A Machine-Learning-Based Approach for Solving Atomic Structures of Nanomaterials Combining Pair Distribution...

Area of Science:

• Solid-state chemistry and physics
• Materials science
• Crystallography

Background:

• Determining crystal structures of nanocrystalline or amorphous materials is challenging.
• Pair distribution function (PDF) analysis aids in structure determination but requires a starting model.
• Existing methods often struggle with complex structures like metastable configurations or stacking disorders.

Purpose of the Study:

• To develop an automated algorithm for determining crystal structures of unknown compounds.
• To overcome the limitations of traditional PDF analysis by eliminating the need for a predefined structural motif.
• To enable the identification of complex structural features such as metastable phases and stacking faults.

Main Methods:

• An on-the-fly trained machine learning model was developed.
• The algorithm combines density functional theory (DFT) calculations with experimental PDF data.
• Global optimization was performed within an artificial landscape constructed from calculated and measured PDFs.

Main Results:

• The algorithm successfully determines crystal structures without requiring an initial structural model.
• It demonstrates the capability to identify metastable configurations, which are often missed by conventional methods.
• The approach effectively detects stacking disorders in crystalline materials.

Conclusions:

• The presented machine learning algorithm offers a powerful new tool for crystal structure determination, particularly for challenging nanocrystalline and amorphous materials.
• This method advances the field of solid-state structure analysis by automating the process and enabling the characterization of complex structural phenomena.
• The algorithm has significant implications for materials discovery and understanding material properties through accurate structural elucidation.