Network analysis of synthesizable materials discovery

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Predicting the synthesizability of inorganic materials computationally remains a significant challenge.
• Material synthesis depends on thermodynamic stability, kinetics, synthesis techniques, and precursor availability, complicating direct first-principles prediction.
• Current computational methods struggle to provide a general theory for predicting experimental synthesis success.

Purpose of the Study:

• To develop a novel computational approach for predicting the synthesizability of inorganic materials.
• To leverage machine learning on a dynamically evolving network of material stability to forecast synthesis outcomes.
• To accelerate the discovery of new inorganic materials by improving prediction of experimental feasibility.

Main Methods:

• Constructed a scale-free network representing inorganic material stability by integrating high-throughput density functional theory calculations of free-energy surfaces.
• Extracted experimental discovery timelines from citation data to inform the network's evolution.
• Applied machine learning algorithms to the time-evolution of network properties to predict material synthesizability.

Main Results:

• Demonstrated that the dynamics of the materials stability network can be used to predict synthesizability.
• Showcased a machine learning model capable of forecasting the likelihood of successful experimental synthesis for hypothetical materials.
• Established an alternative pathway to predicting synthesizability beyond traditional thermodynamic and kinetic assessments.

Conclusions:

• The materials stability network dynamics offer a powerful new paradigm for predicting inorganic material synthesizability.
• Machine learning applied to this network provides a practical tool for accelerating computational materials discovery.
• This approach overcomes limitations of purely first-principles or thermodynamic predictions by incorporating network dynamics and discovery timelines.