What it is
A quantum channel describes how a quantum system evolves, including noise, and 'process tomography' is the task of learning an unknown channel by probing it, a basic step in calibrating and debugging quantum hardware. How many uses of the channel this takes, measured by the diamond distance (the standard worst-case way to tell two quantum processes apart), had been an open question. The authors settle it: a channel with input dimension d_in, output dimension d_out, and Kraus rank k can be learned to any fixed accuracy using a number of channel uses that scales with d_in x d_out x k, the number of parameters the channel actually has, and they prove a matching lower bound showing no method can do better. For a general channel on a d-dimensional system this is d^4 uses, down from the d^6 of the best previous schemes.
Why it matters
Process tomography underpins the calibration and validation of every quantum computer, so knowing the true cost of doing it well sets a hard floor on that overhead. The result closes a gap that had stood for the general-channel case even as the analogous question for quantum states was solved a decade ago. Because channels include quantum states, unitary gates, and measurements as special cases, the authors' protocol is a single framework that covers all of them, rather than a one-off method for channels alone.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed underQuantum Information and Cryptography, Quantum Computing Algorithms and Architecture, Quantum Mechanics and Applications
