[Paper Review] A New Quadratic Bound for the Manickam-Miklós-Singhi Conjecture

Ameera Chowdhury, Ghassan Sarkis+1 more|arXiv (Cornell University)|Mar 7, 2014

Limits and Structures in Graph Theory21 references3 citations

TL;DR

This paper proves the Manickam-Miklós-Singhi conjecture for all $ n \geq 8k^2 $, establishing that any set of $ n $ real numbers with nonnegative sum contains at least $ \binom{n-1}{k-1} $ $ k $-element subsets with nonnegative sum. The proof uses spectral graph theory and inclusion matrices, leveraging eigenvalue bounds and averaging arguments to show that the number of nonnegative $ k $-subsets exceeds the conjectured threshold when $ n \geq 8k^2 $, improving prior polynomial and linear bounds.

ABSTRACT

More than twenty-five years ago, Manickam, Miklos, and Singhi conjectured that for positive integers $n,k$ with $n \geq 4k$, every set of $n$ real numbers with nonnegative sum has at least $\binom{n-1}{k-1}$ $k$-element subsets whose sum is also nonnegative. We verify this conjecture when $n \geq 8k^2$, which simultaneously improves and simplifies a bound of Alon, Huang, and Sudakov and also a bound of Pokrovskiy when $k < 10^{45}$.

Motivation & Objective

• To resolve the Manickam-Miklós-Singhi conjecture for a broad range of $ n $ and $ k $, particularly when $ n \geq 8k^2 $.
• To provide a simplified and improved polynomial bound over previous results, including those by Alon, Huang, and Sudakov and by Pokrovskiy.
• To establish that the extremal case—where exactly $ \binom{n-1}{k-1} $ $ k $-subsets have nonnegative sum—occurs only when the family is a star centered at the largest element.
• To unify and streamline techniques from earlier works, particularly those involving Bose-Mesner algebras and eigenvalue analysis of inclusion matrices.

Proposed method

• The authors use the Bose-Mesner matrix framework, analyzing the action of inclusion matrices $ W_{1k}^T $ on the vector $ \vec{x} $ of real numbers.
• They show that $ W_{1k}^T\vec{x} $ is an eigenvector of the Bose-Mesner matrix $ B_j $ with eigenvalue $ -\binom{k-1}{j-1}\binom{n-j-1}{k-1} $, enabling spectral analysis.
• A key step involves bounding the number of nonnegative $ k $-subsets containing the largest element $ x_1 $, using combinatorial averaging and inclusion-exclusion over structured subsets.
• The proof applies Lemma 4.1, which guarantees at least $ \binom{n-2k}{k-1} \geq \left(1 - \frac{(2k-1)(k-1)}{n-2k+1}\right)\binom{n-1}{k-1} $ nonnegative $ k $-subsets disjoint from any $ k $-subset $ T $ with negative sum.
• By assuming a contradiction—fewer than $ \binom{n-1}{k-1} $ nonnegative $ k $-subsets—the authors combine bounds from subsets containing $ x_1 $ and those disjoint from a negative-sum $ T $, leading to a sum exceeding the threshold.
• The argument relies on the fact that for $ n \geq 8k^2 $, the combined count from both disjoint and containing cases exceeds $ \binom{n-1}{k-1} $, contradicting the assumption.

Experimental results

Research questions

• RQ1 Does the Manickam-Miklós-Singhi conjecture hold for all $ n \geq 8k^2 $, and can this bound be proven with a simpler method than prior approaches?
• RQ2 Can spectral techniques involving inclusion matrices and Bose-Mesner algebras be used to derive tight lower bounds on the number of nonnegative $ k $-subsets?
• RQ3 What is the minimal $ n $ such that every $ n $-tuple of real numbers with nonnegative sum contains at least $ \binom{n-1}{k-1} $ $ k $-subsets with nonnegative sum?
• RQ4 Under what conditions is the extremal case—exactly $ \binom{n-1}{k-1} $ nonnegative $ k $-subsets—achieved, and is it unique?

Key findings

• The Manickam-Miklós-Singhi conjecture is verified for all $ n \geq 8k^2 $, providing a quadratic bound that improves upon the prior cubic bound $ \min\{33k^2, 2k^3\} $ from Alon, Huang, and Sudakov.
• The bound $ n \geq 8k^2 $ simultaneously improves and simplifies the earlier linear bound $ n \geq 10^{46}k $ of Pokrovskiy when $ k < 10^{45} $.
• When $ n \geq 8k^2 $, the number of $ k $-element subsets with nonnegative sum is at least $ \binom{n-1}{k-1} $, and equality holds only when the family is a star centered at the largest element.
• The proof shows that if any $ k $-subset containing $ x_1 $ has negative sum, then the total number of nonnegative $ k $-subsets exceeds $ \binom{n-1}{k-1} $, contradicting the assumption of fewer than $ \binom{n-1}{k-1} $ such subsets.
• The method relies on eigenvalue analysis of inclusion matrices and combinatorial averaging, yielding a clean contradiction when the conjectured threshold is assumed to be violated.
• The result confirms that the extremal configuration is unique and corresponds to the star family $ \{ S \in \binom{X}{k} : x_1 \in S \} $, achieved when one element is $ n-1 $ and the rest are $ -1 $.