Evolutionary Variational Inequalities on the Hellinger-Kantorovich and Spherical Hellinger-Kantorovich spaces

Vaios Laschos, Alexander MielkearXiv 2022 · DOI · arXiv

Abstract

We study the minimizing movement scheme for families of geodesically semi-convex functionals defined on either the Hellinger-Kantorovich or the Spherical Hellinger-Kantorovich space. By exploiting some of the finer geometric properties of those spaces, we prove that the sequence of curves, which are produced by geodesically interpolating the points generated by the minimizing movement scheme, converges to curves that satisfy the Evolutionary Variational Inequality (EVI), when the time step goes to 0.

Introduction

In the realm of optimal transport, the classical Wasserstein distance has long served as a powerful tool for analyzing gradient flows. However, a significant limitation of the Wasserstein framework is the requirement of mass conservation; the total mass of the system must remain constant. This constraint is often violated in real-world physical and biological processes, such as population dynamics, tumor growth, or chemical reactions, where mass is frequently created or destroyed.

To address these "unbalanced" systems, researchers have turned to the Hellinger-Kantorovich (HK) distance. This metric generalizes optimal transport by combining spatial movement with mass growth and decay. While the HK distance provides a robust framework for modeling, defining and analyzing gradient flows within these abstract metric spaces presents substantial mathematical challenges. This paper establishes a rigorous foundation for these flows by proving that discrete optimization steps converge to unique, stable continuous solutions defined by Evolutionary Variational Inequalities (EVI).

The Minimizing Movement Scheme and the HK Space

The authors focus on the Minimizing Movement (MM) scheme, also known as the Jordan-Kinderlehrer-Otto (JKO) scheme. This discrete-time approach produces a sequence of measures by iteratively minimizing a functional that balances the "energy" of the state with the "distance" moved.

To bridge the gap between these discrete steps and a continuous-time gradient flow, the space must possess specific geometric properties. A primary technical hurdle is that the HK space is not globally "well-behaved." Specifically, it is only geodesically semiconcave when restricted to measures with bounded densities. To overcome this, the authors employ a localization approach. They prove that if an initial measure has a bounded density, the MM scheme preserves these bounds across all subsequent time steps—a result referred to as the discrete maximal principle. This ensures the flow remains within a subset of the space where the geometry is sufficiently regular.

Key Insights: Localization and the Power of EVI

A central contribution of this research is the realization that global semiconcavity is not a prerequisite for guaranteeing a unique gradient flow. By demonstrating that the solution is "trapped" within subsets of measures with bounded densities, the authors show that the flow remains stable and unique.

Furthermore, the authors exploit the cone representation of the HK distance. By viewing the HK distance as a standard Wasserstein distance on a cone over the original space, they are able to transfer established results from transport theory to the more complex unbalanced setting.

Proving that the flow satisfies an Evolutionary Variational Inequality (EVI) is a significant milestone. Unlike "weak solutions" to partial differential equations, EVI solutions automatically provide:

  1. Uniqueness: There is only one path the system can take.
  2. Stability: The distance between two different starting points does not grow over time, ensuring the system is robust to small perturbations.

Results and Implications

The paper culminates in a Convergence Theorem, proving that for a broad class of entropy functionals, the geodesically interpolated curves from the discrete MM scheme converge to a unique EVI solution as the time step $\tau$ approaches zero. This provides the rigorous mathematical limit showing that taking infinitely small steps in a JKO-style optimization leads exactly to the continuous-time reaction-diffusion equations used in physics.

Additionally, the authors verified that both HK and Spherical HK (SHK) spaces satisfy the Local Angle Condition (LAC). This verification, combined with the established semiconcavity on density-restricted subsets, fulfills the necessary requirements for EVI theory in these spaces.

Limitations and Future Work

While this work provides a solid theoretical foundation, several avenues for future research remain:

  • Geometric Constraints: The current proofs assume the underlying physical space $X$ is a compact, convex subset of $\mathbb{R}^d$. Extending these results to non-compact spaces or general Riemannian manifolds is an ongoing challenge.
  • Functional Characterization: While the paper identifies which functionals are "convex enough" for HK spaces, a complete characterization for the Spherical HK space remains an open question.
  • Numerical Implementation: The research focuses on existence and convergence proofs; the computational efficiency of implementing these EVI-convergent schemes for high-dimensional biological simulations has yet to be addressed.

In conclusion, this paper bridges the gap between discrete optimization and continuous gradient flows in unbalanced optimal transport, offering a rigorous framework for modeling complex systems where mass is not conserved.

Summary updated

Abstract

We study the minimizing movement scheme for families of geodesically semi-convex functionals defined on either the Hellinger-Kantorovich or the Spherical Hellinger-Kantorovich space. By exploiting some of the finer geometric properties of those spaces, we prove that the sequence of curves, which are produced by geodesically interpolating the points generated by the minimizing movement scheme, converges to curves that satisfy the Evolutionary Variational Inequality (EVI), when the time step goes to 0.
BibTeX
@misc{laschos2022evolutionary,
  author = {Vaios Laschos and Alexander Mielke},
  title = {{Evolutionary Variational Inequalities on the Hellinger-Kantorovich and Spherical Hellinger-Kantorovich spaces}},
  year = {2022},
  eprint = {2207.09815},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2207.09815},
}

Loading related content...

Figure