Theorem 4.7, p. 14 — if Weak PhaseLift is C-stable at x₀, PhaseCut is (2C + 2√2 + 1)-stable at x₀
ProvedPhaseCut.Stable.theorem_4_7Let be a nonzero measurement matrix, , and . Consider the two convex relaxations of phase recovery written as projection problems: PhaseCut,
and Weak PhaseLift with the distance,
An algorithm is -stable at if for all close enough to zero every minimizer for satisfies (Definition 4.5).
Theorem. If Weak PhaseLift is -stable at , then PhaseCut is -stable at .
The theorem says that PhaseCut is "almost as stable as" Weak PhaseLift: whatever stability guarantee is known for the (weak) PhaseLift relaxation transfers to PhaseCut with a constant of the same order.
Formalization Note. Two departures from the printed statement. (i) The page writes ", for all "; here with the column dimension of , as requires. (ii) The hypothesis is added: for and the printed statement is false (, Weak PhaseLift is trivially stable, while (13) has nonzero minimizers and the PhaseCut bound would force them to vanish); the proof itself works under "". The case is not excluded. The exponent of the distance in (14) is a parameter, and the theorem is stated for each such .
Retired 2026-10-27: a Lean-vs-paper fidelity audit found this statement weaker than, or narrower than, the cited result. A corrected statement will be resubmitted.
import Mathlib import Definitions.Def_PhaseCut_Stable_Defs
namespace PhaseCut.Stable
open Matrix
open scoped ComplexOrder
/-- **Theorem 4.7** (p. 14). Let `A ∈ ℂ^{n×p}`, `A ≠ 0`, `x₀ ∈ ℂᵖ`, `C > 0` and `s ≥ 1`. If Weak
PhaseLift (in the form (14) with the `lˢ` distance) is `C`-stable at `x₀`, then PhaseCut (in the form
(13)) is `(2C + 2√2 + 1)`-stable at `x₀`. -/
theorem theorem_4_7 {n p : ℕ} (A : Matrix (Fin n) (Fin p) ℂ) (hA : A ≠ 0) (x₀ : Fin p → ℂ)
(C : ℝ) (hC : 0 < C) (s : ℝ) (hs : 1 ≤ s) (hWPL : WeakPhaseLiftStable A s x₀ C) :
PhaseCutStable A x₀ (2 * C + 2 * Real.sqrt 2 + 1) := by sorry
end PhaseCut.Stable
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.