Core Concepts
This work generalizes the classical binary search problem to a d-dimensional domain, where the target element must be found using a minimum number of queries.
Abstract
The authors study a generalization of the classical binary search problem to a d-dimensional domain S1 × · · · × Sd, where Si = {0, 1, ..., ni - 1} and d ≥ 1. Given a target element (t1, ..., td), the result of a comparison of a selected element (x1, ..., xd) is the sequence of inequalities stating whether ti < xi or ti > xi for each i ∈ {1, ..., d}, where at least one inequality is correct but the algorithm does not know the specific dimension i on which the correct direction to the target is given.
The authors provide a series of upper and lower bounds on the query complexity Q(n1, ..., nd) required to find the target:
For the 2-dimensional case, they show that Q(m, n) ≤ 2n log2(m/n + 1) + 4, which is asymptotically tight.
For the 3-dimensional case, they prove a lower bound of Ω(n2n3 log2(n1/n2 + 1)) and an almost matching upper bound of O(n2n3 log2(n1/n2 + 1)).
For the general d-dimensional case, they show a lower bound of Ω(nd-1/d) and an upper bound of O(nd), leaving an open question about the exact query complexity Q(n×d) for an arbitrary dimension d.
The problem is shown to be equivalent to the classical binary search in the 1-dimensional case and exhibits interesting differences in higher dimensions, such as the fact that if each of the d inequalities must be correct, the search can be completed in log2 max{n1, ..., nd} queries.
Stats
The number of queries needed to search one line segment in the 2-dimensional diagonal is at least 1 + ⌊log2 ⌊m/n⌋⌋.
The number of queries needed to search the 3-dimensional plane H is at least 1/2 (n2 - 1) n3 log2((n1 - 1)/(n2 - 1) + 1).
The number of queries needed to search the d-dimensional hyperplane H is at least 2⌊nd-1/(d-1)⌋.
Quotes
"One way to define the problem is to see it as an adaptive query-reply search game between two players called Algorithm and Adversary."
"Besides a formal statement of our problem, we also give some remarks regarding alternative formulations."
"We note that the majority of works focus solely on the query complexity. However, from the point of view of potential applications, it may be of interest to have an algorithm whose computational complexity per query is low."