Journal of

Electroanalytical
Chemistry
Journal of Electroanalytical Chemistry 607 (2007) 83–89
www.elsevier.com/locate/jelechem

Electrolysis of water on oxide surfaces

a,*
J. Rossmeisl , Z.-W. Qu b, H. Zhu b, G.-J. Kroes b, J.K. Nørskov a

a
Center for Atomic-scale Materials Design, Department of Physics, Nano-DTU, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark
b
Leiden Institute of Chemistry, Gorlaeus Laboratory, Leiden University, P.O. Box 9502, 2300 RA Leiden, The Netherlands

Received 4 July 2006; received in revised form 2 November 2006; accepted 10 November 2006
Available online 16 January 2007

Abstract

In this paper, density functional theory (DFT) calculations are performed to analyze the electrochemical water-splitting process pro-
ducing molecular oxygen (O2) and hydrogen (H2). We investigate the trends in the electro-catalytic properties of (1 1 0) surfaces of three
rutile-type oxides (RuO2, IrO2, and TiO2). The two first of these oxide anodes show lower O2-evolving over-potentials than metal anodes,
due to weak O binding but strong hydroxyl (HO*) binding on the surface. Furthermore, the binding energies of O, HO, and HOO on the
(1 1 0) surfaces fulfill universal linear relations similar to those found on metal surfaces.
 2006 Elsevier B.V. All rights reserved.

Keywords: Electrolysis; Oxide surfaces; Density functional theory; Linear relations

1. Introduction therefore important to find the optimal oxygen-evolving

electro-catalyst in order to minimize the energy loss. In a
Direct electrochemical splitting of water is a non-pollut- previous paper we have investigated the origin of the
ing way of producing pure hydrogen. The products are free over-potential for elemental metal electrodes applying den-
of carbon-mono-oxide, which is a strong poison in proton sity functional theory calculations [3]. It was found that an
exchange membrane (PEM) fuel cells. If the energy needed oxygen layer has to be formed on the metal surface in order
to run the reaction is provided from sustainable sources, for oxygen evolution to occur. This oxygen layer demands
like wind, hydro or solar power, the hydrogen is a clean a high potential to be stabilized.
and CO2-free energy carrier. Several experimental investigations have shown that
Water electrolysis can be done under alkaline conditions oxide surfaces, particularly rutile-type oxides like RuO2,
with nickel electrodes and many commercial electrolysers IrO2 are considerably better as oxygen-evolving electrodes
utilize this method. Acidic water electrolysis is interesting than the elemental metals [4–6]. In the present paper we
for the development of PEM-based electrolysers. With a investigate the reason for this by calculating the stability
PEM system in connection with e.g., a windmill, it is pos- of the intermediates in the reaction. We consider the most
sible to have a single unit forming hydrogen when the load stable (1 1 0) surface of three rutile-type oxides, RuO2, IrO2
is low and producing electricity when the load is high. Such and TiO2, TiO2 being included because it is an interesting
a unitised regenerative fuel cell [1] has for instance been catalyst for the photo-electrolysis of water [7] and additive
used for space applications [2]. to stabilize active RuO2 and/or IrO2 anodes [5].
Unfortunately acidic electrochemical water splitting is The present work can be viewed as a first step towards a
associated with substantial energy loss, mainly due to the theoretical description of these complex systems. The aim is
high over-potentials at the oxygen-evolving anode. It is to determine volcano-type relations of the activity of the
anode materials as a function of one single descriptor.
Knowing this descriptor, future calculations could suggest
*
Corresponding author. Fax: +45 45932399. improved anode materials by determining their place on
E-mail address: [email protected] (J. Rossmeisl). the volcano curve. Subsequent experiments could test

0022-0728/$ - see front matter  2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.jelechem.2006.11.008
84 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89

whether the suggested materials are indeed better than 20 Å is used to separate the slab from its periodic images.
existing anode materials. Our analysis finds that rutile-type We use the 2 · 1 surface unit cell and the 2 · 2 · 1 Monk-
oxides show better oxygen-evolving properties than the horst–Pack type of k-point sampling for slab calculations.
metals, and that there is a linear relationship between the The two bottom layers of the slab are fixed at the optimized
binding energies of O, HO and HOO intermediates on bulk lattice constants while the two top layers as well as
the oxide surfaces. We show that trends are well described possible adsorbates on it are fully relaxed. Ultra-soft
and it is possible to obtain basic insight based on this sim- pseudo-potentials are used to deal with the ion cores [12].
plified description. Therefore the electronic wave-functions can be represented
well by plane wave basis set with a cutoff energy of 340 eV.
2. Water-splitting reaction The electron density is treated on a grid corresponding to a
plane wave cutoff at 500 eV. A Fermi smearing of 0.1 eV
The water-splitting reaction can be written schematically and Pulay mixing is used to ensure the fast convergence
as: of the self-consistent electron density. Atomic positions
are relaxed until the sum of the absolute forces is less than
H2 O ! 12O2 þ H2 ð1Þ
0.05 eV/Å. All calculations are performed using the ASE
The products, hydrogen and oxygen, are evolved on the simulation package [13].
cathode and the anode side, respectively. We will consider We have previously developed a method for modelling
the following reaction mechanism in an acidic the thermo-chemistry of electrochemical reactions. The
environment: challenge is to include the electrochemical potential. In this
Cathode: method the only way the potential affects the relative free
energy is through the chemical potential of the electrons
2Hþ þ 2e ! H þ Hþ þ e ! H2 ð2Þ
in the electrode. This ‘‘first order’’ inclusion of the potential
Anode: has been shown to predict trends for the oxygen reduction
reaction on metal and metal alloys quite well [14,15]. Fur-
2H2 O ! HO þ H2 O þ Hþ þ e
thermore, we have shown that thermo-chemical features
! O þ H2 O þ 2Hþ þ 2e such as phase diagrams in water are also well described
! HOO þ 3Hþ þ 3e ! O2 þ 4Hþ þ 4e ð3Þ by this method [16].
The method can be summarized as follows:
The * represents an active surface site, such as the bridge
site between two fourfold coordinated metal ions (bridge) 1. By setting the reference potential to be that of the stan-
or the coordinately unsaturated site (CUS) on top of a me- dard hydrogen electrode, we can relate the chemical
tal ion fivefold coordinated on the rutile (1 1 0) surface. potential (the free energy per H) for (H+ + e) to that
We are only considering reaction steps involving charge of 1/2H2 in the gas phase. At pH 0 in the electrolyte
transfer, which depend directly on the applied potential. and 1 bar of H2 in the gas phase at 298 K the reaction
The diffusion of species and other surface reactions are free energy of 1/2H2 ! H+ + e is zero at an electrode
expected to depend only weakly on the potential. We con- potential of U = 0. At standard conditions, the free
sider a process where oxygen molecules are formed by an energy DG0 = DG (U = 0, pH 0, p = 1 bar, T = 298 K)
associative mechanism on the anode via a surface HOO* of the reaction *AH ! A + H+ + e, can therefore be
intermediate. Direct recombination of oxygen atoms to calculated as the free energy of the reaction *AH !
form O2 is not considered because a large activation barrier A + 1/2H2.
can be expected for this process [8]. 2. DG0 = DE + DZPE  TDS, is calculated as follows: The
Hydrogen evolution on the cathode could also happen reaction energy DE is calculated using DFT. The differ-
by direct association between surface H*, but that will ence in zero point energies due to the reaction, DZPE,
not change the picture [9]. Under acidic conditions hydro- and the change in entropy DS is calculated using DFT
gen evolution is very fast on the metal cathode, and hence calculations of the vibrational frequencies and standard
we can concentrate on the oxygen-evolving reaction in the tables for gas phase molecules [17].
following. 3. We include the effect of a bias on all states involving an
electron in the electrode, by shifting the energy of this
3. Method state by DGU = eU, where U is the electrode potential
relative to the standard hydrogen electrode.
The DFT calculations are performed using a plane wave 4. At a pH different from 0 we can correct for the free energy
implementation [10] at the generalized gradient approxima- of H+-ions by the concentration dependence of the
tion (GGA) RPBE level [11]. The bulk lattice constants of entropy: DGpH (pH) = kT Æ ln [H+] = kT Æ ln Æ 10 Æ pH.
RuO2, IrO2 and TiO2 are optimized using the 8 · 8 · 8
Monkhorst–Pack type of k-point sampling. A periodically The reaction free energy is then calculated as:
repeating 4-layer slab is chosen for the most stable rutile-
type (1 1 0) surface in our calculations. A vacuum of at least DGðU ; pH;pH2 ¼ 1 bar; TÞ ¼ DG0 þ DGU þ DGpH ð4Þ
 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89 85

H2O and H2 in the gas phase are used as references states, The effect of the water surroundings has been investi-
since they are well described within DFT. The entropy for gated by performing simulations on the RuO2 surface cov-
gas phase water is calculated at 0.035 bars because this is ered with five water molecules. The calculations were
the equilibrium pressure in contact with liquid water at performed on a surface with all bridge site covered with
298 K. The free energy of gas phase water at these condi- oxygen, one CUS site was occupied by the adsorbents
tions is equal to the free energy of liquid water. The free en- (O* or HO*). On the second CUS site one of the five water
ergy change of the total reaction: H2O ! 1/2O2 + H2, is molecules was situated. Water molecules were put above all
fixed at the experimental value of 2.46 eV, in order to avoid bridge and CUS sites. In this structure everything were
the calculation of the O2 bond energy which is hard to cal- relaxed and hydrogen bonds were formed. However, the
culate accurately with DFT. change in adsorption energy at the CUS site due to the
interaction with water is for both O* and HO* very small
4. Results (less than 0.05 eV), and we therefore conclude that the
effect of the solvent in this case can be neglected. On the
4.1. Water splitting on the RuO2(1 1 0) surface other hand this effect is huge on the metal surfaces and is
therefore included in that case [14].
We start with RuO2 as a model system, RuO2 is known On the O*-covered RuO2(1 1 0) surface the intermediate
to be a quite good catalyst for oxygen evolution [5,4,6]. We structures along with the calculated free energies are shown
calculate the reaction intermediates at two different sur- in Fig. 2. Starting with an oxygen vacancy at a CUS site,
faces: one with all bridge and CUS sites occupied by oxy- water is dissociated at the vacancy forming a hydroxyl
gen (O*) and one with these sites occupied by hydroxyl group. Next, the proton is removed, after which another
(HO*). We calculate the relative stability of the different water molecule dissociates resulting in HOO* on the CUS
surface states by considering the reaction free energies as site. Finally, upon the removal of one more proton, O2 des-
follows: orbs leaving a vacancy at the surface.
The reactions and the corresponding reaction free ener-
RuO2 þ 2O þ 2H2 O $ RuO2 þ 4HO ð5Þ
gies under an applied potential U can be written as:
DG ¼ 0:47 eV
RuO2 þ 2O þ 2H2 O $ RuO2 þ 4O þ 4e þ 4Hþ ð6Þ 2H2 O $ H2 O þ HO þ e þ Hþ ;
DG ¼ 5:12 eV  4 eU DG1 ¼ DGHO  DGwater ¼ 1:32 eV  eU ð7Þ
RuO2 + 2O* represents the surface with all bridge sites H2 O þ HO þ e þ Hþ $ H2 O þ O þ 2ðe þ Hþ Þ;
occupied by oxygen and RuO2 + 4O* represents the totally DG2 ¼ DGO  DGHO ¼ 2:71 eV  1:32 eV  eU ð8Þ
O*-covered surface. Note that the first reaction free energy
¼ 1:39 eV  eU
does not and the second does depend on the applied poten-
tial U. The relative stability as function of potential U is H2 O þ O þ 2ðe þ Hþ Þ $ HOO þ 3ðe þ Hþ Þ;
shown in Fig. 1. It can be seen that the O*-covered and DG3 ¼ DGHOO  DGO ¼ 4:31 eV  2:71 eV  eU ð9Þ
the HO*-covered surfaces are very close in stability in the ¼ 1:60 eV  eU
interesting potential range from 1.3 V to 1.7 V. At low
HOO þ 3ðe þ Hþ Þ $ O2 þ 4ðe þ Hþ Þ
potentials the HO*-covered surface is more stable while
the O*-covered surface becomes more stable at potentials DG4 ¼ DGO2  DGHOO ¼ 4:96 eV  4:31 eV  eU ð10Þ
above 1.4 V. ¼ 0:65 eV  eU
Relative stability [eV]

-1

OH covered O covered
-2

0 1
Potential [V]
Fig. 1. The phase-diagram of the RuO2(1 1 0) surface calculated as function of the potential. At a potential of approximately 1.40 V the surface changes
from being totally HO*-covered surface to becoming O*-covered. The top horizontal line represents the free energy of the surface with all bridge site
occupied with O* (DG ” 0), the other horizontal line is the free energy of the surface covered by HO* (DG = 0.47 eV), the tilted line represents the
calculated O*-covered surface (DG = 5.12 eV  4 eU).
86 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89

+2H2O(l) +H2O(l)+½H2(g) +H2O(l)+H2(g) +3/2H2(g) +O2(g)+ 2H2(g)

ΔEHO=0.97 eV ΔEO=2.66 eV ΔEHOO=3.91 eV
ΔGwater=0.0 eV ΔGHO=1.32 eV ΔGO=2.71 eV ΔGHOO=4.31 eV ΔGO2=4.96 eV

Fig. 2. The calculated free energies at zero potential and structures of the intermediates of the oxygen evolution reaction on O*-covered RuO2(1 1 0)
surface are shown. The white dashed box indicates the unit cell. Blue spheres represent metal ions, red spheres represent oxygen and white spheres
represent hydrogen. From the left: frame 1, the surface with one oxygen vacancy on the CUS site, the free energy for this starting point is defined as zero,
frame 2, a water molecule adsorbs on the vacancy and splits of a proton forming HO*. Frame 3, the next proton is transferred to the electrolyte forming
O*. A second water molecule is adsorbed and the two protons transferred consecutively, first leading to HOO* (frame 4), and then to desorbed O2 and a
vacancy (frame 5). DFT binding energies are corrected for change in entropy (at 298 K) and zero point energy. For adsorbed O, OH, and OOH the
corrections added to the binding energy to obtain the free energy are 0.05, 0.35, and 0.40 eV, respectively.

5 tro-catalytic activity of the different surfaces. Notice that

O2(g) U=0 V the lower limit is defined by the equilibrium potential for
-
4 HOO* +4(e +H+) the overall reaction, which in this case is 1.23 V. Fig. 3
O*+H2O(l) +3(e-+H+)
- +
+2(e +H )
illustrates that the first three reaction steps are endothermic
3
for the equilibrium potential, and that an overpotential of
HO*+H2O(l)
0.37 V is necessary to make all steps down-hill in free
ΔG [eV]

2 - +
+e +H
energy.
1 The same calculations are performed on a HO*-covered
2H2O(l)
0 U=1.23 V surface, see Fig. 4. Hydrogen bonds between the hydroxyl
groups tend to stabilize HO* and HOO* slightly compared
-1 to the situation with O*. This surface composition will not
U=1.60 V
be quite as active as the previous one – it is only at poten-
-2
tials above 1.73 V that all steps become down-hill in free
Fig. 3. The free energies of the intermediates on O-*-covered RuO2 at energy. Furthermore, Fig. 1 shows that at such potentials
three different potentials (U = 0, U = 1.23, and U = 1.60 V) are depicted. the O*-covered surface is the relevant stable surface,
At the equilibrium potential (U = 1.23 V) the reaction steps are uphill in whereas the OH*-covered surface is unstable.
free energy. At 1.60 V all reaction steps are downhill in free energy.
As expected, our calculations show that RuO2 is a very
good oxygen-evolving catalyst. It is considerably better
The largest free energy difference (1.60 eV) is found for than the pure metal and oxidized metal surfaces we consid-
the third step, which is splitting the second water molecule. ered in a previous study [3]. Furthermore, we find that the
This means that at potentials above 1.60 V all reaction O*-covered surface shows higher activity than the HO-cov-
steps of oxygen evolution are exothermic (see Fig. 3). We ered surface, and that in any case the reaction requires an
will use the minimum potential at which all reaction steps overpotential such that RuO2 is expected to be covered
become downhill in free energy as the measure of the elec- by O* rather than OH*.

+2H2O(l) +H2O(l)+½H2(g) +H2O(l)+H2(g) +3/2H2(g) +O2(g)+ 2H2(g)

ΔEHO=0.47 eV ΔEO=2.50 eV ΔEHOO=3.74 eV
ΔGwater=0.0 eV ΔGHO=0.82 eV ΔGO=2.55 eV ΔGHOO=4.14 eV ΔGO2=4.96 eV

Fig. 4. The calculated free energies at zero potential and structures of the intermediates of the oxygen evolution reaction are shown. The white dashed box
indicates the unit cell. Blue spheres represent metal ions, red oxygen and white hydrogen. The RuO2(1 1 0) surface is covered with HO* groups. From the
left: frame 1, the surface with one oxygen vacancy on the CUS site, the free energy for this starting point is defined as zero, frame 2, a water molecule
adsorbs on the vacancy and splits of a proton forming HO*. Frame 3, the next proton is transferred to the electrolyte forming O*. A second water molecule
is adsorbed and the two protons transferred consecutively, first leading to HOO* (frame 4), and then to desorbed O2 and a vacancy (frame 5). DFT binding
energies are corrected for change in entropy (at 298 K) and zero point energy. For adsorbed O, OH, and OOH the corrections added to the binding energy
to obtain the free energy are 0.05, 0.35, and 0.40 eV, respectively.
 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89 87

On IrO2 the reaction is very similar to that on RuO2. are approximately the same as the ones (0.5) found for
The IrO2 surface is also covered by oxygen and the forma- metal surfaces [3], reflecting the double bonding nature
tion of HOO* is likewise the rate limiting step, but the over- of O@* and the single bonding nature of HO–* and
potential is slightly larger than what was found for RuO2 HOO–*.
(0.56 V compared to 0.37 V). The binding energy relations on oxide surfaces are
The reaction on TiO2 is considerably different, only shifted down by about 1 eV compared with those on metal
bridge sites are occupied and the step that demands the surfaces. For a metal and a metal oxide surface with the
highest potential is the formation of O* on the CUS site. same O* binding energy both HO* and HOO* are bound
The overpotential is found to be 1.19 V. The trend of considerably stronger on the oxide surface. As a result, dif-
over-potentials can be written as: TiO2 > IrO2 > RuO2. ferent elementary reaction steps may limit the oxygen-
Note that this approach only accounts for the thermo- evolving activity on an oxide anode compared with that
chemistry of the elementary reaction steps – no barriers on a metal anode. This opens a new dimension in rational
are included. This means that the method presented here catalyst design.
represents a necessary, but not a sufficient analysis of the With the established linear relations between the binding
activity. energies of O*, HO*, and HOO* intermediates in hand, it is
possible to calculate the oxygen-evolving activity as func-
tion of only one parameter, i.e., the O* binding energy.
4.2. Linear relations of binding energies
Since all the investigated elementary steps are basically sin-
gle-proton transfers from the adsorbents to the electrolyte,
The analysis described above can be used to predict the
it is reasonable to assume as a first approximation that they
trends of catalytic activity. In a previous study [3] we found
all have the same pre-exponential factor. In case there are
that the binding energies of HO, O and HOO on metal sur-
additional activation barriers they are often found to be
faces are linearly correlated. To establish similar correla-
linearly related to the reaction free energy [8]. Therefore,
tions on oxide surfaces, we calculate binding energies of
the reaction free energy can be used directly as a simple
O, HO and HOO at O*, CUS and bridge sites of three
measure of electro-catalytic activity. The reaction free ener-
rutile-type (1 1 0) surfaces for different coverages of O*
gies can be expressed as functions of the potential U and
and HO* on the surface. In this way we obtain a wide range
the O* binding energy as follows:
of binding energies as shown in Fig. 5. Clearly linear rela-
tions also exist for the binding energies on oxide surfaces,
i.e., the oxide surfaces that bind O* strongly also bind 2H2 O $ H2 O þ HO þ e þ Hþ ;
HO and HOO strongly. It is interesting to note that even DG1 ¼ DGHO  DGwater ¼ DEHO þ 0:35 eV  eU
the slopes (0.6) of the linear relations for oxide surfaces ¼ 0:61DEO  0:55 eV  eU ð11Þ

3 5
TiO2_OH TiO2_2O
TiO2_3O

2 TiO2_2O
4 RuO2_3O
RuO2_3OH
1 RuO2_3O
ΔEHOO [eV]

RuO2_2O
ΔEHO [eV]

RuO2_2O
RuO2_3OH IrO2_3O
RuO2_2OH
IrO2_3O RuO2_2OH 3 IrO2_2OH
IrO2_3OH
0 RuO2
RuO2 IrO2_2O RuO2_O IrO2_2O
RuO2_O IrO _3OH
2 RuO2_OH
IrO2_2OH
IrO2_OH
RuO2_OH IrO2_O
-1 IrO2_O 2 IrO2
IrO2
TiO2_O
TiO2
-2
IrO2_OH
1
-2 0 2 4 6 -1 0 1 2 3 4 5
ΔEO [eV] ΔEO [eV]

Fig. 5. To the left the linear relation between binding energies of O* and HO* on rutile-type oxides (1 1 0) surfaces is depicted. To the right the relation
between O* and HOO* adsorption energies is shown. Calculations are done at different levels of oxidized and reduced surfaces, resulting in O* binding
energies in a range of more than 5 electron volts. 1/2O2 M O* corresponds to DEO = 2.41 eV, DEO = 0 eV is defined by the equilibrium: H2O M O* + H2.
The best fits to the data points are: DEHO = 0.61 DEO  0.90 eV and DEHOO = 0.64 DEO + 2.03 eV, respectively, compared with those of DEHO = 0.50
DEO + 0.05 eV and DEHOO = 0.53 DEO + 3.18 eV on metal (1 1 1) surfaces. The green spots represent the stable surface configuration at water oxidation
potentials.
88 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89

ited by the same reaction steps. Although the best metal

H2 O þ HO þ e þ Hþ $ H2 O þ O þ 2ðe þ Hþ Þ;
anode (the top of activity volcano curve) can be found at
DG2 ¼ DGO  DGHO ¼ DEO þ 0:05 eV  DEHO  0:35 eV  eU the O* binding energy around 3.2 eV, the corresponding
¼ 0:39DEO þ 0:60 eV  eU ð12Þ overpotential is much higher than that of the best oxide
  þ   þ
H2 O þ O þ 2ðe þ H Þ $ HOO þ 3ðe þ H Þ; anode for oxygen evolution. All clean metal surfaces have
DG3 ¼ DGHOO  DGO ¼ DEHOO þ 0:40 eV  DEO  0:05 eV  eU too strong oxygen binding, even gold has a too strong
¼ 0:36DEO þ 2:38 eV  eU ð13Þ binding of 2.75 eV. However, metal surfaces with a high
oxygen coverage have been found to fulfill the same linear
HOO þ 3ðe þ Hþ Þ $ O2 þ 4ðe þ Hþ Þ;
relations as shown above and can therefore be investigated
DG4 ¼ DGO2  DGHOO ¼ 4:92 eV  DEHOO  0:40 eV  eU
with the analysis shown in Fig. 6. Oxygen evolution on
¼ 0:64DEO þ 2:49 eV  eU ð14Þ metals will only occur on surfaces with a high oxygen cov-
The activity, defined as the negative change of the free erage [3]. In reality many metals might form oxides at high
energy, DG, is shown in Fig. 6 for the four reaction steps. oxidation potentials and start to fulfill the oxides linear
Compared with the theoretical equilibrium potential of relations.
1.23 V, it seems possible to find an oxide associated with The volcano curve shown here is in qualitative agree-
a very small over-potential at an O* binding energy around ment with what has been proposed earlier based on exper-
2.3 eV (relative to H2O). For strong (small positive or neg- iments [5]. The main result of this paper is really the
ative DEO) and weak (large positive DEO) O* binding on deduction of the descriptor of the reaction, DEO, (the
the oxide surfaces the oxygen-evolving activity is limited x-axis in Fig. 6), based directly on insight of the reaction
by the *HOO (green line) and the *O (red line) formation at the atomic level. Knowing the reaction descriptor will
step, respectively. As shown above, RuO2 binds oxygen make it possible to search computationally for new materi-
only a little too weakly with an O* binding energy of als with better catalytic properties.
2.66 eV, while IrO2 binds O* too strongly (DEO = 1.66 eV), We note that by turning Fig. 6 up side down the same
leading to an even higher overpotential in agreement with analysis can be done for the reverse reaction, i.e., the oxy-
experiments [5,6]. On the other hand, TiO2 binds O* too gen reduction reaction at low temperatures on cathodes in
weakly and hence is also associated with low oxygen- PEM fuel cell. Now the potential at which all the reverse
evolving activity. reaction steps become exothermic should be as high as pos-
Based on previous calculations [3], we have performed sible, with the upper limit being the equilibrium potential
the same analysis for metal surfaces as also shown in the of 1.23 V. The oxygen reduction activity is limited by the
right panel of Fig. 6. The metal surfaces and oxide are lim- HO* reduction step (black line) and O2 reduction steps

0 0
-ΔG1=-ΔGHO

-ΔG2=-(ΔGO-ΔGHO) -ΔG2=-(ΔGO-ΔGHO)
-1 -1 -ΔG1=-ΔGHO
Activity [eV]

Activity [eV]

-ΔG4=-(ΔGO -ΔGHOO)
-ΔG4=-(ΔGO -ΔGHOO) RuO2_3O 2
2
IrO2_3O

-2 -2
-ΔG3=-(ΔGHOO-ΔGO) -ΔG3=-(ΔGHOO-ΔGO)
TiO2_2O

-3 -3
-2 -1 0 1 2 3 4 5 -2 -1 0 1 2 3 4 5
ΔEO[eV] EO[eV]

Fig. 6. The theoretical activity of the four charge transferring steps of oxygen evolution is depicted as function of the oxygen binding energy. The y-axis is
the activity defined as DG of the different reaction steps, calculated from the linear relations: DG1 = DGHO  DGwater = DEHO + 0.35 eV,
DG2 = DGO  DGHO = DEO + 0.05 eV  DEHO  0.35 eV, DG3 = DGHOO  DGO = DEHOO + 0.40 eV  DEO  0.05 eV, DG4 = DGO2  DGHOO =
4.96 eV  DEHOO  0.40 eV. The resulting volcano is indicated with dash–dot lines and the hatched areas. For the metal surfaces interaction
with water has been taken into account, it results in a stabilization of HO and HOO compared to O. For the oxides the interaction with water is small. The
best possible material would fall on the horizontal dashed line representing the equilibrium potential 1.23 eV. The optimal oxides are very close to this
value.
 J. Rossmeisl et al. / Journal of Electroanalytical Chemistry 607 (2007) 83–89 89

(blue line) at the strong and weak O* bonding, respectively. class of systems) may offer even better catalysts for these
The metal surfaces are almost as good as oxides for this types of processes. It would be particularly interesting to
reaction, Pt is the best metal with a oxygen binding energy find a class of systems where the maximum in activity for
at 1.56 eV, close to the very top of the metal volcano. This both oxygen evolution and oxygen reduction happens at
is consistent with the fact that Pt is known to be a very the same oxygen adsorption energy, that is, for the same
good catalyst for this reaction, which we have also found material. Oxides are in this sense quite good, however,
in a previous study [14]. The reason why oxides in reality the problem is that they are only thermodynamically stable
are not used as fuel cell cathodes is because the oxides under oxidation conditions.
are thermodynamically unstable under reducing conditions
in an acidic environment. Under such conditions they may Acknowledgments
be covered by HO* or dissolve [18], although important
examples do exist showing that RuO2 and IrO2 may be CAMD is founded by the Lundbeck foundation. This
kinetically stable and highly useful cathode materials for, work was supported by EU through Grant No. MRTN-
for instance, H2 evolution [19]. The stability issue is as CT-2006-032474, the Danish Research Council through
important as the activity and this will be investigated in the NABIIT programme, the Danish Center for Scientific
the future. Computing through Grant No. HDW-1103-06, and the
This very simple analysis presented here only takes into NWO/ACTS hydrogen programme.
account the material dependent parameters, namely the
binding energies. It does not take into account effects of
References
electric field and surface charges. Also proton transfer bar-
riers are not included. All these effects are very important [1] T. Ioroi, N. Kitazawa, K. Yasuda, Y. Yamamoto, H. Takenaka,
for the absolute rate of the reaction. However, we would J. Appl. Electrochem. 31 (2001) 1179–1183.
not expect such effects to be as important for the trend [2] F. Mitlitsky, B. Myers, A.H. Weisberg, Energ. Fuels 12 (1998) 56–71.
study as the present. [3] J. Rossmeisl, A. Logadottir, J.K. Nørskov, Chem. Phys. 319 (2005)
178–184.
[4] A. Marshall, M. Tsypkin, B. Børresen, G. Hagen, R. Tunold, Mater.
5. Conclusion Chem. Phys. 94 (2005) 226–232.
[5] (a) S. Trasatti, Electrochim. Acta. 29 (1983) 1503–1512;
First of all, we have established that rutile oxide surfaces (b) S. Trasatti, Electrochim. Acta. 36 (1991) 225–241.
obey linear relations between O* binding energy and the [6] A. Marshall, M. Tsypkin, B. Børresen, G. Hagen, R. Tunold, J. New
Mater. Electrochem. Syst. 7 (2004) 197–204.
binding energy of two other species relevant to the oxida-
[7] M. Gratzel, Nature 414 (2001) 338–344.
tion of water (HO* and HOO*). The relations on oxides [8] J.K. Nørskov et al., J. Catal. 209 (2002) 275–278.
are shifted compared to the similar relations on metals, [9] J.K. Nørskov et al., J. Electrochem. Soc. 152 (2005) 23–26.
which opens up new possibilities for rational anode design. [10] M.C. Payne, M.P. Teter, D.C. Allan, T.A. Arias, J.D. Joannopoulos,
Based on such linear binding energy relations, it is pos- Rev. Mod. Phys. 64 (1992) 1045–1097.
[11] B. Hammer, L.B. Hansen, J.K. Nørskov, Phys. Rev. B 59 (1999)
sible to describe trends in the oxygen-evolving activity
7413–7421.
through a volcano curve showing the activity as a function [12] D. Vanderbilt, Phys. Rev. B 41 (1990) 7892–7895.
of a single parameter, the binding energy of O to the sur- [13] An open source code available at http://wiki.fysik.dtu.dk/ase.
face. In agreement with experiment, it is found that [14] J.K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J. Kitchin,
RuO2 is a very good catalyst with a high activity, while T. Bligaard, H. Jónsson, J. Phys. Chem. B 108 (2004) 17886–17892.
IrO2 shows a higher overpotential. Our theoretical analysis [15] V. Stamenkovic, B.S. Moon, K.J.J. Mayrhofer, P.N. Ross, N.M.
Markovic, J. Rossmeisl, J. Greeley, J.K. Nørskov, Angew. Chem. Int.
suggests that there is some room for improvement: A mate- Ed. 45 (2006) 2897–2901.
rial that binds oxygen a little stronger than the RuO2 is pre- [16] J. Rossmeisl, J.K. Nørskov, C.D. Taylor, M.J. Janik, M. Neurock,
dicted to have a higher catalytic activity, meaning that the J. Phys. Chem. B 110 (2006) 21833–21839.
potential at which such a material would split water is [17] P.W. Atkins, Phys. Chem., sixth ed., Oxford University Press,
Oxford, 1998, pp. 485, 925–927, 942.
slightly lower than the potential needed on RuO2.
[18] Y. Matsui, M. Hirratani, S. Kimura, J. Mater. Sci. 35 (2000) 4093–
We note that other classes of materials, which obey 4098.
other relations between adsorption energies (or, indeed, [19] L. Chen, D. Guay, A. Lasia, J. Electrochem. Soc. 143 (1996) 3576–
possible exceptions to the general correlations within a 3584.