Target localization, identification and sensing using latent symmetries

Technical Summary: Target Localization, Identification, and Sensing Using Latent Symmetries

Problem Statement
The paper addresses the challenge of target localization and identification in three-dimensional open systems using arrays of scatterers. Conventional sensing methods often rely on detecting shifts in eigenfrequencies or exploiting geometric symmetries. However, the authors note that eigenfrequency shifts have physical limits regarding sensitivity, and many practical systems lack the classical geometric symmetries required for traditional symmetry-based sensing. The specific problem tackled here is how to utilize "latent" (or "hidden") symmetries—symmetries present in the eigenfunctions of a system despite the absence of classical geometric symmetries—to detect, identify, and localize an "intruder" scatterer within a resonator array.

Methodology
The authors employ the capacitance matrix as a canonical model for a coupled system of three-dimensional acoustic or electrostatic resonators. Unlike sparse graph models where interactions are local, the capacitance matrix is dense, capturing non-trivial long-range coupling due to the slow decay of the Laplace Green's function.

1. Latent Symmetry Identification: The study begins by identifying resonator configurations where the capacitance matrix exhibits latent symmetries. This is achieved by finding pairs of sites (resonators) where the eigenvectors display parity symmetry (values are equal up to a sign change) despite the lack of geometric symmetry. The authors utilize a graph-theoretic approach, checking for cospectral vertices via the matrix power identity $(C^k)_{ii} = (C^k)_{jj}$ for all $k$.
2. Forward Problem (Perturbation Analysis): The introduction of an intruder scatterer breaks these latent symmetries. The authors define a metric, $F_{pq}$, to quantify the degree of symmetry breaking. This metric represents the average absolute difference between the magnitudes of the eigenvector components at a pair of latently symmetric sites ($p$ and $q$) across all eigenmodes.
   • They analyze the asymptotic behavior of $F_{pq}$ as the intruder's radius ($r$) approaches zero or its distance ($|\tilde{z}|$) approaches infinity, showing that the perturbation vanishes in these limits.
   • Numerical simulations reveal that $F_{pq}$ depends on both the intruder's radius and position. Crucially, the authors observe that for a fixed position, the response curves for different radii are parallel, suggesting a separable form $F_{pq}(r, x, y) = H_{pq}(r)G_{pq}(x, y)$.
3. Inverse Problem (Target Localization): The goal is to predict the intruder's position $(x, y)$ and radius $r$ given measurements of the symmetry breaking ($F_{pq}$).
   • Radius Prediction: For a known position, the relationship between radius and $F_{pq}$ is modeled using an exponential guess function and refined using Gaussian Process Regression (GPR) to handle multiplicative measurement noise.
   • Position Prediction: Due to the multivalued nature of the inverse mapping (caused by system symmetries) and ill-conditioning near resonators, the authors propose three approaches:
     • Dictionary-Based Approach: A lookup method comparing measured values against a pre-computed grid of known positions. This works well for deterministic (noise-free) data, especially when combining data from multiple symmetric pairs to resolve ambiguities.
     • Multi-Layer Perceptron (MLP): A neural network trained to map noisy symmetry-breaking measurements directly to position coordinates.
     • Bayesian Inference: A probabilistic approach using Markov Chain Monte Carlo (MCMC) to estimate the posterior distribution of the intruder's position given noisy observations and a prior distribution.

Key Results

• Discovery of 3D Latent Symmetries: The authors successfully identified resonator arrays in three dimensions that possess latent symmetries without any underlying geometric symmetry. This is the first observation of such symmetries in a dense, 3D open system that cannot be approximated by a sparse graph.
• Sensitivity to Intruders: The introduction of an intruder breaks the latent symmetry, quantifiable via the $F_{pq}$ metric. The magnitude of this breakage correlates with the intruder's size and proximity to the symmetric pair.
• Algorithm Performance:
  • In noise-free scenarios, the dictionary-based approach combined with multiple symmetric pairs achieves exact localization.
  • In the presence of measurement noise (simulated as 20% deviation), the MLP and Bayesian inference models significantly outperform the dictionary-based method.
  • Bayesian inference with a Gaussian prior provided the most accurate localization, effectively correcting for distribution skewness observed in the posterior.
• Separability: The study confirms that the perturbation functions are separable into radial and positional components, simplifying the modeling of the sensing response.

Significance and Claims
The paper claims two primary milestones:

1. First Successful Application to Sensing: To the authors' knowledge, this is the first time latent symmetries have been successfully exploited for sensing problems, specifically for target localization and identification.
2. First Observation in Dense 3D Systems: This work represents the first observation of latent symmetries in a three-dimensional open system characterized by dense, long-range coupling (modeled by the capacitance matrix), distinguishing it from previous studies limited to sparse graphs or tight-binding systems.

The authors conclude that while simple dictionary matching suffices for ideal conditions, sophisticated machine learning and probabilistic methods (MLP and Bayesian inference) are necessary to robustly handle the noise inherent in real-world sensing applications. They suggest future extensions could involve systems with multiple latent symmetry pairs to lift degeneracies further and more generalized symmetry concepts involving orthogonal matrices.