Latest Research
- 2024.12.03
- Tateyama-Ando Group
Data-driven Materials Science and Ion Jamology
In recent years, data-driven research based on data science has made rapid progress, and is becoming increasingly important in the materials and energy fields. Data-driven materials research does not only accelerate conventional research processes, but also has the potential to bring about qualitative changes in research itself. We especially focus on the basic cycle of research and development, and promote fundamental technology on data-driven materials research from theoretical and mathematical perspectives.
Conventional materials research and development can be broadly organized into three basic processes: "data generation", "data accumulation", and "data analysis". Data generation refers to the process of acquiring data that will serve as scientific evidence by measurement, calculation, and synthesis. Data accumulation is the process of "structuring" the generated research data and making them available for reference and further analysis. Data analysis is the process of extracting information from the data and connecting it to the next data generation. Research and development generally has a structure that recursively leads from data generation to data accumulation, from data accumulation to data analysis, and from data analysis to the next data generation, which we call the "data cycle". Our concept of data-driven materials research is to further accelerate this cycle by leveraging machine learning, data science, and robotics, and extending it to qualitatively different research.
![]() |
|
Figure 1 | Conceptual diagram of data cycle consisting of "data generation," "data accumulation," "data analysis" in materials research and development |
We use computational simulation as our primary data generation method. Computational simulation is now an indispensable tool for the study of functional materials such as batteries and catalysts. For instance, we have developed and applied machine learning potentials to improve the efficiency of first-principles molecular dynamics simulations, which are highly accurate but expensive to run. Machine learning potentials can provide both accuracy and efficiency in calculations for disordered systems and surfaces/interfaces, where conventional classical potentials are difficult to apply. We applied the search for all ion diffusion pathways in amorphous materials (Fig. 2(a)). The technique of machine-learning potential can be made from small first-principles computational data for large size simulation of material systems [1,2].
In addition to computational simulations, our group contributes to the efficiency of research data generation for experimental studies. For example, we have used Bayesian optimization and robotics to make the data generation process more efficient and autonomous [3]. Bayesian optimization is a method for estimating optimal experimental conditions from limited data and reducing the number of experiments, especially in costly materials synthesis. The autonomous experiment system (Figure 2(b)) developed in collaboration with the group led by Professor Taro Hitosugi and Associate Professor Ryota Shimizu succeeded in deriving and synthesizing conditions that minimize the electrical resistance of TiO2 thin films, improving the efficiency of experiments by about 10 times and achieving the world's first results for inorganic materials.
Automating data generation is an important approach for data-driven materials research, but it is not enough. Without proper management and analysis from the data, it will not be possible to solve research problems. We adopted the JSON (JavaScript Object Notation) format, which is often used in web applications, as a data management method to create a database of research data, which is difficult to be structured in table format. The JSON format has a hierarchical, or tree structure, making it easy to search and manage data. In the performance evaluation experiments of lithium-ion batteries with different electrolyte solutions, experimental data was structured in JSON to enable efficient retrieval and use. By constructing such a database, we can smoothly connect to "data analysis" based on machine learning and are challenging problems such as predicting material functions and understanding material functionalities.
Finally, we will introduce our efforts to connect the framework of data-driven materials research to the understanding of ion flows, which started in FY2024 under the title of "Ion Jamology: materials design transformation by understanding non-equilibrium and collective ion flow.
![]() |
|
Figure 3 | Schematic diagram of Research for Academic Transformation Area (A) Ion Jamology (Reproduced from Reference 4.) |
In battery materials and in catalytic materials, much research has been done to improve material properties by focusing on interactions between ions and their environment. Several discussions have been proposed in which concerted changes in the crystal lattice or aggregations of atoms promote the flow of ions and enhance their functionalities. However, no scientific theory has been developed to accurately capture, understand and control the collective motion of ions (ion flow) and the various interactions that cause it.
Conventional theories of ion flow are based on the assumption that each ion can move independently. This is true only when the ion density is dilute and cannot account for dense ion flows in real materials. As ion density increases, ion-ion interactions become non-negligible. We turned to "jamology" as a simple framework for understanding these interactions. Jamology is the science that deals with the collective motion and flow of vehicles and people, in other words, it is a field of mathematical science that studies transportation and logistics. "Ion Jamology" is an attempt to replace cars and people with charged particles and to understand them using non-equilibrium statistical mechanics that incorporates the interactions between the particles. We aim for a fundamental change in conventional materials and chemical research by collaborating with Computational and Mathematical Sciences (A01), Materials Sythesis (A02), and Advanced Measurements (A03) to describe more dynamic processes that incorporate interactions between ions and their surroundings, and to develop methods to control ion flows from a broad perspective.
References:
[1] | W. Li, Y. Ando, E. Minamitani, and S. Watanabe, J. Chem. Phys. 147, 214106 (2017). |
[2] | S. Watanabe, W. Li, W. Jeong, D. Lee, K. Shimizu, E. Mimanitani, Y. Ando, and S. Han, J. Phys. Energy 3, 012003 (2021). |
[3] | R. Shimizu, S. Kobayashi, Y. Watanabe, Y. Ando, and T. Hitosugi, APL Materials 8, 111110 (2020). |
[4] | https://ion-jamology.jp |