The paper resolves a long-standing question about how the randomness of a sum of independent random variables grows as you add more terms. Imagine rolling the same unusual die repeatedly and keeping a running total. After each new roll, the total becomes harder to predict. Shannon entropy is the standard mathematical measure of that unpredictability. The question, going back to Claude Shannon himself, is whether the entropy of the running total always increases in a orderly, monotone way as you add more and more independent copies of the same random variable.
A conjecture made by Terence Tao in 2010 predicted a precise lower bound on how much entropy must increase at each step. Specifically, it said that the gain in entropy when going from a sum of n terms to a sum of n+1 terms should be at least roughly one-half times the logarithm of (n+1)/n, at least when the underlying random variable has very high entropy to begin with. This bound mirrors what happens in the continuous setting with Gaussian (bell-curve) distributions, where everything is well understood, and the conjecture essentially says the discrete case cannot be too far behind.
The authors prove this conjecture in full generality, covering random variables taking values in a broad class of mathematical structures called torsion-free abelian groups, which includes ordinary integers and real numbers. The result holds uniformly, meaning it does not depend on the specific shape of the probability distribution or the particular group being used, as long as the entropy of the base random variable is large. This closes a problem that had been open for decades and confirms that entropy growth in repeated summation obeys the same fundamental constraints in the discrete world as in the continuous one.