A Predictive Algorithm for Ansatz Selection
Ronak Tali
Abstract:
Variational Quantum Eigensolver (VQE) is a hybrid quantum-classical algorithm based on the variational principle, which involves minimizing the total system energy over multiple rounds, by varying the parameters of a quantum circuit, to efficiently arrive at the ground state. The parametrization of such circuits is called an ansatz. However, the correct parametrization, is not known a priori. As a result, an experimenter needs to undertake a trial and error approach that involves trying out several ansatze until a correct ansatz is identified for a given problem. Moreover, each of these VQE algorithm runs with a trial ansatz requires a large number of minimization rounds to find out if or not the ground state has been attained. This brute force approach can be very expensive. We propose a machine learning driven algorithm that achieves a reduction in number of minimization rounds through appropriate feature engineering and off the shelf classifiers trained on smaller sized systems and then using these trained models to make predictions on larger sized systems only by monitoring the signals from the first few rounds of the VQE minimization procedure. Results indicate that our algorithm can make promising predictions while significantly cutting down the time.