Tokyo, Japan – Developing materials that have the properties necessary to perform specific functions is a challenge for researchers working in areas from catalysis to solar cells. To accelerate development processes, modeling approaches can be used to predict information to guide refinements. Researchers at the Institute of Industrial Science at Tokyo University have developed a machine learning model to determine properties of bound and adsorbed materials based on parameters of the individual components. Your results will be published in Applied Physics Express.

Factors such as the length and strength of bonds in materials play a crucial role in determining the structures and properties that we experience on a macroscopic scale. The ability to easily predict these properties is therefore valuable in the development of new materials.

The density of states (DOS) is a parameter that can be calculated for individual atoms, molecules and materials. Put simply, it describes the possibilities that are available to the electrons that arrange themselves in a material. A modeling approach that uses this information for selected components and provides useful data for the desired product – without the material having to be manufactured and analyzed – is an attractive tool.

The researchers used a machine learning approach – in which the model refines its response without human intervention – to predict four different properties of products from the DOS information of each component. Although the DOS has been used as a descriptor to set individual parameters before, this is the first time that several different properties have been predicted.

“We were able to quantitatively predict the binding energy, the bond length, the number of covalent electrons and the Fermi energy after binding for three different general system types,” explains first author of the study Eiki Suzuki. “And our predictions were very accurate for all properties.”

Since the computation of the DOS of an isolated state is less complex than for bound systems, the analysis is relatively efficient. In addition, the neural network model used performed well, even if only 20% of the data set was used for training.

“A major advantage of our model is that it is universal and can be applied to a large number of systems,” explains study correspondent Teruyasu Mizoguchi. “We believe that our findings make a significant contribution to numerous development processes, for example in catalysis, and could be particularly useful in newer research areas such as nanoclusters and nanowires.”


The article “Accurate Prediction of Bonding Properties by a Machine Learning-based Model using Isolated States Before Bonding” was published in Applied Physics Express at DOI: 10.35848 / 1882-0786 / ac083b.

Via the Institute of Industrial Science (IIS), the University of Tokyo

Institute of Industrial Science (IIS) at the University of Tokyo is one of the largest university-linked research institutes in Japan.

The IIS comprises more than 120 research laboratories, each headed by a faculty member, with more than 1,200 members, including approximately 400 staff and 800 students who are actively involved in education and research. Our activities cover almost all areas of the engineering discipline. Since its inception in 1949, IIS has worked to fill the large gaps between academic disciplines and real world applications.


Please enter your comment!
Please enter your name here