Bayesian Co-Navigation: Bridging Theory and Experiment to Build Material Digital Twins

The predictive power of theoretical models makes them highly attractive for cost-efficient materials design. From this perspective, digital twins of materials – models tailored to replicate a specific class of materials and capture the evolution of their properties over time and under treatments – are of particular interest. However, creating digital twins capable of reproducing not only macroscopic behavior but also the underlying multiscale physical mechanisms remains a nontrivial challenge.

Here, we introduce a Bayesian co-navigation framework that unifies experimental and theoretical studies within a single optimization workflow. The framework integrates two active-learning loops - one theoretical and one experimental - each driven toward the optimal composition. An additional outer alignment loop iteratively refines model parameters to minimize discrepancies between prediction and experimentally observed properties.

This theoretical model refinement based on the experimental feedback transforming the theoretical model into a digital twin. We demonstrate this approach using PbTiO₃ films, co-navigating between FerroSIM simulations and real Piesoresponse Force Microscopy data. We believe that the proposed framework can streamline the development of digital twins, helping to uncover the origins of targeted functionalities.