Large-scale Parallel Cluster Dynamics Simulation Techniques And Applications For Radiation Damage In Structural Materials

Radiation damage in the structural materials of nuclear reactors directly affects the safety and economy of reactors.Multiscale simulation combined with experiments is the main way to study radiation damage.In multiscale simulation systems,the mesoscale cluster dynamics(CD)is one of the most effective methods to study the long-term behaviors of damage in materials.This paper focuses on the core topic of "massively parallel simulation of CD" and conducts research from deterministic and stochastic approaches to CD.With the help of parallel computing technology and domestic supercomputers,develop massively parallel CD simulation techniques are developed to expand the simulation scale,and carry out engineering applications.The main work and innovation achievements of this paper are as follows:(1)The deterministic simulations of CD require solving a large number of stiff equations.Based on the exponential time differencing(ETD)method,a massively parallel ETD-CD technique for deterministic simulation is proposed in this paper,which can enable efficient,explicit,and parallel solutions of large-scale CD equations.For simple systems,a parallel ETD-CD algorithm based on the highorder ETD4RK scheme is proposed.After analyzing the computational hotspots in the algorithm,the paper optimizes it by conditional branch elimination and SIMD intrinsics on the Sunway TaihuLight supercomputer.The optimized version achieves an average time reduction of 40%across 11 equation scales(210～220).On the Sunway TaihuLight supercomputer,the technique finally achieves an efficient solution of over one billion CD equations on 8,192 processes,and a parallel efficiency of more than 81%is obtained by taking 512 processes as the baseline.(2)For complex systems,the paper focuses on the spatially resolved stochastic cluster dynamics(SRSCD)method.Aiming at the high computational cost of SRSCD in long-scale simulations,a set of massively parallel SRSCD simulation techniques is proposed based on the synchronous parallel(SP)algorithm and synchronous sublattice(SL)algorithm,which can realize large-scale expansion of CD.First,the paper optimizes the existing SP-SRSCD method and develops an optimized SP(OSP)SRSCD method,which expands the simulation scale to over ten million volume elements and tens of thousands of processes.Furthermore,based on the SL algorithm,a synchronous sublattice parallel(SLP)SRSCD method is proposed,and an adaptive synchronous algorithm and an on-demand communication strategy are designed to reduce the communication frequency and eliminate communication redundancy.On the "Eastern" supercomputer,the technique obtains a good performance of over 74%parallel efficiency on 25,600 processes with a 128-million-volume-element(128 μm3),and a parallel efficiency of over 74%is obtained by taking 3,200 processes as the baseline.(3)For the high-efficiency storage access requirements of more than one million and dynamically changing defects and reactions in SRSCD simulations,a defect-reaction associated storage and update technique is proposed in the paper,which can realize on-demand storage and rapid update of defects and their reactions.Based on the analysis of the relationship and storage requirements of defects and their reactions,this paper designs a defect-reaction tree(DRT)data structure,which stores defects in layers and associates them with their reactions while merging similar reactions.It eliminates redundant storage and speeds up data updates.And based on the DRT structure,a double-grouping search strategy is designed to improve the selection speed of reaction events.Compared with the SP-SRSCD,this technique achieves 39%memory savings and 12x computing acceleration.(4)Utilizing the proposed techniques,a massively parallel SRSCD simulation software,MISA-SCD,is developed in the paper.Compared with other similar international software(such as MFVISC,PARASPACE,SP-SRSCD,etc.),it is characterized in terms of versatility,functionality,simulation scale,and efficiency.The accuracy of MISA-SCD is verified by two simulation examples:pure iron under neutron irradiation and iron-copper alloy under electron irradiation.Then,MISA-SCD is applied to the radiation embrittlement prediction of reactor pressure vessel(RPV)steels,realizing the long-term evolution simulation of "complex system+spatial resolution".The changes in yield strength and ductile-to-brittle transition temperature of RPV steels as a function of irradiation dose are obtained.Compared with the experimental data,the relative error is not more than 20%(within the acceptable range of engineering applications).The prediction of the embrittlement trend of RPV steels in service for more than 100 years has been successfully achieved,realizing good application effects.