Extracting Crystal Chemistry from Amorphous Carbon Structures.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Predicting novel carbon allotropes is crucial for materials science.
• Traditional ab initio methods like density functional theory (DFT) are computationally expensive.
• Existing methods struggle with the vast search space of possible structures.

Purpose of the Study:

• To develop a faster and more efficient method for predicting new carbon allotropes.
• To demonstrate the transferability of machine learning models in structure prediction.
• To overcome the computational limitations of DFT for materials discovery.

Main Methods:

• Utilized a novel class of machine-learning-based interatomic potentials.
• Employed random structure searching algorithms.
• Trained models exclusively on structural data from liquid and amorphous carbon.

Main Results:

• Successfully predicted several previously unknown carbon allotropes.
• Demonstrated true transferability of the machine learning model to crystalline phases without prior knowledge.
• Achieved computational speeds orders of magnitude faster than DFT.

Conclusions:

• Machine learning potentials offer a promising, highly efficient alternative to DFT for exploring materials.
• This approach enables large-scale structure searches and accelerates the discovery of novel materials.
• The demonstrated transferability is key for broad applications in chemistry and materials science.