Sectional curvature and matrix displacement convexity

Gautam Aishwarya, Liran Rotem, Yair ShenfeldarXiv 2025 · DOI · arXiv

TL;DR

This paper establishes a deep connection between the geometric property of non-negative sectional curvature on Riemannian manifolds and a novel concept called "matrix displacement convexity" of the entropy functional. This connection leads to new, "intrinsic dimensional" functional inequalities that offer sharper bounds than previous results, particularly in the context of heat flow and optimal transport.

What Problem Does This Paper Address?

The paper builds upon a well-established link between the Ricci curvature of a Riemannian manifold and the "displacement convexity" of its entropy functional. Specifically, non-negative Ricci curvature is equivalent to the entropy being displacement convex along optimal transport geodesics. However, Ricci curvature is a "sum" of sectional curvatures, and the authors sought to understand if a similar characterization exists for sectional curvature itself, which is a more fundamental geometric quantity. The challenge lies in finding a way to "disentangle" the contributions of individual sectional curvatures from the averaged Ricci curvature.

Methods

The core of the paper introduces and utilizes the concept of matrix displacement convexity for the entropy tensor. This is a tensorial generalization of displacement convexity. The authors define an "entropy tensor" derived from Jacobi fields along optimal transport geodesics. They then show that non-negative sectional curvature is equivalent to this entropy tensor being "matrix displacement convex." This tensorial condition is stronger than simple displacement convexity and allows for a more refined analysis.

The proof strategy involves:

  • Lagrangian Picture of Optimal Transport: The authors primarily work with the Lagrangian formulation of optimal transport, which describes the evolution of probability measures via the exponential map of a potential function.
  • Jacobi Fields and Curvature: They leverage the relationship between Jacobi fields (which describe the deviation of geodesics) and the Riemann curvature tensor. Specifically, the evolution of Jacobi fields is governed by an equation involving curvature.
  • Entropy Tensor Definition: The entropy tensor is defined as an integral of a matrix derived from Jacobi fields along an optimal transport geodesic.
  • Matrix Displacement Convexity: This is the key novel concept, defined as a non-negativity condition on the second derivative of the entropy tensor with respect to time, ordered by the Loewner order.
  • Functional Inequalities: As an application, they derive "intrinsic dimensional" evolution variational inequalities and Wasserstein contraction results along heat flows. These inequalities improve upon existing ones by replacing the ambient dimension of the manifold with a more refined notion of dimension implicitly captured by the matrix displacement convexity condition.

Datasets

Not discussed in the paper. The paper focuses on theoretical results on Riemannian manifolds, not empirical evaluation on datasets.

Key Insights and Ideas

  • Sectional Curvature's Tensorial Signature: The paper reveals that non-negative sectional curvature has a specific "tensorial fingerprint" in the form of matrix displacement convexity of the entropy tensor. This is a more precise geometric condition than the scalar displacement convexity associated with Ricci curvature.
  • Entropy Tensor as a Geometric Probe: The entropy tensor, derived from Jacobi fields, acts as a powerful tool to probe the local geometry of the manifold, specifically its sectional curvature.
  • Intrinsic Dimensionality: The matrix displacement convexity condition naturally leads to functional inequalities where the "dimension" involved is intrinsic to the problem (e.g., related to the number of independent directions of curvature), rather than just the ambient dimension of the manifold. This allows for sharper bounds.
  • Bridging Geometry and Optimal Transport: The work further solidifies the profound link between Riemannian geometry and optimal transport, showing how geometric curvature properties can be translated into convexity properties of functionals defined on spaces of probability measures.

Key Results

  • Theorem 1.4 (Main Result): Equivalence between non-negative sectional curvature, matrix displacement convexity of the entropy tensor, and a stronger inequality involving the second derivative of the entropy tensor.
  • Theorem 1.6 & Corollary 1.7: Derivation of "intrinsic dimensional" evolution variational inequalities and Wasserstein contraction along heat flows. These results improve upon existing inequalities by incorporating the intrinsic dimensional aspect, leading to tighter bounds.
  • Example 1.8 (Product Manifolds): Demonstrates how the intrinsic dimensional inequalities can correctly identify the "effective" dimension in product spaces, where the ambient dimension might be misleading.

Limitations and Future Work

  • Compactness Assumption: Theorem 1.6 and Corollary 1.7 are stated for compact manifolds. Remark 1.9 notes that this assumption can be relaxed, but it's made for technical convenience.
  • Approximate Gradients: Remark 2.11 mentions that without the compact support assumption for measures, the Brenier-McCann theorem requires approximate gradients. However, for manifolds with non-negative sectional curvature, these can be replaced by true gradients, allowing the proof to go through.
  • Future Directions: The paper implicitly suggests exploring further applications of these intrinsic dimensional inequalities in various areas of analysis and geometry. The connection between matrix displacement convexity and other geometric notions could also be a fruitful area of research.

Summary updated

Abstract

We show that the sectional curvature of a Riemannian manifold is nonnegative if, and only if, the entropy functional is matrix displacement convex. As an application we obtain intrinsic dimensional evolution variational inequalities, and the corresponding Wasserstein contraction along heat flows, which improve on their dimensional counterparts.
BibTeX
@misc{aishwarya2025sectional,
  author = {Gautam Aishwarya and Liran Rotem and Yair Shenfeld},
  title = {{Sectional curvature and matrix displacement convexity}},
  year = {2025},
  eprint = {2509.23399},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.23399},
}

Loading related content...

Figure