Quantum Monte Carlo Benchmarking of Molecular Adsorption on Graphene-Supported Single Pt Atom

The precise understanding of adsorption energetics and molecular geometry at catalytic sites is fundamental for advancing catalysis, particularly under the constraints of resource efficiency and environmental sustainability. This study benchmarks the performance of density functional theory (DFT) calculations against diffusion Monte Carlo (DMC) calculations for adsorption properties of small gas molecules relevant to CO oxidation – namely O$_2$, CO, CO$_2$, and atomic oxygen – on a single Pt atom supported by pristine graphene. Our findings reveal that DMC calculations provide a significantly different landscape of adsorption energetics compared to DFT results. Notably, DFT predicts different lowest-energy configurations and spin states, particularly for O$_2$, which suggests potential discrepancies in predicting the catalytic behavior. Furthermore, this study identifies the critical issue of CO poisoning, highlighted by the large disparity between the DMC adsorption energies of O$_2$ ($-1.23(2)$ eV) and CO ($-3.37(1)$ eV), which can inhibit the catalytic process. These results emphasize the necessity for more sophisticated computational approaches in catalysis research, aiming to refine the prediction accuracy of reaction mechanisms and to enhance the design of more effective catalysts.

💡 Research Summary

The authors present a systematic benchmark of density‑functional theory (DFT) against diffusion Monte Carlo (DMC) for the adsorption of four small molecules—O₂, CO, CO₂, and atomic O—on a single platinum atom anchored to pristine graphene. The study is motivated by the growing interest in graphene‑supported single‑atom catalysts, where accurate energetics are essential for predicting reaction pathways such as the Eley–Rideal (ER) versus Langmuir–Hinshelwood (LH) mechanisms in CO oxidation, and for assessing the risk of CO poisoning.

Methodologically, the work employs spin‑polarized DFT calculations using the PBE + U functional (U = 3.78 eV) to generate Kohn‑Sham orbitals, which serve as the Slater part of trial wave functions in fixed‑node DMC simulations performed with QMCPACK. A norm‑conserving relativistic pseudopotential describes the 18 valence electrons of Pt, while BFD pseudopotentials are used for C, O, and H. The authors carefully address finite‑size effects by using twist‑averaged boundary conditions (36 twists) and a 1/N extrapolation based on three supercell sizes (1 × 1 × 1, 2 × 2 × 1, 3 × 3 × 1). The time step is set to 0.005 Ha⁻¹, and three‑body Jastrow factors are optimized via variational Monte Carlo.

For O₂ adsorption, three geometries are examined: Z‑mode (molecular axis tilted relative to the graphene plane), A‑mode (axis parallel to the armchair direction), and V′‑mode (end‑on tilt). Both singlet and triplet spin states are considered. DMC finds the singlet Z‑mode to be the most stable with an adsorption energy of –1.23(2) eV; the triplet A‑mode and V′‑mode are less stable at –1.16(1) eV and –1.04(1) eV, respectively. In contrast, PBE predicts the triplet A‑mode as the lowest‑energy configuration and overestimates all adsorption energies by roughly 0.5–0.7 eV. Charge‑density‑difference analysis shows a net transfer of electrons from the Pt‑graphene substrate to O₂, with DFT underestimating the magnitude of this transfer.

CO adsorption is explored in four configurations. The tilted end‑on V′‑C mode yields the strongest binding, –3.37(1) eV, only marginally more favorable than the perfectly vertical end‑on geometry (–3.34(1) eV). Side‑on Z‑mode and a configuration where the carbon atom binds to Pt (V‑O mode) are substantially weaker (–0.82(1) eV and –1.09(2) eV). DFT functionals again over‑bind by 0.2–0.3 eV and underestimate the charge redistribution that accompanies CO binding.

For CO₂, DMC identifies a side‑on adsorption geometry as the most stable (–1.21(2) eV), while end‑on orientations are less favorable. Atomic oxygen prefers the triplet state by 0.5(1) eV, with a very strong adsorption energy of –4.04(1) eV. Using these values, the dissociation energy of O₂ on the Pt‑graphene complex is calculated to be 2.29(3) eV, substantially lower than the gas‑phase O–O bond energy, indicating that the Pt site efficiently activates O₂.

Across all species, DMC consistently yields weaker adsorption energies than PBE and, more importantly, predicts different lowest‑energy spin‑state/geometry combinations. The discrepancies are attributed to the inability of mean‑field DFT to capture many‑body correlation effects that are pronounced in Pt‑O and Pt‑C bonding. Inclusion of van‑der‑Waals corrections (D2, rVV10, etc.) does not materially improve DFT performance, confirming that electronic correlation, rather than dispersion, dominates the energetics.

The authors argue that these differences have practical implications for catalytic modeling. Since NEB calculations rely on accurate initial and final states, using DFT‑derived geometries could lead to erroneous reaction pathways and activation barriers. Moreover, the large gap between O₂ (≈ –1.2 eV) and CO (≈ –3.4 eV) adsorption energies signals a high propensity for CO poisoning of the Pt site, which would block O₂ activation and suppress the ER mechanism.

In conclusion, the paper demonstrates that diffusion Monte Carlo provides a reliable benchmark for adsorption energetics on low‑dimensional, transition‑metal single‑atom catalysts. It highlights the necessity of incorporating high‑level quantum‑chemical methods into the computational design workflow for graphene‑supported Pt catalysts, especially when precise energetics dictate mechanistic preferences and catalyst stability.