Refining quantum computing models by removing gauge freedom

Techniques for quantum device characterization are used to probe the behavior of quantum computational operations, and construct predictive models of them. But these models grow exponentially in complexity with the number of qubits in the system being probed. Reconstructing a fully general model may require an overwhelming, infeasible amount of data. One way to avoid this catastrophe is to identify “reduced” statistical models that have relatively few parameters, yet still accurately describe the observed data. Reduced-model techniques will be essential for characterizing many qubit systems. Finding reduced models by numerically exploring the space of models is made much harder by a pernicious gauge freedom in the “gate set” model of qubit systems. This gauge freedom occurs because infinitely many distinct yet equivalent models predict the same physical outcomes. I have been developing a new parameterization for gate set models, referred to as first-order gaugeinvariant (FOGI), which eliminates most of the barriers to finding reduced models. I will describe

the FOGI framework, and then I will introduce an algorithm that I am developing for automated model selection (AMS), which constructs reduced models by finding and removing parameters that are not necessary to capture the noise affecting a quantum device. Finally, I will demonstrate AMS with FOGI models in practice by applying it on simulated data and experimental data from neutral atom qubits operated at Sandia. In experiment, AMS finds a model of the atomic qubit’s dynamics that has only half as many parameters as the standard fully general model, but loses virtually no accuracy. SNL is managed and operated by NTESS under DOE NNSA contract DENA0003525.