Skip to content
nonconvexcard

Derive the Marchenko–Pastur law via the Stieltjes transform

ProofOlympiad / Extreme⏱ 1 dayRandom Matricesmarchenko-pasturstieltjes transformself-consistent equationresolvent

Problem

Let XX be n×dn \times d with i.i.d. entries of mean 00 and variance 11, and let S=1nX⊤XS = \frac1n X^\top X with d/n→γ∈(0,1]d/n \to \gamma \in (0,1]. Derive the limiting spectral density

ϱMP(x)=(λ+−x)(x−λ−)2πγx,λ±=(1±γ)2,\varrho_{\mathrm{MP}}(x) = \frac{\sqrt{(\lambda_+-x)(x-\lambda_-)}}{2\pi\gamma x}, \qquad \lambda_\pm = (1\pm\sqrt\gamma)^2,

by: defining the Stieltjes transform, deriving the self-consistent equation it satisfies in the limit, solving it, and inverting. State clearly which step is the one requiring concentration (and which this derivation therefore assumes).

Write a complete argument. Every inequality you use should be one you could state precisely.

Hints

Hints — 0 of 3 revealed

Solution

Opening this before you have a real attempt on paper is the fastest way to feel like you understand something you cannot reproduce.

Related problems

All problems
SolvedProblemFieldTypeDifficultyTime
Marchenko–Pastur edges and conditioningRandom MatricesNumericalBeginner15 min
Simulate the Marchenko–Pastur law runRandom MatricesCodingIntermediate1 hour
The interpolation peak is a conditioning phenomenonLearning TheoryProofAdvanced30 min
The Marchenko–Pastur density integrates to oneRandom MatricesProofAdvanced1 hour
Why ordinary least squares degrades as p approaches nRandom MatricesProofAdvanced30 min