Preferred Networks has launched "Matlantis CSP (MTCSP)," a new service within its cloud-based atomistic simulation platform, Matlantis. At the core of this service lies Optuna, an open-source black-box optimization framework led by Preferred Networks. Crystal Structure Prediction (CSP) is a critical step in developing new materials, yet the search space is vast, and traditional Density Functional Theory (DFT) methods require immense computational resources and time. MTCSP addresses this by utilizing the "PFP" universal Neural Network Potential for rapid energy evaluation, while Optuna highly optimizes the entire exploration process.

Within this system, Optuna plays a pivotal role that extends beyond simple parameter tuning. Specifically, it generates candidate crystal structures based on user-defined compositions and conditions, constructing an efficient search loop centered on energy metrics. By leveraging Optuna’s strength in large-scale asynchronous parallel processing, the system can evaluate tens to hundreds of thousands of structural candidates within a realistic timeframe. This breakthrough liberates researchers from tedious trial-and-error, allowing them to focus on more fundamental physical and chemical analysis.

Furthermore, the search algorithm incorporates specialized logic for crystal structure exploration based on "NSGA-II," a standard method in genetic algorithms. To manage data persistence, the company developed a dedicated "Structure Store" optimized for crystal structure data rather than relying on conventional relational databases, achieving both storage efficiency and high-speed access. By fusing cutting-edge AI optimization with domain-specific material science expertise, the platform enhances its position as a vital infrastructure for accelerating next-generation material development. The fact that the evolution of the open-source Optuna directly translates into performance gains for MTCSP creates a powerful and sustainable ecosystem.

Preferred Networks has launched "Matlantis CSP (MTCSP)," a new service within its cloud-based atomistic simulation platform, Matlantis. At the core of this service lies Optuna, an open-source black-box optimization framework led by Preferred Networks. Crystal Structure Prediction (CSP) is a critical step in developing new materials, yet the search space is vast, and traditional Density Functional Theory (DFT) methods require immense computational resources and time. MTCSP addresses this by utilizing the "PFP" universal Neural Network Potential for rapid energy evaluation, while Optuna highly optimizes the entire exploration process.

Within this system, Optuna plays a pivotal role that extends beyond simple parameter tuning. Specifically, it generates candidate crystal structures based on user-defined compositions and conditions, constructing an efficient search loop centered on energy metrics. By leveraging Optuna’s strength in large-scale asynchronous parallel processing, the system can evaluate tens to hundreds of thousands of structural candidates within a realistic timeframe. This breakthrough liberates researchers from tedious trial-and-error, allowing them to focus on more fundamental physical and chemical analysis.

Furthermore, the search algorithm incorporates specialized logic for crystal structure exploration based on "NSGA-II," a standard method in genetic algorithms. To manage data persistence, the company developed a dedicated "Structure Store" optimized for crystal structure data rather than relying on conventional relational databases, achieving both storage efficiency and high-speed access. By fusing cutting-edge AI optimization with domain-specific material science expertise, the platform enhances its position as a vital infrastructure for accelerating next-generation material development. The fact that the evolution of the open-source Optuna directly translates into performance gains for MTCSP creates a powerful and sustainable ecosystem.