Surface Science 602 (2008) 3424–3431
Contents lists available at ScienceDirect
Surface Science
journal homepage: www.elsevier.com/locate/susc
Reactivity descriptors for direct methanol fuel cell anode catalysts
Peter Ferrin a, Anand Udaykumar Nilekar a, Jeff Greeley b, Manos Mavrikakis a, Jan Rossmeisl a,c,*
a
b
c
Department of Chemical and Biological Engineering, University of Wisconsin-Madison, 1415 Engineering Drive, Madison, WI 53706, USA
Center for Nanoscale Materials, Argonne National Laboratory, Argonne, IL 60439, USA
Center for Atomic-scale Materials Design, Department of Physics- Nano-DTU, Technical University of Denmark, Kgs. Lyngby 2800, Denmark
a r t i c l e
i n f o
Article history:
Received 4 July 2008
Accepted for publication 11 August 2008
Available online 28 August 2008
Keywords:
Methanol
Transition metals
DFT
Electrocatalysis
a b s t r a c t
We have investigated the anode reaction in direct methanol fuel cells using a database of adsorption free
energies for 16 intermediates on 12 close-packed transition metal surfaces calculated with periodic, selfconsistent, density functional theory (DFT–GGA). This database, combined with a simple electrokinetic
model of the methanol electrooxidation reaction, yields mechanistic insights that are consistent with previous experimental and theoretical studies on Pt, and extends these insights to a broad spectrum of other
transition metals. In addition, by using linear scaling relations between the adsorption free energies of
various intermediates in the reaction network, we find that the results determined with the full database
of adsorption energies can be estimated by knowing only two key descriptors for each metal surface: the
free energies of OH and CO on the surface. Two mechanisms for methanol oxidation to CO2 are investigated: an indirect mechanism that goes through a CO intermediate and a direct mechanism where methanol is oxidized to CO2 without the formation of a CO intermediate. For the direct mechanism, we find
that, because of CO poisoning, only a small current will result on all non-group 11 transition metals;
of these metals, Pt is predicted to be the most active. For methanol decomposition via the indirect mechanism, we find that the onset potential is limited either by the ability to activate methanol, by the ability
to activate water, or by surface poisoning by CO* or OH*/O*. Among pure metals, there is no obvious candidate for a good anode catalyst, and in order to design a better catalyst, one has to look for bi-functional
surfaces such as the well-studied PtRu alloy.
Ó 2008 Elsevier B.V. All rights reserved.
1. Introduction
Direct methanol fuel cells (DMFCs) are attracting considerable
attention since many of the problems present in PEM-FC technology, especially those concerning hydrogen storage, can be circumvented by using methanol as a feed [1,2]. However, other problems
exist that limit the usefulness of DMFCs. With regards to electrocatalysis there are two main concerns: (1) CO is one of the natural
intermediates of methanol decomposition while, at the same time,
being a strong poison for the platinum anode and thus seriously
compromising its activity and (2) the anode overpotential with
current technology is higher than that of a hydrogen-based fuel
cell, thus reducing the overall cell potential and fuel cell efficiency,
even though the equilibrium potential of the anode reaction is similar to that of the hydrogen fuel cell anode [3]. These two concerns
are not necessarily independent, as the formation of a CO intermediate may require a high overpotential to be removed. These points
suggest the need for improved catalysis via alloy material design
* Corresponding author. Address: Department of Chemical and Biological Engineering, University of Wisconsin-Madison, 1415 Engineering Drive, Madison, WI
53706, USA. Tel.: +1 608 262 9053; fax: +1 45 45932399.
E-mail address: [email protected] (J. Rossmeisl).
0039-6028/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.susc.2008.08.011
for further development of DMFCs. The present anode material of
choice is an alloy of Pt, Ru and sometimes a third or fourth metal
[4–9]. These alloys have higher activity and less severe CO poisoning than monometallic catalysts because of the bi-functionality of
their component metals [10]. However, this is an expensive solution, since both Pt and Ru are very rare metals, and thus further
materials development is highly desirable.
In order to design better materials for this reaction, it is important to have an understanding of the mechanism of the anode reaction in DMFCs. Unfortunately, the complexity of this reaction, with
many possible intermediates and competing reaction pathways,
has limited the amount of mechanistic insight that can be obtained
[1]. On Pt, for example, it has been suggested that a direct reaction
path is possible [11–15], but on other metals, only limited mechanistic information is available [12]. In the direct mechanism, the
reaction path does not involve a CO intermediate, and CO2 is formed
directly from methanol; the indirect mechanism, in contrast, proceeds via a CO intermediate and its subsequent oxidation to CO2.
Density functional theory (DFT) has been used to better understand many aspects of electrochemical reactions. For instance, DFT
has been used to elucidate the reaction mechanisms for the oxygen
reduction reaction [16], the hydrogen evolution reaction [17], electrochemical water splitting [18], and methanol oxidation [11].
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
Using this understanding, one can identify essential characteristics
for good electrocatalysts; careful evaluation of these characteristics, in turn, can suggest new materials for use as electrocatalysts
[19,20]. In addition, DFT has been used to describe other electrochemical phenomena such as metal dissolution [21].
The aim of this paper is to investigate trends in methanol
decomposition on pure metal surfaces. Here, instead of attempting
to model the electrochemical environment with a high degree of
accuracy, we focus on differences in the adsorption behavior of
key intermediates on different metals. This approach should be useful for understanding trends in the activity of the catalytic electrooxidation of methanol across a variety of transition metals. In
particular, we investigate methanol decomposition on 12 transition
metal surfaces: namely, the close-packed facets of Au, Ag, Cu, Pt, Pd,
Ni, Ir, Rh, Co, Os, Ru, and Re. Using periodic self-consistent density
functional theory calculations within the generalized-gradient
approximation (DFT-GGA), we calculate the adsorption free energy
of 16 possible reaction intermediates on each of the 12 surfaces.
Based on these data, we investigate the direct and indirect reaction
mechanisms, and we identify the important parameters (descriptors) that determine: (1) the required reaction overpotential and
(2) the reaction path found on the different metals.
2. Methods
The free energy of different adsorbates on each surface is calculated by employing density functional theory using DACAPO, a total energy code [22,23]. A periodic 3 3 unit cell (corresponding to
1/9 ML coverage of each adsorbate) with three layers of metal
atoms for each slab and at least five equivalent layers of vacuum
between successive slabs is used throughout this study. Metal
atoms are kept fixed at their optimized bulk positions, as selected
calculations showed that surface relaxation has only small effects
on the energetics of the systems studied. The (1 1 1) facet of the
face-centered cubic metals Au, Ag, Cu, Pt, Pd, Ni, Ir, and Rh is used.
The (0 0 0 1) surface of the hexagonal-close-packed metals Co, Os,
Ru, and Re is employed. Adsorption is allowed on only one of the
two exposed surfaces, with the dipole moment adjusted accordingly [24,25]. Ionic cores are described by ultrasoft pseudopotentials [26]. The Kohn–Sham one-electron states are expanded in a
series of plane waves with an energy cutoff of 25 Ry. Based on
the convergence of total energies, the surface Brillouin zone for
non-group 11 transition metals (i.e., all metals considered in this
study expect Au, Ag, and Cu) is sampled at 6 special Chadi–Cohen
k-points, whereas 18 k-points are sampled for Au, Ag, and Cu [27].
The exchange-correlation energy and potential are described selfconsistently, using the PW91 form of the generalized-gradient
approximation (GGA) [28]. The electron density is determined by
iterative diagonalization of the Kohn–Sham Hamiltonian, Fermi
population of the Kohn–Sham states (kBT = 0.1 eV), and Pulay mixing of the resulting electronic density. All total energies have been
extrapolated to kBT = 0 eV. The energetics determined with this approach is sufficiently accurate to determine relative reactivities
(i.e., trends) for the electrooxidation of methanol across the transition metal series. We include change in zero-point energy (ZPE)
upon adsorption calculated from vibrational frequencies relative
to the reference species. Vibrational frequencies are calculated by
numerical differentiation of forces using a second-order finite difference approach with a step-size of 0.015 Å [29]. The Hessian matrix is mass-weighted and diagonalized to yield the frequencies
and normal modes of the adsorbed species. Calculations performed
with selected adsorbates on a relaxed slab showed only minor differences in vibrational frequency and adsorption energetics as
compared with the fixed slab. All calculations involving Co and
Ni are spin-polarized.
3425
We also include change in entropies; for gas and liquid-phase
molecules, tabulated entropies are used [30]. For molecules bound
to the surface, the vibrational entropy is calculated assuming a
quantum mechanical harmonic oscillator with the same vibrational frequencies as for the zero-point energy.
All free energies are calculated relative to H2O(l), CO2(g), and
H2(g); the free energy of methanol is calculated from the reaction
CO2 ðgÞ þ 3H2 ðgÞ ! CH3 OHðgÞ þ H2 OðlÞ;
as follows:
DGCH3 OH ¼ TEH2 O þ TECH3 OH TECO2 3 TEH2 þ ZPEH2 O þ ZPECH3 OH
ZPECO2 3 ZPEH2 TðSH2 O þ SCH3 OH 3 SCO2 SH2 Þ;
where TE is the total energy of reactant and product species calculated by DFT, T is the standard temperature (298 K), ZPE is the zeropoint energy for the species as calculated from the vibrational frequencies, and S is the entropy of the species.
The free energy of HCOOH(g) is calculated from the reaction
CO2 ðgÞ þ H2 ðgÞ ! HCOOHðgÞ :
DGHCOOH ¼ TEHCOOH TECO2 TEH2 þ ZPEHCOOH
ZPECO2 ZPEH2 T ðSHCOOH SH2 SCO2 Þ:
Free energies of surface intermediates are also calculated in a
similar manner, e.g.,
CO2 ðgÞ þ 2H2 ðgÞ þ ! CHOH þ H2 OðlÞ
DGCHOH ¼ TEH2 O þ TECHOH TECO2 2 TEH2 TEclean
þ ZPEH2 O þ ZPECHOH ZPECO2 2 ZPEH2
T ðSH2 O þ SCHOH SCO2 2 SH2 Þ;
where TEclean is the total energy of the clean slab, TECHOH is the total
energy of CHOH adsorbed on a clean slab, ZPECHOH and SCHOH are
the zero-point energy and entropy for the adsorbed CHOH and
the other terms are as above.
To treat the electrochemical potential, we apply the concept of
the computational standard hydrogen electrode (SHE). At standard
conditions, the free energy of protons and electrons at zero potential is equal to the free energy of the hydrogen molecule [16,31].
Thus, the numbers in Table 1 are the free energies of each intermediate at zero potential. Considering a reaction such as
CH3 OHðgÞ ! CH2 OH þ Hþ þ e ;
for example, a change in electrode potential (U) will change the free
energy of the reaction by DG(U) = eU. Therefore, we can identify
the highest positive change in free energy along the reaction path
(at zero potential) with the electrode potential above which all elementary reaction steps are downhill in free energy. As is shown later, even though this potential is well-defined, it may not in all cases
be directly comparable to experimental polarization curves where
poisons, diffusion barriers, and kinetic effects can affect the current.
This method for including the effect of potential has previously
been used successfully to describe the oxygen reduction [16] and
oxygen evolution reactions [32,33].
3. Results and discussion
3.1. Methanol decomposition on Pt
The total anode reaction can be written as
CH3 OHðgÞ þ H2 OðlÞ ! CO2 ðgÞ þ 6Hþ þ 6e :
In this paper, we restrict our investigation to possible intermediates occurring in proton and electron transfer reactions. Also, we
consider only Heyrovsky-type reactions [34], with the exception of
3426
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
Table 1
Calculated free energies (in eV) for methanol decomposition intermediates at 1/9 ML surface coverage at standard conditions (298 K, 1 bar) studied on the closest-packed surface
of 12 metals
Re
Ru
Os
Co
Rh
Ir
Ni
Pd
Pt
Cu
Ag
Au
CH3O
CH2OH
CHOH
H2COOH
CHO
COH
H2COO
C(OH)2
CO
CHOO
COOH
OH
0.46
0.08
0.09
0.12
0.46
0.72
0.27
1.08
1.15
0.63
1.06
1.82
0.30
0.32
0.64
0.91
0.63
0.51
0.96
0.63
0.41
1.59
1.77
1.39
0.23
0.50
0.36
0.79
0.43
0.47
0.80
0.70
0.39
1.96
2.51
2.18
0.25
0.07
0.11
0.49
0.91
0.91
0.67
1.28
1.24
0.99
1.39
2.11
0.02
0.02
0.07
0.66
0.30
0.32
0.53
0.46
0.36
1.58
1.84
1.50
0.35
0.16
0.25
0.18
0.07
0.01
0.21
0.26
0.05
2.03
3.00
2.57
0.22
0.48
0.33
0.98
1.33
1.36
1.25
2.42
2.43
1.77
2.50
3.42
0.60
0.25
0.21
0.95
0.46
0.25
0.92
0.65
0.14
1.64
1.98
1.56
0.58
0.65
0.77
0.50
0.58
0.49
0.56
0.62
0.41
0.62
1.23
1.23
0.36
0.21
0.43
0.19
0.19
0.29
0.39
0.79
0.94
0.49
0.79
1.29
0.07
0.10
0.15
0.65
0.32
0.26
0.65
0.71
0.51
1.38
1.71
1.55
0.46
0.05
0.21
0.14
0.54
0.67
0.29
1.04
1.16
0.57
0.98
1.71
Pt values are in bold. CO2(g), H2O(l) and H2(g) are used as reference, as explained in the text. Zero-point energy and entropy corrections are included; the magnitude of these
corrections can be found in the appendix. Calculated free energies for closed-shell intermediates are: CH3OH = 0.11 eV; CH2O = 0.71 eV; HCOOH = 0.20 eV; CO2 = 0 eV. The
absolute DFT error for these molecules is small; for comparison, the standard table values are: CH3OH(g) = 0.05 eV; CH2O(g) = 0.57 eV; HCOOH(g) = 0.44 eV; CO2(g) = 0 eV.
CO þ OH ! CO2 ðgÞ þ Hþ þ e ;
which we also treat in our analysis. This step involves both CO–O
bond formation and a Heyrovsky proton–electron transfer step
[34]. For all steps, we do not include kinetic barriers. Through the
use of BEP lines, strong links between the thermodynamic and kinetic parameters of a reaction have been well-established in heterogeneous catalysis [35–37], and we expect the general trends
determined below to be valid representations of actual fuel cell
chemistry. In all steps, whenever a particular intermediate is a
closed-shell molecule, we use the gas-phase value for the free energy, which is appropriate since closed-shell molecules have weak
binding to surfaces. The gas-phase free energy, therefore, is lower
than that of the surface, primarily due to entropy effects, resulting
in very low surface coverages of the closed-shell species. We do
not consider hydrogen adsorption, since methanol decomposition
requires an overpotential, and thus the coverage of hydrogen on
the surfaces will be low. However, we note that, to avoid poisoning
the surface with hydrogen, DMFC anodes will require a potential
that is higher than that of the hydrogen fuel cell, regardless of the
catalyst used.
The intermediates and reaction paths we consider are shown in
Fig. 1. The relative free energies for each intermediate on the 12
surfaces studied are given in Table 1.
The first step in the reaction mechanism, i.e., the activation of
the methanol molecule, can take place via hydrogen abstraction
from either the carbon or the oxygen atoms. Further hydrogen
abstraction steps can create formaldehyde or hydroxymethylene
(CHOH), followed by formyl or COH. In the direct mechanism,
rather than stripping off the final hydrogen from COH or CHO to
H2COOH
CH3OH
form CO, a proton/electron pair is stripped off of a water molecule,
and the resulting OH group binds with the carbonaceous species to
form a di-oxygenated species (dihydroxycarbene (C(OH)2) or formic acid (HCOOH), as shown in Fig. 1). This hydroxyl addition is
followed by dehydrogenation to either formate (HCOO) or carboxyl
(COOH), with subsequent dehydrogenation to the final product. An
alternative direct pathway involves the stripping of a proton/electron pair from water and addition of the resulting hydroxyl to
formaldehyde, to form H2COOH, which can then be dehydrogenated to formic acid or dioxymethylene (H2COO). Dioxymethylene can
then be dehydrogenated to formate and ultimately to CO2. In the
indirect mechanism, on the other hand, CHO or COH are directly
dehydrogenated to CO. Water is dissociated separately on the surface to form OH, and the two surface species react together to form
CO2(g) in a manner analogous to the water-gas-shift reaction [38–
41].
We begin our discussion of preferred reaction pathways with Pt.
While the absolute activity of Pt for methanol oxidation to CO2 is
too low for use in practical DMFCs, it is generally regarded as the
most active monometallic catalyst for this reaction [11,42,43],
and it therefore forms a natural starting point for our analysis.
The relative free energies of the important intermediates (Fig. 1)
on Pt(1 1 1) are shown in Fig. 2. The most stable surface intermedi-
H2COO
CH3O
CH2O
CHO
HCOOH
CH2OH
CHOH
COH
C(OH)2
CO
CHOO
COOH
CO2
CO + OH
Fig. 1. Schematic representation of the reaction paths and possible intermediates
considered in this analysis. Green arrows indicate the indirect mechanism to CO2
formation. Arrows to the right also include the generation of a proton/electron pair
(not shown) from either the carbonaceous species or the surrounding H2O (the
latter implies the addition of a hydroxyl group). Dotted arrows indicate reactions
with no generation of a proton/electron pair. We note that liquid-phase reactions
can take place for H2CO and HCOOH, which are not included in this analysis [51].
Fig. 2. Free energies for the different intermediates on the Pt(1 1 1) surface, with
H2(g), CO2(g), and H2O(g) as a reference, as in Table 1. In red are the most stable
intermediates for each step. The energy levels of gas and liquid-phase molecules,
independent of the metal surface are shown in black. Intermediates with a higher
energy are shown in blue. The x-axis indicates how many proton/electron pairs
have been created from the original reactants (e.g., CH3OH + H2O?
HCOOH(g) + 4H+ + 4e).
3427
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
H2COO
H2COOH
Ag,Cu
Ag,Cu,Re,Ru,
Os,Co,Ni,Rh
CH3O
Au
CH3OH
Ir,Pd,Pt,Au
Ag,Cu
Ag,Cu,Co,Re,
Ni,Os,Ru
CH2OH
Ir, Pd,Pt
Au,Co,Re,Ni,
Au,Rh
CH2OOs,Ru,Rh CHO
Co,Re,Ni,
Os,Ru
CHOH
Ir, Pd,Pt
COH
Pt,Ir,Pd
HCOOH
Au,Ag,
Cu,Rh
CHOO
Au,Ag,
Cu,Rh
C(OH)2
Pt,Ir,Pd
COOHPt,Ir,Pd
CO2
Re,Ru,Os,Co,
Ni
CO Re,Ru,Os,Co,CO + OH
Ni
Fig. 3. The minimum-energy paths for methanol decomposition on all pure metals, with the potential-determining step indicated in red. Green arrows refer to steps in the
indirect mechanism, as in Fig. 1. For splitting of the fifth proton we have investigated three possible candidates: CHOO (formate), COOH (carboxyl), and CO* + OH*. For most
non-group 11 metals the latter is the most stable. Pt, Ir, and Pd prefer COOH, whereas Ag, Cu, and Au go through formate.
ates are depicted in red, and gas- or liquid-phase molecules, for
which the free energy is independent of the metal surface, are
shown in black. We find that CH2OH* is more stable than CH3O*.
This result implies that a higher potential is required to strip off
the hydroxyl proton than the carbon proton, in agreement with
experiments [3]. This result can be rationalized by analyzing the
binding characteristics of the two intermediates. CH2OH* binds
through carbon to the surface, whereas CH3O* binds through oxygen; these preferred binding geometries, in turn, result from the
fact that Pt binds carbon relatively strongly compared to oxygen
(see Table 1) [44]. The affinity of Pt for carbon also explains the stability of CHOH* and COH* (both of which bind through carbon), the
next intermediates in the most stable reaction pathway. As we will
show below, other metals with relatively strong carbon binding
also prefer intermediates that bind through the carbon atom
(e.g., CH2OH, COOH), whereas on metals with relatively strong oxygen binding, the preferred pathways involve intermediates that
bind to the surface through the oxygen atom (e.g., CH3O*).
It is also noteworthy that CO* is the most stable of all the intermediates on Pt; this stability, in turn, is the main reason for the
extensive CO poisoning problem that is often seen on Pt. As soon
as the potential is high enough to overcome the thermodynamic
barrier to activate methanol (eU = GCH2OHGCH3OH; U = (0.41–
0.11) eV/e = 0.30 V), all subsequent reaction steps leading to CO*
are downhill in free energy. The step following the formation of
CO*, however, has a very large thermodynamic barrier (Fig. 2). To
overcome this barrier, the surface must activate water, either by
forming OH* or COOH*, which in turn can form CO2. Of these two
possible intermediates, COOH* is more stable than a combination
of CO* and OH* adsorbed separately on Pt(1 1 1), as shown in
Fig. 2. Thus, the difference in free energies of CO* and COOH* defines the potential needed for oxidation through the indirect mechanism to occur. According to Table 1, this potential is
U = (0.51(0.41)) eV/e = 0.92 V versus SHE. At this potential, all
of the steps along the indirect mechanism from methanol to CO2
have a negative DG.
Instead of producing CO2 via CO (the indirect mechanism), CO2
can be created via HCOOH(g), CðOHÞ2 , or CH2OO* intermediates
(the direct mechanism). This path becomes spontaneous at a much
lower potential than the indirect mechanism, as the largest difference in free energy between successive intermediates on the lowest potential pathway is between CðOHÞ2 and COOH* (DG = 0.51–
0.14 eV = 0.37 eV, meaning the overpotential necessary for this
reaction to occur is 0.37 V). This observation suggests that at low
potentials (below 0.92 V), the activity seen on pure Pt comes primarily via the direct mechanism, in agreement with Wieckowski
et al. [11,45]. However, the direct mechanism has a relatively
low rate on Pt. Even though CO is not an intermediate in this path-
way, CO poisoning effectively reduces the rate, as we discuss
below.
3.2. Methanol decomposition on other metals
3.2.1. Direct reaction mechanism
Having seen that our analysis for Pt is consistent with earlier
studies, we apply this analysis to the rest of the metals in our database. Fig. 3 shows the lowest-free-energy reaction path for each of
the 12 metals for both the direct and the indirect mechanisms.
We also determine the step with the largest positive change in
free energy along the lowest-energy reaction pathway for each of
the 12 metals. The corresponding free energy change is the potential that must be applied for the overall reaction to take place spontaneously, hereafter referred to as the ‘onset potential.’ The onset
potential for the direct mechanism on each metal studied is shown
in Fig. 4. We find that, in all cases, the onset potential for the direct
mechanism is similar to – or less than – the potential required for
the indirect mechanism. A small calculated value of the onset potential for the direct mechanism, however, does not necessarily imply that the current density will be high, as the coverage of active
sites for the direct mechanism may be low because of CO poisoning, and because coverage effects are not directly included in our
free energy analysis. We note that such poisoning is not a problem
on the group 11 metals (Cu, Ag, and Au) because, although at
potentials relevant for the direct path it is still favorable to form
CO* on these surfaces, CO* is so weakly bound to these metals that
Fig. 4. The lowest potential needed in order for all steps along the direct path to be
downhill in free energy. The color code is related to the nature of the potentialdetermining step, as described in the legend in the figure. For Os, the final step
(CO + OH ? CO2) has a DG within 0.1 eV of that of the potential-determining step.
For Pt, the initial methanol activation (CH3OH ? CH2OH) has a DG within 0.1 eV of
that of the potential-determining step.
3428
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
further reaction to CO2 is thermodynamically favored under these
conditions (we note that CO2 formation is even more favorable
than simple CO desorption on these metals). For all other metals,
CO* poisoning can significantly reduce the net rate of methanol
electrooxidation via the direct mechanism. In those cases, the formation of CO* becomes possible as soon as the potential is high enough for initial methanol activation. However, further reaction to
CO2 is not necessarily negative in free energy – there is a potential
window where both the direct reaction to CO2 and the CO formation reaction are downhill in free energy but CO2 formation
through the indirect mechanism is not. Therefore, it is not always
possible to determine the rate of the direct path only by evaluating
the potential at which the route becomes feasible, and poisoning
effects must be considered.
To illustrate the above point, we use the following argument:
we first assume that there is a high coverage of CO on the surface.
This is reasonable, as CO is the most stable intermediate and is also
more stable than the end product, CO2(g), over a large potential
range. We further assume that whatever else is on the surface is
in equilibrium with CO*. The rate of the direct path is proportional
to the coverage of the intermediate in the step producing HCOOH,
either CHO* or COH*. Since we have assumed equilibrium with CO*,
the coverage of these reactive intermediates is proportional to the
difference in free energy between CO* and the intermediate in
question, e.g.,
plained in a recent paper by Abild-Pedersen et al. [46]. In that paper, the binding energies of CH13 where shown to scale directly
with the C* binding and that of OH* with O*. It is not surprising
then, that the linear scaling law also holds for methanol decomposition intermediates. The scaling allows us to write the necessary
electrochemical potential for the overall reaction to take place as
a function of only OH* and CO* adsorption free energies, GOH and
GCO . We consider the most important steps and use the linear scaling relations shown below (detailed derivations are given in the
appendix):
HCOH / expððGCOH GCO Þ=kTÞHCO
DG ¼ GCO þ GOH 0:20 eV
for GCO < GCOH. Also, the rate of, e.g., COH + H2O ? HCOOH + H+ + e
is proportional to
This analysis shows that the free energy of CO* ðGCO Þ and the
free energy of OH* ðGOH Þ are two key descriptors for the total reaction scheme shown in Fig. 3, and they can be used to estimate the
free energies for all other reaction intermediates. This allows us to
get a simplified overview of the reaction mechanism. The phase
space is mapped out in Fig. 5, where the different two-dimensional
regions (shown as functions of GCO and GOH ) correspond to partic-
HCOH expðDG=kTÞ;
where D ¼ GHCOOH GCOH . Assuming a high coverage of
CO(HCO 1), this in turn means that the rate is effectively proportional to
CH3 OHðgÞ ! CH2 OH þ Hþ þ e
GCH2 OH ¼ 0:59 GCO þ 0:94 eV
DG ¼ 0:59 GCO þ 0:83 eV
CH3 OHðgÞ ! CH3 O þ Hþ þ e
GCH3 O ¼ 1:08 GOH 0:05 eV
DG ¼ 1:08 GOH 0:16 eV
CH2 OðgÞ ! CHO þ Hþ þ e
GCHO ¼ 0:83 GCO þ 0:77eV
DG ¼ 0:83 GCO þ 0:06 eV
CO þ OH ! CO2 ðgÞ þ Hþ þ e
DG ¼ GCO GOH
HCOOHðgÞ ! CO þ OH þ Hþ þ e
expððGHCOOH GCO Þ=kTÞ;
the exponential of the difference between the free energies of CO*
and HCOOH(g). An analogous argument can be made for pathways
through other, non-CO intermediates, such as C(OH)2. Since CO is
strongly bound to all metals studied (with the exception of Cu,
Ag, and Au), and because of its high coverage on these metals, the
rate of the direct mechanism, and therefore the current, may be
very low even at potentials where all reaction steps are downhill
in free energy. To quantify the effect of this poisoning on the activity is beyond the scope of this work, however, as the current analyses are performed for low adsorbate coverages.
Among the non-group 11 metals studied, CO is most weakly
bound to Pt (see Table 1). This is in accord with the widely held
view that Pt is the best monometallic anode material of all the
transition metals. Based on this analysis, we find that the major
contribution to the activity at low potentials comes via the direct
path, but because of CO poisoning, the current is too small to be
of any practical use in a DMFC. For Ru, Re, and Os, a similar analysis
shows the possibility of poisoning by methoxy rather than CO at
low potentials, due to its relative stability as compared to CH2OH.
From the results above we see that the binding energy of CO*
plays an important role in determining the methanol electrooxidation current densities, and the importance of the OH* binding energy was also illustrated in the case of Pt. In fact, the phase space
and the different mechanisms on the different metals can, to a
large extent, be described by the surface’s affinity for these two
intermediates. All relative free energies of intermediate species
binding through the carbon atom scale linearly with the relative
free energy of CO* on the surface, and all free energies of intermediates binding through the O atom scale linearly with that of OH*.
We note that this scaling relationship has been observed and ex-
Fig. 5. The potential-determining steps for the direct mechanism plotted with GCO
and GOH as descriptors. These two binding energies describe the total reaction
landscape and can be used to estimate the potential through the linear relations as
described in the text. The potential-determining steps for each region are as
follows: (a) CHO ? HCOOH; (b) CH2O ? H2COOH; (c) CH2OH ? H2CO; (d)
H3COH ? CH2OH; (e) H3COH ? H3CO; (f) H3CO ? H2CO. Iso-potential lines are
also included for reference. Each line represents a difference of 0.1 V.
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
ular potential-determining steps. The lines in Fig. 5 represent the
border between two regimes with different potential-determining
step. We note that the onset potential of reaction is only one
important parameter in determining the efficiency of a fuel cell.
For a highly active electrocatalyst, it is also important to avoid
CO* poisoning. In this respect, metals farther to the right (higher
GCO ) are better, with Pt being the non-group 11 metal with the
weakest CO binding. The map also shows that there is an inherent
limit to the improvement of the catalyst performance achievable
via materials design. A catalyst with strong CO* binding (and thus
strong binding of all C-bound intermediates) will dehydrogenate
methanol easily, but the oxidation of CO will require a high electrochemical potential, and CO will block sites for the direct mechanism to take place. Weak CO* binding, on the other hand,
prevents site blockage through CO; however, it requires a higher
electrochemical potential for methanol activation. This Sabatierlike analysis also applies to OH* binding. Strong OH* binding permits water activation at low potentials, but it also makes it difficult
to remove OH* (or formate, in some cases). Moreover, the optimal
value of either one of these descriptors (i.e., the value needed to
obtain the lowest onset potential for this reaction) depends, to
some extent, on the value of the other descriptor. This interaction,
in turn, creates a range of values that give the lowest potential for
methanol activation, rather than just a single optimum, as is generally the case in one-dimensional ‘volcano’ plots [47].
According to this analysis, the lowest potential necessary for the
direct mechanism to have a negative DG for all steps occurs on Ir,
Ni, Rh, and Pt (see Fig. 4). Each of these metals has a different potential-determining step, indicating that several of the regions
within the entire phase space can be explored with monometallic
catalysts. Unfortunately, none of these catalysts are found to have
sufficiently high current densities to be useful at low potentials,
probably because of CO* poisoning. On Ag, Cu, and Au, where the
reaction rate will not be affected by CO* poisoning, the onset potential for methanol oxidation is much higher and may be affected
by other problems not included in this analysis (such as the dissolution potential of the metal). The direct mechanism, therefore, has
inherent limitations as to the minimum electrochemical potential
needed for the reaction to take place, and even at these minimum
electrochemical potentials, other factors may impede the reaction.
Based on this analysis, it seems necessary to also consider the possibility of tuning the catalyst performance through the indirect
mechanism, where the CO* poison is also a reactant.
3.2.2. Indirect reaction path
Having established that anodes in DMFCs must be able to catalyze the oxidation of CO in order for a reasonable rate of reaction to
be observed, we now turn to the indirect path to CO2. As with our
analysis of Pt (see discussion above), we first calculate the lowest
potential at which the reaction can proceed downhill in all steps
from methanol to CO2 on the different metals studied (see
Fig. 6). For Pt, Pd, Ir, Rh, and Au, there is a higher onset potential
for the indirect mechanism than for the direct mechanism, primarily because of the poor activity of these metals for the activation of
water. For Ag and Cu, the higher onset potential for the indirect
method is due to the activation of formaldehyde to a carbon-bound
species. The other metals studied have potential-determining steps
that are shared by both the direct and indirect mechanisms, and
therefore the onset potential for both pathways is identical. On
these metals, water activation to hydroxyl is relatively easy (making the indirect reaction path feasible). However, on some of these
metals, e.g., Co and Ni, further hydroxyl oxidation to form O* on the
surface may occur. This oxygen is too strongly bound to oxidize
CO*, meaning that these metals are too active toward water splitting and the surface will be poisoned by O*. Oxygen poisoning will
be a problem for all metals with very strong OH* binding (i.e., very
3429
Fig. 6. The lowest potential needed in order for all steps along the indirect path to
be downhill in free energy. The color code is related to the nature of the potentialdetermining step, shown in the legend in the figure. For Os, the final step
(CO + OH ? CO2) has a DG within 0.1 eV of that of the potential-determining step.
negative GOH), such as Ni, Os, Re, Ru, and Co. For all of these metals,
water will favor the splitting off of a second proton as soon as OH*
is formed. We note that oxidation of CO* by O* is not explicitly included in our analysis since it is not a proton-transfer reaction, and
it cannot be treated in the same manner as the other reactions
steps.
As before, we can investigate trends in the indirect pathway
using the two reactivity descriptors: GCO and GOH . The same analysis as for the direct mechanism is used here, employing the linear
scaling developed earlier. The correlations used are shown below.
CH3 OHðgÞ ! CH2 OH þ Hþ þ e
GCH2 OH ¼ 0:59 GCO þ 0:94 eV
DG ¼ 0:59 GCO þ 0:83 eV
CH3 OHðgÞ ! CH3 O þ Hþ þ e
GCH3 O ¼ 1:08 GOH 0:05 eV
DG ¼ 1:08 GOH 0:16 eV
H2 OðlÞ ! OH þ Hþ þ e
DG ¼ GOH
CO þ OH ! CO2 ðgÞ þ Hþ þ e
DG ¼ GCO GOH
Based on these equations, the phase space is mapped out in Fig. 7.
The picture is simpler than in the case of the direct mechanism,
since the activation of water is clearly potential-determining in
the region of weak OH* binding. It can be seen that Rh, Ni, and Cu
are closest to the optimum onset potential in three different potential-determining regions (see Fig. 6). As noted previously, however,
the onset potential for the indirect pathway on all metals is always
greater than or equal to the onset potential for the direct pathway.
One effect not taken into account in this analysis is the effect of
hydrogen-bonding in OH-containing intermediates with the water
in the electrolyte, which would make the free energy relatively
more negative for all OH-containing species. This would have the
effect of moving the lines downward with respect to the y-axis. If
the magnitude of this effect is 0.3 eV [48], Fig. 7 shows that the
pure metals are present in each of the four regions defined for
the indirect mechanism whether or not this correction is included.
The regions mentioned above clearly outline the task for a
DMFC anode: first, it has to be active towards forming CO* from
methanol (either through methoxy or H2COH); second, it must
activate water, forming OH*; and third, it has to bind CO* and
OH* only moderately, to avoid CO* poisoning or OH* poisoning/oxidation. As with the direct mechanism, the lowest onset potential
occurs at a set of free energies representing a compromise between
the three necessary tasks. For instance, for the latter two tasks, Cu
3430
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
bound CO*), by the ability to activate water (owing to weakly
bound OH*) or by poisoning by CO* or OH*/O* (owing to very strong
binding of CO* or OH*, respectively). No pure metal catalyst can
simultaneously optimize all of these factors. Hence, there is no
obvious candidate for a good anode material among the pure metals. In order to design a better catalyst, one has to look for the bifunctional surfaces such as the well-known PtRu alloy. We believe
that the volcano behavior presented here can provide meaningful
insights for the reaction mechanism at the anode, and it can be
very helpful in the design of alloy catalysts with improved performance. We are currently working on identifying promising bimetallic surfaces which may improve methanol electrooxidation
catalysis.
Acknowledgements
Work at the University of Wisconsin was funded in part by the
Department of Energy, Office of Basic Energy Sciences, the National
Science Foundation, and the University of Wisconsin. Supercomputing time at NERSC, PNNL, and NCCS and ORNL is gratefully
acknowledged. Use of the Center for Nanoscale Materials at ANL
was supported by the US Department of Energy, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. JR would
like to thank CAMD, which is funded by the Lundbeck foundation.
Fig. 7. The potential-determining steps for the direct mechanism plotted with GCO
and GOH as descriptors. These two binding energies describe the total reaction
landscape and can be used to estimate the potential through the linear relations, as
described in the text. The potential-determining steps for each region are as
follows: (a) H2O ? OH; (b) CO + OH ? CO2; (c) H3CO ? H2CO; (d) H2CO ? CHO.
Iso-potential lines are also included for reference. Each line represents a difference
of 0.1 V.
Appendix
See Fig. A1, Table A1.
is known to be effective, as it is the catalyst of choice for the watergas-shift reaction. However, in the case of water-gas shift, CO is
present in gaseous form, meaning that the catalyst only has to bind
CO and not activate methanol, which is more easily done on a metal with a higher CO binding energy than that of Cu. These delicate
compromises are nicely illustrated in the case of the well-known
Pt–Ru bi-functional catalyst, where easy activation of water appears to be combined with weaker binding of CO* to produce improved catalysts for methanol electrooxidation [49,50]. Other
alloys may also prove to have similar or even better properties; indeed, one could quantitatively study these trends on a series of
transition metals and alloys, using the descriptors introduced in
this study as screening criteria for new materials that possess
superior properties for methanol electrooxidation.
4. Conclusion
We have investigated the anode reaction in direct methanol fuel
cells, namely the oxidation of methanol to CO2, based on a density
functional theory-determined database of free energies for 16
intermediates on 12 transition metals. We are able to rationalize
earlier experimental results for Pt. We have also shown that the
catalyst performance for both the direct and indirect mechanisms
can be characterized by two reactivity descriptors: GCO and GOH .
We find that because of CO poisoning, the direct mechanism will
only result in low current densities on non-group 11 transition
metals at low potentials. Pt has the lowest onset potential for the
direct mechanism among these metals. We have mapped out the
phase space using linear scaling relations between the free energies of all intermediates binding to the surface via carbon and that
of CO*, and between the free energies of all intermediates binding
to the surface via oxygen and that of OH*.
For the indirect mechanism, we find that the onset potential is
limited by either the ability to activate methanol (owing to weakly
Fig. A1. (A) Free energies of O-bound species plotted against GOH (B) Free energies
of C-bound species plotted against GCO Free energies of all surface species binding
through O can be described as a linear function of GOH ; Free energies of all surface
species binding through C can be described as a linear function of GCO .
3431
P. Ferrin et al. / Surface Science 602 (2008) 3424–3431
Table A1
Entropy and zero-point energy of reactants, products and intermediates
Intermediate
TS
ZPE
Associated reaction
TDS
DZPE
CO2(g)
H2(g)
H2O(l)
CH3OH(g)
CH2O(g)
HCOOH(g)
CH2OH*
CH3O*
CHOH*
CHO*
COH*
CO*
OH*
O*
C(OH)2
COOH*
CHOO*
H2COO*
H2COOH*
0.66
0.40
0.67
0.74
0.68
0.77
0.05
0.03
0.03
0.03
0.03
0.05
0.04
0.04
0.05
0.07
0.05
0.02
0.03
0.31
0.27
0.56
1.36
0.70
0.70
1.10
1.04
0.78
0.43
0.47
0.17
0.31
0.07
0.77
0.61
0.61
0.86
1.06
CO2(g) + 3H2(g) ? CH3OH(g) + H2O(l)
CO2(g) + 2H2(g) ? CH2O(g) + H2O(l)
CO2(g) + H2(g) ? HCOOH(g)
CO2(g) + 5/2H2(g) ? CH2OH* + H2O(l)
CO2(g) + 5/2H2(g) ? CH3O* + H2O(l)
CO2(g) + 2H2(g) ? CHOH* + H2O(l)
CO2(g) + 3/2H2(g) ? CHO* + H2O(l)
CO2(g) + 3/2H2(g) ? COH* + H2O(l)
CO2(g) + H2(g) ? CO* + H2O(l)
H2O(l) ? 1/2H2(g) + OH*
H2O(l) ? H2(g) + O*
CO2(g) + H2(g) ? C(OH)2*
CO2(g) + 1/2H2(g) ? COOH*
CO2(g) + 1/2H2(g) ? CHOO*
CO2(g) + H2(g) ? H2COO*
CO2(g) + 3/2H2(g) ? H2COOH*
0.45
0.12
0.29
0.95
0.97
0.76
0.56
0.56
0.34
0.43
0.22
1.01
0.78
0.80
1.04
1.24
0.79
0.40
0.12
0.66
0.61
0.49
0.27
0.31
0.15
-0.11
-0.22
0.19
0.16
0.17
0.28
0.34
Entropies and ZPEs of gas-phase molecules are obtained from standard tables and from calculations, respectively. Zero-point energies and entropies of bound intermediates
are calculated based on calculated vibrational spectra. TS is the standard entropy times temperature; ZPE is zero-point energy. TDS is the temperature times the change in
entropy for a reaction; DZPE is the change in zero-point energy for the reaction. The energy of H2O(l) is obtained via the H2O(g) at the equilibrium pressure 0.035 bar. All
values are at standard conditions (P = 1 bar, T = 298 K).
References
[1] A. Hamnett, Catalysis Today 38 (1997) 445.
[2] E. Reddington, A. Sapienza, B. Gurau, R. Viswanathan, S. Sarangapani, E.S.
Smotkin, T.E. Mallouk, Science 280 (1998) 1735.
[3] T.D. Jarvi, E.M. Stuve, Fundamental aspects of vacuum and electrocatalytic
reactions of methanol and formic acid on platinum surfaces, in: J. Lipkowski, P.
Ross (Eds.), The Science of Electrocatalysis on Bimetallic Surfaces, Wiley-VCH,
Inc., 1998, p. 75.
[4] P. Liu, A. Logadottir, J.K. Nørskov, Electrochimica Acta 48 (2003) 3731.
[5] S. Surampudi, S.R. Narayanan, E. Vamos, H. Frank, G. Halpert, A. LaConti, J.
Kosek, G.K.S. Prakash, G.A. Olah, Journal of Power Sources 47 (1994) 377.
[6] S.R. Brankovic, J.X. Wang, R.R. Adzic, Electrochemical and Solid State Letters 4
(2001) A217.
[7] P. Waszczuk, J. Solla-Gullon, H.S. Kim, Y.Y. Tong, V. Montiel, A. Aldaz, A.
Wieckowski, Journal of Catalysis 203 (2001) 1.
[8] C. Lu, C. Rice, R.I. Masel, P.K. Babu, P. Waszczuk, H.S. Kim, E. Oldfield, A.
Wieckowski, Journal of Physical Chemistry B 106 (2002) 9581.
[9] P. Strasser, Journal of Combinatorial Chemistry 10 (2008) 216.
[10] N.M. Markovic, H.A. Gasteiger, P.N. Ross, X.D. Jiang, I. Villegas, M.J. Weaver,
Electrochimica Acta 40 (1995) 91.
[11] D. Cao, G.Q. Lu, A. Wieckowski, S.A. Wasileski, M. Neurock, Journal of Physical
Chemistry B 109 (2005) 11622.
[12] J. Kua, W.A. Goddard, Journal of the American Chemical Society 121 (1999)
10928.
[13] V.S. Bagotzky, Y.B. Vassiliev, O.A. Khazova, Journal of Electroanalytical
Chemistry 81 (1977) 229.
[14] Y.X. Chen, A. Miki, S. Ye, H. Sakai, M. Osawa, Journal of the American Chemical
Society 125 (2003) 3680.
[15] N.M. Markovic, P.N. Ross, Surface Science Reports 45 (2002) 121.
[16] J.K. Nørskov, J. Rossmeisl, A. Logadottir, L. Lindqvist, J.R. Kitchin, T. Bligaard, H.
Jonsson, Journal of Physical Chemistry B 108 (2004) 17886.
[17] J. Greeley, J.K. Nørskov, L.A. Kibler, A.M. El-Aziz, D.M. Kolb, Chemphyschem 7
(2006) 1032.
[18] J. Rossmeisl, K. Dimitrievski, P. Siegbahn, J.K. Nørskov, Journal of Physical
Chemistry C 111 (2007) 18821.
[19] J. Greeley, T.F. Jaramillo, J. Bonde, I.B. Chorkendorff, J.K. Norskov, Nature
Materials 5 (2006) 909.
[20] A.U. Nilekar, Y. Xu, J.L. Zhang, M.B. Vukmirovic, K. Sasaki, R.R. Adzic, M.
Mavrikakis, Topics in Catalysis 46 (2007) 276.
[21] Z.H. Gu, P.B. Balbuena, Journal of Physical Chemistry A 110 (2006) 9783.
[22] J. Greeley, J.K. Nørskov, M. Mavrikakis, Annual Review of Physical Chemistry
53 (2002) 319.
[23] B. Hammer, L.B. Hansen, J.K. Nørskov, Physical Review B 59 (1999) 7413.
[24] J. Neugebauer, M. Scheffler, Physical Review B 46 (1992) 16067Lp.
[25] L. Bengtsson, Physical Review B 59 (1999) 12301.
[26] D. Vanderbilt, Physical Review B 41 (1990) 7892.
[27] D.J. Chadi, M.L. Cohen, Physical Review B 8 (1973) 5747.
[28] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh, C.
Fiolhais, Physical Review B 46 (1992) 6671.
[29] J. Greeley, M. Mavrikakis, Surface Science 540 (2003) 215.
[30] CRC Handbook of Chemistry and Physics, 76th ed., CRC Press, New York, 1996.
[31] G.S. Karlberg, J. Rossmeisl, J.K. Nørskov, Physical Chemistry Chemical Physics 9
(2007) 5158.
[32] J. Rossmeisl, A. Logadottir, J.K. Nørskov, Chemical Physics 319 (2005) 178.
[33] J. Rossmeisl, Z.W. Qu, H. Zhu, G.J. Kroes, J.K. Nørskov, Journal of
Electroanalytical Chemistry 607 (2007) 83.
[34] J. Lipkowski, P.N. Ross (Eds.), Electrocatalysis, vol. 5, VCH Publishers, New
York, 1998, p. 197.
[35] J.K. Nørskov, T. Bligaard, A. Logadottir, S. Bahn, L.B. Hansen, M. Bollinger, H.
Bengaard, B. Hammer, Z. Sljivancanin, M. Mavrikakis, Y. Xu, S. Dahl, C.J.H.
Jacobsen, Journal of Catalysis 209 (2002) 275.
[36] Y. Xu, A.V. Ruban, M. Mavrikakis, Journal of the American Chemical Society 126
(2004) 4717.
[37] R. Alcala, M. Mavrikakis, J.A. Dumesic, Journal of Catalysis 218 (2003) 178.
[38] A.A. Gokhale, J.A. Dumesic, M. Mavrikakis, Journal of the American Chemical
Society 130 (2008) 1402.
[39] C.V. Ovesen, B.S. Clausen, B.S. Hammershoi, G. Steffensen, T. Askgaard, I.
Chorkendorff, J.K. Nørskov, P.B. Rasmussen, P. Stoltze, P. Taylor, Journal of
Catalysis 158 (1996) 170.
[40] N. Schumacher, A. Boisen, S. Dahl, A.A. Gokhale, S. Kandoi, L.C. Grabow, J.A.
Dumesic, M. Mavrikakis, I. Chorkendorff, Journal of Catalysis 229 (2005) 265.
[41] L.C. Grabow, A.A. Gokhale, S.T. Evans, J.A. Dumesic, M. Mavrikakis, Journal of
Physical Chemistry C 112 (2008) 4608.
[42] J. Greeley, M. Mavrikakis, Journal of the American Chemical Society 126 (2004)
3910.
[43] J. Greeley, M. Mavrikakis, Journal of the American Chemical Society 124 (2002)
7193.
[44] D.C. Ford, Y. Xu, M. Mavrikakis, Surface Science 587 (2005) 159.
[45] G.Q. Lu, W. Chrzanowski, A. Wieckowski, Journal of Physical Chemistry B 104
(2000) 5566.
[46] F. Abild-Pedersen, J. Greeley, F. Studt, J. Rossmeisl, T.R. Munter, P.G. Moses, E.
Skulason, T. Bligaard, J.K. Nørskov, Physical Review Letters (2007) 99.
[47] P. Sabatier, Berichte der Deutschen Chemischen Gesellschaft 44 (1911) 1984.
[48] J. Rossmeisl, J.K. Nørskov, C.D. Taylor, M.J. Janik, M. Neurock, Journal of
Physical Chemistry B 110 (2006) 21833.
[49] M.T.M. Koper, T.E. Shubina, R.A. van Santen, Journal of Physical Chemistry B
106 (2002) 686.
[50] F.B. de Mongeot, M. Scherer, B. Gleich, E. Kopatzki, R.J. Behm, Surface Science
411 (1998) 249.
[51] T.H.M. Housmans, A.H. Wonders, M.T.M. Koper, Journal of Physical Chemistry
B 110 (2006) 10021.