196
Y.-W. Huang et al.
CO adsorption on Pt(111) surface at low coverage [34]. However, theoretically the
density functional theory calculation based on LDA and GGA failed to predict the
correct adsorption site for CO on Pt(111) [9]. Up to now, it is still hard to find a
suitable and complete DFT functional to solve this puzzle. In this study, we would
like to evaluate the performance of computational schemes including 21 different
density functionals, HF and MP2 for CO adsorption on Pt(111) cluster model. The
chemical bonding analyses were also adopted to investigate the effect of DFT functionals to the adsorption behavior of CO on Pt nanocluster.
Investigations of CO on Pt(111) by using density functional theory has been
long back to early 90’s. Philipsen et al. applied the BP86 [48] functional to calculate CO adsorption on Pt(111) surface showing that the adsorption energies
are 0.83 and 1.02 eV for CO adsorption on top site and hollow site [43]. Lynch
et al. applied the PW91 [49] functional with ultrasoft pseudopotentials to calculate the CO/Pt(111) obtaining the adsorption energies of 1.87 and 2.00 eV for
CO adsorption at top and fcc sites, respectively [25]. Both results did not agree
with the experimental observation. In order to check this problem of site preference, Feibelman et al. considered different DFT functionals including LDA [50],
PW91, PBE [51] and RPBE [52], and different numerical condition to calculate
the CO on Pt(111) surface [9]. However, all LDA and GGA functionals predicted
the wrong site preferences and then coined this interesting problem, “CO/Pt(111)
puzzle”. Since then this “puzzle” has been addressed and studied further by several
groups [7, 14, 16, 19, 20, 23, 26, 30, 31, 37, 44].
Grinberg et al. carefully studied the influence of pseudopotentials and DFT functional on the CO binding energy and site preference [16]. Their results show the
wrong prediction of site preference was due to the poor treatment of CO bonding
for LDA and GGA functional. Gil et al. applied several different GGA and hybridGGA B3LYP [53] to calculate CO adsorption on Pt clusters [14]. The results show
that the inaccuracy in site preference may come from the underestimation of unoccupied CO 2π ∗ orbital by DFT-GGA, which leads to the unrealistic strengthening
of the 2π ∗ -d bonding interaction. Kresse and co-workers also report that a linear
relationship between the CO LUMO energy and the top-hollow adsorption energy
difference by using slab model [23]. Mason et al. also found out the relationship
between singlet-triplet excitation and adsorption energy [26]. These results all implied that the CO/Pt(111) puzzle may be attributed to the incorrect underestimation
of CO LUMO energy in the conventional LDA or GGA DFT schemes. Therefore,
the modification of the CO LUMO energy by improving the DFT functional may be
a feasible way to solve this puzzle.
On the other hand, Gil et al. applied the B3LYP and cluster model to study the
CO adsorption on Pt(111) and also compared their results with other LDA/GGA
functional and slab model [14]. They found that both slab and cluster models, irrespective of the functionals used, always favor CO adsorption at hollow site than at
top site. However, the B3LYP functional can largely reduce the adsorption energy
difference between the top site and hollow site. The predicted adsorption energies
under the B3LYP and Pt 12−6 cluster model are 1.42 and 1.49 eV for CO adsorption
at top and fcc sites. They also found that the CO LUMO energy increased when
Y.-W. Huang et al.
CO adsorption on Pt(111) surface at low coverage [34]. However, theoretically the
density functional theory calculation based on LDA and GGA failed to predict the
correct adsorption site for CO on Pt(111) [9]. Up to now, it is still hard to find a
suitable and complete DFT functional to solve this puzzle. In this study, we would
like to evaluate the performance of computational schemes including 21 different
density functionals, HF and MP2 for CO adsorption on Pt(111) cluster model. The
chemical bonding analyses were also adopted to investigate the effect of DFT functionals to the adsorption behavior of CO on Pt nanocluster.
Investigations of CO on Pt(111) by using density functional theory has been
long back to early 90’s. Philipsen et al. applied the BP86 [48] functional to calculate CO adsorption on Pt(111) surface showing that the adsorption energies
are 0.83 and 1.02 eV for CO adsorption on top site and hollow site [43]. Lynch
et al. applied the PW91 [49] functional with ultrasoft pseudopotentials to calculate the CO/Pt(111) obtaining the adsorption energies of 1.87 and 2.00 eV for
CO adsorption at top and fcc sites, respectively [25]. Both results did not agree
with the experimental observation. In order to check this problem of site preference, Feibelman et al. considered different DFT functionals including LDA [50],
PW91, PBE [51] and RPBE [52], and different numerical condition to calculate
the CO on Pt(111) surface [9]. However, all LDA and GGA functionals predicted
the wrong site preferences and then coined this interesting problem, “CO/Pt(111)
puzzle”. Since then this “puzzle” has been addressed and studied further by several
groups [7, 14, 16, 19, 20, 23, 26, 30, 31, 37, 44].
Grinberg et al. carefully studied the influence of pseudopotentials and DFT functional on the CO binding energy and site preference [16]. Their results show the
wrong prediction of site preference was due to the poor treatment of CO bonding
for LDA and GGA functional. Gil et al. applied several different GGA and hybridGGA B3LYP [53] to calculate CO adsorption on Pt clusters [14]. The results show
that the inaccuracy in site preference may come from the underestimation of unoccupied CO 2π ∗ orbital by DFT-GGA, which leads to the unrealistic strengthening
of the 2π ∗ -d bonding interaction. Kresse and co-workers also report that a linear
relationship between the CO LUMO energy and the top-hollow adsorption energy
difference by using slab model [23]. Mason et al. also found out the relationship
between singlet-triplet excitation and adsorption energy [26]. These results all implied that the CO/Pt(111) puzzle may be attributed to the incorrect underestimation
of CO LUMO energy in the conventional LDA or GGA DFT schemes. Therefore,
the modification of the CO LUMO energy by improving the DFT functional may be
a feasible way to solve this puzzle.
On the other hand, Gil et al. applied the B3LYP and cluster model to study the
CO adsorption on Pt(111) and also compared their results with other LDA/GGA
functional and slab model [14]. They found that both slab and cluster models, irrespective of the functionals used, always favor CO adsorption at hollow site than at
top site. However, the B3LYP functional can largely reduce the adsorption energy
difference between the top site and hollow site. The predicted adsorption energies
under the B3LYP and Pt 12−6 cluster model are 1.42 and 1.49 eV for CO adsorption
at top and fcc sites. They also found that the CO LUMO energy increased when