math / data /lean /OAI /Probability /SKBarriers /Dynamics /Uniformization.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
3.86 kB
import Mathlib
import OAI.Probability.SKBarriers.Dynamics.HeatBathStationary
import OAI.Probability.SKBarriers.Dynamics.TVSignBridge
namespace OAI
section
section
noncomputable section
open scoped BigOperators
open MeasureTheory ProbabilityTheory Filter Set
namespace SK.Analytic
def clockWeight (a : ℝ) (k : β„•) : ℝ := Real.exp (-a)*a^k/(k.factorial:ℝ)
theorem clockWeight_nonneg {a : ℝ} (ha : 0 ≀ a) (k : β„•) : 0 ≀ clockWeight a k := by
unfold clockWeight
positivity
theorem clockWeight_hasSum (a : ℝ) : HasSum (clockWeight a) 1 := by
have H := (NormedSpace.expSeries_div_hasSum_exp a).mul_left (Real.exp (-a))
rw [← Real.exp_eq_exp_ℝ,← Real.exp_add,neg_add_cancel,Real.exp_zero] at H
unfold clockWeight
simpa only [mul_div_assoc] using H
theorem clockWeight_sum (a : ℝ) : βˆ‘' k : β„•, clockWeight a k = 1 := (clockWeight_hasSum a).tsum_eq
theorem uniformization_summable {n : β„•} (hn : 0 < n) (Ξ² : ℝ) (J : Disorder n)
{t : ℝ} (ht : 0 ≀ t) (x y : Config n) :
Summable (fun k : β„• => clockWeight ((n:ℝ)*t) k*discreteKernel Ξ² J k x y) := by
refine Summable.of_nonneg_of_le (fun k => mul_nonneg (clockWeight_nonneg (by positivity) k)
(discreteKernel_nonneg Ξ² J k x y)) (fun k => ?_) (clockWeight_hasSum ((n:ℝ)*t)).summable
exact mul_le_of_le_one_right (clockWeight_nonneg (by positivity) k) (discreteKernel_le_one hn Ξ² J k x y)
theorem continuousKernel_eq_clock {n : β„•} (Ξ² : ℝ) (J : Disorder n) (t : ℝ)
(x y : Config n) : continuousKernel Ξ² J t x y =
βˆ‘' k : β„•, clockWeight ((n:ℝ)*t) k*discreteKernel Ξ² J k x y := by
simp only [continuousKernel,clockWeight,neg_mul]
theorem continuousKernel_nonneg {n : β„•} (Ξ² : ℝ) (J : Disorder n) {t : ℝ} (ht : 0 ≀ t)
(x y : Config n) : 0 ≀ continuousKernel Ξ² J t x y := by
rw [continuousKernel_eq_clock]
apply tsum_nonneg
intro k
apply mul_nonneg
Β· exact clockWeight_nonneg (by positivity) k
Β· exact discreteKernel_nonneg Ξ² J k x y
theorem continuousKernel_sum {n : β„•} (hn : 0 < n) (Ξ² : ℝ) (J : Disorder n)
{t : ℝ} (ht : 0 ≀ t) (x : Config n) : βˆ‘ y : Config n, continuousKernel Ξ² J t x y = 1 := by
classical
simp_rw [continuousKernel_eq_clock]
rw [← Summable.tsum_finsetSum (fun y (_ : y ∈ (Finset.univ : Finset (Config n))) =>
uniformization_summable hn Ξ² J ht x y)]
simp_rw [← Finset.mul_sum,discreteKernel_sum hn,mul_one]
exact clockWeight_sum _
theorem continuousKernel_stationary {n : β„•} (hn : 0 < n) (Ξ² : ℝ) (J : Disorder n)
{t : ℝ} (ht : 0 ≀ t) (y : Config n) :
βˆ‘ x : Config n, gibbs Ξ² J x*continuousKernel Ξ² J t x y = gibbs Ξ² J y := by
classical
simp_rw [continuousKernel_eq_clock]
simp_rw [← tsum_mul_left]
rw [← Summable.tsum_finsetSum (fun x (_ : x ∈ (Finset.univ : Finset (Config n))) =>
(uniformization_summable hn Ξ² J ht x y).mul_left (gibbs Ξ² J x))]
have HH (k : β„•) : (βˆ‘ x : Config n, gibbs Ξ² J x*(clockWeight ((n:ℝ)*t) k*discreteKernel Ξ² J k x y)) =
clockWeight ((n:ℝ)*t) k*gibbs Ξ² J y := by
simp_rw [mul_left_comm (gibbs Ξ² J _) (clockWeight ((n:ℝ)*t) k),← Finset.mul_sum,
discreteKernel_stationary hn]
simp_rw [HH]
rw [tsum_mul_right,clockWeight_sum,one_mul]
theorem continuous_sign_mass_of_tv {n : β„•} (hn : 0 < n) (Ξ² : ℝ) (J : Disorder n)
(x v : Config n) {t : ℝ} (ht : 0 ≀ t) (hTV : continuousDistance Ξ² J x t ≀ (1/4:ℝ)) :
(1/4:ℝ) ≀ βˆ‘ y : Config n, if overlap v y ≀ 0 then continuousKernel Ξ² J t x y else 0 := by
have H := finite_event_le_tv (continuousKernel Ξ² J t x) (gibbs Ξ² J)
(continuousKernel_sum hn Ξ² J ht x) (gibbs_sum Ξ² J) (fun y => overlap v y ≀ 0)
have HG := gibbs_nonpositive_overlap_mass Ξ² J v
change _ ≀ continuousDistance Ξ² J x t at H
linarith
end SK.Analytic
end
end
end
end OAI