CDCL is a Collapse Chain: A Formal Equivalence Between Conflict-Driven Clause Learning and Irreversible Uncertainty Reduction

Fuente: Zenodo
Salvato in:
Dettagli Bibliografici
Autore principale: Rao, Huiying
Natura: Recurso digital
Lingua:inglese
Pubblicazione: Zenodo 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866901773311016960
author Rao, Huiying
author_facet Rao, Huiying
contents <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">We prove that the Conflict-Driven Clause Learning (CDCL) algorithm is formally equivalent to a branching collapse chain: a collection of collapse chains that share information through learned clauses, with globally monotone non-decreasing information content despite local backtracking.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The equivalence resolves the backtracking paradox: backtracking does not reverse collapse. It terminates the current chain and initiates a new one from a checkpoint, with the learned clause serving as the irreversible information transfer between chains. The global information content of CDCL is monotone non-decreasing (Information Monotonicity Theorem), consistent with the branching collapse chain definition.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">Three consequences follow: (1) The Information Monotonicity Theorem — each k-literal learned clause contributes n−k bits of information, explaining why modern CDCL implementations prioritise short clauses; (2) The Gradient Impossibility Theorem — no differentiable collapse operator can improve upon CDCL for SAT near the phase transition, providing a principled explanation for the hard wall; (3) Empirical validation — 47 out of 100 variables exhibit intrinsic uncertainty across satisfying assignments at α=4.2, confirming that residual uncertainty after CDCL is structural, not epistemic.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">Part of the NSD / MOEH research programme. Companion papers: "Collapse as a Unifying Language for NP-Hard Optimisation" (Zenodo: 10.5281/zenodo.19574013) and "A Landscape-Aware Classification Framework for NP-Hard Optimisation Problems" (under review at JAIR #22499).</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_19600733
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle CDCL is a Collapse Chain: A Formal Equivalence Between Conflict-Driven Clause Learning and Irreversible Uncertainty Reduction
Rao, Huiying
CDCL
SAT
collapse chain
unit propagation
clause learning
phase transition
uncertainty reduction
information monotonicity
hard wall
NP-hard
<p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">We prove that the Conflict-Driven Clause Learning (CDCL) algorithm is formally equivalent to a branching collapse chain: a collection of collapse chains that share information through learned clauses, with globally monotone non-decreasing information content despite local backtracking.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">The equivalence resolves the backtracking paradox: backtracking does not reverse collapse. It terminates the current chain and initiates a new one from a checkpoint, with the learned clause serving as the irreversible information transfer between chains. The global information content of CDCL is monotone non-decreasing (Information Monotonicity Theorem), consistent with the branching collapse chain definition.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">Three consequences follow: (1) The Information Monotonicity Theorem — each k-literal learned clause contributes n−k bits of information, explaining why modern CDCL implementations prioritise short clauses; (2) The Gradient Impossibility Theorem — no differentiable collapse operator can improve upon CDCL for SAT near the phase transition, providing a principled explanation for the hard wall; (3) Empirical validation — 47 out of 100 variables exhibit intrinsic uncertainty across satisfying assignments at α=4.2, confirming that residual uncertainty after CDCL is structural, not epistemic.</p> <p class="font-claude-response-body break-words whitespace-normal leading-[1.7]">Part of the NSD / MOEH research programme. Companion papers: "Collapse as a Unifying Language for NP-Hard Optimisation" (Zenodo: 10.5281/zenodo.19574013) and "A Landscape-Aware Classification Framework for NP-Hard Optimisation Problems" (under review at JAIR #22499).</p>
title CDCL is a Collapse Chain: A Formal Equivalence Between Conflict-Driven Clause Learning and Irreversible Uncertainty Reduction
topic CDCL
SAT
collapse chain
unit propagation
clause learning
phase transition
uncertainty reduction
information monotonicity
hard wall
NP-hard
url https://doi.org/10.5281/zenodo.19600733