Lipschitz Stability of Travel Time Data

Joonas Ilmavirta, Antti Kykkänen, Matti Lassas, Teemu Saksala, Andrew Shedlock

TL;DR

This paper proves that reconstructing a geometric space from "travel time data"—the distances from every point in the space to a fixed measurement set—is Lipschitz stable. By introducing the concept of $\epsilon$-local isometries, the authors show that small errors in distance measurements lead to only proportionally small errors in the reconstructed shape of the space, even for complex geometries like non-simple manifolds and metric trees.

What Problem Does This Paper Address?

In geometric inverse problems, a classic question is: "Can we determine the shape of an object if we only know the distances from its interior points to its boundary?" This is known as the travel time tomography problem, and it has deep roots in geophysics and medical imaging.

While previous research established that such a reconstruction is possible (uniqueness) for "simple" spaces, the real-world challenge is stability. In practice, measurements are noisy. If a tiny change in the data leads to a massive change in the reconstructed space, the method is useless for applications. This paper tackles the gap between theoretical uniqueness and practical stability by providing quantitative Lipschitz bounds—essentially a mathematical guarantee that the reconstruction process won't "blow up" in the presence of small errors.

Methods

The authors move beyond smooth Riemannian manifolds to the broader category of compact length spaces. Their approach relies on several key mathematical innovations:

  • $\epsilon$-Local Isometries: Instead of requiring a global isometry (which is too restrictive), they define "Forward/Backward Locally Isometrically Embeddable" (FLIE/BLIE) conditions. This means the mapping between the space and the data only needs to preserve distances for points that are relatively close to each other (within a distance $\epsilon$).
  • Truncated Gromov–Hausdorff Distance: To compare spaces without needing global diameter bounds, the authors introduce a "truncated" version of the Gromov–Hausdorff distance. This metric only cares about the structure of the space up to a certain scale $\epsilon$.
  • Midpoint Properties: They provide a data-driven way to verify stability. If the travel time data satisfies a specific "local midpoint property," one can conclude the reconstruction is stable without knowing the underlying space's geometry beforehand.
  • Myers–Steenrod Extension: They prove a new version of the Myers–Steenrod theorem specifically for manifolds with boundaries, which is a crucial tool for showing that distance-preserving maps are actually smooth isometries.

Datasets

As a theoretical mathematics paper, this work does not use external empirical datasets. Instead, the authors evaluate their theorems using synthetic geometric models and counter-examples:

  • Herglotz Manifolds: Spherically symmetric disks with varying "sound speeds" used to demonstrate stability in the presence of interior conjugate points (where geodesics focus).
  • Metric Trees: Graphs where edges have lengths, showing the theory applies to non-manifold structures.
  • Euclidean Domains: Specifically, an annulus in $\mathbb{R}^n$ to show how non-trivial topology affects stability.
  • Spheres ($S^2$): Used to illustrate how "small" measurement sets can fail to provide enough data for a unique reconstruction.

Key Insights and Ideas

  • Stability via Locality: The biggest "aha" moment is that you don't need a global distance-preserving map to ensure stability. If the map is a local isometry at a small enough scale, the global structure is "locked in" Lipschitz-stably.
  • The "Twice the Injectivity Radius" Rule: For Riemannian manifolds, the authors identify a sufficient condition for stability: no geodesic should be longer than twice the injectivity radius. This provides a concrete geometric heuristic for when stability is guaranteed.
  • Busemann G-Spaces: They show that the travel time map is a topological embedding for Busemann G-spaces (a generalization of manifolds where geodesics don't branch), significantly broadening the class of spaces where these inverse problems can be solved.

Key Results

  • Quantitative Stability (Theorem 6): The authors prove that the Gromov–Hausdorff distance between two spaces $X_1$ and $X_2$ is bounded by: $$d_{GH}(X_1, X_2) \leq \left(\frac{2D}{\epsilon} + 1\right) d_H(\text{Travel Time Data})$$ where $D$ is the diameter and $\epsilon$ is the local isometry scale. This provides the explicit Lipschitz constant.

  • Truncated Stability (Theorem 9): They show that the $\epsilon$-truncated Gromov–Hausdorff distance is directly bounded by the Hausdorff distance of the data: $d_{GH}^\epsilon \leq d_H(\text{data})$.

  • Finsler Manifolds: They extend these results to reversible Finsler manifolds, proving that travel time data determines the manifold up to a smooth isometry.

Limitations and Future Work

  • The BLIE Assumption: The main theorems require the "Backward Locally Isometrically Embeddable" property. While the authors provide ways to check this, it remains a strong assumption that might not hold for all complex geometries.
  • Non-Reversible Metrics: The current work focuses on symmetric (reversible) distance functions. Extending this to non-reversible Finsler metrics (where the distance from $A$ to $B$ isn't the same as $B$ to $A$) remains an open challenge.
  • Numerical Implementation: While the paper provides the theoretical stability, it does not provide a specific reconstruction algorithm. Future work could involve developing a computational framework that utilizes these Lipschitz bounds for robust imaging in noisy environments.

Summary updated

Abstract

We prove that the reconstruction of a certain type of length spaces from their travel time data on a closed subset is Lipschitz stable. The travel time data is the set of distance functions from the entire space, measured on the chosen closed subset. The case of a Riemannian manifold with boundary with the boundary as the measurement set appears is a classical geometric inverse problem arising from Gel'fand's inverse boundary spectral problem. Examples of spaces satisfying our assumptions include some non-simple Riemannian manifolds, Euclidean domains with non-trivial topology, and metric trees.
BibTeX
@article{ilmavirta2025lipschitz,
  author = {Joonas Ilmavirta and Antti Kykkänen and Matti Lassas and Teemu Saksala and Andrew Shedlock},
  title = {{Lipschitz Stability of Travel Time Data}},
  journal = {Journal of Geometric Analysis},
  year = {2025},
  doi = {10.1007/s12220-025-02084-3},
}
Figure