Information-Based Complexity vs Computational Complexity in Phaseless Polynomial Interpolation

TL;DR

This paper investigates the complexity of phaseless polynomial interpolation, showing that the number of points needed for polynomial-time reconstruction varies from polynomial to NP-complete depending on the number of points used.

Contribution

It provides a detailed complexity classification of the reconstruction problem based on the number of evaluation points, resolving open questions in the field.

Findings

Reconstruction from 2n-k points is polynomial time for any constant k.

Reconstruction from (1+c)n+2 points is NP-Complete for any constant c in [0,1).

Polynomial-time selection of evaluation points with a unique solution is possible.

Abstract

The authors of ``A note on the complexity of a phaseless polynomial interpolation'' have shown that phaseless polynomial interpolation over $\mathbf{Q}$ is possible with $n+2$ points, where $n$ is the upper-bound on the degree of a polynomial. Nonetheless, their reconstruction algorithm and the method of adaptively choosing evaluation points are exponential time. On the other hand, they have also shown that given $2n+1$ points, the polynomial can be reconstructed in a polynomial time. A conjecture have been put forward, namely that the reconstruction problem from such $n+2$ points is exponential time. Moreover, a question about the number of points sufficient for polynomial time reconstruction have been posed. In this paper, we answer these questions -- we show that (1) reconstruction problem from $2n-k$ for any constant $k$ is polynomial time, (2) reconstruction problem from $(1+c)n+2$ points for any constant $c \in [0, 1)$ is NP-Complete, (3) evaluation points admitting a unique solution can be chosen in polynomial time.