An evaluation of the performances of several known low-cost
methods
for the reproduction of structural features of differently sized Pt
nanoparticles (NPs) is presented. The full density functional theory
PBE-dDsC functional (within the plane-wave formalism) was employed
to benchmark the semiempirical tight-binding DFTB and GFNn-xTB (n = 0, 1, 2) and the reactive force-field
ReaxFF. Performances were evaluated by comparing several size-dependent
features (such as relative stabilities, structural descriptors, and
vibrational features) computed with the different methods. Various
structures (ordered and amorphous) and sizes (from Pt13 to Pt561) were considered in the datasets. ReaxFF molecular
dynamics (MD) was employed to achieve the amorphization of cuboctahedral
Pt147, Pt309, and Pt561 geometries,
which were subsequently optimized with both the low-cost methods
and the DFT reference, within a multilevel modeling approach. The
structures were further annealed with GFN0-xTB MD. While DFTB performs
quite well over all the selected structures, GFN2-xTB and the cheaper
GFN0-xTB show a general predilection for amorphous geometries. The
performances of GFN2-xTB are found to worsen with the increasing size
of the system, while ReaxFF and GFN0-xTB undergo the opposite trend.
We suggest that the semiempirical DFTB (and within certain limitations
GFN0-xTB and ReaxFF) could be suited for fast screening through amorphous
big-sized Pt NPs.