📊 statistics

Bayesian inference for the fractional Calderón problem with a single measurement

This paper establishes the consistency and convergence rates of a Bayesian posterior distribution for the single-measurement fractional Calderón problem by employing rescaled Gaussian sieve priors and overcoming technical challenges related to the limited regularity of fractional elliptic problems.

Original authors: Pu-Zhao Kow, Janne Nurminen, Jesse Railo

Published 2026-09-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Pu-Zhao Kow, Janne Nurminen, Jesse Railo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to see inside a solid object without ever cutting it open or shining a light through it. In the world of physics and engineering, this is the essence of an inverse problem: you measure what happens on the outside and try to work backward to figure out what is hidden inside. A classic version of this challenge involves electricity. If you attach electrodes to the surface of a material and measure how the voltage changes, you can theoretically deduce the material's internal conductivity. This is the basis of medical imaging techniques like electrical impedance tomography, which can help doctors see tumors or other anomalies without surgery. However, real-world measurements are never perfect; they are always tainted by noise, like static on a radio, and often, you can only take a limited number of readings. When the physics governing the object is complex and the data is imperfect, figuring out the hidden interior becomes a mathematical puzzle of extreme difficulty.

In a recent study, researchers tackled a particularly stubborn version of this puzzle involving a phenomenon known as the fractional Calderón problem. This problem deals with materials where the rules of movement and interaction are "fractional," meaning they do not follow the standard, smooth paths we see in everyday life. Instead, particles in these systems can make sudden, long jumps, a behavior that appears in everything from the movement of animals searching for food to the way light behaves in certain quantum experiments. The researchers wanted to know if it was possible to reliably reconstruct the hidden properties of such a material using noisy measurements taken from the outside. They approached this not by trying to solve the equations directly, but by using a statistical method called Bayesian inference. This approach treats the unknown interior as a mystery to be solved by combining what we already know with the new, imperfect data we have collected, gradually narrowing down the possibilities until a clear picture emerges.

The team, led by Pu-Zhao Kow, Janne Nurminen, and Jesse Railo, set out to prove that this statistical method works even in the most difficult scenarios. They built a mathematical model where the hidden properties of the material were represented by a random guess, or "prior," which was then updated as they simulated taking measurements. These measurements were not perfect; they were taken at random points outside the object and included a layer of random noise, mimicking the errors found in real instruments. The researchers used two different types of statistical guesses to start their process: one that assumed the hidden properties followed a smooth, bell-curve distribution, and another that built the image up from a finite collection of wave-like patterns. By running their model thousands of times, they observed how the statistical guess changed as the number of measurements increased.

Their main discovery was that the method is indeed consistent. As the researchers added more and more data points to their model, the statistical guess stopped wandering and began to cluster tightly around the true, hidden value. In other words, the more data they fed into the system, the more confident the method became in its answer, eventually zeroing in on the correct interior structure. They also calculated exactly how fast this convergence happens. They found that the error in their reconstruction shrinks at a logarithmic rate, which is a specific, slow pace typical for these kinds of difficult problems. To put this in perspective, while the method works, it requires a significant amount of data to achieve high precision, much like trying to hear a whisper in a noisy room; you need to listen for a long time to be sure of what was said.

Crucially, the researchers did not just show that their method worked; they also proved that it was as good as it could possibly be. They demonstrated that no other method, no matter how clever, could reconstruct the hidden properties any faster given the same amount of noisy data. This establishes a fundamental limit on what is achievable in this field. The study also highlighted the specific mathematical hurdles they had to overcome. Unlike standard physics problems where the rules are well-behaved and predictable, these fractional problems are "rougher" and less regular, making it much harder to guarantee that the outside measurements accurately reflect the inside. The team had to develop new, refined mathematical tools to bridge this gap, proving that the connection between the outside data and the inside truth, while fragile, is strong enough to be trusted.

To ensure their theoretical findings were not just abstract numbers, the researchers also ran a computer simulation to see the method in action. They created a one-dimensional scenario where they knew the true answer beforehand and then tried to recover it using their Bayesian approach. They generated 150 random measurements with a very small amount of noise and ran a computer algorithm that iteratively improved its guess over five million steps. The result was a reconstruction that closely matched the true hidden value, visually demonstrating the convergence they had predicted. While the simulation showed some limitations due to the precision of the computer's calculations, it successfully illustrated the core principle: the statistical method can indeed peel back the layers of uncertainty to reveal the hidden truth.

This work matters because it provides a rigorous safety net for using advanced statistical methods in critical applications. Whether it is monitoring the health of a bridge, diagnosing a disease, or understanding the behavior of exotic materials, knowing that a method is mathematically guaranteed to converge to the truth is essential. The study confirms that even when the physics is complex and the data is noisy, we can trust these Bayesian tools to guide us toward the correct answer, provided we are willing to gather enough evidence. It turns a daunting mathematical challenge into a reliable procedure, offering a clear path forward for scientists and engineers who need to see the invisible.

Technical Summary: Bayesian Inference for the Fractional Calderón Problem with a Single Measurement

Problem Statement
This paper addresses the statistical inverse problem of recovering a potential function $f$ in the fractional Schrödinger equation from noisy, discrete exterior measurements. The governing equation is the Dirichlet problem:
$((-\Delta)^s + f)u = 0 \quad \text{in } \Omega, \quad u|_{\Omega^e} = \phi,$
where $(-\Delta)^s$ is the fractional Laplacian of order $s \in (0,1)$ , $\Omega \subset \mathbb{R}^d$ is a bounded smooth domain, and $\Omega^e = \mathbb{R}^d \setminus \Omega$ . The goal is to reconstruct $f$ from the exterior Dirichlet-to-Neumann (DN) map $\Lambda_f \phi = (-\Delta)^s u|_D$ , where $D \subset \Omega^e$ is a fixed open set.

Unlike classical Calderón problems where full data is often assumed, this work focuses on the single-measurement setting, where the potential is uniquely determined by a single non-zero exterior source $\phi$ . The statistical model assumes that measurements are taken at $N$ random points $\{X_i\}_{i=1}^N$ uniformly distributed in $D$ , subject to additive Gaussian noise:
$Y_i = G(f)(X_i) + \sigma W_i,$
where $G(f) = \Lambda_f \phi$ and $W_i \sim \mathcal{N}(0,1)$ .

Methodology
The authors employ a Bayesian framework to study the consistency and convergence rates of the posterior distribution as the sample size $N \to \infty$ . The methodology follows the general statistical inverse problem framework established in [GN20], adapted to the nonlocal nature of the fractional Laplacian.

Priors: Two types of priors are considered for the reparametrized potential $F$ (where $f = \Phi \circ F$ via a smooth link function $\Phi$ ):
- Rescaled Gaussian Priors: A centered Gaussian measure on a Banach space, scaled by a factor depending on $N$ to ensure appropriate concentration.
- Gaussian Sieve Priors: Finite-dimensional approximations using truncated Karhunen–Loève expansions based on Daubechies wavelets.
Technical Challenges and Estimates:
A central difficulty is the lack of strong regularity theory for fractional elliptic problems compared to classical ones. To apply the Bayesian machinery, the authors establish refined deterministic estimates:
- Forward Estimates: They prove that the DN map is smooth in the exterior domain and establish a uniform $L^\infty$ bound and a local Lipschitz estimate ( $L^2$ -based) for the forward operator. This involves a delicate analysis using the Alessandrini identity, strong elliptic estimates, and cutoff approximations.
- Inverse Stability: They derive a logarithmic stability estimate for the inverse problem. This refines existing results (e.g., [Rül21]) to explicitly track the dependence on the $C^s$ norms of the potentials, which is crucial for the contraction rate analysis.
Contraction Analysis: The posterior distribution is shown to concentrate around the true parameter $f_0$ in the $L^\infty$ norm. The convergence rates are derived by combining the stability of the inverse problem with the properties of the priors.

Key Results

Posterior Consistency: The posterior distribution concentrates around the true potential $f_0$ as $N \to \infty$ . Specifically, for both rescaled Gaussian and Gaussian sieve priors, the posterior probability that the reconstruction error exceeds a certain threshold decays exponentially.
Convergence Rates: The paper establishes a logarithmic convergence rate for the posterior mean (and the posterior distribution itself) in the $L^\infty$ norm:
$\| \hat{f}_N - f_0 \|_{L^\infty} = O_P((\log N)^{-\mu}).$
This rate is consistent with the inherent ill-posedness of the fractional Calderón problem.
Minimax Optimality: The authors prove that this logarithmic rate is optimal in the statistical minimax sense. They demonstrate that no estimator can achieve a faster rate of convergence uniformly over the class of potentials, based on instability results for the fractional Calderón problem [RS18].
Numerical Demonstration: A one-dimensional numerical experiment using a Metropolis-Hastings MCMC algorithm is presented to illustrate the Bayesian procedure. While the reconstruction is limited by floating-point precision, the simulation confirms the feasibility of the approach.

Significance and Claims
The paper claims to provide the first theoretical guarantees for Bayesian inference in the single-measurement fractional Calderón problem. Its primary contribution is bridging the gap between the deterministic stability theory of nonlocal inverse problems and statistical estimation theory.

The authors emphasize that their results show the Bayesian method is theoretically sound for this specific nonlocal inverse problem, mirroring previous successes in the classical (local) Calderón problem [AN19] and source-to-solution maps [GN20]. They explicitly state that the logarithmic rate is a consequence of the instability of the inverse problem and that their derived rate is tight, meaning it cannot be improved without additional assumptions. The work does not claim to solve the computational challenges of high-dimensional nonlocal problems but rather establishes the statistical consistency and optimality of the Bayesian approach in the asymptotic limit.

Technical Summary: Bayesian Inference for the Fractional Calderón Problem with a Single Measurement

More like this