An ergodic process of zero divergence-distance from the class of all stationary processes (Q1383180)

Let \(A\) be a finite set and \(a_1^n=\{a_1,\dots, a_n\}\) a sequence of elements of \(A\). A process \(P\) is a shift-invariant Borel probability measure on the space of all doubly infinite sequences \(x\) of elements of \(A\). Define \(P(a_1^n)=P(x:x_1^n=a_1^n)\). Further define \[ D_n(P\parallel Q)=n^{-1}\sum_{a_1^n} P(a_1^n)\log [P(a_1^n)/Q(a_1 ^n)]. \] If there exists a limit \(D(P\parallel Q)=\lim D_n(P\parallel Q)\), then it is called the divergence rate of \(P\) with respect to \(Q\) or the divergence-distance of \(Q\) from \(P\). It is known that there exists an ergodic process \(Q\) such that for the coin-tossing process \(P\) the relation \(D(P\parallel Q)=0\) holds although \(Q\neq P\). The author generalizes this result and proves that there exists an ergodic process \(Q\) such that \(D(P\parallel Q)=0\) for every stationary process \(P\). The construction of such a process \(Q\) is based on the cutting and stacking method.