theorem
proved
J_symmetric
show as:
view math explainer →
open explainer
Generate a durable explainer page for this declaration.
open lean source
IndisputableMonolith.Foundation.LedgerForcing on GitHub at line 27.
browse module
All declarations in this module, on Recognition.
explainer page
depends on
used by
formal source
24noncomputable def J (x : ℝ) : ℝ := (x + x⁻¹) / 2 - 1
25
26/-- **J-Symmetry**: J(x) = J(1/x) for all x ≠ 0. -/
27theorem J_symmetric {x : ℝ} (_hx : x ≠ 0) : J x = J (x⁻¹) := by
28 simp only [J, inv_inv]; ring
29
30/-- J-symmetry in ratio form: J(a/b) = J(b/a). -/
31theorem J_symmetric_ratio {a b : ℝ} (ha : a ≠ 0) (hb : b ≠ 0) :
32 J (a / b) = J (b / a) := by
33 have h1 : (a / b)⁻¹ = b / a := by field_simp
34 rw [← h1]
35 exact J_symmetric (div_ne_zero ha hb)
36
37/-! ## Recognition Events -/
38
39/-- A recognition event between two agents. -/
40structure RecognitionEvent where
41 source : ℕ
42 target : ℕ
43 ratio : ℝ
44 ratio_pos : 0 < ratio
45 deriving DecidableEq
46
47/-- The reciprocal event: B recognizes A with inverse ratio. -/
48noncomputable def reciprocal (e : RecognitionEvent) : RecognitionEvent := {
49 source := e.target
50 target := e.source
51 ratio := e.ratio⁻¹
52 ratio_pos := inv_pos.mpr e.ratio_pos
53}
54
55/-- The reciprocal of a reciprocal is the original event. -/
56theorem reciprocal_reciprocal (e : RecognitionEvent) : reciprocal (reciprocal e) = e := by
57 simp only [reciprocal, inv_inv]
papers checked against this theorem (showing 1 of 1)
-
Clipping a single ratio replaces the trust region in policy gradients
"L^CLIP is asymmetric: for Â_t > 0 the clip activates only when r_t > 1+ε; for Â_t < 0 only when r_t < 1−ε. The objective is not invariant under r ↦ 1/r."