Unrestricted matching to the loss derivative gives a stationary point in one update
ProvedPGDRevision20261113.loss_matched_one_stepbregman-geometryconvex-optimizationformalization
Let and satisfy the scalar-profile hypotheses on an open interval , including the surjectivity that gives unique implicit updates. Assume
For every free tilt and every exact one-step trajectory,
the new state satisfies . This is the one-step observation underlying the decision to restrict geometry families; it places no bound on the cost of solving the implicit equation.
Related results. Exact profile performance (Theorem 4.1) · Geometry classification (Theorem 5.1)
Preamble
import Definitions.Def_PGDRevision20261113_Examples open PGDRevision20261113
Formal statement
theorem PGDRevision20261113.loss_matched_one_step : LossMatchedOneStepClaim := by sorry
Source
Author manuscript, Exact Residual Bounds and Optimal Geometry Selection for Proximal Gradient Methods, revised reading edition, 13 November 2026; Section 4.1, final paragraph; sections/04_bregman_framework.tex, Allowing unrestricted.