The paper addresses a classical problem in mathematics called **discrepancy theory**, which studies how evenly a finite collection of points can be distributed inside a region. If you place n points in a unit square and then draw all possible axis-aligned rectangles anchored at the corner, you can count how many points fall inside each rectangle and compare that to the fraction of the square's area the rectangle occupies, scaled by n. The difference between the actual count and the expected count is called the discrepancy. A perfectly uniform distribution would have zero discrepancy everywhere, but it has been known for decades that perfect uniformity is impossible, and the discrepancy must be large somewhere.
The main result of the paper is a precise and sharp statement about just how large the discrepancy must be, and how often it must be that large. The authors prove that for any arrangement of n points, the discrepancy function must have absolute value at least a fixed constant times the square root of the logarithm of n, not just at one special location, but across a region of the square that has a fixed positive area, independent of n. In other words, large discrepancy is not merely a rare spike that could be hidden in a tiny corner of the square; it must persist over a genuinely substantial portion of the domain.
This result is called "sharp" because it matches known upper bounds up to constants, meaning nothing fundamentally better can be proven. A key consequence is that standard measures of the overall size of the discrepancy function, called Lp norms for any positive p, are all bounded below by the same square-root-of-log-n quantity. Previous results often established lower bounds only for specific norms or specific locations, so this distributional lower bound, which controls how spread out the large discrepancy values are, is a meaningful strengthening of what was previously known.