From Alternation to FPRAS: Toward a Complexity Classification of Approximate Counting

TL;DR

This work tackles the challenge of classifying and enabling efficient approximation for counting problems whose decision version is easy but exact counting is hard. It introduces spanALP, a machine-model-based framework built from alternating transducers, to guarantee FPRAS for a broad class of problems, and situates it within the counting landscape between #L and TotP. The authors connect spanALP to context-free grammars (via spanALT and CFG-equivalences under affine reductions) and demonstrate applicability by deriving an FPRAS for Dyck-constrained walks, illustrating how the framework yields new approximability results. They also outline key open questions, including separations between spanL and spanALP and the search for broader practical algorithms arising from this theoretical model.

Abstract

Counting problems are fundamental across mathematics and computer science. Among the most subtle are those whose associated decision problem is solvable in polynomial time, yet whose exact counting version appears intractable. For some such problems, however, one can still obtain efficient randomized approximation in the form of a fully polynomial randomized approximation scheme (FPRAS). Existing proofs of FPRAS existence are often highly technical and problem-specific, offering limited insight into a more systematic complexity-theoretic account of approximability. In this work, we propose a machine-based framework for establishing the existence of an FPRAS beyond previous uniform criteria. Our starting point is alternating computation: we introduce a counting model obtained by equipping alternating Turing machines with a transducer-style output mechanism, and we use it to define a corresponding counting class spanALP. We show that every problem in spanALP admits an FPRAS, yielding a reusable sufficient condition that can be applied via reductions to alternating logspace, polynomial-time computation with output. We situate spanALP in the counting complexity landscape as strictly between #L and TotP (assuming RP $\neq$ NP) and observe interesting conceptual and technical gaps in the current machinery counting complexity. Moreover, as an illustrative application, we obtain an FPRAS for counting answers to counting the answers Dyck-constrained path queries in edge-labeled graphs, i.e., counting the number of distinct labelings realized by s-t walks whose label sequence is well-formed with respect to a Dyck-like language. To our knowledge, no FPRAS was previously known for this setting. We expect the alternating-transducer characterization to provide a broadly applicable tool for establishing FPRAS existence for further counting problems.

Figures (5)

• Figure 1: Counting complexity classes and their relation under affine polynomial-time reductions (except Proposition \ref{['prop:strict']}, which uses $c$-monious log-space reductions). A directed arrow from $A$ to $B$ indicates that $A$ is included in $B$. Color coding indicates easiness based on FPRAS, however spanL still contains hard counting problems (under FP $\neq$ #P): Green classes contain problems that can be efficiently approximated, orange classes are intermediate (easy decision problems) and red classes comprise very hard counting problems.
• Figure 2: Illustration of the outputs in \ref{['ex:cq.atrm']} for $q(x,w)=\exists y\,\exists z\;(S(x,y)\wedge R(y,z)\wedge T(z,w) \wedge U(x))$.
• Figure 3: Visualization of how the ATrM for Dyck-like walks uses log space via universal branching.
• Figure 4: Binarizing a universal node preserves the output under the $\circ$-transparent semantics.
• Figure :

Theorems & Definitions (40)

• Definition 2.1: FPRAS
• Proposition 2.2: Theorem 1 in MeelColnet26
• Corollary 2.3
• proof
• Definition 3.1
• Definition 3.2: Alternating Transducer Machine
• Definition 3.3: ATrM Output
• Definition 3.4: Output Set and Span
• Definition 3.5: The Class $\mathsf{span}\mathsf{ALT}\xspace(S,Z)$
• Proposition 3.6