Researchers working in several fields find it challenging to develop materials that exhibit ideal properties to meet certain functions. The issue is common in areas ranging from catalysis to solar cells.
To expedite the pace of this development process, researchers could rely on modeling methods that are capable of predicting information that can guide further improvements.
Scientists from The University of Tokyo Institute of Industrial Science have designed a machine learning model that can find the properties of adsorbed and bonded materials based on the specifications of individual components.
The study results were published in the journal Applied Physics Express.
Features like strength and length of bonds in materials are of significant importance when identifying the structure and characteristics experienced on the macroscopic scale. Thus, the potential to easily estimate such characteristics makes a significant contribution to designing new materials.
The density of states (DOS) is a parameter that can be estimated for molecules, individual atoms and materials. This refers to the choices available to the electrons to organize themselves inside a material.
A modeling technique capable of processing this data for specific components and generate useful data for the required product, by eliminating the need to make and examine the material, would be an impressive tool.
The team employed a machine learning technique, where the model is capable of refining its response without human intervention. This is done to project four various characteristics of products from the DOS data of the separate components.
Earlier, the DOS has been used as a descriptor to form single parameters. This is the first time it has been utilized to predict multiple different properties.
We were able to quantitatively predict the binding energy, bond length, number of covalent electrons, and the Fermi energy after bonding for three different general types of system.
Eiki Suzuki, Study First Author, Institute of Industrial Science, University of Tokyo
Determining the DOS of an isolated state is less complex compared to determining those for the bonded systems. Thus, the analysis is comparatively efficient. Moreover, the neural network model exhibited a good performance even when a mere 20% of the dataset was involved for training.
A significant advantage of our model is that it is general and can be applied to a wide variety of systems.
Teruyasu Mizoguchi, Study Corresponding Author, Institute of Industrial Science, University of Tokyo
“We believe that our findings could make a significant contribution to numerous development processes, for example in catalysis, and could be particularly useful in newer research areas such as nano clusters and nanowires,” added Teruyasu Mizoguchi.
Suzuki, E., et al. (2021) Accurate prediction of bonding properties by a machine learning–based model using isolated states before bonding. Applied Physics Express. doi.org/10.35848/1882-0786/ac083b.