiple sources. I also introduce an algorithmfor deciding which information source to sample next and
CS13 Abstracts 327
with what input values.
David WolpertNASA Ames Research [email protected]
MS260
Multiscale Methods and Analysis for the Nonlin-ear Klein-Gordon Equation in the NonrelativisticLimit Regime
In this talk, I will review our recent works on numeri-cal methods and analysis for solving the nonlinear Klein-Gordon (KG) equation in the nonrelativistic limit regime,involving a small dimensionless parameter which is in-versely proportional to the speed of light. In this regime,the solution is highly oscillating in time and the energybecomes unbounded, which bring significant difficulty inanalysis and heavy burden in numerical computation. Webegin with four frequently used finite difference time do-main (FDTD) methods and obtain their rigorous error es-timates in the nonrelativistic limit regime by paying par-ticularly attention to how error bounds depend explicitlyon mesh size and time step as well as the small parameter.Then we consider a numerical method by using spectralmethod for spatial derivatives combined with an exponen-tial wave integrator (EWI) in the Gautschi-type for tem-poral derivatives to discretize the KG equation. Rigoriouserror estimates show that the EWI spectral method showmuch better temporal resolution than the FDTD methodsfor the KG equation in the nonrelativistic limit regime. Inorder to design a multiscale method for the KG equation,we establish error estimates of FDTD and EWI spectralmethods for the nonlinear Schrodinger equation perturbedwith a wave operator. Finally, a multiscale method is pre-sented for discretizing the nonlinear KG equation in thenonrelativistic limit regime based on large-small amplitudewave decompostion. This multiscale method converges uni-formly in spatial/temporal discretization with respect tothe small parameter for the nonlinear KG equation in thenonrelativistic limite regime. Finally, applications to sev-eral high oscillatory dispersive partial differential equationswill be discussed.
Weizhu BaoNational University of SingaporeDepartment of [email protected]
MS260
Uniform Error Estimates of An Exponential WaveIntegrator Sine Pseudospectral Method for Nonlin-ear Schrodinger Equation with Wave Operator
We propose an exponential wave integrator sine pseu-dospectral (EWI-SP) method for the nonlinear Schrodingerequation (NLS) with wave operator (NLSW), and carry outrigorous error analysis. NLSW is NLS perturbed by thewave operator with strength described by a dimensionlessparameter ε ∈ (0, 1]. In this work, we show that the pro-posed EWI-SP possesses the optimal uniform error boundsat O(τ 2) and O(τ ) in τ (time step) for well-prepared initialdata and ill-prepared initial data, respectively, and spec-tral accuracy in h (mesh size) for the both cases, in the L2
and semi-H1 norms.
Yongyong CaiNational University of [email protected]
Weizhu BaoNational University of SingaporeDepartment of [email protected]
MS260
Fast Computation of Time-Domain Electromag-netic Scattering Problems with Exact TransparentBoundary Conditions
Wave propagations in unbounded media arise from di-verse applications. Intensive research has been devoted tofrequency-domain simulation, e.g., for the time-harmonicHelmholtz problems and Maxwell’s equations. Here, we areinterested in time-domain computation, which is known tobe more flexible in capturing wide-band signals and model-ing more general material inhomogeneities and nonlineari-ties. In this talk, we shall show how to use tools in complexanalysis to analytically evaluate circular and spherical non-reflecting boundary conditions (NRBCs), and how to effi-ciently deal with time-space globalness of such boundaryconditions. Fast spectral-Galerkin solvers together withstable time integration and techniques for handling generalirregular scatterers will be introduced for the simulation.We intend to demonstrate that the interplay between ana-lytic tools, accurate numerical means and sometimes bruteforce hand calculations can lead to efficient methodologiesfor challenging simulations. This is a joint work with BoWang and Xiaodan Zhao.
Li-Lian WangSchool of Physical & Mathematical SciencesNanyang Technological [email protected]
MS260
Uniformly Correct Multiscale Time Integrators forHighly Oscillatory Second Order Differential Equa-tions
In this talk, two multiscale time integrators (MTIs), moti-vated from two types of multiscale decomposition by eitherfrequency or frequency and amplitude, are proposed andanalyzed for solving highly oscillatory second order differ-ential equations with a dimensionless parameter 0 < ε � 1.In fact, the solution to this equation propagates waves withwavelength at O(ε2) when 0 < ε � 1. We rigorously es-tablish two independent error bounds for the two MTIsas O(τ 2/ε2) and O(ε2) for ε ∈ (0, 1] with τ > 0 as stepsize, which imply that the two MTIs converge uniformlywith linear convergence rate at O(τ ) for ε ∈ (0, 1] and op-timally with quadratic convergence rate at O(τ 2) in theregimes when either ε = O(1) or 0 < ε � τ . Thus themeshing strategy requirement (or ε-scalability) of the twoMTIs is τ = O(1) for 0 < ε � 1, which is significantlyimproved from τ = O(ε3) and τ = O(ε2) requested by fi-nite difference methods and exponential wave integratorsto the equation, respectively. At last numerical results arereported to support the two error bounds.
Xiaofei ZhaoNational University of SingaporeDepartment of [email protected]
Bao Weizhu, Dong XuanchunDepartment of MathematicsNational University of [email protected], [email protected]
328 CS13 Abstracts
MS261
Title Not Available at Time of Publication
Abstract not available at time of publication.
Yalchin EfendievDept of MathematicsTexas A&M [email protected]
MS261
Estimation of Anthropogenic CO2 Emissions fromSparse Observations using a Multiresolution Ran-dom Field Model
We present a method to estimate fossil-fuel CO2 emissionsfrom observed CO2 concentrations at a sparse set of loca-tions. The emission field is represented using wavelets, withnightlight images providing the spatial model. Sparsity-enforced estimation is used to reconstruct the emissionfield; wavelets unconstrained by the observations are set tozero. The method is tested with synthetic data. Emissionsfrom North-eastern US are estimated most accurately, andthe South-west is the least constrained by observations.
Jaideep RaySandia National Laboratories, Livermore, [email protected]
Vineet YadavCarnegie Institution for ScienceStanford, [email protected]
Sean A. MckennaSandia National [email protected]
Anna MichalakCarnegie Institution for ScienceStanford, [email protected]
Bart van Bloeman WaandersSandia National [email protected]
MS261
Quantification of Parameter Uncertainties in Non-linear Distributed Material Models
Piezoelectric, magnetic and shape memory alloy (SMA)materials offer unique capabilities for energy harvestingand reduced energy requirements in aerospace, aeronautic,automotive, industrial and biomedical applications. How-ever, all of these materials exhibit creep, rate-dependenthysteresis, and constitutive nonlinearities that must be in-corporated in models and model-based robust control de-signs to achieve their full potential. Furthermore, mod-els and control designs must be constructed in a mannerthat incorporates parameter and model uncertainties andpermits predictions with quantified uncertainties. In thispresentation, we will discuss Bayesian techniques to quan-tify parameter uncertainties in nonlinear distributed mod-els arising in the context of smart systems. We will alsodiscuss highly efficient techniques to propagate these un-certainties through models to quantify uncertainties asso-
ciated with quantities of interest.
Ralph C. SmithNorth Carolina State UnivDept of Mathematics, [email protected]
Nathanial [email protected]
Zhengzheng HuDepartment of MathematicsNorth carolina State [email protected]
Michael HaysFlorida State [email protected]
William OatesDepartment of Mechanical Engineering, Florida [email protected]
MS261
Calibration of Uncertain Parameters in StochasticTurbulence Models
Abstract not available at time of publication.
Catalin S. TrencheaDepartment of MathematicsUniversity of [email protected]
William LaytonUniversity of [email protected]
Clayton G. WebsterOak Ridge National [email protected]
MS262
Multiscale Modeling of Energy Storage Device
An efficient implementation of the Poisson-Nernst-Plank -classical Density Functional Theory (PNP-cDFT), a mul-tiscale method for the simulations of charge transport innanocomposite materials, is designed and evaluated. Spa-tial decomposition of the multi particle system is employedin the parallelization of classical density functional the-ory (cDFT) algorithm. Furthermore, a truncated strategyis used to reduce the computational complexity of cDFTalgorithm. The simulation results show that the paral-lel implementation has close to linear scalability in parallelcomputing environments for both 1D and 3D systems. Ad-ditionally, we develop robust numerical algorithms to findsteady state fluxes in 2D circular nano-particles made upof rutile TiO2. The equations we solve come from densityfunctional theory coupled with the Poisson-Nernst-Plankformalism. In solving steady state fluxes, we come to abetter understanding of how battery properties are influ-enced by nano-particle arrangement which influence theflux of Lithium ions.
Lin Guang
CS13 Abstracts 329
MS262
Multiscale Computation of Heterogeneous SurfaceCatalytic Reactions
Kinetic Monte Carlo simulations are typically performed todetermine steady state values in kinetic reactions for catal-ysis. We present a full non-linear master equation capableof resolving these dynamics; we also present a linearizedapproximation whose solutions and eigenstructure can befully classified. The equations are simple, yet have manydegrees of freedom. We use symbolic packages to generatecode to solve the resulting ODEs for the master equationand its linear approximation.
Gregory HerschlagDepartment of MathematicsUniversity of North Carolina at Chapel [email protected]
Sorin MitranUniversity of North Carolina Chapel [email protected]
MS262
Numerical Simulation of Reactive Particle Com-pacts
A computational model is developed for the simulation oftransient reactions in heterogeneous porous compacts of re-active multilayered particles. The evolution of the reactionis described in terms of a reduced model formalism that ef-ficiently accounts for thermal transport at the macroscale,and for atomic mixing and chemical heat release at the mi-cro or nanoscale. Computations are used to analyze theeffects of microstructural parameters on the mean proper-ties of the reaction front.
Leen AlawiehJohns Hopkins [email protected]
Ihab SrajDuke [email protected]
Timothy WeihsJohns Hopkins [email protected]
Omar M. KnioDuke [email protected]
MS262
Quantifying Uncertainty in Ionic Flow through aSilica Nanopore
We discuss uncertainty quantification in MD simulations ofconcentration-driven flow through a silica nanopore. Wemodel all components, namely water, silica, sodium andchloride ions, and explore the sensitivity of the flow to thepore diameter and gating charge. A Bayesian inferenceformalism, coupled with polynomial chaos representations,is used for this purpose. In addition to sensitivity analysis,the study focuses on complex effects arising due to system
heterogeneity as well as intrinsic thermal fluctuations.
Francesco RizziDepartment of Mechanical EngineeringJohns Hopkins [email protected]
Reese [email protected]
Bert J. DebusschereEnergy Transportation CenterSandia National Laboratories, Livermore [email protected]
Omar M. KnioDuke [email protected]
MS263
A Global Jacobian Method for Simultaneous Solu-tion of Mortar and Subdomain Variables in Non-linear Porous Media Flow
We describe a new algorithm to perform non-overlappingdomain decomposition with nonlinear model problems,called the Global Jacobian (GJ) method. The are two mainideas: (1) we linearize the global system in both subdomainand interface variables simultaneously to yield a singleNewton iteration; and (2) we algebraically eliminate sub-domain velocities (and optionally, subdomain pressures) tosolve either the 1st or 2nd Schur complement systems. TheGJ method improves upon the previous nonlinear mortaralgorithm, which required two nested Newton iterationsand a Forward Difference (FD) approximation.
Benjamin GanisThe University of Texas at AustinCenter for Subsurface [email protected]
Mika JuntunenUniversity of Texas at Austin, ICES / [email protected]
Gergina PenchevaUniversity of Texas at [email protected]
Mary F. WheelerCenter for Subsurface ModelingUniversity of Texas at [email protected]
Ivan YotovUniveristy of PittsburghDepartment of [email protected]
MS263
Feedback Control of the Boussinesq Equations withApplication to Control of Energy Efficient BuildingSystems
In this talk, we present theoretical and numerical results forfeedback control of the Boussinesq Equations. The prob-lem is motivated by design and control of energy efficient
330 CS13 Abstracts
building systems. In particular, new low energy conceptssuch as chilled beams and radiant heating lead to problemswith Dirichlet, Neumann and Robin type boundary condi-tions. It is natural to consider control formulations thataccount for minimizing energy consumption and providingreasonable performance. We discuss a LQR type controlproblem for this system with Robin/Neumann boundarycontrol inputs and apply the results to a 2D problem toillustrate the ideas and demonstrate the computational al-gorithms.
John A. BurnsVirginia TechInterdisciplinary Center for Applied [email protected]
Xiaoming HeDepartment of Mathematics and StatisticsMissouri University of Science and [email protected]
Weiwei HuUniversity of Southern [email protected]
MS263
Model Reduction of Nonlinear PDEs Using GroupPOD
We propose a new method to reduce the cost of computingnonlinear terms in projection based reduced order mod-els with global basis functions. We develop this methodby extending ideas from the group finite element methodto proper orthogonal decomposition (POD). Numerical re-sults for a scalar 2D Burgers’ equation show that the pro-posed group POD reduced order models are as accurateand are computationally more efficient than standard PODmodels of the Burgers’ equation.
John SinglerMissouri S & TMathematics and [email protected]
Benjamin DickinsonMunitions Directorate, Air Force Research LaboratoryEglin [email protected]
MS263
New Efficient Splitting Methods for Flow Problems
In this talk, we develop new reliable and efficient splittingmethods for simulating fluid flows in two different prob-lems: magnetohydrodynamics and Navier-Stokes equationswith Coriolis force. For the former, we study partitionedmethods which allow us to uncouple the problem by solvingone each of the subphysics problems per time step (with-out iteration). For the latter, we present a fast-slow wavesplitting method with Crank-Nicolson Leap-Frog scheme,which is unconditionally stable. We will discussed the sta-bility and accuracy of our methods and give numerical ex-periments that support the theory. This is the joint workwith Nan Jiang, William Layton and Catalin Trenchea (allfrom University of Pittsburgh).
Hoang A. TranUniversity of PittsburgDepartment of Mathematics
MS264
Quasi-Newton Update of Preconditioners for theLinearized Newton System Arising from 3d Dis-cretizations of Groundwater Flow Models
We discuss quasi-Newton updates of preconditioners forlinear systems arising in 3D groundwater flow problems.
Luca BergamaschiUniversita di [email protected]
MS264
Update Preconditioners and Incomplete Decompo-sitions using Approximate Inverses
In the past ten years we propose a framework for updatingpreconditioners in factorized form for sequences of generallinear systems based on the use of inversion and sparsifica-tion of the incomplete factorizations. Examples of success-ful applications to sequences of systems arising in linearand nonlinear PDEs, image restoration, inexact Newtonsolvers, optimization, approximation of functions of largematrices and some issues of the proposed strategies will beconsidered.S. Bellavia, D. Bertaccini, and B. Morini, Nonsymmetricpreconditioner updates in Newton-Krylov methods for non-linear systems, SIAM J. Sci. Comput., 33 (2011), pp. 2595-2619.M. Benzi and D. Bertaccini, Approximate inverse precon-ditioning for shifted linear systems, BIT Numerical Math-ematics, 43 (2003), pp. 231-244.D. Bertaccini, Efficient preconditioning for sequences ofparametric complex symmetric linear systems, ElectronicTrans. on Num. Anal., 18 (2004), pp. 49-64.D. Bertaccini and F. Sgallari, Updating preconditioners fornonlinear deblurring and denoising image restoration, Ap-plied Numerical Mathematics, 60 (2010), pp. 994-1006.
Daniele BertacciniDepartment of MathematicsUniversita’ di Roma Tor [email protected]
MS264
Incremental Approximate LU Factorizations andApplications
A problem that is common to many applications is to solvea sequence of linear systems whose matrices change onlyslightly from one step to the next. We address the issue ofsolving these systems using an ILU preconditioner withoutrecomputing entirely the ILU factorizations but by updat-ing them instead. Mathematical properties and implemen-tation aspects of these incremental methods are discussedand numerical tests on a collection of linear systems andon the 2D Navier-Stokes equation with variable density arepresented.
Caterina CalgaroLaboratoire Paul Painleve - UMR 8524Universite Lille 1, Sciences et [email protected]
Jean-Paul Chebab
CS13 Abstracts 331
Laboratoire Amienois de Mathematiques Fondamentaleset Appl.Universite de Picardie Jules [email protected]
Yousef SaadDepartment of Computer ScienceUniversity of [email protected]
MS264
Updating Preconditioners for Model Reductionand Other Parameterized Systems
After a brief introduction on various approaches to up-dating preconditioners for sequences of linear systems, wewill discuss our recent efforts for updating preconditionersfor parameterized linear systems. Such systems arise in arange of applications such as acoustics, (parametric) modelreduction, and inverse problems. We will show a verygeneral approach to updating preconditioners for modestchanges in parameters, thereby significantly reducing thecost of computing preconditioners while maintaining goodconvergence.
Eric De SturlerVirginia [email protected]
Serkan GugercinVirginia Tech.Department of [email protected]
MS265
Placing Communicating Tasks Apart to MaximizeEffective Bandwidth
Topology-aware mapping algorithms should take the appli-cation behavior (latency versus bandwidth-bound) and tar-get platform characteristics (routing schemes, implemen-tation of collectives, etc.) into account. When mappingon n-dimensional tori, most previous work has focused onbringing communicating tasks closer on the network. Wepresent a non-intuitive idea of moving tasks farther apartto increase the number of available routes and effectivebandwidth on torus networks. Bandwidth-bound applica-tions can benefit significantly from heuristics that imple-ment such mappings.LLNL-ABS-571694
Abhinav Bhatele, Todd GamblinLawrence Livermore National [email protected], [email protected]
MS265
Improving MPI Process Mapping for CartesianTopologies on Multicore Nodes
We consider the problem of MPI process mapping for mod-ern architectures that contain many cores on a single CPUwith complex NUMA architectures. Although the MPIstandard provides routines for process mapping in Carte-sian topologies, most implementations do nothing to re-duce off-node communications. We evaluate methods toimprove process mapping on multicore nodes. We evaluateperformance for hybrid nodes with accelerators; multipleMPI processes sharing an accelerator are used to overlap
host-device communication with computation.
William M. BrownComputational BiologySandia National [email protected]
MS265
Task Mapping for Noncontiguous Allocations
This talk presents task mapping algorithms for non-contiguously allocated parallel jobs. Previous work on taskmapping either uses a very general model that is hard toapply in practice or assumes that jobs are allocated to becompletely isolated from each other. We pose the mappingproblem for non-contiguous jobs utilizing a simple sten-cil communication pattern. We then devise several taskmapping algorithms for this problem and evaluate theirperformance using simulations and experiments.
David BundeKnox [email protected]
Vitus LeungSandia National [email protected]
MS265
Ensuring Continued Scalability of Mesh Based Hy-drocodes
CTH is a widely used shock hydrodynamics code basedon the bulk synchronous parallel programming model. Atincreasingly large processor counts, we noted that its near-est neighbor communication performance significantly de-graded. MiniGhost, a miniapp in the Mantevo suite, wasconfigured to mimic CTH’s communication patterns, ex-posed the problem, and led to a solution that illustratesand informs a fundamental issue that must be consideredin our preparations for exascale computing.
Courtenay T. VaughanSandia National [email protected]
MS266
Energy Conserving Local Discontinuous GalerkinMethods for the Wave Propagation Problems
Wave propagation problems arise in a wide range of ap-plications. The energy conserving property is one of theguiding principles for numerical algorithms, in order tominimize the phase or shape errors after long time integra-tion. In this presentation, we develop and analyze a localdiscontinuous Galerkin (LDG) method for solving the waveequation. We prove optimal error estimates and the energyconserving property for the semi-discrete formulation. Theanalysis is extended to the fully discrete LDG scheme, withthe energy conserving high order time discretization. Nu-merical experiments have been provided to demonstratethe optimal ratesof convergence. We also show that theshape of the solution, after long time integration, is wellpreserved due to the energy conserving property.
Yulong XingDepartment of MathematicsUniveristy of Tennessee / Oak Ridge National [email protected]
332 CS13 Abstracts
MS266
Energy Stable Numerical Schemes and Simulationsof Two Phase Complex Fluids on Phase FieldMethod
We present an energetic variational phase-field model forthe two-phase Incompressible flow with one phase beingthe nematic liquid crystal. The model leads to a couplednonlinear system satisfying an energy law. An efficient andeasy-to-implement numerical scheme is presented for solv-ing the coupled nonlinear system. We use this scheme tosimulate two benchmark experiments: one is the formationof a bead-on-a-string phenomena, and the other is the dy-namics of drop pinching-off. We investigate the detaileddynamical pinch-off behavior, as well as the formation ofthe consequent satellite droplets, by varying order param-eters of liquid crystal bulk and interfacial anchoring en-ergy constant. Qualitative agreements with experimentalresults are observed.
Xiaofeng YangUniversity of South [email protected]
MS266
Krylov Implicit Integration Factor WENO Meth-ods for Advection-diffusion-reaction Systems
Implicit integration factor (IIF) methods are originallya class of efficient “exactly linear part” time discretiza-tion methods for solving time-dependent partial differen-tial equations with linear high order terms and stiff lowerorder nonlinear terms. In this paper, we developed newKrylov IIF-WENO methods to deal with both semilinearand fully nonlinear advection-diffusion-reaction equations.The methods are free of operator splitting error and canbe designed for arbitrary order of accuracy. The stiffnessof the system is resolved well and the methods are stableby using time step sizes which are just determined by thenon-stiff hyerbolic part of the system.
Tian Jiang, Yongtao ZhangNotre Dame [email protected], [email protected]
MS266
A Simple WENO Limiter for RKDG Methods
We investigate a simple limiter using weighted essen-tially non-oscillatory (WENO) methodology for the Runge-Kutta discontinuous Galerkin (RKDG) methods solvingconservation laws, with the goal of obtaining a robust andhigh order limiting procedure to simultaneously achieveuniform high order accuracy and sharp, non-oscillatoryshock transitions. The main advantage of this limiteris its simplicity in implementation, especially for multi-dimensional meshes.
Xinghui ZhongMichigan State [email protected]
MS267
Bayesian Inference with Processed Data Products
Abstract not available at time of publication.
Kenny ChowdharySandia National Laboratories
MS267
Probabilistic Schwarz Coupling for Fault-Toleranceand Scalability
In the drive towards exascale computing, simulation codesneed to scale effectively to ever more cores, and be resilientagainst faults. We discuss a novel probabilistic Schwarzcoupling approach, which employs a probabilistic represen-tation of the knowledge about the solution at subdomaininterfaces, and as such can naturally account for faults andother nondeterministic events. This probabilistic represen-tation is updated asynchronously with information fromindependent subdomain computations for sampled valuesof the interface distribution.
Bert J. DebusschereEnergy Transportation CenterSandia National Laboratories, Livermore [email protected]
Khachik Sargsyan, Cosmin Safta, Gilbert HendrySandia National [email protected], [email protected],[email protected]
Habib N. NajmSandia National LaboratoriesLivermore, CA, [email protected]
MS267
Stochastic Reduced Models
Abstract not available at time of publication.
Roger GhanemUniversity of Southern CaliforniaAerospace and Mechanical Engineering and [email protected]
MS267
Statistically Robust and Parallel Load BalancedSampling Algorithms for Bayesian Analysis andMultimodal Distributions
The Bayesian analysis of mathematical models often re-quire the evaluation of multidimensional integrals relatedto posterior probability density functions (PDFs) of un-certain model parameters. Such integrals rarely can becomputed analytically, which motivates the approximatecalculation of their values with stochastic simulation meth-ods that sample the corresponding posterior PDFs. In thistalk we will discuss a recently proposed method that is sta-tistically robust and parallel load-balanced. We will alsopresent some numerical results.
Ernesto E. PrudencioInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
Sai Hung CheungSchool of Civil and Environmental EngineeringNanyang Technological University, [email protected]
CS13 Abstracts 333
MS268
Adaptive Multi-resolution Solver for Multi-phaseFlows with Sharp Interface Model and EfficientData Structure
In this work, we present a block-based multi-resolutionsolver coupled with sharp interface model (MR-SIM)for multi-phase flows. While the solver updates theoverlapped-block system according to two separate proce-dures, i.e. tracking interface position and MR analyzingfor each individual phase, it uses a table-based storage-and-operation-splitting data structure. The data structureallows direct searches among blocks on the same or dif-ferent resolution levels with indexes, and splits the blockfrom its associated computational data. Since there is noextra memory required for the data associated with over-lap regions of the blocks, highly efficient memory usage andinter-block communication are achieved.
Luhui Han, Xiangyu Hu, Nikolaus AdamsTechnical University of [email protected], [email protected],[email protected]
MS268
Interface Capturing with High-order AccurateSchemes: Pressure and Temperature Considera-tions
Compressible multifluid simulations are challenging due tothe necessity to accurately represent discontinuities andtransport processes. Direct application of shock captur-ing may produce spurious pressure oscillations and havebeen shown to generate temperature errors. The presentfocus is on high-order accurate interface capturing for mul-tiphase flows. We show that, by appropriately transportingthe relevant parameters of the equation of state, pressure,temperature and conservation errors can be prevented ingas/gas and gas/liquid flows.
Eric JohnsenUniversity of [email protected]
MS268
Simple Interface Sharpening Technique with Hy-perbolic Tangent Function Applied to Compress-ible Two-Fluid Modeling
A very simple interface sharpening technique for compress-ible two fluids model is proposed. A main idea is using hy-perbolic tangent function for reconstruction of volume frac-tion in the formulation of compressible two fluids model,which is similar to THINC or MTHINC scheme recentlywell-discussed in the incompressible multi-phase computa-tion. This very simple technique sharpens the interfacevery much, although just the interpolation of volume frac-tion differs from conventional MUSCL scheme applied totwo-fluids model. Several examples are shown in the pre-sentation.
Taku NonomuraJAXA/Institute of Space and Astronautical [email protected]
Keiichi KitamuraNagoya [email protected]
Kozo FujiiJAXA/[email protected]
MS268
Eulerian Interface-sharpening Algorithms for Com-pressible Flow Problems
Our goal is to describe a novel Eulerian interface-sharpening approach for the efficient numerical resolutionof interfaces arising from inviscid compressible flow in morethan one space dimension. The algorithm uses the com-pressible Euler equations as the model system, and intro-duces auxiliary differential terms to the model so as to neu-tralize numerical diffusion that is inevitable when the orig-inal Euler system is solved by a diffused interface method.A standard fractional-step method is employed to solve theproposed model equations in two steps, yielding an easy im-plementation of the algorithm. Sample numerical resultsare shown to demonstrate the feasibility of the algorithmfor sharpening compressible interfaces numerically.
Keh-Ming ShyueDepartment of Mathematics, National Taiwan [email protected]
MS269
Variational Data Assimilation and Particle Filters
There is a subtle similarity between the 4D-Var and EnKFmethods for linear, Gaussian models, that can be used todesign a hybrid ensemble filter which performs better thanthe regular EnKF. This paper investigates the conjecture,that combining a variational method with particle filteringcan deliver information useful for improving the perfor-mance of particle filters.
Haiyan ChengWillamette [email protected]
MS269
Ensemble Data Assimilation Using An Unstruc-tured Adaptive Mesh Ocean Model
For an unstructured adaptive mesh ocean model, a super-mesh technology is applied to the adapted meshes of theensembles. Mesh adaptivity is also adopted around theobservation locations to improve the efficiency of the dataassimilation process and to address the key problem of therepresentativity of the observations in ocean data assimi-lation.
Juan DuInstitute of Atmospheric Physics, Chinese Academy [email protected]
Fangxin FangDepartment of Earth Science and EngineeringImperial College London, [email protected]
Jiang ZhuInstitute of Atmospheric PhysicsChinese Academy of Sciences,[email protected]
334 CS13 Abstracts
C.C. Pain, P.A. AllisonDepartment of Earth Science and EngineeringImperial College London, [email protected], [email protected]
MS269
POD/DEIM 4-D Var Data Assimilation Applied toa Nonliner Adaptive Mesh Model
A novel POD 4-D Var model has been developed for anolinear adaptive mesh model. For nonlinear problems, aperturbation approach is used to help accelerate the matrixequation assembly process. The discrete empirical interpo-lation method (DEIM) is further applied to the POD modeland achieves a complexity reduction of the high order non-linear term in the full model. A non-linear Petrov-Galerkinmethod is used for improving the stability of POD resultswithout tuning parameters.
Fangxin FangDepartment of Earth Science and EngineeringImperial College London, [email protected]
Dunhui XiaoDepartment of Earth Science and EngineeringImperial Colleg [email protected]
Christopher PainImperial [email protected]
Juan DuInstitute of Atmospheric Physics, Chinese Academy [email protected]
Ionel M. NavonFlorida State UniversityDepartment of Scientific [email protected]
MS269
Data Assimilation 2
Abstract not available at time of publication.
Ionel M. NavonFlorida State UniversityDepartment of Scientific [email protected]
MS270
Landslides Thresholds for Early-Warning Systems:History, Challenges and Perspectives
The earliest thresholds for landslides were derived by vi-sually fitting boundaries to a few triggering events usingground-based measured rainfall. Over the years, thresh-old estimation has faced increasing complexities: largerdatabases, real-time monitoring from both ground-basedand remote sensing products, and demand for optimumperformance for early-warning systems. The evolution ofthe state-of-the-art in landslide thresholds is presented, em-phasizing recent developments in data mining and proba-bilistic classification with case studies in Europe and Cen-
tral America.
Jose CepedaInternational Center for GeohazardsNorwegian Geotechnical Institute, Oslo [email protected]
Dalia KirschbaumHydrological Sciences LaboratoryNASA Goddard Space Flight Center, MD, [email protected]
Zenon Medina-CetinaZachry Department of Civil EngineeringTexas A&M [email protected]
MS270
Groundwater Flow, Poroelastic Waves and Lique-faction
Liquefaction is one of the less well quantified consequencesof earthquakes: engineering assessments of liquefaction riskoften boil down to back of a cigarette packet calculations.Uncertainties are generally not considered. However suchassessments provide a basis for land classification and foun-dation engineering. This talk summarises our attemptsto build rational models of liquefaction and hence bet-ter understand the mechanisms that lead to liquefaction.As a backdrop we consider the case of Christchurch, NewZealand.
Nicholas Dudley WardOtago Computational Modelling [email protected]
Jari KaipioDepartment of MathematicsUniversity of [email protected]
Tiangang [email protected]
MS270
Accelerating Numerical Modeling of Waves Prop-agating Through 2-D Anisotropic Materials Usinga Graphic Processing Card
We present an implementation and analysis of performanceover multi-threaded executions using graphic cards as par-allel computing devices to model numerically (using finitedifferences implemented in PyOpenCL) the behavior ofwaves propagating through 2-D anisotropic materials. Thisapproach is inspired by laboratory experiments where asample plate of different anisotropic materials inside a wa-ter tank is rotated and for every angle of rotation is sub-jected to the emission of an ultrasonic wave.
Ursula IturraranUniversidad Nacional Autonoma de Mexico [email protected]
Miguel MoleroCentro de Acustica Aplicada y Evaluacion No DestructivaCAEND (CSIC-UPM), Madrid, [email protected]
CS13 Abstracts 335
MS270
Diagnosis and Prognosis Analysis of EcologicalManagement under Varying Climate Change Sce-narios
Unsustainable management of rangeland where energy de-velopment systems take place leads to the degradation ofboth the resource base, the value of the commodities, andthe benefits these generate. Sustainable development andits inherent spatio-temporal complexities include interac-tions across social, ecological and economic paradigms, in-cluding climate change, considered a key triggering mech-anism for sudden interventions across these processes. ABayesian mapping and GIS tool is presented as applied toecoregions of shale gas development.
Patricia VarelaTexas A&M Universityvarela [email protected]
Zenon Medina-CetinaZachry Department of Civil EngineeringTexas A&M [email protected]
Bill FoxBlackland Research & Extension CenterTexas AgriLife - Research, TX [email protected]
Jay AngererBlackland Research & Extension CenterTexas AgriLife - Research, TX, [email protected]
Luis Alberto Munoz UbandoPlenum SoftMerida Yucatan, [email protected]
MS271
Exterior Calculus Stuff
Abstract not available at time of publication.
Anil N. HiraniUniversity of Illinois at Urbana-ChampaignDepartment of Computer [email protected]
MS271
Mimetic Finite Difference Method for the StokesEquations
We present two formulations of the mimetic finite differ-ence (MFD) method for the Stokes equations on arbitrarypolygonal meshes with convex and non-convex cells. Theformulations correspond to C0 and C1 continuity of thevelocity field. The numerical and theoretical analysis ofthe stability will be discussed in great details. We showhow flexibility of the MFD method can be used to enforcestability for problematic mesh configurations.
Lourenco Beirao Da VeigaUniversita’ degli Studi di [email protected]
Konstantin Lipnikov, Gianmarco Manzini
Los Alamos National [email protected], [email protected]
MS271
Is the Keller-Box Scheme Mimetic?
The Keller-Box scheme is a method for solving hyperbolicand parabolic PDEs that has been shown to have a numberof attractive physical and mathematical properties. It canhandle discontinuous material coefficients and works wellon distorted or high aspect ratio meshes. Most recentlythe Keller-Box scheme has been shown by Reich to multi-symplectic (2000) and by Frank (2006) to always propagatewaves in the correct direction. By deriving the scheme asa type of exact discretization we show that the differentialoperators in the Keller-Box scheme are mimetic. However,the mimetic gradient operator is shown to be fundamen-tally different from that found in mimetic FE (Nedelec)methods, or commonly used as the basis for SOM meth-ods. And the method is not easily described by algebraictopology or discrete differential forms. The advantages anddisadvantages of this unusual numerical method comparedto other mimetic schemes are discussed.
Blair PerotUniversity of Massachusetts - Amhersttba
MS271
Mimetic Methods Toolkit: An Object-OrientedApi Implementing Mimetic Discretization Methodswith Application Examples in Oil Reservoir Simu-lation.
In this work, we introduce Mimetic Methods Toolkit(MTK), an object-oriented Application Programming In-terface for the implementation of Mimetic DiscretizationMethods in developing computer applications of a scien-tific nature. We present examples on how can MTK beused to simulate pressure distribution in oil reservoirs. Wepresent the fundamentals of the mathematical models foroil reservoirs as porous media, and we present an introduc-tion to Mimetic Discretization Methods, while we discussthe advantages of using both Mimetic Discretization Meth-ods and MTK. Finally, we explain how MTK implementsthese methods, and we compare the results against well-known discretization methods.
Eduardo SanchezSan Diego State [email protected]
MS272
Efficient Nonlinear Model Reduction Approach us-ing Local Reduced bases and Hyper-reduction
A new model reduction approach for nonlinear dynamicalsystems developed. This method is based on three ingre-dients: 1) the definition of local reduced-order bases thatcan better capture the subspace containing the solution ina given regime than a global counterpart, 2) a local hyper-reduction technique resulting in an efficient algorithm and3) an efficient basis update procedure. Applications to thereduction of CFD-based dynamical systems highlight theeffectiveness of the method.
David Amsallem, Kyle WashabaughStanford [email protected], [email protected]
336 CS13 Abstracts
Matthew J. ZahrUniversity of California, BerkeleyStanford [email protected]
Charbel FarhatStanford [email protected]
MS272
Frequency-weighted H2-optimal Model Reduction
In this talk, we discuss different optimality conditions aris-ing in the context of frequency weighted H2-model reduc-tion for linear control systems. In particular, we establisha connection between a recently introduced distributed in-terpolation framework and a special class of linear matrixequations. We further provide a discussion on numericallyefficient methods to construct locally optimal reduced mod-els.
Tobias BreitenMax-Planck-Institute for Dynamics of Complex TechnicalSystems, [email protected]
Christopher A. BeattieVirginia Polytechnic Institute and State [email protected]
Serkan GugercinVirginia Tech.Department of [email protected]
MS272
Greedy Algorithms and Stable Variational Formu-lations
Greedy schemes for computing reduced bases for parame-ter dependent PDEs are typically based on efficiently com-putable surrogates for the actual current distance betweenthe reduced space and the solution set S. It can be shownthat the accuracy provided by the reduced spaces is “rate-optimal’, when compared with the Kolmogorov n-widths ofS, if the surrogate is tight. By this we mean that up to aconstant the surrogate is also a lower bound for the distanceof the reduced space from the solution set S. We show howto obtain such tight surrogates also for non-elliptic prob-lems such as, for instance, pure transport equations or sin-gularly perturbed unsymmetric problems by deriving vari-ational formulations that are in a certain sense uniformlystable and by properly stabilizing the reduced spaces. Theresults are illustrated by some numerical experiments.
Wolfgang Dahmen, Gerrit Welper, Christian PleskenRWTH [email protected], [email protected], [email protected]
MS272
Greedy Algorithms for Eigenvalue Problems
Greedy algorithms are promising algorithms to treat high-dimensional PDEs. Encouraging numerical results have al-ready been obtained in a large number of applications. Re-cently, theoretical convergence results have been obtained
for nonlinear convex optimization problems. The aim ofthis talk is to present new convergence results for this fam-ily of numerical methods regarding high-dimensional eigen-value linear problems, along with some numerical illustra-tions.
Virginie Ehrlacher
CERMICS - Ecole des Ponts Paristech / [email protected]
Tony LelievreEcole des Ponts [email protected]
Eric CancesEcole des Ponts and INRIA, [email protected]
MS273
Multi-fidelity Analysis and Optimization for Low-boom Supersonic Aircraft
Abstract not available at time of publication.
Juan J. AlonsoDepartment of Aeronautics and AstronauticsStanford [email protected]
MS273
Advanced Mathematical Techniques for NewSystems-level Analysis and Optimization
We seek to advance the science of systems integration bysystematically implementing sound scientific principles intothe engineering process, a “physics-based approach.’ Threeareas of recent progress will be highlighted in this pre-sentation. First, a generic dynamic stochastic model wasdeveloped that employs a methodology that predicts andcorrects the time-varying parameters of a dynamic sys-tem. This procedure was employed to reduce the growthof the prediction interval for design predictions and is ex-tensible to all dynamic stochastic mathematical models.Second, a formulation for calculating and measuring dy-namic entropy generation rate was developed that is con-sistent with computational and experimental determina-tions. This formulation allows for direct calculation oflosses and supports the physics-based modeling supportedby experimental validations. The importance of calculat-ing the entropy generation dynamically is illustrated in anumerical example. Third and last, the impact of sam-pling methodology on uncertainty propagation is detailedin a computational example. Psuedo-random sampling iscompared to low-discrepancy-sequence sampling to illus-trate numerically-induced uncertainties that influence un-certainty propagation and sensitivity analysis. Uncertaintyanalysis, uncertainty propagation and sensitivity analysisare all performed for a nonlinear system.
Jose A. CamberosAir Force Research LaboratoryMultidisciplinary Science & Technology [email protected]
John DotyAir Force Research Laboratory,Multidisciplinary Science & Technology [email protected]
CS13 Abstracts 337
Ray KolonayAir Force Research LaboratoryMultidisciplinary Science & Technology [email protected]
MS273
Multifidelity Approaches for Parallel Multidisci-plinary Optimization
This talk formulates two methods to parallelize the op-timization of multidisciplinary systems. The first methoddecomposes the system optimization problem into multiplesubsystem optimizations that are solved in parallel. Thesecond method generates a list of designs at which com-putationally expensive simulations should be run, evalu-ates those designs in parallel, and then solves an inexpen-sive surrogate-based optimization problem. Both methodsenable the use of multifidelity optimization and are high-fidelity gradient-optional, i.e. they exploit high-fidelitysensitivity information when available, but do not requiregradients of the high-fidelity models.
Andrew I. MarchMassachusetts Institute of [email protected]
MS273
Optimization Under Uncertainty Using ControlVariates
Accounting for uncertainty and variability is importantto design robust and reliable systems, but it is often toocomputationally expensive to nest Monte Carlo simulationwithin design optimization. The control variate methodcan take advantage of the correlation between randommodel outputs at two design points to reduce estimatorvariance. By chaining control variates through the se-quence of optimization design points, we can achieve sig-nificant computational savings. Numerical experimentsdemonstrate 80% reduction in model evaluations.
Leo Ng, Karen E. WillcoxMassachusetts Institute of Technologyleo [email protected], [email protected]
MS274
Accelerating MCMC with Local Quadratic Models
In many inference problems, the cost of MCMC analy-ses is dominated by repeated evaluations of expensive for-ward models. Surrogate methods attempt to exploit modelregularity by substituting an inexpensive approximationconstructed from a limited number of model runs. Theconstruction of globally accurate surrogates may be pro-hibitively expensive, however. We therefore propose to in-terleave MCMC with the construction of local quadraticsurrogate models, borrowing from trust-region methodsand approximation results in derivative free optimization.
Patrick R. [email protected]
Youssef M. MarzoukMassachusetts Institute of [email protected]
MS274
Projections on Positive Random Variables in FiniteWiener Chaos Spaces
Abstract not available at time of publication.
Evangelia KalligiannakiUniversity of Southern [email protected]
MS274
Preconditioned Bayesian Regression for StochasticChemical Kinetics
We develop a preconditioned Bayesian regression methodthat enables sparse polynomial chaos representations ofnoisy outputs for stochastic chemical systems with uncer-tain reaction rates. The approach is based on couplinga multiscale transformation of the state variables witha Bayesian regression formalism. Numerical experimentsshow that the approach accommodates large noise levelsand large variability with uncertain parameters, and en-ables efficient and robust recovery of both the transientdynamics and the corresponding noise levels.
Olivier P. Le [email protected]
Alen AlexanderianJohns Hopkins [email protected]
Francesco RizziDepartment of Mechanical EngineeringJohns Hopkins [email protected]
Muruhan RathinamUniversity of Maryland, Baltimore [email protected]
Omar M. KnioDuke [email protected]
MS274
Hybrid Discrete/Continuum Algorithms forStochastic Reaction Networks
Direct solutions Chemical Master Equation (CME) modelsgoverning Stochastic Reaction Networks (SRNs) are gen-erally prohibitively expensive due to excessive numbers ofpossible discrete states in such systems. To enable effi-ciency gains we develop a hybrid approach where stateswith low molecule counts are treated with the discreteCME model while states with large molecule counts aremodeled by the continuum Fokker-Planck equation. Theperformance of this novel hybrid approach is explored incanonical SRNs configurations.
Cosmin Safta, Khachik SargsyanSandia National [email protected], [email protected]
Bert J. DebusschereEnergy Transportation CenterSandia National Laboratories, Livermore CA
338 CS13 Abstracts
Habib N. NajmSandia National LaboratoriesLivermore, CA, [email protected]
PP1
Parameter Mesh Adaptivity for Ill-Posed InverseProblems Based on the Dominant Modes of theMisfit Hessian
In ill-posed parameter identification problems, the ob-served data typically provide different levels of informationabout the parameter in different parts of the domain. Inparticular, the informativeness of the data is reflected inthe eigenstructure of the misfit Hessian. Here, we presenta mesh adaptivity scheme in which the parameter mesh isrefined and coarsened to approximate the dominant (wellinformed) modes of the misfit Hessian.
Nick AlgerThe University of Texas at AustinCenter for Computational Geosciences and [email protected]
Tan Bui-ThanhThe University of Texas at [email protected]
PP1
Flexible Krylov Subspace Methods for Shifted Sys-tems
We discuss methods for solving shifted systems of the formK + σjM = b, j = 1, . . . , nf , where σj ∈ C using anArnoldi-based method, with flexible preconditioners of theform K + τM that are inverted using iterative solvers. Wedemonstrate this method with error estimates and numer-ical examples from Oscillatory Hydraulic Tomography.
Tania BakhosInstitute for Computational and MathematicalEngineeringStanford [email protected]
Arvind SaibabaStanford [email protected]
Peter K. KitanidisDept. of Civil and Environmental EngineeringStanford [email protected]
PP1
Spatiotemporal Dynamics of Cardiac Electrical Al-ternans
We study the behavior of cardiac alternans, a period-2 car-diac rhythm. During spatially discordant alternans, wavesgenerated from long action potentials at the stimulus sitebecome short as they propagate through the tissue and viceversa. Separating these regions are sites called nodes whereaction potentials remain identical from beat to beat. Weanalyze the dynamics of these nodes including their loca-tions and movement with and without spatial gradients in
electrophysiological properties.
Michael BellRochester Institute of [email protected]
Elizabeth M. CherryRochester Institute of TechnologySchool of Mathematical [email protected]
PP1
Mass-Conservative Adaptive Mesh UnrefinementComputation for Shallow Water Flow Simulations
The h-adaptive unstructured mesh refinement (h-AMR)has been widely used to improve mesh resolution locallyin order to achieve better quality of numerical simula-tions while sustaining limited computer power. In AMR,massconservation is often neglected when a node is simplydeleted during coarsening. A local approach to conservemass for unrefinement process is presented. Mathematicalderivation for a higher-order time derivative and its de-tailed implementation are described. Experimental resultsalso demonstrate our successful development.
Ruth ChengUS Army Corps of [email protected]
PP1
Overlapping Local/Global Iteration Scheme forWhole-Core Neutron Transport Calculation
In the overlapping local/global (OLG) iteration schemedescribed in this paper, local transport calculation andglobal diffusion-like calculation are coupled through in-terface boundary conditions. Local calculations are per-formed over half-assembly overlapping subregions since thisoverlapping takes into account the inter-assembly trans-port effect. The global calculation is performed via par-tial current-based coarse-mesh finite difference (p-CMFD)method to obtain the whole-core transport solution. Thistwo-level iterative method is tested on several multi-slaband two-dimensional rectangular geometry problems, withencouraging results.
Seungsu Yuk, Nam Zin ChoKorea Advanced Institute of Science and [email protected], [email protected]
PP1
A GPU-Accelerated Method of RegularizedStokeslets
We present a GPU-accelerated computational method forsimulating the mechanics of poroelastic materials. Theelastic material is described by a collection of point forces,and we develop a parallel implementation of the methodof regularized Stokeslets to solve for the fluid motion. Re-sults from our method are compared with a naive serialalgorithm for two and three spatial dimensions. We demon-strate the performance gain of our algorithm in a simula-tion of the dynamics of cellular bleb formation.
Calina A. CoposUniversity of California [email protected]
CS13 Abstracts 339
Robert D. GuyMathematics DepartmentUniversity of California [email protected]
Wanda StrychalskiDepartment of MathematicsUniversity of California, [email protected]
PP1
Sparse Matrix Operations on GPU Architectures
GPUs have been leveraged in numerous numerical appli-cations to achieve tremendous gains in performance. Al-though operations on dense matrices have been studied ex-tensively their sparse counterparts, with the exception ofSpMV, have developed relatively slowly with few tangibleimplementations. In this work we present work concerningseveral data dependent and irregular operations on sparsematrices on GPU architectures: sparse matrix-matrix mul-tiplication, graph contractions, and graph partitioning.
Steven DaltonUniversity of Illinois at [email protected]
PP1
Matrix Functions and the NAG Library
Functions of matrices are of growing interest in science andengineering due to the concise way they allow problems tobe formulated and solutions to be expressed. The Numeri-cal Linear Algebra Group at the University of Manchesterhas developed many algorithms for computing matrix func-tions. These are currently being implemented in the NAGSoftware Library. This poster will introduce some of thetechniques used in the algorithms and discuss some of theimplementation details.
Edvin DeadmanNumerical Algorithms Group and University [email protected]
PP1
Computational Tools for Digital Holographic Mi-croscopy
Digital holographic microscopy is in principle a fast 3Dimaging technique, but extracting precise 3D informationfrom the 2D holograms is challenging, especially for com-plex samples. A promising approach used for colloidal sys-tems involves fitting scattering models to the holograms.We demonstrate a related approach to imaging biologicalsamples with much more complex shapes and structures.Our method makes use of the discrete dipole approxima-tion and a new mathematical approach to solving inverseproblems.
Thomas G. DimidukHarvard UniversityManoharan [email protected]
Jerome Fung, Rebecca Perry, Vinothan ManoharanHarvard [email protected], [email protected],
PP1
Group Steiner Problem: An Ant Colony Optimiza-tion Based Solution and Practical Applications
The Group Steiner Problem (GSP) is an important gener-alization of some basic NP-hard problems. Many complexreal-world applications require solving the GSP in graphsmodeling the topology of the given problem, such as: thedesign of Very Large Scale Integration (VLSI), the design ofa minimal length irrigation network, and routing problemfor wireless sensor networks. In this talk, we first describetopology models of these applications solved by searchingthe minimum group Steiner tree. Next, we show our designof some new algorithms based on Ant Colony Optimizationmodel to solve the GSP in general graphs. Our experi-mental results show that our method outperforms the bestother heuristic methods for GSP.
Thuan P. [email protected]
Duong NguyenSchool of Information and Communication TechnologyHanoi University of Science and [email protected]
PP1
Block Conjugate Gradient Type Methods for theApproximation of Bilinear form CHA−1B
We consider approximating the bilinear form CHA−1B,where matrices A ∈ Cn×n, C and B ∈ Cn×m, usuallym � n. The common way is first to solve AX = B,linear systems with multiple right-hand sides, by blockKrylov subspace methods. Then, η := CHX can be ob-tained. Inspired by methods in [Strakos and Tichy, SISC,2011], we propose block conjugate gradient type methodsfor CHA−1B. Numerical results will be presented to showthe efficiency of our methods.
Lei DuUniversity of Tsukuba, [email protected]
Yasunori FutamuraUniversity of [email protected]
Tetsuya SakuraiDepartment of Computer ScienceUniversity of [email protected]
PP1
An Asymptotic-Based Numerical Algorithm forAnalyzing Nonlinear Waves
Resolving the complicated fluid-solid interactions of themammalian cochlea requires solving a nonlinear wave prob-lem; however, the geometric and material properties of theear make this difficult to do with numerics alone. In thiswork a hybrid analytic-numeric approach is used. Asymp-totic methods are utilized in conjunction with an iterativenumerical method to achieve an approximate solution to
340 CS13 Abstracts
the nonlinear wave problem.
Kimberly Fessel, Mark HolmesRensselaer Polytechnic [email protected], [email protected]
PP1
Scalable Methods for Large-Scale Bayesian InverseProblems
We address the challenge of large-scale nonlinear statisticalinverse problems by developing an adaptive Hessian-basedGaussian process response surface method to approximatethe posterior pdf solution. We employ an adaptive sam-pling strategy for exploring the parameter space efficientlyto find interpolation points and build a global analyticalresponse surface far less expensive to evaluate than theoriginal. The accuracy and efficiency of the response sur-face is demonstrated with examples, including a subsurfaceflow problem.
H. Pearl FlathInstitute for Computational Engineering and SciencesThe University of Texas at [email protected]
PP1
Parallel in Time Using Multigrid
One of the main challenges facing computational sciencewith future architectures is that faster compute speedswill be achieved through increased concurrency, since clockspeeds are no longer increasing. As a consequence, tradi-tional time marching will eventually become a huge sequen-tial bottleneck, and parallel in time integration methodsare necessary. This poster discusses optimality and par-allelization properties of one approach for parallelizationin time, namely a multigrid-in-time method that is fairlyunintrusive on existing codes.
Stephanie FriedhoffTufts [email protected]
Robert Falgout, Tzanio KolevLawrence Livermore National [email protected], [email protected]
Scott MaclachlanDepartment of MathematicsTufts [email protected]
Jacob SchroderLawrence Livermore National [email protected]
PP1
A Fully Implicit Newton-Krylov Method for EulerEquations
Fully implicit methods for hyperbolic PDEs are relativelyrare. We present a fully nonlinearly implicit in timemethod for the Euler equations of gas dynamics based ona Newton-Krylov approach. Numerical examples are pre-sented for one-dimensional smooth linear wave propoga-tion, and for the Riemann shock-tube problems with dis-continuities. Numerical tests with CFL number as large as
Order(103) demonstrate that the computed solutions arein good agreement with exact solutions. Preconditioningstrategies will be presented.
Wei GaoKing Abdullah University of Science and [email protected]
Ravi [email protected]
PP1
Lighthouse Taxonomy: Delivering Linear AlgebraSolutions
While scientists use linear algebra in a wide variety of appli-cations, they typically lack the training required to develophigh-performance implementations. We narrow this gapwith a taxonomy entitled Lighthouse, which steers practi-tioners from algorithmic descriptions to high-performanceimplementations. Lighthouse unifies a routine database,search capabilities, and code generation and tuning. Fortuning, Lighthouse employs the Build To Order (BTO)compiler that reliably achieves high performance for lin-ear algebra via a variety of optimization techniques.
Paul L. Givens, David Johnson, Javed HossainCU - [email protected], [email protected],[email protected]
Sa-Lin BernsteinComputation Institute, University of [email protected]
Elizabeth JessupCU - [email protected]
Boyanna NorrisArgonne National [email protected]
PP1
On the Relationship between Polynomial ChaosExpansions and Gaussian Process Regression
Gaussian process regression (GPR) and polynomial chaosexpansions (PCE) are often used as surrogates to approxi-mate the outputs of complex computational simulations—for statistical inference, uncertainty propagation, and othertasks. We use Mercer’s theorem to relate the choice of ker-nel and training points in GPR to the choice of polynomialbasis and quadrature scheme in PCE. This relationship isused to show the conditions under which the mean predic-tion of a Gaussian process is equivalent to a polynomialchaos expansion.
Alex A. GorodetskyMassachussets Institute of [email protected]
Tarek [email protected]
Youssef M. Marzouk
CS13 Abstracts 341
Massachusetts Institute of [email protected]
PP1
Effects of Abruptly Reversing Shear Flow on RedBlood Cells
Blood pumps and other cardiovascular devices subject redblood cells to fluctuating shear flow, which leads to shearstresses well beyond their normal physiological values andmay cause hemolysis. Using the lattice Boltzmann andImmersed Boundary methods, we model the red blood cellas a viscoelastic biconcave capsule. The capsule dynam-ics and shape change under abruptly reversing shear floware considered, particularly with respect to dependence onpeak shear stress and capsule characteristics (e.g., bendingstiffness and membrane viscosity).
John GounleyOld Dominion [email protected]
Yan PengDept of Math. & Stat.Old Dominion [email protected]
PP1
Inverse Sensitivity Analysis for Storm Surge Mod-els
We describe a framework for quantifying uncertainty inspatially varying parameters critical to the accuracy ofmodel forecasts of storm surge. We characterize uncer-tainty in Manning’s n (a bottom friction parameter) us-ing an inverse sensitivity analysis applied to the ADCIRCstorm surge model. We model bottom friction as a ran-dom field, and use a Karhunen-Loeve expansion to obtaina finite dimensional approximation. The inverse sensitiv-ity analysis is based on a measure-theoretic approach toinversion.
Lindley GrahamThe University of Austin at TexasInstitute for Computational Engineering and Sciences(ICES)[email protected]
Clint DawsonInstitute for Computational Engineering and SciencesUniversity of Texas at [email protected]
Troy Butler, Don EstepColorado State [email protected], [email protected]
Joannes WesterinkDepartment of Civil Engineering and Geological SciencesUniversity of Notre [email protected]
PP1
A Multigrid Algorithm for Elliptic Type ProblemsUsing the Hierarchical Element Structure
We present a multigrid algorithm for elliptic type problems,where the projection / restriction operators are built using
the hierarchical element structure, in which the basis ofdegree n+1 is obtained as a correction to the degree nbasis. This with respect to a geometric multigrid avoidsthe complication of building recursive grid structures. Thealgorithm is implemented in 2 and 3 dimensions, and willbe extended to more complicated physics, including Navier-Stokes and FSI.
Janitha GunatilakeTexas Tech [email protected]
Eugenio AulisaDepartment of Mathematics and Statistics.Texas Tech [email protected]
PP1
A Bayesian Framework for Uncertainty Quantifica-tion in the Design of Complex Systems
This poster presents a Bayesian framework for the design ofcomplex systems, in which uncertainty in various parame-ters is characterized probabilistically, and updated throughsuccessive design iterations as new estimates become avail-able. Incorporated in the model are methods to quantifysystem complexity and risk, and reduce them through theallocation of resources for redesign and refinement. Thisapproach enables the rigorous quantification and manage-ment of uncertainty, thereby serving to help mitigate tech-nical and programmatic risk.
Qinxian He, Douglas Allaire, John Deyst, Karen E.WillcoxMassachusetts Institute of [email protected], [email protected], [email protected], [email protected]
PP1
Practical Experience with Gaussian Processes inQuantification of Margins and Uncertainties
Quantification of margins and uncertainties is a process bywhich performance and safety of engineered systems areassessed. Computational simulation, combined with ex-perimental testing, plays a key role in this type of analy-sis. However, full system models can be too computation-ally intensive to fully explore the relevant uncertainties,so emulators such as Gaussian processes are often used.We describe our successes and challenges using Gaussianprocesses in assessing margins and uncertainties for an en-gineered system.
Patricia D. Hough, Jeff CrowellSandia National [email protected], [email protected]
Laura SwilerSandia National LaboratoriesAlbuquerque, New Mexico [email protected]
PP1
Adaptive Time Stepping in Moose
The INL’s Multi-physics Object Oriented Simulation En-vironment (MOOSE) had the ability to integrate over timeusing backwards Euler, BDF2, and Crank-Nicolson. Vari-
342 CS13 Abstracts
able time step selection was highly limted. I have addedin an AB2 predictor according to ”Philip Gresho, DavidGriffiths, and David Silvester, Adaptive time-stepping forincompressible flow. Part 1: scalar advection-diffusion”for both Crank-Nicolson and BDF2. I am also adding invariable step-variable order time integration according.
John T. HutchinsBoise State University‘[email protected]
Michael PerniceIdaho National [email protected]
Donna CalhounBoise State [email protected]
PP1
A 3D Pharmacophore-Based Scoring Method forDOCK: Application to HIVgp41
Pharmacophore modeling incorporates geometrical andchemical features of either know inhibitors(s) or the tar-geted binding site in order to rationally design new drugleads. This presentation describes our efforts to encodea three dimensional pharmacophore matching similarity(FMS) scoring function into the program DOCK for usein virtual screening and de novo design. Application tothe drug target HIVgp41 using binding profiles for knownpeptide inhibitors to generate reference pharmacophoreswill be shown.
Lingling JiangRizzo Research Group, Stony Brook [email protected]
Robert RizzoDepartment of Applied Math. & Stat., Stony BrookUniversityLaufer Center for Physical & Quantitative [email protected]
PP1
Optimal Dirichlet Boundary Control for theNavier-Stokes Equations
We consider an optimal Dirichlet boundary control prob-lem for the Navier–Stokes equations. The control is consid-ered in the energy space where the related norm is realizedby the so-called Steklov–Poincare operator. We introducea stabilized finite element method for the optimal controlproblem with focus on lowest order elements. At the end,we present some numerical results demonstrating the dif-ferences between the realization of the control in L2(Γ) and
the energy space H1/2(Γ).
Lorenz JohnInstitute of Computational MathematicsGraz University of [email protected]
Olaf SteinbachTU Graz, Institute of Computational [email protected]
PP1
A Deformed Spectral Quadrilateral Multi-DomainPenalty Model for the Incompressible Navier-Stokes Equations
A penalty method is a variant of a spectral element methodthat weakly enforces continuity between adjacent elementsand weakly enforces continuity at physical boundaries.Furthermore, at the boundaries, the PDE is also partiallysatisfied. The spirit of such a formulation is that in the-ory, a PDE operates arbitrarily close to any measure-zeroboundary. Here, a previous spectral multi-domain penaltymodel for the incompressible Navier-Stokes equations is ex-tended to include deformed boundaries for shoaling-typeproblems encountered in Environmental Fluid Mechanics.Some difficulties addressed include satisfying compatibil-ity conditions in a (pseudo-)pressure Poisson equation thatarises. A previous strategy to satisfy compatibility by useof a null singular vector is presented, and strategies to en-force compatibility for the deformed problem are discussed.Results are shown for standard incompressible flow bench-marks. The primary goal of this work is to model non-linear internal wave propagation along a shallow, slopingbathymetry, as may be characteristic of a continental shelfregion.
Sumedh JoshiCenter for Applied MathematicsCornell [email protected]
Peter DiamessisDepartment of Civil and Environmental EngineeringCornell [email protected]
PP1
A Bayesian Approach to Feed Reconstruction inChemical Processes
We develop a fully Bayesian hierarchical model to esti-mate the detailed chemical composition of a stream ina petroleum refinery from limited measurements (e.g., ofbulk properties and elemental composition). Complexprior information, coupled with positivity and normaliza-tion constraints on the composition parameters, make itdifficult to sample from the resulting high-dimensional pos-terior distribution. We propose a method to efficiently gen-erate posterior samples and validate the method on exam-ple data sets.
Naveen Kartik, Youssef M. MarzoukMassachusetts Institute of [email protected], [email protected]
PP1
Nonlinear Dynamic State Reconstruction WithApplications to Exposure-Based (Bio)ChemicalHazard Assessment
A new approach to exposure-based (bio)chemical hazardassessment is proposed through a nonlinear dynamic statereconstruction method. The formulation of the nonlinearstate estimator problem is realized via a system of invari-ance nonhomogeneous functional equations, a general setof necessary and sufficient conditions for solvability is de-rived and an easily programmable series solution method isdeveloped. Finally, the performance of the proposed state
CS13 Abstracts 343
estimator is evaluated in an illustrative case study.
Nikolaos KazantzisDepartment of Chemical EngineeringWorcester Polytechnic [email protected]
PP1
A Hybrid CPU-GPU Approach to Fourier-BasedImage Stitching
We present a hybrid CPU-GPU system for the Fourier-based stitching of 2D optical microscopy images. The sys-tem uses CPU and GPU resources to achieve interactivestitching rates on large problems. It stitches a grid of59 × 42 tiles in 29.5 s. This is a 42x speedup over anoptimized sequential implementation which takes 20.5 minfor the same workload. This is a major step towards com-putationally steerable experiments in Biology that rely onautomated optical microscopes.
Walid Keyrouz, Bertrand Stivalet, Joe Chalfoun, MaryBradyNational Institute of Standards and [email protected], bertrand,[email protected],[email protected], [email protected]
Timothy Blattner, Shujia ZhouComputer ScienceUniversity of Maryland Baltimore [email protected], [email protected]
PP1
Non-Asymptotic Confidence Regions for Model Pa-rameters of a Hybrid Dynamical System
In this paper, a non-asymptotic method is introduced forevaluating the uncertainties of model parameters of a classof hybrid dynamical systems. The new algorithm for thispurpose is based on the Leave-out Sign-dominant Corre-lation Regions (LSCR) algorithm. This method has beenknown for generating non-conservative confidence regionsof model parameters in system identification using onlya finite number of data points. In order to extend therange of system types to which the LSCR method can beapplied, a simple hybrid dynamical system with uncertainparameters is considered, and the effectiveness of the LSCRmethod is demonstrated using a simulated finite number ofdata points obtained from the system.
Sangho KoSchool of Aerospace and Mechanical EngineeringKorea Aerospace [email protected]
PP1
Mathematical Analysis of Three-Dimensional OpenWater Maneuverability by Mantas MantaBirostris
For aquatic animals, turning maneuvers may not be con-fined to a single coordinate plane, making analysis difficultparticularly in the field. To measure turning performancefor the manta ray, a large open-water swimmer, scaledstereo video recordings (30 fr/s) were collected around Yap,Micronesia. Movements of the cephalic lobes, eye and tailbase were tracked to obtain three-dimensional coordinates.A mathematical analysis was performed on the coordinatedata to calculate the turning rate and curvature (1/turning
radius) by numerically estimating the derivative. Tikhonovregularization was used involving the minimization of acost function with a parameter controlling the balance be-tween data fidelity and regularity of the derivative. Theapproach gives a more accurate estimate of the derivativethan conventional finite-difference schemes which can am-plify noise leading to erroneous results. Data for 30 se-quences of rays performing slow, steady turns showed amedian turning rate of 46.48 deg/s with a median turningradius of 0.39 body lengths. Such turning maneuvers fallwithin the range of performance exhibited by swimmerswith rigid bodies.
Allison KolpasDepartment of Mathematical SciencesUniversity of [email protected]
Frank FishDepartment of BiologyWest Chester University of [email protected]
Alex MeadeDepartment of MathematicsWest Chester University of [email protected]
Michael DudasDudas’ Diving DudsWest Chester, [email protected]
Keith MooredMechanical EngineeringPrinceton [email protected]
PP1
Lqr Optimal Control for a Thermal-Fluid Dynam-ics
In recent years, considerable attention has been devoted toenergy efficient buildings and controlling the thermal-fluiddynamics therein. In this work, we investigate the feasi-bility of POD for the Linear Quadratic Regulator Problem(LQR) of a coupled PDE. However, since the dynamics ofa system (e.g. viscosity of air) is subject to change dur-ing a given time period, we are particularly interested inthe sensitivity of the model to a change in the underlyingReynolds number.
Boris KraemerVirginia [email protected]
John A. BurnsVirginia TechInterdisciplinary Center for Applied [email protected]
PP1
Matrix Interpolation Reduced Order Modeling ForMicrostructure Design
Matrix Interpolation Reduced Order Modeling (MIROM)uses radial basis functions to interpolate reduced matrices,rather than computing them directly as in proper orthogo-
344 CS13 Abstracts
nal decomposition. The resulting interpolation allows for acomputationally cheaper reduced order model evaluation.This work will present results from applying MIROM to amicrostructure design problem, where the design parame-ters are the microstructure dimensions and the outputs ofinterest are the coefficient of thermal expansion and theYoung’s modulus.
Kyle Lange, Dan White, Mark L. StowellLawrence Livermore National [email protected], [email protected], [email protected]
PP1
Large-Scale Stochastic Linear Inversion Using Hi-erarchical Matrices
Large scale inverse problems, which frequently arise inearth-science applications, involve estimating unknownsfrom sparse data. The goal is to evaluate the best estimate,quantify the uncertainty in the solution, and obtain condi-tional realizations, which are computationally intractableusing conventional methods. In this talk, I will discuss thehierarchical matrix approach optimized for a realistic largescale stochastic inverse problem arising from a cross-wellseismic tomography application.
Judith Yue LiStanford [email protected]
Sivaram AmbikasaranInstitute for Computational and MathematicalEngineeringStanford [email protected]
Peter K KitanidisStanford [email protected]
Eric F. DarveStanford UniversityMechanical Engineering [email protected]
PP1
Crosslinks: Connecting Topics Across Mathemat-ics and Engineering Curricula
Crosslinks is a tool to discover and track the relationshipsbetween core concepts in mathematics and the engineer-ing topics that follow. Its primary purpose is to improveknowledge transfer through use by students and facultyalike. The greatest learning moments occur when studentsdiscover connections between topics that once felt distinct;Crosslinks is intended to elicit and record those discoveries.
Chad E. [email protected]
Karen E. WillcoxMassachusetts Institute of [email protected]
Haynes MillerMIT
PP1
Mathematical Modeling with Sensor Data
We present our work on the development of mathematicalmodels used in conjunction with real-time sensor data tosimulate certain physical processes, for example, air flowand heat transfer. In particular, we consider the use ofboth physics-based models (in the form of boundary valueproblems) and data-driven statistical models within a datacenter energy management system and present results fromsimulations conducted as part of case studies.
Vanessa Lopez-Marrero, Hendrik HamannIBM T. J. Watson Research [email protected], [email protected]
Huijing JiangIBM T.J. Watson Research [email protected]
Xinwei DengVirginia [email protected]
PP1
Bovine Lameness Detection Via MultidimensionalTime-Series Force Data
Bovine lameness is a serious humanitarian concern and acostly problem for the US dairy industry. A mechanicallameness detection system developed at UMBC has beencontinuously collecting three dimensional force data from aherd of 750 dairy cows since 2012. To detect lameness, wereduce the raw data to a suitable heuristic representationand then train a hierarchical logistic model on scores froma veterinary examination. Our method achieves sensitivityand specificity above 85 percent.
Jonathan S. McHenryDepartment of Mathematics and [email protected]
Nagaraj NeerchalDepartment of Mathematics and StatisticsUniversity of Maryland, Baltimore [email protected]
Uri Tasch, Jason DunthornUniversity of Maryland, Baltimore [email protected], [email protected]
Robert DyerUniversity of [email protected]
PP1
Matrix Kronecker Products for Easy NumericalImplementation of PDEs and BCs in 2D
We illustrate the use of matrix Kronecker products to for-mulate 2D PDEs involving the laplacian or biharmonic op-erators on a rectangular domain with arbitrary boundaryconditions. For Poisson-type equations, an iterative multi-grid method is developed to solve the resulting Sylvester
CS13 Abstracts 345
system directly. For biharmonic equations that requiretwo boundary conditions on each edge, we explain how touse matrix Kronecker products to implement the boundaryconditions easily in the matrix formulation.
Ruben Glueck, Michael FranklinClaremont Graduate [email protected], [email protected]
Ali NadimClaremont Graduate UniversityDepartment of [email protected]
PP1
Random Ordinary Differential Equations for Multi-Storey Buildings
We investigate the numerical properties of modern meth-ods for the simulation of random differential equations andtheir performance in non-tivial applied settings. In par-ticular, the averaged Euler, the averaged Heun, and a K-RODE-Taylor scheme are used for ground-motion-inducedexcitation of multi-storey buildings subject to the Kanai-Tajimi earthquake model. The generalisation of our ap-proach to random partial differential equations as well as anefficient vectorised implementation on different new CPUarchitectures are subject to current work.
Tobias Neckel, Alfredo Parra, Florian RuppTechnische Universitat [email protected], [email protected],[email protected]
PP1
Boundary Integral Equations for Waves in LinearlyGraded Media
Boundary integral equations (BIEs) are a popular methodfor numerical solution of Helmholtz boundary value prob-lems in a piecewise-uniform medium. We generalize BIEsfor the first time to a continuously-graded medium in twodimensions, where the square of the wavenumber varieslinearly with one coordinate. Applications include opticsand acoustics in thermal/density/index gradients. We usecontour integrals to rapidly evaluate the new fundamentalsolution, and give examples of interior and high-frequencyscattering problems.
Brad Nelson, Alexander H. Barnett, Matt MahoneyDartmouth [email protected], [email protected],[email protected]
PP1
Multiphysical Coupled Problems for Vehicle Simu-lation
For engineering application in controller development forvehicles it is advantageous to be able to simulate efficientlycoupled problems arising in a vehicle efficiently. Thoseproblems are typically from multiphysical domains. Weshow problems, formulation and calculations for couplingof the mechanical system with 1D gasdynamic of the en-gine, cooling circuits, hybrid components and other do-mains needed for vehicle simulation.
Ralf U. PfauIMCC
MathConsult [email protected]
Roman [email protected]
PP1
A Library of Tensors with Order-Oblivious Index-ing: Fast Prototyping and Vectorization of Inter-preted Scientific Codes
Interpreted languages with multi-array and vectorizationsupport are widely used to fast prototype scientific codes.However, vectorized codes are difficult to write, debug,maintain, and optimize. First, we use labeled multi-arrayindices to help write, debug, and maintain scientific codes.Moreover, index positions are dynamic (order-oblivious)and can be optimized a posteriori to improve performance.Finally, we incorporate tensor products and contractions toenhance the applicability of the presented MATLAB andPython library.
Francisco J. Roca, David Moro, Ngoc Cuong Nguyen,Jaime PeraireMassachusetts Institute of [email protected], [email protected], [email protected],[email protected]
PP1
Efficient Solution of the Optimization Problem inModel-Reduced Gradient-Based History Matching
We present preliminary results of a performance evalua-tion study of several gradient-based state-of-the-art opti-mization methods for solving the nonlinear minimizationproblem arising in model-reduced gradient-based historymatching. The issues discussed in this work also apply toother problems, such as production optimization in closed-loop reservoir management.
Marielba Rojas, Slawomir SzklarzDelft University of [email protected], [email protected]
Malgorzata KaletaShell Global Solutions [email protected]
PP1
Adaptive Simulation of Global Mantle Flows
I present adaptive discretization methods and efficient par-allel solvers for the nonlinear Stokes systems arising inglobal mantle flow. The nonlinear Stokes equations are dis-cretized using high-order elements that are discontinuousfor the pressure. The block preconditioner for the Kryloviterations employs a BFBT approximation of the pressureSchur complement and an algebraic multigrid approxima-tion of the viscous operator. Local mesh refinement en-ables resolution of localized features while keeping the sizeof the resulting algebraic systems amenable for solution oncontemporary supercomputers.
Johann RudiICESThe University of Texas at Austin
346 CS13 Abstracts
Carsten BursteddeInstitut fuer Numerische SimulationUniversitaet [email protected]
Omar GhattasUniversity of Texas at [email protected]
Michael [email protected]
Toby IsaacICESThe University of Texas at [email protected]
Georg StadlerUniversity of Texas at [email protected]
Hari SundarInstitute for Computational and Engineering SciencesUniversity of Texas at [email protected]
PP1
A Reduced Basis Method for the Design of Meta-materials Through Optimization
Wave propagation through heterogeneous materials is oftenunintuitive and leads to unique phenomena. The design ofsuch materials is therefore of big interest. The state-of-the-art numerical simulation tools are used for the simu-lation. Hybrid DG methods provide accurate and efficientsolutions but design is only possible if NP-hard binary-optimization problems are formulated. A reduced-basismethod with heuristic discrete programming techniques isintroduced. Applications of this method for cloaking andother wave problems will be presented.
Joel [email protected]
Cuong Ngoc Nguyen, Abby [email protected], [email protected]
Robert M. FreundMITSloan School of [email protected]
Jaume [email protected]
PP1
High-Performance Computing in Simulating Car-bon Dioxide Geologic Sequestration
In Carbon Sequestration, codes that simulate water-rockinteraction and reactive transport sequentially solve mass
balance equations for each control volume representing alithology containing charged aqueous solutes. This is notwell suited for execution on many-core computers. Wepresent the theory and implementation of a numericalscheme whereby solute concentrations in all control vol-umes are solved simultaneously by constructing a largeblock-banded matrix. These matrices are factored withSuperLU DIST (Berkeley Laboratory). Performance met-rics are evaluated.
Eduardo SanchezSan Diego State [email protected]
PP1
Accelerator-Enabled Distributed-Memory Imple-mentation of the Icon Dynamical Core
We present first results from the accelerator-enableddistributed-memory Icosahedral Non-hydrostatic (ICON)dynamical core, which will be part of the ICON climatemodel currently under development at the Max PlanckInstitute for Meteorology. The accelerator implementa-tion utilizes the evolving OpenACC standard for direc-tives augmenting existing Fortran compilers, such as Crayand PGI. An initial single-node prototype implementationhalved time-to-solution on NVIDIA M2090 with respect toIntel Sandybridge for high resolution runs.
William Sawyer
Swiss Centre of Scientific Computing (CSCS)Manno, [email protected]
PP1
Blocking Symmetric Tensors
We introduce a new method of storing symmetric tensors(m-dimensional arrays) based on blocking in linear alge-bra. With this blocked storage scheme, we devise a blockedalgorithm for the ttsm (tensor times same matrix) in allmodes operation based on insights gained from linear al-gebra. The C = BABT operation is the matrix-equivalentof the ttsm in all modes operation and is used as the basisfor our insights.
Martin D. SchatzDepartment of Computer SciencesThe University of Texas at [email protected]
Tamara G. KoldaSandia National [email protected]
PP1
Multi-Scale Atmospheric Chemical TransportModeling with Wavelet-Based Adaptive Mesh Re-finement (WAMR) Numerical Method
Application of non-adaptive numerical techniques for mod-eling of multi-scale atmospheric chemical transport oftenresults in significant numerical errors due to poor spatialand temporal resolution. Here we present an adaptivemulti-scale WAMR method applied to numerical simula-tion of transpacific pollution plume transport. It is shownthat multilevel numerical grid efficiently adapts to the nu-merical solution development and the algorithm accurately
CS13 Abstracts 347
reproduces the plume dynamics at a reasonable compu-tational cost unlike conventional non-adaptive numericalmethods.
Yevgenii RastigejevHarvard Universityye [email protected]
Artem N. SemakinNorth Carolina A&T State [email protected]
PP1
Fluctuating Lipid Bilayer Membranes with Diffus-ing Protein Inclusions: Hybrid Continuum-ParticleNumerical Methods
Many proteins through their geometry and specific interac-tions with lipids induce changes in local membrane materialproperties. To study such phenomena we introduce a newhybrid continuum-particle description for the membrane-protein system that incorporates protein interactions, hy-drodynamic coupling, and thermal fluctuations. We inves-tigate how collective protein effects influence membranemechanical properties. We discuss interesting numericalaspects that are required to obtain good translation in-variance. Finally, we discuss a coarse grained model thatincorporates important hydrodynamics.
Jon Karl SigurdssonDepartment of MathematicsUniversity of California, Santa [email protected]
Paul J. AtzbergerUniversity of California-Santa [email protected]
PP1
Preconditioning Techniques for Stochastic Conser-vation Laws
We derive a preconditioning technique for a class ofstochastic conservation laws. The transformation, basedon a space-time stretching, either pushes the solution fieldinto a low rank manifold or expresses it as a low rank per-turbation of a reference solution. Both cases allow efficienttime integration through reduced basis methods or the gen-eralized spectral decomposition, together with a substan-tial reduction in storage and computational burdens. Thetechnique is particularly suited to long time integrationproblems where polynomial chaos approaches typically fail.An extension to a more general class of SPDEs is also pro-vided.
Alessio SpantiniMassachusetts Institute of [email protected]
Lionel MathelinLIMSI - [email protected]
Youssef M. MarzoukMassachusetts Institute of [email protected]
PP1
Balanced Splitting Methods for MultidimensionalSystems
Splitting methods are frequently used when solving largeODE systems such as those arising in reacting flow simu-lations. For some systems, standard splitting methods canintroduce large steady-state errors. We introduce a newmethod, balanced splitting, which eliminates the steadystate error by adding a balancing constant to each of thesplit terms. We analyze the method’s stability and presentexamples based on typical reacting flow problems.
Raymond L. [email protected]
William H. GreenMIT Chemical [email protected]
PP1
Referenceless Magnetic Resonance TemperatureImaging Approaches
Online temperature monitoring is mandatory for safe andsuccessful thermal therapy treatments of patients. Tworeferenceless magnetic resonance temperature imaging ap-proaches are compared: a l1-minimization edge detectionapproach and a PDE solution approach. Both methodshave high computational costs due to real time require-ments. The robustness of the PDE based approach is vali-dated using polynomial chaos (DAKOTA framework) withadditional high computational costs. Distributed comput-ing and GPU implementations are employed to find solu-tions efficiently.
Wolfgang Stefan, Florian MaierThe University of Texas MD [email protected], [email protected]
Jason Stafford, John HazleMD Anderson Cancer [email protected], [email protected]
PP1
An Adaptive Simplex Cut-cell Method for High-order Discontinuous Galerkin Discretizations of El-liptic Interface Problems
We present a new approach for high-order discretizationsof elliptic interface problems on unfitted meshes. Theapproach consists of a discontinuous Galerkin (DG) dis-cretization and a simplex cut-cell technique. We show thatno modification on DG bilinear form is needed for interfacetreatment. For irregularly shaped interfaces, we combineour strategy with an adaptive scheme to control the effectof geometry-induced singularities. High-order convergenceis demonstrated for elliptic interface problems with regularand irregular interfaces.
Huafei [email protected]
PP1
Stochastic Eulerian-Lagrangian Method with Ther-mal Fluctuations for Fluid-Structure Interactions
348 CS13 Abstracts
with Strong Coupling
We model a fluid-structure system subject to thermal fluc-tuations involving both Eulerian and Lagrangian referenceframes. Although this description arises rather naturally, itpresents both analytic and computational challenges. Wetreat the central issue of coupling between the two frames.Specifically, we simplify the viscous coupling between theimmersed structures and the fluid in the regime of strongcoupling by utilizing an asymptotic reduction on the in-finitesimal generator of the stochastic differential equa-tions.
Gil J. Tabak, Paul AtzbergerUniversity of California at Santa [email protected], [email protected]
PP1
Sport - An Effective Algorithm Towards ImprovingBayesian Network Structure Learning
The Bayesian Network (BN) is a very powerful tool forcausal relationship modeling and probabilistic reasoning.It facilitates deep understanding of very complex, high-dimensional problem domains. A key process of the build-ing the BN is identifying its structure – a directed acyclicgraph (DAG). In the literature, researchers proposed overfifty algorithms to discover the BN structure from data.The recent emerging of many BN-Structure learning algo-rithms (BN-SLAs) calls for generic techniques capable ofimproving BN structures learned by different algorithms.However, currently, these types of generic improvementtechniques are lacking. This study proposes a novel three-phased algorithm called SPORT (Score-based Partial Or-der RefinemenT). Through three phases: Pruning, Thick-ening and Correcting, SPORT leverages Bayesian scorefunction to iteratively simplify and improve BN struc-tures. It is generic and pluggable to different BN-SLAswith marginal computational overhead. Empirical studyapplies SPORT to the BN structures learned by four ma-jor BN-SLAs: PC, TPDA, OR and MMHC. Improvementsup to 64.4 % are observed among all the learned structures,confirming the effectiveness of SPORT towards improvingBN structural learning.
Yan TangHohai [email protected]
Kendra CooperUniversity of Texas at [email protected]
DeShan TangHohai [email protected]
PP1
Simplifying Chemical Kinetic Systems under Un-certainty using Markov Chains
We propose a new approach to the simplification of chem-ical kinetic systems, particularly suited to systems withlarge uncertainties. The method measures the probabilitythat elimination of a chemical species will influence the mi-crostates of the system. We use a Markov process to modelthe transfer of atoms from one molecule to another via el-ementary reactions, and show that the absorption prop-erties of the Markov chain identify chemical species to be
eliminated. Preliminary results will be shown for the com-bustion of methane.
Luca Tosatto, Youssef M. MarzoukMassachusetts Institute of [email protected], [email protected]
PP1
A Parallelized Model Reduction Library: Modred
We present modred, a Python library for model reduc-tion, modal analysis, and system identification of largesystems and datasets. It is parallelized for distributed-memory computing and has a comprehensive suite of au-tomated tests. Its modular design allows it to interfacewith arbitrary data formats. The algorithms implementedinclude the Proper Orthogonal Decomposition (POD),Balanced POD (BPOD), Dynamic Mode Decomposition(DMD), Eigensystem Realization Algorithm (ERA), andObserver/Kalman Filter Identification (OKID).
Jonathan Tu, Brandt BelsonPrinceton [email protected], [email protected]
Clarence RowleyPrinceton UniversityDepartment of Mechanical and Aerospace [email protected]
PP1
An Estimation Theory Approach to Decision Un-der Uncertainty with Application to Wind FarmSiting
In this poster we present a methodology for the quantifica-tion and systematic reduction of risk in the design and de-velopment of complex systems. Our methodology exploitsBayesian estimation theory to track the evolution of proba-bility distributions of critical parameters that characterizethe risk in the development process, as well as sensitivityanalysis techniques to reduce these uncertainties via effi-cient management of resources. The development of themethodology is demonstrated on the decision of whetheror not to site a wind farm.
Fatma D. Ulker, Douglas L. Allaire, John J. Deyst, KarenE. WillcoxMassachusetts Institute of [email protected], [email protected], [email protected], [email protected]
PP1
Updating Singular Subspaces for Latent SemanticIndexing
Latent Semantic Indexing (LSI) is a popular technique forintelligent information retrieval. Computationally, the ap-proach is based on finding dominant singular vectors ofterm-document matrices, which are used to represent datain a lower-dimensional space. In practice, since the data arenot static, the term-document matrices are frequently up-dated. The state-of-the-art algorithm for the correspond-ing updates of singular vectors is due to Zha and Simon[SIAM J. Sci. Comput., 21(2):782-791, 1999]. In this work,we propose a different updating approach that is based onthe Rayleigh-Ritz method. The two updating schemes are
CS13 Abstracts 349
compared for a few standard document collections.
Eugene VecharynskiUniversity of Colorado [email protected]
Yousef SaadDepartment of Computer ScienceUniversity of [email protected]
PP1
Application of Automatic Model Order Reductionto Electromagnetic Interactions
We are interested in electromagnetic coupling problems,an example problem is computing induced currents onprinted circuit board traces given an incident electromag-netic wave. The geometry is parameterized, and we wishto solve for the induced currents for all values of the pa-rameters. We use the frequency domain boundary elementmethod. A new Model Order Reduction technique is devel-oped; this technique combines hierarchical decompositionof parameter space, reduced basis method, and radial basisfunction interpolation.
Daniel White, Kyle LangeLawrence Livermore National [email protected], [email protected]
Matt StephansonOhio State [email protected]
PP1
Inversion of Rheological Parameters of MantleFlow Models from Observed Plate Motions
Modeling the dynamics of the Earth’s mantle is critical forunderstanding the dynamics of the solid earth. Yet, thereremain large uncertainties in the constitutive parametersemployed within mantle convection models. Here we for-mulate an inverse problem to infer the rheological param-eters that minimize the misfit between observed and mod-eled tangential surface velocity fields. The inverse problemis solved using a parallel scalable implementation of anadjoint-based quasi-Newton method.
Jennifer A. Worthen, Georg StadlerUniversity of Texas at [email protected], [email protected]
Noemi PetraInstitute for Computational Engineering and Sciences(ICES)The University of Texas at [email protected]
Michael GurnisCalifornia Institute of [email protected]
Omar GhattasUniversity of Texas at [email protected]
PP1
Hardware-Aware Optimizations for Using ExafmmAs a Preconditioner
The hardware landscape is changing and machine balanceis converging, so that it is timely to revisit algorithms toadapt to this situation. In regards to the fast multipolemethod, there is growing interest in its application as apreconditioner. These two factors are an opportunity toimplement optimizations that can make FMM competitivewith mainstream preconditioners like multigrid. We willshow evidence of this statement, using the exaFMM codeframework.
Rio YokotaKing Abdullah University of Science and [email protected]
Simon LaytonBoston [email protected]
Lorena A. BarbaDepartment of Mechanical EngineeringBoston [email protected]
PP1
A Method of Calculating Stress Intensity Factorsat The Edges of a Crack Located Near a WeldingSeam
We consider a crack located in the influence zone of a weld-ing seam and parallel to it. We compute the stress intensityfactors at the edges of the crack and present a method ofcalculating these factors, in which the seam is regarded asa periodic system of collinear cracks. We also showed that,with a high degree of accuracy, we can view a periodicalsystem of cracks as a chain of three cracks.
Olga ZaydenvargNational Aerospace University ”Kharkiv AviationInstitute”Department of Higher Mathematicsolga [email protected]
PP1
Population Size Effects in Genetic Algorithms forAuto Tuning
Scientific applications often rely on linear algebra compu-tations. We describe the Build To Order (BTO) com-piler which automates the optimization of those compu-tations. BTO takes in a high level linear algebra specifica-tion, searches for the optimal combination of loop fusion,and shared memory parallelism, and outputs it in C. Inthis poster, we present the effect of population size on theoverall performance of the genetic search algorithm, one ofthe search strategies in BTO.
Xing Jie ZhongUniversity of Colorado [email protected]
Thomas NelsonUniversity of Colorado at BoulderArgonne National [email protected]
350 CS13 Abstracts
Elizabeth JessupUniversity of Colorado [email protected]
Jeremy SiekDepartment of Electrical and Computer EngineeringUniversity of Colorado at [email protected]
PP1
Inverse Problems for Basal Boundary Conditions inA Thermomechanically Coupled Nonlinear StokesIce Sheet Model
Modeling the dynamics of polar ice sheets is critical forprojection of future sea level rise. Yet, there remain largeuncertainties in the boundary conditions at the base of theice sheet. Here we study mathematical and computationalissues in inversion of basal sliding coefficient and geother-mal heat flux in a thermomechanically coupled nonlinearStokes model using observations of surface flow velocities.We employ adjoint-based inexact Newton methods to solvethe inverse problem.
Hongyu ZhuInstitute for Computational Engineering and SciencesThe University of Texas at [email protected]
Tobin IsaacUniversity of Texas at [email protected]
Noemi PetraInstitute for Computational Engineering and Sciences(ICES)The University of Texas at [email protected]
Georg StadlerUniversity of Texas at [email protected]
Thomas HughesInstitute for Computational Engineering and SciencesThe University of Texas at [email protected]
Omar GhattasUniversity of Texas at [email protected]
