Development of a Transferable Density-Functional Tight-Binding Model for Organic Molecules at the Water/Platinum Interface

A computationally
efficient and transferable approach
for modeling
reactions at metal/water interfaces could significantly accelerate
our understanding and ultimately the development of new catalytic
transformations, particularly in the context of the emerging field
of biomass conversion. Here, we present a parametrization of Pt–X (X = H, O, C) density-functional tight-binding
(DFTB) for addressing this need. We first constructed Pt–H,
Pt–O, and Pt–C repulsive potential splines. These pairwise
parameters were then augmented to include many-body interactions using
the Chebyshev Interaction Model for Efficient Simulation (ChIMES).
We compare the geometrical and energetic performances of both DFTB
and DFTB/ChIMES methods with DFT reference data across a variety of
organic molecules at the platinum surface from nanoparticles to single-crystal
surfaces. DFTB shows limited transferability between extended crystal
surfaces and small nanoparticles. This transferability is significantly
improved through the introduction of three-body interactions with
Pt in DFTB/ChIMES, which provides consistent results across various
systems, with reductions in the RMSD from around 30 kcal/mol in DFTB
to around 10 kcal/mol. We demonstrate the stability and reliability
of the obtained parameters by performing metadynamic simulations for
the adsorption of phenol on Pt(111). We observe that DFTB itself is
undersolvating the surface, leading to only one or two chemisorbed
water molecules in a c(4 × 6) unit cell. In contrast, DFTB/ChIMES
leads to a coverage of about 0.5 ML and successfully captures the
chemisorbed mode of phenol at both the solid/liquid and the solid/gas
interfaces. Furthermore, in agreement with experimental measurements,
the adsorption at the solid/liquid interface is significantly weaker
than that at the solid/gas interface. Furthermore, we highlight that
even with DFTB, where we can accumulate dynamics for more than 1 ns
for a given system, the simulations are not fully converged.