douka: A universal platform of data assimilation for materials modeling.

Area of Science:

• Materials Science
• Computational Modeling
• Data Assimilation

Background:

• Materials modeling often requires estimating physical properties not directly measurable from experimental data.
• Accurate state estimation and uncertainty quantification are crucial for understanding complex physical processes and refining models.

Purpose of the Study:

• To develop and apply a large-scale, general-purpose data assimilation (DA) platform, named douka, for nonlinear materials models.
• To demonstrate the platform's capability in estimating unobservable physical properties and providing quantified uncertainty.
• To validate the platform's performance and scalability using experimental data from the oxygen evolution reaction.

Main Methods:

• Development of the douka data assimilation platform for materials modeling.
• Application of douka to nonlinear materials models, including experimental images of the oxygen evolution reaction.
• Large-scale ensemble data assimilation performed on the Fugaku supercomputer with up to 8192 ensemble members.

Main Results:

• The douka platform successfully estimated physical properties not directly obtained from observed data.
• State estimation with quantified uncertainty was achieved, offering new insights into physical processes.
• Successful application to oxygen evolution reaction images allowed estimation of oxygen gas injection velocity and bubble contact angle.
• Runtime scaling analysis confirmed computational efficiency following the weak scaling law for large ensemble sizes.

Conclusions:

• The douka platform is effective for estimating unobservable physical properties and quantifying uncertainty in materials modeling.
• The platform enables integration of experimental data with numerical simulations, advancing data-driven materials science.
• The demonstrated scalability ensures computational efficiency for large-scale applications on supercomputing infrastructure.