New updating scheme for Quantum Monte Carlo method suitable for the acceleration using hybrid parallelization

2014 Fiscal Year Final Research Report

• Project/Area Number: 25600156
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Computational science
• Research Institution: Japan Advanced Institute of Science and Technology
• Principal Investigator: MAEZONO RYO 北陸先端科学技術大学院大学, 情報科学研究科, 准教授 (40354146)
• Project Period (FY): 2013-04-01 – 2015-03-31
• Keywords: 並列計算 / 電子状態計算 / 高速化 / GPU / モンテカルロ

Outline of Final Research Achievements

We considered a new updating scheme for particle positions appearing in Monte Carlo methods, which is suitable for the GPGPU acceleration. In the particle-by-particle Metropolis updating scheme, we can evaluate whether a proposed update of a particle position was accepted or rejected base on the configuration of other particles at the previous step, though most of the conventional implementation uses those at the present step. By taking the previous step configuration, we can create the considerable number of degrees for further internal parallel processing. We implemented the internal parallel processing using GPGPU further on the conventional MPI implementation, and achieved the reasonable acceleration.

Free Research Field

物性理論、計算物理学