Bounds And Low-Rank Approximation For Controlled Markov Processes