A Re‐evaluation of How Functional Groups Modify the Electronic Structure of Graphene Oxide

The first 4 eV of the conduction band in graphene oxide is dominated by states from carbon sites that are in close proximity, but not directly bonded, to oxidizing functional groups. The carbon sites that are bonded directly to these groups, such as epoxide and hydroxyl groups, are much higher in energy. Graphene is the first stable two-dimensional (2D) crystal ever found, a feat for which its discovers received the Nobel Prize in 2010.1 This fascinating material has carbon atoms arranged in a hexagonal lattice, and it is of great interest because of its exceptional electrical and mechanical properties. The physical structure of graphene forces the bands near the Fermi level into a Dirac cone with its vertex at the Fermi level, a peculiar state that gives rise to massless charge carriers near the Fermi level.2 Graphene is a zero band gap semiconductor with unprecedented carrier mobility, and thus has been suggested for use in many different electronic and technological applications.3-5 Graphene is, not surprisingly, highly transparent, and one very active avenue of research is developing graphene for use in optoelectronic and photovoltaic applications.6, 7 Although the properties of graphene are exciting, graphene is not perfectly suited for all situations. If one wishes to use graphene for electronic device applications, a band gap must be introduced.8 Using graphene for a transparent conducting electrode in organic solar cells requires that the energy levels of graphene align more favorably with those of the molecules in the active layer of the device to improve overall performance.9 Many different ways to engineer the band structure of graphene have been proposed and successfully demonstrated.10-12 However, the oxidization of graphene to make graphene oxide (GO), although one of the first methods tried, still promises to be first among the methods of graphene manufacture that will provide the highest yields for industrial-level manufacturing. If graphene-based products are indeed going to start from some form of GO, then understanding how the electronic structure of GO changes with chemical treatment from starting stock to final product is therefore an extremely important, and multi­faceted, question. To understand how the electronic structure of GO changes, one must first find an experimental method with which one may probe said electronic structure. A powerful technique for probing the electronic structure is X-ray absorption near-edge spectroscopy (XANES), given that it is site- and element-selective. This allows the experimentalist to individually probe the different non-equivalent sites within graphene oxide. With great success, XANES has been used to study how chemical modification leads to electronic structure modulation and physical structure evolution.13-20 However, having a powerful and appropriate experimental technique is not enough; one must also have a model framework within which one may analyze the experimental results and characterize the system under scrutiny. Up until this point, features in XANES spectra of GO have usually been assigned by correlating the feature positions to binding energy values extracted from X-ray photoemission (XPS) spectra of simpler systems containing similar functional groups. XANES spectra of simpler model systems, like phthalic acid, can be used rather than XPS.21 However, the database of XANES spectra to which one may compare a GO XANES spectrum is significantly smaller, which makes the task more challenging. That being said, the fundamental problem with the model system approach is that one must assume that each functional group remains an isolated unit, and thus retains its spectral identity. In extended, bulk systems like GO, particularly if it is multilayered, this assumption seems specious. Ideally, one would use an ab initio theoretical framework, such as density functional theory (DFT), to model the electronic structure. However, the electronic structure of graphene oxide is notoriously difficult to model because it is a highly chaotic, non-stoichiometric system, and thus a successful simulation of the electronic structure of GO, and subsequent comparison to experimental XANES measurements, has not previously been available. Using DFT, we have simulated the electronic structure of four model structures for the purposes of performing a weighted sum to reproduce an experimental measurement. The simulations were computed using WIEN2k, a full-potential muffin tin density functional theory (DFT) code that uses a linearized augmented plane wave (LAPW) basis set.22 The structures simulated are graphene and three different forms of GO, named sGO1, sGO2, and sGO3. The four simulated structures, as well as details of the computations, are included in the Supporting Information. A simulation of pristine graphene was included to model those parts of GO that are very lightly functionalized. sGO1 is a simple structure consisting of only epoxide groups with a C:O ratio of 2:1; in this structure, all carbon atoms are functionalized. sGO2 is another epoxide-only structure with a C:O ratio of 4:1. Both structures were allowed to relax via internal force minimization. The final structure, sGO3, will be the subject of further research on its own. Beginning as a 4×4 supercell of graphene bonded with hydroxyls in a C:O 2:1 ratio, the structure was allowed to completely relax via force minimization. The unit cell began with 16 hydroxyl groups, but four absconded with hydrogen atoms from neighboring hydroxyl groups to form free-floating water molecules. The four deprotonated hydroxyl groups became epoxide groups. This spontaneous process of water formation from hydroxyl reaction has previously been theorized,23 and has been seen again here. This theoretical evolution of the chemical structure is very interesting; however it is the electronic structure of the final product, rather than the dynamics of the reaction process, that is the focus of this report. Thus, the only functional groups included are epoxides and hydroxyls, which are two of the main groups in the Lerf–Klinowski model of graphene oxide.24, 25 We have calculated the C 1s XANES spectrum that each structure provides and added them to form a composite lineshape, to which we have compared an experimental XANES measurement of a sample of GO in Figure 1. The XANES measurement was performed at the SGM beamline at the Canadian Light Source (CLS);26 a brief explanation is supplied in the Experimental section. The sample was positioned such that its normal made a 30° with respect to the propagation vector of the incident beam. The details of the preparation of the sample are published elsewhere.27 The component simulated lineshapes are also included in Figure 1 to show how each contributed to the summation. The different components were weighted as follows: 30% graphene, 20% sGO1, 20% sGO2, and 30% sGO3. With this mix, the overall theoretical C:O ratio is about 3.6. As one can see, the agreement between theory and experiment in Figure 1 is good, in the sense that almost all of the main experimental features, highlighted and labeled as a through f, appear in the composite spectrum. Even the relative peak intensities are reasonably well-reproduced, except for feature e, at 288.6 eV; this will be discussed later. Let us first break the simulations down to show the contribution from each non-equivalent (NEQ) C site, as is shown in Figure 1b–d, to explore how the C atoms interact. The NEQ spectra are arranged into three distinct categories: 1) those spectra arising from carbon atoms bonded to epoxides, 2) those from carbon atoms bonded to hydroxyls, and 3) those from carbon atoms bonded to no functional group. In each panel is an image of one unit cell for that simulation; the carbon atoms are color-coded so that the color of the atom in the unit cell matches the color of the corresponding calculated spectrum. So, for example, carbon atoms colored blue are not bonded to any functional group, therefore the spectra from non-functionalized NEQ C sites are also blue. Breaking the simulated spectra apart in this manner reveals something unexpected: the C sites that give rise to the spectral features commonly seen in GO C 1s XANES spectra, particularly below 288 eV, are the interstitial C sites that exist between the functional groups, not the C sites bonded to functional groups. Our results suggest that one must think of the electronic structure of GO in spatial terms, above and beyond the oxidation level. To study the spatial aspect of the electronic structure of GO, we have calculated the charge density (CD) at the bottom of the CB for sGO1, sGO2, and sGO3, and displayed the 3D maps of said CD in Figure 2. The charge density maps were calculated by XCrysDen, which took, as input, the simulated electronic structure provided by WIEN2k.28 The charge density surfaces shown in Figure 2 are equiprobability contours within each plot; to compare plots, the contours were chosen such that the charge density bubble around the epoxides in each GO structure looked to enclose approximately the same volume. It is clear from Figure 2 that the bottom of the CB is dominated by two types of sites: the O sites and the non-bonded C sites. The fact that the O sites dominate the bottom of the CB is to be expected: this is a fundamental aspect of all electron-withdrawing groups. However, the non-bonded C sites also have significant weighting at the bottom of the CB, and in the case of sGO2 and sGO3, embryonic charge conduction pathways can be seen (highlighted by the boxes). Functionalized C sites can also supply charge carriers at the bottom of the CB, provided that they are close enough to the pathways. However, that is much less common. Indeed, one can see how a modest reduction in oxidation level can result in a sharp decrease in the band gap and a concomitant increase in conductivity.20 One would not have to remove many functional groups for a much longer, contiguous charge conduction pathway to open. These results directly contradict conventional thinking on how one should label features in GO C 1s XANES spectra. Our simulations show that the states near the main π* resonance at 285.4 eV, which are vital to the electronic and optical performance of GO, are almost entirely due to the non-functionalized C sites. It is clear that these sites are still strongly affected by the presence of the functional groups. It is this last point that provides the bridge between our results and those that have come before: although the feature at 286.5 eV is not due to resonantly exciting a core electron into a C–OH bond, it is nevertheless the presence of that hydroxyl group that makes that transition possible. This means that not only the types of functional groups, but also the distribution of those functional groups, play significant roles in determining the final electronic structure. Thus, the importance of research seeking to control the patterning of functional groups cannot be understated.29, 30 What then about the resonance at 288.6 eV, where the discrepancy between theory and experiment is the strongest? The source of this particular feature has engendered the strongest controversy of any peak that one may find in a C 1s XANES spectrum measured from GO. This debate is encapsulated in the discussion between Pacilé et al., who assigned this feature to interlayer states,31 and Jeong et al., who ascribed it the influence of carboxyl functional groups bonded to the edge of the graphene flakes.32 We submit that the feature at 288.6 eV may be due, at least in part, to the presence of carboxyl. However, features at 287.5 eV and 288.5 eV have been seen in the C 1s XANES spectra of highly oriented pyrolytic graphite (HOPG) and pristine graphene, but not consistently. Brandes et al. found that these states seem to occur only on damaged sections of the graphite structure.33 Schultz et al. also saw features at these energies on pristine graphene, but they could be removed after annealing in vacuum.34 After annealing, the features would return if the sample was once again exposed to ambient atmospheric conditions. Schultz et al. interpreted the features as due to physiosorption of contaminants, and our experiments agree with this conclusion. Consider the C 1s XANES spectra of HOPG, measured in TEY mode on three different beamlines, shown in Figure 3. Features occur at 287.5 and 288.5 eV in only two spectra, and the only common variable is that both were measured at the SGM beamline. Following the reasoning of Schultz et al., these features appear because conditions on the SGM do not promote desorption of the trapped gasses as strongly as either BL8 or REIXS. This is eminently plausible, for two reasons. First, the SGM has the lowest flux density of all three beamlines. The lower flux density means less energy density deposited in the sample, and therefore, lessened heat loading that will promote offgassing of trapped molecules. Second, REIXS and BL8 each have lower pressure in the measurement chamber than SGM. Lesser heat loading and higher pressure, when taken together, means that HOPG will hold more of its physisorbed gasses when studied at SGM. In conclusion, we have analyzed the C 1s XANES spectrum of a typical sample of GO using DFT, and we have found that the changes to the electronic structure of graphene brought about by oxidization through functionalization is more complicated than has been envisioned to date. Our DFT simulations included epoxide and hydroxyl groups, and they reproduce all of the spectral features in a typical GO spectrum, except for the strong resonance at 288.6 eV. This confirms the legitimacy of the Lerf–Klinowski model, and lends credence to the accuracy of our results. However, all spectral features below 288 eV are mostly due to C sites that are not directly bonded to a functional group, yet have their local states modified by the nearby functional groups. On the other hand, states around 288 eV are likely strongly influenced by physiosorbed atmospheric gases, notably water. Our work significantly impacts the efforts of any researcher attempting to accurately engineer the electronic structure of graphene, because the type and distribution of the functional groups impacts the final electronic structure in a way not previously considered. Additionally, physiosorbed atmospheric molecules also play a significant role, the full breadth of which has not yet been studied. The electronic structure of GO is susceptible to changes in ambient conditions such as humidity, so accounting for such changes is a very important design consideration. The C 1s XANES spectra of GO and some of the HOPG spectra in Figure 3 were measured at the Spherical Grating Monochromator (SGM) beamline at the Canadian Light Source.26 The exit slit to the monochromator was set to 25 μm, which gives the incident light an energy resolution better than 0.1 eV. The spectra were measured in both total electron yield mode and total fluorescence yield mode. C 1s XANES spectra of HOPG were also performed at Beamline 8.0.1 at the Advanced Light Source at the Lawrence Berkeley National Laboratory (BL8).35 The monochromator entrance and exit slits were set to 20 μm and 30 μm, respectively, to give a total experimental resolution was 0.1 eV FWHM. The resolution of the REIXS monochromator was approximately 0.05 eV with the exit slit set to 25 μm. For the experiments on all three beamlines, the incident light linearly polarized, and the electric field component of the incoming light lies in the plane of incidence. In all cases but one, the HOPG spectra were normalized by a current generated in a highly transparent gold mesh that is upstream of the sample, thus providing a measure of the incident photon flux. The one exception is the 2012 spectrum measured at the SGM. Instead, this spectrum was instead normalized to the current generated in a photodiode.36 This photodiode current spectrum was not taken simultaneously with the sample spectrum, but rather directly afterwards. This was necessary because the gold mesh on the SGM had become highly contaminated by carbon at that time, and was no longer able to provide an accurate measure of the incident photon flux. The authors acknowledge the efforts of Dmitriy Dikin and his research group at Northwestern University, as they prepared the samples. The authors acknowledge support by the Natural Sciences and Engineering Research Council of Canada (NSERC), the Canada Research Chair program and the Russian Foundation for Basic Research (Projects 11–02–00022). The Canadian Light Source is supported by NSERC, the National Research Council (NSC) Canada, the Canadian Institutes of Health Research (CIHR), the Province of Saskatchewan, Western Economic Diversification Canada, and the University of Saskatchewan. The Advanced Light Source is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02–05CH11231. Finally, the authors acknowledge Compute Canada, because the calculations presented in this work were performed on the Grex high-performance computing cluster, which is part of the Westgrid network. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.