Physics Informed Models and Optimization Schemes for Multi-Parameter Non-convex Optimization Problems

In superconducting quantum chips, the presence of noise and crosstalk makes it challenging to achieve perfect quantum circuits. However, through chip calibration, we can enhance the fidelity of quantum circuits in existing quantum chips. Chip calibration itself constitutes a multi-parameter non-convex optimization problem. In this work, we successfully established an error model using neural networks to predict the true objective function based on experimental results from actual chips. The BCD algorithm was employed to determine the convergence conditions for the optimization algorithm during the optimization process. Numerical simulation results demonstrate that the constructed simulator and optimization algorithm exhibit excellent robustness.