Bayesian inference for the fractional Calderón problem with a single measurement

Pu-Zhao Kow, Janne Nurminen, Jesse Railo · 2025

Read Bayesian inference for the fractional Calderón problem with a single measurement by Pu-Zhao Kow, Janne Nurminen, Jesse Railo (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-2511-11068.txt.

Introduction

Pu-Zhao Kow, Janne Nurminen, and Jesse Railo study whether a Bayesian method can reliably recover an unknown trait from just one noisy measurement in a fractional Calderón problem. That problem asks what can be inferred about a material's inside from outside data, here in a fractional form that spreads influence widely. Priors must be chosen with care so the method does not fool itself.

Their main result shows that the posterior, the updated belief after data, piles up near the true value as readings grow. They also pin down tight rates for how fast the mean estimate's error shrinks. Sharp stability bounds for both directions of the problem drive the proof.

Catalog record

CREATORPu-Zhao Kow, Janne Nurminen, Jesse Railo
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-11068

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bayesian inference for the fractional Calderón problem with a single measurement Pu-Zhao Kow, Janne Nurminen, Jesse Railo Abstract . This paper investigates the consistency of a posterior distribution in the single-measurement fractional Calderón problem with additive Gaussian noise. We consider a Bayesian framework with rescaled and Gaussian sieve priors, using a collection of noisy, discrete observations taken from a suitable exterior domain. Our main result shows that the posterior distribution concentrates around the true parameter as the number of measurements increases. Furthermore, we establish tight convergence rates for the reconstruction error of the posterior mean. A central technical challenge is to obtain refined stability estimates for both the forward and inverse problems. In particular, the required forward estimates are delicate to obtain because the fractional elliptic problems do not enjoy as strong regularity theory as their classical counterparts. 2020 Mathematics Subject Classification 35R11, 35R30, 62G20 Notations Throughout this paper, we shall use the symbol \lesssim and \gtrsim for inequalities holding up to a universal constant. For two real sequences (a_{N}) and b_{N} , we say that \simeq if both a_{N}\lesssim b_{N} and b_{N}\lesssim a_{N} for all sufficiently large N . For a sequence of random variables Z_{N} and a real sequence (a_{N}) , we write Z_{N}=O_{\Pr}(a_{N}) if for all \varepsilon>0 there exists M_{\varepsilon}<\infty such that for all N large enough, \Pr(|Z_{N}|\geq M_{\varepsilon}a_{N})<\varepsilon . Denote {\mathcal{L}}(Z) the law of a random variable Z . We also denote a\vee b=\max\{a,b\} for all a,b\in\mathbb{R} . Let d\geq 1 be an integer. For each \gamma\in\mathbb{R} , let H^{\gamma}(\mathbb{R}^{d}) be the standard L^{2} -based fractional Sobolev space. For each open set U\subset\mathbb{R}^{d} , we denote \tilde{H}^{\gamma}(U) be the closure of C_{c}^{\infty}(U) (i.e. the collection of smooth function supported in \Omega ) and denote H^{\gamma}(U) be the restriction of H^{\gamma}(\mathbb{R}^{d}) in U equipped with the standard quotient norm. The following identifications is well-known for any bounded Lipschitz domain U : (\tilde{H}^{\gamma}(U))^{\prime}=H^{-\gamma}(U),\quad(H^{\gamma}(U))^{\prime}=\tilde{H}^{-\gamma}(U)\quad\text{for all $\gamma\in\mathbb{R}$,} see e.g. [ GSU20 , Section 2A] , [ McL00 , Chapter 3] or [ KLW22 ] and the references therein for more details. 1. Introduction Let \Omega be a nonempty bounded smooth domain in \mathbb{R}^{d} , and let (-\Delta)^{s} denote the fractional Laplacian of order s\in(0,1) , which admits at least ten equivalent definitions under suitable interpretations [ Kwa17 ] . We fix any 0\not\equiv\phi\in C_{c}^{\infty}(\Omega_{e}) with \Omega_{e}:=\mathbb{R}^{d}\setminus\overline{\Omega} , and consider the following Dirichlet problem: (1.1) ((-\Delta)^{s}+f)u=0\text{ in $\Omega$},\quad u|_{\Omega_{e}}=\phi. In this work we study an inverse problem for the fractional Schrödinger equation which concerns the recovery of the potential f from the Dirichlet-to-Neumann map (DN map) (1.2) \Lambda_{f}\phi\equiv\left.(-\Delta)^{s}u_{f}\right|_{\mathcal{D}}, where \mathcal{D}\subset\Omega_{e} is a fixed open set. It was proved in [ GSU20 ] that the potential can be uniquely determined from exterior measurements and, remarkably, a single-point measurement suffices [ GRSU20 ] . The nonlocal Dirichlet problem ( 1.1 ) has been extensively studied in, e.g., [ Gru15 , RO16 ] . Nonlocal problems of this type also arise naturally in various applications. For instance, nonlocal models describe animal movement [ VAB+96 ] , foraging patterns of marine predators [ HQD+10 ] , the dynamics of competing populations sharing the same resource [ MV17 ] and in quantum optics [ KS22 , KS23 , OHKSW24 ] . Moreover, the fractional Laplacian appears in the linear problem of unique continuation from partial data Radon transforms in integral geometry and X-ray tomography [ IM20 , CMR21 , IKS25 ] . Nonlocal problems also play a central role in physics, fluid dynamics, finance, and image processing, see [ RO16 , BV16 ] for further references. For inverse problems related to ( 1.1 ) we mention [ CMRU22 , FGKU25 , FGK+25 ] and the monograph [ LL25 ] . Further nonlinear variants are studied in [ KRZ23 , KMS23 , KW23 ] . We consider a statistical model in which we randomly choose finitely many uniformly distributed points from \mathcal{D} . A priori ( 1.2 ) is not pointwise well-defined but Lemma 3.1 shows it to be smooth when restricted to \mathcal{D} for a smooth exterior value \phi . Then the value of the DN map at these points, with added noise, creates our measurements. We model f with rescaled Gaussian priors and Gaussian sieve priors. Then we study the concentration and contraction rates, also the optimality of these rates, of posterior distributions arising from these priors. This approach to studying the Calderón problem was iniated in [ AN19 ] where the authors introduced a statistical inverse problem for the isotropic conductivity equation. Then the approach was generalized in [ GN20 ] in the context of determining a conductivity from a source-to-solution map. The authors of [ GN20 ] used priors that were inspired by previous consistent inversion of noisy non-abelian X -ray transforms [ MNP21 ] . Further results for various inverse problems with similar techniques can be found in [ FKW24 , GK20 , Kek22 , KW25 , NvdGW20 ] and the references therein. Our main results show that an estimator, here we use the posterior mean, converges to the ground truth that creates the data. The convergence rate is logarithmic and is based on the stability of the inverse problem [ Rül21 ] . Furthermore we prove that this convergence rate is optimal in the statistical minimax sense and this is based on the instability result for the inverse problem [ RS18 ] (see also [ Man01 ] for the local version and [ KRS21 ] for a more general setting for instability). To prove these results we follow the general framework developed in [ GN20 ] . In order to use this framework we need to analyze the forward and inverse problem in more detail. For the forward problem we need our forward map f\mapsto\Lambda_{f}\phi to be uniformly bounded on its domain and a suitable L^{2} -based local Lipschitz forward estimate (see Section 3.1 ). The uniform boundedness is proved via the singular integral definition of the fractional Laplacian. To prove the forward estimate we analyze the DN map using the following observations. Firstly we use that the DN map is smooth in the exterior domain when we assign a smooth exterior value and thus we can use the strong definition of the DN map (see [ GSU20 ] ). We use this together with the weak definition of the DN map and the Alessandrini identity. The trick is that in the Alessandrini identity we can choose up to three different functions to appear in the identity. Choosing these in a clever way, using a stronger elliptic estimate from [ CRZ22 ] and in the end combining this with an approximation argument using cutoffs gives the desired estimate. See Lemma 3.3 for further details. For the inverse problem we need a stability estimate, in particular a stability estimate for a single measurement Calderón problem (for full data this was proven in [ RS20 ] ). A stability estimate is proved in [ Rül21 ] but we need to refine it in order to make visible the expressions for \lVert f_{1}\rVert_{C^{s}(\Omega)} and \lVert f_{2}\rVert_{C^{s}(\Omega)} . Deriving this estimate requires modifying some of the arguments in [ Rül21 , GRSU20 ] and using tools developed in these articles. Once we have obtained the

Read the complete text