Optimizing transmon readout with dynamiqs, a library for GPU-accelerated and differentiable quantum simulations (1/2)

Superconducting qubit readout stands out as a main bottleneck of circuit quantum electrodynamics, with the fidelity and length of the operation typically lagging in comparison to state-of-the-art single and two qubit gates. In this two-part talk, we introduce a constant memory numerical method to compute arbitrary gradients of a master equation, that scales favorably with system size and evolution time. This approach enables the realization of quantum optimal control on arbitrary dissipative quantum systems with a large Hilbert space. Here, we demonstrate it on the optimization of a dispersive transmon readout that includes both a readout resonator and a Purcell filter. We obtain robust and interpretable pulses and show a reduction of readout times by up to a factor 2 compared to standard readout for our realistic system parameters. This optimization is achieved using dynamiqs (https://www.dynamiqs.org/), an open-source Python library powered by PyTorch. The library offers GPU-accelerated and differentiable solvers for closed and open quantum systems. It is designed for large-scale quantum simulations, optimal control, and parameter estimation, thus enabling straightforward replication of this work.

In the first part of this talk, we focus on the constant-memory differentiation method and the optimization of transmon readout.