Sensitivity Analysis of Cluster Models for Calculating Adsorption Energies for Organic Molecules on Mineral Surfaces

We calculated the adsorption energy for ethanol on the magnesite {10.4} surface using density functional theory (DFT) and cluster models for the mineral surface, and we quantified the errors introduced by using a finite size cluster, freezing various parts of the mineral cluster during geometry optimization and for altering the edges of the cluster. We also investigated how the adsorption energy changes for increasingly accurate density functionals, PBE, BLYP, B3LYP, and B2PLYP, also when supplemented with empirical dispersion (-D). We concluded that calculations with clusters large enough to include the surface atoms and groups binding to the adsorbate and their nearest-neighbor ions provide accurate adsorption energies, typically for MgCO3-ethanol systems, which is about 60 atoms. Using B3LYP-D and a finite size cluster of 80 atoms, we found that the adsorption energy was underestimated by 0.17 eV for adsorption from vacuum and by 0.10 eV for adsorption from solution, and we estimated a random error of 0.10 eV in adsorption energy when surface atoms were frozen during geometry optimization. The basis set superposition error leads to an overestimation of the adsorption energy of 0.08 eV, which in solution almost cancels the effects of finite size. We also compared adsorption energies for binding to magnesite clusters of ethanol, water, acetic acid, hydroxyl, and acetate from vacuum and from aqueous solution. When adsorbed from solution, using an implicit solvent model, bonding with water is stronger than for the other molecules. No adsorption of any of the other molecules was predicted. The adsorption energy for the acetate ion is considerably more uncertain than for the other molecules. We recommend B3LYP-D as a reliable and computationally effective method for calculating adsorption energies of organic molecules on mineral surfaces using cluster models.