SAT Encodings for Bandwidth Coloring: A Systematic Design Study

The Bandwidth Coloring Problem (BCP) generalizes graph coloring by enforcing minimum separation constraints between adjacent vertices and arises in frequency assignment applications. While SAT-based approaches have shown promise for exact BCP solving, the encoding design space remains largely unexplored. This paper presents a systematic study of SAT encodings for the BCP, proposing a unified framework with six encoding methods across three categories: one-variable, two-variable, and block encodings. We evaluate the impact of key features including incremental solving and symmetry breaking. While symmetry breaking has been studied for graph coloring, it has not been systematically evaluated for SAT-based BCP solvers. Our analysis reveals significant interaction effects between encoding choices and solver configurations. The proposed framework achieves state-of-the-art performance on GEOM and MS-CAP benchmarks. Block encodings solve GEOM120b, the hardest instance, to proven optimality in approximately 1000 seconds, whereas previous methods could not solve it within a one-hour time limit.

💡 Research Summary

The paper conducts a systematic design study of SAT encodings for the Bandwidth Coloring Problem (BCP), a generalization of graph coloring where each edge carries a minimum separation requirement between the colors of its incident vertices. While SAT‑based exact solvers have shown promise for the classical graph coloring problem, the space of possible encodings for BCP has remained largely unexplored. To fill this gap, the authors propose a unified framework that defines six distinct encodings grouped into three families: one‑variable encodings (1G and 1L), two‑variable encodings (2G and 2L), and block encodings (X and Xa). The “G” variants use a greater‑than semantic (a Boolean variable indicates that the color of a vertex exceeds a given threshold), while the “L” variants use a less‑than semantic. One‑variable encodings assign a Boolean variable to each vertex‑color pair; two‑variable encodings introduce a pair of variables per edge to directly model the distance constraint; block encodings organize variables into structured blocks that enforce a global ordering within each block, with Xa allowing the block width to vary adaptively.

Beyond the raw encodings, the study investigates two configurable solver features: incremental SAT solving and symmetry breaking. Incremental solving starts from a loose upper bound on the span and iteratively tightens it, re‑using learned clauses across iterations. Symmetry breaking adapts techniques from graph coloring (fixing the order of colors, fixing a vertex ordering) and introduces a novel “distance symmetry” that exploits the fact that swapping colors while preserving all distance constraints yields equivalent solutions.

The experimental evaluation uses two benchmark suites, GEOM and MS‑CAP, comprising a total of over thirty instances of varying difficulty. The authors compare all six encodings under four configurations (with/without incremental solving, with/without symmetry breaking) using state‑of‑the‑art SAT solvers. The results reveal several key findings:

1. Block encodings dramatically outperform both one‑variable and two‑variable encodings. On average they achieve a 1.8× speed‑up and solve a higher percentage of instances within the time limit.
2. Incremental solving provides a substantial benefit for block encodings (≈30‑40 % reduction in solving time) but yields negligible gains for one‑variable encodings, indicating that clause reuse is more valuable when the underlying constraint network is dense.
3. Symmetry breaking effectiveness is highly encoding‑dependent. For block encodings, the distance‑symmetry constraints lead to the largest performance boost, whereas for one‑variable encodings the classic color‑order breaking can actually degrade performance due to added overhead.
4. The interaction between incremental solving and symmetry breaking is not additive; in some cases the combination yields synergistic improvements, while in others the benefits cancel out. This underscores the need for a tailored configuration per encoding type.

The most striking result is that the block encoding X, combined with both incremental solving and distance‑symmetry breaking, solves the notoriously hard GEOM120b instance in roughly 1,000 seconds, whereas previous state‑of‑the‑art SAT‑based approaches (including the POP‑S‑B and POPH‑S‑B partial‑ordering encodings) failed to solve it within a one‑hour limit. This demonstrates that the reduction of constraints to O(H·|E|) (instead of O(H·|E|·\bar d)) together with structured variable grouping yields a dramatically more tractable SAT formulation.

The paper concludes by highlighting several avenues for future work: automatic tuning of block widths, dynamic symmetry‑breaking strategies that adapt during search, and portfolio approaches that select the best encoding‑feature combination per instance. It also suggests that the proposed framework could be extended to related problems such as T‑coloring and the bandwidth multicoloring problem (BMCP). Overall, the study provides a comprehensive empirical foundation showing that careful SAT encoding design, coupled with appropriate solver features, can push the frontier of exact BCP solving well beyond what was previously achievable.