Improvement of a multigrid solver for 3D EM diffusion PowerPoint PPT Presentation

presentation player overlay
About This Presentation
Transcript and Presenter's Notes

Title: Improvement of a multigrid solver for 3D EM diffusion


1
Improvement of a multigrid solver for 3D EM
diffusion
  • Research proposal final thesis Applied
    Mathematics, specialism CSE (Computational
    Science and Engineering)

2
Tom JönsthövelTU Delft januari 6, 2006Tutors
Kees Oosterlee (EWI, TUD)Wim Mulder (SIEP,
Shell)
3
Overview presentation
  • Introduction
  • Practical application/problem statement
  • Mathematical model
  • Discretization model equations
  • Short overview Multigrid
  • Problems encoutered with MG solver
  • Possible improvements on MG solver
  • Summary
  • Questions

4
Practical Application
  • Practical goal
  • Image structures that could host potential
    reservoirs
  • Providing evidence of presence of hydrocarbons
  • How?
  • Recovering the conductivity profile from
    measurements of electric and magnetic fields
  • Oil/Gas are more resistive than surrounding
    (gesteente)

5
(3D) Electromagnetic Diffusion
2D example
EM source
receivers
Oil/Gas?
6
(3D) Electromagnetic Diffusion
Amundsen, Johansen Røsten (2004) A Sea Bed
Logging (SBL) calibration survey over the Troll
gas field
7
Mathematical model
Maxwell equations in presence of current source
With,
Ohms law
8
Maxwell equations
Eliminate the magnetic field from the equation
?
9
Maxwell equations
Transform equation from time to frequency
domain
With ? angular frequency. Now,
In practice
10
Maxwell equations
PEC boundary conditions (Perfectly Electrically
Conduction)
domain
11
Discretization model equationsStep 1
  • Choose discretization
  • Finite Integration Technique (Clemens/Weiland
    01)
  • Finite volume generalisation of Yees scheme
    (1966)
  • Error analysis for constant-coeffients (Monk
    Sülli 1994)
  • 2nd order accuracy for electric/magnetic field
    components

12
Discretization model equationsStep 2
Placement EM field components (Yees scheme)
13
Discretization model equationsStep 3
Next steps discretize all components of main
equation
ii)
iii)
14
Discretization model equationsii)
1st 2nd
15
Discretization curl
Stokes
16
Discretization curl
17
Introduce (discreet) residu
Goal solve for r 0 How? Multigrid solver
18
Overview Multigrid
Idea use BIM for solving Axb 1. the error
exex-xapr becomes smooth (not small) 2.
Quantity smooth on fine grid ? approx on
coarser grid (e.g. double mesh size) Concl
error smooth after x relaxation sweeps ? approx
error on coarser grid ? Cheaper/Faster
19
Basis MG
  • Pre-smoothing
  • Coarse grid correction
  • Restriction
  • Compute approximation solution of defect
    equation
  • - Direct/iterative solver
  • - New cycle on coarser grid
  • Prolongation
  • Post-smoothing

20
Basis MG
  • Important choices
  • Coarser grids
  • Restriction operator residu from fine to coarse
  • Prolongation operator correction from coarse to
    fine
  • Smoother

21
MG Components
Coarser grids
22
MG Components
Restriction Full weighting
23
MG Components
Prolongation Linear/bilinear interpolation, is
transpose of restriction
24
MG Components
Smoother Pointwise smoother Symmetric GS-LEX
25
Introduce Test probleem
  • Artificial eigenvalues problem
  • On the domain 0,2p3.
  • This defines the source term Js.
  • Convergence 108

26
Stretching
27
Problems MG Solver
s010 S/m, s11 S/m
cells equidistant equidistant equidistant equidistant Stretched (4) Stretched (4) Stretched (4) Stretched (4)
cells hmax MG MG bi hmax MG MG bi
16 0.39 7 6 6 0.45 8 6 6
32 0.20 8 7 7 0.26 11 8 8
64 0.098 8 7 7 0.17 12 14 14
128 0.049 8 6 6 0.13 81 32 32
28
Anisotropy
2D anisotropic elliptical equations
29
Anisotropy
Discretization in stencil notation
30
Anisotropy
Error averaging with GS-LEX
If e?0,
No smoothing effect in x-direction
31
Anisotropy and stretched grid
2D elliptical equations
Simple stretching
Hence
32
Anisotropy
  • Two possible improvements MG solver
  • Semi coarsening
  • Line-smoother

33
Semicoarsening
34
Line-smoother
Solve all unknowns on line in direction
anisotropy simultaneously. Reason Errors become
smooth if strong connected unknows are updated
collectively
35
Preview results
Combination line-smoother and semi-coarsening
gives good results Factor 5 less MG iterations
needed
36
Summary
  • Oil/Gas reservoir?
  • EM diffusion method ? Maxwell equations
  • Multigrid solver ? Problems when gridstretching
    used
  • Improvements
  • Line-Smoother
  • Semi Coarsening
  • Results are obtained ? more research for
    improvement and generalisation, mathematical
    soundness

37
Questions?
Write a Comment
User Comments (0)
About PowerShow.com