math / data /lean /OAI /Probability /Perceptron /Pressure /EnrichedConcentration.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
2.54 kB
import Mathlib
import OAI.Probability.Perceptron.Variational.GaussianLinearVariance
namespace OAI
noncomputable section
open MeasureTheory ProbabilityTheory Filter Set
open scoped Topology NNReal ENNReal BigOperators BoundedContinuousFunction
namespace SphericalPerceptronFreeEnergy
lemma enrichedTerminal_C1 (n M : β„•) (f : ℝ →ᡇ ℝ)
(g : Fin M β†’ Fin (n+1) β†’ ℝ) (p : Fin (n+1) β†’ β„•) (u : Fin (n+1) β†’ ℝ)
(htop : ℝ) :
βˆƒ D : EnrichedMark (n+1) (n+1) p β†’ StrongDual ℝ (EnrichedMark (n+1) (n+1) p),
Continuous D ∧ βˆ€ x, HasFDerivAt (enrichedTerminal n M f g p u htop) (D x) x := by
let W := normalizedPatternEnergy (n+1) M f g
let V := fun x : NormalizedSpin (n+1) => innerSL ℝ (sourceEnrichedFeature (n+1) p u x)
have hW : Continuous W := (normalizedPatternEnergy_continuous (n+1) M f).comp
(continuous_const.prodMk continuous_id)
have hV : Continuous V := (innerSL ℝ).continuous.comp (enrichedFeature_continuous _ _ _ _ _)
have hWB : βˆ€ x, |W x| ≀ (M:ℝ)*β€–fβ€– := normalizedPatternEnergy_bound _ _ _ _
have hVB : βˆ€ x, β€–V xβ€– ≀ enrichedFeatureBound (n+1) u := fun x => by
change β€–innerSL ℝ (sourceEnrichedFeature (n+1) p u x)β€– ≀ _
rw [innerSL_apply_norm,sourceEnrichedFeature_norm]
refine ⟨vectorLogDerivative (unitSphereLaw (n+1)) W V,
vectorLogDerivative_continuous (unitSphereLaw (n+1)) hW.measurable hV.measurable
(enrichedFeatureBound (n+1) u).coe_nonneg hWB hVB,?_⟩
intro x
exact (vectorLogPartition_hasFDerivAt (unitSphereLaw (n+1)) hW.measurable hV.measurable
(enrichedFeatureBound (n+1) u).coe_nonneg hWB hVB x).sub_const _
lemma enrichedCascadeLog_fixed_patterns_variance (n M k : β„•) (f : ℝ →ᡇ ℝ)
(g : Fin M β†’ Fin (n+1) β†’ ℝ) (p d : Fin (n+1) β†’ β„•) (u : Fin (n+1) β†’ ℝ)
(h : Fin (k+1) β†’ ℝ) (z : Fin k β†’ ℝ)
(hz : StrictMono z) (hz0 : βˆ€ i, 0<z i) (hz1 : βˆ€ i, z i<1) :
MemLp (enrichedCascadeLog n M k f g p d u h) 2 (enrichedDisorderLaw n k p z) ∧
variance (enrichedCascadeLog n M k f g p d u h) (enrichedDisorderLaw n k p z) ≀
cascadeLogFluctuationConstant k z+(Real.pi^2/8)*
((enrichedFeatureBound (n+1) u:ℝ)*β€–enrichedRootMap p d hβ€–)^2 := by
obtain ⟨D,hD,hd⟩ := enrichedTerminal_C1 n M f g p u (h (Fin.last k))
exact gaussianLinearCascadeLog_variance (enrichedIncrementMap p d h) (enrichedRootMap p d h)
(enrichedTerminal_lipschitz n M f g p u _) hd hD k z hz hz0 hz1
end SphericalPerceptronFreeEnergy
end
end OAI