← 研踪 PaperTrailarXiv:2611.24548 · v1
Further simulations between tropical and linear programming
首次提交 2026-11-18 22:52
版本更新 2026-11-18 22:52
雷达观测 2026-11-19 15:04
摘要
Propositional proof systems that apply various kinds of cuts to linear inequalities (such as Cutting Planes [CCT87] and Lift&Project [BCC93,Pud99,Das05]) or operate with disjunctions of linear inequalities (such as Res(CP) [Kra98] and Res(LP) [HK06]) have been in the focus of research for the past 30 years. Recently, tropical proof systems were introduced [AGH25]; they resemble the latter systems but differ from them in important respects. The relations between the traditional systems and the new tropical systems have remained largely unknown, and exponential lower bounds for tropical systems were limited to non-CNF examples. In this paper we prove several new polynomial simulations between these systems. To this end, we introduce Thr-Res (Threshold Resolution), a new natural daglike proof system operating on threshold conditions over the Boolean literals: lines asserting that at least s literals of a given multiset are true. Its main difference from CP is that the rounding rule (without it CP is incomplete) is replaced by a weaker rule that allows one to perform "resolution" steps with linear combinations beyond Boolean clauses; also, in its addition rules, Thr-Res resembles input resolution [Cha70]: a derived line is combined with an input clause or an axiom, whereas CP can add up any two derived lines. By varying the maximum value of s, one obtains systems ranging in strength from Resolution to MP-NSR, the static tropical (min-plus) proof system introduced in [AGH25]. As a consequence, we prove that MP-NSR is polynomially simulated by CP and L&P (even by their unary versions), which automatically yields exponential lower bounds on the size of MP-NSR proofs for formulas in CNF. We complement it with simulations in the converse direction: binary daglike MP-PCR (a dynamic version of MP-NSR) polynomially simulates L&P; it also polynomially simulates CP when division axioms are added.
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-18 22:52 · 最新观测版本
Further simulations between tropical and linear programming
Propositional proof systems that apply various kinds of cuts to linear inequalities (such as Cutting Planes [CCT87] and Lift&Project [BCC93,Pud99,Das05]) or operate with disjunctions of linear inequalities (such as Res(CP) [Kra98] and Res(LP) [HK06]) have been in the focus of research for the past 30 years. Recently, tropical proof systems were introduced [AGH25]; they resemble the latter systems but differ from them in important respects. The relations between the traditional systems and the new tropical systems have remained largely unknown, and exponential lower bounds for tropical systems were limited to non-CNF examples. In this paper we prove several new polynomial simulations between these systems. To this end, we introduce Thr-Res (Threshold Resolution), a new natural daglike proof system operating on threshold conditions over the Boolean literals: lines asserting that at least s literals of a given multiset are true. Its main difference from CP is that the rounding rule (without it CP is incomplete) is replaced by a weaker rule that allows one to perform "resolution" steps with linear combinations beyond Boolean clauses; also, in its addition rules, Thr-Res resembles input resolution [Cha70]: a derived line is combined with an input clause or an axiom, whereas CP can add up any two derived lines. By varying the maximum value of s, one obtains systems ranging in strength from Resolution to MP-NSR, the static tropical (min-plus) proof system introduced in [AGH25]. As a consequence, we prove that MP-NSR is polynomially simulated by CP and L&P (even by their unary versions), which automatically yields exponential lower bounds on the size of MP-NSR proofs for formulas in CNF. We complement it with simulations in the converse direction: binary daglike MP-PCR (a dynamic version of MP-NSR) polynomially simulates L&P; it also polynomially simulates CP when division axioms are added.
查看此版本解读 →查看此版本原文 ↗