Unitary Matrix Integrals and Linear ODEs

• Unitary matrix integrals are integrals over the unitary group with respect to the Haar measure, encoding combinatorial, number-theoretic, and mathematical physics properties.
• They are uniquely characterized by a matrix first-order ODE system and an equivalent scalar ODE of order l+1, enabling precise power series expansions.
• The method unifies enumeration problems and analysis of zeta function moments, while extending to circular β-ensembles for broader applications in random matrix theory.

A unitary matrix integral is an integral over the unitary group $U(l)$ with respect to Haar probability measure, typically encoding objects of combinatorial, number-theoretic, and mathematical physics interest. A canonical example is

$$I_{l,q}(s) := \left\langle (\det U)^q \exp(s^{1/2} \mathrm{Tr}(U+U^\dagger)) \right\rangle_{U(l)}.$$

Recent advances characterize this entire class of integrals in terms of both first-order matrix linear differential equations (for vector-valued generating functions of size $l+1$) and scalar linear ordinary differential equations of order $l+1$. These linear characterizations are computationally superior and conceptually distinct from earlier, nonlinear approaches (such as those using the $\sigma$–Painlevé III$'$ system). The methodology also generalizes to $\beta$-ensembles, unifying classical and deformed unitary averages within the same algebraic framework.

1. Linear Differential Equation Characterizations

(a) Matrix First-Order System

For the average

$$I_{l,q}(s) = \left\langle (\det U)^q \exp\left(s^{1/2}\mathrm{Tr}(U+U^\dagger)\right) \right\rangle_{U(l)},$$

the integral can be embedded in a vector-valued function $\mathbf{H}(x)$ of length $l+1$. The components satisfy a bidiagonal system (cf. Proposition 3.1): $$(n-p) x \hat{H}_{p+1}(x) = ((n-p)x + \tilde{B}_p) \hat{H}_p(x) + x \frac{d}{dx} \hat{H}_p(x) - \tilde{D}_p \hat{H}_{p-1}(x)$$ for $p=0,\dots,l$, with explicitly computable coefficients $\tilde{B}_p$, $\tilde{D}_p$ (the parameters may encode $q$ and $l$).

This is compactly rendered as a matrix first-order ordinary differential equation: $$x \frac{d}{dx} \hat{\mathbf{H}}(x) = (x X + Y) \hat{\mathbf{H}}(x),$$ where $X$ and $Y$ are sparse $(l+1)\times(l+1)$ matrices derived from the recurrences above. The first component of $\hat{\mathbf{H}}(x)$ is $I_{l,q}(s)$ up to normalization and rescaling.

(b) Scalar Higher-Order Differential Equation

By recursive elimination—using the recurrences described—the system above yields a single scalar equation for $I_{l,q}(s)$: $$x^l \frac{d^{l+1} y}{dx^{l+1}} + q_l(x) x^{l-1} \frac{d^{l} y}{dx^{l}} + q_{l-1}(x) x^{l-2} \frac{d^{l-1} y}{dx^{l-1}} + \dots + q_0(x) y = 0,$$ where the $q_j(x)$ are polynomials of degree $[(l-1 + j)/2]$. This (in principle) determines all Taylor coefficients about $x=0$ uniquely.

2. Algorithmic Power Series Solution and Enumerative Implications

The matrix ODE system enables efficient computation of the Taylor coefficients for $I_{l,q}(s)$. Let $\tilde{\mathbf{I}}(x) = \sum_k \mathbf{d}_k x^k$, then the series coefficients satisfy (Equation MD3a0)

$$(kI - \tilde{Y})\mathbf{d}_k = \tilde{X}\, \mathbf{d}_{k-1},$$

with initial condition $\mathbf{d}_0$. This is a backward substitution in a bidiagonal matrix and is highly efficient (linear in the maximum order). Thus, even for large $l$, this method produces long expansions, necessary for probing enumerative asymptotics and random matrix predictions.

Example Applications

• Longest Increasing Subsequence Problem: For $q=0$, coefficients $T_l(N)$ in

$1 + \sum_{N=1}^\infty \frac{T_l(N)}{(N!)^2} z^{2N}$

enumerate the number of permutations of $N$ elements whose longest increasing subsequence is at most $l$.

• Moments of $\zeta'(1/2+it)$: For $q=l$, the corresponding Hankel determinant describes leading-order moments of the first and second derivatives of the Riemann zeta function on the critical line, via analogy to characteristic polynomial derivatives in the unitary group.

3. Comparison with Nonlinear $\sigma$–Painlevé III′ Characterization

Earlier approaches characterize $I_{l,q}(s)$ via the logarithmic derivative $u(s) = s \frac{d}{ds} \log I_{l,q}(s)$, which satisfies a nonlinear equation of the $\sigma$–Painlevé III$'$ family: $(su'')^2 + u'(4u' - 1)(u - su') = 0$ with specific boundary conditions ($$u(s) = -s^{1+l}/(2^{2(1+l)}\Gamma(1+l)\Gamma(2+l)) + O(s^{2+l})$$ for small $s$).

However, this nonlinear ODE does not uniquely specify the power series coefficients beyond order $s^{2l}$ without additional data, hampering its utility in enumerative and moment computation. By contrast, the linear ODEs provide a unique and direct recursion for all coefficients, yielding a conceptually and computationally superior approach.

4. $\beta$-Deformation and General Ensembles

The construction extends to circular $\beta$-ensembles (CE$_{\beta,l}$), where the eigenvalue distribution has the probability density

$$\frac{1}{(2\pi)^l C_{\beta,l}} \prod_{1 \leq j < k \leq l} |e^{i\theta_k} - e^{i\theta_j}|^\beta.$$

The average

$$\langle (\det U)^q \exp(s^{1/2} \mathrm{Tr}(U + U^\dagger)) \rangle_{\text{CE}_{\beta,l}}$$

can then be written (Equation 2.9a) in terms of a Jack polynomial hypergeometric function: $$\propto s^{ql/2} / (C_{\beta,l} B_l(q+1,2/\beta)) \cdot {}_0F_1^{(2/\beta)}(q + l; s^l),$$ with all coefficients and recursions adapted to $\beta$ via the Jack parameter $\alpha = 2 / \beta$. The same class of $(l+1)$-component first-order matrix ODEs and scalar ODEs of order $l+1$ applies, by adjusting for the deformation.

5. Computational and Structural Advantages

The linear differential equation framework possesses numerous advantages:

• Determinacy: The linear system uniquely fixes the full Taylor expansion of the unitary integral generating function, without recourse to extra boundary data or nonlinear inversion.
• Efficiency: The bidiagonal nature of the recurrence matrices yields $O(N)$ complexity for depth-$N$ expansions, enabling high precision enumerative and numerical work even for large $l$.
• Uniformity: The approach smoothly interpolates between the combinatorial enumeration regime ($q=0$) and the random-matrix and analytic number theory regime (moments of the derivatives of the zeta function, $q=l$).
• Generality: Extensions to circular $\beta$-ensembles are systematic via Jack polynomial theory, accommodating symmetric function deformations relevant for quantum transport and critical statistics.

6. Summary Table of Characterizations

Characterization Type	Differential Order	Uniqueness for Series	Suitability for Computation
Matrix ODE (vector, size $l+1$)	1	Yes	High
Scalar ODE (single function)	$l+1$	Yes	Efficient for moderate $l$
Nonlinear $\sigma$–Painlevé III′	2	No (series ambiguity)	Poor for enumeration

7. Significance and Applications

These differential equation characterizations have far-reaching consequences:

• They provide a rigorous, algorithmic foundation for generating explicit enumeration results involving longest increasing subsequences.
• They underlie refined asymptotic analyses of moments of characteristic polynomial derivatives—quantities conjectured to model analogous moments for Riemann zeta derivatives on the critical line.
• By encompassing circular $\beta$-ensembles, the methodology offers a path to unified treatment of symmetric function theory, combinatorial enumeration, and integrable probability in quantum chaos, statistical mechanics, and analytic number theory.

The embedding of classical unitary group averages within a common linear ODE framework, compatible with $\beta$-deformation and Jack polynomial expansions, marks a significant structural advance in the theory of special function solutions to random matrix averages and related enumerative problems (Forrester et al., 28 Aug 2025).