Any true nonzero dual lattice vector lower-bounds the smoothing parameter
ProvedInfoLossLAR.WorstCase.RegevSketch.smoothing_reciprocal_dual_normbdd-stoppingdual-latticegaussian-smoothinglattice
For a full-rank discrete Euclidean lattice L in positive dimension, every positive error epsilon ≤ exp(-pi), and every actual nonzero point y in its dual lattice, one has 1/||y|| ≤ eta_epsilon(L). This uses the actual full-dual Gaussian smoothing definition and an individual genuine dual term. It supports the existing Gram–Schmidt/BDD stopping analysis without introducing the basis encoding length into its geometric threshold. It supplies no LLL polynomial-time certificate or successful BDD algorithm.
Preamble
import Theorems.Thm_RegevLWE_GaussConv_smoothing_remark set_option autoImplicit false
Formal statement
namespace InfoLossLAR.WorstCase.RegevSketch
open RegevLWE.GaussConv
theorem smoothing_reciprocal_dual_norm {dimension : ℕ}
(lattice : Submodule ℤ (EuclideanSpace ℝ (Fin dimension))) [DiscreteTopology lattice] [IsZLattice ℝ lattice]
(error : ℝ) (hdimension : 0 < dimension) (herror : 0 < error) (hsmall : error ≤ Real.exp (-Real.pi))
(point : dual lattice) (hpoint : point ≠ 0) :
1/‖(point : EuclideanSpace ℝ (Fin dimension))‖ ≤ smoothingParam lattice error := by sorry
end InfoLossLAR.WorstCase.RegevSketch
Source
Complete existing checked proofs from Solutions/RegevBridgeSmoothingLowerBound.lean (smoothing_reciprocal_dual_norm), RegevBridgeSmoothingDualSeparation.lean (smoothing_dual_separation), and the three needed helper proofs dual_mass_le_of_smoothing, rho_width_mono, summable_dual_of_smoothing in RegevBridgeGaussianProjectionSmoothing.lean. Official Lean syntax/info-tree oracle preserves exact source command slices. Only the final required global solution is added. No new mathematical argument.