About (18)-(21) in the paper

Hi!
I was wondering why the lower bound always struggles to achieve rates greater than O(log M). I read your paper and was wondering why Q_Y (an empirical distribution for QY, you declare) in equation (20) is expressed as 1/M, with M denoting numbers of samples.
