🔢 mathematics

Combinatorics and loop equations for antisymmetrised and Hermitised matrix product ensembles

This paper investigates the eigenvalue statistics of antisymmetrised and Hermitised matrix product ensembles by constructing combinatorial ribbon graphs and constellations to enumerate their mixed cumulants and deriving loop equation characterizations for their generating functions in the case of order $m=1$ .

Original authors: Stephane Dartois, Anas. A Rahman

Published 2026-09-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Stephane Dartois, Anas. A Rahman

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the vast landscape of modern mathematics and physics, there is a field dedicated to understanding how randomness behaves when it is multiplied together. Imagine a system where you take a large collection of numbers, arrange them into a grid, and then multiply that grid by another, and another, creating a long chain of products. This is the world of random matrix products. For decades, scientists have studied these chains to understand everything from the flow of signals in wireless networks to the stability of complex dynamical systems and even the behavior of neural networks. The core question is simple yet profound: when you multiply many random matrices together, what does the final result look like? Specifically, what are the patterns hidden within the numbers that define the system's state?

Usually, researchers look at these products as they are, but sometimes, to see the underlying structure more clearly, they must change the rules of the game. They might insert a special, fixed matrix into the middle of the chain or force the matrices to have specific symmetries, such as being perfectly mirrored or rotated. These modifications create new, more complex ensembles of matrices. While the standard versions of these problems have been solved, these modified versions have remained elusive, resisting the standard tools used to predict their behavior. Understanding them is crucial because they appear in real-world scenarios where symmetry and specific constraints dictate how information or energy flows through a system.

In this work, two researchers set out to map the hidden geometry of these modified matrix chains. They focused on two specific types: one where the matrices are complex and a special symmetric matrix is inserted into the product, and another where the matrices are real and a special antisymmetric matrix is placed in the middle. These setups are not just abstract puzzles; they are perturbations of well-known models that describe physical phenomena. The researchers wanted to know how the statistics of the numbers inside these matrices change when these specific constraints are applied.

To solve this, the team turned to a powerful method that treats the problem like a game of counting shapes. They realized that the complicated averages of these matrix products could be broken down into simpler pieces, which could then be visualized as ribbon graphs. Imagine a set of colored polygons, each representing a part of the matrix chain. The researchers found that the answer to their questions depended on how many ways these polygons could be glued together along their edges using ribbons. The way these ribbons twist and connect determines the final statistical properties of the system. By counting these specific arrangements, they could translate the difficult algebra of random matrices into a problem of enumerating geometric shapes.

The team discovered that these shapes fall into a strict hierarchy based on their complexity, a concept known as genus, which roughly corresponds to the number of holes in the surface formed by the glued polygons. They proved that the statistical properties of these matrix ensembles can be described by a series of terms, where each term counts a specific type of shape. This provided a complete combinatorial description, meaning they could calculate the behavior of the system by simply counting these geometric configurations. This was a significant step forward, as it turned a chaotic algebraic problem into a structured counting exercise.

However, counting shapes is only the first step. The researchers then needed to find a way to predict the behavior of these shapes without having to count them one by one for every possible scenario. For the simplest case of their problem, where the chain of matrices was short, they derived a set of rules called loop equations. These equations act like a master key, relating the behavior of the whole system to the behavior of its smaller parts. By solving these equations, the researchers were able to find the exact mathematical formulas that describe the large-scale behavior of the system.

The results revealed something surprising about the nature of these modified matrix chains. Unlike the standard models, which follow a simple, well-known pattern, these new ensembles follow a more complex, higher-order rule. The researchers found that the relationship governing the system is a cubic equation, meaning it involves the third power of the variables, rather than the simpler quadratic equations seen in classical models. This higher-order structure implies that the system has a richer, more intricate internal geometry. They were able to solve this cubic equation and express the solution in a rational form, providing a clear, exact description of the system's limits.

Furthermore, the team showed that the statistical properties of these matrices are deeply connected to a famous family of numbers known as Fuss-Catalan numbers. These numbers appear in various counting problems in mathematics, and finding them here suggests a deep, underlying unity between the random matrix products and these classical combinatorial structures. The researchers demonstrated that the distribution of the numbers in their matrices, when the system becomes very large, matches the distribution predicted by these Fuss-Catalan numbers. This connection confirms that their new methods are correct and places these complex matrix ensembles firmly within the broader framework of mathematical knowledge.

In the end, this work provides a complete toolkit for understanding these specific types of random matrix products. By combining the visualization of ribbon graphs with the power of loop equations, the researchers have unlocked the ability to calculate the statistical behavior of these systems with precision. They have shown that even when the rules of the game are changed by inserting special matrices, the system still follows a predictable, albeit more complex, set of laws. This achievement not only solves a specific problem in random matrix theory but also offers a new perspective on how to tackle other complex systems where symmetry and constraints play a critical role. The findings stand as a testament to the power of combining geometric intuition with rigorous algebraic analysis to reveal the hidden order in randomness.

Technical Summary: Combinatorics and Loop Equations for Antisymmetrised and Hermitised Matrix Product Ensembles

Problem Statement
This work investigates the eigenvalue statistics of two specific classes of random matrix product ensembles: the Hermitised matrix product ensemble and the antisymmetrised matrix product ensemble. These ensembles are defined as perturbations of the standard Wishart product ensembles.

The Hermitised ensemble ( $H_m$ ) is formed by the product $X_m^\dagger \cdots X_1^\dagger H X_1 \cdots X_m$ , where $X_i$ are independent complex Ginibre matrices and $H$ is a matrix drawn from the Gaussian Unitary Ensemble (GUE).
The antisymmetrised ensemble ( $J_m$ ) is formed by the product $X_m^T \cdots X_1^T J X_1 \cdots X_m$ , where $X_i$ are independent real Ginibre matrices and $J$ is an elementary antisymmetric matrix.

While recent studies have established connections between these ensembles and Muttalib–Borodin ensembles as well as Harish-Chandra–Itzykson–Zuber (HCIZ) type integrals, a detailed combinatorial and loop equation analysis of their mixed moments and cumulants was lacking. The authors aim to extend the loop equation analysis previously applied to the complex Wishart product ensemble to these perturbed cases.

Methodology
The authors employ a two-pronged approach combining combinatorial enumeration with the loop equation formalism.

Combinatorial Characterisation (Sections 2 & 3):
- Utilizing the Isserlis–Wick theorem, the authors decompose the mixed moments of the ensembles into sums of products of covariances.
- They map these sums to ribbon graphs (maps on surfaces). For the Hermitised case, the graphs are orientable $(m+1)$ -coloured ribbon graphs constructed from amalgamations of Laguerre Unitary Ensemble (LUE) and GUE graphs. For the antisymmetrised case, the graphs are locally orientable $(m+1)$ -coloured ribbon graphs involving twisted ribbons to account for the real Ginibre nature and the antisymmetric matrix $J$ .
- The mixed moments and cumulants are shown to admit genus expansions (topological expansions in powers of $N$ ). The coefficients of these expansions are enumerations of specific ribbon graphs or, equivalently, constellations (tuples of permutations satisfying transitivity and cycle conditions).
- For the antisymmetrised case, a specific "J-weight" is introduced to handle the deterministic antisymmetric matrix $J$ , leading to a formulation involving Euler genus and parity constraints on cycle structures.
Loop Equation Analysis (Sections 4 & 5):
- Specializing to the case $m=1$ (the Laguerre ensembles), the authors derive loop equations (also known as Schwinger–Dyson equations or cut-and-join equations) for the mixed moments.
- These equations are derived via integration by parts on the probability density functions of the underlying Ginibre matrices.
- The authors transform the loop equations for moments into equations for unconnected correlators ( $U_n$ ) and subsequently for connected correlators ( $W_n$ , the generating functions of cumulants).
- By inserting the genus expansion into these equations, they derive closed recursive relations for the coefficients $W_n^l$ (the genus- $l$ contributions to the $n$ -point correlators).

Key Contributions and Results

Genus Expansions: The paper proves that the mixed moments and cumulants of both ensembles possess terminating genus expansions. The coefficients are explicitly enumerated by $(m+1)$ -coloured ribbon graphs (Hermitised) and locally orientable ribbon graphs with J-weights (antisymmetrised).
Constellation Formulations: The authors provide an algebraic reformulation of these combinatorial results using the theory of constellations (tuples of permutations). This links the random matrix statistics to the enumeration of specific permutation tuples with transitivity and cycle constraints.
Loop Equations for $m=1$ :
- The authors derive a hierarchy of loop equations for the antisymmetrised Laguerre ensemble ( $m=1$ ).
- A significant finding is that the spectral curve for this ensemble is a cubic polynomial (third-order) in the generating function $W_1^0(x)$ , contrasting with the quadratic spectral curves typical of classical $\beta$ -ensembles (like Wigner semicircle or Marčenko–Pastur).
- The loop equations exhibit a distinct structure involving denominators of the form $(x_i^2 - x_j^2)$ , reflecting the antisymmetric nature of the underlying matrix $J$ and the vanishing of odd moments.
Explicit Solutions:
- The spectral curve is solved to provide a rational parametrisation of the genus-zero Riemann sphere.
- The leading order correlator $W_1^0(x)$ is shown to generate the Fuss–Catalan numbers (specifically order $d=3$ ), confirming the connection to the complex Wishart product ensemble via the eigenvalues of $iJ_1$ .
- Explicit rational expressions for the genus-zero two-point correlator $W_2^0(x_1, x_2)$ and the genus-one one-point correlator $W_1^1(x_1)$ are derived.

Significance and Claims
The paper claims to provide the first systematic combinatorial and loop equation characterisation of these specific matrix product ensembles.

It demonstrates how perturbing Wishart ensembles with a GUE matrix or an antisymmetric matrix alters the topological expansion and the structure of the loop equations.
The work highlights that these ensembles serve as concrete examples of random matrix models with higher-order spectral curves (order > 2), a feature often studied in abstract multi-matrix models but less common in standard random matrix theory.
The derivation of closed loop equations for the connected correlator coefficients allows for the recursive computation of all higher-order terms in the genus expansion, providing a complete framework for calculating eigenvalue statistics in the large $N$ limit.
The results confirm the asymptotic equivalence of the antisymmetrised ensemble's eigenvalue statistics to those of the Fuss–Catalan distribution, bridging the gap between the combinatorial ribbon graph approach and known results on Muttalib–Borodin ensembles.

The authors do not propose new experimental applications or future physical systems but position this work as a foundational theoretical step in understanding the integrability and combinatorial structure of non-standard random matrix products.

More like this