Sur la contrôlabilité de l’équation de la chaleur avec potentiel
Michel Duprez
Institut de Mathématiques de Marseille
Séminaire Analyse Appliquée (AA)
Marseille, I2M
Problem
How to act on the temperature from a given region ?
Controllability of the heat equation
Let beT >0,ω⊂Ω⊂RN and consider the system ( ∂
ty= ∆y+1ωu inQT := Ω×(0, T), y= 0 on ΣT :=∂Ω×(0, T), y(·,0) =y0 in Ω.
Such a system is said :
approximatively controllableat timeT if
∀y0, yT∈L2(Ω), ε >0∃u∈L2(QT) s.t.ky(T)−yTkL2(Ω)6ε, exactly controllableat timeT if
∀y0, yT ∈L2(Ω)∃u∈L2(QT) s.t.y(T) =yT, null controllableat timeT if
∀y0∈L2(Ω)∃u∈L2(QT) s.t. y(T) = 0.
Remark :
•The system is not exactly contr.
•Null contr.⇒approx. contr.
Controllability of the heat equation
Consider the system ( ∂
ty= ∆y+1ωu inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
Approximate controllability å 58’ Mizohata Null controllabilityN = 1
å 74’ Fattorini and Russell Null controllabilityN >1
å 95’ Lebeau and Robbiano å 95’ Fursikov and Imanuvilov
Controllability of a parabolic equation
Consider the parabolic equation ( ∂
ty= div(d∇y) +g· ∇y+ay+1ωu inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
(G)
wherea∈L∞(QT),g∈L∞(0, T;W∞1(Ω))N and for a constantd0>0
dij∈W∞1(QT), dij=djil inQT,
N
P
i,j=1
dijξiξj>d0|ξ|2 inQT, ∀ξ∈RN.
å System (G) is null controllable at any timeT.
Problem
Let beT >0,ω⊂Ω⊂RN and consider the system ( ∂
ty= ∆y+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
Ifω⊂⊂Ω,∂x1(1ωu) can be seen as a control with average equal to zero
Preliminary study for the Fokker-Planck control problem ( ∂
tρ=ε2∆ρ−div(ρ(v+1ωu)), ρ(·,0) =ρ0,
Supp(u)⊂ω
å Change of variables : div→∂x1
Null controllability when∂Ω∩∂ω6=∅
å Benabdallah, Cristofol, Gaitan, De Teresa, 2014 Null controllability in general
å Duprez, Lissy, 2016
Problem
General system
∂ty= div(d∇y) +g· ∇y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω.
(G)
Toy model
∂ty= ∆y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω.
(S)
We will see thataplay a critical role for the controllability of these system ! ! !
Duality
Consider the system ( ∂
ty= ∆y+ay+1ωu inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
(S)
Proposition
(i) System (S) isnull controllableat timeT iff there existsCobs>0 s.t. for allϕ0∈L2(Ω) the solutionϕ∈W(0, T;H01(Ω), H−1(Ω)) to
( −∂
tϕ= ∆ϕ+aϕ inQT,
ϕ= 0 on ΣT,
ϕ(·, T) =ϕ0 in Ω
(D)
satisfies theinequality of observability
kϕ(·,0)k2L2(Ω)6Cobs
Z T 0
Z
ω
ϕ2.
(ii) System (S) isapproximately controllableat timeT iff for all ϕ0∈L2(Ω) the solutionϕsatisfies
ϕ= 0 inω×(0, T) ⇒ϕ= 0 in Ω×(0, T).
Duality
Consider the system ( ∂
ty= ∆y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
(S)
Proposition
(i) System (S) isnull controllableat timeT iff there existsCobs>0 s.t. for allϕ0∈L2(Ω) the solutionϕ∈W(0, T;H01(Ω), H−1(Ω)) to
( −∂
tϕ= ∆ϕ+aϕ inQT,
ϕ= 0 on ΣT,
ϕ(·, T) =ϕ0 in Ω
(D)
satisfies theinequality of observability
kϕ(·,0)k2L2(Ω)6Cobs
Z T 0
Z
ω
∂x1ϕ2.
(ii) System (S) isapproximately controllableat timeT iff for all ϕ0∈L2(Ω) the solutionϕsatisfies
∂x1ϕ= 0 inω×(0, T) ⇒ϕ= 0 in Ω×(0, T).
Proof
Let
St: L2(Ω) → L2(Ω)
y0 7→ y(t;y0,0) and Lt: L2(QT) → L2(Ω) u 7→ y(t; 0, u).
(i) System (S) is null controllable at timeT iff
∀y0∈L2(Ω), ∃u∈L2(QT) s.t.
STy0=−LTu.
⇔ ImST⊆ImLT.
⇔
∃C >0 : ∀ϕ0 ∈L2(Ω), kST∗ϕ0k2L2(Ω)6CkL∗Tϕ0k2L2(QT). (ii) System (S) is approximatively controllable at timeT iff
∀y0, yT ∈L2(Ω), ε >0, ∃u∈L2(QT) s.t.
kLTu+STy0−yTkL2(Ω)6ε.
⇔ LT(L2(QT)) =L2(Ω).
⇔ ker (L∗T) ={0}
⇔ ∀ϕ0∈L2(Ω) : L∗Tϕ0 ⇒ϕ≡0.
Proof
Lety0, ϕ0∈L2(Ω) andu∈L2(QT) : hSTy0, ϕ0iL2(Ω) = hy(T;y0,0), ϕ(T)iL2(Ω)
= Z T
0
h∂ty(s;y0,0), ϕ(s)iH−1,H01ds
+ Z T
0
h∂tϕ(s), y(s;y0,0)iH−1,H1
0ds+hy0, ϕ(0)iL2(Ω)
= hy0, ϕ(0)iL2(Ω)
and
hLTu, ϕ0iL2(Ω) = hy(T; 0, u), ϕ(T)iL2(Ω)
= Z T
0
n
h∂ty(s; 0, u), ϕ(s)iH−1,H10+h∂tϕ(s), y(s; 0, u)iH−1,H01
o ds
= h∂x1u, ϕiL2(0,T;H−1(Ω)),L2(0,T;H01(Ω))=−hu, ∂x1ϕiL2(QT). Thus
ST∗ : L2(Ω) → L2(Ω)
ϕ0 7→ ϕ(·,0) and L∗T : L2(Ω) → L2(QT).
ϕ0 7→ −∂x1ϕ
Boundary condition
Consider
∂ty= ∆y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω
(S)
Theorem
If∂Ω∩∂ω6=∅, then System (S) isnull controllableat timeT.
Suppose that∂Ω∩∂ω∩(Ox1)⊥6=∅. If (0, R)×eω⊂ω, let (y, v) solution of ( ∂
ty= ∆y+ay+1(0,R)טωv inQT,
y= 0 on ΣT,
y(·,0) =y0, y(·, T) = 0 in Ω.
We take
u:=
Z x1 R
v.
Or equivalently Z
Ω
ϕ(·,0)26C Z
ω×(0,T)
ϕ26C Z
ω×(0,T)
(∂x1ϕ)2.
Boundary condition : General parabolic equation
Consider the parabolic system ( ∂
ty= div(d∇y) +g· ∇y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
(G)
Theorem (Benabdallah, Cristofol, Gaitan, De Teresa, 2014)
Assume that∂Ω∩∂ω6=∅. Then System (G) isnull controllableon the time interval (0, T).
General parabolic operator
Consider the system
∂ty= ∆y+p(x)∂xu+q(x)u in (0, π)×(0, T), y(0,·) =y(π,·) = 0 on (0, T),
y(·,0) =y0 in (0, π),
Supp(u)⊂ω.
(S’)
Theorem (Duprez, 2016) Suppose that
Supp(p, q)∩ω6=∅.
Then System (S’) isnull controllableon the time interval (0, T).
Constant coefficient(s)
Consider the system
∂ty= ∆y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω.
(S)
Theorem ( D., Lissy, 2016)
Ifais constant, then System (S) isnull controllableat timeT. Sketch of proof :
Up to the initial condition and the boundary condition,∂x1ϕsatisfies the same system asϕ:
( −∂
tϕ= ∆ϕ+aϕ inQT,
ϕ= 0 on ΣT,
ϕ(·, T) =ϕ0 in Ω.
Using the Carleman estimates, for some weightsρ1, ρ2, ρ3, Z
Ω
ϕ(·,0)26C Z
QT
ρ1ϕ26C Z
QT
ρ2(∂x1ϕ)26C Z
qT
ρ3(∂x1ϕ)2.
Constant coefficients : General parabolic equation
Consider the system ( ∂
ty= div(d∇y) +g· ∇y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
(G)
Theorem ( D., Lissy, 2016)
Suppose thatd, gandaare constant in space and in time.
Then System (G) isnull controllableon the time interval (0, T).
Algebraic resolvability
Let us recall the system
∂ty= ∆y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂⊂ω.
(S)
Theorem (D., Lissy, 2016, submitted)
Leta∈ C1(QT). If there exists (x0, t0)∈ω×(0, T) such that (∂x1a)(x0, t0)6= 0,
then System (S) isnull controllable(hence approximately controllable) on the time interval (0, T).
Proof : first approach
Fictitious control method :
Letωb⊂⊂ω. Consider (by,bu) a solution to
∂tby= ∆by+ayb+ ˆu inQT,
yb= 0 on ΣT,
y(·) =b y0, y(·, T) = 0 in Ω, Supp(u)b ⊂⊂bω×(0, T).
If we find a solution to
∂tz= ∆z+az+∂x1(v) +ub inQT, Supp(z, v)⊂⊂ω×(0, T), (S1) then(by−z,−v)will be a solution of the initial problem.
Proof : first approach
Algebraic problem : Rewrite (S1) as
L(z, v) =bu where
L(z, v) =∂tz−∆z−az−∂x1(v).
Let us search a differential operatorMsuch that L ◦ M=Id (S2) Thus (z, v) :=M(bu) will be a solution to (S1).
Remark :buhave to be regular enough.
We will solve the formal adjoint to (S2) :
M∗◦ L∗ψ=ψ (S3) where
L∗(ψ) = L∗1ψ
L∗2ψ
=
−∂tψ−∆ψ−aψ
−∂x1ψ
.
Proof : first approach
Resolution of the algebraic problem :
We recall the expression toL∗: L∗(ψ) =
L∗1ψ L∗2ψ
=
−∂tψ−∆ψ−aψ
−∂x1ψ
.
The commutator ofL∗1 andL∗2 is :
[L∗1:L∗2]ψ=∂x1(a)ψ.
So we have
M∗1◦ L∗1+M∗2◦ L∗2=Id, where
M∗1=∂x1(a)−1∂x1 andM∗2=∂x1(a)−1[∂t+ ∆ +a].
Proof : second approach
Resolvability + Inequality of observability :
Letϕthe solution to the dual system ( −∂
tϕ= ∆ϕ+aϕ inQT,
ϕ= 0 on ΣT,
ϕ(·, T) =ϕ0 in Ω.
Using the algebraic resolvability :
ϕ=M∗1◦(−∂t−∆−a)ϕ+M2∗◦(−∂x1)ϕ=M∗2◦(−∂x1)ϕ Hence
kϕ(·,0)k2L2(Ω)6C Z
qT
ϕ2=C Z
qT
(M∗2◦∂x1ϕ)2 Thus we have a solution to the control problem :
( ∂
ty= ∆y+ay+∂x1(M2u) inQT,
y= 0 on ΣT,
y(·,0) =y0, y(·, T) = 0 in Ω.
Proof : first approach
Regular control with the fictitious control method :
Letωb⊂⊂ωand (y,bbu),y,θ,ηsuch that
∂tyb= ∆by+ayb+1ωˆuˆ inQT, by= 0 on ΣT, by(·,0) =y0, by(·, T) = 0 in Ω
∂ty= ∆y+ay inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω
( Supp(θ)⊂ω, θ= 1 inbω, 0≤θ≤1
and
( η= 1 in [0, T /3], η= 0 in [2T /3, T], 0≤η≤1.
Theny:= (1−θ)by+ηθyis solution to
∂ty= ∆y+ay+1ωu inQT,
y= 0 on ΣT,
y(·,0) =y0, y(·, T) = 0 in Ω.
withu:= ∆θby+ 2∇θ· ∇by+∂t(η)θy−η∆θy−2η∇θ· ∇y.
We remark thatyis more regular thanby.
General parabolic equation
Consider the system ( ∂
ty= div(d∇y) +g· ∇y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
(G)
Theorem (D., Lissy, 2016, submitted)
Letdkli , gkij∈ CN2+3(ωT) andaij∈ CN2+2(ωT) for everyi, j∈ {1,2}and k, l∈ {1, ..., N}. Assume∃ωT ⊂qT s.t.
(
eais not an element of theCt,x0 2,...,xN(ωT)-module 1,eg
2, ...,eg
N, d22, ..., dN N
C0t,x
2,...,xN(ωT), where
egi:=gi−
N
P
j=1
∂xjdij,
ea:=−a+ div(g).
Then System (G) isnull controllableon the time interval (0, T).
Sketch of proof
Suppose thatN= 2 ( ∂
ty=∂x1x1(d1y) +∂x2x2(d2y) +∂x1(ge1y) +∂x2(ge2y) +aye +∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω.
Algebraic resolution : Expression toL∗: L∗1ψ
L∗2ψ
=
∂tψ−d1∂x1x1(ψ)−d2∂x2x2(ψ) +ge1∂x1(ψ) +ge2∂x2(ψ) +eaψ
−∂x1ψ
.
Step 1 : The derivatives inx1
L∗3ψ:=L∗1ψ−(d1∂x1+bg1)L
∗
2ψ=∂tψ−d2∂x2x2(ψ) +ge2∂x2(ψ) +eaψ Step 2 : The derivative in time
L∗4ψ:= [L∗3 :L2∗]ψ=−∂x1(d2)∂x2x2(ψ) +∂x1(ge2)∂x2(ψ) +∂x1(ea)ψ
Sketch of proof
Step 3 : The other terms
•If|∂x1(d2)| ≥δ1, then
L∗5ψ:= [∂x1(d2)−1L∗4 :L∗2]ψ=∂x1
∂x1(ge2)
∂x1(d2)
∂x2(ψ) +∂x1
∂x1(ea)
∂x1(d2)
ψ
•If|∂x1
∂x1(ge2)
∂x1(d2)
| ≥δ2, then
L∗6ψ:= [∂x1
∂x1(ge2)
∂x1(d2) −1
L∗5 :L∗2]ψ=∂x1
∂x1
∂x1(
ea)
∂x1(d2)
∂x1
∂x1( ge2)
∂x1(d2)
ψ
Sketch of proof
Step 4 : Conclusion
∂x1
∂x1
∂x1(
ea)
∂x1(d2 )
∂x1
∂x1(ge2 )
∂x1(d2 )
= 0 ⇔
∂x1
∂x1(
ea)
∂x1(d2 )
∂x1
∂x1(ge2 )
∂x1(d2 )
=λ1(t, x2)
⇔ ∂x1
∂x1(
ea−λ1ge2)
∂x1(d2)
= 0
⇔ ∂x1(ea−λ1ge2)
∂x1(d2) =λ2(t, x2)
⇔ ∂x1(ea−λ1ge2−λ2d2) = 0
⇔ ea=λ1ge2+λ2d2+λ3
Summary
Let us recall the system
∂ty= ∆y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂⊂ω.
(S)
aconstant
å Null controllable
There exists (x0, t0)∈ω×(0, T) s.t.∂x1a6= 0 å Null controllable
Is System (S) always null controllable ?
Example of non controllability
Let us recall the system
∂ty= ∆y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω.
(S)
Theorem (D., Lissy, 2016, submitted)
Ifω⊂⊂Ω, then there existsa∈ C∞(Ω) such that System (S) isnot approximately controllable(hence not null controllable) at timeT.
Fattorini criterion
Let us recall the system
∂ty= ∆y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂⊂ω.
(S)
Theorem (Fattorini criterion)
System (S) isapproximately controllableon the time interval (0, T), if and only if for everys∈Cand everyϕ∈D(∆), we have
−∆ϕ−aϕ=sϕ in Ω
∂x1ϕ= 0 inω
⇒ϕ= 0.
å Olive, 2014
Proof
Suppose thatN= 1, Ω := (0, π) andω1:= (2π/5,3π/5).
Consider
a:=−∆ϕ−ϕ
ϕ .
Thus
−∆ϕ−aϕ=ϕ in Ω,
∂x1ϕ= 0 inω, ϕ6= 0.
Using the Fattorini Criterion, the system isnot approximately controllableon the time interval (0, T).
Letϕ∈ C2([0, π]) satisfying
ϕ(x) = sin(x) ∀x∈[0, π/5]∪[4π/5, π], ϕ(x) = 1 ∀x∈[2π/5,3π/5], ϕ > δ >0 in [π/5,4π/5].
0 1 2 3
0 0.5 1
sin(x) 1
Geometrical condition
Let us recall the system
∂ty= ∆y+ay+∂x1(u) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂ω.
(S)
Theorem (D., Lissy, 2016, submitted) There existsa∈ C∞(Ω) such that :
(i) There existsω⊂⊂Ω such that, for allT >0, System (S) isnull controllable(then approximatively controllable) at timeT. (ii) There existsω⊂⊂Ω such that, for allT >0, System (S) isnot
approximatively controllable(then not null controllable) at time T.
Conclusions and perspectives
Let us recall the system
∂ty= ∆y+ay+∂x1(1ωu) inQT,
y= 0 on ΣT,
y(·,0) =y0 in Ω,
Supp(u)⊂⊂ω.
(S)
Conclusions :
System (S) isnull controllableunder one of the following condition : å ∂Ω∩∂ω6=∅
å constant coefficients
å ∂x1a(x0, t0)6= 0 for a (t0, x0)∈qT
There existsa∈ C∞(QT) s.t. System (S) isnot approximately controllable
Perspectives :
Find a general condition...
Study the Fokker-Planck equation