HAL Id: hal-01840081
https://hal.inria.fr/hal-01840081
Preprint submitted on 31 Jul 2018
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished 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.
The MHM method for the second-order elastodynamic
model
Weslley da Silva Pereira, Claire Scheid, Frédéric Valentin
To cite this version:
Weslley da Silva Pereira, Claire Scheid, Frédéric Valentin. The MHM method for the second-order elastodynamic model. 2018. �hal-01840081�
The MHM method for the second-order
elastodynamic model
Weslley da Silva Pereira1
Joint work with
Claire Scheid2 and Frédéric Valentin1 1Laboratório Nacional de Computação Científica (BR)
2Inria Sophia Antipolis - Mediterranée (FR)
July 13, 2018
Outline
Introduction Hybridization procedure The MHM method Numerical results An analytical problem Multiscale problemFinal conclusions and remarks
Introduction
Elastodynamic problem
Equation of motion and the constitutive equationρ∂ttu − ∇· σ =f, σ =Cε , ε :=
∇u + (∇u)T
2 with u(x, 0) = u0(x), ∂tu(x, 0) = v0(x) and u|∂Ω= 0.
Figure : Seismic imaging technique.
Figure : Layered media.
MHM - Multiscale Hybrid-Mixed method
1Figure : TH with elements K .
h
Figure : Local mesh ThK.
1
C. Harder, D. Paredes, and F. Valentin.A family of Multiscale Hybrid-Mixed finite element methods for the Darcy equation with rough coefficients.J. Comput. Phys., 245:107–130, 2013
Hybridization procedure
1. Partition the domain Ω × [0, T ]
V := w ∈ L2(Ω) : w |K ∈ H1(K ) ∀K ∈ TH , Λ := τ nK|∂K, ∀K ∈ TH : τ ∈ H(div; Ω; S) . t t t t t t t t Ω t t 0 1 1 2 n-1 n n n+1 n -1 n T T + + + + -In K I =n (t , t )n-1 n TH H Notation: • (·, ·)K - L2(K ) product;• h·, ·i∂K - dual product
H−12(∂K ) × H 1 2(∂K ); • (·, ·)∂TH:=P K ∈THh·, ·i∂K; • u(n):= u|In. 5/20
2. A hybrid variational problem
Find (u, λ) that, for n = 1, · · · , nT, satisfies dtt(ρu(n), w )K+ (σ(u(n)), ε(w ))K−hλ(n), w i∂K =(f , w )K, ∀w ∈ H1(K ), (µ, u(n))∂TH= 0, ∀µ ∈ Λ, u(n)(tn−1) =u(n−1)(tn−1), dtu(n)(tn−1) =dtu(n−1)(tn−1),
where un:= u(n)(tn) = u(tn) and vn:= dtu(n)(tn) = dtu(tn) .
Theorem
The problem is well-defined for
f ∈ H1(0, T ; L2(Ω)), u0, v0 ∈ H10(Ω) and σ(u0) ∈ H(div; Ω; S).
The solution is unique in
C (0, T ; V) ∩ C1(0, T ; L2(Ω)) × L2(0, T ; Λ). Moreover, u solves the original problem and
λ = σ(u)nK on ∂K , ∀K ∈ TH.
3. Split the solution
u(n)=Tn(λ(n))+Tˆn(zn), zn:= (un−1, vn−1, f ).
where Tn and ˆTn are linear operators
Tn: L2(In; Λ) → L2(In; L2(Ω)), ˆ
Tn: V × L2(Ω) × L2(In; L2(Ω)) → C (In; V) ∩ C1(In; L2(Ω)).
For each K × In, u(n)λ := Tn(λ(n)) is the solution of
( dtt(ρ u (n) λ , w )K+ (σ(u (n) λ ), ε(w ))K= hλ(n), w i∂K, ∀w ∈ H1(K ), u(n)λ (tn−1) = dtu(n)λ (tn−1) = 0,
and u(n)z := ˆTn(zn) is the solution of dtt(ρ u(n)z , w )K+ (σ(u (n) z ), ε(w ))K = (f , w )K, ∀w ∈ H1(K ), u(n)z (tn−1) = un−1, dtu(n)z (tn−1) = vn−1. 7/20
Global-Local formulation
2Global-Local formulation
Find λ that, for n = 1, · · · , nT, satisfies
(µ, Tn(λ(n)))∂TH = −(µ, ˆTn(zn))∂TH, ∀µ ∈ Λ,
where
zn:= (un−1, vn−1, f ),
un:= Tn(λ(n))|tn+ ˆTn(zn)|tn,
vn:= dtTn(λ(n))|tn+ dtTˆn(zn)|tn,
This is equivalent to the hybrid problem using
u(n):= Tn(λ(n)) + ˆTn(zn).
2
A. T. Gomes, D. Paredes, W. Pereira, R. Souto, and F. Valentin.A Multiscale Hybrid-Mixed Method for the Elastodynamic Model with Rough Coefficients.In XXXVIII CILAMCE, 2017
The MHM method
One-level MHM method
Global discretization • ΛH ⊂ Λ and basis {ψi} mΛ i =1. • P0(In; ΛH) ⊂ L2(In; Λ). u(n)H := mΛ X j=1 αnjTn(ψj) + ˆTn(zH,n)For n = 1, · · · , nT , find αn∈ RmΛ such that mΛ X j=1 αnj(ψi, Tn(ψj)|tn)∂TH=−(ψi, ˆTn(zH,n)|tn)∂TH, ∀i = 1, · · · , mΛ, where zH,n:= (uH,n−1, vH,n−1, f ), uH,n:= u (n) H (tn), vH,n:= dtu (n) H (tn)
and initial conditions
uH,0:= u0 and vH,0:= v0.
Two-level MHM method
Discretization in K × In: • Vh(K ) ⊂ H1(K ), ∀K ∈ TH; • Newmark-(γ, β) in In, ∀n = 1, · · · , nt. η(n)j ≈ Tn(ψj)|tn, η (n) 0 ≈ ˆTn(zh,n)|tn, η˙ (n) j ≈ dtTn(ψj)|tn, η˙ (n) 0 ≈ dtTˆn(zh,n)|tn, uh,n |K := mΛ X j=1 αnjη(n)j + η (n) 0 , vh,n |K := mΛ X j=1 αnjη˙(n)j + ˙η (n) 0 , σh,n |τ:= σ(uh,n)|τ. MHM methodFor n = 1, · · · , nT , find αn∈ RmΛ such that mΛ X j=1 αnj(ψi, η (n) j )∂TH =−(ψi, η (n) 0 )∂TH, ∀i = 1, · · · , mΛ, where zh,n:= (uh,n−1, vh,n−1, f ) , uh,0≈ u0 and vh,0≈ v0 . 10/20
Two-level MHM method
• uh,n ∈ H1(TH), ∀n = 1, · · · , nT (H1(Ω)-non-conforming)
• Strong symmetric σh
• High-order local approximations in space Vh(K )
• Use of different local parameters (h, p) and
(∆τ = ∆tnτ, γ, β) to different K × In
• Energy-conservative local method ⇒ energy-conservative
MHM
Numerical results
An analytical problem
Isotropic homogeneous media with exact solution u = [u1, u2, u3]T:
( u
1(t, x , y , z) = 12(1 − cos(ωt)) sin(2πx ) sin(2πy ) sin(2πz) ,
u2(t, x , y , z) = −12(1 − cos(ωt)) sin(2πx ) sin(2πy ) sin(2πz) ,
u3(t, x , y , z) = 12(1 − cos(ωt)) sin(πx ) sin(πy ) sin(πz) .
ΛH :=µH∈ Λ : µH|F ∈ [P`(F )] 2 for all F ∈ ∂TH Vh(K ) := vh∈ C (K ) : vh|τ ∈ [Pk(τ )]3 for all τ ∈ ThK 12/20
Convergence results
Figure : Experiments with ` = 3 and k = 5.
10-6 10-5 10-4 10-3 10-2 10-1 100 101 10-1 100 H ||u-uh||0 ||u-uh||1 ||v-vh||0 ||v-vh||1 ||σ-σh||div O(H) O(H2) O(H3) 10-5 10-4 10-3 10-2 10-1 100 101 10-1 100 Δt ||u-uh||0 ||u-uh||1 ||v-vh||0 ||v-vh||1 ||σ-σh||div O(Δt2)
Error convergence observed numerically:
• O(∆t2) in all norms.
• O(H`+2) in the L2(Ω)-norm, O(H`+1) in the H1(TH)-norm
and O(H`) in the div -norm.
Multiscale scenario
3Figure : The physical domain divided in layers. Problem: • Flat case. • Single shot. • Two-dimensional problem at a slice (inline = 0 m).
• Free surface boundary
condition.
3
J. de la Puente.HPC4E Seismic Test Suite, 2016.Copyright Josep de la Puente (Barcelona Supercomputing Center) 2016. Licenced under the Creative-Commons Attribution-ShareAlike 4.0 International License
MHM configuration
• Edges divided in 4 equally spaced parts.
• P1 approximation in each part of edge.
• Local meshes with 4096 P3 elements (h = H/64).
MHM solution
Figure : Isolines of |uh| solution at t = 0.15.
Global energy
Eh:= 1 2 X K ∈TH (ρvh, vh)K + (σh, ε(uh))K 0 1x10-7 2x10-7 3x10-7 4x10-7 5x10-7 0 0.5 1 1.5 2 t Eh 17/20Comparison with a Galerkin solution
Figure : Isovalues of the von Mises stress for MHM (top) and a reference solution (bottom), at t = 0.15.
Final conclusions and remarks
• The MHM method provides a natural level of parallelism.
• It allows local mesh refinement and high-order approximations.
• Optimal convergence in time and super convergence in space
for displacement, velocity and stress.
• It conserves energy depending on the local level.
• It is robust on non-aligned mesh with interfaces.
Ongoing work:
• Local time step validation.
• Error and stability analysis.
Thank you!
Funding and resources:
[1] J. de la Puente.HPC4E Seismic Test Suite, 2016. Copyright Josep de la Puente (Barcelona Supercomputing Center) 2016. Licenced under the Creative-Commons Attribution-ShareAlike 4.0 International License.
[2] A. T. Gomes, D. Paredes, W. Pereira, R. Souto, and
F. Valentin. A Multiscale Hybrid-Mixed Method for the
Elastodynamic Model with Rough Coefficients. In XXXVIII
CILAMCE, 2017.
[3] C. Harder, A. L. Madureira, and F. Valentin.A Hybrid-Mixed
Method for Elasticity. ESAIM: M2AN, 50(2):311–336, 2016.
[4] C. Harder, D. Paredes, and F. Valentin.A family of Multiscale
Hybrid-Mixed finite element methods for the Darcy equation
with rough coefficients. J. Comput. Phys., 245:107–130, 2013.
[5] C. Harder and F. Valentin.Foundations of the MHM method, volume 114 of Lecture Notes in Computational Science and Engineering, chapter Building Bridges: Connections and Challenges in Modern Approaches to Numerical Partial
Differential Equations, pages 401–433.Springer, 2016.
[6] W. S. Pereira and F. Valentin.A Locking-Free MHM Method
for Elasticity. In Proceeding Series of the Brazilian Society of
Computational and Applied Mathematics, volume 5, 2017.
[7] P. A. Raviart, J. M. Thomas, P. G. Ciarlet, and J.-L. Lions.
Introduction à l’analyse numérique des équations aux dérivées partielles. Masson, Paris, 1983.