発表者名 |
新井 敦郎 |
指導教員名 |
求 幸年 教授 |
発表題目(英語) |
Self-learning Monte Carlo method |
要旨(英語) |
Markov chain Monte Carlo is an unbiased numerical method for studying physical systems. A significant problem is that there exist correlations between configurations in the chain, especially around critical points, which leads to the drastic increase in the computational cost. For some models, Global update algorithms have been proposed to reduce the cost (e.g. Swendsen-Wang, Wolff). For most of the general models, however, any efficient global updating procedure is known.
In this presentation, based on [1], I will introduce the Self-learning Monte Carlo method, in which an effective model is first trained from the training data generated in trial simulation and then utilizing this model for update proposal. Self-learning Monte Carlo method is applied to a spin model at the phase transition point, achieving a 10-20 times speed up.
[1] J. Liu et al., Physical Review B 95, 041101(R) (2017) |
発表言語 |
日本語 |