Observability for parabolic equations from a measurable set in time Kim Dang PHUNG , joint work Gengsheng Wang Université d'Orléans. 27 September 2011, Orléans, GDRE CONEDP EXACT CONTROL FOR THE HEAT EQUATION Let Ω be a bounded smooth domain in Rn , ω ⊂ Ω, T > 0. Given y0 ∈ L2 (Ω), there is (y, f ) satisfying ∂t y − ∆y = 1|ω f y=0 y (·, 0) = y0 y (·, T ) = 0 kf kL2 (Ω×(0,T )) ≤ c ky0 kL2 (Ω) G. Lebeau, L. Robbiano, in Ω × (0, T ) , on ∂Ω × (0, T ) , in Ω , in Ω , . CPDE (1995). A. Fursikov, O. Imanuvilov, parabolic equations Lecture Note Series (1996). Also OK for EQUIVALENTLY in Ω × (0, T ) , ∂t u − ∆u = 0 u=0 on ∂Ω × (0, T ) , 2 u (·, 0) ∈ L (Ω) , ku (·, T )k2L2 (Ω) Z Z ≤c ω 0 T |u (x, t)|2 dxdt . ORIGINAL LEBEAU-ROBBIANO STRATEGY Hölder type Observation for elliptic 1−α α kwkH 1 (Ω×(γ,T −γ)) ≤ C kwkL2 (ω×(0,T )) kwkH 1 (Ω×(0,T )) f or any w ∈ H 2 (Ω × (0, T )) , =⇒ ∂t2 + ∆ w = 0 and w|∂Ω = 0 . Controllability in nite and innite dim =⇒ Observability SHORTCUT OF LEBEAU-ROBBIANO STRATEGY Hölder type Observation for elliptic =⇒ Sum of eigenfunctions estimate =⇒ G. Lebeau, E. Zuazua, L. Miller, DCDS (2010). Observability ARMA (1998). PLAN & AIM I. A similar way for parabolic equations II. Replace (0, T ) by a measurable set E in time, |E| > 0 III. Time optimal bang-bang control NEW OBSERVABILITY INEQUALITY a ∈ L∞ (0, T ; Lq (Ω)), q ≥ 2 and b ∈ L∞ (Ω × (0, T ))n ∂t u − ∆u + au + b · ∇u = 0 in Ω × (0, T ) , u=0 on ∂Ω × (0, T ) , u (·, 0) ∈ L2 (Ω) , Ω convex, ω ⊂ Ω, E ⊂ (0, T ) and |E| > 0 Z Z ku (·, T )kL2 (Ω) ≤ c ω E |u (x, t)| dxdt . STRATEGY FOR PARABOLIC Hölder type Observation from one point in time C ku (·, `)kL2 (Ω) ≤ Ce ` ku (·, `)kL2 (ω) 1−α ku (·, 0)kL2 (Ω) =⇒ Observability for a measurable set in time Z Z ku (·, T )kL2 (Ω) ≤ c |u (x, t)| dxdt ω E α Proof Step I. Observation from one point in time →here Ω convex or star-shaped domain →interpolation inequality, parameter ε > 0 Step II. Usage of set of positive measure, E ⊂ (0, T ) and |E| > 0 →Construction of a sequence of points {`j }j≥1 →parameter z > 1 Step III. Choice of the parameters: ε and z Observation from a point in time ku (·, `)kL2 (Ω) ≤ Ce C ` ku (·, `)kL2 (ω) 1−α ku (·, 0)kL2 (Ω) α =⇒ ku (·, `)kL2 (Ω) ≤ C 1 Ce ` ku (·, `)kL1 (ω) + ε ku (·, 0)kL2 (Ω) εγ ∀ε > 0 Set of positive measure Let T > 0 and E ⊂ (0, T ) be a set of positive measure. Let L be a point of density for E . Then for any z > 1, there exists `1 ∈ (L, T ) such that the decreasing sequence {`j }j≥1 given by `j+1 = L + 1 zj (`1 − L) converges to L and satises `j − `j+1 ≤ 3 |E ∩ (`j+1 , `j )| . End of proof 1/2 Take 0 < `j+2 < `j+1 ≤ t ≤ `j < T , ku (·, t)kL2 (Ω) ≤ ε ku (·, `j+2 )kL2 (Ω) 1 + γ CeC/(t−`j+2 ) ku (·, t)kL1 (ω) ε We know that ku (·, `j )kL2 (Ω) ≤ c ku (·, t)kL2 (Ω) . We integrate over E ∩ (`j+1 , `j ) |E ∩ (`j+1 , `j )| c ku (·, `j )kL2 (Ω) ≤ |E ∩ (`j+1 , `j )| ε ku (·, `j+2 )kL2 (Ω) Z C C/(`j+1 −`j+2 ) + γ e ku (·, t)kL1 (ω) dt ε E∩(`j+1 ,`j ) End of proof 2/2 We use `j h −C εγ e 1 z j+2 `1 −L z(z−1) i h −C ku (·, `j )kL2 (Ω) − εγ+1 e z j+2 1 `1 −L z(z−1) i ku (·, `j+2 )kL2 (Ω Z ≤c E∩(`j+1 ,`j ) Set d = C h ku (·, t)kL1 (ω) dt 1 1 `1 −L z(z−1) We choose ε = e−dz j i j+2 . and (γ + 1) z 2 = (γ + 2) e−d(γ+2)z ku (·, `j )kL2 (Ω) − e−d(γ+2)z Z ≤c ku (·, t)kL1 (ω) dt j+2 ku (·, `j+2 )kL2 (Ω) E∩(`j+1 ,`j ) We take j = 2j 0 and make the sum for j 0 = 1 until innity. Proof of Observation from a point in time 1/3 − Gλ (x, t) = e |x − x0 |2 4 (` − t + λ) (` − t − λ)n/2 R |∇u (x, t)|2 Gλ (x, t) dx Nλ,u (t) = RΩ 2 Ω |u (x, t)| Gλ (x, t) dx Proof of Observation from a point in time 2/3 Ω convex 1 2 d dt Z 2 Z |u| Gλ dx + Ω Z 2 |∇u| Gλ dx = Ω d 1 Nλ,u (t) ≤ Nλ,u (t) + dt `−t+λ u (∂t − ∆) uGλ dx Ω R − ∆) u|2 Gλ dx 2 Ω |u| Gλ dx Ω |(∂ Rt Proof of Observation from a point in time 3/3 Step I. Make appear Br ⊂ ω Z Z 2 16λ n 0| 2 − |x−x 4λ dx ≤ |u (`)|2 dx − 2 λNλ,u (`) + +1 |u (`)| e r 4 Br Ω Step II. Estimate λNλ,u (`) + n λNλ,u (`) + ≤ C 4 n 4 " # R 2 |u (0)| dx 1 λ + 1 eC` ln eC (1+ ` ) RΩ 2 ` Ω |u (`)| dx Step III. Choice of λ 1 16λ = 2 C 2 r " # R 2 λ C (1+ 1` ) Ω |u (0)| dx C` + 1 e ln e R 2 ` Ω |u (`)| dx Exact control and cost imply Bang-bang control For any y0 ∈ L2 (Ω), there exists a control f ∈ L∞ (Ω × E) such that the unique solution y of ∂t y − ∆y + a (x, t) y = 1|ω×E f y=0 y (·, 0) = y0 in Ω × (0, T ) , on ∂Ω × (0, T ) , in Ω , satises y (·, T ) = 0 and kf kL∞ (Ω×E) ≤ c ky0 kL2 (Ω) . Known result Let Ω be a bounded smooth domain in Rn, ω ⊂ Ω. Given y0 ∈ L2 (Ω) and M > 0, if y ∗ , 1|(0,T ∗ ) f ∗ is a couple (y, f ) satisfying ∂t y − ∆y = 1|ω f y=0 y (·, 0) = y0 y (·, T ∗ ) = 0 kf (·, t)kL2 (Ω) ≤ M in Ω × (0, +∞) , on ∂Ω × (0, +∞) , in Ω , in Ω , a.e. t ∈ (0, +∞) , is the smallest of the admissible time among and T ∗ > 0 all the couples (y, f ), then kf ∗ (·, t)kL2 (Ω) = M Further, y ∗ , 1|(0,T ∗ ) f ∗ is unique. a.e. t ∈ (0, T ∗ ) . NEW RESULT Let Ω be a convex or star-shaped domain in Rn , ω ⊂ Ω, T > 0, 2 a ∈ L∞ (Ω × (0, T )). Given y0 ∈ L (Ω) and M > 0, if ∗ ∗ y , 1|(τ ∗ ,T ) f is a couple (y, f ) satisfying ∂t y − ∆y + a (x, t) y = 1|ω f y=0 y (·, 0) = y0 y (·, T ) = 0 kf (·, t)kL2 (Ω) ≤ M in Ω × (0, T ) , on ∂Ω × (0, T ) , in Ω , in Ω , a.e. t ∈ (0, T ) , is the largest admissible time among all the and τ ∗ ∈ [0, T ) couples (y, f ), then kf ∗ (·, t)kL2 (Ω) = M Further, y ∗ , 1|(τ ∗ ,T ) f ∗ is unique. a.e. t ∈ (τ ∗ , T ) . Proof by contradiction If kf ∗ (·, t)kL2 (Ω) = M a.e. t ∈ (τ ∗ , T ) is false, then ∃ ε > 0 and E ⊂ (τ ∗ , T ) , |E| > 0 such that kf ∗ (·, t)kL2 (Ω) ≤ M − ε a.e. t ∈ E . Aim : Contradiction with the optimality of τ∗ i.e., construction of a couple z, 1|(τ ∗ +δ,T ) v solution of ∂t z − ∆z + az = 1|ω×(τ ∗ +δ,T ) v z=0 z (·, 0) = y0 z (·, T ) = 0 kv (·, t)kL2 (Ω) ≤ M in Ω × (0, T ) , on ∂Ω × (0, T ) , in Ω , in Ω , a.e. t ∈ (0, T ) , for a small δ > 0 with E ⊂ (τ ∗ + δo , T ) for a suciently small δ o > 0. Thank You!