`EMLClosure'` at any fixed depth is monotone in the seed set.
ProvedClosureGrowth.EMLClosure_p_depth_mono_setaether-catalogbridges
EMLClosure' at any fixed depth is monotone in the seed set.
theorem ClosureGrowth.EMLClosure'_depth_mono_set(n : ℕ) :
SetMono (EMLClosure' n) := by sorry
Formalization Note Transplanted verbatim from the Aether Catalog source Bridges/EntropyClosureSeparation.lean; the statement is byte-identical to the source declaration, elaborated with autoImplicit disabled in the platform environment.
Preamble
-- Thm stub generated from Bridges/EntropyClosureSeparation.lean import Mathlib import Definitions.Def_Bridges_EntropyClosureSeparation /-! # Entropy-Rate Separation via Closure Growth Dynamics This file formalizes the mathematical core of **closure-growth separation**: the principle that two monotone set transformers (modeling proof-search policies) can be distinguished by a finite witness whenever their iterated closures diverge at any stage. ## Main results ### Abstract closure iteration theory - `closureIter`: iteration of a set transformer via `Nat.iterate` - `closureIter_zero`, `closureIter_succ_apply`: standard recursion lemmas - `closureIter_mono`: monotonicity propagates through iteration - `closureIter_stabilizes`: idempotent closure operators stabilize in one step ### Witness extraction - `finite_witness_of_stage_separation`: any stage where `F^[n] S ⊈ G^[n] S` yields a concrete element witnessing the separation - `finite_witness_of_eventual_growth_gap`: eventual strict inclusion implies a finite separating witness ### Fixed-point invariance - `closure_fixed_points_are_iterative_invariants`: fixed points of a closure operator are invariant under all iterates ### EML instantiation - `fullEMLClosure'_extensive`: seed sets embed into their full EML closure - `fullEMLClosure'_setMono`: full EML closure is monotone in the seed set - `fullEMLClosure'_isClosureOp`: full EML closure is a closure operator - `fullEMLClosure'_iter_stabilizes`: iterates of full EML closure stabilize ## Motivation In neural proof mining, a **proof policy** induces a set transformer on the space of proof states: given a set of reachable states, the policy expands it by one step of search. The **closure filtration** `F^[0] S ⊆ F^[1] S ⊆ ⋯` captures the cumulative reach of the policy from seed set `S`. Two policies `F` and `G` are **separable** if their filtrations diverge: some state is reachable by `F` but not by `G` (or vice versa) within finitely many steps. The **finite witness theorem** (`finite_witness_of_stage_separation`) extracts a concrete certificate of this divergence — the exact mathematical object needed for counterexample-guided training and benchmark generation. The split between **preclosure** dynamics (monotone + extensive, where growth happens) and **closure** saturation (idempotent, where growth halts) is the formal kernel of "thermodynamic proof complexity": entropy lives in the transient filtration, while semantic invariants live in the idempotent hull. -/ open Set Function noncomputable section open ClosureGrowth /-! ## Definitions: Set Monotonicity, Closure, and Preclosure -/ /-! ## Iterated Closure -/ /-! ## Monotonicity of Iterates -/ /-! ## Extensivity of Iterates -/ /-! ## Stabilization for Closure Operators -/ /-! ## Stagewise Reachability and Witness Extraction -/ /-! ## Eventual Growth Gap and Separation -/ /-! ## Fixed-Point Invariance -/ /-! ## EML Closure Infrastructure We reproduce the core EML definitions here for self-containedness, then connect them to the abstract closure theory. The EML (Exponential-Minus-Log) operation generates a closure on `Set ℝ` modeling compositional proof-state transformations. -/
Formal statement
theorem ClosureGrowth.EMLClosure_p_depth_mono_set(n : ℕ) :
SetMono (EMLClosure' n) := by sorrySource