Rotated semicircle laws for permanental roots of Gaussian random matrices

Technical Summary: Rotated Semicircle Laws for Permanental Roots of Gaussian Random Matrices

Problem Statement
The paper addresses the asymptotic distribution of the "permanental roots" of random matrices. For an $N \times N$ matrix $A$, the permanental characteristic polynomial is defined as $P_A(z) = \text{Per}(zI_N - A)$, where $\text{Per}$ denotes the permanent. Unlike the determinant, the permanent lacks a simple combinatorial cancellation structure, making the analysis of its roots significantly more challenging.

The authors focus on matrices $H_N$ drawn from the standard Gaussian Orthogonal Ensemble (GOE, $\beta=1$) and Gaussian Unitary Ensemble (GUE, $\beta=2$). While the empirical eigenvalue measures of these matrices converge to the standard Wigner semicircle law on the real interval $[-2, 2]$, the behavior of the roots of $P_{H_N}(z)$ was previously conjectured by Fyodorov [3]. Fyodorov proposed that these roots, which are generally complex, concentrate on the imaginary axis and follow a semicircle law rotated by $\pi/2$. The primary goal of this work is to prove this conjecture almost surely.

Methodology
The proof strategy combines exact algebraic identities specific to Gaussian ensembles with tools from potential theory and orthogonal polynomials.

1. Duality and Integral Representations:
   The authors utilize Fyodorov's $n$-point duality formulas to express the expectation of products of permanental polynomials as integrals over smaller matrix spaces. Specifically, for $\mu_1, \mu_2 \in \mathbb{C}$, the expectations $E[P_N(\mu_1)P_N(\mu_2)]$ are reduced to integrals over $2 \times 2$ Hermitian matrices (for GUE) and a mix of Hermitian and complex variables (for GOE).
   By specializing to $\mu_1 = ix$ and $\mu_2 = -ix$, these high-dimensional expectations are transformed into two-dimensional scalar Gaussian integrals involving terms of the form $(x - iy_1)^N(x - iy_2)^N$ weighted by a repulsive factor $|y_1 - y_2|^\alpha$ (where $\alpha = 4/\beta$).

2. $L^2$ Norm Analysis:
   The authors define a monic polynomial $R_N(z) = i^{-N}P_N(iz)$. They calculate the expected squared $L^2$ norm of $R_N$ with respect to the Gaussian weight $e^{-Nx^2/2}$ on the real line. Using the integral representations derived in the previous step, they establish that:
   $$E\|R_N\|_N^2 = \kappa_{\beta,N} h_N$$
   where $h_N$ is the squared norm of the scaled monic Hermite polynomial, and $\kappa_{\beta,N}$ is a polynomial factor in $N$ (specifically $\kappa_{1,N} \sim N/2$ and $\kappa_{2,N} \sim N$).

3. Potential Theory and Zero Distribution:
   The core of the convergence proof relies on the theory of weighted Chebyshev polynomials and zero distribution.

   • Lower Bound: Lemma 4.1 establishes that for any monic polynomial $M$ of degree $N$, $\|M\|_N^2 \ge h_N$, with equality only for the scaled Hermite polynomial.
   • Upper Bound: Theorem 4.4 demonstrates that if a sequence of random monic polynomials satisfies $E\|M_N\|_N^2 \le A N^m h_N$ for constants $A, m$, then their normalized zero counting measures converge weakly to the standard semicircle law $\mu_{sc}$ almost surely. This is achieved via the Borel–Cantelli lemma and the relationship between the $L^2$ norm and the weighted supremum norm (via Lemma 4.2).

Key Contributions and Results

• Proof of Fyodorov's Conjecture: The paper rigorously proves that for both GOE and GUE, the normalized zero counting measure $\nu_{P_N}$ of the permanental characteristic polynomial converges almost surely to the push-forward of the standard semicircle law under the map $z \mapsto iz$.
  $$\nu_{P_N} \xrightarrow{w} (m_i)_\# \mu_{sc}$$
  where $m_i(z) = iz$. This confirms that the roots are asymptotically distributed on the imaginary axis according to the density $\frac{1}{2\pi}\sqrt{4-x^2}$.
• Exact Norm Identities: The authors derive exact expressions for the expected $L^2$ norms of permanental polynomials in terms of Hermite polynomial norms, bridging the gap between the combinatorial complexity of permanents and the analytic properties of orthogonal polynomials.
• Almost Sure Convergence: Unlike previous approaches that relied on expected logarithmic potentials (which require justifying the interchange of limits and expectations), this work establishes almost sure convergence directly through norm bounds and Borel–Cantelli arguments.

Significance and Limitations
The paper claims to resolve the global limiting distribution problem for permanental roots of Gaussian ensembles, a problem initiated by Fyodorov. The significance lies in providing a rigorous foundation for the "rotated semicircle law" in the context of permanents, which are computationally hard (#P-complete) and lack the algebraic simplifications of determinants.

The authors explicitly note the limitations and open problems:

• Universality: The proof relies on exact identities specific to Gaussian ensembles. Extending this result to non-Gaussian Wigner matrices remains an open universality problem.
• Local Statistics: The paper does not address local statistics (e.g., sine-kernel universality in the bulk or Airy-kernel behavior at the edges) or the fluctuations of the extremal roots (Tracy–Widom laws).
• Quantitative Confinement: While the global distribution is established, the paper does not provide quantitative estimates for the transverse displacement of roots from the imaginary axis or the overshoot beyond the limiting edges.

In summary, the work provides a definitive proof of the global limiting law for permanental roots in the Gaussian case, utilizing a novel combination of duality identities and weighted polynomial approximation theory, while leaving local and universality questions for future investigation.