Multiscale Simulation Algorithms for Materials: Development & Application
Haifeng Song1,2*, Xingyu Gao1, Jun Fang1, Haifeng Liu1, Yafan Zhao1,2,Deye Lin1,2, Guomin Han1,2, Xueyan Zhu1,2, Yuzhi Zhou1,2
1 Institute of Appllied
Physics and Computational Mathematics, 100094, Beijing
2 CAEP Software Center for
High Performance and Numerical Simulation, 100088, Beijing
ABSTRACT: The materials genome
initiative accelerates the development of materials. In particular, multi-scale
simulation algorithms combined with high-throughput experiments and big data
analysis have been successfully applied to the design and prediction of
materials composition- structure-properties. To achieve spatiotemporal
multiscale simulation of materials, our team has developed a variety of
computational methods to improve the functionality and efficiency of
simulations.
At the electronic scale, a number of first-principles
computing techniques have been developed including three-level parallel approach,
highly scalable FFT methods, accelerated electron interaction calculation and
adaptive ion interaction iteration, which improve the stability and efficiency
of simulations.
At the atomic scale, the modified Z method for
simulating melting has been developed and validated, achieving a realistic
simulation of the evolution of the system from solid phase into the
solid-liquid coexistence. On the other hand, to accelerate structure search,
techniques of random structure generation, repetitive structure identification
and high-throughput computation have been developed based on basin-hopping
algorithm.
At the mesoscale, a phase-field model has been
developed for investigating the irradiation induced void swelling of metals and
the morphological evolution of hydrides. Based on phase-field simulation, the
morphological characteristics of hydrides are identified, which reveals that
the tensile hoop stress induced transition of stacking sequences from along the
circumferential direction to along the radial direction.
A multiscale simulation capability for structure
prediction, dynamical behavior and microstructure evolution has been built,
providing a powerful tool for further material and physical studies. Above
researches have been selected by the National Supercomputing Guangzhou Center
for “2018 Tianhe Star Outstanding Applications”.
Keywords: materials genome; multiscale simulation; computational methods.
