theorem
proved
discreteness_forced
show as:
view math explainer →
open explainer
Read the cached plain-language explainer.
open lean source
IndisputableMonolith.Foundation.DiscretenessForcing on GitHub at line 450.
browse module
All declarations in this module, on Recognition.
explainer page
Derivations using this theorem
depends on
formal source
447
448 Note: The hypothesis includes x > 0 because defect is only meaningful for positive x
449 (J(x) = (x + 1/x)/2 - 1 requires x ≠ 0, and for x < 0, J(x) < 0 ≠ defect minimum). -/
450theorem discreteness_forced :
451 (∀ x : ℝ, 0 < x → defect x = 0 → x = 1) ∧ -- Unique minimum
452 (∀ ε > 0, ∃ y : ℝ, y ≠ 1 ∧ defect y < ε) → -- No isolation in ℝ
453 ¬∃ (x : ℝ), 0 < x ∧ x ≠ 1 ∧ defect x = 0 := by -- No other stable points
454 intro ⟨hunique, _hno_isolation⟩
455 push_neg
456 intro x hx_pos hx_ne hdef
457 exact hx_ne (hunique x hx_pos hdef)
458
459/-! ## RSExists Requires Discreteness -/
460
461/-- A predicate for "stable existence" in the RS sense.
462
463 x RSExists if:
464 1. defect(x) = 0 (it exists)
465 2. x is isolated in configuration space (it's locked in)
466
467 In a continuous space, condition 2 fails for all x. -/
468def RSExists_stable (x : ℝ) (config_space : Set ℝ) : Prop :=
469 defect x = 0 ∧ ∃ ε > 0, ∀ y ∈ config_space, y ≠ x → |y - x| > ε
470
471/-- **Theorem**: RSExists_stable is impossible in connected configuration spaces containing 1.
472
473 If config_space is connected and contains 1, then 1 is not isolated,
474 so RSExists_stable 1 config_space is false. -/
475theorem rs_exists_impossible_continuous
476 (config_space : Set ℝ)
477 (h1 : (1 : ℝ) ∈ config_space)
478 (_hconn : IsConnected config_space)
479 (hdense : ∀ x ∈ config_space, ∀ ε > 0, ∃ y ∈ config_space, y ≠ x ∧ |y - x| < ε) :
480 ¬RSExists_stable 1 config_space := by
papers checked against this theorem (showing 3 of 3)
-
Large chunks unlock efficient test-time training for million-token sequences
"we adopt the opposite strategy and introduce Large Chunk Test-Time Training (LaCT). LaCT leverages extremely large chunk (from 2048 to 1M tokens) as the basic unit to update the fast weight"
-
LLMs search multiple reasoning paths via continuous hidden states
"Coconut utilizes the last hidden state of the LLM as a representation of the reasoning state, termed “continuous thought.” Instead of decoding this state into words, we feed it back to the model as the next input embedding directly in the continuous space. This latent reasoning paradigm enables an advanced reasoning pattern, where continuous thoughts can encode multiple alternative next steps, allowing the model to perform a breadth-first search (BFS)"
-
Mem0 raises LLM memory accuracy 26% while cutting latency 91%
"Mem0 attains a 91% lower p95 latency and saves more than 90% token cost, offering a compelling balance between advanced reasoning capabilities and practical deployment constraints."