Kamien, Politzer, and Wise (1988) proved that Hermitian random-matrix ensembles with probability weights that are smooth on the scale of the mean level spacing — not necessarily Gaussian, independent, or invariant under global orthogonal/unitary conjugation — exhibit the universal sine-kernel correlations of eigenvalues in the bulk of the spectrum for the unitary class (Dyson index β=2\beta=2). They explicitly presumed the analogous statement for the orthogonal (β=1\beta=1) and symplectic (β=4\beta=4) classes, conjectured it for a continuous range of β\beta including 1 and 4, argued that adding a fixed matrix of intermediate strength preserves the smoothness of the ensemble, and observed that eigenvector-dependent statistics (e.g., those governing nuclear level widths) are not universal in this sense. The β=1\beta=1, β=4\beta=4, and general-β\beta bulk universality has since been rigorously proved for the standard classes (invariant ensembles with analytic one-cut potentials, Wigner matrices, deformed Wigner matrices, and log-gases (invariant ensembles whose eigenvalue weight is one-body, S[E]=−βN∑iV(Ei)S[E]=-\beta N\sum_i V(E_i) with potential VV) at the gap-distribution level), results that do not rest on the 1988 method's compensating-construction and kernel-decay limitations. A precise open core survives at the source's own generality: kernel-level universality for symplectic ensembles with general analytic potentials, non-analytic and many-body smooth weights, correlated matrix elements, and, separately, a systematic theory of eigenvector statistics for symmetry-breaking ensembles.

Background

Random matrix theory models the statistics of energy levels of complex quantum systems (nuclei, atoms, molecules, electrons in disordered conductors) by treating the Hamiltonian as a large random Hermitian matrix. Wigner's Gaussian ensembles — the Gaussian orthogonal, unitary, and symplectic ensembles (GOE, GUE, GSE) for matrices whose independent entries are real, complex, or quaternionic — are labeled by Dyson's index β=1,2,4\beta=1,2,4, which also appears as the power of the level-repulsion Jacobian ∏i<j∣Ei−Ej∣β\prod_{i<j}|E_i-E_j|^{\beta} when the eigenvalue measure is written in terms of the eigenvalues EiE_i. Spectral statistics are encoded in the nn-level correlation functions Rn(x1,…,xn)R_n(x_1,\dots,x_n), the probability densities of finding levels at nn specified energies; for the GUE these are determinants of a kernel K(x,y)K(x,y), which in the bulk of the spectrum — after unfolding, i.e., rescaling energies by the local mean density ρ(x)\rho(x) so the mean spacing is 1 — converges to the sine kernel sin⁡(π(x−y))/(π(x−y))\sin(\pi(x-y))/(\pi(x-y)). For β=1\beta=1 and β=4\beta=4 the bulk limits are Pfaffian analogues built from the same sine kernel: the correlation functions become Pfaffians (the antisymmetric-matrix analogue of a determinant, Pf(A)2=det⁡A\mathrm{Pf}(A)^2=\det A) of matrices of sine kernels rather than determinants. One may equivalently study the gap (spacing) distribution, the law of the difference between consecutive unfolded levels. Universality is the statement, supported by nuclear data, spectra of quantum systems with chaotic classical limits, and numerics, that these local bulk statistics do not depend on the detailed ensemble whenever the level density varies slowly on the scale of the mean spacing.

Before 1988 universality under smooth changes of the eigenvalue measure was established essentially only in the β=2\beta=2 case and for special (Gaussian or similarly rigid) ensembles. Kamien, Politzer, and Wise considered general Hermitian ensembles whose probability weight on the matrix elements is smooth on the scale of the mean level spacing — not necessarily Gaussian, not independent, and not invariant under a global orthogonal or unitary conjugation — and, after integrating over the diagonalizing rotations, reduced the eigenvalue measure to the repulsion Jacobian times a smooth symmetric function S[E]S[E] that may depend on many eigenvalues at once (their worked perturbation is a three-level potential h(Ei,Ej,Ek)h(E_i,E_j,E_k)). The most widely studied special case is the log-gas, an invariant ensemble whose weight is one-body, S[E]=−βN∑iV(Ei)S[E]=-\beta N\sum_i V(E_i) with a confining potential VV; the 1988 setting is strictly broader, since SS need not factorize over eigenvalues. For β=2\beta=2 they proved that all such ensembles have the universal sine-kernel bulk correlations up to O(1/N)O(1/N) corrections. The proof is perturbative: a compensating potential u′u', fixed by a mean-field equation, is added so that the perturbation does not change the mean density; the authors note that without this compensation naive power counting produces factors LnL^n (with LL the smoothness length in units of the spacing) at order nn, making the method insufficient for smooth potential changes in general, and they assume rather than prove that the kernel decays like ∣z−z′∣−1|z-z'|^{-1} outside the local universal regime. They explicitly presumed, but did not prove, the analogous results for β=1\beta=1 and β=4\beta=4, and conjectured them for a continuous range of β\beta including 1 and 4. They further argued that adding a single fixed matrix ϵH′\epsilon H' to every member of a smooth ensemble — with ϵ\epsilon in the intermediate regime where the elements of ϵH′\epsilon H' are comparable to those of a typical ensemble member — typically induces a smooth new eigenvalue weight, so that Wigner's statistics persist, and they observed that eigenvector statistics, which control observables such as nuclear level widths, are not generic in this sense: a fixed ϵH′\epsilon H' picks out preferred directions and destroys the orthogonal or unitary symmetry. The question is what remains to be proved of this universality paradigm at its stated generality.

Problem Statement

Prove or refute, at the full generality of the 1988 claim, that Wigner–Dyson bulk statistics are universal: for Hermitian random-matrix ensembles in the orthogonal (β=1\beta=1) and symplectic (β=4\beta=4) symmetry classes — and, as conjectured by the source, for a continuous range of β\beta including β=1\beta=1 and β=4\beta=4 — whose probability weight on the matrix elements is smooth on the scale of the local mean level spacing (so that, after integrating over the diagonalizing rotations, the eigenvalue measure is ∏i<j∣Ei−Ej∣βeS[E]\prod_{i<j}|E_i-E_j|^{\beta} e^{S[E]} with SS a smooth symmetric function of the eigenvalues, allowed to depend on many eigenvalues at once; for β≠2\beta \neq 2 the formulation is at the eigenvalue-measure level, with β\beta the strength of the pairwise logarithmic repulsion; the matrix-element distributions are not required to be Gaussian, independent, or invariant under global orthogonal or unitary conjugation), where smoothness is understood in the source's sense: the variation of SS (equivalently of the weight) and of the induced mean level density ρ\rho is characterized by a long length LL in units of the local mean spacing, with N/L=O(1)N/L = O(1) as N→∞N \to \infty, N=∫ρN = \int \rho being the number of levels within the scale of variation; show that, under the source's conditions (1) ρ−2dρ/dx≪1\rho^{-2} d\rho/dx \ll 1, ensuring the existence of intervals of approximately constant density containing many levels, (2) the nn correlation arguments lie within an interval small compared with the scale of variation of SS and ρ\rho, and (3) energies are unfolded in units of the local ρ\rho, the nn-level correlation functions converge in the limit of matrix size N→∞N \to \infty to the GOE form (for β=1\beta=1), the GSE form (for β=4\beta=4), and the corresponding Gaussian β\beta-ensemble (log-gas) correlation-function form for β\beta in that conjectured continuous range, with the same O(1/N)O(1/N)-type deviation control as the 1988 β=2\beta=2 proof; and correspondingly establish, or delimit by explicit counterexample, the persistence of these statistics when a fixed deterministic matrix ϵH′\epsilon H' whose elements are comparable to those of a typical ensemble member is added to every member of the ensemble. The requirements just listed — universality for β=1\beta=1, universality for β=4\beta=4, universality across the continuous range of β\beta, and persistence under the fixed ϵH′\epsilon H' — are not mutually independent: the β=1\beta=1 and β=4\beta=4 cases are the corresponding instances of the continuous-range family, so a proof of the continuous-β\beta statement resolves the β=1\beta=1 and β=4\beta=4 requirements as well, and a counterexample at β=1\beta=1 or β=4\beta=4 refutes the continuous-β\beta claim too. The persistence requirement under the fixed ϵH′\epsilon H' stands or falls on its own.

References

  1. Randall D. Kamien, H. David Politzer, and Mark B. Wise, 'Universality of Random-Matrix Predictions for the Statistics of Energy Levels', Physical Review Letters 60, no. 20, 1995–1998 (1988), DOI: 10.1103/PhysRevLett.60.1995, https://doi.org/10.1103/PhysRevLett.60.1995 (open-access scanned full text: Caltech Authors repository, https://authors.library.caltech.edu/records/0zqh8-crh75)
  2. L. A. Pastur, 'On the universality of the level spacing distribution for some ensembles of random matrices', Letters in Mathematical Physics 25, no. 4, 259–265 (1992), DOI: 10.1007/BF00398398, https://doi.org/10.1007/BF00398398
  3. A. D. Mirlin and Y. V. Fyodorov, 'Universality of level correlation function of sparse random matrices', Journal of Physics A: Mathematical and General 24, no. 10, 2273–2286 (1991), DOI: 10.1088/0305-4470/24/10/016, https://doi.org/10.1088/0305-4470/24/10/016
  4. E. Brézin and A. Zee, 'Universality of the correlations between eigenvalues of large random matrices', Nuclear Physics B 402, 613–627 (1993), DOI: 10.1016/0550-3213(93)90121-5, https://doi.org/10.1016/0550-3213(93)90121-5
  5. L. Pastur and M. Shcherbina, 'Universality of the local eigenvalue statistics for a class of unitary invariant random matrix ensembles', Journal of Statistical Physics 86, 109–147 (1997), DOI: 10.1007/BF02180200, https://doi.org/10.1007/BF02180200
  6. P. Deift, T. Kriecherbauer, K. T.-R. McLaughlin, S. Venakides, and X. Zhou, 'Uniform asymptotics for polynomials orthogonal with respect to varying exponential weights and applications to universality questions in random matrix theory', Communications on Pure and Applied Mathematics 52, no. 11, 1335–1425 (1999), DOI: 10.1002/(sici)1097-0312(199911)52:11<1335::aid-cpa1>3.0.co;2-1, https://doi.org/10.1002/(sici)1097-0312(199911)52:11<1335::aid-cpa1>3.0.co;2-1
  7. L. Pastur and M. Shcherbina, 'Bulk Universality and Related Properties of Hermitian Matrix Models', Journal of Statistical Physics 130, 205–250 (2007), DOI: 10.1007/s10955-007-9434-6, https://doi.org/10.1007/s10955-007-9434-6
  8. P. Deift and D. Gioev, 'Universality in Random Matrix Theory for orthogonal and symplectic ensembles', International Mathematics Research Papers, article rpm004 (published online 2010), DOI: 10.1093/imrp/rpm004, https://doi.org/10.1093/imrp/rpm004 (preprint: arXiv:math-ph/0411075, https://arxiv.org/abs/math-ph/0411075)
  9. P. Deift, D. Gioev, T. Kriecherbauer, and M. Vanlessen, 'Universality for Orthogonal and Symplectic Laguerre-Type Ensembles', Journal of Statistical Physics 129, 949–1053 (2007), DOI: 10.1007/s10955-007-9325-x, https://doi.org/10.1007/s10955-007-9325-x
  10. M. Shcherbina, 'On Universality for Orthogonal Ensembles of Random Matrices', Communications in Mathematical Physics 285, 957–974 (2009; published online 2008), DOI: 10.1007/s00220-008-0648-5, https://doi.org/10.1007/s00220-008-0648-5
  11. Percy Deift and Dimitri Gioev, 'Random Matrix Theory: Invariant Ensembles and Universality', Courant Lecture Notes in Mathematics 18, American Mathematical Society (2009), DOI: 10.1090/cln/018, https://doi.org/10.1090/cln/018
  12. Terence Tao and Van Vu, 'Random matrices: Universality of local eigenvalue statistics', Acta Mathematica 206, 127–204 (2011), DOI: 10.1007/s11511-011-0061-3, https://doi.org/10.1007/s11511-011-0061-3
  13. Paul Bourgade, László Erdős, and Horng-Tzer Yau, 'Universality of general β-ensembles', Duke Mathematical Journal 163, no. 6, 1127–1190 (2014), DOI: 10.1215/00127094-2649752, https://doi.org/10.1215/00127094-2649752
  14. Paul Bourgade, László Erdős, and Horng-Tzer Yau, 'Bulk universality of general β-ensembles with non-convex potential', Journal of Mathematical Physics 53, 095221 (2012), DOI: 10.1063/1.4751478, https://doi.org/10.1063/1.4751478
  15. László Erdős and Horng-Tzer Yau, 'Gap Universality of Generalized Wigner and β-Ensembles', Journal of the European Mathematical Society 17, no. 8, 1927–2036 (2015), DOI: 10.4171/JEMS/548, https://doi.org/10.4171/JEMS/548
  16. Ji Oon Lee, Kevin Schnelli, Ben Stetler, and Horng-Tzer Yau, 'Bulk universality for deformed Wigner matrices', The Annals of Probability 44, no. 3, 2349–2425 (2016), DOI: 10.1214/15-AOP1023, https://doi.org/10.1214/15-AOP1023
  17. L. Benigni, 'Eigenvectors distribution and quantum unique ergodicity for deformed Wigner matrices', Annales de l'Institut Henri Poincaré, Probabilités et Statistiques 56, no. 4, 2822–2867 (2020), DOI: 10.1214/20-AIHP1060, https://doi.org/10.1214/20-AIHP1060