Finite element programming by FreeFem++ advanced course

Size: px
Start display at page:

Download "Finite element programming by FreeFem++ advanced course"

Transcription

1 1 / 37 Finite element programming by FreeFem++ advanced course Atsushi Suzuki 1 1 Cybermedia Center, Osaka University atsushi.suzuki@cas.cmc.osaka-u.ac.jp Japan SIAM tutorial 4-5, June 2016

2 2 / 37 Outline Basics of FreeFem++ by examples from the Poisson equation Poisson equation with mixed boundary conditions variational forms and stiffness matrix in FreeFem++ Schwarz iterative method overlapping two subdomains and Jacobi-Schwarz method parallel implementation by MPI in FreeFem++ Schwarz algorithm as preconditioner for global Krylov iteration overlapping subdomains and RAS/ASM Krylov subspace method with preconditioner 2-level algorithm with a coarse space Iterative substructuring method / Balancing N-N Schur complement method with preconditioner coarse space constructed from kernel of local problems FETI : Finite element Tearing and Interconnecting interface problem by Lagrange multiplier CG method on the image space with orthogonal projection

3 Numerical simulation with finite element method mathematical modeling discretization of time for evolution problem discretization scheme for the space mesh generation / adaptive mesh refinement stiffness matrix from finite elements and variational formulation linear solver CG, GMRES, direct solver: UMFPACK, MUMPS FreeFem++ provides vast amounts of tools distributed parallelization Krylov subspace method with sophisticated preconditioner by domain decomposition 2-level additive Schwarz method with coarse space iterative method on the interface problem between subdomains iterative substructurinrg method and balancing domain decomposition FETI method 3 / 37

4 4 / 37 Poisson equation with mixed B.C. and a weak formulation Ω R 2, Ω = Γ D Γ N u = f in Ω, u = g on Γ D, u n = h on Γ N. weak formulation function space : V = H 1 (Ω), affine space: V (g) = {u V ; u = g on Γ D }. Find u V (g) s.t. u vdx = Ω a(, ) : V V R : bilinear form F ( ) : V R : functional Ω f vdx + h vds Γ N v V (0)

5 FreeFem++ script to solve Poisson equation finite element basis, span[ϕ 1,..., ϕ N ] = V h V, u h V h u h = 1 i N u iϕ i Dirichlet data : u(p j ) = g(p j ) P j Γ D Find u h V h (g) s.t. u h v h dx = f v h dx + h v h ds v h V h (0). Ω Ω Γ N mesh Th=square(20,20); fespace Vh(Th,P1); Vh uh,vh; func f = 5.0/4.0*pi*pi*sin(pi*x)*sin(pi*y/2.0); func g = sin(pi*x)*sin(pi*y/2.0); func h = (-pi)/2.0 * sin(pi * x); solve poisson(uh,vh)= int2d(th)( dx(uh)*dx(vh)+dy(uh)*dy(vh) ) - int2d(th)( f*vh ) - int1d(th,1)( h *vh ) + on(2,3,4,uh=g); // boundary 1 : (x,0) plot(uh); 5 / 37

6 6 / 37 discretization and matrix formulation FE basis, span[ϕ 1,..., ϕ N ] = V h V, Λ = {1, 2,, N}. u h V h u h = i Λ u iϕ i Dirichlet data : g k = g(p k )k Λ D Λ Find {u j } j Λ s.t. a(ϕ j, ϕ i )u j = F (ϕ i ) i Λ \ Λ D j u k = g k k Λ D mesh Th=square(20,20); fespace Vh(Th,P1); Vh u,v; varf aa(u,v)=int2d(th)( dx(u)*dx(v)+dy(u)*dy(v) ) +on(2,3,4,u=g); // boundary 1 : (x,0) varf external(u,v)=int2d(th)(f*v)+int1d(th,1)(h*v) +on(2,3,4,u=g); real tgv=1.0e+30; matrix A = aa(vh,vh,tgv=tgv,solver=cg); // boundary 1 : (x,0) real[int] ff = external(0,vh,tgv=tgv); u[] = A^-1 * ff; // u : fem unknown, u[] : vector

7 penalty method to solve inhomogeneous Dirichlet problem modification of diagonal entries of A where index k Λ D penalization parameter τ = 1/ε; tgv τ u k = τg k, k Λ D [A] i j = a(ϕ j, ϕ i ) u i f i τu k + a k j u j = τg k u k g k = ε( a k j u j ), j k j k a i j u j = f i i {1,..., N} \ Λ D. j keeping symmetry of the matrix without changing index numbering. 7 / 37

8 8 / 37 alternative Schwarz algorithm (1/2) Ω 2 u 0 1, u0 2 : given loop n = 0, 1, 2, u n+1 1 = f in Ω 1 u n+1 2 = f in Ω 2 u n+1 1 = 0 on Ω 1 Ω u n+1 2 = 0 on Ω 2 Ω u n+1 1 = u n 2 on Ω 1 Ω 2 u n+1 2 = u n+1 1 on Ω 2 Ω 1 essentially sequential interpolation between difference meshes requires Ω 1

9 alternative Schwarz algorithm (2/3) : FreeFem++ script int interface = 2; int original = 1; border a(t=1,2){x=t;y=0;label=original;}; border b(t=0,1){x=2;y=t;label=original;}; border c(t=2,0){x=t;y=1;label=original;}; border d(t=1,0){x= 1-t;y=t;label=interface;}; border e(t=0,pi/2){x= cos(t);y=sin(t);label=interface;}; border e1(t=pi/2,2*pi){x=cos(t);y=sin(t); label=original;}; int n=5; mesh[int] th(2); th[0]=buildmesh(a(5*n)+b(5*n)+c(10*n)+d(5*n)); th[1]=buildmesh(e(5*n)+e1(25*n) ); 9 / 37

10 10 / 37 alternative Schwarz algorithm (3/3) : FreeFem++ script fespace Vh0(th[0],P1); fespace Vh1(th[1],P1); Vh0 u0, v0; Vh1 u1, v1; macro Grad(u) [dx(u),dy(u)] // EOM int i; problem pb0(u0,v0,init=i,solver=umfpack)= int2d(th[0])(grad(u0) *Grad(v0)) -int2d(th[0])(-v0)+on(interface, u0=u1) +on(original,u0=0); problem pb1(u1,v1,init=i,solver=umfpack)= int2d(th[1])(grad(u1) *Grad(v1)) -int2d(th[1])(-v1)+on(interface, u1=u0) +on(original,u1=0); for (i=0 ;i< 10; i++) { pb0; pb1; };

11 11 / 37 Schwarz algorithm : 1/2 Jacobi-Schwarz algorithm u 0 1, u0 2 : given loop n = 0, 1, 2, u n+1 1 = f in Ω 1 u n+1 2 = f in Ω 2 u n+1 1 = 0 on Ω 1 Ω u n+1 2 = 0 on Ω 2 Ω u n+1 1 = u n 2 on Ω 1 Ω 2 u n+1 2 = u n 1 on Ω 2 Ω 1 parallel computation, extendable to more than two subdomains overlapping subdmains

12 12 / 37 Schwarz algorithm : implementation with MPI 1/2 MPI (message passing interface) for distributed parallel computation almost all MPI functions are available in FreeFem++ MPI communicator: mpicomm() broadcasting data from one processor to all others : brodcast() asynchronous data sending/receiving: Isend(),Irecv() synchronization of processes synchronization with message passing requests: mpiwaitany() synchronization of all processes: mpibarrier() reduction operation among processes: mpiallreduce()

13 Schwarz algorithm : implementation with MPI 2/2 mpicomm comm(mpicommworld,0,0); int myrank=mpirank(comm); int orank=(myrank+1)%2; if (myrank==0) th[0]=buildmesh(a(5*n)+b(5*n)+c(10*n)+d(5*n)); else th[1]=buildmesh(e(5*n)+e1(25*n)); broadcast(processor(0,comm),th[0]); broadcast(processor(1,comm),th[1]); mesh tth=th[myrank]; int i; problem pb(u,v,init=i,solver=umfpack) = int2d(tth)(grad(u) *Grad(v)) - int2d(tth)(-v)+on(inside,u=u1)+on(outside,u=0); mpirequest[int] rq(2); for (i=0;i<10; i++) { pb; Irecv(processor(orank,comm,rq[0]),u1[]); Isend(processor(orank,comm,rq[1]),u[]); mpiwaitany(rq); }; 13 / 37

14 Restricted Additive Schwarz algorithm : 1/2 partition of unity u = 2 i=1 E i(χ i u i ) E i : extension from Ω i to Ω χ i : characteristic function in Ω i loop n = 0, 1, 2, = f in Ω i w n+1 i wi n+1 wi n+1 u n+1 = 2 i=1 E i(χ i w n+1 i ) = 0 on Ω i Ω = u n on Ω i Ω j by substituting u n that satisfies u n = 0 on Ω i Ω w n+1 i + u n = f + u n in Ω i wi n+1 u n = 0 on Ω i Ω, wi n+1 u n = 0 on Ω i Ω j u n+1 u n = 2 i=1 E i(χ i w n+1 i ) 2 i=1 E i(χ i u n i ) = 2 i=1 E i(χ i (w n+1 i u n )) 14 / 37

15 15 / 37 Restricted Additive Schwarz algorithm : 2/2 Restricted additive Schwarz (RAS) algorithm u 0 1, u0 2 : given loop n = 0, 1, 2, r n = f + u n vi n+1 = r n in Ω i = 0 on Ω i u n+1 = u n + 2 i=1 v n+1 i E i (χ i v n+1 i ) RAS Jacobi-Schwarz method proof by induction (u n + v n i ) = (u n ) + r n = f in Ω i u n + v n i = u n on Ω i Ω. u n = E 1 (χ 1 u n 1 ) + E 2 (χ 2 u n 2 ) u n = E 1 (0 u n 1 ) + E 2 (1 u n 2 ) = u n 2 on Ω i Ω 2.

16 16 / 37 Additive Schwarz algorithm loop n = 0, 1, 2, wi n+1 wi n+1 wi n+1 u n+1 = 2 i=1 E i(w n+1 i ) = f in Ω i = 0 on Ω i Ω = u n on Ω i Ω j Additive Schwarz Method (ASM) u 0 1, u0 2 : given loop n = 0, 1, 2, r n = f + u n vi n+1 = r n in Ω i = 0 on Ω i u n+1 = u n + 2 i=1 v n+1 i E i (v n+1 i )

17 preconditioned fixed-point iteration cf. V. Dolean, P Jolivet, F. Nataf, An Introduction to Domain Decomposition Methods Algorithms, Theory, and Parallel Implementation, SIAM, ISBN M: preconditioner Ax = b, Mx n+1 = Mx n + (b Ax n ) x n+1 = x n + M 1 (b Ax n ). by setting P = M A, x n+1 = x n + M 1 (b Ax n ) = (I M 1 (M P ))x n + M 1 b = M 1 P x n + M 1 b = x + M 1 P (x n x) x n+1 = x 0 + (M 1 P ) i M 1 (b Ax 0 ) 1 i n x n+1 x 0 span[m 1 r 0, (M 1 P )M 1 r 0, (M 1 P ) n M 1 r 0 ] x n+1 x 0 = s n (M 1 P )r 0 s n (t) = 1 + t + + t n : polynomial Krylov subspace method preconditioned fixed-point iteration 17 / 37

18 conjugate gradient method A u = f. preconditioner Q A 1 Krylov subspace : K n (Q r 0, QA) = span[q r 0, QAQ r 0,..., (QA) n 1 Q r 0 ] Find u n K n (Q r 0, QA) + u 0 s.t. (A u n f, v) = 0 v K n (Q r 0, QA). Preconditioned CG method u 0 : initial step for CG. r 0 = f A u 0 p 0 = Q r 0. loop n = 0, 1,... α n = (Q r n, r n )/(A p n, p n ), u n+1 = u n + α n p n, r n+1 = r n α n A p n, if r n+1 < ɛ exit loop. β n = (Q r n+1, r n+1 )/(Q r n, r n ), p n+1 = Q r n+1 + β n p n. LinearCG(opA,u,f,precon=opQ,nbiter=100,eps=1.0e-10) 18 / 37

19 GMRES method : 1/2 Krylov subspace : K n ( r 0, A) = span[ r 0, A r 0,..., A n 1 r 0 ] Find u n K n ( r 0, A) + u 0 s.t. A u n f A v n f v K n ( r 0, A) + u 0. V m : Arnoldi basis generated by Gram-Schmidt orthogonization for Krylov vectors. u = V m y, y R m J( y) := AV m y r 0 = V T m+1(av m y r 0 ) = (V T m+1av m ) y (V T m+1 r 0 ) = H m y β e 1. (β = r 0 ) Find y R m J( y) J( z) z R m. minimization problem with Hessenberg matrix H m R (m+1) m is solved by Givens rotation. LinearGMRES(opA,u,f,precon=opQ,nbiter=100, eps=1.0e-10) 19 / 37

20 GMRES method : 2/2 Arnoldi method (Gram-Schmidt method on Krylov subspace) v 1 = 1; do j = 1, 2,..., m do i = 1, 2,..., j w j := A v j 1 i j h i j v i, h i j := (A v j, v i ) v j+1 := w j /h j+1 j, h j+1 j := w j Givens rotation matrices Ω i R (m+1) (m+1) Ω i := I i 1 c i s i, c s i c 1 := i Im i Q m := Ω m Ω m 1 Ω 1 R (m+1) (m+1), R m := Q m Hm : upper triangular, ḡ m := Q [ m (βe 1 ) = ] [ γ 1 γ 2 γ m+1 ] T, Rm R m := (R 0 0 m R m m ), ḡ m := h 1 1, s h h 2 1 := 2 2 [ gm γ m+1 h 2 1. h h ] (g m R m ). min βe 1 H m y = min ḡ m R m y = γ m+1 = s 1 s 2 s m β. y m = Rm 1 g m attains the minimum. Remark : Rm 1 (1 m M) for all non-singular matrix A. 20 / 37

21 21 / 37 Schwarz methods as preconditioner ASM preconditioner M 1 M ASM = Rp T (R p ARp T ) 1 R p p=1 ASM does not converge as fixed point iteration, but M 1 ASM is symmetric and works well as a preconditioner for CG method. RAS preconditioner M 1 M RAS = Rp T (R p D p ARp T ) 1 R p p=1 RAS does converge but M 1 RAS is not symmetric and then works as a preconditioner for GMRES method. convergence : slow for many subdomains coarse space

22 22 / 37 2-level Schwarz methods with a coarse space coarse space by Nicolaides D p : discrete representation of the partition of unity M p=1 RT p D p R p = I N, { z p } R N : basis of coarse space, Z = [ z 1,, z M ]. R 0 = Z T. 2-level ASM preconditioner z p = R T p D p R p 1, M 1 ASM,2 = RT 0 (R 0 AR T 0 ) 1 R level RAS preconditioner M 1 RAS,2 = RT 0 (R 0 AR T 0 ) 1 R 0 + M Rp T (R p ARp T ) 1 R p p=1 M Rp T (R p D p ARp T ) 1 R p p=1

23 non-overlapping subdomains and Schur complement method non-overlapping domain decomposition : Ω = Ω 1 Ω 2, Γ = Ω 1 Ω 2, Λ = Λ 1 Λ 2 Λ 3 Λ D : decomposition of DOF, Ω 1, Ω 2, Γ, Ω. A 11 A 13 u 1 f 1 A 22 A 23 u 2 = f 2 + B.C. u = g A 31 A 32 A 33 u 3 f 3 interface problem : S 33 u 3 = (A 33 A 31 A 1 11 A 13 A 32 A 1 22 A 23) u 3 = f 3 A 31 A 1 11 f 1 A 32 A22 1 f 2 subdomain solver: [ A11 A 13 I 3 ] [ ] [ ] w1 0 = w 3 u 3 Dirichlet to Neumann map: u 3 v 3 = (A (1) 33 A 31A 1 11 A 13) u 3 [ ] [ ] [ ] v1 A11 A 13 w1 = v 3 A 13 A (1) w / 37

24 Neumann-Neumann preconditioner S 33 = (A (1) 33 A 31A 1 11 A 13) + (A (2) 33 A 32A 1 22 A 23) Interface problem : Sym. Positive Preconditioned CG Neumann subproblem: [ ] [ ] [ ] A11 A 13 w1 g1 A 13 A (1) = w 33 3 g 3 Note : Neumann problem may be singular in floating subdomain : Ω i Ω =. 1-level Neumann-Neumann preconditioner Q NN = (A (1) 33 A 31A 1 11 A 13) 1 + (A (2) 33 A 32A 1 22 A 23) 1 many subdomains : Λ = Λ I Λ B = (Λ (p) I Λ (p) B ) Q NN = M p=1 RT p D p (A (p) BB A(p BI A(p) II 1 A (p) IB ) D p R p condition nunmer of Q NN S depends on # of subdomains. 24 / 37

25 Balancing Neumann-Neumann preconditioner : 1/3 cf. J. Mandel, Balancing domain decomposition. Commun. Numer. Meth. Engng., 9: DOI: /cnm V : DOF on the skeleton, S : V V, sym. positive definite W : coarse space constructed by vectors Z p in each subdomain including kernel of Neumann problem : [ ] A11 A 13 spanz p Ker A 13 A (1) 33 coarse space on the skeleton M W = { v V ; v = Rp T D p u p, u p spanz p } p=1 P : S-orthogonal projection onto W v = P u : find v W (S( v u), w) = 0 w W 2-level Neumann-Neumann preconditioner Q BNN = (I P )Q NN (I P T ) + P S 1 25 / 37

26 26 / 37 Balancing Neumann-Neumann preconditioner : 2/3 P : S-orthogonal projection onto W coarse grid solver u : given, find v W (S v, w) = (S u, w) find ν q R dimzq Zp T D p R p SRq T D q Z q ν q = Zp T D p R p S u 1 q M p v = 1 q M ZT q D q R q ν q z = Q BNN r find λ q Zp T D p R p SRq T D q Z q λq = Zp T D p R p r p 1 q M s = r S 1 q M RT q D q Z q λq, s p = D p R p s find u p S p u p = s p find µ q Zp T ( D p R p r SR T q D q ( u q + Z q µ q ) ) = 0 1 q M z = 1 q M RT q D q ( u q + Z q µ q ) p

27 Balancing Neumann-Neumann preconditioner : 3/3 z = S 1 r find v W (S( v z ), w) = 0 w W v = P z. s = r S v = S( z v) = S(I P ) z v W u = Q NN s + v find y W (S( z Q NN s ( y + v )), w) = 0 w W y + v = P ( z Q NN s) z = u + y = Q NN s + y + v = Q NN s + P ( z Q NN s) = (I P )Q NN s + P z = (I P )Q NN S(I P ) z + P z = ((I P )Q NN S(I P ) + P ) S 1 r R 0 = [R T 1 D 1Z 1,, R T M D MZ M ] T : restriction operator V W P = R T 0 (R 0 SR T 0 ) 1 R 0 S, P S 1 = R T 0 (R 0 SR T 0 ) 1 R 0, P 2 = R T 0 (R 0 SR T 0 ) 1 R 0 SR T 0 (R 0 SR T 0 ) 1 R 0 S = P, SP = S R T 0 (R 0 SR T 0 ) 1 R 0 S = P T S. 27 / 37

28 28 / 37 Estimation of condition number of BDD Theorem C = sup { 1 q M R q 1 p M RT p D p u p 2 S p 1 p M u p 2 ; S q u p KerS p u p spanz Sp p }. (S u, u) (SM 1 Su, u) C(Su, u) cond(m, S) C. Theorem T pˆx T p ŷ C 1 H ˆx ŷ ˆx, ŷ ˆΩ Tp 1 x Tp 1 y C 2 H 1 x y x, y Ω p Ω p is mapped by T p from a reference square ˆΩ, H = max Ω p. Theorem cond(m, S) c(1 + log 2 H h )

29 p non-overlapping subdomains and Lagrange multiplier : 1/4 cf. C. Farhat, F.-X. Roux., Implicit parallel processing in structural mechanics, Computational Mechanics Advances, 2 (1994) A p u p = f p Bp T λ p Ω B p u p = 0 u p Γ pq = u q Γ (p) Ω (q) pq p u p = A p(f p Bp T λ) + R p α p, span R p = KerA p [ ] [ ] [ A11 A A p = 12, A A 1 p = 11 0 A 1, R A 21 A p = 11 A ] 12 I 2 compatibility condition for external force: R T p (A p u p ) = R T p (f p B T p λ) = 0 continuity on the interface: ( Bp A p(f p Bp T ) λ) + B p R p α p = 0 29 / 37 λ

30 30 / 37 non-overlapping subdomains and Lagrange multiplier : 2/4 p B p A pb T p λ + p B p R p α p = p B p A pf p R T p B T p λ = R T p f p by setting F = p B pa pb T p, G = [B 1 R 1, B 2 R 2,, B M R M ] [ ] [ ] F G λ G T = 0 {α p } p [ fλ f α ] f λ = p B pa pf p f α, p = R T p f p F : positive semi-definite due to cross point of decomposition originally non-floating structure G has full column rank proof: assume that G does not have full column rank γ = {γ p } 0 s.t. p B P R p γ p = 0. Ω p Ω, 0 Dirichlet b.c. R p γ p = 0 then R q γ q = 0 q.

31 non-overlapping subdomains and Lagrange multiplier : 3/4 P : orthogonal projection onto KerG T, P = I G(G T G) 1 G T. reformulating interface problem of Lagrange multiplier on KerG: find λ : [ ] [ ] [ ] F G λ + λ fλ = G T 0 [ F G G T 0 f α [ fλ α ] [ ] ] [ ] [ ] λ F G λ = α f α G T 0 0 [ ] [ ] fλ F λ = fλ F λ f α + G T = λ 0 λ = G(G T G) 1 [f T 1 R 1,, f T M R M] T. CG method on KerG T by the orthogonal projection P, P F λ = P (f λ F λ ). F λ = (f λ F λ ) + Gα α = (G T G) 1 G T (F (λ + λ ) f λ ). finally, u p = A p(f p B T p (λ + λ )) + R p α p = ū p + R p α p = ū p R p [(G T G) 1 G T q B qū q ] p. 31 / 37

32 conjugate gradient method with orthogonal projection orthogonal projection P : R N KerG T find u KerG T P A u = P f u 0 KerG T, r 0 = P f P A u 0 KerG T Krylov subspace : K n ( r 0, P A) = span[ r 0, P A r 0,..., (P A) n r 0 ] Find u n K n ( r 0, P A) + u 0 s.t. (A u n f, v) = 0 v K n ( r 0, P A). Projected CG method u 0 KerG T : initial step for CG. r 0 = P ( f A u 0 ) p 0 = r 0. loop n = 0, 1,... α n = ( r n, r n )/(P A p n, p n ), u n+1 = u n + α n p n, r n+1 = r n α n P A p n, if r n+1 < ɛ exit loop. β n = ( r n+1, r n+1 )/( r n, r n ), p n+1 = r n+1 + β n p n. 32 / 37

33 non-overlapping subdomains and Lagrange multiplier : 4/4 F = p B pa pb T p is positive semi-definite cross point, e.g. Q; B q u q (Q) + B p u p (Q) = 0 B p u p (Q) + B r u r (Q) = 0 B r u r (Q) + B s u s (Q) = 0 B s u s (Q) + B q u q (Q) = 0 removing redundant DOF of Lagrange multiplier: Ω (r) Ω (s) Q Ω (p) Ω (q) λ f λ = p B pa pf p ImF guarantees convergence of CG in ImF. 33 / 37

34 preconditioned CG method with orthogonal projection : 1/2 orthogonal projection P : R N KerG T find u KerG T P A u = P f, u 0 KerG T, r 0 = P f P A u 0 Q : preconditioner, Q A 1. Krylov subspace : K n ((P Q) r 0, P QP A) = span[(p Q) r 0, (P QP A)(P Q) r 0,..., (P QP A) n (P Q) r 0 ] Find u n K n ((P Q) r 0, P QP A) + u 0 s.t. (A u n f, v) = 0 v K n ((P Q) r 0, P QP A). Projected PCG method u 0 KerG T : initial step for CG. r 0 = P ( f A u 0 ) p 0 = P Q r 0. loop n = 0, 1,... α n = (P Q r n, r n )/(P A p n, p n ), u n+1 = u n + α n p n, r n+1 = r n α n P A p n, if r n+1 < ɛ exit loop. β n = (P Q r n+1, r n+1 )/(P Q r n, r n ), p n+1 = P Q r n+1 + β n p n. 34 / 37

35 35 / 37 preconditioned CG method with orthogonal projection : 2/2 Dirichlet preconditioner Q D = [ ] 0 0 B p Bp T, 0 S bb p S bb = A bb A bi A 1 ii A ib lumped preconditioner Q L = p [ ] 0 0 B p Bp T 0 A bb FETI method converges fast when the domain is cut with smooth interafce segments. Dirichlet preconditioner is better than lumped one but two local solvers are necessary.

36 36 / 37 extension of FreeFem++ by dynamic loading capability dynamic loading functions are in examples++-load source is written by C++ and called from edp script by using load "PARDISO" etc. several capabilities are extended without changing the original FreeFem++ sources linear solver : PARDISO, MUMPS mesh decomposer : metis, scotch finite elements : Element_P3, Element_Mixte output : VTK_writer The simplest examples in examples++-load myfunction.cpp and PARDISO.cpp

37 PARDISO in MKL by Intel void pardiso(_mkl_dss_handle_t pt,// pardiso handler MKL_INT *maxfct, // #of factorization(=1) MKL_INT *mnum, // 1<=mnum<=maxfct (=1) MKL_INT *mtype, // MKL_INT *phase, // MKL_INT *n, // size of matrix void *a, // coefficients MKL_INT *ia, // pointer where i-th row MKL_INT *ja, // column index of CSR MKL_INT *perm, MKL_INT *nrhs, // #of RHS vectors MKL_INT *iparm, MKL_INT *msglv, // message level void *b, // RHS void *x, // solution MKL_INT *error); mtype real structurally symm. SPD sym. indefinite phase symbolic factorization symbolic + numeric fw/bw solve 37 / 37

Toward black-box adaptive domain decomposition methods

Toward black-box adaptive domain decomposition methods Toward black-box adaptive domain decomposition methods Frédéric Nataf Laboratory J.L. Lions (LJLL), CNRS, Alpines Inria and Univ. Paris VI joint work with Victorita Dolean (Univ. Nice Sophia-Antipolis)

More information

Une méthode parallèle hybride à deux niveaux interfacée dans un logiciel d éléments finis

Une méthode parallèle hybride à deux niveaux interfacée dans un logiciel d éléments finis Une méthode parallèle hybride à deux niveaux interfacée dans un logiciel d éléments finis Frédéric Nataf Laboratory J.L. Lions (LJLL), CNRS, Alpines Inria and Univ. Paris VI Victorita Dolean (Univ. Nice

More information

Indefinite and physics-based preconditioning

Indefinite and physics-based preconditioning Indefinite and physics-based preconditioning Jed Brown VAW, ETH Zürich 2009-01-29 Newton iteration Standard form of a nonlinear system F (u) 0 Iteration Solve: Update: J(ũ)u F (ũ) ũ + ũ + u Example (p-bratu)

More information

Recent advances in HPC with FreeFem++

Recent advances in HPC with FreeFem++ Recent advances in HPC with FreeFem++ Pierre Jolivet Laboratoire Jacques-Louis Lions Laboratoire Jean Kuntzmann Fourth workshop on FreeFem++ December 6, 2012 With F. Hecht, F. Nataf, C. Prud homme. Outline

More information

Some examples in domain decomposition

Some examples in domain decomposition Some examples in domain decomposition Frédéric Nataf nataf@ann.jussieu.fr, www.ann.jussieu.fr/ nataf Laboratoire J.L. Lions, CNRS UMR7598. Université Pierre et Marie Curie, France Joint work with V. Dolean

More information

Adaptive Coarse Space Selection in BDDC and FETI-DP Iterative Substructuring Methods: Towards Fast and Robust Solvers

Adaptive Coarse Space Selection in BDDC and FETI-DP Iterative Substructuring Methods: Towards Fast and Robust Solvers Adaptive Coarse Space Selection in BDDC and FETI-DP Iterative Substructuring Methods: Towards Fast and Robust Solvers Jan Mandel University of Colorado at Denver Bedřich Sousedík Czech Technical University

More information

Optimal Left and Right Additive Schwarz Preconditioning for Minimal Residual Methods with Euclidean and Energy Norms

Optimal Left and Right Additive Schwarz Preconditioning for Minimal Residual Methods with Euclidean and Energy Norms Optimal Left and Right Additive Schwarz Preconditioning for Minimal Residual Methods with Euclidean and Energy Norms Marcus Sarkis Worcester Polytechnic Inst., Mass. and IMPA, Rio de Janeiro and Daniel

More information

AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 23: GMRES and Other Krylov Subspace Methods; Preconditioning

AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 23: GMRES and Other Krylov Subspace Methods; Preconditioning AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 23: GMRES and Other Krylov Subspace Methods; Preconditioning Xiangmin Jiao SUNY Stony Brook Xiangmin Jiao Numerical Analysis I 1 / 18 Outline

More information

An introduction to Schwarz methods

An introduction to Schwarz methods An introduction to Schwarz methods Victorita Dolean Université de Nice and University of Geneva École thematique CNRS Décomposition de domaine November 15-16, 2011 Outline 1 Introduction 2 Schwarz algorithms

More information

Parallelism in FreeFem++.

Parallelism in FreeFem++. Parallelism in FreeFem++. Guy Atenekeng 1 Frederic Hecht 2 Laura Grigori 1 Jacques Morice 2 Frederic Nataf 2 1 INRIA, Saclay 2 University of Paris 6 Workshop on FreeFem++, 2009 Outline 1 Introduction Motivation

More information

Topics. The CG Algorithm Algorithmic Options CG s Two Main Convergence Theorems

Topics. The CG Algorithm Algorithmic Options CG s Two Main Convergence Theorems Topics The CG Algorithm Algorithmic Options CG s Two Main Convergence Theorems What about non-spd systems? Methods requiring small history Methods requiring large history Summary of solvers 1 / 52 Conjugate

More information

A dissection solver with kernel detection for unsymmetric matrices in FreeFem++

A dissection solver with kernel detection for unsymmetric matrices in FreeFem++ . p.1/21 11 Dec. 2014, LJLL, Paris FreeFem++ workshop A dissection solver with kernel detection for unsymmetric matrices in FreeFem++ Atsushi Suzuki Atsushi.Suzuki@ann.jussieu.fr Joint work with François-Xavier

More information

Conjugate gradient method. Descent method. Conjugate search direction. Conjugate Gradient Algorithm (294)

Conjugate gradient method. Descent method. Conjugate search direction. Conjugate Gradient Algorithm (294) Conjugate gradient method Descent method Hestenes, Stiefel 1952 For A N N SPD In exact arithmetic, solves in N steps In real arithmetic No guaranteed stopping Often converges in many fewer than N steps

More information

OUTLINE ffl CFD: elliptic pde's! Ax = b ffl Basic iterative methods ffl Krylov subspace methods ffl Preconditioning techniques: Iterative methods ILU

OUTLINE ffl CFD: elliptic pde's! Ax = b ffl Basic iterative methods ffl Krylov subspace methods ffl Preconditioning techniques: Iterative methods ILU Preconditioning Techniques for Solving Large Sparse Linear Systems Arnold Reusken Institut für Geometrie und Praktische Mathematik RWTH-Aachen OUTLINE ffl CFD: elliptic pde's! Ax = b ffl Basic iterative

More information

Multilevel spectral coarse space methods in FreeFem++ on parallel architectures

Multilevel spectral coarse space methods in FreeFem++ on parallel architectures Multilevel spectral coarse space methods in FreeFem++ on parallel architectures Pierre Jolivet Laboratoire Jacques-Louis Lions Laboratoire Jean Kuntzmann DD 21, Rennes. June 29, 2012 In collaboration with

More information

Two-level Domain decomposition preconditioning for the high-frequency time-harmonic Maxwell equations

Two-level Domain decomposition preconditioning for the high-frequency time-harmonic Maxwell equations Two-level Domain decomposition preconditioning for the high-frequency time-harmonic Maxwell equations Marcella Bonazzoli 2, Victorita Dolean 1,4, Ivan G. Graham 3, Euan A. Spence 3, Pierre-Henri Tournier

More information

Multipréconditionnement adaptatif pour les méthodes de décomposition de domaine. Nicole Spillane (CNRS, CMAP, École Polytechnique)

Multipréconditionnement adaptatif pour les méthodes de décomposition de domaine. Nicole Spillane (CNRS, CMAP, École Polytechnique) Multipréconditionnement adaptatif pour les méthodes de décomposition de domaine Nicole Spillane (CNRS, CMAP, École Polytechnique) C. Bovet (ONERA), P. Gosselet (ENS Cachan), A. Parret Fréaud (SafranTech),

More information

Some Geometric and Algebraic Aspects of Domain Decomposition Methods

Some Geometric and Algebraic Aspects of Domain Decomposition Methods Some Geometric and Algebraic Aspects of Domain Decomposition Methods D.S.Butyugin 1, Y.L.Gurieva 1, V.P.Ilin 1,2, and D.V.Perevozkin 1 Abstract Some geometric and algebraic aspects of various domain decomposition

More information

Short title: Total FETI. Corresponding author: Zdenek Dostal, VŠB-Technical University of Ostrava, 17 listopadu 15, CZ Ostrava, Czech Republic

Short title: Total FETI. Corresponding author: Zdenek Dostal, VŠB-Technical University of Ostrava, 17 listopadu 15, CZ Ostrava, Czech Republic Short title: Total FETI Corresponding author: Zdenek Dostal, VŠB-Technical University of Ostrava, 17 listopadu 15, CZ-70833 Ostrava, Czech Republic mail: zdenek.dostal@vsb.cz fax +420 596 919 597 phone

More information

Multispace and Multilevel BDDC. Jan Mandel University of Colorado at Denver and Health Sciences Center

Multispace and Multilevel BDDC. Jan Mandel University of Colorado at Denver and Health Sciences Center Multispace and Multilevel BDDC Jan Mandel University of Colorado at Denver and Health Sciences Center Based on joint work with Bedřich Sousedík, UCDHSC and Czech Technical University, and Clark R. Dohrmann,

More information

An Efficient FETI Implementation on Distributed Shared Memory Machines with Independent Numbers of Subdomains and Processors

An Efficient FETI Implementation on Distributed Shared Memory Machines with Independent Numbers of Subdomains and Processors Contemporary Mathematics Volume 218, 1998 B 0-8218-0988-1-03024-7 An Efficient FETI Implementation on Distributed Shared Memory Machines with Independent Numbers of Subdomains and Processors Michel Lesoinne

More information

ASM-BDDC Preconditioners with variable polynomial degree for CG- and DG-SEM

ASM-BDDC Preconditioners with variable polynomial degree for CG- and DG-SEM ASM-BDDC Preconditioners with variable polynomial degree for CG- and DG-SEM C. Canuto 1, L. F. Pavarino 2, and A. B. Pieri 3 1 Introduction Discontinuous Galerkin (DG) methods for partial differential

More information

SOLVING SPARSE LINEAR SYSTEMS OF EQUATIONS. Chao Yang Computational Research Division Lawrence Berkeley National Laboratory Berkeley, CA, USA

SOLVING SPARSE LINEAR SYSTEMS OF EQUATIONS. Chao Yang Computational Research Division Lawrence Berkeley National Laboratory Berkeley, CA, USA 1 SOLVING SPARSE LINEAR SYSTEMS OF EQUATIONS Chao Yang Computational Research Division Lawrence Berkeley National Laboratory Berkeley, CA, USA 2 OUTLINE Sparse matrix storage format Basic factorization

More information

Master Thesis Literature Study Presentation

Master Thesis Literature Study Presentation Master Thesis Literature Study Presentation Delft University of Technology The Faculty of Electrical Engineering, Mathematics and Computer Science January 29, 2010 Plaxis Introduction Plaxis Finite Element

More information

On the choice of abstract projection vectors for second level preconditioners

On the choice of abstract projection vectors for second level preconditioners On the choice of abstract projection vectors for second level preconditioners C. Vuik 1, J.M. Tang 1, and R. Nabben 2 1 Delft University of Technology 2 Technische Universität Berlin Institut für Mathematik

More information

Preface to the Second Edition. Preface to the First Edition

Preface to the Second Edition. Preface to the First Edition n page v Preface to the Second Edition Preface to the First Edition xiii xvii 1 Background in Linear Algebra 1 1.1 Matrices................................. 1 1.2 Square Matrices and Eigenvalues....................

More information

The All-floating BETI Method: Numerical Results

The All-floating BETI Method: Numerical Results The All-floating BETI Method: Numerical Results Günther Of Institute of Computational Mathematics, Graz University of Technology, Steyrergasse 30, A-8010 Graz, Austria, of@tugraz.at Summary. The all-floating

More information

18. Balancing Neumann-Neumann for (In)Compressible Linear Elasticity and (Generalized) Stokes Parallel Implementation

18. Balancing Neumann-Neumann for (In)Compressible Linear Elasticity and (Generalized) Stokes Parallel Implementation Fourteenth nternational Conference on Domain Decomposition Methods Editors: smael Herrera, David E Keyes, Olof B Widlund, Robert Yates c 23 DDMorg 18 Balancing Neumann-Neumann for (n)compressible Linear

More information

A Balancing Algorithm for Mortar Methods

A Balancing Algorithm for Mortar Methods A Balancing Algorithm for Mortar Methods Dan Stefanica Baruch College, City University of New York, NY 11, USA Dan Stefanica@baruch.cuny.edu Summary. The balancing methods are hybrid nonoverlapping Schwarz

More information

Preconditioning of Saddle Point Systems by Substructuring and a Penalty Approach

Preconditioning of Saddle Point Systems by Substructuring and a Penalty Approach Preconditioning of Saddle Point Systems by Substructuring and a Penalty Approach Clark R. Dohrmann 1 Sandia National Laboratories, crdohrm@sandia.gov. Sandia is a multiprogram laboratory operated by Sandia

More information

Lecture 9 Approximations of Laplace s Equation, Finite Element Method. Mathématiques appliquées (MATH0504-1) B. Dewals, C.

Lecture 9 Approximations of Laplace s Equation, Finite Element Method. Mathématiques appliquées (MATH0504-1) B. Dewals, C. Lecture 9 Approximations of Laplace s Equation, Finite Element Method Mathématiques appliquées (MATH54-1) B. Dewals, C. Geuzaine V1.2 23/11/218 1 Learning objectives of this lecture Apply the finite difference

More information

From Direct to Iterative Substructuring: some Parallel Experiences in 2 and 3D

From Direct to Iterative Substructuring: some Parallel Experiences in 2 and 3D From Direct to Iterative Substructuring: some Parallel Experiences in 2 and 3D Luc Giraud N7-IRIT, Toulouse MUMPS Day October 24, 2006, ENS-INRIA, Lyon, France Outline 1 General Framework 2 The direct

More information

Numerical simulation of the Gross-Pitaevskii equation by pseudo-spectral and finite element methods comparison of GPS code and FreeFem++

Numerical simulation of the Gross-Pitaevskii equation by pseudo-spectral and finite element methods comparison of GPS code and FreeFem++ . p.1/13 10 ec. 2014, LJLL, Paris FreeFem++ workshop : BECASIM session Numerical simulation of the Gross-Pitaevskii equation by pseudo-spectral and finite element methods comparison of GPS code and FreeFem++

More information

ITERATIVE METHODS BASED ON KRYLOV SUBSPACES

ITERATIVE METHODS BASED ON KRYLOV SUBSPACES ITERATIVE METHODS BASED ON KRYLOV SUBSPACES LONG CHEN We shall present iterative methods for solving linear algebraic equation Au = b based on Krylov subspaces We derive conjugate gradient (CG) method

More information

20. A Dual-Primal FETI Method for solving Stokes/Navier-Stokes Equations

20. A Dual-Primal FETI Method for solving Stokes/Navier-Stokes Equations Fourteenth International Conference on Domain Decomposition Methods Editors: Ismael Herrera, David E. Keyes, Olof B. Widlund, Robert Yates c 23 DDM.org 2. A Dual-Primal FEI Method for solving Stokes/Navier-Stokes

More information

Multilevel and Adaptive Iterative Substructuring Methods. Jan Mandel University of Colorado Denver

Multilevel and Adaptive Iterative Substructuring Methods. Jan Mandel University of Colorado Denver Multilevel and Adaptive Iterative Substructuring Methods Jan Mandel University of Colorado Denver The multilevel BDDC method is joint work with Bedřich Sousedík, Czech Technical University, and Clark Dohrmann,

More information

Dissection sparse direct solver for indefinite finite element matrices and application to a semi-conductor problem

Dissection sparse direct solver for indefinite finite element matrices and application to a semi-conductor problem 10th FreeFem++ days 2018, 13 Dec. Dissection sparse direct solver for indefinite finite element matrices and application to a semi-conductor problem Atsushi Suzuki 1 1 Cybermedia Center, Osaka University

More information

The quadratic eigenvalue problem (QEP) is to find scalars λ and nonzero vectors u satisfying

The quadratic eigenvalue problem (QEP) is to find scalars λ and nonzero vectors u satisfying I.2 Quadratic Eigenvalue Problems 1 Introduction The quadratic eigenvalue problem QEP is to find scalars λ and nonzero vectors u satisfying where Qλx = 0, 1.1 Qλ = λ 2 M + λd + K, M, D and K are given

More information

Parallel Numerics, WT 2016/ Iterative Methods for Sparse Linear Systems of Equations. page 1 of 1

Parallel Numerics, WT 2016/ Iterative Methods for Sparse Linear Systems of Equations. page 1 of 1 Parallel Numerics, WT 2016/2017 5 Iterative Methods for Sparse Linear Systems of Equations page 1 of 1 Contents 1 Introduction 1.1 Computer Science Aspects 1.2 Numerical Problems 1.3 Graphs 1.4 Loop Manipulations

More information

Domain Decomposition Algorithms for an Indefinite Hypersingular Integral Equation in Three Dimensions

Domain Decomposition Algorithms for an Indefinite Hypersingular Integral Equation in Three Dimensions Domain Decomposition Algorithms for an Indefinite Hypersingular Integral Equation in Three Dimensions Ernst P. Stephan 1, Matthias Maischak 2, and Thanh Tran 3 1 Institut für Angewandte Mathematik, Leibniz

More information

Algebra C Numerical Linear Algebra Sample Exam Problems

Algebra C Numerical Linear Algebra Sample Exam Problems Algebra C Numerical Linear Algebra Sample Exam Problems Notation. Denote by V a finite-dimensional Hilbert space with inner product (, ) and corresponding norm. The abbreviation SPD is used for symmetric

More information

Efficient domain decomposition methods for the time-harmonic Maxwell equations

Efficient domain decomposition methods for the time-harmonic Maxwell equations Efficient domain decomposition methods for the time-harmonic Maxwell equations Marcella Bonazzoli 1, Victorita Dolean 2, Ivan G. Graham 3, Euan A. Spence 3, Pierre-Henri Tournier 4 1 Inria Saclay (Defi

More information

Lecture 8: Fast Linear Solvers (Part 7)

Lecture 8: Fast Linear Solvers (Part 7) Lecture 8: Fast Linear Solvers (Part 7) 1 Modified Gram-Schmidt Process with Reorthogonalization Test Reorthogonalization If Av k 2 + δ v k+1 2 = Av k 2 to working precision. δ = 10 3 2 Householder Arnoldi

More information

Domain Decomposition solvers (FETI)

Domain Decomposition solvers (FETI) Domain Decomposition solvers (FETI) a random walk in history and some current trends Daniel J. Rixen Technische Universität München Institute of Applied Mechanics www.amm.mw.tum.de rixen@tum.de 8-10 October

More information

ADDITIVE SCHWARZ FOR SCHUR COMPLEMENT 305 the parallel implementation of both preconditioners on distributed memory platforms, and compare their perfo

ADDITIVE SCHWARZ FOR SCHUR COMPLEMENT 305 the parallel implementation of both preconditioners on distributed memory platforms, and compare their perfo 35 Additive Schwarz for the Schur Complement Method Luiz M. Carvalho and Luc Giraud 1 Introduction Domain decomposition methods for solving elliptic boundary problems have been receiving increasing attention

More information

Contents. Preface... xi. Introduction...

Contents. Preface... xi. Introduction... Contents Preface... xi Introduction... xv Chapter 1. Computer Architectures... 1 1.1. Different types of parallelism... 1 1.1.1. Overlap, concurrency and parallelism... 1 1.1.2. Temporal and spatial parallelism

More information

Scalable Domain Decomposition Preconditioners For Heterogeneous Elliptic Problems

Scalable Domain Decomposition Preconditioners For Heterogeneous Elliptic Problems Scalable Domain Decomposition Preconditioners For Heterogeneous Elliptic Problems Pierre Jolivet, F. Hecht, F. Nataf, C. Prud homme Laboratoire Jacques-Louis Lions Laboratoire Jean Kuntzmann INRIA Rocquencourt

More information

Projected Schur Complement Method for Solving Non-Symmetric Saddle-Point Systems (Arising from Fictitious Domain Approaches)

Projected Schur Complement Method for Solving Non-Symmetric Saddle-Point Systems (Arising from Fictitious Domain Approaches) Projected Schur Complement Method for Solving Non-Symmetric Saddle-Point Systems (Arising from Fictitious Domain Approaches) Jaroslav Haslinger, Charles University, Prague Tomáš Kozubek, VŠB TU Ostrava

More information

Coupled FETI/BETI for Nonlinear Potential Problems

Coupled FETI/BETI for Nonlinear Potential Problems Coupled FETI/BETI for Nonlinear Potential Problems U. Langer 1 C. Pechstein 1 A. Pohoaţǎ 1 1 Institute of Computational Mathematics Johannes Kepler University Linz {ulanger,pechstein,pohoata}@numa.uni-linz.ac.at

More information

Deflated Krylov Iterations in Domain Decomposition Methods

Deflated Krylov Iterations in Domain Decomposition Methods 305 Deflated Krylov Iterations in Domain Decomposition Methods Y.L.Gurieva 1, V.P.Ilin 1,2, and D.V.Perevozkin 1 1 Introduction The goal of this research is an investigation of some advanced versions of

More information

4.8 Arnoldi Iteration, Krylov Subspaces and GMRES

4.8 Arnoldi Iteration, Krylov Subspaces and GMRES 48 Arnoldi Iteration, Krylov Subspaces and GMRES We start with the problem of using a similarity transformation to convert an n n matrix A to upper Hessenberg form H, ie, A = QHQ, (30) with an appropriate

More information

Algebraic two-level preconditioners. for the Schur complement method. L. M. Carvalho, L. Giraud and P. Le Tallec

Algebraic two-level preconditioners. for the Schur complement method. L. M. Carvalho, L. Giraud and P. Le Tallec Algebraic two-level preconditioners for the Schur complement method L. M. Carvalho, L. Giraud and P. Le Tallec June 1998 TR/PA/98/18 Algebraic two-level preconditioners for the Schur complement method

More information

HPDDM une bibliothèque haute performance unifiée pour les méthodes de décomposition de domaine

HPDDM une bibliothèque haute performance unifiée pour les méthodes de décomposition de domaine HPDDM une bibliothèque haute performance unifiée pour les méthodes de décomposition de domaine P. Jolivet and Frédéric Nataf Laboratory J.L. Lions, Univ. Paris VI, Equipe Alpines INRIA-LJLL et CNRS joint

More information

Applied Mathematics 205. Unit V: Eigenvalue Problems. Lecturer: Dr. David Knezevic

Applied Mathematics 205. Unit V: Eigenvalue Problems. Lecturer: Dr. David Knezevic Applied Mathematics 205 Unit V: Eigenvalue Problems Lecturer: Dr. David Knezevic Unit V: Eigenvalue Problems Chapter V.4: Krylov Subspace Methods 2 / 51 Krylov Subspace Methods In this chapter we give

More information

Adaptive Coarse Spaces and Multiple Search Directions: Tools for Robust Domain Decomposition Algorithms

Adaptive Coarse Spaces and Multiple Search Directions: Tools for Robust Domain Decomposition Algorithms Adaptive Coarse Spaces and Multiple Search Directions: Tools for Robust Domain Decomposition Algorithms Nicole Spillane Center for Mathematical Modelling at the Universidad de Chile in Santiago. July 9th,

More information

AMG for a Peta-scale Navier Stokes Code

AMG for a Peta-scale Navier Stokes Code AMG for a Peta-scale Navier Stokes Code James Lottes Argonne National Laboratory October 18, 2007 The Challenge Develop an AMG iterative method to solve Poisson 2 u = f discretized on highly irregular

More information

Domain decomposition on different levels of the Jacobi-Davidson method

Domain decomposition on different levels of the Jacobi-Davidson method hapter 5 Domain decomposition on different levels of the Jacobi-Davidson method Abstract Most computational work of Jacobi-Davidson [46], an iterative method suitable for computing solutions of large dimensional

More information

A Balancing Algorithm for Mortar Methods

A Balancing Algorithm for Mortar Methods A Balancing Algorithm for Mortar Methods Dan Stefanica Baruch College, City University of New York, NY, USA. Dan_Stefanica@baruch.cuny.edu Summary. The balancing methods are hybrid nonoverlapping Schwarz

More information

2.29 Numerical Fluid Mechanics Spring 2015 Lecture 9

2.29 Numerical Fluid Mechanics Spring 2015 Lecture 9 Spring 2015 Lecture 9 REVIEW Lecture 8: Direct Methods for solving (linear) algebraic equations Gauss Elimination LU decomposition/factorization Error Analysis for Linear Systems and Condition Numbers

More information

FEM and sparse linear system solving

FEM and sparse linear system solving FEM & sparse linear system solving, Lecture 9, Nov 19, 2017 1/36 Lecture 9, Nov 17, 2017: Krylov space methods http://people.inf.ethz.ch/arbenz/fem17 Peter Arbenz Computer Science Department, ETH Zürich

More information

Lecture 17: Iterative Methods and Sparse Linear Algebra

Lecture 17: Iterative Methods and Sparse Linear Algebra Lecture 17: Iterative Methods and Sparse Linear Algebra David Bindel 25 Mar 2014 Logistics HW 3 extended to Wednesday after break HW 4 should come out Monday after break Still need project description

More information

C. Vuik 1 R. Nabben 2 J.M. Tang 1

C. Vuik 1 R. Nabben 2 J.M. Tang 1 Deflation acceleration of block ILU preconditioned Krylov methods C. Vuik 1 R. Nabben 2 J.M. Tang 1 1 Delft University of Technology J.M. Burgerscentrum 2 Technical University Berlin Institut für Mathematik

More information

Auxiliary space multigrid method for elliptic problems with highly varying coefficients

Auxiliary space multigrid method for elliptic problems with highly varying coefficients Auxiliary space multigrid method for elliptic problems with highly varying coefficients Johannes Kraus 1 and Maria Lymbery 2 1 Introduction The robust preconditioning of linear systems of algebraic equations

More information

J.I. Aliaga 1 M. Bollhöfer 2 A.F. Martín 1 E.S. Quintana-Ortí 1. March, 2009

J.I. Aliaga 1 M. Bollhöfer 2 A.F. Martín 1 E.S. Quintana-Ortí 1. March, 2009 Parallel Preconditioning of Linear Systems based on ILUPACK for Multithreaded Architectures J.I. Aliaga M. Bollhöfer 2 A.F. Martín E.S. Quintana-Ortí Deparment of Computer Science and Engineering, Univ.

More information

cfl Jing Li All rights reserved, 22

cfl Jing Li All rights reserved, 22 NYU-CS-TR83 Dual-Primal FETI Methods for Stationary Stokes and Navier-Stokes Equations by Jing Li A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy

More information

Inexact Data-Sparse BETI Methods by Ulrich Langer. (joint talk with G. Of, O. Steinbach and W. Zulehner)

Inexact Data-Sparse BETI Methods by Ulrich Langer. (joint talk with G. Of, O. Steinbach and W. Zulehner) Inexact Data-Sparse BETI Methods by Ulrich Langer (joint talk with G. Of, O. Steinbach and W. Zulehner) Radon Institute for Computational and Applied Mathematics Austrian Academy of Sciences http://www.ricam.oeaw.ac.at

More information

Lab 1: Iterative Methods for Solving Linear Systems

Lab 1: Iterative Methods for Solving Linear Systems Lab 1: Iterative Methods for Solving Linear Systems January 22, 2017 Introduction Many real world applications require the solution to very large and sparse linear systems where direct methods such as

More information

A simple FEM solver and its data parallelism

A simple FEM solver and its data parallelism A simple FEM solver and its data parallelism Gundolf Haase Institute for Mathematics and Scientific Computing University of Graz, Austria Chile, Jan. 2015 Partial differential equation Considered Problem

More information

Chapter 7 Iterative Techniques in Matrix Algebra

Chapter 7 Iterative Techniques in Matrix Algebra Chapter 7 Iterative Techniques in Matrix Algebra Per-Olof Persson persson@berkeley.edu Department of Mathematics University of California, Berkeley Math 128B Numerical Analysis Vector Norms Definition

More information

Adaptive algebraic multigrid methods in lattice computations

Adaptive algebraic multigrid methods in lattice computations Adaptive algebraic multigrid methods in lattice computations Karsten Kahl Bergische Universität Wuppertal January 8, 2009 Acknowledgements Matthias Bolten, University of Wuppertal Achi Brandt, Weizmann

More information

Parallel sparse linear solvers and applications in CFD

Parallel sparse linear solvers and applications in CFD Parallel sparse linear solvers and applications in CFD Jocelyne Erhel Joint work with Désiré Nuentsa Wakam () and Baptiste Poirriez () SAGE team, Inria Rennes, France journée Calcul Intensif Distribué

More information

On deflation and singular symmetric positive semi-definite matrices

On deflation and singular symmetric positive semi-definite matrices Journal of Computational and Applied Mathematics 206 (2007) 603 614 www.elsevier.com/locate/cam On deflation and singular symmetric positive semi-definite matrices J.M. Tang, C. Vuik Faculty of Electrical

More information

AN INTRODUCTION TO DOMAIN DECOMPOSITION METHODS. Gérard MEURANT CEA

AN INTRODUCTION TO DOMAIN DECOMPOSITION METHODS. Gérard MEURANT CEA Marrakech Jan 2003 AN INTRODUCTION TO DOMAIN DECOMPOSITION METHODS Gérard MEURANT CEA Introduction Domain decomposition is a divide and conquer technique Natural framework to introduce parallelism in the

More information

Stabilization and Acceleration of Algebraic Multigrid Method

Stabilization and Acceleration of Algebraic Multigrid Method Stabilization and Acceleration of Algebraic Multigrid Method Recursive Projection Algorithm A. Jemcov J.P. Maruszewski Fluent Inc. October 24, 2006 Outline 1 Need for Algorithm Stabilization and Acceleration

More information

The Conjugate Gradient Method

The Conjugate Gradient Method The Conjugate Gradient Method Classical Iterations We have a problem, We assume that the matrix comes from a discretization of a PDE. The best and most popular model problem is, The matrix will be as large

More information

FEM and Sparse Linear System Solving

FEM and Sparse Linear System Solving FEM & sparse system solving, Lecture 7, Nov 3, 2017 1/46 Lecture 7, Nov 3, 2015: Introduction to Iterative Solvers: Stationary Methods http://people.inf.ethz.ch/arbenz/fem16 Peter Arbenz Computer Science

More information

The amount of work to construct each new guess from the previous one should be a small multiple of the number of nonzeros in A.

The amount of work to construct each new guess from the previous one should be a small multiple of the number of nonzeros in A. AMSC/CMSC 661 Scientific Computing II Spring 2005 Solution of Sparse Linear Systems Part 2: Iterative methods Dianne P. O Leary c 2005 Solving Sparse Linear Systems: Iterative methods The plan: Iterative

More information

CME342 Parallel Methods in Numerical Analysis. Matrix Computation: Iterative Methods II. Sparse Matrix-vector Multiplication.

CME342 Parallel Methods in Numerical Analysis. Matrix Computation: Iterative Methods II. Sparse Matrix-vector Multiplication. CME342 Parallel Methods in Numerical Analysis Matrix Computation: Iterative Methods II Outline: CG & its parallelization. Sparse Matrix-vector Multiplication. 1 Basic iterative methods: Ax = b r = b Ax

More information

An Introduction to Domain Decomposition Methods: algorithms, theory and parallel implementation

An Introduction to Domain Decomposition Methods: algorithms, theory and parallel implementation An Introduction to Domain Decomposition Methods: algorithms, theory and parallel implementation Victorita Dolean, Pierre Jolivet, Frédéric ataf To cite this version: Victorita Dolean, Pierre Jolivet, Frédéric

More information

A High-Performance Parallel Hybrid Method for Large Sparse Linear Systems

A High-Performance Parallel Hybrid Method for Large Sparse Linear Systems Outline A High-Performance Parallel Hybrid Method for Large Sparse Linear Systems Azzam Haidar CERFACS, Toulouse joint work with Luc Giraud (N7-IRIT, France) and Layne Watson (Virginia Polytechnic Institute,

More information

Domain Decomposition Preconditioners for Spectral Nédélec Elements in Two and Three Dimensions

Domain Decomposition Preconditioners for Spectral Nédélec Elements in Two and Three Dimensions Domain Decomposition Preconditioners for Spectral Nédélec Elements in Two and Three Dimensions Bernhard Hientzsch Courant Institute of Mathematical Sciences, New York University, 51 Mercer Street, New

More information

Architecture Solutions for DDM Numerical Environment

Architecture Solutions for DDM Numerical Environment Architecture Solutions for DDM Numerical Environment Yana Gurieva 1 and Valery Ilin 1,2 1 Institute of Computational Mathematics and Mathematical Geophysics SB RAS, Novosibirsk, Russia 2 Novosibirsk State

More information

4.6 Iterative Solvers for Linear Systems

4.6 Iterative Solvers for Linear Systems 4.6 Iterative Solvers for Linear Systems Why use iterative methods? Virtually all direct methods for solving Ax = b require O(n 3 ) floating point operations. In practical applications the matrix A often

More information

Incomplete Cholesky preconditioners that exploit the low-rank property

Incomplete Cholesky preconditioners that exploit the low-rank property anapov@ulb.ac.be ; http://homepages.ulb.ac.be/ anapov/ 1 / 35 Incomplete Cholesky preconditioners that exploit the low-rank property (theory and practice) Artem Napov Service de Métrologie Nucléaire, Université

More information

The Deflation Accelerated Schwarz Method for CFD

The Deflation Accelerated Schwarz Method for CFD The Deflation Accelerated Schwarz Method for CFD J. Verkaik 1, C. Vuik 2,, B.D. Paarhuis 1, and A. Twerda 1 1 TNO Science and Industry, Stieltjesweg 1, P.O. Box 155, 2600 AD Delft, The Netherlands 2 Delft

More information

The Lanczos and conjugate gradient algorithms

The Lanczos and conjugate gradient algorithms The Lanczos and conjugate gradient algorithms Gérard MEURANT October, 2008 1 The Lanczos algorithm 2 The Lanczos algorithm in finite precision 3 The nonsymmetric Lanczos algorithm 4 The Golub Kahan bidiagonalization

More information

Multilevel low-rank approximation preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota

Multilevel low-rank approximation preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota Multilevel low-rank approximation preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota SIAM CSE Boston - March 1, 2013 First: Joint work with Ruipeng Li Work

More information

An Introduction to Domain Decomposition Methods

An Introduction to Domain Decomposition Methods An Introduction to Domain Decomposition Methods Algorithms, Theory, and Parallel Implementation Victorita Dolean Pierre Jolivet Frédéric Nataf An Introduction to Domain Decomposition Methods An Introduction

More information

Finite element programming by FreeFem++ intermediate course

Finite element programming by FreeFem++ intermediate course 1 / 73 Finite element programming by FreeFem++ intermediate course Atsushi Suzuki 1 1 Cybermedia Center, Osaka University atsushi.suzuki@cas.cmc.osaka-u.ac.jp Japan SIAM tutorial 11-12, February 2016 2

More information

Linear Solvers. Andrew Hazel

Linear Solvers. Andrew Hazel Linear Solvers Andrew Hazel Introduction Thus far we have talked about the formulation and discretisation of physical problems...... and stopped when we got to a discrete linear system of equations. Introduction

More information

FETI-DPH: A DUAL-PRIMAL DOMAIN DECOMPOSITION METHOD FOR ACOUSTIC SCATTERING

FETI-DPH: A DUAL-PRIMAL DOMAIN DECOMPOSITION METHOD FOR ACOUSTIC SCATTERING Journal of Computational Acoustics, c IMACS FETI-DPH: A DUAL-PRIMAL DOMAIN DECOMPOSITION METHOD FOR ACOUSTIC SCATTERING Charbel Farhat, Philip Avery and Radek Tezaur Department of Mechanical Engineering

More information

Computational Linear Algebra

Computational Linear Algebra Computational Linear Algebra PD Dr. rer. nat. habil. Ralf-Peter Mundani Computation in Engineering / BGU Scientific Computing in Computer Science / INF Winter Term 2018/19 Part 4: Iterative Methods PD

More information

Solving Symmetric Indefinite Systems with Symmetric Positive Definite Preconditioners

Solving Symmetric Indefinite Systems with Symmetric Positive Definite Preconditioners Solving Symmetric Indefinite Systems with Symmetric Positive Definite Preconditioners Eugene Vecharynski 1 Andrew Knyazev 2 1 Department of Computer Science and Engineering University of Minnesota 2 Department

More information

A Robust Preconditioner for the Hessian System in Elliptic Optimal Control Problems

A Robust Preconditioner for the Hessian System in Elliptic Optimal Control Problems A Robust Preconditioner for the Hessian System in Elliptic Optimal Control Problems Etereldes Gonçalves 1, Tarek P. Mathew 1, Markus Sarkis 1,2, and Christian E. Schaerer 1 1 Instituto de Matemática Pura

More information

Uncertainty analysis of large-scale systems using domain decomposition

Uncertainty analysis of large-scale systems using domain decomposition Center for Turbulence Research Annual Research Briefs 2007 143 Uncertainty analysis of large-scale systems using domain decomposition By D. Ghosh, C. Farhat AND P. Avery 1. Motivation and objectives A

More information

Iterative Methods for Solving A x = b

Iterative Methods for Solving A x = b Iterative Methods for Solving A x = b A good (free) online source for iterative methods for solving A x = b is given in the description of a set of iterative solvers called templates found at netlib: http

More information

BETI for acoustic and electromagnetic scattering

BETI for acoustic and electromagnetic scattering BETI for acoustic and electromagnetic scattering O. Steinbach, M. Windisch Institut für Numerische Mathematik Technische Universität Graz Oberwolfach 18. Februar 2010 FWF-Project: Data-sparse Boundary

More information

Preconditioned inverse iteration and shift-invert Arnoldi method

Preconditioned inverse iteration and shift-invert Arnoldi method Preconditioned inverse iteration and shift-invert Arnoldi method Melina Freitag Department of Mathematical Sciences University of Bath CSC Seminar Max-Planck-Institute for Dynamics of Complex Technical

More information

Parallel scalability of a FETI DP mortar method for problems with discontinuous coefficients

Parallel scalability of a FETI DP mortar method for problems with discontinuous coefficients Parallel scalability of a FETI DP mortar method for problems with discontinuous coefficients Nina Dokeva and Wlodek Proskurowski University of Southern California, Department of Mathematics Los Angeles,

More information

Various ways to use a second level preconditioner

Various ways to use a second level preconditioner Various ways to use a second level preconditioner C. Vuik 1, J.M. Tang 1, R. Nabben 2, and Y. Erlangga 3 1 Delft University of Technology Delft Institute of Applied Mathematics 2 Technische Universität

More information