In this proposal, we show that the construction and utilization of a "digital twin for materials creation" is extremely effective in accelerating the materials R&D process and achieving innovation. In the digital twin, the entire materials development process is modeled in cyberspace, and the development process is carried out only in cyberspace as much as possible.

In most industrial fields, there is a growing expectation to improve the properties of the materials used in each field. However, the materials currently in use have been discovered because of improvement long time to provide the best performance for each application, and it has become extremely difficult to discover new materials that surpass the performance of these materials through conventional materials research and development methods. Materials Informatics (MI), which uses AI including machine learning to search for materials based on data obtained from experiments and materials science calculations, has been expected to be a key technique to overcome such situation since around 2010. Although MI has successfully provided its effectiveness in some fields, it has not yet fundamentally improved the efficiency of materials research and development, as it often fails to synthesize predicted materials or, when it does synthesize them, fails to achieve the expected performance. A reason for this is that the results of the material science calculations that MI uses for learning do not currently coincide the actual material properties with the desired accuracy.

The material development process in real space is executed by repeating the "design-synthesizeevaluate" cycle. Since the digital twin for material creation requires the cycle to be modeled in cyberspace, it is necessary to create appropriate model of not only the material itself, but also the environment surrounding the material and the interaction between the environment and the material. To achieve this, the following issues in current computational materials science must be resolved.

R&D Subject 1: Development of Modeling Methods for Realistic Systems

The first target is to build an appropriate model of the real system in cyberspace. In computational materials science, some kinds of computational processing are performed on models of atoms, molecules, or groups of atoms or molecules created in cyberspace to obtain information about the state of matter and its time variation. When proper modeling is obtained, the calculation results should predict the result of physical property measurement and state changes of materials in the real world. The appropriate modeling methods depend on the target material and the information we want to know, so modeling methods need to be developed for each purpose.

R&D Subject 2: Development of computational methods for frontier areas

In first-principles calculations, which are the basis of computational materials science, there are certain objects and phenomena for which it is difficult to obtain realistic calculation results. Among such difficult targets and phenomena, the particularly important "electronic correlation," "excited states," "dynamics," and "multiscale" are called frontier areas in this proposal. To handle frontier areas in cyberspace, it is necessary to develop new computational principles and methods.

R&D Subject 3: Development of Multi modal Method

For objects that are difficult to handle with a single calculation method or modality, it is important to combine them. The combination approach includes the combination of different computational methods in computational materials science (e.g., mixing multiple frontiers), the combination of computation with experimental and data science (e.g., data assimilation, machine learning models), the combination of different types of computing machines (e.g., CPU and GPU, classical and quantum computers).

To promote R&D in this area, national research programs to develop the methods listed in the subjects 1 to 3 will be necessary. Moreover, a research center, which is a long-lived organization that takes in new software developed in R&D projects and continuously improves it and provides user support, will play a crucial role. Since developing new materials science theories, computational methods, and software and making various technical contributions to their dissemination require different kinds of skills, motivation, and time management, it is effective to have both short-term intensive R&D projects and long-term sustained research center operations. Furthermore, to ensure the permanence of developed software, commercialization in the broad sense of the term is necessary. It is necessary to take measures to support business development for social implementation, for example, by creating a framework for venture support and a licensing mechanism, as well as efforts to foster the industry. It is also extremely important to conduct the necessary human resource development within each of these measures.