Lemma 6.8 (used in the proof of Theorem 6.5) — weighted-phase solution count
OpenCNFLearning873.Valiant.Supplement.lemma_6_8cnf-learningcombinatoricsp2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
Let , and , with the oracle learner's specified sample count and round count. Fix a satisfiable -CNF and learning samples that are solutions of and satisfy every survivor's violation bound from Lemma 6.7. Over only the ideal internal randomness remaining in Step 2,
The statement includes Step 1's empty-survivor True return. For nonempty survivors, the formula is the best candidate actually stored by the weighted-sampling phase. The ideal samplers are untruncated here; the implementation's separate abort budget is included by Theorem 6.5.
Differences from the source:
- encoding: Boolean assignments on a named n-variable set → assignments on Fin n, using the published CNF and nonempty uniform-solution laws; reason: variable relabeling changes no instance; output satisfiability prevents an undefined uniform law
- encoding: condition on the good event and fix learning samples before Lemma 6.8 → solution-valued fixed sample sequence and GoodSamples hypotheses, with the actual ideal phase law; reason: makes the inherited p.29 context explicit; no new restriction on the source executions
Preamble
import Definitions.Def_CNFLearning873_Valiant_Supplement_OracleLearner
Formal statement
namespace CNFLearning873.Valiant.Supplement
/-- Lemma 6.8, pp.29-30 = PDF pp.29-30. The p.29 conditioning paragraph is explicit;
probability is over the ideal remaining internal samplers, with the learning samples fixed. -/
theorem lemma_6_8 (k : ℕ) (hk : 3 ≤ k) :
∃ n₀ : ℕ, 3 ≤ n₀ ∧ ∀ d : ℕ, 3 ≤ d → ∀ n : ℕ, n₀ ≤ n →
∀ ε δ : ℝ, 0 < ε → ε < 1 → 0 < δ → δ < 1 →
∀ F : CNF n, KD F k d → ∀ h : (solutions F).Nonempty,
∀ xs : Fin (learningSamples n k d ε δ) → Assignment n,
(∀ i, xs i ∈ solutions F) → GoodSamples F h k (rounds n k d ε δ) ε xs →
1 - δ / 3 ≤ probability (ideal (phase2 k d ε δ xs))
(fun st => (solutions (phaseFormula st)).card ≤ 2 * (solutions F).card) := by sorry
end CNFLearning873.Valiant.Supplement
Source
Weiming Feng, Yixiao Yu, Yiyao Zhang, STOC 2027 submission #873 (October 2026), Lemma 6.8, printed p.29 = 1-based PDF p.29; https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/873.pdf