PreprintCC BY 4.0arXiv
Three Standard Deviations Suffice While One Does Not
Abstract
Spencer's 1985 ``six standard deviations suffice'' theorem shows that every A ∈ [−1,1]n × n has a sign vector x ∈ {−1,1}n with |Ax|∞ ≤ 6√n. We show the upper bound √(3arsinh(10))√n+4 < 2.9992 √n + 4 by directly rounding the minimizer of a potential function to a vertex of the cube. We also show that, for every power of two n ≥ 250, there exists a matrix A ∈ {−1,1}n × n such that |Ax|∞ >1.0000002√n for every choice of signs x ∈ {−1,1}n. The construction simply replaces a 2−22 fraction of the columns of a Hadamard matrix with independent random sign vectors. This is the first improvement over the √n lower bound of Olson and Spencer (1978), which uses a Hadamard matrix.
Related
- Matrix Spencer: Eight Standard Deviations Suffice and an Almost-Linear Time Algorithm for Dense Input
- Cost-Aware Stopping for Freeze-Thaw Bayesian Optimization: When Does One-Step Lookahead Suffice?
- Three Machines Suffice for Local Hardening of Electromechanical Oscillations
- When Weak Submodularity Meets Matroids: How Does Standard Greedy Behave?
- What Does a Discrete Diffusion Model Learn? One Reverse Process in Three Coordinates
BibTeX
@misc{reis2026three,
title = {{Three Standard Deviations Suffice While One Does Not}},
author = {Victor Reis and Zhao Song},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.34471},
eprint = {2609.34471},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.34471},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Three Standard Deviations Suffice While One Does Not
AU - Victor Reis
AU - Zhao Song
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.34471
UR - https://arxiv.org/abs/2609.34471
AB - Spencer's 1985 ``six standard deviations suffice'' theorem shows that every A ∈ [−1,1]^n × n has a sign vector x ∈ {−1,1}^n with |Ax|_∞ ≤ 6√n. We show the upper bound √(3arsinh(10))√n+4 < 2.9992 √n + 4 by directly rounding the minimizer of a potential function to a vertex of the cube. We also show that, for every power of two n ≥ 2^50, there exists a matrix A ∈ {−1,1}^n × n such that |Ax|_∞ >1.0000002√n for every choice of signs x ∈ {−1,1}^n. The construction simply replaces a 2^−22 fraction of the columns of a Hadamard matrix with independent random sign vectors. This is the first improvement over the √n lower bound of Olson and Spencer (1978), which uses a Hadamard matrix.
ER -