Census drop forces a same-death pair or a value-break pin (the SLIVER dichotomy)
ProvedPersistClust.Count.census_drop_forces_pair_or_breakLet a tame 0-dimensional persistence module be given by a filtration law (stage, J), and let (b, d) be a real point with d < b at which the bar multiplicity is nonzero. Let delta0 be a positive window excluding every critical level from b and from d, and let eta be a probe with 0 < eta < delta0 and eta < (b - d)/2. Assume two descent-regularity facts at the probe level d + eta: (hstrict) every component class C whose birth-supremum at d + eta equals b still has birth-supremum exactly b after descending to level d; and (hshare) two distinct such b-valued classes at d + eta descend to distinct classes at d. Then the following dichotomy holds: EITHER there are two DISTINCT component classes M1 and M2 at level d, both of birth-supremum b, whose further descents to level d - eta coincide (a same-death-at-d pair: both threads die together at d), OR some component class M at level d with birth-supremum b has its birth-supremum strictly INCREASE above b when descended to level d - eta (a break-at-d pin: the thread's value breaks upward below d). The mechanism is a pigeonhole census count: the nonzero multiplicity forces the census of b-valued classes at d + eta to strictly exceed that at d - eta (the eta-window census drop, via the combination t3_comb at small eta); the descent-to-d map is injective on the b-valued classes at d + eta by hshare, so it injects the above-census into the d-census; and if neither alternative held, the descent-to-(d - eta) map would be injective on the b-valued d-classes (no shared below-descent) and would land inside the b-valued classes at d - eta (no value break, because birth-suprema only grow under descent), chaining the three cardinalities into |VAL| <= |BELOW'| < |PIN'| <= |VAL| - an impossibility.
import Mathlib import Definitions.Def_PersistClust_Count_Setting import Definitions.Def_PersistClust_Count_Diagram import Definitions.Def_PersistClust_Count_FiltrationLaw import Definitions.Def_PersistClust_Count_MatchingCensus open PersistClust.Count open scoped ENNReal
namespace PersistClust.Count
theorem census_drop_forces_pair_or_break {ι : Type*} {stage : ℝ → Set ι} {J : ℝ → ι → ι → Prop}
(hLaw : FiltrationLaw stage J) (hTame : IsTameModule stage J)
{b d δ₀ η : ℝ} (hbd : d < b)
(hgap : (∀ τ ∈ critSet hTame, τ ≠ b → δ₀ < |τ - b|) ∧
(∀ τ ∈ critSet hTame, τ ≠ d → δ₀ < |τ - d|))
(hη : η ∈ Set.Ioo (0 : ℝ) δ₀) (hηbd : η < (b - d) / 2)
(hpx : mult (rankFn stage J) ((b : EReal), (d : EReal)) ≠ 0)
(hle1 : d ≤ d + η) (hle2 : d - η ≤ d)
(hstrict : ∀ C : comps stage J (d + η), birthSup hLaw (d + η) C = b →
birthSup hLaw d (structMap hLaw hle1 C) = b)
(hshare : ∀ (P1 P2 : comps stage J (d + η)), P1 ≠ P2 →
birthSup hLaw (d + η) P1 = b → birthSup hLaw (d + η) P2 = b →
structMap hLaw hle1 P1 ≠ structMap hLaw hle1 P2) :
(∃ (M₁ M₂ : comps stage J d), birthSup hLaw d M₁ = b ∧
birthSup hLaw d M₂ = b ∧ M₁ ≠ M₂ ∧
structMap hLaw hle2 M₁ = structMap hLaw hle2 M₂) ∨
(∃ M : comps stage J d, birthSup hLaw d M = b ∧
b < birthSup hLaw (d - η) (structMap hLaw hle2 M))
:= by sorry