Method and system to determine an optimal set of atom centered symmetry functions (ACSFs)

A system, comprising:

• a memory storing instructions;
• one or more communication interfaces;
• and
• one or more hardware processor(s) coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processor(s) are configured by the instructions to:
• receive, a plurality of parameters associated with a plurality of atom centered symmetry functions (ACSFs);
• generate, an initial set of atom centered symmetry functions (ACSFs) by varying the plurality of parameters associated with the plurality of ACSFs;
• compute, a value of each of the initial set of ACSFs for each local environment in a training dataset, wherein each local environment corresponds to a region within a cut-off radius (Rc) from a central atom;
• construct, a histogram with a prespecified bin size to obtain a distribution of the value of each of the initial set of ACSFs;
• prune, the initial set of ACSFs to obtain a pruned list of ACSFs based on a width and a maximum value of the distribution of the value of each of the initial set of ACSFs;
• sort, the pruned list of ACSFs in a decreasing order of a spread to obtain a sorted list of atom centered symmetry functions (ACSFs);
• and
• determine, an optimal set of a plurality of shortlisted ACSFs by traversing through the sorted list of ACSFs, and wherein a pair-wise distance between a distribution of each of the plurality of shortlisted ACSFs is greater than a predefined threshold.

One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:

• receiving, a plurality of parameters associated with a plurality of atom centered symmetry functions (ACSFs);
• generating, an initial set of atom centered symmetry functions (ACSFs) by varying the plurality of parameters associated with the plurality of ACSFs;
• computing, a value of each of the initial set of ACSFs for each local environment in a training dataset, wherein each local environment corresponds to a region within a cut-off radius (Rc) from a central atom;
• constructing, a histogram with a prespecified bin size to obtain a distribution of the value of each of the initial set of ACSFs;
• pruning, the initial set of ACSFs to obtain a pruned list of ACSFs based on a width and a maximum value of the distribution of the value of each of the initial set of ACSFs;
• sorting, the pruned list of ACSFs in a decreasing order of a spread to obtain a sorted list of atom centered symmetry functions (ACSFs);
• and
• determining, an optimal set of a plurality of shortlisted ACSFs by traversing through the sorted list of ACSFs, and wherein a pair-wise distance between a distribution of each of the plurality of shortlisted ACSFs is greater than a predefined threshold.