Black-Box Optimization Techniques with Adaptive Learning for Multi-Qubit Gate Optimization

Abstract

This investigation details the formulation and implementation of a black-box descent optimization algorithm with adaptive scaling and momentum (AdamSPSA) to optimize the generation of a high-fidelity Cross Resonance (CR) gate on two indirectly coupled transmon superconducting qubits in an open quantum system, via the pulse-shaping of control microwave fields for single-qubit control amidst ongoing environmental interactions. Analysis of the computational simulation is presented in the form of quantified gate infidelity and quantum process tomography. The study finds the model successful in reducing initial gate infidelity for both Gaussian and cubic spline paramaterizations by over 30%. However, final objectives remained insufficient relative to computation-caliber gates, implying a stronger model or additional human aid in situ is still required for suitable outcomes in multi-qubit quantum computation.