Nov. 17, 2008D. McCune1 Status of TRANSP and PTRANSP Presented at APS-DPP 2008, Nov. 17--21, 2008 *Supported by U.S. DOE Contract No. DE-AC02-76CH03073

Nov. 17, 2008 D. McCune 1

Status of TRANSP and PTRANSP

Presented at APS-DPP 2008,

Nov. 17--21, 2008

*Supported by U.S. DOE Contract No. DE-AC02-76CH03073


Abstract – BP6.0055D. McCune, R. Andre, E. Feibush, K. Indireshkumar, C. Ludescher-Furth, L. Randerson, PPPL*, H. St. John, General Atomics, G. Bateman, F. Halpern, A. Kritz, Lehigh University, L. Lodestro, W. Meyer, D. Pearlstein, LLNL**– The PPPL TRANSP code suite is a set of tools for time dependent simulation of axisymmetric tokamak plasmas. While the code has historically been used primarily for analysis of experimental results, predictive modeling enhancements to TRANSP have been carried out under the PTRANSP project. TRANSP and PTRANSP are now both deployed as Fusion Grid services at PPPL, supporting an international user base. The status of TRANSP and PTRANSP code development, as well as trends in production use, are presented here. New and developing features, such as the solution of free boundary MHD equilibria, the use of advanced solvers for prediction of the evolution of plasma density, angular momentum, and temperature profiles, are emphasized. The status of MPI parallelizatoin of TRANSP components is described. A recent positive development has been the beneficial collaboration between TRANSP/ PTRANSP and SciDAC FSP prototype projects SWIM and FACETs, with code components being shared effectively across all of these efforts.

*PPPL work performed under auspices of DOE contract DE-AC02-76CH03073**LLNL work performed under auspices of DOE contract DE-AC52-07NA27344


TRANSP: Vision Statement

Provide a comprehensive end-to-end modeling

capability for magnetic confinement fusion energy experiments of today and

tomorrow.


Traditional TRANSP: Overview

Preliminary dataAnalysis andPreparation

(largely automated)

DiagnosticHardware

Experiments (Asdex-U, C-Mod, DIII-D, ITER, JET, KSTAR, MAST, NSTX)

20-50 signals {f(t), f(x,t)}Plasma position, Shape,Temperatures, DensitiesField, Current, RF andBeam Injected Powers.

TRANSP Analysis*:Current diffusion, MHD equilibrium, fast ions, heating, current drive;

power, particle and momentum balance.

TRANSP Analysis*:Current diffusion, MHD equilibrium, fast ions, heating, current drive;

power, particle and momentum balance.

Experiment simulationOutput Database

~1000-2000 signals{f(t), f(x,t)}

Visualization

Load RelationalDatabases

Detailed (3d) time-slice physics simulations: GS2, ORBIT, M3D… *FusionGrid TRANSP

on PPPL servers

Pre- and Post-processing at the experimental site… MDS+

MDS+


PPPL TRANSP Run Production*

0

1000

2000

3000

4000

5000

Completed Runs

FY-2005

FY-2006

FY-2007

FY-2008

Fusion Grid TRANSPFusion Grid TRANSP

*Not shown in graphic: independent run production at JET site (approximately 1000 runs); quasi-production use of development versions in PTRANSP project (estimated to be 100s of runs).


PPPL TRANSP SERVICEFY-2005 through FY-2008

0

500

1000

1500

2000

2500

3000

3500

4000

4500

Total Runs, by Tokamak

NSTX

MAST

DIII-D

C-Mod

Asdex-U

JET*

ITER

HL-2A

other

*~3000 additional JET runs on JET production server.

12868 Total Runs


FY-2008 PPPL TRANSP Team

Name TRANSP PTRANSP SciDAC Other

Andre 80% 20%

Kumar 65% 35%

McCune 25% 35% 20% 20%

Feibush 20% 20% 60%

Ludescher 80% 20%

Randerson 40% 60%

Total FTE 3.1 0.55 0.75 1.6

Color code: Physics, Visualization, Engineering/ Operational Support


TRANSP Staffing Level• Staff levels found to be insufficient:

– Cannot promptly meet code development requests of production users;

– Long term development projects have lagged.

• Main reason: production system support.– 100s of crashed TRANSP runs per year, each

requiring individual, detailed investigation.

• New hire in FY-2009: Marina Gorelenkova.– Financed by experimental projects.


TRANSP Development• Plasma State Module (collaboration with

SWIM SciDAC).

• MPI Parallelization of code components.

• NUBEAM Upgrades.

• Monte Carlo RF Operator.

• TORIC.

• GENRAY / CQL3D.

• User interfaces (ElVis; web services).

Nov. 17, 2008 D. McCune 10

SciDAC Plasma State Module

• Standardized repository for tokamak simulation data:– MHD Equilibrium– Profiles– Species Lists, etc.

• Used in PTRANSP, SWIM, FACETS.

• NUBEAM coupled.• GENRAY in progress.

Simulation Driver

NBImodule

ICRFmodule

ECH/LHModule

Instance of Plasma State

Control ChannelData Channel

Nov. 17, 2008 D. McCune 11

Common Coupling to NUBEAM• Based on SWIM SciDAC Plasma State:

– call nubeam_ctrl_init(ps_in,init_ctrl,status)– call nubeam_ctrl_step(ps_in,ps_out,step_ctrl,status)

– Plasma State objects: ps_in, ps_out.– NUBEAM-specific control objects: init_ctrl, step_ctrl.

– integer return code: status.

• Interface used now by TRANSP.– Expected to be used by SWIM & FACETs

SciDAC projects.

Nov. 17, 2008 D. McCune 12

Plasma State Provides Infrastructure for Module Coupling

• TRANSP now provides Plasma States for use by heating and current drive codes.

• GENRAY and CQL3D use this interface now in SWIM project.

• Plan to leverage this development to import GENRAY and CQL3D into TRANSP.

• Beneficial collaboration with SciDACs!

Nov. 17, 2008 D. McCune 13

TRANSP MPI Parallelization Methods Investigated

• Client-Server: Single MPI server serves multiple serial clients:– Advantage: shared use of MPI machines.– Drawbacks: complexity, performance (file transport).

• MPI Subprocess Coupled by Plasma State:– Advantage: simplicity, process level code sharing.– Drawback: performance (use of file I/O); load balance*.

• Single MPI image:– Advantage: MPI module performance (avoid file I/O).– Drawback: interface / build complexity; load balance*.

*MPI machines idle during execution of serial sections of TRANSP.

Nov. 17, 2008 D. McCune 14

NUBEAM Upgrades

• MPI Parallelization – BP6.0057 (Kumar)• Planned: Diagnostic Simulation Upgrades:

– GC f(E,vpll/v,x,theta) based NPA simulator.– 3d Halo Neutral Density.

• Investigations:– Orbit averaging techniques (reduce variance

in momentum sources and charge exchange loss model).

• Beam-resolved outputs & other requests…

Nov. 17, 2008 D. McCune 15

MPI for more Particles, Better Statistics in Distribution Functions

10K particles 100K particlesPlots provided by MAST Team

Nov. 17, 2008 D. McCune 16

Monte Carlo RF Operator in NUBEAM

• Method based on TORIC wave fields was investigated by visiting Post Doc last year.

• Model was able to couple power to MC fast ion population.

• Statistics: MPI absolutely required.

• Post Doc (Dr. Jae-Min Kwon) returned to Korea.

• We don’t have the staff to pursue this…

Nov. 17, 2008 D. McCune 17

TORIC Full Wave ICRF Code

• Collaboration with CMod, JET, NSTX, and TORIC authors at IPP/Garching.

• TRANSP now coupled to IPP/Garching TORIC svn repository; up-to-date TORIC version now available in TRANSP.

• Plasma State upgrade– HHFW TORIC can now be run:– Support NSTX experiments– Permit scenarios other than minority heating.

Nov. 17, 2008 D. McCune 18

GENRAY/CQL3D Coupling

• Plasma State interface built for standalone GENRAY and CQL3D operation.

• Effort by B. Harvey supported by PPPL TRANSP team as part of SWIM SciDAC Collaboration.

• Foundation Laid for coupling GENRAY and CQL3D into TRANSP.

Nov. 17, 2008 D. McCune 19

Development of User Interfaces

• Web-based NSTX Between Shots Analysis user interface.

• Prototypes for time-slice post-analysis:– Linear stability: Balloon, Pest, Nova-K.– Turbulent transport: GTS– Methods based on extraction of TRANSP time

slices as Plasma States; MHD equilibrium refinement by JSolver.

• Poster by E. Feibush: BP6.0054

Nov. 17, 2008 D. McCune 20

Between Shots TRANSP Service

Nov. 17, 2008 D. McCune 21

PTRANSP Developments: General Atomics, Lehigh U., LLNL, PPPL

• Production Use.

• MHD Equilibrium Solvers:– Free Boundary Solutions; rotation pressure

modification.– Coupled Poloidal Field Diffusion: NTCC

module under development by LLNL.

• Transport Equation Solvers:– Density Prediction.– Incorporation of TGLF Transport Model.

Nov. 17, 2008 D. McCune 22

PTRANSP Production Use• 263 Full Discharge ITER PTRANSP

Simulations In FY-2008.– Accounted for nearly 60% of cpu-time usage of

entire PPPL TRANSP service.

• PTRANSP-based Publications:– Four ITER papers in FY-2007 and FY-2008.– Four more currently in preparation.– IAEA validation study using JET, DIII-D data.– Authors: R. Budny, G. Bateman, F. Halpern, A.

Kritz (PPPL / Lehigh University collaboration).

Nov. 17, 2008 D. McCune 23

PTRANSP Implementation of TEQ Free Boundary Solver– see BP6.056 (Andre).

• TRANSP/PTRANSP has successfully been using the TEQ fixed boundary solver for over two years.

– Reasonably robust and very accurate– Has become the preferred fixed boundary solver

• The TEQ free boundary solver is invoked over the fixed solver by a single namelist change.

– NTEQ_MODE=102 causes <J.B> to be used in a free boundary solution when TEQ (LEVGEO=11) is being used.

– The prescribed boundary given as input to PTRANSP is used to select fuzzy boundary points for TEQ. The coil currents are constrained in a least squares manner so that the plasma boundary lies on the fuzzy boundary points.

– To startup the run an existing free boundary solution of the tokamak is read into TEQ and perturbed through multiple invocations of TEQ to the starting conditions of the shot.

– A q mode free boundary solution will be made available after further development of the magnetic field diffusion in PTRANSP.

• An alternate free boundary solution is available in TRANSP through the PPPL ISolver code. This has been used to test the effects related to plasma toroidal rotation.

Coil

Limiter

Plasma

X point

Fuzzy Boundary Point

Nov. 17, 2008 D. McCune 24

• 1D transport (C++) and its connection logic to F90. Includes the

diffusion equations’ numerical coef.’s and sources; and 1D solver.• Corsica F90 source (tor.f90)

– Provides the transformations btw the equilibrium coordinates

(poloidal flux) and the independent transport coordinate.– Logic to obtain the conductivity and bootstrap current using either

neoHS (LLNL) or NCLASS (NTC).• Corsica run-time scripts --> transStep.f90

– Provides time step logic, restart capability, and time history of

selected variables. In addition, contains the circuit-equation logic.

Work in Progress: Parts 1 and 2 are complete. Part 3 remains.

LLNL Report: RMHD (Resistive MHD Code) Coupling TEQ & Bpol Diffusion.

Nov. 17, 2008 D. McCune 25

New PTRANSP Features Added by Lehigh Group - 1

• Expanded choice of theory-based transport models:– New MMM08 model for particle, momentum and thermal transport

• Posters BP6.00059 and BP6.00060 Monday morning by F.D. Halpern, A.H. Kritz, et al.

– GLF23 particle, momentum and thermal transport– MMM95 particle and thermal transport – NCLASS neoclassical particle and thermal transport as well as computation of

radial electric field and neoclassical electrical resistivity• Improved choice of predictive sawtooth oscillation models:

– Porcelli model for triggering sawtooth crashes and partial mixing• Porcelli sawtooth trigger can also be used with older Kadomtsev sawtooth

mixing model– Park-Monticello model for triggering sawtooth crashes – Prescribed sawtooth period

Nov. 17, 2008 D. McCune 26

New PTRANSP Features Added by Lehigh Group - 2

• Predictive transport modeling for density, momentum and temperature– Unified system of predictive transport equations are advanced in time

using consistent numerical algorithms and time-step control• See poster BP6.00058 Monday morning by G. Bateman et al.

– Boundary conditions from predictive pedestal model or experimental data– Can select subset of profiles to be obtained from experimental data

• Gas puffing feedback loop implemented for predictive particle transport

– Controls average electron density with prescribed Zeff profile vs time

• Implemented choice of techniques in the new combined system of transport equations to control numerical problems :

– Pereverzev-Corrigan numerical technique provides fast, simple controlsee Comput. Phys. Comm. 179 (2008) 579

• Tends to slow down transient behavior– Newton’s method, see Jardin et al., J. Comp. Phys. 227 (2008) 8769

• More computational effort than the Pereverzev-Corrigan method, but Newton’s method can provide improved treatment of transient conditions

– Older time-averaging methods

Nov. 17, 2008 D. McCune 27

Access to TGLF via GCNMP (General Atomics)

• Holger St. John – poster TP6.0026

• Will need to use TRANSP’s new Plasma State based MPI module coupling method.– TGLF very computationally expensive!

• Ongoing PPPL/GA collaboration: Plasma State pairs define prediction problem.– Human iterations needed to complete

definition and agree on interface methods.

Nov. 17, 2008 D. McCune 28

Summary

• Important new capabilities– MPI, Free MHD Boundary, PTRANSP, NUBEAM Upgrades.

• Ambitious development program– more MPI, PTRANSP, access to SciDAC RF codes via Plasma State.

• Labor Constraint: budget << $25M/year!– All fusion transport codes = FSP wannabes.– Operation of production system delivers V&V of

(P)TRANSP simulations to user community but imposes a heavy labor burden (FSP take note!).

Documents

Nov. 17, 2008D. McCune1 Status of TRANSP and PTRANSP Presented at APS-DPP 2008, Nov. 17--21, 2008 *Supported by U.S. DOE Contract No. DE-AC02-76CH03073