A true short independent lattice family gives logarithmic smoothing
ProvedInfoLossLAR.WorstCase.RegevSketch.smoothing_le_short_independentgaussian-smoothinglatticeregevsuccessive-minima
Let L be a real Euclidean lattice of dimension n>0, with full-rank discrete dual. Suppose L contains n real-linearly-independent vectors, each of norm at most a positive radius r. For every error epsilon>0, its actual smoothing parameter satisfies eta_epsilon(L) ≤ r sqrt(log(4 n (1+1/epsilon))/pi). The vectors are mathematical witnesses; the theorem neither assumes nor produces an efficient short-basis algorithm. The smoothing parameter and Gaussian use the existing GaussConv pi/width convention.
Preamble
import Mathlib import Definitions.Def_RegevLWE_GaussConv_Lattice import Theorems.Thm_SchwartzMap_exists_euclideanSpace_apply_eq_exp_neg_pi_mul_norm_sq import Theorems.Thm_SchwartzMap_tsum_eq_tsum_fourier_euclideanSpace import Theorems.Thm_ZLattice_tsum_translate_eq_inv_covolume_mul_tsum_fourierIntegral import Theorems.Thm_RegevLWE_GaussConv_fourier_rhoShift import Theorems.Thm_ZLattice_summable_fourierIntegral_mul_fourierChar_dualSubmodule set_option autoImplicit false
Formal statement
namespace InfoLossLAR.WorstCase.RegevSketch
open RegevLWE.GaussConv
theorem smoothing_le_short_independent {dimension : ℕ}
(lattice : Submodule ℤ (EuclideanSpace ℝ (Fin dimension)))
[DiscreteTopology (dual lattice)] [IsZLattice ℝ (dual lattice)]
(directions : Fin dimension → lattice)
(hindependent : LinearIndependent ℝ (fun index => (directions index : EuclideanSpace ℝ (Fin dimension))))
(radius error : ℝ) (hdimension : 0 < dimension) (hradius : 0 < radius) (herror : 0 < error)
(hbound : ∀ index,‖(directions index : EuclideanSpace ℝ (Fin dimension))‖ ≤ radius) :
smoothingParam lattice error ≤
Real.sqrt (Real.log (4*(dimension : ℝ)*(1+1/error))/Real.pi)*radius := by sorry
end InfoLossLAR.WorstCase.RegevSketch
Source
Complete existing checked proof InfoLossLAR.WorstCase.GaussianPoissonSeries.smoothing_le_short_independent in Solutions/RegevBridgeSharpSmoothingFamily.lean, together with its minimal mathematical proof closure. Official Lean declaration and syntax/info-tree oracles select source slices; the public statement only delta-expands shortSmoothingFactor and the Euclidean Space alias. No new mathematical argument. This sharp smoothing prerequisite supports Regev 3.3→3.1 and 3.15–3.17; it is not itself a complete DGS/SIVP/GapSVP reduction.