Jump to content

Dyson Brownian motion

From Wikipedia, the free encyclopedia

In mathematics, the Dyson Brownian motion is a real-valued continuous-time stochastic process named for Freeman Dyson.[1] Dyson studied this process in the context of random matrix theory.

There are several equivalent definitions:[2][3]

Definition by stochastic differential equation:dλi=dBi+∑1≤j≤n:j≠idtλi−λjwhere B1,...,Bn are different and independent Wiener processes. Start with a Hermitian matrix with eigenvalues λ1(0),λ2(0),...,λn(0), then let it perform Brownian motion in the space of Hermitian matrices. Its eigenvalues constitute a Dyson Brownian motion. This is defined within the Weyl chamber Wn:={(x1,…,xn)∈ℝn:x1<⋯<xn}, as well as any coordinate-permutation of it.

Start with n independent Wiener processes started at different locations λ1(0),λ2(0),...,λn(0), then condition on those processes to be non-intersecting for all time. The resulting process is a Dyson Brownian motion starting at the same λ1(0),λ2(0),...,λn(0).[4]

Random matrix theory

[edit]

In Random Matrix Theory, the Gaussian unitary ensemble is a fundamental ensemble. It is defined as a probability distribution over the space of A∈ℝn×n Hermitian matrices, with probability density function ρ(A)∝e−12tr(A2).

Consider a Hermitian matrix A∈ℝn×n. The space of Hermitian matrices can be mapped to the space of real vectors ℝn2: A↦(A11,…,Ann,2Re(A12),…,2Re(An−1,n),2Im(A12),…,2Im(An−1,n))This is an isometry, where the matrix norm is Frobenius norm. By reversing this process, a standard Brownian motion in ℝn2 maps back to a Brownian motion in the space of n×n Hermitian matrices:dA=[dB1112(dB12+idB12′)12(dB13+idB13′)⋯12(dB1n+idB1n′)12(dB12−idB12′)dB2212(dB23+idB23′)⋯12(dB2n+idB2n′)12(dB13−idB13′)12(dB23−idB23′)dB33⋯12(dB3n+idB3n′)⋮⋮⋮⋱⋮12(dB1n−idB1n′)12(dB2n−idB2n′)12(dB3n−idB3n′)⋯dBnn]The claim is that the eigenvalues of A evolve according to[3]dλi=dBi+∑1≤j≤n:j≠idtλi−λj

Proof
Proof

Since each dB is on the order of O(dt), we can equivalently write dA=dtG, where G is a random Hermitian matrix where each entry is on the order of O(1). By construction of the standard Brownian motion, dA is independent of A, so G is independent of A, and can be written as dA=[X1112(X12+iY12)12(X13+iY13)⋯12(X1n+iY1n)12(X12−iY12)X2212(X23+iY23)⋯12(X2n+iY2n)12(X13−iY13)12(X23−iY23)X33⋯12(X3n+iY3n)⋮⋮⋮⋱⋮12(X1n−iY1n)12(X2n−iY2n)12(X3n−iY3n)⋯Xnn] where each random variable Xij,Yij is standard normal. In other words, G is distributed according to the GUE(n).

By the first and second Hadamard variation formulas and Ito’s lemma, we have λi(t+dt)=λi(t)+dtui∗Gui+dt∑j≠i𝔼[|ui∗Guj|2]λi−λj

Since G is sampled from GUE(n), it is rotationally symmetric. Also, by assumption, the eigenvector ui has norm 1, so dtui∗Gui has the same distribution as dte1∗Ge1, which is distributed as dB.

Similarly, 𝔼[|ui∗Guj|2]=1.

Infinitesimal generator

[edit]

Define the adjoint Dyson operator:D∗F:=12∑i=1n∂λi2F+∑1≤i,j≤n:i≠j∂λiFλi−λj.For any smooth function F:ℝn→ℝ with bounded derivatives, by direct differentiation, we have the Kolmogorov backward equation ∂t𝔼[F]=𝔼[D∗F]. Therefore, the Kolmogorov forward equation for the eigenspectrum is ∂ρ=Dρ, where D is the Dyson operator byDρ:=12∑i=1n∂λi2ρ−∑1≤i,j≤n:i≠j∂λi(ρλi−λj)Let ρ(t,λ)=Δn(λ)u(t,λ), where Δn:=∏i<j(λi−λj) is the Vandermonde determinant, then the time-evolution of eigenspectrum is equivalent to the time-evolution of u, which happens to satisfy the heat equation ∂tu=12∑i∂i2u,

This can be proven by starting with the identity ∂λiΔn=Δn∑1≤j≤n:i≠j1λi−λj, then apply the fact that the Vandermonde determinant is harmonic: ∑i∂i2Δn=0.

Johansson formula

[edit]

Each Hermitian matrix with exactly two eigenvalues equal is stabilized by U(2)×U(1)n−2, so its orbit under the action of U(n) has dim(U(n))−dim(U(2)×U(1)n−2)=n2−n−2 dimensions. Since the space of n−1 different eigenvalues is (n−1)-dimensional, the space of Hermitian matrix with exactly two eigenvalues equal has n2−3 dimensions.

By a dimension-counting argument, ρ vanishes at sufficiently high order on the border of the Weyl chamber, such that u can be extended to all of ℝn by antisymmetry, and this extension still satisfies the heat equation.

Now, suppose the random matrix walk begins at some deterministic A(0). Let its eigenspectrum be ν=λ(A(0)), then we have u(0,λ)=1Δn(ν)∑σ∈Sn(−1)|σ|δ(λ−σ(ν)), so by the solution to the heat equation, and the Leibniz formula for determinants, we have[5]

Johansson formula—Let A0 be a Hermitian matrix with simple spectrum ν=(ν1,…,νn), let t>0, and let At=A0+t1/2G where G is drawn from GUE. Then the spectrum λ=(λ1,…,λn) of At has probability density function

ρ(t,λ)=1(2πt)n/2Δn(λ)Δn(ν)det⁡(e−(λi−νj)2/2t)1≤i,j≤n

on the Weyl chamber.

Harish-Chandra-Itzykson-Zuber integral formula

[edit]

Dyson Brownian motion allows a short proof of the Harish-Chandra-Itzykson-Zuber integral formula.[6][7][8]

Harish-Chandra-Itzykson-Zuber integral formula—If A,B have no repeated eigenvalues, and t is a nonzero complex number, then - ∫U(n)exp⁡(ttr⁡(AUBU∗))dU=cndet⁡[exp⁡(tλi(A)λj(B))]1≤i,j≤nt(n2−n)/2Δn(λ(A))Δn(λ(B))

where U is integrated over the Haar probability measure of the unitary group U(n), and cn=∏i=1ni!.

Proof
Proof

Let the GUE(n) probability distribution over Vn be defined as ρ(M)dM, where dM=dM11…dMnnd(Re(M12))…d(Im(Mn,n−1)), and ρ(M)=Cne−12tr(M2) and Cn is a constant. Similarly, the eigenvalue distribution for the GUE(n) is ρ(λ)=1(2π)n/2cnΔn(λ)2e−‖λ‖22/2dλ where dλ=dλ1…dλn, and cn is another constant., and Δn=∏1≤i<j≤n(λi−λj) is the Vandermonde determinant.

If f:Vn→ℂ is unitarily invariant, with sufficient regularity and decay, then it can be decomposed as f(Mn)=f(λ(Mn)). By Riesz representation theorem, there exists some function w such that ∫f(M)dM=∫f(λ)w(λ)dλ, which by the above argument equals w(λ)=ρ(λ)/ρ(M)=Δn(λ)2cnCn(2π)n/2

Given two such unitarily invariant functions f,g with sufficient regularity and decay, then consider their heat kernel convolution X=∬dAdBf(A)Cntn2/2e−tr(A−B)22tg(B)

We compute X in one way.

Let ft(A):=f(A)e−tr(A2)2t,gt(B):=g(B)e−tr(B2)2t, then the quantity is X=Cntn2/2∫Vn∫Vnft(A)e−tr(AB)tgt(B)dAdB=Cntn2/2∫Vn∫Vn∫U(n)ft(A)etr(AUBU∗)tgt(B)dAdBdU=Cntn2/2∫Vn∫Vnft(A)Kt(A,B)gt(B)dAdB where we integrate over the Haar measure of the unitary group U(n), and use the fact that gt(B) is unitarily invariant, and we define the kernel
Kt(A,B):=∫U(n)etr(AUBU∗)tdU

Since Kt,ft,gt are all unitarily invariant, we have X=1Cn(2π)ncn2tn2/2∫Wn∫Wnft(λ)gt(ν)Kt(λ,ν)Δn(λ)2Δn(ν)2dλdν

We compute X in another way.

Fix B, then set G=(A−B)/t, then we have X=∫Vn∫Vnf(B+tG)Cne−12tr(G2)dGdB

Apply the Johansson formula, and convert the domain of integral to the Weyl chamber: X=1Cncn(2π)ntn/2∬f(λ)g(ν)Δn(λ)Δn(ν)det⁡[e−(λi−νj)22t]i,jdλdν

Equate the two results, and simplify, we obtain the equality.

Ginibre formula

[edit]

Ginibre formula (Tao 2012, page 251)—ρ(λ)=1(2π)n/21!…n!e−|λ|2/2|Δn(λ)|2 on the Weyl chamber.

Proof
Proof

The GUE is constructed as the t=1 distribution when starting with A0. So we take t=1 and ν→0 in the Johansson formula.

ρ(λ)=Δn(λ)(2π)n/2limν→01Δn(ν)det⁡(e−12(λi−νj)2)1≤i,j≤n

Since e−(λi−νj)2/2=e−λi2/2e−νj2/2eλiνj, we have det⁡(e−(λi−νj)2/2)1≤i,j≤n=e−|λ|2/2e−|ν|2/2det⁡(eλiνj)1≤i,j≤n

Now by a property of Vandermonde matrix, at the ν→0 limit,

det⁡(eλiνj)1≤i,j≤n=11!…n!Δn(λ)Δn(ν)+o(Δn(ν)) uniformly in ν.

References

[edit]
  1. ↑ Dyson, Freeman J. (1962-11-01). "A Brownian-Motion Model for the Eigenvalues of a Random Matrix". Journal of Mathematical Physics. 3 (6): 1191–1198. doi:10.1063/1.1703862. ISSN 0022-2488.
  2. ↑ Bouchaud, Jean-Philippe; Potters, Marc, eds. (2020), "Dyson Brownian Motion", A First Course in Random Matrix Theory: for Physicists, Engineers and Data Scientists, Cambridge: Cambridge University Press, pp. 121–135, ISBN 978-1-108-48808-2, retrieved 2023-11-25
  3. 1 2 Tao, Terence (2010-01-19). "254A, Notes 3b: Brownian motion and Dyson Brownian motion". What's new. Retrieved 2023-11-25.
  4. ↑ Grabiner, David J. (1999). "Brownian motion in a Weyl chamber, non-colliding particles, and random matrices". Annales de l'I.H.P. Probabilités et statistiques. 35 (2): 177–204. ISSN 1778-7017.
  5. ↑ Johansson, Kurt Johansson (2001-01-01). "Universality of the local spacing distribution in certain ensembles of Hermitian Wigner matrices". Communications in Mathematical Physics. 215 (3): 683–705. arXiv:math-ph/0006020. doi:10.1007/s002200000328. ISSN 1432-0916.
  6. ↑ Harish-Chandra (1957). "Differential Operators on a Semisimple Lie Algebra". American Journal of Mathematics. 79 (1): 87–120. doi:10.2307/2372387. ISSN 0002-9327.
  7. ↑ Itzykson, C.; Zuber, J.-B. (1980-03-01). "The planar approximation. II". Journal of Mathematical Physics. 21 (3): 411–421. doi:10.1063/1.524438. ISSN 0022-2488.
  8. ↑ Tao, Terence (2013-02-09). "The Harish-Chandra-Itzykson-Zuber integral formula". What's new. Retrieved 2025-01-30.