1. Introduction
While promising non-noble metal catalysts for the oxygen reduction reaction (ORR) in proton-exchange membrane (PEM) hydrogen fuel cells have been synthesized over the years [
1,
2,
3,
4,
5], their stability in fuel cells remains the main obstacle to their widespread use [
6,
7]. One of the most promising non-noble metal catalysts synthesized to date is FeN
x-doped disorganized carbon [
8,
9,
10,
11]. There are several experimental and theoretical pieces of evidence that the Fe atom is the site where the ORR takes place [
11,
12,
13,
14,
15,
16,
17,
18,
19]. It has been observed that this type of catalyst suffers from a decrease of almost half of its activity in a few hours of operation in fuel cells, followed by a much slower decrease thereafter. The current delivered by the fuel cell versus time can be fitted by a double exponential decay [
20]. Few hypotheses have been put forward to explain the first rapid decay of catalytic activity. These include demetallation of the metal catalytic sites [
20,
21,
22,
23] and chemical reactions with H
2O
2 [
24,
25,
26,
27]. The slower decay has not attracted as much interest as the fast one to date. Recent simulation work suggests that planar M
3(C
6O
6)
2 [
28] and M
3(C
6S
3O
3)
2 [
29] structures, where M is a transition metal, may also be promising candidates but these have not yet passed the test of experiment.
There are indications that the fluorination of materials in acidic media improves their oxidative stability. Examples of such systems include Nafion ionomer, Pt/C, and platinum group metal-free catalysts for PEM fuel cells [
30,
31,
32]. Recently, we also fluorinated a highly active FeN
x-doped carbon catalyst in the hopes that fluorine would increase its stability in PEM fuel cells [
33]. However, even after a short (2 min) exposure to a room-temperature F
2:N
2 (1:1;
vol.) gas stream, fluorination considerably inhibited the catalyst performance in H
2/O
2 PEM fuel cells. Even if these experiments did not yield the expected results, they enabled several important observations to be made regarding the properties of the catalytic material under study:
- (1)
The catalytic activity of the fluorinated Fe/N/C catalyst became similar to that of Fe-free nitrogen-doped carbon catalysts;
- (2)
The XPS F1s spectra revealed that most Fe sites were associated with a single F atom and fewer were associated with two F atoms;
- (3)
A total of 70% of the initial activity could be recovered after a heat treatment of the F-poisoned catalyst at 900 °C in Ar.
The observations made in the context of these fluorination experiments have in fact provided a unique opportunity to improve our understanding of the nature of our FeNx-doped carbon catalysts and of the decay mechanisms of their catalytic activity in PEM fuel cells. In order to support and deepen the conclusions of our experimental study, in this paper we report density functional theory (DFT) calculations, based on the current understanding of the atomic structure of the catalytic sites and processes, and study the catalytic properties of these sites in the absence/presence of adsorbed fluorine.
Several theoretical studies have already focused on MN
x-doped carbon catalysts, most often Fe [
34,
35,
36,
37,
38], Co [
35], Mn [
35,
39], and Ni [
35]. Per the indications of several experimental studies [
34,
39,
40,
41], they conclude that the catalytic site is, specifically, the M atom within a functional group MN
x embedded in a planar carbon layer. It is generally thought that the ORR catalyzed on these sites follows the four-electron exchange process [
42,
43,
44,
45]
where * denotes the adsorption site and the labels I to VI refer to the six reaction steps.
For several MN
x-doped carbon structures, it has been found that, at low enough potentials, the free energy of each step of the reaction sequence (1) decreases uniformly from the first to the last step, indicating that the reaction sequence (1) is thermodynamically viable at these potentials. Other possible pathways, such as those involving spontaneous O
2 dissociation or H
2O
2 formation, are less likely due to the increase in free energy at some stage of the process [
24]. Several DFT studies have also been carried out for catalysts without metal [
45,
46,
47,
48,
49,
50,
51]. These generally consider N-doped carbon structures and assume that the reaction sequence (1) still takes place at low enough potentials. These catalysts appear to be thermodynamically viable for some carbon sites near a nitrogen atom. However, O
2 adsorbs weakly or not at all on the catalytic sites (step II in the sequence (1)). This characteristic likely explains, at least in part, the much lower activity of these sites compared to the higher activity obtained with metal sites.
In a recent work, we theoretically studied the fluorination of two single-layer porous FeN
4-doped carbon structures, one with pyrrolic nitrogen atoms and the other with pyridinic nitrogen atoms at the FeN
4 sites, and we assumed that the catalytic reaction took place through the sequence (1) [
52]. Subsequent work has investigated ORR for various adsorbates bound to several transition metals on an MN
4 site [
53,
54]. However, actual catalysts most likely contain many embodiments of MN
x moieties as well as M-free N atoms in carbon layers, and involve more than one carbon layer. In order to provide a more complete picture of the fluorination process and of its influence on ORR, we performed DFT calculations for nine additional atomic structures of FeN
x-doped carbon sites with x between 1 and 4 as well as for two N-doped metal-free carbon structures. We also examined the possibility of whether the ORR can be catalyzed on an Fe site between two parallel carbon layers in the presence of a F atom bound on Fe on the opposite side.
It is generally believed that the catalytic sites for the ORR of the FeN
x-doped carbon materials consist of planar FeN
x moieties located in a carbon structure which is generally approximated by a single carbon layer.
Figure 1 shows some of the possible embodiments of such structures. Of course, the set of structures shown in
Figure 1 is not exhaustive. Many variants of each structure are possible, such as, for example, pores in the carbon layer (as in
Figure 1j), the FeN
x moiety being near the edge of the carbon layer (as in
Figure 1g,i), and N atoms being randomly distributed in the carbon layer (as in
Figure 1e). Also, the N atoms surrounding the Fe ion may be either of a pyridinic type (
Figure 1a–i) or of a pyrrolic type (
Figure 1j). We will consider the 10 structures shown in
Figure 1, expecting that they will be representative of the main effects of fluorination on the properties of the catalysts. For the purpose of the following discussion, special attention will be paid to the structure in
Figure 1j. For this structure, the F–N bond length is 2.00 Å and the N–Fe–N angles are 197.19° and 82.81°.
It can be found in the literature that the free energy at zero potential of the steps of the reaction sequence (1) for the pristine structures of
Figure 1c [
55], 1h [
56], and 1j [
52] is uniformly descending at each reaction step (see
Section 2.1). The structure of
Figure 1b has also been investigated and was found to be not thermodynamically viable [
55] (because the free energy presents a minimum at step V of the reaction sequence (1)—see
Section 2.1). The other sites shown in
Figure 1 have, to the best of our knowledge, never been investigated so far.
Figure 2 shows the two metal-free nitrogen-doped carbon catalytic sites considered in this work. As reported above, the activity of the fluorinated Fe/N/C catalysts became similar to that of Fe-free nitrogen-doped carbon catalysts. It is assumed that the reaction sequence is given by (1) where * now denotes an active carbon site. Other nitrogen-doped carbon structures have been investigated and shown to be potential catalysts for ORR [
46,
47,
48,
49]. The structures shown in
Figure 2 were selected for this work because they turn out to have the lowest formation energies [
45] and are, therefore, most likely to be found in actual catalysts. Only two of these structures are considered in this work because we show that their catalytic properties are immune to fluorination and this result is sufficient for our purpose.
3. Computational Methods
All DFT calculations reported here were done using the Vienna ab initio software package (VASP) [
61,
62,
63,
64]. The calculations were performed using the generalized gradient approximation (GGA) with the Perdew–Burke–Ernzerhof (PBE) functional [
65]. The convergence criterion on the relative energy was set to 10
−5 and the plane wave energy cut-off was set to 500 eV for all calculations. The Brillouin zone was sampled on regular 4 × 4 × 4 gamma grids. A graphene sheet with cell dimensions of a = 20.22 Å and b = 14.88 Å was used as a model for the carbon support. A void of 15 Å was included in the normal direction to avoid interactions between the periodic FeN
x-doped carbon layers. The doped carbon structures were created by substituting carbon atoms of the graphene sheet by FeN
x groups or by N atoms in the case of metal-free catalysts. The positions of all atoms were fully relaxed, except in the case of the two carbon layers, where the positions of the carbon atoms were fixed to prevent the planes from moving relative to each other. However, for the calculation of the activation energy of O
2 transiting between the two planes, in relation to
Figure 5a, we used constraints where the edges of the planes were fixed along the
x-axis while keeping the edges along the
y-axis fixed, and vice versa. The activation energy was almost the same (1.5 eV) in both cases. The binding energy of an adsorbate on a given site was calculated using
where
is the energy of the carbon-doped catalyst with the adsorbate,
is the energy of the catalyst alone, and
is the energy of the adsorbate far from the catalyst. The energy of
is taken as half the energy of the H
2 molecule, since H
2 is at an equilibrium with its dissociated form
at the anode [
43]. The molecules of the gas phases considered in this work, namely O
2, H
2, and F
2, are assumed to be non-interacting with each other, which implies that only single molecules have been considered. Each step of the catalytic sequence (1) corresponds to a free energy given by
where
is the energy of the structure per cell,
ZPE is the zero point energy,
TdS is the entropy term, and
is the solvation energy arising by the aqueous medium. As was done in some of our previous works [
52,
54], for simplicity we assumed that the sum of the last three contributions nearly cancels, in agreement with [
43,
66]. However, corrections were brought to the intermediate state
and
of the reaction sequence (1), which were inferred to be +0.4 and −0.6 eV, respectively [
43].
4. Conclusions
We used DFT to examine the consequences of fluorination of the FeNx-doped and N-doped carbon catalysts used for ORR at the cathode of H2/O2 fuel cells. The main objectives of these calculations were to rationalize some of the experimental observations and to verify our conceptual representation of the catalytic sites and processes. We have considered several moieties of catalytic sites of FeNx-doped carbon with x ranging from 1 to 4. Most of them seem to be suitable catalysts for ORR because the free energy of the supposed catalytic sequence decreases regularly at zero potential. When the FeNx sites are located on a single graphene layer, it turns out that F2 binds to Fe at FeNx sites, with a binding energy of approximately −2 eV, but is subject to dissociation, leaving a single F on Fe with a binding energy of approximately −4 eV, which is stronger than the typical binding energy of O2 on Fe. In these conditions, ORR cannot happen on the F-poisoned FeNx side, but is still possible on the other side of the F–FeNx site, even transforming some otherwise poor un-poisoned FeNx catalytic configurations into better F–FeNx active ones. In addition, two F atoms can also bind to Fe on both sides of the carbon layer with almost twice the binding energy of a single F. When this happens, the Fe site is completely poisoned on both sides and is no longer able to catalyze ORR.
The occurrence of single graphene layers in actual catalysts is probably quite exceptional. Those are certainly better represented by several stacks of disorganized graphene layers forming a network of connected micropores and mesopores. Therefore, we have also examined the double graphene layer case where there is a second parallel carbon layer at a distance of 3.6 Å from the upper carbon layer carrying the FeNx sites. We found that O2 adsorption on Fe between the two carbon layers is stable and that OOH dissociates spontaneously into O adsorbed on Fe and OH adsorbed on the opposite carbon layer. Because O2 adsorption increases the free energy by about 1 eV (and needs an activation energy of around 1.5 eV) relative to free O2, the catalytic process is unlikely in this case, even though the free energy of subsequent steps decreases monotonically. On the other hand, we found that F2 can adsorb on Fe between the two carbon layers without energy expenditure, making this process more likely than for O2. These results suggest complete poisoning of the FeNx sites through extensive fluorination of the catalyst, in agreement with the experimental observations.
We then focused on the residual catalytic activity after fluorination by considering Fe-free N-doped carbon armchair and zigzag structures for which previous DFT calculations suggested a viable catalytic process although O2 hardly adsorbs on these structures. For both structures, the active catalytic site is a carbon atom near a N atom. We found that these catalytic structures are not poisoned by F or F2, thus justifying a residual ORR catalytic activity similar to that of the Fe-free catalysts observed for fluorinated Fe/N/C catalysts.
Finally, we provided an explanation for the recovery of ORR upon heating to 900 °C after fluorination. This explanation is based on the presence of radicals or small molecules released from the catalyst surface upon heat treatment. Most of the calculations presented in this work are based on free energy levels that only indicate whether a catalytic process is thermodynamically viable or not. A more thorough study would include the determination of activation energies. These calculations are very computationally demanding and will be the subject of future work.