A Newton-Galerkin-ADI Method for Large-Scale Algebraic Riccati Equations

Size: px
Start display at page:

Download "A Newton-Galerkin-ADI Method for Large-Scale Algebraic Riccati Equations"

Transcription

1 A Newton-Galerkin-ADI Method for Large-Scale Algebraic Riccati Equations Peter Benner Max-Planck-Institute for Dynamics of Complex Technical Systems Computational Methods in Systems and Control Theory Group Magdeburg, Germany Jens Saak Technische Universität Chemnitz Fakultät für Mathematik Mathematik in Industrie und Technik Chemnitz, Germany Applied Linear Algebra 2010 GAMM Workshop Applied and Numerical Linear Algebra Novi Sad, May 27, /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

2 Outline 1 Introduction 2 LRCF-ADI with Galerkin-Projection-Acceleration 3 LRCF-NM for the ARE 2/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

3 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: 0 = R(X ) := A T X + XA XGX + W. 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

4 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: G = 0 = Lyapunov equation: 0 = R(X ) := A T X + XA XGX + W. 0 = L(X ) := A T X + XA + W. 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

5 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: G = 0 = Lyapunov equation: 0 = R(X ) := A T X + XA XGX + W. 0 = L(X ) := A T X + XA + W. Typical situation in model reduction and optimal control problems for semi-discretized PDEs: n = (= unknowns!), 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

6 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: G = 0 = Lyapunov equation: 0 = R(X ) := A T X + XA XGX + W. 0 = L(X ) := A T X + XA + W. Typical situation in model reduction and optimal control problems for semi-discretized PDEs: n = (= unknowns!), A has sparse representation (A = M 1 S for FEM), 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

7 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: G = 0 = Lyapunov equation: 0 = R(X ) := A T X + XA XGX + W. 0 = L(X ) := A T X + XA + W. Typical situation in model reduction and optimal control problems for semi-discretized PDEs: n = (= unknowns!), A has sparse representation (A = M 1 S for FEM), G, W low-rank with G, W {BB T, C T C}, where B R n m, m n, C R p n, p n. 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

8 Introduction Large-Scale Algebraic Lyapunov and Riccati Equations General form of algebraic Riccati equation (ARE) for A, G = G T, W = W T R n n given and X R n n unknown: G = 0 = Lyapunov equation: 0 = R(X ) := A T X + XA XGX + W. 0 = L(X ) := A T X + XA + W. Typical situation in model reduction and optimal control problems for semi-discretized PDEs: n = (= unknowns!), A has sparse representation (A = M 1 S for FEM), G, W low-rank with G, W {BB T, C T C}, where B R n m, m n, C R p n, p n. Standard (eigenproblem-based) O(n 3 ) methods are not applicable! 3/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

9 Introduction Low-Rank Approximation Consider spectrum of ARE solution (analogous for Lyapunov equations). Example: Linear 1D heat equation with point control, Ω = [ 0, 1 ], FEM discretization using linear B-splines, h = 1/100 = n = /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

10 Introduction Low-Rank Approximation Consider spectrum of ARE solution (analogous for Lyapunov equations). Example: Linear 1D heat equation with point control, Ω = [ 0, 1 ], FEM discretization using linear B-splines, h = 1/100 = n = 101. Idea: X = X T 0 = X = ZZ T = n λ k z k zk T Z (r) (Z (r) ) T = k=1 r λ k z k zk T. k=1 = Goal: compute Z (r) R n r directly w/o ever forming X! 4/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

11 Introduction Review: LRCF-ADI for Lyapunov Equations Consider FX + XF T = GG T ADI iteration for the Lyapunov equation (LE) [Wachspress 95] For j = 1,..., J X 0 = 0 (F + p j I )X j 1 = GG T X 2 j 1 (F T p j I ) (F + p j I )Xj T = GG T X T (F T p j 1 j I ) 2 5/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

12 Introduction Review: LRCF-ADI for Lyapunov Equations Consider FX + XF T = GG T ADI iteration for the Lyapunov equation (LE) [Wachspress 95] For j = 1,..., J X 0 = 0 (F + p j I )X j 1 = GG T X 2 j 1 (F T p j I ) (F + p j I )Xj T = GG T X T (F T p j 1 j I ) 2 Rewrite as one step iteration and factorize X i = Z i Zi T, i = 0,..., J Z 0 Z0 T = 0 Z j Zj T = 2p j (F + p j I ) 1 GG T (F + p j I ) T +(F + p j I ) 1 (F p j I )Z j 1 Zj 1 T ji ) T (F + p j I ) T 5/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

13 Introduction Review: LRCF-ADI for Lyapunov Equations Z j = [ 2p j (F + p j I ) 1 G, (F + p j I ) 1 (F p j I )Z j 1 ] [Penzl 00] 6/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

14 Introduction Review: LRCF-ADI for Lyapunov Equations Z j = [ 2p j (F + p j I ) 1 G, (F + p j I ) 1 (F p j I )Z j 1 ] [Penzl 00] Observing that (F p i I ), (F + p k I ) 1 commute, we rewrite Z J as Z J = [z J, P J 1 z J, P J 2 (P J 1 z J ),..., P 1 (P 2 P J 1 z J )], where and P i := z J = 2p J (F + p J I ) 1 G 2pi 2pi+1 [ I (pi + p i+1 )(F + p i I ) 1]. [Li/White 02] 6/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

15 Introduction Review: LRCF-ADI for Lyapunov Equations Algorithm 1 Low-rank Cholesky factor ADI iteration (LRCF-ADI) [Penzl 97/ 00, Li/White 99/ 02, B./Li/Penzl 99/ 08] Input: F,G defining FX + XF T = GG T and shifts {p 1,..., p imax } Output: Z = Z imax C n t imax, such that ZZ H X 1: For V 1 solve (F + p 1 I ) V 1 = 2 Re (p 1 )G 2: Z 1 = V 1 3: for i = 2, 3,..., i max do 4: For Ṽ solve (F + p i I )Ṽ = V i 1 5: V i = ( ) Re (p i )/ Re (p i 1 ) V i 1 (p i + p i 1 )Ṽ 6: Z i = [Z i 1 V i ] 7: end for 7/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

16 Introduction Review: LRCF-ADI for Lyapunov Equations Algorithm 1 General. Low-rank Cholesky factor ADI iteration (G-LRCF-ADI) [B. 04, B./Saak 09, S. 09] Input: E,F,G defining FXE T +EXF T = GG T and shifts {p 1,..., p imax } Output: Z = Z imax C n t imax, such that ZZ H X 1: For V 1 solve (F + p 1 E) V 1 = 2 Re (p 1 )G 2: Z 1 = V 1 3: for i = 2, 3,..., i max do 4: For Ṽ solve (F + p i E)Ṽ = EV i 1 5: V i = ( ) Re (p i )/ Re (p i 1 ) V i 1 (p i + p i 1 )Ṽ 6: Z i = [Z i 1 V i ] 7: end for 7/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

17 Introduction Krylov Subspace Based Solvers for Lyapunov Equations Consider Schur/singular value decomposition X = UΣU T, U R n n, U T U = I, Σ = diag (σ 1,..., σ n ) and σ 1 σ 2 σ n. The best rank-m Frobenius-norm approximation to X is thus given by [ ] Σm 0 X m := U U T = U 0 0 m Σ m Um T. 8/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

18 Introduction Krylov Subspace Based Solvers for Lyapunov Equations Consider Schur/singular value decomposition X = UΣU T, U R n n, U T U = I, Σ = diag (σ 1,..., σ n ) and σ 1 σ 2 σ n. The best rank-m Frobenius-norm approximation to X is thus given by [ ] Σm 0 X m := U U T = U 0 0 m Σ m Um T. Krylov projection idea [Saad 90, Jaimoukha/Kasenally 94] Solve (Um T FU m )Y m + Y m (Um T F T U m ) = Um T GG T U m, (1) on colspan(u m ) and get X m as X m = U m Y m Um T. 8/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

19 Introduction Krylov Subspace Based Solvers for Lyapunov Equations Consider Schur/singular value decomposition X = UΣU T, U R n n, U T U = I, Σ = diag (σ 1,..., σ n ) and σ 1 σ 2 σ n. The best rank-m Frobenius-norm approximation to X is thus given by [ ] Σm 0 X m := U U T = U 0 0 m Σ m Um T. Note that a factorization Z m Z T m = X m can easily be computed from a Cholesky factorization of Y m = Z m Z T m as Z m = U m Zm. 8/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

20 Introduction Krylov Subspace Based Solvers for Lyapunov Equations Algorithm 2 Basic Krylov Subspace Method for the Lyapunov Equation Input: F,G defining FX + XF T = GG T, an initial Krylov subspace V, e.g., V = K p (F, G) with orthogonal basis V C n p. Output: Z C n t, such that ZZ H X repeat if not first step then increase dimension of V and update V. end if Solve the small LE for Z with a classical solver: (V T FV ) Z Z T + Z Z T (V T F T V ) = V T GG T V, Lift Z to the full space: Z = U Z until res(z)< TOL 9/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

21 LRCF-ADI with Galerkin-Projection-Acceleration ADI and Rational Krylov [Li 00; Theorem 2] interprets the column span of the ADI solution as a certain rational Krylov subspace 8 < 1 L(F, G, p) := span :..., Y (F + p i I ) 1 G,..., (F + p 2 I ) 1 (F + p 1 I ) 1 G, i= j (F + p 1 I ) 1 G, G, (F + p 1 I )G, 9 jy = (F + p 2 I )(F + p 1 I )G,..., (F + p i I )G... ; i=1 10/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

22 LRCF-ADI with Galerkin-Projection-Acceleration ADI and Rational Krylov [Li 00; Theorem 2] interprets the column span of the ADI solution as a certain rational Krylov subspace 8 < 1 L(F, G, p) := span :..., Y (F + p i I ) 1 G,..., (F + p 2 I ) 1 (F + p 1 I ) 1 G, i= j (F + p 1 I ) 1 G, G, (F + p 1 I )G, 9 jy = (F + p 2 I )(F + p 1 I )G,..., (F + p i I )G... ; i=1 Idea Solve on current subspace of L(F, G, p) in the ADI step to increase the quality of the iterate. 10/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

23 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step Projected ADI Step LRCF-ADI-GP [B./Li/Truhar 09, Saak 09, B./Saak 10] 1 Compute the LRCF-ADI iterate Z i 2 Compute orthogonal basis via QR factorization: Q i R i Π i = Z i a 3 Solve (for Z) the projected Lyapunov equation (Qi T FQ i ) Z Z T + Z Z T (Qi T F T Q i ) = Qi T GG T Q i 4 Update Z i according to Z i := Q i Z a economy size QR with column pivoting; crucial to compute correct subspace if Z i rank deficient. 11/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

24 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step Projected ADI Step LRCF-ADI-GP [B./Li/Truhar 09, Saak 09, B./Saak 10] 1 Compute the LRCF-ADI iterate Z i 2 Compute orthogonal basis via QR factorization: Q i R i Π i = Z i 3 Solve (for Z) the projected Lyapunov equation (Q T i FQ i ) Z Z T + Z Z T (Q T i F T Q i ) = Q T i GG T Q i 4 Update Z i according to Z i := Q i Z Need to ensure that projected systems remain stable, e.g., F + F T < 0 may perform projected ADI step only every k-th step (e.g. k = 5) restarted ADI with shifts Λ(Q T i FQ i ). 11/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

25 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step Projected ADI Step G-LRCF-ADI-GP [B./Li/Truhar 09, Saak 09, B./Saak 10] 1 Compute the G-LRCF-ADI iterate Z i 2 Compute orthogonal basis via QR factorization: Q i R i Π i = Z i 3 Solve (for Z) the projected Lyapunov equation (Q T i FQ i ) Z Z T (Q T i E T Q i ) + (Q T i EQ i ) Z Z T (Q T i F T Q i ) = Q T i GG T Q i 4 Update Z i according to Z i := Q i Z 11/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

26 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step F Z Z T + Z Z T F T = G G T Legend: new factor old factor original matrix original rhs projected matrix projected Cholesky factor projected rhs 12/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

27 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step F Z Z T + Z Z T F T = G G T F m Fm T G Gm T m Legend: new factor old factor original matrix original rhs projected matrix projected Cholesky factor projected rhs 12/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

28 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step F m Cm C T m + C T m F T m G m Cm = G T m Legend: new factor old factor original matrix original rhs projected matrix projected Cholesky factor projected rhs 12/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

29 LRCF-ADI with Galerkin-Projection-Acceleration Projected ADI Step F m Cm C T m + C T m F T m G m Cm = G T m F Z Z T + Z Z T F T = G G T Legend: new factor old factor original matrix original rhs projected matrix projected Cholesky factor projected rhs 12/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

30 LRCF-ADI with Galerkin-Projection-Acceleration Test Example: Optimal Cooling of Steel Profiles Mathematical model: boundary control for linearized 2D heat equation. c ρ t x = λ x, ξ Ω λ n x = κ(u k x), ξ Γ k, 1 k 7, x = 0, ξ Γ0. n = q = 7, p = FEM Discretization, different models for initial mesh (n = 371), 1, 2, 3, 4 steps of mesh refinement n = 1 357, 5 177, , Source: Physical model: courtesy of Mannesmann/Demag. Math. model: Tröltzsch/Unger 99/ 01, Penzl 99, S /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

31 LRCF-ADI with Galerkin-Projection-Acceleration Numerical Results steel profile n= good shifts 10 0 Iteration history for controllability gramian no projection every step every 5 steps 10 0 Iteration history for observability gramian no projection every step every 5 steps 10 2 normalized residual normalized residual iteration number iteration number 14/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

32 LRCF-ADI with Galerkin-Projection-Acceleration Numerical Results steel profile n= good shifts 100 Computation times time in seconds galerkin projection frequency 14/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

33 LRCF-ADI with Galerkin-Projection-Acceleration Numerical Results steel profile n= bad shifts 10 0 Iteration history for controllability gramian Iteration history for observability gramian normalized residual normalized residual no projection every step every 5 steps iteration number 10 6 no projection every step every 5 steps iteration number 15/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

34 LRCF-ADI with Galerkin-Projection-Acceleration Numerical Results steel profile n= bad shifts 2500 Computation times 2000 time in seconds galerkin projection frequency 15/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

35 LRCF-NM for the ARE 1 Introduction 2 LRCF-ADI with Galerkin-Projection-Acceleration 3 LRCF-NM for the ARE Newton s Method for AREs Low-Rank Newton-ADI (LRCF-NM) for AREs Test Examples Test Results (ADI-loop) Test Results (both-loops) Computation Time Scaling with Problem Size 16/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

36 LRCF-NM for the ARE Newton s Method for AREs Consider R(X ) := C T C + A T X + XA XBB T X = 0 Newton s Iteration for the ARE R X (N l ) = R(X l ), X l+1 = X l + N l, l = 0, 1,... where the Frechét derivative of R at X is the Lyapunov operator R X : Q (A BB T X ) T Q + Q(A BB T X ), i.e., in every Newton step solve a Lyapunov Equation [Kleinman 68] (A BB T X l ) T X l+1 + X l+1 (A BB T X l ) = C T C X l BB T X l. 17/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

37 LRCF-NM for the ARE Newton s Method for AREs Consider R(X ) := C T C + A T X + XA XBB T X = 0 Newton s Iteration for the ARE R X (N l ) = R(X l ), X l+1 = X l + N l, l = 0, 1,... where the Frechét derivative of R at X is the Lyapunov operator R X : Q (A BB T X ) T Q + Q(A BB T X ), i.e., in every Newton step solve a Lyapunov Equation [Kleinman 68] F T l X l+1 + X l+1 F l = G l G T l. 17/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

38 LRCF-NM for the ARE Newton s Method for AREs Consider R(X ) := C T C + A T X E + E T XA E T XBB T X E = 0 Newton s Iteration for the ARE R X (N l ) = R(X l ), X l+1 = X l + N l, l = 0, 1,... where the Frechét derivative of R at X is the Lyapunov operator R X : Q (A BB T X E) T QE + E T Q(A BB T X E), i.e., in every Newton step solve a Lyapunov Equation [Kleinman 68] F T l X l+1 E + E T X l+1 F l = G l G T l. 17/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

39 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Factored Newton-Kleinman Iteration [Benner/Li/Penzl 99/ 08] F l = A BB T X l =: A BK l G l = [C T Kl T ] is sparse + low rank is low rank factor 18/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

40 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Factored Newton-Kleinman Iteration [Benner/Li/Penzl 99/ 08] F l = A BB T X l =: A BK l G l = [C T Kl T ] apply LRCF-ADI in every Newton step is sparse + low rank is low rank factor exploit structure of F l using Sherman-Morrison-Woodbury formula 18/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

41 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Factored Newton-Kleinman Iteration [Benner/Li/Penzl 99/ 08] F l = A BB T X l E =: A BK l G l = [C T Kl T ] apply LRCF-ADI in every Newton step is sparse + low rank is low rank factor exploit structure of F l using Sherman-Morrison-Woodbury formula 18/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

42 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Algorithm 3 Low-Rank Cholesky Factor Newton Method (LRCF-NM) Input: A, B, C, K (0) for which A BK (0)T is stable Output: Z = Z (kmax ), such that ZZ H approximates the solution X of C T C + A T X + XA XBB T X = 0. 1: for k = 1, 2,..., k max do 2: Determine (sub)optimal ADI shift parameters p (k) 1, p(k) 2,... with respect to the matrix F (k) = A T K (k 1) B T. 3: G (k) = [ C T K (k 1) ] 4: Compute Z (k) using Algorithm 1 (LRCF-ADI) such that F (k) Z (k) Z (k)h + Z (k) Z (k)h F (k)t G (k) G (k)t. 5: K (k) = Z (k) (Z (k)h B) 6: end for 19/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

43 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Algorithm 3 Low-Rank Cholesky Factor Newton Method (G-LRCF-NM) Input: E, A, B, C, K (0) for which A BK (0)T is stable Output: Z = Z (kmax ), such that ZZ H approximates the solution X of C T C + A T X E + E T XA E T XBB T X E = 0. 1: for k = 1, 2,..., k max do 2: Determine (sub)optimal ADI shift parameters p (k) 1, p(k) 2,... with respect to the matrix F (k) = A T E T K (k 1) B T E T. 3: G (k) = [ C T K (k 1) ] 4: Compute Z (k) using Algorithm 1 (G-LRCF-ADI) such that F (k) Z (k) Z (k)h E + E T Z (k) Z (k)h F (k)t G (k) G (k)t. 5: K (k) = E T (Z (k) (Z (k)h B)) 6: end for 19/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

44 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Algorithm 3 Low-Rank Cholesky Factor Newton Method (LRCF-NM) Input: A, B, C, K (0) for which A BK (0)T is stable Output: Z = Z (kmax ), such that ZZ H approximates the solution X of C T C + A T X + XA XBB T X = 0. 1: for k = 1, 2,..., k max do 2: Determine (sub)optimal ADI shift parameters p (k) 1, p(k) 2,... with respect to the matrix F (k) = A T K (k 1) B T. 3: G (k) = [ C T K (k 1) ] 4: Compute Z (k) using Algorithm 1 (LRCF-ADI) or (LRCF-ADI-GP) such that F (k) Z (k) Z (k)h + Z (k) Z (k)h F (k)t G (k) G (k)t. 5: K (k) = Z (k) (Z (k)h B) 6: end for 20/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

45 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Algorithm 4 Simpl. Low-Rank Cholesky Factor Newton Method (LRCF-NM-S) Input: A, B, C, K (0) for which A BK (0)T is stable Output: Z = Z (kmax ), such that ZZ H approximates the solution X of C T C + A T X + XA XBB T X = 0. 1: Determine (sub)optimal ADI shift parameters p 1, p 2,... with respect to the matrix F (k) = A T K (0) B T. 2: for k = 1, 2,..., k max do 3: G (k) = [ C T K (k 1) ] 4: Compute Z (k) using Algorithm 1 (LRCF-ADI) or (LRCF-ADI-GP) such that F (k) Z (k) Z (k)h + Z (k) Z (k)h F (k)t G (k) G (k)t. 5: K (k) = Z (k) (Z (k)h B) 6: end for 20/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

46 LRCF-NM for the ARE Low-Rank Newton-ADI (LRCF-NM) for AREs Algorithm 5 Low-Rank Cholesky Factor Galerkin-Newton Method (LRCF-NM-GP) Input: A, B, C, K (0) for which A BK (0)T is stable Output: Z = Z (kmax ), such that ZZ H approximates the solution X of C T C + A T X + XA XBB T X = 0. 1: for k = 1, 2,..., k max do 2: Determine (sub)optimal ADI shift parameters p (k) 1, p(k) 2,... with respect to the matrix F (k) = A T K (k 1) B T. 3: G (k) = [ C T K (k 1) ] 4: Compute Z (k) using Algorithm 1 (LRCF-ADI) or (LRCF-ADI-GP) such that F (k) Z (k) Z (k)h + Z (k) Z (k)h F (k)t G (k) G (k)t. 5: Project ARE, solve and prolongate solution 6: K (k) = Z (k) (Z (k)h B) 7: end for 20/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

47 LRCF-NM for the ARE Test Examples Example 1: 3d Convection-Diffusion Equation FDM for 3D convection-diffusion equation on [0, 1] 3 proposed in [Simoncini 07], q = p = 1 non-symmetric A R n n, n = Example 2: 2d Convection-Diffusion Equation FDM for 2D convection-diffusion equations on [0, 1] 2 LyaPack benchmark, q = p = 1, e.g., demo l1 non-symmetric A R n n, n = shift parameters Penzl s heuristic from 50/25 Ritz/harmonic Ritz values of A 21/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

48 LRCF-NM for the ARE Test Results (ADI-loop): Example 1 Newton-ADI Newton-Galerkin-ADI LRCF-ADI-GP(5) NWT rel. change rel. residual ADI NWT rel. change rel. residual ADI CPU time: sec. CPU time: sec. test system: Intel Xeon GHz ; 16 GB RAM; 64Bit-MATLAB (R2010a) using threaded BLAS (romulus) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

49 LRCF-NM for the ARE Test Results (ADI-loop): Example 2 Newton-ADI Newton-Galerkin-ADI LRCF-ADI-GP(5) NWT rel. change rel. residual ADI NWT rel. change rel. residual ADI CPU time: sec. CPU time: sec. test system: Intel Core 2 Quad Q GHz; 4 GB RAM; 64Bit-MATLAB (R2009a) using threaded BLAS (reynolds) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

50 LRCF-NM for the ARE Test Results (both-loops): Example 1 Newton-ADI NWT rel. change rel. residual ADI CPU time: sec. NG-ADI inner= 5, outer= 1 NWT rel. change rel. residual ADI CPU time: sec. NG-ADI inner= 1, outer= 1 NWT rel. change rel. residual ADI CPU time: sec. NG-ADI inner= 0, outer= 1 NWT rel. change rel. residual ADI CPU time: sec. test system: Intel Xeon GHz ; 16 GB RAM; 64Bit-MATLAB (R2010a) using threaded BLAS (romulus) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

51 LRCF-NM for the ARE Test Results (both-loops): Example 2 Newton-ADI NWT rel. change rel. residual ADI CPU time: sec. NG-ADI inner= 5, outer= 1 NWT rel. change rel. residual ADI CPU time: 24.1 sec. NG-ADI inner= 1, outer= 1 NWT rel. change rel. residual ADI CPU time: 26.8 sec. NG-ADI inner= 0, outer= 1 NWT rel. change rel. residual ADI CPU time: 24.0 sec. test system: Intel Core 2 Quad Q GHz; 4 GB RAM; 64Bit-MATLAB (R2009a) using threaded BLAS (reynolds) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

52 LRCF-NM for the ARE Computation Time Scaling with Problem Size (0, 1) (1, 1) tx(ξ, t) = x(ξ, t) in Ω Ω Γ c νx = b(ξ) u(t) x νx = x on Γ c on Ω \ Γ c x(ξ, 0) = 1 (0, 0) (1, 0) Note: Here b(ξ) = 4 (1 ξ 2 ) ξ 2 for ξ Γ c and 0 otherwise, thus t R >0, we have u(t) R. 26/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

53 LRCF-NM for the ARE Computation Time Scaling with Problem Size (0, 1) (1, 1) tx(ξ, t) = x(ξ, t) in Ω Ω Γ c νx = b(ξ) u(t) x νx = x on Γ c on Ω \ Γ c x(ξ, 0) = 1 (0, 0) (1, 0) Note: Here b(ξ) = 4 (1 ξ 2 ) ξ 2 for ξ Γ c and 0 otherwise, thus t R >0, we have u(t) R. B h = M Γ,h b. 26/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

54 LRCF-NM for the ARE Computation Time Scaling with Problem Size (0, 1) (1, 1) Ω Γ c tx(ξ, t) = x(ξ, t) νx = b(ξ) u(t) x νx = x x(ξ, 0) = 1 in Ω on Γ c on Ω \ Γ c (0, 0) (1, 0) Consider: output equation y = Cx, where C : L 2 (Ω) x(ξ, t) R y(t) = x(ξ, t) dξ. Ω 26/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

55 LRCF-NM for the ARE Computation Time Scaling with Problem Size (0, 1) (1, 1) Ω Γ c tx(ξ, t) = x(ξ, t) νx = b(ξ) u(t) x νx = x x(ξ, 0) = 1 in Ω on Γ c on Ω \ Γ c (0, 0) (1, 0) Consider: output equation y = Cx, where C : L 2 (Ω) R x(ξ, t) y(t) = Ω x(ξ, t) dξ, C h = 1 M h. 26/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

56 LRCF-NM for the ARE Computation Time Scaling with Problem Size (0, 1) (1, 1) tx(ξ, t) = x(ξ, t) in Ω Ω Γ c νx = b(ξ) u(t) x νx = x on Γ c on Ω \ Γ c x(ξ, 0) = 1 (0, 0) (1, 0) Cost Function: J (u) = 0 y 2 (t) + u 2 (t) dt. 26/27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

57 LRCF-NM for the ARE Computation Time Scaling with Problem Size simplified Low Rank Newton-Galerkin ADI generalized state space form implementation Penzl shifts (16/50/25) with respect to initial matrices projection acceleration in every outer iteration step projection acceleration in every 5-th inner iteration step test system: Intel Xeon 3.00 GHz; 16 GB RAM; 64Bit-MATLAB (R2010a) using threaded BLAS (romulus) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

58 LRCF-NM for the ARE Computation Time Scaling with Problem Size Computation Times discretization level problem size time in seconds test system: Intel Xeon 3.00 GHz; 16 GB RAM; 64Bit-MATLAB (R2010a) using threaded BLAS (romulus) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

59 LRCF-NM for the ARE Computation Time Scaling with Problem Size Scaling of CPU time time in seconds refinement level test system: Intel Xeon 3.00 GHz; 16 GB RAM; 64Bit-MATLAB (R2010a) using threaded BLAS (romulus) stopping criterion tolerances: /27 Peter Benner, Jens Saak Newton-Galerkin-ADI for AREs

Numerical Solution of Descriptor Riccati Equations for Linearized Navier-Stokes Control Problems

Numerical Solution of Descriptor Riccati Equations for Linearized Navier-Stokes Control Problems MAX PLANCK INSTITUTE Control and Optimization with Differential-Algebraic Constraints Banff, October 24 29, 2010 Numerical Solution of Descriptor Riccati Equations for Linearized Navier-Stokes Control

More information

Model Reduction for Dynamical Systems

Model Reduction for Dynamical Systems Otto-von-Guericke Universität Magdeburg Faculty of Mathematics Summer term 2015 Model Reduction for Dynamical Systems Lecture 8 Peter Benner Lihong Feng Max Planck Institute for Dynamics of Complex Technical

More information

Efficient Implementation of Large Scale Lyapunov and Riccati Equation Solvers

Efficient Implementation of Large Scale Lyapunov and Riccati Equation Solvers Efficient Implementation of Large Scale Lyapunov and Riccati Equation Solvers Jens Saak joint work with Peter Benner (MiIT) Professur Mathematik in Industrie und Technik (MiIT) Fakultät für Mathematik

More information

The Newton-ADI Method for Large-Scale Algebraic Riccati Equations. Peter Benner.

The Newton-ADI Method for Large-Scale Algebraic Riccati Equations. Peter Benner. The Newton-ADI Method for Large-Scale Algebraic Riccati Equations Mathematik in Industrie und Technik Fakultät für Mathematik Peter Benner benner@mathematik.tu-chemnitz.de Sonderforschungsbereich 393 S

More information

S N. hochdimensionaler Lyapunov- und Sylvestergleichungen. Peter Benner. Mathematik in Industrie und Technik Fakultät für Mathematik TU Chemnitz

S N. hochdimensionaler Lyapunov- und Sylvestergleichungen. Peter Benner. Mathematik in Industrie und Technik Fakultät für Mathematik TU Chemnitz Ansätze zur numerischen Lösung hochdimensionaler Lyapunov- und Sylvestergleichungen Peter Benner Mathematik in Industrie und Technik Fakultät für Mathematik TU Chemnitz S N SIMULATION www.tu-chemnitz.de/~benner

More information

BALANCING-RELATED MODEL REDUCTION FOR LARGE-SCALE SYSTEMS

BALANCING-RELATED MODEL REDUCTION FOR LARGE-SCALE SYSTEMS FOR LARGE-SCALE SYSTEMS Professur Mathematik in Industrie und Technik Fakultät für Mathematik Technische Universität Chemnitz Recent Advances in Model Order Reduction Workshop TU Eindhoven, November 23,

More information

Passivity Preserving Model Reduction for Large-Scale Systems. Peter Benner.

Passivity Preserving Model Reduction for Large-Scale Systems. Peter Benner. Passivity Preserving Model Reduction for Large-Scale Systems Peter Benner Mathematik in Industrie und Technik Fakultät für Mathematik Sonderforschungsbereich 393 S N benner@mathematik.tu-chemnitz.de SIMULATION

More information

MODEL REDUCTION BY A CROSS-GRAMIAN APPROACH FOR DATA-SPARSE SYSTEMS

MODEL REDUCTION BY A CROSS-GRAMIAN APPROACH FOR DATA-SPARSE SYSTEMS MODEL REDUCTION BY A CROSS-GRAMIAN APPROACH FOR DATA-SPARSE SYSTEMS Ulrike Baur joint work with Peter Benner Mathematics in Industry and Technology Faculty of Mathematics Chemnitz University of Technology

More information

arxiv: v1 [math.na] 4 Apr 2018

arxiv: v1 [math.na] 4 Apr 2018 Efficient Solution of Large-Scale Algebraic Riccati Equations Associated with Index-2 DAEs via the Inexact Low-Rank Newton-ADI Method Peter Benner a,d, Matthias Heinkenschloss b,1,, Jens Saak a, Heiko

More information

Numerical solution of large-scale Lyapunov equations, Riccati equations, and linear-quadratic optimal control problems

Numerical solution of large-scale Lyapunov equations, Riccati equations, and linear-quadratic optimal control problems NUMERICAL LINEAR ALGEBRA WITH APPLICATIONS Numer. Linear Algebra Appl. 2008; 15:755 777 Published online in Wiley InterScience (www.interscience.wiley.com)..622 Numerical solution of large-scale Lyapunov

More information

Numerical Solution of Matrix Equations Arising in Control of Bilinear and Stochastic Systems

Numerical Solution of Matrix Equations Arising in Control of Bilinear and Stochastic Systems MatTriad 2015 Coimbra September 7 11, 2015 Numerical Solution of Matrix Equations Arising in Control of Bilinear and Stochastic Systems Peter Benner Max Planck Institute for Dynamics of Complex Technical

More information

BALANCING-RELATED MODEL REDUCTION FOR DATA-SPARSE SYSTEMS

BALANCING-RELATED MODEL REDUCTION FOR DATA-SPARSE SYSTEMS BALANCING-RELATED Peter Benner Professur Mathematik in Industrie und Technik Fakultät für Mathematik Technische Universität Chemnitz Computational Methods with Applications Harrachov, 19 25 August 2007

More information

Accelerating Model Reduction of Large Linear Systems with Graphics Processors

Accelerating Model Reduction of Large Linear Systems with Graphics Processors Accelerating Model Reduction of Large Linear Systems with Graphics Processors P. Benner 1, P. Ezzatti 2, D. Kressner 3, E.S. Quintana-Ortí 4, Alfredo Remón 4 1 Max-Plank-Institute for Dynamics of Complex

More information

Order reduction numerical methods for the algebraic Riccati equation. V. Simoncini

Order reduction numerical methods for the algebraic Riccati equation. V. Simoncini Order reduction numerical methods for the algebraic Riccati equation V. Simoncini Dipartimento di Matematica Alma Mater Studiorum - Università di Bologna valeria.simoncini@unibo.it 1 The problem Find X

More information

Moving Frontiers in Model Reduction Using Numerical Linear Algebra

Moving Frontiers in Model Reduction Using Numerical Linear Algebra Using Numerical Linear Algebra Max-Planck-Institute for Dynamics of Complex Technical Systems Computational Methods in Systems and Control Theory Group Magdeburg, Germany Technische Universität Chemnitz

More information

Factorized Solution of Sylvester Equations with Applications in Control

Factorized Solution of Sylvester Equations with Applications in Control Factorized Solution of Sylvester Equations with Applications in Control Peter Benner Abstract Sylvester equations play a central role in many areas of applied mathematics and in particular in systems and

More information

Structure preserving Krylov-subspace methods for Lyapunov equations

Structure preserving Krylov-subspace methods for Lyapunov equations Structure preserving Krylov-subspace methods for Lyapunov equations Matthias Bollhöfer, André Eppler Institute Computational Mathematics TU Braunschweig MoRePas Workshop, Münster September 17, 2009 System

More information

A POD Projection Method for Large-Scale Algebraic Riccati Equations

A POD Projection Method for Large-Scale Algebraic Riccati Equations A POD Projection Method for Large-Scale Algebraic Riccati Equations Boris Kramer Department of Mathematics Virginia Tech Blacksburg, VA 24061-0123 Email: bokr@vt.edu John R. Singler Department of Mathematics

More information

Towards high performance IRKA on hybrid CPU-GPU systems

Towards 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

Balanced Truncation Model Reduction of Large and Sparse Generalized Linear Systems

Balanced Truncation Model Reduction of Large and Sparse Generalized Linear Systems Balanced Truncation Model Reduction of Large and Sparse Generalized Linear Systems Jos M. Badía 1, Peter Benner 2, Rafael Mayo 1, Enrique S. Quintana-Ortí 1, Gregorio Quintana-Ortí 1, A. Remón 1 1 Depto.

More information

Parametrische Modellreduktion mit dünnen Gittern

Parametrische Modellreduktion mit dünnen Gittern Parametrische Modellreduktion mit dünnen Gittern (Parametric model reduction with sparse grids) Ulrike Baur Peter Benner Mathematik in Industrie und Technik, Fakultät für Mathematik Technische Universität

More information

Model Order Reduction of Continuous LTI Large Descriptor System Using LRCF-ADI and Square Root Balanced Truncation

Model Order Reduction of Continuous LTI Large Descriptor System Using LRCF-ADI and Square Root Balanced Truncation , July 1-3, 15, London, U.K. Model Order Reduction of Continuous LI Large Descriptor System Using LRCF-ADI and Square Root Balanced runcation Mehrab Hossain Likhon, Shamsil Arifeen, and Mohammad Sahadet

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

ADI-preconditioned FGMRES for solving large generalized Lyapunov equations - A case study

ADI-preconditioned FGMRES for solving large generalized Lyapunov equations - A case study -preconditioned for large - A case study Matthias Bollhöfer, André Eppler TU Braunschweig Institute Computational Mathematics Syrene-MOR Workshop, TU Hamburg October 30, 2008 2 / 20 Outline 1 2 Overview

More information

ON THE NUMERICAL SOLUTION OF LARGE-SCALE RICCATI EQUATIONS

ON THE NUMERICAL SOLUTION OF LARGE-SCALE RICCATI EQUATIONS ON THE NUMERICAL SOLUTION OF LARGE-SCALE RICCATI EQUATIONS VALERIA SIMONCINI, DANIEL B. SZYLD, AND MARLLINY MONSALVE Abstract. The inexact Newton-Kleinman method is an iterative scheme for numerically

More information

SYSTEM-THEORETIC METHODS FOR MODEL REDUCTION OF LARGE-SCALE SYSTEMS: SIMULATION, CONTROL, AND INVERSE PROBLEMS. Peter Benner

SYSTEM-THEORETIC METHODS FOR MODEL REDUCTION OF LARGE-SCALE SYSTEMS: SIMULATION, CONTROL, AND INVERSE PROBLEMS. Peter Benner SYSTEM-THEORETIC METHODS FOR MODEL REDUCTION OF LARGE-SCALE SYSTEMS: SIMULATION, CONTROL, AND INVERSE PROBLEMS Professur Mathematik in Industrie und Technik Fakultät für Mathematik Technische Universität

More information

Matrix Equations and and Bivariate Function Approximation

Matrix Equations and and Bivariate Function Approximation Matrix Equations and and Bivariate Function Approximation D. Kressner Joint work with L. Grasedyck, C. Tobler, N. Truhar. ETH Zurich, Seminar for Applied Mathematics Manchester, 17.06.2009 Sylvester matrix

More information

Linear and Nonlinear Matrix Equations Arising in Model Reduction

Linear and Nonlinear Matrix Equations Arising in Model Reduction International Conference on Numerical Linear Algebra and its Applications Guwahati, January 15 18, 2013 Linear and Nonlinear Matrix Equations Arising in Model Reduction Peter Benner Tobias Breiten Max

More information

Karhunen-Loève Approximation of Random Fields Using Hierarchical Matrix Techniques

Karhunen-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 information

Krylov Subspace Type Methods for Solving Projected Generalized Continuous-Time Lyapunov Equations

Krylov Subspace Type Methods for Solving Projected Generalized Continuous-Time Lyapunov Equations Krylov Subspace Type Methods for Solving Proected Generalized Continuous-Time Lyapunov Equations YUIAN ZHOU YIQIN LIN Hunan University of Science and Engineering Institute of Computational Mathematics

More information

Max Planck Institute Magdeburg Preprints

Max Planck Institute Magdeburg Preprints Peter Benner Patrick Kürschner Jens Saak Real versions of low-rank ADI methods with complex shifts MAXPLANCKINSTITUT FÜR DYNAMIK KOMPLEXER TECHNISCHER SYSTEME MAGDEBURG Max Planck Institute Magdeburg Preprints

More information

THE solution of the absolute value equation (AVE) of

THE solution of the absolute value equation (AVE) of The nonlinear HSS-like iterative method for absolute value equations Mu-Zheng Zhu Member, IAENG, and Ya-E Qi arxiv:1403.7013v4 [math.na] 2 Jan 2018 Abstract Salkuyeh proposed the Picard-HSS iteration method

More information

Iterative methods for positive definite linear systems with a complex shift

Iterative methods for positive definite linear systems with a complex shift Iterative methods for positive definite linear systems with a complex shift William McLean, University of New South Wales Vidar Thomée, Chalmers University November 4, 2011 Outline 1. Numerical solution

More information

Model order reduction of large-scale dynamical systems with Jacobi-Davidson style eigensolvers

Model order reduction of large-scale dynamical systems with Jacobi-Davidson style eigensolvers MAX PLANCK INSTITUTE International Conference on Communications, Computing and Control Applications March 3-5, 2011, Hammamet, Tunisia. Model order reduction of large-scale dynamical systems with Jacobi-Davidson

More information

Model reduction of coupled systems

Model reduction of coupled systems Model reduction of coupled systems Tatjana Stykel Technische Universität Berlin ( joint work with Timo Reis, TU Kaiserslautern ) Model order reduction, coupled problems and optimization Lorentz Center,

More information

Inexact inverse iteration with preconditioning

Inexact 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 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

On the numerical solution of large-scale linear matrix equations. V. Simoncini

On the numerical solution of large-scale linear matrix equations. V. Simoncini On the numerical solution of large-scale linear matrix equations V. Simoncini Dipartimento di Matematica, Università di Bologna (Italy) valeria.simoncini@unibo.it 1 Some matrix equations Sylvester matrix

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 rational Krylov subspace for parameter dependent systems. V. Simoncini

The rational Krylov subspace for parameter dependent systems. V. Simoncini The rational Krylov subspace for parameter dependent systems V. Simoncini Dipartimento di Matematica, Università di Bologna valeria.simoncini@unibo.it 1 Given the continuous-time system Motivation. Model

More information

Solving Large-Scale Matrix Equations: Recent Progress and New Applications

Solving Large-Scale Matrix Equations: Recent Progress and New Applications 19th Conference of the International Linear Algebra Society (ILAS) Seoul, August 6 9, 2014 Solving Large-Scale Matrix Equations: Recent Progress and New Applications Peter Benner Max Planck Institute for

More information

Exploiting off-diagonal rank structures in the solution of linear matrix equations

Exploiting off-diagonal rank structures in the solution of linear matrix equations Stefano Massei Exploiting off-diagonal rank structures in the solution of linear matrix equations Based on joint works with D. Kressner (EPFL), M. Mazza (IPP of Munich), D. Palitta (IDCTS of Magdeburg)

More information

In this article we discuss sparse matrix algorithms and parallel algorithms, Solving Large-Scale Control Problems

In this article we discuss sparse matrix algorithms and parallel algorithms, Solving Large-Scale Control Problems F E A T U R E Solving Large-Scale Control Problems EYEWIRE Sparsity and parallel algorithms: two approaches to beat the curse of dimensionality. By Peter Benner In this article we discuss sparse matrix

More information

MPC/LQG for infinite-dimensional systems using time-invariant linearizations

MPC/LQG for infinite-dimensional systems using time-invariant linearizations MPC/LQG for infinite-dimensional systems using time-invariant linearizations Peter Benner 1 and Sabine Hein 2 1 Max Planck Institute for Dynamics of Complex Technical Systems, Sandtorstr. 1, 39106 Magdeburg,

More information

arxiv: v1 [hep-lat] 2 May 2012

arxiv: v1 [hep-lat] 2 May 2012 A CG Method for Multiple Right Hand Sides and Multiple Shifts in Lattice QCD Calculations arxiv:1205.0359v1 [hep-lat] 2 May 2012 Fachbereich C, Mathematik und Naturwissenschaften, Bergische Universität

More information

arxiv: v2 [math.na] 8 Apr 2017

arxiv: v2 [math.na] 8 Apr 2017 A LOW-RANK MULTIGRID METHOD FOR THE STOCHASTIC STEADY-STATE DIFFUSION PROBLEM HOWARD C. ELMAN AND TENGFEI SU arxiv:1612.05496v2 [math.na] 8 Apr 2017 Abstract. We study a multigrid method for solving large

More information

LYAPACK. A MATLAB Toolbox for Large Lyapunov and Riccati Equations, Model Reduction Problems, and Linear Quadratic Optimal Control Problems

LYAPACK. A MATLAB Toolbox for Large Lyapunov and Riccati Equations, Model Reduction Problems, and Linear Quadratic Optimal Control Problems 1 LYAPACK A MATLAB Toolbox for Large Lyapunov and Riccati Equations, Model Reduction Problems, and Linear Quadratic Optimal Control Problems Users Guide (Version 1.0) Thilo Penzl Preface Control theory

More information

Lecture 3: Inexact inverse iteration with preconditioning

Lecture 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 information

Balancing-Related Model Reduction for Large-Scale Systems

Balancing-Related Model Reduction for Large-Scale Systems Balancing-Related Model Reduction for Large-Scale Systems Peter Benner Professur Mathematik in Industrie und Technik Fakultät für Mathematik Technische Universität Chemnitz D-09107 Chemnitz benner@mathematik.tu-chemnitz.de

More information

Numerical Methods for Large-Scale Nonlinear Systems

Numerical Methods for Large-Scale Nonlinear Systems Numerical Methods for Large-Scale Nonlinear Systems Handouts by Ronald H.W. Hoppe following the monograph P. Deuflhard Newton Methods for Nonlinear Problems Springer, Berlin-Heidelberg-New York, 2004 Num.

More information

On Solving Large Algebraic. Riccati Matrix Equations

On Solving Large Algebraic. Riccati Matrix Equations International Mathematical Forum, 5, 2010, no. 33, 1637-1644 On Solving Large Algebraic Riccati Matrix Equations Amer Kaabi Department of Basic Science Khoramshahr Marine Science and Technology University

More information

arxiv: v3 [math.na] 9 Oct 2016

arxiv: v3 [math.na] 9 Oct 2016 RADI: A low-ran ADI-type algorithm for large scale algebraic Riccati equations Peter Benner Zvonimir Bujanović Patric Kürschner Jens Saa arxiv:50.00040v3 math.na 9 Oct 206 Abstract This paper introduces

More information

Structured Krylov Subspace Methods for Eigenproblems with Spectral Symmetries

Structured Krylov Subspace Methods for Eigenproblems with Spectral Symmetries Structured Krylov Subspace Methods for Eigenproblems with Spectral Symmetries Fakultät für Mathematik TU Chemnitz, Germany Peter Benner benner@mathematik.tu-chemnitz.de joint work with Heike Faßbender

More information

Lecture 1: Numerical Issues from Inverse Problems (Parameter Estimation, Regularization Theory, and Parallel Algorithms)

Lecture 1: Numerical Issues from Inverse Problems (Parameter Estimation, Regularization Theory, and Parallel Algorithms) Lecture 1: Numerical Issues from Inverse Problems (Parameter Estimation, Regularization Theory, and Parallel Algorithms) Youzuo Lin 1 Joint work with: Rosemary A. Renaut 2 Brendt Wohlberg 1 Hongbin Guo

More information

MPI at MPI. Jens Saak. Max Planck Institute for Dynamics of Complex Technical Systems Computational Methods in Systems and Control Theory

MPI at MPI. Jens Saak. Max Planck Institute for Dynamics of Complex Technical Systems Computational Methods in Systems and Control Theory MAX PLANCK INSTITUTE November 5, 2010 MPI at MPI Jens Saak Max Planck Institute for Dynamics of Complex Technical Systems Computational Methods in Systems and Control Theory FOR DYNAMICS OF COMPLEX TECHNICAL

More information

SOLUTION OF ALGEBRAIC RICCATI EQUATIONS ARISING IN CONTROL OF PARTIAL DIFFERENTIAL EQUATIONS

SOLUTION OF ALGEBRAIC RICCATI EQUATIONS ARISING IN CONTROL OF PARTIAL DIFFERENTIAL EQUATIONS SOLUTION OF ALGEBRAIC RICCATI EQUATIONS ARISING IN CONTROL OF PARTIAL DIFFERENTIAL EQUATIONS Kirsten Morris Department of Applied Mathematics University of Waterloo Waterloo, CANADA N2L 3G1 kmorris@uwaterloo.ca

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

Finite Element Methods with Linear B-Splines

Finite Element Methods with Linear B-Splines Finite Element Methods with Linear B-Splines K. Höllig, J. Hörner and M. Pfeil Universität Stuttgart Institut für Mathematische Methoden in den Ingenieurwissenschaften, Numerik und geometrische Modellierung

More information

Low Rank Approximation Lecture 7. Daniel Kressner Chair for Numerical Algorithms and HPC Institute of Mathematics, EPFL

Low Rank Approximation Lecture 7. Daniel Kressner Chair for Numerical Algorithms and HPC Institute of Mathematics, EPFL Low Rank Approximation Lecture 7 Daniel Kressner Chair for Numerical Algorithms and HPC Institute of Mathematics, EPFL daniel.kressner@epfl.ch 1 Alternating least-squares / linear scheme General setting:

More information

On the numerical solution of large-scale sparse discrete-time Riccati equations

On the numerical solution of large-scale sparse discrete-time Riccati equations Advances in Computational Mathematics manuscript No (will be inserted by the editor) On the numerical solution of large-scale sparse discrete-time Riccati equations Peter Benner Heike Faßbender Received:

More information

DELFT UNIVERSITY OF TECHNOLOGY

DELFT UNIVERSITY OF TECHNOLOGY DELFT UNIVERSITY OF TECHNOLOGY REPORT 16-02 The Induced Dimension Reduction method applied to convection-diffusion-reaction problems R. Astudillo and M. B. van Gijzen ISSN 1389-6520 Reports of the Delft

More information

Parallel Solution of Large-Scale and Sparse Generalized algebraic Riccati Equations

Parallel Solution of Large-Scale and Sparse Generalized algebraic Riccati Equations Parallel Solution of Large-Scale and Sparse Generalized algebraic Riccati Equations José M. Badía 1, Peter Benner 2, Rafael Mayo 1, and Enrique S. Quintana-Ortí 1 1 Depto. de Ingeniería y Ciencia de Computadores,

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

Solving the Stochastic Steady-State Diffusion Problem Using Multigrid

Solving the Stochastic Steady-State Diffusion Problem Using Multigrid Solving the Stochastic Steady-State Diffusion Problem Using Multigrid Tengfei Su Applied Mathematics and Scientific Computing Advisor: Howard Elman Department of Computer Science Sept. 29, 2015 Tengfei

More information

SOLVING DISCRETE-TIME ALGEBRAIC RICCATI EQUATIONS USING MODIFIED NEWTON S METHOD

SOLVING DISCRETE-TIME ALGEBRAIC RICCATI EQUATIONS USING MODIFIED NEWTON S METHOD PHYSCON 2013, San Luis Potosí, México, 26 29th August, 2013 SOLVING DISCRETE-TIME ALGEBRAIC RICCATI EQUATIONS USING MODIFIED NEWTON S METHOD Vasile Sima Advanced Research National Institute for Research

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

Rational Krylov methods for linear and nonlinear eigenvalue problems

Rational Krylov methods for linear and nonlinear eigenvalue problems Rational Krylov methods for linear and nonlinear eigenvalue problems Mele Giampaolo mele@mail.dm.unipi.it University of Pisa 7 March 2014 Outline Arnoldi (and its variants) for linear eigenproblems Rational

More information

Trust-Region SQP Methods with Inexact Linear System Solves for Large-Scale Optimization

Trust-Region SQP Methods with Inexact Linear System Solves for Large-Scale Optimization Trust-Region SQP Methods with Inexact Linear System Solves for Large-Scale Optimization Denis Ridzal Department of Computational and Applied Mathematics Rice University, Houston, Texas dridzal@caam.rice.edu

More information

Max Planck Institute Magdeburg Preprints

Max Planck Institute Magdeburg Preprints Thomas Mach Computing Inner Eigenvalues of Matrices in Tensor Train Matrix Format MAX PLANCK INSTITUT FÜR DYNAMIK KOMPLEXER TECHNISCHER SYSTEME MAGDEBURG Max Planck Institute Magdeburg Preprints MPIMD/11-09

More information

Adaptive rational Krylov subspaces for large-scale dynamical systems. V. Simoncini

Adaptive rational Krylov subspaces for large-scale dynamical systems. V. Simoncini Adaptive rational Krylov subspaces for large-scale dynamical systems V. Simoncini Dipartimento di Matematica, Università di Bologna valeria@dm.unibo.it joint work with Vladimir Druskin, Schlumberger Doll

More information

On the numerical solution of large-scale sparse discrete-time Riccati equations CSC/ Chemnitz Scientific Computing Preprints

On the numerical solution of large-scale sparse discrete-time Riccati equations CSC/ Chemnitz Scientific Computing Preprints Peter Benner Heike Faßbender On the numerical solution of large-scale sparse discrete-time Riccati equations CSC/09-11 Chemnitz Scientific Computing Preprints Impressum: Chemnitz Scientific Computing Preprints

More information

BDF Methods for Large-Scale Differential Riccati Equations

BDF Methods for Large-Scale Differential Riccati Equations BDF Methods for Large-Scale Differential Riccati Equations Peter Benner Hermann Mena Abstract We consider the numerical solution of differential Riccati equations. We review the existing methods and investigate

More information

University of Maryland Department of Computer Science TR-5009 University of Maryland Institute for Advanced Computer Studies TR April 2012

University of Maryland Department of Computer Science TR-5009 University of Maryland Institute for Advanced Computer Studies TR April 2012 University of Maryland Department of Computer Science TR-5009 University of Maryland Institute for Advanced Computer Studies TR-202-07 April 202 LYAPUNOV INVERSE ITERATION FOR COMPUTING A FEW RIGHTMOST

More information

Daniel B. Szyld and Jieyong Zhou. Report June 2016

Daniel B. Szyld and Jieyong Zhou. Report June 2016 Numerical solution of singular semi-stable Lyapunov equations Daniel B. Szyld and Jieyong Zhou Report 16-06-02 June 2016 Department of Mathematics Temple University Philadelphia, PA 19122 This report is

More information

SPARSE SOLVERS POISSON EQUATION. Margreet Nool. November 9, 2015 FOR THE. CWI, Multiscale Dynamics

SPARSE 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 information

Balancing-Related Model Reduction for Parabolic Control Systems

Balancing-Related Model Reduction for Parabolic Control Systems Balancing-Related Model Reduction for Parabolic Control Systems Peter Benner Max Planck Institute for Dynamics of Complex Technical Systems, Magdeburg, Germany (e-mail: benner@mpi-magdeburg.mpg.de). Abstract:

More information

Algebraic Multigrid as Solvers and as Preconditioner

Algebraic 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 information

DELFT UNIVERSITY OF TECHNOLOGY

DELFT UNIVERSITY OF TECHNOLOGY DELFT UNIVERSITY OF TECHNOLOGY REPORT 15-05 Induced Dimension Reduction method for solving linear matrix equations R. Astudillo and M. B. van Gijzen ISSN 1389-6520 Reports of the Delft Institute of Applied

More information

Computation of a Compact State Space Model for an Adaptive Spindle Head Configuration with Piezo Actuators using Balanced Truncation

Computation of a Compact State Space Model for an Adaptive Spindle Head Configuration with Piezo Actuators using Balanced Truncation Prod. Eng. Res. Devel. manuscript No. (will be inserted by the editor) Computation of a Compact State Space Model for an Adaptive Spindle Head Configuration with Piezo Actuators using Balanced Truncation

More information

A Tuned Preconditioner for Inexact Inverse Iteration Applied to Hermitian Eigenvalue Problems

A Tuned Preconditioner for Inexact Inverse Iteration Applied to Hermitian Eigenvalue Problems A Tuned Preconditioner for Applied to Eigenvalue Problems Department of Mathematical Sciences University of Bath, United Kingdom IWASEP VI May 22-25, 2006 Pennsylvania State University, University Park

More information

APPLIED NUMERICAL LINEAR ALGEBRA

APPLIED NUMERICAL LINEAR ALGEBRA APPLIED NUMERICAL LINEAR ALGEBRA James W. Demmel University of California Berkeley, California Society for Industrial and Applied Mathematics Philadelphia Contents Preface 1 Introduction 1 1.1 Basic Notation

More information

Krylov Subspace Methods for Nonlinear Model Reduction

Krylov Subspace Methods for Nonlinear Model Reduction MAX PLANCK INSTITUT Conference in honour of Nancy Nichols 70th birthday Reading, 2 3 July 2012 Krylov Subspace Methods for Nonlinear Model Reduction Peter Benner and Tobias Breiten Max Planck Institute

More information

Key words. preconditioned conjugate gradient method, saddle point problems, optimal control of PDEs, control and state constraints, multigrid method

Key words. preconditioned conjugate gradient method, saddle point problems, optimal control of PDEs, control and state constraints, multigrid method PRECONDITIONED CONJUGATE GRADIENT METHOD FOR OPTIMAL CONTROL PROBLEMS WITH CONTROL AND STATE CONSTRAINTS ROLAND HERZOG AND EKKEHARD SACHS Abstract. Optimality systems and their linearizations arising in

More information

Reduced Order Methods for Large Scale Riccati Equations

Reduced Order Methods for Large Scale Riccati Equations Reduced Order Methods for Large Scale Riccati Equations Miroslav Karolinov Stoyanov Dissertation submitted to the Faculty of the Virginia Polytechnic Institute and State University in partial fulfillment

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

University of Maryland Department of Computer Science TR-4975 University of Maryland Institute for Advanced Computer Studies TR March 2011

University of Maryland Department of Computer Science TR-4975 University of Maryland Institute for Advanced Computer Studies TR March 2011 University of Maryland Department of Computer Science TR-4975 University of Maryland Institute for Advanced Computer Studies TR-2011-04 March 2011 LYAPUNOV INVERSE ITERATION FOR IDENTIFYING HOPF BIFURCATIONS

More information

A reduced basis method and ROM-based optimization for batch chromatography

A reduced basis method and ROM-based optimization for batch chromatography MAX PLANCK INSTITUTE MedRed, 2013 Magdeburg, Dec 11-13 A reduced basis method and ROM-based optimization for batch chromatography Yongjin Zhang joint work with Peter Benner, Lihong Feng and Suzhou Li Max

More information

Boundary Value Problems - Solving 3-D Finite-Difference problems Jacob White

Boundary 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 information

Time-Invariant Linear Quadratic Regulators Robert Stengel Optimal Control and Estimation MAE 546 Princeton University, 2015

Time-Invariant Linear Quadratic Regulators Robert Stengel Optimal Control and Estimation MAE 546 Princeton University, 2015 Time-Invariant Linear Quadratic Regulators Robert Stengel Optimal Control and Estimation MAE 546 Princeton University, 15 Asymptotic approach from time-varying to constant gains Elimination of cross weighting

More information

Efficiency improving implementation techniques for large scale matrix equation solvers CSC/ Chemnitz Scientific Computing Preprints

Efficiency improving implementation techniques for large scale matrix equation solvers CSC/ Chemnitz Scientific Computing Preprints Martin Köhler Jens Saak Efficiency improving implementation techniques for large scale matrix equation solvers CSC/09-10 Chemnitz Scientific Computing Preprints Impressum: Chemnitz Scientific Computing

More information

First published on: 23 February 2011 PLEASE SCROLL DOWN FOR ARTICLE

First published on: 23 February 2011 PLEASE SCROLL DOWN FOR ARTICLE This article was downloaded by: [Benner, Peter][MPI Fuer Dynamik Komplexer] On: 25 February 20 Access details: Access Details: [subscription number 93087330] Publisher Taylor & Francis Informa Ltd Registered

More information

Fast Frequency Response Analysis using Model Order Reduction

Fast Frequency Response Analysis using Model Order Reduction Fast Frequency Response Analysis using Model Order Reduction Peter Benner EU MORNET Exploratory Workshop Applications of Model Order Reduction Methods in Industrial Research and Development Luxembourg,

More information

A 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 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 information

Time-Invariant Linear Quadratic Regulators!

Time-Invariant Linear Quadratic Regulators! Time-Invariant Linear Quadratic Regulators Robert Stengel Optimal Control and Estimation MAE 546 Princeton University, 17 Asymptotic approach from time-varying to constant gains Elimination of cross weighting

More information

Preliminary/Qualifying Exam in Numerical Analysis (Math 502a) Spring 2012

Preliminary/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 information

Downloaded 05/27/14 to Redistribution subject to SIAM license or copyright; see

Downloaded 05/27/14 to Redistribution subject to SIAM license or copyright; see SIAM J. MATRIX ANAL. APPL. Vol. 31, No. 2, pp. 272 288 c 2009 Society for Industrial and Applied Mathematics INEXACT KLEINMAN NEWTON METHOD FOR RICCATI EQUATIONS F. FEITZINGER, T. HYLLA, AND E. W. SACHS

More information

PROJECTED GMRES AND ITS VARIANTS

PROJECTED GMRES AND ITS VARIANTS PROJECTED GMRES AND ITS VARIANTS Reinaldo Astudillo Brígida Molina rastudillo@kuaimare.ciens.ucv.ve bmolina@kuaimare.ciens.ucv.ve Centro de Cálculo Científico y Tecnológico (CCCT), Facultad de Ciencias,

More information

TECHNISCHE UNIVERSITÄT BERLIN

TECHNISCHE UNIVERSITÄT BERLIN TECHNISCHE UNIVERSITÄT BERLIN On best rank one approximation of tensors S. Friedland V. Mehrmann R. Pajarola S.K. Suter Preprint 2012/07 Preprint-Reihe des Instituts für Mathematik Technische Universität

More information

From Completing the Squares and Orthogonal Projection to Finite Element Methods

From Completing the Squares and Orthogonal Projection to Finite Element Methods From Completing the Squares and Orthogonal Projection to Finite Element Methods Mo MU Background In scientific computing, it is important to start with an appropriate model in order to design effective

More information