Stochastic lists: Sampling multi-variable functions with population methods

Abstract

We introduce the method of stochastic lists to deal with a multi-variable positive function, defined by a self-consistent equation, typical for certain problems in physics and mathematics. In this approach, the function's properties are represented statistically by lists containing a large collection of sets of coordinates (or "walkers") that are distributed according to the function's value. The coordinates are generated stochastically by the Metropolis algorithm and may replace older entries according to some protocol. While stochastic lists offer a solution to the impossibility of efficiently computing and storing multi-variable functions without a systematic bias, extrapolation in the inverse of the number of walkers is usually difficult, even though in practice very good results are found already for short lists. This situation is reminiscent of diffusion Monte Carlo, and is hence generic for all population-based methods. We illustrate the method by computing the lowest-order vertex corrections in Hedin's scheme for the Fröhlich polaron and the ground state energy and wavefunction of the Heisenberg model in two dimensions.

arXiv comment7 pages, 5 figures

References in corpus (5)

• Finite-Size Scaling of the Ground State Parameters of the Two-Dimensional Heisenberg Model
• Polaron Problem by Diagrammatic Quantum Monte Carlo
• Population size bias in Diffusion Monte Carlo
• Diffusion Monte Carlo: Exponential scaling of computational cost for large systems
• Lecture notes on Diagrammatic Monte Carlo for the Fröhlich polaron

Cited by in corpus (5)

• Projective quantum Monte Carlo simulations guided by unrestricted neural network states
• Self-learning projective quantum Monte Carlo simulations guided by restricted Boltzmann machines
• Simulating disordered quantum systems via dense and sparse restricted Boltzmann machines
• Tunneling in projective quantum Monte Carlo simulations with guiding wave functions
• Zero-temperature Monte Carlo simulations of two-dimensional quantum spin glasses guided by neural network states