Next-order asymptotic expansion for N-marginal optimal transport with Coulomb and Riesz costs

Codina Cotar, Mircea PetracheAdvances 2019 · DOI · arXiv

Introduction

In the fields of many-body physics and quantum chemistry, Density Functional Theory (DFT) serves as a cornerstone for understanding the behavior of large systems of interacting particles. A central challenge in this domain is the $N$-marginal optimal transport (OT) problem, which acts as a mathematical proxy for the "strictly correlated" limit of these systems. While the first-order behavior—the mean-field or Hartree energy—is well-established, the underlying physics is often dictated by the next-order term: the exchange-correlation energy. This term accounts for the fine-grained correlations between particles that the mean-field approximation fails to capture.

Historically, a sharp asymptotic expansion for this exchange-correlation term was only rigorously available for the specific case of Coulomb interactions ($s=1$) in three dimensions ($d=3$). This paper provides a significant advancement by deriving a unified, rigorous asymptotic expansion for a general class of Riesz costs ($|x-y|^{-s}$) across all dimensions where the energy remains integrable ($0 < s < d$).

Methodology: The Fefferman-Gregg Decomposition

To address the singular nature of Riesz kernels, the authors employ a sophisticated "divide and conquer" strategy known as the Fefferman-Gregg decomposition. This approach allows for the localization of interaction energy, treating a global problem as a collection of smaller, independent sub-problems.

Kernel Decomposition and Positive Definiteness

The authors break down the singular interaction kernel into a sum of "nice" parts—specifically those that are positive definite and have a finite range—plus a manageable error term. A critical innovation in this work is ensuring that every component of the decomposition remains "balanced-positive-definite." This property is essential for ensuring that local energy contributions converge correctly to the mean-field limit, providing a more robust framework than the Graf-Schenker decomposition used in previous literature.

Geometric Localization: Swiss Cheese Packings

To manage the spatial distribution of particles, the authors utilize a "Swiss cheese" geometric strategy. By packing the domain with balls of varying scales, they can localize the interaction energy. To overcome the technical difficulty of maintaining a fixed number of particles $N$ within these small localized regions, they adopt a grand-canonical relaxation. In this formulation, the number of particles is fixed only on average, which simplifies the mathematical treatment of fluctuations.

Key Insights

Universality of the Correlation Constant

A major finding of this research is the universality of the constant $C(s,d)$ in the second-order term. The expansion shows that this term scales as $N^{1+s/d}$ and depends solely on the cost exponent $s$ and the dimension $d$, rather than the specific geometry of the particle distribution $\mu$. This suggests a fundamental structural property of correlated systems across different physical contexts.

Beyond Rotational Invariance

Unlike many classical methods in statistical mechanics that rely on radial symmetry or rotational invariance, the Fefferman-Gregg approach is applicable to non-rotationally-invariant costs. This flexibility is vital for future studies involving anisotropic materials or systems situated in curved spaces, where traditional methods often fail.

Main Results

The primary contribution of the paper is the derivation of the sharp asymptotic expansion:

$$F_{N,c}^{OT}(\mu) = N^2 \iint \frac{\rho(x)\rho(y)}{|x-y|^s} dx dy - C(s,d) N^{1+s/d} \int \rho^{1+s/d}(x) dx + o(N^{1+s/d})$$

This formula explicitly decouples the macroscopic mean-field energy from the microscopic correlation energy. Furthermore, the authors establish that for a specific range of exponents, the constant $C(s,d)$ in the OT problem is identical to the constant found in the "Jellium" model—a classic plasma model—thereby bridging optimal transport theory with classical statistical mechanics.

Limitations and Future Directions

While the paper covers a broad range of Riesz interactions ($0 < s < d$), the authors identify a potential "gap" at the Coulomb exponent boundary ($s = d-2$). While the energies for the Uniform Electron Gas (UEG) and Jellium models agree for $s \in (d-2, d)$, the behavior exactly at $s = d-2$ remains a subtle point for future investigation.

Additionally, the current framework is optimized for long-range interactions. Extending these sharp asymptotics to short-range or hypersingular kernels ($s \geq d$) would require the development of entirely different mathematical tools. The authors suggest that their extended Fefferman-Gregg decomposition may eventually provide a simplified proof for the Jellium problem itself, a prospect they intend to explore in subsequent work.

Summary updated

BibTeX
@inproceedings{cotar2019next,
  author = {Codina Cotar and Mircea Petrache},
  title = {{Next-order asymptotic expansion for N-marginal optimal transport with Coulomb and Riesz costs}},
  booktitle = {Advances},
  year = {2019},
  doi = {10.1016/j.aim.2018.12.008},
}
Figure