Abstract
We present a simple and self-contained approach to establish the unique continuation property for some classical evolution equations of second order in a cylindrical domain. We namely discuss this property for wave, parabolic and Schödinger operators with time-independent principal part. Our method is builds on two-parameter Carleman inequalities combined with unique continuation across a pseudo-convex hypersurface with respect to the space variable. The most results we demonstrate in this work are more or less classical. Some of them are not stated exactly as in their original form.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let D be a domain of \({\mathbb {R}}^d\), \(d\ge 1\). We recall that \(f\in C^\infty (D)\) is said real-analytic if its Taylor series around any arbitrary point of D converges in a ball centered at this point. It is known that a real analytic function possesses the unique continuation property which means that if f vanishes in a nonempty open subset \(D_0\) of D then f must vanishes identically (see for instance [22, Theorem, page 65]). Functions satisfying this unique continuation property are also called quasi-analytic.
A classical result shows that a strong solution of an elliptic operator with smooth principal coefficients is quasi-analytic. In the present work we consider the analogue of this property for second order evolution PDEs. The right property in this context should be the following: if a solution of an evolution equation of second order in the cylindrical domain \(D\times (t_1,t_2)\) vanishes in \(D_0\times (t_1,t_2)\), for some nonempty open subset \(D_0\) of D, then this solution must vanishes identically. Unfortunately this property holds only in the parabolic case. Less optimal result still holds for the Schödinger case. The worst case is that for wave equations for which we have only a weak version of unique continuation property, even for a large time interval.
We provide a simple and self contained approach to show the unique continuation property for wave, parabolic and Schrödinger equations. The approach we carry out is quite classical and it is based on two-parameter Carleman inequalities. The unique continuation property in each case is obtained as a consequence of the property of unique continuation across a non characteristic hypersurface satisfying in addition a pseudo-convexity condition in the case of wave and Schrödinger equations.
The core of our analysis consists in establishing two-parameter Carleman inequalities. We follow a classical scheme for obtaining these \(L^2\)-weighted energy estimates, essentially based on conjugating the original operator with a well chosen exponential function, splitting the resulting operator into its self-adjoint part and skew-adjoint part and finally making integrations by parts. The main assumption on the weight function is a pseudo-convexity condition with respect to the operator under consideration. A systematic approach was considered by Hörmander [19, Section 28.2, page 234] for a general operator P of an arbitrary order m where the pseudo-convexity condition is expressed in term of the principal symbol of P. The method we develop in this work is more simple and does not appeal to fine analysis of PDEs and our results have no pretension nor for generality neither for optimality.
It is worth remarking that splitting the conjugated operator into self-adjoint and skew-adjoint parts is not the best possible way to get two-parameter Carleman inequalities for elliptic and parabolic operators. There is a particular way to split the conjugated operator into two parts. However this particular decomposition is not applicable for wave and Schrödinger equations. But this is not really surprising since solutions of wave and Schrödinger equations do not enjoy the same regularity properties of solutions of elliptic and parabolic equations.
We choose to start with the more subtle case corresponding to the wave equation. Since most calculations to get Carleman inequalities are common for different type of equations, Carleman inequalities for parabolic and Schrödinger equations are obtained by making some modifications in the proof of the Carleman inequality for the wave equation. We also added a short section for the elliptic case whose analysis is almost similar to that of the parabolic case.
One can find in the literature two-parameter Carleman inequalities with degenerate weight function. We refer for instance to [8, 12, 13, 15, 17, 25] for parabolic operators and [3,4,5, 26, 27] for Schrödinger operators.
A Carleman inequality for wave equations on compact Riemannian manifold can be found in [7] and quite recently Huang [20] proved a Carleman inequality for a general wave operators with time-dependent principal part. The interested reader is referred to [16] for a unified approach establishing Carleman inequalities for second order PDEs and their applications to control theory and inverse problems.
We point out that it is possible to derive Carleman inequalities with single parameter in the weight function. We refer for instance to [18, 19, 23, 38, 39] and references therein for more details. Introducing a second parameter in the Carleman weight is a simple way to guarantee pseudo-convexity condition as it is remarked in [18]. It certainly contributed to the development of Carleman inequalities with two-parameter. Two-parameter Carleman inequalities appear to be more flexible then one-parameter Carleman inequalities since we have to our disposal two parameters. Of course, one-parameter Carleman inequalities can be deduced from two-parameter Carleman inequalities (see for instance Theorem 3.3 for the case of the wave equation).
The unique continuation property for elliptic and parabolic operators with unbounded lower order coefficients was obtained in [32,33,34,35]. The analysis in these references combine both classical tools used for establishing the property of unique continuation together with interpolation inequalities. Uniqueness and non-uniqueness for general operators were discussed in [1, 2, 40] (see also the references therein). We also mention [21, 28] as additional references on uniqueness of Cauchy problems.
We also discuss briefly observability inequalities which can be seen as the quantification of unique continuation for the Cauchy problem associated to IBVPs. We refer to [24] for general observability inequalities for wave and Schrödinger equations with arbitrary interior or boundary observation region. The reader can find in this work a detailed introduction to explain the main steps to get the property of unique continuation for an intermediate case between Holmgren’s analytic case and Hörmander’s general case for operators with partially analytic coefficients.
A more difficult problem consists in quantifying the property of unique continuation from an interior subdomain or the Cauchy data on a sub-boundary. The elliptic case is now almost completely solved with optimal results for \(C^{1,\alpha }\)-solutions and \(C^{0,1}\)-domains [11] or \(H^2\)-solutions and \(C^{1,1}\)-domains [9]. A non optimal result for \(H^2\)-solutions and \(C^{0,1}\)-domains was obtained in [10]. These kind of results can be obtained by a method based on three-sphere inequality which is deduced itself from a Carleman inequality. The case of parabolic and wave equations is extremely more difficult than the elliptic case. Concerning parabolic equations, a first result was obtained in [8] with Cauchy data in a particular subboundary. This result is based on a global Carleman inequality. The general case was tackled in [13] where a non optimal result was established using a three-cylinder inequality. A result for the wave equation was recently proved in [6]. This result was obtained via Fourier–Bros–Iagolnitzer transform allowing to transfer the quantification of unique continuation of an elliptic equation to that of the wave equation.
2 Preliminaries
2.1 Main notations and assumptions
Throughout \(\Omega \) is bounded Lipschitz domain of \({\mathbb {R}}^n\), \(n\ge 2\), with boundary \(\Gamma \). The unit normal exterior vector field on \(\Gamma \) is denoted by \(\nu \).
We set \(Q=\Omega \times (t_1,t_2)\) and \(\Sigma =\Gamma \times (t_1,t_2)\), where \(t_1,t_2\in {\mathbb {R}}\) are fixed so that \(t_1<t_2\).
\(A=(a_{k\ell })\) will denote a symmetric matrix with coefficients \(a_{k\ell }\in C^{2,1}({\overline{\Omega }})\), \(1\le k,\ell \le n\), and there exist two constants \({\mathfrak {m}}>0\) and \(\varkappa \ge 1\) so that
and
The set of such matrices will denoted in the rest of this text by \({\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\).
It is worth mentioning that, according to Rademacher’s theorem, \(C^{2,1}({\overline{\Omega }})\) is continuously embedded in \(W^{3,\infty }(\Omega )\).
For \(\xi ,\eta \in {\mathbb {C}}^n\), \((\xi |\eta )\) and \(\xi \otimes \eta \) are defined as usual respectively by
The Jacobian matrix and the Hessian matrix are denoted respectively by
The norm (resp. the scalar product) of a Banach (resp. a Hilbert) space E is always denoted by \(\Vert \cdot \Vert _E\) (resp. \(\langle \cdot |\cdot \rangle _E\)).
We recall that the anisotropic Sobolev space \(H^{2,1}(Q)\) is defined as follows
The following notations will be useful in the sequel, where \(S=\Omega \) or \(S=\Gamma \),
We readily obtain from the above definitions the following identity
Furthermore, the Green type formula
holds for any \(u\in H^2(\Omega ;{\mathbb {C}} )\) and \(v\in H^1(\Omega ;{\mathbb {C}})\).
We need also to introduce the following notations
The \(n\times n\) identity matrix will denoted by \(\mathbf{I }\).
We shall use for notational convenience the following notation
Finally, we equip \(\partial Q\) with following measure
where \({ d}\sigma (x)\) is the Lebesgue measure on \(\Gamma \) and \(\delta _t\) is the Dirac measure at t.
We used above \(\mathbbm {1}_X\) to denote the characteristic function of the measurable set X:
2.2 Pseudo-convexity condition
Define, for \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\),
where \(x\in {\overline{\Omega }}\) and \(1\le k,\ell ,m\le n\).
We associate to \(h\in C^1({\overline{\Omega }})\) the matrix \(\Upsilon _A (h)\) given by
Note that \(\Upsilon _A (h)\) is not necessarily symmetric.
Inspired by the definition introduced in [19, Section 28.2, page 234] we consider the following one:
Definition 2.1
We say that \(h\in C^2({\overline{\Omega }})\) is A-pseudo-convex with constant \(\kappa >0\) in \(\Omega \) if \(\nabla h(x)\ne 0\) for any \(x\in {\overline{\Omega }}\) and if
where
It worth noticing that \(A\rightarrow \Theta _A\) is positively homogenous of degree two:
Since \(\Theta _{\mathbf{I }}(h)=2\nabla ^2h\), h is \(\mathbf{I }\)-pseudo-convex in \(\Omega \) if \(\nabla h(x)\ne 0\) and \(\nabla ^2h(x)\) is positive definite for any \(x\in {\overline{\Omega }}\). In other words, when \(A=\mathbf{I }\) pseudo-convexity is reduced to local strict convexity.
2.3 Carleman weights
It will be convenient to define the notion of Carleman weight for different kind of operators we are interested in. In the rest of this paper \(\psi =\psi (x,t)\) is a function of the form
and \(\phi =e^{\lambda \psi }\), \(\lambda >0\).
Definition 2.2
-
(a)
Let \(0\le \psi _0 \in C^4({\overline{\Omega }})\). We say that \(\phi _0=e^{\lambda \psi _0}\), \(\lambda >0\), is a weight function for the elliptic operator \({\mathcal {L}}_{A,0}^e\) if \(\nabla \psi _0(x)\ne 0\) for any \(x\in {\overline{\Omega }}\).
-
(b)
If \(0\le \psi \in C^4({\overline{Q}})\) and \(\nabla \psi _0(x)\ne 0\), for any \(x\in {\overline{\Omega }}\), we say that \(\phi \) is a weight function for the parabolic operator \({\mathcal {L}}_{A,0}^p\).
-
(c)
Assume that \(0\le \psi \in C^4({\overline{Q}})\). Then \(\phi \) is said a weight function for the Schrödinger operator \({\mathcal {L}}_{A,0}^s\) if \(\psi _0\) is A-pseudo-convex in \(\Omega \).
-
(d)
We say that \(\phi \), with \(0\le \psi \in C^4({\overline{Q}})\), is a weight function for the wave operator \({\mathcal {L}}_{A,0}^w\) if \(\psi _0\) is A-pseudo-convex with constant \(\kappa >0\) in \(\Omega \) and if, in addition, the following two conditions hold:
$$\begin{aligned}&\min _{{\overline{Q}}}\left[ |\nabla \psi _0|_A^2-(\partial _t\psi _1)^2\right] ^2>0, \end{aligned}$$(2.2)$$\begin{aligned}&|\partial _t^2\psi _1|\le \varkappa ^{-1}\kappa /4. \end{aligned}$$(2.3)
Example 2.1
Fix \(t_0\in {\mathbb {R}}\), \(x_0\in {\mathbb {R}}^n{{\setminus }} {\overline{\Omega }}\) and set, for \(\gamma \in {\mathbb {R}}\),
where the constant C is chosen sufficiently large in order to guarantee that \(\psi \ge 0\). In that case
Let us first discuss A-pseudo-convexity condition of \(\psi _0\) in different cases.
-
(i)
Assume that \(\Omega =B(0,r)\) and \(x_0\in B(0,2r){\setminus } {\overline{B}}(0,r)\). We can then choose r sufficiently small in such a way that
$$\begin{aligned} (\Upsilon _A(\psi _0)\xi |\xi )\ge -\varkappa ^2 |\xi |^2, \end{aligned}$$from which we deduce that
$$\begin{aligned} (\Theta _A(\psi _0)\xi |\xi )\ge \varkappa ^2|\xi |^2. \end{aligned}$$ -
(ii)
As the mapping
$$\begin{aligned} A \in C^{2,1}({\overline{\Omega }} ,{\mathbb {R}}^{n\times n})\mapsto \Upsilon _A(\psi _0)\in C^{1,1}({\overline{\Omega }} ,{\mathbb {R}}^{n\times n}) \end{aligned}$$is continuous in a neighborhood of \(\mathbf{I }\) and \(\Upsilon _{\mathbf{I }}(\psi _0)=0\), we conclude that there exists \({\mathcal {N}}\), a neighborhood of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }} ,{\mathbb {R}}^{n\times n})\), so that, for any \(A\in {\mathcal {N}}\), we have
$$\begin{aligned} (A(x)\xi |\xi )&\ge |\xi |^2/2,\quad x\in {\overline{\Omega }} ,\; \xi \in {\mathbb {R}}^n, \\ (\Upsilon _A(\psi _0)\xi |\xi )&\ge -|\xi |^2/4,\quad x\in {\overline{\Omega }} ,\; \xi \in {\mathbb {R}}^n. \end{aligned}$$Whence
$$\begin{aligned} (\Theta _A(\psi _0)\xi |\xi )\ge |\xi |^2/4,\quad x\in {\overline{\Omega }} ,\; \xi \in {\mathbb {R}}^n, \end{aligned}$$provided that \(A\in {\mathcal {N}}\).
-
(iii)
Consider the particular case in which \(A=a\mathbf{I }\) with \(a\in C^{2,1}({\overline{\Omega }})\) satisfying \(a\ge \varkappa \). Simple computations then yield
$$\begin{aligned} \Upsilon _A(\psi _0)= -a\left( \nabla a |x-x_0\right) \mathbf{I }+2a\nabla a\otimes (x-x_0). \end{aligned}$$In consequence
$$\begin{aligned} (\Theta _A(\psi _0)\xi |\xi )\ge \varkappa \left( 2\varkappa -3\left| \nabla a\right| |x-x_0|\right) |\xi |^2. \end{aligned}$$Hence a condition guaranteeing that \(\psi _0\) is \(a\mathbf{I }\)-pseudo-convex is
$$\begin{aligned} \left| \nabla a\right| |x-x_0|<2\varkappa /3. \end{aligned}$$This condition is achieved for instance if \(\Omega \) has sufficiently small diameter and \(x_0\) is close to \(\Omega \) or else \(|\nabla a|\) is small enough.
Next, we discuss a bound on \(\gamma \) for which (2.2) and (2.3) hold simultaneously. If \(d=\text{ dist }(x_0,{\overline{\Omega }})\) \((>0)\) and \(\sigma =\Vert t+t_0\Vert _{L^\infty ((t_1,t_2))}\), then (2.2) is satisfied whenever
As \(\partial _t^2\psi =\gamma \), we see that both (2.2) and (2.3) are satisfied when
2.4 Pseudo-convex hypersurface
We begin by a lemma concerning the action of an orthogonal transformation on \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\). If \({\mathcal {O}}\) is an orthogonal transformation and \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\), we set \(A_{{\mathcal {O}}}(y)={\mathcal {O}}A({\mathcal {O}}^ty){\mathcal {O}}^t\). Here \({\mathcal {O}}^t\) denotes the transposed matrix of \({\mathcal {O}}\).
The proof of the following lemma is straightforward.
Lemma 2.1
Let \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\) and \({\mathcal {O}}\) an orthogonal transformation. Then \(A_{{\mathcal {O}}}\in {\mathscr {M}}({\mathcal {O}}\Omega ,\varkappa ,{\mathfrak {m}}')\), where \({\mathfrak {m}}'={\mathfrak {m}}'(n,{\mathfrak {m}})>0\) is a constant.
The gradient with respect to the variable \(x'\in {\mathbb {R}}^{n-1}\) or \(y'\in {\mathbb {R}}^{n-1}\) is denoted hereafter by \(\nabla '\).
Let \(\theta \) be a \(C^{3,1}\)-function defined in a neighborhood \({\mathcal {U}}\) of \({\tilde{x}}\) in \(\Omega \) with \(\nabla \theta ({\tilde{x}})\ne 0\). Consider then the hypersurface
Making a translation and change of coordinates we may assume that \({\tilde{x}}=0\), \(\theta ({\tilde{x}})=0\), \(\nabla '\theta (0)=0\) and \(\partial _n\theta (0)\ne 0\). With the help of the implicit function theorem \(\theta (x)=0\) near 0 may rewritten as \(x_n=\vartheta (x')\) with \(\vartheta (0)=0\) and \(\nabla '\vartheta (0) =0\).
Let \({\hat{A}}\) be the matrix obtained after this transformations. According to Lemma 2.1, \({\hat{A}}\in {\mathscr {M}}({\mathcal {O}}({\mathcal {U}}+{\tilde{x}}),\varkappa ,{\mathfrak {m}}')\), \({\mathfrak {m}}'={\mathfrak {m}}'(n,{\mathfrak {m}})>0\) is a constant, where \({\mathcal {O}}\) is the orthogonal transformation corresponding to the above change of coordinates. Also, note that \(\hat{\mathbf{I }}=\mathbf{I }\).
Consider, in some neighborhood of 0, the mapping
Elementary calculations yield
with \(g_k(x')=-\partial _k\vartheta (x')+2x_k'\), \(0\le k\le n-1\). Whence
Since
there exists a neighborhood \(\omega \) of 0, only depending of \(\vartheta \), so that
In consequence
Whence (2.5) together with Cauchy-Schwarz’s inequality yield
Let \({\tilde{\omega }}=\varphi (\omega )\) (hence \(\varphi \) is a diffeomorphism from \(\omega \) onto \({\tilde{\omega }}\)) and define
In light of (2.6) we obtain
Also, by straightforward computations we get
with \(\tilde{{\mathfrak {m}}}\) only depending of n, \({\mathfrak {m}}\) and \(\vartheta \).
We observe that the role of \(\varphi \) is to transform the hypersurface \(\{x_n=\vartheta (x')\}\) in a neighborhood of the origin into the convex hypersurface \(\{y_n=|y'|^2\}\) in another neighborhood of the origin.
Define \({\tilde{\psi }}_0\) as follows
The matrix \({\tilde{A}}\) appearing in (2.7) is denoted hereafter by \(A_H\). The following definition is motivated by the classical procedure used to establish the unique continuation property of an elliptic operator across the convex hypersurface \(\{y_n=|y'|^2\}\).
Definition 2.3
We say that the hypersurface H is A-pseudo-convex if \({\tilde{\psi }}_0\) is \(A_H\)-pseudo-convex in \({\tilde{\omega }}\).
Lemma 2.2
-
(a)
There exists \({\mathcal {N}}\), a neighborhood of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }} ;{\mathbb {R}}^{n\times n})\), so that, for any \(A\in {\mathcal {N}}\), H is A-pseudo-convex.
-
(b)
There exists \({\mathcal {N}}_0\), a neighborhood of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }} ;{\mathbb {R}}^{n\times n})\), so that, for any \(A\in {\mathcal {N}}_0\) and any orthogonal transformation \({\mathcal {O}}\), we have \(A_{\mathcal {O}}\in {\mathcal {N}}\).
Proof
-
(a)
Let us first discuss the case where \(A=\mathbf{I }\). Note that it is not hard to check that
$$\begin{aligned} \varphi ^{-1}(y',y_n)=\left( y',y_n+\vartheta (y')-|y'|^2\right) \end{aligned}$$and
$$\begin{aligned} \mathbf{I }_H(y)=\tilde{\mathbf{I }} (y)=({\tilde{a}}_{ij}(y'))=\left( \begin{array} {cccc} 1 &{}\ldots &{}0 &{}{\tilde{g}}_1(y') \\ \vdots &{}\ddots &{}0 &{}\vdots \\ 0 &{}\ldots &{}1 &{}{\tilde{g}}_{n-1}(y') \\ {\tilde{g}}_1(y') &{}\ldots &{}{\tilde{g}}_{n-1}(y') &{}{\tilde{g}}_n(y') \end{array} \right) , \end{aligned}$$with \({\tilde{g}}_k(y')=\partial _k\vartheta (y')-2y_k'\), \(0\le k\le n-1\) and \({\tilde{g}}_n=|\nabla \vartheta (y')-2y'|^2+1\).
We have clearly \({\tilde{A}} (0)=\mathbf{I }\) and, for \(1\le p\le n-1\),
$$\begin{aligned} \partial _p{\tilde{a}}_{k\ell }(y')= \left\{ \begin{array}{ll} 0,\quad 1\le k,\ell \le n-1, \\ \\ \partial _{pk}^2\vartheta (y')-2\delta _{pk}, \quad 1\le k\le n-1,\; \ell =n, \\ \\ 2\sum _{\alpha =1}^{n-1}(\partial _\alpha \vartheta (y')-2y_\alpha )(\partial _{p\alpha }^2 \vartheta (y')-2\delta _{p\alpha }) ,\quad k=n,\; \ell =n. \end{array} \right. \end{aligned}$$Therefore
$$\begin{aligned} \partial _p{\tilde{a}}_{k\ell }(0)= \left\{ \begin{array}{ll} 0, &{}\quad 1\le k,\ell \le n-1, \\ \\ \partial _{pk}^2\vartheta (0)-2\delta _{pk} &{}1\le k\le n-1,\; \ell =n, \\ \\ 0 &{} k=n,\; \ell =n. \end{array} \right. \end{aligned}$$Let \({\tilde{\Lambda }}_{k\ell }^m\) given by
$$\begin{aligned} {\tilde{\Lambda }}_{k,\ell }^m(y)=-\sum _{p=1}^n\partial _p{\tilde{a}}_{k\ell }(y){\tilde{a}}_{pm}(y)+2\sum _{p=1}^n{\tilde{a}}_{kp}(y)\partial _p{\tilde{a}}_{\ell m}(y) \end{aligned}$$and define \({\tilde{\Upsilon }}(y)=({\tilde{\Upsilon }}_{k\ell }(y))\) as follows
$$\begin{aligned} {\tilde{\Upsilon }}_{k\ell }(y)=\sum _{m=1}^n{\tilde{\Lambda }}_{k\ell }^m(y)\partial _m {\tilde{\psi }}_0(y) . \end{aligned}$$It is then straightforward to check that
$$\begin{aligned} {\tilde{\Upsilon }}_{k\ell }(0)=\sum _{m=1}^n\Lambda _{k,\ell }^m(0)\partial _m {\tilde{\psi }}_0(0)=-2\Lambda _{k,\ell }^n(0)=0,\quad 0\le k,\ell \le n. \end{aligned}$$Since
$$\begin{aligned} {\tilde{\Theta }}(y)=\Theta _{\mathbf{I }}({\tilde{\psi }}_0)(y)= 2\nabla ^2{\tilde{\psi }}_0(y)+{\tilde{\Upsilon }}(y), \end{aligned}$$we get
$$\begin{aligned} {\tilde{\Theta }}(0)= 4\mathbf{I }, \end{aligned}$$and hence
$$\begin{aligned} ({\tilde{\Theta }}(0)\xi |\xi )\ge 4|\xi |^2,\quad \xi \in {\mathbb {R}}^n. \end{aligned}$$Continuity argument, first with respect to y and then with respect to A, shows that, by reducing \({\tilde{\omega }}\) if necessary,
$$\begin{aligned} (\Theta _{{\tilde{A}}}(\psi _0)(y)\xi |\xi )\ge 2|\xi |^2,\quad y\in {\tilde{\omega }},\; \xi \in {\mathbb {R}}^n. \end{aligned}$$ -
(b)
Immediate from Lemma 2.1. \(\square \)
3 The wave equation
3.1 Carleman inequality
In this subsection \(\psi (x,t)=\psi _0(x)+\psi _1(t)\) is a weight function for the wave operator \({\mathcal {L}}_{A,0}^w\) with A-pseudo-convexity constant \(\kappa >0\). We set
and \(\phi =e^{\lambda \psi }\).
We use for notational convenience \({\mathfrak {d}}=(\Omega , t_1, t_2,\varkappa , {\mathfrak {m}}, \kappa ,\delta ,{\mathfrak {b}})\) with \({\mathfrak {b}}\ge \Vert \psi \Vert _{C^4({\overline{Q}})}\), and \(\mathbf{D }_A=(\nabla _A\, \cdot , \partial _t\, \cdot )\).
Theorem 3.1
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that
for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(Q,{\mathbb {R}})\).
Proof
In this proof, \(\aleph _j\), \(\lambda _j\) and \(\tau _j\), \(j=0,1,\ldots \), are positive generic constants only depending on \({\mathfrak {d}}\).
Set \(\Phi =e^{-\tau \phi }\) with \(\tau >0\). Elementary computations then give
The preceding two first formulas can be rewritten in the following form
For \(w\in H^2 (Q;{\mathbb {R}})\), we obtain
Also,
We decompose \(L=\Phi ^{-1}{\mathcal {L}}_{A,0}^w\Phi \) into its self-adjoint part and skew-adjoint part:
Simple calculations show that
with
We have
where
The most integrations by parts, with respect to the space variable, we use in this proof are often based on Green’s formula (2.1). A first integration by parts then yields
Whence
Now as
we find
where \(C=(\text{ div }(a_{k\ell }B))\).
Let
We get by putting (3.4) into (3.3)
For \(I_2\), we obtain by making integrations by parts
As A is symmetric, we have
Hence
We have also
Let \(J_1=I_1+I_2+I_3\) and
Putting together (3.5)–(3.7), we find
Straightforward computations show that
Whence
We obtain, by using again an integration by parts,
Hence
But
Therefore
In consequence
For \(I_5\), we have
Also,
and hence
Let \(J_2=I_4+I_5+I_6\) and define
A combination of (3.9)–(3.11) gives
Let us observe that we have by straightforward computations
Hence
Let \({\tilde{J}}=J_1+J_2\). In light of (3.5) and (3.12) we deduce that
where
As \(\phi =e^{\lambda \psi }\), we have
This and the fact that \(\nabla \partial _t\psi =0\) imply
That is we have
from which we deduce, by noting that \(\Upsilon _A(\phi )=\lambda \phi \Upsilon _A(\psi )\),
Here
For \(I_7\), we have
Finally,
Define \({\hat{J}}=I_7+I_8+I_9\) and
Then we have from (3.13) to (3.15)
We prove (see details in the end of this proof) that
Whence
for \(\lambda \ge \lambda _1\) and \(\tau \ge \tau _1\), with
On the other hand, we find by making twice integration by parts
with
Let \(\epsilon >0\) to be determined later. Then Cauchy–Schwarz’s inequality yields
Using that \(a=\tau ^2\lambda ^2\phi ^2\left( |\nabla \psi |_A^2-(\partial _t\psi )^2\right) \) we get
Hence
Here \({\mathscr {B}}(w)={\mathscr {B}}_0(w)+\frac{\epsilon \tau \lambda }{2}{\mathscr {B}}_1(w)\).
We take \(\epsilon =2\varkappa ^{-1}\kappa \) in the preceding inequality and we use inequality (2.3). We obtain, by noting that in the right hand side of the last inequality, the fourth, fifth and sixth terms can be absorbed by the third term,
We find by making elementary calculations
for \(\lambda \ge \lambda _3\) and \(\mu \ge \mu _3\). This and
imply
We take in this inequality \(w=\Phi ^{-1}u\) with \(u\in H^2 (Q;{\mathbb {R}})\). In light of the identities
we obtain an inequality similar to (3.1) which leads to (3.1) by observing that the additional terms in the right hand side appearing in this intermediate inequality can be absorbed by the terms in left hand side.
Proof of (3.17). Set
We have
That is
In light of (3.19) we get
Since
we find
Also, as
we obtain
We get by putting together (3.20) and (3.21)
On the other hand,
Set
A combination of (3.22) and (3.23) yields
We have again from (3.19)
Therefore
where we used that \(\chi ^2\ge \delta \).
Inequality (3.17) then follows by combining (3.24) and (3.25) and using that
\(\square \)
Define \(\partial _{\nu _A}\psi _0\) by
and set
Let \(w\in H^2(Q,{\mathbb {R}})\) satisfying \(w=0\) on \(\Sigma \). In that case it is straightforward to check that \({\hat{g}}(w)\) defined in the preceding proof takes the form
Furthermore, if \(u=\Phi w\) then
In light of identities (3.26) and (3.27), slight modifications of the last part of the preceding proof enable us to establish the following result.
Theorem 3.2
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(Q,{\mathbb {R}})\) satisfying \(u=0\) on \(\Sigma \) and \(u=\partial _tu=0\) in \(\Omega \times \{t_1,t_2\}\), we have
Let us see that Theorem 3.1 remains valid whenever we add to \({\mathcal {L}}_{A,0}^w\) a first order operator. Consider then the operator
where \(q_0,\ldots ,q_n\) and p belong to \(L^\infty (Q,{\mathbb {C}})\) and satisfy
Let \(u=v+iw\in H^2(Q,{\mathbb {C}})\) and apply Theorem 3.1 to both v and w. We obtain by adding side by side the inequalities we obtain by taking in (3.1) \(u=v\) and then \(u=w\)
Since
and the term
can be absorbed by the left hand side of (3.29), we get the following result:
Corollary 3.1
We find three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(Q,{\mathbb {C}})\), we have
Finally, we note that the preceding arguments allow us also to prove the following corollary.
Corollary 3.2
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(Q,{\mathbb {C}})\) satisfying \(u=0\) on \(\Sigma \) and \(u=\partial _tu=0\) in \(\Omega \times \{t_1,t_2\}\), we have
We close this subsection by a one-parameter Carleman inequality that we obtain as a special case of Corollary 3.1 in which we fixed \(\lambda \ge \lambda ^*\).
Theorem 3.3
There exist two constants \(\aleph =\aleph ({\mathfrak {d}})>0\) and \(\tau ^*=\tau ^*({\mathfrak {d}})>0\) so that, for any \(\tau \ge \tau ^*\) and \(u\in C_0^\infty (Q,{\mathbb {C}})\), we have
3.2 Geometric form of the Carleman inequality
Pick \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\) and let \((a^{k\ell }(x))=(a_{k\ell }(x))^{-1}\), \(x\in {\overline{\Omega }}\). Consider then on \({\overline{\Omega }}\) the Riemannian metric g defined as follows
Set then
As usual define on \(T_x{\overline{\Omega }}={\mathbb {R}}^n\) the inner product
where \((\partial _1,\ldots ,\partial _n)\) is the dual basis of the Euclidean basis of \({\mathbb {R}}^n\). Set
We use for notational convenience \((X|Y)_g\) and \(|X|_g\) instead of \((X|Y)_{g(x)}\) and \(|X|_{g(x)}\).
Recall that the gradient of \(u\in H^1(\Omega )\) is the vector field given by
and the divergence of a vector field \(X=\sum _{\ell =1}^nX_\ell \partial _k\) with \(X_\ell \in H^1(\Omega )\), \(1\le \ell \le n\), is defined as follows
The usual Laplace–Betrami operator associated to the metric g is given, for \(u\in H^2(\Omega )\), by
Straightforward computations show that \(\Delta _Au=\sqrt{|g|}\Delta _gu\), from which we deduce the following identity
We assume in this subsection that \(\psi \) is a weight function for the wave operator \(\Delta _{\sqrt{|g|}^{-1}A}-\partial _t^2\) with \(\sqrt{|g|}^{-1}A\)-pseudo convexity constant \(\kappa >0\). Hereafter
where \(P=\sum _{\ell =1}^nP_i\partial _i\), \(q_0\) and q are so that \(P_1,\ldots ,P_n, q_0,q_1\) belong to \(L^\infty (Q;{\mathbb {C}})\) and satisfy
As in the preceding section \({\mathfrak {d}}=(\Omega , t_1,t_2,{\mathfrak {m}}, \varkappa ,\kappa ,\delta ,{\mathfrak {b}})\) with \({\mathfrak {b}}\ge \Vert \psi \Vert _{C^4({\overline{Q}})}\). In light of (3.32) we deduce the following Carleman inequality from Corollary 3.1 in which
\({ d}V_{{\overline{\Omega }}} =\sqrt{|g|}{ d}x_1\ldots { d}x_n\) is the volume form and
where \({ d}V_\Gamma \) is the volume form on \(\Gamma \).
Theorem 3.4
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(Q,{\mathbb {C}})\), we have
3.3 Unique continuation
We assume, for simplicity, in the present subsection that \(t_1=-{\mathfrak {t}}\) and \(t_2={\mathfrak {t}}\) with \({\mathfrak {t}}>0\).
We start with unique continuation across a particular convex hypersurface. To this end, we set
where \({\tilde{x}}=({\tilde{x}}',{\tilde{x}}_n)\in {\mathbb {R}}^n\) and \(c>0\).
Theorem 3.5
Suppose, for some \(r>0\), that \(B({\tilde{x}},r)\Subset \Omega \). We find \(0<c^*=c^*(\varkappa ,{\mathfrak {m}})\) with the property that, for any \(0<c<c^*\), there exist \(\tilde{{\mathfrak {t}}}=\tilde{{\mathfrak {t}}}(c,\varkappa )\) and \(0<\rho =\rho (c,\varkappa )<r\) so that, if \({\mathfrak {t}}\ge \tilde{{\mathfrak {t}}}\) then there exists \(0<{\mathfrak {t}}_0={\mathfrak {t}}_0(\varkappa ,{\mathfrak {m}},{\mathfrak {t}})\le {\mathfrak {t}}\) satisfying: for any \(u\in H^2({\mathcal {Q}};{\mathbb {C}})\), with \({\mathcal {Q}}=B({\tilde{x}},r)\times (-{\mathfrak {t}},{\mathfrak {t}})\), such that \({\mathcal {L}}_A^wu=\) in \({\mathcal {Q}}\) and \(\text{ supp }(u(\cdot ,t))\cap B({\tilde{x}},r)\subset E_+({\tilde{x}},c)\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\), we have \(u=0\) in \(B({\tilde{x}},\rho )\times (-{\mathfrak {t}}_0,{\mathfrak {t}}_0)\).
Proof
Let \({\tilde{x}}\in \Omega \) and \({\mathfrak {e}}_n=(0,1)\in {\mathbb {R}}^{n-1}\times {\mathbb {R}}\). Set \(x_0={\tilde{x}}+c{\mathfrak {e}}_n\) and let \(0<r_0\le \min (r,c/2)\). Define
As
we find in a straightforward manner that, for some constant \(\aleph =\aleph ({\mathfrak {m}})\), we have
We fix \(0<c<c^*=\varkappa ^2/\aleph \). With this choice of c we get
It is then not difficult to check that, where \(E_+=E_+({\tilde{x}},c)\),
Pick \(\chi \in C_0^\infty (B({\tilde{x}},r_0))\) satisfying \(\chi =1\) in \(B({\tilde{x}},\rho _1)\), for some fixed \( 0<\rho _1<r_0\). Fix then \(\epsilon >0\) in such a way that
Also, choose \(0<\rho _0<\rho _1\) such that
We are going to apply Corollary 3.1 with Q substituted by \({\mathcal {Q}}=B({\tilde{x}},r)\times (-{\mathfrak {t}},{\mathfrak {t}})\). Let
where the constant \(C >0\) is chosen sufficiently large in order to guarantee that \(\psi \ge 0\). From Example 2.1 we can easily see that \(\phi =e^{\lambda \psi }\) is a weight function for \({\mathcal {L}}_{A,0}^w\) in \({\mathcal {Q}}\) provided that \(0<\gamma <\min (c/(2{\mathfrak {t}}),\varkappa /4)\).
Suppose that \({\mathfrak {t}}\) is chosen sufficiently large in such a way \(c/(2{\mathfrak {t}})<\varkappa /4\). In that case we can take \(\gamma =c/(4{\mathfrak {t}})\).
Fix \(\gamma \) as above and let \(0<\varrho \le 1\) to be determined later. Let \(\lambda ^*\) and \(\tau ^*\) be as in Corollary 3.1. In the sequel we fix \(\lambda \ge \lambda ^*\). Set
Then straightforward computations show
In these inequalities we substitute \(\gamma {\mathfrak {t}}\) by c/4 in order to get
If \({\mathfrak {t}}\ge 2\epsilon /c\) we choose \(\varrho \) so that \(c\varrho ^2{\mathfrak {t}}=2\epsilon \). In that case we have
In consequence \(c_1<c_0\). Furthermore, if \({\mathfrak {t}}> 24\epsilon /c\) then we have also \(c_2<c_0\).
Let \(u\in H^2({\mathcal {Q}};{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^wu=0\) in \({\mathcal {Q}}\) and \(\text{ supp }(u(\cdot ,t))\cap B({\tilde{x}},r)\subset E_+\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\). Define \(v=\chi (x)\vartheta (t)u\) with \(\vartheta \in C_0^\infty ((-{\mathfrak {t}},{\mathfrak {t}}))\) satisfying \(\vartheta =1\) in \([- {\mathfrak {t}}/2,{\mathfrak {t}}/2]\).
As \({\mathcal {L}}_A^wu=0\) in \({\mathcal {Q}}\), elementary computations show that \({\mathcal {L}}_A^wv=f_1+f_2\) in \({\mathcal {Q}}\), where
Taking into account that \(\text{ supp }(f_1)\subset \mathbf{Q }_1\) and \(\text{ supp }(f_2)\subset \mathbf{Q }_2\), we find by applying Corollary 3.1, where \(\tau \ge \tau ^*\),
Passing then to the limit, as \(\tau \) tends to \(\infty \), in (3.34) in order to obtain that \(u=0\) in \(B({\tilde{x}},\rho _0)\times (-\varrho {\mathfrak {t}},\varrho {\mathfrak {t}})\). \(\square \)
Definition 3.1
We will say that \({\mathcal {L}}_A^w\) has the weak unique continuation property in Q if we can find \(0<\tau \le {\mathfrak {t}}\) so that, for any open subset \({\mathcal {O}}\) with \(\overline{{\mathcal {O}}}\subsetneqq \Omega \), there exists an open subset \({\mathcal {O}}_0\) with \({\mathcal {O}}_0\supsetneqq \overline{{\mathcal {O}}}\) so that, for any \(u\in H^2(Q;{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^wu=0\) in Q and \(u=0\) in \({\mathcal {O}}\times (-{\mathfrak {t}},{\mathfrak {t}})\), we have \(u=0\) in \({\mathcal {O}}_0\times (-\tau ,\tau )\).
Theorem 3.6
There exist a universal constant \({\mathfrak {t}}^*>0\) and a neighborhood \({\mathcal {N}}\) of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }};{\mathbb {R}}^n\times {\mathbb {R}}^n)\) so that, for each \({\mathfrak {t}}\ge {\mathfrak {t}}^*\) and \(A\in {\mathcal {N}}\), \({\mathcal {L}}_A^w\) has the weak unique continuation property in Q.
Proof
Set
Let \({\mathcal {N}}_0\) be the neighborhood of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }};{\mathbb {R}}^n\times {\mathbb {R}}^n)\) given by Lemma 2.2. Pick \(A\in {\mathcal {N}}_0\) and let \(u\in H^2(Q;{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^w=0\) in Q and \(u=0\) in \({\mathcal {O}}\times (-{\mathfrak {t}},{\mathfrak {t}})\), where \({\mathcal {O}}\) is an open subset satisfying \(\overline{{\mathcal {O}}}\subsetneqq \Omega \).
Fix \(y\in \partial {\mathcal {O}}\cap \Omega \) and \(r>0\) so that \(B(y,r)\subset \Omega \). Pick then \(y_0\in {\mathcal {O}}\cap B(y,r)\) sufficiently close to y in such a way that \(\partial B(y_0, d)\cap \partial {\mathcal {O}}\ne \emptyset \), with \(d=\text{ dist }(y_0,\partial {\mathcal {O}})\). Let then \(z\in \partial B(y_0, d)\cap \partial {\mathcal {O}}\). Making a translation and a change of coordinates if necessary, we may assume that \(z=0\) and \(B(y_0,d)\subset \{(y',y_n)\in {\mathbb {R}}^n;\; x_n<0\}\). We still denote, for notational convenience, the new matrix obtained after this translation and this change of coordinates by A. In that case, according to Lemma 2.2, A belongs to the neighborhood \({\mathcal {N}}\) appearing in this lemma. Whence, \(\text{ supp }(u(\cdot , t))\cap B(z,\rho )\subset H_+\), for some \(\rho >0\), with
Let \(\varphi \) given by (2.4) with \(\vartheta =0\) and \({\tilde{A}}=A_H\). If v is defined in a neighborhood \({\tilde{\omega }}\) by \(v(y,\cdot )=u(\varphi ^{-1}(y),\cdot )\) then straightforward computations give \({\mathcal {L}}^w_{{\tilde{A}}}v=0\) in \({\tilde{\omega }}\) and \(\text{ supp }(v(\cdot ,t))\subset E_+\) with
We complete the proof similarly to that of Theorem 3.5, with \({\tilde{x}}=0\), \(B(0,r)\Subset {\tilde{\omega }}\), \(\varkappa =1/4\) and \(c=1\). Note that in the present case
We get that there exists a universal constant \({\mathfrak {t}}^*\) so that, for any \({\mathfrak {t}}\ge {\mathfrak {t}}^*\), we find \(0<\tau \le {\mathfrak {t}}\) for which \(v=0\) in \(\tilde{{\mathcal {U}}}\times (-\tau ,\tau )\), for some neighborhood \(\mathcal {{\tilde{U}}}\) of 0. In consequence \(u=0\) in \({\mathcal {U}}\times (-\tau ,\tau )\), where \({\mathcal {U}}\) is a neighborhood of z in \(\Omega \). In other words we proved that \(u=0\) in \({\mathcal {O}}_0\times (-\tau ,\tau )\), where \({\mathcal {O}}_0={\mathcal {O}}\cup {\mathcal {U}}\supsetneqq {\mathcal {O}}\). The proof is then complete. \(\square \)
The result of Theorem 3.6 is false without the condition that \({\mathfrak {t}}\) is sufficiently large as shows the non uniqueness result in [2]. The authors show that, in the two dimensional case, there exists \({\mathcal {U}}\), a neighborhood of the origin in \({\mathbb {R}}^2\times {\mathbb {R}}\), \(p\in C^\infty ({\mathcal {U}})\), \(u\in C^\infty ({\mathcal {U}})\) so that \((\Delta -\partial _t^2+p(x,t))u=0\) in \({\mathcal {U}}\) and \(\text{ supp }(u)\subset {\mathcal {U}}\cap \{(x_1,x_2,t)\in {\mathbb {R}}^2\times {\mathbb {R}};\; x_2\ge 0\}\).
A better result than that in Theorem 3.6 can be obtained in the case of operators with time-independent coefficients. Let \(\dot{{\mathcal {L}}}_A^w\) be the operator \({\mathcal {L}}_A^w\) when \(q_0=0\), \(q_j=q_j(x)\), \(j=1,\ldots ,n\), and \(p=p(x)\).
Theorem 3.7
[31] There exits \({\mathfrak {t}}^*={\mathfrak {t}}(\Omega ,\varkappa , {\mathfrak {m}})\) so that, for any \({\mathfrak {t}}>{\mathfrak {t}}^*\) and \(\omega \Subset \Omega \), if \(u\in H^2(Q)\) satisfies \(\dot{{\mathcal {L}}}_A^w=0\) in Q and \(u=0\) in \(\omega \times (-{\mathfrak {t}},{\mathfrak {t}})\) then \(u=0\) in \(\Omega \times (-\tau ,\tau )\), where \(\tau ={\mathfrak {t}}-{\mathfrak {t}}^*\).
The main idea in the proof of Theorem 3.7 consists in transforming, via the Fourier–Bros–Iagolnitzer transform, the wave operator \(\dot{{\mathcal {L}}}_A^w\) into an elliptic operator for which uniqueness of continuation property result is known.
Theorem 3.6 can be used to establish a result on local uniqueness of continuation from Cauchy data on a part of the boundary.
Corollary 3.3
Let \({\mathfrak {t}}^*\) and \({\mathcal {N}}\) be as in Theorem 3.6 with \(\Omega \) substituted by larger domain \({\hat{\Omega }}\Supset \Omega \). Let \(\Gamma _0\) be a nonempty open subset of \(\Gamma \) and \(\Sigma _0=\Gamma _0\times (-{\mathfrak {t}},{\mathfrak {t}})\) with \({\mathfrak {t}}\ge {\mathfrak {t}}^*\). There exist \({\mathcal {U}}\) a neighborhood of a point of \(\Gamma _0\) in \(\Omega \) and \(0<\tau \le {\mathfrak {t}}\) so that if \(A\in {\mathcal {N}}\), and if \(u\in H^2(Q;{\mathbb {C}})\) satisfies \({\mathcal {L}}_A^wu=0\) in Q and \(u=\partial _\nu u=0\) on \(\Sigma _0\) then \(u=0\) in \({\mathcal {U}}\times (-\tau ,\tau )\).
Proof
Pick \(A\in {\mathcal {N}}\) and \(u\in H^2(Q;{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^wu=0\) in Q and \(u=\partial _\nu u=0\) on \(\Sigma _0\). Then there exists \({\mathcal {V}}\subset {\hat{\Omega }}\), a neighborhood of a point in \(\Gamma _0\), so that \({\hat{u}}\), the extension of u by zero in \({\mathbb {R}}^n{\setminus } {\overline{\Omega }}\), belongs to \(H^2(\Omega '\times (-{\mathfrak {t}},{\mathfrak {t}}))\), with \(\Omega '=\Omega \cup {\mathcal {V}}\), satisfies \({\mathcal {L}}_A^w{\hat{u}}=0\) in \(\Omega '\times (-{\mathfrak {t}},{\mathfrak {t}})\) and \({\hat{u}}=0\) in \([(\Omega '{\setminus } {\overline{\Omega }}) \cap {\mathcal {V}}]\times (-{\mathfrak {t}},{\mathfrak {t}})\). Theorem 3.6 allows us to conclude that there exist \({\mathcal {U}}\), a neighborhood of a point of \(\Gamma _0\) in \(\Omega \), and \(0<\tau \le {\mathfrak {t}}\) so that \(u=0\) in \({\mathcal {U}}\times (-\tau ,\tau )\). \(\square \)
It is worth mentioning that the following global unique continuation result from boundary data can be deduced from [6, Theorem 1.1].
Theorem 3.8
Let \(\Gamma _0\) an arbitrary non empty open subset of \(\Gamma \). There exits \({\mathfrak {t}}^*={\mathfrak {t}}(\Omega ,\varkappa , {\mathfrak {m}})\) so that, for any \({\mathfrak {t}}>{\mathfrak {t}}^*\) we find \(0<{\mathfrak {t}}_0<{\mathfrak {t}}\) with the property that if \(u\in C^\infty ({\overline{\Omega }}\times [-{\mathfrak {t}},{\mathfrak {t}}])\) satisfies \({\mathcal {L}}_{A,0}^wu=0\) and
then u is identically equal to zero.
We end this subsection by remarking that we can proceed similarly to Theorem 3.6 to prove the unique continuation property across a pseudo-convex hypersurface.
Theorem 3.9
Let \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\) and \(H=\{x\in \omega ;\;\theta (x)=\theta ({\tilde{x}})\}\) be a A-pseudo-convex hypersurface defined in a neighborhood \(\omega \) of \({\tilde{x}}\in \Omega \) with \(\theta \in C^{3,1}({\overline{\omega }})\). Then there exist \({\mathcal {B}}\), a neighborhood of \({\tilde{x}}\), and \({\mathfrak {t}}^*>0\) so that, for each \({\mathfrak {t}}\ge {\mathfrak {t}}^*\), we find \(0<\tau \le {\mathfrak {t}}\) with the property that if \(u\in H^2(\omega \times (-{\mathfrak {t}},{\mathfrak {t}}))\) satisfies \({\mathcal {L}}_A^wu=0\) in \(\omega \times (-{\mathfrak {t}},{\mathfrak {t}})\) and \(\text{ supp }(u(\cdot ,t))\subset H_+=\{ x\in \omega ;\; \theta (x)\ge \theta ({\tilde{x}})\}\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\), then \(u=0\) in \({\mathcal {B}}\times (-\tau ,\tau )\).
3.4 Observability inequality
We suppose in this subsection that \(t_1=0\) and \(t_2={\mathfrak {t}}>0\).
We shall need in the sequel the following technical lemma.
Lemma 3.1
Fix \(0<\alpha <1\) and let \(0\le \psi _0\in C^1({\overline{\Omega }})\) satisfying
Let \(\mathbf{m }=\Vert \psi _0\Vert _{L^\infty (\Omega )}\) and define, for an arbitrary constant \(C>0\),
If \({\mathfrak {t}}> {\mathfrak {t}}_\alpha =\max \left( \delta _0^{-1/[2(1-\alpha )]}, (64{\mathfrak {m}}/2)^{1/\alpha }\right) \) then
Proof
If \({\mathfrak {t}}\ge {\mathfrak {t}}_\alpha \) then
That is we proved (3.36).
Inequality (3.37) is straightforward. On the other hand, we have
if \((x,t)\in {\overline{\Omega }}\times \left( [0,{\mathfrak {t}}/4]\cup [3{\mathfrak {t}}/4,{\mathfrak {t}}]\right) \). That is we proved (3.38). \(\square \)
We consider in this subsection the following wave operator
with \(p,q\in L^\infty (\Omega ;{\mathbb {C}})\). We associate to \({\mathcal {L}}_A^w\) the IBVP
According to the semigroup theory, for all \((u_0,u_1)\in H_0^1(\Omega )\times L^2(\Omega )\), the IBVP (3.39) admits unique solution
We also know that \(\partial _\nu u\in L^2(\Sigma )\) (hidden regularity).
From usual energy estimate for wave equations, if
then
where the constant \(\aleph _0>0\) only depends of \(\Omega \), A, \({\mathfrak {t}}\), p and q.
We apply (3.41) to \(v(\cdot ,t)=u(\cdot ,s-t)\), with fixed \(0<t\le s\). We find
where the constant \(\aleph _1>0\) only depends of \(\Omega \), A, \({\mathfrak {t}}\), p and q. We have in particular
In light of (3.41) and (3.42) we get
for some constant \(\aleph >1\) only depending of \(\Omega \), A, \({\mathfrak {t}}\), p and q.
We recall that \(\Sigma _+=\Gamma _+\times (0,{\mathfrak {t}})\).
Theorem 3.10
Fix \(0<\alpha <1\) and assume that \(0\le \psi _0\in C^4({\overline{\Omega }})\) is A-pseudo-convex with constant \(\kappa >0\) and let \(\Gamma _+=\{x\in \Gamma ;\; \partial _{\nu _A}\psi _0(x)>0\}\). If \(\tilde{{\mathfrak {t}}}_\alpha =\min \left( {\mathfrak {t}}_\alpha , (8\varkappa /\kappa )^{1/(2-\alpha )}\right) \), where \({\mathfrak {t}}_\alpha \) be as in Lemma 3.1, then, for any \({\mathfrak {t}}\ge \tilde{{\mathfrak {t}}}_\alpha \) and \((u_0,u_1)\in H_0^1(\Omega )\times L^2(\Omega )\), we have
where the constant \(\aleph >0\) only depends of \(\Omega \), \({\mathfrak {t}}\), \(\varkappa \), \(\kappa \), \(\Gamma _+\) and u is the solution of the IBVP (3.39) corresponding to \((u_0,u_1)\).
Proof
Fix \({\mathfrak {t}}>\tilde{{\mathfrak {t}}}_\alpha \) and let \(\psi \) defined as in (3.35) in which the constant \(C>0\) is chosen sufficiently large to guarantee that \(\psi \ge 0\). In that case we easily check that \(\phi =e^{\lambda \psi }\) is a weight function for the operator \({\mathcal {L}}_{A,0}^w\) in Q.
Pick \(\varrho \in C_0^\infty ({\mathfrak {t}}/8,7{\mathfrak {t}}/8)\) so that \(\varrho =1\) in \([{\mathfrak {t}}/4,3{\mathfrak {t}}/4]\). Clearly, a simple density argument shows that Corollary 3.2 remains valid for \(\varrho u\) for any solution u of (3.39) with \((u_0,u_1)\in H_0^1(\Omega )\times L^2(\Omega )\). According to this Corollary we have, for fixed \(\lambda \ge \lambda ^*\) and any \(\tau \ge \tau ^*\),
But
Hence (3.44) together with Poincaré’s inequality (\(u(\cdot , t)\in H_0^1(\Omega )\)) give
Define
If \({\mathcal {E}}_u\) is given by (3.40) then we get from (3.37), (3.38) and (3.45), where \(\tau \ge \tau ^*\),
This inequality together with (3.43) imply
As \(c_0 >c_1\), we fix \(\tau \) sufficiently large in such a way that \({\tilde{\aleph }}=\aleph e^{\tau c_0} -e^{\tau c_1}>0\). That is we have
The expected inequality follows readily from (3.46). \(\square \)
From the calculations in Example 2.1 when \(\psi _0(x)=|x-x_0|^2/2\), with \(x_0\in {\mathbb {R}}^n{\setminus }{\overline{\Omega }}\), there exists a neighborhood \({\mathcal {N}}\) of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }};{\mathbb {R}}^{n\times n})\) so that that for any \(A\in {\mathcal {N}}\), \(\varkappa =1/2\) and \(\psi _0\) is A-pseudo-convex with constant \(\kappa =1/4\). In this case
with \(d_0=\text{ dist }(x_0,{\overline{\Omega }})\) and \(d=\text{ diam }(\Omega )\).
A result in the variable coefficients case was already established in [37, Theorem 1.1]. This result is based on a generalization of the multiplier method in which a vector field is used as an alternative to the multiplier. This vector field satisfies a certain convexity condition. Note however that the lower bound in \({\mathfrak {t}}\) appearing in [37, Theorem 1.1] is not easily comparable to that we used in Theorem 3.10. The minimal time guaranteeing observability was estimated in precise way in [14] for wave equations with \(C^1\) variable coefficients. Recently an observability result was established in [36] for constant coefficients wave equation in the case of time-dependent domains. The minimal time in [36] is explicit.
4 Elliptic equations
We show briefly how we can modify the calculations we carried out for wave equations in order to retrieve Carleman inequalities for elliptic equations and the corresponding property of unique continuation. In this section
where \(p_1,\ldots p_n\) and q belong to \(L^\infty (\Omega ;{\mathbb {C}})\) and satisfy
Also, \(0\le \psi \in C^4({\overline{\Omega }})\) is fixed so that
for some constant \(\delta >0\).
4.1 Carleman inequality
Let \(\phi =e^{\lambda \psi }\) and set \({\mathfrak {d}}=(\Omega ,\varkappa , \delta ,{\mathfrak {m}} )\).
Theorem 4.1
We find three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^2(\Omega ,{\mathbb {C}})\), we have
Proof
In this proof, \(\lambda _k\) and \(\tau _k\), \(k=1,2,\ldots \), denote generic constants only depending on \({\mathfrak {d}}\).
Let \(\Phi =e^{-\tau \phi }\), \(\tau >0\). We have from the calculations of the preceding section
We decompose L in the following special form
with, for \(w\in H^2(\Omega ,{\mathbb {R}})\),
The coefficients of \(L_0\) and \(L_1\) and c are given as follows
We have
where
Let
Then straightforward modifications of the computations of the preceding section yield
Identities (4.3)–(4.5) in (4.2) give
with
We have
where \(\tilde{{\mathfrak {A}}}\) is a matrix depending only on A and \(\psi \). Therefore
We have also
where the reminder term \(\tilde{{\mathfrak {a}}}\) contains, as for the wave equation, only terms with factors \(\tau ^k\lambda ^\ell \phi ^m\), \(1\le k,\ell ,m \le 3\) and terms with factor \(\tau \lambda ^4\phi \). Hence
The rest of the proof is almost similar to that of the wave equation. \(\square \)
Remark 4.1
The symbol of the principal part of the operator \({\mathcal {L}}_A^e\) is given by
Therefore if \(\phi \in C^4({\overline{\Omega }})\) then we have \(p(x,\xi +i\tau \nabla \phi )=p_0+ip_1\) with
When \(\phi =e^{\lambda \psi }\) we find, for \(\tau \ge 1\),
In consequence \(\phi \) satisfies the sub-ellipticity condition in [18, Theorem 8.3.1, page 190] if \(\lambda \) is sufficiently large and hence the following Carleman inequality holds: there exist \(\aleph >0\) and \(\tau ^*>0\) only depending on \(\Omega \) and bounds on the coefficients of \({\mathcal {L}}_A^e\) so that
In other words, if \(\phi =e^{\lambda \psi }\) is a weight function for the elliptic operator \({\mathcal {L}}_{A,0}^e\) then \(\phi \) possesses the sub-ellipticity condition for large \(\lambda \).
4.2 Unique continuation
We use a similar method as for the wave equation. For sake of completeness, we provide some details.
We start with a unique continuation result across a convex hypersurface. To this end, we set
As \(|\nabla \psi (0,0)|=2\), there exists \(r>0\) so that \(|\nabla \psi |\ge 1\) in B(0, r). Consider then the set
We have clearly
Pick \(\chi \in C_0^\infty (B(0,r))\) satisfying \(\chi =1\) in \(B(0,\rho _1)\), where \(0<\rho _1 <r\) is fixed arbitrary. Let then \(\epsilon >0\) so that
Lemma 4.1
There exists \(0<\rho _0<\rho _1\) so that if \(u\in H^2(B(0,r);{\mathbb {C}})\) satisfies \({\mathcal {L}}_A^eu=0\) in B(0, r) and \(\text{ supp }(u)\subset E_+\) then \(u=0\) in \(B(0,\rho _0)\).
Proof
Let us choose \(0<\rho _0 <\rho _1\) in such a way that
Pick \(u\in H^2(B(0,r);{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^eu=0\) in B(0, r) and \(\text{ supp }(u)\subset E_+\). Let \(v=\chi u\), and \(\lambda ^*\) and \(\tau ^*\) be as in Theorem 4.1. Fix then \(\lambda \ge \lambda ^*\) and set
Theorem 4.1 yields
Noting that \(c_0 >c_1\), we obtain that \(u=0\) in \(B(0,\rho _0)\) by taking in the right hand side of the last inequality the limit, as \(\tau \) tends to \(\infty \). \(\square \)
Let \(\vartheta =\vartheta (x')\) be in \(C^{3,1}({\overline{B}}(0,r))\) satisfying \(\vartheta (0)=0\) and \(\nabla '\vartheta (0)=0\), and consider
As \(\varphi '(0,0)=\mathbf{I }\), we deduce that \(\varphi \) is a diffeomorphism from \(\omega \subset B(0,r)\times {\mathbb {R}}\), a neighborhood of 0 in \({\mathbb {R}}^n\), onto \({\tilde{\omega }}=\varphi (\omega )\).
Pick \(u\in H^2(\omega ,{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^eu=0\) in \(\omega \) and \(\text{ supp }(u)\subset \omega _+=\{(x',x_n)\in \omega ;\; x_n\ge \vartheta (x')\}\). Define v by \(v(y',y_n)=u(\varphi ^{-1}(y',y_n))\), \((y',y_n)\in {\tilde{\omega }}\). Then it is straightforward to check that \({\mathcal {L}}_{{\tilde{A}}}^e v=0\) in \({\tilde{\omega }}\). Here \({\mathcal {L}}_{{\tilde{A}}}^e\) is of the same form as \({\mathcal {L}}_A^e\). Its principal part is given by
with
Furthermore, \(\text{ supp }(v)\subset {\tilde{\omega }}_+=\{ (y',y_n)\in {\tilde{\omega }};\; y_n\ge |y'|^2\}\).
Similar calculations as in Sect. 2.4 show, by reducing \(\omega \) if necessary, that
We apply Lemma 4.1 in order to get \(v=0\) in \(\tilde{{\mathcal {V}}}\), where \(\tilde{{\mathcal {V}}}\) is a neighborhood of the origin, and hence \(u=0\) in \({\mathcal {V}}\) with \({\mathcal {V}}=\varphi ^{-1}(\tilde{{\mathcal {V}}})\). In other words, we proved the following result.
Lemma 4.2
There exists a neighborhood \({\mathcal {V}}\) of the origin in \(\omega \) so that if \(u\in H^2(\omega ;{\mathbb {C}} )\) satisfies \({\mathcal {L}}_A^eu=0\) in \(\omega \) and \(\text{ supp }(u)\subset \omega _+\) then \(u=0\) in \({\mathcal {V}}\).
The global uniqueness of continuation result is based on the following lemma.
Lemma 4.3
Let \(\Omega _0\subset \Omega \) so that \(\partial \Omega _0\cap \Omega \ne \emptyset \). There exists \(z\in \partial \Omega _0\cap \Omega \) and \({\mathcal {W}}\) a neighborhood of z in \(\Omega \) so that if \(u\in H^2(\Omega ;{\mathbb {C}})\) satisfies \({\mathcal {L}}_A^eu=0\) in \(\Omega \) together with \(u=0\) in \(\Omega _0\) then \(u=0\) in \({\mathcal {W}}\).
Proof
Fix \(y\in \partial \Omega _0\cap \Omega \) and \(r>0\) so that \(B(y,r)\subset \Omega \). Pick then \(y_0\in \Omega _0\cap B(y,r)\) sufficiently close to y in such a way that \(\partial B(y_0, d)\cap \partial \Omega _0\ne \emptyset \), with \(d=\text{ dist }(y_0,\partial \Omega _0)\). Pick then \(z\in \partial B(y_0, d)\cap \partial \Omega _0\). Making a translation we may assume that \(z=0\). As the \(\partial B(y_0, d)\) can be represented locally by a graph \(x_n=\vartheta (x')\). Making a change of coordinates we may assume that \(\text{ supp }(u)\subset \{(x',x_n);\; x_n\ge \vartheta (x')\}\). This orthogonal transformation modify A, but the new matrix has the same properties as A. We then complete the proof by using Lemma 4.2 with A substituted by this new matrix. \(\square \)
Theorem 4.2
Let \(u\in H^2(\Omega ;{\mathbb {C}})\) satisfying \({\mathcal {L}}_A^eu=0\) in \(\Omega \) and \(u=0\) in \(\omega \), for some nonempty open subset \(\omega \) of \(\Omega \). Then \(u=0\) in \(\Omega \).
Proof
Let \(\Omega _0\) be the maximal domain in which \(u=0\). If \(\Omega {\setminus } \overline{\Omega _0}\ne \emptyset \) then we would have \(\partial \Omega _0\cap \Omega \ne \emptyset \). Therefore we would find, by Lemma 4.3, \(z\in \partial \Omega _0\cap \Omega \) and \({\mathcal {W}}\) a neighborhood of z in \(\Omega \) so that \(u=0\) in \({\mathcal {W}}\). That is \(u=0\) in \(\Omega _0\cup {\mathcal {W}}\) which contains strictly \(\Omega _0\) and hence contradicts the maximality of \(\Omega _0\). \(\square \)
It is worth mentioning that Theorem 4.2 can also be obtained as a consequence [11, Proposition 2.28, page 28] that quantifies the uniqueness of continuation from a subdomain of \(\Omega \) to another subdomain of \(\Omega \). The proof of [11, Proposition 2.28, page 28] relies on three-ball inequality which is itself a consequence of the Carleman inequality of Theorem 4.1.
As an immediate consequence of Theorem 4.2 we get uniqueness of continuation from the Cauchy data on a subboundary.
Corollary 4.1
Assume that \({\mathcal {L}}_A^e\) is defined in \({\hat{\Omega }}\Supset \Omega \). Let \(\Gamma _0\) be an arbitrary nonempty open subset of \(\Gamma \). If \(u\in H^2(\Omega ;{\mathbb {C}})\) satisfies \({\mathcal {L}}_A^eu=0\) in \(\Omega \) together with \(u=\partial _\nu u= 0\) in \(\Gamma _0\) then \(u=0\) in \(\Omega \).
Proof
If \(u\in H^2(\Omega ;{\mathbb {C}})\) satisfies \(u=\partial _\nu u= 0\) in \(\Gamma _0\) then we find \({\mathcal {V}}\subset {\hat{\Omega }}\), a neighborhood of a point in \(\Gamma _0\), so that \({\tilde{u}}\), the extension of u by zero in \({\mathbb {R}}^n{\setminus } {\overline{\Omega }}\), belongs to \(H^2(\Omega \cup {\mathcal {V}})\), satisfies \({\mathcal {L}}_A^e{\tilde{u}}=0\) in \(\Omega \cup {\mathcal {V}}\) and \({\tilde{u}}=0\) in \({\mathcal {V}}{\setminus } {\overline{\Omega }}\). Theorem 4.2 allows us to conclude that \({\tilde{u}}=0\) and hence \(u=0\). \(\square \)
We observe once again that Corollary 4.1 can be deduced from [11, Proposition 2.28 in page 28 and Proposition 2.30 in page 29]. We point out that [11, Proposition 2.30 in page 29] quantifies the uniqueness of continuation from the Cauchy data on a subboundary to an interior subdomain.
5 Parabolic equations
We fix in this section \(0\le \psi \in C^4({\overline{Q}})\) of the form \(\psi (x,t)=\psi _0(x)+\psi _1(t)\), where
for some constant \(\delta >0\). Let \(\phi =e^{\lambda \psi }\), \(\lambda >0\), and consider the parabolic operator
where \(A\in {\mathscr {M}}(\Omega ,\varkappa ,{\mathfrak {m}})\), \(p_1,\ldots p_n\) and q belong to \(L^\infty (Q ;{\mathbb {C}})\) and satisfy
5.1 Carleman inequality
Recall that \({\mathcal {L}}_{A,0}^p\) represents the principal part of \({\mathcal {L}}_A^p\):
We will use in the sequel the notation \({\mathfrak {d}}=(\Omega ,t_1,t_2,\varkappa , \delta ,{\mathfrak {m}} )\).
Theorem 5.1
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^{2,1}(Q,{\mathbb {C}})\), we have
Proof
As for the wave equation, we set \(\Phi =e^{-\tau \phi }\), \(\tau >0\). We also recall that
We have, for \(w\in H^{2,1}(Q,{\mathbb {R}})\),
Also,
We decompose \(L=\Phi ^{-1}{\mathcal {L}}_{A,0}^p\Phi \) as in the elliptic case. That is in the form
with
where
We have
Quantities \(I_j\), \(1\le j\le 6\), are given as follows
Let \(C=(\text{ div }(a_{k\ell }B))\) and
We already proved that
On the other hand inequality (3.6) with \(d=-1\) gives
\(I_3\) is the same as in (3.7):
Let \(J_1=I_1+I_2+I_3\) and
We find by putting together (5.3)–(5.5)
\(I_4\) was calculated in (3.14). Precisely, we have
On the other hand an integration by parts, with respect to t, gives
Set
and let \(J_2=I_4+I_5+I_6\). In light of (5.7) and (5.8) we get
Putting together (5.2), (5.6) and (5.9) in order to obtain
where \({\mathfrak {a}}=a_1+a_2\), \(g=g_1+g_2\) and \(h=h_1+h_2\).
As \(\partial _t a=0\) (which is a consequence of \(\partial _t\nabla \psi =0\)), we see that \({\mathfrak {A}}\) and \({\mathfrak {a}}\) has exactly the same form as in the elliptic case. Therefore we can mimic the proof of the elliptic case to complete the proof. \(\square \)
We already defined \(\Gamma _+=\Gamma _+^{\psi _0}=\{x\in \Gamma ;\; \partial _{\nu _A} \psi _0>0\}\) and \(\Sigma _+=\Sigma _+^{\psi _0}=\Gamma _+\times (t_1,t_2)\). Similarly to the wave equation we have the following result.
Theorem 5.2
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^{2,1}(Q,{\mathbb {C}})\) satisfying \(u=0\) on \(\Sigma \) and \(u(\cdot ,t)=0\), \(t\in \{t_1,t_2\}\), we have
5.2 Unique continuation
An adaptation of the proof in the case of wave equations enables us to establish the following result.
Theorem 5.3
Let \(u\in H^{2,1}(Q)\) satisfying \({\mathcal {L}}_A^pu=0\) in Q and \(u=0\) in \(\omega \times (t_1,t_2)\), for some nonempty open subset \(\omega \) of \(\Omega \). Then \(u=0\) in Q.
We remark that Theorem 5.3 can be also obtained from [13, Proposition 3.2].
Similarly to the case of wave equations Theorem 5.3 can serve to prove uniqueness of continuation from the Cauchy data on a subboundary.
Corollary 5.1
Assume that \({\mathcal {L}}_A^p\) is defined in \({\hat{\Omega }}\Supset \Omega \). Let \(\Gamma _0\) be an arbitrary nonempty open subset of \(\Gamma \). If \(u\in H^1((t_1,t_2), H^2(\Omega ))\) satisfies \({\mathcal {L}}_A^pu=0\) in Q and \(u=\partial _\nu u= 0\) in \(\Sigma _0=\Gamma _0\times (t_1,t_2)\) then \(u=0\) in Q.
We remark that Corollary 5.1 follows also from [13, Proposition 3.2 and Proposition 4.1].
5.3 Final time observability inequality
We assume in the present subsection that \(t_1=0\) and \(t_2={\mathfrak {t}}>0\), and we consider the IBVP
Define the unbounded operator \({\mathscr {A}}:L^2(\Omega )\rightarrow L^2(\Omega )\) by
It is known that \(-{\mathscr {A}}\) generates an analytic semigroup \(e^{-t{\mathscr {A}}}\). In particular, for any \((u_0,f)\in L^2(\Omega )\times L^1((0,T);L^2(\Omega ))\), the IBVP (5.12) has a unique (mild) solution \(u={\mathscr {S}}(u_0,f)\in C([0,T];L^2(\Omega ))\) so that
This solution is given by Duhamel’s formula
Note that if \(u_0\in D({\mathscr {A}})\) then \(u={\mathcal {S}}(u_0,0)\) satisfies
We refer to [30, Chapter 11] for a concise introduction to semigroup theory.
Lemma 5.1
Let \(f\in L^2(Q)\) and \(\zeta \in C_0^\infty ([0,{\mathfrak {t}}))\). Then \(u({\mathfrak {t}})={\mathscr {S}}(0,\zeta f)({\mathfrak {t}})\in H_0^1(\Omega )\) and
where \(\aleph >0\) is a constant only depending on \(\Omega \), A, \({\mathfrak {t}}\) and \(\zeta \).
Proof
We have from [29, Theorem 8.1 in page 254] that \(D({\mathscr {A}}^{1/2})=H_0^1(\Omega )\). Therefore, in light of (5.14), we obtain
But
where the constant \(\aleph _0>0\) only depends on \(\Omega \), A. If \(\text{ supp }(\zeta )\subset [0,{\mathfrak {t}}-\epsilon ]\), \(\epsilon >0\), we find
Whence Cauchy-Schwarz’s inequality yields
The expected inequality then follows. \(\square \)
Theorem 5.4
Let \(0\le \psi _0 \in C^4(\Omega )\) with no critical point in \({\overline{\Omega }}\). Set \(\Gamma _+=\{x\in \Gamma ;\; \partial _{\nu _A}\psi _0(x)>0\}\) and \(\Sigma _+=\Gamma _+\times (0,{\mathfrak {t}})\). For any \(u=e^{-t{\mathscr {A}}}u_0\) with \(u_0\in D({\mathscr {A}})\) we have
where the constant \(\aleph >0\) only depends of \(\Omega \), A and \({\mathfrak {t}}\).
Proof
Let \(\chi \in C_0^\infty ((0,{\mathfrak {t}}))\) satisfying \(\chi =1\) in \([{\mathfrak {t}}/4,3{\mathfrak {t}}/4]\). As \((\Delta _A-\partial _t)(\chi u)=-\chi 'u\), we get by applying Theorem 5.2, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\),
where the notations are those of Theorem 5.2.
Since the first term in the right hand side of this inequality can be absorbed by the left hand side, provided that \(\lambda \) and \(\tau \) are sufficiently large, we obtain in a straightforward manner
Pick \(\varphi \in C^\infty ([0,{\mathfrak {t}}])\) so that that \(\varphi =0\) in \([0,{\mathfrak {t}}/4]\) and \(\varphi =1\) in \([3{\mathfrak {t}}/4,{\mathfrak {t}}]\). We easily check that \(\varphi u={\mathscr {S}}(0,\varphi 'u)\). In light of Lemma 5.1, we then conclude that
This and (5.16) imply the expected inequality. \(\square \)
6 Schrödinger equations
Let \(\phi =e^{\lambda \psi }\) be a weight function for the Schrödinger operator
and set
where \(p_1,\ldots p_n\) and q belong to \(L^\infty (Q ;{\mathbb {C}})\) and satisfy
6.1 Carleman inequality
Let
and \({\mathfrak {d}}=(\Omega ,t_1,t_2,\varkappa , \delta ,{\mathfrak {m}} )\).
Theorem 6.1
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^{2,1}(Q,{\mathbb {C}})\), we have
Proof
In this proof, \(\aleph \), \(\lambda _j\), \(\tau _j\), \(j=1,2,\ldots \), denote positive generic constants only depending on \({\mathfrak {d}}\).
As in the preceding section, if \(\Phi =e^{-\tau \phi }\), \(\tau >0\), then
We have, where \(w\in H^2 (Q;{\mathbb {C}})\),
and
We decompose \(L=\Phi ^{-1}{\mathcal {L}}_{A,0}^s\Phi \) as follows
with
where we set
We have
with
Some parts of the proof are quite similar to that of the wave equation and therefore we omit their details. We have
where
with \(C=(\text{ div }(a_{k\ell }B))\).
Also,
But \(\mathfrak {R}({\overline{w}}\nabla w)=\nabla |w|^2/2\). Therefore
Let \(J=\mathfrak {R}(I_1+I_2)\) and define
We combine (6.2) and (6.3) in order to obtain
We have once again from the calculations we done for the wave equation
This identity together with the following ones
imply
As \(A(\nabla \psi _0\otimes \nabla \psi _0)A\) is non negative and \(\psi _0\) is A-pseudo-convex with constant \(\kappa >0\), we get
This inequality in (6.4) yields
We find, by making an integration by parts with respect to t and then with respect to x,
We then obtain, by noting that \(\text{ div }(B)=2b\),
From the identity
we deduce that
Or equivalently
Observing that
we obtain
with
We find, by using once again the identity \(\mathfrak {R}w\nabla {\overline{w}}=\nabla |w|^2/2\),
and hence
with
Let
We can carry out the same calculations as for the wave equation in order to obtain
We end up getting, by combining (6.1), (6.5), (6.6) and (6.7), the following inequality
where we set \(g=g_1+g_2\).
Let \(\epsilon >0\). Then an elementary convexity inequality yields
In consequence the third term in (6.8) can be absorbed by the first two ones, provided that \(\lambda \ge \lambda _2\) and \(\tau \ge \tau _2\). That is we have
The rest of the proof is almost similar to that of the wave equation. \(\square \)
Recall that \(\Gamma _+=\Gamma _+^{\psi _0}=\{x\in \Gamma ;\; \partial _{\nu _A}\psi _0>0\}\) and \(\Sigma _+=\Sigma _+^\psi =\Gamma _+\times (t_1,t_2)\). As for the wave equation we have
Theorem 6.2
There exist three constants \(\aleph =\aleph ({\mathfrak {d}})\), \(\lambda ^*=\lambda ^*({\mathfrak {d}})\) and \(\tau ^*=\tau ^*({\mathfrak {d}})\) so that, for any \(\lambda \ge \lambda ^*\), \(\tau \ge \tau ^*\) and \(u\in H^{2,1}(Q,{\mathbb {C}})\) satisfying \(u=0\) on \(\Sigma \) and \(u(\cdot ,t)=0\), \(t\in \{t_1,t_2\}\), we have
6.2 Unique continuation
In this subsection, \(t_1=-{\mathfrak {t}}\) and \(t_2={\mathfrak {t}}\), where \({\mathfrak {t}}>0\) is fixed. We recall that
with \({\tilde{x}}\in \Omega \), \(c>0\).
Theorem 6.3
Suppose that \(B({\tilde{x}},r)\Subset \Omega \), for some \(r>0\). There exists \(c^*=c^*(\varkappa ,{\mathfrak {m}})\) with the property that, for any \(0<c<c^*\), we find \(0<\rho =\rho (c,\varkappa )<r\) so that if \(u\in H^{2,1}(Q)\) satisfies \({\mathcal {L}}_A^su=\) in Q and \(\text{ supp }(u(\cdot ,t))\cap B({\tilde{x}},r)\subset E_+({\tilde{x}},c)\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\), then \(u=0\) in \(B({\tilde{x}},\rho )\times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\).
Proof
We proceed similarly to the proof of Theorem 3.5. We keep the same notations as in Theorem 3.5. Let \(c^*=c^*(\varkappa ,{\mathfrak {m}})\) defined as in Theorem 3.5 and \(0<c<c^*\).
Fix \(0<\eta <1\) and take instead of \(\mathbf{Q }_j\), \(j=0,1,2\), in Theorem 3.5 the following sets
Also, the constants \(c_j\), \(j=0,1,2\) are substituted by the following ones
Straightforward computations show that choosing \(\gamma \) so that
guarantee that \(c_1<c_0\) and \(c_2<c_0\). We can then mimic the last part of Theorem 3.5 to derive that if \(u\in H^{2,1}(Q)\) satisfies \({\mathcal {L}}_A^su=\) in Q and \(\text{ supp }(u(\cdot ,t))\cap B({\tilde{x}},r)\subset E_+({\tilde{x}},c)\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\), then \(u=0\) in \(B({\tilde{x}},\rho _0)\times (-\eta {\mathfrak {t}}/2,\eta {\mathfrak {t}}/2)\). Since \(0<\eta <1\) is chosen arbitrarily we get, as expected, \(u=0\) in \(B({\tilde{x}},\rho _0)\times (-{\mathfrak {t}}/2, {\mathfrak {t}}/2)\). \(\square \)
We say that \({\mathcal {L}}_A^s\) has the reduced unique continuation property if, for any non empty open subset \({\mathcal {O}}\subset \Omega \) and for any \(u\in H^{2,1}(Q)\) satisfying \({\mathcal {L}}_A^su=0\) in Q and \(u=0\) in \({\mathcal {O}} \times (-{\mathfrak {t}},{\mathfrak {t}})\), we must have \(u=0\) in \(\Omega \times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\).
Theorem 6.4
There exists a neighborhood \({\mathcal {N}}\) of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }};{\mathbb {R}}^n\times {\mathbb {R}}^n)\) so that \({\mathcal {L}}_A^s\) has the reduced unique continuation property for any \(A\in {\mathcal {N}}\).
Proof
Let \({\mathcal {N}}\) be the neighborhood of \(\mathbf{I }\) in \(C^{2,1}({\overline{\Omega }},{\mathbb {R}}^{n\times n})\) given in Lemma 2.2. Pick \(u\in H^{2,1}(Q)\) satisfying \({\mathcal {L}}_A^su=0\) in Q and \(u=0\) in \({\mathcal {O}} \times (-{\mathfrak {t}},{\mathfrak {t}})\) for some non empty open subset \({\mathcal {O}}\subset \Omega \). Define \(\Omega _0\) as the maximal subdomain of \(\Omega \) so that \(u=0\) in \(\Omega _0\times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\). We claim that \(\Omega {\setminus } \overline{\Omega _0}\) is empty which is sufficient to give the expected result. Indeed if \(\Omega {\setminus } \overline{\Omega _0}\) is nonempty then we can proceed as in the proof of Theorem 3.6 to derive that u vanishes in \({\mathcal {U}}\times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\), for some \({\mathcal {U}}\), a neighborhood of a point in \(\partial \Omega _0\cap \Omega \). But this contradicts the maximality of \(\Omega _0\). \(\square \)
The uniqueness of continuation from the Cauchy data on a subboundary is given in the following corollary.
Corollary 6.1
Let \({\mathcal {N}}\) be as in Theorem 6.4 with \(\Omega \) substituted by larger domain \({\hat{\Omega }}\Supset \Omega \). Let \(\Gamma _0\) a nonempty open subset of \(\Gamma \) and \(\Sigma _0=\Gamma _0\times (-{\mathfrak {t}},{\mathfrak {t}})\). For \(A\in {\mathcal {N}}\), let \(u\in H^{2,1}(Q)\) satisfying \({\mathcal {L}}_A^wu=0\) in Q and \(u=\partial _\nu u=0\) on \(\Sigma _0\). Then \(u=0\) in \(\Omega \times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\).
Also, the unique continuation across a A-pseudo-convex hypersurface is contained in the following theorem.
Theorem 6.5
Let \(H=\{x\in \omega ;\;\theta (x)=\theta ({\tilde{x}})\}\) be a A-pseudo-convex hypersurface defined in a neighborhood of \({\tilde{x}}\in \Omega \) with \(\theta \in C^{3,1}({\overline{\omega }})\). Then there exists \({\mathcal {B}}\), a neighborhood of \({\tilde{x}}\), so that if \(u\in H^{2,1}(\omega \times (-{\mathfrak {t}},{\mathfrak {t}}))\) satisfies \({\mathcal {L}}_A^wu=0\) in \(\omega \times (-{\mathfrak {t}},{\mathfrak {t}})\) and \(\text{ supp }(u(\cdot ,t))\subset H_+=\{ x\in \omega ;\; \theta (x)\ge \theta ({\tilde{x}})\}\), \(t\in (-{\mathfrak {t}},{\mathfrak {t}})\), then \(u=0\) in \({\mathcal {B}}\times (-{\mathfrak {t}}/2,{\mathfrak {t}}/2)\).
6.3 Observability inequality
In this subsection \(t_1=0\) and \(t_2={\mathfrak {t}}>0\).
Let \({\mathscr {A}}:L^2(\Omega )\rightarrow L^2(\Omega )\) be the unbounded operator introduced in the preceding section. That is
It is known that \(u(t)=e^{it{\mathscr {A}}}u_0\), \(u_0\in L^2(\Omega )\) is the solution of the following IBVP
Furthermore, u belongs to \(C([0,{\mathfrak {t}}];D({\mathscr {A}}))\cap C^1([0,{\mathfrak {t}}];L^2(\Omega ))\) whenever \(u_0\in D({\mathscr {A}})\) and, for \(0\le t\le {\mathfrak {t}}\), we have
Theorem 6.6
Suppose that \(0\le \psi _0\in C^4({\overline{\Omega }})\) is A-pseudo-convex with constant \(\kappa >0\) and let \(\Gamma _+=\{x\in \Gamma ;\; \partial _{\nu _A}\psi _0(x)>0\}\). Then there exists a constant \(\aleph \) only depending \(\Omega \), \({\mathfrak {t}}\), \(\varkappa \), \(\kappa \) and \(\Gamma _+\), so that, for any \(u_0\in D({\mathscr {A}})\), we have
where \(\Sigma _+=\Gamma _+\times (0,{\mathfrak {t}})\) and \(u=e^{it{\mathscr {A}}}u_0\).
Proof
In light of Theorem 6.1 and identities (6.11), the expected inequality can be proved by modifying slightly that of the wave equation. \(\square \)
Remark 6.1
It is worth mentioning that the results for the elliptic, wave and Schrödinger equations can be extended to the case where \(\Delta _A\) is substituted by the associated magnetic operator defined by
with \(\mathbf{b }=(b_1,\ldots ,b_n)\in W^{1,\infty }(\Omega ;{\mathbb {R}}^n)\).
Note that \(\Delta _{A,\mathbf{b }}u\) can be rewritten in the following form
References
Alinhac, S.: Non-unicité du problème de Cauchy. Ann. Math. (2) 117, 77–108 (1983)
Alinhac, S., Baouendi, M.S.: A non uniqueness result for operators of principal type. Math. Z. 220, 561–568 (1995)
Baudouin, L., Puel, J.-P.: Détermination du potentiel dans l’équation de Schrödinger à partir de mesures sur une partie du bord. C. R. Math. Acad. Sci. Paris 334(11), 967–972 (2002)
Baudouin, L., Puel, J.-P.: Uniqueness and stability in an inverse problem for the Schrödinger equation. Inverse Probl. 18(6), 1537–1554 (2002)
Baudouin, L., Puel, J.-P.: Corrigendum: “Uniqueness and stability in an inverse problem for the Schrödinger equation” [Inverse Problems 18 (6) (2002), 1537–1554]. Inverse Probl. 23(3), 1327–1328 (2007)
Bellassoued, M., Choulli, M.: Global logarithmic stability of the Cauchy problem for anisotropic wave equations. arXiv:1902.05878
Bellassoued, M., Yamamoto, M.: Carleman Estimates and Applications to Inverse Problems for Hyperbolic Systems, Springer Monographs in Mathematics, pp. xii+260. Springer, Tokyo (2017)
Bourgeois, L.: Quantification of the unique continuation property for the heat equation. Math. Control Relat. Fields 7(3), 347–367 (2017)
Bourgeois, L.: About stability and regularization of ill-posed elliptic Cauchy problems: the case of C\(^{1,1}\)-domains, M2AN Math. Model. Numer. Anal. 44(4), 715–735 (2010)
Choulli, M.: New global logarithmic stability result for the Cauchy problem for elliptic equations. Bull. Aust. Math. Soc. 101(1), 141–145 (2020)
Choulli, M.: Applications of Elliptic Carleman Inequalities to Cauchy and Inverse Problems, SpringerBriefs in Mathematics, BCAM SpringerBriefs, pp. ix+81. Springer, Bilbao (2016)
Choulli, M.: Une Introduction Aux Problèmes Inverses Elliptiques et Paraboliques, Mathématiques & Applications, vol. 65, pp. xxii+249. Springer, Berlin (2009)
Choulli, M., Yamamoto, M.: Logarithmic global stability of parabolic Cauchy problems. J. Inverse Ill Posed Probl. (to appear)
Duyckaerts, T., Zhang, X., Zuazua, E.: On the optimality of the observability inequalities for parabolic and hyperbolic systems with potentials. Annales de l’Institut Henri Poincaré (C), Analyse Non Linéaire 25, 1–41 (2008)
Fernàndez-Cara, E., Guerrero, S.: Global Carleman inequalities for parabolic systems and applications to controllability. SIAM J. Control. Optim. 45(4), 1399–1446 (2006)
Fu, X., Lü, Q., Zhang, X.: Carleman Estimates for Second Order Partial Differential Operators and Applications. A Unified Approach, SpringerBriefs in Mathematics. BCAM SpringerBriefs, pp. xi+127. Springer (2019)
Fursikov, A.V., Imanuvilov, O.Yu.: Controllability of Evolution Equations, Lecture Notes Series. Seoul National Univ. (1996)
Hörmander, L.: Linear Partial Differential Operators, Fourth Printing, p. 285. Springer, Berlin (1976)
Hörmander, L.: The Analysis of Linear Partial Differential Operators IV. Fourier Integral Operators, Reprint of the 1994 Edition, Classics in Mathematics, pp. viii+352. Springer, Berlin (2009)
Huang, X.: Carleman Estimate for a General Second-order Hyperbolic Equation, Inverse Problems and Related Topics, Springer Proceedings in Mathematics and Statistics, pp. 149–165. Springer, Singapore (2020)
Isakov, V.: Inverse Problems for Partial Differential Equations, Applied Mathematical Sciences, vol. 127, 3rd edn, pp. xv+406. Springer, Cham (2017)
John, F.: Partial Differential Equations, Applied Mathematical Sciences, vol. 1, 4th edn, pp. x+249. Springer, New York (1986)
Lasiecka, I., Triggiani, R., Zhang, X.: Global uniqueness, observability and stabilization of nonconservative Schrödinger equations via pointwise Carleman estimates. Part: I \(H^1(\Omega )\)-estimates. J. Inverse Ill Posed Probl. 12, 43–123 (2004)
Laurent, C., Léautaud, M.: Quantitative unique continuation for operators with partially analytic coefficients. Application to approximate control for waves. J. Eur. Math. Soc. 21(4), 957–1069 (2019)
Le Rousseau, J., Lebeau, G.: On Carleman estimates for elliptic and parabolic operators. Applications to unique continuation and control of parabolic equations. ESAIM Control Optim. Calc. Var. 18(3), 712–747 (2012)
Mercado, A., Osses, A., Rosier, L.: Carleman inequalities and inverse problems for the Schrödinger equation. C. R. Math. Acad. Sci. Paris 346(1–2), 53–58 (2008)
Mercado, A., Osses, A., Rosier, L.: Inverse problems for the Schrödinger equation via Carleman inequalities with degenerate weights. Inverse Probl. 24(1), 015017 (2008). (18 pp)
Nirenberg, L.: Uniqueness in Cauchy problems for differential equations with constant leading coefficients. Commun. Pure Appl. Math. 10, 89–105 (1957)
Ouhabaz, E. M.: Analysis of Heat Equations on Domains, London Mathematical Society Monographs Series, vol. 31, pp. xiv+284. Princeton University Press, Princeton (2005)
Renardy, M., Rogers, R.C.: An Introduction to Partial Differential Equations, Texts in Applied Mathematics, vol. 13, pp. xiv+428. Springer, New York (1993)
Robbiano, L.: Théorème d’unicité adapté au contrôle des solutions des problèmes hyperboliques. Commun. Partial Differ. Equ. 16(4–5), 789–800 (1991)
Saut, J.-C., Scheurer, B.: Un théorème de prolongement unique pour des opérateurs elliptiques dont les coefficients ne sont pas localement bornés. C. R. Acad. Sci. Paris Sér. A-B 290(13), A595–A598 (1980)
Saut, J.-C., Scheurer, B.: Sur l’unicité du problème de Cauchy et le prolongement unique pour des équations elliptiques à coefficients non localement bornés. J. Differ. Equ. 43(1), 28–43 (1982)
Saut, J.-C., Scheurer, B.: Remarques sur un théorème de prolongement unique de Mizohata. C. R. Acad. Sci. Paris Sér. I Math. 296(6), 307–310 (1983)
Saut, J.-C., Scheurer, B.: Unique continuation for some evolution equations. J. Differ. Equ. 66(1), 118–139 (1987)
Shao, A.: On Carleman and observability estimates for wave equations on time-dependent domains. Proc. Lond. Math. Soc. 119, 998–1064 (2019)
Yao, P.-F.: On the observability inequalities for exact controllability of wave equations with variable coefficients. SIAM J. Control. Optim. 37(5), 1568–1599 (1999)
Zhang, X.: Explicit observability estimate for the wave equation with potential and its application. R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 456, 1101–1115 (2000)
Zhang, X.: Exact controllability of the semilinear plate equations. Asymptot. Anal. 27, 95–125 (2001)
Zuily, C.: Uniqueness and Nonuniqueness in the Cauchy Problem, Progress in Mathematics, vol. 33. Birkhüser Boston, Inc., Boston, pp. xi+168 (1983)
Author information
Authors and Affiliations
Corresponding author
Additional information
This article is part of the section “Theory of PDEs” edited by Eduardo Teixeira.
The author is supported by the grant ANR-17-CE40-0029 of the French National Research Agency ANR (project MultiOnde).
Rights and permissions
About this article
Cite this article
Choulli, M. The unique continuation property for second order evolution PDEs. Partial Differ. Equ. Appl. 2, 67 (2021). https://doi.org/10.1007/s42985-021-00123-6
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s42985-021-00123-6
Keywords
- Wave equation
- Parabolic equation
- Schrödinger equation
- Elliptic equation
- Carleman inequality
- Pseudo-convexity condition
- Non characteristic hypersurface
- Property of unique continuation
- Observability inequality