The paper studies a basic question in number theory: which whole numbers can be written as a prime number plus a power of two (like 2, 4, 8, 16, and so on)? Numbers that can be written this way are called Romanoff numbers, named after the mathematician who proved in the 1930s that such numbers make up a positive fraction of all whole numbers. So while Romanoff numbers are fairly common, they do not cover everything, and the paper investigates just how large the gaps between them can get.
Specifically, the authors want to know how long a stretch of consecutive integers can go without containing a single Romanoff number, among all such stretches up to some large bound X. They prove that the longest such gap grows at least as fast as the iterated logarithm of X (roughly, the logarithm of the logarithm of X). This improves on the previous best result by Kalmynin and Konyagin, which had an extra slowly growing factor in the denominator making it slightly weaker. The improvement is modest in absolute terms, but closing even this small gap requires a genuinely new technical idea.
The key innovation is a more efficient way of handling many integer "shifts" simultaneously when doing a type of number-theoretic sieving argument. Previous approaches ran into trouble when trying to cover many candidate positions at once, but the authors show that troublesome prime factors can be removed at essentially no cost, allowing all the shifts to be dealt with together. The result holds unconditionally, meaning it requires no unproven assumptions like the Riemann Hypothesis, and it extends naturally to numbers of the form prime plus a power of any fixed integer greater than one, not just powers of two.