Lipschitz stability and reconstruction in inverse problems for semi-discrete parabolic operators
Rodrigo Lecaros, Juan López-Ríos, Ariel A. Pérez · 2025
Read Lipschitz stability and reconstruction in inverse problems for semi-discrete parabolic operators by Rodrigo Lecaros, Juan López-Ríos, Ariel A. Pérez (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-01143.txt.
Introduction
Rodrigo Lecaros, Juan López-Ríos, and Ariel Pérez solve an inverse problem for a heat-type equation on a grid. Their 2025 paper seeks the right-hand side of a semi-discrete parabolic equation from readings taken partway through and inside part of the domain. The result then steadies a harder task: recovering a place-dependent potential.
A reconstruction method follows, with a proof that it converges. The engine is a new Carleman estimate tuned to the grid, whose parameter respects mesh size. Discrete terms in the bound naturally add an error tied to the starting state. The authors frame the work as a step toward reliable recovery from partial data in grid-based heat models.
Catalog record
| CREATOR | Rodrigo Lecaros, Juan López-Ríos, Ariel A. Pérez |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-01143
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Lipschitz stability and reconstruction in inverse problems for semi-discrete parabolic operators Rodrigo Lecaros, Juan López-Ríos, Ariel A. Pérez Abstract . This work addresses an inverse problem for a semi-discrete parabolic equation, consisting of identifying the right-hand side of the equation from solution measurements at an intermediate time and within a spatial subdomain. We apply this result to establish a stability estimate for a coefficient inverse problem involving the recovery of a spatially dependent potential function. Furthermore, we present a reconstruction algorithm for recovering this coefficient and provide a proof of its convergence. Our approach relies on a novel semi-discrete Carleman estimate in which the parameter is constrained by the mesh size. Due to the discrete terms arising in the Carleman inequality, this method naturally introduces an error term associated with the solution’s initial condition. 2020 Mathematics Subject Classification 35R30, 35R20, 35K20, 35K10 1. Introduction Let d\geq 1 , T>0 , and \displaystyle\Omega:=\prod_{i=1}^{d}(0,1)\subset\mathbb{R}^{d} , with \omega\Subset\Omega denoting an arbitrary subdomain. We consider the following parabolic system (1.1) \begin{cases}\partial_{t}y(t,x)-\mathcal{A}y(t,x)=g(t,x),&(t,x)\in(0,T)\times\Omega,\\ y(t,x)=0,&(t,x)\in(0,T)\times\partial\Omega,\\ y(0,x)=y_{ini}(x),&x\in\Omega,\end{cases} where \mathcal{A} is a uniformly elliptic second-order operator defined by (1.2) \mathcal{A}y(t,x)=\sum_{i=1}^{d}\frac{\partial}{\partial x_{i}}\left(\gamma_{i}(t,x)\frac{\partial y}{\partial x_{i}}(t,x)\right)-\sum_{i=1}^{d}b_{i}(t,x)\frac{\partial y}{\partial x_{i}}(t,x)-c(t,x)y(t,x). Here, \gamma_{i}(t,x)>0 for all (t,x)\in(0,T)\times\Omega , and g\in H^{1}((0,T),L^{2}(\Omega)) . In this framework, a classical inverse problem consists of determining the source term g(t,x) from observations of y within the subdomain \omega . Specifically, for a fixed time \vartheta\in(0,T) , we consider the observation operator \Lambda_{\vartheta}:H^{1}((0,T),L^{2}(\Omega))\to H^{2}(\Omega)\times H^{1}((0,T),L^{2}(\omega)), given by \Lambda_{\vartheta}(g):=(y|_{t=\vartheta},y|_{\omega\times(0,T)}), where y denotes the solution of ( 1.1 ). The stability of the inverse problem corresponds to the Lipschitz inequality (1.3) \|g\|_{H^{1}((0,T),L^{2}(\Omega))}\leq C\|\Lambda_{\vartheta}(g)\|:=C\left(\|y|_{t=\vartheta}\|_{H^{2}(\Omega)}+\|y\|_{H^{1}((0,T),L^{2}(\omega))}\right), for some constant C>0 . Several works have addressed this inverse problem in the literature; see, for instance, [ 15 , 16 , 14 , 30 ] . As noted in [ 15 ] , most results in this area are obtained when the observation time \vartheta is lies in (0,T) , following the method introduced by Bukhgeim and Klibanov [ 8 , 9 , 17 ] . In [ 14 ] , the authors applied this method to prove uniqueness and Lipschitz stability of the inverse problem, while in [ 16 ] , they established conditional Lipschitz stability and uniqueness for the case \vartheta=T . In contrast, the (semi)discrete setting has been explored primarily in the context of controllability problems for parabolic operators; see [ 7 , 10 , 26 ] for the spatial semi-discrete setting, [ 5 ] for the time semi-discrete case, and [ 12 , 24 ] for the fully discrete setting. Recently, the time semi-discrete setting for an inverse problem was studied in [ 19 ] . In this regard, the authors did not discuss extending the analysis to the spatial semi-discrete framework for parabolic operators. Moreover, the only results on inverse problems in the spatial semi-discrete setting are [ 3 , 32 ] and [ 34 ] , concerning the wave and Schrödinger equations, respectively. Hence, our objective is to fill this gap by studying an inverse source problem for a spatial semi-discretization of the system ( 1.1 ), establishing its stability (see Theorem 1.1 ), and providing a reconstruction algorithm to recover a zeroth-order spatially dependent coefficient (see Algorithm 1 ). Let us introduce some notation related to the spatial semi-discrete framework. Given N\in\mathbb{N} , let h=\frac{1}{N+1} be small enough to represent the size of the mesh. We define the Cartesian grid of [0,1]^{d} as (1.4) \mathcal{K}_{h}:=\left\{x\in[0,1]^{d}\mid\exists k\in\mathbb{Z}^{d}\text{ such that }x=hk\right\}. We set the mesh \mathcal{W}:=\Omega\cap\mathcal{K}_{h} and denote by C(\mathcal{W}) the set of functions defined on \mathcal{W} . Moreover, we define the average and difference operators as (1.5) \begin{split}A_{i}u(x)&:=\frac{1}{2}\left(\tau_{+}iu(x)+\tau_{-i}u(x)\right),\\ D_{i}u(x)&:=\frac{1}{h}\left(\tau_{+i}u(x)-\tau_{-i}u(x)\right),\end{split} where \tau_{\pm i}y(x):=y(x\pm\frac{h}{2}e_{i}) , being \{e_{i}\}_{i=1}^{d} the canonical basis of \mathbb{R}^{d} . Thus, by denoting Q:=(0,T)\times\mathcal{W} , the spatial semi-discrete approximation of the system ( 1.1 ) is given by (1.6) \begin{cases}\partial_{t}y(t,x)-\mathcal{A}_{h}y(t,x)=g(t,x),\quad&(t,x)\in Q,\\ y(t,x)=0,&(t,x)\in(0,T)\times\partial\mathcal{W},\\ y(0,x)=y_{ini}(x),&x\in\mathcal{W},\end{cases} with \mathcal{A}_{h} being the finite difference space approximation of the continuous operator ( 1.2 ), given by (1.7) \mathcal{A}_{h}y(t,x):=\sum_{i=1}^{d}D_{i}\left(\gamma_{i}(t,x)D_{i}y(t,x)\right)-\sum_{i=1}^{d}b_{i}(t,x)D_{i}A_{i}y(t,x)-c(t,x)y(t,x). Our inverse problem consists of determining the right-hand side of the system ( 1.6 ), known as an inverse source problem, from the knowledge of the data \left(y\Bigr|_{t=\vartheta},\Bigl.y\Bigr|_{(0,T)\times\omega}\right) , where \omega\subset\mathcal{W} is an arbitrary subdomain. That is, we investigate the semi-discrete analogue of ( 1.3 ). Assume that \gamma_{i},b_{i}\in C^{1}([0,T]\times\overline{\Omega}) , for all i=1,\ldots,d , that c\in C^{1}([0,T]\times\overline{\Omega}) and that y_{ini}\in C^{0}(\overline{\Omega}) . Let \Gamma(t,x):=\mathrm{Diag}(\gamma_{1}(t,x),\gamma_{2}(t,x),\ldots,\gamma_{d}(t,x)), such that \gamma_{i}(t,x)>0 for all i=1,\ldots,d , and it holds \mbox{reg}(\Gamma):=\esssup_{\begin{array}[]{c}(t,x)\in[0,T]\times\overline{\Omega}\\ i=1,\ldots,d\end{array}}\left(\gamma_{i}(t,x)+\frac{1}{\gamma_{i}(t,x)}+|\nabla_{x}\gamma_{i}(t,x)|+|\partial_{t}\gamma_{i}(t,x)|\right)<+\infty. Given \mbox{reg}^{0}>0 , henceforth \Gamma is such that \mathrm{reg}(\Gamma)\leq\mbox{reg}^{0} . Moreover, assume that, for some constant C>0 , the function g\in C^{1}([0,T];L^{\infty}(\overline{\Omega})) satisfies the estimate (1.8) |\partial_{t}g(t,x)|\leq C|g(\vartheta,x)|,\quad\text{for almost all }(t,x)\in(0,T)\times\overline{\Omega}. Our first main result is the following stability estimate. The detailed notation is introduced in the next section. Theorem 1.1 . Let \psi satisfy ( 2.19 ) and let \varphi be given by ( 2.20 ). Assume that g satisfies ( 1.8 ), and let y\in\mathcal{C}^{1}([0,T],\overline{\mathcal{W}}) be the solution of the system ( 1.6 ). Then there exist positive constants C , C^{\prime\prime} , \tau_{0}\geq 1 , h_{0}>0 , \varepsilon>0 , depending on \omega , \mathrm{reg}^{0} and T , such that for all \tau\geq\tau_{0}(T+T^{2}) , 0<h\leq h_{0} , there exists 0<\delta(h)\leq 1/2 satisfying \tau h(\delta T^{2})^{-1}\leq\varepsilon , and the estimate \displaystyle\|g\|_{L_{h}^{2}(Q)}\leq \displaystyle Ce^{\frac{C^{\prime\prime}}{T^{2}}\tau}\left(\|y|_{t=\vartheta}\|_{H_{h}^{2}(\mathcal{W})}+\|e^{s\varphi}\partial_{t}y\|_{L_{h}^{2}(Q_{\omega})}+\|e^{s\varphi}y\|_{L_{h}^{2}(Q_{\omega})}\right) \displaystyle+Ce^{-\frac{C^{\prime\prime}}{h}}\left(\|y|_{t=0}\|_{L^{2}_{h}(\mathcal{W})}+\|\partial_{t}y|_{t=0}\|_{L^{2}_{h}(\mathcal{W})}\right), holds for Q_{\omega}:=(0,T)\times\omega . In the inequality of the above Theorem, there is an error term e^{-\frac{C^{\prime\prime}}{h}}\left(\|y|_{t=0}\|_{L^{2}_{h}(\mathcal{W})}+\|\partial_{t}y|_{t=0}\|_{L^{2}_{h}(\mathcal{W})}\right), which arises from the discrete phenomenon and tends to zero as h\to 0 . Moreover, if we assume y(0)=\partial_{t}y(0)=0 , we recover the classical inequality for the continuous case as in [ 14 ] . The proof of Theorem 1.1 is based on the new Carleman estimate ( 1.9 ) established for the operator in ( 1.6 ). To our knowledge, the only known Carleman estimate available in the literature for semi-discrete parabolic operators in arbitrary dimensions is that of [ 7 ] . However, it is not suitable for studying the inverse problem because it lacks a term involving the second-order spatial operator. In this work, we address this issue by establishing Carleman estimates for the solution of system ( 1.6 ) and ( 3.1 ), corresponding to the cases q=0 and 2q=1 , respectively. These results are summarized below. Theorem 1.2 . Let \psi satisfy ( 2.19 ) and let \varphi be given by ( 2.20 ). Let y\in\mathcal{C}^{1}([0,T],\overline{\mathcal{W}}) be the solution of system ( 1.6 ). For \lambda\geq 1 sufficiently large, there exist constants C , \tau_{0}\geq 1 , h_{0}>0 , \varepsilon>0 , depending on \omega , \omega_{0} , \mathrm{reg}^{0} , T , and \lambda , such that (1.9) \displaystyle I_{2q}(y)+J_{2q}(y)\leq \displaystyle C\left(\int_{Q}e^{2\tau\theta\varphi}(\tau\theta)^{2q}|g|^{2}+\int_{(0,T)\times\omega}(\tau\theta)^{{2q}+3}e^{2\tau\theta\varphi}|y|^{2}\right) \displaystyle+Ch^{-2}\int_{\mathcal{W}}(\tau\theta(0))^{2q}\left(\Big|y|_{t=0}\Big|^{2}+\Big|y|_{t=T}\Big|^{2}\right)e^{2\tau\theta(0)\varphi}, where \displaystyle I_{{2q}}(y):=\int_{Q}(\tau\theta)^{2q-1} \displaystyle|\partial_{t}y|^{2}e^{2\tau\theta\varphi}+\sum_{i,j\in\llbracket 1,d\rrbracket}\int_{Q^{\ast}_{ij}}(\tau\theta)^{2q-1}\gamma_{i}\gamma_{j}e^{2\tau\theta\varphi}|D_{ij}y|^{2}, and \displaystyle J_{2q}(y):= \displaystyle\tau^{2q+1}\sum_{i\in\llbracket 1,d\rrbracket}\left(\left\|\theta^{1/2+q}e^{\tau\theta\varphi}D_{i}y\right\|^{2}_{L^{2}_{h}(Q^{\ast}_{i})}+\left\|\theta^{1/2+q}e^{\tau\theta\varphi}A_{i}D_{i}y\right\|^{2}_{L^{2}_{h}(Q)}\right) \displaystyle+\tau^{3+2q}\left\|\theta^{3/2+q}e^{\tau\theta\varphi}y\right\|^{2}_{L^{2}_{h}(Q)}, for any q\in\mathbb{R} , \tau\geq\tau_{0}(T+T^{2}) , 0<h\leq h_{0} , 0<\delta\leq 1/2 , \tau h(\delta T^{2})^{-1}\leq\varepsilon . Finally, assume that the coefficients \gamma_{i},b_{i} , i=1,\ldots,d , and c , are independent of time. Our final main result is a reconstruction algorithm for the inverse problem of identifying the potential p\in L^{\infty}_{h}({\mathcal{W}}) from the measurements of y , the solution of (1.10) \begin{cases}\partial_{t}y(t,x)-\mathcal{A}_{h}y(t,x)+p(x)y(t,x)=g(t,x),\quad&(t,x)\in Q,\\ y(0,x)=y_{ini}(x),&x\in\mathcal{W},\\ y(t,x)=f(t,x),&(t,x)\in(0,T)\times\partial\mathcal{W},\end{cases} where f,g and y_{ini} are given functions. More precisely, we consider the measurement operator \Lambda_{p}:=(y_{p}|_{t=T/2},\partial_{t}y_{p}|_{Q_{\omega}},\partial_{t}y_{p}|_{\{t=0\}\times\mathcal{W}}), where y_{p} is the solution of ( 1.10 ) associated with p . We also define, for m>0 , the set \mathcal{X}_{m}:=\{p\in L^{\infty}_{h}(\mathcal{W}):\;\|p\|_{L^{\infty}_{h}}\leq