Komlós conjecture (constant ℓ∞ discrepancy for unit-norm columns)

Establish the Komlós conjecture: Determine a universal constant C such that for every m×n real matrix M whose columns have ℓ2-norm at most 1, there exist signs x in {−1,1}^n with ||Mx||_∞ ≤ C, independent of m and n.

Background

The Komlós conjecture is a central question in discrepancy theory and generalizes the Beck–Fiala setting. It asks for a dimension- and size-independent constant bound on the ℓ∞-discrepancy when columns have unit ℓ2-norm.

Non-constructively, Banaszczyk established an O(√log n) bound; more recently, Bansal and Jiang improved the bound to O(log{1/4} n). This paper contributes a related local mean squared discrepancy guarantee for row subsets, but the conjectured absolute-constant ℓ∞ bound remains open.

References

The central open problems in this area are the Beck-Fiala conjecture and more generally the Komlos conjecture, where given a set of $n$ vectors in the unit ball in $m$, the goal is to assign signs (also called ``colors") to the vectors so as to bound the $\ell_{\infty}$ norm of the final signed sum, by a constant independent of $m$ and $n$.

— Constructive l2-Discrepancy Minimization with Additive Deviations  (2508.21423 - Dutta, 29 Aug 2025) in Section 1, Introduction, paragraph “Discrepancy Minimisation and The Bansal-Garg Algorithm”

Find a deterministic algorithm for the Komlós problem with $O(mn+n{\omega+2-c})$ operations for a constant $c>0$, or with $o(n2)$ updates. The example above suggests what is needed: a potential that controls the averages of $W$ over every set of active coordinates and is smooth at a scale coarser than $\eta$.

— Fast Spectral Signing for Vector Balancing  (2609.30044 - Li, 24 Sep 2026) in Section 7, Discussion, paragraph “Open problems,” item 1

Find a polynomial-time algorithm with discrepancy at most $36$, the bound of the elementary proof, or even $3\sqrt{2\pi}$; an efficient Banaszczyk-type induction is one route.

— Fast Spectral Signing for Vector Balancing  (2609.30044 - Li, 24 Sep 2026) in Section 7, Discussion, paragraph “Open problems,” item 2