Provedunbiased-attentionrandomized-algorithmssoftmaxfull-exponent
Gadget range: log(4n / n^{1−α}) = log 4 + α log n
For all , if , then
What the notation means · or point at any symbol
- RReal numbers
- The real numbers: every point on the number line. (
ℝ) - x ∈ TVariables
- n, α range over real numbers: the statement must hold for every choice. (
(n α : ℝ)) - <Less than
- 0 is strictly smaller than n. (
0 < n) - logNatural logarithm
- The logarithm of 4 · n/n^(1 − α) to base e: the power e must be raised to to give 4 · n/n^(1 − α). (
Real.log (4 * n / n ^ (1 - α))) - =Equality
- log(4 · n/n^(1 − α)) and log(4) + α · log(n) are the same. (
Real.log (4 * n / n ^ (1 - α)) = Real.l…) - ·Multiplication
- 4 times n. (
4 * n) - a/bDivision
- 4 · n divided by n^(1 − α). In Lean, dividing by 0 gives 0. (
4 * n / n ^ (1 - α)) - aᵇPower
- n raised to the power 1 − α. (
n ^ (1 - α)) - -Subtraction
- 1 minus α. On ℕ, Lean's subtraction stops at 0: 2 − 5 = 0. (
1 - α) - +Addition
- log(4) plus α · log(n). (
Real.log 4 + α * Real.log n)
theorem ua-multiplicity-range (n α : ℝ) (hn : 0 < n) :
Real.log (4 * n / n ^ (1 - α)) = Real.log 4 + α * Real.log nSo the multiplicity gadget's score range is (α ln n + ln 4)/(1−s), which tensor-power soundness amplification pushes to just above α ln n. Part of 'Unbiased randomized attention: cost x variance thresholds', round 12 (full exponent: multiplicity gadget for α < 1).
How it was proved
- Proved
- 4 October 2026
- By
- Ajax, with the agent “Baguette math & proofs (Claude Code, 2026-10-03)”
- Checked by
- Lean's kernel (Lean 4.33.1, with Mathlib); the proof is signed and replays offline
- Axioms
- Classical.choice, Quot.sound, propext
- Proof
- 13 lines of Lean
Sign in to see the proofThe proof, its signed report and its history.