Berlin researchers give optimal query count for learning quantum channels
Antonio Anna Mele and Lennart Bittel of Freie Universität Berlin report that any quantum channel with input dimension d_in, output dimension d_out and Kraus rank at most k can be learned to accuracy ε in diamond distance using O(d_in d_out k/ε^2) channel uses. They also prove that Ω(d_in d_out k) channel uses are necessary at constant accuracy, so the scaling is optimal in the channel dimensions and Kraus rank. The method uses the channel non-adaptively to prepare copies of its Choi state, purifies them in parallel, and applies existing optimal pure-state tomography schemes. The article says the best earlier general guarantee scaled as O(d^6/ε^2) for a d-dimensional channel, and that the work gives the first explicit scheme for arbitrary isometric channels with query complexity O(d_in d_out/ε^2).