Hilbert Uniqueness Method and regularity

Similar documents
The Wave Equation: Control and Numerics

A remark on the observability of conservative linear systems

Control of Waves: Theory and Numerics

Nonlinear stabilization via a linear observability

Infinite dimensional controllability

Inégalités spectrales pour le contrôle des EDP linéaires : groupe de Schrödinger contre semigroupe de la chaleur.

Evolution problems involving the fractional Laplace operator: HUM control and Fourier analysis

Wave operators with non-lipschitz coefficients: energy and observability estimates

Controllability of linear PDEs (I): The wave equation

Optimal and Approximate Control of Finite-Difference Approximation Schemes for the 1D Wave Equation

Wave propagation in discrete heterogeneous media

Global Carleman inequalities and theoretical and numerical control results for systems governed by PDEs

On the observability of time-discrete conservative linear systems

The wave equation. Paris-Sud, Orsay, December 06

Sharp observability estimates for heat equations

Numerical control of waves

Minimal time issues for the observability of Grushin-type equations

Numerical control and inverse problems and Carleman estimates

Asymptotic Behavior of a Hyperbolic-parabolic Coupled System Arising in Fluid-structure Interaction

Stabilization of the wave equation with localized Kelvin-Voigt damping

Stability of nonlinear locally damped partial differential equations: the continuous and discretized problems. Part II

Spectrum and Exact Controllability of a Hybrid System of Elasticity.

A quantitative Fattorini-Hautus test: the minimal null control time problem in the parabolic setting

Optimal shape and location of sensors. actuators in PDE models

Conservative Control Systems Described by the Schrödinger Equation

A survey of controllability and stabilization results for partial differential equations

The effect of Group Velocity in the numerical analysis of control problems for the wave equation

Introduction to Exact Controllability and Observability; Variational Approach and Hilbert Uniqueness Method 1

Optimal shape and location of sensors or actuators in PDE models

Switching, sparse and averaged control

arxiv: v1 [math.oc] 6 Apr 2018

On the bang-bang property of time optimal controls for infinite dimensional linear systems

On the Resolvent Estimates of some Evolution Equations and Applications

arxiv: v1 [math.ap] 8 Jul 2013

A spectral approach for the exact observability of infinite-dimensional systems with skew-adjoint generator

Averaged control and observation of parameter-depending wave equations

Long Time Behavior of a Coupled Heat-Wave System Arising in Fluid-Structure Interaction

Control and Observation for Stochastic Partial Differential Equations

EXACT NEUMANN BOUNDARY CONTROLLABILITY FOR SECOND ORDER HYPERBOLIC EQUATIONS

Controllability of partial differential equations

Optimal shape and placement of actuators or sensors for PDE models

A spectral approach for the exact observability of infinite dimensional systems with skew-adjoint generator.

The heat equation. Paris-Sud, Orsay, December 06

Global Carleman estimates for waves and applications

Local null controllability of the N-dimensional Navier-Stokes system with N-1 scalar controls in an arbitrary control domain

Some new results related to the null controllability of the 1 d heat equation

Contrôlabilité de quelques systèmes gouvernés par des equations paraboliques.

Dispersive Equations and Hyperbolic Orbits

Control from an Interior Hypersurface

A regularity property for Schrödinger equations on bounded domains

ROBUST NULL CONTROLLABILITY FOR HEAT EQUATIONS WITH UNKNOWN SWITCHING CONTROL MODE

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

METHODS AND APPLICATIONS OF ANALYSIS. Vol. 17, No. 4, pp , December 2010

Observability and measurable sets

Stability of an abstract wave equation with delay and a Kelvin Voigt damping

Uniform polynomial stability of C 0 -Semigroups

Optimal location of controllers for the one-dimensional wave equation

UNIVERSITY OF MANITOBA

From averaged to simultaneous controllability of parameter dependent finite-dimensional systems

Optimal shape and position of the support for the internal exact control of a string

arxiv: v1 [math.ap] 11 Jun 2007

u n 2 4 u n 36 u n 1, n 1.

Class Meeting # 1: Introduction to PDEs

Pierre Lissy. 19 mars 2015

Neatest and Promptest Manner. E d i t u r ami rul)lihher. FOIt THE CIIILDIIES'. Trifles.

The Output Stabilization problem: a conjecture and counterexample

1 Introduction. Controllability and observability

Stabilization of second order evolution equations with unbounded feedback with delay

Some recent results on controllability of coupled parabolic systems: Towards a Kalman condition

The Fattorini-Hautus test

Controllability of parabolic PDES : old and new

Research Article On a Hyperbolic Coefficient Inverse Problem via Partial Dynamic Boundary Measurements

Global stabilization of a Korteweg-de Vries equation with saturating distributed control

Strichartz Estimates in Domains

Exponential stability of abstract evolution equations with time delay feedback

Half of Final Exam Name: Practice Problems October 28, 2014

z x = f x (x, y, a, b), z y = f y (x, y, a, b). F(x, y, z, z x, z y ) = 0. This is a PDE for the unknown function of two independent variables.

Stabilization of a Boussinesq system of KdV KdV type

Problem List MATH 5173 Spring, 2014

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

Averaged controllability of parameter dependent wave equations

Lecture 13 - Wednesday April 29th

Topics in Harmonic Analysis Lecture 6: Pseudodifferential calculus and almost orthogonality

Null-controllability of the heat equation in unbounded domains

Single input controllability of a simplified fluid-structure interaction model

arxiv: v1 [math.oc] 23 Aug 2017

Giuseppe Floridia Department of Matematics and Applications R. Caccioppoli, University of Naples Federico II

CUTOFF RESOLVENT ESTIMATES AND THE SEMILINEAR SCHRÖDINGER EQUATION

Periodic functions: simple harmonic oscillator

New estimates for the div-curl-grad operators and elliptic problems with L1-data in the half-space

Carleman estimates for the Euler Bernoulli plate operator

Compact perturbations of controlled systems

The PML Method: Continuous and Semidiscrete Waves.

Partial Differential Equations

THEODORE VORONOV DIFFERENTIABLE MANIFOLDS. Fall Last updated: November 26, (Under construction.)

Control, Stabilization and Numerics for Partial Differential Equations

Final Exam May 4, 2016

UNIVERSITY of LIMERICK OLLSCOIL LUIMNIGH

Hardy inequalities, heat kernels and wave propagation

Finite-time singularity formation for Euler vortex sheet

Transcription:

Hilbert Uniqueness Method and regularity Sylvain Ervedoza 1 Joint work with Enrique Zuazua 2 1 Institut de Mathématiques de Toulouse & CNRS 2 Basque Center for Applied Mathematics Institut Henri Poincaré 29

Outline of the talk 1 Introduction: The Hilbert Uniqueness Method 2 An alternate HUM type method The η-weighted HUM Main result Regularity of the controlled trajectory 3 Applications Distributed controls Boundary control 4 Conclusion

1 Introduction: The Hilbert Uniqueness Method 2 An alternate HUM type method The η-weighted HUM Main result Regularity of the controlled trajectory 3 Applications Distributed controls Boundary control 4 Conclusion

An abstract control problem Let A be the generator of a group on a Hilbert space X. Consider the following model: y (t) = A y(t) + Bv(t), y() = y X, where B L(U, D(A) ) and v L 2 (, T ; U). Assumption For all v L 2 (, T ; U), solutions can be defined in the sense of transposition in C ([, T ]; X). Goal : Exact controllability Fix a time T > and y X. Can we find v L 2 (, T ; U) such that y(t ) =?

Hypotheses Main Assumption A : D(A) X generates a group. Consequences: one can solve the equations forward and backward the same holds for the adjoint equation.

Examples Wave equation in a bounded domain+ BC with distributed control u u + a(x). u + b(x)u + c(x)u = χ ω v, (t, x) R Ω, u Ω =, (u(), u()) = (u, u 1 ) H 1 (Ω) L2 (Ω), ( A = Id a(x). b c(x) ( B = χ ω ), X = H 1 (Ω) L2 (Ω), ), U = L 2 (ω). Wave equation in a bounded domain with boundary control Schrödinger equation A = i + BC, Linearized KdV A = xxx + BC, Maxwell equation,...

Duality Use the adjoint system to characterize the controls! For all z solution of z = A z, z(t ) = z T X, we have y(t ), z T X y, z() X = T v(t), B z(t) U dt. In particular, v is a control if and only if z T X = T v(t), B z(t) U dt + y, z X.

Fundamental hypotheses B : D(A) U, B L(D(A), U). Definition B is admissible if T >, K T >, T B z(t) 2 U dt K T z T 2 X, z D(A). Definition B is exactly observable at time T > if k >, T k z() 2 X B z(t) 2 U dt, z X.

Important remark If B is admissible and exactly observable in some time T, then the following norms are equivalent: z T X, z() X z T 2 obs,1 = T B z(t) 2 U dt T z T 2 obs,2 = η(t) B z(t) 2 U dt, for η, η α > on some interval of length T.

Important remark If B is admissible and exactly observable in some time T, then the following norms are equivalent: z T X, z() X z T 2 obs,1 = T B z(t) 2 U dt T z T 2 obs,2 = η(t) B z(t) 2 U dt, for η, η α > on some interval of length T.

Important remark If B is admissible and exactly observable in some time T, then the following norms are equivalent: z T X, z() X z T 2 obs,1 = T B z(t) 2 U dt T z T 2 obs,2 = η(t) B z(t) 2 U dt, for η, η α > on some interval of length T.

The Hilbert Uniqueness Method (Lions 86) Let T T. Define, for z T X, J(z T ) = 1 2 T B z(t) 2 U dt + y, z(), where z satisfies z = A z, z(t ) = z T. Observability Existence and Uniqueness of a minimizer Z T. Then V = B Z is such that the solution y of y = Ay + BV, y() = y, satisfies y(t ) =. Besides, V is the control of minimal L 2 (, T ; U)-norm.

On the Hilbert Uniqueness Method Recall the following facts: the Hilbert Uniqueness Method does not prove observability/controllability, but prove their equivalence. Observability/controllability properties have to be proved first: several possible methods: multipliers, see Lions, Komornik, Alabau... non-harmonic Fourier series, see Ingham, Haraux, Tucsnak... Carleman estimates, see Puel, Osses, Zhang, etc pseudo-differential calculus, see Bardos, Lebeau, Rauch, Burq, Gerard,... Resolvent estimates, see Burq, Zworski, Miller, Tucsnak...

A regularity problem On the regularity If y D(A), Does the function Z T computed that way belongs to D(A)? Is the controlled solution (y, v) a strong solution? Here, strong solutions means: y C 1 ([, T ]; X) C([, T ]; X 1 ), for some space X 1 smoother than X (for instance D(A)). General Answer : NO!

Consider the wave equation y tt y xx =, < x < 1, < t < T, y(, t) =, y(1, t) = v(t), < t < T, (y(x, ), y t (x, )) = (y (x), y 1 (x)) L 2 (, 1) H 1 (, 1). The adjoint problem is z tt z xx =, z(, t) = z(1, t) =, (z, z 1 ) H 1 (, 1) L2 (, 1), and the solutions write ( z(x, t) = 2 k 1 ẑ k cos(kπt) + ẑk 1 kπ sin(kπt) Controllability in time T = 4 : If ( y (x), y 1 (x) ) = 2 k 1 (ŷ k, ŷ k 1 ) sin(kπx), Ẑ k = ŷ k 1 4k 2 π 2, Ẑ k 1 = ŷk 4. ) sin(kπx),

In particular, the HUM control can be computed explicitly v(t) = Z x (1, t) ( = 1 ŷ ( 1) k k kπ 1 4 k 2 π 2 cos(kπt) ŷ ) k kπ sin(kπt). k 1 = v() = 1 ( 1) k ŷ 1 k 4 kπ! k 1 = If y H 1 (, 1), the controlled solution is not a strong solution in general because of the failure of the compatibility conditions y (1) = v() =.

Main question Goal To design a control method, and only one, which respects the regularity of the solutions. If y D(A), we want Z T D(A ), the controlled equation y = Ay + BV is satisfied in a strong sense. Related result - Dehman Lebeau 9 and Lebeau Nodet 9: The wave equation with distributed control B = χ ω where χ ω is smooth, and where the HUM operator is modified by a weight function η(t) vanishing at t {, T }.

The η-weighted HUM Main result Trajectory 1 Introduction: The Hilbert Uniqueness Method 2 An alternate HUM type method The η-weighted HUM Main result Regularity of the controlled trajectory 3 Applications Distributed controls Boundary control 4 Conclusion

The η-weighted HUM Main result Trajectory The modified HUM method: The η-weighted HUM Let y X, and δ > such that T 2δ T, where T is the time of observability. Define, for z T X, J(z T ) = 1 2 T η(t) B z(t) 2 U dt + y, z(), where z satisfies z = A z, z(t ) = z T and { on (, ] [T, ) η = 1 on [δ, T δ] η. Observability Existence and Uniqueness of a minimizer Z T. Then V = ηb Z is such that the solution y of y = Ay + BV, y() = y, satisfies y(t ) =. Besides, V is the control of minimal L 2 ((, T ), dt/η; U)-norm.

The η-weighted HUM Main result Trajectory Main result Theorem (SE Zuazua 9) Assume that admissibility and observability property hold. Also assume that η C 1 (R). If y D(A), then the minimizer Z T and the control function V = ηb Z computed by the η-weighted HUM are more regular: Z T D(A ), V H 1 (, T ; U). Moreover, there exists a constant C such that Z T D(A ) + V H 1 (,T ;U) C y D(A).

The η-weighted HUM Main result Trajectory Before the proof First remark that, due to the classical observability property, Z T X + v L 2 (,T ;U) C y X. Also remark that admissibility and observability properties yield T k z T D(A ) k z() D(A ) η(t) B z (t) 2 U dt K z T D(A ) It is sufficient to prove that T η(t) B Z (t) 2 U dt <. Indeed, this implies Z T D(A ) and v H 1 (, T ; U).

The η-weighted HUM Main result Trajectory Idea of the proof-i Write the characterization of the control V = ηb Z : = T η(t) B Z (t), B z(t) U dt + y, z() X, for all z solution of z = A z, z(t ) = z T. Then take formally z = Z = (A ) 2 Z : But T and η(t) B Z (t) 2 U dt = Ay, A Z () X T η (t) B Z (t), B Z (t) U dt. Ay, A Z () X C Ay X A Z T X T A Z T X C η(t) B Z (t) 2 U dt.

The η-weighted HUM Main result Trajectory Idea of the proof-ii But T η (t) B Z (t), B Z (t) U dt ( T C B Z (t) ) 1/2 ( ) 1/2 T 2 U dt B Z (t) 2 U dt Using observability, T C A Z T X Z T X C A Z T X y X η(t) B Z (t) 2 U dt C y 2 D(A).

The η-weighted HUM Main result Trajectory More general results Theorem (SE-Zuazua 9) Let s N and assume that η C s (R). If the initial datum y D(A s ), then the minimizer Z T and the control function V given by the η-weighted HUM satisfy Z T D((A ) s ) V H s (, T ; U). Besides, there exists a positive constant C s = C s (η, k, K T ) independent of y D(A s ) T Z T 2 D((A ) s ) + V (s) (t) 2 dt C s y 2 D(A U s ).

The η-weighted HUM Main result Trajectory Comments If η C (R), the η-weighted HUM map defined by { X X L Θ : 2 (, T ; dt/η, U) y (Z T, V ) defines by restriction a map on D(A s ): Θ : D(A s ) D((A ) s ) H s (, T ; U). No smoothness assumption on B, or on [A, BB ]. generalizes Dehman Lebeau 9 to the cases B = χ ω not smooth and boundary control cases, but less precise: pseudo-differential estimates on Θ for the wave equation with a distributed controlled. We have not make precise yet the regularity of the control trajectory!!!!

The η-weighted HUM Main result Trajectory Comments If η C (R), the η-weighted HUM map defined by { X X L Θ : 2 (, T ; dt/η, U) y (Z T, V ) defines by restriction a map on D(A s ): Θ : D(A s ) D((A ) s ) H s (, T ; U). No smoothness assumption on B, or on [A, BB ]. generalizes Dehman Lebeau 9 to the cases B = χ ω not smooth and boundary control cases, but less precise: pseudo-differential estimates on Θ for the wave equation with a distributed controlled. We have not make precise yet the regularity of the control trajectory!!!!

The η-weighted HUM Main result Trajectory And the regularity of the controlled trajectory? Corollary if the initial datum y D(A s ), then the controlled trajectory y with control function V given by the η-weighted HUM satisfies y(t) s k= C k ([, T ]; Z s k ), where the spaces Z j are defined by induction by Z = X, Z j = A 1 (Z j 1 + BB D((A ) j )). Important Remark In several situations, these spaces can be computed.

The η-weighted HUM Main result Trajectory Idea of the proof Z T D((A ) s ) and Z solves Z = A Z = Z s k= C k ([, T ]; D((A ) s k ) = V = ηb Z s k= C k ([, T ]; B D((A ) s k ) = BV = ηbb Z s k= C k ([, T ]; BB D((A ) s k ) Besides, V H s (, T ; U) since (A ) s Z T X. Hence the result by standard regularity results.

Distributed controls Boundary control 1 Introduction: The Hilbert Uniqueness Method 2 An alternate HUM type method The η-weighted HUM Main result Regularity of the controlled trajectory 3 Applications Distributed controls Boundary control 4 Conclusion

Distributed controls Boundary control Distributed Controls Let Ω R N and ω Ω. Wave equation: y y = vχ ω, y =, (y(), y ()) = (y, y 1 ) in Ω (, ), on Ω (, ), H 1(Ω) L2 (Ω), v L 2 ((, T ) ω) is the control function χ ω (x) supported in ω, strictly positive on some ω ω (no smoothness assumption)

Distributed controls Boundary control Geometric Control Condition Assume GCC in time T : (Bardos Lebeau Rauch 92 & Burq Gerard 96) All the rays of Geometric Optics enter in the control region ω before the time T >. { on (, ] [T, ) Take T 2δ > T, and η = 1 on [δ, T δ] η.

Distributed controls Boundary control The η-weighted HUM Our functional reads: T J(z, z 1 ) = 1 η(t)χ 2 2 ω(x) z(x, t) 2 dxdt ω + y, z (, ) H 1 (Ω) H 1 (Ω) y 1 (x) z(x, ) dx Ω where z is the solution of z z =, z =, (z(t ), z (T )) = (z, z 1 ) in Ω (, ), on Ω (, ), L 2 (Ω) H 1 (Ω).

Distributed controls Boundary control Results Theorem-I Given any (y, y 1 ) H 1 (Ω) L2 (Ω), there exists a unique minimizer (Z, Z 1 ) of J over L 2 (Ω) H 1 (Ω). The function V (x, t) = η(t)χ ω (x)z (x, t) is the control function of minimal L 2 (, T ; dt/η; L 2 (ω))-norm, defined by T v 2 L 2 (,T ;dt/η;l 2 (ω)) = ω v(x, t) 2 dx dt η(t). If (y, y 1 ) D(A s ) for some s N, (Z, Z 1 ) D((A ) s ) = D(A s 1 ) and V H s (, T ; L2 (ω)).

Distributed controls Boundary control Results Theorem-II If (y, y 1 ) D(A s ) for some s N and χ ω is smooth, the η-weighted HUM yields V H s (, T ; L2 (ω)) C k ([, T ]; H s k k= (ω)) and the controlled trajectory satisfies y s s k= C k ([, T ]; H s+1 k (Ω) H s k (Ω)). Yields another proof of the one in Dehman Lebeau 9.

Distributed controls Boundary control Boundary control Let Ω R N, Γ = Ω, Γ Γ satisifying GCC in time T. Let ξ Γ be defined on Ω, non-vanishing on Γ. We now consider the following wave equation: y y =, in Ω (, ), y = ξ Γ v, on Γ (, ), (y(), y ()) = (y, y 1 ) L 2 (Ω) H 1 (Ω). η(t) as before.

Distributed controls Boundary control Our method The functional J reads: J(z, z 1 ) = 1 2 + T 1 Γ η(t)ξ Γ (x) 2 n z(x, t) 2 dγdt y (x) z (x, ) dx y 1, z(, ) H 1 (Ω) H 1 (Ω), where z is the solution of z z =, z =, (z(t ), z (T )) = (z, z 1 ) in Ω (, ), on Ω (, ), H 1(Ω) L2 (Ω).

Distributed controls Boundary control Results Theorem Assume that Γ, T satisfies GCC. Given any (y, y 1 ) L 2 (Ω) H 1 (Ω), there exists a unique minimizer (Z, Z 1 ) of J over H 1 (Ω) L2 (Ω). The function V (x, t) = η(t)ξ Γ (x) n Y (x, t) Γ1 is the control method of minimal L 2 (, T ; dt/η; L 2 (Γ))-norm, defined by T v 2 L 2 (,T ;dt/η;l 2 (Γ)) = Γ v(x, t) 2 dγ dt η(t). If (y, y 1 ) D(A s ) for some s N, then (Z, Z 1 ) D((A ) s ) = D(A s+1 ) and V H s (, T ; L2 (Γ)).

Distributed controls Boundary control Results Theorem Assume that the function ξ Γ is smooth. If (y, y 1 ) D(A s ) for some s N, the η- weighted HUM yields: s V H s (, T ; L2 (Γ 1 )) C k ([, T ]; H s k 1/2 (Γ 1 )) k= and (Z, Z 1 ) D((A ) s ) = D(A s+1 ). In particular, the controlled trajectory y satisfies y s k= C k ([, T ]; H s k (Ω) H s 1 k (Ω)).

1 Introduction: The Hilbert Uniqueness Method 2 An alternate HUM type method The η-weighted HUM Main result Regularity of the controlled trajectory 3 Applications Distributed controls Boundary control 4 Conclusion

Further applications Orders of convergence of discrete controls Joint work with Enrique Zuazua The exact controls computed with the η-weighted HUM converge at a rate h 2/3 to the continuous η-hum control. Controllability of semi-linear wave equations See Dehman-Lebeau 9 for such examples.

Open problem-i Exact controllability What happens for the exact controllability when A does not generate a group? Typical example: Linearized KdV equation Can we design a HUM type method which yields smooth controlled trajectory for smooth initial data? Do we have such regularity properties for the non linear KdV equation? Work in progress...

Open problem-ii Exact controllability to trajectories What happens when A does not generate a group? Typical example: The heat equation. Can we design a HUM type method which yields smooth controlled trajectory for smooth initial data? Widely open.

Thank you for your attention! Articles available on http://www.math.uvsq.fr/ ervedoza/ http://www.math.univ-toulouse.fr/ ervedoza/