Fragment-based embedding of correlated wavefunction methods for CO adsorption on TiO2(110)

Abstract

Obtaining quantitatively reliable adsorption energies for molecules on oxide surfaces remains challenging, because the binding involves a delicate balance of local chemical interactions and long-range correlation effects that are expensive to capture with canonical correlated wavefunction methods in periodic systems. In this thesis, CO adsorption on rutile TiO2(110) at low coverage is studied with a fragment-based embedding strategy that combines periodic random-phase approximation (RPA) with high-level CCSD(T)calculations on localized fragments. A periodic slab model is employed and adsorption energies are defined from bound and unbound configurations in the same simulation cell. Starting from periodic DFT and subsequent HF and RPA calculations, occupied orbitals are localized to enable the construction of correlated fragments centered on the adsorption site. Several fragment selection schemes are investigated—geometric (GEO)and charge-difference based (CHG, CH2), including a hole-filling variant (MIX)—and fragment-size convergence is analyzed as a function of the number of occupied orbitals in the fragment. Periodic CCSD(T) adsorption energies are approximated by an RPA based embedding correction that combines periodic RPA with fragment CCSD(T) and fragment RPA; for relaxed geometries, a scaled correction is additionally used to improve robustness with respect to fragment size. For the (4×2) four-layer surface model, the scaled CCSD(T):RPA embedding approach converges consistently across fragment definitions to an electronic adsorption energy of approximately −0.51 to −0.52 eV, with a residual uncertainty of about ±0.02 eV. Finite size effects associated with slab thickness and lateral cell size were quantified separately and yield a combined correction of approximately −0.21 eV. Applying this correction results in an electronic adsorption energy of approximately −0.72 eV. Including a zero point vibrational correction of +0.04 eV gives a 0 K adsorption energy of ≈ −0.68 eV. The sizable finite-size contribution highlights the sensitivity of correlated adsorption energies to the underlying periodic model and underscores the importance of systematic convergence tests for quantitative predictions. The resulting adsorption energy is noticeably stronger than values typically inferred from temperature-programmed desorption experiments, pointing to an interesting discrepancy between high-level theory and experiment.