contrôlabilité d’équations aux di˙érencesmazanti/...contrôlabilité d’équations aux...

71
Contrôlabilité d’équations aux diérences Guilherme Mazanti CRAN, Nancy octobre Laboratoire de Mathématiques d’Orsay Université Paris-Sud Université Paris-Saclay France

Upload: others

Post on 29-Mar-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Contrôlabilité d’équations aux di�érences

Guilherme Mazanti

CRAN, Nancy15 octobre 2018

Laboratoire de Mathématiques d’OrsayUniversité Paris-SudUniversité Paris-SaclayFrance

Page 2: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Outline

1 Introduction

2 Explicit solution

3 A word on stability

4 Relative controllability

5 Approximate and exact controllability

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 3: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionControlled di�erence equations

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

• Λ = (Λ1, . . . , ΛN ) ∈ (0,+∞)N : vector of positive delays;

• x (t ) ∈ Ãd , u(t ) ∈ Ãm (or Òd , Òm );• Notation: Λmin = minj Λj , Λmax = maxj Λj .

Goal: study the controllability of Σ(A,B , Λ): given initial and �nalstates, can one �nd a control u steering the system from theinitial to the �nal state?

The state at time t can be either x (t ) ∈ Ãd or xt = x (t + ·)|[−Λmax,0)according to the type of controllability we consider.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 4: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionControlled di�erence equations

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

• Λ = (Λ1, . . . , ΛN ) ∈ (0,+∞)N : vector of positive delays;

• x (t ) ∈ Ãd , u(t ) ∈ Ãm (or Òd , Òm );• Notation: Λmin = minj Λj , Λmax = maxj Λj .

Goal: study the controllability of Σ(A,B , Λ): given initial and �nalstates, can one �nd a control u steering the system from theinitial to the �nal state?

The state at time t can be either x (t ) ∈ Ãd or xt = x (t + ·)|[−Λmax,0)according to the type of controllability we consider.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 5: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

Hyperbolic PDEs→ di�erence equations: [Cooke, Krumme; 1968],[Slemrod; 1971], [Greenberg, Li; 1984], [Coron, Bastin, d’Andréa Novel;2008], [Fridman, Mondié, Saldivar; 2010], [Gugat, Sigalotti; 2010]...

∂t yi (t , x ) + ∂x yi (t , x ) + αi (x )yi (t , x ) = 0,

yi (t , 0) =N∑j=1

mi j yj (t , Λj ) +m∑j=1

bi ju j (t ),

i ∈ û1,N ü,

t ∈ Ò+,

x ∈ [0, Λi ].

Method of characteristics: for t ≥ Λmax and j ∈ û1,N ü, one hasyj (t , Λj ) = e

−r Λj0 αj (x )dx yj (t − Λj , 0),

and then y (t ) = (yi (t , 0))i ∈û1,N ü satis�es a di�erence equationwith control u(t ) = (ui (t ))i ∈û1,mü:

yi (t , 0) =N∑j=1

mi j e−

r Λj0 αj (x )dx yj (t − Λj , 0) +

m∑j=1

bi ju j (t ).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 6: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

Hyperbolic PDEs→ di�erence equations: [Cooke, Krumme; 1968],[Slemrod; 1971], [Greenberg, Li; 1984], [Coron, Bastin, d’Andréa Novel;2008], [Fridman, Mondié, Saldivar; 2010], [Gugat, Sigalotti; 2010]...

∂t yi (t , x ) + ∂x yi (t , x ) + αi (x )yi (t , x ) = 0,

yi (t , 0) =N∑j=1

mi j yj (t , Λj ) +m∑j=1

bi ju j (t ),

i ∈ û1,N ü,

t ∈ Ò+,

x ∈ [0, Λi ].

Method of characteristics: for t ≥ Λmax and j ∈ û1,N ü, one hasyj (t , Λj ) = e

−r Λj0 αj (x )dx yj (t − Λj , 0),

and then y (t ) = (yi (t , 0))i ∈û1,N ü satis�es a di�erence equationwith control u(t ) = (ui (t ))i ∈û1,mü:

yi (t , 0) =N∑j=1

mi j e−

r Λj0 αj (x )dx yj (t − Λj , 0) +

m∑j=1

bi ju j (t ).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 7: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

Hyperbolic PDEs→ di�erence equations: [Cooke, Krumme; 1968],[Slemrod; 1971], [Greenberg, Li; 1984], [Coron, Bastin, d’Andréa Novel;2008], [Fridman, Mondié, Saldivar; 2010], [Gugat, Sigalotti; 2010]...

∂t yi (t , x ) + ∂x yi (t , x ) + αi (x )yi (t , x ) = 0,

yi (t , 0) =N∑j=1

mi j yj (t , Λj ) +m∑j=1

bi ju j (t ),

i ∈ û1,N ü,

t ∈ Ò+,

x ∈ [0, Λi ].

Method of characteristics: for t ≥ Λmax and j ∈ û1,N ü, one hasyj (t , Λj ) = e

−r Λj0 αj (x )dx yj (t − Λj , 0),

and then y (t ) = (yi (t , 0))i ∈û1,N ü satis�es a di�erence equationwith control u(t ) = (ui (t ))i ∈û1,mü:

yi (t , 0) =N∑j=1

mi j e−

r Λj0 αj (x )dx yj (t − Λj , 0) +

m∑j=1

bi ju j (t ).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 8: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

Λ1

Λ2

Λ3

ΛN

Edges: EVertices: VEdges incidenton q ∈ V: Eq

∂2t t yi (t , x ) = ∂

2xx yi (t , x ), i ∈ E, t ∈ Ò+, x ∈ [0, Li ],

yi (t , q ) = yj (t , q ), q ∈ V, i , j ∈ Eq ,

+ conditions and control on vertices.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 9: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

Λ1

Λ2

Λ3

ΛN

Edges: EVertices: VEdges incidenton q ∈ V: Eq

∂2t t yi (t , x ) = ∂

2xx yi (t , x ), i ∈ E, t ∈ Ò+, x ∈ [0, Li ],

yi (t , q ) = yj (t , q ), q ∈ V, i , j ∈ Eq ,

+ conditions and control on vertices.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 10: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

D’Alembert decomposition on travelling waves:

System of 2N transport equations.Can be reduced to a system of di�erence equations.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 11: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation I: hyperbolic PDEs

D’Alembert decomposition on travelling waves:

System of 2N transport equations.Can be reduced to a system of di�erence equations.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 12: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMotivation II: NFDEs

Consider a neutral functional di�erential equation of the form

d

d t

x (t ) −N∑j=1

Aj x (t − Λj )

= Lxt + Bu(t ), (NFDE)

where xt = x (t + ·)|[−Λmax,0] and L : C([−Λmax, 0],Òd ) → Òd is alinear map.

• Under no control, the stability of Σ(A,B , Λ) is a necessarycondition for that of (NFDE) (see [Henry; 1974] and [Hale,Verduyn Lunel; 2002]).• The stabilizability of Σ(A,B , Λ) by linear feedback laws is anecessary condition for that of (NFDE) (see [Hale, VerduynLunel; 2002]).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 13: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionRational dependence

Rational dependence of the delays Λ1, . . . , ΛN plays a key role inthe behavior of Σ(A,B , Λ).

De�nitionLet Λ = (Λ1, . . . , ΛN ) ∈ (0,+∞)N .• We say that Λ is rationally dependent if there existn1, . . . , nN ∈ Ú (or Ñ) not all zero such that

n1Λ1 + · · · + nNΛN = 0.Otherwise, we say that Λ is rationally independent.• We say that Λ is commensurable if there exist λ > 0 andk1, . . . , kN ∈ Î

∗ such thatΛj = λk j , [j ∈ û1,N ü.

Otherwise, we say that Λ is incommensurable.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 14: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionThe commensurable case

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

The controllability problem for Σ(A,B , Λ) is easy if Λ1, . . . , ΛNare commensurable. Assume to simplify that Λj = j λ:x (t ) = A1x (t − λ) + A2x (t − 2λ) + · · · + AN x (t − Nλ) + Bu(t ).

Set

X (t ) =

©­­­­«x (t )

x (t − λ)...

x (t − (N − 1)λ)

ª®®®®¬.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 15: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionThe commensurable case

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

The controllability problem for Σ(A,B , Λ) is easy if Λ1, . . . , ΛNare commensurable. Assume to simplify that Λj = j λ:x (t ) = A1x (t − λ) + A2x (t − 2λ) + · · · + AN x (t − Nλ) + Bu(t ).

Set

X (t ) =

©­­­­«x (t )

x (t − λ)...

x (t − (N − 1)λ)

ª®®®®¬.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 16: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionThe commensurable case

Then

X (t ) =

©­­­­­­«

A1 A2 · · · AN−1 ANId 0 · · · 0 00 Id · · · 0 0...

.... . .

......

0 0 · · · Id 0

ª®®®®®®¬X (t − λ) +

©­­­­­­«

B

00...

0

ª®®®®®®¬u(t ).

System in discrete time. Controllability can be studied through aKalman-like criterion.

Does not work in the incommensurable case.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 17: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionThe commensurable case

Then

X (t ) =

©­­­­­­«

A1 A2 · · · AN−1 ANId 0 · · · 0 00 Id · · · 0 0...

.... . .

......

0 0 · · · Id 0

ª®®®®®®¬X (t − λ) +

©­­­­­­«

B

00...

0

ª®®®®®®¬u(t ).

System in discrete time. Controllability can be studied through aKalman-like criterion.

Does not work in the incommensurable case.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 18: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMain problems

Goal: characterize relative, approximate, and exact controllabilityof Σ(A,B , Λ).

• Relative controllability in timeT : given x0 : [−Λmax, 0) → Ãd

and xf ∈ Ãd , does there exist u : [0,T ] → Ãm such that thesolution x : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initialcondition x0 and control u satis�es x (T ) = xf ?

• Approximate/exact controllability in timeT : givenx0 : [−Λmax, 0) → Ãd and xf : [−Λmax, 0) → Ãd , does thereexist u : [0,T ] → Ãm such that the solutionx : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initial condition x0and control u is such that x (T + ·)|[−Λmax,0) is close to xf in acertain norm / satis�es x (T + ·)|[−Λmax,0) = xf ?

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 19: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMain problems

Goal: characterize relative, approximate, and exact controllabilityof Σ(A,B , Λ).

• Relative controllability in timeT : given x0 : [−Λmax, 0) → Ãd

and xf ∈ Ãd , does there exist u : [0,T ] → Ãm such that thesolution x : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initialcondition x0 and control u satis�es x (T ) = xf ?• Approximate/exact controllability in timeT : givenx0 : [−Λmax, 0) → Ãd and xf : [−Λmax, 0) → Ãd , does thereexist u : [0,T ] → Ãm such that the solutionx : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initial condition x0and control u is such that x (T + ·)|[−Λmax,0) is close to xf in acertain norm / satis�es x (T + ·)|[−Λmax,0) = xf ?

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 20: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

IntroductionMain problems

Previous results:• Stability analysis for the uncontrolled system: [Cruz, Hale;1970], [Henry; 1974], [Melvin; 1974], [Hale; 1975], [Silkowski; 1976],[Avellar, Hale; 1980], [Hale, Verduyn Lunel; 1993], [Michiels et al;2009], [Chitour, M., Sigalotti; 2016]...• Stabilization by linear feedbacks: [Hale, Verduyn Lunel; 2002],[Hale, Verduyn Lunel; 2003].• Relative controllability: introduced in [Chyung; 1970], [Olbrot;1972], [Klamka; 1976], originally for delays in control. Results inthe case N = 2 and integer delays in [Diblík, Khusainov,Růžičková; 2008] and [Pospíšil, Diblík, Fečkan; 2015].• Spectral and approximate controllability inLp ([−Λmax, 0],Ã

d ): [Salamon; 1984] (for more general NFDEs,using duality) and [O’Connor, Tarn; 1983].

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 21: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionWell-posedness

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

• Solution with initial condition x0 : [−Λmax, 0) → Ãd andcontrol u : [0,T ] → Ãm : x satisfying Σ(A,B , Λ) for t ≥ 0 andx (t ) = x0(t ) for −Λmax ≤ t < 0.• Notation: xt = x (t + ·)|[−Λmax,0).• Existence and uniqueness follow easily; no regularityassumptions!

• If x0,u ∈ Lp for some p , then x ∈ Lp , xt ∈ Lp .• Ck solutions exist under a compatibility condition:

limt→0

x(r )0 (t ) =

N∑j=1

Aj x(r )0 (−Λj ) + Bu

(r )(0), r ∈ û0, k ü.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 22: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionWell-posedness

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

• Solution with initial condition x0 : [−Λmax, 0) → Ãd andcontrol u : [0,T ] → Ãm : x satisfying Σ(A,B , Λ) for t ≥ 0 andx (t ) = x0(t ) for −Λmax ≤ t < 0.• Notation: xt = x (t + ·)|[−Λmax,0).• Existence and uniqueness follow easily; no regularityassumptions!• If x0,u ∈ Lp for some p , then x ∈ Lp , xt ∈ Lp .• Ck solutions exist under a compatibility condition:

limt→0

x(r )0 (t ) =

N∑j=1

Aj x(r )0 (−Λj ) + Bu

(r )(0), r ∈ û0, k ü.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 23: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionWell-posedness

Two ways to see the solution: x : [−Λmax,T ] → Ãd orxt = x (t + ·)|[−Λmax,0) : [−Λmax, 0) → Ãd .

Example

x (t ) =

(18 1

0 18

)x (t − 1) +

(17

14

− 1417

)x

(t −√22

)+

(19 0

− 1219

)x

(t − 1

π

)+

(01

)u(t ),

u(t ) = sin(4t ) +1

2cos(5t ).

t

x (t ) = (x1(t ), x2(t ))

−1 10

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 24: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

Technique to study controllability: explicit formula for thesolution.

Proposition ([Chitour, M., Sigalotti; 2016], [M.; 2017])

Let x0 : [−Λmax, 0) → Ãd and u : [0,T ] → Ãm . The solutionx : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initial condition x0 andcontrol u is given for t ∈ [0,T ] byx (t ) =

∑(n,j )∈ÎN×û1,N ü−Λj ≤t−Λ ·n<0

Ξn−e jAj x0(t − Λ · n) +∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n),

where Ξn is de�ned inductively for n ∈ ÎN by Ξ0 = Idd andΞn =

∑Nk=1nk ≥1

AkΞn−ek for n = (n1, . . . , nN ) ∈ ÎN \ {0}.

• Term due to the initial condition + term due to the control.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 25: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

Technique to study controllability: explicit formula for thesolution.

Proposition ([Chitour, M., Sigalotti; 2016], [M.; 2017])

Let x0 : [−Λmax, 0) → Ãd and u : [0,T ] → Ãm . The solutionx : [−Λmax,T ] → Ãd of Σ(A,B , Λ) with initial condition x0 andcontrol u is given for t ∈ [0,T ] byx (t ) =

∑(n,j )∈ÎN×û1,N ü−Λj ≤t−Λ ·n<0

Ξn−e jAj x0(t − Λ · n) +∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n),

where Ξn is de�ned inductively for n ∈ ÎN by Ξ0 = Idd andΞn =

∑Nk=1nk ≥1

AkΞn−ek for n = (n1, . . . , nN ) ∈ ÎN \ {0}.

• Term due to the initial condition + term due to the control.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 26: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

Technique to study controllability: explicit formula for thesolution.

Proposition ([Chitour, M., Sigalotti; 2016], [M.; 2017])

Let u : [0,T ] → Ãm . The solution x : [−Λmax,T ] → Ãd ofΣ(A,B , Λ) with initial condition 0 and control u is given fort ∈ [0,T ] by

x (t ) =∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n),

where Ξn is de�ned inductively for n ∈ ÎN by Ξ0 = Idd andΞn =

∑Nk=1nk ≥1

AkΞn−ek for n = (n1, . . . , nN ) ∈ ÎN \ {0}.

• Term due to the initial condition + term due to the control.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 27: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

Idea for the explicit formula:

x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ),

x (t − Λj ) =N∑k=1

Ak x (t − Λj − Λk ) + Bu(t − Λj ).

Then

x (t ) =N∑j=1

N∑k=1

AjAk x (t − Λj − Λk ) +N∑j=1

AjBu(t − Λj ) + Bu(t ).

Proceed inductively until x is evaluated at times in [−Λmax, 0).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 28: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

Idea for the explicit formula:

x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ),

x (t − Λj ) =N∑k=1

Ak x (t − Λj − Λk ) + Bu(t − Λj ).

Then

x (t ) =N∑j=1

N∑k=1

AjAk x (t − Λj − Λk ) +N∑j=1

AjBu(t − Λj ) + Bu(t ).

Proceed inductively until x is evaluated at times in [−Λmax, 0).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 29: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

x (t ) =∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

One may have di�erent n yielding the same Λ · n. Goal: groupthem in the above formula.

• Equivalence relation on ÎN : n ∼ n′ ⇐⇒ Λ · n = Λ · n′.• Equivalence classes: [n]Λ or simply [n].Remark: [n] = {n} if Λ is rationally independent!• NΛ = Î

N /∼.

Set Ξ̂Λ[n] =

∑n′∈[n]

Ξn′ . Then x (t ) =∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

In particular, di�erent [n] ⇐⇒ di�erent Λ · n.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 30: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

x (t ) =∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

One may have di�erent n yielding the same Λ · n. Goal: groupthem in the above formula.

• Equivalence relation on ÎN : n ∼ n′ ⇐⇒ Λ · n = Λ · n′.• Equivalence classes: [n]Λ or simply [n].Remark: [n] = {n} if Λ is rationally independent!• NΛ = Î

N /∼.

Set Ξ̂Λ[n] =

∑n′∈[n]

Ξn′ . Then x (t ) =∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

In particular, di�erent [n] ⇐⇒ di�erent Λ · n.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 31: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

x (t ) =∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

One may have di�erent n yielding the same Λ · n. Goal: groupthem in the above formula.

• Equivalence relation on ÎN : n ∼ n′ ⇐⇒ Λ · n = Λ · n′.• Equivalence classes: [n]Λ or simply [n].Remark: [n] = {n} if Λ is rationally independent!• NΛ = Î

N /∼.

Set Ξ̂Λ[n] =

∑n′∈[n]

Ξn′ . Then x (t ) =∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

In particular, di�erent [n] ⇐⇒ di�erent Λ · n.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 32: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Explicit solutionExplicit solution

x (t ) =∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

One may have di�erent n yielding the same Λ · n. Goal: groupthem in the above formula.

• Equivalence relation on ÎN : n ∼ n′ ⇐⇒ Λ · n = Λ · n′.• Equivalence classes: [n]Λ or simply [n].Remark: [n] = {n} if Λ is rationally independent!• NΛ = Î

N /∼.

Set Ξ̂Λ[n] =

∑n′∈[n]

Ξn′ . Then x (t ) =∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

In particular, di�erent [n] ⇐⇒ di�erent Λ · n.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 33: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

A word on stabilityHale–Silkowski criterion

Stability for u = 0:

Σ0(A, Λ) : x (t ) =N∑j=1

Aj x (t − Λj ), t ≥ 0.

Theorem ([Hale; 1975], [Silkowski; 1976])The following are equivalent:

• max(θ1,...,θN )∈[0,2π]N

ρ©­«N∑j=1

Aj ei θj ª®¬ < 1;

• Σ0(A, Λ) is exponentially stable for some Λ ∈ (0,+∞)N withrationally independent components;• Σ0(A, Λ) is exponentially stable for every Λ ∈ (0,+∞)N .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 34: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

A word on stabilityHale–Silkowski criterion

• Obtained by spectral methods. Stability depends on the realparts of the roots of det

(Id−

∑Nj=1 Aj e

−sΛj)= 0 (exponential

polynomial, see [Avellar, Hale; 1980]).• For rationally dependent delays: [Michiels et al.; 2009].

• Our previous explicit formula can be adapted for the case oftime-varying matrices Aj (t ):

x (t ) =N∑j=1

Aj (t )x (t − Λj ), t ≥ 0.

One can then generalize Hale–Silkowski criterion: [Chitour, M.,Sigalotti; 2016]

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 35: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

A word on stabilityHale–Silkowski criterion

• Obtained by spectral methods. Stability depends on the realparts of the roots of det

(Id−

∑Nj=1 Aj e

−sΛj)= 0 (exponential

polynomial, see [Avellar, Hale; 1980]).• For rationally dependent delays: [Michiels et al.; 2009].• Our previous explicit formula can be adapted for the case oftime-varying matrices Aj (t ):

x (t ) =N∑j=1

Aj (t )x (t − Λj ), t ≥ 0.

One can then generalize Hale–Silkowski criterion: [Chitour, M.,Sigalotti; 2016]

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 36: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityDe�nition

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

De�nitionWe say that Σ(A,B , Λ) is relatively controllable in timeT > 0 if,for every x0 : [−Λmax, 0) → Ãd and xf ∈ Ãd , there existsu : [0,T ] → Ãm such that the unique solution x of Σ(A,B , Λ) withinitial condition x0 and control u satis�es x (T ) = xf .

It su�ces to consider zero initial conditions. Thenx (t ) =

∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 37: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityDe�nition

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

De�nitionWe say that Σ(A,B , Λ) is relatively controllable in timeT > 0 if,for every x0 : [−Λmax, 0) → Ãd and xf ∈ Ãd , there existsu : [0,T ] → Ãm such that the unique solution x of Σ(A,B , Λ) withinitial condition x0 and control u satis�es x (T ) = xf .

It su�ces to consider zero initial conditions. Thenx (t ) =

∑[n]∈NΛΛ ·n≤t

Ξ̂Λ[n]Bu(t − Λ · n).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 38: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRelative controllability criterion

Theorem ([M.; 2017])The following statements are equivalent:

• Σ(A,B , Λ) is relatively controllable in timeT ;

• Span{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}= Ãd ;

• \ε0 > 0 s.t., [ε ∈ (0, ε0), [x0 : [−Λmax, 0) → Ãd , and[xf : [0, ε] → Ãd , \u : [0,T + ε] → Ãm s.t. the solution x ofΣ(A,B , Λ) with initial condition x0 and control u satis�esx (T + ·)|[0,ε] = xf ;

• \ε0 > 0 s.t., [p ∈ [1,+∞], [ε ∈ (0, ε0), [x0 ∈ Lp ((−Λmax, 0),Ãd ),and [xf ∈ Lp ((0, ε),Ãd ), \u ∈ Lp ((0,T + ε),Ãm ) s.t. the solution xof Σ(A,B , Λ) with initial condition x0 and control u satis�esx ∈ Lp ((−Λmax,T + ε),Ãd ) and x (T + ·)|[0,ε] = xf .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 39: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRelative controllability criterion

Theorem ([M.; 2017])The following statements are equivalent:

• Σ(A,B , Λ) is relatively controllable in timeT ;

• Span{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}= Ãd ;

• \ε0 > 0 s.t., [ε ∈ (0, ε0), [x0 : [−Λmax, 0) → Ãd , and[xf : [0, ε] → Ãd , \u : [0,T + ε] → Ãm s.t. the solution x ofΣ(A,B , Λ) with initial condition x0 and control u satis�esx (T + ·)|[0,ε] = xf ;

• \ε0 > 0 s.t., [p ∈ [1,+∞], [ε ∈ (0, ε0), [x0 ∈ Lp ((−Λmax, 0),Ãd ),and [xf ∈ Lp ((0, ε),Ãd ), \u ∈ Lp ((0,T + ε),Ãm ) s.t. the solution xof Σ(A,B , Λ) with initial condition x0 and control u satis�esx ∈ Lp ((−Λmax,T + ε),Ãd ) and x (T + ·)|[0,ε] = xf .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 40: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRelative controllability criterion

Theorem ([M.; 2017])The following statements are equivalent:

• Σ(A,B , Λ) is relatively controllable in timeT ;

• Span{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}= Ãd ;

• \ε0 > 0 s.t., [ε ∈ (0, ε0), [x0 : [−Λmax, 0) → Ãd , and[xf : [0, ε] → Ãd , \u : [0,T + ε] → Ãm s.t. the solution x ofΣ(A,B , Λ) with initial condition x0 and control u satis�esx (T + ·)|[0,ε] = xf ;

• \ε0 > 0 s.t., [p ∈ [1,+∞], [ε ∈ (0, ε0), [x0 ∈ Lp ((−Λmax, 0),Ãd ),and [xf ∈ Lp ((0, ε),Ãd ), \u ∈ Lp ((0,T + ε),Ãm ) s.t. the solution xof Σ(A,B , Λ) with initial condition x0 and control u satis�esx ∈ Lp ((−Λmax,T + ε),Ãd ) and x (T + ·)|[0,ε] = xf .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 41: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRelative controllability criterion

Remarks:

• Can also be generalized to other spaces (e.g., Ck ).• Generalizes Kalman criterion: for x (t ) = Ax (t − 1) + Bu(t ),one hasSpan

{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}

= Ran(B AB A2B · · · A bT cB

).

• Relative controllability 6=⇒ stabilizability by a lineartime-invariant feedback law.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 42: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRational dependence of the delays

What happens to the relative controllability of Σ(A,B , Λ) if wechange Λ?• Z (Λ) = {n ∈ ÚN | Λ · n = 0}.• We write Λ 4 L if Z (Λ) ⊂ Z (L) (L is more rationally dependentthan Λ).• We write Λ ≈ L if Λ 4 L and L 4 Λ (L is as rationallydependent as Λ).• Remark: Z (Λ) = {0} ⇐⇒ Λ is rationally independent.

Theorem ([M.; 2017])If Λ 4 L and Σ(A,B , L) is relatively controllable in timeT , thenΣ(A,B , Λ) is relatively controllable in time κT , withκ = maxj ∈û1,N ü

ΛjLj.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 43: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRational dependence of the delays

What happens to the relative controllability of Σ(A,B , Λ) if wechange Λ?• Z (Λ) = {n ∈ ÚN | Λ · n = 0}.• We write Λ 4 L if Z (Λ) ⊂ Z (L) (L is more rationally dependentthan Λ).• We write Λ ≈ L if Λ 4 L and L 4 Λ (L is as rationallydependent as Λ).• Remark: Z (Λ) = {0} ⇐⇒ Λ is rationally independent.

Theorem ([M.; 2017])If Λ 4 L and Σ(A,B , L) is relatively controllable in timeT , thenΣ(A,B , Λ) is relatively controllable in time κT , withκ = maxj ∈û1,N ü

ΛjLj.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 44: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRational dependence of the delays

The converse is false, but we have the following.

Theorem ([M.; 2017])Let Λ = (Λ1, . . . , ΛN ) ∈ (0,+∞)N andT > 0. For every ε > 0, thereexists a commensurable L = (L1, . . . , LN ) ∈ (0,+∞)N satisfyingL < Λ and 1 ≤ Λj

Lj< 1 + ε for every j ∈ û1,N ü such that, if

Σ(A,B , Λ) is relatively controllable in timeT , then Σ(A,B , L) isalso relatively controllable in timeT .

Idea of the proof: Construct L as a small perturbation of Λ sothat, if Λ · n, L · n ∈ [0,T ], then

Λ · n = Λ · n′ ⇐⇒ L · n = L · n′.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 45: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityRational dependence of the delays

The converse is false, but we have the following.

Theorem ([M.; 2017])Let Λ = (Λ1, . . . , ΛN ) ∈ (0,+∞)N andT > 0. For every ε > 0, thereexists a commensurable L = (L1, . . . , LN ) ∈ (0,+∞)N satisfyingL < Λ and 1 ≤ Λj

Lj< 1 + ε for every j ∈ û1,N ü such that, if

Σ(A,B , Λ) is relatively controllable in timeT , then Σ(A,B , L) isalso relatively controllable in timeT .

Idea of the proof: Construct L as a small perturbation of Λ sothat, if Λ · n, L · n ∈ [0,T ], then

Λ · n = Λ · n′ ⇐⇒ L · n = L · n′.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 46: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

• For x (t ) = Ax (t − 1) + Bu(t ), one has

Span{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}

= Ran(B AB A2B · · · A bT cB

).

• In this case, by Cayley–Hamilton theorem, relativecontrollability in some timeT > 0 ⇐⇒ relativecontrollability in time d − 1.

Theorem ([M.; 2017])If Σ(A,B , Λ) is relatively controllable in some timeT > 0, then itis also relatively controllable in timeT = (d − 1)Λmax.

Idea: prove �rst for commensurable delays using stateaugmentation + Kalman criterion + Cayley–Hamilton. Conclude bythe previous results.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 47: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

• For x (t ) = Ax (t − 1) + Bu(t ), one has

Span{Ξ̂Λ[n]Bw

��� [n] ∈ NΛ, Λ · n ≤ T , w ∈ Ãm}

= Ran(B AB A2B · · · A bT cB

).

• In this case, by Cayley–Hamilton theorem, relativecontrollability in some timeT > 0 ⇐⇒ relativecontrollability in time d − 1.

Theorem ([M.; 2017])If Σ(A,B , Λ) is relatively controllable in some timeT > 0, then itis also relatively controllable in timeT = (d − 1)Λmax.

Idea: prove �rst for commensurable delays using stateaugmentation + Kalman criterion + Cayley–Hamilton. Conclude bythe previous results.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 48: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

Assume to simplify that Λ is rationally independent. Criterion:Span

{ΞnBw

�� n ∈ ÎN , Λ · n ≤ (d − 1)Λmax, w ∈ Ãm}= Ãd .

Finitely many vectors, but their number tends to +∞ asΛmaxΛmin→ +∞.

Theorem ([M.; 2017])If Λ is rationally independent, then Σ(A,B , Λ) is relativelycontrollable in some timeT > 0 if and only if

Span{ΞnBe j

�� n ∈ ÎN , |n|1 ≤ d − 1, j ∈ û1,mü} = Ãd .Number of vectors to compute: m

(N+d−1d−1

). Independent of Λ.

Idea of the proof: Λ ≈ Lε where Lε is a rationally independentperturbation of (1, 1, . . . , 1).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 49: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

Assume to simplify that Λ is rationally independent. Criterion:Span

{ΞnBe j

�� n ∈ ÎN , Λ · n ≤ (d − 1)Λmax, j ∈ û1,mü} = Ãd .

Finitely many vectors, but their number tends to +∞ asΛmaxΛmin→ +∞.

Theorem ([M.; 2017])If Λ is rationally independent, then Σ(A,B , Λ) is relativelycontrollable in some timeT > 0 if and only if

Span{ΞnBe j

�� n ∈ ÎN , |n|1 ≤ d − 1, j ∈ û1,mü} = Ãd .Number of vectors to compute: m

(N+d−1d−1

). Independent of Λ.

Idea of the proof: Λ ≈ Lε where Lε is a rationally independentperturbation of (1, 1, . . . , 1).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 50: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

Assume to simplify that Λ is rationally independent. Criterion:Span

{ΞnBe j

�� n ∈ ÎN , Λ · n ≤ (d − 1)Λmax, j ∈ û1,mü} = Ãd .Finitely many vectors, but their number tends to +∞ asΛmaxΛmin→ +∞.

Theorem ([M.; 2017])If Λ is rationally independent, then Σ(A,B , Λ) is relativelycontrollable in some timeT > 0 if and only if

Span{ΞnBe j

�� n ∈ ÎN , |n|1 ≤ d − 1, j ∈ û1,mü} = Ãd .Number of vectors to compute: m

(N+d−1d−1

). Independent of Λ.

Idea of the proof: Λ ≈ Lε where Lε is a rationally independentperturbation of (1, 1, . . . , 1).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 51: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Relative controllabilityMinimal controllability time

Assume to simplify that Λ is rationally independent. Criterion:Span

{ΞnBe j

�� n ∈ ÎN , Λ · n ≤ (d − 1)Λmax, j ∈ û1,mü} = Ãd .Finitely many vectors, but their number tends to +∞ asΛmaxΛmin→ +∞.

Theorem ([M.; 2017])If Λ is rationally independent, then Σ(A,B , Λ) is relativelycontrollable in some timeT > 0 if and only if

Span{ΞnBe j

�� n ∈ ÎN , |n|1 ≤ d − 1, j ∈ û1,mü} = Ãd .Number of vectors to compute: m

(N+d−1d−1

). Independent of Λ.

Idea of the proof: Λ ≈ Lε where Lε is a rationally independentperturbation of (1, 1, . . . , 1).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 52: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityDe�nition

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

Recall that xt = x (t + ·)|[−Λmax,0).

De�nition

• We say that Σ(A,B , Λ) is approximately controllable in timeT > 0 if, [x0, xf ∈ L2((−Λmax, 0),Ãd ) and [ε > 0,\u ∈ L2((0,T ),Ãm) s.t. the solution x of Σ(A,B , Λ) withinitial condition x0 and control u satis�es ‖xT − xf ‖L2 < ε.

• We say that Σ(A,B , Λ) is exactly controllable in timeT > 0 if,[x0, xf ∈ L

2((−Λmax, 0),Ãd ), \u ∈ L2((0,T ),Ãm) s.t. the

solution x of Σ(A,B , Λ) with initial condition x0 and controlu satis�es xT = xf .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 53: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityDe�nition

Σ(A,B , Λ) : x (t ) =N∑j=1

Aj x (t − Λj ) + Bu(t ), t ≥ 0.

Recall that xt = x (t + ·)|[−Λmax,0).

De�nition

• We say that Σ(A,B , Λ) is approximately controllable in timeT > 0 if, [x0, xf ∈ L2((−Λmax, 0),Ãd ) and [ε > 0,\u ∈ L2((0,T ),Ãm) s.t. the solution x of Σ(A,B , Λ) withinitial condition x0 and control u satis�es ‖xT − xf ‖L2 < ε.• We say that Σ(A,B , Λ) is exactly controllable in timeT > 0 if,[x0, xf ∈ L

2((−Λmax, 0),Ãd ), \u ∈ L2((0,T ),Ãm) s.t. the

solution x of Σ(A,B , Λ) with initial condition x0 and controlu satis�es xT = xf .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 54: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityDe�nition

If x0 = 0, our explicit formula for x readsx (t ) =

∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

ForT ≥ 0, let E (T ) be the operator that maps a control u toxT = x (T + ·)|[−Λmax,0), the state at timeT with initial conditionzero:

E (T ) :

L2((0,T ),Ãm) → L2((−Λmax, 0),Ã

d )

u 7→∑n∈ÎN

Λ ·n≤T +·

ΞnBu(T + · − Λ · n).

• Exactly controllable ⇐⇒ E (T ) is surjective

⇐⇒ \c > 0 s.t. ‖E (T )∗x ‖L2 ≥ c‖x ‖L2 [x .

• Approximately controllable ⇐⇒ RanE (T ) is dense

⇐⇒ E (T )∗ is injective.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 55: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityDe�nition

If x0 = 0, our explicit formula for x readsx (t ) =

∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

ForT ≥ 0, let E (T ) be the operator that maps a control u toxT = x (T + ·)|[−Λmax,0), the state at timeT with initial conditionzero:

E (T ) :

L2((0,T ),Ãm) → L2((−Λmax, 0),Ã

d )

u 7→∑n∈ÎN

Λ ·n≤T +·

ΞnBu(T + · − Λ · n).

• Exactly controllable ⇐⇒ E (T ) is surjective

⇐⇒ \c > 0 s.t. ‖E (T )∗x ‖L2 ≥ c‖x ‖L2 [x .

• Approximately controllable ⇐⇒ RanE (T ) is dense

⇐⇒ E (T )∗ is injective.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 56: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityDe�nition

If x0 = 0, our explicit formula for x readsx (t ) =

∑n∈ÎNΛ ·n≤t

ΞnBu(t − Λ · n).

ForT ≥ 0, let E (T ) be the operator that maps a control u toxT = x (T + ·)|[−Λmax,0), the state at timeT with initial conditionzero:

E (T ) :

L2((0,T ),Ãm) → L2((−Λmax, 0),Ã

d )

u 7→∑n∈ÎN

Λ ·n≤T +·

ΞnBu(T + · − Λ · n).

• Exactly controllable ⇐⇒ E (T ) is surjective⇐⇒ \c > 0 s.t. ‖E (T )∗x ‖L2 ≥ c‖x ‖L2 [x .

• Approximately controllable ⇐⇒ RanE (T ) is dense⇐⇒ E (T )∗ is injective.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 57: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityCase N = d = 2, m = 1

No known criterion in the general case. We consider hereΣ(A,B , Λ) : x (t ) = A1x (t − Λ1) + A2x (t − Λ2) + Bu(t ),

with x (t ) ∈ Ã2, u(t ) ∈ Ã, and assume that Λ1 > Λ2.

(Recall that C(A,B) = (B ,AB , . . . ,Ad−1B) and that the pair (A,B) is controllable if and

only if C(A,B) has full rank.)

Theorem ([Chitour, M., Sigalotti; Preprint])

LetT ≥ 0.

• (A1,B) and (A2,B) not controllable =⇒ Σ(A,B , Λ) neitherapproximately nor exactly controllable in timeT .• RanA1 ⊂ RanB =⇒ Σ(A,B , Λ) neither approximately norexactly controllable in timeT .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 58: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityCase N = d = 2, m = 1

No known criterion in the general case. We consider hereΣ(A,B , Λ) : x (t ) = A1x (t − Λ1) + A2x (t − Λ2) + Bu(t ),

with x (t ) ∈ Ã2, u(t ) ∈ Ã, and assume that Λ1 > Λ2.(Recall that C(A,B) = (B ,AB , . . . ,Ad−1B) and that the pair (A,B) is controllable if and

only if C(A,B) has full rank.)

Theorem ([Chitour, M., Sigalotti; Preprint])

LetT ≥ 0.

• (A1,B) and (A2,B) not controllable =⇒ Σ(A,B , Λ) neitherapproximately nor exactly controllable in timeT .• RanA1 ⊂ RanB =⇒ Σ(A,B , Λ) neither approximately norexactly controllable in timeT .

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 59: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityCase N = d = 2, m = 1

Theorem ([Chitour, M., Sigalotti; Preprint])

• RanA1 1 RanB and exactly one of the pairs (A1,B) and(A2,B) is controllable =⇒ the following are equivalent:• Σ(A,B , Λ) is exactly controllable in timeT .• Σ(A,B , Λ) is approximately controllable in timeT .• T ≥ 2Λ1.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 60: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityCase N = d = 2, m = 1

Theorem ([Chitour, M., Sigalotti; Preprint])

• If (A1,B) and (A2,B) are controllable, let B⊥ ∈ Ã2 be theunique vector such that det(B ,B⊥) = 1 and BTB⊥ = 0. Set

β =detC(A1,B)detC(A2,B)

, α = det(B (A1 − βA2)B

⊥).

Let S ⊂ Ã be the set of all possible complex values of theexpression β + α1−

Λ2Λ1 .

• Σ(A,B , Λ) is approximately controllable in timeT if and onlyifT ≥ 2Λ1 and 0 < S.

• Σ(A,B , Λ) is exactly controllable in timeT if and only ifT ≥ 2Λ1 and 0 < S.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 61: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityCase N = d = 2, m = 1

S ={β + α

1−Λ2Λ1

}.

β

|α |1−

Λ2Λ1

Λ2Λ1∈ Ñ

β

|α |1−

Λ2Λ1

Λ2Λ1< Ñ

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 62: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

The non-trivial case is when (A1,B) and (A2,B) are controllable.One can reduce to the normal form

x (t ) =

(α β

0 0

)x (t − 1) +

(0 10 0

)x (t − L) +

(01

)u(t )

with L = Λ2Λ1∈ (0, 1). Moreover, approximately/exactly

controllable in timeT ⇐⇒ in time 2.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 63: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

Approximate controllability: injectivity of E (2)∗. Reduction to�nite dimension and state augmentation when L ∈ Ñ. WhenL < Ñ, if E (2)∗x = 0, then x1 satis�es

x1(t ) = −1

β

(αχ(−1,−L)(t ) + χ(−L,0)(t )

)x1 ◦ϕ(t )

(χ : characteristic functions; ϕ: translation by L modulo 1).

Letting y (t ) = eγt x1(t ), where eγ(1−L) = −β , we gety (t ) =

(αe−γχ(−1,−L)(t ) + χ(−L,0)(t )

)y ◦ϕ(t ). (Y)

• 0 ∈ S ⇐⇒ αe−γ = 1 ⇐⇒ y = y ◦ϕ ⇐⇒ y is constant(ergodicity of ϕ). One gets non-trivial elements in KerE (2)∗.• Otherwise, we have(

1 − |αe−γ |2) w 0

L−1|y (t )|2 dt = 0.

Ok if |αe−γ | , 1. Otherwise, go back to (Y) and expandy |(L−1,0) in Fourier series to obtain that y vanishes.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 64: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

Approximate controllability: injectivity of E (2)∗. Reduction to�nite dimension and state augmentation when L ∈ Ñ. WhenL < Ñ, if E (2)∗x = 0, then x1 satis�es

x1(t ) = −1

β

(αχ(−1,−L)(t ) + χ(−L,0)(t )

)x1 ◦ϕ(t )

(χ : characteristic functions; ϕ: translation by L modulo 1).Letting y (t ) = eγt x1(t ), where eγ(1−L) = −β , we get

y (t ) =(αe−γχ(−1,−L)(t ) + χ(−L,0)(t )

)y ◦ϕ(t ). (Y)

• 0 ∈ S ⇐⇒ αe−γ = 1 ⇐⇒ y = y ◦ϕ ⇐⇒ y is constant(ergodicity of ϕ). One gets non-trivial elements in KerE (2)∗.• Otherwise, we have(

1 − |αe−γ |2) w 0

L−1|y (t )|2 dt = 0.

Ok if |αe−γ | , 1. Otherwise, go back to (Y) and expandy |(L−1,0) in Fourier series to obtain that y vanishes.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 65: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

Approximate controllability: injectivity of E (2)∗. Reduction to�nite dimension and state augmentation when L ∈ Ñ. WhenL < Ñ, if E (2)∗x = 0, then x1 satis�es

x1(t ) = −1

β

(αχ(−1,−L)(t ) + χ(−L,0)(t )

)x1 ◦ϕ(t )

(χ : characteristic functions; ϕ: translation by L modulo 1).Letting y (t ) = eγt x1(t ), where eγ(1−L) = −β , we get

y (t ) =(αe−γχ(−1,−L)(t ) + χ(−L,0)(t )

)y ◦ϕ(t ). (Y)

• 0 ∈ S ⇐⇒ αe−γ = 1 ⇐⇒ y = y ◦ϕ ⇐⇒ y is constant(ergodicity of ϕ). One gets non-trivial elements in KerE (2)∗.• Otherwise, we have(

1 − |αe−γ |2) w 0

L−1|y (t )|2 dt = 0.

Ok if |αe−γ | , 1. Otherwise, go back to (Y) and expandy |(L−1,0) in Fourier series to obtain that y vanishes.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 66: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

Exact controllability: show that \c > 0 s.t. [x ,‖E (2)∗x ‖L2 ≥ c‖x ‖L2 . (X)

• If L ∈ Ñ, equivalent to approximate controllability andcharacterized by detM , 0 for some matrix M , withc = |M −1 |−12 .

• If L < Ñ and 0 < S, approximate L by rationals and obtain anuniform upper bound on |M −1 |2 along this approximation.• If L < Ñ and 0 ∈ S, one can construct a sequence of points inS tending to 0 and obtain a sequence (xn )n∈Î with L2 norm 1such that E (2)∗xn → 0, contradicting (X).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 67: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityIdeas of the proof

Exact controllability: show that \c > 0 s.t. [x ,‖E (2)∗x ‖L2 ≥ c‖x ‖L2 . (X)

• If L ∈ Ñ, equivalent to approximate controllability andcharacterized by detM , 0 for some matrix M , withc = |M −1 |−12 .• If L < Ñ and 0 < S, approximate L by rationals and obtain anuniform upper bound on |M −1 |2 along this approximation.• If L < Ñ and 0 ∈ S, one can construct a sequence of points inS tending to 0 and obtain a sequence (xn )n∈Î with L2 norm 1such that E (2)∗xn → 0, contradicting (X).

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 68: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityControllability to constants

What if we want to go from x0 = 0 to a constant state in timeT ?

Theorem ([Chitour, M., Sigalotti; Preprint])

Σ(A,B , Λ) is approximately controllable if and only if it isapproximately controllable to constants (i.e., every constant statecan be approximately reached by a suitable control from a zeroinitial condition).

• Main idea: if Σ(A,B , Λ) is approximately controllable toconstants, using a kind of commutation between E (T ) andintegration (inspired by the �nite-dimensional ideas of[Gohberg, Shalom; 1989]), one can approximate anypolynomial.• Does it also hold for exact controllability? True forN = d = 2 and m = 1, unknown in general.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 69: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityControllability to constants

What if we want to go from x0 = 0 to a constant state in timeT ?

Theorem ([Chitour, M., Sigalotti; Preprint])

Σ(A,B , Λ) is approximately controllable if and only if it isapproximately controllable to constants (i.e., every constant statecan be approximately reached by a suitable control from a zeroinitial condition).

• Main idea: if Σ(A,B , Λ) is approximately controllable toconstants, using a kind of commutation between E (T ) andintegration (inspired by the �nite-dimensional ideas of[Gohberg, Shalom; 1989]), one can approximate anypolynomial.

• Does it also hold for exact controllability? True forN = d = 2 and m = 1, unknown in general.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 70: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Approximate and exact controllabilityControllability to constants

What if we want to go from x0 = 0 to a constant state in timeT ?

Theorem ([Chitour, M., Sigalotti; Preprint])

Σ(A,B , Λ) is approximately controllable if and only if it isapproximately controllable to constants (i.e., every constant statecan be approximately reached by a suitable control from a zeroinitial condition).

• Main idea: if Σ(A,B , Λ) is approximately controllable toconstants, using a kind of commutation between E (T ) andintegration (inspired by the �nite-dimensional ideas of[Gohberg, Shalom; 1989]), one can approximate anypolynomial.• Does it also hold for exact controllability? True forN = d = 2 and m = 1, unknown in general.

Contrôlabilité d’équations aux di�érences Guilherme Mazanti

Page 71: Contrôlabilité d’équations aux di˙érencesmazanti/...Contrôlabilité d’équations aux di˙érences Guilherme Mazanti CRAN, Nancy 15 octobre 2018 Laboratoire de Mathématiques

Introduction Explicit solution Stability Relative controllability Approximate and exact controllability

Contrôlabilité d’équations aux di�érences Guilherme Mazanti