One smooth bump feature gives distinct optimal geometries
OpenPGDRevision20261113.finite_feature_nonuniqueness_counterexamplebregman-geometryconvex-optimizationformalization
On let
The corresponding loss and penalty can be taken as
The feature is , vanishes outside , and has . These data satisfy the finite-feature design hypotheses.
Over every and , the error norm is at least . Every attains that bound at :
Different feature parameters give different curvature values at , so this is nonuniqueness of geometries. The statement asserts optimality for all nonpositive parameters, without claiming that these exhaust every possible optimal positive parameter.
Related results. Geometry classification (Theorem 5.1) · Finite-feature design · polynomial quadratic geometry examples
Preamble
import Definitions.Def_PGDRevision20261113_Examples open PGDRevision20261113
Formal statement
theorem PGDRevision20261113.finite_feature_nonuniqueness_counterexample : FiniteFeatureNonuniquenessClaim := 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, paragraph Uniqueness can fail outside ordered families; sections/04_bregman_framework.tex, Uniqueness can fail outside ordered families.