observability for heat equations · 2018-12-10 · this talk describes di⁄erent approaches to get...
TRANSCRIPT
Observability for heat equations�
Kim Dang PHUNGYangtze Center of Mathematics, Sichuan University,
Chengdu 610064, China.Université d�Orléans, Laboratoire de
Mathématiques - Analyse, Probabilités, Modélisation - Orléans,CNRS FR CNRS 2964, 45067 Orléans cedex 2, France.
E-mail: [email protected]
Abstract
This talk describes di¤erent approaches to get the observability forheat equations without the use of Carleman inequalities.
Contents
1 The heat equation and observability 2
2 Our motivation 3
3 Our strategy 43.1 Proof of "Hölder continuous dependence from one point in time"
) "Sum of Laplacian eigenfunctions" . . . . . . . . . . . . . . . 43.2 Proof of "Hölder continuous dependence from one point in time"
) "Observability" . . . . . . . . . . . . . . . . . . . . . . . . . . 43.3 Proof of "Hölder continuous dependence from one point in time"
) "Re�ned Observability" . . . . . . . . . . . . . . . . . . . . . 7
4 What I hope 94.1 Logarithmic convexity method . . . . . . . . . . . . . . . . . . . 94.2 Weighted logarithmic convexity method . . . . . . . . . . . . . . 10
5 What I can do 125.1 The frequency function . . . . . . . . . . . . . . . . . . . . . . . . 125.2 The frequency function with weight . . . . . . . . . . . . . . . . . 145.3 The heat equation with space-time potential . . . . . . . . . . . . 19
�This talk was done when the author visited School of Mathematics & Statistics, NortheastNormal University, Changchun, China, (July 4-21, 2011).
1
6 What already exists 216.1 Monotonicity formula . . . . . . . . . . . . . . . . . . . . . . . . 21
6.1.1 Proof of Lemma B . . . . . . . . . . . . . . . . . . . . . . 226.1.2 Proof of Lemma A . . . . . . . . . . . . . . . . . . . . . . 236.1.3 Proof of Lemma C . . . . . . . . . . . . . . . . . . . . . . 24
6.2 Quantitative unique continuation property for the Laplacian . . . 286.3 Quantitative unique continuation property for the elliptic opera-
tor @2t +� . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296.4 The heat equation and the Hölder continuous dependence from
one point in time . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1 The heat equation and observability
We consider the heat equation in the solution u = u(x; t)8<: @tu��u = 0 in � (0;+1) ,u = 0 on @� (0;+1) ,
u (�; 0) 2 L2() ,(1.1)
living in a bounded open set in Rn, n � 1, either convex or C2 and connected,with boundary @. It is well-known that the above problem is well-posed andhave a unique solution u 2 C
�[0; T ] ;L2 ()
�\ L2
�0; T ;H1
0 ()�for all T > 0.
The observability problem consists in proving the following estimateZ
ju (x; T )j2 dx � CZ T
0
Z!
ju (x; t)j2 dxdt
for some constant C > 0 independent on the initial data. Here, T > 0 and ! isa non-empty open subset in .
2
In the literature, two ways allow to prove such observability estimate. Oneis due to the work of Fursikov and Imanuvilov based on global Carleman in-equalities (see [FI]). The other proof is established by Lebeau and Robbiano(see [LR]. See [Le] for an english version). We resume the Lebeau-Robbianostrategy as follows.
ku (�; T )k2L2() � CZ T
0
Z!
ju (x; t)j2 dxdt
*controllability in �nite and in�nite dimension
*
Xj=1;::;N
jaj j2 � CeCp�N
Z!
������X
j=1;::;N
ajej (x)
������2
dx
for any fajg where (ej ; �j) solves the eigenvalue problem with Dirichlet bound-ary conditions (0 < �1 � �2 � � � � being the corresponding eigenvalues).
*
For any > 0 and any non-trivial ' 2 C10 (� (0; T )), there are C > 0 and� 2 (0; 1), such that for any w 2 H2 (� (0; T )) with
�@2t +�
�w = f and
wj@ = 0, it holds
kwkH1(�( ;T� )) � C kwk�H1(�(0;T ))
�kfkL2(�(0;T )) + k'wkL2(�(0;T ))
�1��The above interpolation inequality is proved using Carleman inequalities.
Recently, a shortcut of the Lebeau-Robbiano strategy is given in [M].
2 Our motivation
In application to bang-bang control (see [W]), we need the following re�nedobservability estimate from measure set in time.
ku (�; T )kL2() � CZE
Z!
ju (x; t)j dxdt
for some constant C > 0 independent on the initial data. Here, E � (0; T ) is ameasurable set of positive measure and ! is a non-empty open subset in .
Further, we want to be able to extend the proof to heat equations withspace-time potentials.
The approach describes in this talk is linked to parabolic quantitative uniquecontinuation (see [BT], [Li], [L], [P], [EFV], [K], [KT] and references therein).
3
3 Our strategy
We are able to prove that
Hölder continuous dependence from one point in time
=) Observability from a measure set in time.
More precisely,
ku (�; t)kL2() � C ku (�; 0)k�L2()
�eC=t ku (�; t)kL2(!)
�1��8t > 0
+
ku (�; T )kL2() � CZE
Z!
ju (x; t)j dxdt
3.1 Proof of "Hölder continuous dependence from one pointin time" ) "Sum of Laplacian eigenfunctions"
Choose u (x; t) =P
j=1;::;N bje��jtej (x). Then it holds for any t > 0
Xj=1;::;N
��bje��jt��2 �0@C X
j=1;::;N
jbj j21A�
0B@CeC=t Z!
������X
j=1;::;N
bje��jtej (x)
������2
dx
1CA1��
which implies when t = 1=p�N
Pj=1;::;N
���bje��j=p�N ���2 ��Ce2
p�NP
j=1;::;N
���bje��j=p�N ���2����CeC
p�NR!
���Pj=1;::;N bje��j=
p�N ej (x)
���2 dx�1�� .
This gives the desired estimate with bje��j=p�N = aj .
3.2 Proof of "Hölder continuous dependence from one pointin time" ) "Observability"
Let ` 2 [0; T ) and `1 2 (`; T ]. Let z > 1. Introduce the decreasing sequencef`mgm�1, which converges to ` given by
`m+1 = `+1
zm(`1 � `) .
4
Then
`m � `m+1 =1
zm(z � 1) (`1 � `) > 0 .
We start with the following interpolation estimate. Let ! be a nonempty opensubset of . For any 0 � t1 < t2 � T ,
ku (�; t2)kL2() �C1" e
C2t2�t1 ku (�; t2)kL2(!) + " ku (�; t1)kL2() 8" > 0 .
Let 0 < `m+2 < `m+1 � t < `m < T . We get
ku (�; t)kL2() �C1" e
C2t�`m+2 ku (�; t)kL2(!) + " ku (�; `m+2)kL2() 8" > 0 .
Recall thatku (�; `m)kL2() � ku (�; t)kL2() .
Therefore,
ku (�; `m)kL2() �C1" e
C2t�`m+2 ku (�; t)kL2(!) + " ku (�; `m+2)kL2() 8" > 0 .
Integrating it over t 2 (`m+1; `m), it gives
(`m � `m+1) ku (�; `m)kL2() � C1" e
C2`m+1�`m+2
R `m`m+1
ku (�; t)kL2(!) dt+" (`m � `m+1) ku (�; `m+2)kL2() 8" > 0 .
That is
ku (�; `m)kL2() �h
1`1�`
zm
z�1
iC1" e
C2h
1`1�`
zm+1
z�1
i R `m`m+1
ku (�; t)kL2(!) dt+" ku (�; `m+2)kL2() 8" > 0 .
Then
ku (�; `m)kL2() � 1"
1zC1C2e2C2
h1
`1�`zm+1
z�1
i R `m`m+1
ku (�; t)kL2(!) dt+" ku (�; `m+2)kL2() 8" > 0 .
Take d = 2C2h
1`1�`
1z(z�1)
i. It gives
" e�dzm+2 ku (�; `m)kL2() � "1+ e�dz
m+2 ku (�; `m+2)kL2()� 1
zC1C2
R `m`m+1
ku (�; t)kL2(!) dt 8" > 0 .
Take " = e�dzm+2
, then
e�( +1)dzm+2 ku (�; `m)kL2() � e�(2+ )dz
m+2 ku (�; `m+2)kL2()� 1
zC1C2
R `m`m+1
ku (�; t)kL2(!) dt .
5
Take z =q
+2 +1 , then
e�(2+ )dzm ku (�; `m)kL2() � e�(2+ )dz
m+2 ku (�; `m+2)kL2()�q
+1 +2
C1C2
R `m`m+1
ku (�; t)kL2(!) dt .
Change m to 2m0 and sum the above from m0 = 1 to in�nity give the desiredresult with E = (0; T ). When E = (0; T ), we take ` = 0 and `1 = T , and get
e�(2+ )2C2
24 1T
r +2 +1r
+2 +1
�1
35ku (�; T )kL2()
� e�(2+ )dz2 ku (�; `2)kL2()�P
m0�1
�e�(2+ )dz
2m0ku (�; `2m0)kL2() � e�(2+ )dz
2m0+2 ku (�; `2m0+2)kL2()�
�P
m0�1
q +1 +2
C1C2
R `2m0`2m0+1
ku (�; t)kL2(!) dt
�q
+1 +2
C1C2
R `20ku (�; t)kL2(!) dt .
Finally,
ku (�; T )kL2() �r + 1
+ 2
C1C2e2C2
1T (2+ )
p +2[
p +2+
p +1]
Z T
0
ku (�; t)kL2(!) dt .
In particular,
ku (�; 1)kL2() �r + 1
+ 2
C1C2e2C2(2+ )
p +2[
p +2+
p +1]
Z 1
0
ku (�; t)kL2(!) dt
and
ku (�;m)kL2() �r + 1
+ 2
C1C2e2C2(2+ )
p +2[
p +2+
p +1]
Z m
m�1ku (�; t)kL2(!) dt
for any m � 1. Now, take T such that M < T �M + 1 for some M � 1,T2 ku (�; T )kL2()
�M ku (�;M)kL2()�PM
m=1 ku (�;m)kL2()�q
+1 +2
C1C2e2C2(2+ )
p +2[
p +2+
p +1]PM
m=1
Rmm�1 ku (�; t)kL2(!) dt
�q
+1 +2
C1C2e2C2(2+ )
p +2[
p +2+
p +1] RM
0ku (�; t)kL2(!) dt
�q
+1 +2
C1C2e2C2(2+ )
p +2[
p +2+
p +1] R T
0ku (�; t)kL2(!) dt .
We conclude that for any T � 1,
ku (�; T )kL2() �1
T
r + 1
+ 2
C1C2e2C2
1T (2+ )
p +2[
p +2+
p +1]
Z T
0
ku (�; t)kL2(!) dt ,
6
and for any T > 1
ku (�; T )kL2() �2
T
r + 1
+ 2
C1C2e2C2(2+ )
p +2[
p +2+
p +1]
Z T
0
ku (�; t)kL2(!) dt .
Combining the case T � 1 and the case T > 1, we get the following observability.
ku (�; T )kL2() �3
T
r + 1
+ 2
C1C2e4C2(1+
1T )(2+ )
2
Z T
0
ku (�; t)kL2(!) dt
for any T > 0.
3.3 Proof of "Hölder continuous dependence from one pointin time" ) "Re�ned Observability"
In this proof, C denotes a positive constant which may change of value fromline to line. Let ` 2 [0; T ) and `1 2 (`; T ]. Let z > 1. Introduce the decreasingsequence f`mgm�1, which converges to ` given by
`m+1 = `+1
zm(`1 � `) .
Then`m � `m+1 =
1
zm(z � 1) (`1 � `) > 0 .
We start with the following interpolation estimate. Introduce e! b ! � . Forany 0 � t1 < t2 � T ,
ku (�; t2)kL2() �C1"�e
C2t2�t1 ku (�; t2)kL2(e!) + " ku (�; t1)kL2() 8" > 0 .
By Nash and Poincare inequality,
ku (�; t2)kL2(e!) � C3"n=2
ku (�; t2)kL1(!) + " kru (�; t2)kL2() 8" > 0 .
By an energy method,
kru (�; t2)kL2() �C4
(t2 � t1)1=2ku (�; t1)kL2() .
Therefore, from the above three estimate, denoting
= ��1 +
n
2
�+n
2,
C5 = C1C3
�C1C4pC2
�n=2, C6 = (n+ 1)C2 ,
7
we get
ku (�; t2)kL2() �C5" e
C6t2�t1 ku (�; t2)kL1(!) + " ku (�; t1)kL2() 8" > 0 .
Let 0 < `m+2 < `m+1 � t < `m < T . We get
ku (�; t)kL2() �C5" e
C6t�`m+2 ku (�; t)kL1(!) + " ku (�; `m+2)kL2() 8" > 0 .
Recall thatku (�; `m)kL2() � ku (�; t)kL2() .
Therefore,
ku (�; `m)kL2() �C5" e
C6t�`m+2 ku (�; t)kL1(!) + " ku (�; `m+2)kL2() 8" > 0 .
Integrating it over t 2 (`m+1; `m), it gives
(`m � `m+1) ku (�; `m)kL2() � C5" e
C6`m+1�`m+2
R `m`m+1
ku (�; t)kL1(!) dt+" (`m � `m+1) ku (�; `m+2)kL2() 8" > 0 .
That is
ku (�; `m)kL2() �h
1`1�`
zm
z�1
iC5" e
C6h
1`1�`
zm+1
z�1
i R `m`m+1
ku (�; t)kL1(!) dt+" ku (�; `m+2)kL2() 8" > 0 .
Then
ku (�; `m)kL2() � 1"
1zC5C6e2C6
h1
`1�`zm+1
z�1
i R `m`m+1
ku (�; t)kL1(!) dt+" ku (�; `m+2)kL2() 8" > 0 .
Take d = 2C6h
1`1�`
1z(z�1)
i. It gives
" e�dzm+2 ku (�; `m)kL2() � "1+ e�dz
m+2 ku (�; `m+2)kL2()� 1
zC5C6
R `m`m+1
ku (�; t)kL1(!) dt 8" > 0 .
Take " = e�dzm+2
, then
e�( +1)dzm+2 ku (�; `m)kL2() � e�(2+ )dz
m+2 ku (�; `m+2)kL2()� 1
zC5C6
R `m`m+1
ku (�; t)kL1(!) dt .
Take z =q
+2 +1 , then
e�(2+ )dzm ku (�; `m)kL2() � e�(2+ )dz
m+2 ku (�; `m+2)kL2()� 1
zC5C6
R `m`m+1
ku (�; t)kL1(!) dt .
8
Change m to 2m0 and sum the above from m0 = 1 to in�nity give the desiredresult with E = (0; T ). By using a property of density points, we are ableto replace (0; T ) by a measurable set of positive measure (see [PW2]). Forsimplicity, when E = (0; T ), we take ` = 0 and `1 = T , and get
e�(2+ )2C6
24 1T
r +2 +1r
+2 +1
�1
35ku (�; T )kL2()
� e�(2+ )dz2 ku (�; `2)kL2()�P
m0�1
he�(2+ )dz
2m0ku (�; `2m0)kL2() � e�(2+ )dz
2m0+2 ku (�; `2m0+2)kL2()i
�P
m0�11zC5C6
R `2m0`2m0+1
ku (�; t)kL1(!) dt� 1
zC5C6
R T0ku (�; t)kL1(!) dt .
That is
ku (�; T )kL2() �r + 2
+ 1
C5C6e2C6
1T (2+ )
p +2[
p +2+
p +1]
Z T
0
Z!
ju (x; t)j dxdt .
4 What I hope
Recall the following two energy identities. For t 2 (0; T ],
1
2
d
dt
Z
ju (x; t)j2 dx+Z
jru (x; t)j2 dx = 0 ,
1
2
d
dt
Z
jru (x; t)j2 dx+Z
j�u (x; t)j2 dx = 0 .
4.1 Logarithmic convexity method
Recall that if t 7!logf is a convex function, then for any t1 < t2 and any� 2 (0; 1),
logf (�t2 + (1� �) t1) � �logf (t2) + (1� �) logf (t1)
which impliesf (�t2 + (1� �) t1) � [f (t2)]� [f (t1)]1�� .
Further, when f 2 C2, logf is a convex function if and only if (logf)00 =f 00f�(f 0)
2
f2 � 0.
9
Now take t = �T , t1 = 0, t2 = T and f (t) =Rju (x; t)j2 dx. Therefore,
(logf (t))00 = 1f2
�4Rj�u (x; t)j2 dx
Rju (x; t)j2 dx�
��2Rjru (x; t)j2 dx
�2�� 0 by Cauchy-Schwarz inequality,
which impliesZ
ju (x; t)j2 dx ��Z
ju (x; T )j2 dx�t=T �Z
ju (x; 0)j2 dx�1�t=T
.
This estimate has similar form than a quantitative Hölder continuous depen-dence. We refer to [FEF] for more advanced computation with this technique.
4.2 Weighted logarithmic convexity method
We would like reproduce similar computation than the previous subsection butwith
f (t) =
Z
ju (x; t)j2G (x; t) dx ,
where G is a suitable weighted function.
Step 1 .- Make appear Br � .- Let Br be the ball of radius r and centerx0, and contained in . Denote m0 = max
x2jx� x0j2. Let � > 0. Take G (x; t) =
e�jx�x0j2T�t+� . If logf is a convex function, then
Rju (x; t)j2 e�
jx�x0j2T�t+� dx
��Rju (x; T )j2 e�
jx�x0j2� dx
�t=T �Rju (x; 0)j2 e�
jx�x0j2T+� dx
�1�t=T��
1"(1��)=�
Rju (x; T )j2 e�
jx�x0j2� dx
�� �"Rju (x; 0)j2 e�
jx�x0j2T+� dx
�1��� 1
"(1��)=�
Rju (x; T )j2 e�
jx�x0j2� dx+ "
Rju (x; 0)j2 e�
jx�x0j2T+� dx 8" > 0 .
On the other hand,
Rju (x; T )j2 e�
jx�x0j2� dx
=RBrju (x; T )j2 e�
jx�x0j2� dx+
RnBr
ju (x; T )j2 e�jx�x0j2
� dx
�RBrju (x; T )j2 dx+ e� r2
�
RnBr
ju (x; T )j2 dx
�RBrju (x; T )j2 dx+ e� r2
� em0T�t
Rju (x; t)j2 e�
jx�x0j2T�t+� dx .
10
Take t = T=2 i.e., � = 1=2 and � > 0 such that 1"(1��)=�
e�r2
� em0T=2 = 1
2 . ThenRju (x; T=2)j2 e�
jx�x0j2T=2+� dx � 1
"
RBrju (x; T )j2 dx
+ 12
Rju (x; T=2)j2 e�
jx�x0j2T=2+� dx
+"Rju (x; 0)j2 e�
jx�x0j2T+� dx 8" > 0 .
It impliesRju (x; T=2)j2 dx � e
m0T=2Rju (x; T=2)j2 e�
jx�x0j2T=2+� dx
� 2em0T=2 1
"
RBrju (x; T )j2 dx+ 2e
m0T=2 "
Rju (x; 0)j2 dx ,
for any " > 0. Take " = 12
hRju (x; T=2)j2 dx
i h2e
m0T=2Rju (x; 0)j2 dx
i�1. ThenR
ju (x; T )j2 dx �
Rju (x; T=2)j2 dx
� 4em0T=2
hRBrju (x; T )j2 dx
i1=2 hRju (x; 0)j2 dx
i1=2.
Step 2 .- Get good sign for a derivative. The choice of the weighted functionG can be seen as follows. The quantityZ
(@t ��)�ju (x; t)j2
�G (x; t) dx+
Z
ju (x; t)j2 (@t +�)G (x; t) dx
where G 2 C1, has two expressionsZ
d
dt
�ju (x; t)j2G (x; t)
�dx�
Z@
@�
�ju (x; t)j2
�G (x; t)�ju (x; t)j2 @�G (x; t) d�
and
�2Z
jru (x; t)j2G (x; t) dx+Z
ju (x; t)j2 (@t +�)G (x; t) dx .
When (@t +�)G (x; t) = 0 and u = 0 on @,
1
2
d
dt
Z
ju (x; t)j2G� (x; t) dx+Z
jru (x; t)j2G� (x; t) dx = 0 .
This can be compared with the �rst energy identity. The natural choice for Gis given by
G� (x; t) =1
(T � t+ �)n=2e�
jx�x0j24(T�t+�) ,
which satis�es (@t +�)G� (x; t) = 0 and G� (x; T ) = 1�n=2
e�jx�x0j2
4� .
Of course, we will need to compute ddt
Rjru (x; t)j2G� (x; t) dx.
11
Conclusion .- There are two parts: "Make appear Br � " and "Get goodsign for a derivative". With the weighted logarithmic convexity method, "Makeappear Br � " was easy but "Get good sign for a derivative" is di¢ cult andactually unsolved.
Extension .- The above computation can be extended to the heat equationwith second member. We consider the heat equation in the solution U = U(x; t)with a suitable second member F = F (x; t)8<: @tU ��U = F in � (0; T ) ,
U = 0 on @� (0; T ) ,U (�; 0) 2 L2() .
The two energy identities become as follows. For t 2 (0; T ],
1
2
d
dt
Z
jU (x; t)j2 dx+Z
jrU (x; t)j2 dx =Z
F (x; t)U (x; t) dx ,
1
2
d
dt
Z
jrU (x; t)j2 dx+Z
j�U (x; t)j2 dx = �Z
F (x; t)�U (x; t) dx .
Further,
12ddt
RjU (x; t)j2G� (x; t) dx+
RjrU (x; t)j2G� (x; t) dx
=RU (x; t)F (x; t)G� (x; t) dx .
5 What I can do
5.1 The frequency function
De�ne for t 2 (0; T ],
N (t) =
Rjru (x; t)j2 dxRju (x; t)j2 dx
, wheneverZ
ju (x; t)j2 dx 6= 0 .
Then the �rst energy identity becomes as follows.
1
2
d
dt
Z
ju (x; t)j2 dx+N (t)Z
ju (x; t)j2 dx = 0 .
On the other hand, we study the sign N 0 (t).
d
dtN (t) = 2
�Rj�u (x; t)j2 dx
Rju (x; t)j2 dx+
�Rjru (x; t)j2 dx
�2�R
ju (x; t)j2 dx
�2 � 0
12
by Cauchy-Schwarz inequality. Therefore, for t 2 [�; T ],
N (T ) � N (t) � N (�) .
Backward uniqueness uses this technique when u (�; 0) 2 H10 () with N (t) �
N (0). We have to solve
0 � 1
2
d
dt
Z
ju (x; t)j2 dx+N (0)Z
ju (x; t)j2 dx .
That is
0 � d
dt
�e2tN(0)
Z
ju (x; t)j2 dx�.
Integrating over (0; t),Z
ju (x; 0)j2 dx � e2tN(0)Z
ju (x; t)j2 dx .
In our problem, we use N (T ) � N (t). We have to solve
1
2
d
dt
Z
ju (x; t)j2 dx+N (T )Z
ju (x; t)j2 dx � 0 .
That isd
dt
�e2tN(T )
Z
ju (x; t)j2 dx�� 0 .
Integrating over (0; T ),
e2TN(T )Z
ju (x; T )j2 dx �Z
ju (x; 0)j2 dx .
That is
2TN (T ) � logRju (x; 0)j2 dxR
ju (x; T )j2 dx
.
Extension .- The above computation can be extended to the heat equationwith second member. We consider the heat equation in the solution U = U(x; t)with a suitable second member F = F (x; t)8<: @tU ��U = F in � (0; T ) ,
U = 0 on @� (0; T ) ,U (�; 0) 2 L2() .
The following inequality holds (see [BT]).
d
dtN (t) �
RjF (x; t)j2 dxR
jU (x; t)j2 dx
.
13
Indeed,
ddtN (t) = 2
[�Rj�U j2�
RF�Udx]
RjU j2dx+(
RjrU j2dx)
2�RFUdx
RjrU j2dx
(RjU j2dx)
2 .
But �RjrU j2 dx
�2�RFUdx
RjrU j2 dx
=�R
jrU j2 dx� 1
2
RFUdx
�2� 1
4
�RFUdx
�2=��R�UUdx� 1
2
RFUdx
�2 � 14
�RFUdx
�2=��R
��U + 1
2F�Udx
�2 � 14
�RFUdx
�2�R
���U + 12F��2 dx R
jU j2 dx� 1
4
�RFUdx
�2by Cauchy-Schwarz inequality. Finally,�R
jrU j2 dx
�2�RFU (x; t) dx
RjrU j2 dx
�R
���U + 12F��2 dx R
jU j2 dx
=Rj�U j2 dx
RjU j2 dx+
R�UFdx
RjU j2 dx+
R
�� 12F��2 dx R
jU j2 dx .
5.2 The frequency function with weight
Let � > 0. Recall that for t 2 [0; T ]
G� (x; t) =1
(T � t+ �)n=2e�
jx�x0j24(T�t+�) .
De�ne for t 2 (0; T ],
N� (t) =
Rjru (x; t)j2G� (x; t) dxRju (x; t)j2G� (x; t) dx
, wheneverZ
ju (x; t)j2 dx 6= 0 .
Step 1 .- Get good sign for a derivative. We claim that if is convex orstar-shaped w.r.t. x0, then
d
dtN� (t) �
1
T � t+ �N� (t) .
That isd
dt[(T � t+ �)N� (t)] � 0 .
Integrating over (t; T ),
�N� (T ) � (T + �)N� (t) .
14
Recall that
1
2
d
dt
Z
ju (x; t)j2G� (x; t) dx+Z
jru (x; t)j2G� (x; t) dx = 0 .
That is
1
2
d
dt
Z
ju (x; t)j2G� (x; t) dx+N� (t)Z
ju (x; t)j2G� (x; t) dx = 0 .
Therefore,
d
dt
Z
ju (x; t)j2G� (x; t) dx+2�N� (T )
T + �
Z
ju (x; t)j2G� (x; t) dx � 0 .
That isd
dt
�e2t�N�(T )
T+�
Z
ju (x; t)j2G� (x; t) dx�� 0 .
Integrating over (0; T=2),
eT�N�(T )
T+�
Z
ju (x; T=2)j2G� (x; T=2) dx �Z
ju (x; 0)j2G� (x; 0) dx .
But Rju (x; T )j2 dx �
Rju (x; T=2)j2 dx
� em02T
Rju (x; T=2)j2 e�
jx�x0j24(T=2+�) dx .
where m0 = maxx2
jx� x0j2. Therefore,
eT�N�(T )
T+� �Rju(x;0)j2G�(x;0)dxR
ju(x;T=2)j2G�(x;T=2)dx
�Rju(x;0)j2dxR
ju(x;T=2)j2e�
jx�x0j24(T=2+�) dx
� em02T
Rju(x;0)j2dxR
ju(x;T )j2dx .
That is
�N� (T ) ��1 +
�
T
�logem02T
Rju (x; 0)j2 dxR
ju (x; T )j2 dx
.
On the other hand,
n
4� n
4
�1 +
�
T
�loge1+
m02T
Rju (x; 0)j2 dxR
ju (x; T )j2 dx
.
Finally,
n
4+ �N� (T ) �
�n4+ 1��1 +
�
T
�loge1+
m02T
Rju (x; 0)j2 dxR
ju (x; T )j2 dx
.
15
Now, we prove the claim saying that if is convex or star-shaped w.r.t. x0,then
d
dtN� (t) �
1
T � t+ �N� (t) .
First, by an integration by parts and by usingrG� (x; t) = � x�x02(T�t+�)G� (x; t),
we haveRjruj2G�dx = �
R(�uuG� +ruurG�) dx+
R@@�uuG�d�
= �R
�@tu� x�x0
2(T�t+�) � ru�uG�dx .
Next, we compute ddt
Rjruj2G�dx.
ddt
Rjruj2G�dx =
R
�2ru@truG� + jruj2 @tG�
�dx
= �2R(�u@tuG� +ru@turG�) dx�
Rjruj2�G�dx
= �2Rj@tuj2G�dx+
R@tu
x�x0T�t+� � ruG�dx
+Rr�jruj2
�rG�dx+
R@jruj2 x�x0
2(T�t+�) � �G�d� .
But Rr�jruj2
�rG�dx =
R@i
�j@juj2
�@iG�dx
= 2R@ju@
2iju@iG�dx
= �2R
�@2j u@iu@iG� + @ju@iu@
2ijG�
�dx
+2R@@ju@iu�j@iG�d�
= �2R
��ururG� + @ju@iu
h�@j(xi�x0i)
2(T�t+�) +(xi�x0i)(xj�x0j)
4(T�t+�)2
iG�
�dx
�R@j@�uj2 x�x0
(T�t+�) � �G�d�
=R@tu
x�x0T�t+� � ruG�dx+
1(T�t+�)
Rjruj2G�dx� 2
R
�x�x0
2(T�t+�) � ru�2G�dx
�R@j@�uj2 x�x0
(T�t+�) � �G�d� .
Therefore,
ddt
Rjruj2G�dx
= �2Rj@tuj2G�dx+ 2
R@tu
x�x0T�t+� � ruG�dx� 2
R
�x�x0
2(T�t+�) � ru�2G�dx
+ 1(T�t+�)
Rjruj2G�dx
�R@j@�uj2 x�x0
(T�t+�) � �G�d� +R@jruj2 x�x0
2(T�t+�) � �G�d� .
That is
ddt
Rjruj2G�dx = �2
R
���@tu� x�x02(T�t+�) � ru
���2G�dx+ 1(T�t+�)
Rjruj2G�dx
�R@j@�uj2 x�x0
2(T�t+�) � �G�d� .
16
Finally, we compute ddtN� (t)
ddtN� (t) =
ddt
Rjruj2G�dx
Rjuj2G�dx�
Rjruj2G�dx
ddt
Rjuj2G�dx
(Rjuj2G�dx)
2
= 1
(Rjuj2G�dx)
2
��2R
���@tu� x�x02(T�t+�) � ru
���2G�dx R juj2G�dx�+ 1
(Rjuj2G�dx)
2
�1
(T�t+�)Rjruj2G�dx
Rjuj2G�dx
�+ 1
(Rjuj2G�dx)
2
��R@j@�uj2 x�x0
2(T�t+�) � �G�d�Rjuj2G�dx
�+ 1
(Rjuj2G�dx)
2 2�R
jruj2G�dx
�2= 1
(Rjuj2G�dx)
2
��2R
���@tu� x�x02(T�t+�) � ru
���2G�dx R juj2G�dx�+ 1(T�t+�)N� (t)�
R@j@�uj2 x�x0
2(T�t+�) ��G�d�Rjuj2G�dx
+ 1
(Rjuj2G�dx)
2 2��R
�@tu� x�x0
2(T�t+�) � ru�uG�dx
�2� 1
(T�t+�)N� (t)
by Cauchy-Schwarz inequality and the fact that (x� x0) � � � 0 for convexdomain .
Step 2 .- Make appear Br � .Rju (x; T )j2 e�
jx�x0j24� dx
�RBrju (x; T )j2 e�
jx�x0j24� dx+
RnBr
ju (x; T )j2 e�jx�x0j2
4� dx
�RBrju (x; T )j2 dx+ 1
r2
Rjx� x0j2 ju (x; T )j2 e�
jx�x0j24� dx .
On the other hand,Rjx� x0j2 ju (x; T )j2 e�
jx�x0j24� dx
=R(x� x0) ju (x; T )j2 � (�2�)re�
jx�x0j24� dx
= �2�R@((x� x0) � �) ju (x; T )j2 e�
jx�x0j24� d� + 2�n
Rju (x; T )j2 e�
jx�x0j24� dx
+4�R(x� x0)u (x; T ) � ru (x; T ) e�
jx�x0j24� dx by integration by parts
� 2�nRju (x; T )j2 e�
jx�x0j24� dx
+ 12
R16�2 jru (x; T )j2 e�
jx�x0j24� dx+ 1
2
Rjx� x0j2 ju (x; T )j2 e�
jx�x0j24� dx ,
by Cauchy-Schwarz inequality. Therefore,Rju (x; T )j2 e�
jx�x0j24� dx
�RBrju (x; T )j2 dx
+ 1r2
�4�n
Rju (x; T )j2 e�
jx�x0j24� dx+
R16�2 jru (x; T )j2 e�
jx�x0j24� dx
��RBrju (x; T )j2 dx+ 16�
r2
�n4 + �N� (T )
� Rju (x; T )j2 e�
jx�x0j24� dx .
17
By step 1, if is convex,Rju (x; T )j2 e�
jx�x0j24� dx
�RBrju (x; T )j2 dx
+ 16�r2
��n4 + 1
� �1 + �
T
�log
e1+m02T
Rju(x;0)j2dxR
ju(x;T )j2dx
� Rju (x; T )j2 e�
jx�x0j24� dx .
Take
� = �T +
264T 2 + 4 r216
T
2�n4 + 1
�log
e1+m02T
Rju(x;0)j2dxR
ju(x;T )j2dx
3751=2
in order that
16�
r2
"�n4+ 1��1 +
�
T
�loge1+
m02T
Rju (x; 0)j2 dxR
ju (x; T )j2 dx
#=1
2.
Then Rju (x; T )j2 dx � e
m04�
Rju (x; T )j2 e�
jx�x0j24� dx
� 2em04�
RBrju (x; T )j2 dx .
But
m0
4� = m0
4
0B@�T +24T 2 + 4 r216 T
2(n4+1)loge1+
m02T
Rju(x;0)j2dxR
ju(x;T )j2dx
351=21CA�1
� m0
4
��1 + r2
(n+4)m0
�1=2+ 1
�2r2 (n+ 4) log
e1+m02T
Rju(x;0)j2dxR
ju(x;T )j2dx .
Finally,Rju (x; T )j2 dx
� 2�e1+
m02T
Rju(x;0)j2dxR
ju(x;T )j2dx
�(n4+1)��1+ r2
(n+4)m0
�1=2+1
�2m0r2 R
Brju (x; T )j2 dx .
That isZ
ju (x; T )j2 dx ��e1+
m02T
Z
ju (x; 0)j2 dx���
2
ZBr
ju (x; T )j2 dx�1��
,
where
� =
�n4 + 1
���1 + r2
(n+4)m0
�1=2+ 1
�2m0
r2
1 +�n4 + 1
���1 + r2
(n+4)m0
�1=2+ 1
�2m0
r2
.
18
It impliesRju (x; T )j2 dx
��"Rju (x; 0)j2 dx
�� �1
"�=(1��)2e(1+
m02T )
�1��
RBrju (x; T )j2 dx
�1��� 1
"�2e(1+
m02T )�
RBrju (x; T )j2 dx+ "
Rju (x; 0)j2 dx 8" > 0 ,
where
� =�
1� � =�n4+ 1� �
1 +r2
(n+ 4)m0
�1=2+ 1
!2m0
r2.
Conclusion .- There are two parts: "Make appear Br � " and "Get goodsign for a derivative". With the frequency function with weight, "Make appearBr � " is unusual "Get good sign for a derivative" is more usual.
Now, recall that
ku (�; T )kL2() �C1" eC2T ku (�; T )kL2(!) + " ku (�; 0)kL2() 8" > 0 ,
implies
ku (�; T )kL2() �3
T
r + 1
+ 2
C1C2e4C2(1+
1T )(2+ )
2
Z T
0
ku (�; t)kL2(!) dt .
Take ! = Br, = �,
C1 = 2e� and C2 =
m0
2� =
�n4+ 1� �
1 +r2
(n+ 4)m0
�1=2+ 1
!m20
r2
then using the fact that r2 � m0 = maxx2
jx� x0j2,
ku (�; T )kL2() � C(n)1
T
1
m0eC(n)m0(1+ 1
T )(1+m0r2)3Z T
0
ku (�; t)kL2(Br)dt ,
whenever is convex.
5.3 The heat equation with space-time potential
The above computation can be extended to the heat equation with second mem-ber. We consider the heat equation in the solution U = U(x; t) with a suitablesecond member F = F (x; t)8<: @tU ��U = F in � (0; T ) ,
U = 0 on @� (0; T ) ,U (�; 0) 2 L2() .
19
The following inequality holds (see [PW]). If is convex,
d
dtN� (t) �
1
T � t+ �N� (t) +RjF (x; t)j2G� (x; t) dxR
jU (x; t)j2G� (x; t) dx
.
Consider the heat equation with potentials a 2 L1 (� (0; T )) and b 2L1 (� (0; T ))n .8<: @tv ��v + av + b � rv = 0 in � (0; T ) ,
v = 0 on @� (0; T ) ,v (�; 0) 2 L2() .
Take F = av + b � rv. Then
d
dtN� (t) �
1
T � t+ �N� (t) + kakL1(�(0;T )) + kbkL1(�(0;T ))N� (t) ,
whenever is convex. We are able to reproduce the above treatment to get are�ned observability estimate (see [PW2]).
In order to drop the convexity hypothesis, a local study is necessary (seenext section). Take w = �u where � 2 C10 (BR) with 0 � � � 1 and � = 1 inBR=2. Then8<: @tw ��w = �2r�ru���u in BR � (0; T ) ,
w = 0 on @BR � (0; T ) ,w (�; 0) 2 L2(BR) .
Take F = �2r�ru���u. Then
d
dtN� (t) �
1
T � t+ �N� (t) + 2Rj2r�ruj2G�dxRj�uj2G�dx
+ 2
Rj��uj2G�dxRj�uj2G�dx
.
We need to bound Rj2r�ruj2G�dxRj�uj2G�dx
+
Rj��uj2G�dxRj�uj2G�dx
by an easy function on time in order to reproduce the above treatment. Here,we can choose t small to do so. Actually, it is not yet done but ideas alreadymay be set up in [EFV].
20
6 What already exists
Here, we recall most of the material from the works of I. Kukavica [Ku2] and L.Escauriaza [E] for the elliptic equation and its application to the heat equation.
In the original paper dealing with doubling property and frequency function,N. Garofalo and F.H. Lin [GaL] study the monotonicity property of the followingquantity
rRB0;r
jrv (y)j2 dyR@B0;r
jv (y)j2 d� (y).
However, it seems more natural in our context to consider the monotonicityproperties of the frequency function (see [Ze]) de�ned byR
B0;rjrv (y)j2
�r2 � jyj2
�dyR
B0;rjv (y)j2 dy
.
6.1 Monotonicity formula
Following the ideas of I. Kukavica ([Ku2], [Ku], [KN], see also [E], [AE]), oneobtains the following three lemmas. Detailed proofs are given in [Ph3].
Lemma A .- Let D � RN+1, N � 1, be a connected bounded open setsuch that Byo;Ro � D with yo 2 D and Ro > 0. If v = v (y) 2 H2 (D) is asolution of �yv = 0 in D, then
� (r) =
RByo;r
jrv (y)j2�r2 � jy � yoj2
�dyR
Byo;rjv (y)j2 dy
is non-decreasing on 0 < r < Ro , and
d
drlnZByo;r
jv (y)j2 dy = 1
r(N + 1 + � (r)) .
Lemma B .- Let D � RN+1, N � 1, be a connected bounded open setsuch that Byo;Ro
� D with yo 2 D and Ro > 0. Let r1, r2, r3 be three realnumbers such that 0 < r1 < r2 < r3 < Ro. If v = v (y) 2 H2 (D) is a solutionof �yv = 0 in D, thenZ
Byo;r2
jv (y)j2 dy � Z
Byo;r1
jv (y)j2 dy!� Z
Byo;r3
jv (y)j2 dy!1��
,
21
where � = 1ln r2r1
�1
ln r2r1+ 1
ln r3r2
��12 (0; 1).
The above two results are still available when we are closed to a part � ofthe boundary @ under the homogeneous Dirichlet boundary condition on �,as follows.
Lemma C .- Let D � RN+1, N � 1, be a connected bounded open setwith boundary @D. Let � be a non-empty Lipschitz open subset of @D. Letro, r1, r2, r3, Ro be �ve real numbers such that 0 < r1 < ro < r2 < r3 < Ro.Suppose that yo 2 D satis�es the following three conditions:
i). Byo;r \D is star-shaped with respect to yo 8r 2 (0; Ro) ,ii). Byo;r � D 8r 2 (0; ro) ,iii). Byo;r \ @D � � 8r 2 [ro; Ro) .
If v = v (y) 2 H2 (D) is a solution of �yv = 0 in D and v = 0 on �, then
ZByo;r2
\Djv (y)j2 dy �
ZByo;r1
jv (y)j2 dy!� Z
Byo;r3\Djv (y)j2 dy
!1��
where � = 1ln r2r1
�1
ln r2r1+ 1
ln r3r2
��12 (0; 1).
6.1.1 Proof of Lemma B
Let
H (r) =
ZByo;r
jv (y)j2 dy .
By applying Lemma A, we know that
d
drlnH (r) =
1
r(N + 1 + � (r)) .
Next, from the monotonicity property of �, one deduces the following two in-equalities
ln�H(r2)H(r1)
�=R r2r1
N+1+�(r)r dr
� (N + 1 + � (r2)) ln r2r1 ,
ln�H(r3)H(r2)
�=R r3r2
N+1+�(r)r dr
� (N + 1 + � (r2)) ln r3r2 .
Consequently,
ln�H(r2)H(r1)
�ln r2r1
� (N + 1) + � (r2) �ln�H(r3)H(r2)
�ln r3r2
,
22
and therefore the desired estimate holds
H (r2) � (H (r1))� (H (r3))1�� ,
where � = 1ln r2r1
�1
ln r2r1+ 1
ln r3r2
��1.
6.1.2 Proof of Lemma A
We introduce the following two functions H and D for 0 < r < Ro :
H (r) =RByo;r
jv (y)j2 dy ,D (r) =
RByo;r
jrv (y)j2�r2 � jy � yoj2
�dy .
First, the derivative of H (r) =R r0
RSNjv (�s+ yo)j2 �Nd�d� (s) is given by
H 0 (r) =R@Byo;r
jv (y)j2 d� (y). Next, recall the Green formula
R@Byo;r
jvj2 @�Gd� (y)�R@Byo;r
@�
�jvj2�Gd� (y)
=RByo;r
jvj2�Gdy �RByo;r
��jvj2�Gdy .
We apply it with G (y) = r2 � jy � yoj2 where Gj@Byo;r= 0, @�Gj@Byo;r
= �2r,and �G = �2 (N + 1). It gives
H 0 (r) = 1r
RByo;r
(N + 1) jvj2 dy + 12r
RByo;r
��jvj2��r2 � jy � yoj2
�dy
= N+1r H (r) + 1
r
RByo;r
div (vrv)�r2 � jy � yoj2
�dy
= N+1r H (r) + 1
r
RByo;r
�jrvj2 + v�v
��r2 � jy � yoj2
�dy .
Consequently, when �yv = 0,
H 0 (r) =N + 1
rH (r) +
1
rD (r) , (A.1)
that is H0(r)H(r) =
N+1r + 1
rD(r)H(r) the second equality in Lemma A.
Now, we compute the derivative of D (r).
D0 (r) = ddr
�r2R r0
RSN
���(rv)j�s+yo ���2 �Nd�d� (s)��RSNr2���(rv)jrs+yo ���2 rNd� (s)
= 2rR r0
RSN
���(rv)j�s+yo ���2 �Nd�d� (s)= 2r
RByo;r
jrvj2 dy .
(A.2)
23
On the other hand, we have by integrations by parts that
2rRByo;r
jrvj2 dy = N+1r D (r) + 4
r
RByo;r
j(y � yo) � rvj2 dy� 1r
RByo;r
rv � (y � yo)�v�r2 � jy � yoj2
�dy .
(A.3)
Therefore,
(N + 1)RByo;r
jrvj2�r2 � jy � yoj2
�dy
= 2r2RByo;r
jrvj2 dy � 4RByo;r
j(y � yo) � rvj2 dy+2RByo;r
(y � yo) � rv�v�r2 � jy � yoj2
�dy ,
and this is the desired estimate (A.3). Consequently, from (A.2) and (A.3), weobtain, when �yv = 0, the following formula
D0 (r) =N + 1
rD (r) +
4
r
ZByo;r
j(y � yo) � rvj2 dy . (A.4)
The computation of the derivative of � (r) = D(r)H(r) gives
�0 (r) =1
H2 (r)[D0 (r)H (r)�D (r)H 0 (r)] ,
which implies using (A.1) and (A.4) that
H2 (r) �0 (r) =1
r
4
ZByo;r
j(y � yo) � rvj2 dyH (r)�D2 (r)
!� 0 ,
indeed, thanks to an integration by parts and using Cauchy-Schwarz inequality,we have
D2 (r) = 4�R
Byo;rvrv � (y � yo) dy
�2� 4
�RByo;r
j(y � yo) � rvj2 dy��R
Byo;rjvj2 dy
�� 4
�RByo;r
j(y � yo) � rvj2 dy�H (r) .
Therefore, we have proved the desired monotonicity for � and this completesthe proof of Lemma A.
6.1.3 Proof of Lemma C
Under the assumption Byo;r \ @D � � for any r 2 [ro; Ro), we extend v by zeroin Byo;Ro nD and denote by v its extension. Since v = 0 on �, we have8<: v = v1D in Byo;Ro ,
v = 0 on Byo;Ro \ @D ,rv = rv1D in Byo;Ro
.
24
Now, we denote r = Byo;r\D, when 0 < r < Ro. In particular, r = Byo;r,when 0 < r < ro. We introduce the following three functions:
H (r) =Rrjv (y)j2 dy ,
D (r) =Rrjrv (y)j2
�r2 � jy � yoj2
�dy ,
and
� (r) =D (r)
H (r)� 0 .
Our goal is to show that � is a non-decreasing function. Indeed, we will provethat the following equality holds
d
drlnH (r) = (N + 1)
d
drlnr +
1
r� (r) . (C.1)
Therefore, from the monotonicity of �, we will deduce (in a similar way than inthe proof of Lemma A) that
ln�H(r2)H(r1)
�ln r2r1
� (N + 1) + � (r2) �ln�H(r3)H(r2)
�ln r3r2
,
and this will imply the desired estimateZr2
jv (y)j2 dy � Z
Byo;r1
jv (y)j2 dy!� Z
r3
jv (y)j2 dy!1��
,
where � = 1ln r2r1
�1
ln r2r1+ 1
ln r3r2
��1.
First, we compute the derivative of H (r) =RByo;r
jv (y)j2 dy.
H 0 (r) =RSNjv (rs+ yo)j2 rNd� (s)
= 1r
RSNjv (rs+ yo)j2 rs � srNd� (s)
= 1r
RByo;r
div�jv (y)j2 (y � yo)
�dy
= 1r
RByo;r
�(N + 1) jv (y)j2 +r jv (y)j2 � (y � yo)
�dy
= N+1r H (r) + 2
r
Rrv (y)rv (y) � (y � yo) dy .
(C.2)
Next, when �yv = 0 in D and vj� = 0, we remark that
D (r) = 2
Zr
v (y)rv (y) � (y � yo) dy , (C.3)
indeed,Rrjrvj2
�r2 � jy � yoj2
�dy
=Rrdivhvrv
�r2 � jy � yoj2
�idy �
Rrvdiv
hrv�r2 � jy � yoj2
�idy
= �Rrv�v
�r2 � jy � yoj2
�dy �
Rrvrv � r
�r2 � jy � yoj2
�dy
because on @Byo;r, r = jy � yoj and vj� = 0= 2
Rrvrv � (y � yo) dy because �yv = 0 in D .
25
Consequently, from (C.2) and (C.3), we obtain
H 0 (r) =N + 1
rH (r) +
1
rD (r) , (C.4)
and this is (C.1).
On another hand, the derivative of D (r) is
D0 (r) = 2rR r0
RSN
���(rv)j�s+yo ���2 �Nd�d� (s)= 2r
Rrjrv (y)j2 dy .
(C.5)
Here, when �yv = 0 in D and vj� = 0, we will remark that
2rRrjrv (y)j2 dy = N+1
r D (r) + 4r
RByo;r
j(y � yo) � rv (y)j2 dy+ 1r
R�\Byo;r
j@�vj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
(C.6)indeed,
(N + 1)Rrjrvj2
�r2 � jy � yoj2
�dy
=Rrdiv�jrvj2
�r2 � jy � yoj2
�(y � yo)
�dy
�Rrr�jrvj2
�r2 � jy � yoj2
��� (y � yo) dy
=R�\Byo;r
jrvj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
�Rr@yi
�jrvj2
�r2 � jy � yoj2
��(yi � yoi) dy
=R�\Byo;r
jrvj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
�Rr2rv@yirv
�r2 � jy � yoj2
�(yi � yoi) dy
+2Rrjrvj2 jy � yoj2 dy ,
and �Rr@yjv@
2yiyjv
�r2 � jy � yoj2
�(yi � yoi) dy
= �Rr@yj
�(yi � yoi) @yjv@yiv
�r2 � jy � yoj2
��dy
+Rr@yj (yi � yoi) @yjv@yiv
�r2 � jy � yoj2
�dy
+Rr(yi � yoi) @2yjv@yiv
�r2 � jy � yoj2
�dy
+Rr(yi � yoi) @yjv@yiv@yj
�r2 � jy � yoj2
�dy
= �R�\Byo;r
�j
�(yi � yoi) @yjv@yiv
�r2 � jy � yoj2
��d� (y)
+Rrjrvj2
�r2 � jy � yoj2
�dy
+0 because �yv = 0 in D�Rr2 j(y � yo) � rvj2 dy .
26
Therefore, when �yv = 0 in D, we have
(N + 1)Rrjrvj2
�r2 � jy � yoj2
�dy
=R�\Byo;r
jrvj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
�2R�\Byo;r
@yjv�j ((yi � yoi) @yiv)�r2 � jy � yoj2
�d� (y)
+2r2Rrjruj2 dy � 4
Rrj(y � yo) � rvj2 dy .
By using the fact that vj� = 0, we get rv = (rv � �) � on � and we deduce that
(N + 1)Rrjrvj2
�r2 � jy � yoj2
�dy
= �R�\Byo;r
j@�vj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
+2r2Rrjrvj2 dy � 4
Rrj(y � yo) � rvj2 dy ,
and this is (C.6). Consequently, from (C.5) and (C.6), when �yv = 0 in D andvj� = 0, we have
D0 (r) = N+1r D (r) + 4
r
Rrj(y � yo) � rv (y)j2 dy
+ 1r
R�\Byo;r
j@�vj2�r2 � jy � yoj2
�(y � yo) � �d� (y) .
(C.7)
The computation of the derivative of � (r) = D(r)H(r) gives
�0 (r) =1
H2 (r)[D0 (r)H (r)�D (r)H 0 (r)] ,
which implies from (C.4) and (C.7), that
H2 (r) �0 (r) = 1r
�4Rrj(y � yo) � rv (y)j2 dy H (r)�D2 (r)
�+H(r)
r
R�\Byo;r
j@�vj2�r2 � jy � yoj2
�(y � yo) � �d� (y)
Thanks to (C.3) and Cauchy-Schwarz inequality, we obtain that
0 � 4Zr
j(y � yo) � rv (y)j2 dy H (r)�D2 (r) .
The inequality 0 � (y � yo) � � on � holds when Byo;r \D is star-shaped withrespect to yo for any r 2 (0; Ro). Therefore, we get the desired monotonicityfor � which completes the proof of Lemma C.
27
6.2 Quantitative unique continuation property for the Lapla-cian
Let D � RN+1, N � 1, be a connected bounded open set with boundary @D.Let � be a non-empty Lipschitz open part of @D. We consider the Laplacian inD, with a homogeneous Dirichlet boundary condition on � � @:8<: �yv = 0 in D ,
v = 0 on � ,v = v (y) 2 H2 (D) .
(D.1)
The goal of this section is to describe interpolation inequalities associated tosolutions v of (D.1).
Theorem D .- Let ! be a non-empty open subset of D. Then, for anyD1 � D such that @D1 \ @D b � and D1 n(� \ @D1) � D, there exist C > 0and � 2 (0; 1) such that for any v solution of (D.1), we haveZ
D1
jv (y)j2 dy � C�Z
!
jv (y)j2 dy���Z
D
jv (y)j2 dy�1��
.
Or in a equivalent way,
Theorem D�.- Let ! be a non-empty open subset of D. Then, for anyD1 � D such that @D1 \ @D b � and D1 n(� \ @D1) � D, there exist C > 0and � 2 (0; 1) such that for any v solution of (D.1), we haveZ
D1
jv (y)j2 dy � C�1
"
� 1���Z!
jv (y)j2 dy + "ZD
jv (y)j2 dy 8" > 0 .
Proof of Theorem D .- We divide the proof into two steps.
Step 1 .- We apply Lemma B, and use a standard argument (see e.g., [Ro])which consists to construct a sequence of balls chained along a curve. Moreprecisely, we claim that for any non-empty compact sets in D, K1 and K2, suchthat meas(K1) > 0, there exists � 2 (0; 1) such that for any v = v (y) 2 H2 (D),solution of �yv = 0 in D, we haveZ
K2
jv (y)j2 dy ��Z
K1
jv (y)j2 dy���Z
D
jv (y)j2 dy�1��
. (D.2)
Step 2 .- We apply Lemma C, and choose yo in a neighborhood of the part� such that the conditions i, ii, iii, hold. Next, by an adequate partition ofD, we deduce from (D.2) that for any D1 � D such that @D1 \ @D b �
28
and D1 n(� \ @D1) � D, there exist C > 0 and � 2 (0; 1) such that for anyv = v (y) 2 H2 (D) such that �yv = 0 on D and v = 0 on �, we haveZ
D1
jv (y)j2 dy � C�Z
!
jv (y)j2 dy���Z
D
jv (y)j2 dy�1��
.
This completes the proof.
6.3 Quantitative unique continuation property for the el-liptic operator @2t +�
In this section, we present the following result.
Theorem E .- Let be a bounded open set in Rn, n � 1, either convexor C2 and connected. We choose T2 > T1 and � 2 (0; (T2 � T1) =2). Letf 2 L2 (� (T1; T2)). We consider the elliptic operator of second order in� (T1; T2) with a homogeneous Dirichlet boundary condition on @� (T1; T2),8<: @2tw +�w = f in � (T1; T2) ,
w = 0 on @� (T1; T2) ,w = w (x; t) 2 H2 (� (T1; T2)) .
(E.1)
Then, for any ' 2 C10 (� (T1; T2)), ' 6= 0, there exist C > 0 and � 2 (0; 1)such that for any w solution of (E.1), we haveR T2��
T1+�
Rjw (x; t)j2 dxdt
� C�R T2
T1
Rjw (x; t)j2 dxdt
�1���R T2T1
Rj'w (x; t)j2 dxdt+
R T2T1
Rjf (x; t)j2 dxdt
��.
Proof .- First, by a di¤erence quotient technique and a standard extension at� fT1; T2g, we check the existence of a solution u 2 H2 (� (T1; T2)) solving�
@2t u+�u = f in � (T1; T2) ,u = 0 on @� (T1; T2) [ � fT1; T2g ,
such thatkukH2(�(T1;T2)) � c kfkL2(�(T1;T2)) ,
for some c > 0 only depending on (; T1; T2). Next, we apply Theorem D withD = � (T1; T2), � (T1 + �; T2 � �) � D1, y = (x; t), �y = @2t + �, andv = w � u.
29
6.4 The heat equation and the Hölder continuous depen-dence from one point in time
In this section, we presents the following result.
Theorem F .- Let be a bounded open set in Rn, n � 1, either convexor C2 and connected. Let ! be a nonempty open subset of , and T > 0. Thenthere are C > 0 and � 2 (0; 1) such that any solution to8<: @tu��u = 0 in � (0; T ) ,
u = 0 on @� (0; T ) ,u (�; 0) 2 L2 () ,
(F.1)
satis�es, for any to 2 (0; T ),Z
ju (x; to)j2 dx � C�eCto
Z
ju (x; 0)j2 dx�1���Z
!
ju (x; to)j2 dx��
.
Proof .- We divide the proof into three steps.
Step 1 .- Let �1,�2,� � � and e1,e2,� � � be the eigenvalues and eigenfunctionsof the Laplacian �� in H1
0 (), constituting an orthonormal basis in L2 ().
For any uo = u (�; 0) =Pj�1
�jej in L2 () with �j =Ruoejdx, the solution
u to (F.1), can be written as u (x; t) =Pj�1
�jej (x) e��jt. Let to 2 (0; T ). We
introduce (see [L] or [CRV]) the function
w (x; t) =Xj�1
�jej (x) e��jtoch
�p�jt�� � (x)
Xj�1
�jej (x) e��jto ,
where � 2 C10 (!), � = 1 in e! b !. Recall that cht = (et + e�t) =2. Then, wsolves 8>><>>:
@2tw +�w = �f in � (0; T ) ,w = 0 on @� (0; T ) ,w = @tw = 0 in e! � f0g ,w = w (x; t) 2 H2 (� (0; T )) ,
where f = ��u (�; to) 2 H20 ().
We denote by w the extension of w to zero in e! � (�T; 0). Then, w solves8<: @2tw +�w = �f1�(0;T ) in � (0; T ) [ e! � (�T; 0) ,w = 0 in e! � (�T; 0) ,w = 0 on @� (0; T ) .
Now, we construct D, an open connected set in RN+1, satisfying the followingsix conditions:
30
i). � (�; T � �) � D with � 2 (0; T=2) ;ii). @� (�; T � �) � @D ;iii). D � � (0; T ) [ e! � (�T; 0) ;iv). there is an nonempty open !o b D \ e! � (�To; 0) with To 2 (0; T ) ;v). D 2 C2 when is of class C2 and connected ;vi). D is convex with an adequate choice of the pair (�; To) when is convex
.
In particular, w 2 H2 (D).
Step 2 .- We claim that there is g 2 H2 (� (�T; T ))\H10 (� (�T; T )) �
H2 (D) such that�@2t g +�g = �f1�(0;T ) in � (�T; T ) ,g = 0 on @ (� (�T; T )) = @� (�T; T ) [ � f�Tg ,
andkgkL2(D) � kfkL2(�(0;T )) . (F.2)
Indeed, we will proceed in six substep when is of class C2 and connected (thecase where is convex is well-known because � (�T; T ) is then convex). Wede�ne h = �f1�(0;T ) 2 L2 (� (�T; T )).Substep 1: recall that h 2 L2 (� (�T; T )) implies the existence of such
g 2 H10 (� (�T; T )).
Substep 2: thanks to the interior regularity theorem of elliptic systems, forany D0 b � (�T; T ), g 2 H2 (D0).Substep 3: thanks to the boundary regularity theorem of elliptic systems,
but far for � f�T; Tg, g is locally in H2 because is of class C2.Substep 4: we extend the solution at t = T as follows.
Let h (x; t) = h (x; t) when (x; t) 2 � (�T; T ) and h (x; t) = �h (x; 2T � t)when (x; t) 2 � (T; 3T ). Then h 2 L2 (� (�T; 3T )). Let g (x; t) = g (x; t)when (x; t) 2 � [�T; T ) and g (x; t) = �g (x; 2T � t) when (x; t) 2 � [T; 3T ].Then, g solves�
@2t g +�g = h in � (�T; 3T ) ,g = 0 on @ (� (�T; 3T )) = @� (�T; 3T ) [ � f�T; 3Tg .
By applying the boundary regularity theorem of elliptic systems as in substep3, we get g 2 H2 (� (0; 2T )). In particular, g 2 H2 (� (0; T )).Substep 5: we extend in a similar way the solution at t = �T in order to
conclude that g 2 H2 (� (�T; 0)).Substep 6: Finally, we multiply @2t g+�g = h by (��)
�1g and integrate by
parts over � (�T; T ), to getR T�T k@tgk
2H�1() dt+ kgk
2L2(�(�T;T ))
=R T0
R��f (x) (��)�1 g (x; t) dxdt
=R T0
Rf (x) (��) (��)�1 g (x; t) dxdt because f 2 H2
0 ()� kfkL2(�(0;T )) kgkL2(�(�T;T )) by Cauchy-Schwarz .
31
This gives the desired inequality (F.2).
Step 3 .- Finally, we apply Theorem D with �y = @2t +�, v = w � g in Dwhere � = @ � (�; T � �) and � (�=2 + T=4; 3T=4� �=2) � D1 � D suchthat @D1 \ @D b � and D1 n(� \ @D1) � D in order that
ZD1
jw � gj2 dy � C�1
"
� 1���Z!o
jw � gj2 dy + "ZD
jw � gj2 dy 8" > 0 ,
which impliesZD1
jwj2 dy � C�1
"
� 1���ZD
jgj2 dy + "ZD
jwj2 dy 8" 2 (0; 1) ,
where we used the fact that w = 0 in !o. By (F.2), we conclude that there existC > 0 and � 2 (0; 1) such thatR 3T=4��=2
�=2+T=4
Rjw (x; t)j2 dxdt � C
�1"
� 1���R T0
Rjf (x)j2 dxdt
+"R T0
Rjw (x; t)j2 dxdt 8" > 0 .
Consequently, there exist � 2 (0; T=2), C > 0 and � 2 (0; 1) such thatR T���
Rjw (x; t)j2 dxdt � C
�1"
� 1���R T0
Rjf (x)j2 dxdt
+"R T0
Rjw (x; t)j2 dxdt 8" > 0 .
On the other hand, the following four inequalities hold.Z T
0
Z
jf (x)j2 dxdt � TZ!
j� (x)u (x; to)j2 dx ,
Z T
0
Z
jw (x; t)j2 dxdt � 2TXj�1
�2je�2(�jto�
p�jT) + 2T
Z!
j� (x)u (x; to)j2 dx ,
Pj�1
�2je�2(�jto�
p�jT) =
Pfj�1;
p�j� T
tog�2je
�2(�jto�p�jT)
+P
fj�1;p�j>
Ttog�2je
�2(�jto�p�jT)
� e2T2
to
Pj�1
�2j ,
R T���
R
�����Pj�1�jej (x) e��jtoch �p�jt������2
dxdt � 2R T���
Rjw (x; t)j2 dxdt
+2TR!j� (x)u (x; to)j2 dx ,
32
We deduce from the above �ve inequalities that for any " > 0,
(T � 2�)Pj�1
�2je�2�jto � (T � 2�)
Pj�1
�2je�2(�jto�
p�j�)
�R T���
R
�����Pj�1�jej (x) e��jtoch �p�jt������2
dxdt
� 2TC�1"
� 1���R!j� (x)u (x; to)j2 dx
+4T"
e2
T2
to
Pj�1
�2j +R!j� (x)u (x; to)j2 dx
!+2T
R!j� (x)u (x; to)j2 dx .
Finally, there is C > 0 such that for any to > 0,Rju (x; to)j2 dx =
Pj�1
�2je�2�jto
� C�1"
� 1���R!ju (x; to)j2 dx+ "e
Cto
Rju (x; 0)j2 dx 8" 2 (0; 1) ,
which implies the desired estimate of Theorem F,Z
ju (x; to)j2 dx � C�eC(1��)
to
�Z
ju (x; 0)j2 dx�1���Z
!
ju (x; to)j2 dxdt��
.
References
[AE] V. Adolfsson and L. Escauriaza, C1;� domains and unique continuationat the boundary, Comm. Pure Appl. Math. 50 (1997) 935-969.
[BT] C. Bardos, L. Tartar, Sur l�unicité retrograde des équations paraboliqueset quelques questions voisines, Arch. Rational Mech. Anal. 50 (1973)10-25.
[CRV] B. Canuto, E. Rosset, S. Vessella, Quantitative estimates of unique con-tinuation for parabolic equations and inverse initial-boundary value prob-lems with unknown boundaries, Trans. Amer. Math. Soc. 354 (2001) 491-535.
[E] L. Escauriaza, Doubling property and linear combinations of eigenfunc-tions, manuscript.
[EFV] L. Escauriaza, F.J. Fernandez, S. Vessella, Doubling properties of caloricfunctions, Applicable analysis 85 (1-2) (2006) 205-223.
33
[FI] A.V. Fursikov, O.Yu. Imanuvilov, Controllability of evolution equations.Lecture Notes Series, 34. Seoul National University, Research Instituteof Mathematics, Global Analysis Research Center, Seoul, 1996.
[FEF] X-L. Feng, L. Eldén, C-L Fu, Stability and regularization of a backwardparabolic PDE with variable coe¢ cients, Journal of Inverse and Ill-PosedProblems 18 (2010) 217-243.
[GaL] N. Garofalo and F H. Lin, Monotonicity properties of variational inte-grals, Ap-weights and unique continuation, Indiana Univ. Math. J. 35(1986) 245-268.
[K] C. Kenig, Quantitative unique continuation, logarithmic convexity ofgaussian means and Hardy�s uncertainty principle, Proc. of Symp. ofPure Math., volume in celebration of V. Mazya�s 70 birthday, 79, 2008,p.207-227.
[KT] H. Koch, D. Tataru, Carleman estimates and unique continuation forsecond order parabolic equations with non smooth coe¢ cients, Comm.Partial Di¤erential Equations 34 (4) (2009) 305-366.
[Ku] I. Kukavica, Level sets for the stationary Ginzburg-Landau equation,Calc. Var. 5 (1997) 511-521.
[Ku2] I. Kukavica, Quantitative uniqueness for second order elliptic operators,Duke Math. J. 91 (1998) 225-240.
[KN] I. Kukavica, K. Nyström, Unique continuation on the boundary for Dinidomains, Proc. of Amer. Math. Soc. 126 (1998) 441-446.
[Le] M. Léautaud, Spectral inequalities for non-selfadjoint elliptic operatorsand application to the null-controllability of parabolic systems, Journalof Functional Analysis 258 (2010) 2739�2778.
[LR] G. Lebeau, L. Robbiano, Contrôle exact de l�équation de la chaleur,Comm. Part. Di¤. Eq. 20 (1995) 335-356.
[Li] F.H. Lin, Remarks on a backward parabolic problem, Methods and Ap-plications of Analysis 10 (2) (2003) 245-252.
[L] F.H. Lin, A uniqueness theorem for the parabolic equations, Comm. PureAppl. Math. 43 (1990) 127-136.
[M] L.Miller, A direct Lebeau-Robbiano strategy for the observability of heat-like semigroups, Discrete and Continuous Dynamical Systems Series B14 (4) (2010) 1465-1485.
[PW] K. D. Phung , G. Wang , Quantitative unique continuation for the semi-linear heat equation in a convex domain, Journal of Functional Analysis259 (5) (2010) 1230-1247.
34
[PW2] K. D. Phung , G. Wang, An observability for parabolic equations frommeasurable set in time and its applications, in preparation.
[P] C.C. Poon, Unique continuation for parabolic equations, Comm. PartialDi¤erential Equations 21 (3-4) (1996) 521-539.
[Ph3] K.-D. Phung, Observation et stabilisation d�ondes: géométrie et coût ducontrôle, Habilitation à diriger des recherches, (2007).
[Ro] L. Robbiano, Théorème d�unicité adapté au contrôle des solutions desproblèmes hyperboliques, Comm. Part. Di¤. Eq. 16 (1991) 789-800.
[W] G. Wang, L1-null controllability for the heat equation and its conse-quences for the time optimal control problem, SIAM J. Control Optim.47 (2008) 1701-1720.
[Ze] S. Zelditch, Local and global analysis of eigenfunctions on Riemannianmanifolds, arXiv:09033420.
35