Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Discovery of Novel Reticular Materials for Carbon Dioxide Capture using GFlowNets

Flaviu Cipcigan, Jonathan Booth, Rodrigo Neumann Barros Ferreira, Carine Ribeiro dos Santos, Mathias Steiner

TL;DR

A new Python package (matgfn) is introduced to generate the matgfn-rm dataset of novel and diverse reticular materials with gravimetric surface area above 5000 m$^2$/g, and 15 materials outperforming all materials in CoRE2019 are discovered.

Abstract

Artificial intelligence holds promise to improve materials discovery. GFlowNets are an emerging deep learning algorithm with many applications in AI-assisted discovery. By using GFlowNets, we generate porous reticular materials, such as metal organic frameworks and covalent organic frameworks, for applications in carbon dioxide capture. We introduce a new Python package (matgfn) to train and sample GFlowNets. We use matgfn to generate the matgfn-rm dataset of novel and diverse reticular materials with gravimetric surface area above 5000 m$^2$/g. We calculate single- and two-component gas adsorption isotherms for the top-100 candidates in matgfn-rm. These candidates are novel compared to the state-of-art ARC-MOF dataset and rank in the 90th percentile in terms of working capacity compared to the CoRE2019 dataset. We discover 15 materials outperforming all materials in CoRE2019.

Discovery of Novel Reticular Materials for Carbon Dioxide Capture using GFlowNets

TL;DR

A new Python package (matgfn) is introduced to generate the matgfn-rm dataset of novel and diverse reticular materials with gravimetric surface area above 5000 m/g, and 15 materials outperforming all materials in CoRE2019 are discovered.

Abstract

Artificial intelligence holds promise to improve materials discovery. GFlowNets are an emerging deep learning algorithm with many applications in AI-assisted discovery. By using GFlowNets, we generate porous reticular materials, such as metal organic frameworks and covalent organic frameworks, for applications in carbon dioxide capture. We introduce a new Python package (matgfn) to train and sample GFlowNets. We use matgfn to generate the matgfn-rm dataset of novel and diverse reticular materials with gravimetric surface area above 5000 m/g. We calculate single- and two-component gas adsorption isotherms for the top-100 candidates in matgfn-rm. These candidates are novel compared to the state-of-art ARC-MOF dataset and rank in the 90th percentile in terms of working capacity compared to the CoRE2019 dataset. We discover 15 materials outperforming all materials in CoRE2019.
Paper Structure (7 sections, 1 equation, 17 figures)

This paper contains 7 sections, 1 equation, 17 figures.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Generating reticular frameworks with GFlowNets
  4. The matgfn-rm dataset
  5. Conclusion
  6. Contributions
  7. Acknowledgements

Figures (17)

  • Figure 1: Regression of simulated high pressure CO2 uptake to gravimetric surface area
  • Figure 2: Two dimensional T-SNE embedding of the average minimum distance of ARC-MOF (green), the top-100,000 (gray) and top-100 (orange) materials from matgfn-rm.
  • Figure 3: Simulated CO2 working capacity and CO2 / N2 selectivity for the top-100 matgfn-rm materials. The (red) dashed line represents the highest working capacity found in the CoRE2019 dataset, which is surpassed by 15 of the top-100 matgfn-rm materials.
  • Figure 4: A render of the relaxed structure of 005-ffc-10217, the highest performing structure in the matgfn-rm dataset.
  • Figure 5: trajectory balance losses for training a GFlowNet on the ASC topology without edges. Losses are smoothed with a 1,000 episode window moving average due to the discovery of a high performing MOF causing a one-episode long spike in the loss.
  • ...and 12 more figures