Repulsive Ensembles of N Particles

• Repulsive ensembles of N particles are systems characterized by singular, diverging interactions that prevent particle collision.
• They employ pairwise log-gas or power-law potentials and encompass models in random matrix theory, integrable systems, and machine learning.
• Analytical methods, including equilibrium measure analysis and fluctuation statistics, elucidate universal behavior and finite-size effects.

Repulsive Ensembles of N Particles

Repulsive ensembles of N particles constitute a fundamental class of statistical and dynamical systems in which particles interact through mutual repulsion, typically encoded via pairwise or many-body potentials that diverge or become strongly positive as particles approach coincidence. These ensembles appear across mathematical physics, random matrix theory, integrable systems, condensed matter, and probabilistic models, with rigorous formulations spanning both discrete and continuous (and non-Archimedean) settings. The study of repulsive particle ensembles leverages analytic, probabilistic, and combinatorial tools to characterize equilibrium distributions, fluctuation properties, dynamical behavior, and connections to underlying algebraic and integrable structures.

1. Core Models and Definition

Repulsive N-particle ensembles are specified by the configuration space $x = (x_1, \dots, x_N)$, an interaction energy $E(x)$ exhibiting singular repulsion for coinciding particles, and a joint probability measure (often of Gibbs form). The prototypical example is the log-gas or $\beta$-ensemble on $\mathbb{R}$: $$p_N(x_1,\dots,x_N) = \frac{1}{Z_N} \prod_{1 \leq i < j \leq N} |x_i - x_j|^\beta \prod_{i=1}^N e^{-N V(x_i)},$$ with $V$ a confining potential and $\beta > 0$ controlling the repulsion (Kriecherbauer et al., 2015, Kuijlaars, 2015, Olshanski, 2020, Borot et al., 23 Jan 2026).

Generalizations include:

• Riesz or power-law repulsion, $V_{\text{pair}}(x_i, x_j) = J |x_i - x_j|^{-k}$ for $k>-2$ (Agarwal et al., 2019, Mughal, 2013).
• Polynomial and Macdonald-level ensembles, where repulsion is encoded via Vandermonde or $(q,t)$-deformed determinants (Kuijlaars, 2015, Olshanski, 2020).
• Discrete ensembles on lattices and non-Archimedean fields, where the interaction is via the $p$-adic absolute value of the Vandermonde determinant (Sinclair, 2020).

Physical realizations arise in Wigner-Dyson random matrices, classical Coulomb gases, zero-range plasma models, repelling eigenvalue problems, and synthetic ensembles in machine learning (repulsive deep/final-layer ensembles) (d'Angelo et al., 2021, Steger et al., 2024).

2. Interaction Energies and Algebraic Structures

The interaction energy in repulsive ensembles typically possesses the following generic features:

• Singular repulsion: Divergence (often logarithmic or power-law) as interparticle distances vanish, enforcing effective non-collision.
• Algebraic form: Many ensembles express the interaction as $E(x) = -\log |\Delta_N(x)|^2$, where $\Delta_N(x)$ is a Vandermonde determinant (for polynomial ensembles) or its generalizations (e.g., $|\Delta_N(x)|_p$ in the $p$-adic setting) (Sinclair, 2020, Kuijlaars, 2015).
• Factorization properties: In discrete and ultrametric models (such as on $\mathbb{Z}_p$), non-interacting coset structure arises directly from the strong triangle inequality (Sinclair, 2020).
• Multi-component and multi-charge: Interaction matrices can be extended to particles with distinct species/charges, where cross-terms involve products of Vandermonde-like factors and their analogues (Sinclair, 2020).

Algebraic manipulations enabled by determinants, hypergeometric weights, or difference operators (e.g., in Macdonald-level or $q,t$-deformed settings), support recursion relations, explicit computation of partition functions, and emergent integrable structure (Olshanski, 2020, Kuijlaars, 2015).

3. Equilibrium Measures and Macroscopic Profiles

The macroscopic behavior for $N \gg 1$ is governed by a mean-field or variational (energy minimization) principle applied to the empirical density $$\rho_N(x) = \frac{1}{N} \sum_{i=1}^N \delta(x - x_i)$$. For log-gas and Riesz ensembles in confinement:

• The equilibrium measure $\rho_*(x)$ arises from the solution to

$$\rho_*(x) = \arg\min_{\rho} \left\{ \iint_{\mathbb{R}^2} V_{\text{pair}}(x,y) \rho(x) \rho(y)\,dx\,dy + \int V(x)\rho(x)\,dx \right\},$$

subject to $\int \rho = 1$ and nonnegativity (Agarwal et al., 2019, Mughal, 2013).

• For $|x_i - x_j|^{-k}$ repulsion, density profiles exhibit regime transitions at $k=1$ and $k=-1$: semicircular law (log-gas), uniform density (1d OCP), and algebraic edge singularities for generic $k$ (Agarwal et al., 2019).
• In spherical geometries, the interplay between inverse-power repulsion and hard wall confinement yields shell/skin transitions, divergence at the boundary (for moderate $k$), and uniform filling for very steep repulsion (large $k$) (Mughal, 2013).

In random matrix ensembles, filling constraints and saturation effects induce bands, voids, and saturated regions characterized by variational (Euler-Lagrange) inequalities and modified equilibrium conditions (Borot et al., 23 Jan 2026).

4. Fluctuations, Finite-Size Corrections, and Edge Statistics

Beyond the mean profile, repulsive ensembles exhibit universal and non-universal fluctuation behavior:

• Central limit and law of large numbers: Linear statistics about equilibrium obey the law of large numbers and central limit theorem, with explicit covariance structures depending on the strength and range of repulsion (Borot et al., 23 Jan 2026).
• Edge statistics: The distribution of extreme particles (largest/smallest) exhibits Tracy–Widom universality in certain scaling regimes, becoming modified by the non-determinantal or kernel-averaged nature of generalized repulsive ensembles (Kriecherbauer et al., 2015).
• Large/moderate deviations: The probabilities of atypical fluctuations (large excursions of, e.g., the rightmost particle) are governed by large deviation rate functions computable from the equilibrium problem, with prefactors sensitive to the model's fine structure (Kriecherbauer et al., 2015).
• Boundary layers: In presence of hard walls or barriers, a boundary layer emerges where $o(N)$ particles deviate from the bulk equilibrium description, with scaling and profile determined by a first-order $\Gamma$-expansion and nonlocal integral equations (Meurs, 2021).

In discrete ensembles and in the presence of distinct filling fractions, additional discrete Gaussian components may perturb the central limit theorem, and modifications of the standard Kenyon-Okounkov conjecture are required to describe non-orientable domains (Borot et al., 23 Jan 2026).

5. Dynamical and Integrable Aspects

Repulsive N-particle systems also appear as Hamiltonian models, with dynamics governed by Newtonian or generalized Hamilton's equations under repulsive forces:

• Long-time dispersive dynamics: For purely repulsive Coulomb $n$-body systems in $\mathbb{R}^3$, each pairwise distance grows linearly in $t$, and each particle's trajectory approaches a straight line with limiting velocity distinct from all others (Rein, 2017).
• Integrability and global reduction: On the line with sufficiently rapidly decaying or inverse power–law repulsive potentials $f_{ij}(q_i-q_j)$ (including Calogero–Moser $r=2$), one constructs global action–angle variables (asymptotic velocities and phases), obtaining canonical reduction to free motion (Gorni et al., 2012).
• Cone-potential theory: The class of cone potentials permits robust integrability even under compact perturbations, with explicit correspondence between original and normal-form coordinates, and transparent description of scattering data and phase shifts.

In the configuration/occupation space of non-Archimedean ensembles (e.g., over $\mathbb{Z}_p$), strong triangle inequalities induce highly nonlocal non-dynamical partitioning effects, and recursive decomposition yields partition functions computable in finite time (Sinclair, 2020).

6. Generalizations and Applications

Repulsive ensembles are structurally versatile and admit significant extensions:

• Multi-species/charge/multicomponent systems: Partition functions and interaction matrices generalize naturally to several species with arbitrary cross repulsion coefficients, altering the equilibrium laws and fluctuations (Sinclair, 2020, Borot et al., 23 Jan 2026).
• Macdonald and $(q,t)$-deformed discrete ensembles: Inclusion of Macdonald parameters and $q$-lattice support introduces additional algebraic and harmonic-analytic structure, facilitating projective systems and measure-theoretic boundaries for infinite-particle limits (Olshanski, 2020).
• Non-Archimedean fields: On ultrametric spaces (e.g., $\mathbb{Q}_p$), occupation independence between cosets is enforced, and the partition function is constructed recursively leveraging Haar measure factorization (Sinclair, 2020).
• Function-space and deep learning ensembles: In Bayesian deep learning, "repulsive deep ensembles" and "repulsive last-layer ensembles" utilize kernel-based diversity-inducing functionals to enforce non-collapse among neural predictors, aligning the ensemble with approximations to the true posterior or Wasserstein gradient flows (d'Angelo et al., 2021, Steger et al., 2024).

For many-body quantum systems in $D \geq 2$, repulsive $\delta$-potentials are ineffective: rigorous analysis shows no spectral effect or phase shift unless finite-range or suitably renormalized pseudopotentials are employed (Simenog et al., 2017).

7. Many-Body Interactions and Exotic Regimes

Beyond pairwise repulsive models, explicit higher-order many-body terms are relevant in dispersion interactions and certain materials physics settings:

• Alternating-sign N-body interactions: For parallel polarizable filaments, N-body dispersion terms alternate in sign, with odd-$N$ clusters manifesting net repulsion that opposes further clustering, favoring even-N bundling (Pal et al., 2024).
• Geometric and combinatorial regimes: Random tilings, partitions, and models with complex domain topology introduce correlated groups of particles with variable or constrained repulsion intensities, modifying the equilibrium problem and fluctuation behavior (Borot et al., 23 Jan 2026).

Limiting care is essential in specifying the regime of interaction, the class of admissible configurations, and the asymptotic or scaling limits applied, as these details dictate universality classes, fluctuation phenomena, and connections to underlying algebraic structures.

---

The study of repulsive ensembles of $N$ particles provides a unifying framework linking random matrix theory, integrable systems, mathematical physics, combinatorics, and modern probabilistic learning, with a rich interplay of algebraic, analytical, and dynamical tools underpinning both universal and model-specific phenomena (Sinclair, 2020, Kriecherbauer et al., 2015, Agarwal et al., 2019, Borot et al., 23 Jan 2026, d'Angelo et al., 2021).