Towards modeling at all scales

Modeling porous solids is at the sale time predicting in a theoretical way their physicochemical properties, understanding them and building new structures.

By bringing together a community of Ile-de-France modelers, with different specialties, Respore brings together their know-how to understand and predict the properties of porous matrices. This will guide the development of new architectures and processes, through the combination of existing modeling techniques and the development of new approaches.

For modeling to be optimal, it is necessary to combine approaches at different scales, from microscopic to macroscopic scale. Each modeling specific to a restricted time and space scale only allows us to understand the impact of a phenomenon on this scale on certain properties.

However, multi-scale modeling must also take into account the complexity of the real system in its environment (thermal, chemical, mechanical stresses, etc.): within the pores (host-matrix interactions, confined fluids, diffusion, electronic properties) at the interfaces (predominant role in composite materials) or at long distance. It is necessary to combine electronic, photophysical and photochemical, mechanical and thermodynamic properties. This applies not only to pure phases but also to heterogeneous systems: solid / fluid interfaces, heterogeneous / composite materials (crystalline or amorphous and multivariate / multifunctional structures).

Developing a digital database of real or hypothetical materials will also be a major asset. This "materials genome" will allow to quickly identify the structures with the desired properties among a large number of candidates.

A first approach consists in developing, validating, and using simple descriptors showing good correlations with the desired physical, chemical, textural ... properties. Another approach, purely combinatorial, based on screening, can also be supplemented by automatic learning (machine learning) or deep learning methods. These methods do not require the prior definition of suitable descriptors and use all the raw data produced by the computational methods.