Title: Multiscale techniques and Machine Learning methods for improving and predicting the properties of nanoporous materials

Abstract

Metal Organic Frameworks, MOFs, have attracted the last two decades a major interest from scientific community due to their wide range of applications. With the appropriate selection of their primary and secondary building units, one can create structures with tunable properties according to the application they are aimed to be used for, leading to almost infinity number of candidate structures. Gas adsorption, and separation and drug delivery are among the applications that have been extensively studied. By combining accurate Quantum calculations and fast classical methods one can study possible modifications on exiting materials and their impact on their performance concerning the aforementioned applications. The last few years a number of MOFs databases appeared in the literature offering the ability by using high throughput calculations to correlate the structural details of the materials with their performance on gas adsorption and separation. Due to the large number of the materials these databases include and in order to effectively succeed this correlation machine learning methods are also applied.

In this presentation it is presented the impact of functionalization of already existing materials on their performance concerning gas adsorption and separation together with controlled released of drugs from their host structures for pharmaceutical use. In addition, a series of MOFs databases are presented and the correlation of their properties with their performance is studied by using Machine Learning, ML, techniques. Some new descriptors for improving the ML accuracy are introduced. Finally, a new scheme for faster high throughput calculations is presented having as a goal to reduce the time needed for conducting simulations for every single structure of such large databases.