Speaker
Description
Amorphous porous materials play a central role in energy and environmental technologies, including direct air capture of CO₂ and heterogeneous catalysis. Their performance is governed by strong local heterogeneity at the atomic scale, where variations in coordination, topology, and chemical environment control adsorption, reaction energetics, and transport. Capturing these effects with atomistic simulations is challenging, as amorphous systems exhibit large statistical variability and require extensive sampling to obtain meaningful structure–property relationships.
We present a multiscale, data-driven framework that addresses this challenge by constructing surrogate models linking local atomic structure to key quantities of interest. Atomistic simulations are used to generate representative ensembles of amorphous configurations, from which local atomic environments are described using physically motivated descriptors. Supervised learning techniques, in particular partial least squares (PLS), are employed to identify low-dimensional representations that retain the dominant correlations between structure and material response.
These reduced descriptors serve as inputs to surrogate models based on Gaussian process regression (GPR/kriging), enabling fast prediction of properties such as adsorption energies, grafting energies of metal dopants, or energy barriers along selected catalytic pathways. Importantly, the probabilistic nature of these surrogates provides uncertainty estimates, which are exploited through active-learning strategies to guide additional atomistic calculations and systematically improve model accuracy at minimal computational cost.
By replacing expensive brute-force sampling with uncertainty-aware surrogate models, the proposed framework enables efficient exploration of heterogeneous amorphous materials while preserving physical interpretability. The approach provides a practical route to quantify structure–property relationships in disordered porous media and supports the rational design of materials for energy and climate-relevant applications.
| Country | Switzerland |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
