The number system hidden inside the Boolean satisfiability problem

This paper gives a novel approach to analyze SAT problem more deeply. First, I define new elements of Boolean formula such as dominant variable, decision chain, and chain coupler. Through the analysis of the SAT problem using the elements, I prove that we can construct a k-SAT (k>2) instance where the coefficients of cutting planes take exponentially large values in the input size. This exponential property is caused by the number system formed from the calculation of coefficients. In addition, I show that 2-SAT does not form the number system and Horn-SAT partially forms the number system according to the feasible value of the dominant variable. Whether or not the coefficients of cutting planes in cutting plane proof are polynomially bounded was open problem. Many researchers believed that cutting plane proofs with large coefficients are highly non-intuitive20. However, we can construct a k-SAT (k>2) instance in which cutting planes take exponentially large coefficients by the number system. In addition, this exponential property is so strong that it gives definite answers for several questions: why Horn-SAT has the intermediate property between 2-SAT and 3-SAT; why random-SAT is so easy; and why k-SAT (k>2) cannot be solved with the linear programming technique. As we know, 2-SAT is NL-complete, Horn-SAT is P-complete, and k-SAT (k>2) is NP-complete. In terms of computational complexity, this paper gives a clear mathematical property by which SAT problems in three different classes are distinguished.

💡 Research Summary

The paper proposes a novel structural analysis of Boolean satisfiability (SAT) by introducing three new notions: dominant variable, decision chain, and chain coupler. A dominant variable is a literal whose assignment alone determines the satisfiability of the whole formula. A decision chain is a linear sequence of variables linked by logical dependencies, each variable’s value being forced by the preceding one. A chain coupler connects two or more decision chains, allowing their logical flows to interact.

Using these constructs, the author shows how to build a family of k‑SAT instances (k > 2) whose cutting‑plane proofs necessarily involve linear inequalities with coefficients that grow exponentially in the size of the input. The key observation is that when decision chains of length ℓ are linked via couplers, the coefficients contributed by each chain are powers of two (≈ 2^ℓ). By arranging O(n) such chains, the resulting coefficients become on the order of 2^{Θ(n)}. Consequently, any cutting‑plane derivation for these instances must manipulate numbers of exponential magnitude; the coefficients cannot be bounded by any polynomial in n.

The paper contrasts this behavior with two well‑studied subclasses of SAT. In 2‑SAT every clause contains at most two literals, which precludes the formation of decision chains longer than a constant; thus cutting‑plane coefficients remain polynomially bounded. Horn‑SAT, where each clause contains at most one positive literal, can admit limited chain structures, but the presence of a single positive literal restricts the dominant variable’s feasible values. As a result, Horn‑SAT exhibits an intermediate growth: coefficients may become super‑polynomial but not fully exponential (e.g., 2^{Θ(√n)}). This intermediate growth aligns with the known complexity classification—Horn‑SAT is P‑complete, sitting between NL‑complete 2‑SAT and NP‑complete k‑SAT.

Beyond the technical contribution, the author draws several broader implications. First, the long‑standing open question of whether cutting‑plane proofs can be kept within polynomially bounded coefficients receives a negative answer for k‑SAT (k > 2). Second, the “number system” perspective explains why random SAT formulas are empirically easy: random clauses rarely generate the long decision chains required for exponential coefficient blow‑up, so cutting‑plane methods remain tractable in practice. Third, the result clarifies why linear‑programming (LP) relaxations cannot solve general k‑SAT efficiently; LP solvers assume coefficients of manageable size, while the constructed instances force coefficients beyond any polynomial bound, breaking the LP approach.

The paper’s strengths lie in its fresh terminology and the clear construction that links combinatorial structure (chains and couplers) to arithmetic growth (exponential coefficients). It connects three distinct complexity classes (2‑SAT, Horn‑SAT, k‑SAT) through a unified algebraic lens, offering an intuitive explanation for their differing algorithmic behaviors.

However, several limitations temper the impact. The definitions of dominant variable, decision chain, and chain coupler are presented at a high level without algorithmic procedures for detecting or extracting them from arbitrary SAT instances. Consequently, it is unclear how often such structures appear in natural or benchmark formulas beyond the deliberately engineered examples. The paper also does not quantify how the exponential coefficient growth interacts with proof length; existing lower‑bound results (e.g., Håstad’s exponential resolution lower bounds) already imply hardness, but the added dimension of coefficient size needs a more rigorous comparative analysis. Moreover, the claim that the number system explains the ease of random SAT lacks a probabilistic model showing that long decision chains have negligible probability in the uniform random clause distribution. Finally, while the work references the open problem about polynomially bounded cutting‑plane coefficients, it does not discuss related work on coefficient growth in integer programming (e.g., Cook et al.’s bounds on facet coefficients) that could situate the contribution within a broader literature.

In summary, the paper introduces a compelling algebraic framework that demonstrates exponential coefficient blow‑up in cutting‑plane proofs for k‑SAT (k > 2) and uses this to distinguish the three major SAT subclasses. The construction is elegant and the high‑level insights are valuable, but the practical relevance and theoretical integration would benefit from more detailed algorithmic descriptions, probabilistic analysis of random instances, and a deeper comparison with existing lower‑bound techniques in proof complexity and integer programming.