Adversary lower bounds for nonadaptive quantum algorithms

We present general methods for proving lower bounds on the query complexity of nonadaptive quantum algorithms. Our results are based on the adversary method of Ambainis.

💡 Research Summary

The paper “Adversary lower bounds for nonadaptive quantum algorithms” develops general techniques for proving lower bounds on the query complexity of quantum algorithms that are nonadaptive, i.e., algorithms that issue all their queries simultaneously rather than sequentially based on previous answers. The authors build upon the weighted adversary method introduced by Ambainis and extend it to the nonadaptive setting, providing two complementary frameworks: a direct “super‑query” method and a primal‑dual (minimax) method.

First, the authors formalize the black‑box model for quantum computation. An input is a function x from a finite domain Γ to a finite alphabet Σ. An adaptive quantum algorithm is described by a sequence of unitary operators interleaved with oracle calls O_x, while a nonadaptive algorithm is described by a pair of unitaries (U,V) that apply a single “batch” oracle O_T^x which simultaneously queries T positions i_1,…,i_T. This formulation makes it possible to treat the whole batch of T queries as a single query to a new “super‑input” k x defined on the product domain Γ^k.

Using this viewpoint, the authors apply Ambainis’s weighted adversary theorem (Theorem 1) to the super‑query. They define a weight function w on pairs of inputs (x,y) that are distinguished by the function F, and derive a lower bound (Theorem 3) of the form

 Q_na,ε(F) ≥ C_{2ε} · L_na(F),

where L_na(F) = max_w max_{s∈S′} min_{x: F(x)=s, i} wt(x)/v(x,i). Here wt(x) is the total weight incident to x, and v(x,i) is the total weight of pairs (x,y) that differ on coordinate i. This bound is essentially the same as Ambainis’s bound for adaptive algorithms, but the factor C_{2ε} reflects the error tolerance and the fact that the algorithm is nonadaptive.

The second framework is a primal‑dual (minimax) approach. For each input x the authors consider the average probability p_x(i) that a nonadaptive algorithm queries index i. Theorem 5 (citing Laplante and Magniez) shows that the query complexity is at least

 C_ε · max_{x≠y, F(x)≠F(y)} 1 / ∑_{i: x_i≠y_i} p_x(i) p_y(i).

They then prove (Theorem 6) that this minimax quantity DL(F) coincides with the maximum over all valid weight functions of the expression used in the weighted adversary bound, i.e., DL(F)=PL(F). Consequently, the primal‑dual bound is never weaker (up to constant factors) than the direct weighted‑adversary bound.

Armed with these general tools, the authors derive tight Ω(N) lower bounds for several canonical problems:

• Unordered Search: The input is a Boolean string of length N, and the goal is to compute the OR. By choosing X as all‑zero strings and Y as strings with a single 1, and defining a relation R that pairs each x∈X with each y∈Y differing in exactly one position, they obtain m=N, ℓ=ℓ′=m′=1, yielding Q_na,ε ≥ Ω(N).

• Element Distinctness: X consists of all injective functions f: