Effect of Water on the Manifestation of the Reaction Selectivity of Nitrogen-Doped Graphene Nanoclusters toward Oxygen Reduction Reaction.

We investigated the selectivity of N-doped graphene nanoclusters (N-GNCs) toward the oxygen reduction reaction (ORR) using first-principles calculations within the density functional theory. The results show that the maximum electrode potentials (U Max) for the four-electron (4e-) pathway are higher than those for the two-electron (2e-) pathway at almost all of the reaction sites. Thus, the N-GNCs exhibit high selectivity for the 4e- pathway, that is, the 4e- reduction proceeds preferentially over the 2e- reduction. Such high selectivity results in high durability of the catalyst because H2O2, which corrodes the electrocatalyst, is not generated. For the doping sites near the edge of the cluster, the value of U Max greatly depends on the reaction sites. However, for the doping sites around the center of the cluster, the reaction-site dependence is hardly observed. The GNC with a nitrogen atom around the center of the cluster exhibits higher ORR catalytic capability compared with the GNC with a nitrogen atom in the vicinity of the edge. The results also reveal that the water molecule generated by the ORR enhances the selectivity toward the 4e- pathway because the reaction intermediates are significantly stabilized by water.

Introduction

Fuel cells have recently attracted much attention as eco-friendly energy systems. In a fuel cell, energy is obtained through an electrochemical reaction of hydrogen fuel with oxygen. The main problem impeding the widespread implementation of fuel cells is that the oxygen reduction reaction (ORR) at the cathode is less efficient than the hydrogen oxidation reaction at the anode. Currently, platinum-based materials are being commonly used as effective catalysts for the ORR. However, platinum-based materials have many problems, including high cost and low durability. Therefore, a platinum-free catalyst with the high ORR activity is required. N-doped graphene is expected to be one such electrocatalyst materials. Recently, N-doped graphene has been experimentally confirmed to exhibit high ORR activity.[1−9] Although various aspects of the ORR mechanism have been proposed, such as the type of C–N bonding, the presence of local active sites, and the effects of edges, little consensus exists about the mechanism that N-doped graphene follows to promote the ORR at the atomic scale. Many researchers have suggested on the basis of experimentation that the N-doped graphene containing pyridinic-N exhibits high ORR activity.[2−6] Wu et al. have reported that N-doped graphene can be synthesized from g-C3N4 and have suggested that the pyridinic-N tends to be the most active N functional group to facilitate ORR at low overpotential.[4] Zeng has suggested that highly graphitized carbon with high contents of pyridinic-N and graphitic-N at the edge exhibits higher ORR activity.[6] With respect to the active site, a C atom adjacent to a pyridinic-N can potentially function as an active site where an O2 molecule can be adsorbed in the initial step of the ORR.[5] In contrast, Geng et al. have claimed that graphitic-N species appear to play a predominant role in determining the ORR activity.[9] Our previous calculations[10] have also suggested that the graphitic-N is capable of providing reaction sites for ORR.

To improve the electrocatalytic performances of N-doped graphene, increasing the number of active sites is critical. In this respect, small graphene surrounded by edges, that is, graphene nanoclusters (GNCs), have attracted much attention. High-symmetry GNCs with various shapes and sizes have been fabricated experimentally.[11−17] Because N atoms get preferentially doped at edges that exhibit high chemical reactivity,[18−21] numerous theoretical studies on the ORR activity at edges and defects of N-doped graphene have been reported.[22−28] The C atoms near the N atoms at the edge have been identified as the ORR active sites.[24−28] For example, Saidi has suggested that both pyridinic- and graphitic-N are dominant active sites at the edge of the N-doped nanocarbons.[28] Furthermore, the electronic structure near the doped N atom has been suggested to drastically change depending on the distance of the N atom from the edge.[20,21,29] However, little attention has been devoted to the dependence of the ORR activity on doping and active sites in the vicinity of the edge.

In general, the ORR mainly proceeds via two pathways: a two-electron (2e–) pathway wherein O2 molecules are reduced to hydrogen peroxides (H2O2) and a direct four-electron (4e–) pathway wherein the final product of the reaction is water.[4,26,30−36] H2O2 generated by the 2e– reaction might corrode a carbon-based electrocatalyst causing poor device durability. Thus, the 4e– pathway is more desirable than the 2e– pathway; hence, selectivity for the 4e– pathway is required for the carbon-based catalyst. The discussion of reaction selectivity has thus far focused on metal-based catalysts. The reaction selectivity on metal surfaces strongly depends on the metals and their surface orientations,[37] and it is dominated by the stability of reaction intermediates adsorbed on the metal surface. However, the onset of the reaction selectivity is not fully understood. The stability of reaction intermediates has been suggested to be affected by environmental water. For example, ORR intermediates have been demonstrated to be stabilized by the environmental water on the surface of a Pt catalyst.[38,39] For N-doped graphene, the stability of adsorbed O2 molecules is strongly affected by environmental water.[36] Further, it has also been reported that the ORR activity is strongly influenced by water molecules generated by the ORR itself.[40] However, the effects of water on the reaction selectivity have not been clarified.

To comprehensively evaluate the catalytic activity for ORR, it is necessary to understand the reaction kinetics as well as thermodynamics. However, in the first place, the thermodynamic properties of ORR should be revealed before the reaction kinetics because a potential of the ORR activity is governed by the thermodynamical stability for each reaction process. In the present work, we use first-principles calculations based on the density functional theory (DFT) to investigate the effects of doping and reaction sites on the ORR activity of the N-doped graphene nanoclusters (N-GNCs) in view of the thermodynamical selectivity for the reaction pathway. We focus on the effect of the water molecule generated by the reaction on the ORR activity. We show that high selectivity toward the 4e– pathway is attained for GNCs with N atoms located at the inner doping sites of the cluster rather than at the edge sites. We also suggest that the water molecule plays a critical role in the manifestation of the reaction selectivity.

Figure shows the model of the GNC, in which a C atom is substituted by a N atom. We adopted the hexagonal N-GNCs with zigzag edges which have been fabricated experimentally.[17] The doping and the reaction sites are denoted by numbers and letters, respectively (see Figure ). We assumed that the reaction sites are C atoms adjacent to the N atom, onto which the reaction intermediates (O, OH, and OOH) for the ORR are adsorbed (see Figure ). We define model “x–y” as a model for doping site “x” and reaction site “y”.

Figure 1

Model of the N-GNC (C95H24N). The white and gray balls indicate H and C atoms, respectively. (a) Numbers 1–7 and (b) symbols a–g indicate doping sites and reaction sites, respectively.

Figure 2

Model of the ORR intermediates on the N-GNC. The white, gray, blue, and red balls indicate H, C, N, and O atoms, respectively.

Results and Discussion

Figure shows the relative formation energies of the N-GNCs. The N atom preferentially located at the edge, at sites 1 and 1′, rather than at the inner site of the cluster, which agrees well with previous results.[19−21] The edge-localized states[41] on zigzag edges strongly affect the doping stability as follows:[21] for N-doping on the zigzag graphene nanoribbon (GNR), the formation energy increased as a function of the distance of N from the edge and two independent curves arose for odd- and even-numbered sites. Because the wave function of the edge state has amplitude only at odd-numbered sites, only the N 2p orbital at odd-numbered sites can resonate with the edge state. As a result, the eigenvalue of the edge state for the odd-numbered-site doping became lower than that for the even-numbered-site doping. For GNCs, the edge states apparently emerged at the edges of GNCs larger than C96H24.[29] For our model (C96H24), the number of edge atoms was too small to clearly produce the localized state at the edge. Therefore, such an odd–even trend in the formation energy was not observed distinctly in our cluster model.

Figure 3

Relative formation energy of the H-terminated N-GNC for each doping site. The total energy for site 1 is set to be zero.

As evident in Figure , the N atom was preferentially located at site 1′ rather than at site 1. N atoms have been reported to preferentially locate at the zigzag edges rather than at armchair edges.[20] Site 1′ was deemed to be a pure zigzag edge, whereas site 1 was regarded as either the side or the corner of the cluster—namely, zigzag as well as armchair edges. Therefore, site 1′ became more stable than site 1. In this study, each N atom located at site 1 or 1′ was terminated by a H atom, despite the apparent improbability of such an arrangement. The N atom of a pyridine molecule is rarely accompanied by a H atom because the N atom of pyridine has a stable lone pair of electrons. However, the N atom at the zigzag edge preferred to be hydrogenated, maintaining the planar C–N–C structure. The formation of a stable N–H bond at the zigzag edge of GNR corroborated the charge-transfer model:[21] if the extra electron transfers to the edge state, one of the electrons of N contributes to the formation of the covalent bonding between N and H, and then the π-conjugated network near the edge is preserved even with N-doping. The stability of the N–H bond for N-GNC cannot be discussed the same way as that of the N–H bond for GNR because the edge state hardly emerged in our (C96H24) model. In fact, at doping site 1, the pyridinic-N was more stable than N terminated by a H atom. The quantitative stability of the N–H bond was shown in the Supporting Information.

In the present work, the value of UMax with respect to a standard hydrogen electrode (SHE) and the reaction selectivity were evaluated using free-energy diagrams. Figure shows the free-energy diagrams for model 7-f as an example. In these diagrams, the free energies of the ORR intermediates are plotted from O2 toward the final product of the ORR at zero-cell potential (U = 0 V), the equilibrium potential (Ueq), and the maximum potential (UMax), wherein all reaction steps are exothermic. At U = 0 V, the diagram for the 4e– pathway becomes downhill: all of the reactions spontaneously proceed toward the generation of H2O. The UMax calculated for the 4e– pathway is 0.68 V. By contrast, for the 2e– pathway, the diagram becomes uphill from the OOH adsorption to the H2O2 generation at U = 0 V, which causes a negative value of UMax (−0.13 V). Thus, the ORR for the 2e– pathway stops at the OOH adsorption step, and H2O2 generation is thereby suppressed. All of the corresponding reaction diagrams are shown in the Supporting Information.

Figure 4

Free-energy diagrams of the model 7-f under acidic conditions (a) for the 4e– pathway at zero cell potential (U = 0 V), the equilibrium potential (Ueq = 1.23 V), and the maximum potential (UMax = +0.68 V) wherein all reaction steps are exothermic and (b) for the 2e– pathway at U = 0 V, Ueq = 0.68 V, and UMax = −0.13 V.

We evaluated the ORR activity of the N-GNC for each doping site and reaction site by estimating UMax in terms of the output voltage or the durability, that is, the reaction selectivity, of fuel cells. Figure shows the variation of UMax depending on the doping and the reaction sites. The UMax for the 4e– pathway is higher than that for the 2e– one for each doping and each reaction site except for the 1-b, pyridinic 1-b, and 2-a models. The highest UMax for the 4e– pathway for any doping sites is ≈0.7–0.8 V, comparable to that for platinum (UMax = 0.9 V[42]). Although the values of UMax are greatly scattered near the edge, the reaction site variation of UMax is reduced for the doping sites inside the cluster (≥site 3). The average UMax for inner doping sites (≥site 3) of the clusters is 0.76 and 0.01 V for the 4e– and the 2e– pathways (see dashed and dotted lines in Figure ), respectively. Notably, the values of UMax for the 2e– pathway are approximately negative numbers except for those corresponding to doping at the edge. When UMax becomes negative, the reaction does not occur unless the reverse voltage is applied. Thus, the N-GNCs have high selectivity for the 4e– pathway, that is, the ORR via the 4e– pathway proceeds preferentially. However, for the doping sites in the vicinity of the edge (ction sites. For example, for the model 1-a″, the UMax for the 4e– pathway becomes a negative value. In this case, the ORR is aborted at the OH adsorption step, and the reaction site then becomes inactive for the ORR. Furthermore, for 1-b and 1′-b, the UMax values for the 2e– pathway are relatively large and positive as compared with cases for deeper site doping. In such cases, H2O2 molecules generated by the 2e– reaction corrode the electrocatalyst, resulting in poor durability of the fuel cell. Hence, we conclude that high capability for the ORR is more likely to be achieved via the doping of N into a deeper site of the cluster.

Figure 5

UMax of the N-GNCs for the models. The circles and the crosses show the UMax for the 4e– and the 2e– pathways, respectively. The UMax for the pyridinic-N models are indicated by the squares for the 4e– pathway and by the plus marks for the 2e– pathway, respectively. The dashed and dotted lines show averaged values of UMax over deeper doping sites (≥site 3) for the 4e– and the 2e– pathways, respectively.

The N-GNR has been reported to exhibit high ORR activity in the vicinity of the edge.[27] In our results, the N-GNC also shows a high UMax at the edge sites (1′-b, 2-c, etc.). However, in the vicinity of the edge (ction sites, and the selectivity for the 4e– reduction is low. Therefore, the ORR at the edge of the N-GNC is undesirable from the viewpoint of achieving a stable output and a durable fuel cell. Because N atoms thermodynamically prefer to locate near the edges, fabricating N-GNCs containing N atoms inside the clusters in a thermal equilibrium state would be experimentally difficult. However, if the N-GNC containing N atoms inside the cluster can be fabricated using nonequilibrium processes, such as a solution plasma process,[43] the N-GNC becomes an epoch-making catalyst for the ORR.

UMax is determined by the relative free energies of the intermediates, ΔGOOH, ΔGO, and ΔGOH. For the 4e– pathway, the state of equilibrium between OH adsorption and H2O generation dominates the value of UMax for almost all of the models. Figure shows the ΔG values for the models. For inner doping sites (≥site 3) of the clusters, the ΔG values are almost the same, regardless of the reaction site. By contrast, for the doping sites in the vicinity of the edge (ction site, which leads to a substantial dependence of UMax on the reaction site. Furthermore, for the pyridinic-N models, the values of ΔG are higher than those for the graphitic-N models, that is, each intermediate for the pyridinic-N models is relatively unstable. When the reaction site is just at the edge, ΔGOH becomes a negative value for site 1-a″, 2-a′, and 2′-a′. This result means that the reaction is aborted at the step of OH adsorption, and that these reaction sites are then no longer active.

Figure 6

ΔG values of the reaction intermediates for the models. The squares, rhombuses, and triangles represent ΔGOOH, ΔGO, and ΔGOH values for the 4e– pathway, respectively. The circles represent ΔGOOH values for the 2e– pathway. The free energies for pyridinic-N models are indicated by filled marks. The dashed lines show the averaged ΔG values for the reaction intermediates over deeper doping sites (≥site 3).

We explicitly investigated the influence of water (ΔGW) on the adsorption of a water molecule onto reaction intermediates to evaluate the ORR activity because the adsorption energy of reaction intermediates is strongly affected by water. Hydrogen bonding between a reaction intermediate and a water molecule stabilizes the intermediates of the ORR. Water molecules have been reported to affect the stability of the OOH and the OH adsorbed onto the surface of the metal catalyst, whereas the adsorption energy of O has been reported to be hardly affected by water.[38] However, for the N-GNCs, the adsorbed O is stabilized by a water molecule as well as the adsorption of OOH or OH. Figure shows the ΔGW for all models. The values of ΔGW(OOH) and ΔGW(OH) are approximately −0.5 and −0.4 eV, respectively. The value of ΔGW(OOH) and ΔGW(OH) are roughly independent of the doping site. However, ΔGW(O) in the vicinity of the edge varies depending on the doping sites, whereas the values of ΔGW(O) for inner doping sites (≥site 4) of the clusters are settled. As shown in Figure , the maximum electrode potentials without the adsorption of a water molecule, denoted by the value of UMaxw/o for the 4e– pathway, are approximately the same as UMax (Figure ). However, the UMaxw/o for the 2e– pathways are substantially greater than the UMax; in addition, the UMaxw/o of the 4e– pathway is still greater than those for the 2e– pathway except for 2-a and 3-b. Thus, almost all models show selectivity for the 4e– pathway (such as the ORR with ΔGW). However, the UMaxw/o values of the 2e– pathway are positive and relatively large values ≈0.5 V because the ΔGOOH, which dominates the maximum electrode potential for the 2e– pathway, increases in the absence of water. As a result, the 2e– reduction as well as the 4e– reduction must proceed. Hence, the water plays an important role in the onset of the reaction selectivity. It has been reported that the removal of water molecules at the Pt-catalyst surface enhances the reactivity of oxygen species.[44] Furthermore, for Pt-based catalysts, water molecules have been reported to corrode the catalyst and to substantially decrease the electrocatalytic performance for the ORR.[40] Therefore, to improve the ORR activity of Pt-based catalysts, water molecules should be removed in the reaction field. On the other hand, for the N-GNCs, the water rather improves the reaction selectivity, resulting in high durability of the fuel cell. This durability will be a great advantage of N-GNCs over metal-based catalysts.

Figure 7

ΔGW of the reaction intermediates for each doping site and reaction site. The squares, rhombuses, and triangles represent ΔGW(OOH), ΔGW(O), and ΔGW(OH), respectively. The ΔGW for pyridinic-N models are indicated by the filled marks.

Figure 8

UMaxw/o of the N-GNCs for the models; maximum electrode potentials without the adsorption of a water molecule. The circles and the crosses show UMaxw/o for the 4e– and the 2e– pathways, respectively. The UMaxw/o for the pyridinic-N models are indicated by squares and the plus marks for the 4e– and the 2e– pathways, respectively. The dashed and dotted lines show the average UMaxw/o for the 4e– and the 2e– pathways over deeper doping sites (≥site 3), respectively.

Conclusions

The ORR activity for the hexagonal N-GNCs was investigated. The GNC with a nitrogen atom around the center of the cluster exhibits a high UMax and high selectivity for the 4e– pathway, that is, it exhibits high durability. For the doping sites near the edge, the value of UMax depends strongly on both the doping and the reaction sites, resulting in low selectivity for the 4e– pathway. However, for the doping sites inside the cluster, the lack of a significant dependence of UMax on the doping and the reaction sites is confirmed. Furthermore, we clarified that the reaction intermediates are stabilized by the water molecule, resulting in high selectivity for the 4e– reduction. This behavior contrasts sharply with those of metal-based catalysts, where water molecules corrode and substantially decrease its ORR electrocatalytic performance.[40] We concluded that from the viewpoint of durability under a water environment, the N-GNC with a N atom inside the cluster is a potential electrocatalyst for next-generation fuel cells.

Computational Methods

For all DFT calculations, we used the Gaussian 09 code[45] employing the hybrid B3LYP functional[46,47] and the 6-31G (d,p) basis set in this package. Structural optimization with respect to ionic positions was performed until each component of the interatomic force was <0.0003 Ha/bohr. The ORR for the 4e– pathway is ideally facilitated just below the equilibrium potential (1.23 V). For the 4e– pathway, O2 is reduced to H2O through the intermediates OOH, O, and OH. We considered the following associative mechanismwhere “*” denotes that the ORR intermediates are adsorbed onto the surface of the catalyst. The ORR for the 2e– pathway is ideally facilitated just below the equilibrium potential (0.68 V). For the 2e– pathway, O2 molecules are reduced to H2O2 through the intermediate of OOH. An associative mechanism was also considered for the 2e– pathway

For the 2e– pathway, the process of H2O2 reduction is excluded because a H2O2 molecule is desorbed from the N-GNC without an activation barrier. The electrocatalytic activities for the 4e– and 2e– reductions are dominated by the stability of adsorption of the intermediates OOH, O, and OH.

The ORR electrocatalytic activities were evaluated on the basis of the computational hydrogen electrode model proposed by Nørskov.[48] The chemical potential for (H+ + e–) is equivalent to that of 1/2H2 in the gas phase with the SHE. We considered the free-energy difference (ΔG) of the ORR processes with a pressure of 1 bar, pH = 0, and T = 298 K.

The ΔG was constructed as followswhere ΔG0 is the Gibbs free energy, ΔGU corresponds to the electrode potential, ΔGpH represents the effect of the solvent, ΔGW is the stabilization energy by water, and ΔGfield is the effect of the local electric field at the electrode. The ΔG0 was calculated considering the following termswhere ΔE is the energy difference between reaction intermediates and final products and ΔZPE and TΔS means the zero-point energy and the entropy, respectively. ΔZPE and TΔS were computed by the vibrational frequency calculation. Here, ΔE values were calculated for the 4e– pathwayand for the 2e– pathway

We assumed that ΔGW is the stabilization energy by a water molecule. Here, we optimized the hydrogen-bonding configuration between each reaction intermediate and a water molecule for all models. Typical values of ΔGW are −0.49 (OOH), −0.52 (O), and −0.39 eV (OH) for the model 7-g′, which agree well with the results for the effect of several water layers, −0.49 (OOH), −0.53 (O), and −0.42 eV (OH),[36] respectively. We set ΔGpH = 0, that is, the acidic condition (pH = 0). We ignored the term ΔGfield because the absolute value of ΔGfield was estimated to be very small (≈10–2 eV).[49] By changing ΔGU, we estimated the maximum electrode potential (UMax) wherein all reaction steps are exothermic for the 4e– and the 2e– pathways.