Dyson Index Research Articles

Phase transitions in the condition-number distribution of Gaussian random matrices.

We study the statistics of the condition number κ=λ_{max}/λ_{min} (the ratio between largest and smallest squared singular values) of N×M Gaussian random matrices. Using a Coulomb fluid technique, we derive analytically and for large N the cumulative P(κ<x) and tail-cumulative P(κ>x) distributions of κ. We find that these distributions decay as P(κ<x)≈exp[-βN^{2}Φ_{-}(x)] and P(κ>x)≈exp[-βNΦ_{+}(x)], where β is the Dyson index of the ensemble. The left and right rate functions Φ_{±}(x) are independent of β and calculated exactly for any choice of the rectangularity parameter α=M/N-1>0. Interestingly, they show a weak nonanalytic behavior at their minimum 〈κ〉 (corresponding to the average condition number), a direct consequence of a phase transition in the associated Coulomb fluid problem. Matching the behavior of the rate functions around 〈κ〉, we determine exactly the scale of typical fluctuations ∼O(N^{-2/3}) and the tails of the limiting distribution of κ. The analytical results are in excellent agreement with numerical simulations.

Spectral order statistics of Gaussian random matrices: large deviations for trapped fermions and associated phase transitions.

We compute the full order statistics of a one-dimensional gas of spinless fermions (or, equivalently, hard bosons) in a harmonic trap at zero temperature, including its large deviation tails. The problem amounts to computing the probability distribution of the kth smallest eigenvalue λ(k) of a large dimensional Gaussian random matrix. We find that this probability behaves for large N as P[λ(k)=x]≈exp[-βN(2)ψ(k/N,x)], where β is the Dyson index of the ensemble. The rate function ψ(c,x), computed explicitly as a function of x in terms of the intensive label c=k/N, has a quadratic behavior modulated by a weak logarithmic singularity at its minimum. This is shown to be related to phase transitions in the associated Coulomb gas problem. The connection with statistics of extreme eigenvalues and order stastistics of random matrices is also discussed. We find that, as a function of c and keeping the value of x fixed, the rate function ψ(c,x) describes the statistics of the shifted index number, generalizing known results on its typical fluctuations; as a function of x and keeping the fraction c=k/N fixed, the rate function ψ(c,x) also describes the statistics of the kth eigenvalue in the bulk, generalizing as well the results on its typical fluctuations. Moreover, for k=1 (respectively, for k=N), the rate function captures both the fluctuations to the left and to the right of the typical value of λ(1) (respectively, of λ(N)).

Dirac spectra of two-dimensional QCD-like theories

We analyze Dirac spectra of two-dimensional QCD like theories both in the continuum and on the lattice and classify them according to random matrix theories sharing the same global symmetries. The classification is different from QCD in four dimensions because the anti-unitary symmetries do not commute with $\gamma_5$. Therefore in a chiral basis, the number of degrees of freedom per matrix element are not given by the Dyson index. Our predictions are confirmed by Dirac spectra from quenched lattice simulations for QCD with two or three colors with quarks in the fundamental representation as well as in the adjoint representation. The universality class of the spectra depends on the parity of the number of lattice points in each direction. Our results show an agreement with random matrix theory that is qualitatively similar to the agreement found for QCD in four dimensions. We discuss the implications for the Mermin-Wagner-Coleman theorem and put our results in the context of two-dimensional disordered systems.

Universal covariance formula for linear statistics on random matrices.

We derive an analytical formula for the covariance cov(A,B) of two smooth linear statistics A=[under ∑]ia(λ_{i}) and B=[under ∑]ib(λ_{i}) to leading order for N→∞, where {λ_{i}} are the N real eigenvalues of a general one-cut random-matrix model with Dyson index β. The formula, carrying the universal 1/β prefactor, depends on the random-matrix ensemble only through the edge points [λ_{-},λ_{+}] of the limiting spectral density. For A=B, we recover in some special cases the classical variance formulas by Beenakker and by Dyson and Mehta, clarifying the respective ranges of applicability. Some choices of a(x) and b(x) lead to a striking decorrelation of the corresponding linear statistics. We provide two applications-the joint statistics of conductance and shot noise in ideal chaotic cavities, and some new fluctuation relations for traces of powers of random matrices.

Phase transitions and edge scaling of number variance in Gaussian random matrices.

We consider N × N Gaussian random matrices, whose average density of eigenvalues has the Wigner semicircle form over [-√2],√2]. For such matrices, using a Coulomb gas technique, we compute the large N behavior of the probability P(N,L)(N(L)) that N(L) eigenvalues lie within the box [-L,L]. This probability scales as P(N,L)(N(L) = κ(L)N) ≈ exp(-βN(2)ψ(L)(κ(L))), where β is the Dyson index of the ensemble and ψ(L)(κ(L)) is a β-independent rate function that we compute exactly. We identify three regimes as L is varied: (i) N(-1)≪L < √2 (bulk), (ii) L∼√2 on a scale of O(N(-2/3)) (edge), and (iii) L > sqrt[2] (tail). We find a dramatic nonmonotonic behavior of the number variance V(N)(L) as a function of L: after a logarithmic growth ∝ln(NL) in the bulk (when L∼O(1/N)), V(N)(L) decreases abruptly as L approaches the edge of the semicircle before it decays as a stretched exponential for L > sqrt[2]. This "dropoff" of V(N)(L) at the edge is described by a scaling function V(β) that smoothly interpolates between the bulk (i) and the tail (iii). For β = 2 we compute V(2) explicitly in terms of the Airy kernel. These analytical results, verified by numerical simulations, directly provide for β = 2 the full statistics of particle-number fluctuations at zero temperature of 1D spinless fermions in a harmonic trap.

The Interpolating Airy Kernels for the $$\beta =1$$ β = 1 and $$\beta =4$$ β = 4 Elliptic Ginibre Ensembles

We consider two families of non-Hermitian Gaussian random matrices, namely the elliptical Ginibre ensembles of asymmetric N-by-N matrices with Dyson index beta=1 (real elements) and with beta=4 (quaternion-real elements). Both ensembles have already been solved for finite N using the method of skew-orthogonal polynomials, given for these particular ensembles in terms of Hermite polynomials in the complex plane. In this paper we investigate the microscopic weakly non-Hermitian large-N limit of each ensemble in the vicinity of the largest or smallest real eigenvalue. Specifically, we derive the limiting matrix-kernels for each case, from which all the eigenvalue correlation functions can be determined. We call these new kernels the "interpolating" Airy kernels, since we can recover -- as opposing limiting cases -- not only the well-known Airy kernels for the Hermitian ensembles, but also the complementary error function and Poisson kernels for the maximally non-Hermitian ensembles at the edge of the spectrum. Together with the known interpolating Airy kernel for beta=2, which we rederive here as well, this completes the analysis of all three elliptical Ginibre ensembles in the microscopic scaling limit at the spectral edge.

Index distribution of Cauchy random matrices

Using a Coulomb gas technique, we compute analytically the probability that a large N × N Cauchy random matrix has N+ positive eigenvalues, where N+ is called the index of the ensemble. We show that this probability scales for large N as , where β is the Dyson index of the ensemble. The rate function ψC(κ) is computed in terms of single integrals that are easily evaluated numerically and amenable to an asymptotic analysis. We find that the rate function, around its minimum at κ = 1/2, has a quadratic behavior modulated by a logarithmic singularity. As a consequence, the variance of the index scales for large N as Var(N+) ∼ σCln N, where σC = 2/(βπ2) is twice as large as the corresponding prefactor in the Gaussian and Wishart cases. The analytical results are checked by numerical simulations and against an exact finite N formula which, for β = 2, can be derived using orthogonal polynomials.

A Real Quaternion Spherical Ensemble of Random Matrices

One can identify a tripartite classification of random matrix ensembles into geometrical universality classes corresponding to the plane, the sphere and the anti-sphere. The plane is identified with Ginibre-type (iid) matrices and the anti-sphere with truncations of unitary matrices. This paper focusses on an ensemble corresponding to the sphere: matrices of the form Y=A −1 B, where A and B are independent N×N matrices with iid standard Gaussian real quaternion entries. By applying techniques similar to those used for the analogous complex and real spherical ensembles, the eigenvalue joint probability density function and correlation functions are calculated. This completes the exploration of spherical matrices using the traditional Dyson indices β=1,2,4. We find that the eigenvalue density (after stereographic projection onto the sphere) has a depletion of eigenvalues along a ring corresponding to the real axis, with reflective symmetry about this ring. However, in the limit of large matrix dimension, this eigenvalue density approaches that of the corresponding complex ensemble, a density which is uniform on the sphere. This result is in keeping with the spherical law (analogous to the circular law for iid matrices), which states that for matrices having the spherical structure Y=A −1 B, where A and B are independent, iid matrices the (stereographically projected) eigenvalue density tends to uniformity on the sphere.

Large deviations of the top eigenvalue of large Cauchy random matrices

We compute analytically the large deviation tails of the probability density function (pdf) of the top eigenvalue λmax in rotationally invariant and heavy-tailed Cauchy ensembles of N × N matrices for any Dyson index β > 0, where β = 1, 2, 4 correspond, respectively, to orthogonal, unitary and symplectic ensembles. Furthermore, we show that these large deviation tails flank a central non-Gaussian regime for on both sides. By matching these tails with the central regime, we obtain the exact leading asymptotic behaviors for any β of the pdf in the central regime, which generalizes the Tracy–Widom distribution known for Gaussian ensembles. Our analytical results are confirmed by numerical simulations.

Invariant β-Wishart ensembles, crossover densities and asymptotic corrections to the Marčenko–Pastur law

We construct a diffusive matrix model for the β-Wishart (or Laguerre) ensemble for general continuous β ∈ [0, 2], which preserves invariance under the orthogonal/unitary group transformation. Scaling the Dyson index β with the largest size M of the data matrix as β = 2c/M (with c a fixed positive constant), we obtain a family of spectral densities parametrized by c. As c is varied, this density interpolates continuously between the Marčenko–Pastur (c → ∞ limit) and the Gamma law (c → 0 limit). Analyzing the full Stieltjes transform (resolvent) equation, we obtain as a byproduct the correction to the Marčenko–Pastur density in the bulk up to order 1/M for all β and up to order 1/M2 for the particular cases β = 1, 2.

Number of Relevant Directions in Principal Component Analysis and Wishart Random Matrices

We compute analytically, for large N, the probability P(N+,N) that a N×N Wishart random matrix has N+ eigenvalues exceeding a threshold Nζ, including its large deviation tails. This probability plays a benchmark role when performing the principal component analysis of a large empirical data set. We find that P(N+,N)≈exp[-βN2ψζ(N+/N)], where β is the Dyson index of the ensemble and ψζ(κ) is a rate function that we compute explicitly in the full range 0≤κ≤1 and for any ζ. The rate function ψζ(κ) displays a quadratic behavior modulated by a logarithmic singularity close to its minimum κ⋆(ζ). This is shown to be a consequence of a phase transition in an associated Coulomb gas problem. The variance Δ(N) of the number of relevant components is also shown to grow universally (independent of ζ) as Δ(N)∼(βπ2)-1 lnN for large N.

Mixing of orthogonal and skew-orthogonal polynomials and its relation to Wilson RMT

The unitary Wilson random matrix theory is an interpolation between the chiral Gaussian unitary ensemble and the Gaussian unitary ensemble. This new way of interpolation is also reflected in the orthogonal polynomials corresponding to such a random matrix ensemble. Although the chiral Gaussian unitary ensemble and the Gaussian unitary ensemble are associated with the Dyson index β = 2, the intermediate ensembles exhibit a mixing of orthogonal polynomials and skew-orthogonal polynomials. We consider the Hermitian and the non-Hermitian Wilson random matrix and derive the corresponding polynomials, their recursion relations, Christoffel–Darboux-like formulas, Rodrigues formulas and representations as random matrix averages in a unifying way. With the help of these results, we derive the unquenched k-point correlation function of the Hermitian and the non-Hermitian Wilson random matrix in terms of two-flavor partition functions only. This representation is due to a Pfaffian factorization. It drastically simplifies the expressions, which can be easily numerically evaluated. It also serves as a good starting point for studying the Wilson–Dirac operator in the ϵ-regime of lattice quantum chromodynamics.

Surprising Pfaffian factorizations in random matrix theory with Dyson index β = 2

In recent decades, determinants and Pfaffians were found for eigenvalue correlations of various random matrix ensembles. These structures simplify the average over a large number of ratios of characteristic polynomials to integrations over one and two characteristic polynomials only. Up to now it was thought that determinants occur for ensembles with Dyson index β = 2, whereas Pfaffians only for ensembles with β = 1, 4. We derive a non-trivial Pfaffian determinant for β = 2 random matrix ensembles which is similar to the one for β = 1, 4. Thus, it unveils a hidden universality of this structure. We also give a general relation between the orthogonal polynomials related to the determinantal structure and the skew-orthogonal polynomials corresponding to the Pfaffian. As a particular example, we consider the chiral unitary ensembles in great detail.

How many eigenvalues of a Gaussian random matrix are positive?

We study the probability distribution of the index N(+), i.e., the number of positive eigenvalues of an N×N Gaussian random matrix. We show analytically that, for large N and large N(+) with the fraction 0≤c=N(+)/N≤1 of positive eigenvalues fixed, the index distribution P(N(+)=cN,N)~exp[-βN(2)Φ(c)] where β is the Dyson index characterizing the Gaussian ensemble. The associated large deviation rate function Φ(c) is computed explicitly for all 0≤c≤1. It is independent of β and displays a quadratic form modulated by a logarithmic singularity around c=1/2. As a consequence, the distribution of the index has a Gaussian form near the peak, but with a variance Δ(N) of index fluctuations growing as Δ(N)~lnN/βπ(2) for large N. For β=2, this result is independently confirmed against an exact finite-N formula, yielding Δ(N)=lnN/2π(2)+C+O(N(-1)) for large N, where the constant C for even N has the nontrivial value C=(γ+1+3ln2)/2π(2)≃0.185 248… and γ=0.5772… is the Euler constant. We also determine for large N the probability that the interval [ζ(1),ζ(2)] is free of eigenvalues. Some of these results have been announced in a recent letter [Phys. Rev. Lett. 103, 220603 (2009)].

Nonintersecting Brownian interfaces and Wishart random matrices

We study a system of N nonintersecting (1+1)-dimensional fluctuating elastic interfaces ("vicious bridges") at thermal equilibrium, each subject to periodic boundary condition in the longitudinal direction and in presence of a substrate that induces an external confining potential for each interface. We show that, for a large system and with an appropriate choice of the external confining potential, the joint distribution of the heights of the N nonintersecting interfaces at a fixed point on the substrate can be mapped to the joint distribution of the eigenvalues of a Wishart matrix of size N with complex entries (Dyson index beta=2), thus providing a physical realization of the Wishart matrix. Exploiting this analogy to random matrix, we calculate analytically (i) the average density of states of the interfaces, (ii) the height distribution of the uppermost and lowermost interfaces (extrema), and (iii) the asymptotic (large-N) distribution of the center of mass of the interfaces. In the last case, we show that the probability density of the center of mass has an essential singularity around its peak, which is shown to be a direct consequence of a phase transition in an associated Coulomb gas problem.

Superstatistical generalizations of Wishart–Laguerre ensembles of random matrices

Using Beck and Cohen's superstatistics, we introduce in a systematic way a family of generalized Wishart–Laguerre ensembles of random matrices with Dyson index β = 1, 2 and 4. The entries of the data matrix are Gaussian random variables whose variances η fluctuate from one sample to another according to a certain probability density f(η) and a single deformation parameter γ. Three superstatistical classes for f(η) are usually considered: χ2-, inverse χ2- and log-normal distributions. While the first class, already considered by two of the authors, leads to a power-law decay of the spectral density, we here introduce and solve exactly a superposition of Wishart–Laguerre ensembles with inverse χ2-distribution. The corresponding macroscopic spectral density is given by a γ-deformation of the semi-circle and Marčenko–Pastur laws, on a non-compact support with exponential tails. After discussing in detail the validity of Wigner's surmise in the Wishart–Laguerre class, we introduce a generalized γ-dependent surmise with stretched-exponential tails, which well approximates the individual level spacing distribution in the bulk. The analytical results are in excellent agreement with numerical simulations. To illustrate our findings we compare the χ2- and inverse χ2-classes to empirical data from financial covariance matrices.

Extended studies of separability functions and probabilities and the relevance of Dyson indices

We report substantial progress in the study of separability functions and their application to the computation of separability probabilities for the real, complex and quaternionic qubit–qubit and qubit–qutrit systems. We expand our recent work [P.B. Slater, J. Phys. A 39 (2006) 913], in which the Dyson indices of random matrix theory played an essential role, to include the use of not only the volume element of the Hilbert–Schmidt (HS) metric, but also that of the Bures (minimal monotone) metric as measures over these finite-dimensional quantum systems. Further, we now employ the Euler-angle parameterization of density matrices ( ρ ), in addition to the Bloore parameterization. The Euler-angle separability function for the minimally degenerate complex two-qubit states is well-fitted by the sixth-power of the participation ratio, R ( ρ ) = 1 Tr ρ 2 . Additionally, replacing R ( ρ ) by a simple linear transformation of the Verstraete–Audenaert–De Moor function [F. Verstraete, K. Audenaert, B.D. Moor, Phys. Rev. A 64 (2001) 012316], we find close adherence to Dyson-index behaviour for the real and complex (nondegenerate) two-qubit scenarios. Several of the analyses reported help to fortify our conjectures that the HS and Bures separability probabilities of the complex two-qubit states are 8 33 ≈ 0.242424 and 1680 ( 2 − 1 ) π 8 ≈ 0.733389 , respectively. Employing certain regularized beta functions in the role of Euler-angle separability functions, we closely reproduce–consistently with the Dyson-index ansatz–several HS two-qubit separability probability conjectures.

On invariant 2×2 β -ensembles of random matrices

We introduce and solve exactly a family of invariant 2×2 random matrices, depending on one parameter η , and we show that rotational invariance and real Dyson index β are not incompatible properties. The probability density for the entries contains a weight function and a multiple trace–trace interaction term, which corresponds to the representation of the Vandermonde-squared coupling on the basis of power sums. As a result, the effective Dyson index β eff of the ensemble can take any real value in an interval. Two weight functions (Gaussian and non-Gaussian) are explored in detail and the connections with β -ensembles of Dumitriu–Edelman and the so-called Poisson–Wigner crossover for the level spacing are respectively highlighted. A curious spectral twinning between ensembles of different symmetry classes is unveiled: as a consequence, the identification between symmetry group (orthogonal, unitary or symplectic) and the exponent of the Vandermonde ( β = 1 , 2 , 4 ) is shown to be potentially deceptive. The proposed technical tool more generically allows for designing actual matrix models which (i) are rotationally invariant; (ii) have a real Dyson index β eff ; (iii) have a pre-assigned confining potential or alternatively level-spacing profile. The analytical results have been checked through numerical simulations with an excellent agreement. Eventually, we discuss possible generalizations and further directions of research.

Dyson indices and Hilbert–Schmidt separability functions and probabilities

A confluence of numerical and theoretical results leads us to conjecture that the Hilbert–Schmidt separability probabilities of the 15- and 9-dimensional convex sets of complex and real two-qubit states (representable by 4 × 4 density matrices ρ) are and , respectively. Central to our reasoning are the modifications of two ansätze, recently advanced by Slater (2007 Phys. Rev. A 75 032326), involving incomplete beta functions Bν(a, b), where . We, now, set the separability function proportional to . Then, in the complex case—conforming to a pattern we find, manifesting the Dyson indices (β = 1, 2, 4) of random matrix theory—we take proportional to . We also investigate the real and complex qubit–qutrit cases. Now, there are two variables, , but they appear to remarkably coalesce into the product , so that the real and complex separability functions are again univariate in nature.

Finite volume chiral partition functions and the replica method

In the framework of chiral perturbation theory we demonstrate the equivalence of the supersymmetric and the replica methods in the symmetry breaking classes of Dyson indices \beta=1 and \beta=4. Schwinger-Dyson equations are used to derive a universal differential equation for the finite volume partition function in sectors of fixed topological charge, \nu. All dependence on the symmetry breaking class enters through the Dyson index \beta. We utilize this differential equation to obtain Virasoro constraints in the small mass expansion for all \beta and in the large mass expansion for \beta=2 with arbitrary \nu. Using quenched chiral perturbation theory we calculate the first finite volume correction to the chiral condensate demonstrating how, for all \betathere exists a region in which the two expansion schemes of quenched finite volume chiral perturbation theory overlap.