Null controllability of the Grushin equation : non-rectangular control region
M. Duprez
1, A. Koenig
2,
1Laboratoire Jacques-Louis Lions, Paris
2Laboratoire Jean-Alexandre Dieudonné, Nice
Séminaire d’analyse
Institut de Mathématiques de Bordeaux 18/10/18
Outline
1 Controllability of the heat equation
2 Controllability of the Grushin equation
3 Comments and open problems
Notions of controllability
Problematic
We searchu, calledcontrol, such that the solutionfto system
∂tf=A(f) +B(u) f(0) =f0
satisfies :
ε
f0
f1 f(T)
• fnear a given target at timeT
∀ε >0, f0, f1,∃u s.t. d(f(T), f1)6ε.
â Approximate controllability
• freach a target at timeT
∀f0, f1, ∃us.t.f(T) =f1. â Exact controllability
• freach a target at timeT
∀f0, f1, ∃us.t.f(T) = 0.
â Null controllability
1
Internal null controllability of the heat equation
LetΩbe an open bounded set ofRn, ωan non-empty open set ofΩ andT >0.
Consider the heat equation ( ∂
tf−∆f=1ωu, f|∂Ω= 0,
f(0) =f0.
(1)
ω Ω
Theorem (Internal null controllability of the heat equation)
The heat equation(1)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×ω)such that solutionfto (1) satisfiesf(T) = 0.
→Fattorini-Russell 71’ (dim 1) : moment method
→Fursikov-Imanuvilov / Lebeau-Robbiano 95’ : Carleman inequality / spectral inequality
Remark
• No minimal timeof controllability
• No geometrical conditiononω
2
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
Ω γ
Remark
• Equivalence between internal and boundarynull controllability
3
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
Ω˜ Ω
γ
Remark
• Equivalence between internal and boundarynull controllability
3
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
ω Ω˜ Ω
γ
Remark
• Equivalence between internal and boundarynull controllability
3
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
ω Ω˜ Ω
γ
Remark
• Equivalence between internal and boundarynull controllability
3
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
ω Ω˜ Ω
γ
Remark
• Equivalence between internal and boundarynull controllability
3
Boundary null controllability of the heat equation
LetΩbe an open bounded set ofRn,γan non-empty open set of∂ΩandT >0.
( ∂
tf−∆f= 0, f|∂Ω=1γu, f(0) =f0.
(2)
Corollary (Boundary null controllability of the heat equation)
The heat equation(2)isnull controllableat timeT,i.e.for each initial condition f0∈L2(Ω), there exists a controlu∈L2([0, T]×γ)such that solutionfto (2) satisfiesf(T) = 0.
Sketch of proof: the fictitious control method
• ExtensionΩ˜ofΩthroughγ
• ω⊂Ω˜
• f˜: = solution of the int. contr. problem onΩ∪Ω˜
• u:= ˜f|γ
ω Ω˜ Ω
γ
Remark
• Equivalence between internal and boundarynull controllability
3
Duality
Denote byf(t;f0,u)the solution to
∂tf−∆f=1ωu, f|∂Ω= 0, f(0) =f0.
(1)
LetQT:= Ω×(0, T)and FT : L2(Ω) → L2(Ω),
f0 7→ f(T;f0,0)
GT : L2(QT) → L2(Ω), u 7→ f(T; 0,u) One has :
Null controllability of (1) ⇔ ∀f0,∃u:FT(f0) +GT(u) = 0
⇔ Im(FT)⊂Im(GT)
⇔ kFT∗ϕTk6CkG∗TϕTk,∀ϕT Approximate controllability of (1) ⇔ Im(GT) =L2(Ω)
⇔ G∗TϕT = 0⇒ϕT= 0 ,∀ϕT
4
Duality
∂tf−∆f=1ωu, f|∂Ω= 0,
f(0) =f0.
(1)
−∂tϕ= ∆ϕ inQT, ϕ= 0 onΣT, ϕ(T) =ϕT inΩ.
Proposition (see Coron’s book 2007 or Tucsnak-Weiss’s book 2009)
1 System(1)is null controllableon(0, T), if and only if there existsCobs>0 such that for allϕT ∈L2(Ω)
kϕ(0)kL2(Ω)6Cobs
Z T 0
kϕkL2(ω)dt
withϕthe solution to the dual system.
2 System(1)isapprox. controllableat timeT, if and only if for allϕT∈L2(Ω) the solution to the dual system satisfies
ϕ= 0inω×(0, T) ⇒ϕ= 0inQT.
5
Approximate controllability of the heat equation
∂tf−∆f=1ωu inQT, f|∂Ω= 0 onΣT, f(0) =f0 inΩ.
(1)
−∂tϕ= ∆ϕ inQT, ϕ= 0 onΣT, ϕ(T) =ϕT inΩ.
Since (1) is null controllable,
kϕ(0)kL2(Ω)6Cobs
Z T 0
kϕkL2(ω)dt
Hence
ϕ= 0inω×(0, T) ⇒ ϕ(0) = 0inΩ
⇒ ϕ= 0inΩ×(0, T)
Corollary (Approximate controllability of the heat equation) The heat equation(1)isapprox. controllableat timeT, i.e. for each
f0, f1∈L2(Ω), there exists a controlu∈L2([0, T]×ω)such that solutionfto (1)satisfieskf(T)−f1k6ε.
6
The heat equation
Conclusion
• Non exact controllability
âRegularizing effect of the heat equation
• Null controllability
âEquivalence of int. and bound. null contr. thanks to fictitious control method
• Approximate controllability
âCome from the null controllability thanks to the duality
Question
How a modification of−∆can influx these notions of controllability ?
7
The Grushin equation
Consider ( (∂
t−∂x2−x2∂y2)f(t, x, y) =1ωu t∈[0, T],(x, y)∈Ω := (−1,1)×(0,1), f(t, x, y) = 0 t∈[0, T],(x, y)∈∂Ω,
f(0, x, y) =f0(x, y) (x, y)∈Ω,
x y
a a)
x y
a b)
x y
c)
a) minimal time of internal null controllabilityT∗=a2/2 âBeauchard-Miller-Morancey 2015
b) minimal time of internal null controllabilityT∗> a2/2 âBeauchard-Cannarsa-Guglielmi 2014
minimal time of boundary null controllabilityT∗=a2/2 âBeauchard-Dardé-Ervedoza 2018
c) No internal null controllability âKoenig 2017
8
Positive result
−a
x y
ω0 γ
Theorem (D.-Koenig)
Assume that there existε >0andγ∈C0([0,1],Ω)withγ(0)∈(−1,1)× {0}and γ(1)∈(−1,1)× {π}such that
ω0:={z∈Ω,distance(z,Range(γ))< ε} ⊂ω, then the Grushin equation isnull-controllablein any timeT>a2/2, where a:= sup
(x,y)∈Ω\ω0
{|x|:∃x0∈(−1,1),|x|<|x0|,sgn(x) = sgn(x0),(x0, y)∈ω0}.
9
Positive result : sketch of proof
Proof :
−a a
x y
ω0
γ ωleft
ωright
Consider that controlled solutions [Beauchard Darde Ervedoza 2018]
( (∂
t−∂x2−x2∂y2)fleft=1ωleftuleft on[0, T]×Ω
(∂t−∂x2−x2∂y2)fright=1ωrighturight on[0, T]×Ω
fleft(0) =fright(0) =f0, fleft(T) =fright(T) = 0 onΩ There exists a functionθ∈C∞( ¯Ω)such that
θ= 0onωleft\ω0, θ= 1onωright\ω0,supp(∇θ)⊂ω0. We remark thatf:=θfleft+ (1−θ)fright,solves
(∂t−∂x2−x2∂y2)f= 1ω0u on[0, T]×Ω, f(0) =f0, f(T) = 0 onΩ,
whereu:=θ1ωleftuleft+ (1−θ)1ωrighturight+ (fright−fleft)(∂2x+x2∂y2)θ
+2∂x(fright−fleft)∂xθ+ 2x2∂y(fright−fleft)∂yθ.
10
Positive result : sketch of proof
Proof :
−a a
x y
ω0
γ ωleft
ωright
Consider that controlled solutions [Beauchard Darde Ervedoza 2018]
( (∂
t−∂x2−x2∂y2)fleft=1ωleftuleft on[0, T]×Ω
(∂t−∂x2−x2∂y2)fright=1ωrighturight on[0, T]×Ω
fleft(0) =fright(0) =f0, fleft(T) =fright(T) = 0 onΩ There exists a functionθ∈C∞( ¯Ω)such that
θ= 0onωleft\ω0, θ= 1onωright\ω0,supp(∇θ)⊂ω0. We remark thatf:=θfleft+ (1−θ)fright,solves
(∂t−∂x2−x2∂y2)f= 1ω0u on[0, T]×Ω, f(0) =f0, f(T) = 0 onΩ,
whereu:=θ1ωleftuleft+ (1−θ)1ωrighturight+ (fright−fleft)(∂2x+x2∂y2)θ
+2∂x(fright−fleft)∂xθ+ 2x2∂y(fright−fleft)∂yθ.
10
Positive result : sketch of proof
Proof :
−a a
x y
ω0
γ ωleft
ωright
Consider that controlled solutions [Beauchard Darde Ervedoza 2018]
( (∂
t−∂x2−x2∂y2)fleft=1ωleftuleft on[0, T]×Ω
(∂t−∂x2−x2∂y2)fright=1ωrighturight on[0, T]×Ω
fleft(0) =fright(0) =f0, fleft(T) =fright(T) = 0 onΩ There exists a functionθ∈C∞( ¯Ω)such that
θ= 0onωleft\ω0, θ= 1onωright\ω0,supp(∇θ)⊂ω0. We remark thatf:=θfleft+ (1−θ)fright,solves
(∂t−∂x2−x2∂y2)f= 1ω0u on[0, T]×Ω, f(0) =f0, f(T) = 0 onΩ,
whereu:=θ1ωleftuleft+ (1−θ)1ωrighturight+ (fright−fleft)(∂2x+x2∂y2)θ
+2∂x(fright−fleft)∂xθ+ 2x2∂y(fright−fleft)∂yθ.
10
Negative result
ω
y0
−a a
x y
Theorem (D.-Koenig)
If for somey0∈(0, π)anda >0one has
{(x, y0),−a < x < a} ∩ω=∅,
then the Grushin equation isnot null controllablein timeT<a2/2.
Remark :Since the Grushin equation is not null-controllable whenωis the complement of a horizontal strip [Koenig 2017], our Assumption is quasi optimal.
11
Observability
Proposition (see Coron’s book 2007 or Tucsnak-Weiss’s book 2009)
The Grushin equation isnull controllabilityif, and only if, there existsC >0such that for allf0inL2(Ω), the solutionfto
( (∂
t−∂x2−x2∂y2)f(t, x, y) = 0 t∈[0, T],(x, y)∈Ω, f(t, x, y) = 0 t∈[0, T],(x, y)∈∂Ω, f(0, x, y) =f0(x, y) (x, y)∈Ω,
satisfies
Z
Ω
|f(T, x, y)|2dxdy≤C Z
[0,T]×ω
|f(t, x, y)|2dtdxdy.
12
Negative result : sketch of proof
Assume thatΩ =R×T. Forn >0,
e−nx2/2is the first eigenfunction of −∂2x+ (nx)2 and with associated eigenvaluen.
Then
e−nx2/2einyis an eigenfunction of −∂x2−x2∂y2 and with eigenvaluen.
For the functions
f(t, x, y) =X
ane−nt+iny−nx2/2, the observability inequality reads :
X 2π n
1/2
|an|2e−2nT ≤C Z
[0,T]×ω
Xane−nt+iny−nx2/2
2
dtdxdy
13
Negative result : sketch of proof
Assume thatΩ =R×T. Forn >0,
e−nx2/2is the first eigenfunction of −∂2x+ (nx)2 and with associated eigenvaluen.
Then
e−nx2/2einyis an eigenfunction of −∂x2−x2∂y2 and with eigenvaluen.
For the functions
f(t, x, y) =X
ane−nt+iny−nx2/2, the observability inequality reads :
X 2π n
1/2
|an|2e−2nT ≤C Z
[0,T]×ω
Xane−nt+iny−nx2/2
2
dtdxdy
13
Negative result : sketch of proof
Consider the change of variablezx(t, y) := e−t+iy−x2/2 and denoteDx:={e−x2/2−t+iy,0< t < T,(x, y)∈ω}:
Z
[0,T]×ω
Xaneiny−nt−nx2/2
2
dtdxdy= Z 1
−1
Z
Dx
X
n>0
anzn−1
2
dλ(z) dx.
ω x
|x|> a y |x|< a
−a a
Dx
Dx
14
Negative result : sketch of proof
Thus Z
D(0,e−T)
X
n>0
anzn−1
2
dλ(z)≤C
X
n>0
anzn−1 L∞(K)
.
e−a2/2 K=S1
x=−1Dx
15
Negative result : sketch of proof
Z
D(0,e−T)
X
n>0
anzn−1
2
dλ(z)≤C
X
n>0
anzn−1
2 L∞(K)
.
Letz0∈D(0, e−T)\K and
f:z∈C\z0[1,+∞)7→(z−z0)−1.
K
D(0, e−T)
z0
Proposition (Runge’s theorem)
LetKbe a connected and simply connected open subset ofC, and letfbe a holomorphic function onK. There exists a sequence(pk)of polynomials that converges uniformly on every compact subsets ofKtof.
Then, the familypkis a counter example to the inequality on entire polynomials.
16
Negative result : sketch of proof
Assume now thatΩ = (−1,1)×(0,1).
Forn >0, let us note
vnthe first eigenfunction of −∂2x+ (nx)2
with Dirichlet boundary condition on(−1,1), and with associated eigenvalueλn. Then
vn(x) sin(ny)is an eigenfunction of −∂x2−x2∂2y
with Dirichlet boundary condition on∂Ω, and with eigenvalueλn. For the functions
f(t, x, y) =X
anvn(x)e−λntsin(ny), the observability inequality reads :
X|an|2|vn|2L2e−2λnT≤C Z
[0,T]×ω
Xanvn(x)e−λntsin(ny)
2
dtdxdy Let us estimate the right hand-side of this inequality
17
Negative result : sketch of proof
•Notingω˜=ω∪ {(x,−y),(x, y)∈ω}, this implies Z
[0,T]×ω
Xanvn(x)e−λntsin(ny)
2
dtdxdy
≤C Z
[0,T]טω
Xanvn(x)einy−λnt
2
dtdxdy.
˜ ω
x y
−a0 a0
18
Negative result : sketch of proof
•Consider the change of variable
λn=n+ρn
vn(x) = e−(1−ε)nx2/2wn(x) zx(t, y) = e−t+iy−(1−ε)x2/2
Then Z
[0,T]טω
Xanvn(x)einy−λnt
2
dtdxdy
= Z
[0,T]טω
X
n>0
anwn(x)e−ρntzx(t, y)n
2
dtdxdy.
Remark (lemma)
The eigenfunctionvnof−∂x2+ (nx)2for the eigenvalue is closed to the eigenfunction of same operator onRfor the eigenvaluen,i.e.
vnclosed to n 2π
1/4
e−nx2/2 and λnclosed ton.
19
Negative result : sketch of proof
•DenotingDx={e−(1−ε)x2/2e−t+iy,0< t < T,(x, y)∈ω}, we get˜ Z
[0,T]טω
X
n>0
anwn(x)e−ρntzx(t, y)n
2
dtdxdy
= Z 1
−1
Z
Dx
X
n>0
anwn(x)e−ρntzn−1
2
dλ(z) dx.
ω˜ x
|x|> a0 y |x|< a0
−a0 a0
Dx
Dx
20
Negative result : sketch of proof
•Sinceρn,wnare closed to0,1, Z 1
−1
X
n>N
anwn(x)e−ρntzn−1
2 L∞(Dx)
dx≤C Z 1
−1
X
n>N
anzn−1 L∞(U)
.
U
e−(1−)a02/2 K=S1
x=−1Dx
21
Negative result : sketch of proof
Thus Z
D(0,e−T)
X
n>0
anzn−1
2
dλ(z)≤C
X
n>N
anzn−1 L∞(U)
.
U
e−(1−)a02/2 K=S1
x=−1Dx
22
Negative result : sketch of proof
Then Z
D(0,e−T)
X
n>N
anzn−1
2
dλ(z)≤C
X
n>N
anzn−1
2 L∞(U)
.
Letz0∈D(0, e−T)\U and
f:z∈C\z0[1,+∞)z7→(z−z0)−1.
U
D(0, e−T)
z0
Proposition (Runge’s theorem)
LetUbe a connected and simply connected open subset ofC, and letfbe a holomorphic function onU. There exists a sequence(˜pk)of polynomials that converges uniformly on every compact subsets ofUtof.
Then, the familypk(z) :=zN+1p˜k(z)is a counter example to the inequality on entire polynomials.
23
Minimal time
ω0 ω
y0
−a a
x y
Corollary (T0=a2/2)
Assume that there existε >0andγ∈C0([0,1],Ω)withγ(0)∈(−1,1)× {0}and γ(1)∈(−1,1)× {π}such thatω0:={z∈Ω,distance(z,Range(γ))< ε} ⊂ω.
Assume that somey0∈(0, π),{(x, y0),−a < x < a}is disjoint fromωwhere a:= sup
(x,y)∈Ω\ω0
{|x|:∃x0∈(−1,1),|x|<|x0|,sgn(x) = sgn(x0),(x0, y)∈ω0}.
One has
• IfT>a2/2, then the Grushin equation isnull controllablein timeT.
• IfT<a2/2, then the Grushin equation isnot null controllablein timeT.
24
Open problems
ω
ω
−a
ω ω
−a We can’t go from the bottom boundary
to the top boundary while staying in- sideω.
T∗>a2/2
A pinched domain.
T∗>a2/2
ω
−a
−b
A cave.
T∗∈[a2/2, b2/2]
25
The Grushin equation
Conclusion
• Non exact controllability
âRegularizing effect of the heat equation
• Approximate controllability âBeauchard-Cannarsa-Guglielmi 2014
• Minimal time null controllability
âDetailed description of the influence of the control region
26
Comments
Other example : Parabolic coupled systems Consider
∂tf1−∆f1+q(x)f2= 0,
∂tf2−∆f2=1ωu, f|∂Ω= 0, f(0) =f0.
(1) System (1) is approx. contr. if and only if
Ik(q) = Z π
0
q(x)φk(x)2dx6= 0, ∀k >0. (2) whereφk:=p
2/πsin(kx).
If (2) is satisfied and Supp(q)∩ω=∅, then the minimal time of null controllability of system (1) is given by
T∗= lim sup−log|Ik(q)|
k2 . Ammar Khodja-Benabdallah-Gonzalez-Burgos-De Teresa, 2015 Question
Is there ageneral theoryto determine the minimal time of null controllability for parabolic systems ?
∂tf+Af=Bu
27
Comments
Other example : Parabolic coupled systems Consider
∂tf1−∆f1+q(x)f2= 0,
∂tf2−∆f2=1ωu, f|∂Ω= 0, f(0) =f0.
(1) System (1) is approx. contr. if and only if
Ik(q) = Z π
0
q(x)φk(x)2dx6= 0, ∀k >0. (2) whereφk:=p
2/πsin(kx).
If (2) is satisfied and Supp(q)∩ω=∅, then the minimal time of null controllability of system (1) is given by
T∗= lim sup−log|Ik(q)|
k2 . Ammar Khodja-Benabdallah-Gonzalez-Burgos-De Teresa, 2015 Question
Is there ageneral theoryto determine the minimal time of null controllability for parabolic systems ?
∂tf+Af=Bu
27
Thanks for your attention !
27