analysis of the performance of the interior penality discontinuous galerkin method c. baldassari, h....
TRANSCRIPT
![Page 1: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/1.jpg)
Analysis of the performance of the Interior Penality Discontinuous
Galerkin method
C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ
![Page 2: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/2.jpg)
The seismic exploration
The Reverse Time Migration technique (RTM):– Propagation of the wave field– Retro-propagation of the data– Application of an imaging condition
![Page 3: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/3.jpg)
Motivation of the work
• Goal of the thesis: Migration including topography effects with wave equation in 3D (RTM)
• Both the accuracy and the computational cost of the numerical method to solve the direct problem are crucial
• Our choice: a finite element method which uses meshes adapted to the topography of the domain: the Interior Penality Discontinuous Galerkin method (IPDG)
![Page 4: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/4.jpg)
Outline
• Presentation of the IPDG method
• Comparison 1D with a spectral finite element method (SEM)
• Comparison 2D with SEM and analytic solution
• Results of the propagation in an irregular top domain with IPDG
• Conclusion and ongoing works
![Page 5: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/5.jpg)
Initial problem
2
2
1 1.
0, 0
0, 0
0
0
1 10
D
N
abs
uu f in I
t
u x in
ux in
tu on
u n on
uu n on
t
![Page 6: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/6.jpg)
Notations
•
• • • • •
D N abs
0, : a finite time interval
: a bounded domain in IR , 2,3d
I T
d
: the density of the medium satisfying 0<
: the compressibility satisfying 0<
: the unit outward normal of
x
x
n
![Page 7: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/7.jpg)
The IPDG method
• Method proposed by Douglas and Dupont in the 70’s
• Applied to the wave equation by Grote, Schneebeli and Schötzau in 2005
• Continuity is weakly enforced across interfaces by adding bilinear forms, so-called fluxes
• Method based on meshes made of triangles in 2D or tetrahedra in 3D
![Page 8: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/8.jpg)
Notations -1-
i
b
: a shape-regular mesh of composed by
elements (triangles) ;
an interior face of define by two elements ;
F the set of all
a boundary face of ;
F the s
h
h
K T
i h i
i
b h b
T
K K
F T F K K
F
F T F K
h i b
et of all
F F F
the unit outward vectors on the boundaries
the traces of a fonction on
bF
n K
v v K
![Page 9: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/9.jpg)
Notations -2-
i
i
b
b
: the jump of at F ; F F
/ 2 : the average of at F ; F F
: the jump of at F ; F F
: the average of at F ; F F
For a smooth vector-valued function, we
v v n v n v x
v v v v x
v vn v x
v v v x
2
analogously define
the jump and the average
: : the finite element spacehl l hKV v L v P K K T
![Page 10: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/10.jpg)
Space discretization -1-
2
2
After space discretization, we obtain the scheme:
where
is the vector defined by the composants of
The mass matrix is block-diagonal and its coefficients are:
1
h hh h
hh h l
ij iK K
U UM B KU F
t t
U u V
M
M v
b
abs
F
, ,
The matrix is equal to zero except for 's elements:
1, ,
h
b abs b
hj i j l
T
hij i j i j l
F F
v dx v v V
B
B v v dF v v V
![Page 11: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/11.jpg)
Space discretization -2-
i
i i
F
F F
The stiffness matrix is symmetric and has for entries the terms:
1 1
1
h i i
i ii i
ij i j i jK T FK F
j i i jF FF F
i
K
K v v dx v v dF
v v dF v v dF
v
b b
b
F F
F
1 1
, ,
b bb b
b b
j j iF FF F
hi j i j l
F F
v dF v v dF
v v dF v v V
![Page 12: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/12.jpg)
Space discretization -3-
1h i b max minF
,
The function penalizes the jump of and on 's faces.
It is defined by: F F F F ,
The source vector has for components the terms:
h
i j h
h
h i iK T K
v v T
c h
F
F fv dx
![Page 13: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/13.jpg)
IPDG’s advantages
• Meshes made of triangles in 2D or tetrahedra in 3D. Thus the topography of the computational domain is easily discretized.
• The representation of the solution is quasi-explicit because the mass matrix is block-diagonal.
To compute easily its coefficients, we use an exact quadrature formula which does not hamper the order of convergence.
![Page 14: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/14.jpg)
Comparison 1D IPDG versus SEM
ORDER 2 Uniform mesh Random mesh
IPDG SEM IPDG SEM
nb_ddl = 90 dx = 0.200dt = 0.0808
err = 24.9565
dx = 0.100dt = 0.0990err = 0.2700
dx = 0.200dt = 0.0761
err = 24.9720
dx = 0.100dt = 0.0913err = 1.7403
nb_ddl = 180 dx = 0.100dt = 0.0404
err = 11.0781
dx = 0.0500dt = 0.0495err = 0.0676
dx = 0.100dt = 0.0380
err = 10.9472
dx = 0.0500dt = 0.0448err = 0.5149
nb_ddl = 360 dx = 0.0500dt = 0.0202err = 2.9120
dx = 0.0250dt = 0.0247err = 0.0169
dx = 0.0500dt = 0.0190err = 2.8747
dx = 0.0250dt = 0.0224err = 0.1261
nb_ddl = 540 dx = 0.0333dt = 0.0135err = 1.3004
dx = 0.0167dt = 0.0165err = 0.0075
dx = 0.0333dt = 0.0127err = 1.2836
dx = 0.0167dt = 0.0149err = 0.0565
nb_ddl = 720 dx = 0.0250dt = 0.0101err = 0.7325
dx = 0.0125dt = 0.0124err = 0.0042
dx = 0.0250dt = 0.0095err = 0.7219
dx = 0.0125dt = 0.0112err = 0.0318
![Page 15: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/15.jpg)
Comparison 1D IPDG versus SEM
ORDER 3 Uniform mesh Random mesh
IPDG SEM IPDG SEM
nb_ddl = 180 dx = 0.1500dt = 0.0347err = 1.2187
dx = 0.100dt = 0.0400err = 1.5961
dx = 0.1500dt = 0.0320err = 1.0594
dx = 0.100dt = 0.0363err = 1.3152
nb_ddl = 360 dx = 0.0750dt = 0.0174err = 0.3040
dx = 0.0500dt = 0.0200err = 0.4018
dx = 0.0750dt = 0.0159err = 0.2570
dx = 0.0500dt = 0.0182err = 0.3327
nb_ddl = 720 dx = 0.0375dt = 0.0087err = 0.0760
dx = 0.0250dt = 0.0100err = 0.1007
dx = 0.0375dt = 0.0079err = 0.0632
dx = 0.0250dt = 0.0091err = 0.0827
nb_ddl = 1080 dx = 0.0250dt = 0.0058err = 0.0338
dx = 0.0167dt = 0.0067err = 0.0448
dx = 0.0250dt = 0.0053err = 0.0281
dx = 0.0167dt = 0.0060err = 0.0368
nb_ddl = 1440 dx = 0.0187dt = 0.0043err = 0.0190
dx = 0.0125dt = 0.0050err = 0.0252
dx = 0.0187dt = 0.0040err = 0.0159
dx = 0.0125dt = 0.0125err = 0.0207
![Page 16: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/16.jpg)
Comparison 1D IPDG versus SEM
ORDER 4 Uniform mesh Random mesh
IPDG SEM IPDG SEM
nb_ddl = 180 dx = 0.200dt = 0.0307err = 0.4865
dx = 0.1500dt = 0.0348err = 0.6263
dx = 0.200dt = 0.0305err = 0.4795
dx = 0.1500dt = 0.0335err = 0.5916
nb_ddl = 360 dx = 0.100dt = 0.0153err = 0.1215
dx = 0.0750dt = 0.0174err = 0.1565
dx = 0.100dt = 0.0152err = 0.1195
dx = 0.0750dt = 0.0168err = 0.1460
nb_ddl = 720 dx = 0.0500dt = 0.0077err = 0.0304
dx = 0.0375dt = 0.0087err = 0.0391
dx = 0.0500dt = 0.0076err = 0.0299
dx = 0.0375dt = 0.0084err = 0.0362
nb_ddl = 1080 dx = 0.0333dt = 0.0051err = 0.0135
dx = 0.0250dt = 0.0058err = 0.0174
dx = 0.0333dt = 0.0051err = 0.0132
dx = 0.0250dt = 0.0056err = 0.0163
nb_ddl = 1440 dx = 0.0250dt = 0.0038err = 0.0076
dx = 0.0187dt = 0.0044err = 0.0098
dx = 0.0250dt = 0.0038err =0.0074
dx = 0.0187dt = 0.0042err = 0.0091
![Page 17: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/17.jpg)
Comparison 2D IPDG/SEM/Exact solution
• [0 1400] x [0 2100]
• C1=1500m/s
• C2=3000m/s
• Source position: (700,1050)
• First derivative of a Gaussian
• f =20Hz
• Dirichlet condition on the top, absorbing condition elsewhere
• Time propagation: 0.9s
• Position of the receivers:
(5*i,1050) i=1,…,280
![Page 18: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/18.jpg)
Results for a fine mesh (52 pts/ ) -1-
![Page 19: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/19.jpg)
Results for a fine mesh (52 pts/ ) -2-
![Page 20: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/20.jpg)
Results for a coarse mesh (20 pts/ ) -1-
![Page 21: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/21.jpg)
Results for a coarse mesh (20 pts/ ) -2-
![Page 22: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/22.jpg)
The foothill case
• [0 1440] x [0 730]
• C1=1500m/s
• C2=3000m/s
• Source position: (300,530)
• First derivative of a Gaussian
• f =20Hz
• Dirichlet condition on the top, absorbing condition elsewhere
• Time propagation: 1s
• Position of the receivers:
(5*i,530) i=1,…,288
![Page 23: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/23.jpg)
Influence of the size of the mesh for IPDG
Mesh n°1 Mesh n°2
Mesh n°3
![Page 24: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/24.jpg)
Seismograms
Mesh n°1 Mesh n°2 Mesh n°3
100 pts/ 30 pts/ 15 pts/
![Page 25: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/25.jpg)
Conclusions and ongoing works
• Conclusions:– Absorbing conditions must be improved– It’s only necessary to have a fine mesh at the surface of
the domain
• Ongoing works:– Analysis of the numerical dispersion– Improvement of the absorbing conditions– Implementation into the MigWE code– Local time stepping
![Page 26: Analysis of the performance of the Interior Penality Discontinuous Galerkin method C. BALDASSARI, H. BARUCQ, H. CALANDRA, B. DENEL, J. DIAZ](https://reader030.vdocument.in/reader030/viewer/2022032702/56649cf75503460f949c72ab/html5/thumbnails/26.jpg)
iF
1
i i
i jF F
v v dF
bF
1
b b
j iF F
v v dF
iFi i
i jF F
v v dF
bFb b
i jF F
v v dF
i
b
F
F
Fonction test telle que:
10 et 0
1
1
1
h
i i
b b
j
j j
ij i jK T K
j iF F
i jF F
v
v v
K v v dx
v v dF
v v dF