first_neumann_normal_projected_centered_iterate_eq_negative_linear_correction
Provedcandes-rechtdual-certificatelean4matrix-completionneumann-seriesnormal-projectiontangent-space
Source: Candes-Recht 2008, PDF p. 26, Section 6.2, the display and equation (6.8).
This theorem is the normal-projected version of the first-order expansion in equation (6.8). With the local convention
the centered first iterate carries a minus sign relative to the displayed correction in the paper. The statement says
where linearNeumannDiagonalContribution is the diagonal part and linearNeumannOffDiagonalContribution is the off-diagonal part from equation (6.8).
Preamble
import Definitions.Def_matrix_completion_neumann open MatrixCompletion
Formal statement
theorem first_neumann_normal_projected_centered_iterate_eq_negative_linear_correction
{n₁ n₂ r : ℕ} {M : Matrix (Fin n₁) (Fin n₂) ℝ}
(S : SVD M r) (Omega : Finset (Fin n₁ × Fin n₂)) (p : ℝ) :
normalProjection S
(centeredSamplingFluctuation Omega p (neumannIterate Omega S p 1)) =
-normalProjection S
(linearNeumannDiagonalContribution Omega S p +
linearNeumannOffDiagonalContribution Omega S p) := by
sorrySource
Candes, Emmanuel, and Benjamin Recht. "Exact matrix completion via convex optimization." Communications of the ACM 55.6 (2012): 111-119.