Yusuf Shaidu
(University of California & Lawrence Berkeley National Laboratory)

The Mg2(dobpdc) is a metal organic framework (MOF) made up of Mg metal ions connected via a dobpdc4- (dobpdc4-= 4,4'-dihydroxy-(1,1'-biphenyl)-3,3'-dicarboxylic acid) organic molecule to form a hexagonal porous material along the c crystallographic axis. The diamine-functionalized variant of Mg2(dobpdc) has been demonstrated as a transformative material for carbon capture applications. The functionalized materials form carbamate species upon CO2 insertion and exhibit cooperative adsorption mechanisms leading to a step-shaped isotherm that enables a full CO2 capacity to be accessed with a minimal temperature swing^[1]. Thermal properties, CO2 adsorption kinetics and mechanisms of CO2 insertion are understudied due to the complexity of the materials. Finite temperature simulations of these materials are prohibitively costly because of the large number of atoms in a unit cell. Previous studies have been based on empirical force-fields whose parameters are not optimized to reproduce the density functional theory (DFT) energies and forces of these systems. Here, we present our work on the development of a reactive interatomic potential based on neural network techniques for amine-appended Mg2(dobpdc) systems. The neural network potentials (NNPs) belong to a class of reactive force-fields that allow to simulate bond breaking and formation which are prevalent during a finite temperature dynamics. The interatomic potentials here are constructed using the active learning approach combining the artificial neural network approaches and dispersion-corrected DFT. We show that the NNPs are accurate in predicting adsorption energy, mechanical, vibrational and thermal properties. In addition, we demonstrate that these potentials can be combined with a simulated annealing approach to find better starting structures for DFT energy minimization needed to compute CO2 binding energy.
 

Zoom registration link will follow in due time.