Lemma 5.8 — classical progress against the injective run
OpenCompressedOracleDuality.LabelSymmetric.Supplement.lemma_5_8classical-computationcollision-findingcomplexity-theorycryptographyp2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let , , and let be any deterministic label-symmetric branching program of length , with at most vertices in each internal layer and unrestricted finite output leaves. Fix the canonical injective oracle and define from that fixed execution. Let be the uniform probability, over all functions , that the actual output reports and . Then
There is no success-threshold hypothesis and no square-root restriction on or on the summation indices. The additive is retained. At the sum is empty and the bound applies to the fixed output. The fixed injective execution determines every summand independently of , so the bound controls actual average collision success.
Differences from the source:
- encoding: one carrier [N] for inputs and labels → published Fin (2^n) inputs and n-bit labels, identified by the fixed most-significant-bit-first bijection e; reason: e transports every oracle and label permutation bijectively; input ordering is the published strict order
- encoding: a singleton start layer, exactly 2^S internal vertices, and arbitrarily many terminal leaves → finite layer types with a Unique initial layer and at most 2^S vertices only for 1 <= t < T; reason: padding unreachable internal vertices recovers the printed width without altering executions, symmetry, or success; T=0 retains one fixed output
- encoding: the unique minimum matched set for t < N/2 → a minimum-cardinality selector for 2*t < 2^n, with empty-set totalization otherwise; reason: finiteness ensures a minimizer and label symmetry ensures uniqueness in the used regime; no matched-set property is assumed
- encoding: H*(x)=x and L_t on the common input/label carrier → H(x)=e(x) and label subsets L*_t from the fixed canonical binary identification; reason: e is a bijection between the published bases, so it represents the same fixed injective-oracle comparison
- encoding: T < N/2 and the sum over t<T → 2*T < 2^n and a real finite sum indexed by Fin T; reason: the time range and every summand are unchanged; no terminal-layer width restriction is added
Preamble
import Definitions.Def_CompressedOracleDuality_LabelSymmetric_Supplement_MatchedLabels open scoped BigOperators
Formal statement
namespace CompressedOracleDuality.LabelSymmetric.Supplement
/-- Lemma 5.8, (84), p.29: the full T<N/2 regime and the additive guessed-collision term. -/
theorem lemma_5_8 {n S T : ℕ} (B : ClassicalProgram n T)
(hn : 6 ≤ n) (hS : n ≤ S) (hwidth : B.SpaceBound S)
(hsym : ClassicalLabelSymmetric B) (hT : 2 * T < 2 ^ n) :
classicalSuccess B ≤
(1 + ∑ t : Fin T, ((injectiveLabels B t.castSucc).card : ℝ)) / (2 : ℝ) ^ n := by sorry
end CompressedOracleDuality.LabelSymmetric.Supplement
Source
Ding and Zhandry, Compressed Oracle Duality: A simple framework for knowing what quantum adversaries know, October 30, 2026 preprint, arXiv:submit/8306656 [quant-ph], STOC 2027 submission #94; source.pdf SHA-256 6614c6a707d45eff1e475fa25c9c0d22747adf304a0680ebf79d2bd118122daa; https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/94.pdf, Lemma 5.8, (84), p.29; proof pp.29-30; standing assumptions p.11