File size: 4,912 Bytes
4a56cd6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
import Mathlib
import OAI.Probability.SKGap.Terminal.PairGibbsBound
import OAI.Probability.SKGap.Entropy.SparseWeightedForm
import OAI.Probability.SKGap.Terminal.SmallFieldCount

namespace OAI

section
noncomputable section
namespace SKGap
open Matrix Real
open scoped BigOperators RealInnerProductSpace Matrix.Norms.L2Operator
variable {n : β„•}

lemma one_sub_tanh_sq (x : ℝ) : 1-tanh x^2=1/cosh x^2 := by
  rw [tanh_eq_sinh_div_cosh]
  have hc := (cosh_pos x).ne'
  have he := cosh_sq_sub_sinh_sq x
  field_simp
  nlinarith only [he]

lemma variance_tanh_le_cosh {t x : ℝ} (ht : 0 ≀ t) (hx : t ≀ |x|) :
    1-tanh x^2 ≀ 1/cosh t^2 := by
  rw [one_sub_tanh_sq]
  have hcosh : cosh t ≀ cosh x := cosh_le_cosh.mpr (by simpa only [abs_of_nonneg ht] using hx)
  have hs := (sq_le_sqβ‚€ (cosh_pos t).le (cosh_pos x).le).mpr hcosh
  exact one_div_le_one_div_of_le (sq_pos_of_pos (cosh_pos t)) hs

def terminalU (g : Disorder n) : Matrix (Fin n) (Fin n) ℝ := Matrix.of (fun i j=>tanh (coupling g i j))
def terminalQ (g : Disorder n) : Matrix (Fin n) (Fin n) ℝ := Matrix.of (fun i j=>tanh (coupling g i j)^2)

theorem terminal_deterministic (g : Disorder n) (h : Fin n→ℝ) {M t Ξ± Ξ΅ R : ℝ}
    (hM : 0 ≀ M) (ht : 0 < t) (hΞ΅ : 0 ≀ Ξ΅)
    (hJ : β€–Matrix.of (coupling g)β€– ≀ M)
    (hU : β€–terminalU gβ€– ≀ M) (hQ : β€–terminalQ gβ€– ≀ M)
    (hc : βˆ€ i j,|tanh (coupling g i j)| ≀ 1/1000)
    (hR : βˆ€ i,βˆ‘ j,tanh (coupling g i j)^4 ≀ R)
    (hsparse : βˆ€ S : Finset (Fin n),(S.card:ℝ) ≀ Ξ±*(n:ℝ) β†’
      βˆ€ w : EuclideanSpace ℝ (Fin n),(βˆ€ i,i βˆ‰ S β†’ w i=0) β†’
      |βŸͺw,Matrix.toEuclideanCLM (π•œ := ℝ) (n := Fin n) (terminalU g) w⟫| ≀ Ξ΅*β€–wβ€–^2 ∧
      |βŸͺw,Matrix.toEuclideanCLM (π•œ := ℝ) (n := Fin n) (terminalQ g) w⟫| ≀ Ξ΅*β€–wβ€–^2)
    (hB : ((Finset.univ.filter (fun i=>|h i| < 2*t)).card:ℝ) ≀ Ξ±*(n:ℝ)/2)
    (hsmall : M^2 ≀ t^2*Ξ±/2)
    (hnum : 800000*R+400001*(Ξ΅+M*(2*sqrt (1/cosh t^2)+1/cosh t^2)) ≀ 1/2) :
    βˆ€ f : Spin n→ℝ,variance g h f ≀ 2*dirichlet g h f := by
  have hΞ· : (0:ℝ) ≀ 1/cosh t^2 := by positivity
  have hS := all_configurations_small_fields g h hM ht hJ hB hsmall
  let k := Ξ΅+M*(2*sqrt (1/cosh t^2)+1/cosh t^2)
  have hcurv (f : Spin n→ℝ) : (1/2:ℝ)*dirichlet g h f ≀
      expectation g h (fun x=>heatBathGenerator g h f x^2) := by
    let q := fun x=>βˆ‘ i,siteVariance g h i x*spinGradient i f x^2
    have hq : expectation g h q=dirichlet g h f := by
      dsimp [q]
      rw [expectation_sum,dirichlet_weighted]
    have hform (x : Spin n) :
        (βˆ‘ i,βˆ‘ j,tanh (coupling g i j)*weightedGradient g h i f x*weightedGradient g h j f x) ≀ k*q x ∧
        (βˆ‘ i,βˆ‘ j,tanh (coupling g i j)^2*|weightedGradient g h i f x*weightedGradient g h j f x|) ≀ k*q x := by
      let S := Finset.univ.filter (fun i=>|localField g h i x| < t)
      have hv (i : Fin n) : 0 ≀ siteVariance g h i x := (siteVariance_pos g h i x).le
      have hvsmall (i : Fin n) (hi : i βˆ‰ S) : siteVariance g h i x ≀ 1/cosh t^2 := by
        apply variance_tanh_le_cosh ht.le
        have hi' : Β¬|localField g h i x| < t := by simpa only [S,Finset.mem_filter,Finset.mem_univ,true_and] using hi
        exact le_of_not_gt hi'
      have hboundU := weighted_sparse_form (terminalU g) S (fun i=>siteVariance g h i x)
        (fun i=>spinGradient i f x) hΞ· hΞ΅ hM hv (fun i=>siteVariance_le_one g h i x) hvsmall hU
        (fun w hw=>(hsparse S (hS x) w hw).1)
      have hboundQ := weighted_sparse_form (terminalQ g) S (fun i=>siteVariance g h i x)
        (fun i=>|spinGradient i f x|) hΞ· hΞ΅ hM hv (fun i=>siteVariance_le_one g h i x) hvsmall hQ
        (fun w hw=>(hsparse S (hS x) w hw).2)
      constructor
      Β· exact (le_abs_self _).trans hboundU
      Β· simp only [sq_abs] at hboundQ
        have he : (βˆ‘ i,βˆ‘ j,tanh (coupling g i j)^2*|weightedGradient g h i f x*weightedGradient g h j f x|)=
            βˆ‘ i,βˆ‘ j,terminalQ g i j*(siteVariance g h i x*|spinGradient i f x|)*
              (siteVariance g h j x*|spinGradient j f x|) := by
          apply Finset.sum_congr rfl
          intro i _
          apply Finset.sum_congr rfl
          intro j _
          simp only [terminalQ,Matrix.of_apply,weightedGradient,abs_mul,abs_of_nonneg (hv i),abs_of_nonneg (hv j)]
          ring
        rw [he]
        exact (le_abs_self _).trans hboundQ
    have hu := expectation_mono g h (fun x=>(hform x).1)
    have hq' := expectation_mono g h (fun x=>(hform x).2)
    rw [expectation_const_mul,hq] at hu hq'
    have hl := gibbs_signed_curvature g h f hc hR
    have hd := dirichlet_nonneg g h f
    have hn : 800000*R+400001*k ≀ 1/2 := hnum
    have hh := mul_le_mul_of_nonneg_right hn hd
    nlinarith only [hu,hq',hl,hh]
  intro f
  have hp := heatBath_curvature_poincare g h (by norm_num : (0:ℝ) < 1/2) hcurv f
  linarith only [hp]
end SKGap

end
end

end OAI