Prove2Me
Navigate
DiscoverFormalpediaBlogsUsersMomentumMy Missions+
Prove2Me
⌕
Log in
← Formalpedia

The 1/n1/n1/n ball lies in the hull

Proved
ConvexOptimization.lowner_john_ball_subset_hull_of_kkt

by Shuze Chen · Aug 13, 2026 · Mathlib 0df444a (Lean v4.33.1)

convexgeometryconvexoptimizationellipsoids

The geometric core of Löwner–John rounding: a John decomposition of the identity forces the ball of radius 1/n1/n1/n to lie inside the convex hull of the points.

Let n≥1n \ge 1n≥1, let x1,…,xm∈Rnx_1,\dots,x_m \in \mathbb{R}^nx1​,…,xm​∈Rn, and let λ1,…,λm≥0\lambda_1,\dots,\lambda_m \ge 0λ1​,…,λm​≥0 satisfy the normalized optimality identities

∑i=1mλi xixiT=I,∑i=1mλi xi=0,λi(1−xiTxi)=0(i=1,…,m),\sum_{i=1}^{m} \lambda_i\, x_i x_i^{T} = I, \qquad \sum_{i=1}^{m} \lambda_i\, x_i = 0, \qquad \lambda_i\bigl(1 - x_i^{T} x_i\bigr) = 0 \quad (i = 1,\dots,m),i=1∑m​λi​xi​xiT​=I,i=1∑m​λi​xi​=0,λi​(1−xiT​xi​)=0(i=1,…,m),

where xixiTx_i x_i^{T}xi​xiT​ is the rank-one outer product and III the n×nn \times nn×n identity; the third family is complementary slackness, so every xix_ixi​ carrying a nonzero weight lies on the unit sphere. Then every v∈Rnv \in \mathbb{R}^nv∈Rn whose Euclidean norm satisfies ∥v∥2≤1/n\lVert v \rVert_2 \le 1/n∥v∥2​≤1/n, i.e. vTv≤1/n2v^{T} v \le 1/n^{2}vTv≤1/n2, lies in the hull of the points:

{ v∈Rn:∥v∥2≤1/n }  ⊆  conv⁡{x1,…,xm}.\{\, v \in \mathbb{R}^n : \lVert v\rVert_2 \le 1/n \,\} \;\subseteq\; \operatorname{conv}\{x_1,\dots,x_m\}.{v∈Rn:∥v∥2​≤1/n}⊆conv{x1​,…,xm​}.

This is the step that produces the dimension-dependent constant in Löwner–John rounding, and it is stated separately from the optimization problem so that it applies to any family satisfying a John decomposition of the identity, whatever its origin — the same statement is the workhorse behind John's theorem in Banach-space geometry. The constant is sharp: for the regular simplex inscribed in the unit sphere, the largest centred ball inside the simplex has radius exactly 1/n1/n1/n.

Formalization Note Both norms are written through the dot product ⬝ᵥ on Fin n → ℝ, so the hypothesis on vvv reads v ⬝ᵥ v ≤ 1 / n ^ 2; the dimension is called nn and is assumed positive. The fourth KKT identity ∑iλi=n\sum_i \lambda_i = n∑i​λi​=n is not needed and is deliberately absent from the hypotheses. Source: Boyd & Vandenberghe §8.4.1, pp. 412–413.

Preamble
import Mathlib

open scoped RealInnerProductSpace ENNReal
open MeasureTheory

Formal statement
theorem ConvexOptimization.lowner_john_ball_subset_hull_of_kkt {nn m : ℕ} (hnn : 0 < nn)
    (x : Fin m → Fin nn → ℝ) (lam : Fin m → ℝ) (hlam : ∀ i, 0 ≤ lam i)
    (hI : (∑ i, lam i • Matrix.vecMulVec (x i) (x i)) = (1 : Matrix (Fin nn) (Fin nn) ℝ))
    (hz : (∑ i, lam i • x i) = 0)
    (hcs : ∀ i, lam i * (1 - x i ⬝ᵥ x i) = 0)
    (v : Fin nn → ℝ) (hv : v ⬝ᵥ v ≤ 1 / (nn : ℝ) ^ 2) :
    v ∈ convexHull ℝ (Set.range x) := by
  sorry
Source
Boyd & Vandenberghe 2004, Convex Optimization, Cambridge University Press (seventh printing with corrections, 2009), https://web.stanford.edu/~boyd/cvxbook/, pp. 412-413, §8.4.1 (the convex-combination step that produces the ball of radius 1/n)

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 paper
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me