Tau-excision: no sample vertex at tau kills all Rips bars born at tau with death at most 0
ProvedPersistClust.Count.theorem_4_8_corrected_tau_excisionLet pts be a finite sample (n points) of a metric space X, let f be a real-valued function on X, and let tau > 0. Suppose no sample value equals tau (f (pts i) is different from tau for every index i). Then the Rips persistence diagram of the sample at any scale delta has multiplicity ZERO at every point born at tau whose death coordinate is at most 0 (this includes the essential case where the death coordinate is bottom). The reason: Rips persistence births are vertex values, and since the sample is finite, the values |f (pts i) - tau| have a positive minimum gamma; for every epsilon < gamma no vertex value lies in the window [tau - epsilon, tau + epsilon), so the stage sets at the two levels tau - epsilon and tau + epsilon coincide, and the rank functions entering the diagram's infimum formula agree at both levels. Every rank difference in that formula therefore collapses at such an epsilon, forcing the multiplicity to vanish. In the corrected Theorem 4.8 sample-count bound this is the tau-excision bridge: configurations whose diagram has a bar at (tau, d) with d at most 0 must contain a vertex with value exactly tau - a measure-zero event under the null hypothesis - so those configurations are negligible in the probability bound.
import Mathlib import Definitions.Def_PersistClust_Count_Setting import Definitions.Def_PersistClust_Count_Diagram import Definitions.Def_PersistClust_Count_Rips open PersistClust.Count
namespace PersistClust.Count
theorem theorem_4_8_corrected_tau_excision {X : Type*} [MetricSpace X] {n : ℕ} (f : X → ℝ) (τ : ℝ) (δ : ℝ)
(pts : Fin n → X) (hτ : 0 < τ) (hne : ∀ i : Fin n, f (pts i) ≠ τ)
(d : EReal) (hd : d ≤ ((0 : ℝ) : EReal)) :
ripsDiagram (fun i j => dist (pts i) (pts j)) (f ∘ pts) δ (((τ : ℝ) : EReal), d) = 0
:= by sorry