The central limit theorem (CLT) is a fundamental result in probability theory saying that when you add up many independent random variables, the distribution of the sum approaches a bell curve (normal distribution). A natural follow-up question is: how fast does this convergence happen, and can we measure it precisely? This paper tackles that question for situations where the random variables are not fully independent but only "weakly dependent," meaning their influence on each other fades over time. The authors use a mathematical technique called Stein's method to derive upper bounds on how far the distribution of a sum is from the normal distribution, where distance is measured using a tool called the second order Wasserstein metric, which captures differences between distributions in a geometrically meaningful way.
The main achievement is proving that the rate of convergence is on the order of 1 divided by the square root of N, where N is the number of terms being added. This is the best possible rate, matching what is known for sums of fully independent random variables. Crucially, the authors achieve this optimal rate under a condition involving only fourth moments (a measure of how heavy the tails of the distribution are), which is the weakest condition known to give optimal rates even in the independent case. They apply their results to several concrete settings including chaotic dynamical systems, which are deterministic systems that behave unpredictably, and to sequences satisfying various technical mixing conditions that quantify how quickly dependence between observations decays.
A particularly notable aspect is that these are the first optimal CLT convergence rates of this kind for weakly dependent sequences of unbounded random variables under general mixing conditions. The authors also handle non-stationary sequences, where the statistical properties of the variables can change over time, and obtain optimal rates without requiring any assumptions about how fast the variance grows. Additional results cover more exotic settings like nonconventional sums, where the terms being added involve the sequence evaluated at multiple time points simultaneously, and random dynamical systems where the rules governing the system change randomly at each step.