 Chain rule for relative entropy

For probability distributions $P$ and $Q$ on a product space, the Kullback-Leibler divergence decomposes into the divergence of a marginal and the expected divergence of the corresponding conditional distributions. Iteration gives