On the growth of variances in a central limit theorem for strongly mixing sequences (Q1283384)

Let \(X=(X_k,k\in Z)\) be a strictly stationary sequence of random variables on the probability space \((\Omega,F,P)\) and a \(\sigma\)-field of events generated by a given family of random variables on the same probability space. For each positive integer \(n\), there are defined the mixing coefficients \(\alpha (n)\), \(\rho(n)\), \(\beta(n)\), and \(\rho^*(n)\). The random sequence \(X\) is said to be strongly mixing if \(\alpha(n)\to 0\) and \(n\to\infty\). Similar definitions are obtained when the other coefficients converge to zero; \(X\) is \(\rho\)-mixing \((\rho(n)\to 0)\), \(X\) is absolutely regular \((\beta(n)\to 0)\), and \(X\) is \(\rho^*\)-mixing \((\rho^*(n)\to 0)\). The main result of the paper proves that, if a stationary Gaussian sequence \(X\) is strongly mixing and its spectral density is bounded between two positive constants, the spectral density can still fail to be continuous. As the author points out, in a central limit theorem under certain strong mixing conditions, ``one does not quite have an asymptotic linear growth of the variance of the partial sums''.