Mechanisms for Oxidative Unzipping and Cutting of Graphene

Nov 23, 2011 - current models of graphene oxidative unzipping and cutting. We find that these processes are rate limited by O diffusion and driven by ...
0 downloads 0 Views 2MB Size
Download PDF
Letter pubs.acs.org/NanoLett

Mechanisms for Oxidative Unzipping and Cutting of Graphene Tao Sun†,‡,§ and Stefano Fabris*,†,‡ †

CNR-IOM Democritos, Theory@Elettra group, Istituto Officina dei Materiali, s.s. 14 km 163,5 AREA Science Park, 34149 Trieste, Italy ‡ SISSA, Scuola Internazionale Superiore di Studi Avanzati, via Bonomea 265, I-34136 Trieste, Italy S Supporting Information *

ABSTRACT: We identify mechanisms and surface precursors for the nucleation and growth of extended defects on oxidized graphene. Density functional theory calculations show that the formation of surface structures capable to initiate the unzipping and cracking of the oxidized C network is strongly influenced by the constraint of the graphitic lattice on the surface functional groups. Accounting for this effect on the preferential spatial patterning of O adsorbates allows us to revise and extend the current models of graphene oxidative unzipping and cutting. We find that these processes are rate limited by O diffusion and driven by the local strain induced by the O adspecies. Adsorbate mobility is ultimately recognized as a key factor to control and to prevent the C-network breakdown during thermal processing of oxidized graphene. KEYWORDS: Graphene, DFT, graphene oxide, defects

O

of a graphitic hexagon and break the underlying C−C bonds. The resulting unzipped ether (ET) pair was identified as the nucleation site for the growth of fault lines and cracks on oxidized graphitic surfaces.16 Z. Y. Li et al.21 further demonstrated how the graphene surfaces are cut along the ET chains. They found that O adatoms distributed on both sides of the graphene sheet can form facing EP pairs that protrude on opposite sides of the surface. These EP structures can then transform into a more stable structure containing two semiquinones, leading to the fracture of the graphene sheet. The preferential formation of the unzipped linear defects was established by approximating graphene with small organic C24H12 and C54H18 molecules.3 Although such practice is widely adopted in modeling the oxidation of graphitic materials,17,22 it does not fully capture the lattice constraint due to the surrounding C network on the O adsorbates. We show that this constraint is particularly important for the nucleation of extended linear defects on oxidized graphene. Its inclusion into the DFT calculations reverses the adsorbate energetics governing the unzipping of the C network and predicts a different surface structure as the nucleation center for the growth of these linear defects. The formation of this structure turns out to be controlled by the surface diffusion of the O adatoms, which is also the rate-limiting step for the growth of the linear defects and cracks. These findings suggest a more general mechanism for crack formation, valid also for

xidation is one of the main routes to functionalize and manipulate graphitic materials.1−3 For instance, this reaction allows for tuning the band gap and transport properties of graphene,4−7 thus widening its applications in electronics and optoelectronics.8,9 Moreover, oxidation is at the basis of cost-efficient approaches for the mass-production of graphene: It turns the basal surfaces of graphite from highly hydrophobic to hydrophilic, and graphite dissolves in oxidative solution as individual layers, which can then be reduced and deoxygenated with chemical or thermal processes.10−14 During oxidation, O atoms are included into the C network. This leads to O-containing surface groups and defects, which induce various degrees of deformation or damage to the honeycomb lattice.5 Understanding how the O atoms break up graphitic surfaces is critical to preserve the graphene network during oxidation and reduction processes as well as to develop new nanofabrication techniques based on oxidative cutting.15,16 Furthermore, this knowledge would also help to elucidate more complex processes, such as carbon combustion17,18 and plasma etching.19,20 There is presently no consensus on the formation of extended defects and cracks on oxidized graphitic surfaces. In this Letter, we provide mechanisms for the nucleation and growth of these defects, whose formation is shown to be rate limited by O surface diffusion and driven by the local strain around the O adspecies. In a seminal paper, J. L. Li et al. explained the formation of cracks in oxidized graphene and nanotubes on the basis of a cooperative alignment of epoxy (EP) groups that leads to an Odriven unzipping of the C network.3 A surface EP consists of an O adatom above a bridging C−C site and is the most stable O adsorbate on defect-free graphene. According to these DFT calculations, pairs of EP groups aggregate at the opposite ends © 2011 American Chemical Society

Received: August 2, 2011 Revised: November 15, 2011 Published: November 23, 2011 17

dx.doi.org/10.1021/nl202656c | Nano Lett. 2012, 12, 17−21

Nano Letters

Letter

Figure 1. Energetics of diffusion for a pair of O adsorbates on a coronene molecule, C24H12, (a) and on graphene (b). Red and blue lines represent the diffusion path leading to a NNN EP and to an unzipped ET pair, respectively. The data for the C24H12 molecule are taken from ref 3 and are marked by a “*”. Black and red circles represent C and O, respectively. All energies are in eV. Figure 2. (a−c) Energetics for the formation of the unzipped NNN ET trimer from the diffusion of an O adsorbate toward a NNN epoxy pair. (d) Comparison with an unzipped linear epoxy trimer. Side views of the local surface structures in the NNN EP (e) and unzipped ET (f) trimers. All energies are in eV.

graphitic materials exposing only one side of the C network, such as multiwalled nanotubes or graphite. This work is based on DFT simulations carried out using the Perdew−Burke−Ernzerhof functional23 and plane-wave ultrasoft pseudopotential method, as implemented in the PWscf code of the Quantum ESPRESSO distribution.24,25 The planewave basis set and density representation were limited by energy cutoff of 30 and 300 Ry, respectively. Graphene was modeled with a 7 × 7 periodic supercell slab separated by 14 Å of vacuum, while the Brillouin zone integrations were performed on a 4 × 4 × 1 k-point mesh. All structures were optimized until the ionic force on each atom was smaller than 10−3 Ry/Bohr. Reaction mechanisms and transition barriers were studied with the climbing image nudged elastic band method, employing 8−10 system replicas.26,27 Stable structures and transition states of particular significance were further validated by normal-mode analysis. Overlap population analysis28 is performed by using the SIESTA electronic structure code.29 Further details of the calculations and of the computational supercells used to model the graphitic surfaces are included in the Supporting Information. The calculated energetics of two O adatoms on a C24H12 molecule (Figure 1a, data from ref 3 marked with a “*”) shows that the lowest-energy configuration consists of two O adatoms at the opposite ends of a hexagonal cell (blue line). The elastic strain associated to this adsorbates’ configuration forces a collective breaking of the underlying C−C bonds, leading to two unzipped ETs. This structure was reported to be 1.0 eV lower in energy than a pair of next nearest-neighbor (NNN) EP.3 In addition, the activation energy for the formation of the unzipped structure via adsorbate diffusion was found to be lower (0.83 eV, blue lines in Figure 1a) than the one to form the NNN EP pair (1.10 eV, blue lines).3 This energy trend turns out to depend strongly on the size of the graphene flake and to reverse in the case of an extended graphene surface. Our calculations confirm that for the coronene molecule, the ET pair is more stable than the EP one by 1.08 eV. However, this energy difference decreases to a mere 0.05 eV for the larger C54H18 molecule. When graphene is modeled as an extended surface (Figure 1b), the lowest-energy configuration is now the NNN EP pair, while the unzipped ETs are 0.2 eV higher in energy. Moreover, also kinetic considerations favor the formation of the former with respect to the latter: The activation energies for the formation of the

NNN EP and of the unzipped ET pairs are 0.73 and 0.96 eV (red and blue lines in Figure 1b), correspondingly. Note that the activation energy for the diffusion of an isolated EP group on pristine graphene is 0.73 eV (in good agreement with the literature),30 hence the NNN EP pairs can form as soon as the thermal diffusion of the O adatoms is activated. This reversed relative energetics of O-adsorbates (Figure 1) originates from the lattice constraint caused by the surrounding C network, which is not properly accounted for in small molecular models of a graphitic surface. For small molecules, the lattice strain associated with bond unzipping can be readily released at the edges of the molecules, which bends considerably, thus underestimating the strain energy (see Supporting Information, Figure S1). Once the lattice constraint is taken into account, forming the unzipped ET pair is no longer energetically and kinetically favorable. Instead, the two EP groups preferentially form NNN EP pairs, which preserve the graphene C network. This latter conclusion is insensitive to the type of functional employed in the calculation, as a recent computational study31 found the same energy trend using local density approximation (LDA) for the exchange−correlation functional. Starting from these considerations, we investigate in the following what drives C−C unzipping on an extended graphene surface. It turns out that the extended linear faults initiates from a nucleation center that is closely related to the thermodynamically most stable NNN EP pairs. Figure 2 shows that the NNN EP pair can capture and trap an additional O adsorbate nearby, forming a very stable NNN EP trimer (Figure 2b). This is 0.59 eV lower in energy than the initial state (a NNN EP pair and a neighboring epoxide, Figure 2a). The activation energy for the diffusion of an O adatom close to the NNN EP pair is 0.68 eV, actually lower than for EP diffusion on the pristine surface (0.73 eV). Hence, these structures can be formed as soon as O diffusion is activated on graphene. The NNN EP trimer is also more stable than the linear unzipped chain of three ETs (Figure 2d, blue line). We propose this structure to be the nucleation center for the oxidative cutting of graphene. The NNN EP trimer can easily 18

dx.doi.org/10.1021/nl202656c | Nano Lett. 2012, 12, 17−21

Nano Letters

Letter

Figure 4. Energetics of an O adsorbate in the presence of an unzipped ET chain governing the oxidative cutting of graphene. All energies in eV.

corresponding difference in bond lengths, 1.38 Å in ET and 1.45 Å in EP. We then address the changes in the C−C bonding before and after the EP−ET transformation. In the NNN EP trimer, each C atom remains chemically bonded with its three nearestneighboring C atoms. Lattice strain induced by the EP groups causes considerable weakening of the C−C bonds close to the O adatoms. This is shown in the C−C OP analysis reported in Figure 5a. The C−C bond beneath an O adatom is the weakest, with an OP of 0.62, 37% smaller than that of the perfect C−C bond in pristine graphene (1.02). The OPs of its two neighboring C−C bonds are also quite low (0.80 and 0.88). Correspondingly, all these C−C bonds have longer bond lengths (1.52, 1.51, and 1.47 Å) comparing to the unperturbed value (1.42 Å). The strain in the lattice is greatly relieved after the EP−ET transformation. As shown in the side views of these two surface isomers (Figure 2e−f), the NNN ET trimer exhibits a much greater convexity compared to the NNN EP trimer. This outward displacement allows to accommodate the larger C−O−C bond angle in the ETs and to optimize the C− C bonds in the surrounding lattice. The calculated OP reported in Figure 5b,d reveals that the EP−ET transformation leads to an overall strengthening of the C−C bonds. In particular, the C−C bonds between the ETs have remarkably high OP (1.12). Together with ETs they form a very stable and symmetric ninemember ring, where the distance between every two adjacent atoms is 1.38 Å. In summary, the NNN EP-to-ET transformation can be seen as a competition in which the energy gain from creating stronger C−O bonds, relieving the lattice strain, and forming highly stable nine-member ring outweighs the energy cost of breaking the three weak C−C bonds underneath the EP O adatoms. This highly stable nine-member ring is an unique feature of the NNN ET trimer. Its C3v symmetry is more commensurate to the graphene lattice than the linear ET trimer (Figure 2d). As shown in the OP distribution of the two structures (Figure 5d), except for three bonds with highest OP values, most C−C bonds in the NNN ET trimer have OPs close to the unperturbed value (within 4%). In contrast, OPs of C−C bonds in the linear ET trimer are more spread, indicating a higher degree of lattice strain. Also, the C−O bonds in the linear ET trimer are not identical to each other. Those at the two ends have a smaller OP (0.53) than the ones at the center

Figure 3. (a,b) Energetics for the formation and growth of an extended linear defect from the O diffusion to the NNN ET trimer (c,d). Schematic representation of further nucleation and growth of Odriven unzipping in graphene.

transform into a trimer of NNN ET groups (Figure 2b,c), breaking three C−C bonds. This local structural transformation releases 0.50 eV and requires an activation energy of 0.66 eV, again comparable to (and lower than) that one for O diffusion on pristine graphene. Quite importantly, the NNN ET trimer is 0.60 eV lower in energy than the linear ET chain (Figure 2d). This further supports the conclusion that these extended linear defects do not originate from aligned ET groups. The distinct energetic preference of the NNN ET trimer can be attributed to several factors, which are manifest in a chemical bonding analysis based on overlap population (OP). Such analysis is widely used to evaluate the relative strength of chemical bonding between a given pair of atoms.32,33 Because absolute OP values depend on the basis set and are not very meaningful, we focus on the relative magnitude of the OP calculated for the same type of chemical bonds (C−C or C−O) with the same basis set. This provides an effective way to understand different bonding properties of the EP and ET structures under analysis. We start with a Mulliken charge population analysis of the C and O atoms involved in the C−O bonds: The calculated number of valence electrons for the C/O atoms in the EP and ET groups are 3.94/6.22 and 3.90/6.25, respectively. The difference in the C and O charges shows that the C−O bondings in both EP and ET groups are covalent in nature and have almost the same ionic polarization. Hence, a charge population analysis alone can not explain the relative energetics among these structures. Instead, the OP analysis presents a distinct difference between the C−O bonds in the EP and ET groups. The calculated OP for the C−O bonds are 0.35 (NNN EP trimer) and 0.57 (NNN ET trimer). The OP value for the C−O bond in the ET groups is 60% larger than that in the EP ones. This indicates that the C−O bonds in ETs are substantially stronger than epoxides, as also supported by the 19

dx.doi.org/10.1021/nl202656c | Nano Lett. 2012, 12, 17−21

Nano Letters

Letter

Figure 5. OPs of the C−C bonds in the NNN EP (a), NNN ET (b), and linear ET (c) trimers and in corresponding bond lengths (in the parentheses). Distribution of the OPs of the C−C bonds in the 7 × 7 supercell (d). Values of OP and bond lengths (in Å) are indicated only for the symmetry equivalent bonds. Colors in (a−c) refer to the value of the OP reported in (d).

formation of unzipped ET chains does not directly lead to cutting of the graphene sheet.21 Indeed, although the C−C bond is broken after unzipping, the C atoms are still strongly connected via C−O−C bonds, and the mechanical properties are only marginally altered.34 Z. Y. Li et al. proposed a mechanism for explaining the breaking of graphene flakes into smaller pieces upon oxidation.21 The key intermediate of their mechanism is a pair of EP groups facing on opposite sides of the graphene sheet and is bound to the same two C atoms. According to this mechanism, oxidative cutting necessarily requires the active participation of O adsorbates on opposite sides of the graphene surface, and it therefore does not apply to other relevant graphitic surfaces, such as graphite or multiwalled nanotubes. The energetics of an O adsorbate in the presence of an ET chain is displayed in Figure 4a,b. The linear defect attracts a neighboring epoxide with a driving force of 0.91 eV and with an activation energy for diffusion of 0.51 eV. This leads to the formation of a local EP−ET group (Figure 4b) that transforms, almost barrierless, into two semiquinones (SQ, Figure 4c), opening up a crack with O-terminated edges. This final state is further lower in energy than the epoxide−ether group by 1.21 eV. We conclude that the unzipped ET chains play a key role

(0.57). This bonding analysis is consistent with the result that linear ET trimer has higher energy than the NNN ET one. Due to the large energy gain in the epoxy−ether rearrangement of the trimer, all the participating O adsorbates become immobile and the C−C network is irreversibly modified. The calculated energetics (Figure 3a,b) shows that also these unzipped ET trimers attract and trap the O adatoms nearby and lead to extended linear defects. The presence of the unzipped ET trimer favors the diffusion of the O adsorbate in front of the ET (Figure 3a, red arrow) rather than to the NNN site (Figure 3a, blue arrow). The C−C bond below the O adsorbate breaks during diffusion, thus increasing the length of the linear defect by one hexagonal unit. The energy barrier for the growth of the unzipped ET chain is 0.60 eV, again lower than the activation energy for epoxy diffusion on pristine graphene. We note that because of the C3v symmetry of the NNN ET trimer, every seed can nucleate up to three extended linear defects, leading to different morphologies, as schematically shown in Figure 3c−e. Having established the thermodynamic and atomistic origins for the formation and growth of the linear defects, we now address the mechanism of oxidative graphene cutting and its relationship with these unzipped structures. Note that the 20

dx.doi.org/10.1021/nl202656c | Nano Lett. 2012, 12, 17−21

Nano Letters

Letter

(9) Loh, K. P.; Bao, Q. L.; Eda, G.; Chhowalla, M. Nat. Chem. 2010, 2, 1015. (10) Stankovich, S.; Piner, R. D.; Chen, X. Q.; Wu, N. Q.; Nguyen, S. T.; Ruoff, R. S. J. J. Mater. Chem. 2006, 16, 155. (11) Schniepp, H. C.; Li, J. L.; McAllister, M. J.; Sai, H.; HerreraAlonso, M.; Adamson, D. H.; Prud’homme, R. K.; Car, R.; Saville, D. A.; Aksay, I. A. J. Phys. Chem. B 2006, 110, 8535. (12) Gilje, S.; Han, S.; Wang, M.; Wang, K. L.; Kaner, R. B. Nano Lett. 2007, 7, 3394. (13) Bagri, A.; Grantab, R.; Medhekar, N. V.; Shenoy, V. B. J. Phys. Chem. C 2010, 114, 12053. (14) Larciprete, R.; Fabris, S.; Sun, T.; Lacovig, P.; Baraldi, A.; Lizzit, S. J. Am. Chem. Soc. 2011, 133, 17315. (15) Pan, D. Y.; Zhang, J. C.; Li, Z.; Wu, M. H. Adv. Mater. 2009, 22, 734. (16) Fujii, S.; Enoki, T. J. Am. Chem. Soc. 2010, 132, 10034. (17) Radovic, L. J. Am. Chem. Soc. 2009, 131, 17166. (18) Carlsson, J. M.; Hanke, F.; Linic, S.; Scheffler, M. Phys. Rev. Lett. 2009, 102, 166104. (19) Fredriksson, H.; Chakarov, D.; Kasemo, B. Carbon 2009, 47, 1335. (20) Sun, T.; Fabris, S.; Baroni, S. J. Phys. Chem. C 2011, 115, 4730. (21) Li, Z. Y.; Zhang, W. H.; Luo, Y.; Yang, J. L.; Hou, J. G. J. Am. Chem. Soc. 2009, 131, 6320. (22) Gao, W.; Alemany, L. B.; Ci, L. J.; Ajayan, P. M. Nat. Chem. 2009, 1, 403. (23) Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, 77, 3865. (24) Giannozzi, P.; et al. J. Phys.: Condens. Matter 2009, 21, 395502. (25) Ultrasoft pseudopotentials C.pbe-rrkjus.UPF and O.pberrkjus.UPF, from the Quantum ESPRESSO distribution were used to describe the electron−ion interaction. (26) Jónsson, H.; Mills, G.; Jacobsen, K. W. In Classical and Quantum Dynamics in Condensed Phase Simulations; Berne, B. J., Ciccotti, G., Coker, D. F., Eds.; World Scientific: Hackensack, NJ, 1998; Chapter 16. (27) These calculations were performed with a 2 × 2 × 1 k-point mesh, and at convergence, refined with a 4 × 4 × 1 k-point mesh. (28) Mulliken, R. S. J. Chem. Phys. 1955, 23, 1833. (29) Soler, J. M.; Artacho, E.; Gale, J. D.; García, A.; Junquera, J.; Ordejón, P.; Sánchez-Portal, D. J. Phys.: Condens. Matter 2002, 14, 2745. (30) Suarez, A. M.; Radovic, L. R.; Bar-Ziv, E.; Sofo, J. O. Phys. Rev. Lett. 2011, 106, 146802. (31) Kawai, T.; Miyamoto, Y. Curr. Appl. Phys. 2011, 11, s50. (32) Fernández-Serra, M. V.; Artacho, E. Phys. Rev. Lett. 2006, 96, 016404. (33) López-Corral, I.; Germán, E.; Juan, A.; Volpe, M. A.; Brizuela, G. P. J. Phys. Chem. C 2011, 115, 4315. (34) Paci, J. T.; Belytschko, T.; Schatz, G. C. J. Phys. Chem. C 2007, 111, 18099. (35) Barinov, A.; Baris, M. O.; Fabris, S.; Sun, T.; Gregoratti, L.; Dalmiglio, M.; Kiskinova, M. J. Phys. Chem. C 2009, 113, 9009. (36) Ghaderi, N.; Peressi, M. J. Phys. Chem. C 2010, 114, 21625.

into oxidative cutting of graphene: They attract the diffusing O adsorbates, which weaken C−O−C bonds, and guide crack formation into the C network. The process does not require the presence of adsorbates on opposite sides of the graphene surface and leads to the formation of O-saturated edges that are well characterized on oxidized graphitic surfaces.35 A sequence of these elementary steps can ultimately break down graphitic surfaces into smaller flakes. In conclusion, we have identified the surface precursors and the atomistic mechanisms for the oxidative unzipping and cutting of graphene. Lattice constraint is shown to be critical for the aggregation of O species into surface structures capable to initiate C−C unzipping. Our calculations establish the central role of epoxy diffusion in the nucleation and growth of extended linear defects as well as in the oxidative cutting of graphene. O diffusion is the elementary rate-limiting step that controls and unifies these events. In perspective, it will be important to unravel how other constituents of graphene oxide affect epoxide diffusion, particularly hydroxyls and point defects.36 This knowledge will be useful for the fabrication of nanoscale structures from graphitic materials15,16 and for achieving a controlled oxidation of graphene and reduction of graphene oxide.

■

ASSOCIATED CONTENT S Supporting Information * Computational details, overlap population analysis, top and side views of oxidized coronene molecules (Figure S1). This material is available free of charge via the Internet at http:// pubs.acs.org.

■

AUTHOR INFORMATION Corresponding Author *E-mail: [email protected]. Present Address § Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, Minnesota 55455, United States.

■

ACKNOWLEDGMENTS Stefano Baroni is kindly acknowledged for a critical reading of the manuscript and for support. We are grateful to Marivi Fernández-Serra for help in the overlap population analysis and to O. B. Malcioglu, R. Larciprete, S. Lizzit, C. Ma, D. B. Zhang, R. M. Wentzcovitch, and A. Baraldi for useful discussions. This work was supported by Regione FVG with the project “NANOCAT”.

■

REFERENCES

(1) Ajayan, P. M.; Yakobson, B. I. Nature 2006, 441, 15. (2) Dreyer, D. R.; Park, S.; Bielawski, C. W.; Ruoff, R. S. Chem. Soc. Rev. 2010, 39, 228. (3) Li, J. L.; Kudin, K.; McAllister, M.; Prud’homme, R.; Aksay, I.; Car, R. Phys. Rev. Lett. 2006, 96, 176101. (4) McAllister, M. J.; Li, J.-L.; Adamson, D. H.; Schniepp, H. C.; Abdala, A. A.; Liu, J.; Herrera-Alonso, M.; Milius, D. L.; Car, R.; Prud’homme, R. K.; Aksay, I. A. Chem. Mater. 2007, 19, 4396−4404. (5) Mkhoyan, K. A.; Contryman, A. W.; Silcox, J.; Stewart, D. A.; Eda, G.; Mattevi, C.; Miller, S.; Chhowalla, M. Nano Lett. 2009, 9, 1058. (6) Xu, Z.; Xue, K. Nanotechnology 2010, 21, 045704. (7) Yan, J. A.; Chou, M. Phys. Rev. B 2010, 82, 125403. (8) Eda, G.; Chhowalla, M. Adv. Mater. 2010, 22, 2392. 21

dx.doi.org/10.1021/nl202656c | Nano Lett. 2012, 12, 17−21