[Paper Review] Recovering boundary conditions in inverse Sturm-Liouville problems

Norbert Roehrl|arXiv (Cornell University)|Jan 2, 2006

Spectral Theory in Mathematical Physics9 references3 citations

TL;DR

This paper presents a variational algorithm that simultaneously recovers the potential and boundary conditions in inverse Sturm-Liouville problems using two given spectra, without requiring prior knowledge of the potential's mean. The method employs a least squares functional minimized via conjugate gradient descent, and theoretical analysis proves the absence of strict local minimizers, ensuring convergence to the global solution even with noisy data or poor initial guesses.

ABSTRACT

We introduce a variational algorithm, which solves the classical inverse Sturm-Liouville problem when two spectra are given. In contrast to other approaches, it recovers the potential as well as the boundary conditions without a priori knowledge of the mean of the potential. Numerical examples show that the algorithm works quite reliable, even in the presence of noise. A proof of the absence of strict local minimizers of the functional supports the observation, that a good initial guess is not essential.

Motivation & Objective

• To develop a numerical method that recovers both the potential and boundary conditions in inverse Sturm-Liouville problems from two given spectra.
• To eliminate the need for prior knowledge of the potential's mean, which is required in many existing approaches.
• To ensure robustness against noisy spectral data and insensitivity to initial guesses through theoretical and numerical validation.
• To prove the absence of strict local minimizers in the functional, supporting reliable convergence of optimization algorithms.

Proposed method

• A least squares functional G(𝑞) is defined to measure the discrepancy between computed and given eigenvalues, with weights applied to spectral data.
• The functional is minimized using the Polak-Ribière conjugate gradient descent algorithm to iteratively update the parameters (h₀, h₁, h₂, q).
• Gradients of eigenvalues with respect to h₀, h₁, h₂, and q are computed using eigenfunction squared values at the endpoints and within the domain.
• Theoretical justification relies on the linear independence of eigenvalue gradients in R³ × L², proven via a bilinear form and Wronskian identities.
• The method uses normalized eigenfunctions and boundary conditions to compute sensitivity information for optimization.
• Numerical experiments include noise injection and periodic reinitialization of the potential to improve convergence.

Experimental results

Research questions

• RQ1 Can a variational approach recover both the potential and boundary conditions in inverse Sturm-Liouville problems without requiring the mean of the potential as input?
• RQ2 Does the absence of strict local minimizers in the functional ensure convergence to the global solution regardless of initial guess?
• RQ3 How robust is the algorithm under noisy spectral data, and can reinitialization of the potential improve convergence?
• RQ4 Can two interlacing sequences of eigenvalues uniquely determine both the potential and the boundary conditions?
• RQ5 What is the role of gradient linear independence in ensuring the convergence of the conjugate gradient method?

Key findings

• The algorithm successfully recovers the potential and boundary conditions with high accuracy, achieving G(𝑞) ≈ 1.39 × 10⁻⁴ and Δ₂ ≈ 0.669 with noise level r = 0.005.
• With noise level r = 0.01, the method achieved G(𝑞) ≈ 1.50 × 10⁻⁴ and Δ₂ ≈ 0.471 after 237 iterations, showing robustness to noise.
• Reinitializing the potential to zero at iteration 89 significantly improved convergence, reducing the final error and accelerating progress.
• Theoretical analysis proves that the functional has no strict local minimizers, meaning ∇G(𝑞) = 0 if and only if G(𝑞) = 0.
• The gradients of eigenvalues are linearly independent in R³ × L², which underpins the absence of spurious local minima.
• The method converges reliably even with poor initial guesses, as confirmed by numerical experiments with noisy data.