Approximate Counting for Complex-Weighted Boolean Constraint Satisfaction Problems

Constraint satisfaction problems (or CSPs) have been extensively studied in, for instance, artificial intelligence, database theory, graph theory, and statistical physics. From a practical viewpoint, it is beneficial to approximately solve those CSPs. When one tries to approximate the total number of truth assignments that satisfy all Boolean-valued constraints for (unweighted) Boolean CSPs, there is a known trichotomy theorem by which all such counting problems are neatly classified into exactly three categories under polynomial-time (randomized) approximation-preserving reductions. In contrast, we obtain a dichotomy theorem of approximate counting for complex-weighted Boolean CSPs, provided that all complex-valued unary constraints are freely available to use. It is the expressive power of free unary constraints that enables us to prove such a stronger, complete classification theorem. This discovery makes a step forward in the quest for the approximation-complexity classification of all counting CSPs. To deal with complex weights, we employ proof techniques of factorization and arity reduction along the line of solving Holant problems. Moreover, we introduce a novel notion of T-constructibility that naturally induces approximation-preserving reducibility. Our result also gives an approximation analogue of the dichotomy theorem on the complexity of exact counting for complex-weighted Boolean CSPs.

💡 Research Summary

The paper tackles the approximation‑complexity classification of Boolean constraint satisfaction problems (CSPs) when constraints are allowed to carry complex‑valued weights. In the unweighted (or real‑weighted) Boolean setting a celebrated trichotomy theorem classifies every counting CSP into exactly one of three categories under approximation‑preserving (AP) reductions: (i) problems admitting a fully polynomial‑time randomized approximation scheme (FPRAS), (ii) problems that are #BIS‑hard (believed to be intermediate), and (iii) #P‑hard problems. The authors ask whether a similarly clean classification exists for the richer class of complex‑weighted Boolean CSPs.

The main result is a dichotomy theorem: assuming that all complex‑valued unary constraints are freely available, every complex‑weighted Boolean CSP either (a) admits an FPRAS (hence is AP‑easy) or (b) is AP‑#P‑hard (hence AP‑intractable). No intermediate AP‑hard class appears. The availability of arbitrary unary constraints is crucial; it provides enough expressive power to “normalize’’ any instance by scaling variables and adjusting phases, which eliminates the subtle intermediate cases that arise in the unweighted setting.

To achieve this classification the authors blend techniques from the Holant framework with new algebraic tools. First, they embed a given complex‑weighted CSP into a Holant problem, where each variable corresponds to an edge and each constraint to a vertex function. This translation enables two structural transformations:

1. Factorization – the complex weight of a constraint is decomposed into a product of simpler factors, each of which can be realized by a combination of unary and binary functions.
2. Arity reduction – high‑arity constraints are simulated by a network of binary constraints without changing the overall partition function.

Both transformations preserve approximation‑preserving reducibility.

The central conceptual contribution is the notion of T‑constructibility. A constraint set 𝔽 is T‑constructible from another set 𝔾 if every function in 𝔽 can be approximated (within any desired multiplicative error) by a polynomial‑size gadget built from functions in 𝔾 together with the free unary constraints. T‑constructibility induces an AP‑reduction: if 𝔽 is T‑constructible from 𝔾, then #CSP(𝔽) ≤_AP #CSP(𝔾). This notion is stronger than the traditional notion of “realizability’’ used in exact counting dichotomies, because it explicitly respects approximation error.

Armed with T‑constructibility, the authors prove that any complex‑weighted Boolean CSP falls into one of two regimes:

• FPRAS‑eligible regime – the constraint set can be T‑constructed from a family of “tractable’’ functions, namely affine functions, product‑type functions, or functions that are essentially a product of a complex scalar and a real‑valued Boolean function. In this case the Holant representation reduces to a problem known to admit an FPRAS (e.g., the complex‑valued Ising model with ferromagnetic interactions after suitable phase adjustments).

• #P‑hard regime – the constraint set contains a function that cannot be expressed as a T‑constructible combination of the tractable families. By separating the real and imaginary parts of the complex weights, the authors reduce the counting problem to a pair of real‑valued counting CSPs, each of which is already known to be #P‑hard. The reduction preserves approximation, establishing AP‑#P‑hardness.

Because the free unary constraints allow arbitrary scaling and phase rotation of variables, any non‑tractable function can be “amplified’’ into a form that directly encodes a known #P‑hard problem, thereby eliminating the possibility of an intermediate AP‑hard class such as #BIS.

The paper also presents concrete examples to illustrate the dichotomy. For instance, the complex‑weighted Ising model with edge weight e^{iθ} is FPRAS‑easy when θ is a multiple of π (the model reduces to a real‑valued ferromagnetic Ising model) but becomes AP‑#P‑hard for generic θ because the phase cannot be eliminated by unary gadgets. Similarly, a Boolean circuit where each gate carries a complex amplitude is either approximable (if the amplitudes factor through a product‑type gadget) or AP‑#P‑hard (if the amplitudes encode a non‑realizable phase).

In addition to the classification, the authors discuss the relationship between their approximation dichotomy and the known exact‑counting dichotomy for complex‑weighted Boolean CSPs (which also hinges on the presence of all unary constraints). The parallelism underscores that the same algebraic boundary separating tractable from intractable functions governs both exact and approximate counting when the full unary toolbox is available.

Implications and future work – The result advances the broader program of classifying the approximation complexity of all counting CSPs. It suggests that, at least in the presence of unrestricted unary constraints, the landscape collapses to a simple two‑way split even for complex weights. Open directions include extending the dichotomy to settings where unary constraints are restricted (where intermediate AP‑hardness may reappear), exploring T‑constructibility for larger domains (e.g., q‑ary CSPs), and applying the Holant‑based techniques to quantum‑inspired counting problems where amplitudes are inherently complex.

Overall, the paper delivers a rigorous, technically deep, and conceptually elegant dichotomy for the approximation of complex‑weighted Boolean CSPs, introducing T‑constructibility as a powerful tool for AP‑reductions and bridging the gap between exact and approximate counting in the complex domain.