Approximation SEM-DG pour les problèmes d ondes elasto-acoustiques

Transcription

1 Approximation SEM-DG pour les problèmes d ondes elasto-acoustiques Helene Barucq, Henri Calandra, Aurélien Citrain, Julien Diaz, Christian Gout To cite this version: Helene Barucq, Henri Calandra, Aurélien Citrain, Julien Diaz, Christian Gout. Approximation SEM- DG pour les problèmes d ondes elasto-acoustiques. Journée Ondes du Sud Ouest 2019, Mar 2019, Le Barp, France. <hal > HAL Id: hal Submitted on 18 Mar 2019 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

2 Approximation SEM-DG pour les problèmes d ondes elasto-acoustiques Hélène Barucq 1, Henri Calandra 2, Aurélien Citrain 3,1, Julien Diaz 1 and Christian Gout 3 1 Team project Magique.3D, INRIA, E2S UPPA, CNRS, Pau, France. 2 TOTAL SA, CSTJF, Pau, France. 3 INSA Rouen-Normandie Université, LMI EA 3226, 76000, Rouen. JOSO 2019 The authors thank the M2NUM project which is co-financed by the European Union with the European regional development fund (ERDF, HN ) and by the Normandie Regional Council. Aurélien Citrain Couplage DG/SEM JOSO / 26

3 Seismic imaging?

4 Why using hybrid meshes? Useful when the use of unstructured grid is non-sense (e.g. medium with a layer of water). Well suited for the coupling of numerical methods in order to reduce the computational cost and improve the accuracy. Aurélien Citrain Couplage DG/SEM JOSO / 26

5 Why using hybrid meshes? water water sand sandstone salt Useful when the use of unstructured grid is non-sense (e.g. medium with a layer of water). Well suited for the coupling of numerical methods in order to reduce the computational cost and improve the accuracy. Aurélien Citrain Couplage DG/SEM JOSO / 26

6 Elastodynamic system x Ω R d, t [0, T ], T > 0 : ρ(x) v (x, t) t σ (x, t) t = σ(x, t), = C(x)ɛ(v(x, t)). With: ρ(x) the density, C(x) the elasticity tensor, ɛ(x, t) the deformation tensor, v(x, t), the wavespeed, σ(x, t) the strain tensor. Aurélien Citrain Couplage DG/SEM JOSO / 26

7 Elasticus software Software written in Fortran for wave propagation simulation in the time domain Features Simulation: on various types of meshes (unstructured triangles and tetrahedra), on heterogeneous media (acoustic, elastic and elasto-acoustic). Discontinuous Galerkin (DG) based on unstructured triangles and unstructured tetrahedra, with various time-schemes : Runge-Kutta (2 or 4), Leap-Frog, with multi-order computation(p-adaptivity)... Aurélien Citrain Couplage DG/SEM JOSO / 26

8 Table of contents 1 Numerical Methods 2 Comparison DG/SEM on structured quadrangle mesh 3 DG/SEM coupling 4 DGSEM vs DG 5 3D extension Aurélien Citrain Couplage DG/SEM JOSO / 26

9 1 Numerical Methods Discontinous Galerkin Method (DG) Spectral Element Method (SEM) Advantages of each method Aurélien Citrain Couplage DG/SEM JOSO / 26

10 Discontinuous Galerkin Method Use discontinuous functions : Degrees of freedom necessary on each cell : Aurélien Citrain Couplage DG/SEM JOSO / 26

11 Spectral Element Method General principle Finite Element Method (FEM) discretization + Gauss-Lobatto quadrature, Gauss-Lobatto points as degrees of freedom (gives us exponential convergence on L 2 -norm). N+1 f (x)dx ω j f (ξ j ), ϕ i (ξ j ) = δ ij. j=1 Aurélien Citrain Couplage DG/SEM JOSO / 26

12 Advantages of each method DG Element per element computation ( hp-adaptivity). Time discretization quasi explicit (block diagonal mass matrix). Simple to parallelize. Robust to brutal changes of physics and geometry SEM Couples the flexibility of FEM with the accuracy of the pseudo-spectral method. Simplifies the mass and stiffness matrices (mass matrix diagonal). Aurélien Citrain Couplage DG/SEM JOSO / 26

13 2 Comparison DG/SEM on structured quadrangle mesh Description of the test cases Comparative tables Aurélien Citrain Couplage DG/SEM JOSO / 26

14 Description of the test cases Physical parameters General context Acoustic homogeneous medium. Four different meshes : cells, cells, cells, cells. CFL computed using power iteration method. Leap-Frog time scheme. Eight threads parallel execution with OpenMP. P wavespeed Density 1000 m.s 1 1 kg.m 3 Second order Ricker Source in Pwave (f peak = 10Hz) Aurélien Citrain Couplage DG/SEM JOSO / 26

15 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0) 3.18e-3 2e SEM 4.9e-3 5e Figure: DG not penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0) 2.12e-3 7e SEM 3.26e-3 4e Figure: DG not penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

16 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0) 3.18e-3 2e SEM 4.9e-3 5e Figure: DG not penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0) 2.12e-3 7e SEM 3.26e-3 4e Figure: DG not penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

17 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0) 3.18e-3 2e SEM 4.9e-3 5e Figure: DG not penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0) 2.12e-3 7e SEM 3.26e-3 4e Figure: DG not penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

18 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0.5) 2e-3 3e SEM 4.9e-3 5e Figure: DG penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 1.33e-3 2e SEM 3.26e-3 4e Figure: DG penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

19 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0.5) 2e-3 3e SEM 4.9e-3 5e Figure: DG penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 1.33e-3 2e SEM 3.26e-3 4e Figure: DG penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

20 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0.5) 2e-3 3e SEM 4.9e-3 5e Figure: DG penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 1.33e-3 2e SEM 3.26e-3 4e Figure: DG penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

21 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time Nb of time steps DG(α = 0.5) 2e-3 3e SEM 4.9e-3 5e Figure: DG penalized and SEM comparison on the cells case CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 1.33e-3 2e SEM 3.26e-3 4e Figure: DG penalized and SEM comparison on the cells case Aurélien Citrain Couplage DG/SEM JOSO / 26

22 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 2e-3 3e SEM 2e-3 3e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh CFL L2-error CPU-time(s) Nb of time steps DG (α = 0.5) 1.33e-3 2e SEM 1.33e-3 2e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh Aurélien Citrain Couplage DG/SEM JOSO / 26

23 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 2e-3 3e SEM 2e-3 3e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh CFL L2-error CPU-time(s) Nb of time steps DG (α = 0.5) 1.33e-3 2e SEM 1.33e-3 2e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh Aurélien Citrain Couplage DG/SEM JOSO / 26

24 Comparative tables Error computed as the difference between an analytical and a numerical solution for each method. Three cases considered : DG without penalization terms, DG with penalization terms and SEM. CFL L2-error CPU-time(s) Nb of time steps DG(α = 0.5) 2e-3 3e SEM 2e-3 3e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh CFL L2-error CPU-time(s) Nb of time steps DG (α = 0.5) 1.33e-3 2e SEM 1.33e-3 2e Figure: DG penalized and SEM comparison using the same CFL on a thousands cells mesh Aurélien Citrain Couplage DG/SEM JOSO / 26

25 Advantages of each method DG Element per element computation ( hp-adaptivity). Time discretization quasi explicit (block diagonal mass matrix). Simple to parallelize. Robust to brutal changes of physics and geometry SEM Couples the flexibility of FEM with the accuracy of the pseudo-spectral method. Simplifies the mass and stiffness matrices (mass matrix diagonal). Reduces the computational costs on structured quadrangle cells in comparison with DG Aurélien Citrain Couplage DG/SEM JOSO / 26

26 3 DG/SEM coupling Hybrid meshes structures Variational formulation Aurélien Citrain Couplage DG/SEM JOSO / 26

27 Hybrid meshes structures Aim at coupling P k and Q k structures. Need to extend or split some structures (e.g. neighbour indexes). Define new face matrices: M K,L ij = φ K i φ L j, K L MK,L ij = ψi K ψj L, K L MK,L ij = φ K i ψj L. K L Aurélien Citrain Couplage DG/SEM JOSO / 26

28 Variational formulation Global context Domain in two parts : Ω h,1 (structured quadrangles + SEM), Ω h,2 (unstructured triangles + DG). Aurélien Citrain Couplage DG/SEM JOSO / 26

29 Variational formulation Definitions Jump and average. [[u]] n = (u K2 u K1 ) n {{u}} = 1 2 (u K 2 + u K1 ) Aurélien Citrain Couplage DG/SEM JOSO / 26

30 Variational formulation SEM variational formulation : ρ tv 1 w 1 = σ 1 w 1 + (σ 1 n 1 ) w 1, Ω h,1 Ω h,1 Γ out,1 tσ 1 : ξ 1 = ( (Cξ 1 )) v 1 + (Cξ 1 n 1 ) v 1. Ω h,1 Ω h,1 Γ out,1 DG variational formulation : ρ tv 2 w 2 = σ 2 w 2 + (σ 2 n 2 ) w 2 + {{σ 2 }}[[w 2 ]] n 2, Ω h,2 Ω h,2 Γ out,2 Γ int tσ 2 : ξ 2 = ( (Cξ 2 )) v 2 + Ω h,2 Ω h,2 Γ out,2 (Cξ 2 n 2 ) v 2 + {{v 2 }}[[Cξ 2 ]] n 2. Γ int Aurélien Citrain Couplage DG/SEM JOSO / 26

31 Variational formulation ρ tv 1 w 1 + ρ tv 2 w 2 = σ 1 w 1 σ 2 w 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (σ 1 n 1 ) w 1 + (σ 2 n 2 ) w 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + [[σw]] n, Γ 1/2 tσ 1 : ξ 1 + tσ 2 : ξ 2 = ( (Cξ 1 )) v 1 ( (Cξ 2 )) v 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (Cξ 1 n 1 ) v 1 + (Cξ 2 n 2 ) v 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + [[(Cξ)v]] n. Γ 1/2 Aurélien Citrain Couplage DG/SEM JOSO / 26

32 Variational formulation ρ tv 1 w 1 + ρ tv 2 w 2 = σ 1 w 1 σ 2 w 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (σ 1 n 1 ) w 1 + (σ 2 n 2 ) w 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + {{σ}}[[w]] n + [[σ]]{{w}} n, Γ 1/2 tσ 1 : ξ 1 + tσ 2 : ξ 2 = ( (Cξ 1 )) v 1 ( (Cξ 2 )) v 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (Cξ 1 n 1 ) v 1 + (Cξ 2 n 2 ) v 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + [[Cξ]]{{v}} n + {{Cξ}}[[v]] n. Γ 1/2 Aurélien Citrain Couplage DG/SEM JOSO / 26

33 Variational formulation ρ tv 1 w 1 + ρ tv 2 w 2 = σ 1 w 1 σ 2 w 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (σ 1 n 1 ) w 1 + (σ 2 n 2 ) w 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + {{σ}}[[w]] n +[[σ]]{{w}} n, Γ 1/2 tσ 1 : ξ 1 + tσ 2 : ξ 2 = ( (Cξ 1 )) v 1 ( (Cξ 2 )) v 2 Ω h,1 Ω h,2 Ω h,1 Ω h,2 + (Cξ 1 n 1 ) v 1 + (Cξ 2 n 2 ) v 2 + {{σ 2 }}[[w 2 ]] n 2 Γ out,1 Γ out,2 Γ int + [[Cξ]]{{v}} n +{{Cξ}}[[v]] n. Γ 1/2 Aurélien Citrain Couplage DG/SEM JOSO / 26

34 Aurélien Citrain Couplage DG/SEM JOSO / 26

35 Aurélien Citrain Couplage DG/SEM JOSO / 26

36 Aurélien Citrain Couplage DG/SEM JOSO / 26

37 Aurélien Citrain Couplage DG/SEM JOSO / 26

38 Aurélien Citrain Couplage DG/SEM JOSO / 26

39 Aurélien Citrain Couplage DG/SEM JOSO / 26

40 DGSEM vs DG Two dimensional Mesh with cells : triangles and quadrangles. Aurélien Citrain Couplage DG/SEM JOSO / 26

41 DGSEM vs DG Two dimensional Mesh with cells : triangles and quadrangles. Elastic isotropic case with order 2 on triangles (DG) and order 8 on the quadrangles (DG or SEM), point source at the center of the 3000x3000 domain. CFL CPU-time Nb of time steps DG(α = 0.5) 1e DG/SEM 1e Figure: DG and DG/SEM comparison on an elastic isotropic case Aurélien Citrain Couplage DG/SEM JOSO / 26

42 DGSEM vs DG Two dimensional Mesh with cells : triangles and quadrangles. Elastic isotropic case with order 2 on triangles (DG) and order 8 on the quadrangles (DG or SEM), point source at the center of the 3000x3000 domain. Salt dome case with order 2 on the triangles (DG) and order 8 on the quadrangles (DG or SEM), point source on the top layer of water. CFL CPU-time Nb of time steps DG(α = 0.5) 1e DG/SEM 1e Figure: DG and DG/SEM comparison on an elastic isotropic case CFL CPU-time Nb of time steps DG(α = 0.5) 3e DG/SEM 3e Figure: DG and DG/SEM comparison on a salt dome case Aurélien Citrain Couplage DG/SEM JOSO / 26

43 General settings Figure: Hexa/Tet boundary configuration Only deal with a simple case of 3D hybrid meshes : one hexahedron has only two tetrahedra as neighbour. Extend SEM in 3D (basis functions...). Require introducing a new matrix which handles the rotation cases between two elements. Aurélien Citrain Couplage DG/SEM JOSO / 26

44 General settings Figure: Hexa/Tet boundary configuration Only deal with a simple case of 3D hybrid meshes : one hexahedron has only two tetrahedra as neighbour. Extend SEM in 3D (basis functions...). Require introducing a new matrix which handles the rotation cases between two elements. Aurélien Citrain Couplage DG/SEM JOSO / 26

45 General settings Figure: Hexa/Tet boundary configuration Only deal with a simple case of 3D hybrid meshes : one hexahedron has only two tetrahedra as neighbour. Extend SEM in 3D (basis functions...). Require introducing a new matrix which handles the rotation cases between two elements. Aurélien Citrain Couplage DG/SEM JOSO / 26

46 General settings Figure: Hexa/Tet boundary configuration Only deal with a simple case of 3D hybrid meshes : one hexahedron has only two tetrahedra as neighbour. Require introducing a new matrix which handles the rotation cases between two elements. Aurélien Citrain Couplage DG/SEM JOSO / 26

47 Aurélien Citrain Couplage DG/SEM JOSO / 26

48 Aurélien Citrain Couplage DG/SEM JOSO / 26

49 Aurélien Citrain Couplage DG/SEM JOSO / 26

50 Aurélien Citrain Couplage DG/SEM JOSO / 26

51 Aurélien Citrain Couplage DG/SEM JOSO / 26

52 Aurélien Citrain Couplage DG/SEM JOSO / 26

53 Aurélien Citrain Couplage DG/SEM JOSO / 26

54 Aurélien Citrain Couplage DG/SEM JOSO / 26

55 Conclusion and perspectives Conclusion 1 SEM is more efficient on structured quadrangle mesh than DG 2 Build a variational formulation for DG/SEM coupling and find a CFL condition that ensures stability 3 Show the utility of using hybrid meshes and method coupling (reduce computational cost,...) Perspectives Implement DG/SEM coupling on the code (2D) Develop DG/SEM coupling in 3D Develop PML in the hexahedral part Add a local time-stepping scheme Aurélien Citrain Couplage DG/SEM JOSO / 26

56 Thank you for your attention! Questions? Aurélien Citrain Couplage DG/SEM JOSO / 26

57 Error-order graphic 10 0 DG SEM Target L 2 -error order Figure: L 2 -error comparison on a cells mesh Aurélien Citrain Couplage DG/SEM JOSO / 26

58 Error-order graphic 10 0 DG SEM Target L 2 -error order Figure: L 2 -error comparison on a cells mesh CFL L2-error CPU-time Nb of time steps DG 2e-3 3e SEM(order five) 2.13e-3 3e Figure: SEM and DG comparison with fixed error Aurélien Citrain Couplage DG/SEM JOSO / 26