OAI/Analysis/LipschitzHilbert/Estimates/StoppedScalarExpansion.lean

GitHub source file

4 declarations, 0 contain sorry, 0 declared as axioms

Finite scalar algebra for a stopped path. The terminal coefficient is retained, so a small positive terminal mass is never replaced by zero.

  1. def OAI.LipschitzHilbert.stoppedScalarExpansion line 13
  2. theorem OAI.LipschitzHilbert.stoppedScalarExpansion_eq line 19

    Source documentation

    Unroll up to the first inactive index, keeping its exact terminal value.

    theorem stoppedScalarExpansion_eq (active : ℕ → Prop) (beta terminal : ℕ → ℝ)
        (count stop j : ℕ) (hstop : stop ≤ count)
        (ha : ∀ i < stop, active (j+i)) (hz : ¬active (j+stop)) :
        stoppedScalarExpansion active beta terminal count j =
          terminal (j+stop)*∏ i ∈ Finset.range stop, beta (j+i) := by
    …

    Mentions: OAI.LipschitzHilbert.stoppedScalarExpansion

  3. theorem OAI.LipschitzHilbert.normalized_path_parent_bound line 54

    Source documentation

    A normalized path reaches its final parent with probability at most one. This is the scalar form of the subunit child transition bound.

    theorem normalized_path_parent_bound (V D : ℕ → ℝ) (N : ℕ)
        (hV : ∀ j ≤ N, 0 < V j) (hD : ∀ j ≤ N, 0 < D j)
        (hchild : ∀ j < N, V (j+1) ≤ D j) :
        (∏ j ∈ Finset.range (N+1), V j/D j) ≤ V 0/D N := by
    …
  4. theorem OAI.LipschitzHilbert.normalized_terminal_weight_bound line 74

    Source documentation

    An arbitrary terminal coefficient with a weight bound gives the required per-top probability bound, with no depth factor.

    theorem normalized_terminal_weight_bound (V D : ℕ → ℝ) (N : ℕ)
        (e C w : ℝ) (hV : ∀ j ≤ N, 0 < V j) (hD : ∀ j ≤ N, 0 < D j)
        (hchild : ∀ j < N, V (j+1) ≤ D j) (he : 0 ≤ e) (hCw : e ≤ C*w)
        (hfinal : 1 ≤ D N) :
        e*(∏ j ∈ Finset.range (N+1), V j/D j)/V 0 ≤ C*w := by
    …