Distance Matrix-Based Crystal Structure Prediction Using Evolutionary Algorithms.

Area of Science:

• Materials Science
• Computational Chemistry
• Crystallography

Background:

• Crystal structure prediction (CSP) is a major challenge in materials science.
• Current ab initio methods (e.g., DFT-based) are computationally expensive and limited to small systems.
• Protein structure prediction utilizes knowledge-based approaches, unlike the primarily ab initio focus in CSP.

Purpose of the Study:

• To explore the use of known geometric constraints, specifically pairwise atomic distances, for crystal structure prediction.
• To develop a novel algorithm, DMCrystal, inspired by knowledge-based approaches in bioinformatics.
• To improve the efficiency and accuracy of inorganic crystal structure prediction.

Main Methods:

• Developed DMCrystal, a genetic algorithm (GA)-based crystal structure reconstruction algorithm.
• Utilized predicted pairwise atomic distances as geometric constraints.
• Compared DMCrystal's performance against CMCrystal, an atomic contact map-based CSP algorithm.

Main Results:

• Predicted pairwise atomic distances significantly aid in crystal structure reconstruction.
• DMCrystal demonstrates substantially improved performance compared to CMCrystal.
• The approach highlights the benefit of incorporating learned atomic interaction information.

Conclusions:

• Knowledge-based geometric constraints, like atomic distances, can effectively enhance inorganic CSP.
• DMCrystal offers a more efficient and accurate method for predicting crystal structures.
• This work suggests a promising direction for advancing materials discovery through computational methods.