Planted Random Number Partitioning Problem

TL;DR

This paper introduces a planted version of the number partitioning problem, analyzes its statistical and geometric properties, and demonstrates how the multi Overlap Gap Property (m-OGP) can be used to show the limitations of stable algorithms in finding near-optimal solutions.

Contribution

It is the first to establish and leverage the m-OGP framework for a planted model, revealing fundamental algorithmic limitations in the planted number partitioning problem.

Findings

Planting does not significantly lower the minimal objective value compared to the unplanted case.

Complete characterization of the minimal objective value at fixed distances from the planted solution.

Demonstration that stable algorithms fail to find solutions with small objective values due to the m-OGP.

Abstract

We consider the random number partitioning problem (\texttt{NPP}): given a list $X\sim \mathcal{N}(0,I_n)$ of numbers, find a partition $\sigma\in\{-1,1\}^n$ with a small objective value $H(\sigma)=\frac{1}{\sqrt{n}}\left|\langle \sigma,X\rangle\right|$. The \texttt{NPP} is widely studied in computer science; it is also closely related to the design of randomized controlled trials. In this paper, we propose a planted version of the \texttt{NPP}: fix a $\sigma^*$ and generate $X\sim \mathcal{N}(0,I_n)$ conditional on $H(\sigma^*)\le 3^{-n}$. The \texttt{NPP} and its planted counterpart are statistically distinguishable as the smallest objective value under the former is $\Theta(\sqrt{n}2^{-n})$ w.h.p. Our first focus is on the values of $H(\sigma)$. We show that, perhaps surprisingly, planting does not induce partitions with an objective value substantially smaller than $2^{-n}$: $\min_{\sigma \ne \pm \sigma^*}H(\sigma) = \widetilde{\Theta}(2^{-n})$ w.h.p. Furthermore, we completely characterize the smallest $H(\sigma)$ achieved at any fixed distance from $\sigma^*$. Our second focus is on the algorithmic problem of efficiently finding a partition $\sigma$, not necessarily equal to $\pm\sigma^*$, with a small $H(\sigma)$. We show that planted \texttt{NPP} exhibits an intricate geometrical property known as the multi Overlap Gap Property ($m$-OGP) for values $2^{-\Theta(n)}$. We then leverage the $m$-OGP to show that stable algorithms satisfying a certain anti-concentration property fail to find a $\sigma$ with $H(\sigma)=2^{-\Theta(n)}$. Our results are the first instance of the $m$-OGP being established and leveraged to rule out stable algorithms for a planted model. More importantly, they show that the $m$-OGP framework can also apply to planted models, if the algorithmic goal is to return a solution with a small objective value.