Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Proc. R. Soc. A
doi:10.1098/rspa.2010.0561
Published online
The decomposition of H2O2 over the
components of Au/TiO2 catalysts
BY ADAM THETFORD, GRAHAM J. HUTCHINGS, STUART H. TAYLOR
AND DAVID J. WILLOCK*
Cardiff Catalysis Institute, School of Chemistry, Cardiff University,
Main Building, Park Place, Cardiff CF10 3AT, UK
Hydrogen peroxide is an important oxidant that is increasingly being employed in
selective oxidation reactions over support metal catalysts. We present a density functional
theory study of the adsorption of H2 O2 to the components of a model Au/TiO2 system
based on Au10 nanoclusters and the rutile TiO2 (110) surface. We find that H2 O2
decomposes easily to 2OH on the metal nanoparticles while the interaction with surface
hydroxyls on TiO2 (110) gives a low barrier to a surface OOH species. This work suggests
that the production of H2 O2 takes place at the interface between the particle and
oxide and we further show how this interface region is influenced by the hydroxylation
of the surface.
Keywords: heterogeneous catalysis; hydrogen peroxide; nanoparticles; density functional theory
1. Introduction
Hydrogen peroxide is a widely used chemical with major long-standing
applications in cleaning and bleaching. More recently, H2 O2 is increasingly
being employed as the oxygen source in selective partial oxidation, for example
the production of propylene oxide from propene over titanium silicate (TS-1)
catalysts. For this reaction, calculations by To et al. (2008) have shown that
H2 O2 reacts to give a surface hydroperoxy species at the framework Ti site by
H–OOH bond activation on the catalyst surface.
An important aspect of any process generating hydrogen peroxide is the control
of its decomposition, which clearly leads to loss of efficiency. Hydrogen peroxide
is currently produced via the sequential hydrogenation and oxidation of an alkyl
anthraqinone. By cycling hydrogenation and oxidation stages, the mixing of H2
and O2 at compositions near to their explosive limit can be avoided and the
selectivity to H2 O2 from hydrogen is practically 100 per cent. However, this
process is only economic on a large scale, requiring a distribution network for
this highly corrosive reagent to be delivered to end users.
*Author for correspondence ([email protected]).
One contribution to a Special feature ‘High-performance computing in the chemistry and physics
of materials’.
Received 29 October 2010
Accepted 17 January 2011
1
This journal is © 2011 The Royal Society
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
2
A. Thetford et al.
The in situ production of hydrogen peroxide from O2 and H2 promises an
alternative, localized, small-scale production method with lower environmental
and cost implications. To be competitive, loss of H2 O2 through decomposition
must be minimized. Recently, a promising new catalyst based on Au and Au/Pd
alloys supported on oxides such as TiO2 has shown impressive activity for the
direct synthesis reaction using mixtures of water and methanol as solvent (Landon
et al. 2002; Okumura et al. 2003; Edwards et al. 2009; Piccinini et al. 2010).
These catalysts can also be active for oxidation reactions such as the
epoxidation of propene either with Au/TiO2 or Au/TS-1 (Corma & Garcia 2008).
In these examples, hydrogen peroxide is consumed as it is produced. However,
one drawback is the atom efficiency with regard to H2 . In oxidation carried out
with these catalysts this is usually only around 80 per cent. Hydrogen is the
expensive reagent and so higher selectivities are desirable to compete with the
current process. Careful catalyst preparation, involving acid pre-treatments, can
give H2 efficiencies as high as 95 per cent (Edwards et al. 2009) and part of the
aim of this paper is to use theoretical methods to consider the role of the acidic
environment on the reaction.
The decomposition of H2 O2 to give water is an important reaction pathway
that leads to lower utilization of H2 . In this contribution, we consider the
decomposition pathways for H2 O2 on the components of Au/TiO2 catalysts using
density functional theory (DFT) calculations. Decomposition of H2 O2 via scission
of the HO–OH bond results in hydroxyl radical species that ultimately lead to
H2 O as a side product, reducing the efficiency of H2 O2 production. In the first
set of results presented below, we find that this process readily occurs on Au
particles. We then discuss the TiO2 (110) rutile surface as a model of the support
material and its surface structure in the presence of water, the common solvent
for hydrogen peroxide chemistry. Here, we find that the alternative H–OOH
cleavage, which preserves the peroxy species, takes place. We also discuss the
importance of water and acidity on the structure of the support surface and the
metal particles upon it.
2. Calculation method
All calculations for isolated Au particles, the TiO2 (110) surface and
Au/TiO2 (110) were carried out using the VASP code from Kresse & Hafner (1993,
1994) with optimization and molecular dynamics (MD) runs at the Perdew–
Burke–Ernzerhof (PBE; Perdew et al. 1996) functional level along with a 400 eV
planewave cut-off to define the basis set. To give an estimate of adsorption or
reaction energies, further optimization using the revision of PBE for surface
adsorbate calculations, RPBE (Hammer et al. 1999), was carried out. The
Monkhorst–Pack sampling scheme (Monkhorst & Pack 1976) was applied for
k-point grids. The bulk unit cell volume was optimized with a k-point grid of
9 × 9 × 9. The optimal cell volume was calculated by plotting the cell energy
versus an expansion factor (between −4 and +6% of the experimental cell volume)
and locating the minimum in the resulting quadratic plot. This procedure gave
cell dimensions of a = b = 4.633 Å and c = 3.004 Å close to the experimental values
of a = b = 4.594 Å and c = 2.959 Å (Abrahams & Bernstien 1971).
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Decomposition of H2 O2 over Au and TiO2
(a)
(111)
(100)
(b) O
2c
3
(111)
O3c
Ti5c
Ti6c
Figure 1. (a) An Au10 cluster from the face-centred cubic crystal structure with the faces marked
according to the bulk plane Miller indices. (b) A p(4 × 2) simulation slab of the rutile TiO2 (110)
surface with the distinct surface Ti and O positions marked according to their coordination. (Online
version in colour.)
For surface calculations, slabs were cut from the optimized bulk and expanded
to give supercells of suitably large area for which k-point sampling at the 3 × 3 × 1
level was sufficient. For the hydroxylation of the TiO2 (110) surface p(3 × 2), we
used cells with dimensions 9.0111 × 13.1898 × 24.1623 Å, while for all models
of supported Au10 clusters p(4 × 2) cells with dimensions 12.0148 × 13.1898 ×
24.1623 Å were employed. In each case, the slab consisted of three Ti layers with
a 15 Å vacuum gap to expose the surface. More details on the selection of the
slab dimensions are given in the relevant result section.
For calculations on isolated Au10 clusters, an initial FCC arrangement of the
atoms to form a Au(7,3) structure with (111) and (100) facets was employed, as
illustrated in figure 1a. The seven-atom layer of the Au(7,3) cluster was aligned
with the ab plane of the simulation cell, and in this layer each Au atom had the
component of its coordinate in the c-direction frozen to simulate interaction with
the support. A simulation cell equivalent to a p(6 × 3) TiO2 surface supercell was
employed, calculations were carried out with only one k-point and the energies
quoted are based on the RPBE functional. Clusters of this size in isolation
have low-energy planar structures (Xiao et al. 2006); however, the interaction
with the support can alter the ordering to favour two-layer structures (Boronat
et al. 2009) and high resolution electron microscopy also suggests that particles
on this scale form two-layer structures in Au supported on oxide catalysts
(Herzing et al. 2008).
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
4
A. Thetford et al.
Adsorption energies per adsorbate molecule, Eads , were calculated using
the equation
E(slab + nA) − (E(slab) + nE(A))
,
(2.1)
Eads =
n
with E(slab + nA) the energy of the optimized surface slab with n adsorbates,
E(slab) the energy of the slab optimized without adsorbate and E(A) the energy
of the free adsorbates optimized in the gas phase. To arrive at Eads , all three
calculations are carried out using the same periodic box, with the same planewave
cut-off and the same set of fixed coordinates defined for the slab with and
without adsorbates. Using this definition, negative adsorption energies indicate
that adsorption is favourable.
3. H2 O2 dissociation over isolated Au10 clusters
Figure 2 shows the molecular orbitals and related energy levels in the immediate
vicinity of the highest occupied (HOMO) and lowest unoccupied molecular orbital
(LUMO) energy levels for H2 O2 calculated at the B3LYP level with a 6-31G(d,p)
basis set using the Gaussian03 package (Frisch et al. 2004). The LUMO level
consists of an anti-bonding interaction for the HO–OH bond so that electron
donation into this orbital would be expected to weaken the bond between the
two oxygen atoms.
We initially considered the adsorption of H2 O2 to the isolated Au10 cluster by
placing the molecule at different locations on the cluster surface and carrying
out geometry optimization. The resulting geometries and adsorption energies are
shown in figure 3a. Placing the molecule on the three atom (111) facet gave a
very low adsorption energy (6 kJ mol−1 ), with the molecule moving away from the
cluster. For all other positions starting from molecular H2 O2 , bond scission occurs
on geometry optimization to produce two OH groups. On the edge of the cluster
with one OH bonded to an Au atom in the three layer and the other to an atom in
the base layer an adsorption energy of −209 kJ mol−1 is found. If both OH groups
are bonded to base layer Au atoms, lower adsorption energies are calculated:
−228 kJ mol−1 for the Au(100) facet and −237 kJ mol−1 for the Au(111) facet.
From the molecular orbitals shown in figure 2, these structures are expected to
result from electron donation into the LUMO state of the molecule from the Au
atoms. Bader analysis shows that the overall charge on the OH groups is negative
(−0.37 e and −0.46 e). The spin density for a representative structure is given in
figure 3b. The spin density on the O atoms shows that this is the main location
of the donated charge density. The Au10 cluster alone has an even number of
electrons with no spin polarization. On adsorption and bond cleavage of H2 O2 ,
the Au atoms bonded to the adsorbed OH species also show spin polarization
owing to partial transfer of an electron to the OH groups. A similar donation
effect is observed for adsorption of molecular O2 adsorbed to the base atoms
on Au clusters, which results in a superoxo (O−
2 ) adsorbate (Yoon et al. 2003).
In that work, variation in oxygen adsorption energy was seen with cluster size,
particularly for anionic Au clusters, which show differences of around 50 kJ mol−1
between odd- and even-numbered clusters. However, the variation with cluster
size for neutral clusters of the type studied here was less marked.
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Decomposition of H2 O2 over Au and TiO2
5
orbital energy (eV)
0.33
–1.63
–6.37
LUMO
HOMO
–7.67
Figure 2. The HOMO and LUMO energy levels for H2 O2 calculated at the B3LYP/6-31G(d,p)
level. (Online version in colour.)
(a)
29
11
–16
6
–209
–228
–237
79
(b)
Figure 3. (a) Comparison of structures and adsorption energies for H2 O2 on isolated Au10 .
(b) Calculated spin density for the lowest energy structure from (a) plotted at an isosurface value
of 4 × 10−4 e Å−3 . Atom colours follow: Au, yellow; O, red and H, white. (Online version in colour.)
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
6
A. Thetford et al.
The remaining structures shown in figure 3a are products of HOO–H bond
cleavage that were set up to test this alternative. However, all give positive
adsorption energies relative to gas phase H2 O2 .
For comparison with these results, we have also considered the adsorption
of H2 O2 and 2OH to the surfaces of the bulk metal. The adsorption of 2OH to
Au(111) using a five-layer periodic slab model has a lower adsorption energy than
does the adsorption of a molecular species (−143 cf. −66 kJ mol−1 ), although
bond cleavage does not occur on optimization. For the stepped (625) surface,
the adsorption energy for 2OH (−204 kJ mol−1 ) is closer to the values found for
Au10 clusters.
In summary, these calculations show that hydrogen peroxide decomposition
over Au10 clusters via HO–OH bond scission occurs without a barrier and leads
to adsorption states some 200 kJ mol−1 lower in energy than the alternatives,
including those formed by breaking of the H–OOH bond. For the bulk
surfaces, similar stabilization of 2OH groups is found for surfaces containing low
coordination atoms. The metallic component of the catalyst can clearly facilitate
loss of H2 O2 either during its production or use in oxidation reactions. We next
consider the role of the titania support material in the reaction chemistry of
hydrogen peroxide.
4. TiO2 (110) surface
As a model of the oxide support for the Au/TiO2 catalyst, we consider the rutile
TiO2 (110) surface. In the bulk structure all Ti atoms are six coordinate with four
equivalent equatorial Ti–O bonds (1.97 Å (PBE)) slightly shorter than the two
axials (2.00 Å (PBE)). The preferred termination of the (110) surface is shown
in figure 1b, the surface presents two distinct cation and two distinct anion sites.
Sixfold titanium (Ti6c ) cations are joined by edge sharing of their equatorial
oxygen neighbours, one half of which form a row of surface bridging oxygen atoms
(O2c ). The other half is below the surface plane and also axially coordinated to
Ti centres in the second cation layer. In the surface, there are also fivefold (Ti5c )
coordinate surface cations that have each lost an equatorial oxygen neighbour.
Two Ti5c and one Ti6c cations are linked by a three coordinate surface anion (O3c )
that is equatorial for the Ti5c and axial to the Ti6c sites. The TiO2 (110) cleavage
plane shown in figure 1b can be formed by only breaking the longer axial Ti–O
bonds of the bulk, leading to a low surface energy.
The line of bridging oxygen atoms on the bulk termination is mirrored
below the surface plane and so a neutral stacking unit consists of an atomic
tri-layer. For the simulation of the (110) surface, the choice of the number of
these tri-layers in the slab is important. Bredow et al. (2004) have pointed
out that slabs with an odd number of tri-layers have a symmetry plane in the
middle of the slab which is absent for even numbers of stacking units. They
found that relaxation of even-layered slabs leads to a rumpling of the surface,
whereas odd-layered slabs remain close to the bulk termination structure. They
showed that the interaction of the coordinatively unsaturated Ti5c surface atoms
with their axial oxygen neighbours in the second layer was stronger than the
corresponding interaction between tri-layers in the bulk. For the even slabs,
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Decomposition of H2 O2 over Au and TiO2
7
this leads to a tendency to form sets of paired tri-layers, which give rise to the
observed rumpling, whereas for odd slabs a more regular spacing of the tri-layers
is achieved since the layer in the centre of the slab would be left isolated by
such pairing of layers from the two surfaces. Thompson & Lewis (2006) have
demonstrated that PW91 calculations on odd-layered slabs give good agreement
with the available experimental data on the surface structure of TiO2 (110) from
LEED-IV experiments (Lindsay et al. 2005).
For the calculations presented here, a large surface area slab was required to
accommodate a Au10 cluster with good separation between periodic images and
so three-layer slabs have been employed. The calculated surface energy for the
three-layer slab is 0.88 J m−2 and tests with thicker slabs (from five to 13 layers)
showed variations within 9 per cent. This surface energy value is also in good
agreement with literature estimates (Swamy et al. 2002).
5. Surface hydroxylation
The exact nature of the TiO2 surface in the presence of Au and the aqueous
environment for the H2 O2 production experiment is difficult to determine. Perron
et al. (2007) have used a combination of X-ray photoelectron spectroscopy
and periodic DFT calculations to show that water adsorption results in some
dissociation to give surface hydroxyl groups. Their calculations suggest that
the optimal surface coverage consists of two dissociated and four molecular
waters on a p(2 × 3) surface with all Ti5c sites either hydroxylated or capped by
molecular water. Arrouvel et al. (2004) have also found partial hydroxylation to be
thermodynamically favourable for the surfaces of the anatase phase of TiO2 . This
conclusion is based on surface energy calculations for alternative combinations of
molecular and dissociated water, which can be used to find the preferred relative
concentrations by balancing the chemical potential of an isolated water molecule.
However, for the production of hydrogen peroxide over Au/TiO2 , it has been
demonstrated that the activity of the catalyst can be improved by an acid prewash of the support prior to metal deposition (Edwards et al. 2009). The hydrogen
peroxide generation reaction is also carried out in a mixed water/methanol solvent
in an autoclave using CO2 as a diluent to give typical gas mixtures of 5 per cent
H2 /CO2 and 25 per cent O2 /CO2 , suggesting that the reaction mixture too will be
at a low pH. The TiO2 isoelectric point is at pH 6 and so it is likely that the surface
will become more heavily hydroxylated under reaction conditions than would
be expected from surface science experiments under vacuum or thermodynamic
arguments based on a molecular water reference (Moreau & Bond 2007).
To examine the influence of surface hydroxyl groups on the adsorption of
H2 O2 on the catalyst support and on the structure of the Au10 cluster, we first
generated a series of surfaces by considering the dissociation of water at Ti5c sites
by proton transfer to a neighbouring O2c atom. Dissociating one water molecule
in this way on a p(3 × 2) surface gives the structures shown in figure 4a. These
energy minima have one of the OH groups orientated towards the oxygen atom
of the other hydroxyl group. This can occur for either the bridging oxygen OH
group (O2c –H . . . O(H)Ti5c ) or for the OH group capping the Ti5c site (Ti5c O–
H . . . O2c (H)). While these orientations are preferred, the distances involved are
longer than usually considered for formal hydrogen bonding interactions with
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
8
A. Thetford et al.
(a)
(b)
–78
–99
(c)
–57
Figure 4. Hydroxylated surfaces of TiO2 (110) and calculated adsorption energies (kJ mol−1 )
relative to the clean p(3 × 2) surface and H2 O(g). Hydroxylation from (a) one water molecule,
(b) two water molecules and (c) six water molecules, which constitutes a complete hydroxylation.
Atom colours follow: Ti, grey; O, red and H, white. (Online version in colour.)
H . . . O separations of 2.65 and 2.94 Å. The calculated adsorption energies are
within 4 kJ mol−1 of one another, indicating that these interactions between
neighbouring hydroxyl species are quite weak.
For two dissociated water molecules (figure 4b), there is also the possibility of
a Ti5c O–H . . . O(H)Ti5c interaction. This gives a much closer H . . . O separation
than seen for the dissociation of a single water molecule, with an interaction
distance of 1.70 Å. These interactions also have a significant influence on the
calculated adsorption energy: allowing only (Ti5c O–H . . . O2c (H)) interactions
gives a value of −78 kJ mol−1 compared with −99 kJ mol−1 when the geometry
includes a Ti5c O–H . . . O(H)Ti5c motif. Adsorption of two molecules of H2 O at
adjacent Ti5c sites is also stabilized by hydrogen bonding, and the calculated
adsorption energy per water molecule (−103 kJ mol−1 ) is slightly higher in
magnitude than the dissociated states. This importance of hydrogen bonding
in the preferred arrangement of surface hydroxyls and adsorbed water is in line
with the work of Perron et al. (2007).
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
9
Decomposition of H2 O2 over Au and TiO2
7
relative potential energy (eV)
6
5
4
3
2
1
0
50
100
150
200
simulation time (fs)
250
300
Figure 5. The relative potential energy as a function of time for molecular dynamics (MD)
simulations of the fully hydroxylated surface of TiO2 (110). The inset structures are snapshots
at 0, 75 and 300 fs as indicated by the arrows. Atom colours follow: Ti, grey; O, red and H, white.
(Online version in colour.)
Complete coverage of all surface Ti5c sites following this method requires the
dissociation of six water molecules per p(3 × 2) surface to give the structure shown
in figure 4c. The hydrogen bonding pattern was chosen to be quite regular so
that each OH group over a Ti5c site acts as a hydrogen bond donor and acceptor
with neighbouring OH(Ti5c ) groups to maximize the number of short hydrogen
bonds on the surface. The hydroxyl groups formed on protonation of bridging O2c
atoms are also orientated towards the oxygen atoms of neighbouring OH(Ti5c ).
The calculated adsorption energy per water molecule for this structure is less
than found at the lower coverages at −57 kJ mol−1 .
To investigate the stability of the fully hydroxylated surface, we have also
performed short MD simulations starting from the relaxed fully hydroxylated
surface. The goal of this MD simulation was to assess whether any low-energy
barriers exist for the re-arrangement of the hydroxyl groups between the various
patterns of surface hydrogen bonds that can be envisaged or for the conversion
of hydroxyl groups to molecular water. Born–Oppenheimer (BO) MD with the
microcanonical ensemble (NVE) was carried out for a total simulation time of
300 fs using a 1 fs time step. The initial temperature was set to 500 K but over
the first 50 fs of the simulation this dropped to 270 K and then remained close
to this value (268 ± 15 K) for the remainder of the simulation. The potential
energy of the system as a function of time is plotted in figure 5. After an initial
jump of about 5 eV as kinetic energy is introduced into the system, the potential
energy is centred on a constant value for the duration of the simulation. Figure 5
also has inserts showing the structure of the surface at the start, after 75 fs
and at the end of the MD run. At 75 fs, a transfer process is observed (bottom
left of inset) in which the hydrogen atom of an O2c H group moves close to the
neighbouring oxygen atom of an OH(Ti5c ). This results in the formation of a water
molecule adsorbed at the Ti5c site, which can be seen in the final frame. Under the
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
10
A. Thetford et al.
experimental conditions of low pH, we expect the hydroxyl surface coverage to
be higher than expected from our adsorption energy calculations or from earlier
work assuming a surface exposed to gas phase H2 O or neutral solutions. The MD
run shows that these hydroxl groups can react to reform water on the time scale
of a few tens of femtoseconds so that the dynamic exchange with solution phase
molecules will be rapid.
6. Adsorption of H2 O2 over the hydroxylated TiO2 (110) surface
Under reaction conditions, we have suggested that the TiO2 (110) surface will
be hydroxylated and shown that this layer can interchange between hydroxyl
and water forms on a short time scale. The structure from the end of the MD
run in figure 5 was re-optimized to give a water molecule at a Ti5c site with all
other surface sites hydroxylated. The calculated adsorption energy for this water
molecule referenced to the partially hydroxylated surface left by its removal is
−44 kJ mol−1 . Replacing the water molecule with H2 O2 gives the structure shown
in figure 6c. The calculated adsorption energy for H2 O2 is −68 kJ mol−1 , so that
it is energetically favourable for hydrogen peroxide to displace water when the
surface is hydroxylated.
7. Reaction of H2 O2 over TiO2 (110) hydroxylated surface
The adsorption geometry for H2 O2 in figure 6c shows the oxygen atom of the
hydrogen peroxide over a basic Ti5c site with hydrogen bonding to two adjacent
Ti5c OH oxygen atoms. We will next consider the hydrogen transfer for one of the
H atoms to form water with a neighbouring hydroxyl group and so generate a
surface-bound hydroperoxy species. To achieve this, we used the nudged elastic
band (NEB) method, with the initial image structures produced using a grouped
interpolation method described below.
(a) Nudged elastic band group method
In the NEB method, a set of structures are defined that lie on a pathway
between the start and endpoint structures, which are usually minima obtained
from geometry optimization. For example, in the H2 O2 deprotonation set out in
figure 6, the startpoint structure was an optimized H2 O2 placed so that one of the
O atoms is near to an exposed Ti5c site on the partially hydroxylated surface. The
endpoint structure was an optimization with the H nearest the surface transferred
to the adjacent hydroxyl group. The easiest way to obtain the intervening set of
structures is by linear interpolation, i.e.
i
(7.1)
(sN − s0 ),
N
where si is the 3Natoms vector for the positions of the ith structure with the
zeroth being the startpoint and the N th the endpoint. The whole set of images
are then simultaneously optimized under the additional constraint of a harmonic
spring force between each image and its neighbours, with the start and endpoints
si = s0 +
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
11
Decomposition of H2 O2 over Au and TiO2
relative energy (kJ mol–1)
(a) 10
5
(c)
0
RC = 0.00
–5
–10
–15
–20
–25
(d)
RC = 0.54
(b) 2.4
Ti5c...OOH
2.2
distance (Å)
2.0
1.8
1.6
O–O
(e)
1.4
1.2
RC = 1.00
HOO–H
1.0
0
0.2
0.4
0.6
0.8
reaction coordinate
1.0
Figure 6. A nudged elastic band (NEB) calculation for the deprotonation of H2 O2 over a Ti5c site
on the hydroxylated surface of TiO2 (110). (a) Relative energy versus reaction coordinate (RC),
(b) atom–atom distances versus RC. Structures shown to the right are at the RC values indicated.
Atom colours follow: Ti, grey; O, red and H, white. (Online version in colour.)
fixed. The harmonic springs prevent the images collapsing onto the minima
endpoints while allowing optimization perpendicular to the RC defined by the
set of structures on the band.
However, for many surface chemical reactions, linear interpolation does not
give a chemically reasonable set of images. When the reaction involves a bond
bending or molecular rotation, the intermediate images tend to over-shorten the
inter-atomic distances, which leads to slow convergence of the NEB optimization.
To counter this, we have used a method of user-defined atom groups. Within a
group, G, a central atom is nominated, and the distances of all members of the
group from the central atom are measured for the start and end structures. For
the image structures, the linear interpolation position of each atom in the group
is adjusted to ensure that their distance to the central atom varies smoothly in
the interpolation:
sji = sj0 +
Proc. R. Soc. A
i
i
(sjN − sj0 ) − rjci + (djcN − djc0 )r̂jci ,
N
N
j ∈ G.
(7.2)
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
12
A. Thetford et al.
Now, sji is the three component vector of the ith structure belonging to atom j,
rjci is the vector from the group centre atom to the jth atom after the ith linear
interpolation has been performed, djc0 and djcN are the distances between the
group centre and the jth atom at the start and endpoints, respectively, and the
hat symbol is used to signify a unit vector.
For the deprotonation reaction of H2 O2 on the hydroxylated TiO2 (110) surface,
two groups were defined. The first consisted of the OOH species with the O atom
nearest to the Ti5c site as the group centre. The second group contained the
hydrogen leaving the H2 O2 adsorbate and the receiving surface OH moiety with
the O atom as the group centre. A first estimate of the barrier was obtained with
an NEB calculation in which only these grouped atoms were allowed to move and
all other atom positions were fixed. This was then used as a starting point for
a second NEB calculation in which the entire top layer of the slab is allowed to
optimize geometry. The results of the second scan are presented in figure 6.
Each of the optimized NEB images was compared with the start structure to
calculate the distance that the system has moved
3Nat
(sij − s0j )2 ,
(7.3)
Di = j=1
where sij is the jth cartessian coordinate of the ith structure for the Nat atoms
of the system. The position of the optimized image along the RC was then
defined as the ratio between the image and the distance between the start and
endpoints, (Di /DN ).
After converging the resulting NEB images, figure 6a shows an estimated
barrier for the H2 O2 deprotonation reaction of 8 kJ mol−1 . The transition state
occurs at a reaction coordinate value of 0.54, as confirmed by a vibrational
frequency calculation for this image, which showed a single imaginary mode.
A plot of the inter-atomic distances as a function of the reaction coordinate
(figure 6b) shows that the barrier involves movement of the molecular hydrogen
peroxide towards the Ti5c site, which places one of the peroxide H atoms near
the receiving hydroxyl group. At RC = 0.54 (figure 6d), the peroxide HOO–H
bond has stretched to 1.18 Å as the transfer takes place. Finally at RC = 1.00
(figure 6e), a surface hydroperoxy species is formed with a neighbouring molecular
adsorbed water molecule. This is favoured over the starting structure by
19 kJ mol−1 . During the reaction, the HO–O bond changes only marginally from
1.42 to 1.47 Å.
8. Au/TiO2 (110)
The TiO2 (110) surface has been widely studied by computational chemists, and a
number of papers have already reported on the interaction of Au with the perfect
surface. For single Au atoms, the lowest energy position is over the surface O2c
atoms with calculated binding energies of around −0.5 eV (Giordano et al. 2001;
Worz et al. 2005). However, the interaction between a continuous slab of Au as
a representation of a large Au particle and the TiO2 (110) surface is found to
be negligibly small (Lopez & Nørskov 2002). The interaction for atomic species
is much stronger (−1.7 eV) when surface O2c defects are introduced as sites for
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Decomposition of H2 O2 over Au and TiO2
(a)
13
(b)
Figure 7. (a) Au cluster supported on the clean TiO2 (110) surface and (b) the hydroxylated
TiO2 (110) surface. Atom colours follow: Au, yellow; Ti, grey, O, red and H, white. (Online version
in colour.)
Au adsorption. Wang & Hammer (2006) have looked at a Au7 cluster on the
TiO2 (110) surface under alkaline conditions as represented by an excess surface
OH group over a Ti5c site on their simulation slab. In this case, Au atoms at
the oxide interface become positively charged and adhesion energies in the range
−2.1 to −2.4 eV are obtained.
Considering the acid environment of the reaction in H2 O2 production, we have
compared the structure of an Au10 particle supported on the clean TiO2 (110)
surface with that on the hydroxylated surface. In the latter case, a patch of
hydroxyl groups was first cleared, since Au deposition will involve a condensation
reaction between Au species in solution and surface hydroxyl groups (Moreau &
Bond 2007). The two structures are compared in figure 7. It is noticeable that the
Au cluster on the clean TiO2 (110) surface makes only minimal contact with the
row of O2c atoms (figure 7a). In contrast, on the hydroxylated surface, the Au
cluster can be seen to spread over the surface to give close Au–OH interactions.
9. Conclusions
We have used periodic DFT to study the adsorption of hydrogen peroxide to
Au10 clusters and to the surface of hydroxylated rutile TiO2 (110). These models
represent the two components of Au/TiO2 catalysts. The calculations show that
hydrogen peroxide undergoes facile HO–OH cleavage on Au10 metal clusters to
form hydroxyl groups that would react further to form water. Calculations on
the bulk surfaces of Au indicate that this process will also be favourable for
larger particles, particularly when low coordination atoms are present on the
surface. HO–OH cleavage is a loss mechanism for hydrogen peroxide over these
catalysts and so the calculations suggest that high H2 selectivities will be attained
by limiting the metal content of the catalyst to a minimum. In contrast, on the
hydroxylated TiO2 (110) surface, there is a small barrier to deprotonation of H2 O2
producing a surface hydroperoxide of the type expected to be an active species
for oxidation reactions.
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
14
A. Thetford et al.
It has been shown using periodic DFT methods that H2 can be dissociated on
Au catalyst systems (Boronat et al. 2009). Since O2 adsorbs molecularly at the
edge of Au clusters, peroxy species could be created at the interface between metal
and support. These observations, along with those presented here on the stability
of H2 O2 over components of the catalyst, suggest that the production of hydrogen
peroxide relies on the interplay of metal and support sites with hydroperoxy
species being stabilized on the TiO2 support. Mapping out this reaction scheme
in greater detail is the focus of ongoing calculations.
The initial work on the required Au/TiO2 (110) combined model shows that the
hydroxylation state of the surface will be critical in describing the interface region.
We would like to thank Dow Chemicals for support under the methane challenge project. Computer
resources were provided by the ARCCA division at Cardiff University and from HECToR through
the Materials Consortium within the reactivity theme.
References
Abrahams, S. C. & Bernstein, J. L. 1971 Rutile: normal probability plot analysis and accurate
measurement of crystal structure. J. Chem. Phys. 55, 3206–3211. (doi:10.1063/1.1676569)
Arrouvel, C., Digne, M., Breysse, M., Toulhoat, H. & Raybaud, P. 2004 Effects of morphology on
surface hydroxyl concentration: a DFT comparison of anatase-TiO2 and g-alumina catalytic
supports. J. Catal. 222, 152–166. (doi:10.1016/j.cat.2003.10.016)
Boronat, M., Illas, F. & Corma, A. 2009 Active sites for H2 adsorption and activation in Au/TiO2
and the role of the support. J. Phys. Chem. A 113, 3750–3757. (doi:10.1021/jp808271y)
Bredow, T., Giordano, L., Cinquini, F. & Pacchioni, G. 2004 Electronic properties of rutile TiO2
ultrathin films: odd–even oscillations with the number of layers. Phys. Rev. B 70, 035419.
(doi:10.1103/PhysRevB.70.035419)
Corma, A. & Garcia, H. 2008 Supported gold nanoparticles as catalysts for organic reactions.
Chem. Soc. Rev. 37, 2096–2126. (doi:10.1039/B707314N)
Edwards, J. K., Ntainjua, E., Carley, A. F., Herzing, A. A., Kiely, C. J. & Hutchings, G. J. 2009
Direct synthesis of H2 O2 from H2 and O2 over gold, palladium, and gold–palladium catalysts
supported on acid-pretreated TiO2 . Angew. Chem.-Int. Ed. 48, 8512–8515. (doi:10.1002/anie.
200904115)
Frisch, M. J. et al. 2004 Gaussian 03, Revision E.01. Wallingford, CT: Gaussian, Inc.
Giordano, L., Pacchioni, G., Bredow, T. & Fdez-Sanz, J. 2001 Cu, Ag, and Au atoms adsorbed
on TiO2 (110): cluster and periodic calculations. Surf. Sci. 471, 21–31. (doi:10.1016/S00396028(00)00879-7)
Hammer, B., Hansen, L. B. & Nørskov, J. K. 1999 Improved adsorption energetics within
density-functional theory using revised Perdew–Burke–Ernzerhof functional. Phys. Rev. B 59,
7413–7421. (doi:10.1103/PhysRevB.59.7413)
Herzing, A. A., Kiely, C. J., Carley, A. F., Landon, P. & Hutchings, G. J. 2008 Identification
of active gold nanoclusters on iron oxide supports for CO oxidation. Science 321, 1331–1335.
(doi:10.1126/science.1159639)
Kresse, G. & Hafner, J. 1993 Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47,
558–561. (doi:10.1103/PhysRevB.47.558)
Kresse, G. & Hafner, J. 1994 Ab initio molecular-dynamics simulation of the liquid-metal–
amorphous-semiconductor transition in germanium. Phys. Rev. B 49, 14 251–14 269. See
http://cms.mpi.univie.ac.at/vasp. (doi:10.1103/PhysRevB.49.14251)
Landon, P., Collier, P. J., Papworth, A. J., Kiely, C. J. & Hutchings, G. J. 2002 Direct
formation of hydrogen peroxide from H2 O2 using a gold catalyst, Chem.Commun. 2058–2059.
(doi:10.1039/B205248M)
Proc. R. Soc. A
Downloaded from http://rspa.royalsocietypublishing.org/ on June 16, 2017
Decomposition of H2 O2 over Au and TiO2
15
Lindsay, R., Wander, A., Ernst, A., Montanari, B., Thornton, G. & Harrison, N. M. 2005 Revisiting
the surface structure of TiO2 (110): a quantitative low-energy electron diffraction study. Phys.
Rev. Lett. 94, 246102. (doi:10.1103/PhysRevLett.94.246102)
Lopez, N. & Nørskov, J. K. 2002 Theoretical study of the Au/TiO2 (110) interface. Surf. Sci. 515,
175–186. (doi:10.1016/S0039-6028(02)01873-3)
Monkhorst, H. J. & Pack, J. D. 1976 Special points for Brillouin-zone integrations. Phys. Rev. B
13, 5188–5192. (doi:10.1103/PhysRevB.13.5188)
Moreau, F. & Bond, G. C. 2007 Preparation and reactivation of Au/TiO2 catalysts. Catal. Today
122, 260–265. (doi:10.1016/j.cattod.2007.01.017)
Okumura, M., Kitagawa, Y., Yamagcuhi, K., Akita, T., Tsubota, S. & Haruta, M. 2003 Direct
production of hydrogen peroxide from H2 and O2 over highly dispersed Au catalysts. Chem.
Letts. 32, 822–823. (doi:10.1246/cl.2003.822)
Perdew, J. P., Burke, K. & Ernzerhof, M. 1996 Generalized gradient approximation made simple.
Phys. Rev. Lett. 77, 3865–3868. (doi:10.1103/PhysRevLett.77.3865)
Perron, H., Domain, C., Roques, J., Drot, R., Simoni, E. & Catalette, H. 2007 Optimisation
of accurate rutile TiO2 (110), (100), (101), and (001) surface models from periodic DFT
calculations. Theor. Chem. Acc. 117, 565–574. (doi:10.1007/s00214-006-0189-y)
Piccinini, M., Ntainjua, E. N., Edwards, J. K., Carley, A. F., Moulijn, J. A. & Hutchings, G. J.
2010 Effect of the reaction conditions on the performance of Au-Pd/TiO2 catalyst for the
direct synthesis of hydrogen peroxide. Phys. Chem. Chem. Phys. 12, 2488–2492. (doi:10.1039/
b921815g)
Swamy, V., Muscat, J., Gale, J. D. & Harrison, N. M. 2002 Simulation of low index rutile surfaces
with a transferable variable-charge Ti–O interatomic potential and comparison with ab initio
results. Surf. Sci. 504, 115–124. (doi:10.1016/S0039-6028(01)01925-2)
Thompson, S. J. & Lewis, S. P. 2006 Revisiting the (110) surface structure of TiO2 : a theoretical
analysis. Phys. Rev. B 73, 073403. (doi:10.1103/PhysRevB.73.073403)
To, J., Sokol, A. A., French, S. A. & Catlow, C. R. A. 2008 Hybrid QM/MM investigations into the
structure and properties of oxygen-donating species in TS-1. J. Phys. Chem. C 112, 7173–7185.
(doi:10.1021/jp066448p)
Wang, J. G. & Hammer, B. 2006 Role of Au+ in supporting and activating Au7 on TiO2 (110).
Phys. Rev. Lett. 97, 136107. (doi:10.1103/PhysRevLett.97.136107)
Worz, A. S., Heiz, U., Cinquini, F. & Pacchioni, G. 2005 Charging of Au Atoms on TiO2 thin films
from CO vibrational spectroscopy and DFT calculations. J. Phys. Chem. B 109, 18 418–18 426.
(doi:10.1021/jp054093o)
Xiao, L., Tollberg, B., Hu, X. & Wang, L. 2006 Structural study of gold clusters. J. Chem. Phys.
124, 114309. (doi:10.1063/1.2179419)
Yoon, B., Häkkinen, H. & Landman, U. 2003 Interaction of O2 with Au clusters: molecular and
dissociative adsorption. J. Phys. Chem. A 107, 4066–4071. (doi:10.1021/jp027596s)
Proc. R. Soc. A