Multilevel Low-Rank Preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota. Modelling 2014 June 2,2014
|
|
- Allan Dickerson
- 5 years ago
- Views:
Transcription
1 Multilevel Low-Rank Preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota Modelling 24 June 2,24 Dedicated to Owe Axelsson at the occasion of his 8th birthday
2 Acknowledments Joint work with Ruipeng Li Work supported by NSF-DMS Modelling4 6/2/24 p. 2
3 Intro: ILU-type preconditioners Problem: To solve linear systems Ax = b Common approach: [ grey-box solvers] Krylov subspace accelerator (e.g., GMRES, BiCGSTAB) + Preconditioner Common preconditioners: Incomplete LU factorizations; Relaxation-type; AMG;... Common difficulties of ILUs: Often fail for indefinite problems Not too good for highly parallel environments [GPUs] Modelling4 6/2/24 p. 3
4 Alternatives to ILU preconditioners Time to think about (radical) alternatives? Preconditioners requiring few irregular computations..... that trade volume of computations for speed,.. and, if possible, more robust for indefinite case Possible candidates: Methods based on Multilevel Low- Rank (MLR) approximations Low-rank approximation techniques can be seen everywhere in computational sciences Common approach: truncated SVD.... and more often now : random sampling Modelling4 6/2/24 p. 4
5 Related work: Work on H-matrices [Hackbusch and co-workers, B. Khoromskij, L. Grasedyck, S. Leborne, + many others..] Work on HSS matrices [e.g., J. XIA, S. CHANDRASEKARAN, M. GU, AND X-S. LI 2.] Work on balanced incomplete factorizations (R. Bru et al.) Work on sweeping preconditioners by Engquist and Ying. Work on computing the diagonal of a matrix inverse [Jok Tang and YS (2)..] Modelling4 6/2/24 p. 5
6 MULTI-LEVEL LOW-RANK PRECONDITIONERS
7 Low-rank Multilevel Approximations Starting point: symmetric matrix derived from a 5-point discretization of a 2-D Pb on n x n y grid A = A = A D 2 D 2 A 2... D D α A α D α+ D α+ A α D ny A ny ( ) A A 2 A 2 A 22 ( A A22) + ( A 2 A 2 ) Modelling4 6/2/24 p. 7
8 Corresponding splitting on FD mesh: Modelling4 6/2/24 p. 8
9 A R m m, A 22 R (n m) (n m) In the simplest case second matrix is: ( ) A A 2 = A 2 A 22 ( A A22) + I I Write 2nd matrix as: I I = + I + I I I I I E T = I I E E T Modelling4 6/2/24 p. 9
10 Thus: A = (A + EE T ) }{{} B EE T Note: E R n n x, X R n x n x n x = separator = [O(n /2 ) in 2-D, O(n 2/3 ) in 3-D] B := ( B B2) A = B EE T, R n n, E := ( E E 2 ) R n n x, Next step: use Sherman- Morrison formula: A = B + (B E)X (B E) T X = I E T B E Modelling4 6/2/24 p.
11 Multilevel Low-Rank (MLR) algorithm Use in a recursive framework [apply recursively to B, B 2 ] Next step: lowrank approx. for B E B E U k V T k, U k R n k, V k R n x k, Replace B E by U k Vk T in X = I (ET B )E: X G k = I V k U T k E, ( Rn x n x ) Leads to... Preconditioner M = B + U k H k U T k Use recursivity Can show : H k = (I U T k EV k) and H T k = H k Modelling4 6/2/24 p.
12 Other options explored Another thought : approximate X (only) and exploit recursivity B [v + E X E T B v] However won t work: cost explodes with # levels [recursivity] We will see later how we can use this in DD framework Another possibility: approximate B E on one side only: M = B + B EG k V ku T k = B [I + EG k V ku T k ] However, can show that this is equivalent to the previous method Modelling4 6/2/24 p. 2
13 Recursive multilevel framework ( A i = B i + E i Ei T, B Bi i. Bi2) Next level, set A i B i and A i2 B i2 Repeat on A i, A i2 Last level, factor A i (IC, ILU) Binary tree structure: Modelling4 6/2/24 p. 3
14 Generalization: Domain Decomposition framework Domain partitioned into 2 domains with an edge separator Ω 2 Ω Matrix can be permuted to: P AP T = ˆB ˆF ˆF T C X X T ˆF 2 T C 2 ˆB 2 ˆF 2 Interface nodes in each domain are listed last. Modelling4 6/2/24 p. 4
15 Each matrix ˆB i is of size n i n i (interior var.) and the matrix C i is of size m i m i (interface var.) P AP T = Let: E α = ( B B2) and α used for balancing αi X T α EE T with B i = then we have: { D = α 2 I D 2 = α 2 X T X. ( ˆB i ˆF Better results when using diagonals instead of αi ˆF T i C i + D i ) Modelling4 6/2/24 p. 5
16 Theory: 2-level analysis for model problem Interested in eigenvalues γ j of A B = B EX E T B when A = Pure Laplacean.. They are: γ j = β j = α j = β j α j, j =,, n x with: n y /2 k= n y /2 k= 4 ( sin 2 sin 2 sin 2 n y kπ n y + kπ n y + + sin2 sin 2 n y kπ n y + kπ n y + + sin2 ) 2, jπ 2(n x +) jπ 2(n x +). Modelling4 6/2/24 p. 6
17 Decay of the γ j s when nx = ny = β j sqrt(β j ) α j γ j Note β j are the singular values of B E. In this particular case 3 eigenvectors will capture 92 % of the inverse whereas 5 eigenvectors will capture 97% of the inverse. Modelling4 6/2/24 p. 7
18 EXPERIMENTS
19 A few MATLAB experiments Matlab first Small problems ; real tests later Helmholtz-like problem: 2 u x 2 u 2 y ρu = 6 ρ ( 2x 2 + y 2) in Ω, 2 + Boundary conditions so solution is known ρ = constant selected to make problem more or less difficult Finite differences on a mesh (matrix size 4,96). MLR starts converging for k = 2. ρ = 845 selected so original Laplacean is shifted by.2 6 negative eigenvalues, smallest = Modelling4 6/2/24 p. 9
20 Comparison with ILUTP 2 MLLR k=2 MLLR k=3 MLLR k=5 ILUTP(.) Log of residual norm GMRES(4) iterations ILUTP vs. MLR (E) # levels = 7 for MLR Modelling4 6/2/24 p. 2
21 k nlev=7 nlev=6 nlev=5 nlev=4 nlev= MLR: GMRES(4) iteration counts and fill ratios Modelling4 6/2/24 p. 2
22 Helmoltz-like equation - a 3D case Similar set-up to 2D case grid size n = 24 3 = 3, 824 ρ = 32.5 selected so the shift is.5 - making the problem very indefinite [6 negative eigenvalues, λ min = ] GMRES(4)-MLR iteration counts and fill ratios k nlev=6 nlev=5 nlev= ILUTP fails even for very small values of droptol (large fill) Modelling4 6/2/24 p. 22
23 General matrices 7 matrices from the Univ. Florida sparse matrix collection + one from a shell problem. 7 matrices are SPD Size varies from n =, 224 (HB/bcsstm27) to n = 9, (AG-Monien/3elt dual) nnz varies from nnz = 5, 3 (HB/bcspwr6) to nnz = 355, 46 (Boeing/bcsstk38). Only indefinite cases shown
24 MATRICES (Non SPD) MLR ICT/ILUTP nlev k fill-ratio #its fill-ratio #its HB/bcsstm HB/bcspwr F HB/bcspwr F HB/bcspwr F HB/blckhole F HB/jagmesh Boeing/nasa AG-Monien/3elt_dual F AG-Monien/airfoil_dual F AG-Monien/ukerbe_dual F SHELL/COQUE8E F MLR vs. ICT/ILUTP Modelling4 6/2/24 p. 24
25 Real tests Experimental setting Hardware: Intel Xeon X5675 processor (2 MB Cache, 3.6 GHz, 6-core) C/C++; Intel Math Kernel Library (MKL,version.2) Stop when: r i 8 r or its exceeds 5 Model Problems in 2-D/3-D: u cu = g in Ω + B.C. 2-D: g(x, y) = ( x 2 + y 2 + c ) e xy ; Ω = (, ) 3. 3-D: g(x, y, z) = 6 c ( x 2 + y 2 + z 2) ; Ω = (, ) 3. F.D. Differences discret. Modelling4 6/2/24 p. 25
26 Symmetric indefinite cases c > in u cu; i.e., shifted by si. 2D case: s =., 3D case: s =.5 MLR + GMRES(4) compared to ILDLT + GMRES(4) 2-D problems: #lev= 4, rank= 5, 7, 7 3-D problems: #lev= 5, rank= 5, 7, 7 ILDLT failed for most cases Difficulties in MLR: #lev cannot be large, [no convergence] inefficient factorization at the last level (memory, CPU time)
27 Grid ILDLT-GMRES MLR-GMRES fill p-t its i-t fill p-t its i-t F F F F F F F Modelling4 6/2/24 p. 27
28 General symmetric matrices - Test matrices MATRIX N NNZ SPD DESCRIPTION Andrews/Andrews 6, 76,54 yes computer graphics pb. Williams/cant 62,45 4,7,383 yes FEM cantilever UTEP/Dubcova2 65,25,3,225 yes 2-D/3-D PDE pb. Rothberg/cfd 7,656,825,58 yes CFD pb. Schmid/thermal 82, ,458 yes thermal pb. Rothberg/cfd2 23,44 3,85,46 yes CFD pb. Schmid/thermal2,228,45 8,58,33 yes thermal pb. Cote/vibrobox 2,328 3,7 no vibroacoustic pb. Cunningham/qa8fk 66,27,66,579 no 3-D acoustics pb. Koutsovasilis/F2 7,55 5,294,285 no structural pb. Modelling4 6/2/24 p. 28
29 MATRIX ICT/ILDLT MLR-CG/GMRES fill p-t its i-t k lev fill p-t its i-t Andrews cant F Dubcova cfd thermal cfd F thermal Modelling4 6/2/24 p. 29
30 MATRIX ICT/ILDLT MLR-CG/GMRES fill p-t its i-t k lev fill p-t its i-t vibrobox F qa8fk F F Modelling4 6/2/24 p. 3
31 Avoiding recursivity: standard DD framework Work in progress Goal: avoid recursivity Consider a domain partition in p domains using vertex- based partitioniong (edge-separation) Interface nodes in each domain are listed last Modelling4 6/2/24 p. 3
32 Local view: External interface points Local interface points Interior points ( ) ( ) Bi F i ui Ei T C i y i + ( ) j N i E ij y j = ( fi g i ) Modelling4 6/2/24 p. 32
33 The global system: Global view Global system can be permuted to the form u i s internal variables y interface variables External interface points B... ˆF B 2... ˆF B p ˆF p Ê2 T... ÊT p C Ê T u u 2. y u p = b Local interface points Interior points ˆF i maps local interface points to interior points in domain Ω i Êi T does the reverse operation Modelling4 6/2/24 p. 33
34 Example: nz = 438 Modelling4 6/2/24 p. 34
35 Splitting Split as: Define: [ B Ê T ˆF C ( ) ( ( ) B ˆF B ˆF A Ê T = + C C) Ê T ( ) ( ) α ˆF α F ; E Ê Then: αi αi ] [ ] B + α = 2 ˆF Ê T C + α 2 F E T. I Property: ˆF Ê T is local, i.e., no inter-domain couplings [ B + α A 2 ˆF Ê T C + α 2 I = block-diagonal ] Modelling4 6/2/24 p. 35
36 Low-Rank Approximation DD preconditioners Sherman-Morrison A = A + A F G E T A G I E T A F Options: (a) Approximate A F, E T A, G [as before] (b) Approximate only G [new] (b) requires 2 solves with A. Let G G k Preconditioner M = A + A F G k ET A Modelling4 6/2/24 p. 36
37 Symmetric Positive Definite case Recap: Let G I E T A E I H. Then A = A + A EG E T A Approximate G by G k Matrix A is SPD M = A + (A Can show: λ j (H) <. preconditioner: E)G k (ET A ) Modelling4 6/2/24 p. 37
38 Next, approximate H as H U DU T Then: (I U DU T ) = I + U[(I D) I]U T. Now take rank-k approximation to H: H U k D k U T k G k = I U k D k U T k G k (I U k D k U T k ) = I + U k [(I D k ) I]U T k Observation: A = M + A E[G G ]ET A G k : k largest eigenvalues of G matched others set == Result: AM has n s + k eigenvalues == All others between and k Modelling4 6/2/24 p. 38
39 Alternative: reset lowest eigenvalues to constant Let H = UΛU T = exact (full) diagonalization of H We replaced Λ by: Alternative: replace Λ by λ λ λ 2 λ λ k λ k θ θ Interesting case: θ = λ k+ Question: related approximation to G? Modelling4 6/2/24 p. 39
40 Result: Let γ = /( θ). Then approx. to G is: G k,θ γi + U k[(i D k ) γi]u T k G k : k largest eigenvalues of G matched others set == θ θ = yields previous case When λ k+ θ < we get Result: AM has n s + k eigenvalues == All others Next: An example for a 9 9 Laplacean, 4 domains, s = 9 Modelling4 6/2/24 p. 4
41 k = 5 Eigenvalues of AM for the case θ = 2.5 x Modelling4 6/2/24 p. 4
42 k = 5 Eigenvalues of AM for the case θ = λ k+ 4 x Modelling4 6/2/24 p. 42
43 k = 5 Eigenvalues of AM for the case θ = λ k+ 2 x
44 Proposition Assume θ is so that λ k+ θ <. Then the eigenvalues η i of AM satisfy: η i + θ A/2 A E 2 2. Experiments: For the Laplacean (FD) and when α =, A /2 A E 2 2 = ET A AA E 2 4 regardless of the mesh-size. Being investigated. Best upper bound for θ = λ k+ Assume above is true and set θ = λ k+. Then κ(am ) constant, if k large enough so that λ k+ constant. i.e., need to capture sufficient part of spectrum Modelling4 6/2/24 p. 44
45 The symmetric indefinite case Appeal of this approach over ILU: approximate inverse Not as sensitive to indefiniteness Part of the results shown still hold But λ i (H) can be > now. Parameter α now plays a more important role Work still in progress Modelling4 6/2/24 p. 45
46 Example: Take Laplacean on a 3 3 FD grid. Subtract.4I result: 26 negative eigenvalues λ min = , λ max = Use α = 4., θ =.9; We do test for k = and then k = 5 Modelling4 6/2/24 p. 46
47 k = Eigenvalues of AM [θ =.9, α = 4] x k = All eigenvalues are > Modelling4 6/2/24 p. 47
48 k = 5 Eigenvalues of AM [θ =.9, α = 4] 6 x 2 k = eigenvalues are < Modelling4 6/2/24 p. 48
49 Parallel implementations Recall : [ ] M = A I + EG k,θ ET A G k,θ = γi + U k[(i D k ) γi]u T k Steps involved in applying M to a vector x : ALGORITHM : Preconditioning operation. z = A x // ˆB i -solves and C α solve 2. y = E T z // Interior points to interface (Loc.) 3. y k = G k,θ y // Use Low-Rank approx. 4. z k = Ey k // Interface to interior points (Loc.) 5. u = A (x + z k ) // ˆB i -solves and C α solve Modelling4 6/2/24 p. 49
50 A Solves Note: A = ˆB ˆB 2... ˆB p C α Recall ˆB i = B i + α 2 E i E T i A solve with A amounts to all p ˆB i -solves and a C α -solve Can replace C α Can use any solver for the ˆB i s by a low degree polynomial [Chebyshev] Modelling4 6/2/24 p. 5
51 Parallel tests: Itasca (MSI) HP linux cluster- with Xeons 556 ( Nehalem ) processors 2-D 3-D Mesh Nproc Rank #its Prec-t Iter-t Mesh Nproc Rank #its Prec-t Iter-t Modelling4 6/2/24 p. 5
52 Mixing Divide & Conquer and standard DD Modelling4 6/2/24 p. 52
53 Mixing Divide & Conquer and standard DD Must use a two-sided approximation Back to recursive version Recall A = A + (A E)G (A E)T G I E T (A E) Use a 2-domain partitioning + recursion Approximate A E by a low-rank matrix - get related approximation to G : A E U k Vk T M = A + U k X k X k = I U T k EV k Advantage: Natural way of splitting - no need for balancing U T k Modelling4 6/2/24 p. 53
54 A A is nearly low-rank Similar to experiment shown earlier earlier Eigenvalues of A A & 3 zooms closing in on 5 Original Zoom / Zoom /25 Zoom / Modelling4 6/2/24 p. 54
55 Conclusion Promising alternatives to ILUs can be found in new forms of approximate inverse techniques Seek data-sparsity instead of regular sparsity DD approch easier to implement, easier to understand than recursive approach Advantages of Multilevel Low-Rank preconditioners: Approximate inverses less sensitive to indefiniteness Exploit dense computations Easy to update Modelling4 6/2/24 p. 55
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 informationParallel Multilevel Low-Rank approximation preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota
Parallel Multilevel Low-Rank approximation preconditioners Yousef Saad Department of Computer Science and Engineering University of Minnesota ICME, Stanford April 2th, 25 First: Partly joint work with
More informationThe new challenges to Krylov subspace methods Yousef Saad Department of Computer Science and Engineering University of Minnesota
The new challenges to Krylov subspace methods Yousef Saad Department of Computer Science and Engineering University of Minnesota SIAM Applied Linear Algebra Valencia, June 18-22, 2012 Introduction Krylov
More informationMultilevel preconditioning techniques with applications Yousef Saad Department of Computer Science and Engineering University of Minnesota
Multilevel preconditioning techniques with applications Yousef Saad Department of Computer Science and Engineering University of Minnesota Syracuse University, Sep. 20, 2013 First: Joint work with Ruipeng
More information1. Introduction. We consider the problem of solving the linear system. Ax = b, (1.1)
SCHUR COMPLEMENT BASED DOMAIN DECOMPOSITION PRECONDITIONERS WITH LOW-RANK CORRECTIONS RUIPENG LI, YUANZHE XI, AND YOUSEF SAAD Abstract. This paper introduces a robust preconditioner for general sparse
More informationDomain Decomposition-based contour integration eigenvalue solvers
Domain Decomposition-based contour integration eigenvalue solvers Vassilis Kalantzis joint work with Yousef Saad Computer Science and Engineering Department University of Minnesota - Twin Cities, USA SIAM
More information9.1 Preconditioned Krylov Subspace Methods
Chapter 9 PRECONDITIONING 9.1 Preconditioned Krylov Subspace Methods 9.2 Preconditioned Conjugate Gradient 9.3 Preconditioned Generalized Minimal Residual 9.4 Relaxation Method Preconditioners 9.5 Incomplete
More informationFine-grained Parallel Incomplete LU Factorization
Fine-grained Parallel Incomplete LU Factorization Edmond Chow School of Computational Science and Engineering Georgia Institute of Technology Sparse Days Meeting at CERFACS June 5-6, 2014 Contribution
More informationSolving 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 informationIncomplete 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 informationITERATIVE METHODS FOR SPARSE LINEAR SYSTEMS
ITERATIVE METHODS FOR SPARSE LINEAR SYSTEMS YOUSEF SAAD University of Minnesota PWS PUBLISHING COMPANY I(T)P An International Thomson Publishing Company BOSTON ALBANY BONN CINCINNATI DETROIT LONDON MADRID
More informationPreface 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 informationDivide and conquer algorithms for large eigenvalue problems Yousef Saad Department of Computer Science and Engineering University of Minnesota
Divide and conquer algorithms for large eigenvalue problems Yousef Saad Department of Computer Science and Engineering University of Minnesota PMAA 14 Lugano, July 4, 2014 Collaborators: Joint work with:
More informationSchur complement-based domain decomposition preconditioners with low-rank corrections
NUMERICAL LINEAR ALGEBRA WITH APPLICATIONS Numer. Linear Algebra Appl. (206) Published online in Wiley Online Library (wileyonlinelibrary.com)..205 Schur complement-based domain decomposition preconditioners
More informationFine-Grained Parallel Algorithms for Incomplete Factorization Preconditioning
Fine-Grained Parallel Algorithms for Incomplete Factorization Preconditioning Edmond Chow School of Computational Science and Engineering Georgia Institute of Technology, USA SPPEXA Symposium TU München,
More informationA robust multilevel approximate inverse preconditioner for symmetric positive definite matrices
DICEA DEPARTMENT OF CIVIL, ENVIRONMENTAL AND ARCHITECTURAL ENGINEERING PhD SCHOOL CIVIL AND ENVIRONMENTAL ENGINEERING SCIENCES XXX CYCLE A robust multilevel approximate inverse preconditioner for symmetric
More informationNumerical Methods I Non-Square and Sparse Linear Systems
Numerical Methods I Non-Square and Sparse Linear Systems Aleksandar Donev Courant Institute, NYU 1 donev@courant.nyu.edu 1 MATH-GA 2011.003 / CSCI-GA 2945.003, Fall 2014 September 25th, 2014 A. Donev (Courant
More informationAMS Mathematics Subject Classification : 65F10,65F50. Key words and phrases: ILUS factorization, preconditioning, Schur complement, 1.
J. Appl. Math. & Computing Vol. 15(2004), No. 1, pp. 299-312 BILUS: A BLOCK VERSION OF ILUS FACTORIZATION DAVOD KHOJASTEH SALKUYEH AND FAEZEH TOUTOUNIAN Abstract. ILUS factorization has many desirable
More informationMultigrid absolute value preconditioning
Multigrid absolute value preconditioning Eugene Vecharynski 1 Andrew Knyazev 2 (speaker) 1 Department of Computer Science and Engineering University of Minnesota 2 Department of Mathematical and Statistical
More informationSolving Ax = b, an overview. Program
Numerical Linear Algebra Improving iterative solvers: preconditioning, deflation, numerical software and parallelisation Gerard Sleijpen and Martin van Gijzen November 29, 27 Solving Ax = b, an overview
More informationLecture 3: Inexact inverse iteration with preconditioning
Lecture 3: Department of Mathematical Sciences CLAPDE, Durham, July 2008 Joint work with M. Freitag (Bath), and M. Robbé & M. Sadkane (Brest) 1 Introduction 2 Preconditioned GMRES for Inverse Power Method
More informationFINE-GRAINED PARALLEL INCOMPLETE LU FACTORIZATION
FINE-GRAINED PARALLEL INCOMPLETE LU FACTORIZATION EDMOND CHOW AND AFTAB PATEL Abstract. This paper presents a new fine-grained parallel algorithm for computing an incomplete LU factorization. All nonzeros
More informationThe flexible incomplete LU preconditioner for large nonsymmetric linear systems. Takatoshi Nakamura Takashi Nodera
Research Report KSTS/RR-15/006 The flexible incomplete LU preconditioner for large nonsymmetric linear systems by Takatoshi Nakamura Takashi Nodera Takatoshi Nakamura School of Fundamental Science and
More informationInexact inverse iteration with preconditioning
Department of Mathematical Sciences Computational Methods with Applications Harrachov, Czech Republic 24th August 2007 (joint work with M. Robbé and M. Sadkane (Brest)) 1 Introduction 2 Preconditioned
More informationFINE-GRAINED PARALLEL INCOMPLETE LU FACTORIZATION
FINE-GRAINED PARALLEL INCOMPLETE LU FACTORIZATION EDMOND CHOW AND AFTAB PATEL Abstract. This paper presents a new fine-grained parallel algorithm for computing an incomplete LU factorization. All nonzeros
More informationAMS526: 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 informationLinear 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 informationOUTLINE 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 informationAn Adaptive Hierarchical Matrix on Point Iterative Poisson Solver
Malaysian Journal of Mathematical Sciences 10(3): 369 382 (2016) MALAYSIAN JOURNAL OF MATHEMATICAL SCIENCES Journal homepage: http://einspem.upm.edu.my/journal An Adaptive Hierarchical Matrix on Point
More informationPreconditioning Techniques for Large Linear Systems Part III: General-Purpose Algebraic Preconditioners
Preconditioning Techniques for Large Linear Systems Part III: General-Purpose Algebraic Preconditioners Michele Benzi Department of Mathematics and Computer Science Emory University Atlanta, Georgia, USA
More informationParallelization of Multilevel Preconditioners Constructed from Inverse-Based ILUs on Shared-Memory Multiprocessors
Parallelization of Multilevel Preconditioners Constructed from Inverse-Based ILUs on Shared-Memory Multiprocessors J.I. Aliaga 1 M. Bollhöfer 2 A.F. Martín 1 E.S. Quintana-Ortí 1 1 Deparment of Computer
More informationILAS 2017 July 25, 2017
Polynomial and rational filtering for eigenvalue problems and the EVSL project Yousef Saad Department of Computer Science and Engineering University of Minnesota ILAS 217 July 25, 217 Large eigenvalue
More informationFEM 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 informationThe EVSL package for symmetric eigenvalue problems Yousef Saad Department of Computer Science and Engineering University of Minnesota
The EVSL package for symmetric eigenvalue problems Yousef Saad Department of Computer Science and Engineering University of Minnesota 15th Copper Mountain Conference Mar. 28, 218 First: Joint work with
More informationThe 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 informationA Numerical Study of Some Parallel Algebraic Preconditioners
A Numerical Study of Some Parallel Algebraic Preconditioners Xing Cai Simula Research Laboratory & University of Oslo PO Box 1080, Blindern, 0316 Oslo, Norway xingca@simulano Masha Sosonkina University
More informationc 2015 Society for Industrial and Applied Mathematics
SIAM J. SCI. COMPUT. Vol. 37, No. 2, pp. C169 C193 c 2015 Society for Industrial and Applied Mathematics FINE-GRAINED PARALLEL INCOMPLETE LU FACTORIZATION EDMOND CHOW AND AFTAB PATEL Abstract. This paper
More informationNotes on PCG for Sparse Linear Systems
Notes on PCG for Sparse Linear Systems Luca Bergamaschi Department of Civil Environmental and Architectural Engineering University of Padova e-mail luca.bergamaschi@unipd.it webpage www.dmsa.unipd.it/
More informationDepartment of Computer Science and Engineering
High performance numerical linear algebra: trends and new challenges. Yousef Saad Department of Computer Science and Engineering University of Minnesota HPC days in Lyon April 8, 216 First: A personal
More informationA parameter tuning technique of a weighted Jacobi-type preconditioner and its application to supernova simulations
A parameter tuning technique of a weighted Jacobi-type preconditioner and its application to supernova simulations Akira IMAKURA Center for Computational Sciences, University of Tsukuba Joint work with
More informationLecture 18 Classical Iterative Methods
Lecture 18 Classical Iterative Methods MIT 18.335J / 6.337J Introduction to Numerical Methods Per-Olof Persson November 14, 2006 1 Iterative Methods for Linear Systems Direct methods for solving Ax = b,
More informationSPARSE SOLVERS POISSON EQUATION. Margreet Nool. November 9, 2015 FOR THE. CWI, Multiscale Dynamics
SPARSE SOLVERS FOR THE POISSON EQUATION Margreet Nool CWI, Multiscale Dynamics November 9, 2015 OUTLINE OF THIS TALK 1 FISHPACK, LAPACK, PARDISO 2 SYSTEM OVERVIEW OF CARTESIUS 3 POISSON EQUATION 4 SOLVERS
More informationRobust Preconditioned Conjugate Gradient for the GPU and Parallel Implementations
Robust Preconditioned Conjugate Gradient for the GPU and Parallel Implementations Rohit Gupta, Martin van Gijzen, Kees Vuik GPU Technology Conference 2012, San Jose CA. GPU Technology Conference 2012,
More informationJ.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 informationEffective matrix-free preconditioning for the augmented immersed interface method
Effective matrix-free preconditioning for the augmented immersed interface method Jianlin Xia a, Zhilin Li b, Xin Ye a a Department of Mathematics, Purdue University, West Lafayette, IN 47907, USA. E-mail:
More informationAMS526: Numerical Analysis I (Numerical Linear Algebra for Computational and Data Sciences)
AMS526: Numerical Analysis I (Numerical Linear Algebra for Computational and Data Sciences) Lecture 19: Computing the SVD; Sparse Linear Systems Xiangmin Jiao Stony Brook University Xiangmin Jiao Numerical
More informationSolving Large Nonlinear Sparse Systems
Solving Large Nonlinear Sparse Systems Fred W. Wubs and Jonas Thies Computational Mechanics & Numerical Mathematics University of Groningen, the Netherlands f.w.wubs@rug.nl Centre for Interdisciplinary
More informationLecture 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 informationTopics. 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 informationChapter 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 informationIterative Methods and Multigrid
Iterative Methods and Multigrid Part 3: Preconditioning 2 Eric de Sturler Preconditioning The general idea behind preconditioning is that convergence of some method for the linear system Ax = b can be
More informationUsing an Auction Algorithm in AMG based on Maximum Weighted Matching in Matrix Graphs
Using an Auction Algorithm in AMG based on Maximum Weighted Matching in Matrix Graphs Pasqua D Ambra Institute for Applied Computing (IAC) National Research Council of Italy (CNR) pasqua.dambra@cnr.it
More informationConjugate 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 informationScientific Computing with Case Studies SIAM Press, Lecture Notes for Unit VII Sparse Matrix
Scientific Computing with Case Studies SIAM Press, 2009 http://www.cs.umd.edu/users/oleary/sccswebpage Lecture Notes for Unit VII Sparse Matrix Computations Part 1: Direct Methods Dianne P. O Leary c 2008
More informationAlgebraic Multigrid as Solvers and as Preconditioner
Ò Algebraic Multigrid as Solvers and as Preconditioner Domenico Lahaye domenico.lahaye@cs.kuleuven.ac.be http://www.cs.kuleuven.ac.be/ domenico/ Department of Computer Science Katholieke Universiteit Leuven
More informationContents. 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 informationSOLVING 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 informationFAST STRUCTURED EIGENSOLVER FOR DISCRETIZED PARTIAL DIFFERENTIAL OPERATORS ON GENERAL MESHES
Proceedings of the Project Review, Geo-Mathematical Imaging Group Purdue University, West Lafayette IN, Vol. 1 2012 pp. 123-132. FAST STRUCTURED EIGENSOLVER FOR DISCRETIZED PARTIAL DIFFERENTIAL OPERATORS
More informationIncomplete LU Preconditioning and Error Compensation Strategies for Sparse Matrices
Incomplete LU Preconditioning and Error Compensation Strategies for Sparse Matrices Eun-Joo Lee Department of Computer Science, East Stroudsburg University of Pennsylvania, 327 Science and Technology Center,
More informationBoundary Value Problems - Solving 3-D Finite-Difference problems Jacob White
Introduction to Simulation - Lecture 2 Boundary Value Problems - Solving 3-D Finite-Difference problems Jacob White Thanks to Deepak Ramaswamy, Michal Rewienski, and Karen Veroy Outline Reminder about
More informationVariants of BiCGSafe method using shadow three-term recurrence
Variants of BiCGSafe method using shadow three-term recurrence Seiji Fujino, Takashi Sekimoto Research Institute for Information Technology, Kyushu University Ryoyu Systems Co. Ltd. (Kyushu University)
More informationFast Structured Spectral Methods
Spectral methods HSS structures Fast algorithms Conclusion Fast Structured Spectral Methods Yingwei Wang Department of Mathematics, Purdue University Joint work with Prof Jie Shen and Prof Jianlin Xia
More informationJae Heon Yun and Yu Du Han
Bull. Korean Math. Soc. 39 (2002), No. 3, pp. 495 509 MODIFIED INCOMPLETE CHOLESKY FACTORIZATION PRECONDITIONERS FOR A SYMMETRIC POSITIVE DEFINITE MATRIX Jae Heon Yun and Yu Du Han Abstract. We propose
More informationSome 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 informationPreconditioned Locally Minimal Residual Method for Computing Interior Eigenpairs of Symmetric Operators
Preconditioned Locally Minimal Residual Method for Computing Interior Eigenpairs of Symmetric Operators Eugene Vecharynski 1 Andrew Knyazev 2 1 Department of Computer Science and Engineering University
More informationPreliminary/Qualifying Exam in Numerical Analysis (Math 502a) Spring 2012
Instructions Preliminary/Qualifying Exam in Numerical Analysis (Math 502a) Spring 2012 The exam consists of four problems, each having multiple parts. You should attempt to solve all four problems. 1.
More informationA Chebyshev-based two-stage iterative method as an alternative to the direct solution of linear systems
A Chebyshev-based two-stage iterative method as an alternative to the direct solution of linear systems Mario Arioli m.arioli@rl.ac.uk CCLRC-Rutherford Appleton Laboratory with Daniel Ruiz (E.N.S.E.E.I.H.T)
More informationAMG 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 informationOn the Preconditioning of the Block Tridiagonal Linear System of Equations
On the Preconditioning of the Block Tridiagonal Linear System of Equations Davod Khojasteh Salkuyeh Department of Mathematics, University of Mohaghegh Ardabili, PO Box 179, Ardabil, Iran E-mail: khojaste@umaacir
More informationEfficient Solvers for Stochastic Finite Element Saddle Point Problems
Efficient Solvers for Stochastic Finite Element Saddle Point Problems Catherine E. Powell c.powell@manchester.ac.uk School of Mathematics University of Manchester, UK Efficient Solvers for Stochastic Finite
More informationKarhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques
Institut für Numerische Mathematik und Optimierung Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques Oliver Ernst Computational Methods with Applications Harrachov, CR,
More informationMaster 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 informationCLASSICAL ITERATIVE METHODS
CLASSICAL ITERATIVE METHODS LONG CHEN In this notes we discuss classic iterative methods on solving the linear operator equation (1) Au = f, posed on a finite dimensional Hilbert space V = R N equipped
More informationEFFECTIVE AND ROBUST PRECONDITIONING OF GENERAL SPD MATRICES VIA STRUCTURED INCOMPLETE FACTORIZATION
EFFECVE AND ROBUS PRECONDONNG OF GENERAL SPD MARCES VA SRUCURED NCOMPLEE FACORZAON JANLN XA AND ZXNG XN Abstract For general symmetric positive definite SPD matrices, we present a framework for designing
More informationPASC17 - Lugano June 27, 2017
Polynomial and rational filtering for eigenvalue problems and the EVSL project Yousef Saad Department of Computer Science and Engineering University of Minnesota PASC17 - Lugano June 27, 217 Large eigenvalue
More informationStabilization 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 informationParallel 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 informationPreconditioning techniques to accelerate the convergence of the iterative solution methods
Note Preconditioning techniques to accelerate the convergence of the iterative solution methods Many issues related to iterative solution of linear systems of equations are contradictory: numerical efficiency
More informationImplementation of a preconditioned eigensolver using Hypre
Implementation of a preconditioned eigensolver using Hypre Andrew V. Knyazev 1, and Merico E. Argentati 1 1 Department of Mathematics, University of Colorado at Denver, USA SUMMARY This paper describes
More informationA Robust Preconditioned Iterative Method for the Navier-Stokes Equations with High Reynolds Numbers
Applied and Computational Mathematics 2017; 6(4): 202-207 http://www.sciencepublishinggroup.com/j/acm doi: 10.11648/j.acm.20170604.18 ISSN: 2328-5605 (Print); ISSN: 2328-5613 (Online) A Robust Preconditioned
More informationSparse Linear Systems. Iterative Methods for Sparse Linear Systems. Motivation for Studying Sparse Linear Systems. Partial Differential Equations
Sparse Linear Systems Iterative Methods for Sparse Linear Systems Matrix Computations and Applications, Lecture C11 Fredrik Bengzon, Robert Söderlund We consider the problem of solving the linear system
More informationPerformance Evaluation of GPBiCGSafe Method without Reverse-Ordered Recurrence for Realistic Problems
Performance Evaluation of GPBiCGSafe Method without Reverse-Ordered Recurrence for Realistic Problems Seiji Fujino, Takashi Sekimoto Abstract GPBiCG method is an attractive iterative method for the solution
More informationNumerical tensor methods and their applications
Numerical tensor methods and their applications 8 May 2013 All lectures 4 lectures, 2 May, 08:00-10:00: Introduction: ideas, matrix results, history. 7 May, 08:00-10:00: Novel tensor formats (TT, HT, QTT).
More informationPreconditioning for Nonsymmetry and Time-dependence
Preconditioning for Nonsymmetry and Time-dependence Andy Wathen Oxford University, UK joint work with Jen Pestana and Elle McDonald Jeju, Korea, 2015 p.1/24 Iterative methods For self-adjoint problems/symmetric
More informationTwo-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 informationMathematics and Computer Science
Technical Report TR-2007-002 Block preconditioning for saddle point systems with indefinite (1,1) block by Michele Benzi, Jia Liu Mathematics and Computer Science EMORY UNIVERSITY International Journal
More informationOptimal multilevel preconditioning of strongly anisotropic problems.part II: non-conforming FEM. p. 1/36
Optimal multilevel preconditioning of strongly anisotropic problems. Part II: non-conforming FEM. Svetozar Margenov margenov@parallel.bas.bg Institute for Parallel Processing, Bulgarian Academy of Sciences,
More informationKasetsart University Workshop. Multigrid methods: An introduction
Kasetsart University Workshop Multigrid methods: An introduction Dr. Anand Pardhanani Mathematics Department Earlham College Richmond, Indiana USA pardhan@earlham.edu A copy of these slides is available
More informationSolving Sparse Linear Systems: Iterative methods
Scientific Computing with Case Studies SIAM Press, 2009 http://www.cs.umd.edu/users/oleary/sccs Lecture Notes for Unit VII Sparse Matrix Computations Part 2: Iterative Methods Dianne P. O Leary c 2008,2010
More informationSolving Sparse Linear Systems: Iterative methods
Scientific Computing with Case Studies SIAM Press, 2009 http://www.cs.umd.edu/users/oleary/sccswebpage Lecture Notes for Unit VII Sparse Matrix Computations Part 2: Iterative Methods Dianne P. O Leary
More informationRobust solution of Poisson-like problems with aggregation-based AMG
Robust solution of Poisson-like problems with aggregation-based AMG Yvan Notay Université Libre de Bruxelles Service de Métrologie Nucléaire Paris, January 26, 215 Supported by the Belgian FNRS http://homepages.ulb.ac.be/
More informationAn Introduction to Algebraic Multigrid (AMG) Algorithms Derrick Cerwinsky and Craig C. Douglas 1/84
An Introduction to Algebraic Multigrid (AMG) Algorithms Derrick Cerwinsky and Craig C. Douglas 1/84 Introduction Almost all numerical methods for solving PDEs will at some point be reduced to solving A
More informationMultilevel Preconditioning of Graph-Laplacians: Polynomial Approximation of the Pivot Blocks Inverses
Multilevel Preconditioning of Graph-Laplacians: Polynomial Approximation of the Pivot Blocks Inverses P. Boyanova 1, I. Georgiev 34, S. Margenov, L. Zikatanov 5 1 Uppsala University, Box 337, 751 05 Uppsala,
More informationBlock Iterative Eigensolvers for Sequences of Dense Correlated Eigenvalue Problems
Mitglied der Helmholtz-Gemeinschaft Block Iterative Eigensolvers for Sequences of Dense Correlated Eigenvalue Problems Birkbeck University, London, June the 29th 2012 Edoardo Di Napoli Motivation and Goals
More informationParallelism 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 informationSparse factorization using low rank submatrices. Cleve Ashcraft LSTC 2010 MUMPS User Group Meeting April 15-16, 2010 Toulouse, FRANCE
Sparse factorization using low rank submatrices Cleve Ashcraft LSTC cleve@lstc.com 21 MUMPS User Group Meeting April 15-16, 21 Toulouse, FRANCE ftp.lstc.com:outgoing/cleve/mumps1 Ashcraft.pdf 1 LSTC Livermore
More informationAMS526: Numerical Analysis I (Numerical Linear Algebra)
AMS526: Numerical Analysis I (Numerical Linear Algebra) Lecture 24: Preconditioning and Multigrid Solver Xiangmin Jiao SUNY Stony Brook Xiangmin Jiao Numerical Analysis I 1 / 5 Preconditioning Motivation:
More informationResidual iterative schemes for largescale linear systems
Universidad Central de Venezuela Facultad de Ciencias Escuela de Computación Lecturas en Ciencias de la Computación ISSN 1316-6239 Residual iterative schemes for largescale linear systems William La Cruz
More informationPreconditioned 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 informationTowards high performance IRKA on hybrid CPU-GPU systems
Towards high performance IRKA on hybrid CPU-GPU systems Jens Saak in collaboration with Georg Pauer (OVGU/MPI Magdeburg) Kapil Ahuja, Ruchir Garg (IIT Indore) Hartwig Anzt, Jack Dongarra (ICL Uni Tennessee
More information