Monomials in arithmetic circuits: Complete problems in the counting hierarchy

We consider the complexity of two questions on polynomials given by arithmetic circuits: testing whether a monomial is present and counting the number of monomials. We show that these problems are complete for subclasses of the counting hierarchy which had few or no known natural complete problems. We also study these questions for circuits computing multilinear polynomials.

💡 Research Summary

This paper investigates two fundamental decision problems concerning polynomials represented by arithmetic circuits: (1) determining whether a given monomial appears with a non‑zero coefficient (the Zero Monomial Coefficient problem, ZMC) and (2) counting how many distinct monomials the circuit’s output polynomial contains (the CountMon problem). The authors place these problems within the counting hierarchy (CH) and prove completeness results for several of its subclasses, thereby providing natural complete problems for levels of CH that previously lacked such examples.

For ZMC, the input is an arithmetic circuit C and a monomial m (encoded by the binary exponents of its variables). The authors first show that when C is multiplicatively disjoint (or a formula), ZMC is C = P‑complete. The reduction is from the permanent‑equality problem (PER =), which asks whether the permanent of a {‑1,0,1} matrix equals a given integer d. By constructing a product‑of‑sums circuit Q whose monomial Y₁Y₂…Yₙ has coefficient exactly per(A), they reduce PER = to checking whether the coefficient of Y₁…Yₙ in Q − d·Y₁…Yₙ is zero. To place ZMC in C = P they use parse‑tree analysis: each parse tree contributes +1 or –1 to the coefficient depending on the parity of −1‑labeled leaves, and the coefficient is zero exactly when the numbers of positively and negatively contributing trees are equal. This equality can be expressed as a language L in P, yielding a log‑space many‑one reduction and establishing C = P‑completeness.

For unrestricted circuits the authors prove ZMC lies in coRP·PP. They employ the CoeffSLP oracle (computing a monomial’s coefficient modulo a prime) which is in FP#P. By randomly selecting a prime p < 2^{c·n} (c from a number‑theoretic lemma guaranteeing many suitable primes) and checking whether the coefficient is 0 mod p, they obtain a one‑sided error algorithm. Hence ZMC ∈ coRP·PP.

When the circuit is monotone (constants are only 1), ZMC becomes coNP‑complete. Hardness follows from a reduction of Exact‑3‑Cover: a monotone formula F = ∏{i=1}^m (1 + ∏{j∈C_i} X_j) contains the monomial X₁X₂…Xₙ iff the instance has an exact cover. Membership in coNP is shown via “parse‑tree types”: a monomial has a non‑zero coefficient iff there exists a valid parse‑tree type, a condition checkable in polynomial time, giving a coNP algorithm.

The second problem, CountMon, asks whether the output polynomial of a circuit C has at least d monomials. The authors first study the auxiliary problem ExistExtMon: given C and a monomial m, does there exist a monomial M that extends m (i.e., M = m·m′ with disjoint variable sets)? They prove ExistExtMon is PP‑complete by a reduction from a PP‑complete language. Using this, they show CountMon is PP^{PP}‑complete (equivalently, CC^{≠}P‑complete), placing it at the second level of CH. The reduction essentially embeds a PP‑complete decision into the existence of an extending monomial and then asks whether the total number of monomials reaches a threshold.

Finally, the authors consider the special case of multilinear circuits (each variable appears with exponent at most one). In this setting ZMC becomes equivalent to the classic arithmetic circuit identity testing (ACIT) problem, which lies in coRP, and CountMon drops to PP‑complete. The multilinear restriction eliminates the need to track signs of parse trees, simplifying the complexity.

Overall, the paper provides the first natural complete problems for the classes C = P, PP^{NP}, and PP^{PP} within the counting hierarchy, using elegant reductions based on permanents, parse‑tree parity, and extensions of monomials. It also highlights how structural restrictions on circuits (monotonicity, multilinearity) affect the complexity, offering a nuanced view of the interplay between algebraic representation and counting complexity.