math / data /lean /OAI /Probability /LogConcave /Sampling /CenteringOutputLaw.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
1.76 kB
import Mathlib
import OAI.Probability.LogConcave.Dynamics.FlowRestrict
namespace OAI
section
section
noncomputable section
namespace LogConcaveSampling
open Set Function MeasureTheory ProbabilityTheory
open scoped NNReal RealInnerProductSpace
theorem centering_output_law {d : ℕ} {μ : Measure (Point d)} [IsProbabilityMeasure μ]
{K : Point d → Point d →L[ℝ] Point d} {M : Point d → Point d}
(hM : Continuous M) (u : Point d) {s : ℝ} (hs : 0<s)
{Ψ : (Point d × Point d) → ℝ → (Point d × Point d)}
(hΨ : ∀p,Continuous (Ψ p) ∧ Ψ p 0=p ∧ ∀t∈Icc (0:ℝ) 1,
HasDerivWithinAt (Ψ p) (skewCenteringField K (fun y => M y-u) s (Ψ p t)) (Icc (0:ℝ) 1) t)
(hΨc : Continuous (fun p => Ψ p 1))
(hΨlaw : (μ.prod (stdGaussian (Point d))).map (fun p => Ψ p 1)=μ.prod (stdGaussian (Point d))) :
(μ.prod (stdGaussian (Point d))).map
(fun p => s • p.2+∫t in (0:ℝ)..1,M (Ψ p t).1)=
(stdGaussian (Point d)).map (fun g => u+s • g) := by
have he : (fun p => s • p.2+∫t in (0:ℝ)..1,M (Ψ p t).1)=
(fun p => u+s • (Ψ p 1).2) := by
funext p
simpa only [(hΨ p).2.1] using centering_momentum_identity hM u hs (hΨ p).1 (hΨ p).2.2
rw [he]
have hm : Measurable (fun p : Point d × Point d => u+s • p.2) := by fun_prop
change (μ.prod (stdGaussian (Point d))).map
((fun p : Point d × Point d => u+s • p.2) ∘ (fun p => Ψ p 1)) = _
rw [←Measure.map_map hm hΨc.measurable,hΨlaw]
rw [show (fun p : Point d × Point d => u+s • p.2)=
(fun g : Point d => u+s • g) ∘ Prod.snd from rfl,
←Measure.map_map (by fun_prop) measurable_snd,Measure.map_snd_prod,
measure_univ,one_smul]
end LogConcaveSampling
end
end
end
end OAI