Basis-Free Study of Confined Atoms and Molecules Based on a Neural Network Approach

One of the main problems in the analysis of cavity systems is the necessity to construct special basis sets that meet boundary conditions. The task often involves variational optimization of parameters for each molecular system, each type of potential, and cavity size, which makes the procedure very time-consuming. Here, we present a solution to this problem by using the variational quantum Monte Carlo method based on a neural networks approach, which needs no basis set and, hence, can be used "out-of-the-box" for a broad range of systems and potentials. This approach is particularly efficient for the calculation of systems in cavities of variable sizes, which is important in the study of high-pressure effects. Even for small systems, it provides results comparable to highly accurate variational estimates. For many-electron systems, it often outperforms methods based on Gaussian basis sets.