Genarris 3.0: Generating Close-Packed Molecular Crystal Structures with Rigid Press.

Area of Science:

• Crystallography
• Computational Chemistry
• Materials Science

Background:

• Molecular crystal polymorphism significantly impacts material properties and performance.
• Crystal structure prediction (CSP) is crucial for computationally exploring polymorphs.
• Existing methods require significant computational resources for exploring the crystal structure landscape.

Purpose of the Study:

• To introduce Genarris 3.0, an enhanced open-source code for molecular crystal structure prediction.
• To accelerate the exploration of potential energy landscapes using machine-learned interatomic potentials (MLIPs).
• To develop and validate a new clustering and down-selection workflow for efficient polymorph ranking.

Main Methods:

• Genarris 3.0 incorporates a novel "Rigid Press" algorithm for geometric compression.
• Integration with MACE-OFF23(L) MLIPs for accelerated geometry optimization and energy ranking.
• Application of a clustering and down-selection workflow for efficient data processing.

Main Results:

• Genarris 3.0 successfully predicted crystal structures for six diverse targets, including energetic materials.
• Analysis revealed variability in MLIP performance, particularly for energetic materials, which was mitigated by the new workflow.
• The code effectively generates molecular crystal data sets suitable for training machine learning models.

Conclusions:

• Genarris 3.0 offers an efficient and effective platform for crystal structure prediction.
• The integration of MLIPs and a robust workflow accelerates the discovery of stable polymorphs.
• The generated data sets are valuable for advancing machine learning applications in materials science.