golden_bpe_bias_rates_agree
At the golden bias p_*, the population rate (1 - p_*)^2 / (2 - p_*) of nonoverlapping 00 pairs equals the population rate p_* * (1 - p_*) of 01 pairs.
Statement
theorem golden_bpe_bias_rates_agree :
(1 - golden_bpe_bias) ^ 2 / (2 - golden_bpe_bias) = golden_bpe_bias * (1 - golden_bpe_bias)- Related work
- knowledgebase
- Verified
- 12 Nov 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 2
- golden_bpe_biasThe golden critical bias p_* is defined as (3 - √5) / 2.
- golden_bpe_bias_polyThe golden bias p_* satisfies the quadratic polynomial equation p_*^2 - 3 p_* + 1 = 0.
Read back from the Lean
The real number (1 - golden_bpe_bias)^2 / (2 - golden_bpe_bias) equals golden_bpe_bias * (1 - golden_bpe_bias), where golden_bpe_bias is the real number (3 - √5) / 2.
Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.
The proof, explained
The proof establishes that the asymptotic rate $(1 - p_*)^2 / (2 - p_*)$ of 00 pairs equals the rate $p_*(1 - p_*)$ of 01 pairs at the golden bias $p_*$. It first verifies that the denominator $2 - p_*$ is nonzero by showing that $p_* = 2$ would contradict the quadratic equation $p_*^2 - 3p_* + 1 = 0$. Clearing the denominator reduces the goal to $(1 - p_*)^2 = p_*(1 - p_*)(2 - p_*)$, which holds because their difference is precisely $(1 - p_*)(p_*^2 - 3p_* + 1) = 0$.
The golden BPE bias $p_*$ satisfies the quadratic relation $p_*^2 - 3p_* + 1 = 0$.
Retrieve the characteristic quadratic polynomial relation `golden_bpe_bias_poly` for $p_*$, naming it hypothesis $hp$ for subsequent algebraic steps.
uses
golden_bpe_bias_polyhave hp := golden_bpe_bias_polyThe denominator $2 - p_*$ is nonzero.
Assume for contradiction that $2 - p_* = 0$, which yields $p_* = 2$. Substituting $p_* = 2$ into the quadratic relation $p_*^2 - 3p_* + 1 = 0$ yields $2^2 - 3(2) + 1 = -1 = 0$, an arithmetic contradiction closed by `nlinarith`.
have h2 : 2 - golden_bpe_bias ≠ 0 := by intro h have h_val : golden_bpe_bias = 2 := by linarith nlinarith
The target identity is equivalent to clearing the denominator: $(1 - p_*)^2 = p_*(1 - p_*)(2 - p_*)$.
Because the denominator $2 - p_*$ is nonzero, rewrite the division on the left-hand side using `div_eq_iff`.
uses
div_eq_iffrw [div_eq_iff h2]The cleared identity $(1 - p_*)^2 = p_*(1 - p_*)(2 - p_*)$ holds, completing the proof.
The difference between the two sides factors as $(1 - p_*)^2 - p_*(1 - p_*)(2 - p_*) = (1 - p_*)(p_*^2 - 3p_* + 1)$. Since $p_*^2 - 3p_* + 1 = 0$ by hypothesis $hp$, this algebraic combination vanishes identically.
linear_combination (1 - golden_bpe_bias) * hp
golden_bpe_bias_poly- The golden BPE bias satisfies the quadratic relation $p_*^2 - 3p_* + 1 = 0$.
div_eq_iff- In a group with zero (such as the real numbers), if $b \\neq 0$, then $a / b = c$ if and only if $a = c \\cdot b$.
Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
/-- At the golden bias p_*, the asymptotic nonoverlapping merge rate (1 - p_*)^2 / (2 - p_*) of pair 00 equals the rate p_*(1 - p_*) of pair 01. -/theorem golden_bpe_bias_rates_agree : (1 - golden_bpe_bias) ^ 2 / (2 - golden_bpe_bias) = golden_bpe_bias * (1 - golden_bpe_bias) := by have hp := golden_bpe_bias_poly have h2 : 2 - golden_bpe_bias ≠ 0 := by intro h have h_val : golden_bpe_bias = 2 := by linarith nlinarith rw [div_eq_iff h2] linear_combination (1 - golden_bpe_bias) * hp