OAI/Probability/DilutedSpin/ConcentrationRate.lean
16 declarations, 0 contain sorry, 0 declared as axioms
- def
OAI.DilutedSpinGlass.scoreScaleline 13Source documentation
The scale prescribed by the manuscript; no enlarged moment hypothesis is used in the bounded-model concentration step.
noncomputable def scoreScale (N : ℕ) : ℝ := (N:ℝ)^((3:ℝ)/4)
- def
OAI.DilutedSpinGlass.scoreMarginline 14 - lemma
OAI.DilutedSpinGlass.sqrt_scoreScale_divline 16 - lemma
OAI.DilutedSpinGlass.sqrt_div_scoreMargin_scaleline 22 - theorem
OAI.DilutedSpinGlass.score_envelope_normalizeline 29Source documentation
Exact normalization of the four-term concentration envelope.
theorem score_envelope_normalize {x K T ell w : ℝ} (hx : 0 < x) (hK : 0 ≤ K) (hT : 0 ≤ T) (hell : 0 < ell) (hw : 0 < w) : (Real.sqrt (T*x^((3:ℝ)/4))/ell+ 4*Real.sqrt (K*x)/(x^(-(1:ℝ)/8))+ 12*(x^(-(1:ℝ)/8))*(x^((3:ℝ)/4))/ell+ Real.sqrt (x^((3:ℝ)/4)))/(w*x^((3:ℝ)/4)) = ((Real.sqrt T/ell+1)/w)*x^(-(3:ℝ)/8)+ ((4*Real.sqrt K+12/ell)/w)*x^(-(1:ℝ)/8) := by … - theorem
OAI.DilutedSpinGlass.score_envelope_boundline 47Source documentation
Monotonicity version used with the physical variance envelope.
theorem score_envelope_bound {x K T ell w v R : ℝ} (hx : 0 < x) (hK : 0 ≤ K) (hT : 0 ≤ T) (hell : 0 < ell) (hw : 0 < w) (hR : R ≤ K*x) (hv : v ≤ Real.sqrt (T*x^((3:ℝ)/4))/ell+ 4*Real.sqrt R/(x^(-(1:ℝ)/8))+ 12*(x^(-(1:ℝ)/8))*(x^((3:ℝ)/4))/ell+Real.sqrt (x^((3:ℝ)/4))) : v/(w*x^((3:ℝ)/4)) ≤ ((Real.sqrt T/ell+1)/w)*x^(-(3:ℝ)/8)+ ((4*Real.sqrt K+12/ell)/w)*x^(-(1:ℝ)/8) := by … - lemma
OAI.DilutedSpinGlass.scoreScale_leline 59 - lemma
OAI.DilutedSpinGlass.physical_score_variance_linearline 65Source documentation
The variance bound required in normalization holds for any reservoir with interaction rate at most linear in N, and for s_N=N^(3/4).
lemma physical_score_variance_linear {N : ℕ} (hN : 1 ≤ N) {r C H R : ℝ} (hr : r ≤ R*N) : 3*C^2*r+4*scoreScale N+2*H^2*N ≤ (3*C^2*R+4+2*H^2)*N := by …Mentions: OAI.DilutedSpinGlass.scoreScale
- theorem
OAI.DilutedSpinGlass.tendsto_score_rateline 74Source documentation
Both prescribed normalized score rates vanish, including the thermal term; the constants may depend on the fixed type and depth.
theorem tendsto_score_rate (A B : ℝ) : Tendsto (fun N : ℕ => A*(N:ℝ)^(-(3:ℝ)/8)+B*(N:ℝ)^(-(1:ℝ)/8)) atTop (𝓝 0) := by … - lemma
OAI.DilutedSpinGlass.unit_limsup_nonnegline 91 - lemma
OAI.DilutedSpinGlass.unit_limsup_le_oneline 99 - lemma
OAI.DilutedSpinGlass.limsup_finite_sqrt_boundline 106Source documentation
A finite-family limsup estimate, proved by taking the size limit before removing the positive scalar slack. No uniform size rate is used.
lemma limsup_finite_sqrt_bound {ι : Type*} [Fintype ι] (a e : ℕ → ℝ) (f : ι → ℕ → ℝ) {η b c k : ℝ} (hη : 0<η) (hc : 0≤c) (hk : 0≤k) (ha0 : ∀ n, 0≤a n) (ha1 : ∀ n, a n≤1) (hf1 : ∀ j n, f j n≤1) (he : Tendsto e atTop (𝓝 0)) (hbound : ∀ n, η*a n≤b+2*e n+c*Real.sqrt (k*∑ j, Real.sqrt (f j n))) : η*limsup a atTop≤b+c*Real.sqrt (k*∑ j, Real.sqrt (limsup (f j) atTop)) := by … - lemma
OAI.MeasureTheory.integrable_triple_rotateline 160 - lemma
OAI.MeasureTheory.integrable_triple_middle_swapline 167 - lemma
OAI.MeasureTheory.integral_triple_rotateline 175 - lemma
OAI.MeasureTheory.integral_triple_middle_swapline 188