https://arxiv.org/abs/2005.12702 (or view on SciRate)
We introduce maximum likelihood fragment tomography (MLFT) as an improved circuit cutting technique for running clustered quantum circuits on quantum devices with a limited number of qubits. In addition to minimizing the classical computing overhead of circuit cutting methods, MLFT finds the most likely probability distribution for the output of a quantum circuit, given the measurement data obtained from the circuit's fragments. We demonstrate the benefits of MLFT for accurately estimating the output of a fragmented quantum circuit with numerical experiments on random unitary circuits. Finally, we show that circuit cutting can estimate the output of a clustered circuit with higher fidelity than full circuit execution, thereby motivating the use of circuit cutting as a standard tool for running clustered circuits on quantum hardware.
This paper extends Maximum Likelihood State Tomography to find maximum likelihood distributions from fragments of cut circuits (called maximum likelihood fragment tomography or MLFT in the paper). These cut circuits allow for running larger (more qubits) circuits by cutting into subcircuits, running each subcirctuit individually, and combining the results with classical algorithms. This general technique was already known, but this new work allows for skipping some naive steps in the older technique like removing negative probability amplitudes and re-normalizing the outcome by virtue of the MLFT algorithm.
Runtime wasn’t measured as it is known to be unfavorable (exponential in both the number of cuts made on the circuit and number of qubits involved) but the metric looked at in numerical experiments was fidelity/infidelity. In general, MLFT had better fidelity than full or direct simulation for larger number of qubits. In the plot it is a bit hard to tell if there is any change in S (the number of samples) when qubit # (Q) is held constant leads to a realm in which MLFT is better/worse because the qubit # is held at Q=18 where, as seen in the lower plots, MLFT tends to perform better.