Parameterized Complexity Results in Symmetry Breaking

Symmetry is a common feature of many combinatorial problems. Unfortunately eliminating all symmetry from a problem is often computationally intractable. This paper argues that recent parameterized complexity results provide insight into that intractability and help identify special cases in which symmetry can be dealt with more tractably

💡 Research Summary

The paper investigates why breaking symmetry in combinatorial problems is often computationally intractable and shows how recent results from parameterized complexity theory shed light on this difficulty. After reviewing the prevalence of symmetry in SAT, CSP, ILP, and related domains, the authors introduce the key concepts of fixed‑parameter tractability (FPT), the W‑hierarchy, and para‑NP‑hardness, and then define several natural parameters that capture the structure of a symmetry group: the number of generators (k), the maximum cycle length (l), the symmetry impact on variables (d), and the overall problem size (n).

The core technical contributions are three theorems. The first theorem proves that when the number of generators k is bounded by a constant, a symmetry‑breaking predicate can be constructed in time O(f(k)·poly(n)), i.e., the problem is FPT with respect to k. The algorithm builds a minimal representative set from the generators and adds a Lex‑Leader style ordering constraint for each representative, avoiding the exponential blow‑up of classic global methods.

The second theorem deals with Abelian symmetry groups whose longest cycle length l grows at most logarithmically with the instance size. By decomposing the group into independent cycles and imposing local ordering constraints, the authors obtain a total constraint size O(l·n) and guarantee that the search tree depth is bounded by O(log n). This result shows that even relatively large groups become tractable when their cyclic structure is shallow.

The third theorem focuses on the “symmetry impact” d, defined as the maximum number of variables that any symmetry maps onto each other. When d is small, a custom branching strategy that interleaves variable reordering with domain splitting can collapse symmetric sub‑trees into a single node. The authors prove that the overall search tree height becomes O(d·log n), a dramatic improvement over the naïve O(n) bound for unrestricted symmetry.

Conversely, the paper also establishes hardness results: if the combined parameter k·l·d is unbounded (i.e., grows super‑polynomially with n), the symmetry‑breaking problem becomes W