--- title: Wigner’s Semicircle Law url: https://www.emergentmind.com/topics/wigner-s-semicircle-law type: topic --- # Wigner’s Semicircle Law Wigner’s semicircle law is the statement that the empirical spectral distribution of a suitably normalized Hermitian or real symmetric random matrix converges to a deterministic probability measure supported on \([-2,2]\) with density \[ \rho_{sc}(x)=\frac{1}{2\pi}\sqrt{4-x^2}\,\mathbf 1_{[-2,2]}(x). \] In its classical form, the law governs Wigner matrices with independent upper-triangular entries of mean zero and variance \(1/n\); in modern formulations it extends to non-identically distributed variance profiles, local spectral statistics, sparse graphs, dependent-entry ensembles, algebraic constructions from linear codes, and free-probability analogues [1907.12170][1601.04055][2105.09519]. ## 1. Classical formulation and canonical objects Let \(H=(H_{ij})_{1\le i,j\le N}\) be an \(N\times N\) Hermitian random matrix whose upper-triangular entries are independent, centered, and satisfy \(\mathrm{Var}(H_{ij})=1/N\) for \(i\neq j\), with the diagonal typically of order \(O(1/N)\). If \(\lambda_1\le\cdots\le \lambda_N\) are its eigenvalues, the empirical spectral distribution is \[ \mu_N=\frac1N\sum_{k=1}^N\delta_{\lambda_k}. \] Its Stieltjes transform is \[ s_N(z)=\int\frac{\mu_N(dx)}{x-z}=\frac1N\mathrm{Tr}(H-z)^{-1},\qquad z\in\mathbb C^+. \] The global semicircle law states that \(s_N(z)\to m_{sc}(z)\) in probability for each fixed \(z\in\mathbb C^+\), equivalently \(\mu_N\) converges weakly to the semicircle distribution [1601.04055]. The deterministic Stieltjes transform \(m_{sc}\) is characterized by \[ m_{sc}(z)^2+z\,m_{sc}(z)+1=0, \qquad m_{sc}(z)=\frac{-z+\sqrt{z^2-4}}{2}, \] with the branch chosen so that \(\Im m_{sc}(z)>0\) for \(\Im z>0\). The support of the limiting measure is \([-2,2]\), and its even moments are Catalan numbers while odd moments vanish: \[ \int x^{2\ell}\,\mu_{sc}(dx)=C_\ell,\qquad \int x^{2\ell+1}\,\mu_{sc}(dx)=0, \] where \(C_\ell=\frac1{\ell+1}\binom{2\ell}{\ell}\) [1907.12170][1601.04055]. A strengthened global formulation no longer requires identical distributions or exact variance \(1/n\). For Hermitian matrices \(W_n=(w_{ij}^{(n)})\) with independent upper-triangular entries, mean zero, row-variance control \(\sum_j \sigma_{ij}^{2(n)}\le C/n\), and a Lindeberg condition, the semicircle law still holds when \[ \sum_{i,j=1}^n\bigl|n\,\sigma_{ij}^{2(n)}-1\bigr|\to 0 \] as \(n\to\infty\) [1907.12170]. ## 2. Moment method, resolvents, and the mechanism of the law The classical proof expands \[ \frac1n\,\mathbb E\,\mathrm{Tr}\,W_n^k = \frac1n\sum_{i_0,\dots,i_k=1}^n \mathbb E\Bigl[\prod_{s=1}^k w_{i_{s-1},i_s}^{(n)}\Bigr], \qquad i_k=i_0, \] and interprets each summand as a closed walk. Because the entries are independent and centered, any edge traversed exactly once contributes zero. The leading contribution comes from “double-tree” walks: each edge is traversed twice, and the canonical representatives are in bijection with Dyck paths. Their number is the Catalan number, which reproduces the moments of \(\mu_{sc}\) [1907.12170]. The resolvent method encodes the same limit through a self-consistent equation. Writing \[ s_n(z)=\frac1n\,\mathbb E\,\mathrm{Tr}(W_n-zI)^{-1}, \] Schur-complement identities lead to \[ s_n(z)=-\frac{1}{z+s_n(z)}+o(1), \] and passing to the limit yields the quadratic equation for \(m_{sc}\). Stieltjes inversion then recovers the density \(\rho_{sc}\) [1907.12170]. The two methods emphasize different structural features. The moment method exposes Catalan combinatorics and survives in many dependent or algebraic settings through walk-counting. The resolvent method is the basis of local semicircle laws and of quantitative estimates for eigenvalue rigidity and eigenvector delocalization [1601.04055][1511.00862]. An orthogonal-polynomial manifestation makes the same mechanism explicit. For the \(n\)-th monic Hermite polynomial \(H_n\), if \(\xi^{(n)}_1>\cdots>\xi^{(n)}_n\) are its zeros and \(\lambda_j^{(n)}=\xi_j^{(n)}/(2\sqrt n)\), then the empirical measure \[ \mu_n=\frac1n\sum_{j=1}^n\delta_{\lambda_j^{(n)}} \] converges weakly to the semicircle distribution on \([-1,1]\) with density \(\frac{2}{\pi}\sqrt{1-x^2}\mathbf 1_{[-1,1]}(x)\), equivalently to the standard \([-2,2]\) semicircle after rescaling by \(\sqrt n\). Kornyik and Michaletzky gave both a generating-function proof via the fixed-point equation \(1=h(z)-z^2h(z)^2\) and a moment-polynomial proof in which the leading coefficient is Catalan [1512.03724]. ## 3. Local semicircle laws and fine spectral control The global law controls spectral mass on macroscopic \(O(1)\) scales. A local semicircle law controls the resolvent down to spectral windows containing slightly more than one eigenvalue. For Wigner matrices, with \[ G(z)=(H-z)^{-1},\qquad s_N(z)=\frac1N\mathrm{Tr}\,G(z), \] the local law on the spectral domain \[ S_N(\tau):=\{z=E+i\eta:\ |E|\le \tau^{-1},\ \tau N^{-1+\tau}\le \eta\le \tau^{-1}\} \] takes the form \[ |s_N(z)-m_{sc}(z)|\prec \frac1{N\eta}, \] and, entrywise, \[ |G_{ii}(z)-m_{sc}(z)|\prec \Psi(z),\qquad |G_{ij}(z)|\prec \Psi(z)\quad (i\neq j), \] where \[ \Psi(z):=\sqrt{\frac{\Im m_{sc}}{N\eta}+\frac1{N\eta}}. \] In the bulk \(\Im m_{sc}\sim 1\), so \(\Psi\sim (N\eta)^{-1/2}\); near the edge \(E\approx \pm 2\), the same framework gives optimal control down to \(\eta\gg N^{-2/3}\) [1601.04055]. Under only a finite \((4+\delta)\)-th moment assumption, \[ \mathbb E|X_{11}|^{4+\delta}<\infty, \] one still obtains quantitative convergence. Götze, Naumov, and Tikhomirov showed that the typical Kolmogorov distance between the empirical spectral distribution \(F_n\) of \(n^{-1/2}X\) and \(F_{sc}\) is of order \(1/n\) up to logarithmic correction, more precisely \[ \sup_x |F_n(x)-F_{sc}(x)| = O_p\!\left(\frac{\log^\kappa n}{n}\right), \] with \(\kappa=1+2/(1-2\alpha)\) and \(\alpha=2/(4+\delta)\). They also proved a short-scale law, \[ |m_n(z)-s(z)|\le \mathcal O\!\left(\frac{\log^2 n}{n\eta}\right) \] with overwhelming probability for \(\eta\gg n^{-1}\log^{2/(1-2\alpha)}n\), along with eigenvalue rigidity and eigenvector delocalization [1511.00862]. A sharper local result under the same finite \((4+\delta)\)-moment regime gives \[ \mathbb E\,|m_n(E+iv)-s(E+iv)|^p\le \Bigl(\frac{Cp}{nv}\Bigr)^p \] for \(v\ge A_0\log n/n\), leading to \[ |m_n(z)-s(z)|\lesssim \frac{\log n}{nv} \] with overwhelming probability, \(\Delta_n^*=O(n^{-1}\log^2 n)\) for the Kolmogorov distance, the optimal expectation bound \(\sup_x|\mathbb EF_n(x)-G_{sc}(x)|=O(1/n)\), rigidity at scale \[ n^{-2/3}\,[\min(j,n-j+1)]^{-1/3}, \] and delocalization bounds of GOE/GUE type [1602.03073]. The same local philosophy extends beyond classical Wigner matrices. For the Hermite \(\beta\) ensemble, Bao and Su proved that if \(x\in(-2,2)\) and \(t_n\to 0\) with \(t_n\sqrt{\log n}\to\infty\), then the average number of states in \[ I_n(x)=[x\sqrt n-t_n/2,\ x\sqrt n+t_n/2] \] converges in probability to \(n\rho_{sc}(x)\), and the index \(N_+(n)=\#\{i:\lambda_i>0\}\) satisfies \[ \frac{N_+(n)-n/2}{\sqrt{(\log n)/(\pi^2\beta)}}\xrightarrow{d}N(0,1) \] [1104.3431]. ## 4. General conditions, dependence, and heavy tails A precise finite-variance characterization is available. For Hermitian matrices \(W_n=(w_{ij})\) with independent upper-triangular entries, mean zero, variances \(\sigma_{ij}^2=\mathbb E|w_{ij}|^2\), Lindeberg condition \[ \frac1n\sum_{i,j=1}^n \mathbb E[|w_{ij}|^2\,\mathbf 1_{|w_{ij}|>\varepsilon}]\to 0, \] and margin-smallness \[ \frac1n\sum_{i\in J_n}\sum_{j=1}^n \sigma_{ij}^2\to 0 \quad\text{whenever } |J_n|/n\to 0, \] the empirical spectral distribution converges weakly almost surely to \(\mu_{sc}\) if and only if \[ \frac1n \sum_{i=1}^n \Bigl|\sum_{j=1}^n \sigma_{ij}^2-1\Bigr|\to 0. \] This identifies semicircle convergence with the condition that most rows have total variance near \(1\) [2105.09519]. The finite-variance hypothesis is not definitive. Zhou showed that if the entries are in the normal-domain-of-attraction and the truncated second moment \[ l(x)=E[|X|^2I(|X|\le x)] \] is slowly varying, then Wigner’s semicircle law for the raw empirical spectral distribution \(F^{W_n}\) still holds almost surely, even when \(E|X|^2=\infty\) [1107.2714]. The necessary-and-sufficient framework of [2105.09519] likewise extends to infinite second moments through truncated second moments \(E[|w_{ij}|^2\mathbf 1_{|w_{ij}|\le 1}]\) and weak Lindeberg-type assumptions. Dependence does not automatically destroy the law, but it changes the admissible hypotheses. Friesen and Löwe introduced an “approximately uncorrelated” criterion for correlated symmetric ensembles \(A_N=\frac1{\sqrt N}X_N\). Under this criterion the empirical measure converges weakly in probability to the semicircle law. For Curie–Weiss ensembles, the criterion holds for \(\beta<1\), while at \(\beta=1\) there is a transition in the top eigenvalue: below criticality \(\|A_N\|\to 2\) in probability, whereas above criticality the top eigenvalue becomes macroscopic [1401.6636]. Other dependence patterns are also compatible with semicircle behavior. Alt proved a local semicircle law for matrices with the fourfold symmetry \[ h_{xy}=h_{yx}=h_{-y,-x}=h_{-x,-y}, \] including the Fourier transform of a GOE and the flip matrix model, with optimal-scale control down to \(\eta\ge (M\log N)^{-1}\) up to logarithmic corrections [1506.04683]. Krajewski, Tanasa, and Vu established a dependent-entry semicircle law through cumulant scaling bounds and a renormalisation-group argument based on replicas: if the non-Gaussian cumulants satisfy the prescribed \(N\)-power bounds, then only the Gaussian two-point contractions survive in the large-\(N\) limit, yielding the semicircle law [1609.01873][1710.05685]. Sparse graph models occupy an intermediate regime between independent-entry Wigner matrices and strongly structured combinatorial objects. For centered, rescaled Erdős–Rényi adjacency matrices, Erdős, Knowles, Yau, and Yin proved that as long as \(pN\to\infty\) with at least logarithmic speed, the density of eigenvalues is given by the semicircle law for spectral windows of length larger than \(N^{-1}\) up to logarithmic corrections, and the eigenvectors are completely delocalized in the sense that their \(\ell^\infty\)-norms are at most of order \(N^{-1/2}\) with very high probability [1103.1919]. ## 5. Orthogonal polynomials, \(\beta\)-ensembles, and free probability The semicircle law is not confined to matrix entries. For monic Hermite polynomials \(H_n\), the appropriately normalized roots converge weakly to the same distribution. The generating function of the moments of the roots satisfies a fixed-point equation whose unique power-series solution is the Catalan generating function \(c(z^2)\), and the direct moment expansion shows that for even \(k\), \[ M_n(2k)=C_k\,n^{k+1}-\Bigl(2^{2k-1}-\binom{2k-1}{k}\Bigr)n^k+O(n^{k-1}), \] with \(M_n(2k)=\sum_j (\xi_j^{(n)})^{2k}\). Kornyik and Michaletzky also recalled the Forrester–Gamburd identity \[ \mathbb E\,\det(\lambda I-A)=H_n(\lambda) \] for a Wigner matrix \(A\), which makes the relation between Hermite zeros and expected characteristic polynomials exact [1512.03724]. For the Hermite \(\beta\) ensemble, the Dumitriu–Edelman tridiagonal model gives a concrete matrix realization whose eigenvalues have the log-gas density \[ P_n(x_1,\dots,x_n)\propto \prod_{i0\). Bao and Su extended this to a local law in the bulk and obtained a Gaussian fluctuation result for the index with variance \((\log n)/(\pi^2\beta)\) [1104.3431]. A free-probability analogue replaces matrices by free self-adjoint noncommutative random variables. For self-normalized sums \[ U_n=V_n^{-1/2}S_nV_n^{-1/2},\qquad S_n=\sum_{i=1}^n X_i,\quad V_n^2=\sum_{i=1}^n X_i^2, \] Neufeld proved weak convergence \(\mu_{U_n}\Rightarrow \mathfrak w\), where \(\mathfrak w\) is the standard semicircle distribution. In the bounded identically distributed case this gives \[ \Delta(\mu_{U_n},\mathfrak w)=O(n^{-1/2}\log n), \] while in the unbounded i.i.d. case with finite fourth moment it yields \[ \Delta(\mu_{U_n},\mathfrak w)=O(n^{-1/4}) \] for the Kolmogorov distance [2406.13601]. ## 6. Structured ensembles, estimators, and contemporary variants Linear-code ensembles provide a non-classical source of semicircle behavior. If \(C\subset \mathbb F_q^n\) is an \([n,k]_q\) linear code with dual distance \(d^\perp\ge 5\), \(\Phi\) is formed by choosing \(p\) distinct codewords uniformly at random and mapping them via the canonical additive character, and \[ G=\frac1n\Phi\Phi^*,\qquad M=\sqrt{\frac np}\,(G-I_p), \] then under \(p\to\infty\) and \(n/p\to\infty\) the empirical spectral distribution of \(M\) converges in probability to the standard semicircle law [1808.09129]. Chan and Xiong strengthened this by proving that the dual distance condition \(d^\perp\ge 5\) is sufficient for convergence via the Stieltjes-transform method and that \[ |F_M(I)-\int_I \rho_{sc}(x)\,dx|=O_p(n^{-\beta}) \] uniformly over intervals \(I\subset \mathbb R\), for some \(0<\beta<1\); if \(p\sim n^\theta\) with \(0<\gamma_1\le \theta\le \gamma_2<1\), then \[ \beta=\min\{\gamma_1/4,\ (1-\gamma_2)/8\}. \] The same work states that codes of dual distance \(4\) fail to converge to the semicircle law in general, so the condition \(d^\perp\ge 5\) is not merely technical [1907.00323]. Weighted-network models yield a shifted semicircle law for normalized Laplacians. For random weighted graphs with normalized Laplacian \[ N=I-D^{-1/2}AD^{-1/2}, \] the empirical distribution of the nontrivial eigenvalues converges to a density centered at \(1\), \[ \rho(\lambda)= \begin{cases} \displaystyle \frac{2}{\pi r^2}\sqrt{r^2-(1-\lambda)^2},& |\,\lambda-1\,|