View
213
Download
0
Tags:
Embed Size (px)
Citation preview
5-11-2005 Lamps of Aladdin 2
Introduction
• The Meshing Problem– Discretize a Spatial Domain– Minimize Size (Number of Triangles)– Maximize Quality (Shape of Triangles)
• Mesh Adaptation– Introduce More Elements– Base Decisions on the Function being Interpolated
• How do we make such a decision?
5-11-2005 Lamps of Aladdin 6
The Basic Setup
• We can adapt the Mesh to Better Approximate the Function
5-11-2005 Lamps of Aladdin 7
Background: Local Feature Size
• What is Local Feature Size?– Roughly, the Size of Triangles We Should Be Using– The Smallest Distance to Two Geometric Domain Features
• Local Feature Size Should be k-Lipschitz– Doesn’t change too fast– For all p,q lfs(p) <= lfs(q) + k*dist(p,q)– This is like having derivative bounded by k.
5-11-2005 Lamps of Aladdin 8
Optimal Meshing and Adaptive Meshing
• Optimal Meshing– Consider the LFS of the Input Domain– Output a Mesh
• Triangle Size is within a constant of LFS• Triangle Aspect Ratio is bounded
– Note the Impossabilty of this if LFS is not Lipschitz• Adaptive Meshing
– LFS Accounts for Other ‘Features’ Provided by Some Oracle
– Introduce Geometric Features to Accommodate– Perturb the Original Mesh to obtain a new Optimal Mesh
5-11-2005 Lamps of Aladdin 11
• Modify the Geometric Domain to Capture a New Feature
Example of Modifying Local Feature Size
5-11-2005 Lamps of Aladdin 13
How can we Define an Oracle?
• Function-Angle Based Refinement– Some Function is Approximated by the Mesh– This Embeds the Mesh as a surface in one-higher
dimension– Observe the angle between faces of this surface– Introduce Features Where This Angle is Small
• Insert Circumcenters, Split Edges, etc.
5-11-2005 Lamps of Aladdin 19
Will this work?
• Original Feature Size must be Small Enough
• Actual Function must be Differentiable
5-11-2005 Lamps of Aladdin 20
Practicalities
• Real Mesh Adaptation Requires Coarsening As Well– Sometimes Functional Features Move or Disappear– Thrashing is a dirty word
• Real Solutions aren’t always differentiable– Give up after a Minimum Feature Size
5-11-2005 Lamps of Aladdin 21
Why is this so great for Moving Meshes?
• Unlike the One Dimensional Movie– Fixed Mesh ( Grid points didn’t move left/right )– Constantly ‘Chasing’ the Feature
• Mesh moves with the Velocity Function– Hence, Mesh moves with the Velocity features– After features are captured, very little Mesh Adaptation
5-11-2005 Lamps of Aladdin 22
How does this compare to other methods?
• Sensitivity Analysis and a posteriori Error Estimates– Double the Mesh Size Everywhere, compare solutions– Require multiple meshes of different sizing be maintained– Require multiple solves at every iteration– Don’t take advantage of Moving Mesh
• (Persistance of Features)