A Superpolynomial Lower Bound on the Size of Uniform Non-constant-depth Threshold Circuits for the Permanent

We show that the permanent cannot be computed by DLOGTIME-uniform threshold or arithmetic circuits of depth o(log log n) and polynomial size.

💡 Research Summary

The paper establishes a super‑polynomial lower bound on the size of DLOGTIME‑uniform threshold and arithmetic circuits that compute the permanent, even when the circuit depth is allowed to grow sub‑log‑logarithmically. The permanent, a classic #P‑complete function, has long been a benchmark for proving circuit lower bounds. Prior work showed that constant‑depth TC⁰ or AC⁰ circuits cannot compute the permanent with polynomial size, but the situation for slightly deeper circuits—specifically those of depth o(log log n)—remained open.

The authors first formalize the models under consideration. A DLOGTIME‑uniform threshold circuit is a Boolean circuit composed of majority (threshold) gates whose wiring and gate functions can be described by a deterministic log‑time algorithm; similarly, a DLOGTIME‑uniform arithmetic circuit uses only addition and multiplication gates and is described uniformly in log‑time. The depth restriction o(log log n) means that as the input size n grows, the number of layers grows slower than any constant multiple of log log n.

The main theorem states: The permanent cannot be computed by any DLOGTIME‑uniform threshold or arithmetic circuit of depth o(log log n) and polynomial size. In other words, any such circuit family must have size at least 2^{Ω(n^{c})} for some constant c>0 that depends on the depth bound. This result lifts the known constant‑depth lower bounds to a whole new regime of “sub‑log‑log” depth, showing that even modest depth increases do not suffice to bring the permanent into polynomial‑size uniform circuits.

The proof proceeds in two major phases. The first phase introduces a novel random‑restriction technique tailored to preserve DLOGTIME‑uniformity while drastically simplifying the circuit. Traditional Håstad‑style restrictions fix a large fraction of inputs and analyze the residual circuit; here the authors design a restriction that fixes enough gates to reduce the effective depth to a constant, yet leaves enough randomness to argue about the remaining structure. Crucially, the restriction can be described by a log‑time algorithm, ensuring that the uniformity condition is not violated.

The second phase shows that the restricted circuit cannot capture the global combinatorial structure of the permanent. The permanent of an n×n matrix is the sum over all n! perfect matchings, each term being a product of n entries. Even after fixing many inputs, any surviving subcircuit must still be able to compute a polynomial that encodes a large number of monomials with non‑trivial coefficients. By analyzing the degree‑d polynomial representation of the permanent and applying known approximation limits for threshold gates (e.g., Razborov’s lower bounds on the sign‑representation of the majority function), the authors prove that a circuit of the given depth and size cannot approximate the permanent’s coefficients within the required accuracy. Consequently, the circuit would need to be of size 2^{Ω(n^{c})}, contradicting the polynomial‑size assumption.

For arithmetic circuits, the argument is adapted by focusing on the multiplicative depth needed to realize the nested products inherent in the permanent’s definition. With depth limited to o(log log n), any arithmetic circuit can only compose a sub‑exponential number of multiplication layers, which is insufficient to generate the full set of n‑fold products required. A counting argument shows that the number of distinct monomials that can be produced is far smaller than n!, leading again to a super‑polynomial size requirement.

The paper discusses several implications. First, it provides a new barrier for uniform circuit complexity of #P‑complete problems, indicating that even modest depth extensions do not collapse the permanent into polynomial‑size uniform models. Second, the random‑restriction method that respects DLOGTIME‑uniformity may be applicable to other circuit classes, such as ACC⁰ or MOD p circuits, potentially yielding analogous lower bounds. Third, the result narrows the gap between known upper bounds (e.g., quasipolynomial‑size circuits for the permanent) and lower bounds, suggesting that the true complexity lies somewhere between polynomial and quasipolynomial for uniform models.

Finally, the authors outline open directions: determining the exact threshold depth at which polynomial‑size uniform circuits become possible, extending the technique to non‑uniform models, and exploring whether similar bounds hold for other #P‑complete functions like the determinant over finite fields. The work thus advances our understanding of the permanent’s inherent difficulty and enriches the toolkit for proving uniform circuit lower bounds.