The mixing advantage is less than 2 (Q626271): Difference between revisions

Let \(X_1,\dots,X_n\) be independent random variables, and let \(X_i^1\),\dots,\(X_i^n\) be i.i.d. copies of \(X_i\). The authors derive upper and lower bounds for \(M=E\max_{1\leq i\leq n} X_i\) in terms of \(M_i=E\max_{1\leq j\leq n} X_i^j\). The main result is that \[ \bar M\leq M\leq \bar M+{n-1\over n}\max_{1\leq i\leq n} M_i, \] where \(\bar M=\sum_{i=1}^n M_i/n\). So, the value of the ``mixing factor'' \(\vartheta_n=M/\max_i M_i\) does not exceed \(2-1/n\), and, in the case where \(M_1=\dots=M_n\), it is at least 1.