A Novel Machine Learning Model for the Accurate Design of Highly Selective Zeolitic-Imidazolate Frameworks

Krokidas P., Karozis S.

Zeolitic-imidazolate frameworks (ZIFs) have the potential to make highly selective membranes, due to their functionalization capabilities that can take place on the molecular level and affect considerably their microscopic properties. However, their application in gas separations is limited due to the lack of knowledge on the modification-properties correlation.

Our approach is based on the premise that the aperture connecting the cages of ZIFs controls the diffusion rate of guest species. In our recent works, we showed that the aperture’s size and flexibility can change, by replacing the organic linker, the metal center or the functional group of the framework. This in turn provides control over the diffusion-based selectivity of a given pair.1,2,3 This knowledge shapes a chemical basis which we used in a novel machine learning (ML) tool towards developing a complete modification/selectivity correlation for ZIFs, as shown schematically in Figure 1. We built numerous new ZIFs of finely discretized aperture sizes, by replacing various units, and simulated the diffusion of guest molecules of varying in size (He, H2 up to n-butane) in these structures, based on force fields developed based on density functional theory (DFT) calculations. The simulations account for the flexibility of the structure, to assess the framework’s response as a function of the gas penetrant’s size. Then we trained a supervised ML model on descriptors with physical (size and weight of the linkers; charge and length of the functional group) and chemical significance (ionic radius of the metal center and information on the metal-nitrogen bond), as well as information on the response of the framework upon a given gas penetrant’s presence.

Overall, our ML tool enables the prediction of the diffusivity of any gas on any ZIF structure and of the diffusivity of any gas in it without the need for molecular level simulations. The only information needed is basic chemical information from DFT calculations and the size of the gas penetrant. This approach can further promote the fast design of process tailored new ZIFs, by providing as input information only the desired diffusivity/selectivity of a given mixture.

1. Krokidas, P. et al. On the Efficient Separation of Gas Mixtures with the Mixed-Linker Zeolitic-Imidazolate Framework-7-8. ACS Appl. Mater. Interfaces 10, 39631–39644 (2018).

2. Krokidas, P., Moncho, S., Brothers, E. N., Castier, M. & Economou, I. G. Tailoring the gas separation efficiency of metal organic framework ZIF-8 through metal substitution: A computational study. Phys. Chem. Chem. Phys. 20, 4879–4892 (2018).

3. Krokidas, P., Moncho, S., Brothers, E. N. & Economou, I. G. Defining New Limits in Gas Separations Using Modified ZIF Systems. ACS Appl. Mater. Interfaces 12, 20536–20547 (2020).

Figure 1. Outline of our ML model for the fast design of top-performing ZIFs with desired performance.

A Deep Learning approach for spatial and time error correction of numerical weather prediction simulation data

Karozis S., Klampanos I.

Meteorological data are produced in various spatial and time sizes, depending on the application they will be used. The data are the result of Numerical Weather Prediction (NWP) simulations, e.g. the solution of mass & energy balance equations, concerning the fluids of the atmosphere. The uncertainty of the solution (time-series of domain) becomes higher as the prediction goes further to the future, thus, limiting the applicability of hour per hour resolution of such models, to 2 to 5 days ahead. In case of further in time predictions, statistical measurements (mean, man, min, etc) or indicators (tropical days or nights etc) of the whole day(s), week(s) or month(s), are taken into account. 

In the current study, a deep learning approach, based on convolutional autoencoders, is explored in order to effectively correct the error of the simulation result in space and time domain, hence performing a result similar to statistical downscaling methods. The result can be used in NWP simulations as initial and boundary condition, thus, enabling more accurate longer prediction. The current study is an attempt to improve the global seasonal forecast (3-6 month ahead) data accuracy, for the Greece area, with a more reliable reanalysis dataset, that incorporates observations, satellite imaging etc and has better spatial resolution. Specifically the Meteo France Seasonal and the NCEP FNL data are utilized and they are publicly available from Copernicus platform and the National Oceanic and Atmospheric Administration (NOAA) accordingly. Moreover, external information is used as evidence transfer, concerning the time conditions (month, day, season) and the simulation characteristics (initialization of simulation). It is found that the convolutional autoencoders help improve the resolution of the seasonal data and successfully correct the error of NWP data for 6 month ahead forecasting. Interestingly, the season evidence yields the best results which indicates a seasonal  (winter, spring, summer, autumn) dependence of the performance.