Optimal interpolation-based coordinate descent method for parameterized quantum circuits (2503.04620v1)

Abstract: Parameterized quantum circuits appear ubiquitously in the design of many quantum algorithms, such as variational quantum algorithms, where the optimization of parameters is crucial for algorithmic efficiency. In this work, we propose an Optimal Interpolation-based Coordinate Descent (OICD) method to solve the parameter optimization problem that arises in parameterized quantum circuits. Our OICD method employs an interpolation technique to approximate the cost function of a parameterized quantum circuit, effectively recovering its trigonometric characteristics, then performs an argmin update on a single parameter per iteration on a classical computer. We determine the optimal interpolation nodes in our OICD method to mitigate the impact of statistical errors from quantum measurements. Additionally, for the case of equidistant frequencies -- commonly encountered when the Hermitian generators are Pauli operators -- we show that the optimal interpolation nodes are equidistant nodes, and our OICD method can simultaneously minimize the mean squared error, the condition number of the interpolation matrix, and the average variance of derivatives of the cost function. We perform numerical simulations of our OICD method using Qiskit Aer and test its performance on the maxcut problem, the transverse field Ising model, and the XXZ model. Numerical results imply that our OICD method is more efficient than the commonly used stochastic gradient descent method and the existing random coordinate descent method.