Abstract
Machine learning has the potential to accelerate materials discovery by accurately predicting materials properties at a low computational cost. However, the model inputs remain a key stumbling block. Current methods typically use descriptors constructed from knowledge of either the full crystal structure — therefore only applicable to materials with already characterised structures — or structure-agnostic fixed-length representations hand-engineered from the stoichiometry. We develop a machine learning approach that takes only the stoichiometry as input and automatically learns appropriate and systematically improvable descriptors from data. Our key insight is to treat the stoichiometric formula as a dense weighted graph between elements. Compared to the state of the art for structure-agnostic methods, our approach achieves lower errors with less data.
Highlights
Machine learning has the potential to accelerate materials discovery by accurately predicting materials properties at a low computational cost
A critical challenge exists for materials discovery in that expanding these ab initio efforts to look at novel compounds requires one to first predict the likely crystal structure for each compound
The use of structure-based descriptions means that the resulting models are limited by the same structure bottlenecks as ab initio approaches when searching for novel compounds
Summary
Machine learning has the potential to accelerate materials discovery by accurately predicting materials properties at a low computational cost. In doing so we give up the ability to handle polymorphs for the ability to enumerate over a design space of novel compounds This exchange empowers a new stage in materials discovery workflows where desirable and computationally cheap pre-processing models can be used, without knowledge of the crystal structure, to triage more time consuming and expensive calculations or experiments in a statistically principled manner. The computational complexity of this approach scales exponentially with the number of constituting elements and is not applicable to materials with different numbers of elements or dopants To address this shortcoming, general-purpose material descriptors, hand-curated from the weighted statistics of chosen atomic properties for the elements in a material, have been proposed[24,25,26]. The power of these general-purpose descriptors is circumscribed by the intuitions behind their construction
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
