Proof of Theorem 3.14, p. 282 — A_k is symmetric and 0 ⪯ A_k ⪯ 4Iₙ via the quadratic-form identity
OpenConvexOptAlg.LowerBounds.thm_3_14_tridiag_psdconvex-optimizationlower-boundsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-bookp2o-v1tridiagonal-matrix
Let and let be the tridiagonal matrix with on the first diagonal entries, on the adjacent off-diagonal entries inside the leading block, and elsewhere. Then is symmetric and for every
and , that is .
The bound is what makes the hard instance of Theorem 3.14 -smooth, and makes it convex.
Formalization Note is the book's 1-based coordinate (coord x i). The case is excluded because the identity refers to and .
Preamble
import Mathlib import Definitions.Def_ConvexOptAlg_LowerBounds_Defs open scoped InnerProductSpace
Formal statement
namespace ConvexOptAlg.LowerBounds
/-- Bubeck, arXiv:1405.4980v2, proof of Theorem 3.14, p. 282: for `1 ≤ k ≤ n` the tridiagonal
matrix `A_k` is symmetric and satisfies `0 ⪯ A_k ⪯ 4Iₙ`, because
`xᵀA_k x = 2 Σ_{i=1}^{k} x(i)² − 2 Σ_{i=1}^{k−1} x(i)x(i+1) = x(1)² + x(k)² + Σ_{i=1}^{k−1} (x(i) − x(i+1))²`.
Coordinates `x(i)` are the book's 1-based ones (`coord x i` is the `Fin n` entry `i − 1`). -/
theorem thm_3_14_tridiag_psd (n k : ℕ) (hk : 1 ≤ k) (hkn : k ≤ n) :
(tridiag n k).IsSymm ∧
∀ x : EuclideanSpace ℝ (Fin n),
quadForm (tridiag n k) x =
2 * ∑ i ∈ Finset.Icc 1 k, coord x i ^ 2
- 2 * ∑ i ∈ Finset.Icc 1 (k - 1), coord x i * coord x (i + 1) ∧
2 * ∑ i ∈ Finset.Icc 1 k, coord x i ^ 2
- 2 * ∑ i ∈ Finset.Icc 1 (k - 1), coord x i * coord x (i + 1) =
coord x 1 ^ 2 + coord x k ^ 2
+ ∑ i ∈ Finset.Icc 1 (k - 1), (coord x i - coord x (i + 1)) ^ 2 ∧
0 ≤ quadForm (tridiag n k) x ∧ quadForm (tridiag n k) x ≤ 4 * ‖x‖ ^ 2 := by sorry
end ConvexOptAlg.LowerBounds
Source
Bubeck, arXiv:1405.4980v2, proof of Theorem 3.14, p. 282