Off-diagonal observable elements from random matrix theory: distributions, fluctuations, and eigenstate thermalization

Kernel methods are an extremely popular set of techniques used for many important machine learning and data analysis applications. In addition to having good practical performance, these methods are supported by a well-developed theory. Kernel methods use an implicit mapping of the input data into a high dimensional feature space defined by a kernel function, i.e., a function returning the inner product between the images of two data points in the feature space. Central to any kernel method is the kernel matrix, which is built by evaluating the kernel function on a given sample dataset. In this paper, we initiate the study of non-asymptotic spectral properties of random kernel matrices. These are n x n random matrices whose (i,j)th entry is obtained by evaluating the kernel function on x_i and x_j, where x_1,..,x_n are a set of n independent random high-dimensional vectors. Our main contribution is to obtain tight upper bounds on the spectral norm (largest eigenvalue) of random kernel matrices constructed by using common kernel functions such as polynomials and Gaussian radial basis. As an application of these results, we provide lower bounds on the distortion needed for releasing the coefficients of kernel ridge regression under attribute privacy, a general privacy notion which captures a large class of privacy definitions. Kernel ridge regression is standard method for performing non-parametric regression that regularly outperforms traditional regression approaches in various domains. Our privacy distortion lower bounds are the first for any kernel technique, and our analysis assumes realistic scenarios for the input, unlike all previous lower bounds for other release problems which only hold under very restrictive input settings.

The eigenstate thermalization hypothesis (ETH) is one of the cornerstones of contemporary quantum statistical mechanics. The extent to which ETH holds for nonlocal operators is an open question that we partially address in this Letter. We report on the construction of highly nonlocal operators, behemoths, that are building blocks for various kinds of local and nonlocal operators. The behemoths have a singular distribution and width w∼D^{-1} (D being the Hilbert space dimension). From there, one may construct local operators with the ordinary Gaussian distribution and w∼D^{-1/2} in agreement with ETH. Extrapolation to even larger widths predicts sub-ETH behavior of typical nonlocal operators with w∼D^{-δ}, 0<δ<1/2. This operator construction is based on a deep analogy with random matrix theory and shows striking agreement with numerical simulations of nonintegrable many-body systems.

We study the transport through a resonant level coupled to two leads with the latter being described by Wigner’s random matrices. By taking appropriate thermodynamic limit before taking the long time limit, we obtain the stationary current as a function of voltage bias. The I–V curve is similar to that of single impurity Anderson model. On the other hand, the current matrix and initial density matrix in our model look like random matrices in the eigenbasis of Hamiltonian. They satisfy the description of eigenstate thermalization hypothesis (ETH) and nonequilibrium steady state hypothesis (NESSH), respectively. A statistical formula of current has been derived based on ETH and NESSH (Wang 2017 J. Stat. Mech. 093105). We check this formula in our model and find it to predict the stationary current to a high precision. The shape of I–V curve is explained by the peak structure in the characteristic function of NESSH, which is reminiscent of the transmission coefficient.

In this paper, we apply tools from random matrix theory (RMT) to estimates of correlations across the volatility of various assets in the S&P 500. The volatility inputs are estimated by modelling price fluctuations as a GARCH(1,1) process. The corresponding volatility correlation matrix is then constructed. It is found that the distribution of a significant number of eigenvalues of the volatility correlation matrix matches with the analytical result from RMT. Furthermore, the empirical estimates of short- and long-range correlations amongst eigenvalues, which are within RMT bounds, match with the analytical results for the Gaussian Orthogonal ensemble of RMT. To understand the information content of the largest eigenvectors, we estimate the contribution of the Global Industry Classification Standard industry groups to each eigenvector. In comparison with eigenvectors of correlation matrix for price fluctuations, only few of the largest eigenvectors of the volatility correlation matrix are dominated by a single industry group. We also study correlations between ‘volatility returns’ and log-volatility to find similar results.

We explicate the need and pathways for systematic stock market characterisation and development (SSMCD) and the role of Random Matrix Theory (RMT) in SSMCD research. This is the first time SSMCD, itself a nascent area of empirical finance introduced by the first author, is linked to RMT. Our focus is on the Nigerian Stock Market, particularly using RMT techniques to correlate respective asset prices in the NSM. The resulting insights are combined with those from related works on stochastic-times series analyses of stylised facts and six main market issues typically explored in empirical finance (efficiency, bubbles, anomalies, volatility, valuation and predictability), to illustrate SSMCD pathways in the NSM. Specifically, the RMT analyses focus on the cross-correlation matrix C of the stock index returns in the Nigerian Stock Market (NSM) from the period 2009 to 2013. Within this purview, we test the eigenvalues of the selected assets from the NSM and use their respective eigenvectors and inverse participation ratios to determine the stocks that drive the market dynamics. A method of obtaining a realistic implied correlation matrix for a hypothetical portfolio of some given assets selected from those considered in the empirical correlation matrix of the assets is considered. The positive implied correlation matrix shows that the corresponding assets in the NSM move in the same direction, meaning that portfolio diversification is not an optimal investment strategy. Hence, investing on derivative assets like call and put options is recommended. Further SSMCD implications of the analyses are foreshadowed. Also, we develop the links among SSMCD, macroeconomic modelling of national and global economic trends amid cycles of booms and crises. Highlights of an econome derived from analogies with sequencing and editing of genomes in molecular biology are provided.

Random matrix theory yields valuable insights into the universal features of quantum many-body chaotic systems. Although all-to-all interactions are traditionally studied, many interesting dynamical questions, such as transport of a conserved density, require a notion of spatially local interactions. We study the transport of the energy, the most basic conserved density, in few-body and 1D chains of nearest-neighbor random matrix terms that square to one, which we dub the “Haar-Ising” local random matrix model. In the few-body but large local Hilbert space dimension case, we develop a mapping for the energy dynamics to a single-particle hopping picture. This allows for the computation of the energy density autocorrelators and an out-of-time-ordered correlator of the energy density. In the 1D chain, we numerically study the energy transport for a small local Hilbert space dimension. Throughout, we discuss the density of states and touch upon the relation to free probability theory.

In contrast to the neatly bounded spectra of densely populated large random matrices, sparse random matrices often exhibit unbounded eigenvalue tails on the real and imaginary axis, called Lifshitz tails. In the case of asymmetric matrices, concise mathematical results have proved elusive. In this work, we present an analytical approach to characterising these tails. We exploit the fact that eigenvalues in the tail region have corresponding eigenvectors that are exponentially localised on highly-connected hubs of the network associated to the random matrix. We approximate these eigenvectors using a series expansion in the inverse connectivity of the hub, where successive terms in the series take into account further sets of next-nearest neighbours. By considering the ensemble of such hubs, we are able to characterise the eigenvalue density and the extent of localisation in the tails of the spectrum in a general fashion. As such, we classify a number of different asymptotic behaviours in the Lifshitz tails, as well as the leading eigenvalue and the inverse participation ratio. We demonstrate how an interplay between matrix asymmetry, network structure, and the edge-weight distribution leads to the variety of observed behaviours.

Eigenstate thermalization is widely accepted as the mechanism behind thermalization in generic isolated quantum systems. Using the example of a single magnetic defect embedded in the integrable spin-1/2 XXZ chain, we show that locally perturbing an integrable system can give rise to eigenstate thermalization. Unique to such setups is the fact that thermodynamic and transport properties of the unperturbed integrable chain emerge in properties of the eigenstates of the perturbed (nonintegrable) one. Specifically, we show that the diagonal matrix elements of observables in the perturbed eigenstates follow the microcanonical predictions for the integrable model, and that the ballistic character of spin transport in the integrable model is manifest in the behavior of the off-diagonal matrix elements of the current operator in the perturbed eigenstates.

It is commonly believed that quantum isolated systems satisfying the eigenstate thermalization hypothesis (ETH) are diffusive. We show that this assumption is too restrictive since there are systems that are asymptotically in a thermal state yet exhibit anomalous, subdiffusive thermalization. We show that such systems satisfy a modified version of the ETH ansatz and derive a general connection between the scaling of the variance of the off-diagonal matrix elements of local operators, written in the eigenbasis of the Hamiltonian, and the dynamical exponent. We find that for subdiffusively thermalizing systems the variance scales more slowly with system size than expected for diffusive systems. We corroborate our findings by numerically studying the distribution of the coefficients of the eigenfunctions and the off-diagonal matrix elements of local operators of the random field Heisenberg chain, which has anomalous transport in its thermal phase. Surprisingly, this system also has non-Gaussian distributions of the eigenfunctions, thus, directly violating Berry's conjecture.

We study low rank matrix recovery from undersampled measurements via nuclear norm minimization. We aim to recover an n 1 x n 2 matrix X from m measurements (Frobenius inner products) 〈X, A j 〉, j = 1…m. We consider different scenarios of independent random measurement matrices A j and derive bounds for the minimal number of measurements sufficient to uniformly recover any rank r matrix X with high probability. Our results are stable under passing to only approximately low rank matrices and under noise on the measurements. In the first scenario the entries of the A j are independent mean zero random variables of variance 1 with bounded fourth moments. Then any X of rank at most r is stably recovered from m measurements with high probability provided that m ≥ Cr max{n 1 , n 2 }. The second scenario studies the physically important case of rank one measurements. Here, the matrix X to recover is Hermitian of size n × n and the measurement matrices A j are of the form A j = a j a∗ j for some random vectors a j . If the a j are independent standard Gaussian random vectors, then we obtain uniform stable and robust rank-r recovery with high probability provided that m ≥ crn. Finally we consider the case that the a j are independently sampled from an (approximate) 4-design. Then we require m ≥ crn log n for uniform stable and robust rank-r recovery. In all cases, the results are shown via establishing a stable and robust version of the rank null space property. To this end, we employ Mendelson's small ball method.

We investigate the random-matrix configurations for two or three interacting electrons in one-dimensional disordered systems. In a suitable noninteracting localized electron basis we obtain a sparse random matrix with very long tails which is different from the superimposed random-band-matrix description usually thought to be valid. The number of nonzero off-diagonal matrix elements decays very weakly from the matrix diagonal and their distribution is a Lorentzian having long tails. The corresponding random matrix for three interacting electrons is similar but even more sparse.

We test the eigenstate thermalization hypothesis (ETH) in 1 + 1 -dimensional SU(2) lattice gauge theory (LGT) with one flavor of dynamical fermions. Using the loop-string-hadron framework of the LGT with a bosonic cutoff, we exactly diagonalize the Hamiltonian for finite size systems and calculate matrix elements (MEs) in the eigenbasis for both local and nonlocal operators. We analyze different indicators to identify the parameter space for quantum chaos at finite lattice sizes and investigate how the ETH behavior emerges in both the diagonal and off-diagonal MEs. Our investigations allow us to study various timescales of thermalization and the emergence of random matrix behavior, and highlight the interplays of the several diagnostics with each other. Furthermore, from the off-diagonal MEs, we extract a smooth function that is closely related to the spectral function for both local and nonlocal operators. We find numerical evidence of the spectral gap and the memory peak in the nonlocal operator case. Finally, we investigate aspects of subsystem ETH in the lattice gauge theory and identify certain features in the subsystem reduced density matrix that are unique to gauge theories.

In this paper, we investigate the spectral behavior of matrices constructed over generalized topological vector spaces (GTVS) using a framework inspired by random geometric graphs and random matrix theory. By assigning random weights to adjacency matrices and systematically varying the connection radius, we analyze transitions in both topological and spectral properties across different graph sizes. We observe that key topological indices the number of nonisolated vertices V×​(G), the Randić index R(G), and the harmonic index H(G) are strongly correlated with spectral measures such as the Shannon entropy S(G), the inverse participation ratio (IPR), and the eigenvalue spacing ratio rrr. Crucially, all these measures scale with a universal parameter ξ=ε/ε∗, where ε∗∼n−1/2. This scaling collapse suggests a unifying mechanism governing spectral complexity in matrix models defined on GTVS.

Parameter-dependent statistical properties of spectra of totally connected irregular quantum graphs with Neumann boundary conditions are studied. The autocorrelation functions of level velocities c(x) and c[over ](omega,x) as well as the distributions of level curvatures and avoided crossing gaps are calculated. The numerical results are compared with the predictions of random matrix theory for Gaussian orthogonal ensemble (GOE) and for coupled GOE matrices. The application of coupled GOE matrices was justified by studying localization phenomena in graphs' wave functions Psi(x) using the inverse participation ratio and the amplitude distribution P(Psi(x)) .

We calculate analytically, for finite-size matrices, joint probability densities of ratios of level spacings in ensembles of random matrices characterized by their associated confining potential. We focus on the ratios of two spacings between three consecutive real eigenvalues, as well as certain generalizations such as the overlapping ratios. The resulting formulas are further analyzed in detail in two specific cases: the β-Hermite and the β-Laguerre cases, for which we offer explicit calculations for small N. The analytical results are in excellent agreement with numerical simulations of usual random matrix ensembles, and with the level statistics of a quantum many-body lattice model and zeros of the Riemann zeta function.