Random matrix

In probability theory and mathematical physics, a random matrix is a matrix-valued random variable—that is, a matrix in which some or all of its entries are sampled randomly from a probability distribution. Random matrix theory (RMT) is the study of properties of random matrices, often as they become large. RMT provides techniques like mean-field theory, diagrammatic methods, the cavity method, or the replica method to compute quantities like traces, spectral densities, or scalar products between eigenvectors. Many physical phenomena, such as the spectrum of nuclei of heavy atoms, the thermal conductivity of a lattice, or the emergence of quantum chaos, can be modeled mathematically as problems concerning large, random matrices.

In probability theory and mathematical physics, a random matrix is a matrix-valued random variable—that is, a matrix in which some or all of its entries are sampled randomly from a probability distribution. Random matrix theory (RMT) is the study of properties of random matrices, often as they become large. RMT provides techniques like mean-field theory, diagrammatic methods, the cavity method, or the replica method to compute quantities like traces, spectral densities, or scalar products between eigenvectors. Many physical phenomena, such as the spectrum of nuclei of heavy atoms, the thermal conductivity of a lattice, or the emergence of quantum chaos, can be modeled mathematically as problems concerning large, random matrices. ==Applications== ===Physics=== In nuclear physics, random matrices were introduced by Eugene Wigner to model the nuclei of heavy atoms. Wigner postulated that the spacings between the lines in the spectrum of a heavy atom nucleus should resemble the spacings between the eigenvalues of a random matrix, and should depend only on the symmetry class of the underlying evolution. In solid-state physics, random matrices model the behaviour of large disordered Hamiltonians in the mean-field approximation. In quantum chaos, the Bohigas–Giannoni–Schmit (BGS) conjecture asserts that the spectral statistics of quantum systems whose classical counterparts exhibit chaotic behaviour are described by random matrix theory. In quantum optics, transformations described by random unitary matrices are crucial for demonstrating the advantage of quantum over classical computation (see, e.g., the boson sampling model). Moreover, such random unitary transformations can be directly implemented in an optical circuit, by mapping their parameters to optical circuit components (that is beam splitters and phase shifters). ===Mathematical statistics and numerical analysis=== In multivariate statistics, random matrices were introduced by John Wishart, who sought to estimate covariance matrices of large samples. Chernoff-, Bernstein-, and Hoeffding-type inequalities can typically be strengthened when applied to the maximal eigenvalue (i.e. the eigenvalue of largest magnitude) of a finite sum of random Hermitian matrices. Random matrix theory is used to study the spectral properties of random matrices—such as sample covariance matrices—which is of particular interest in high-dimensional statistics. Random matrix theory also saw applications in neural networks and deep learning, with recent work utilizing random matrices to show that hyper-parameter tunings can be cheaply transferred between large neural networks without the need for re-training. In numerical analysis, random matrices have been used since the work of John von Neumann and Herman Goldstine to describe computation errors in operations such as matrix multiplication. Although random entries are traditional "generic" inputs to an algorithm, the concentration of measure associated with random matrix distributions implies that random matrices will not test large portions of an algorithm's input space. ===Number theory=== In number theory, the distribution of zeros of the Riemann zeta function (and other L-functions) is modeled by the distribution of eigenvalues of certain random matrices. The connection was first discovered by Hugh Montgomery and Freeman Dyson. It is connected to the Hilbert–Pólya conjecture. ===Free probability=== The relation of free probability with random matrices is a key reason for the wide use of free probability in other subjects. Voiculescu introduced the concept of freeness around 1983 in an operator algebraic context; at the beginning there was no relation at all with random matrices. This connection was only revealed later in 1991 by Voiculescu; he was motivated by the fact that the limit distribution which he found in his free central limit theorem had appeared before in Wigner's semi-circle law in the random matrix context. ===Computational neuroscience=== In the field of computational neuroscience, random matrices are increasingly used to model the network of synaptic connections between neurons in the brain. Dynamical models of neuronal networks with random connectivity matrix were shown to exhibit a phase transition to chaos when the variance of the synaptic weights crosses a critical value, at the limit of infinite system size. Results on random matrices have also shown that the dynamics of random-matrix models are insensitive to mean connection strength. Instead, the stability of fluctuations depends on connection strength variation and time to synchrony depends on network topology. In the analysis of massive data such as fMRI, random matrix theory has been applied in order to perform dimension reduction. When applying an algorithm such as PCA, it is important to be able to select the number of significant components. The criteria for selecting components can be multiple (based on explained variance, Kaiser's method, eigenvalue, etc.). Random matrix theory in this content has its representative the Marchenko-Pastur distribution, which guarantees the theoretical high and low limits of the eigenvalues associated with a random variable covariance matrix. This matrix calculated in this way becomes the null hypothesis that allows one to find the eigenvalues (and their eigenvectors) that deviate from the theoretical random range. The components thus excluded become the reduced dimensional space (see examples in fMRI ). ===Optimal control=== In optimal control theory, the evolution of n state variables through time depends at any time on their own values and on the values of k control variables. With linear evolution, matrices of coefficients appear in the state equation (equation of evolution). In some problems the values of the parameters in these matrices are not known with certainty, in which case there are random matrices in the state equation and the problem is known as one of stochastic control. A key result in the case of linear-quadratic control with stochastic matrices is that the certainty equivalence principle does not apply: while in the absence of multiplier uncertainty (that is, with only additive uncertainty) the optimal policy with a quadratic loss function coincides with what would be decided if the uncertainty were ignored, the optimal policy may differ if the state equation contains random coefficients. ===Computational mechanics=== In computational mechanics, epistemic uncertainties underlying the lack of knowledge about the physics of the modeled system give rise to mathematical operators associated with the computational model, which are deficient in a certain sense. Such operators lack certain properties linked to unmodeled physics. When such operators are discretized to perform computational simulations, their accuracy is limited by the missing physics. To compensate for this deficiency of the mathematical operator, it is not enough to make the model parameters random, it is necessary to consider a mathematical operator that is random and can thus generate families of computational models in the hope that one of these captures the missing physics. Random matrices have been used in this sense, with applications in vibroacoustics, wave propagations, materials science, fluid mechanics, heat transfer, etc. ===Engineering=== Random matrix theory can be applied to the electrical and communications engineering research efforts to study, model and develop Massive Multiple-Input Multiple-Output (MIMO) radio systems. The most-commonly studied random matrix distributions are the Gaussian ensembles: GOE, GUE and GSE. They are often denoted by their Dyson index, β = 1 for GOE, β = 2 for GUE, and β = 4 for GSE. This index counts the number of real components per matrix element. \frac{1}{Z_{\text{GUE}(n)}} e^{- \frac{n}{2} \mathrm{tr} H^2} on the space of n \times n Hermitian matrices H = (H_{ij})_{i,j=1}^n . Here Z_{\text{GUE}(n)} = 2^{n/2} \left(\frac{\pi}{n}\right)^{\frac{1}{2}n^2} is a normalization constant, chosen so that the integral of the density is equal to one. The term unitary refers to the fact that the distribution is invariant under unitary conjugation. The Gaussian unitary ensemble models Hamiltonians lacking time-reversal symmetry. The Gaussian orthogonal ensemble \text{GOE}(n) is described by the Gaussian measure with density \frac{1}{Z_{\text{GOE}(n)}} e^{- \frac{n}{4} \mathrm{tr} H^2} on the space of n × n real symmetric matrices H = (H ij ) . Its distribution is invariant under orthogonal conjugation, and it models Hamiltonians with time-reversal symmetry. Equivalently, it is generated by H = (G+G^T)/\sqrt{2n} , where G is an n\times n matrix with IID samples from the standard normal distribution. The Gaussian symplectic ensemble \text{GSE}(n) is described by the Gaussian measure with density \frac{1}{Z_{\text{GSE}(n)}} e^{- n \mathrm{tr} H^2} on the space of n × n Hermitian quaternionic matrices, e.g. symmetric square matrices composed of quaternions, . Its distribution is invariant under conjugation by the symplectic group, and it models Hamiltonians with time-reversal symmetry but no rotational symmetry. Point correlation functions The ensembles as defined here have Gaussian distributed matrix elements with mean ⟨H ij ⟩ = 0, and two-point correlations given by \langle H_{ij} H_{mn}^* \rangle = \langle H_{ij} H_{nm} \rangle = \frac{1}{n} \delta_{im} \delta_{jn} + \frac{2 - \beta}{n \beta}\delta_{in}\delta_{jm} , from which all higher correlations follow by Isserlis' theorem. The moment generating function for the GOE is E[e^{tr(VH)}] = e^{\frac{1}{4N}\|V + V^T\|_F^2} where \|\cdot \|_F is the Frobenius norm. {{NumBlk|| \frac{1}{Z_{\beta, n}} \prod_{k=1}^n e^{-\frac{\beta}{4}\lambda_k^2}\prod_{i | }} where Z β,n is a normalization constant which can be explicitly computed, see Selberg integral. In the case of GUE (β = 2), the formula (1) describes a determinantal point process. Eigenvalues repel as the joint probability density has a zero (of \beta th order) for coinciding eigenvalues \lambda_j = \lambda_i , and Z_{2, n} = (2\pi)^{n/2} \prod_{k=1}^n k! . More succinctly, \frac{1}{Z_{\beta, n}} e^{-\frac{\beta}{4} \|\lambda\|_2^2} |\Delta_n(\lambda)|^\beta where \Delta_n is the Vandermonde determinant. The distribution of the largest eigenvalue for GOE, and GUE, are explicitly solvable. They converge to the Tracy–Widom distribution after shifting and scaling appropriately. The spectrum, divided by \sqrt{N\sigma^2} , converges in distribution to the semicircular distribution on the interval [-2, +2] : \rho(x) = \frac{1}{2 \pi}\sqrt{4-x^2} . Here \sigma^2 is the variance of off-diagonal entries. The variance of the on-diagonal entries do not matter. ===Wishart matrices=== Wishart matrices are n × n random matrices of the form , where X is an n × m random matrix (m ≥ n) with independent entries, and X * is its conjugate transpose. In the important special case considered by Wishart, the entries of X are identically distributed Gaussian random variables (either real or complex). The limit of the empirical spectral measure of Wishart matrices was found by Vladimir Marchenko and Leonid Pastur. The distribution of the largest and smallest eigenvalue for Wishart and double-Wishart (multivariate beta) matrices are explicitly solvable. It is shown that for Wishart, GOE, and GUE, the probability that all eigenvalues are within the limiting spectral support (given by the Marchenko-Pastur or the Wigner semicircle laws) tends for large dimensions to the universal values 0.6921 and 0.9397 for the real and complex cases, respectively. In one dimension, this means that H_{ij} = 0 if |i - j| > W , where W is the band width. Physically, this means that the amount by which particles i and j interact is 0 if their separation is over W. In more than one dimension, i and j are no longer integers but nd vectors with integer components, and H_{ij} = 0 if |i - j|_{L^1} , where |\cdot|_{L^1} indicates the taxicab distance between the two locations. H_{ij} = H_{ji} for all i,j and nonzero values of H_{ij} have variances \sigma_{ij}^2 of the same order of magnitude, normalized such that \sum_j \sigma_{ij}^2 = 1 for each value of j. ===Random unitary matrices=== ===Non-Hermitian random matrices=== ==Spectral theory== The spectral theory of random matrices studies the distribution of the eigenvalues as the size of the matrix goes to infinity. The empirical spectral measure \mu_H of H is defined by \mu_{H}(A) = \frac{1}{n} \, \# \left\{ \text{eigenvalues of }H\text{ in }A \right\} = N_{1_A, H}, \quad A \subset \mathbb{R}. or more succinctly, if \lambda_1,\ldots, \lambda_n are the eigenvalues of H \mu_{H}(d\lambda) = \frac 1n \sum_i \delta_{\lambda_i}(d\lambda). Usually, the limit of \mu_{H} is a deterministic measure; this is a particular case of self-averaging. The cumulative distribution function of the limiting measure is called the integrated density of states and is denoted N(λ). If the integrated density of states is differentiable, its derivative is called the density of states and is denoted ρ(λ). Given a matrix ensemble, we say that its spectral measures converge weakly to \rho iff for any measurable set A , the ensemble-average converges: \lim_{n \to \infty} \mathbb E_H[\mu_H(A)] = \rho(A) Convergence weakly almost surely: If we sample H_1, H_2, H_3, \dots independently from the ensemble, then with probability 1, \lim_{n \to \infty} \mu_{H_n}(A) = \rho(A) for any measurable set A . In another sense, weak almost sure convergence means that we sample H_1, H_2, H_3, \dots , not independently, but by "growing" (a stochastic process), then with probability 1, \lim_{n \to \infty} \mu_{H_n}(A) = \rho(A) for any measurable set A . For example, we can "grow" a sequence of matrices from the Gaussian ensemble as follows: Sample an infinite doubly infinite sequence of standard random variables \{G_{i, j}\}_{i, j = 1, 2, 3, \dots} . Define each H_n = (G_n+G_n^T)/\sqrt{2n} where G_n is the matrix made of entries \{G_{i, j}\}_{i, j = 1, 2,\dots, n} . Note that generic matrix ensembles do not allow us to grow, but most of the common ones, such as the three Gaussian ensembles, do allow us to grow. ===Global regime=== In the global regime, one is interested in the distribution of linear statistics of the form N_{f, H} = n^{-1} \text{tr} f(H) . The limit of the empirical spectral measure for Wigner matrices was described by Eugene Wigner; see Wigner semicircle distribution and Wigner surmise. As far as sample covariance matrices are concerned, a theory was developed by Marčenko and Pastur. The limit of the empirical spectral measure of invariant matrix ensembles is described by a certain integral equation which arises from potential theory. ====Fluctuations==== For the linear statistics , one is also interested in the fluctuations about ∫ f(λ) dN(λ). For many classes of random matrices, a central limit theorem of the form \frac{N_{f,H} - \int f(\lambda) \, dN(\lambda)}{\sigma_{f, n}} \overset{D}{\longrightarrow} N(0, 1) is known. \mathrm{d}\mu_N(\mu)=\frac{1}{\widetilde{Z}_N}e^{-H_N(\lambda)}\mathrm{d}\lambda,\qquad H_N(\lambda)=-\sum\limits_{j\neq k}\ln|\lambda_j-\lambda_k|+N\sum\limits_{j=1}^N Q(\lambda_j), where Q(M) is the potential of the ensemble and let \nu be the empirical spectral measure. We can rewrite H_N(\lambda) with \nu as H_N(\lambda)=N^2\left[-\int\int_{x\neq y}\ln |x-y|\mathrm{d}\nu(x)\mathrm{d}\nu(y)+\int Q(x)\mathrm{d}\nu(x)\right], the probability measure is now of the form \mathrm{d}\mu_N(\mu)=\frac{1}{\widetilde{Z}_N}e^{-N^2 I_Q(\nu)}\mathrm{d}\lambda, where I_Q(\nu) is the above functional inside the squared brackets. Let now M_1(\mathbb{R})=\left\{\nu:\nu\geq 0,\ \int_{\mathbb{R}}\mathrm{d}\nu = 1\right\} be the space of one-dimensional probability measures and consider the minimizer E_Q=\inf\limits_{\nu \in M_1(\mathbb{R})}-\int\int_{x\neq y} \ln |x-y|\mathrm{d}\nu(x)\mathrm{d}\nu(y)+\int Q(x)\mathrm{d}\nu(x). For E_Q there exists a unique equilibrium measure \nu_{Q} through the Euler-Lagrange variational conditions for some real constant l 2\int_\mathbb{R}\log |x-y|\mathrm{d}\nu(y)-Q(x)=l,\quad x\in J 2\int_\mathbb{R}\log |x-y|\mathrm{d}\nu(y)-Q(x)\leq l,\quad x\in \mathbb{R}\setminus J where J=\bigcup\limits_{j=1}^q[a_j,b_j] is the support of the measure and define q(x)=-\left(\frac{Q'(x)}{2}\right)^2+\int \frac{Q'(x)-Q'(y)}{x-y}\mathrm{d}\nu_{Q}(y) . The equilibrium measure \nu_{Q} has the following Radon–Nikodym density \frac{\mathrm{d}\nu_{Q}(x)}{\mathrm{d}x}=\frac{1}{\pi}\sqrt{q(x)}. inside the unite disk, if the disks have area A_n = O(n^{-1+\epsilon }) , the conditional distribution of the spectrum inside the disks also converges to a uniform distribution. That is, if we cut the shrinking disks along with the spectrum falling inside the disks, then scale the disks up to unit area, we would see the spectra converging to a flat distribution in the disks. ===Local regime=== In the local regime, one is interested in the limit distribution of eigenvalues in a set that shrinks so fast that the number of eigenvalues remains O(1) . Typically this means the study of spacings between eigenvalues, and, more generally, in the joint distribution of eigenvalues in an interval of length of order 1/n. One distinguishes between bulk statistics, pertaining to intervals inside the support of the limiting spectral measure, and edge statistics, pertaining to intervals near the boundary of the support. ====Bulk statistics==== Formally, fix \lambda_0 in the interior of the support of N(\lambda) . Then consider the point process \Xi(\lambda_0) = \sum_j \delta\Big({\cdot} - n \rho(\lambda_0) (\lambda_j - \lambda_0) \Big)~, where \lambda_j are the eigenvalues of the random matrix. The point process \Xi(\lambda_0) captures the statistical properties of eigenvalues in the vicinity of \lambda_0 . For the Gaussian ensembles, the limit of \Xi(\lambda_0) is known; thus, for GUE it is a determinantal point process with the kernel K(x, y) = \frac{\sin \pi(x-y)}{\pi(x-y)} (the sine kernel). The universality principle postulates that the limit of \Xi(\lambda_0) as n \to \infty should depend only on the symmetry class of the random matrix (and neither on the specific model of random matrices nor on \lambda_0 ). Rigorous proofs of universality are known for invariant matrix ensembles and Wigner matrices. ====Edge statistics==== One example of edge statistics is the Tracy–Widom distribution. As another example, consider the Ginibre ensemble. It can be real or complex. The real Ginibre ensemble has i.i.d. standard Gaussian entries \mathcal N(0, 1) , and the complex Ginibre ensemble has i.i.d. standard complex Gaussian entries \mathcal N(0, 1/2) + i\mathcal N(0, 1/2) . Now let G_n be sampled from the real or complex ensemble, and let \rho(G_n) be the absolute value of its maximal eigenvalue: \rho(G_n) := \max_j |\lambda_j| We have the following theorem for the edge statistics: {{Math theorem | name = Edge statistics of the Ginibre ensemble | note = | math_statement = For G_n and \rho\left(G_n\right) as above, with probability one, \lim _{n \rightarrow \infty} \frac{1}{\sqrt{n}} \rho\left(G_n\right)=1 Moreover, if \gamma_n=\log \left(\frac{n}{2 \pi}\right)-2 \log (\log (n)) and Y_n:=\sqrt{4 n \gamma_n}\left(\frac{1}{\sqrt{n}} \rho\left(G_n\right)-1-\sqrt{\frac{\gamma_n}{4 n}}\right), then Y_n converges in distribution to the Gumbel law, i.e., the probability measure on \mathbb{R} with cumulative distribution function F_{\mathrm{Gum}}(x)=e^{-e^{-x}} . }} This theorem refines the circular law of the Ginibre ensemble. In words, the circular law says that the spectrum of \frac{1}{\sqrt{n}} G_n almost surely falls uniformly on the unit disc. and the edge statistics theorem states that the radius of the almost-unit-disk is about 1-\sqrt{\frac{\gamma_n}{4 n}} , and fluctuates on a scale of \frac{1}{\sqrt{4 n \gamma_n}} , according to the Gumbel law. Z_n = \int_{M \in \mathbf{H}^{n \times n}} d\mu_0(M)e^{\text{tr}(V(M))} where V(x):=\sum_{j=1}^\infty v_j x^j and d\mu_0(M) is the standard Lebesgue measure on the space \mathbf{H}^{n \times n} of Hermitian n \times n matrices, is given by p_{n,V}(x_1,\dots, x_n) = \frac{1}{Z_{n,V}}\prod_{i The k -point correlation functions (or marginal distributions) are defined as R^{(k)}_{n,V}(x_1,\dots,x_k) = \frac{n!}{(n-k)!} \int_{\mathbf{R}}dx_{k+1} \cdots \int_{\R} dx_{n} \, p_{n,V}(x_1,x_2,\dots,x_n), which are skew symmetric functions of their variables. In particular, the one-point correlation function, or density of states, is R^{(1)}_{n,V}(x_1) = n\int_{\R}dx_{2} \cdots \int_{\mathbf{R}} dx_{n} \, p_{n,V}(x_1,x_2,\dots,x_n). Its integral over a Borel set B \subset \mathbf{R} gives the expected number of eigenvalues contained in B : \int_{B} R^{(1)}_{n,V}(x)dx = \mathbf{E}\left(\#\{\text{eigenvalues in }B\}\right). The following result expresses these correlation functions as determinants of the matrices formed from evaluating the appropriate integral kernel at the pairs (x_i, x_j) of points appearing within the correlator. Theorem [Dyson-Mehta] For any k , 1\leq k \leq n the k -point correlation function R^{(k)}_{n,V} can be written as a determinant R^{(k)}_{n,V}(x_1,x_2,\dots,x_k) = \det_{1\leq i,j \leq k}\left(K_{n,V}(x_i,x_j)\right), where K_{n,V}(x,y) is the n th Christoffel-Darboux kernel K_{n,V}(x,y) := \sum_{k=0}^{n-1}\psi_k(x)\psi_k(y), associated to V , written in terms of the quasipolynomials \psi_k(x) = {1\over \sqrt{h_k}}\, p_k(z)\, e^{- V(z) / 2} , where \{p_k(x)\}_{k\in \mathbf{N}} is a complete sequence of monic polynomials, of the degrees indicated, satisfying the orthogonality conditions \int_{\mathbf{R}} \psi_j(x) \psi_k(x) dx = \delta_{jk}. Wigner matrices are random Hermitian matrices H_n = (H_n(i,j))_{i,j=1}^n such that the entries \left\{ H_n(i, j)~, \, 1 \leq i \leq j \leq n \right\} above the main diagonal are independent random variables with zero mean and have identical second moments. The Gaussian ensembles can be extended for \beta \neq 1, 2, 4 using the Dumitriu-Edelman tridiagonal trick. These are called the beta ensembles. Invariant matrix ensembles are random Hermitian matrices with density on the space of real symmetric/Hermitian/quaternionic Hermitian matrices, which is of the form \frac{1}{Z_n} e^{- n V(\mathrm{tr}(H))}~, where the function is called the potential. The Gaussian ensembles are the only common special cases of these two classes of random matrices. This is a consequence of a theorem by Porter and Rosenzweig. Heavy tailed distributions generalize to random matrices as heavy tailed matrix ensembles. ==Selected bibliography== ==External links==

Text derived from https://en.wikipedia.org/w/index.php?oldid=1378391057 under CC BY-SA 4.0; modifications have been made.