Accelerating Advanced Material Design Through Versatile Atomistic Scale AI Simulator MATLANTIS

In recent years, materials informatics (MI) has attracted significant attention for an efficient material design. Atomistic simulations, such as the density functional theory (DFT) and molecular dynamics (MD) methods, have been recognized as powerful tools in MI to explore innovative materials. For organic materials, e.g., rubbery materials and lubricant, microscopic properties at an interface between organic additives and inorganic materials significantly contribute to macroscopic material properties. However, there are several challenges when applying the DFT and the MD to organic-inorganic complex interfaces. Generally, the computationally expensive DFT calculation is not suitable for rapid materials exploration. On the other hand, the MD calculation runs much faster than the DFT, making it suitable for large-scale systems. However, the MD calculation usually requires empirical force fields, which are well-validated for a specific interfacial system. Empirical force fields require elaborate and rigid validation specifically for materials systems of our interest, resulting in the lack of versatility towards other systems. One of the solutions for these issues is to use a neural network potential. Recently, we have developed “MATLANTIS^®“ software, which implements a versatile neural network potential called PreFerred Potential (PFP). The PFP is well-trained using large DFT datasets including not only energetically stable crystals and molecules, but also surfaces, disordered structures, and those energetically unstable. Thus, the variety of the training data realizes the versality of PFP. Because of its high-speed calculation and the versatility, MATLANITS offers a powerful tool to achieve the materials discovery of complex systems that requires massive iterative calculations, as represented by organic-inorganic interfaces. In our presentation, two case studies of MATLANTIS are presented.<br/>The first case is an exploration of adsorption structures of coupling agents in a silica filled rubber. Silica is often compounded into a polymeric material in order to improve their mechanical properties such as an elastic modulus and a toughness compared to those of a pure rubber material. In the silica-filled rubber, a tough interface is formed between the silica phase and the polymer phase by the silane coupling agents, which have reaction sites in both the silica and the polymer chain. To measure the properties of the interface, deformation simulations of the coupling agent connected to the silica surface were performed by MATLANTIS for a system consisting of the silica slab, the coupling agent, and the polymers.<br/>The second is to elucidate an adsorption mechanism of oiliness additives in lubricant. Lubricant basically consists of base oil and additives. The composition and the structure of the reaction film, formed by the chemical reaction of additives used in lubricant on metal surfaces, contribute significantly to the lubricant performance. MATLANTIS can elucidate that the adsorption of additives on metal surfaces is influenced by coarseness and density of the oil film, which results from differences in the structure of the base oil. Furthermore, PFP-based MD simulations on MATLANTIS extracts some important features from liquid structures in the oil film at the metal interface under the high pressure. The extracted features were then used for the additive search in our materials discovery processes.<br/>Finally, our recent challenges will be raised to develop MATLANTIS into a more versatile platform for materials scientists. One is to develop a new machine-learning potential called LightPFP which is computationally less expensive than original PFP and enables larger-scale calculations including 10⁵-10⁶ atoms. Another topic is mesoscale dissipative particle dynamics simulations based on PFP.