Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 3.1, p. 12 — without sparsity information the unique minimizer of (5) is X⋆ = Σ x⋆_i v_i v_iᵀ, x⋆_i from (16a) at the positive root γ⋆ of (16b)

Proved
WassInvCov.Shrinkage.theorem_3_1

by mikedeng1 · Oct 18, 2026 · Mathlib 0df444a (Lean v4.33.1)

distributionally-robust-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1precision-matrixshrinkage-estimatorwasserstein-distance

Let p≥1p\ge1p≥1 and ρ>0\rho>0ρ>0. Consider the distributionally robust estimation problem (5) without structural information, X=S++p\mathcal X=\mathbb S^p_{++}X=S++p​:

min⁡X≻0 {−log⁡det⁡X+sup⁡Q∈Pρ0EQ[⟨ξξ⊤,X⟩]},\min_{X\succ0}\ \Big\{-\log\det X+\sup_{\mathbb Q\in\mathcal P^0_\rho}\mathbb E^{\mathbb Q}\big[\langle\xi\xi^\top,X\rangle\big]\Big\},X≻0min​ {−logdetX+Q∈Pρ0​sup​EQ[⟨ξξ⊤,X⟩]},

where Pρ0\mathcal P^0_\rhoPρ0​ is the set of centred normal distributions on Rp\mathbb R^pRp within type-2 Wasserstein distance ρ\rhoρ of N(0,Σ^)\mathcal N(0,\widehat\Sigma)N(0,Σ). Let Σ^∈S+p\widehat\Sigma\in\mathbb S^p_+Σ∈S+p​ (possibly singular) have the spectral decomposition Σ^=∑i=1pλivivi⊤\widehat\Sigma=\sum_{i=1}^p\lambda_iv_iv_i^\topΣ=∑i=1p​λi​vi​vi⊤​ with orthonormal eigenvectors v1,…,vpv_1,\dots,v_pv1​,…,vp​. Then:

  1. the equation
(ρ2−12∑i=1pλi)γ−p+12∑i=1pλi2γ2+4λiγ=0(16b)\Big(\rho^2-\frac12\sum_{i=1}^p\lambda_i\Big)\gamma-p+\frac12\sum_{i=1}^p\sqrt{\lambda_i^2\gamma^2+4\lambda_i\gamma}=0\tag{16b}(ρ2−21​i=1∑p​λi​)γ−p+21​i=1∑p​λi2​γ2+4λi​γ​=0(16b)

has exactly one positive solution γ⋆\gamma^\starγ⋆; 2. with

xi⋆=γ⋆[1−12(λi2(γ⋆)2+4λiγ⋆−λiγ⋆)],i≤p,(16a)x^\star_i=\gamma^\star\Big[1-\frac12\Big(\sqrt{\lambda_i^2(\gamma^\star)^2+4\lambda_i\gamma^\star}-\lambda_i\gamma^\star\Big)\Big],\qquad i\le p,\tag{16a}xi⋆​=γ⋆[1−21​(λi2​(γ⋆)2+4λi​γ⋆​−λi​γ⋆)],i≤p,(16a)

the matrix X⋆=∑i=1pxi⋆vivi⊤X^\star=\sum_{i=1}^px^\star_iv_iv_i^\topX⋆=∑i=1p​xi⋆​vi​vi⊤​ is positive definite, attains the minimum of (5) over all positive definite XXX, and every positive definite minimizer of (5) equals X⋆X^\starX⋆.

The estimator X⋆X^\starX⋆ shares the eigenvectors of Σ^\widehat\SigmaΣ and shrinks its eigenvalues by an explicit nonlinear map; it is invertible even when Σ^\widehat\SigmaΣ is singular (more dimensions than samples). This is the paper's analytical solution of the robust estimation problem.

Formalization Note The hypothesis p≥1p\ge1p≥1 is added: for p=0p=0p=0, (16b) has no positive solution. The eigendecomposition is arbitrary (any orthonormal family vvv and reals λ\lambdaλ with Σ^=∑iλivivi⊤\widehat\Sigma=\sum_i\lambda_iv_iv_i^\topΣ=∑i​λi​vi​vi⊤​), so the statement also asserts independence of the choice of eigenbasis. Part 2 is stated for every positive root, which by part 1 is γ⋆\gamma^\starγ⋆. The objective is the measure-level one of (5), with the worst-case expectation computed as an extended-real supremum over Pρ0\mathcal P^0_\rhoPρ0​; minimality is IsMinOn over the positive definite matrices.

Preamble
import Mathlib
import Definitions.Def_WassInvCov_Shrinkage_Setting

open MeasureTheory ProbabilityTheory Matrix
Formal statement
namespace WassInvCov.Shrinkage

/-- Theorem 3.1, p. 12 (arXiv:1805.07194v1). Without sparsity information (`𝒳 = 𝕊^p_{++}`), for
`ρ > 0` and `Σ̂ ⪰ 0` with an orthonormal eigendecomposition `Σ̂ = Σᵢ λᵢ vᵢ vᵢᵀ`, equation (16b) has a
unique positive solution `γ⋆`, and `X⋆ = Σᵢ x⋆ᵢ vᵢ vᵢᵀ` with `x⋆ᵢ` from (16a) is the unique minimizer
of (5). The hypothesis `0 < p` is added (for `p = 0`, (16b) has no positive solution). -/
theorem theorem_3_1 {p : ℕ} (hp : 0 < p) (ρ : ℝ) (hρ : 0 < ρ) (Sh : Matrix (Fin p) (Fin p) ℝ)
    (hSh : Sh.PosSemidef) (lam : Fin p → ℝ) (v : Fin p → EuclideanSpace ℝ (Fin p))
    (hv : Orthonormal ℝ v)
    (hdecomp : Sh = ∑ i, lam i • Matrix.vecMulVec (v i).ofLp (v i).ofLp) :
    (∃! γ : ℝ, 0 < γ ∧ phi16b ρ lam γ = 0) ∧
    ∀ γs : ℝ, 0 < γs → phi16b ρ lam γs = 0 →
      (shrinkageEstimator lam v γs).PosDef ∧
      IsMinOn (robustObjective ρ Sh) {X | X.PosDef} (shrinkageEstimator lam v γs) ∧
      ∀ X : Matrix (Fin p) (Fin p) ℝ, X.PosDef →
        IsMinOn (robustObjective ρ Sh) {X | X.PosDef} X → X = shrinkageEstimator lam v γs := by sorry

end WassInvCov.Shrinkage
Source
Nguyen, Kuhn & Mohajerin Esfahani, Distributionally Robust Inverse Covariance Estimation, arXiv:1805.07194v1, p. 12, Theorem 3.1, (16a), (16b)
Human review
  • Endorsed by Shuze Chen · Oct 18, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 18, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me