Reduced-Order Methods in PDE Constrained Optimization

Size: px
Start display at page:

Download "Reduced-Order Methods in PDE Constrained Optimization"

Transcription

1 Reduced-Order Methods in PDE Constrained Optimization M. Hinze (Hamburg), K. Kunisch (Graz), F. Tro ltzsch (Berlin) and E. Grimm, M. Gubisch, S. Rogg, A. Studinger, S. Trenz, S. Volkwein University of Konstanz, Department of Mathematics and Statistics Recent Trends and Future Developments in Computational Science & Engineering Koppelsberg, Plo n, March -, 5 Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

2 Introduction Motivation Motivation for our Research Areas Problem: time-variant, nonlinear, parametrized PDE systems Efficient and reliable numerical simulation in multi-query cases finite element or finite volume discretizations too complex Multi-query examples fast simulation for different parameters on small computers parameter estimation, optimal design and feedback control usage of a reduced-order SURROGATE MODEL Time-variant, nonlinear coupled PDEs methods from linear system theory not directly applicable Nonlinear model-order reduction proper orthogonal decomposition and reduced-basis method Error control for reduced-order model new a-priori and a-posteriori error analysis [Martin Gubisch, Stefan Trenz, Andrea Wesche] reliable PDE constrained multi-objective optimization [Laura Iapichino, Stefan Trenz] economic model predicitive control [Lars Grüne, Thomas Meurer, Sabrina Rogg] certified closed-loop strategies [Alessandro Alla, Maurizio Falcone] PDE Partial Differential Equation Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

3 Introduction Introductory Example Minimization with Inequality Constraints 5 Minimiere f(x)=x Problem : min {f (x) = x : < x < } 5 f(x)=x Optimality condition: f (x )! = with f (x) = x Solution: x = Problem: min {f (x) = x : x } Lagrange functional: L(x, µ a, µ b ) = f (x) + µ a ( x) + µ b (x ) Optimality condition: f (x ) µ a +µ b µ a ( x ) µ b (x ) =! f (x )(x x ) x [,] }{{} (x ) 5 * x = * f (x )= x Achse Solution: x =, µ a =, µ b = x Achse Minimiere f(x)=x mit x aus [,] f(x)=x x * = f (x * )=x * Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

4 Introduction Outline of the Talk Outline of the talk Semilinear optimal control problems Proper orthogonal decomposition (POD) A-posteriori error for linear-quadratic optimal control Basis update by optimality-system POD (OS-POD) Extension to semilinear optimal control problems Conclusions and outlook Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

5 Semilinear Optimal Control Problems Semilinear Optimal Control Problems Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 5 / 6

6 Semilinear Optimal Control Problems Problem Formulation Optimal Control of Semilinear Parabolic Problems [Tröltzsch ] PDE constrained optimization problem: Minimize the quadratic cost J(y,u) = α tf Q y(t,x) yq (t,x) dxdt + α Ω y(tf,x) y Ω (x) dx t Ω Ω + κd tf u d (t,x) u d (t,x) dxdt + κb tf u b (t,s) u b (t,s) dsdt t Ω t Ω with respect to (y, u) subject to the semilinear evolution equation c p y t (t,x) y(t,x) + N (y(t,x)) = f (t,x) + u d (t,x), y n (t,s) + qy(t,s) = ub (t,s), and the bilateral control constraints (t,x) Q = (t,t f ) Ω (t,s) Σ = (t,t f ) Ω y(t,x) = y (x), x Ω R n, n {,,} U ad = { u = (u d,u b ) ua d u d ub d in Q and ub a u b ub d on Σ} Assumption: for any control u U ad there exists a unique state y = y(u) satisfying (SE) Reduced problem: Setting Ĵ(u) = J(y(u),u) we consider the reduced problem (SE) minĵ(u) subject to u U ad (ˆP) nonconvex problem possibly many local optimal solutions Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 6 / 6

7 Semilinear Optimal Control Problems Characterization of Solutions by First-Order Optimality Conditions First-Order Necessary Optimality Condition [Hinze/Pinnau/Ulbrich/Ulbrich 9, Ito/Kunisch 8, Tröltzsch ] Reduced problem: Setting Ĵ(u) = J(y(u),u) we consider the reduced problem minĵ(u) subject to u U ad (ˆP) Existence of local optimal controls ū: properties of N, regularity hypotheses Optimality conditions: based on constraint qualifications () State equation [ ȳ = y(ū) ] c p ȳ t ȳ + N (ȳ) = f + ū d in Q = (t,t f ) Ω ȳ + qȳ = ūb n on Σ = (t,t f ) Ω ȳ(t ) = y in Ω () Adjoint equation [ p = p(ū) ] c p p t p + N (ȳ) p = α Q ( yq ȳ ), p n + q p = on Σ ( p(t f ) = α Ω yω ȳ(t f ) ) in Ω () Variational inequality [ ū = (ū d,ū b ] ) U ad tf ( κ d( ū d u d ) ) (u p ū d ) tf ( dxdt + κ b( ū b u b ) ) (u p b ū b) dsdt t Ω t Ω for all (admissible) u = (u d,u b ) U ad in Q Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 7 / 6

8 Semilinear Optimal Control Problems Projected Newton-CG Method with Armijo Linesearch Globalization Computation of the Reduced Hessian [Hinze/Pinnau/Ulbrich/Ulbrich 9] Reduced problem: Setting Ĵ(u) = J(y(u),u) we consider the reduced problem minĵ(u) subject to u U ad (ˆP) ( [κ Reduced gradient: Ĵ (u) = d ( u d u d ) ] p Q [ κ b ( u b u b ) ] ) p Σ Evaluation of the reduced hessian Ĵ at u U ad in a direction u δ = ( uδ d ),ub δ : () solve the linearized state equation [ y δ = y δ (u δ ;u) with y = y(u) ] c p (y δ ) t y δ + N (y)y δ = u d δ in Q y δ n + qy δ = uδ b on Σ y δ (t ) = in Ω () solve the linearized adjoint equation [ p δ = p δ (u δ ;u) with y = y(u), p = p(u) ] c p (p δ ) t p δ + N (y)p δ = α Q y δ N (y)y δ p, in Q p δ n + qp δ = p δ (t f ) = α Ω y δ (t f ) ( [κ () Set Ĵ (u)u δ = d uδ d p ] Q [ δ κ b uδ b p ] ) Σ δ on Σ in Ω Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 8 / 6

9 Semilinear Optimal Control Problems Locally Second-Order Optimization Method Numerical Solution Algorithm Reduced problem: Setting Ĵ(u) = J(y(u),u) we consider the reduced problem minĵ(u) subject to u U ad (ˆP) Newton s method: solve at u k with y k = y(u k ) and p k = p(u k ) minĵ(uk ) + Ĵ (u k )u δ + Ĵ (u k ) ( u δ,u δ ) subject to u δ with u k + u δ U ad with a (truncated) conjugate gradient method [Nocedal/Wright 6] ( [κ Reduced gradient: Ĵ (u k ) = d ( u k,d u d ) p k ] [ Q κ b ( u k,b u b ) p k ] ) Σ Evaluation of the reduced hessian at u U ad in direction u δ = ( uδ d ),ub δ : () solve the linearized state equation for y δ = y δ (u δ ;u k ) () solve the linearized ( adjoint equation for p δ = p δ (u δ ;u k ) [κ () Set Ĵ (u k )u δ = d uδ d p ] [ Q δ κ b uδ b p ] ) Σ δ Globalization: negative curvature test and Armijo line search Control constraints: projected BFGS/Newton method [Kelley 99] Numerical realization: finite elements (FE) and implicit Euler method Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 9 / 6

10 Semilinear Optimal Control Problems Numerical Test Experiment Numerical Example: Distributed Optimal Control [Kammann/Tröltzsch/V., Trenz 5] Control input: u d (t,x) = i= u i(t)w i (x) and u b, start with u () (t,x) = i=.w i(x) Bilateral control bounds: u d a = and u d b = Nonlinearity: N (y) = y with N (y) = y Discretization: N = 79 FE unknown and N t = time instances Exact optimal state: ȳ(t,x) = cos(x )cos(x ) = y (x) for x = (x,x ) Ω = (,π) Exact optimal control: ū = ū and ū = ū in [t,t f ] = [,].5 Initial state y(x,x) Weighting functions w, w, w, w Analytical optimal control u i (t) ua,ub i = i = i = i =.5 x x wi(x,x) 6 x x t Newton-CG BFGS BFGS-Inv # Iterations 6 Time 59 s s 8 s J(ȳ FE,ū FE ).e-.e-.e- ū ū FE 6.69e-.5e- 5.7e- Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

11 Proper Orthogonal Decomposition (POD) Proper Orthogonal Decomposition (POD) Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

12 Proper Orthogonal Decomposition (POD) Motivation of the Reduced-Order Modeling Galerkin-Based Reduced-Order Modeling Model problem (weak form of our PDE): y(t ) = y and ( c p y t (t),ϕ + y(t) ϕ + N (y(t))ϕ dx = f (t) + u d (t) ) ϕ dx + u b (t)ϕ ds Ω Ω Ω for all ϕ V in (t,t f ], where we write y(t) = y(t, ) etc. Typical choices: H (Ω) V H (Ω) and H = L (Ω) Finite element approximation: y N (t) V N = span{ϕ,...,ϕ N } V with y N (t ) = P N y and c p yt N (t),ϕ + y N (t) ϕ + N (y N ( (t))ϕ dx = f (t) + u d (t) ) ϕ dx + u b (t)ϕ ds Ω Ω Ω for all ϕ V N in (t,t f ] Alternatives: finite volume or finite difference schemes Reduced-order model: y l (t) V l = span{ψ N,...,ψN l } V N and l N with y l (t ) = P l y and c p yt l (t),ψ + y l (t) ψ + N (y l ( (t))ψ dx = f (t) + u d (t) ) ψ dx + u b (t)ψ ds Ω Ω Ω for all ψ V l in (t,t f ] Reduced-order subspace V l : Proper Orthogonal Decomposition or Reduced-Basis Nonlinear problems: (Discrete) Empirical Interpolation Method (D)EIM Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

13 Proper Orthogonal Decomposition (POD) A Continuous Variant of Proper Orthogonal Decomposition (POD) Proper Orthogonal Decomposition (POD) Dynamical system in separable Hilbert space X (e.g., X = H or V ): ẏ(t) = F(t,y(t); µ(t)) in (t,t f ], y(t ) = y X with given parameter or control µ(t), and data y, f Given multiple snapshots: solutions y k (t) X, e.g., for parameters {µ k } k= Snapshot subspace: V = span { y k (t) t [t,t f ] and k } X Continuous variant of POD: for every l solve tf y min k l (t) k= t y k (t),ψ i X ψ i dt s.t. {ψ i} l i= X, ψ i,ψ j X X = δ ij, i,j l i= Integral operator: define R : X X as Rψ = k= tf t ψ,y k (t) X y k (t)dt for ψ X Theorem [Hilbert-Schmidt, Riesz-Schauder; Perturbation theory for the spectrum] ( ) ) R is linear, compact, selfadjoint and nonnegative. ) There are eigenfunctions { ψ i } i= and eigenvalues { λ i } i= with ) { ψ i } l i= solves ( ) and k= R ψ i = λ i ψ i, λ λ..., lim i λi = t f t y k (t) l i= y k (t), ψ i X ψ i dt = λi X i=l+ Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

14 Proper Orthogonal Decomposition (POD) Numerical Test Experiment Numerical Example: POD-Basis Computation [Trenz 5] Initial state y (x, x ) Eigenvalues Control input: u d (t, x) = i=.wi (x).8.6 Nonlinearity: N (y) = y..5. λi 6.5 Discretization: N = 79 FEs, Nt = POD topology: X = L (Ω).8 x Number i x Snapshots of the state y at t =.5,.5,.75, and : Snapshot: state y(x, t) at t =.5 Snapshot: state y(x, t) at t =.5.5 Snapshot: state y(x, t) at t = Snapshot: state y(x, t) at t = x x x x x x x x First four POD basis functions: POD basis function ψ POD basis function ψ. POD basis function ψ.. POD basis function ψ x Stefan Volkwein x. x x x x Reduced-Order Methods in PDE Constrained Optimization x x / 6

15 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control A-Posteriori Error for Linear-Quadratic Optimal Control Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 5 / 6

16 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Problem Formulation Linear-Quadratic, Time-Variant Optimal Control Problems Quadratic programming (QP) problem: min x=(y,u) J(x) = y(t f) y tf H + κ subject to the linear evolution problem tf t u(t) U dt y t (t),ϕ + a(t;y(t),ϕ) = (f + Bu)(t),ϕ ϕ V in (t,t f ] with y(t ) = y and to bilateral control constraints u U ad = { v U u a (t) v(t) u b (t) in [t,t f ] } State: y(t) V H with Hilbert spaces V, H Control (Hilbert) space: U = L (t,t f ;U) with U = R Nu, U = L (Ω) or U = L (Γ) Input/control: u U ad (boundary or distributed control) Bilinear form: a(t;, ) continuous and a(t;ϕ,ϕ) γ ϕ V γ ϕ H, e.g. a(t;ϕ,ψ) = ϕ ψ + N (y(t))ϕψ dx for ϕ,ψ V Ω with N (y) = y or N (y) = sinhy Control operator: B : U L (t,t f ;V ) linear, bounded Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 6 / 6

17 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Characterization of Solutions by First-Order Optimality Conditions First-Order Necessary and Sufficient Optimality Conditions Quadratic programming (QP) problem: min x=(y,u) J(x) = y(t f ) y tf H + κ tf t u(t) U dt s.t. y t (t),ϕ + a(t;y(t),ϕ) = (f + Bu)(t),ϕ ϕ V in (t,t f ] y(t ) = y and u a (t) u(t) u b (t), t [t,t f ] Optimal control ū U ad = {u u a u u b in [t,t f ]}, associated state ȳ = y(ū) Adjoint/dual equation: p t (t),ϕ + a(t;ϕ, p(t)) = ϕ V in [t,t f ), p(t f ) = ȳ(t f ) y tf Variational inequality: tf t κū(t) (B p)(t),u(t) ū(t) dt u U ad (VI) Reduced cost: Ĵ(u) = J(y(u),u) with Ĵ (ū) = κū B p U, i.e., (VI) reads Ĵ (ū),u(t) ū(t) u U ad Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 7 / 6

18 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control POD Galerkin Discretization POD Galerkin Scheme for the State and Dual Variable State equation: y t (t),ϕ + a(t;y(t),ϕ) = (f + Bu)(t),ϕ ϕ V in (t,t f ], y(t ) = y POD space: V l = span{ψ,...,ψ l } V Orthogonal projection: P l ϕ = l i= ϕ,ψ i X ψ i V l for ϕ V POD state: y l = y l (u) with y l (t) V l in [t,t f ] solves yt l (t),ψ + a(t;yl (t),ψ) = (f + Bu)(t),ψ ψ V l in (t,t f ], y l (t ) = P l y Adjoint/dual equation: p = p(y(u)) solves p t (t),ϕ + a(t;ϕ,p(t)) = ϕ V in [t,t f ), p(t f ) = y(t f ) y tf POD dual: p l = p l (y l (u)) with p l (t) V l in [t,t f ] solves pt l (t),ψ + a(t;ψ,pl (t)) = ψ V l in [t,t f ), p l (t f ) = y l (t f ) y tf same POD basis for state and adjoint variable Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 8 / 6

19 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Perturbation Argument POD A-Posteriori Error Analysis [Malanowski/Büskens/Maurer 97, Tröltzsch/V. 9] Variational inequality: tf t κū(t) (B p)(t),u(t) ū(t) dt u U ad (VI) Optimal POD solution: ū l U ad, associated state ȳ l = y l (ū l ) and dual p l = p l (y l (ū l )) POD variational inequality: tf Misfit in the variational inequality: tf with ỹ l = y(ū l ) and p l = p(y(ū l )) t κū l (t) (B p l )(t),u(t) ū l (t) dt u U ad (VI l ) t κū l (t) (B p l )(t),u(t) ū l (t) dt u U ad Perturbation analysis: there exists a perturbation ζ U satisfying tf t κū l (t) (B p l )(t) + ζ (t),u(t) ū l (t) dt u U ad (ṼIl ) A-posteriori analysis: choose u = ū l in (VI), u = ū in (ṼIl ) and add A-posteriori error estimate for the control: ū ū l ζ /κ Reduced-basis method: related work [Dede, Grepl/Kärcher, Negri/Rozza/Manzoni/ Quarteroni ] Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 9 / 6

20 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Algorithmic Realization Algorithm with POD A-Posteriori Analysis A-posteriori error estimate: ū ū l ζ /κ and p l = p(y(ū l )) ( κū l (t) (B p l )(t) ) if u a (t) < ū l (t) < u b (t) Computation of ζ : ζ (t) = min (,κū l (t) (B p l )(t) ) if ū l (t) = u a (t) max (,κū l (t) (B p l )(t) ) if ū l (t) = u b (t) Algorithmus (Optimal control with a-posteriori error estimation) : Choose POD basis {ψ i } l i= for the Galerkin approximation of the QP problem; : Determine the reduced-order model for the QP problem; : Calculate suboptimal control ū l U ad, e.g., by a semismooth Newton method; : Compute perturbation ζ l = ζ (ū l ); 5: if ζ l /κ > TOL then 6: Enlarge l and go back to Step ; 7: else 8: Stop; 9: end if Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

21 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Numerical Test Experiment Numerical Example: Problem Formulation and Optimal Control [Studinger/V. ] Consider: min J(y, u) = Z Ω y(tf ) yd dx + tf Z Z u dsdt Ω y y + = u on Σ, n.5 u on Σ = (, tf ) Ω y() = y in Ω = (, ) s.t. yt. y = in Q, Method & discretization: semismooth Newton & implizit Euler, finite elements x = Course of the FE optimal control u*fe Course of the control uref used for snapshot generation and thus POD basis determination x = y=.5 y= x axis.5 time axis x axis.5 time axis.5.5 Stefan Volkwein.5 x axis.5 time axis time axis x axis x=.5.8 time axis x= x = x = x axis x axis.5 time axis y axis time axis Reduced-Order Methods in PDE Constrained Optimization.5.5 y axis.5 time axis / 6

22 A-Posteriori Error Estimation for Linear-Quadratic Optimal Control Numerical Test Experiment Numerical Example: POD Error Analysis [Studinger/V. ] Course of the control uref used for snapshot generation and thus POD basis determination y= y= x axis time axis x axis x= y axis.5.5 x axis.5 time axis time axis.5 x = x = time axis.9.8 y axis.5 x axis time axis time axis x=.5 λi / trace x = W=M W=M+A. Course of the FE optimal control u*fe x = decay of normalized eigen values on semilog scale, Variant eigs index i kζ (u ` )k [Tro ltzsch/v. 9] κ FE A-posteriori error within FE: ku FE u ` k kζ FE (u ` )k =: εape κ x axis.5 time axis x axis..5 time axis A-posteriori error: ku u ` k [Gubisch/Neitzel/V. 5] ` FE εape u FE u ` FE εape ku FE u ` k FE εape u FE u ` FE εape ku FE u ` k e- 5.9e-.e-.e-.e- 9.e-.e-.e-.e- 9.7e e- 7.5e- 8.e-5.e-5.7e-6 5.6e- 7.e- 8.e-5.e-5.7e Difficulty: choice of expected control for computation of the POD basis Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

23 Basis Update by Optimality-System POD Basis Update by Optimality-System POD Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

24 Basis Update by Optimality-System POD Idea of the POD Basis Update Optimality-System POD (OS-POD) [Kunisch/V. 8] Optimal control problem: minj(y,u) s.t. (y,u) Y ad U ad, ẏ(t) = F(t,y(t),u(t)) in (t,t f ], y(t ) = y (P) POD-Galerkin approximation: {ψ i } l i= POD basis of rank l minj l (y l,u) s.t. (y l,u) Y l ad U ad, ẏ l (t) = F l (t,y l (t),u(t)) in (t,t f ], y l (t ) = y l (P l ) POD basis: eigenvalue problem with λ λ... Rψ i = tf y = y(u), i.e., ψ i = ψ i (u) and λ i = λ i (u) t y(t),ψ i y(t)dt = λ i ψ i, i =,...,l ( ) A-priori analysis for linear-quadratic, time-variant problems [Hinze/V. 8]: ) ū ū l = O( λ i (ū) for ψ i = ψ i (ū) i=l+ but ū ū l l with no rate otherwise [Tröltzsch/V. 9] OS-POD: augment (P l ) by the additional constraints ( ) improve the quality of the initial basis by applying a few gradient steps proceed with a fixed basis and utilize a-posteriori error [Gubisch/V., Tröltzsch/V. 9] Optimization method: variants of semismooth Newton methods Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

25 Basis Update by Optimality-System POD Numerical Test Experiment OS-POD for Linear-Quadratic, Control Constrained Control [Grimm, Grimm/Gubisch/V. 5] Optimal control problem: minj(y,u) = y(t f ) y Ω κ tf dx + u dsdt Ω Ω s.t. c p y t y = in Q, n y + qy = u on Σ, y() = y in Ω = (,) u a = u = u b on Σ = (,t f ) Ω initial state y desired final state yω optimal final state ȳ(tf) 6.5 y (x,x).5 yω(x,x) ȳ(tf,x,x).5 x.. x k = k = with ū FE req. l CPU 7. s 8. s.5 s εape FE.e-.8e-.9e- ū FE ū l 9.5e-.6e-.9e- ū FE ū l / ū FE 7.7e-.5e-.59e- x.. x x.. x.6.8 k = k = with ū FE different u a () different u b 8 6 (89) Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 5 / 6

26 Basis Update by Optimality-System POD x x x x x x Numerical Test Experiment x x x x x x x x x OS-POD for Linear-Quadratic, Control Constrained Control [Grimm, Grimm/Gubisch/V. 5] First four POD basis functions generated from u = : POD basis function ψ POD basis function ψ POD basis function ψ POD basis function ψ x x x First four POD basis functions generated from the optimal control ū FE : POD basis function ψ POD basis function ψ POD basis function ψ POD basis function ψ x x x First four POD basis functions generated after k = OS-POD gradient steps: POD basis function ψ POD basis function ψ POD basis function ψ POD basis function ψ x x x Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 6 / 6

27 Basis Update by Optimality-System POD State Constraints State and Control Constrained Optimal Control Problem Quadratic programming (QP) problem: min x=(y,u) J(x) = y(t f) y d L (Ω) + κ subject to the linear evolution problem tf t u(t) U dt y t (t),ϕ + a(t;y(t),ϕ) = (f + Bu)(t),ϕ ϕ V in (t,t f ] with y(t ) = y and to bilateral control constraints u U ad = { v U u a (t) v(t) u b (t) in [t,t f ] } Lavrentiev regularization: ε > y Y ad = { z L (Q) y a (t,x) z(t,x) y b (t,x) in Q } tf J(x,w) = y(t f) y d L (Ω) + κ u(t) U dt + σ w(t) t L t (Ω) dt (y,w) { z L (Q) y a (t,x) εw(t,x) + z(t,x) y b (t,x) in Q } Regular Lagrange multipliers [Tröltzsch 5]: formulation as a control constrained problem A-posteriori error [Grimm/Gubisch/V. 5]: extension from the control-constrained case OS-POD and semismooth Newton method also applicable tf Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 7 / 6

28 Basis Update by Optimality-System POD Numerical Test Experiment x x Numerical Example [Grimm/Gubisch/V. 5] optimal state optimal control Two-dimensional heat equation.5..5 Distributed control Semismooth Newton method y(t,x,x)... Bu(T,x,x).5.5 direction x direction x First four POD basis functions generated from the optimal control ū FE : st pod basis element nd pod basis element rd pod basis element th pod basis element ψ(x,x) ψ6(x,x) ψ(x,x) ψ(x,x) direction x direction x direction x direction x direction x direction x direction x direction x First four POD basis functions generated after k = OS-POD gradient steps: st pod basis element nd pod basis element rd pod basis element th pod basis element ψ(x,x) ψ6(x,x) ψ(x,x) ψ(x,x) direction x direction x direction x direction x direction x direction x direction x direction x Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 8 / 6

29 Extension to Semilinear Optimal Control Problems Extension to Semilinear Optimal Control Problems Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 9 / 6

30 Extension to Semilinear Optimal Control Problems POD A-Posteriori Error for Semilinear Optimal Control Problems A-Posteriori Analysis for Semilinear Optimal Control Problems [Kammann/Tröltzsch/V., DFG grant) Reduced problem: min u U ad Ĵ(u) with hessian Ĵ (u) First-order optimality conditions: Ĵ (ū)(u ū) for all u U ad Second-order sufficient optimality conditions: there is a constant η = η(ū) > with Ĵ (ū)(u,u) η u u Ĵ (ũ)(u,u) η u u, ũ U ad provided ũ ū sufficiently small Theorem [Kammann/Tröltzsch/V. ] ū optimal control, ū l suboptimal control. Then, second-order sufficient optimality implies ū ū l ε ape with ε ape = η ζ (ūl ) provided ū ū l is sufficiently small Computation of ζ (ū l ): based on state y = y(ū l ) and p = p(ū l ) Problem: estimate η = η(ū) smallest eigenvalue of Ĵ (ũ) Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

31 Extension to Semilinear Optimal Control Problems Numerical Test Experiment Numerical Example: Distributed Optimal Control [Kammann/Tröltzsch/V., Trenz 5] Control input: u d (t,x) = i= u i(t)w i (x) and u b Bilateral control bounds: u d a = and u d b = Nonlinearity: N (y) = y with N (y) = y Discretization: N = 79 FE unknowns, N t = time instances and l = 7 PODs Exact optimal state: ȳ(t,x) = cos(x )cos(x ) = y (x) for x = (x,x ) Ω = (,π) Exact optimal control: ū = ū and ū = ū in [t,t f ] = [,].8.6. Analytical optimal control u i (t) ua,ub i = i = i = i = Error u i (t) uh, i (t) (NEWTCG) i = 9 x u i (t) ul, i (t) (NEWTCG), l = i = t t t POD (l = 7) Newton-CG BFGS BFGS-Inv # Iterations 9 Time.7 s 6.5 s 7. s J(ȳ l,ū l ).5e-.5e- -5e- ū ū l.6e- 5.7e- 5.9e- εape.7e-.7e- (.6e-).7e- (.9e-) λ min 5.7e- 5.7e- (.7e-) 5.7e- (.e-) FE (N = 79) Newton-CG # Iterations 6 Time 59 s J(ȳ FE,ū FE ).e- ū ū FE 6.69e- Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

32 Extension to Semilinear Optimal Control Problems Trust-Region POD Methods POD Basis Update by Trust-Region Optimization [Arian/Fahl/Sachs, Schu, Rogg ] QP problem in Newton s method: minq k (u δ ) = Ĵ(uk ) + Ĵ (u k ),u δ + Ĵ (u k )(u δ,u δ ) s.t. u a u k + u δ u b Expensive parts: computation of gradient Ĵ (u k ) and hessian Ĵ (u k ) utilze POD approximations Ĵ l (uk ) and Ĵ l (uk ) in a trust region around u k Trust-region POD subproblem: minĵ(uk ) + Ĵ l (uk ),u δ + Ĵ l (uk )(u δ,u δ ) s.t. u a u k + u δ u b and u δ k Convergence criterium: Carter condition Ĵ (u k ) Ĵ l (uk ) γ Ĵ l (uk ) with γ (,) Numerical experiments [Rogg ] - Steihaug CG [Nocedal/Wright 6], global convergence - faster for two-dimensional, semilinear parabolic PDE Combination with a-posteriori error analysis: use a-posteriori error for the dual Ĵ (u k ) Ĵ l (uk ) = B (p(y(u k )) p l (y l (u k )) Cerr dual (u k ) efficient computation by evaluating primal and dual residuals at u k [Rogg/Trenz/V. 5] Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

33 Extension to Semilinear Optimal Control Problems Numerical Test Experiment Numerical Example: Boundary Optimal Control [Rogg ] Boundary control: u b (t,x) = i= u i(t)w i (x) and u d Nonlinearity: N (y) = y with N (y) = y Discretization: N = 77 FE unknowns, N t = time instances and l = 7 PODs ū h 8 6 ū h ū h ū h ū h l Ĵ(u k ) Ĵ (u k ) # CG it. k.7667.e e e e time t Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization / 6

34 Extension to Semilinear Optimal Control Problems Numerical Test Experiment Numerical Example [Rogg 5] First four POD basis functions generated from the u = : Basis x x x x Basis x x ψ (x) ψ (x) ψ (x) x x x x 6 x. Basis.. u FE : Basis... ψ (x) First four POD basis functions generated from the optimal control Basis x ψ (x). Basis Basis ψ (x). ψ (x) ψ (x) Basis x x x. x First four POD basis functions generated after one trust-region step: Basis Basis x ψ (x) Stefan Volkwein.. x. Basis ψ (x) ψ (x) Basis ψ (x).9 x x x x Reduced-Order Methods in PDE Constrained Optimization x. x / 6

35 Conclusions and Outlook Summary and Ongoing Research POD method for constrained optimal control problems A-priori and a-posteriori error analysis for the controls Change of the POD basis by OS-POD and TR-POD POD basis updates by optimality conditions (OS-POD) POD basis updates by trust region constraints (TR-POD) Combination of SQP and OS-POD [Metzdorf/V.] Reduced-order method in PDE constrained multiobjective optimization [Iapichino/Trenz/V.] Model predictive control in pharmaceutical application [Renal Research Institute New York & Rogg/V.] A-priori error analysis for closed-loop control (Hamilton-Jacobi-Bellman) [Alla/Falcone/V. 5] Reduced Basis Methods Model Reduction for Parametrized Systems Balanced Truncation Are YOU interested in... POD Methods Optimization Low Rank Tensor Approximation Reduced Basis Summer School Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 5 / 6

36 References Literature P. Benner, E.W. Sachs, S.V.: Model order reduction for PDE constrained optimization. ISNM, C. Gräßle: POD based inexact SQP methods for optimal control problems governed by a semilinear heat equation. Diploma thesis, University of Konstanz, E. Grimm, M. Gubisch, S.V.: Numerical analysis of optimality-system POD for constrained optimal control. LNCSE, 5 E. Grimm: Optimality system POD and a-posteriori error analysis for linear-quadratic optimal control problems. Master thesis, University of Konstanz, M. Gubisch, S.V.: Proper orthogonal decomposition for linear-quadratic optimal control. Submitted, M. Gubisch, S.V.: POD a-posteriori error analysis for optimal control problems with mixed control-state constraints. COAP, M. Hinze, S. V.: Error estimates for abstract linear-quadratic optimal control problems using POD. COAP, 8 E. Kammann, F. Tröltzsch, S.V.: A method of a-posteriori error estimation with application to POD. ESAIM: MAN, K. Kunisch, S.V.: POD for optimality systems. ESAIM: MAN, 8 S. Rogg: Trust region POD for optimal boundary control of a semilinear heat equation. Diploma thesis, University of Konstanz, A. Studinger, S.V.: Numerical analysis of POD a-posteriori error estimation for optimal control. ISNM, F. Tröltzsch, S.V.: POD a-posteriori error estimates for linear-quadratic optimal control problems. COAP, 9 S.V.: Optimality system POD and a-posteriori error analysis for linear-quadratic problems. Control and Cybernetics, Stefan Volkwein Reduced-Order Methods in PDE Constrained Optimization 6 / 6

POD for Parametric PDEs and for Optimality Systems

POD for Parametric PDEs and for Optimality Systems POD for Parametric PDEs and for Optimality Systems M. Kahlbacher, K. Kunisch, H. Müller and S. Volkwein Institute for Mathematics and Scientific Computing University of Graz, Austria DMV-Jahrestagung 26,

More information

Proper Orthogonal Decomposition for Optimal Control Problems with Mixed Control-State Constraints

Proper Orthogonal Decomposition for Optimal Control Problems with Mixed Control-State Constraints Proper Orthogonal Decomposition for Optimal Control Problems with Mixed Control-State Constraints Technische Universität Berlin Martin Gubisch, Stefan Volkwein University of Konstanz March, 3 Martin Gubisch,

More information

Reduced-Order Multiobjective Optimal Control of Semilinear Parabolic Problems. Laura Iapichino Stefan Trenz Stefan Volkwein

Reduced-Order Multiobjective Optimal Control of Semilinear Parabolic Problems. Laura Iapichino Stefan Trenz Stefan Volkwein Universität Konstanz Reduced-Order Multiobjective Optimal Control of Semilinear Parabolic Problems Laura Iapichino Stefan Trenz Stefan Volkwein Konstanzer Schriften in Mathematik Nr. 347, Dezember 2015

More information

An optimal control problem for a parabolic PDE with control constraints

An optimal control problem for a parabolic PDE with control constraints An optimal control problem for a parabolic PDE with control constraints PhD Summer School on Reduced Basis Methods, Ulm Martin Gubisch University of Konstanz October 7 Martin Gubisch (University of Konstanz)

More information

Proper Orthogonal Decomposition. POD for PDE Constrained Optimization. Stefan Volkwein

Proper Orthogonal Decomposition. POD for PDE Constrained Optimization. Stefan Volkwein Proper Orthogonal Decomposition for PDE Constrained Optimization Institute of Mathematics and Statistics, University of Constance Joined work with F. Diwoky, M. Hinze, D. Hömberg, M. Kahlbacher, E. Kammann,

More information

Proper Orthogonal Decomposition in PDE-Constrained Optimization

Proper Orthogonal Decomposition in PDE-Constrained Optimization Proper Orthogonal Decomposition in PDE-Constrained Optimization K. Kunisch Department of Mathematics and Computational Science University of Graz, Austria jointly with S. Volkwein Dynamic Programming Principle

More information

Suboptimal Open-loop Control Using POD. Stefan Volkwein

Suboptimal Open-loop Control Using POD. Stefan Volkwein Institute for Mathematics and Scientific Computing University of Graz, Austria PhD program in Mathematics for Technology Catania, May 22, 2007 Motivation Optimal control of evolution problems: min J(y,

More information

Hamburger Beiträge zur Angewandten Mathematik

Hamburger Beiträge zur Angewandten Mathematik Hamurger Beiträge zur Angewandten Mathematik POD asis updates for nonlinear PDE control Carmen Gräßle, Martin Guisch, Simone Metzdorf, Sarina Rogg, Stefan Volkwein Nr. 16-18 August 16 C. Gräßle, M. Guisch,

More information

Numerical approximation for optimal control problems via MPC and HJB. Giulia Fabrini

Numerical approximation for optimal control problems via MPC and HJB. Giulia Fabrini Numerical approximation for optimal control problems via MPC and HJB Giulia Fabrini Konstanz Women In Mathematics 15 May, 2018 G. Fabrini (University of Konstanz) Numerical approximation for OCP 1 / 33

More information

OPTIMALITY CONDITIONS AND POD A-POSTERIORI ERROR ESTIMATES FOR A SEMILINEAR PARABOLIC OPTIMAL CONTROL

OPTIMALITY CONDITIONS AND POD A-POSTERIORI ERROR ESTIMATES FOR A SEMILINEAR PARABOLIC OPTIMAL CONTROL OPTIMALITY CONDITIONS AND POD A-POSTERIORI ERROR ESTIMATES FOR A SEMILINEAR PARABOLIC OPTIMAL CONTROL O. LASS, S. TRENZ, AND S. VOLKWEIN Abstract. In the present paper the authors consider an optimal control

More information

Eileen Kammann Fredi Tröltzsch Stefan Volkwein

Eileen Kammann Fredi Tröltzsch Stefan Volkwein Universität Konstanz A method of a-posteriori error estimation with application to proper orthogonal decomposition Eileen Kammann Fredi Tröltzsch Stefan Volkwein Konstanzer Schriften in Mathematik Nr.

More information

Model order reduction & PDE constrained optimization

Model order reduction & PDE constrained optimization 1 Model order reduction & PDE constrained optimization Fachbereich Mathematik Optimierung und Approximation, Universität Hamburg Leuven, January 8, 01 Names Afanasiev, Attwell, Burns, Cordier,Fahl, Farrell,

More information

Proper Orthogonal Decomposition (POD) for Nonlinear Dynamical Systems. Stefan Volkwein

Proper Orthogonal Decomposition (POD) for Nonlinear Dynamical Systems. Stefan Volkwein Proper Orthogonal Decomposition (POD) for Nonlinear Dynamical Systems Institute for Mathematics and Scientific Computing, Austria DISC Summerschool 5 Outline of the talk POD and singular value decomposition

More information

Multiobjective PDE-constrained optimization using the reduced basis method

Multiobjective PDE-constrained optimization using the reduced basis method Multiobjective PDE-constrained optimization using the reduced basis method Laura Iapichino1, Stefan Ulbrich2, Stefan Volkwein1 1 Universita t 2 Konstanz - Fachbereich Mathematik und Statistik Technischen

More information

Martin Gubisch Stefan Volkwein

Martin Gubisch Stefan Volkwein Universität Konstanz POD a-posteriori error analysis for optimal control problems with mixed control-state constraints Martin Gubisch Stefan Volkwein Konstanzer Schriften in Mathematik Nr. 34, Juni 3 ISSN

More information

The HJB-POD approach for infinite dimensional control problems

The HJB-POD approach for infinite dimensional control problems The HJB-POD approach for infinite dimensional control problems M. Falcone works in collaboration with A. Alla, D. Kalise and S. Volkwein Università di Roma La Sapienza OCERTO Workshop Cortona, June 22,

More information

An Inexact Newton Method for Nonlinear Constrained Optimization

An Inexact Newton Method for Nonlinear Constrained Optimization An Inexact Newton Method for Nonlinear Constrained Optimization Frank E. Curtis Numerical Analysis Seminar, January 23, 2009 Outline Motivation and background Algorithm development and theoretical results

More information

Reduced-Order Greedy Controllability of Finite Dimensional Linear Systems. Giulia Fabrini Laura Iapichino Stefan Volkwein

Reduced-Order Greedy Controllability of Finite Dimensional Linear Systems. Giulia Fabrini Laura Iapichino Stefan Volkwein Universität Konstanz Reduced-Order Greedy Controllability of Finite Dimensional Linear Systems Giulia Fabrini Laura Iapichino Stefan Volkwein Konstanzer Schriften in Mathematik Nr. 364, Oktober 2017 ISSN

More information

K. Krumbiegel I. Neitzel A. Rösch

K. Krumbiegel I. Neitzel A. Rösch SUFFICIENT OPTIMALITY CONDITIONS FOR THE MOREAU-YOSIDA-TYPE REGULARIZATION CONCEPT APPLIED TO SEMILINEAR ELLIPTIC OPTIMAL CONTROL PROBLEMS WITH POINTWISE STATE CONSTRAINTS K. Krumbiegel I. Neitzel A. Rösch

More information

An Inexact Newton Method for Optimization

An Inexact Newton Method for Optimization New York University Brown Applied Mathematics Seminar, February 10, 2009 Brief biography New York State College of William and Mary (B.S.) Northwestern University (M.S. & Ph.D.) Courant Institute (Postdoc)

More information

Asymptotic Stability of POD based Model Predictive Control for a semilinear parabolic PDE

Asymptotic Stability of POD based Model Predictive Control for a semilinear parabolic PDE Asymptotic Stability of POD based Model Predictive Control for a semilinear parabolic PDE Alessandro Alla Dipartimento di Matematica, Sapienza Università di Roma, P.le Aldo Moro, 85 Rome, Italy Stefan

More information

Solving Distributed Optimal Control Problems for the Unsteady Burgers Equation in COMSOL Multiphysics

Solving Distributed Optimal Control Problems for the Unsteady Burgers Equation in COMSOL Multiphysics Excerpt from the Proceedings of the COMSOL Conference 2009 Milan Solving Distributed Optimal Control Problems for the Unsteady Burgers Equation in COMSOL Multiphysics Fikriye Yılmaz 1, Bülent Karasözen

More information

Proper Orthogonal Decomposition Surrogate Models for Nonlinear Dynamical Systems: Error Estimates and Suboptimal Control

Proper Orthogonal Decomposition Surrogate Models for Nonlinear Dynamical Systems: Error Estimates and Suboptimal Control Proper Orthogonal Decomposition Surrogate Models for Nonlinear Dynamical Systems: Error Estimates and Suboptimal Control Michael Hinze 1 and Stefan Volkwein 1 Institut für Numerische Mathematik, TU Dresden,

More information

Adaptive methods for control problems with finite-dimensional control space

Adaptive methods for control problems with finite-dimensional control space Adaptive methods for control problems with finite-dimensional control space Saheed Akindeinde and Daniel Wachsmuth Johann Radon Institute for Computational and Applied Mathematics (RICAM) Austrian Academy

More information

Comparison of the Reduced-Basis and POD a-posteriori Error Estimators for an Elliptic Linear-Quadratic Optimal Control

Comparison of the Reduced-Basis and POD a-posteriori Error Estimators for an Elliptic Linear-Quadratic Optimal Control SpezialForschungsBereich F 3 Karl Franzens Universität Graz Technische Universität Graz Medizinische Universität Graz Comparison of the Reduced-Basis and POD a-posteriori Error Estimators for an Elliptic

More information

Inexact Newton Methods and Nonlinear Constrained Optimization

Inexact Newton Methods and Nonlinear Constrained Optimization Inexact Newton Methods and Nonlinear Constrained Optimization Frank E. Curtis EPSRC Symposium Capstone Conference Warwick Mathematics Institute July 2, 2009 Outline PDE-Constrained Optimization Newton

More information

Lectures Notes Algorithms and Preconditioning in PDE-Constrained Optimization. Prof. Dr. R. Herzog

Lectures Notes Algorithms and Preconditioning in PDE-Constrained Optimization. Prof. Dr. R. Herzog Lectures Notes Algorithms and Preconditioning in PDE-Constrained Optimization Prof. Dr. R. Herzog held in July 2010 at the Summer School on Analysis and Numerics of PDE Constrained Optimization, Lambrecht

More information

Numerical Optimization of Partial Differential Equations

Numerical Optimization of Partial Differential Equations Numerical Optimization of Partial Differential Equations Part I: basic optimization concepts in R n Bartosz Protas Department of Mathematics & Statistics McMaster University, Hamilton, Ontario, Canada

More information

POD/DEIM 4DVAR Data Assimilation of the Shallow Water Equation Model

POD/DEIM 4DVAR Data Assimilation of the Shallow Water Equation Model nonlinear 4DVAR 4DVAR Data Assimilation of the Shallow Water Equation Model R. Ştefănescu and Ionel M. Department of Scientific Computing Florida State University Tallahassee, Florida May 23, 2013 (Florida

More information

Manuel Matthias Baumann

Manuel Matthias Baumann Delft University of Technology Faculty of Electrical Engineering, Mathematics and Computer Science Delft Institute of Applied Mathematics Nonlinear Model Order Reduction using POD/DEIM for Optimal Control

More information

POD-Based Economic Optimal Control of Heat-Convection Phenomena. Luca Mechelli Stefan Volkwein

POD-Based Economic Optimal Control of Heat-Convection Phenomena. Luca Mechelli Stefan Volkwein Universität Konstanz POD-Based Economic Optimal Control of Heat-Convection Phenomena Luca Mechelli Stefan Volkwein Konstanzer Schriften in Mathematik Nr. 365, Oktober 2017 ISSN 1430-3558 Konstanzer Online-Publikations-System

More information

ON THE OPTIMAL CONTROL OF THE SCHLÖGL-MODEL

ON THE OPTIMAL CONTROL OF THE SCHLÖGL-MODEL ON THE OPTIMAL CONTROL OF THE SCHLÖGL-MODEL RICO BUCHHOLZ, HARALD ENGEL, EILEEN KAMMANN, AND FREDI TRÖLTZSCH Abstract. Optimal control problems for a class of 1D semilinear parabolic equations with cubic

More information

minimize x subject to (x 2)(x 4) u,

minimize x subject to (x 2)(x 4) u, Math 6366/6367: Optimization and Variational Methods Sample Preliminary Exam Questions 1. Suppose that f : [, L] R is a C 2 -function with f () on (, L) and that you have explicit formulae for

More information

ON CONVERGENCE OF A RECEDING HORIZON METHOD FOR PARABOLIC BOUNDARY CONTROL. fredi tröltzsch and daniel wachsmuth 1

ON CONVERGENCE OF A RECEDING HORIZON METHOD FOR PARABOLIC BOUNDARY CONTROL. fredi tröltzsch and daniel wachsmuth 1 ON CONVERGENCE OF A RECEDING HORIZON METHOD FOR PARABOLIC BOUNDARY CONTROL fredi tröltzsch and daniel wachsmuth Abstract. A method of receding horizon type is considered for a simplified linear-quadratic

More information

Lecture Notes on PDEs

Lecture Notes on PDEs Lecture Notes on PDEs Alberto Bressan February 26, 2012 1 Elliptic equations Let IR n be a bounded open set Given measurable functions a ij, b i, c : IR, consider the linear, second order differential

More information

INTERPRETATION OF PROPER ORTHOGONAL DECOMPOSITION AS SINGULAR VALUE DECOMPOSITION AND HJB-BASED FEEDBACK DESIGN PAPER MSC-506

INTERPRETATION OF PROPER ORTHOGONAL DECOMPOSITION AS SINGULAR VALUE DECOMPOSITION AND HJB-BASED FEEDBACK DESIGN PAPER MSC-506 INTERPRETATION OF PROPER ORTHOGONAL DECOMPOSITION AS SINGULAR VALUE DECOMPOSITION AND HJB-BASED FEEDBACK DESIGN PAPER MSC-506 SIXTEENTH INTERNATIONAL SYMPOSIUM ON MATHEMATICAL THEORY OF NETWORKS AND SYSTEMS

More information

Priority Programme The Combination of POD Model Reduction with Adaptive Finite Element Methods in the Context of Phase Field Models

Priority Programme The Combination of POD Model Reduction with Adaptive Finite Element Methods in the Context of Phase Field Models Priority Programme 1962 The Combination of POD Model Reduction with Adaptive Finite Element Methods in the Context of Phase Field Models Carmen Gräßle, Michael Hinze Non-smooth and Complementarity-based

More information

A Sobolev trust-region method for numerical solution of the Ginz

A Sobolev trust-region method for numerical solution of the Ginz A Sobolev trust-region method for numerical solution of the Ginzburg-Landau equations Robert J. Renka Parimah Kazemi Department of Computer Science & Engineering University of North Texas June 6, 2012

More information

Part 3: Trust-region methods for unconstrained optimization. Nick Gould (RAL)

Part 3: Trust-region methods for unconstrained optimization. Nick Gould (RAL) Part 3: Trust-region methods for unconstrained optimization Nick Gould (RAL) minimize x IR n f(x) MSc course on nonlinear optimization UNCONSTRAINED MINIMIZATION minimize x IR n f(x) where the objective

More information

Numerical Optimal Control Part 3: Function space methods

Numerical Optimal Control Part 3: Function space methods Numerical Optimal Control Part 3: Function space methods SADCO Summer School and Workshop on Optimal and Model Predictive Control OMPC 2013, Bayreuth Institute of Mathematics and Applied Computing Department

More information

AM 205: lecture 19. Last time: Conditions for optimality, Newton s method for optimization Today: survey of optimization methods

AM 205: lecture 19. Last time: Conditions for optimality, Newton s method for optimization Today: survey of optimization methods AM 205: lecture 19 Last time: Conditions for optimality, Newton s method for optimization Today: survey of optimization methods Quasi-Newton Methods General form of quasi-newton methods: x k+1 = x k α

More information

A Recursive Trust-Region Method for Non-Convex Constrained Minimization

A Recursive Trust-Region Method for Non-Convex Constrained Minimization A Recursive Trust-Region Method for Non-Convex Constrained Minimization Christian Groß 1 and Rolf Krause 1 Institute for Numerical Simulation, University of Bonn. {gross,krause}@ins.uni-bonn.de 1 Introduction

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

A-posteriori error estimation of discrete POD models for PDE-constrained optimal control

A-posteriori error estimation of discrete POD models for PDE-constrained optimal control Wegelerstraße 6 53115 Bonn Germany phone +49 228 73-3427 fax +49 228 73-7527 www.ins.uni-bonn.de M. Gubisch, I. Neitzel, S. Volkwein A-posteriori error estimation of discrete POD models for PDE-constrained

More information

Constrained Minimization and Multigrid

Constrained Minimization and Multigrid Constrained Minimization and Multigrid C. Gräser (FU Berlin), R. Kornhuber (FU Berlin), and O. Sander (FU Berlin) Workshop on PDE Constrained Optimization Hamburg, March 27-29, 2008 Matheon Outline Successive

More information

POD A-POSTERIORI ERROR BASED INEXACT SQP METHOD FOR BILINEAR ELLIPTIC OPTIMAL CONTROL PROBLEMS

POD A-POSTERIORI ERROR BASED INEXACT SQP METHOD FOR BILINEAR ELLIPTIC OPTIMAL CONTROL PROBLEMS ESAIM: M2AN 46 22) 49 5 DOI:.5/m2an/26 ESAIM: Mathematical Modelling and Numerical Analysis www.esaim-m2an.org POD A-POSTERIORI ERROR BASED INEXACT SQP METHOD FOR BILINEAR ELLIPTIC OPTIMAL CONTROL PROBLEMS

More information

E5295/5B5749 Convex optimization with engineering applications. Lecture 8. Smooth convex unconstrained and equality-constrained minimization

E5295/5B5749 Convex optimization with engineering applications. Lecture 8. Smooth convex unconstrained and equality-constrained minimization E5295/5B5749 Convex optimization with engineering applications Lecture 8 Smooth convex unconstrained and equality-constrained minimization A. Forsgren, KTH 1 Lecture 8 Convex optimization 2006/2007 Unconstrained

More information

Proper Orthogonal Decomposition (POD)

Proper Orthogonal Decomposition (POD) Intro Results Problem etras Proper Orthogonal Decomposition () Advisor: Dr. Sorensen CAAM699 Department of Computational and Applied Mathematics Rice University September 5, 28 Outline Intro Results Problem

More information

IPAM Summer School Optimization methods for machine learning. Jorge Nocedal

IPAM Summer School Optimization methods for machine learning. Jorge Nocedal IPAM Summer School 2012 Tutorial on Optimization methods for machine learning Jorge Nocedal Northwestern University Overview 1. We discuss some characteristics of optimization problems arising in deep

More information

Stabilization of Heat Equation

Stabilization of Heat Equation Stabilization of Heat Equation Mythily Ramaswamy TIFR Centre for Applicable Mathematics, Bangalore, India CIMPA Pre-School, I.I.T Bombay 22 June - July 4, 215 Mythily Ramaswamy Stabilization of Heat Equation

More information

Towards parametric model order reduction for nonlinear PDE systems in networks

Towards parametric model order reduction for nonlinear PDE systems in networks Towards parametric model order reduction for nonlinear PDE systems in networks MoRePas II 2012 Michael Hinze Martin Kunkel Ulrich Matthes Morten Vierling Andreas Steinbrecher Tatjana Stykel Fachbereich

More information

Algorithms for constrained local optimization

Algorithms for constrained local optimization Algorithms for constrained local optimization Fabio Schoen 2008 http://gol.dsi.unifi.it/users/schoen Algorithms for constrained local optimization p. Feasible direction methods Algorithms for constrained

More information

SF2822 Applied Nonlinear Optimization. Preparatory question. Lecture 9: Sequential quadratic programming. Anders Forsgren

SF2822 Applied Nonlinear Optimization. Preparatory question. Lecture 9: Sequential quadratic programming. Anders Forsgren SF2822 Applied Nonlinear Optimization Lecture 9: Sequential quadratic programming Anders Forsgren SF2822 Applied Nonlinear Optimization, KTH / 24 Lecture 9, 207/208 Preparatory question. Try to solve theory

More information

Infeasibility Detection and an Inexact Active-Set Method for Large-Scale Nonlinear Optimization

Infeasibility Detection and an Inexact Active-Set Method for Large-Scale Nonlinear Optimization Infeasibility Detection and an Inexact Active-Set Method for Large-Scale Nonlinear Optimization Frank E. Curtis, Lehigh University involving joint work with James V. Burke, University of Washington Daniel

More information

Optimal control of partial differential equations with affine control constraints

Optimal control of partial differential equations with affine control constraints Control and Cybernetics vol. 38 (2009) No. 4A Optimal control of partial differential equations with affine control constraints by Juan Carlos De Los Reyes 1 and Karl Kunisch 2 1 Departmento de Matemática,

More information

arxiv: v1 [math.oc] 21 Jan 2019

arxiv: v1 [math.oc] 21 Jan 2019 An Efficient Parallel-in-Time Method for Optimization with Parabolic PDEs Sebastian Götschel Michael L. Minion arxiv:1901.06850v1 [math.oc] 21 Jan 2019 January 17, 2019 Abstract To solve optimization problems

More information

Nonlinear Optimization: What s important?

Nonlinear Optimization: What s important? Nonlinear Optimization: What s important? Julian Hall 10th May 2012 Convexity: convex problems A local minimizer is a global minimizer A solution of f (x) = 0 (stationary point) is a minimizer A global

More information

Parameterized Partial Differential Equations and the Proper Orthogonal D

Parameterized Partial Differential Equations and the Proper Orthogonal D Parameterized Partial Differential Equations and the Proper Orthogonal Decomposition Stanford University February 04, 2014 Outline Parameterized PDEs The steady case Dimensionality reduction Proper orthogonal

More information

Multigrid and stochastic sparse-grids techniques for PDE control problems with random coefficients

Multigrid and stochastic sparse-grids techniques for PDE control problems with random coefficients Multigrid and stochastic sparse-grids techniques for PDE control problems with random coefficients Università degli Studi del Sannio Dipartimento e Facoltà di Ingegneria, Benevento, Italia Random fields

More information

PDE-Constrained and Nonsmooth Optimization

PDE-Constrained and Nonsmooth Optimization Frank E. Curtis October 1, 2009 Outline PDE-Constrained Optimization Introduction Newton s method Inexactness Results Summary and future work Nonsmooth Optimization Sequential quadratic programming (SQP)

More information

Numerical Analysis of State-Constrained Optimal Control Problems for PDEs

Numerical Analysis of State-Constrained Optimal Control Problems for PDEs Numerical Analysis of State-Constrained Optimal Control Problems for PDEs Ira Neitzel and Fredi Tröltzsch Abstract. We survey the results of SPP 1253 project Numerical Analysis of State-Constrained Optimal

More information

PDE Constrained Optimization selected Proofs

PDE Constrained Optimization selected Proofs PDE Constrained Optimization selected Proofs Jeff Snider jeff@snider.com George Mason University Department of Mathematical Sciences April, 214 Outline 1 Prelim 2 Thms 3.9 3.11 3 Thm 3.12 4 Thm 3.13 5

More information

Part 5: Penalty and augmented Lagrangian methods for equality constrained optimization. Nick Gould (RAL)

Part 5: Penalty and augmented Lagrangian methods for equality constrained optimization. Nick Gould (RAL) Part 5: Penalty and augmented Lagrangian methods for equality constrained optimization Nick Gould (RAL) x IR n f(x) subject to c(x) = Part C course on continuoue optimization CONSTRAINED MINIMIZATION x

More information

Written Examination

Written Examination Division of Scientific Computing Department of Information Technology Uppsala University Optimization Written Examination 202-2-20 Time: 4:00-9:00 Allowed Tools: Pocket Calculator, one A4 paper with notes

More information

Algorithms for PDE-Constrained Optimization

Algorithms for PDE-Constrained Optimization GAMM-Mitteilungen, 31 January 2014 Algorithms for PDE-Constrained Optimization Roland Herzog 1 and Karl Kunisch 2 1 Chemnitz University of Technology, Faculty of Mathematics, Reichenhainer Straße 41, D

More information

Scientific Computing: An Introductory Survey

Scientific Computing: An Introductory Survey Scientific Computing: An Introductory Survey Chapter 6 Optimization Prof. Michael T. Heath Department of Computer Science University of Illinois at Urbana-Champaign Copyright c 2002. Reproduction permitted

More information

Scientific Computing: An Introductory Survey

Scientific Computing: An Introductory Survey Scientific Computing: An Introductory Survey Chapter 6 Optimization Prof. Michael T. Heath Department of Computer Science University of Illinois at Urbana-Champaign Copyright c 2002. Reproduction permitted

More information

On a Data Assimilation Method coupling Kalman Filtering, MCRE Concept and PGD Model Reduction for Real-Time Updating of Structural Mechanics Model

On a Data Assimilation Method coupling Kalman Filtering, MCRE Concept and PGD Model Reduction for Real-Time Updating of Structural Mechanics Model On a Data Assimilation Method coupling, MCRE Concept and PGD Model Reduction for Real-Time Updating of Structural Mechanics Model 2016 SIAM Conference on Uncertainty Quantification Basile Marchand 1, Ludovic

More information

1. INTRODUCTION 2. PROBLEM FORMULATION ROMAI J., 6, 2(2010), 1 13

1. INTRODUCTION 2. PROBLEM FORMULATION ROMAI J., 6, 2(2010), 1 13 Contents 1 A product formula approach to an inverse problem governed by nonlinear phase-field transition system. Case 1D Tommaso Benincasa, Costică Moroşanu 1 v ROMAI J., 6, 2(21), 1 13 A PRODUCT FORMULA

More information

Reduced basis method for efficient model order reduction of optimal control problems

Reduced basis method for efficient model order reduction of optimal control problems Reduced basis method for efficient model order reduction of optimal control problems Laura Iapichino 1 joint work with Giulia Fabrini 2 and Stefan Volkwein 2 1 Department of Mathematics and Computer Science,

More information

Lecture 11 and 12: Penalty methods and augmented Lagrangian methods for nonlinear programming

Lecture 11 and 12: Penalty methods and augmented Lagrangian methods for nonlinear programming Lecture 11 and 12: Penalty methods and augmented Lagrangian methods for nonlinear programming Coralia Cartis, Mathematical Institute, University of Oxford C6.2/B2: Continuous Optimization Lecture 11 and

More information

AM 205: lecture 19. Last time: Conditions for optimality Today: Newton s method for optimization, survey of optimization methods

AM 205: lecture 19. Last time: Conditions for optimality Today: Newton s method for optimization, survey of optimization methods AM 205: lecture 19 Last time: Conditions for optimality Today: Newton s method for optimization, survey of optimization methods Optimality Conditions: Equality Constrained Case As another example of equality

More information

CME 345: MODEL REDUCTION

CME 345: MODEL REDUCTION CME 345: MODEL REDUCTION Proper Orthogonal Decomposition (POD) Charbel Farhat & David Amsallem Stanford University cfarhat@stanford.edu 1 / 43 Outline 1 Time-continuous Formulation 2 Method of Snapshots

More information

Priority Programme 1962

Priority Programme 1962 Priority Programme 1962 Set-Oriented Multiobjective Optimal Control of PDEs using Proper Orthogonal Decomposition Dennis Beermann, Michael Dellnitz, Sebastian Peitz, Stefan Volkwein Non-smooth and Complementarity-based

More information

Constrained optimization: direct methods (cont.)

Constrained optimization: direct methods (cont.) Constrained optimization: direct methods (cont.) Jussi Hakanen Post-doctoral researcher jussi.hakanen@jyu.fi Direct methods Also known as methods of feasible directions Idea in a point x h, generate a

More information

Reduced Basis Method for Parametric

Reduced Basis Method for Parametric 1/24 Reduced Basis Method for Parametric H 2 -Optimal Control Problems NUMOC2017 Andreas Schmidt, Bernard Haasdonk Institute for Applied Analysis and Numerical Simulation, University of Stuttgart andreas.schmidt@mathematik.uni-stuttgart.de

More information

Arc Search Algorithms

Arc Search Algorithms Arc Search Algorithms Nick Henderson and Walter Murray Stanford University Institute for Computational and Mathematical Engineering November 10, 2011 Unconstrained Optimization minimize x D F (x) where

More information

An Inexact Sequential Quadratic Optimization Method for Nonlinear Optimization

An Inexact Sequential Quadratic Optimization Method for Nonlinear Optimization An Inexact Sequential Quadratic Optimization Method for Nonlinear Optimization Frank E. Curtis, Lehigh University involving joint work with Travis Johnson, Northwestern University Daniel P. Robinson, Johns

More information

NECESSARY OPTIMALITY CONDITIONS FOR AN OPTIMAL CONTROL PROBLEM OF 2D

NECESSARY OPTIMALITY CONDITIONS FOR AN OPTIMAL CONTROL PROBLEM OF 2D NECESSARY OPTIMALITY CONDITIONS FOR AN OPTIMAL CONTROL PROBLEM OF 2D g-navier-stokes EQUATIONS Nguyen Duc Loc 1 Abstract Considered here is the optimal control problem of 2D g-navier-stokes equations.

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

Unconstrained optimization

Unconstrained optimization Chapter 4 Unconstrained optimization An unconstrained optimization problem takes the form min x Rnf(x) (4.1) for a target functional (also called objective function) f : R n R. In this chapter and throughout

More information

ALADIN An Algorithm for Distributed Non-Convex Optimization and Control

ALADIN An Algorithm for Distributed Non-Convex Optimization and Control ALADIN An Algorithm for Distributed Non-Convex Optimization and Control Boris Houska, Yuning Jiang, Janick Frasch, Rien Quirynen, Dimitris Kouzoupis, Moritz Diehl ShanghaiTech University, University of

More information

Proper Orthogonal Decomposition: Theory and Reduced-Order Modelling

Proper Orthogonal Decomposition: Theory and Reduced-Order Modelling Proper Orthogonal Decomposition: Theory and Reduced-Order Modelling S. Volkwein Lecture Notes (August 7, 13) University of Konstanz Department of Mathematics and Statistics Contents Chapter 1. The POD

More information

An Introduction to Variational Inequalities

An Introduction to Variational Inequalities An Introduction to Variational Inequalities Stefan Rosenberger Supervisor: Prof. DI. Dr. techn. Karl Kunisch Institut für Mathematik und wissenschaftliches Rechnen Universität Graz January 26, 2012 Stefan

More information

DELFT UNIVERSITY OF TECHNOLOGY

DELFT UNIVERSITY OF TECHNOLOGY DELFT UNIVERSITY OF TECHNOLOGY REPORT -09 Computational and Sensitivity Aspects of Eigenvalue-Based Methods for the Large-Scale Trust-Region Subproblem Marielba Rojas, Bjørn H. Fotland, and Trond Steihaug

More information

RICE UNIVERSITY. Nonlinear Model Reduction via Discrete Empirical Interpolation by Saifon Chaturantabut

RICE UNIVERSITY. Nonlinear Model Reduction via Discrete Empirical Interpolation by Saifon Chaturantabut RICE UNIVERSITY Nonlinear Model Reduction via Discrete Empirical Interpolation by Saifon Chaturantabut A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy

More information

CONSTRAINED NONLINEAR PROGRAMMING

CONSTRAINED NONLINEAR PROGRAMMING 149 CONSTRAINED NONLINEAR PROGRAMMING We now turn to methods for general constrained nonlinear programming. These may be broadly classified into two categories: 1. TRANSFORMATION METHODS: In this approach

More information

Suppose that the approximate solutions of Eq. (1) satisfy the condition (3). Then (1) if η = 0 in the algorithm Trust Region, then lim inf.

Suppose that the approximate solutions of Eq. (1) satisfy the condition (3). Then (1) if η = 0 in the algorithm Trust Region, then lim inf. Maria Cameron 1. Trust Region Methods At every iteration the trust region methods generate a model m k (p), choose a trust region, and solve the constraint optimization problem of finding the minimum of

More information

A Posteriori Error Estimation for Reduced Order Solutions of Parametrized Parabolic Optimal Control Problems

A Posteriori Error Estimation for Reduced Order Solutions of Parametrized Parabolic Optimal Control Problems A Posteriori Error Estimation for Reduced Order Solutions of Parametrized Parabolic Optimal Control Problems Mark Kärcher 1 and Martin A. Grepl Bericht r. 361 April 013 Keywords: Optimal control, reduced

More information

Chapter Two: Numerical Methods for Elliptic PDEs. 1 Finite Difference Methods for Elliptic PDEs

Chapter Two: Numerical Methods for Elliptic PDEs. 1 Finite Difference Methods for Elliptic PDEs Chapter Two: Numerical Methods for Elliptic PDEs Finite Difference Methods for Elliptic PDEs.. Finite difference scheme. We consider a simple example u := subject to Dirichlet boundary conditions ( ) u

More information

Finite-dimensional spaces. C n is the space of n-tuples x = (x 1,..., x n ) of complex numbers. It is a Hilbert space with the inner product

Finite-dimensional spaces. C n is the space of n-tuples x = (x 1,..., x n ) of complex numbers. It is a Hilbert space with the inner product Chapter 4 Hilbert Spaces 4.1 Inner Product Spaces Inner Product Space. A complex vector space E is called an inner product space (or a pre-hilbert space, or a unitary space) if there is a mapping (, )

More information

Numerical Methods for PDE-Constrained Optimization

Numerical Methods for PDE-Constrained Optimization Numerical Methods for PDE-Constrained Optimization Richard H. Byrd 1 Frank E. Curtis 2 Jorge Nocedal 2 1 University of Colorado at Boulder 2 Northwestern University Courant Institute of Mathematical Sciences,

More information

Greedy control. Martin Lazar University of Dubrovnik. Opatija, th Najman Conference. Joint work with E: Zuazua, UAM, Madrid

Greedy control. Martin Lazar University of Dubrovnik. Opatija, th Najman Conference. Joint work with E: Zuazua, UAM, Madrid Greedy control Martin Lazar University of Dubrovnik Opatija, 2015 4th Najman Conference Joint work with E: Zuazua, UAM, Madrid Outline Parametric dependent systems Reduced basis methods Greedy control

More information

Adaptive discretization and first-order methods for nonsmooth inverse problems for PDEs

Adaptive discretization and first-order methods for nonsmooth inverse problems for PDEs Adaptive discretization and first-order methods for nonsmooth inverse problems for PDEs Christian Clason Faculty of Mathematics, Universität Duisburg-Essen joint work with Barbara Kaltenbacher, Tuomo Valkonen,

More information

Sufficiency for Essentially Bounded Controls Which Do Not Satisfy the Strengthened Legendre-Clebsch Condition

Sufficiency for Essentially Bounded Controls Which Do Not Satisfy the Strengthened Legendre-Clebsch Condition Applied Mathematical Sciences, Vol. 12, 218, no. 27, 1297-1315 HIKARI Ltd, www.m-hikari.com https://doi.org/1.12988/ams.218.81137 Sufficiency for Essentially Bounded Controls Which Do Not Satisfy the Strengthened

More information

Suboptimal feedback control of PDEs by solving Hamilton-Jacobi Bellman equations on sparse grids

Suboptimal feedback control of PDEs by solving Hamilton-Jacobi Bellman equations on sparse grids Suboptimal feedback control of PDEs by solving Hamilton-Jacobi Bellman equations on sparse grids Jochen Garcke joint work with Axel Kröner, INRIA Saclay and CMAP, Ecole Polytechnique Ilja Kalmykov, Universität

More information

5 Handling Constraints

5 Handling Constraints 5 Handling Constraints Engineering design optimization problems are very rarely unconstrained. Moreover, the constraints that appear in these problems are typically nonlinear. This motivates our interest

More information

Reduced-Order Model Development and Control Design

Reduced-Order Model Development and Control Design Reduced-Order Model Development and Control Design K. Kunisch Department of Mathematics University of Graz, Austria Rome, January 6 Laser-Oberflächenhärtung eines Stahl-Werkstückes [WIAS-Berlin]: workpiece

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

REGULAR LAGRANGE MULTIPLIERS FOR CONTROL PROBLEMS WITH MIXED POINTWISE CONTROL-STATE CONSTRAINTS

REGULAR LAGRANGE MULTIPLIERS FOR CONTROL PROBLEMS WITH MIXED POINTWISE CONTROL-STATE CONSTRAINTS REGULAR LAGRANGE MULTIPLIERS FOR CONTROL PROBLEMS WITH MIXED POINTWISE CONTROL-STATE CONSTRAINTS fredi tröltzsch 1 Abstract. A class of quadratic optimization problems in Hilbert spaces is considered,

More information