[Paper Review] Ab initio surface chemistry with chemical accuracy

Hong‐Zhou Ye, Timothy C. Berkelbach|arXiv (Cornell University)|Sep 26, 2023

Electron and X-Ray Spectroscopy Techniques10 citations

TL;DR

The paper demonstrates solving the many-electron Schrödinger equation for molecules on surfaces at chemical accuracy using local correlation and periodic CCSD(T), applied to water on Al2O3 and TiO2 surfaces. It benchmarks against DFT and experiments to reveal accurate adsorption, dissociation energetics, and barriers.

ABSTRACT

First-principles calculations are a cornerstone of modern surface science and heterogeneous catalysis. However, accurate reaction energies and barrier heights are frequently inaccessible due to the approximations demanded by the large number of atoms. Here we combine developments in local correlation and periodic correlated wavefunction theory to solve the many-electron Schrödinger equation for molecules on surfaces with chemical accuracy, commonly defined as 1~kcal/mol. As a demonstration, we study water on the surface of \ce{Al2O3} and \ce{TiO2}, two prototypical and industrially important metal oxides for which we obtain converged energies at the level of coupled-cluster theory with single, double, and perturbative triple excitations [CCSD(T)], commonly known as the "gold-standard" in molecular quantum chemistry. We definitively resolve the energetics associated with water adsorption and dissociation, enabling us to address recent experiments and to analyze the errors of more commonly used approximate theories.

Motivation & Objective

• Motivate the need for chemically accurate energetics in surface chemistry beyond standard DFT.
• Develop and apply a local-correlation, periodic CCSD(T) framework to solids and large surface models.
• Converge adsorption, dissociation energies, and reaction barriers with respect to basis set, slab thickness, and surface size.
• Compare high-level results to common DFT functionals and experimental data to assess accuracy and limitations.

Proposed method

• Compute the Hartree-Fock energy in large periodic supercells under periodic boundary conditions.
• Express correlation energy as a sum over localized occupied orbitals using local natural orbitals (LNOs) with periodic boundary conditions.
• Converge CCSD(T) energies by expanding the LNO subspace until reaching chemical accuracy for adsorption/dissociation energetics.
• Use ZPE corrections from PBE to obtain vibrationally corrected energies.
• Benchmark against DFT (PBE+D3, SCAN+D3, PBE0+D3, MP2) and experimental data for water on Al2O3 and TiO2.
• Apply harmonic transition state theory to estimate reaction rates from CCSD(T) barriers, including zero-point and vibrational corrections.

Figure 1: (A) Atomic structure of a single water molecule adsorbed on the $\alpha$ -Al 2 O 3 (0001) surface. The molecularly adsorbed water molecule (left) may transfer a hydrogen to a neighboring surface oxygen via the transition state (middle), resulting in OH ads and OH surf fragments (right). (B

Experimental results

Research questions

• RQ1 What are the adsorption energies and dissociation barriers for water on α-Al2O3(0001) and rutile TiO2(110) surfaces at CCSD(T) accuracy?
• RQ2 How do CCSD(T) results compare to DFT functionals and MP2 for these surface reactions, and what does this imply for functional development?
• RQ3 To what extent can local correlation (LNO) CCSD(T) be converged with respect to basis set, slab thickness, and surface size in periodic solids?
• RQ4 What are the implications of high-accuracy barriers on predicted kinetics relative to experimental observations?
• RQ5 Do high-level results indicate irreversibility or competing mechanisms not captured by simple adsorption/dissociation pathways?

Key findings

• CCSD(T) predicts adsorption energies and dissociation barriers with chemical accuracy for water on Al2O3 and TiO2 surfaces.
• For Al2O3, CCSD(T) gives adsorption energy ≈ 26.5 kcal/mol and a dissociation barrier ≈ 6.6 kcal/mol after ZPE correction, with dissociation thermodynamically favorable by ~8 kcal/mol relative to molecular adsorption.
• For TiO2, CCSD(T) yields a ZPE-corrected adsorption energy of 20.8 kcal/mol and a dissociation barrier of 4.3 kcal/mol, with dissociative adsorption slightly favored by ~1.9 kcal/mol.
• PBE+D3 tends to overestimate adsorption energy and underestimate barriers relative to CCSD(T); MP2 aligns more closely with CCSD(T) for barriers and dissociation energies on TiO2.
• PBE+U destabilizes molecular adsorption and underestimates the barrier for TiO2 depending on U, showing competing improvements and deteriorations for band gaps and reaction energetics.
• The study highlights that discrepancies between experiment and theory at the DFT level are not due to CCSD(T) inaccuracies but may arise from more complex reaction networks or experimental conditions.

Figure 2: (A) Zero-point energy corrected reaction energetics associated with the water adsorption and dissociation reaction calculated using different electronic structure methods. (B) Inverse rate for the water dissociation reaction in the temperature range of 80 K to 300 K evaluated using harmoni