Comparing van der Waals Density Functionals for CO2 Adsorption in Metal Organic Frameworks

The accuracy of five recently proposed van der Waals (vdW) density functionals (optB86b, optB88, optPBE, revPBE, and rPW86), the semiempirical vdW method of Grimme (DFT-D2), and conventional local (LDA) and gradient-corrected (GGA-PBE) density functionals are assessed with respect to experimental enthalpies (Delta H) for CO2 adsorption in four prototypical metal organic frameworks (MOFs) containing coordinatively unsaturated metal sites: M/DOBDC (M = Mg, Ni, and Co) and Cu-HKUST-1. Although the LDA and GGA functionals partially capture trends, they significantly overbind (LDA) and underbind (GGA) CO2 with respect to the experimental enthalpies. The addition of a semiempirical r(-6) dispersion term to the GGA exchange-correlation energy using "off the shelf" DFT-D2 parameters results in a substantial improvement both in trends and in the magnitude of the adsorption enthalpies. However, on average this approach still underbinds CO2 as compared to the experimental data by similar to 7 kJ/mol (18%). Better accuracy is obtained with some of the nonempirical vdW density functionals, with the revPBE-based functional of Dion et al. [Phys. Rev. Lett. 2004, 92, 246401] yielding an average error of only similar to 2 kJ/mol (4%) relative to experiment. This improvement in energetics is accompanied by a slight decrease in the accuracy of predicted structures, as the revPBE functional overestimates the metal-CO2 bond length by about 10%. The identification of an efficient vdW density functional capable of predicting the thermodynamics of CO2 ;adsorption will facilitate rapid computational screening for optimal CO2 adsorbents.