math / data /lean /OAI /MeasureTheory /Falconer /Recursion /ActualDistanceRecursionBound.lean
introvoyz045's picture
Migrated from GitHub
4a56cd6 verified
Raw History Blame Contribute Delete
6.05 kB
import OAI.MeasureTheory.Falconer.Angular.ActualAngularRecursionBound
namespace OAI
namespace PlanarFalconer
open Set MeasureTheory
open scoped ContDiff
theorem PositiveDistanceData.mul_prefixLowMaximum_le (d : PositiveDistanceData)
{c M : ℝ} (hc : 0 < c) (hM : 0 ≀ M)
(hb : βˆ€ v ∈ d.prefixLowDepths,
c*({d with t := v,i := d.i+1} : PositiveDistanceData).coincidence.toReal ≀ M) :
c*d.prefixLowMaximum ≀ M := by
have hh := d.prefixLowMaximum_le (div_nonneg hM hc.le)
(fun v hv => (le_div_iffβ‚€ hc).mpr (by simpa only [mul_comm] using hb v hv))
exact (mul_le_mul_of_nonneg_left hh hc.le).trans_eq (by field_simp)
theorem uniform_actual_distance_recursion_bound {lam sig kap eps T : ℝ}
(hlam : 0 ≀ lam) (hsig : 0 < sig) (hkap : 0 < kap)
(hgap : sig < kap/1000) (heps : 0 < eps) (hT : 0 < T)
{Ξ¨ : ℝ β†’ ℝ} (hΞ¨ : ContDiff ℝ ∞ Ξ¨) (hpos : βˆ€ x, 0 ≀ Ξ¨ x)
(hs : βˆ€ x, Ξ¨ x β‰  0 β†’ x ∈ Ioo (1/4:ℝ) 4)
(hone : βˆ€ x ∈ Icc (1/2:ℝ) 2, 1 ≀ Ξ¨ x) :
βˆƒ Kinv Nβ‚€ : β„•, βˆ€ K β‰₯ Kinv, βˆ€ d : PositiveDistanceData,
d.lam = lam β†’ d.sig = sig β†’ d.q = sourcePolicyBlock d.N kap β†’
d.J = sourceSmoothingLevels sig kap β†’ Nβ‚€ ≀ d.N β†’
d.Q.side = 1 β†’ d.ρ d.Q.carrierᢜ = 0 β†’ d.RecursiveCells β†’
d.branchState.Valid d.N d.sig β†’ d.a+100*d.q < d.t β†’ d.i+1 ≀ d.J β†’
βˆ€ B : β„•, βˆ€ Ξ΄ : ℝ, RegularizedProfileData d.Q d.ρ d.N B Ξ΄ d.lam d.mass β†’
βˆ€ M : ℝ, 0 ≀ M β†’ ((2:ℝ)^d.N)^(12*lam+eps) ≀ M β†’
(βˆ€ t i l, BranchStep d.branchProfile d.N d.q d.sig d.branchState l (.D d.a t i) β†’
((2:ℝ)^d.N)^(branchExponent d.branchProfile lam sig eps d.branchState l (.D d.a t i)) *
({d with t := t,i := i} : PositiveDistanceData).coincidence.toReal ≀ M) β†’
(βˆ€ p v, βˆ€ row ∈ sourceProductVertices d.Q d.Q d.ρ d.ρ d.X d.Y (p-d.a),
BranchStep d.branchProfile d.N d.q d.sig d.branchState .high (.H d.a d.t p v d.i) β†’
βˆ€ k : Fin 2 β†’ PolicyDescriptor β†’ β„•, (βˆ€ l, βˆ‘ e ∈ d.prefixRemainingList p, k l e ≀ K) β†’
βˆ€ h : Fin 2 β†’ Plane β†’ β„‚, (βˆ€ l, Measurable (h l)) β†’ (βˆ€ l x, β€–h l xβ€– ≀ 1) β†’
((2:ℝ)^d.N)^(branchExponent d.branchProfile lam sig eps d.branchState .high (.H d.a d.t p v d.i)) *
((d.angularSuccessor p v row).energy k h Ξ¨).toReal ≀ M) β†’
d.coincidence β‰  ⊀ ∧ d.coincidence.toReal ≀ 3*M+((2:ℝ)^d.N)^(eps-T) := by
obtain ⟨N₁,hβ‚βŸ© := source_positive_distance_suffix hlam hsig hkap heps
obtain ⟨Kinv,Nβ‚‚,hβ‚‚βŸ© := source_positive_distance_prefix hlam hsig hkap hgap heps hT hΞ¨ hpos hs hone
refine ⟨Kinv,max N₁ Nβ‚‚,?_⟩
intro K hK d hdlam hdsig hdq hdJ hN hQ hcarry hcells hv hw hi B Ξ΄ hprof M hM hfloor hD hH
let := hprof.probability
have hfinite := d.coincidence_ne_top
change d.a ≀ d.t ∧ d.t ≀ d.N at hv
let p := policyDepth d.branchProfile d.N d.q d.a d.t
have hRp : 0 < (2:ℝ)^d.N := by positivity
by_cases hp : d.a+d.t < 2*p
Β· have hbound := h₁ d hdlam hdsig hdq hdJ (by omega) hQ
(hcells.stronglySeparated hQ) hcells.1 hcells.2.1 B Ξ΄ hprof hw hv.2 hi hp
have hstep : BranchStep d.branchProfile d.N d.q d.sig d.branchState .suffix (.D d.a p (d.i+1)) :=
BranchStep.suffix hw hp
have hm := hD p (d.i+1) .suffix hstep
have hboundR := ENNReal.toReal_mono
(ENNReal.mul_ne_top ENNReal.ofReal_ne_top
({d with t := p,i := d.i+1} : PositiveDistanceData).coincidence_ne_top) hbound
rw [ENNReal.toReal_mul,ENNReal.toReal_ofReal (by positivity)] at hboundR
simp only [branchExponent,branchCost,PositiveDistanceData.branchState,BranchState.start,BranchState.finish,
sub_self,mul_zero,zero_add,reduceCtorEq,ite_false] at hm
have hboundM : d.coincidence.toReal ≀ M := hboundR.trans hm
refine ⟨hfinite,hboundM.trans ?_⟩
linarith [Real.rpow_pos_of_pos hRp (eps-T)]
Β· have hp' : 2*p ≀ d.a+d.t := by omega
have hbound := hβ‚‚ d hdlam hdsig hdq hdJ (by omega) hQ hcarry hcells.1 hcells.2.1
hcells.2.2.1 hcells.2.2.2 B Ξ΄ hprof hw hv.2 hi hp'
have hlow : ((2:ℝ)^d.N)^eps*d.prefixLowMaximum ≀ M := by
apply d.mul_prefixLowMaximum_le (by positivity) hM
intro v hv
have hm := hD v (d.i+1) .low (d.low_branch_transition hw hv)
simpa only [branchExponent,branchCost,PositiveDistanceData.branchState,BranchState.start,sub_self,mul_zero,
neg_zero,zero_add,ite_true] using hm
let E := -2*(d.branchProfile p-d.branchProfile d.a)+height d.branchProfile d.a p+4000*(lam+sig)
have hhigh : ((2:ℝ)^d.N)^(E+eps)*d.prefixSuccessorSupremum Kinv p Ξ¨ ≀ M := by
have hsup := d.prefixSuccessorSupremum_le Kinv p Ξ¨
(div_nonneg hM (by positivity : 0 ≀ ((2:ℝ)^d.N)^(E+eps))) (by
intro v hv hg row hr k hk h hh hhb
have hm := hH p v row hr (d.high_branch_transition hw hp' hv hg row) k
(fun l => (hk l).trans hK) h hh hhb
apply (le_div_iffβ‚€ (by positivity : 0 < ((2:ℝ)^d.N)^(E+eps))).mpr
simpa only [branchExponent,branchCost,PositiveDistanceData.branchState,BranchState.start,
reduceCtorEq,ite_false,E,mul_comm] using hm)
exact (mul_le_mul_of_nonneg_left hsup (by positivity)).trans_eq (by field_simp)
have hfloor' : ((2:ℝ)^d.N)^eps*((2:ℝ)^d.N)^(12*lam) ≀ M := by
rw [← Real.rpow_add hRp,add_comm eps]
exact hfloor
have hhigh' : ((2:ℝ)^d.N)^eps*(((2:ℝ)^d.N)^E*d.prefixSuccessorSupremum Kinv p Ξ¨) ≀ M := by
rw [← mul_assoc,← Real.rpow_add hRp,add_comm eps]
exact hhigh
have herr : ((2:ℝ)^d.N)^eps*((2:ℝ)^d.N)^(-T) = ((2:ℝ)^d.N)^(eps-T) := by
rw [← Real.rpow_add hRp,sub_eq_add_neg]
refine ⟨hfinite,hbound.trans ?_⟩
change ((2:ℝ)^d.N)^eps * (((2:ℝ)^d.N)^(12*lam)+d.prefixLowMaximum+
((2:ℝ)^d.N)^E*d.prefixSuccessorSupremum Kinv p Ξ¨+((2:ℝ)^d.N)^(-T)) ≀ _
simp only [mul_add]
rw [herr]
linarith
end PlanarFalconer
end OAI