31Our ansatz is justified by the analytic calculations in  [70], which compute the corrections to the logL  behavior. These corrections are shown to be given by integer powers of 1∕L  : c = alogL + ∑ ∞ ck∕Lk
             k=0  .