Jacobian hits circuits: Hitting-sets, lower bounds for depth-D occur-k formulas & depth-3 transcendence degree-k circuits

We present a single, common tool to strictly subsume all known cases of polynomial time blackbox polynomial identity testing (PIT) that have been hitherto solved using diverse tools and techniques. In particular, we show that polynomial time hitting-set generators for identity testing of the two seemingly different and well studied models - depth-3 circuits with bounded top fanin, and constant-depth constant-read multilinear formulas - can be constructed using one common algebraic-geometry theme: Jacobian captures algebraic independence. By exploiting the Jacobian, we design the first efficient hitting-set generators for broad generalizations of the above-mentioned models, namely: (1) depth-3 (Sigma-Pi-Sigma) circuits with constant transcendence degree of the polynomials computed by the product gates (no bounded top fanin restriction), and (2) constant-depth constant-occur formulas (no multilinear restriction). Constant-occur of a variable, as we define it, is a much more general concept than constant-read. Also, earlier work on the latter model assumed that the formula is multilinear. Thus, our work goes further beyond the results obtained by Saxena & Seshadhri (STOC 2011), Saraf & Volkovich (STOC 2011), Anderson et al. (CCC 2011), Beecken et al. (ICALP 2011) and Grenet et al. (FSTTCS 2011), and brings them under one unifying technique. In addition, using the same Jacobian based approach, we prove exponential lower bounds for the immanant (which includes permanent and determinant) on the same depth-3 and depth-4 models for which we give efficient PIT algorithms. Our results reinforce the intimate connection between identity testing and lower bounds by exhibiting a concrete mathematical tool - the Jacobian - that is equally effective in solving both the problems on certain interesting and previously well-investigated (but not well understood) models of computation.

💡 Research Summary

This paper introduces a unified algebraic‑geometric framework based on the Jacobian matrix to simultaneously address two historically separate lines of research in arithmetic circuit complexity: black‑box polynomial identity testing (PIT) and lower‑bound proofs. The authors observe that the Jacobian captures algebraic independence of a set of polynomials: if the Jacobian of a subset of size r is non‑zero, those r polynomials form a transcendence basis. By constructing a linear map ϕ that preserves the non‑zeroness of the Jacobian, they show that evaluating a circuit C(T₁,…,Tₘ) on the images ϕ(Tᵢ) preserves zero‑ness, thereby reducing PIT for circuits whose input polynomials have bounded transcendence degree to PIT on a constant‑size instance.

Using this insight, the paper yields the first polynomial‑time hitting‑set generators for two broad families that previously required distinct techniques: (1) depth‑3 ΣΠΣ circuits where the product gates compute linear polynomials whose transcendence degree is a constant r (no top‑fan‑in restriction), and (2) constant‑depth formulas where each variable appears in at most k leaf polynomials—a model they call “occur‑k” formulas, which generalizes the earlier “read‑k” restriction and removes the multilinearity requirement. For a depth‑D occur‑k formula of size s, a hitting set of size polynomial in s and R = (2k)^{2D}·2^{D} can be constructed in polynomial time (over fields of characteristic zero or larger than R). In the special case of depth‑4 occur‑k formulas, the dependence improves to poly(s·k²).

The same Jacobian technique is turned around to prove exponential lower bounds for the immanant family (including determinant and permanent) on exactly the models for which the PIT algorithms are given. The authors show that if a depth‑4 occur‑k formula computes Immₙ, then its size must be at least 2^{Ω(n/k²)}; if a circuit composed of sparse polynomials of sparsity s and transcendence degree r computes Immₙ, then s ≥ 2^{Ω(n/r)}; and if a circuit of the form C(T₁,…,Tₘ) with linear‑polynomial products Tᵢ computes Immₙ, then the transcendence degree of {T₁,…,Tₘ} must be Ω(n). These bounds improve upon earlier results that applied only to more restricted depth‑4 multilinear circuits.

Technically, the paper’s contribution lies in showing that the Jacobian’s partial‑derivative structure “linearizes’’ the product gates, allowing the construction of the map ϕ that both reduces the effective number of variables for PIT and forces strong algebraic constraints for lower‑bound arguments. The authors also provide a concise proof of the “faithful homomorphism” property of such maps, building on prior work but without invoking heavy algebraic geometry (Krull’s Hauptidealsatz).

Overall, the work demonstrates that a single mathematical object—the Jacobian—can serve as a powerful tool for both algorithmic derandomization (via hitting‑set construction) and complexity‑theoretic separation (via lower‑bound proofs), unifying previously disparate techniques and extending them to more general circuit classes such as constant‑occur formulas and circuits with bounded transcendence degree. This represents a significant step toward a deeper understanding of the interplay between PIT and circuit lower bounds.