Approximation Complexity of Complex-Weighted Degree-Two Counting Constraint Satisfaction Problems

Constraint satisfaction problems have been studied in numerous fields with practical and theoretical interests. In recent years, major breakthroughs have been made in a study of counting constraint satisfaction problems (or #CSPs). In particular, a computational complexity classification of bounded-degree #CSPs has been discovered for all degrees except for two, where the “degree” of an input instance is the maximal number of times that each input variable appears in a given set of constraints. Despite the efforts of recent studies, however, a complexity classification of degree-2 #CSPs has eluded from our understandings. This paper challenges this open problem and gives its partial solution by applying two novel proof techniques–T_{2}-constructibility and parametrized symmetrization–which are specifically designed to handle “arbitrary” constraints under randomized approximation-preserving reductions. We partition entire constraints into four sets and we classify the approximation complexity of all degree-2 #CSPs whose constraints are drawn from two of the four sets into two categories: problems computable in polynomial-time or problems that are at least as hard as #SAT. Our proof exploits a close relationship between complex-weighted degree-2 #CSPs and Holant problems, which are a natural generalization of complex-weighted #CSPs.

💡 Research Summary

The paper tackles a long‑standing open problem in the theory of counting constraint satisfaction problems (#CSPs): the approximation complexity of degree‑2 instances when constraints are allowed to have arbitrary complex weights. While a complete classification exists for bounded‑degree #CSPs of degree three or higher, the case of degree two has resisted prior techniques because non‑symmetric (asymmetric) signatures can appear arbitrarily often, breaking the symmetry‑based arguments that underlie most known dichotomies.

The authors introduce two novel technical tools designed specifically for degree‑2 problems. The first, T₂‑constructibility, is a constructive reduction that, given any ternary signature f, produces a new ternary signature Sym(f) that is symmetric. The reduction preserves approximation complexity under AP‑reductions, i.e., #CSP₂(f) is AP‑equivalent to #CSP₂(Sym(f)). The second tool, parameterized symmetrization, extends this idea by also constructing a binary symmetric signature SymL(f) together with a family of parameters that simultaneously handle infinitely many related signatures. This richer symmetrization is needed for a particular subclass of signatures (the set SIG₁) where a single ternary symmetrization is insufficient.

The paper partitions all possible signatures into four families, denoted SIG₀, SIG₁, SIG₂, and SIG₃. The main results concern signatures outside SIG (Theorem 3.4) and those inside SIG₁ (Theorem 3.5).

Theorem 3.4 states that for any ternary signature f not belonging to the union SIG = SIG₀ ∪ SIG₁ ∪ SIG₂ ∪ SIG₃, the problem #CSP₂(f) is AP‑hard for the complex‑valued counting satisfiability problem (#SAT₍ℂ₎). The proof proceeds by applying T₂‑constructibility to obtain Sym(f), which is symmetric, and then invoking the known dichotomy for symmetric Holant* problems (Cai‑Lu‑Xia 2013). Because #CSP₂(f) and the corresponding Holant* problem are AP‑equivalent, the hardness transfers.

Theorem 3.5 gives a complete classification for signatures drawn from SIG₁. Within this set the authors define a special subclass called DUP (short for “duplicate”), consisting of signatures that can be expressed as products of unary functions applied to each variable. If a set F of signatures is contained in DUP, then #CSP₂(F) is solvable in polynomial time; otherwise, #CSP₂(F) is AP‑hard for the complex‑valued #P class denoted #PC (a natural complex analogue of #P). The proof again uses T₂‑constructibility to produce Sym(f) and SymL(f). The authors then formulate a system of low‑degree multivariate polynomial equations that would need to have a common solution in ℂ for the problem to be easy. By a straightforward algebraic argument they show that no such solution exists unless the signatures belong to DUP, thereby establishing hardness.

A crucial structural observation underlying the whole work is the AP‑equivalence between degree‑2 #CSPs and Holant problems* (the latter allow free unary signatures). This equivalence lets the authors translate results from the Holant framework—where powerful holographic transformations and dichotomies are already known—into the #CSP setting. The paper also revisits the notion of holographic transformation, showing how it can be combined with the new constructibility techniques to preserve approximation complexity while reshaping signatures.

The contributions can be summarized as follows:

1. New proof techniques – T₂‑constructibility and parameterized symmetrization – that enable the conversion of arbitrary degree‑2 signatures into symmetric ones without losing AP‑equivalence.
2. A complete dichotomy for a large class of degree‑2 #CSPs – either polynomial‑time solvable (when the signatures lie in DUP) or AP‑hard for #SAT₍ℂ₎/#PC.
3. Bridging #CSP and Holant theories – establishing AP‑equivalence for degree‑2 instances and leveraging the existing symmetric Holant dichotomy to obtain new #CSP results.
4. An elementary algebraic argument – the hardness proof reduces to showing that a certain system of low‑degree polynomial equations has no common complex solution, avoiding heavy algebraic geometry.

The paper thus resolves the approximation complexity of a substantial portion of degree‑2 complex‑weighted #CSPs, filling the gap left by previous classifications for higher degrees. It opens several avenues for future work: extending T₂‑constructibility to higher degrees, exploring analogous classifications for non‑Boolean domains (q‑ary variables), and applying the parameterized symmetrization framework to other counting problems such as quantum circuit simulation or partition functions of statistical‑physics models.