Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Outline of the proof of Theorem 1 — φγ\varphi_\gammaφγ​ for the square loss and the cost Nq2N_q^2Nq2​ (corrected at γ=∥β∥p2\gamma = \|\beta\|_p^2γ=∥β∥p2​)

Proved
RWPI.SqrtLasso.phi_Nq_closed_form

by mikedeng1 · Oct 5, 2026 · Mathlib 0df444a (Lean v4.33.1)

distributionally-robust-optimizationlinear-regressionp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1

Fix q∈(1,∞]q\in(1,\infty]q∈(1,∞] and let ppp satisfy 1/p+1/q=11/p+1/q = 11/p+1/q=1. Let β∈Rd\beta\in\mathbb R^dβ∈Rd and (x,y)∈Rd×R(x,y)\in\mathbb R^d\times\mathbb R(x,y)∈Rd×R, and write a=y−βTxa = y-\beta^Txa=y−βTx. For the square loss and the cost Nq2N_q^2Nq2​ of (14), let

φγ(x,y;β)=sup⁡x′∈Rd, y′∈R{(y′−βTx′)2−γ Nq((x′,y′),(x,y))2},\varphi_\gamma(x,y;\beta) = \sup_{x'\in\mathbb R^d,\ y'\in\mathbb R}\Big\{ (y'-\beta^Tx')^2 - \gamma\,N_q\big((x',y'),(x,y)\big)^2 \Big\},φγ​(x,y;β)=x′∈Rd, y′∈Rsup​{(y′−βTx′)2−γNq​((x′,y′),(x,y))2},

where points with y′≠yy'\ne yy′=y (infinite cost) do not contribute. Then for every γ≥0\gamma\ge0γ≥0

φγ(x,y;β)={γγ−∥β∥p2 a2,if γ>∥β∥p2,a2,if γ=∥β∥p2 and (a=0 or β=0),+∞,otherwise.\varphi_\gamma(x,y;\beta) = \begin{cases} \dfrac{\gamma}{\gamma-\|\beta\|_p^2}\,a^2, & \text{if } \gamma>\|\beta\|_p^2,\\[2mm] a^2, & \text{if } \gamma = \|\beta\|_p^2 \text{ and } (a = 0 \text{ or } \beta = 0),\\[1mm] +\infty, & \text{otherwise.}\end{cases}φγ​(x,y;β)=⎩⎨⎧​γ−∥β∥p2​γ​a2,a2,+∞,​if γ>∥β∥p2​,if γ=∥β∥p2​ and (a=0 or β=0),otherwise.​

With this formula and Proposition 1, the worst-case square loss of Theorem 1 becomes a one-dimensional minimization over γ\gammaγ.

Correction of the printed statement. The page gives +∞+\infty+∞ for every γ≤∥β∥p2\gamma\le\|\beta\|_p^2γ≤∥β∥p2​ (and writes λ\lambdaλ for γ\gammaγ and φγ(x,y;β)\varphi_\gamma(x,y;\beta)φγ​(x,y;β) for φγ(Xi,Yi;β)\varphi_\gamma(X_i,Y_i;\beta)φγ​(Xi​,Yi​;β)). At γ=∥β∥p2\gamma = \|\beta\|_p^2γ=∥β∥p2​ the supremum is finite when a=0a = 0a=0 (value 000) or when β=0\beta = 0β=0 (then γ=0\gamma = 0γ=0 and the value is y2y^2y2). The correction does not affect Theorem 1.

Formalization Note φγ\varphi_\gammaφγ​ is the mission's phi with the cost Nq2N_q^2Nq2​; the convention that infinite-cost points never contribute holds for every γ≥0\gamma\ge0γ≥0, including γ=0\gamma = 0γ=0. Values are in [0,∞][0,\infty][0,∞].

Preamble
import Mathlib
import Definitions.Def_RWPI_SqrtLasso_phi
import Definitions.Def_RWPI_SqrtLasso_squareLoss
import Definitions.Def_RWPI_SqrtLasso_Nq
Formal statement
namespace RWPI.SqrtLasso

/-- Outline of the proof of Theorem 1, last display (App. A.1, p. 29), corrected at the boundary:
`φ_γ` for the square loss and the cost `N_q²`. With `a = y − βᵀx`, `1/p + 1/q = 1` and `γ ≥ 0`:
`φ_γ((x, y); β) = γ/(γ − ‖β‖_p²) · a²` if `γ > ‖β‖_p²`; `a²` if `γ = ‖β‖_p²` and (`a = 0` or
`β = 0`); `+∞` otherwise. (The page prints `+∞` for all `γ ≤ ‖β‖_p²`.) -/
theorem phi_Nq_closed_form {d : ℕ} (q p : ENNReal) (hq : 1 < q) (hpq : p.HolderConjugate q)
    (β : Fin d → ℝ) (z : (Fin d → ℝ) × ℝ) (γ : ℝ) (hγ : 0 ≤ γ) :
    phi (fun u w => Nq q u w ^ 2) (squareLoss β) γ z =
      if ‖WithLp.toLp p β‖ ^ 2 < γ then
        ENNReal.ofReal (γ / (γ - ‖WithLp.toLp p β‖ ^ 2) * (z.2 - ∑ j, β j * z.1 j) ^ 2)
      else if γ = ‖WithLp.toLp p β‖ ^ 2 ∧ (z.2 - ∑ j, β j * z.1 j = 0 ∨ β = 0) then
        ENNReal.ofReal ((z.2 - ∑ j, β j * z.1 j) ^ 2)
      else ⊤ := by sorry

end RWPI.SqrtLasso
Source
Blanchet, Kang & Murthy, Robust Wasserstein Profile Inference and Applications to Machine Learning, arXiv:1610.05627v4, p. 29, App. A.1, outline of a proof of Theorem 1, last display
Human review
  • Endorsed by Shuze Chen · Oct 5, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 5, 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