The paper studies a classical question in probability theory: when does a Central Limit Theorem (CLT) hold for a Markov chain? A Markov chain is a random process where the future depends only on the present state. The CLT, loosely speaking, says that if you add up many independent random quantities and rescale, you get a bell-shaped normal distribution. For Markov chains the variables are not independent, so getting a CLT requires extra conditions on how the chain behaves over time. The "annealed" CLT considered here averages over random starting points drawn from the chain's long-run equilibrium distribution.
The main technical setting involves operators on function spaces. Associated with any Markov chain is a transition operator that tells you the expected value of a function one step later. When this operator is "normal" (a condition meaning it commutes with its adjoint, which includes the common symmetric case), the authors show something sharp and somewhat surprising: for the CLT to hold simultaneously for every square-integrable function with mean zero, the chain must be "uniformly ergodic," meaning it converges to its equilibrium at a controlled, uniform rate across all starting points. This is a necessity result, pinning down exactly how strong a condition you need to get a universal CLT.
The paper also explores what happens when the normality assumption is relaxed. Under so-called Ritt-type conditions, which control how the spectrum of the operator behaves near certain boundary points, the authors still obtain meaningful CLT results. The work contributes to a long line of research connecting the spectral properties of Markov operators to probabilistic limit theorems, and the results are proved in full generality without restricting to finite or countable state spaces.