Introduction
Formulation of the problem
For ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 , let Ω ⊂ R ℓ \Omega\subset\mathbb{R}^{\ell} Ω ⊂ R ℓ be an open, bounded, and simply connected set with smooth boundary ∂ Ω \partial\Omega ∂ Ω . Given T > 0 T > 0 T > 0 , we denote by Ω T : = ( 0 , T ) × Ω \Omega_T := (0,T) \times\Omega Ω T := ( 0 , T ) × Ω and its lateral boundary by ∂ Ω T : = ( 0 , T ) × ∂ Ω \partial\Omega_T := (0,T) \times\partial\Omega ∂ Ω T := ( 0 , T ) × ∂ Ω . In this article, we investigate the inverse problem associated with identifying the coefficients of a coupled nonlinear Klein-Gordon operator given by
L P , Q u ⃗ : = [ ∂ t 2 u 1 − Δ u 1 + p 11 ( x ) u 1 + p 12 ( x ) u 2 + q 11 ( x ) u 1 3 + q 12 ( x ) u 1 2 u 2 ∂ t 2 u 2 − Δ u 2 + p 21 ( x ) u 1 + p 22 ( x ) u 2 + q 21 ( x ) u 2 2 u 1 + q 22 ( x ) u 2 3 ] . (1) \mathcal{L}_{\mathbb{P},\mathbb{Q}} \vec{u} :=
\begin{bmatrix}
\partial_t^2 u_1 - \Delta u_1 + p_{11}(x)u_1 + p_{12}(x)u_2 + q_{11}(x)u_1^3 + q_{12}(x)u_1^2u_2 \\
\partial_t^2 u_2 - \Delta u_2 + p_{21}(x)u_1 + p_{22}(x)u_2 + q_{21}(x)u_2^2u_1 + q_{22}(x)u_2^3
\end{bmatrix}. \tag*{(1)} L P , Q u := [ ∂ t 2 u 1 − Δ u 1 + p 11 ( x ) u 1 + p 12 ( x ) u 2 + q 11 ( x ) u 1 3 + q 12 ( x ) u 1 2 u 2 ∂ t 2 u 2 − Δ u 2 + p 21 ( x ) u 1 + p 22 ( x ) u 2 + q 21 ( x ) u 2 2 u 1 + q 22 ( x ) u 2 3 ] . (1)
where u ⃗ : = ( u 1 , u 2 ) T \vec{u} := (u_1,u_2)^{\mathsf{T}} u := ( u 1 , u 2 ) T represents the displacement vector and P , Q \mathbb{P}, \mathbb{Q} P , Q denote the matrix-valued potentials given by
P ( x ) : = [ p 11 ( x ) p 12 ( x ) p 21 ( x ) p 22 ( x ) ] and Q ( x ) : = [ q 11 ( x ) q 12 ( x ) q 21 ( x ) q 22 ( x ) ] . \mathbb{P}(x) :=
\begin{bmatrix}
p_{11}(x) & p_{12}(x) \\
p_{21}(x) & p_{22}(x)
\end{bmatrix}
\quad\text{and}\quad
\mathbb{Q}(x) :=
\begin{bmatrix}
q_{11}(x) & q_{12}(x) \\
q_{21}(x) & q_{22}(x)
\end{bmatrix}. P ( x ) := [ p 11 ( x ) p 21 ( x ) p 12 ( x ) p 22 ( x ) ] and Q ( x ) := [ q 11 ( x ) q 21 ( x ) q 12 ( x ) q 22 ( x ) ] .
Physically, the Klein-Gordon equation describes the dynamics of waves or particles at the subatomic level. The cubic nonlinearity comes into play when quartic scalar fields interact. Moreover, this model is used to study solitary-wave interactions [38 ] , energy transfer [18 ] , and the stability of localized waves [28 , 45 ] .
In this article, we consider the following initial boundary value problem (IBVP) associated with the aforementioned Klein-Gordon operator
{ L P , Q u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (2) \begin{cases}
\mathcal{L}_{\mathbb{P},\mathbb{Q}}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(t,x)=\vec{f}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)=\vec{0},\quad\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega. \tag*{(2)}
\end{cases} ⎩ ⎨ ⎧ L P , Q u ( t , x ) = 0 , u ( t , x ) = f ( t , x ) , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (2)
We plan to establish the unique recovery of the coefficients P , Q ∈ C c ∞ ( Ω ) \mathbb{P},\mathbb{Q}\in C^\infty_c(\Omega) P , Q ∈ C c ∞ ( Ω ) appearing in (2) , from the knowledge of Dirichlet-to-Neumann maps measured on whole of ∂ Ω T \partial\Omega_T ∂ Ω T and a suitable open subset of it.
Throughout this article, the boldface notation is used to indicate that the corresponding object is componentwise in the underlying function space. For example, writing P ∈ C c ∞ ( Ω ) P\in C^\infty_c(\Omega) P ∈ C c ∞ ( Ω ) means that each component of P P P belongs to C c ∞ ( Ω ) C^\infty_c(\Omega) C c ∞ ( Ω ) . In all of our discussion, we use the notation □ \square □ interchangeably to denote the standard wave operator ∂ t 2 − Δ \partial_t^2-\Delta ∂ t 2 − Δ . In addition, the notation “∘ \circ ∘ ” is used to denote the component-wise multiplication of two vectors, that is,
[ v 1 v 2 ] ∘ [ w 1 w 2 ] : = [ v 1 w 1 v 2 w 2 ] . \begin{bmatrix}v_1\\v_2\end{bmatrix}\circ
\begin{bmatrix}w_1\\w_2\end{bmatrix}
:=
\begin{bmatrix}v_1w_1\\v_2w_2\end{bmatrix}. [ v 1 v 2 ] ∘ [ w 1 w 2 ] := [ v 1 w 1 v 2 w 2 ] .
We make use of the above notation to rewrite L P , Q u ⃗ \mathcal{L}_{\mathbb{P},\mathbb{Q}}\vec{u} L P , Q u as
L P , Q u ⃗ = ( ∂ t 2 − Δ ) ( I 2 ) u ⃗ + P u ⃗ + ( Q u ⃗ ) ∘ ( u ⃗ ∘ u ⃗ ) , \mathcal{L}_{\mathbb{P},\mathbb{Q}}\vec{u}
=(\partial_t^2-\Delta)(I_2)\vec{u}+\mathbb{P}\vec{u}+(\mathbb{Q}\vec{u})\circ(\vec{u}\circ\vec{u}), L P , Q u = ( ∂ t 2 − Δ ) ( I 2 ) u + P u + ( Q u ) ∘ ( u ∘ u ) ,
where I 2 I_2 I 2 represents the identity matrix of order 2. For f ⃗ ∈ F m + 1 σ \vec{f}\in F^\sigma_{m+1} f ∈ F m + 1 σ , where F m + 1 σ F^\sigma_{m+1} F m + 1 σ is given by (A.6), the IBVP (2) admits a unique solution u ⃗ ∈ B R ( 0 ) ⊂ D m + 1 \vec{u}\in B_R(0)\subset\mathcal{D}_{m+1} u ∈ B R ( 0 ) ⊂ D m + 1 (we refer the reader to Appendix § 6 for the details). Based on the well-posedness of the forward model (2) , we define the Dirichlet-to-Neumann (DN) map Λ P , Q ( f ⃗ ) : F m + 1 σ → H m ( ∂ Ω T ) \Lambda_{\mathbb{P},\mathbb{Q}}(\vec{f}):F^\sigma_{m+1}\to H^m(\partial\Omega_T) Λ P , Q ( f ) : F m + 1 σ → H m ( ∂ Ω T ) , given by
Λ P , Q ( f ⃗ ) : = ∂ ν u ⃗ ∣ ∂ Ω T , \Lambda_{\mathbb{P},\mathbb{Q}}(\vec{f}):=\partial_\nu\vec{u}\big|_{\partial\Omega_T}, Λ P , Q ( f ) := ∂ ν u ∂ Ω T ,
where ν \nu ν denotes the unit normal vector pointing outward to ∂ Ω \partial\Omega ∂ Ω and u ⃗ \vec{u} u is a solution to IBVP (2) . Set
P k ( x ) : = [ p 11 k ( x ) p 12 k ( x ) p 21 k ( x ) p 22 k ( x ) ] and Q k ( x ) : = [ q 11 k ( x ) q 12 k ( x ) q 21 k ( x ) q 22 k ( x ) ] , \mathbb{P}^k(x):=
\begin{bmatrix}
p^k_{11}(x)&p^k_{12}(x)\\
p^k_{21}(x)&p^k_{22}(x)
\end{bmatrix}
\quad\text{and}\quad
\mathbb{Q}^k(x):=
\begin{bmatrix}
q^k_{11}(x)&q^k_{12}(x)\\
q^k_{21}(x)&q^k_{22}(x)
\end{bmatrix}, P k ( x ) := [ p 11 k ( x ) p 21 k ( x ) p 12 k ( x ) p 22 k ( x ) ] and Q k ( x ) := [ q 11 k ( x ) q 21 k ( x ) q 12 k ( x ) q 22 k ( x ) ] ,
for k ∈ { 1 , 2 } k\in\{1,2\} k ∈ { 1 , 2 } , where k k k denotes whether the coefficients belong to P k \mathbb{P}^k P k or Q k \mathbb{Q}^k Q k ; this notation should be interpreted as an index, not an exponentiation operation. We now present our first result, which establishes the unique identification of the coefficients from the full DN map and is stated as follows.
Theorem 1.1. Let T > diam ( Ω ) T>\operatorname{diam}(\Omega) T > diam ( Ω ) , m > ℓ + 1 m>\ell+1 m > ℓ + 1 with ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and P k , Q k ∈ C c ∞ ( Ω ) \mathbb{P}^k,\mathbb{Q}^k\in C^\infty_c(\Omega) P k , Q k ∈ C c ∞ ( Ω ) for k ∈ { 1 , 2 } k\in\{1,2\} k ∈ { 1 , 2 } . Now if
Λ P 1 , Q 1 ( f ⃗ ) = Λ P 2 , Q 2 ( f ⃗ ) for all f ⃗ ∈ F m + 1 σ , \Lambda_{\mathbb{P}^1,\mathbb{Q}^1}(\vec{f})=\Lambda_{\mathbb{P}^2,\mathbb{Q}^2}(\vec{f})\quad\text{for all }\vec{f}\in F^\sigma_{m+1}, Λ P 1 , Q 1 ( f ) = Λ P 2 , Q 2 ( f ) for all f ∈ F m + 1 σ , then P 1 = P 2 \mathbb{P}^1=\mathbb{P}^2 P 1 = P 2 , and Q 1 = Q 2 \mathbb{Q}^1=\mathbb{Q}^2 Q 1 = Q 2 in Ω \Omega Ω .
Next, we extend the above result and establish the unique identification of the coefficient from the partial DN map Λ P , Q ♯ ( f ⃗ ) \Lambda^\sharp_{\mathbb{P},\mathbb{Q}}(\vec{f}) Λ P , Q ♯ ( f ) , which is defined by
Λ P , Q ♯ ( f ⃗ ) : = ∂ ν u ⃗ ∣ ( 0 , T ) × Γ , for all f ⃗ ∈ F m + 1 σ , (3) \Lambda^\sharp_{\mathbb{P},\mathbb{Q}}(\vec{f}):=\partial_\nu\vec{u}\big|_{(0,T)\times\Gamma},\quad\text{for all }\vec{f}\in F^\sigma_{m+1}, \tag*{(3)} Λ P , Q ♯ ( f ) := ∂ ν u ( 0 , T ) × Γ , for all f ∈ F m + 1 σ , (3)
where Γ ⊂ ∂ Ω \Gamma\subset\partial\Omega Γ ⊂ ∂ Ω be a nonempty open set, T > 0 T > 0 T > 0 is sufficiently large and u ⃗ \vec{u} u be the solution to (2) . Now, to formalize the exact statement of the desired problem, we first consider a nonempty relatively open neighborhood of ∂ Ω \partial\Omega ∂ Ω in Ω ‾ \overline{\Omega} Ω , say U ⊆ Ω ‾ \mathcal{U} \subseteq\overline{\Omega} U ⊆ Ω . Also, we introduce an admissible class P \mathcal{P} P , given by
P : = { ( P , Q ) ∈ C ∞ ( Ω ‾ ) × C ∞ ( Ω ‾ ) : P = Q = 0 , in U } . \mathcal{P}:=\left\{(\mathbb{P},\mathbb{Q}) \in C^\infty(\overline{\Omega}) \times C^\infty(\overline{\Omega}): \mathbb{P}=\mathbb{Q}=0,\ \text{in }\mathcal{U}\right\}. P := { ( P , Q ) ∈ C ∞ ( Ω ) × C ∞ ( Ω ) : P = Q = 0 , in U } .
Now we describe our next result related to the unique recovery of coefficients appearing in (2) from the partial DN map and is stated as follows:
Theorem 1.2. Let T T T is sufficiently large, m > ℓ + 1 m>\ell+1 m > ℓ + 1 with ℓ ≥ 2 \ell\geq2 ℓ ≥ 2 and P k , Q k ∈ C c ∞ ( Ω ) \mathbb{P}^{k},\mathbb{Q}^{k}\in C^\infty_{\mathrm{c}}(\Omega) P k , Q k ∈ C c ∞ ( Ω ) . Now if
Λ P 1 , Q 1 ♯ ( f ⃗ ) = Λ P 2 , Q 2 ♯ ( f ⃗ ) for every f ⃗ ∈ F c m + 1 , \Lambda^{\sharp}_{\mathbb{P}^{1},\mathbb{Q}^{1}}(\vec{f})=\Lambda^{\sharp}_{\mathbb{P}^{2},\mathbb{Q}^{2}}(\vec{f})\ \text{for every }\vec{f}\in\mathcal{F}^{m+1}_{c}, Λ P 1 , Q 1 ♯ ( f ) = Λ P 2 , Q 2 ♯ ( f ) for every f ∈ F c m + 1 , then P 1 = P 2 \mathbb{P}^{1}=\mathbb{P}^{2} P 1 = P 2 and Q 1 = Q 2 \mathbb{Q}^{1}=\mathbb{Q}^{2} Q 1 = Q 2 in Ω \Omega Ω , provided ( P 1 − P 2 , Q 1 − Q 2 ) ∈ P (\mathbb{P}^{1}-\mathbb{P}^{2},\mathbb{Q}^{1}-\mathbb{Q}^{2})\in\mathcal{P} ( P 1 − P 2 , Q 1 − Q 2 ) ∈ P .
Remark 1.3. A sufficiently large T T T guarantees the existence of diam ( Ω ) < T ′ < T \operatorname{diam}(\Omega)<T'<T diam ( Ω ) < T ′ < T , used in § 5.2 and § 5.3 to establish the uniqueness of the coefficients given in Theorem 1.2 .
Existing papers
In recent years, the study of inverse problems related to systems of equations has grabbed the attention of researchers. In 1993, Isakov [22 ] introduced the technique of higher-order linearization and solved an inverse problem associated with a nonlinear parabolic partial differential equation (PDE). Since then, this methodology has enabled researchers to address inverse problems associated with nonlinear equations. In [44 ] , the authors considered polynomial nonlinearity in the wave operator and established simultaneous recovery of these coefficients from boundary measurements. In [6 ] , Bhardwaj et al. considered a coefficient-determination inverse problem for the semilinear damped wave operator with polynomial nonlinearity of order ℓ \ell ℓ , assuming coefficients are time-independent. They derived pointwise reconstruction formulae for the damping coefficient, linear and nonlinear potentials from the DN map. For some studies on the scalar nonlinear wave equation, we refer readers to [40 , 41 ] in the Euclidean setup, and [1 , 9 , 14 , 15 , 20 , 23 , 24 , 27 , 33 , 36 , 37 , 43 , 50 ] in the manifold setup, where the authors used an input–output map, or DN map, or source-to-solution map to determine the coefficients. For a one-dimensional nonlinear Klein-Gordon equation, we refer to [48 ] , where a time-dependent coefficient multiplied to nonlinear coefficients is uniquely determined using the finite difference scheme.
In comparison to the scalar wave equation, the inverse problem associated with the system of equations is studied less. The unique recovery of the time-independent potential in a wave equation has been studied in [3 ] . EskIn and Ralston [12 ] uniquely determined both the zeroth-order and first-order matrix-valued coefficients appearing in Schrödinger equation from the DN map. Uniqueness of a time-dependent matrix potential is demonstrated in [25 ] in one spatial dimension. Waters [51 ] established a unique identification of sound speed appearing in a coupled semilinear wave system from the source-to-solution map. In [39 ] , Mishra and Vashisth established unique recovery of a zeroth-order matrix-valued potential appearing in a perturbed linear system of wave equations using partial boundary data. Kumar et al. [32 ] studied an analogous system in an infinite waveguide and derived the stability estimate. Recently, [5 , 30 , 31 ] discussed the stability in determining the time-dependent matrix potential for the wave equation from partial boundary measurements.
On a global hyperbolic manifold, Filippas et al. [16 ] uniquely identified the time-dependent matrix-valued potential in the presence of a connection 1 1 1 -form (time-independent) using a source-to-solution map. In a Lorentzian manifold, Gomes et al. [17 ] study the unique determination of a connection and potential (up to a natural gauge) appearing in a wave operator from the Dirichlet-to-Neumann map. For some other works related to the systems, we refer the reader to [2 , 8 , 9 , 10 , 19 , 26 , 49 ] .
To the best of the authors knowledge, the present work is the first to deal with inverse problems related to simultaneous recovery of time-independent matrix-valued potentials and nonlinear potentials in a nonlinear coupled wave system from the boundary measurements. In this article, we uniquely identify the time-independent coefficients from both full and partial DN maps. For the full data setup, we use the geometric optics solutions given in Lemma 2.1 and 2.2 to establish the unique recovery of zeroth-order coefficients. However, to establish the unique identification of nonlinear coefficients, we use the asymptotic solutions given in Proposition 2.3 and 2.4. In the case of the partial DN map, we use the unique continuation argument given in Theorem 4.1 to convert the problem into some sort of full data problem and then use the previous analysis to establish the unique determination of the coefficients. The study of the Klein-Gordon equation in [47 ] and the inverse problem discussed in [4 , 6 , 29 ] motivates us to consider the problems stated in Theorem 1.1 and 1.2.
The rest of this article is arranged as follows: § 3 is devoted to the study of geometric optics solutions. Next, we establish the unique recovery of the zeroth-order potential and nonlinear coefficients from the full DN map in § 4 . In § 5 , we uniquely determined the coefficients from the partial DN map using the unique continuation principle. Finally, in Appendix § 6 , we discussed the well-posedness of the forward model (2) .
Acknowledgments
The first author acknowledges financial support from the Council of Scientific and Industrial Research (CSIR), India (File No. 09/1005(19269)/2024-EMR-I). The second author was supported by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 258734477 – SFB 1173. The last author acknowledges the Anusandhan National Research Foundation (ANRF), Government of India, for financial support provided through the ARG-MATRICS Grant (File No. ANRF/ARGM/2025/002368/MTR). The authors further acknowledge partial financial support from the Department of Science and Technology (DST), Government of India, under the FIST Programme (Reference No. SR/FST/MS-I/2018/22(C)).
Special solutions
In this section, we recall and extend some existing results needed for the upcoming analysis. We use the term special solutions for geometric optics solutions and asymptotic solutions. We split this section into two subsections: first, we invoke the geometric optics solutions from [21 Lemma (2,3)] and extend them to system of equations in § 3.1 . Next, in § 3.2 , we establish the asymptotic solution associated to initial-value problem (9) and (10) ; However, in contrast to geometric optics solutions (6) and (8) , the asymptotic solutions are not directly obtained by multiplying with a constant vector to that one derived for scalar equation [6 ] , see § 3.2 below.
Geometric optics solutions
We begin with introducing the notations used in the forthcoming study. Let Φ \Phi Φ solve the transport equation
( ∂ t − θ ⋅ ∇ ) Φ ( t , x ) = 0. (4) (\partial_t-\theta\cdot\nabla)\Phi(t,x)=0. \tag*{(4)} ( ∂ t − θ ⋅ ∇ ) Φ ( t , x ) = 0. (4)
In particular, we choose Φ \Phi Φ as follows: For φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 , define
Φ : = φ ( x + t θ ) \Phi:=\varphi(x+t\theta) Φ := φ ( x + tθ )
Then Φ \Phi Φ solves the transport equation (4) . Also, let L P \mathcal{L}_{\mathbf{P}} L P denotes the linear part of the operator L P , Q \mathcal{L}_{\mathbf{P},\mathbf{Q}} L P , Q given in (1) , that is,
L P u ⃗ = ( ∂ t 2 − Δ ) ( I 2 ) u ⃗ + P u ⃗ , \mathcal{L}_{\mathbf{P}}\vec{u}=(\partial_t^2-\Delta)(I_2)\vec{u}+\mathbf{P}\vec{u}, L P u = ( ∂ t 2 − Δ ) ( I 2 ) u + P u ,
and ϕ ( t , x ) = t + x ⋅ θ \phi(t,x)=t+x\cdot\theta ϕ ( t , x ) = t + x ⋅ θ . Then by direct computations, we observe that the geometric optics (GO) solution for the initial value problem (IVP)
{ L P u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( 0 , x ) = ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , (5) \begin{cases}
\mathcal{L}_{\mathbf{P}}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(0,x)=\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega, \tag*{(5)}
\end{cases} { L P u ( t , x ) = 0 , u ( 0 , x ) = ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , (5)
can be derived by extending the GO solution established in [21 Lemma 2] for a scalar wave equation, as stated in the following lemma.
Lemma 2.1. Let ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and let Ω ⊂ R ℓ \Omega\subset\mathbb{R}^{\ell} Ω ⊂ R ℓ be a bounded, open, and connected set with smooth boundary ∂ Ω \partial\Omega ∂ Ω . For any λ > 0 \lambda>0 λ > 0 , any φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) such that supp ( φ ) ∩ Ω = ∅ \operatorname{supp}(\varphi)\cap\Omega=\varnothing supp ( φ ) ∩ Ω = ∅ , and any θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 , the IVP (5) admits a solution of the form
u ⃗ ( + ) ( t , x ) = exp ( i λ ϕ ) Φ ( t , x ) K ⃗ ( + ) + R ⃗ K ⃗ ( + ) ( t , x , λ ) , (6) \vec{u}^{(+)}(t,x)=\exp(i\lambda\phi)\Phi(t,x)\vec{K}^{(+)}+\vec{R}_{\vec{K}^{(+)}}(t,x,\lambda), \tag*{(6)} u ( + ) ( t , x ) = exp ( iλ ϕ ) Φ ( t , x ) K ( + ) + R K ( + ) ( t , x , λ ) , (6)
where K ⃗ ( + ) : = ( K 1 + , K 2 + ) T ∈ R 2 \vec{K}^{(+)}:=(K_1^+,K_2^+)^\mathsf{T}\in\mathbb{R}^2 K ( + ) := ( K 1 + , K 2 + ) T ∈ R 2 is constant vector and the remainder term R ⃗ K ⃗ ( + ) ( t , x , λ ) \vec{R}_{\vec{K}^{(+)}}(t,x,\lambda) R K ( + ) ( t , x , λ ) satisfy the following
R ⃗ K ⃗ ( + ) ( 0 , x , λ ) = ∂ t R ⃗ K ⃗ ( + ) ( 0 , x , λ ) = 0 ⃗ , in Ω \vec{R}_{\vec{K}^{(+)}}(0,x,\lambda)=\partial_t\vec{R}_{\vec{K}^{(+)}}(0,x,\lambda)=\vec{0},\quad\text{in }\Omega R K ( + ) ( 0 , x , λ ) = ∂ t R K ( + ) ( 0 , x , λ ) = 0 , in Ω
and
λ ∥ R ⃗ K ⃗ ( + ) ∥ L 2 ( Ω T ) + ∥ ∂ t R ⃗ K ⃗ ( + ) ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) \lambda\|\vec{R}_{\vec{K}^{(+)}}\|_{L^2(\Omega_T)}+\|\partial_t\vec{R}_{\vec{K}^{(+)}}\|_{L^2(\Omega_T)}\le C\|\varphi\|_{H^3(\mathbb{R}^{\ell})} λ ∥ R K ( + ) ∥ L 2 ( Ω T ) + ∥ ∂ t R K ( + ) ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ )
for some constant C > 0 C>0 C > 0 , independent of λ \lambda λ .
Also, the GO solution associated with the backward problem given by
{ L P u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( T , x ) = ∂ t u ⃗ ( T , x ) = 0 ⃗ , x ∈ Ω , (7) \begin{cases}
\mathcal{L}_{\mathbf{P}}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(T,x)=\partial_t\vec{u}(T,x)=\vec{0}, & x\in\Omega, \tag*{(7)}
\end{cases} { L P u ( t , x ) = 0 , u ( T , x ) = ∂ t u ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , (7)
can be derived by extending the GO solution established in [21 Lemma 3] for a scalar wave equation as stated in the following lemma.
Lemma 2.2. Let ℓ ≥ 2 \ell\ge2 ℓ ≥ 2 and Ω ⊂ R ℓ \Omega\subset\mathbb{R}^{\ell} Ω ⊂ R ℓ be a bounded, open, and connected set with smooth boundary ∂ Ω \partial\Omega ∂ Ω . For any λ > 0 \lambda>0 λ > 0 , any φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) such that ( supp ( φ ) ± T θ ) ∩ Ω = ∅ (\operatorname{supp}(\varphi)\pm T\theta)\cap\Omega=\varnothing ( supp ( φ ) ± T θ ) ∩ Ω = ∅ , and any θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 , then the backward problem (7) admits a solution of the form
u ⃗ ( − ) ( t , x ) = exp ( − i λ ϕ ) Φ ( t , x ) K ⃗ ( − ) + R ⃗ K ⃗ ( − ) ( t , x , λ ) , (8) \vec{u}^{(-)}(t,x)=\exp(-i\lambda\phi)\Phi(t,x)\vec{K}^{(-)}+\vec{R}_{\vec{K}^{(-)}}(t,x,\lambda), \tag*{(8)} u ( − ) ( t , x ) = exp ( − iλ ϕ ) Φ ( t , x ) K ( − ) + R K ( − ) ( t , x , λ ) , (8)
where K ⃗ ( − ) : = ( K 1 − , K 2 − ) T ∈ R 2 \vec{K}^{(-)}:=(K_1^-,K_2^-)^\mathsf{T}\in\mathbb{R}^2 K ( − ) := ( K 1 − , K 2 − ) T ∈ R 2 is constant vector and the remainder term R ⃗ K ⃗ ( − ) ( t , x , λ ) \vec{R}_{\vec{K}^{(-)}}(t,x,\lambda) R K ( − ) ( t , x , λ ) satisfying
R ⃗ K ⃗ ( − ) ( T , x , λ ) = ∂ t R ⃗ K ⃗ ( − ) ( T , x , λ ) = 0 ⃗ in Ω . \vec{R}_{\vec{K}^{(-)}}(T,x,\lambda)=\partial_t\vec{R}_{\vec{K}^{(-)}}(T,x,\lambda)=\vec{0}\quad\text{in }\Omega. R K ( − ) ( T , x , λ ) = ∂ t R K ( − ) ( T , x , λ ) = 0 in Ω.
Moreover, R ⃗ K ⃗ ( − ) ( t , x , λ ) \vec{R}_{\vec{K}^{(-)}}(t,x,\lambda) R K ( − ) ( t , x , λ ) satisfies the following estimate
λ ∥ R ⃗ K ⃗ ( − ) ∥ L 2 ( Ω T ) + ∥ ∂ t R ⃗ K ⃗ ( − ) ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . \lambda\|\vec{R}_{\vec{K}^{(-)}}\|_{L^2(\Omega_T)}+\|\partial_t\vec{R}_{\vec{K}^{(-)}}\|_{L^2(\Omega_T)}\le C\|\varphi\|_{H^3(\mathbb{R}^{\ell})}. λ ∥ R K ( − ) ∥ L 2 ( Ω T ) + ∥ ∂ t R K ( − ) ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) .
We observe that the unique reconstruction of the matrix-valued potential P \mathbf{P} P is manageable with the help of GO solutions derived in Lemmas 2.1 and 2.2, however, as discussed in [6 §3] , the recovery of the nonlinear potential Q \mathbf{Q} Q is not feasible due to the products of three or more L 2 L^2 L 2 functions, which need not be dense in L 1 L^1 L 1 . This motivates us to look for asymptotic solutions to resolve this issue.
Asymptotic solutions
In this subsection, we derive the asymptotic solutions to the IVP given by
{ L P u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( 0 , x ) = ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , (9) \begin{cases}
\mathcal{L}_{P}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(0,x)=\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega, \tag*{(9)}
\end{cases} { L P u ( t , x ) = 0 , u ( 0 , x ) = ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , (9)
and its formal L 2 L^2 L 2 -adjoint given by
{ L P T u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( T , x ) = ∂ t u ⃗ ( T , x ) = 0 ⃗ , x ∈ Ω . (10) \begin{cases}
\mathcal{L}_{P^{\mathrm T}}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(T,x)=\partial_t\vec{u}(T,x)=\vec{0}, & x\in\Omega. \tag*{(10)}
\end{cases} { L P T u ( t , x ) = 0 , u ( T , x ) = ∂ t u ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω. (10)
These solutions are generalization of the asymptotic solutions established in [6 §3] for the scalar equations. In contrast to the geometric optics result above, simply multiplying the constant vector by the one derived in [6 ] for the scalar equation doesn’t work here. So, we begin by following ansatz
u ⃗ ( t , x ) = exp ( i λ ϕ ) ( z 0 ( t , x ) K ⃗ + Z 1 , K ⃗ ( t , x ) λ + Z 2 , K ⃗ ( t , x ) λ 2 + ⋯ + Z N , K ⃗ ( t , x ) λ N ) + R ⃗ K ⃗ ( t , x , λ ) , (11) \vec{u}(t,x)=\exp(i\lambda\phi)\left(z_0(t,x)\vec{K}+\frac{Z_{1,\vec{K}}(t,x)}{\lambda}+\frac{Z_{2,\vec{K}}(t,x)}{\lambda^2}+\cdots+\frac{Z_{N,\vec{K}}(t,x)}{\lambda^N}\right)+\vec{R}_{\vec{K}}(t,x,\lambda), \tag*{(11)} u ( t , x ) = exp ( iλ ϕ ) ( z 0 ( t , x ) K + λ Z 1 , K ( t , x ) + λ 2 Z 2 , K ( t , x ) + ⋯ + λ N Z N , K ( t , x ) ) + R K ( t , x , λ ) , (11)
as solution to the IVP (9) , where K ⃗ ∈ R 2 \vec{K}\in\mathbb{R}^2 K ∈ R 2 is constant vector, Z j , K ⃗ Z_{j,\vec{K}} Z j , K , R ⃗ K ⃗ \vec{R}_{\vec{K}} R K are 2 × 1 2\times1 2 × 1 column vectors for j = 1 , … , N j=1,\ldots,N j = 1 , … , N and λ > 1 \lambda>1 λ > 1 is a real parameter. Let us denote
U ⃗ = [ U 1 U 2 ] : = z 0 K ⃗ + Z 1 , K ⃗ λ + Z 2 , K ⃗ λ 2 + ⋯ + Z N , K ⃗ λ N . \vec{U}=\begin{bmatrix}U_1\\U_2\end{bmatrix}:=z_0\vec{K}+\frac{Z_{1,\vec{K}}}{\lambda}+\frac{Z_{2,\vec{K}}}{\lambda^2}+\cdots+\frac{Z_{N,\vec{K}}}{\lambda^N}. U = [ U 1 U 2 ] := z 0 K + λ Z 1 , K + λ 2 Z 2 , K + ⋯ + λ N Z N , K .
Through a direct computation, one can see that operating L P \mathcal{L}_{P} L P on above ansatz yields
0 ⃗ = L P u ⃗ = L P ( exp ( i λ ϕ ) U ⃗ + R ⃗ K ⃗ ) = exp ( i λ ϕ ) [ [ □ U 1 + 2 i λ ∂ t U 1 − 2 i λ θ ⋅ ∇ U 1 + p 11 U 1 + p 12 U 2 □ U 2 + 2 i λ ∂ t U 2 − 2 i λ θ ⋅ ∇ U 2 + p 21 U 1 + p 22 U 2 ] ] + L P R ⃗ K ⃗ . \begin{aligned}
\vec{0}=\mathcal{L}_{P}\vec{u}=\mathcal{L}_{P}\left(\exp(i\lambda\phi)\vec{U}+\vec{R}_{\vec{K}}\right) \\
&=\exp(i\lambda\phi)\left[
\begin{bmatrix}
\Box U_1+2i\lambda\partial_tU_1-2i\lambda\theta\cdot\nabla U_1+p_{11}U_1+p_{12}U_2\\
\Box U_2+2i\lambda\partial_tU_2-2i\lambda\theta\cdot\nabla U_2+p_{21}U_1+p_{22}U_2
\end{bmatrix}
\right]+\mathcal{L}_{P}\vec{R}_{\vec{K}}.
\end{aligned} 0 = L P u = L P ( exp ( iλ ϕ ) U + R K ) = exp ( iλ ϕ ) [ [ □ U 1 + 2 iλ ∂ t U 1 − 2 iλ θ ⋅ ∇ U 1 + p 11 U 1 + p 12 U 2 □ U 2 + 2 iλ ∂ t U 2 − 2 iλ θ ⋅ ∇ U 2 + p 21 U 1 + p 22 U 2 ] ] + L P R K .
We now choose z 0 K ⃗ : = Z 0 z_0\vec{K}:=Z_0 z 0 K := Z 0 and Z j , K ⃗ Z_{j,\vec{K}} Z j , K be the solution to the transport equations
{ T z 0 ( t , x ) : = ( ∂ t − θ ⋅ ∇ ) z 0 ( t , x ) = 0 , ( t , x ) ∈ Ω T , z 0 ( 0 , x ) = ∂ t z 0 ( 0 , x ) = 0 x ∈ Ω , \begin{cases}
\mathcal{T}z_0(t,x):=(\partial_t-\theta\cdot\nabla)z_0(t,x)=0, & (t,x)\in\Omega_T,\\
z_0(0,x)=\partial_tz_0(0,x)=0 & x\in\Omega,
\end{cases} { T z 0 ( t , x ) := ( ∂ t − θ ⋅ ∇ ) z 0 ( t , x ) = 0 , z 0 ( 0 , x ) = ∂ t z 0 ( 0 , x ) = 0 ( t , x ) ∈ Ω T , x ∈ Ω ,
and
{ T Z j , K ⃗ ( t , x ) = − 1 2 i L P Z j − 1 , K ⃗ ( t , x ) , ( t , x ) ∈ Ω T , Z j , K ⃗ ( 0 , x ) = ∂ t Z j , K ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{T}Z_{j,\vec{K}}(t,x)=-\frac{1}{2i}\mathcal{L}_{P}Z_{j-1,\vec{K}}(t,x), & (t,x)\in\Omega_T,\\
Z_{j,\vec{K}}(0,x)=\partial_tZ_{j,\vec{K}}(0,x)=\vec{0}, & x\in\Omega,
\end{cases} { T Z j , K ( t , x ) = − 2 i 1 L P Z j − 1 , K ( t , x ) , Z j , K ( 0 , x ) = ∂ t Z j , K ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω ,
for each j = 1 , … , N j=1,\ldots,N j = 1 , … , N . Note that the solution of the above transport equations exist and are given by
z 0 ( t , x ) : = φ ( x + t θ ) , z_0(t,x):=\varphi(x+t\theta), z 0 ( t , x ) := φ ( x + tθ ) ,
and
Z j , K ⃗ ( t , x ) = − 1 2 i ∫ 0 t L P Z j − 1 , K ⃗ ( s , x + ( t − s ) θ ) d s . Z_{j,\vec{K}}(t,x)=-\frac{1}{2i}\int_0^t\mathcal{L}_{P}Z_{j-1,\vec{K}}(s,x+(t-s)\theta)\,ds. Z j , K ( t , x ) = − 2 i 1 ∫ 0 t L P Z j − 1 , K ( s , x + ( t − s ) θ ) d s .
respectively, where φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) with supp φ ∩ Ω = ∅ \operatorname{supp}\varphi\cap\Omega= \varnothing supp φ ∩ Ω = ∅ , and θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 . From the above transport equations, we observe that all λ − k \lambda^{-k} λ − k terms vanish and R ⃗ K ⃗ ( t , x , λ ) \vec{R}_{\vec{K}}(t,x,\lambda) R K ( t , x , λ ) solves
L P R ⃗ K ⃗ ( t , x ) = − λ − N exp ( i λ ϕ ) L P Z N , K ⃗ ( t , x ) . \mathcal{L}_{\mathbb{P}}\vec{R}_{\vec{K}}(t,x)=-\lambda^{-N}\exp(i\lambda\phi)\mathcal{L}_{\mathbb{P}}Z_{N,\vec{K}}(t,x). L P R K ( t , x ) = − λ − N exp ( iλ ϕ ) L P Z N , K ( t , x ) .
Since we require that u ⃗ \vec{u} u defined by (11) solves the IVP (9) , the problem reduces to establishing the existence of R ⃗ K ⃗ \vec{R}_{\vec{K}} R K satisfying
{ L P R ⃗ K ⃗ ( t , x ) = − λ − N exp ( i λ ϕ ) L P Z N , K ⃗ ( t , x ) , ( t , x ) ∈ Ω T , R ⃗ K ⃗ ( 0 , x ) = ∂ t R ⃗ K ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{R}_{\vec{K}}(t,x)=-\lambda^{-N}\exp(i\lambda\phi)\mathcal{L}_{\mathbb{P}}Z_{N,\vec{K}}(t,x), & (t,x)\in\Omega_T,\\
\vec{R}_{\vec{K}}(0,x)=\partial_t\vec{R}_{\vec{K}}(0,x)=\vec{0}, & x\in\Omega.
\end{cases} { L P R K ( t , x ) = − λ − N exp ( iλ ϕ ) L P Z N , K ( t , x ) , R K ( 0 , x ) = ∂ t R K ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω.
Using the well-posedness result given by Proposition A.3 and Sobolev embedding, we have R ⃗ K ⃗ ∈ C ∞ ( Ω T ) \vec{R}_{\vec{K}}\in C^\infty(\Omega_T) R K ∈ C ∞ ( Ω T ) satisfying
∥ R ⃗ K ⃗ ∥ H m ( Ω T ) ≤ C λ m − N . \|\vec{R}_{\vec{K}}\|_{H^m(\Omega_T)}\le C\lambda^{m-N}. ∥ R K ∥ H m ( Ω T ) ≤ C λ m − N .
Again, we use the Sobolev embedding theorem (m > ℓ + 1 m>\ell+1 m > ℓ + 1 ) to conclude that
∥ R ⃗ K ⃗ ∥ L ∞ ( Ω T ) ≤ C ∥ R ⃗ K ⃗ ∥ H m ( Ω T ) ≤ C λ m − N . \|\vec{R}_{\vec{K}}\|_{L^\infty(\Omega_T)}\le C\|\vec{R}_{\vec{K}}\|_{H^m(\Omega_T)}\le C\lambda^{m-N}. ∥ R K ∥ L ∞ ( Ω T ) ≤ C ∥ R K ∥ H m ( Ω T ) ≤ C λ m − N .
For N ≥ m + 1 N\ge m+1 N ≥ m + 1 , we have
∥ R ⃗ K ⃗ ∥ L ∞ ( Ω T ) ≤ C λ − 1 , \|\vec{R}_{\vec{K}}\|_{L^\infty(\Omega_T)}\le C\lambda^{-1}, ∥ R K ∥ L ∞ ( Ω T ) ≤ C λ − 1 ,
for any λ > 1 \lambda>1 λ > 1 . On the basis of the preceding analysis, we have established the following proposition.
Proposition 2.3 (Asymptotic solutions for forward problem). Let P ∈ C c ∞ ( Ω ) \mathbb{P}\in C^\infty_c(\Omega) P ∈ C c ∞ ( Ω ) and T > 0 T>0 T > 0 be fixed. Then, the IVP (9) has an asymptotic solution of the form
u ⃗ ( t , x ) = exp ( i λ ϕ ) ( z 0 ( t , x ) K ⃗ + Z 1 , K ⃗ ( t , x ) λ + Z 2 , K ⃗ ( t , x ) λ 2 + ⋯ + Z N , K ⃗ ( t , x ) λ N ) + R ⃗ K ⃗ ( t , x , λ ) . \vec{u}(t,x)=\exp(i\lambda\phi)\left(z_0(t,x)\vec{K}+\frac{Z_{1,\vec{K}}(t,x)}{\lambda}+\frac{Z_{2,\vec{K}}(t,x)}{\lambda^2}+\cdots+\frac{Z_{N,\vec{K}}(t,x)}{\lambda^N}\right)+\vec{R}_{\vec{K}}(t,x,\lambda). u ( t , x ) = exp ( iλ ϕ ) ( z 0 ( t , x ) K + λ Z 1 , K ( t , x ) + λ 2 Z 2 , K ( t , x ) + ⋯ + λ N Z N , K ( t , x ) ) + R K ( t , x , λ ) . where K ⃗ ∈ R 2 \vec{K}\in\mathbb{R}^2 K ∈ R 2 is constant vector, z 0 z_0 z 0 solves the transport equation
{ T z 0 ( t , x ) : = ( ∂ t − θ ⋅ ∇ ) z 0 ( t , x ) = 0 , ( t , x ) ∈ Ω T , z 0 ( 0 , x ) = ∂ t z 0 ( 0 , x ) = 0 , x ∈ Ω , (12) \begin{cases}
\mathcal{T}z_0(t,x):=(\partial_t-\theta\cdot\nabla)z_0(t,x)=0, & (t,x)\in\Omega_T,\\
z_0(0,x)=\partial_tz_0(0,x)=0, & x\in\Omega, \tag*{(12)}
\end{cases} { T z 0 ( t , x ) := ( ∂ t − θ ⋅ ∇ ) z 0 ( t , x ) = 0 , z 0 ( 0 , x ) = ∂ t z 0 ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , (12) and the finite sequence of vectors Z j , K ⃗ Z_{j,\vec{K}} Z j , K solves the system of transport equations:
{ T Z j , K ⃗ ( t , x ) = − 1 2 i L P Z j − 1 , K ⃗ ( t , x ) , ( t , x ) ∈ Ω T , Z j , K ⃗ ( 0 , x ) = ∂ t Z j , K ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{T}Z_{j,\vec{K}}(t,x)=-\frac{1}{2i}\mathcal{L}_{\mathbb{P}}Z_{j-1,\vec{K}}(t,x), & (t,x)\in\Omega_T,\\
Z_{j,\vec{K}}(0,x)=\partial_tZ_{j,\vec{K}}(0,x)=\vec{0}, & x\in\Omega,
\end{cases} { T Z j , K ( t , x ) = − 2 i 1 L P Z j − 1 , K ( t , x ) , Z j , K ( 0 , x ) = ∂ t Z j , K ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , for j = 1 , … , N j=1,\ldots,N j = 1 , … , N and Z 0 : = z 0 K ⃗ Z_0:=z_0\vec{K} Z 0 := z 0 K . Also, the correction vector R ⃗ K ⃗ ( t , x , λ ) \vec{R}_{\vec{K}}(t,x,\lambda) R K ( t , x , λ ) solves
{ L P R ⃗ K ⃗ ( t , x ) = − λ − N exp ( i λ ϕ ) L P Z N , K ⃗ ( t , x ) , ( t , x ) ∈ Ω T , R ⃗ K ⃗ ( 0 , x ) = ∂ t R ⃗ K ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{R}_{\vec{K}}(t,x)=-\lambda^{-N}\exp(i\lambda\phi)\mathcal{L}_{\mathbb{P}}Z_{N,\vec{K}}(t,x), & (t,x)\in\Omega_T,\\
\vec{R}_{\vec{K}}(0,x)=\partial_t\vec{R}_{\vec{K}}(0,x)=\vec{0}, & x\in\Omega,
\end{cases} { L P R K ( t , x ) = − λ − N exp ( iλ ϕ ) L P Z N , K ( t , x ) , R K ( 0 , x ) = ∂ t R K ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , and satisfies
∥ R ⃗ K ⃗ ∥ L ∞ ( Ω T ) ≤ C λ − 1 , for any λ > 1 , \|\vec{R}_{\vec{K}}\|_{L^\infty(\Omega_T)}\le C\lambda^{-1},\quad\text{for any }\lambda>1, ∥ R K ∥ L ∞ ( Ω T ) ≤ C λ − 1 , for any λ > 1 , where C > 0 C>0 C > 0 is independent of λ \lambda λ and N ≥ m + 1 N\ge m+1 N ≥ m + 1 .
Using a similar approach, one can derive the asymptotic solutions for the backward problem, as stated in the following proposition.
Proposition 2.4 (Asymptotic solutions for backward problem). Let P ∈ C c ∞ ( Ω ) \mathrm{P} \in C^\infty_c(\Omega) P ∈ C c ∞ ( Ω ) and T > 0 T>0 T > 0 be fixed. Then, the backward problem (10) has an asymptotic solution of the form
u ⃗ ( t , x ) = exp ( − 3 i λ ϕ ) ( w 0 K ⃗ ( − ) + W 1 , K ⃗ ( − ) λ + W 2 , K ⃗ ( − ) λ 2 + ⋯ + W N , K ⃗ ( − ) λ N ) + R ⃗ K ⃗ ( − ) ( t , x , λ ) . \vec{u}(t,x)=\exp(-3i\lambda\phi)\left(w_0\vec{K}^{(-)}+\frac{W_{1,\vec{K}^{(-)}}}{\lambda}+\frac{W_{2,\vec{K}^{(-)}}}{\lambda^2}+\cdots+\frac{W_{N,\vec{K}^{(-)}}}{\lambda^N}\right)+\vec{R}_{\vec{K}^{(-)}}(t,x,\lambda). u ( t , x ) = exp ( − 3 iλ ϕ ) ( w 0 K ( − ) + λ W 1 , K ( − ) + λ 2 W 2 , K ( − ) + ⋯ + λ N W N , K ( − ) ) + R K ( − ) ( t , x , λ ) . where K ⃗ ( − ) ∈ R 2 \vec{K}^{(-)}\in\mathbb{R}^2 K ( − ) ∈ R 2 is constant vector and W j , K ⃗ ( − ) W_{j,\vec{K}^{(-)}} W j , K ( − ) , R ⃗ K ⃗ ( − ) \vec{R}_{\vec{K}^{(-)}} R K ( − ) are 2 × 1 2\times1 2 × 1 column vectors for j = 1 , … , N j=1,\ldots,N j = 1 , … , N . Here, w 0 w_0 w 0 solves the transport equation
{ T w 0 ( t , x ) : = ( ∂ t − θ ⋅ ∇ ) w 0 ( t , x ) = 0 , ( t , x ) ∈ Ω T , w 0 ( T , x ) = ∂ t w 0 ( T , x ) = 0 , x ∈ Ω , (13) \begin{cases}
\mathcal{T}w_0(t,x):=(\partial_t-\theta\cdot\nabla)w_0(t,x)=0, & (t,x)\in\Omega_T,\\
w_0(T,x)=\partial_t w_0(T,x)=0, & x\in\Omega, \tag*{(13)}
\end{cases} { T w 0 ( t , x ) := ( ∂ t − θ ⋅ ∇ ) w 0 ( t , x ) = 0 , w 0 ( T , x ) = ∂ t w 0 ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , (13) and the finite sequence of vectors W j , K ⃗ ( − ) W_{j,\vec{K}^{(-)}} W j , K ( − ) solves the system of transport equations:
{ T W j , K ⃗ ( − ) ( t , x ) = 1 6 i L P ⃗ T W j − 1 , K ⃗ ( − ) ( t , x ) , ( t , x ) ∈ Ω T , W j , K ⃗ ( − ) ( T , x ) = ∂ t W j , K ⃗ ( − ) ( T , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{T}W_{j,\vec{K}^{(-)}}(t,x)=\frac{1}{6i}\mathcal{L}_{\vec{P}^{\mathsf{T}}}W_{j-1,\vec{K}^{(-)}}(t,x), & (t,x)\in\Omega_T,\\
W_{j,\vec{K}^{(-)}}(T,x)=\partial_tW_{j,\vec{K}^{(-)}}(T,x)=\vec{0}, & x\in\Omega,
\end{cases} { T W j , K ( − ) ( t , x ) = 6 i 1 L P T W j − 1 , K ( − ) ( t , x ) , W j , K ( − ) ( T , x ) = ∂ t W j , K ( − ) ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , for j = 1 , … , N j=1,\ldots,N j = 1 , … , N and W 0 : = w 0 K ⃗ ( − ) W_0:=w_0\vec{K}^{(-)} W 0 := w 0 K ( − ) . Also, the correction vector R ⃗ K ⃗ ( − ) ( t , x , λ ) \vec{R}_{\vec{K}^{(-)}}(t,x,\lambda) R K ( − ) ( t , x , λ ) solves
{ L P ⃗ T R ⃗ K ⃗ ( − ) ( t , x ) = − λ − N exp ( − 3 i λ ϕ ) L P ⃗ T W N , K ⃗ ( − ) ( t , x ) , ( t , x ) ∈ Ω T , R ⃗ K ⃗ ( − ) ( T , x ) = ∂ t R ⃗ K ⃗ ( − ) ( T , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{L}_{\vec{P}^{\mathsf{T}}}\vec{R}_{\vec{K}^{(-)}}(t,x)=-\lambda^{-N}\exp(-3i\lambda\phi)\mathcal{L}_{\vec{P}^{\mathsf{T}}}W_{N,\vec{K}^{(-)}}(t,x), & (t,x)\in\Omega_T,\\
\vec{R}_{\vec{K}^{(-)}}(T,x)=\partial_t\vec{R}_{\vec{K}^{(-)}}(T,x)=\vec{0}, & x\in\Omega,
\end{cases} { L P T R K ( − ) ( t , x ) = − λ − N exp ( − 3 iλ ϕ ) L P T W N , K ( − ) ( t , x ) , R K ( − ) ( T , x ) = ∂ t R K ( − ) ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω , and satisfies
∥ R ⃗ K ⃗ ( − ) ∥ L ∞ ( Ω T ) ≤ C λ − 1 , for any λ > 1 , \left\|\vec{R}_{\vec{K}^{(-)}}\right\|_{L^\infty(\Omega_T)}\le C\lambda^{-1},\quad\text{for any }\lambda>1, R K ( − ) L ∞ ( Ω T ) ≤ C λ − 1 , for any λ > 1 , where C > 0 C>0 C > 0 is independent of λ \lambda λ and $N\ge m+1.
Proof of Theorem 1.1
This section is devoted to proving the unique recovery of the coefficients (both the zeroth-order potential and the nonlinear potential) from the full DN map. We divide this section into two subsections, where §3.1 and §3.2 address the unique identification of the zeroth-order and nonlinear coefficients, respectively.
Unique determination of matrix-valued potential
In the following, we start by linearizing the problem using a first-order linearization method. This technique is commonly practiced in the study of nonlinear equations; see, for instance, [6 , 11 , 29 , 34 , 41 ] . Given f 11 , f 12 , f 13 , f 21 , f 22 , f 23 ∈ H m + 1 ( ∂ Ω T ) f_{11},f_{12},f_{13},f_{21},f_{22},f_{23}\in H^{m+1}(\partial\Omega_T) f 11 , f 12 , f 13 , f 21 , f 22 , f 23 ∈ H m + 1 ( ∂ Ω T ) , we choose ε : = ( ε 1 , ε 2 , ε 3 ) \varepsilon:=(\varepsilon_1,\varepsilon_2,\varepsilon_3) ε := ( ε 1 , ε 2 , ε 3 ) , where ε i ≥ 0 \varepsilon_i\ge0 ε i ≥ 0 for all i = 1 , 2 , 3 i=1,2,3 i = 1 , 2 , 3 , so that
ε f ⃗ : = [ ε 1 f 11 + ε 2 f 12 + ε 3 f 13 ε 1 f 21 + ε 2 f 22 + ε 3 f 23 ] ∈ F m + 1 σ , \varepsilon\vec{f}:=\begin{bmatrix}
\varepsilon_1f_{11}+\varepsilon_2f_{12}+\varepsilon_3f_{13}\\
\varepsilon_1f_{21}+\varepsilon_2f_{22}+\varepsilon_3f_{23}
\end{bmatrix}\in F^{\sigma}_{m+1}, ε f := [ ε 1 f 11 + ε 2 f 12 + ε 3 f 13 ε 1 f 21 + ε 2 f 22 + ε 3 f 23 ] ∈ F m + 1 σ ,
for σ > 0 \sigma>0 σ > 0 sufficiently small. Let u ⃗ ε f ⃗ : = u ⃗ \vec{u}_{\varepsilon\vec{f}}:=\vec{u} u ε f := u be the solution to the IBVP
{ L P , Q u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = ε f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (14) \begin{cases}
\mathcal{L}_{\mathrm{P},\mathrm{Q}}\vec{u}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{u}(t,x)=\varepsilon\vec{f}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)=\vec{0},\quad\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega. \tag*{(14)}
\end{cases} ⎩ ⎨ ⎧ L P , Q u ( t , x ) = 0 , u ( t , x ) = ε f ( t , x ) , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (14)
The corresponding DN map is given by
Λ P , Q ( ε f ⃗ ) : = ∂ ν u ⃗ ∣ ∂ Ω T . (15) \Lambda_{\mathbb{P},Q}(\varepsilon\vec{f}) := \left.\partial_{\nu}\vec{u}\right|_{\partial\Omega_T}. \tag*{(15)} Λ P , Q ( ε f ) := ∂ ν u ∣ ∂ Ω T . (15)
For i ∈ { 1 , 2 , 3 } i \in\{1,2,3\} i ∈ { 1 , 2 , 3 } , we apply ∂ ε i \partial_{\varepsilon_i} ∂ ε i to (14) and evaluate at ε = 0 \varepsilon=0 ε = 0 to get
{ L P v ⃗ ( i ) ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , v ⃗ ( i ) ( t , x ) = ( f 1 i , f 2 i ) T ( t , x ) , ( t , x ) ∈ ∂ Ω T , v ⃗ ( i ) ( 0 , x ) = ∂ t v ⃗ ( i ) ( 0 , x ) = 0 ⃗ , x ∈ Ω , (16) \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{v}^{(i)}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{v}^{(i)}(t,x)=(f_{1i},f_{2i})^{\mathsf{T}}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{v}^{(i)}(0,x)=\partial_t\vec{v}^{(i)}(0,x)=\vec{0}, & x\in\Omega, \tag*{(16)}
\end{cases} ⎩ ⎨ ⎧ L P v ( i ) ( t , x ) = 0 , v ( i ) ( t , x ) = ( f 1 i , f 2 i ) T ( t , x ) , v ( i ) ( 0 , x ) = ∂ t v ( i ) ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω , (16)
where ∂ ε i u ⃗ ∣ ε = 0 : = v ⃗ ( i ) = ( v 1 ( i ) , v 2 ( i ) ) T \left.\partial_{\varepsilon_i}\vec{u}\right|_{\varepsilon=0}:=\vec{v}^{(i)}=(v^{(i)}_1,v^{(i)}_2)^{\mathsf{T}} ∂ ε i u ∣ ε = 0 := v ( i ) = ( v 1 ( i ) , v 2 ( i ) ) T . Also, the first-order linearization of the DN map (15) becomes
1 Λ P ( f 1 i , f 2 i ) ∣ ∂ Ω T : = ( ∂ ε i Λ P , Q ( ε f ⃗ ) ∣ ε = 0 ) ∣ ∂ Ω T = ∂ ν v ⃗ ( i ) ∣ ∂ Ω T . (17) {}^{1}\Lambda_{\mathbb{P}}(f_{1i},f_{2i})\big|_{\partial\Omega_T}:=\left.\left(\partial_{\varepsilon_i}\Lambda_{\mathbb{P},Q}(\varepsilon\vec{f})\big|_{\varepsilon=0}\right)\right|_{\partial\Omega_T}=\left.\partial_{\nu}\vec{v}^{(i)}\right|_{\partial\Omega_T}. \tag*{(17)} 1 Λ P ( f 1 i , f 2 i ) ∂ Ω T := ( ∂ ε i Λ P , Q ( ε f ) ε = 0 ) ∂ Ω T = ∂ ν v ( i ) ∂ Ω T . (17)
To avoid the cumbersomeness, we denote v ⃗ ( i ) : = v ⃗ \vec{v}^{(i)}:=\vec{v} v ( i ) := v , f 1 i : = f f_{1i}:=f f 1 i := f , and f 2 i : = g f_{2i}:=g f 2 i := g throughout the establishment of unique recovery of zeroth-order potential. As a result, (16) and (17) transformed into
{ L P v ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , v ⃗ ( t , x ) = ( f , g ) T ( t , x ) , ( t , x ) ∈ ∂ Ω T , v ⃗ ( 0 , x ) = ∂ t v ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , (18) \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{v}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{v}(t,x)=(f,g)^{\mathsf{T}}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{v}(0,x)=\partial_t\vec{v}(0,x)=\vec{0}, & x\in\Omega, \tag*{(18)}
\end{cases} ⎩ ⎨ ⎧ L P v ( t , x ) = 0 , v ( t , x ) = ( f , g ) T ( t , x ) , v ( 0 , x ) = ∂ t v ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω , (18)
and
1 Λ P ( f , g ) ∣ ∂ Ω T = ∂ ν v ⃗ ∣ ∂ Ω T , f , g ∈ H m + 1 ( ∂ Ω T ) , (19) {}^{1}\Lambda_{\mathbb{P}}(f,g)\big|_{\partial\Omega_T}=\left.\partial_{\nu}\vec{v}\right|_{\partial\Omega_T},\qquad f,g\in H^{m+1}(\partial\Omega_T), \tag*{(19)} 1 Λ P ( f , g ) ∂ Ω T = ∂ ν v ∣ ∂ Ω T , f , g ∈ H m + 1 ( ∂ Ω T ) , (19)
respectively. Next, we assert that the matrix P \mathbb{P} P can be uniquely determined from the boundary measurement (19) . To begin, we denote v ⃗ j \vec{v}^{j} v j be the solution to the IBVP
{ L P j v ⃗ j ( t , x ) = 0 ⃗ , ( t , x ) ∈ Ω T , v ⃗ j ( t , x ) = ( f , g ) T ( t , x ) , ( t , x ) ∈ ∂ Ω T , v ⃗ j ( 0 , x ) = ∂ t v ⃗ j ( 0 , x ) = 0 ⃗ , x ∈ Ω . (20) \begin{cases}
\mathcal{L}_{\mathbb{P}^{j}}\vec{v}^{j}(t,x)=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{v}^{j}(t,x)=(f,g)^{\mathsf{T}}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{v}^{j}(0,x)=\partial_t\vec{v}^{j}(0,x)=\vec{0}, & x\in\Omega. \tag*{(20)}
\end{cases} ⎩ ⎨ ⎧ L P j v j ( t , x ) = 0 , v j ( t , x ) = ( f , g ) T ( t , x ) , v j ( 0 , x ) = ∂ t v j ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (20)
We denote v ⃗ 12 : = v ⃗ 1 − v ⃗ 2 \vec{v}^{12}:=\vec{v}^{1}-\vec{v}^{2} v 12 := v 1 − v 2 , and P 12 : = P 1 − P 2 \mathbb{P}^{12}:=\mathbb{P}^{1}-\mathbb{P}^{2} P 12 := P 1 − P 2 , where for each j = 1 , 2 j=1,2 j = 1 , 2 , v ⃗ j \vec{v}^{j} v j is solution to (20) and
P j : = [ p 11 j p 12 j p 21 j p 22 j ] . \mathbb{P}^{j}:=
\begin{bmatrix}
p^{j}_{11} & p^{j}_{12}\\
p^{j}_{21} & p^{j}_{22}
\end{bmatrix}. P j := [ p 11 j p 21 j p 12 j p 22 j ] .
Thus, we have
{ □ v ⃗ 12 + P 1 v ⃗ 12 = − P 12 v ⃗ 2 , ( t , x ) ∈ Ω T , v ⃗ 12 ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , v ⃗ 12 ( 0 , x ) = ∂ t v ⃗ 12 ( 0 , x ) = 0 ⃗ , x ∈ Ω . (21) \begin{cases}
\Box\vec{v}^{12}+\mathbb{P}^{1}\vec{v}^{12}=-\mathbb{P}^{12}\vec{v}^{2}, & (t,x)\in\Omega_T,\\
\vec{v}^{12}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\vec{v}^{12}(0,x)=\partial_t\vec{v}^{12}(0,x)=\vec{0}, & x\in\Omega. \tag*{(21)}
\end{cases} ⎩ ⎨ ⎧ □ v 12 + P 1 v 12 = − P 12 v 2 , v 12 ( t , x ) = 0 , v 12 ( 0 , x ) = ∂ t v 12 ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (21)
Since Λ P 1 , Q 1 ( ε f ⃗ ) = Λ P 2 , Q 2 ( ε f ⃗ ) \Lambda_{\mathbb{P}^{1},Q_{1}}(\varepsilon\vec{f})=\Lambda_{\mathbb{P}^{2},Q_{2}}(\varepsilon\vec{f}) Λ P 1 , Q 1 ( ε f ) = Λ P 2 , Q 2 ( ε f ) , the first order linearization then leads us to
1 Λ P 1 ( f , g ) ∣ ∂ Ω T = 1 Λ P 2 ( f , g ) ∣ ∂ Ω T , f , g ∈ H m + 1 ( ∂ Ω T ) . {}^{1}\Lambda_{\mathbb{P}^{1}}(f,g)\big|_{\partial\Omega_T}={}^{1}\Lambda_{\mathbb{P}^{2}}(f,g)\big|_{\partial\Omega_T},\qquad f,g\in H^{m+1}(\partial\Omega_T). 1 Λ P 1 ( f , g ) ∂ Ω T = 1 Λ P 2 ( f , g ) ∂ Ω T , f , g ∈ H m + 1 ( ∂ Ω T ) .
Consequently, we have ∂ ν v ⃗ 12 ∣ ∂ Ω T = 0 ⃗ \left.\partial_{\nu}\vec{v}^{12}\right|_{\partial\Omega_T}=\vec{0} ∂ ν v 12 ∂ Ω T = 0 . Next, we consider v ⃗ 0 : = ( v 01 , v 02 ) T \vec{v}_{0}:=(v_{01},v_{02})^{\mathsf{T}} v 0 := ( v 01 , v 02 ) T to be the solution to the backward problem
{ □ v ⃗ 0 + ( P 1 ) T v ⃗ 0 = 0 ⃗ , ( t , x ) ∈ Ω T , v ⃗ 0 ( T , x ) = ∂ t v ⃗ 0 ( T , x ) = 0 ⃗ , x ∈ Ω . (22) \begin{cases}
\Box\vec{v}_{0}+(\mathbb{P}^{1})^{\mathsf{T}}\vec{v}_{0}=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{v}_{0}(T,x)=\partial_t\vec{v}_{0}(T,x)=\vec{0}, & x\in\Omega. \tag*{(22)}
\end{cases} { □ v 0 + ( P 1 ) T v 0 = 0 , v 0 ( T , x ) = ∂ t v 0 ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω. (22)
To derive the desired integral identity, we then multiply (18) by v ⃗ 0 \vec{v}_0 v 0 and integrate over Ω T \Omega_T Ω T , which leads us to
∫ Ω T ( □ v ⃗ 12 + P 1 v ⃗ 12 ) ⋅ v ⃗ 0 d x d t = − ∫ Ω T P 12 v ⃗ 2 ⋅ v ⃗ 0 d x d t . \int_{\Omega_T}\left(\Box\vec{v}^{12}+\mathbb{P}^{1}\vec{v}^{12}\right)\cdot\vec{v}_0\,dxdt=-\int_{\Omega_T}\mathbb{P}^{12}\vec{v}^{2}\cdot\vec{v}_0\,dxdt. ∫ Ω T ( □ v 12 + P 1 v 12 ) ⋅ v 0 d x d t = − ∫ Ω T P 12 v 2 ⋅ v 0 d x d t .
Armed with the initial and final time information available in (21) and (22) respectively, the integration by parts formula yields
∫ Ω T P 12 v ⃗ 2 ⋅ v ⃗ 0 d x d t = 0. (23) \int_{\Omega_T}\mathbb{P}^{12}\vec{v}^{2}\cdot\vec{v}_0\,dxdt=0. \tag*{(23)} ∫ Ω T P 12 v 2 ⋅ v 0 d x d t = 0. (23)
We are now in position to utilize the GO solutions established in § 3.1 to conclude about the unique identification of the coefficients. Although to work in a common class of functions φ \varphi φ satisfying the transport equation (4) with supp ( φ ) ∩ Ω = ∅ \operatorname{supp}(\varphi)\cap\Omega=\varnothing supp ( φ ) ∩ Ω = ∅ and ( supp ( φ ) ± T θ ) ∩ Ω = ∅ (\operatorname{supp}(\varphi)\pm T\theta)\cap\Omega=\varnothing ( supp ( φ ) ± T θ ) ∩ Ω = ∅ , we introduce a function space C c ∞ ( R ℓ ) C_c^\infty(\mathbb{R}^{\ell}) C c ∞ ( R ℓ ) , a subspace of C ∞ ( R ℓ ) C^\infty(\mathbb{R}^{\ell}) C ∞ ( R ℓ ) , and it is given by
C c ∞ ( R ℓ ) : = { φ ∈ C c ∞ ( R ℓ ) : supp ( φ ) ∩ Ω = ∅ and ( supp ( φ ) ± T θ ) ∩ Ω = ∅ , ∀ θ ∈ S ℓ − 1 } . \mathcal{C}_c^\infty(\mathbb{R}^{\ell}):=\{\varphi\in C_c^\infty(\mathbb{R}^{\ell}):\operatorname{supp}(\varphi)\cap\Omega=\varnothing\ \text{and}\ (\operatorname{supp}(\varphi)\pm T\theta)\cap\Omega=\varnothing,\ \forall\ \theta\in S^{\ell-1}\}. C c ∞ ( R ℓ ) := { φ ∈ C c ∞ ( R ℓ ) : supp ( φ ) ∩ Ω = ∅ and ( supp ( φ ) ± T θ ) ∩ Ω = ∅ , ∀ θ ∈ S ℓ − 1 } .
For K ⃗ ± : = ( K 1 ± , K 2 ± ) T ∈ R 2 \vec{K}^{\pm}:=(K_1^{\pm},K_2^{\pm})^{\mathsf T}\in\mathbb{R}^{2} K ± := ( K 1 ± , K 2 ± ) T ∈ R 2 , φ ∈ C c ∞ ( R ℓ ) \varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) , choose (see Lemma 2.1 and Lemma 2.2)
v ⃗ 2 ( t , x ) = φ ( x + t θ ) exp ( i λ ϕ ) K ⃗ ( + ) + R ⃗ K ⃗ ( + ) v 2 ( t , x , λ ) , (24) \vec{v}^{2}(t,x)=\varphi(x+t\theta)\exp(i\lambda\phi)\vec{K}^{(+)}+\vec{R}_{\vec{K}^{(+)}}^{v^{2}}(t,x,\lambda), \tag*{(24)} v 2 ( t , x ) = φ ( x + tθ ) exp ( iλ ϕ ) K ( + ) + R K ( + ) v 2 ( t , x , λ ) , (24)
and
v ⃗ 0 ( t , x ) = φ ( x + t θ ) exp ( − i λ ϕ ) K ⃗ ( − ) + R ⃗ K ⃗ ( − ) v 0 ( t , x , λ ) , (25) \vec{v}_0(t,x)=\varphi(x+t\theta)\exp(-i\lambda\phi)\vec{K}^{(-)}+\vec{R}_{\vec{K}^{(-)}}^{v_0}(t,x,\lambda), \tag*{(25)} v 0 ( t , x ) = φ ( x + tθ ) exp ( − iλ ϕ ) K ( − ) + R K ( − ) v 0 ( t , x , λ ) , (25)
the solutions to IVP (5) and backward problem (7) . Now we substitute (24) and (25) in (23) to arrive at
∫ Ω T φ 2 ( x + t θ ) ( p 11 12 K 1 + K 1 − + p 12 12 K 2 + K 1 − + p 21 12 K 1 + K 2 − + p 22 12 K 2 + K 2 − ) d x d t + ∫ Ω T φ ( x + t θ ) exp ( i λ ϕ ) ( p 11 12 K 1 + R 1 v 0 + p 12 12 K 2 + R 1 v 0 + p 21 12 K 1 + R 2 v 0 + p 22 12 K 2 + R 2 v 0 ) d x d t + ∫ Ω T φ ( x + t θ ) exp ( − i λ ϕ ) ( p 11 12 K 1 − R 1 v 2 + p 12 12 K 2 − R 1 v 2 + p 21 12 K 1 − R 2 v 2 + p 22 12 K 2 − R 2 v 2 ) d x d t + ∫ Ω T ( p 11 12 R 1 v 2 R 1 v 0 + p 12 12 R 2 v 2 R 1 v 0 + p 21 12 R 1 v 2 R 2 v 0 + p 22 12 R 2 v 2 R 2 v 0 ) d x d t = 0 , \begin{aligned}
&\int_{\Omega_T}\varphi^2(x+t\theta)\left(p_{11}^{12}K_1^{+}K_1^{-}+p_{12}^{12}K_2^{+}K_1^{-}+p_{21}^{12}K_1^{+}K_2^{-}+p_{22}^{12}K_2^{+}K_2^{-}\right)\,dxdt\\
&+\int_{\Omega_T}\varphi(x+t\theta)\exp(i\lambda\phi)\left(p_{11}^{12}K_1^{+}R_1^{v_0}+p_{12}^{12}K_2^{+}R_1^{v_0}+p_{21}^{12}K_1^{+}R_2^{v_0}+p_{22}^{12}K_2^{+}R_2^{v_0}\right)\,dxdt\\
&+\int_{\Omega_T}\varphi(x+t\theta)\exp(-i\lambda\phi)\left(p_{11}^{12}K_1^{-}R_1^{v^2}+p_{12}^{12}K_2^{-}R_1^{v^2}+p_{21}^{12}K_1^{-}R_2^{v^2}+p_{22}^{12}K_2^{-}R_2^{v^2}\right)\,dxdt\\
&+\int_{\Omega_T}\left(p_{11}^{12}R_1^{v^2}R_1^{v_0}+p_{12}^{12}R_2^{v^2}R_1^{v_0}+p_{21}^{12}R_1^{v^2}R_2^{v_0}+p_{22}^{12}R_2^{v^2}R_2^{v_0}\right)\,dxdt=0,
\end{aligned} ∫ Ω T φ 2 ( x + tθ ) ( p 11 12 K 1 + K 1 − + p 12 12 K 2 + K 1 − + p 21 12 K 1 + K 2 − + p 22 12 K 2 + K 2 − ) d x d t + ∫ Ω T φ ( x + tθ ) exp ( iλ ϕ ) ( p 11 12 K 1 + R 1 v 0 + p 12 12 K 2 + R 1 v 0 + p 21 12 K 1 + R 2 v 0 + p 22 12 K 2 + R 2 v 0 ) d x d t + ∫ Ω T φ ( x + tθ ) exp ( − iλ ϕ ) ( p 11 12 K 1 − R 1 v 2 + p 12 12 K 2 − R 1 v 2 + p 21 12 K 1 − R 2 v 2 + p 22 12 K 2 − R 2 v 2 ) d x d t + ∫ Ω T ( p 11 12 R 1 v 2 R 1 v 0 + p 12 12 R 2 v 2 R 1 v 0 + p 21 12 R 1 v 2 R 2 v 0 + p 22 12 R 2 v 2 R 2 v 0 ) d x d t = 0 ,
where R ⃗ K ⃗ ( + ) v 2 = ( R 1 v 2 , R 2 v 2 ) T \vec{R}_{\vec{K}^{(+)}}^{v^2}=(R_1^{v^2},R_2^{v^2})^{\mathsf T} R K ( + ) v 2 = ( R 1 v 2 , R 2 v 2 ) T and R ⃗ K ⃗ ( − ) v 0 = ( R 1 v 0 , R 2 v 0 ) T \vec{R}_{\vec{K}^{(-)}}^{v_0}=(R_1^{v_0},R_2^{v_0})^{\mathsf T} R K ( − ) v 0 = ( R 1 v 0 , R 2 v 0 ) T satisfying
λ ∥ R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ) + ∥ ∂ t R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) \lambda\|\vec{R}_{\vec{K}^{(+)}}^{v^2}\|_{L^2(\Omega_T)}+\|\partial_t\vec{R}_{\vec{K}^{(+)}}^{v^2}\|_{L^2(\Omega_T)}\leq C\|\varphi\|_{H^3(\mathbb{R}^{\ell})} λ ∥ R K ( + ) v 2 ∥ L 2 ( Ω T ) + ∥ ∂ t R K ( + ) v 2 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ )
λ ∥ R ⃗ K ⃗ ( − ) v 0 ∥ L 2 ( Ω T ) + ∥ ∂ t R ⃗ K ⃗ ( − ) v 0 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . (26) \lambda\|\vec{R}_{\vec{K}^{(-)}}^{v_0}\|_{L^2(\Omega_T)}+\|\partial_t\vec{R}_{\vec{K}^{(-)}}^{v_0}\|_{L^2(\Omega_T)}\leq C\|\varphi\|_{H^3(\mathbb{R}^{\ell})}. \tag*{(26)} λ ∥ R K ( − ) v 0 ∥ L 2 ( Ω T ) + ∥ ∂ t R K ( − ) v 0 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . (26)
Taking λ → ∞ \lambda\to\infty λ → ∞ along with (26) , we obtain that
∫ Ω T φ 2 ( x + t θ ) ( p 11 12 K 1 + K 1 − + p 12 12 K 2 + K 1 − + p 21 12 K 1 + K 2 − + p 22 12 K 2 + K 2 − ) d x d t = 0 , (27) \int_{\Omega_T}\varphi^2(x+t\theta)\left(p_{11}^{12}K_1^{+}K_1^{-}+p_{12}^{12}K_2^{+}K_1^{-}+p_{21}^{12}K_1^{+}K_2^{-}+p_{22}^{12}K_2^{+}K_2^{-}\right)\,dxdt=0, \tag*{(27)} ∫ Ω T φ 2 ( x + tθ ) ( p 11 12 K 1 + K 1 − + p 12 12 K 2 + K 1 − + p 21 12 K 1 + K 2 − + p 22 12 K 2 + K 2 − ) d x d t = 0 , (27)
holds for any φ ∈ C c ∞ ( R ℓ ) \varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) , K ⃗ ( + ) \vec{K}^{(+)} K ( + ) , K ⃗ ( − ) ∈ R 2 \vec{K}^{(-)}\in\mathbb{R}^{2} K ( − ) ∈ R 2 and θ ∈ S ℓ − 1 \theta\in S^{\ell-1} θ ∈ S ℓ − 1 . Next to ensure that p i j 12 = 0 p_{ij}^{12}=0 p ij 12 = 0 in Ω \Omega Ω for each 1 ≤ i , j ≤ 2 1\leq i,j\leq2 1 ≤ i , j ≤ 2 , we prescribe the specific choices of constant vectors K ⃗ ( + ) \vec{K}^{(+)} K ( + ) and K ⃗ ( − ) \vec{K}^{(-)} K ( − ) . In particular, using (26) along with K ⃗ ( + ) = K ⃗ ( − ) = ( 1 , 0 ) T \vec{K}^{(+)}=\vec{K}^{(-)}=(1,0)^{\mathsf T} K ( + ) = K ( − ) = ( 1 , 0 ) T , (27) boils down to
∫ Ω T φ 2 ( x + t θ ) p 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . (28) \int_{\Omega_T}\varphi^2(x+t\theta)p_{11}^{12}(x)\,dxdt=0,\quad\forall\ \varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell})\ \text{and}\ \theta\in S^{\ell-1}. \tag*{(28)} ∫ Ω T φ 2 ( x + tθ ) p 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . (28)
We extend p 11 12 p_{11}^{12} p 11 12 by zero to R ℓ ∖ Ω \mathbb{R}^{\ell}\setminus\Omega R ℓ ∖ Ω , and continue to denote the resulting function by p 11 12 p_{11}^{12} p 11 12 . It follows that p 11 12 ∈ C c ∞ ( R ℓ ) p_{11}^{12}\in C^\infty_c(\mathbb{R}^{\ell}) p 11 12 ∈ C c ∞ ( R ℓ ) . Thus, we have
∫ R ℓ ∫ 0 T φ 2 ( x + t θ ) p 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . \int_{\mathbb{R}^{\ell}}\int_0^T \varphi^2(x+t\theta)p_{11}^{12}(x)\,dxdt=0,\quad\forall\varphi\in C^\infty_c(\mathbb{R}^{\ell})\text{ and }\theta\in\mathbb{S}^{\ell-1}. ∫ R ℓ ∫ 0 T φ 2 ( x + tθ ) p 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 .
The change of variables x + t θ : = y x+t\theta:=y x + tθ := y then further leads us to
∫ R ℓ ∫ 0 T φ 2 ( y ) p 11 12 ( y − t θ ) d t d y = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . (29) \int_{\mathbb{R}^{\ell}}\int_0^T \varphi^2(y)p_{11}^{12}(y-t\theta)\,dtdy=0,\quad\forall\varphi\in C^\infty_c(\mathbb{R}^{\ell})\text{ and }\theta\in\mathbb{S}^{\ell-1}. \tag*{(29)} ∫ R ℓ ∫ 0 T φ 2 ( y ) p 11 12 ( y − tθ ) d t d y = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . (29)
The next step is to invert the integral transform above by choosing a suitable φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) . Although this step is standard and has been discussed in [6 , 21 , 29 ] , we present here for the sake of completeness. Let ρ ∈ C c ∞ ( B 1 ( 0 ) ) \rho\in C^\infty_c(\mathbb{B}_1(0)) ρ ∈ C c ∞ ( B 1 ( 0 )) be a cutoff function satisfying 0 ≤ ρ ≤ 1 0\le\rho\le1 0 ≤ ρ ≤ 1 with ∫ R ℓ ρ 2 ( x ) d x = 1 \int_{\mathbb{R}^{\ell}}\rho^2(x)dx=1 ∫ R ℓ ρ 2 ( x ) d x = 1 , where B 1 ( 0 ) \mathbb{B}_1(0) B 1 ( 0 ) is a unit ball centered at 0 and radius 1. Also, let Ω r ⊂ R ℓ \Omega_r\subset\mathbb{R}^{\ell} Ω r ⊂ R ℓ be a open set, given by
Ω r : = { x ∈ R ℓ ∖ Ω ‾ : dist ( x , Ω ) < r } , \Omega_r:=\left\{x\in\mathbb{R}^{\ell}\setminus\overline{\Omega}:\operatorname{dist}(x,\Omega)<r\right\}, Ω r := { x ∈ R ℓ ∖ Ω : dist ( x , Ω ) < r } ,
for r ∈ ( 0 , min { 1 , T − diam ( Ω ) 3 } ) r\in\left(0,\min\left\{1,\frac{T-\operatorname{diam}(\Omega)}{3}\right\}\right) r ∈ ( 0 , min { 1 , 3 T − diam ( Ω ) } ) . Further, for y 0 ∈ Ω r y_0\in\Omega_r y 0 ∈ Ω r , we define
ρ δ ( y ) = 1 δ ℓ / 2 ρ ( y − y 0 δ ) . (30) \rho_\delta(y)=\frac{1}{\delta^{\ell/2}}\rho\left(\frac{y-y_0}{\delta}\right). \tag*{(30)} ρ δ ( y ) = δ ℓ /2 1 ρ ( δ y − y 0 ) . (30)
Then the sequence { ρ δ } δ > 0 ⊂ C c ∞ ( R ℓ ) \{\rho_\delta\}_{\delta>0}\subset C^\infty_c(\mathbb{R}^{\ell}) { ρ δ } δ > 0 ⊂ C c ∞ ( R ℓ ) , whenever 0 < δ < r 0<\delta<r 0 < δ < r . Using (30) in (29) leads to
1 δ ℓ ∫ B δ ( y 0 ) ∫ 0 T p 11 12 ( y − t θ ) ρ 2 ( y − y 0 δ ) d t d y = 0 , for all 0 < δ < r and y 0 ∈ Ω r . \frac{1}{\delta^\ell}\int_{B_\delta(y_0)}\int_0^T p_{11}^{12}(y-t\theta)\rho^2\left(\frac{y-y_0}{\delta}\right)\,dtdy=0,\quad\text{for all }0<\delta<r\text{ and }y_0\in\Omega_r. δ ℓ 1 ∫ B δ ( y 0 ) ∫ 0 T p 11 12 ( y − tθ ) ρ 2 ( δ y − y 0 ) d t d y = 0 , for all 0 < δ < r and y 0 ∈ Ω r .
As δ → ∞ \delta\to\infty δ → ∞ , the aforementioned expression reduces to
∫ 0 T p 11 12 ( y 0 − t θ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 . (31) \int_0^T p_{11}^{12}(y_0-t\theta)\,dt=0,\quad\text{for all }y_0\in\Omega_r\text{ and }\theta\in\mathbb{S}^{\ell-1}. \tag*{(31)} ∫ 0 T p 11 12 ( y 0 − tθ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 . (31)
Further, we can be express the above integral as
∫ 0 ∞ p 11 12 ( y 0 − t θ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 , (32) \int_0^\infty p_{11}^{12}(y_0-t\theta)\,dt=0,\quad\text{for all }y_0\in\Omega_r\text{ and }\theta\in\mathbb{S}^{\ell-1}, \tag*{(32)} ∫ 0 ∞ p 11 12 ( y 0 − tθ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 , (32)
because p 11 12 ( y 0 − t θ ) = 0 p_{11}^{12}(y_0-t\theta)=0 p 11 12 ( y 0 − tθ ) = 0 , for every t ≥ T t\ge T t ≥ T . Indeed, because of T > diam ( Ω ) T>\operatorname{diam}(\Omega) T > diam ( Ω ) , one has
( Ω r ± s θ ) ∩ Ω = ∅ , for all s ≥ T and θ ∈ S ℓ − 1 . (\Omega_r\pm s\theta)\cap\Omega=\varnothing,\quad\text{for all }s\ge T\text{ and }\theta\in\mathbb{S}^{\ell-1}. ( Ω r ± s θ ) ∩ Ω = ∅ , for all s ≥ T and θ ∈ S ℓ − 1 .
From here, we aim to show that the X X X -ray transform of p 11 12 p_{11}^{12} p 11 12 vanishes. To this end, we first observe that (32) holds for all θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 , hence we arrive at
∫ − ∞ ∞ p 11 12 ( y 0 − t θ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 . \int_{-\infty}^{\infty}p_{11}^{12}(y_0-t\theta)\,dt=0,\quad\text{for all }y_0\in\Omega_r\text{ and }\theta\in\mathbb{S}^{\ell-1}. ∫ − ∞ ∞ p 11 12 ( y 0 − tθ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 .
This tells us that the X X X -ray transform is vanishing on every line passing through Ω r \Omega_r Ω r . Following [46 ] , for any line y + R θ y+\mathbb{R}\theta y + R θ where θ ∈ S ℓ − 1 \theta\in\mathbb{S}^{\ell-1} θ ∈ S ℓ − 1 and y ∈ θ ⊥ y\in\theta^\perp y ∈ θ ⊥ , if ( y + R θ ) ∩ Ω ≠ ∅ (y+\mathbb{R}\theta)\cap\Omega\ne\varnothing ( y + R θ ) ∩ Ω = ∅ , there exist y 0 ∈ ( y + R θ ) ∩ Ω r y_0\in(y+\mathbb{R}\theta)\cap\Omega_r y 0 ∈ ( y + R θ ) ∩ Ω r , which results into
∫ − ∞ ∞ p 11 12 ( y − t θ ) d t = 0 , for all ( θ , y ) ∈ T S ℓ − 1 . (33) \int_{-\infty}^{\infty}p_{11}^{12}(y-t\theta)\,dt=0,\quad\text{for all }(\theta,y)\in T\mathbb{S}^{\ell-1}. \tag*{(33)} ∫ − ∞ ∞ p 11 12 ( y − tθ ) d t = 0 , for all ( θ , y ) ∈ T S ℓ − 1 . (33)
However, if ( y + R θ ) ∩ Ω = ∅ (y+\mathbb{R}\theta)\cap\Omega=\varnothing ( y + R θ ) ∩ Ω = ∅ , then (33) is trivially true, as supp ( p 11 12 ) ⊂ Ω \operatorname{supp}(p_{11}^{12})\subset\Omega supp ( p 11 12 ) ⊂ Ω . Finally, invoking the inversion of X X X -ray transform from Theorem 1.1 of [42 ] leads to the establishment of the unique recovery of the coefficients, that is p 11 12 = 0 p_{11}^{12}=0 p 11 12 = 0 in Ω \Omega Ω , which simply tells us that p 11 1 = p 11 2 p_{11}^{1}=p_{11}^{2} p 11 1 = p 11 2 in Ω \Omega Ω . For the other coefficients of the matrix P 12 \mathbf{P}^{12} P 12 , again we need to choose the particular values of the constant vectors K ⃗ ( + ) \vec{\mathcal{K}}^{(+)} K ( + ) and K ⃗ ( − ) \vec{\mathcal{K}}^{(-)} K ( − ) . Specifically, to recover p i j 12 p_{ij}^{12} p ij 12 , we set K ⃗ ( + ) = e j \vec{\mathcal{K}}^{(+)}=e_j K ( + ) = e j and K ⃗ ( − ) = e i \vec{\mathcal{K}}^{(-)}=e_i K ( − ) = e i , where e i e_i e i denotes the i i i -th standard basis vector. Subsequently, the preceding analysis completes the proof.
Unique determination of nonlinear potential
To do so, we first observe that the second-order partial derivative ∂ ε i ε j 2 \partial^2_{\varepsilon_i\varepsilon_j} ∂ ε i ε j 2 does not lead to any extra information. So, to this end, we apply ∂ ε 1 ε 2 ε 3 3 \partial^3_{\varepsilon_1\varepsilon_2\varepsilon_3} ∂ ε 1 ε 2 ε 3 3 to the IBVP (14) and evaluate at ε = 0 \varepsilon=0 ε = 0 , to get
{ L P w ⃗ ( t , x ) = − b ⃗ , ( t , x ) ∈ Ω T , w ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , w ⃗ ( 0 , x ) = ∂ t w ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{w}(t,x)=-\vec{b}, & (t,x)\in\Omega_T,\\
\vec{w}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\vec{w}(0,x)=\partial_t\vec{w}(0,x)=\vec{0}, & x\in\Omega.
\end{cases} ⎩ ⎨ ⎧ L P w ( t , x ) = − b , w ( t , x ) = 0 , w ( 0 , x ) = ∂ t w ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω.
where ∂ ε 1 ε 2 ε 3 3 u ⃗ ∣ ε = 0 = w ⃗ : = ( w 1 , w 2 ) T \left.\partial^3_{\varepsilon_1\varepsilon_2\varepsilon_3}\vec{u}\right|_{\varepsilon=0}=\vec{w}:=(w_1,w_2)^{\mathsf{T}} ∂ ε 1 ε 2 ε 3 3 u ε = 0 = w := ( w 1 , w 2 ) T , and
b ⃗ = [ b 1 b 2 ] = [ 6 q 11 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] . \vec{b}=
\begin{bmatrix}
b_1\\
b_2
\end{bmatrix}
=
\begin{bmatrix}
6q_{11}(x)v_1^{(1)}v_1^{(2)}v_1^{(3)}+2q_{12}(x)v_1^{(1)}v_1^{(3)}v_2^{(2)}+2q_{12}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+2q_{12}(x)v_2^{(1)}v_1^{(2)}v_1^{(3)}\\
2q_{21}(x)v_2^{(1)}v_2^{(3)}v_1^{(2)}+2q_{21}(x)v_2^{(1)}v_2^{(2)}v_1^{(3)}+2q_{21}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+6q_{22}(x)v_2^{(1)}v_2^{(2)}v_2^{(3)}
\end{bmatrix}. b = [ b 1 b 2 ] = [ 6 q 11 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] .
A third-order linearization of the DN map (15) is
3 Λ P , Q ∣ ∂ Ω T : = ( ∂ ε 1 ε 2 ε 3 3 Λ P , Q ( ε f ⃗ ) ) ∣ ε = 0 ∣ ∂ Ω T = ∂ ν w ⃗ ∣ ∂ Ω T . {}^3\Lambda_{\mathbb{P},\mathbb{Q}}\big|_{\partial\Omega_T}:=\left.\left(\partial^3_{\varepsilon_1\varepsilon_2\varepsilon_3}\Lambda_{\mathbb{P},\mathbb{Q}}(\varepsilon\vec{f})\right)\right|_{\varepsilon=0}\bigg|_{\partial\Omega_T}=\left.\partial_\nu\vec{w}\right|_{\partial\Omega_T}. 3 Λ P , Q ∂ Ω T := ( ∂ ε 1 ε 2 ε 3 3 Λ P , Q ( ε f ) ) ε = 0 ∂ Ω T = ∂ ν w ∣ ∂ Ω T .
To prove uniqueness, for i = 1 , 2 i=1,2 i = 1 , 2 , we consider w ⃗ i \vec{w}^{\,i} w i be the solution to the IBVP
{ L P w ⃗ i ( t , x ) = − b ⃗ i , ( t , x ) ∈ Ω T , w ⃗ i ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , w ⃗ i ( 0 , x ) = ∂ t w ⃗ i ( 0 , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{w}^{\,i}(t,x)=-\vec{b}^{\,i}, & (t,x)\in\Omega_T,\\
\vec{w}^{\,i}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\vec{w}^{\,i}(0,x)=\partial_t\vec{w}^{\,i}(0,x)=\vec{0}, & x\in\Omega.
\end{cases} ⎩ ⎨ ⎧ L P w i ( t , x ) = − b i , w i ( t , x ) = 0 , w i ( 0 , x ) = ∂ t w i ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω.
where
b ⃗ i = [ 6 q 11 i ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 i ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 i ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 i ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 i ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 i ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 i ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 i ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] . \vec{b}^{\,i}=
\begin{bmatrix}
6q_{11}^{i}(x)v_1^{(1)}v_1^{(2)}v_1^{(3)}+2q_{12}^{i}(x)v_1^{(1)}v_1^{(3)}v_2^{(2)}+2q_{12}^{i}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+2q_{12}^{i}(x)v_2^{(1)}v_1^{(2)}v_1^{(3)}\\
2q_{21}^{i}(x)v_2^{(1)}v_2^{(3)}v_1^{(2)}+2q_{21}^{i}(x)v_2^{(1)}v_2^{(2)}v_1^{(3)}+2q_{21}^{i}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+6q_{22}^{i}(x)v_2^{(1)}v_2^{(2)}v_2^{(3)}
\end{bmatrix}. b i = [ 6 q 11 i ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 i ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 i ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 i ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 i ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 i ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 i ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 i ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] .
A third-order linearization of the DN map (15) corresponding to P = P i \mathbb{P}=\mathbb{P}^{i} P = P i and Q = Q i \mathbb{Q}=\mathbb{Q}^{i} Q = Q i is given by
3 Λ P , Q i ∣ ∂ Ω T = ∂ ν w ⃗ i ∣ ∂ Ω T . (34) {}^3\Lambda_{\mathbb{P},\mathbb{Q}^{i}}\big|_{\partial\Omega_T}=\left.\partial_\nu\vec{w}^{\,i}\right|_{\partial\Omega_T}. \tag*{(34)} 3 Λ P , Q i ∂ Ω T = ∂ ν w i ∂ Ω T . (34)
The matrix P \mathbb{P} P remains unchanged, since it’s unique recovery has already been established. That’s why a single notation P = P 1 = P 2 \mathbb{P}=\mathbb{P}^{1}=\mathbb{P}^{2} P = P 1 = P 2 is used. Also, we denote
w ⃗ 12 : = w ⃗ 1 − w ⃗ 2 , Q 12 = Q 1 − Q 2 . \vec{w}^{\,12}:=\vec{w}^{\,1}-\vec{w}^{\,2},\qquad\mathbb{Q}^{12}=\mathbb{Q}^{1}-\mathbb{Q}^{2}. w 12 := w 1 − w 2 , Q 12 = Q 1 − Q 2 .
Then we have
{ L P w ⃗ 12 ( t , x ) = − b ⃗ 12 , ( t , x ) ∈ Ω T , w ⃗ 12 ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , w ⃗ 12 ( 0 , x ) = ∂ t w ⃗ 12 ( 0 , x ) = 0 ⃗ , x ∈ Ω . (35) \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{w}^{\,12}(t,x)=-\vec{b}^{\,12}, & (t,x)\in\Omega_T,\\
\vec{w}^{\,12}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\vec{w}^{\,12}(0,x)=\partial_t\vec{w}^{\,12}(0,x)=\vec{0}, & x\in\Omega. \tag*{(35)}
\end{cases} ⎩ ⎨ ⎧ L P w 12 ( t , x ) = − b 12 , w 12 ( t , x ) = 0 , w 12 ( 0 , x ) = ∂ t w 12 ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (35)
where
b ⃗ 12 = [ 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] . \vec{b}^{\,12}=
\begin{bmatrix}
6q_{11}^{12}(x)v_1^{(1)}v_1^{(2)}v_1^{(3)}+2q_{12}^{12}(x)v_1^{(1)}v_1^{(3)}v_2^{(2)}+2q_{12}^{12}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+2q_{12}^{12}(x)v_2^{(1)}v_1^{(2)}v_1^{(3)}\\
2q_{21}^{12}(x)v_2^{(1)}v_2^{(3)}v_1^{(2)}+2q_{21}^{12}(x)v_2^{(1)}v_2^{(2)}v_1^{(3)}+2q_{21}^{12}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+6q_{22}^{12}(x)v_2^{(1)}v_2^{(2)}v_2^{(3)}
\end{bmatrix}. b 12 = [ 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] .
Using Λ P 1 , Q 1 ( ε f ⃗ ) = Λ P 2 , Q 2 ( ε f ⃗ ) \Lambda_{\mathbb{P}^{1},\mathbb{Q}^{1}}(\varepsilon\vec{f})=\Lambda_{\mathbb{P}^{2},\mathbb{Q}^{2}}(\varepsilon\vec{f}) Λ P 1 , Q 1 ( ε f ) = Λ P 2 , Q 2 ( ε f ) , in conjunction with (34) , we have ∂ ν w ⃗ 12 ∣ ∂ Ω T = 0 ⃗ \left.\partial_\nu\vec{w}^{\,12}\right|_{\partial\Omega_T}=\vec{0} ∂ ν w 12 ∂ Ω T = 0 . Let v ⃗ 0 : = ( v 01 , v 02 ) T \vec{v}_0:=(v_{01},v_{02})^{\mathsf{T}} v 0 := ( v 01 , v 02 ) T be the solution to the backward problem
{ □ v ⃗ 0 + P T v ⃗ 0 = 0 ⃗ , ( t , x ) ∈ Ω T , v ⃗ 0 ( T , x ) = ∂ t v ⃗ 0 ( T , x ) = 0 ⃗ , x ∈ Ω . (36) \begin{cases}
\Box\vec{v}_0+\mathbb{P}^{\mathsf{T}}\vec{v}_0=\vec{0}, & (t,x)\in\Omega_T,\\
\vec{v}_0(T,x)=\partial_t\vec{v}_0(T,x)=\vec{0}, & x\in\Omega. \tag*{(36)}
\end{cases} { □ v 0 + P T v 0 = 0 , v 0 ( T , x ) = ∂ t v 0 ( T , x ) = 0 , ( t , x ) ∈ Ω T , x ∈ Ω. (36)
Multiply (35) by v ⃗ 0 \vec{v}_0 v 0 and integrate over Ω T \Omega_T Ω T to get
− ∫ Ω T ( □ w ⃗ 12 + P w ⃗ 12 ) ⋅ v ⃗ 0 d x d t = ∫ Ω T ( b 1 12 v 01 + b 2 12 v 02 ) d x d t = ∫ Ω T ( 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) ) v 01 d x d t + ∫ Ω T ( 2 q 21 12 ( x ) v 2 ( 2 ) v 1 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ) v 02 d x d t , \begin{aligned}
-\int_{\Omega_T}\left(\Box\vec{w}^{12}+\mathsf{P}\vec{w}^{12}\right)\cdot\vec{v}_0\,dxdt \\
&=\int_{\Omega_T}\left(b^{12}_1v_{01}+b^{12}_2v_{02}\right)\,dxdt \\
&=\int_{\Omega_T}\left(6q^{12}_{11}(x)v^{(1)}_1v^{(2)}_1v^{(3)}_1+2q^{12}_{12}(x)v^{(1)}_1v^{(3)}_1v^{(2)}_2+2q^{12}_{12}(x)v^{(1)}_1v^{(2)}_1v^{(3)}_2+2q^{12}_{12}(x)v^{(1)}_2v^{(2)}_1v^{(3)}_1\right)v_{01}\,dxdt \\
&+\int_{\Omega_T}\left(2q^{12}_{21}(x)v^{(2)}_2v^{(3)}_1v^{(2)}_1+2q^{12}_{21}(x)v^{(1)}_2v^{(2)}_2v^{(3)}_1+2q^{12}_{21}(x)v^{(1)}_1v^{(2)}_2v^{(3)}_2+6q^{12}_{22}(x)v^{(1)}_2v^{(2)}_2v^{(3)}_2\right)v_{02}\,dxdt,
\end{aligned} − ∫ Ω T ( □ w 12 + P w 12 ) ⋅ v 0 d x d t = ∫ Ω T ( b 1 12 v 01 + b 2 12 v 02 ) d x d t = ∫ Ω T ( 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) ) v 01 d x d t + ∫ Ω T ( 2 q 21 12 ( x ) v 2 ( 2 ) v 1 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ) v 02 d x d t ,
where b ⃗ 12 : = ( b 1 12 , b 2 12 ) ⊤ \vec{b}^{12}:=(b^{12}_1,b^{12}_2)^\top b 12 := ( b 1 12 , b 2 12 ) ⊤ . The integration by parts formula, along with initial/final time information and boundary data in (35) and (36) leads to
∫ Ω T ( 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) ) v 01 d x d t + ∫ Ω T ( 2 q 21 12 ( x ) v 2 ( 2 ) v 1 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ) v 02 d x d t + ∫ Ω T 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) v 02 d x d t = 0. (37) \begin{aligned}
\int_{\Omega_T}\left(6q^{12}_{11}(x)v^{(1)}_1v^{(2)}_1v^{(3)}_1+2q^{12}_{12}(x)v^{(1)}_1v^{(3)}_1v^{(2)}_2+2q^{12}_{12}(x)v^{(1)}_1v^{(2)}_1v^{(3)}_2+2q^{12}_{12}(x)v^{(1)}_2v^{(2)}_1v^{(3)}_1\right)v_{01}\,dxdt \\
&+\int_{\Omega_T}\left(2q^{12}_{21}(x)v^{(2)}_2v^{(3)}_1v^{(2)}_1+2q^{12}_{21}(x)v^{(1)}_2v^{(2)}_2v^{(3)}_1+2q^{12}_{21}(x)v^{(1)}_1v^{(2)}_2v^{(3)}_2\right)v_{02}\,dxdt \\
&+\int_{\Omega_T}6q^{12}_{22}(x)v^{(1)}_2v^{(2)}_2v^{(3)}_2v_{02}\,dxdt=0. \tag*{(37)}
\end{aligned} ∫ Ω T ( 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) ) v 01 d x d t + ∫ Ω T ( 2 q 21 12 ( x ) v 2 ( 2 ) v 1 ( 3 ) v 1 ( 2 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ) v 02 d x d t + ∫ Ω T 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) v 02 d x d t = 0. (37)
Next, we choose (see Proposition 2.3 and 2.4)
v ⃗ ( i ) ( t , x ) = exp ( i λ ϕ ) ( z 0 ( t , x ) K ⃗ + , i + Z 1 K ⃗ + , i ( t , x ) λ + Z 2 K ⃗ + , i ( t , x ) λ 2 + ⋯ + Z N K ⃗ + , i ( t , x ) λ N ) + R ⃗ K ⃗ + , i , \vec{v}^{(i)}(t,x)=\exp(i\lambda\phi)\left(z_0(t,x)\vec{K}^{+ ,i}+\frac{Z_1\vec{K}^{+ ,i}(t,x)}{\lambda}+\frac{Z_2\vec{K}^{+ ,i}(t,x)}{\lambda^2}+\cdots+\frac{Z_N\vec{K}^{+ ,i}(t,x)}{\lambda^N}\right)+\vec{R}_{\vec{K}^{+ ,i}}, v ( i ) ( t , x ) = exp ( iλ ϕ ) ( z 0 ( t , x ) K + , i + λ Z 1 K + , i ( t , x ) + λ 2 Z 2 K + , i ( t , x ) + ⋯ + λ N Z N K + , i ( t , x ) ) + R K + , i ,
and
v ⃗ 0 ( t , x ) = exp ( − 3 i λ ϕ ) ( w 0 ( t , x ) K ⃗ − + W 1 K ⃗ − ( t , x ) λ + W 2 K ⃗ − ( t , x ) λ 2 + ⋯ + W N K ⃗ − ( t , x ) λ N ) + R ⃗ K ⃗ − , \vec{v}_0(t,x)=\exp(-3i\lambda\phi)\left(w_0(t,x)\vec{K}^{-}+\frac{W_1\vec{K}^{-}(t,x)}{\lambda}+\frac{W_2\vec{K}^{-}(t,x)}{\lambda^2}+\cdots+\frac{W_N\vec{K}^{-}(t,x)}{\lambda^N}\right)+\vec{R}_{\vec{K}^{-}}, v 0 ( t , x ) = exp ( − 3 iλ ϕ ) ( w 0 ( t , x ) K − + λ W 1 K − ( t , x ) + λ 2 W 2 K − ( t , x ) + ⋯ + λ N W N K − ( t , x ) ) + R K − ,
where v ⃗ ( i ) \vec{v}^{(i)} v ( i ) and v ⃗ 0 \vec{v}_0 v 0 are the asymptotic solutions to (16) and (36) , respectively. Let φ ∈ C c ∞ ( R ℓ ) \varphi\in\mathcal{C}^{\infty}_{c}(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) then z 0 = w 0 = φ ( x + t θ ) z_0=w_0=\varphi(x+t\theta) z 0 = w 0 = φ ( x + tθ ) solve the transport equation (12) and (13) . The remainder terms R ⃗ K ⃗ + , i ( t , x , λ ) \vec{R}_{\vec{K}^{+ ,i}}(t,x,\lambda) R K + , i ( t , x , λ ) for 1 ≤ i ≤ 3 1\leq i\leq3 1 ≤ i ≤ 3 and R ⃗ K ⃗ − ( t , x , λ ) \vec{R}_{\vec{K}^{-}}(t,x,\lambda) R K − ( t , x , λ ) , satisfy the following estimates:
∥ R ⃗ K ⃗ + , i ∥ L ∞ ( Ω T ) ≤ C λ − 1 and ∥ R ⃗ K ⃗ − ∥ L ∞ ( Ω T ) ≤ C λ − 1 , (38) \left\|\vec{R}_{\vec{K}^{+ ,i}}\right\|_{L^\infty(\Omega_T)}\leq C\lambda^{-1}
\quad\text{and}\quad
\left\|\vec{R}_{\vec{K}^{-}}\right\|_{L^\infty(\Omega_T)}\leq C\lambda^{-1}, \tag*{(38)} R K + , i L ∞ ( Ω T ) ≤ C λ − 1 and R K − L ∞ ( Ω T ) ≤ C λ − 1 , (38)
where C > 0 C>0 C > 0 is independent of λ \lambda λ . Also, we have
R ⃗ K ⃗ + , i ( 0 , x , λ ) = ∂ t R ⃗ K ⃗ + , i ( 0 , x , λ ) = 0 ⃗ and R ⃗ K ⃗ − ( T , x , λ ) = ∂ t R ⃗ K ⃗ − ( T , x , λ ) = 0 ⃗ , \vec{R}_{\vec{K}^{+ ,i}}(0,x,\lambda)=\partial_t\vec{R}_{\vec{K}^{+ ,i}}(0,x,\lambda)=\vec{0}
\quad\text{and}\quad
\vec{R}_{\vec{K}^{-}}(T,x,\lambda)=\partial_t\vec{R}_{\vec{K}^{-}}(T,x,\lambda)=\vec{0}, R K + , i ( 0 , x , λ ) = ∂ t R K + , i ( 0 , x , λ ) = 0 and R K − ( T , x , λ ) = ∂ t R K − ( T , x , λ ) = 0 ,
for x ∈ Ω x\in\Omega x ∈ Ω and 1 ≤ i ≤ 3 1\leq i\leq3 1 ≤ i ≤ 3 . For 1 ≤ j ≤ N 1\leq j\leq N 1 ≤ j ≤ N and 1 ≤ i ≤ 3 1\leq i\leq3 1 ≤ i ≤ 3 , we denote
z 0 K ⃗ + , i = [ z 0 K 1 + , i z 0 K 2 + , i ] , w 0 K ⃗ + , i = [ w 0 K 1 − w 0 K 2 − ] , Z j , K ⃗ + , i = [ Z 1 , ( j , K ⃗ + , i ) Z 2 , ( j , K ⃗ + , i ) ] , z_0\vec{K}^{+ ,i}=
\begin{bmatrix}
z_0K^{+ ,i}_1\\
z_0K^{+ ,i}_2
\end{bmatrix},
\qquad
w_0\vec{K}^{+ ,i}=
\begin{bmatrix}
w_0K^{-}_1\\
w_0K^{-}_2
\end{bmatrix},
\qquad
Z_{j,\vec{K}^{+ ,i}}=
\begin{bmatrix}
Z_{1,(j,\vec{K}^{+ ,i})}\\
Z_{2,(j,\vec{K}^{+ ,i})}
\end{bmatrix}, z 0 K + , i = [ z 0 K 1 + , i z 0 K 2 + , i ] , w 0 K + , i = [ w 0 K 1 − w 0 K 2 − ] , Z j , K + , i = [ Z 1 , ( j , K + , i ) Z 2 , ( j , K + , i ) ] ,
W j , K ⃗ − = [ W 1 , ( j , K ⃗ − ) W 2 , ( j , K ⃗ − ) ] , R ⃗ K ⃗ + , i = [ R 1 , K ⃗ + , i R 2 , K ⃗ + , i ] , R ⃗ K ⃗ − = [ R 1 , K ⃗ − R 2 , K ⃗ − ] , W_{j,\vec{K}^{-}}=
\begin{bmatrix}
W_{1,(j,\vec{K}^{-})}\\
W_{2,(j,\vec{K}^{-})}
\end{bmatrix},
\qquad
\vec{R}_{\vec{K}^{+ ,i}}=
\begin{bmatrix}
R_{1,\vec{K}^{+ ,i}}\\
R_{2,\vec{K}^{+ ,i}}
\end{bmatrix},
\qquad
\vec{R}_{\vec{K}^{-}}=
\begin{bmatrix}
R_{1,\vec{K}^{-}}\\
R_{2,\vec{K}^{-}}
\end{bmatrix}, W j , K − = [ W 1 , ( j , K − ) W 2 , ( j , K − ) ] , R K + , i = [ R 1 , K + , i R 2 , K + , i ] , R K − = [ R 1 , K − R 2 , K − ] ,
Z i = [ Z 1 , i Z 2 , i ] : = [ ∑ j = 1 N Z 1 , ( j , K ⃗ + , i ) λ j ∑ j = 1 N Z 2 , ( j , K ⃗ + , i ) λ j ] , and W = [ W 1 W 2 ] : = [ ∑ j = 1 N W 1 , ( j , K ⃗ − ) λ j ∑ j = 1 N W 2 , ( j , K ⃗ − ) λ j ] . Z_i=
\begin{bmatrix}
Z_{1,i}\\
Z_{2,i}
\end{bmatrix}
:=
\begin{bmatrix}
\sum_{j=1}^{N}\frac{Z_{1,(j,\vec{K}^{+ ,i})}}{\lambda^j}\\
\sum_{j=1}^{N}\frac{Z_{2,(j,\vec{K}^{+ ,i})}}{\lambda^j}
\end{bmatrix},
\qquad
\text{and}\quad
\mathcal{W}=
\begin{bmatrix}
W_1\\
W_2
\end{bmatrix}
:=
\begin{bmatrix}
\sum_{j=1}^{N}\frac{W_{1,(j,\vec{K}^{-})}}{\lambda^j}\\
\sum_{j=1}^{N}\frac{W_{2,(j,\vec{K}^{-})}}{\lambda^j}
\end{bmatrix}. Z i = [ Z 1 , i Z 2 , i ] := [ ∑ j = 1 N λ j Z 1 , ( j , K + , i ) ∑ j = 1 N λ j Z 2 , ( j , K + , i ) ] , and W = [ W 1 W 2 ] := [ ∑ j = 1 N λ j W 1 , ( j , K − ) ∑ j = 1 N λ j W 2 , ( j , K − ) ] .
For 1 ≤ k , l ≤ 2 1 \le k,l \le2 1 ≤ k , l ≤ 2 , the above notation in conjunction with (38) leads to
Case 1: k ≠ l k \ne l k = l
v k ( 1 ) v l ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + t θ ) K k + , 1 K l + , 2 K k + , 3 K − + O ( λ − 1 ) , v_k^{(1)}v_l^{(2)}v_k^{(3)}v_{0k}=\varphi^4(x+t\theta)K_k^{+,1}K_l^{+,2}K_k^{+,3}K_-+O(\lambda^{-1}), v k ( 1 ) v l ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + tθ ) K k + , 1 K l + , 2 K k + , 3 K − + O ( λ − 1 ) ,
v k ( 1 ) v l ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + t θ ) K k + , 1 K l + , 2 K k + , 3 K − + O ( λ − 1 ) , v_k^{(1)}v_l^{(2)}v_k^{(3)}v_{0k}=\varphi^4(x+t\theta)K_k^{+,1}K_l^{+,2}K_k^{+,3}K_-+O(\lambda^{-1}), v k ( 1 ) v l ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + tθ ) K k + , 1 K l + , 2 K k + , 3 K − + O ( λ − 1 ) ,
v l ( 1 ) v k ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + t θ ) K l + , 1 K k + , 2 K k + , 3 K − + O ( λ − 1 ) , v_l^{(1)}v_k^{(2)}v_k^{(3)}v_{0k}=\varphi^4(x+t\theta)K_l^{+,1}K_k^{+,2}K_k^{+,3}K_-+O(\lambda^{-1}), v l ( 1 ) v k ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + tθ ) K l + , 1 K k + , 2 K k + , 3 K − + O ( λ − 1 ) ,
Case 2: k = l k=l k = l
v k ( 1 ) v k ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + t θ ) K k + , 1 K k + , 2 K k + , 3 K − + O ( λ − 1 ) . v_k^{(1)}v_k^{(2)}v_k^{(3)}v_{0k}=\varphi^4(x+t\theta)K_k^{+,1}K_k^{+,2}K_k^{+,3}K_-+O(\lambda^{-1}). v k ( 1 ) v k ( 2 ) v k ( 3 ) v 0 k = φ 4 ( x + tθ ) K k + , 1 K k + , 2 K k + , 3 K − + O ( λ − 1 ) .
Here we have used that z 0 = w 0 = φ ( x + t θ ) z_0=w_0=\varphi(x+t\theta) z 0 = w 0 = φ ( x + tθ ) , where φ ∈ C c ∞ ( R ℓ ) \varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) . Order λ − 1 \lambda^{-1} λ − 1 appears from the residual terms beyond the leading term in the aforementioned asymptotic expansion and/or from the estimates correction terms, possibly in combination with other lower-order terms. The substitution of aforementioned GO solution in (37) and use of the above expression yields
∫ Ω T φ 4 ( x + t θ ) ( 6 q 11 12 ( x ) K 1 + , 1 K 1 + , 2 K 1 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 2 + , 2 K 1 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 1 + , 2 K 2 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 1 + , 2 K 1 + , 3 K − + 2 q 21 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − + 2 q 21 12 ( x ) K 1 + , 1 K 2 + , 2 K 2 + , 3 K − + 2 q 21 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − + 6 q 22 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − ) d x d t = 0 , (39) \begin{aligned}
\int_{\Omega_T}\varphi^4(x+t\theta)\big(&6q_{11}^{12}(x)K_1^{+,1}K_1^{+,2}K_1^{+,3}K_-+2q_{12}^{12}(x)K_1^{+,1}K_2^{+,2}K_1^{+,3}K_-\\
&+2q_{12}^{12}(x)K_1^{+,1}K_1^{+,2}K_2^{+,3}K_-+2q_{12}^{12}(x)K_1^{+,1}K_1^{+,2}K_1^{+,3}K_-\\
&+2q_{21}^{12}(x)K_2^{+,1}K_2^{+,2}K_2^{+,3}K_-+2q_{21}^{12}(x)K_1^{+,1}K_2^{+,2}K_2^{+,3}K_-\\
&+2q_{21}^{12}(x)K_2^{+,1}K_2^{+,2}K_2^{+,3}K_-+6q_{22}^{12}(x)K_2^{+,1}K_2^{+,2}K_2^{+,3}K_-\big)\,dxdt=0, \tag*{(39)}
\end{aligned} ∫ Ω T φ 4 ( x + tθ ) ( 6 q 11 12 ( x ) K 1 + , 1 K 1 + , 2 K 1 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 2 + , 2 K 1 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 1 + , 2 K 2 + , 3 K − + 2 q 12 12 ( x ) K 1 + , 1 K 1 + , 2 K 1 + , 3 K − + 2 q 21 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − + 2 q 21 12 ( x ) K 1 + , 1 K 2 + , 2 K 2 + , 3 K − + 2 q 21 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − + 6 q 22 12 ( x ) K 2 + , 1 K 2 + , 2 K 2 + , 3 K − ) d x d t = 0 , (39)
as λ → ∞ \lambda\to\infty λ → ∞ . To ensure the unique identification of each coefficient of nonlinearity, we prescribe a specific choice of the constant vectors K ⃗ + , i \vec K^{+,i} K + , i and K ⃗ − \vec K^- K − . In particular, we choose K ⃗ + , 1 = K ⃗ + , 2 = K ⃗ + , 3 = K ⃗ − = e 1 \vec K^{+,1}=\vec K^{+,2}=\vec K^{+,3}=\vec K^-=e_1 K + , 1 = K + , 2 = K + , 3 = K − = e 1 , where e i e_i e i denotes the i i i -th standard basis vector of R 2 \mathbb{R}^2 R 2 . Using these choices of constant vector in (39) leads to
∫ Ω T φ 4 ( x + t θ ) q 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . \int_{\Omega_T}\varphi^4(x+t\theta)q_{11}^{12}(x)\,dxdt=0,\quad\forall\varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell})\text{ and }\theta\in\mathbb{S}^{\ell-1}. ∫ Ω T φ 4 ( x + tθ ) q 11 12 ( x ) d x d t = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 .
We implement an argument analogous to that used to establish the uniqueness of the coefficient from the integral identity (28) to show that q 11 12 = 0 q_{11}^{12}=0 q 11 12 = 0 . However, establishing the unique recovery of the coefficients of the nonlinearity requires a suitable choice of φ ∈ C c ∞ ( R ℓ ) \varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) . Extend q 11 12 q_{11}^{12} q 11 12 by zero to R ℓ ∖ Ω \mathbb{R}^{\ell}\setminus\Omega R ℓ ∖ Ω , and use the change of variable to get
∫ R ℓ ∫ 0 T φ 4 ( y ) q 11 12 ( y − t θ ) d t d y = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 . \int_{\mathbb{R}^{\ell}}\int_0^T\varphi^4(y)q_{11}^{12}(y-t\theta)\,dt\,dy=0,\quad\forall\varphi\in\mathcal{C}_c^\infty(\mathbb{R}^{\ell})\text{ and }\theta\in\mathbb{S}^{\ell-1}. ∫ R ℓ ∫ 0 T φ 4 ( y ) q 11 12 ( y − tθ ) d t d y = 0 , ∀ φ ∈ C c ∞ ( R ℓ ) and θ ∈ S ℓ − 1 .
For y 0 ∈ Ω r y_0\in\Omega_r y 0 ∈ Ω r , we define
ρ δ ( y ) = 1 δ ℓ ρ ( y − y 0 δ ) , \rho_\delta(y)=\frac{1}{\delta^\ell}\rho\left(\frac{y-y_0}{\delta}\right), ρ δ ( y ) = δ ℓ 1 ρ ( δ y − y 0 ) ,
where ρ ∈ C c ∞ ( B 1 ( 0 ) ) \rho\in C_c^\infty(B_1(0)) ρ ∈ C c ∞ ( B 1 ( 0 )) is a cutoff function lies in [ 0 , 1 ] [0,1] [ 0 , 1 ] with ∫ R ℓ ρ 4 ( x ) d x = 1 \int_{\mathbb{R}^{\ell}}\rho^4(x)\,dx=1 ∫ R ℓ ρ 4 ( x ) d x = 1 . It follows from the above construction that the sequence { ρ δ } δ > 0 ⊂ C c ∞ ( R ℓ ) \{\rho_\delta\}_{\delta>0}\subset C_c^\infty(\mathbb{R}^{\ell}) { ρ δ } δ > 0 ⊂ C c ∞ ( R ℓ ) whenever 0 < δ < r 0<\delta<r 0 < δ < r . Thus, we have
1 δ ℓ ∫ B δ ( y 0 ) ∫ 0 T q 11 12 ( y − t θ ) ρ 4 ( y − y 0 δ ) d t d y = 0 , for all 0 < δ < r and y 0 ∈ Ω r , \frac{1}{\delta^\ell}\int_{B_\delta(y_0)}\int_0^Tq_{11}^{12}(y-t\theta)\rho^4\left(\frac{y-y_0}{\delta}\right)\,dt\,dy=0,\quad\text{for all }0<\delta<r\text{ and }y_0\in\Omega_r, δ ℓ 1 ∫ B δ ( y 0 ) ∫ 0 T q 11 12 ( y − tθ ) ρ 4 ( δ y − y 0 ) d t d y = 0 , for all 0 < δ < r and y 0 ∈ Ω r ,
which then leads to
∫ 0 T q 11 12 ( y 0 − t θ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 , \int_{0}^{T} q_{11}^{12}(y_{0}-t\theta)\,\mathrm{d}t=0,\quad\text{for all }y_{0}\in\Omega_{r}\text{ and }\theta\in S^{\ell-1}, ∫ 0 T q 11 12 ( y 0 − tθ ) d t = 0 , for all y 0 ∈ Ω r and θ ∈ S ℓ − 1 ,
as δ → ∞ \delta\to\infty δ → ∞ . The foregoing integral is now reduced to (31) from which we concluded that p 11 12 = 0 p_{11}^{12}=0 p 11 12 = 0 , the only change is that the coefficient q 11 12 q_{11}^{12} q 11 12 appears in place of p 11 12 p_{11}^{12} p 11 12 . Since both coefficients belong to the same function class C 0 ∞ ( Ω ) C_{0}^{\infty}(\Omega) C 0 ∞ ( Ω ) , we have q 11 12 = 0 q_{11}^{12}=0 q 11 12 = 0 in Ω \Omega Ω , which simply tells that q 11 = q 11 12 q_{11}=q_{11}^{12} q 11 = q 11 12 in Ω \Omega Ω . For the other coefficients of the matrix Q 12 Q^{12} Q 12 , again we need to choose the particular values of the constant vectors K ⃗ + , i \vec{K}^{+,i} K + , i and K ⃗ − \vec{K}^{-} K − . Specifically, to recover the remaining coefficients, we consider the suitable choices of constant vectors given in Table 3.1 . Subsequently, the preceding analysis completes the proof.
Table 3.1. A particular choice of K ⃗ + , i \vec{K}^{+,i} K + , i and K ⃗ − \vec{K}^{-} K −
Proof of Theorem 1.2
In the following, we establish the unique identification of the coefficients appearing in (14) from the partial DN map given by
Λ P , Q ♯ ( ε f ⃗ ) : = ∂ ν u ⃗ ∣ ( 0 , T ) × Γ , \Lambda_{\mathbb{P},\mathbb{Q}}^{\sharp}(\varepsilon\vec{f}):=\partial_{\nu}\vec{u}|_{(0,T)\times\Gamma}, Λ P , Q ♯ ( ε f ) := ∂ ν u ∣ ( 0 , T ) × Γ ,
where u ⃗ \vec{u} u is a solution to (14) and Γ ⊂ ∂ Ω \Gamma\subset\partial\Omega Γ ⊂ ∂ Ω is a nonempty open set. First, in § 5.1 , we recall the unique continuation principle established in [5 ] , which is then used in converting the partial-data problem to a full-data problem. Next, in § 5.2 and § 5.3 , we discuss the unique identification of matrix-valued potential and nonlinear potential through the partial DN map, respectively.
Unique continuation principle
We begin with defining some notations to state the unique continuation principle (UCP) associated with the linear hyperbolic system
{ L P u ⃗ ( t , x ) = F ⃗ , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω , (40) \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{u}(t,x)=\vec{F}, & (t,x)\in\Omega_{T},\\
\vec{u}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_{T},\\
\vec{u}(0,x)=\vec{0},\quad\partial_{t}\vec{u}(0,x)=\vec{0}, & x\in\Omega, \tag*{(40)}
\end{cases} ⎩ ⎨ ⎧ L P u ( t , x ) = F , u ( t , x ) = 0 , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω , (40)
where P ∈ P \mathbb{P}\in\mathcal{P} P ∈ P and F ⃗ ∈ H 2 ( Ω T ) \vec{F}\in\mathbf{H}^{2}(\Omega_{T}) F ∈ H 2 ( Ω T ) . For the open set U U U in Ω ‾ \overline{\Omega} Ω , we introduce a family of relatively compact open subsets of U U U in Ω ‾ \overline{\Omega} Ω . To be more precise, we consider the open sets U j ⊆ U U_{j}\subseteq U U j ⊆ U , j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 , such that U ‾ j + 1 ⊂ U j \overline{U}_{j+1}\subset U_{j} U j + 1 ⊂ U j for j = 1 , 2 j=1,2 j = 1 , 2 and ∂ Ω ⊂ ∂ U j \partial\Omega\subset\partial U_{j} ∂ Ω ⊂ ∂ U j for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 . We also assume that U ‾ 1 ∩ ( ∂ U ∖ ∂ Ω ) = ∅ \overline{U}_{1}\cap(\partial U\setminus\partial\Omega)=\varnothing U 1 ∩ ( ∂ U ∖ ∂ Ω ) = ∅ . Now, we recall the following results from [5 ] .
Theorem 4.1. [5 Lemma 3.1] Let Γ ⊂ ∂ Ω \Gamma\subset\partial\Omega Γ ⊂ ∂ Ω be a nonempty open set, T > 0 T>0 T > 0 be sufficiently large and P ∈ P \mathbb{P}\in\mathcal{P} P ∈ P . Also, let u ⃗ \vec{u} u be the solution to the IBVP (40) . Then there exist positive constants T ′ < T T'<T T ′ < T , C C C , A A A and η 0 \eta_{0} η 0 such that
∥ u ⃗ ∥ H 1 ( ( 0 , T ′ ) × ( U 2 ∖ U ‾ 3 ) ) ≤ C η 1 / 2 ∥ u ⃗ ∥ H 2 ( Ω T ) + e A η ( ∥ F ⃗ ∥ L 2 ( ( 0 , T ) × U ) + ∥ ∂ ν u ⃗ ∥ L 2 ( ( 0 , T ) × Γ ) ) , \|\vec{u}\|_{\mathbf{H}^{1}((0,T')\times(U_{2}\setminus\overline{U}_{3}))}\leq\frac{C}{\eta^{1/2}}\|\vec{u}\|_{\mathbf{H}^{2}(\Omega_{T})}+e^{A\eta}\left(\|\vec{F}\|_{\mathbf{L}^{2}((0,T)\times U)}+\|\partial_{\nu}\vec{u}\|_{\mathbf{L}^{2}((0,T)\times\Gamma)}\right), ∥ u ∥ H 1 (( 0 , T ′ ) × ( U 2 ∖ U 3 )) ≤ η 1/2 C ∥ u ∥ H 2 ( Ω T ) + e A η ( ∥ F ∥ L 2 (( 0 , T ) × U ) + ∥ ∂ ν u ∥ L 2 (( 0 , T ) × Γ ) ) , holds for any η > η 0 \eta>\eta_{0} η > η 0 .
Corollary 4.2. In addition to the assumptions of Theorem 4.1, suppose that supp ( F ⃗ ) ⊂ ( 0 , T ) × ( Ω ∖ U ) \operatorname{supp}(\vec{F}) \subset(0,T) \times(\Omega\setminus U) supp ( F ) ⊂ ( 0 , T ) × ( Ω ∖ U ) and ∂ ν u ⃗ = 0 ⃗ \partial_\nu\vec{u} = \vec{0} ∂ ν u = 0 on ( 0 , T ) × Γ (0,T) \times\Gamma ( 0 , T ) × Γ . Then u ⃗ = 0 ⃗ \vec{u} = \vec{0} u = 0 in ( 0 , T ′ ) × ( U 2 ∖ U 3 ) (0,T') \times(U_2 \setminus U_3) ( 0 , T ′ ) × ( U 2 ∖ U 3 ) .
Recovery of matrix-valued potential
In this subsection, we give a proof for converting the partial data problem to a full data problem. We start by recalling the IBVP (see (21) )
{ □ v ⃗ 12 + P 1 v ⃗ 12 = − P 12 v ⃗ 2 , ( t , x ) ∈ Ω T , v ⃗ 12 ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , v ⃗ 12 ( 0 , x ) = ∂ t v ⃗ 12 ( 0 , x ) = 0 ⃗ , x ∈ Ω . (41) \begin{cases}
\square\vec{v}^{12} + \mathbb{P}^{1}\vec{v}^{12} = -\mathbb{P}^{12}\vec{v}^{2}, & (t,x) \in\Omega_T, \\
\vec{v}^{12}(t,x) = \vec{0}, & (t,x) \in\partial\Omega_T, \\
\vec{v}^{12}(0,x) = \partial_t\vec{v}^{12}(0,x) = \vec{0}, & x \in\Omega. \tag*{(41)}
\end{cases} ⎩ ⎨ ⎧ □ v 12 + P 1 v 12 = − P 12 v 2 , v 12 ( t , x ) = 0 , v 12 ( 0 , x ) = ∂ t v 12 ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (41)
where v ⃗ j \vec{v}^{j} v j is solution to (20) , v ⃗ 12 = v ⃗ 1 − v ⃗ 2 \vec{v}^{12} = \vec{v}^{1} - \vec{v}^{2} v 12 = v 1 − v 2 , and P 12 = P 1 − P 2 \mathbb{P}^{12} = \mathbb{P}^{1} - \mathbb{P}^{2} P 12 = P 1 − P 2 . Since Λ P 1 , Q 1 ♯ ( ε f ⃗ ) = Λ P 2 , Q 2 ♯ ( ε f ⃗ ) \Lambda_{\mathbb{P}^{1},Q_{1}}^{\sharp}(\varepsilon\vec{f}) = \Lambda_{\mathbb{P}^{2},Q_{2}}^{\sharp}(\varepsilon\vec{f}) Λ P 1 , Q 1 ♯ ( ε f ) = Λ P 2 , Q 2 ♯ ( ε f ) , the first order linearization of partial DN map leads to
1 Λ P 1 ♯ ( f , g ) ∣ ( 0 , T ) × Γ = 1 Λ P 2 ♯ ( f , g ) ∣ ( 0 , T ) × Γ , f , g ∈ H m + 1 ( ∂ Ω T ) . {}^{1}\Lambda_{\mathbb{P}^{1}}^{\sharp}(f,g)|_{(0,T)\times\Gamma} = {}^{1}\Lambda_{\mathbb{P}^{2}}^{\sharp}(f,g)|_{(0,T)\times\Gamma}, \qquad f,g \in H^{m+1}(\partial\Omega_T). 1 Λ P 1 ♯ ( f , g ) ∣ ( 0 , T ) × Γ = 1 Λ P 2 ♯ ( f , g ) ∣ ( 0 , T ) × Γ , f , g ∈ H m + 1 ( ∂ Ω T ) .
Consequently, we have ∂ ν v ⃗ 12 ∣ ( 0 , T ) × Γ = 0 ⃗ \partial_\nu\vec{v}^{12}|_{(0,T)\times\Gamma} = \vec{0} ∂ ν v 12 ∣ ( 0 , T ) × Γ = 0 . Let Ξ ∈ C ∞ ( Ω ‾ , [ 0 , 1 ] ) \Xi\in C^\infty(\overline{\Omega},[0,1]) Ξ ∈ C ∞ ( Ω , [ 0 , 1 ]) be the smooth cutoff function given by
Ξ : = { 0 in U 3 , 1 in Ω ∖ U ‾ 2 . (42) \Xi:= \begin{cases}
0 & \text{in } U_3,\\
1 & \text{in } \Omega\setminus\overline{U}_2.
\end{cases} \tag*{(42)} Ξ := { 0 1 in U 3 , in Ω ∖ U 2 . (42)
Set v ⃗ ~ : = Ξ v ⃗ 12 \widetilde{\vec{v}} := \Xi\vec{v}^{12} v := Ξ v 12 , then v ⃗ ~ \widetilde{\vec{v}} v solves
{ □ v ⃗ ~ + P 1 v ⃗ ~ = Ξ ( − P 12 v ⃗ 2 ) − ( Δ Ξ ) v ⃗ ~ − 2 ∇ Ξ ⋅ ∇ v ⃗ ~ , ( t , x ) ∈ Ω T , v ⃗ ~ ( t , x ) = 0 , ( t , x ) ∈ ∂ Ω T , v ⃗ ~ ( 0 , x ) = 0 , ∂ t v ⃗ ~ ( 0 , x ) = 0 , x ∈ Ω . (43) \begin{cases}
\square\widetilde{\vec{v}} + \mathbb{P}^{1}\widetilde{\vec{v}} = \Xi(-\mathbb{P}^{12}\vec{v}^{2}) - (\Delta\Xi)\widetilde{\vec{v}} - 2\nabla\Xi\cdot\nabla\widetilde{\vec{v}}, & (t,x) \in\Omega_T, \\
\widetilde{\vec{v}}(t,x) = 0, & (t,x) \in\partial\Omega_T, \\
\widetilde{\vec{v}}(0,x) = 0,\quad\partial_t\widetilde{\vec{v}}(0,x) = 0, & x \in\Omega. \tag*{(43)}
\end{cases} ⎩ ⎨ ⎧ □ v + P 1 v = Ξ ( − P 12 v 2 ) − ( ΔΞ ) v − 2∇Ξ ⋅ ∇ v , v ( t , x ) = 0 , v ( 0 , x ) = 0 , ∂ t v ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (43)
where v ⃗ 12 \vec{v}^{12} v 12 is a solution to IBVP (41) . Since P 1 = P 2 \mathbb{P}^{1} = \mathbb{P}^{2} P 1 = P 2 in U U U , we have P 12 = 0 \mathbb{P}^{12} = 0 P 12 = 0 in U U U . From the construction given by (42) , we have Ξ P 12 = P 12 \Xi\mathbb{P}^{12} = \mathbb{P}^{12} Ξ P 12 = P 12 in Ω ‾ \overline{\Omega} Ω . In addition, we have − ( Δ Ξ ) v ⃗ ~ − 2 ∇ Ξ ⋅ ∇ v ⃗ ~ ≡ 0 ⃗ -(\Delta\Xi)\widetilde{\vec{v}} - 2\nabla\Xi\cdot\nabla\widetilde{\vec{v}} \equiv\vec{0} − ( ΔΞ ) v − 2∇Ξ ⋅ ∇ v ≡ 0 in ( 0 , T ) × ( U 3 ∪ ( Ω ∖ U ‾ 2 ) ) (0,T) \times(U_3 \cup(\Omega\setminus\overline{U}_2)) ( 0 , T ) × ( U 3 ∪ ( Ω ∖ U 2 )) . Thus, F ⃗ : = Ξ ( − P 12 v ⃗ 2 ) − ( Δ Ξ ) v ⃗ ~ − 2 ∇ Ξ ⋅ ∇ v ⃗ ~ \vec{F} := \Xi(-\mathbb{P}^{12}\vec{v}^{2}) - (\Delta\Xi)\widetilde{\vec{v}} - 2\nabla\Xi\cdot\nabla\widetilde{\vec{v}} F := Ξ ( − P 12 v 2 ) − ( ΔΞ ) v − 2∇Ξ ⋅ ∇ v is supported in ( 0 , T ) × ( Ω ∖ U ) (0,T) \times(\Omega\setminus U) ( 0 , T ) × ( Ω ∖ U ) . Since v ⃗ ~ ≡ 0 ⃗ \widetilde{\vec{v}} \equiv\vec{0} v ≡ 0 in U 3 U_3 U 3 , we have ∂ ν v ⃗ ~ ∣ ∂ Ω T = 0 \partial_\nu\widetilde{\vec{v}}|_{\partial\Omega_T} = 0 ∂ ν v ∣ ∂ Ω T = 0 . From Corollary 4.2, we then have v ⃗ ~ = 0 ⃗ \widetilde{\vec{v}} = \vec{0} v = 0 in ( 0 , T ′ ) × ( U 2 ∖ U 3 ) (0,T') \times(U_2 \setminus U_3) ( 0 , T ′ ) × ( U 2 ∖ U 3 ) . Next, we multiply (43) with v ⃗ 0 \vec{v}_0 v 0 , where v ⃗ 0 \vec{v}_0 v 0 solves
{ □ v ⃗ 0 + ( P 1 ) T v ⃗ 0 = 0 ⃗ , ( t , x ) ∈ Ω T ′ : = ( 0 , T ′ ) × Ω , v ⃗ 0 ( T ′ , x ) = ∂ t v ⃗ 0 ( T ′ , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\square\vec{v}_0 + (\mathbb{P}^{1})^{\mathsf T}\vec{v}_0 = \vec{0}, & (t,x) \in\Omega_{T'} := (0,T') \times\Omega, \\
\vec{v}_0(T',x) = \partial_t\vec{v}_0(T',x) = \vec{0}, & x \in\Omega.
\end{cases} { □ v 0 + ( P 1 ) T v 0 = 0 , v 0 ( T ′ , x ) = ∂ t v 0 ( T ′ , x ) = 0 , ( t , x ) ∈ Ω T ′ := ( 0 , T ′ ) × Ω , x ∈ Ω.
and integrate over Ω T ′ \Omega_{T'} Ω T ′ to get
∫ Ω T ′ ( P 12 v ⃗ 2 ) ⋅ v ⃗ 0 d x d t = 0 ⃗ . \int_{\Omega_{T'}} (\mathbb{P}^{12}\vec{v}^{2}) \cdot\vec{v}_0 \, dxdt = \vec{0}. ∫ Ω T ′ ( P 12 v 2 ) ⋅ v 0 d x d t = 0 .
Choose (see Lemma 2.1 and Lemma 2.2)
v ⃗ 2 ( t , x ) = φ ( x + t θ ) exp ( i λ ϕ ) K ⃗ ( + ) + R ⃗ K ⃗ ( + ) ( t , x , λ ) , \vec{v}^{2}(t,x) = \varphi(x+t\theta)\exp(i\lambda\phi)\vec{\mathcal{K}}^{(+)} + \vec{\mathcal{R}}_{\vec{\mathcal{K}}^{(+)}}(t,x,\lambda), v 2 ( t , x ) = φ ( x + tθ ) exp ( iλ ϕ ) K ( + ) + R K ( + ) ( t , x , λ ) ,
and
v ⃗ 0 ( t , x ) = φ ( x + t θ ) exp ( − i λ ϕ ) K ⃗ ( − ) + R ⃗ K ⃗ ( − ) v ⃗ 0 ( t , x , λ ) . \vec{v}_0(t,x) = \varphi(x+t\theta)\exp(-i\lambda\phi)\vec{\mathcal{K}}^{(-)} + \vec{\mathcal{R}}_{\vec{\mathcal{K}}^{(-)}}^{\vec{v}_0}(t,x,\lambda). v 0 ( t , x ) = φ ( x + tθ ) exp ( − iλ ϕ ) K ( − ) + R K ( − ) v 0 ( t , x , λ ) .
for all φ ∈ C c ∞ ( R ℓ ) \varphi\in C^\infty_c(\mathbb{R}^{\ell}) φ ∈ C c ∞ ( R ℓ ) , where
C c ∞ ( R ℓ ) : = { φ ∈ C ∞ ( R ℓ ) : supp ( φ ) ∩ Ω = ∅ and ( supp ( φ ) ± T ′ θ ) ∩ Ω = ∅ , ∀ θ ∈ S ℓ − 1 } C^\infty_c(\mathbb{R}^{\ell}) := \{\varphi\in C^\infty(\mathbb{R}^{\ell}) : \operatorname{supp}(\varphi) \cap\Omega= \varnothing\text{ and } (\operatorname{supp}(\varphi) \pm T'\theta) \cap\Omega= \varnothing,\ \forall\theta\in\mathbb{S}^{\ell-1}\} C c ∞ ( R ℓ ) := { φ ∈ C ∞ ( R ℓ ) : supp ( φ ) ∩ Ω = ∅ and ( supp ( φ ) ± T ′ θ ) ∩ Ω = ∅ , ∀ θ ∈ S ℓ − 1 } .
Here the remainder term R ⃗ K ⃗ ( + ) v 2 ( t , x , λ ) \vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}(t,x,\lambda) R K ( + ) v 2 ( t , x , λ ) satisfies
R ⃗ K ⃗ ( + ) v 2 ( 0 , x , λ ) = ∂ t R ⃗ K ⃗ ( + ) v 2 ( 0 , x , λ ) = 0 ⃗ in Ω , \vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}(0,x,\lambda) = \partial_t\vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}(0,x,\lambda) = \vec{0}\text{ in }\Omega, R K ( + ) v 2 ( 0 , x , λ ) = ∂ t R K ( + ) v 2 ( 0 , x , λ ) = 0 in Ω ,
with
λ ∥ R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ) + ∥ ∂ t R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . \lambda\|\vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}\|_{L^2(\Omega_T)}+\|\partial_t\vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}\|_{L^2(\Omega_T)}\le C\|\varphi\|_{H^3(\mathbb{R}^{\ell})}. λ ∥ R K ( + ) v 2 ∥ L 2 ( Ω T ) + ∥ ∂ t R K ( + ) v 2 ∥ L 2 ( Ω T ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) .
From the last relation, we observe that
λ ∥ R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ′ ) + ∥ ∂ t R ⃗ K ⃗ ( + ) v 2 ∥ L 2 ( Ω T ′ ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . \lambda\|\vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}\|_{L^2(\Omega_{T'})}+\|\partial_t\vec{\mathcal{R}}_{\vec{K}_{(+)}^{v_2}}\|_{L^2(\Omega_{T'})}\le C\|\varphi\|_{H^3(\mathbb{R}^{\ell})}. λ ∥ R K ( + ) v 2 ∥ L 2 ( Ω T ′ ) + ∥ ∂ t R K ( + ) v 2 ∥ L 2 ( Ω T ′ ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) .
Also, the correction term R ⃗ K ⃗ ( − ) v 0 \vec{\mathcal{R}}_{\vec{K}_{(-)}^{v_0}} R K ( − ) v 0 satisfies
R ⃗ K ⃗ ( − ) v 0 ( T ′ , x , λ ) = ∂ t R ⃗ K ⃗ ( − ) v 0 ( T ′ , x , λ ) = 0 ⃗ in Ω , \vec{\mathcal{R}}_{\vec{K}_{(-)}^{v_0}}(T',x,\lambda)=\partial_t\vec{\mathcal{R}}_{\vec{K}_{(-)}^{v_0}}(T',x,\lambda)=\vec{0}\text{ in }\Omega, R K ( − ) v 0 ( T ′ , x , λ ) = ∂ t R K ( − ) v 0 ( T ′ , x , λ ) = 0 in Ω ,
with
λ ∥ R ⃗ K ⃗ ( − ) v 0 ∥ L 2 ( Ω T ′ ) + ∥ ∂ t R ⃗ K ⃗ ( − ) v 0 ∥ L 2 ( Ω T ′ ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) . \lambda\|\vec{\mathcal{R}}_{\vec{K}_{(-)}^{v_0}}\|_{L^2(\Omega_{T'})}+\|\partial_t\vec{\mathcal{R}}_{\vec{K}_{(-)}^{v_0}}\|_{L^2(\Omega_{T'})}\le C\|\varphi\|_{H^3(\mathbb{R}^{\ell})}. λ ∥ R K ( − ) v 0 ∥ L 2 ( Ω T ′ ) + ∥ ∂ t R K ( − ) v 0 ∥ L 2 ( Ω T ′ ) ≤ C ∥ φ ∥ H 3 ( R ℓ ) .
Proceeding as in §{3.1}, we obtain the unique identification of P \mathbb{P} P , that is, P 1 = P 2 \mathbb{P}^{1}=\mathbb{P}^{2} P 1 = P 2 in Ω \Omega Ω .
Recovery of nonlinear potential
We start by recalling the IBVP (see (35) )
{ L P w ⃗ 12 ( t , x ) = − b ⃗ 12 , ( t , x ) ∈ Ω T , w ⃗ 12 ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , w ⃗ 12 ( 0 , x ) = ∂ t w ⃗ 12 ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\mathcal{L}_{\mathbb{P}}\vec{w}^{12}(t,x)=-\vec{b}^{12}, & (t,x)\in\Omega_T,\\
\vec{w}^{12}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\vec{w}^{12}(0,x)=\partial_t\vec{w}^{12}(0,x)=\vec{0}, & x\in\Omega,
\end{cases} ⎩ ⎨ ⎧ L P w 12 ( t , x ) = − b 12 , w 12 ( t , x ) = 0 , w 12 ( 0 , x ) = ∂ t w 12 ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω ,
where
b ⃗ 12 = [ 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 3 ) v 2 ( 1 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] . \vec{b}^{12}=
\begin{bmatrix}
6q_{11}^{12}(x)v_1^{(1)}v_1^{(2)}v_1^{(3)}+2q_{12}^{12}(x)v_1^{(1)}v_1^{(3)}v_2^{(2)}+2q_{12}^{12}(x)v_1^{(1)}v_1^{(2)}v_2^{(3)}+2q_{12}^{12}(x)v_2^{(1)}v_1^{(2)}v_1^{(3)}\\
2q_{21}^{12}(x)v_1^{(1)}v_2^{(3)}v_2^{(1)}+2q_{21}^{12}(x)v_2^{(1)}v_2^{(2)}v_1^{(3)}+2q_{21}^{12}(x)v_1^{(1)}v_2^{(2)}v_2^{(3)}+6q_{22}^{12}(x)v_2^{(1)}v_2^{(2)}v_2^{(3)}
\end{bmatrix}. b 12 = [ 6 q 11 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 3 ) v 2 ( 2 ) + 2 q 12 12 ( x ) v 1 ( 1 ) v 1 ( 2 ) v 2 ( 3 ) + 2 q 12 12 ( x ) v 2 ( 1 ) v 1 ( 2 ) v 1 ( 3 ) 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 3 ) v 2 ( 1 ) + 2 q 21 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 1 ( 3 ) + 2 q 21 12 ( x ) v 1 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) + 6 q 22 12 ( x ) v 2 ( 1 ) v 2 ( 2 ) v 2 ( 3 ) ] .
P = P 1 = P 2 \mathbb{P}=\mathbb{P}^{1}=\mathbb{P}^{2} P = P 1 = P 2 , w ⃗ 12 = w ⃗ 1 − w ⃗ 2 \vec{w}^{12}=\vec{w}^{1}-\vec{w}^{2} w 12 = w 1 − w 2 , and Q 12 = Q 1 − Q 2 \mathbb{Q}^{12}=\mathbb{Q}^{1}-\mathbb{Q}^{2} Q 12 = Q 1 − Q 2 . Since Λ P 1 , Q 1 ♯ ( ε f ⃗ ) = Λ P 2 , Q 2 ♯ ( ε f ⃗ ) \Lambda^\sharp_{\mathbb{P}^{1},\mathbb{Q}^{1}}(\varepsilon\vec{f})=\Lambda^\sharp_{\mathbb{P}^{2},\mathbb{Q}^{2}}(\varepsilon\vec{f}) Λ P 1 , Q 1 ♯ ( ε f ) = Λ P 2 , Q 2 ♯ ( ε f ) , the third-order linearization of the partial DN map leads to ∂ ν w ⃗ 12 ∣ ( 0 , T ) × Γ = 0 ⃗ \partial_\nu\vec{w}^{12}|_{(0,T)\times\Gamma}=\vec{0} ∂ ν w 12 ∣ ( 0 , T ) × Γ = 0 . Set w ⃗ ~ : = Ξ w ⃗ 12 \widetilde{\vec{w}}:=\Xi\vec{w}^{12} w := Ξ w 12 , then w ⃗ ~ \widetilde{\vec{w}} w solves the IBVP
{ □ w ⃗ ~ + P w ⃗ ~ = − Ξ b ⃗ 12 − ( Δ Ξ ) w ⃗ − 2 ∇ Ξ ⋅ ∇ w ⃗ , ( t , x ) ∈ Ω T , w ⃗ ~ ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , w ⃗ ~ ( 0 , x ) = ∂ t w ⃗ ~ ( 0 , x ) = 0 ⃗ , x ∈ Ω , (44) \begin{cases}
\Box\widetilde{\vec{w}}+\mathbb{P}\widetilde{\vec{w}}=-\Xi\vec{b}^{12}-(\Delta\Xi)\vec{w}-2\nabla\Xi\cdot\nabla\vec{w}, & (t,x)\in\Omega_T,\\
\widetilde{\vec{w}}(t,x)=\vec{0}, & (t,x)\in\partial\Omega_T,\\
\widetilde{\vec{w}}(0,x)=\partial_t\widetilde{\vec{w}}(0,x)=\vec{0}, & x\in\Omega, \tag*{(44)}
\end{cases} ⎩ ⎨ ⎧ □ w + P w = − Ξ b 12 − ( ΔΞ ) w − 2∇Ξ ⋅ ∇ w , w ( t , x ) = 0 , w ( 0 , x ) = ∂ t w ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω , (44)
Since Q 1 = Q 2 \mathbb{Q}^{1}=\mathbb{Q}^{2} Q 1 = Q 2 in U \mathcal{U} U , Ξ q i j 12 = q i j 12 \Xi q_{ij}^{12}=q_{ij}^{12} Ξ q ij 12 = q ij 12 in Ω ‾ \overline{\Omega} Ω for every i , j ∈ { 1 , 2 } i,j\in\{1,2\} i , j ∈ { 1 , 2 } . Also, we have − ( Δ Ξ ) w ⃗ − 2 ∇ Ξ ⋅ ∇ w ⃗ = 0 ⃗ -(\Delta\Xi)\vec{w}-2\nabla\Xi\cdot\nabla\vec{w}=\vec{0} − ( ΔΞ ) w − 2∇Ξ ⋅ ∇ w = 0 in ( 0 , T ) × ( U 3 ∪ ( Ω ∖ U ‾ 2 ) ) (0,T)\times(\mathcal{U}_{3}\cup(\Omega\setminus\overline{\mathcal{U}}_{2})) ( 0 , T ) × ( U 3 ∪ ( Ω ∖ U 2 )) , due to the construction of Ξ \Xi Ξ . As a result, F ⃗ : = − Ξ b ⃗ 12 − ( Δ Ξ ) w ⃗ − 2 ∇ Ξ ⋅ ∇ w ⃗ \vec{F}:=-\Xi\vec{b}^{12}-(\Delta\Xi)\vec{w}-2\nabla\Xi\cdot\nabla\vec{w} F := − Ξ b 12 − ( ΔΞ ) w − 2∇Ξ ⋅ ∇ w is supported in ( 0 , T ) × ( Ω ∖ U ) (0,T) \times(\Omega\setminus\mathcal{U}) ( 0 , T ) × ( Ω ∖ U ) . From w ⃗ ≡ 0 ⃗ \vec{w} \equiv\vec{0} w ≡ 0 in U 3 \mathcal{U}_3 U 3 , we have ∂ ν w ⃗ ∣ ∂ Ω T = 0 ⃗ \partial_\nu\vec{w}|_{\partial\Omega_T} = \vec{0} ∂ ν w ∣ ∂ Ω T = 0 . Now using Corollary 4.2, we have w ⃗ = 0 ⃗ \vec{w} = \vec{0} w = 0 in ( ( 0 , T ′ ) × ( U 2 ∖ U 3 ) ) ((0,T') \times(\mathcal{U}_2 \setminus\mathcal{U}_3)) (( 0 , T ′ ) × ( U 2 ∖ U 3 )) . Next, we multiply (44) with w ⃗ 0 \vec{w}_0 w 0 , where w ⃗ 0 \vec{w}_0 w 0 solves
{ □ w ⃗ 0 + P T w ⃗ 0 = 0 ⃗ , ( t , x ) ∈ Ω T ′ , w ⃗ 0 ( T ′ , x ) = ∂ t w ⃗ 0 ( T ′ , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\square\vec{w}_0 + P^{\mathsf{T}}\vec{w}_0 = \vec{0}, & (t,x) \in\Omega_{T'},\\
\vec{w}_0(T',x) = \partial_t\vec{w}_0(T',x) = \vec{0}, & x \in\Omega.
\end{cases} { □ w 0 + P T w 0 = 0 , w 0 ( T ′ , x ) = ∂ t w 0 ( T ′ , x ) = 0 , ( t , x ) ∈ Ω T ′ , x ∈ Ω.
and integrate over Ω T ′ \Omega_{T'} Ω T ′ to get
∫ Ω T ′ b ⃗ 12 ⋅ w ⃗ 0 d x d t = 0 ⃗ . \int_{\Omega_{T'}} \vec{b}^{12} \cdot\vec{w}_0\,dxdt = \vec{0}. ∫ Ω T ′ b 12 ⋅ w 0 d x d t = 0 .
Next, we choose (see Proposition 2.3 and 2.4)
v ⃗ ( i ) ( t , x ) = exp ( i λ ϕ ) ( z 0 ( t , x ) K ⃗ + , i + Z 1 , K ⃗ + , i ( t , x ) λ + Z 2 , K ⃗ + , i ( t , x ) λ 2 + ⋯ + Z N , K ⃗ + , i ( t , x ) λ N ) + R ⃗ K ⃗ + , i , \vec{v}^{(i)}(t,x) = \exp(i\lambda\phi)\left(z_0(t,x)\vec{K}_{+,i} + \frac{Z_{1,\vec{K}_{+,i}}(t,x)}{\lambda} + \frac{Z_{2,\vec{K}_{+,i}}(t,x)}{\lambda^2} + \cdots+ \frac{Z_{N,\vec{K}_{+,i}}(t,x)}{\lambda^N}\right) + \vec{R}_{\vec{K}_{+,i}}, v ( i ) ( t , x ) = exp ( iλ ϕ ) ( z 0 ( t , x ) K + , i + λ Z 1 , K + , i ( t , x ) + λ 2 Z 2 , K + , i ( t , x ) + ⋯ + λ N Z N , K + , i ( t , x ) ) + R K + , i ,
and
w ⃗ 0 ( t , x ) = exp ( − 3 i λ ϕ ) ( w 0 ( t , x ) K ⃗ − + W 1 , K ⃗ − ( t , x ) λ + W 2 , K ⃗ − ( t , x ) λ 2 + ⋯ + W N , K ⃗ − ( t , x ) λ N ) + R ⃗ K ⃗ − , \vec{w}_0(t,x) = \exp(-3i\lambda\phi)\left(w_0(t,x)\vec{K}_{-} + \frac{W_{1,\vec{K}_{-}}(t,x)}{\lambda} + \frac{W_{2,\vec{K}_{-}}(t,x)}{\lambda^2} + \cdots+ \frac{W_{N,\vec{K}_{-}}(t,x)}{\lambda^N}\right) + \vec{R}_{\vec{K}_{-}}, w 0 ( t , x ) = exp ( − 3 iλ ϕ ) ( w 0 ( t , x ) K − + λ W 1 , K − ( t , x ) + λ 2 W 2 , K − ( t , x ) + ⋯ + λ N W N , K − ( t , x ) ) + R K − ,
where v ⃗ ( i ) \vec{v}^{(i)} v ( i ) and w ⃗ 0 \vec{w}_0 w 0 are the asymptotic solutions to (16) and (44) , respectively. Here ϕ ∈ C ∞ ( R ℓ ) \phi\in C^\infty(\mathbb{R}^{\ell}) ϕ ∈ C ∞ ( R ℓ ) be arbitrary, z 0 = w 0 = ϕ ( x + t θ ) z_0 = w_0 = \phi(x+t\theta) z 0 = w 0 = ϕ ( x + tθ ) solving the transport equation (12) and (13) . The remainder terms R ⃗ K ⃗ + , i ( t , x , λ ) \vec{R}_{\vec{K}_{+,i}}(t,x,\lambda) R K + , i ( t , x , λ ) for 1 ≤ i ≤ 3 1 \leq i \leq3 1 ≤ i ≤ 3 and R ⃗ K ⃗ − ( t , x , λ ) \vec{R}_{\vec{K}_{-}}(t,x,\lambda) R K − ( t , x , λ ) , satisfies the following estimates:
∥ R ⃗ K ⃗ + , i ∥ L ∞ ( Ω T ) ≤ C λ − 1 and ∥ R ⃗ K ⃗ − ∥ L ∞ ( Ω T ′ ) ≤ C λ − 1 , (45) \|\vec{R}_{\vec{K}_{+,i}}\|_{L^\infty(\Omega_T)} \leq C\lambda^{-1}
\quad\text{and}\quad
\|\vec{R}_{\vec{K}_{-}}\|_{L^\infty(\Omega_{T'})} \leq C\lambda^{-1}, \tag*{(45)} ∥ R K + , i ∥ L ∞ ( Ω T ) ≤ C λ − 1 and ∥ R K − ∥ L ∞ ( Ω T ′ ) ≤ C λ − 1 , (45)
where C > 0 C>0 C > 0 is independent of λ \lambda λ . Also, we have
R ⃗ K ⃗ + , i ( 0 , x , λ ) = ∂ t R ⃗ K ⃗ + , i ( 0 , x , λ ) = 0 ⃗ and R ⃗ K ⃗ − ( T ′ , x , λ ) = ∂ t R ⃗ K ⃗ − ( T ′ , x , λ ) = 0 ⃗ , \vec{R}_{\vec{K}_{+,i}}(0,x,\lambda) = \partial_t\vec{R}_{\vec{K}_{+,i}}(0,x,\lambda) = \vec{0}
\quad\text{and}\quad
\vec{R}_{\vec{K}_{-}}(T',x,\lambda) = \partial_t\vec{R}_{\vec{K}_{-}}(T',x,\lambda) = \vec{0}, R K + , i ( 0 , x , λ ) = ∂ t R K + , i ( 0 , x , λ ) = 0 and R K − ( T ′ , x , λ ) = ∂ t R K − ( T ′ , x , λ ) = 0 ,
for x ∈ Ω x \in\Omega x ∈ Ω and 1 ≤ i ≤ 3 1 \leq i \leq3 1 ≤ i ≤ 3 . From (45) , we observe that
∥ R ⃗ K ⃗ + , i ∥ L ∞ ( Ω T ′ ) ≤ C λ − 1 , \|\vec{R}_{\vec{K}_{+,i}}\|_{L^\infty(\Omega_{T'})} \leq C\lambda^{-1}, ∥ R K + , i ∥ L ∞ ( Ω T ′ ) ≤ C λ − 1 ,
for 1 ≤ i ≤ 3 1 \leq i \leq3 1 ≤ i ≤ 3 . Proceeding as in §3.2, we obtain the unique identification of Q \mathbb{Q} Q , that is, Q 1 = Q 2 \mathbb{Q}^1 = \mathbb{Q}^2 Q 1 = Q 2 in Ω \Omega Ω .
Well-posedness of the forward Klein-Gordon system
In the following, we establish the well-posedness of the forward problem (2) under smallness assumptions on the Dirichlet data. A similar well-posedness result for the scalar equation appears in [6 , 35 , 36 ] . Although the proof relies on quite standard arguments, we collect the proof here for the sake of completeness of the article. We move ahead by defining the function spaces and notation used in the analysis below. By referring [13 ] , for any m ∈ N m \in\mathbb{N} m ∈ N and multi-index β \beta β satisfying ∣ β ∣ ≤ m − 1 |\beta| \leq m-1 ∣ β ∣ ≤ m − 1 , we define the function space
H 0 m ( Ω ) : = { u ∈ H m ( Ω ) : ∂ β u = 0 on ∂ Ω , for ∣ β ∣ ≤ m − 1 } , H^m_0(\Omega) := \left\{u \in H^m(\Omega) : \partial^\beta u = 0 \text{ on } \partial\Omega,\ \text{for } |\beta| \leq m-1\right\}, H 0 m ( Ω ) := { u ∈ H m ( Ω ) : ∂ β u = 0 on ∂ Ω , for ∣ β ∣ ≤ m − 1 } ,
as the completion of ( C ∞ ( Ω ) , ∥ ⋅ ∥ H m ( Ω ) ) (C^\infty(\Omega),\|\cdot\|_{H^m(\Omega)}) ( C ∞ ( Ω ) , ∥ ⋅ ∥ H m ( Ω ) ) with H m H^m H m representing the standard Sobolev space. In the case of vector-valued functions, we define the function space
H 0 m ( Ω ) : = { u ⃗ ∈ H m ( Ω ) : ∂ β u ⃗ = 0 ⃗ on ∂ Ω , for ∣ β ∣ ≤ m − 1 } \mathbf{H}^m_0(\Omega) := \left\{\vec{u} \in\mathbf{H}^m(\Omega) : \partial^\beta\vec{u} = \vec{0} \text{ on } \partial\Omega,\ \text{for } |\beta| \leq m-1\right\} H 0 m ( Ω ) := { u ∈ H m ( Ω ) : ∂ β u = 0 on ∂ Ω , for ∣ β ∣ ≤ m − 1 }
as the completion of ( C c ∞ ( Ω ) , ∥ ⋅ ∥ H m ( Ω ) ) (C^\infty_c(\Omega), \|\cdot\|_{H^m(\Omega)}) ( C c ∞ ( Ω ) , ∥ ⋅ ∥ H m ( Ω ) ) , where ∥ ⋅ ∥ H m ( Ω ) \|\cdot\|_{H^m(\Omega)} ∥ ⋅ ∥ H m ( Ω ) is defined as
∥ u ⃗ ∥ H m ( Ω ) 2 : = ∥ u 1 ∥ H m ( Ω ) 2 + ∥ u 2 ∥ H m ( Ω ) 2 . \|\vec{u}\|_{H^m(\Omega)}^2 := \|u_1\|_{H^m(\Omega)}^2+\|u_2\|_{H^m(\Omega)}^2. ∥ u ∥ H m ( Ω ) 2 := ∥ u 1 ∥ H m ( Ω ) 2 + ∥ u 2 ∥ H m ( Ω ) 2 .
Consider the linear inhomogeneous initial-boundary value problem (IBVP)
{ □ u ( t , x ) = G ( t , x ) , ( t , x ) ∈ Ω T , u ( t , x ) = f ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , x ∈ Ω . (46) \begin{cases}
\Box u(t,x)=G(t,x), & (t,x)\in\Omega_T,\\
u(t,x)=f(t,x), & (t,x)\in\partial\Omega_T,\\
u(0,x)=0,\quad\partial_tu(0,x)=0, & x\in\Omega. \tag*{(46)}
\end{cases} ⎩ ⎨ ⎧ □ u ( t , x ) = G ( t , x ) , u ( t , x ) = f ( t , x ) , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (46)
Let D m D_m D m denote the energy space, which is given by
D m : = ⋂ s = 0 m C s ( [ 0 , T ] ; H m − s ( Ω ) ) , D_m:=\bigcap_{s=0}^{m}C^s([0,T];H^{m-s}(\Omega)), D m := s = 0 ⋂ m C s ([ 0 , T ] ; H m − s ( Ω )) ,
equipped with the norm
∥ u ∥ D m 2 : = sup t ∈ [ 0 , T ] ∑ s = 0 m ∥ ∂ t s u ( t , ⋅ ) ∥ H m − s ( Ω ) 2 . \|u\|_{D_m}^2:=\sup_{t\in[0,T]}\sum_{s=0}^{m}\|\partial_t^s u(t,\cdot)\|_{H^{m-s}(\Omega)}^2. ∥ u ∥ D m 2 := t ∈ [ 0 , T ] sup s = 0 ∑ m ∥ ∂ t s u ( t , ⋅ ) ∥ H m − s ( Ω ) 2 .
By referring to [35 Corollary 2] and [36 Lemma 3.2] , we have the following well-posedness result stated below.
Lemma A.1. [35 Corollary 2] Let T > 0 T>0 T > 0 , m > ℓ + 1 m>\ell+1 m > ℓ + 1 . Then for any f ∈ H m + 1 ( ∂ Ω T ) f\in H^{m+1}(\partial\Omega_T) f ∈ H m + 1 ( ∂ Ω T ) with ∂ t k f ( 0 , ⋅ ) = 0 \partial_t^k f(0,\cdot)=0 ∂ t k f ( 0 , ⋅ ) = 0 on ∂ Ω \partial\Omega ∂ Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m and G ∈ D m + 1 G\in D_{m+1} G ∈ D m + 1 with ∂ t k G ( 0 , ⋅ ) = 0 \partial_t^kG(0,\cdot)=0 ∂ t k G ( 0 , ⋅ ) = 0 in Ω \Omega Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m , the IBVP (A.1) admits a unique solution u ∈ D m + 1 u\in D_{m+1} u ∈ D m + 1 satisfying ∂ ν u ∈ H m ( ∂ Ω T ) \partial_\nu u\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k G ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) , \|u\|_{D_{m+1}}+\|\partial_\nu u\|_{H^m(\partial\Omega_T)}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_t^kG\|_{L^1(0,T;H^{m-k}(\Omega))}+\|f\|_{H^{m+1}(\partial\Omega_T)}\right), ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k G ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) , where C > 0 C>0 C > 0 , depends only on Ω \Omega Ω .
Clearly, for a decoupled linear inhomogeneous wave system
{ □ u ⃗ ( t , x ) = F ⃗ ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (47) \begin{cases}
\Box\vec{u}(t,x)=\vec{F}(t,x), & (t,x)\in\Omega_T,\\
\vec{u}(t,x)=\vec{f}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)=\vec{0},\quad\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega. \tag*{(47)}
\end{cases} ⎩ ⎨ ⎧ □ u ( t , x ) = F ( t , x ) , u ( t , x ) = f ( t , x ) , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (47)
one can extend the above result and obtain
Lemma A.2. Let T > 0 T>0 T > 0 , m > ℓ + 1 m>\ell+1 m > ℓ + 1 . Then for any f ⃗ ∈ H m + 1 ( ∂ Ω T ) \vec{f}\in H^{m+1}(\partial\Omega_T) f ∈ H m + 1 ( ∂ Ω T ) with ∂ t k f ⃗ ( 0 , ⋅ ) = 0 ⃗ \partial_t^k\vec{f}(0,\cdot)=\vec{0} ∂ t k f ( 0 , ⋅ ) = 0 on ∂ Ω \partial\Omega ∂ Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m and F ⃗ ∈ D m + 1 \vec{F}\in D_{m+1} F ∈ D m + 1 with ∂ t k F ⃗ ( 0 , ⋅ ) = 0 ⃗ \partial_t^k\vec{F}(0,\cdot)=\vec{0} ∂ t k F ( 0 , ⋅ ) = 0 in Ω \Omega Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m , the IBVP (A.2) admits a unique solution u ⃗ ∈ D m + 1 \vec{u}\in D_{m+1} u ∈ D m + 1 satisfying ∂ ν u ⃗ ∈ H m ( ∂ Ω T ) \partial_\nu\vec{u}\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k F ⃗ ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ) , \|\vec{u}\|_{D_{m+1}}+\|\partial_\nu\vec{u}\|_{H^m(\partial\Omega_T)}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_t^k\vec{F}\|_{L^1(0,T;H^{m-k}(\Omega))}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_T)}\right), ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k F ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) , where C C C is a positive constant, depends only on Ω \Omega Ω and energy space D m D_m D m is given by
D m : = ⋂ s = 0 m C s ( [ 0 , T ] ; H m − s ( Ω ) ) , D_m:=\bigcap_{s=0}^{m}C^s([0,T];H^{m-s}(\Omega)), D m := s = 0 ⋂ m C s ([ 0 , T ] ; H m − s ( Ω )) , endowed with the norm
∥ u ⃗ ∥ D m 2 : = sup t ∈ [ 0 , T ] ∑ s = 0 m ∑ i = 1 2 ∥ ∂ t s u i ( t , ⋅ ) ∥ H m − s ( Ω ) 2 . \|\vec{u}\|_{D_m}^2:=\sup_{t\in[0,T]}\sum_{s=0}^{m}\sum_{i=1}^{2}\|\partial_t^s u_i(t,\cdot)\|_{H^{m-s}(\Omega)}^2. ∥ u ∥ D m 2 := t ∈ [ 0 , T ] sup s = 0 ∑ m i = 1 ∑ 2 ∥ ∂ t s u i ( t , ⋅ ) ∥ H m − s ( Ω ) 2 .
Next, we introduce a zeroth-order perturbation to (A.2) and seek the well-posedness of the IBVP
{ L P u ⃗ ( t , x ) = F ⃗ ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (48) \left\{
\begin{aligned}
\mathcal{L}_{\mathbb{P}}\vec{u}(t,x)&=\vec{F}(t,x), &&(t,x)\in\Omega_T,\\
\vec{u}(t,x)&=\vec{f}(t,x), &&(t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)&=\partial_t\vec{u}(0,x)=\vec{0}, &&x\in\Omega.
\end{aligned}
\right. \tag*{(48)} ⎩ ⎨ ⎧ L P u ( t , x ) u ( t , x ) u ( 0 , x ) = F ( t , x ) , = f ( t , x ) , = ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (48)
which is established in the proposition given below.
Proposition A.3. In addition to notations and assumption of Lemma A.2, we further assume that P = { p i j } 1 ≤ i , j ≤ 2 ∈ C c ∞ ( Ω ) \mathbb{P}=\{p_{ij}\}_{1\leq i,j\leq2}\in C^\infty_c(\Omega) P = { p ij } 1 ≤ i , j ≤ 2 ∈ C c ∞ ( Ω ) . Then the IBVP (A.3) admits a unique solution u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 satisfying ∂ ν u ⃗ ∈ H m ( ∂ Ω T ) \partial_\nu\vec{u}\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k F ⃗ ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ) , \|\vec{u}\|_{\mathcal{D}_{m+1}}+\|\partial_\nu\vec{u}\|_{H^m(\partial\Omega_T)}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_t^k\vec{F}\|_{L^1(0,T;H^{m-k}(\Omega))}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_T)}\right), ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k F ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) , where C > 0 C>0 C > 0 , depends only on Ω \Omega Ω and P \mathbb{P} P .
Proof. For a given v ⃗ ∈ D m + 1 \vec{v}\in\mathcal{D}_{m+1} v ∈ D m + 1 with ∂ t k v ⃗ ( 0 , ⋅ ) = 0 \partial_t^k\vec{v}(0,\cdot)=0 ∂ t k v ( 0 , ⋅ ) = 0 in Ω \Omega Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m , we consider the IBVP
{ □ u ⃗ ( t , x ) = F ⃗ ( t , x ) − P ( x ) v ⃗ ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (49) \left\{
\begin{aligned}
\square\vec{u}(t,x)&=\vec{F}(t,x)-\mathbb{P}(x)\vec{v}(t,x), &&(t,x)\in\Omega_T,\\
\vec{u}(t,x)&=\vec{f}(t,x), &&(t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)&=\vec{0},\quad\partial_t\vec{u}(0,x)=\vec{0}, &&x\in\Omega.
\end{aligned}
\right. \tag*{(49)} ⎩ ⎨ ⎧ □ u ( t , x ) u ( t , x ) u ( 0 , x ) = F ( t , x ) − P ( x ) v ( t , x ) , = f ( t , x ) , = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (49)
The idea is to use the contraction mapping theorem to establish local existence of the solution. We can obtain global existence by rescaling with respect to the time variable. For m > ℓ + 1 m>\ell+1 m > ℓ + 1 , D m + 1 \mathcal{D}_{m+1} D m + 1 is an Banach algebra (see [7 Definition 3.5] ). Given P ∈ C c ∞ ( Ω ) \mathbb{P}\in C^\infty_c(\Omega) P ∈ C c ∞ ( Ω ) and F ⃗ ∈ D m + 1 \vec{F}\in\mathcal{D}_{m+1} F ∈ D m + 1 with ∂ t k F ⃗ ( 0 , ⋅ ) = 0 ⃗ \partial_t^k\vec{F}(0,\cdot)=\vec{0} ∂ t k F ( 0 , ⋅ ) = 0 in Ω \Omega Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m , we have F ⃗ − P v ⃗ ∈ D m + 1 \vec{F}-\mathbb{P}\vec{v}\in\mathcal{D}_{m+1} F − P v ∈ D m + 1 . Although, to apply Lemma A.2, we further need to establish that ∂ t k ( F ⃗ ( 0 , ⋅ ) − P v ⃗ ( 0 , ⋅ ) ) = 0 ⃗ \partial_t^k(\vec{F}(0,\cdot)-\mathbb{P}\vec{v}(0,\cdot))=\vec{0} ∂ t k ( F ( 0 , ⋅ ) − P v ( 0 , ⋅ )) = 0 in Ω \Omega Ω for k = 0 , 1 , … , m k=0,1,\ldots,m k = 0 , 1 , … , m , which holds trivially due to choice of v ⃗ \vec{v} v . Thus the inhomogeneous IBVP (A.4) admits a unique solution u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 with ∂ ν u ⃗ ∈ H m ( ∂ Ω T ) \partial_\nu\vec{u}\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k ( F ⃗ − P v ⃗ ) ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ) . (50) \|\vec{u}\|_{\mathcal{D}_{m+1}}+\|\partial_\nu\vec{u}\|_{H^m(\partial\Omega_T)}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_t^k(\vec{F}-\mathbb{P}\vec{v})\|_{L^1(0,T;H^{m-k}(\Omega))}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_T)}\right). \tag*{(50)} ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k ( F − P v ) ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) . (50)
Set
N = { v ⃗ ∈ D m + 1 : ∂ t k v ⃗ ( 0 , ⋅ ) = 0 , k = 0 , 1 , … , m } . \mathcal{N}=\{\vec{v}\in\mathcal{D}_{m+1}:\partial_t^k\vec{v}(0,\cdot)=0,\ k=0,1,\ldots,m\}. N = { v ∈ D m + 1 : ∂ t k v ( 0 , ⋅ ) = 0 , k = 0 , 1 , … , m } .
Define a map T : N → N \mathcal{T}:\mathcal{N}\to\mathcal{N} T : N → N mapping
T ( v ⃗ ) = u ⃗ , \mathcal{T}(\vec{v})=\vec{u}, T ( v ) = u ,
where u ⃗ \vec{u} u is the solution to the IBVP (A.4) corresponds to v ⃗ ∈ N \vec{v}\in\mathcal{N} v ∈ N . Although u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 , the compatibility condition of (A.4) forces u ⃗ \vec{u} u to lie in N \mathcal{N} N , hence, T \mathcal{T} T is well-defined. As a next step, we establish that T \mathcal{T} T is a contraction map. To this end, let u ⃗ i \vec{u}^{i} u i be the solution corresponds to the IBVP
{ □ u ⃗ i ( t , x ) = F ⃗ ( t , x ) − P ( x ) v ⃗ i ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ i ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ i ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ i ( 0 , x ) = 0 ⃗ , x ∈ Ω . \left\{
\begin{aligned}
\square\vec{u}^{i}(t,x)&=\vec{F}(t,x)-\mathbb{P}(x)\vec{v}^{i}(t,x), &&(t,x)\in\Omega_T,\\
\vec{u}^{i}(t,x)&=\vec{f}(t,x), &&(t,x)\in\partial\Omega_T,\\
\vec{u}^{i}(0,x)&=\vec{0},\quad\partial_t\vec{u}^{i}(0,x)=\vec{0}, &&x\in\Omega.
\end{aligned}
\right. ⎩ ⎨ ⎧ □ u i ( t , x ) u i ( t , x ) u i ( 0 , x ) = F ( t , x ) − P ( x ) v i ( t , x ) , = f ( t , x ) , = 0 , ∂ t u i ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω.
for i = 1 , 2 i=1,2 i = 1 , 2 , and v ⃗ i ∈ N \vec{v}^{i}\in\mathcal{N} v i ∈ N . The subtraction then leads us to
{ □ ( w ⃗ 1 − w ⃗ 2 ) ( t , x ) = − P ( x ) ( v ⃗ 1 − v ⃗ 2 ) ( t , x ) , ( t , x ) ∈ Ω T , ( w ⃗ 1 − w ⃗ 2 ) ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , ( w ⃗ 1 − w ⃗ 2 ) ( 0 , x ) = 0 ⃗ , ∂ t ( w ⃗ 1 − w ⃗ 2 ) ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\Box(\vec{w}^{1}-\vec{w}^{2})(t,x)=-\mathbb{P}(x)(\vec{v}^{1}-\vec{v}^{2})(t,x), & (t,x)\in\Omega_{T},\\
(\vec{w}^{1}-\vec{w}^{2})(t,x)=\vec{0}, & (t,x)\in\partial\Omega_{T},\\
(\vec{w}^{1}-\vec{w}^{2})(0,x)=\vec{0},\quad\partial_{t}(\vec{w}^{1}-\vec{w}^{2})(0,x)=\vec{0}, & x\in\Omega,
\end{cases} ⎩ ⎨ ⎧ □ ( w 1 − w 2 ) ( t , x ) = − P ( x ) ( v 1 − v 2 ) ( t , x ) , ( w 1 − w 2 ) ( t , x ) = 0 , ( w 1 − w 2 ) ( 0 , x ) = 0 , ∂ t ( w 1 − w 2 ) ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω ,
and the energy estimate
∥ w ⃗ 1 − w ⃗ 2 ∥ D m + 1 + ∥ ∂ ν ( w ⃗ 1 − w ⃗ 2 ) ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ P ∂ t k ( v ⃗ 1 − v ⃗ 2 ) ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) ) ≤ C T exp ( C T ) ( ∑ k = 0 m ∥ P ∂ t k ( v ⃗ 1 − v ⃗ 2 ) ∥ C ( 0 , T ; H m − k ( Ω ) ) ) ≤ C C 1 T exp ( C T ) ∥ P ∥ H m ( Ω ) ∥ ( v ⃗ 1 − v ⃗ 2 ) ∥ D m + 1 , \begin{aligned}
\|\vec{w}^{1}-\vec{w}^{2}\|_{\mathcal{D}_{m+1}}+\|\partial_{\nu}(\vec{w}^{1}-\vec{w}^{2})\|_{H^{m}(\partial\Omega_{T})}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\mathbb{P}\partial_{t}^{k}(\vec{v}^{1}-\vec{v}^{2})\|_{L^{1}(0,T;H^{m-k}(\Omega))}\right) \\
&\leq CT\exp(CT)\left(\sum_{k=0}^{m}\|\mathbb{P}\partial_{t}^{k}(\vec{v}^{1}-\vec{v}^{2})\|_{C(0,T;H^{m-k}(\Omega))}\right) \\
&\leq CC_{1}T\exp(CT)\|\mathbb{P}\|_{H^{m}(\Omega)}\|(\vec{v}^{1}-\vec{v}^{2})\|_{\mathcal{D}_{m+1}},
\end{aligned} ∥ w 1 − w 2 ∥ D m + 1 + ∥ ∂ ν ( w 1 − w 2 ) ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ P ∂ t k ( v 1 − v 2 ) ∥ L 1 ( 0 , T ; H m − k ( Ω )) ) ≤ C T exp ( C T ) ( k = 0 ∑ m ∥ P ∂ t k ( v 1 − v 2 ) ∥ C ( 0 , T ; H m − k ( Ω )) ) ≤ C C 1 T exp ( C T ) ∥ P ∥ H m ( Ω ) ∥ ( v 1 − v 2 ) ∥ D m + 1 ,
by utilizing the estimate ∥ w ⃗ ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) ≤ T ∥ w ⃗ ∥ C ( 0 , T ; H m − k ( Ω ) ) \|\vec{w}\|_{L^{1}(0,T;H^{m-k}(\Omega))}\leq T\|\vec{w}\|_{C(0,T;H^{m-k}(\Omega))} ∥ w ∥ L 1 ( 0 , T ; H m − k ( Ω )) ≤ T ∥ w ∥ C ( 0 , T ; H m − k ( Ω )) . For a sufficient small T T T , say T = T ∗ T=T^{*} T = T ∗ , we have C C 1 T ∗ ∥ P ∥ H m ( Ω ) exp ( C T ∗ ) < 1 CC_{1}T^{*}\|\mathbb{P}\|_{H^{m}(\Omega)}\exp(CT^{*})<1 C C 1 T ∗ ∥ P ∥ H m ( Ω ) exp ( C T ∗ ) < 1 , resulting in
∥ u ⃗ 1 − u ⃗ 2 ∥ D m + 1 ( [ 0 , T ∗ ] ) < ∥ ( v ⃗ 1 − v ⃗ 2 ) ∥ D m + 1 ( [ 0 , T ∗ ] ) . \|\vec{u}^{1}-\vec{u}^{2}\|_{\mathcal{D}_{m+1}([0,T^{*}])}<\|(\vec{v}^{1}-\vec{v}^{2})\|_{\mathcal{D}_{m+1}([0,T^{*}])}. ∥ u 1 − u 2 ∥ D m + 1 ([ 0 , T ∗ ]) < ∥ ( v 1 − v 2 ) ∥ D m + 1 ([ 0 , T ∗ ]) .
where [ 0 , T ∗ ] [0,T^{*}] [ 0 , T ∗ ] with D m \mathcal{D}_{m} D m indicated that the initial and final time are 0 0 0 and T ∗ T^{*} T ∗ respectively in the aforementioned definition of D m \mathcal{D}_{m} D m . Thus T \mathcal{T} T is a contraction map, and by using the Banach fixed point theorem, there exists a unique fixed point v ⃗ ( 0 ) ∈ N \vec{v}^{(0)}\in\mathcal{N} v ( 0 ) ∈ N , that is T ( v ⃗ ( 0 ) ) = v ⃗ ( 0 ) = u ⃗ \mathcal{T}(\vec{v}^{(0)})=\vec{v}^{(0)}=\vec{u} T ( v ( 0 ) ) = v ( 0 ) = u . As a result, (A.3) admits a unique solution whenever 0 ≤ t ≤ T ∗ 0\leq t\leq T^{*} 0 ≤ t ≤ T ∗ . Moreover, using the fixed point v ⃗ \vec{v} v , and the estimate (A.5), we have
∥ u ⃗ ∥ D m + 1 ( [ 0 , T ∗ ] ) + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ∗ ) ≤ C exp ( C T ∗ ) ( ∑ k = 0 m ∥ ∂ t k ( F ⃗ − P u ⃗ ) ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ∗ ) ) ≤ C exp ( C T ∗ ) ( ∑ k = 0 m ∥ ∂ t k F ⃗ ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω ) ) + C 1 T ∗ ∥ P ∥ H m ( Ω ) ∥ u ⃗ ∥ D m + 1 ( [ 0 , T ∗ ] ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ∗ ) ) . \begin{aligned}
\|\vec{u}\|_{\mathcal{D}_{m+1}([0,T^{*}])}+\|\partial_{\nu}\vec{u}\|_{H^{m}(\partial\Omega_{T^{*}})}
\leq C\exp(CT^{*})\left(\sum_{k=0}^{m}\|\partial_{t}^{k}(\vec{F}-\mathbb{P}\vec{u})\|_{L^{1}(0,T^{*};H^{m-k}(\Omega))}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_{T^{*}})}\right) \\
&\leq C\exp(CT^{*})\left(\sum_{k=0}^{m}\|\partial_{t}^{k}\vec{F}\|_{L^{1}(0,T^{*};H^{m-k}(\Omega))}+C_{1}T^{*}\|\mathbb{P}\|_{H^{m}(\Omega)}\|\vec{u}\|_{\mathcal{D}_{m+1}([0,T^{*}])}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_{T^{*}})}\right).
\end{aligned} ∥ u ∥ D m + 1 ([ 0 , T ∗ ]) + ∥ ∂ ν u ∥ H m ( ∂ Ω T ∗ ) ≤ C exp ( C T ∗ ) ( k = 0 ∑ m ∥ ∂ t k ( F − P u ) ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ∗ ) ) ≤ C exp ( C T ∗ ) ( k = 0 ∑ m ∥ ∂ t k F ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω )) + C 1 T ∗ ∥ P ∥ H m ( Ω ) ∥ u ∥ D m + 1 ([ 0 , T ∗ ]) + ∥ f ∥ H m + 1 ( ∂ Ω T ∗ ) ) .
From the choice of T ∗ T^{*} T ∗ , the aforementioned inequality boils down to
∥ u ⃗ ∥ D m + 1 ( [ 0 , T ∗ ] ) + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ∗ ) ≤ C exp ( C T ∗ ) ( ∑ k = 0 m ∥ ∂ t k F ⃗ ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ∗ ) ) . \|\vec{u}\|_{\mathcal{D}_{m+1}([0,T^{*}])}+\|\partial_{\nu}\vec{u}\|_{H^{m}(\partial\Omega_{T^{*}})}
\leq C\exp(CT^{*})\left(\sum_{k=0}^{m}\|\partial_{t}^{k}\vec{F}\|_{L^{1}(0,T^{*};H^{m-k}(\Omega))}+\|\vec{f}\|_{H^{m+1}(\partial\Omega_{T^{*}})}\right). ∥ u ∥ D m + 1 ([ 0 , T ∗ ]) + ∥ ∂ ν u ∥ H m ( ∂ Ω T ∗ ) ≤ C exp ( C T ∗ ) ( k = 0 ∑ m ∥ ∂ t k F ∥ L 1 ( 0 , T ∗ ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ∗ ) ) .
Since the IBVP (A.3) is linear, rescaling in the time variable (see [6 Lemma 3.2] ) completes the proof. □
Next, we introduce the small Dirichlet data f ⃗ ∈ F m + 1 σ \vec{f}\in F_{m+1}^{\sigma} f ∈ F m + 1 σ , where
F m + 1 σ : = { f ⃗ ∈ H m + 1 ( ∂ Ω T ) : ∂ t k f ⃗ ( 0 , ⋅ ) = 0 ⃗ on ∂ Ω and ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) < σ , for k = 0 , 1 , … , m } . (51) F_{m+1}^{\sigma}:=\left\{\vec{f}\in H^{m+1}(\partial\Omega_{T}):\partial_{t}^{k}\vec{f}(0,\cdot)=\vec{0}\text{ on }\partial\Omega\text{ and }\|\vec{f}\|_{H^{m+1}(\partial\Omega_{T})}<\sigma,\text{ for }k=0,1,\ldots,m\right\}. \tag*{(51)} F m + 1 σ := { f ∈ H m + 1 ( ∂ Ω T ) : ∂ t k f ( 0 , ⋅ ) = 0 on ∂ Ω and ∥ f ∥ H m + 1 ( ∂ Ω T ) < σ , for k = 0 , 1 , … , m } . (51)
with σ > 0 \sigma>0 σ > 0 chosen sufficiently small. Further, we denote
B R ( 0 ) : = { u ⃗ ∈ D m + 1 : ∥ u ⃗ ∥ D m + 1 < R } B_{R}(0):=\left\{\vec{u}\in\mathcal{D}_{m+1}:\|\vec{u}\|_{\mathcal{D}_{m+1}}<R\right\} B R ( 0 ) := { u ∈ D m + 1 : ∥ u ∥ D m + 1 < R }
for some R > 0 R>0 R > 0 .
Using the preceding analysis, we establish a unique solution u ⃗ ∈ B R ( 0 ) \vec{u}\in B_{R}(0) u ∈ B R ( 0 ) for the IBVP (2) , whenever f ⃗ ∈ F m + 1 σ \vec{f}\in F_{m+1}^{\sigma} f ∈ F m + 1 σ , which is stated formally below.
Theorem A.4. Let T > 0 T>0 T > 0 , m > ℓ + 1 m>\ell+1 m > ℓ + 1 , ( P , Q ) ∈ C c ∞ ( Ω ) × C c ∞ ( Ω ) (\mathbb{P},\mathbb{Q})\in C^\infty_c(\Omega)\times C^\infty_c(\Omega) ( P , Q ) ∈ C c ∞ ( Ω ) × C c ∞ ( Ω ) , and f ⃗ ∈ F m + 1 \vec{f}\in\mathbb{F}^{m+1} f ∈ F m + 1 . Then the IBVP (2) admits a unique solution u ⃗ ∈ B R ( 0 ) \vec{u}\in\mathbb{B}_R(0) u ∈ B R ( 0 ) satisfying ∂ ν u ⃗ ∈ H m ( ∂ Ω T ) \partial_\nu\vec{u}\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) , \lVert\vec{u}\rVert_{\mathcal{D}_{m+1}}+\lVert\partial_\nu\vec{u}\rVert_{H^m(\partial\Omega_T)}\le C\exp(CT)\lVert\vec{f}\rVert_{H^{m+1}(\partial\Omega_T)}, ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ∥ f ∥ H m + 1 ( ∂ Ω T ) , where C > 0 C>0 C > 0 , depends only on Ω \Omega Ω , P \mathbb{P} P and Q \mathbb{Q} Q .
Proof. We begin by introducing a closed subset of D m + 1 \mathcal{D}_{m+1} D m + 1 which is given by
K : = { v ⃗ ∈ B R ( 0 ) : ∂ t k v ⃗ ( 0 , ⋅ ) = 0 , k = 0 , 1 , ⋯ , m } , \mathcal{K}:=\left\{\vec{v}\in\mathbb{B}_R(0):\partial_t^k\vec{v}(0,\cdot)=0,\ k=0,1,\cdots,m\right\}, K := { v ∈ B R ( 0 ) : ∂ t k v ( 0 , ⋅ ) = 0 , k = 0 , 1 , ⋯ , m } ,
where R > 0 R>0 R > 0 is specified later. Given v ⃗ ∈ K \vec{v}\in\mathcal{K} v ∈ K , we consider the IBVP
{ □ u ⃗ ( t , x ) + P ( x ) u ⃗ ( t , x ) = − ( Q v ⃗ ) ∘ ( v ⃗ ∘ v ⃗ ) ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ ( 0 , x ) = 0 ⃗ , x ∈ Ω . (52) \begin{cases}
\Box\vec{u}(t,x)+\mathbb{P}(x)\vec{u}(t,x)=-(\mathbb{Q}\vec{v})\circ(\vec{v}\circ\vec{v})(t,x), & (t,x)\in\Omega_T,\\
\vec{u}(t,x)=\vec{f}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{u}(0,x)=\vec{0},\quad\partial_t\vec{u}(0,x)=\vec{0}, & x\in\Omega. \tag*{(52)}
\end{cases} ⎩ ⎨ ⎧ □ u ( t , x ) + P ( x ) u ( t , x ) = − ( Q v ) ∘ ( v ∘ v ) ( t , x ) , u ( t , x ) = f ( t , x ) , u ( 0 , x ) = 0 , ∂ t u ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω. (52)
Since v ⃗ ∈ K \vec{v}\in\mathcal{K} v ∈ K and P , Q ∈ C c ∞ ( Ω ) \mathbb{P},\mathbb{Q}\in C^\infty_c(\Omega) P , Q ∈ C c ∞ ( Ω ) , it follows from Proposition A.3 that the IBVP (A.7) admits a unique solution u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 with ∂ ν u ⃗ ∈ H m ( ∂ Ω T ) \partial_\nu\vec{u}\in H^m(\partial\Omega_T) ∂ ν u ∈ H m ( ∂ Ω T ) and
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k ( Q v ⃗ ) ∘ ( v ⃗ ∘ v ⃗ ) ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ) . \lVert\vec{u}\rVert_{\mathcal{D}_{m+1}}+\lVert\partial_\nu\vec{u}\rVert_{H^m(\partial\Omega_T)}
\le C\exp(CT)\left(\sum_{k=0}^{m}\lVert\partial_t^k(\mathbb{Q}\vec{v})\circ(\vec{v}\circ\vec{v})\rVert_{L^1(0,T;H^{m-k}(\Omega))}+\lVert\vec{f}\rVert_{H^{m+1}(\partial\Omega_T)}\right). ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k ( Q v ) ∘ ( v ∘ v ) ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) .
where C > 0 C>0 C > 0 , depends only on Ω \Omega Ω , P \mathbb{P} P and Q \mathbb{Q} Q . As discussed earlier, we now define a map T : K → K \mathcal{T}:\mathcal{K}\to\mathcal{K} T : K → K , which is a contraction. For a given v ⃗ ∈ K \vec{v}\in\mathcal{K} v ∈ K , let T v ⃗ = u ⃗ \mathcal{T}\vec{v}=\vec{u} T v = u , where u ⃗ \vec{u} u is a solution to the IBVP (A.7). However, the well-posedness result implies that u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 . So, to ensure that the map T \mathcal{T} T is a contraction from K \mathcal{K} K to itself, we need to impose some conditions on R R R and σ \sigma σ . From the above energy estimate, we have
∥ T v ⃗ ∥ D m + 1 = ∥ u ⃗ ∥ D m + 1 ≤ C exp ( C T ) ( C 1 R 3 ∥ Q ∥ H m ( Ω ) + σ ) . (53) \lVert\mathcal{T}\vec{v}\rVert_{\mathcal{D}_{m+1}}=\lVert\vec{u}\rVert_{\mathcal{D}_{m+1}}\le C\exp(CT)\left(C_1R^3\lVert\mathbb{Q}\rVert_{H^m(\Omega)}+\sigma\right). \tag*{(53)} ∥ T v ∥ D m + 1 = ∥ u ∥ D m + 1 ≤ C exp ( C T ) ( C 1 R 3 ∥ Q ∥ H m ( Ω ) + σ ) . (53)
To ensure u ⃗ ∈ K \vec{u}\in\mathcal{K} u ∈ K , we need to have ∥ u ⃗ ∥ D m + 1 ≤ R \lVert\vec{u}\rVert_{\mathcal{D}_{m+1}}\le R ∥ u ∥ D m + 1 ≤ R . To make this happen, we choose
R < 1 3 C C 1 ∥ Q ∥ H m ( Ω ) exp ( C T ) and σ < R 3 C exp ( C T ) . R<\frac{1}{\sqrt{3CC_1\lVert\mathbb{Q}\rVert_{H^m(\Omega)}\exp(CT)}}\quad\text{and}\quad\sigma<\frac{R}{3C\exp(CT)}. R < 3 C C 1 ∥ Q ∥ H m ( Ω ) exp ( C T ) 1 and σ < 3 C exp ( C T ) R .
Using the above choices of R R R and σ \sigma σ in (A.8) leads to
∥ T v ⃗ ∥ D m + 1 ≤ 2 3 R < R . \lVert\mathcal{T}\vec{v}\rVert_{\mathcal{D}_{m+1}}\le\frac{2}{3}R<R. ∥ T v ∥ D m + 1 ≤ 3 2 R < R .
From now on, we will work only with the aforementioned choice of R R R and σ \sigma σ . Next, we set u ⃗ i : = T v ⃗ i \vec{u}^{i}:=\mathcal{T}\vec{v}^{i} u i := T v i for v ⃗ i ∈ K \vec{v}^{i}\in\mathcal{K} v i ∈ K and let u ⃗ i \vec{u}^{i} u i be the solution corresponds to the IBVP
{ □ u ⃗ i ( t , x ) + P ( x ) u ⃗ i ( t , x ) = − ( Q v ⃗ i ) ∘ ( v ⃗ i ∘ v ⃗ i ) ( t , x ) , ( t , x ) ∈ Ω T , u ⃗ i ( t , x ) = f ⃗ ( t , x ) , ( t , x ) ∈ ∂ Ω T , u ⃗ i ( 0 , x ) = 0 ⃗ , ∂ t u ⃗ i ( 0 , x ) = 0 ⃗ , x ∈ Ω . \begin{cases}
\Box\vec{u}^{i}(t,x)+\mathbb{P}(x)\vec{u}^{i}(t,x)=-(\mathbb{Q}\vec{v}^{i})\circ(\vec{v}^{i}\circ\vec{v}^{i})(t,x), & (t,x)\in\Omega_T,\\
\vec{u}^{i}(t,x)=\vec{f}(t,x), & (t,x)\in\partial\Omega_T,\\
\vec{u}^{i}(0,x)=\vec{0},\quad\partial_t\vec{u}^{i}(0,x)=\vec{0}, & x\in\Omega.
\end{cases} ⎩ ⎨ ⎧ □ u i ( t , x ) + P ( x ) u i ( t , x ) = − ( Q v i ) ∘ ( v i ∘ v i ) ( t , x ) , u i ( t , x ) = f ( t , x ) , u i ( 0 , x ) = 0 , ∂ t u i ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω.
for i = 1 , 2 i=1,2 i = 1 , 2 . The subtraction then leads us to
{ □ ( w ⃗ 1 − w ⃗ 2 ) + P ( w ⃗ 1 − w ⃗ 2 ) = − ( Q w ⃗ 1 ) ∘ ( v ⃗ 1 ∘ v ⃗ 1 ) + ( Q w ⃗ 2 ) ∘ ( v ⃗ 2 ∘ v ⃗ 2 ) , ( t , x ) ∈ Ω T , ( w ⃗ 1 − w ⃗ 2 ) ( t , x ) = 0 ⃗ , ( t , x ) ∈ ∂ Ω T , ( w ⃗ 1 − w ⃗ 2 ) ( 0 , x ) = 0 ⃗ , ∂ t ( w ⃗ 1 − w ⃗ 2 ) ( 0 , x ) = 0 ⃗ , x ∈ Ω , \begin{cases}
\Box(\vec{w}^{1}-\vec{w}^{2})+\mathrm{P}(\vec{w}^{1}-\vec{w}^{2})=-(\mathrm{Q}\vec{w}^{1})\circ(\vec{v}^{1}\circ\vec{v}^{1})+(\mathrm{Q}\vec{w}^{2})\circ(\vec{v}^{2}\circ\vec{v}^{2}), & (t,x)\in\Omega_{T},\\
(\vec{w}^{1}-\vec{w}^{2})(t,x)=\vec{0}, & (t,x)\in\partial\Omega_{T},\\
(\vec{w}^{1}-\vec{w}^{2})(0,x)=\vec{0},\quad\partial_{t}(\vec{w}^{1}-\vec{w}^{2})(0,x)=\vec{0}, & x\in\Omega,
\end{cases} ⎩ ⎨ ⎧ □ ( w 1 − w 2 ) + P ( w 1 − w 2 ) = − ( Q w 1 ) ∘ ( v 1 ∘ v 1 ) + ( Q w 2 ) ∘ ( v 2 ∘ v 2 ) , ( w 1 − w 2 ) ( t , x ) = 0 , ( w 1 − w 2 ) ( 0 , x ) = 0 , ∂ t ( w 1 − w 2 ) ( 0 , x ) = 0 , ( t , x ) ∈ Ω T , ( t , x ) ∈ ∂ Ω T , x ∈ Ω ,
and the energy estimate
∥ w ⃗ 1 − w ⃗ 2 ∥ D m + 1 + ∥ ∂ ν ( w ⃗ 1 − w ⃗ 2 ) ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k [ − ( Q v ⃗ 1 ) ∘ ( v ⃗ 1 ∘ v ⃗ 1 ) + ( Q v ⃗ 2 ) ∘ ( v ⃗ 2 ∘ v ⃗ 2 ) ] ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) ) . (54) \begin{aligned}
\|\vec{w}^{1}-\vec{w}^{2}\|_{\mathcal{D}_{m+1}}+\|\partial_{\nu}(\vec{w}^{1}-\vec{w}^{2})\|_{\mathbf{H}^{m}(\partial\Omega_{T})} \\
&\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_{t}^{k}[-(\mathrm{Q}\vec{v}^{1})\circ(\vec{v}^{1}\circ\vec{v}^{1})+(\mathrm{Q}\vec{v}^{2})\circ(\vec{v}^{2}\circ\vec{v}^{2})]\|_{L^{1}(0,T;\mathbf{H}^{m-k}(\Omega))}\right). \tag*{(54)}
\end{aligned} ∥ w 1 − w 2 ∥ D m + 1 + ∥ ∂ ν ( w 1 − w 2 ) ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k [ − ( Q v 1 ) ∘ ( v 1 ∘ v 1 ) + ( Q v 2 ) ∘ ( v 2 ∘ v 2 )] ∥ L 1 ( 0 , T ; H m − k ( Ω )) ) . (54)
If we denote v ⃗ i : = ( v 1 i , v 2 i ) ⊤ \vec{v}^{i}:=(v_{1}^{i},v_{2}^{i})^{\top} v i := ( v 1 i , v 2 i ) ⊤ then
− ( Q v ⃗ 1 ) ∘ ( v ⃗ 1 ∘ v ⃗ 1 ) + ( Q v ⃗ 2 ) ∘ ( v ⃗ 2 ∘ v ⃗ 2 ) = − ( q 11 ( ( v 1 1 ) 3 − ( v 1 2 ) 3 ) + q 12 ( ( v 1 1 ) 2 ( v 2 1 ) − ( v 1 2 ) 2 ( v 2 2 ) ) q 21 ( ( v 2 1 ) 2 ( v 1 1 ) − ( v 2 2 ) 2 ( v 1 2 ) ) + q 22 ( ( v 2 1 ) 3 − ( v 2 2 ) 3 ) ) = − ( q 11 ( v 1 1 − v 1 2 ) ( ( v 1 1 ) 2 + ( v 2 1 ) 2 + v 1 1 v 1 2 ) + q 12 ( v 1 1 − v 1 2 ) ( v 1 1 + v 1 2 ) v 2 1 + q 12 ( v 2 1 − v 2 2 ) ( v 2 1 ) 2 q 21 ( v 1 1 − v 1 2 ) ( v 2 1 ) 2 + q 21 ( v 2 1 − v 2 2 ) ( v 2 1 + v 2 2 ) ( v 2 1 + v 2 2 ) + q 22 ( v 2 1 − v 2 2 ) ( ( v 2 1 ) 2 + ( v 2 2 ) 2 + v 2 1 v 2 2 ) ) . \begin{aligned}
-(\mathrm{Q}\vec{v}^{1})\circ(\vec{v}^{1}\circ\vec{v}^{1})+(\mathrm{Q}\vec{v}^{2})\circ(\vec{v}^{2}\circ\vec{v}^{2}) \\
&=-\begin{pmatrix}
q_{11}((v_{1}^{1})^{3}-(v_{1}^{2})^{3})+q_{12}((v_{1}^{1})^{2}(v_{2}^{1})-(v_{1}^{2})^{2}(v_{2}^{2}))\\
q_{21}((v_{2}^{1})^{2}(v_{1}^{1})-(v_{2}^{2})^{2}(v_{1}^{2}))+q_{22}((v_{2}^{1})^{3}-(v_{2}^{2})^{3})
\end{pmatrix} \\
&=-\begin{pmatrix}
q_{11}(v_{1}^{1}-v_{1}^{2})((v_{1}^{1})^{2}+(v_{2}^{1})^{2}+v_{1}^{1}v_{1}^{2})+q_{12}(v_{1}^{1}-v_{1}^{2})(v_{1}^{1}+v_{1}^{2})v_{2}^{1}+q_{12}(v_{2}^{1}-v_{2}^{2})(v_{2}^{1})^{2}\\
q_{21}(v_{1}^{1}-v_{1}^{2})(v_{2}^{1})^{2}+q_{21}(v_{2}^{1}-v_{2}^{2})(v_{2}^{1}+v_{2}^{2})(v_{2}^{1}+v_{2}^{2})+q_{22}(v_{2}^{1}-v_{2}^{2})((v_{2}^{1})^{2}+(v_{2}^{2})^{2}+v_{2}^{1}v_{2}^{2})
\end{pmatrix}.
\end{aligned} − ( Q v 1 ) ∘ ( v 1 ∘ v 1 ) + ( Q v 2 ) ∘ ( v 2 ∘ v 2 ) = − ( q 11 (( v 1 1 ) 3 − ( v 1 2 ) 3 ) + q 12 (( v 1 1 ) 2 ( v 2 1 ) − ( v 1 2 ) 2 ( v 2 2 )) q 21 (( v 2 1 ) 2 ( v 1 1 ) − ( v 2 2 ) 2 ( v 1 2 )) + q 22 (( v 2 1 ) 3 − ( v 2 2 ) 3 ) ) = − ( q 11 ( v 1 1 − v 1 2 ) (( v 1 1 ) 2 + ( v 2 1 ) 2 + v 1 1 v 1 2 ) + q 12 ( v 1 1 − v 1 2 ) ( v 1 1 + v 1 2 ) v 2 1 + q 12 ( v 2 1 − v 2 2 ) ( v 2 1 ) 2 q 21 ( v 1 1 − v 1 2 ) ( v 2 1 ) 2 + q 21 ( v 2 1 − v 2 2 ) ( v 2 1 + v 2 2 ) ( v 2 1 + v 2 2 ) + q 22 ( v 2 1 − v 2 2 ) (( v 2 1 ) 2 + ( v 2 2 ) 2 + v 2 1 v 2 2 ) ) .
The use of foregoing expression along with the fact that v ⃗ i ∈ K \vec{v}^{i}\in\mathcal{K} v i ∈ K in (A.9) leads to
∥ T v ⃗ 1 − T v ⃗ 2 ∥ D m + 1 = ∥ w ⃗ 1 − w ⃗ 2 ∥ D m + 1 ≤ C C 1 R 2 ∥ Q ∥ H m ( Ω ) exp ( C T ) ∥ v ⃗ 1 − v ⃗ 2 ∥ D m + 1 . \|\mathcal{T}\vec{v}^{1}-\mathcal{T}\vec{v}^{2}\|_{\mathcal{D}_{m+1}}=\|\vec{w}^{1}-\vec{w}^{2}\|_{\mathcal{D}_{m+1}}\leq CC_{1}R^{2}\|\mathrm{Q}\|_{\mathbf{H}^{m}(\Omega)}\exp(CT)\|\vec{v}^{1}-\vec{v}^{2}\|_{\mathcal{D}_{m+1}}. ∥ T v 1 − T v 2 ∥ D m + 1 = ∥ w 1 − w 2 ∥ D m + 1 ≤ C C 1 R 2 ∥ Q ∥ H m ( Ω ) exp ( C T ) ∥ v 1 − v 2 ∥ D m + 1 .
If we choose
R < 1 2 C C 1 ∥ Q ∥ H m ( Ω ) exp ( C T ) , R<\frac{1}{2\sqrt{CC_{1}\|\mathrm{Q}\|_{\mathbf{H}^{m}(\Omega)}\exp(CT)}}, R < 2 C C 1 ∥ Q ∥ H m ( Ω ) exp ( C T ) 1 ,
then
∥ T v ⃗ 1 − T v ⃗ 2 ∥ D m + 1 < 1 2 ∥ v ⃗ 1 − v ⃗ 2 ∥ D m + 1 . \|\mathcal{T}\vec{v}^{1}-\mathcal{T}\vec{v}^{2}\|_{\mathcal{D}_{m+1}}<\frac{1}{2}\|\vec{v}^{1}-\vec{v}^{2}\|_{\mathcal{D}_{m+1}}. ∥ T v 1 − T v 2 ∥ D m + 1 < 2 1 ∥ v 1 − v 2 ∥ D m + 1 .
Thus, the Banach fixed point theorem ensures the existence of a unique v ⃗ ∈ K \vec{v}\in\mathcal{K} v ∈ K such that T ( v ⃗ ) = v ⃗ \mathcal{T}(\vec{v})=\vec{v} T ( v ) = v .
Thus there exist a unique u ⃗ ∈ D m + 1 \vec{u}\in\mathcal{D}_{m+1} u ∈ D m + 1 solving (2) , such that
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( ∑ k = 0 m ∥ ∂ t k ( Q u ⃗ ) ∘ ( u ⃗ ∘ u ⃗ ) ∥ L 1 ( 0 , T ; H m − k ( Ω ) ) + ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ) ≤ C exp ( C T ) ∥ ( Q u ⃗ ) ∘ ( u ⃗ ∘ u ⃗ ) ∥ D m + C exp ( C T ) ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) \begin{aligned}
\|\vec{u}\|_{\mathcal{D}_{m+1}}+\|\partial_{\nu}\vec{u}\|_{\mathbf{H}^{m}(\partial\Omega_{T})}
\leq C\exp(CT)\left(\sum_{k=0}^{m}\|\partial_{t}^{k}(\mathrm{Q}\vec{u})\circ(\vec{u}\circ\vec{u})\|_{L^{1}(0,T;\mathbf{H}^{m-k}(\Omega))}+\|\vec{f}\|_{\mathbf{H}^{m+1}(\partial\Omega_{T})}\right) \\
&\leq C\exp(CT)\|(\mathrm{Q}\vec{u})\circ(\vec{u}\circ\vec{u})\|_{\mathcal{D}_{m}}+C\exp(CT)\|\vec{f}\|_{\mathbf{H}^{m+1}(\partial\Omega_{T})}
\end{aligned} ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ( k = 0 ∑ m ∥ ∂ t k ( Q u ) ∘ ( u ∘ u ) ∥ L 1 ( 0 , T ; H m − k ( Ω )) + ∥ f ∥ H m + 1 ( ∂ Ω T ) ) ≤ C exp ( C T ) ∥ ( Q u ) ∘ ( u ∘ u ) ∥ D m + C exp ( C T ) ∥ f ∥ H m + 1 ( ∂ Ω T )
≤ 2 C C 1 R 2 ∥ Q ∥ H m ( Ω ) exp ( C T ) ∥ u ⃗ ∥ D m + 1 + C exp ( C T ) ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) ≤ 1 2 ∥ u ⃗ ∥ D m + 1 + C exp ( C T ) ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) . \begin{aligned}
\leq2CC_{1}R^{2}\|\mathrm{Q}\|_{\mathbf{H}^{m}(\Omega)}\exp(CT)\|\vec{u}\|_{\mathcal{D}_{m+1}}+C\exp(CT)\|\vec{f}\|_{\mathbf{H}^{m+1}(\partial\Omega_{T})} \\
&\leq\frac{1}{2}\|\vec{u}\|_{\mathcal{D}_{m+1}}+C\exp(CT)\|\vec{f}\|_{\mathbf{H}^{m+1}(\partial\Omega_{T})}.
\end{aligned} ≤ 2 C C 1 R 2 ∥ Q ∥ H m ( Ω ) exp ( C T ) ∥ u ∥ D m + 1 + C exp ( C T ) ∥ f ∥ H m + 1 ( ∂ Ω T ) ≤ 2 1 ∥ u ∥ D m + 1 + C exp ( C T ) ∥ f ∥ H m + 1 ( ∂ Ω T ) .
Thus, we have
∥ u ⃗ ∥ D m + 1 + ∥ ∂ ν u ⃗ ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ∥ f ⃗ ∥ H m + 1 ( ∂ Ω T ) . \|\vec{u}\|_{\mathcal{D}_{m+1}}+\|\partial_{\nu}\vec{u}\|_{\mathbf{H}^{m}(\partial\Omega_{T})}\leq C\exp(CT)\|\vec{f}\|_{\mathbf{H}^{m+1}(\partial\Omega_{T})}. ∥ u ∥ D m + 1 + ∥ ∂ ν u ∥ H m ( ∂ Ω T ) ≤ C exp ( C T ) ∥ f ∥ H m + 1 ( ∂ Ω T ) .
This completes the proof.