Materials Screening Through GPU Accelerated Topological Mapping

Abstract

Selecting materials with properties tailored to a specific application may be accelerated through materials informatics, but kinetic properties whose calculations are too computationally intensive to be incorporated into materials screening must be replaced with appropriate descriptors. Here we present a highly optimized method for general processing on graphics processing hardware through which we map the interstitial subspace of atomic structures that are used as a qualitative predictor for diffusivity. Additionally, analytical methods that determine the largest channel diameter and identify the optimal path through a material are proposed to characterize this topology. Analysis of the interstitial subspace, along with the theoretical capacities for Li ions, has lead us to select high-capacity lithium ion battery (LIB) materials that display both promising capacities and migration pathways able to support lithium insertion and removal. The result is the identification of Li₂MgSi, a LIB anode material with a theoretical capacity of 1023 Ah/kg, from an unfiltered set of 1,754 structures.

Keywords:

• Accessible volume; CUDA; High throughput; GPU; Lithium ion batteries; Materials informatics; Pore characterization; Single-instruction multiple-data (SIMD)

ACKNOWLEDGEMENTS

We thank Professor Nirupam Chakraborti for the invitation to write this article. Computational resources from the Golden Energy Computing Organization are gratefully acknowledged.

Notes

¹The conditions for CONNECTS ALONG to return true are only satisfied if SPANS is also true, therefore, the test on the 13th line of pseudocode is not strictly necessary, but is included for clarity.