Quantum Search on Computation Trees

TL;DR

This paper generalizes quantum walk algorithms for search in backtracking trees with variable vertex computation times, leading to optimal query complexity algorithms for variable time search and improved geometric intersection detection.

Contribution

It introduces a simple, explicit quantum algorithm framework that handles variable vertex computation times, resolving open questions and improving existing algorithms.

Findings

Achieves $ ext{O}( ext{sqrt}(TD))$ query complexity for detection in variable-time trees.

Provides an optimal $ ext{O}( ext{sqrt}(T ext{log} ext{min}(n,t_{max})))$ algorithm for variable time search.

Improves geometric intersection detection to $ ilde{O}(n)$ time.

Abstract

We show a simple generalization of the quantum walk algorithm for search in backtracking trees by Montanaro (ToC 2018) to the case where vertices can have different times of computation. If a vertex $v$ in the tree of depth $D$ is computed in $t_v$ steps from its parent, then we show that detection of a marked vertex requires $\text{O}(\sqrt{TD})$ queries to the steps of the computing procedures, where $T = \sum_v t_v^2$. This framework provides an easy and convenient way to re-obtain a number of other quantum frameworks like variable time search, quantum divide & conquer and bomb query algorithms. The underlying algorithm is simple, explicitly constructed, and has low poly-logarithmic factors in the complexity. As a corollary, this gives a quantum algorithm for variable time search with unknown times with optimal query complexity $\text{O}(\sqrt{T \log \min(n,t_{\max})})$, where $T = \sum_i t_i^2$ and $t_{\max} = \max_i t_i$ if $t_i$ is the number of steps required to compute the $i$-th variable. This resolves the open question of the query complexity of variable time search, as the matching lower bound was recently shown by Ambainis, Kokainis and Vihrovs (TQC'23). As another result, we obtain an $\widetilde{\text{O}}(n)$ time algorithm for the geometric task of determining if any three lines among $n$ given intersect at the same point, improving the $\text{O}(n^{1+\text{o}(1)})$ algorithm of Ambainis and Larka (TQC'20).