[Paper Review] Predicting emergence of crystals from amorphous matter with deep learning

Muratahan Aykol, Amil Merchant+3 more|arXiv (Cornell University)|Oct 2, 2023

Machine Learning in Materials ScienceMaterials Science3 citations

TL;DR

This paper introduces a deep learning framework that predicts metastable crystal phases emerging from amorphous precursors by modeling local atomic structural motifs using universal deep learning potentials. It achieves high accuracy in identifying nucleation pathways across diverse inorganic systems, enabling predictive access to new metastable materials via mechanistic application of Ostwald's rule of stages.

ABSTRACT

Crystallization of the amorphous phases into metastable crystals plays a fundamental role in the formation of new matter, from geological to biological processes in nature to synthesis and development of new materials in the laboratory. Predicting the outcome of such phase transitions reliably would enable new research directions in these areas, but has remained beyond reach with molecular modeling or ab-initio methods. Here, we show that crystallization products of amorphous phases can be predicted in any inorganic chemistry by sampling the crystallization pathways of their local structural motifs at the atomistic level using universal deep learning potentials. We show that this approach identifies the crystal structures of polymorphs that initially nucleate from amorphous precursors with high accuracy across a diverse set of material systems, including polymorphic oxides, nitrides, carbides, fluorides, chlorides, chalcogenides, and metal alloys. Our results demonstrate that Ostwald's rule of stages can be exploited mechanistically at the molecular level to predictably access new metastable crystals from the amorphous phase in material synthesis.

Motivation & Objective

• To predict the crystalline phases that emerge from amorphous precursors in inorganic materials.
• To overcome the limitations of traditional molecular dynamics and ab initio methods in simulating complex crystallization pathways.
• To establish a predictive framework for accessing metastable polymorphs through local structural motif analysis.
• To demonstrate the applicability of deep learning potentials in capturing the mechanistic basis of Ostwald's rule of stages at the atomic level.

Proposed method

• The method employs universal deep learning potentials trained on diverse atomic configurations to model energy and forces at the atomistic scale.
• Local structural motifs extracted from amorphous phases are used as input to predict their crystallization pathways.
• The model samples multiple crystallization pathways by exploring structural transformations from amorphous to crystalline states.
• It leverages the concept of local atomic environment stability to identify the most probable nucleation pathways.
• The approach is validated across a broad range of materials, including oxides, nitrides, carbides, fluorides, chlorides, chalcogenides, and metal alloys.
• Ostwald's rule of stages is applied mechanistically by identifying low-energy intermediate phases that precede final crystal structures.

Experimental results

Research questions

• RQ1 Which crystal phases are most likely to nucleate from a given amorphous precursor in inorganic materials?
• RQ2 Can deep learning potentials accurately predict the sequence of metastable phases during amorphous-to-crystalline transitions?
• RQ3 To what extent can local atomic structural motifs predict the final crystalline polymorphs formed from amorphous phases?
• RQ4 How can Ostwald's rule of stages be systematically applied at the atomistic level to guide material synthesis?
• RQ5 Can this framework be generalized across diverse chemical systems, including complex oxides and intermetallic alloys?

Key findings

• The model accurately predicts the initial nucleation phases of polymorphs across a wide range of inorganic materials, including oxides, nitrides, carbides, and chalcogenides.
• The framework successfully identifies metastable crystal structures that emerge from amorphous precursors, even in complex systems with multiple possible polymorphs.
• The approach demonstrates high predictive accuracy in capturing the sequence of phase transitions consistent with Ostwald's rule of stages.
• Deep learning potentials enable efficient and reliable sampling of crystallization pathways that are computationally prohibitive with ab initio methods.
• The method reveals that local structural motifs in amorphous phases are strong predictors of the resulting crystalline polymorphs.
• The framework enables the targeted design of synthesis protocols to access specific metastable phases by predicting the most likely nucleation pathways.