DOI: 10.1002/cssc.201301006

Identifying Active Functionalities on Few-LayeredGraphene Catalysts for Oxidative Dehydrogenation ofIsobutaneGopi Krishna Phani Dathar,[a] Yu-Tung Tsai,[a] Kamil Gierszal,[a] Ye Xu,[a] Chengdu Liang,[a]

Adam J. Rondinone,[a] Steven H. Overbury,[a, b] and Viviane Schwartz*[a]

Introduction

Oxidative dehydrogenation (ODH) of alkanes is an economicalalternative to nonoxidative thermal dehydrogenation foralkane-to-alkene conversion. Traditionally, supported metaloxide catalysts were used as catalysts for ODH.[1] Metal oxidecatalysts are attractive because of their high conversion capa-bility, but are limited by reduced selectivity at higher conver-sions due to increased combustion side reactions and structur-al complexity that hinders tailoring them for selective dehydro-genation. Besides, metal oxides are potentially toxic. Hence,there is a need for alternative catalysts for ODH that can behighly selective to alkenes and that are environmentally friend-ly and sustainable in nature.

Nanostructured carbon catalysts for ODH of alkanes arepromising alternatives to metal oxides because of the inexpen-sive, abundant, sustainable, and renewable nature of carbonmaterials.[2] The relatively simple bonding network of nano-

structured carbons also provides an opportunity for engi-neered catalysts. Various oxygen-functionalized graphiticcarbon nanostructures, such as carbon nanotubes,[3] fullerene-like carbons,[4] carbon nanofibers,[5] onion-like carbon, nanodia-monds,[3a, 6] and few-layered graphenes (GPs),[7] were reportedas active catalysts for alkane ODH. The general consensus inthese studies was that the oxygen functionalities were theactive sites responsible for the catalytic activity exhibited bythese carbocatalysts. However, the active oxygen functionali-ties responsible for ODH and the reaction mechanism at thosefunctionalities are still unclear. Our goal is to identify the mostactive and selective sites on carbon-based catalysts for ODH ofalkanes, which will enable the engineering of active functional-ities in nanostructured carbons.

The proposed mechanism of ODH that occurs on theoxygen functionalities on carbon catalysts can be summarizedinto two steps: 1) abstraction of hydrogen from the alkane andconcomitant hydrogenation of oxygen functionalities, and2) reaction of hydroxylated functionalities with O2, resulting inH2O and regeneration of catalytic sites.[2, 8] The hydrogen ab-straction by dissociation of C�H bonds was reported as therate-limiting step in the ODH of n-butane with an activationbarrier in the order of 89 kJ mol�1 from the total energy calcu-lations performed, independent of reaction mechanism.[3b]

Recent computational studies on ODH of light alkanes onzigzag edge carbon clusters have reported activation barriersin the range of 27–94 kJ mol�1, with the lowest activation barri-

The general consensus in the studies of nanostructured carboncatalysts for oxidative dehydrogenation (ODH) of alkanes toolefins is that the oxygen functionalities generated during syn-thesis and reaction are responsible for the catalytic activity ofthese nanostructured carbons. Identification of the highlyactive oxygen functionalities would enable engineering ofnanocarbons for ODH of alkanes. Few-layered graphenes wereused as model catalysts in experiments to synthesize reducedgraphene oxide samples with varying oxygen concentrations,to characterize oxygen functionalities, and to measure the acti-vation energies for ODH of isobutane. Periodic density func-tional theory calculations were performed on graphene nano-ribbon models with a variety of oxygen functionalities at theedges to calculate their thermal stability and to model reaction

mechanisms for ODH of isobutane. Comparing measured andcalculated thermal stability and activation energies leads tothe conclusion that dicarbonyls at the zigzag edges and qui-nones at armchair edges are appropriately balanced for highactivity, relative to other model functionalities consideredherein. In the ODH of isobutane, both dehydrogenation andregeneration of catalytic sites are relevant at the dicarbonyls,whereas regeneration is facile compared with dehydrogenationat quinones. The catalytic mechanism involves weakly ad-sorbed isobutane reducing functional oxygen and leaving asisobutene, and O2 in the feed, weakly adsorbed on the hydro-genated functionality, reacting with that hydrogen and regen-erating the catalytic sites.

[a] Dr. G. K. P. Dathar, Dr. Y.-T. Tsai, Dr. K. Gierszal, Dr. Y. Xu, Dr. C. Liang,Dr. A. J. Rondinone, Dr. S. H. Overbury, Dr. V. SchwartzCenter for Nanophase Materials SciencesOak Ridge National LaboratoryOne Bethel Valley Road Oak Ridge, TN 37831 (USA)E-mail : schwartzv@ornl.gov

[b] Dr. S. H. OverburyChemical Sciences Division, Oak Ridge National LaboratoryOne Bethel Valley Road Oak Ridge, TN 37831 (USA)

Supporting Information for this article is available on the WWW underhttp://dx.doi.org/10.1002/cssc.201301006.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 483

CHEMSUSCHEMFULL PAPERS

er for hydrogen abstraction from ethylbenzene and the high-est from methane, among the three hydrocarbons studied.[9]

This study reported higher activation barriers for reoxidation ofzigzag edge clusters by gas-phase oxygen, and decreasing acti-vation barriers with increasing hydrogen content on the clus-ters. It was also suggested that reoxidation was a complex pro-cess that might involve a concerted multicenter reaction.[9]

Other groups have reported measured activation energies inthe order of 75[10] and 65–70 kJ mol�1[11] for the ODH of ethyl-benzene and isobutane, respectively. Although the above stud-ies provide some insights, there is still the need to explore thedetailed reaction mechanism and to identify the active oxygenfunctionality during the ODH cycle. Previous studies in the lit-erature have reported the role of quinone groups, in whichthe quinone/hydroquinone redox couple could be responsiblefor reducing gas-phase O2.[8, 12] However, the exact mechanismfor O2 reacting with hydrogenated functionalities is un-known.[2, 3b] Furthermore, the mechanism for the regenerationof catalytic sites is still unclear. Hence, both steps need to beexplored to understand the origins of catalytic activity and tocomplete the ODH cycle.

Herein, we use a combined experimental and theoretical ap-proach to provide further insights into the role of oxygen func-tionalities in the ODH of isobutane on oxygen-functionalizedfew-layered graphene (GP) catalysts. Our approach was basedon the investigation of oxygen-functionalized, few-layered GPas a simplified model system for the identification of the activesites on graphitic carbon. Surface oxygen functionalities werecreated during the synthesis of the GP oxide starting materi-al.[7] The synthetic method is known to generate a variety ofnucleophilic oxygen functionalities on the basal planes, at theedges, and at defect sites.[13] Because thermal stability and re-activity vary among the generated oxygen functionalities, cal-culations of thermal decomposition and chemical reactivity ofthe functional groups are essential to explain their existenceafter pretreatment and under the reaction conditions, andtheir role in the complete ODH cycle.

We performed temperature-programmed desorption (TPD)to characterize the oxygen functionalities that existed on thesamples. To deconvolute the desorption profiles and identifythe oxygen functionalities, we calculated decomposition tem-peratures of model oxygen functionalities on GP edges byusing DFT and compared them with experiments. We also rancatalytic performance tests on the GP samples for isobutaneODH to measure the activation energies. We modeled two re-action pathways for ODH at the model functionalities and cal-culated the activation barriers for dehydrogenation of isobu-tane and regeneration of catalytic sites. Finally, we comparedthe calculated activation barriers from model reaction profileswith measured activation energies to identify the activeoxygen functionalities in the ODH of isobutane on oxygen-functionalized GP edges.

Results and Discussion

Reductive treatment of GP oxide in H2 at different tempera-tures was performed to selectively desorb oxygen functionali-

ties with thermal stability less than the reducing temperature.Consequently, reduced GP oxide samples with varying oxygencontent were obtained after the reductive treatment (seeTable S1 in the Supporting Information) and they were latercharacterized by TPD. In addition to tuning the oxygen con-centration on samples, H2 present during reduction is expectedto chemisorb on the under-coordinated edges and defectsthat result from decomposed functionalities. Hydrogen-cappedcarbon edges remain stable and stop further deterioration ofthe catalyst under reducing temperatures. They are also less re-active and exclude the possibility of bare carbon edges partici-pating in the ODH reaction.

The measured CO and CO2 desorption profiles for the seriesof reduced GP samples, GP-H2-X, (X represents the reductiontemperature and is equal to 500, 600, 800, 900, and 1000 8C),are shown in Figures 1 and 2, respectively. The CO profile for

the GP precursor has a broad peak centered at approximately760 8C with a shoulder at about 920 8C. The GP-H2-500Csample has a large CO peak about 785 8C with a shoulder at880 8C, whereas the GP-H2-600C peak centers at about 895 8C.The amount of CO desorbed diminishes drastically after GP re-duction temperatures higher than 800 8C and desorption tem-peratures shift to values equal to or higher than 1000 8C.A progressive shift in the CO desorption peak is observed fromTPD in GP-H2-X samples with increasing reduction temperature.For example, the desorption peak shifts to higher temperaturewhen comparing GP with GP-H2-600C and GP-H2-600C withGP-H2-800C. Such a shift can be rationalized by looking at the

Figure 1. CO desorption profiles of five samples, GP-H2-X, from TPD and cal-culated decomposition temperatures of model functionalities: 1) isolatedcarbonyls on armchair edges, 2) dicarbonyl, phenols on armchair (3) andzigzag edges (4), 5) anhydride, 6) hydroquinone, and 7) quinone. Dicarbonylsand quinones are marked in red.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 484

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

decreasing oxygen concentration in samples; this results fromany or all of these possibilities: 1) desorption of thermally un-stable oxygen functionalities below a certain temperature,2) increased stability of certain functionalities with decreasedcoverage, and 3) change in the thermal stability upon reactionwith hydrogen.

CO2 desorption is quantitatively less than 20 % of the CO de-sorption for most of the samples. The series show a main peakcentered at 660 8C for all samples. The GP sample also hasa clear shoulder at 500 8C, which can be more or less pro-nounced for the other samples. Interestingly, a shift in thepeak is not observed in CO2, but the amount of CO2 desorp-tion falls drastically in samples that undergo reduction at tem-peratures higher than 600 8C. Desorption peaks with loweredintensity between 600 and 700 8C in samples reduced at andabove 800 8C might be due to readsorption of CO2 from at-mosphere. Further discussion on the type of oxygen function-alities and their thermal stability is presented later with DFTresults.

Oxygen functionalities on GP catalysts

The oxygen functionalities can be identified by TPD deconvo-lution because the functionalities differ in their thermal stabili-ty and the oxygen groups desorb as either CO or CO2 basedon their structure and coordination with carbon atoms.[14]

Hence, we modeled the decomposition reactions based on thefollowing assumptions: carbonyls, quinones, and ethers desorbas CO; phenols and hydroquinones desorb as CO + H2; lac-tones desorb as CO2; carboxylic acids desorb as CO2 + H2 ; andanhydrides desorb as CO + CO2. Although epoxides and hy-droxyls on the basal plane resulting from the synthesis of GP

oxide[13] are kinetically stable at room temperature; they arethermodynamically unstable with respect to O2(g).

[13b, 15] Anneal-ing above 250 8C results in decomposition of epoxides to formgas-phase products or conversion to ethers or carbonyls.[13b]

Hence, we do not expect epoxides and hydroxyls to be pres-ent on our catalysts after the reductive treatment at high tem-peratures and/or during catalytic performance tests performedat 400 8C.

The decomposition reaction for carbonyls, quinones andethers was modeled by using Equation (1):[16]

DGðT ,mCÞ ¼ EðGPÞ þ nEðCOÞ�EðGP,nOÞ�nCmC þ DEvib ð1Þ

in which DG(T,mC) represents the free energy of decomposi-tion, E(GP,nO) represents the GP sheet with one edge site oc-cupied with an oxygen functionality and remaining edge sitescapped with hydrogen, E(GP) represents the GP sheet after de-composition of the oxygen functionality with the unsaturatededge site, nCmC is the number of carbon atoms exchanged withthe reservoir mC (basal plane of GP),[17] and DEvib represents thevibrational contribution to free energy. The partial pressure ofCO(g) is set to 0.001 bar and the entropic contributions to freeenergy of CO(g) is taken from the CCSDB database from NIST.[18]

An increase in the partial pressure of CO(g) and/or increase inthe chemical potential of the reservoir increases the decompo-sition temperature. The model reactions and schematics forother functionalities are shown in Figures S1 and S2 in theSupporting Information. The temperature for decomposition offunctionalities is equal to the temperature at which the freeenergy of the reactions equals zero (Td). We compare calculat-ed decomposition temperatures and TPD measurements tocharacterize the oxygen functional groups based on their ther-mal stability.

The reported decomposition temperatures for functionalitieson carbon from previous studies[14, 19] using TPD measurementsare as follows: carbonyls and quinones (CO, 700–900 8C), phe-nols (CO, 600–700 8C), ethers (827 8C), carboxylic groups(CO2, 100–400 8C), anhydrides (CO, CO2, 800–900 8C), lactones(CO2, 350–667 8C), peroxides (CO2, 550–600 8C), and pyrones(CO2, 1000 8C). Our calculations agree with the reported orderof stability of various oxygen functionalities with respect toCO/CO2. An isolated carbonyl (or ketone) group on a zigzagedge is more stable (Td�1227 8C) than that of a similar groupon an armchair edge (Td�676 8C). The stability of the isolatedcarbonyl group on the zigzag edge decreases upon addinga second oxygen atom to the adjacent C6 ring to form a dicar-bonyl group (Td�764 8C). On an armchair edge, two oxygenatoms can bond to the same C6 ring to form a quinone groupthat is stable up to Td�938 8C. Phenols (hydrogenated carbon-yls) on armchair and zigzag edges and hydroquinones arestable up to Td<850 8C. Anhydrides decompose at Td�820 8C.Ethers and lactones at the zigzag edge remain stable up toTd>1000 8C, but lactones at armchair edges decompose below400 8C. COOH on zigzag and armchair edges decomposearound Td�400 8C.

Calculated desorption temperatures were compared withthose obtained by TPD to deconvolute the peaks and gain an

Figure 2. CO2 desorption profiles of five samples, GP-H2-XC, from TPD andcalculated decomposition temperatures of model functionalities: COOH onarmchair (1) and zigzag edges (3), 2) lactone on armchair edges, and 4) an-hydride.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 485

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

insight into the existence of specific functionalities at a certaintemperature. The important features observed in the measureddesorption profiles were changes in the onset temperature ofdesorption and peak positions in samples reduced at increas-ing temperatures. In GP samples (shown in Figure 1), a smalldesorption peak at low temperature, around 200 8C, anda broad peak with an onset at around 400 8C and a maximumat around 760 8C are prominent in the CO desorption profile.Based on calculated models, the broad peak between 600 and1000 8C should result from a combination of functionalities—carbonyls, phenols, anhydrides, and quinones. The onset tem-perature of around 400 8C could result from kineticsand a lower temperature of desorption of the modelfunctionalities at high coverage plus desorption offunctionalities not included in this paper. As notedearlier in this paper, the onset temperature of de-sorption for a certain functionality changes with theextent of coverage of that species on the samples.For example, our calculations show that the isolatedcarbonyl group on the zigzag edge is stable up toa temperature 1227 8C, but its desorption tempera-ture drops to 764 8C with one adjacent carbonyloxygen (dicarbonyl) and further decreases to 697 8Cwith adjacent carbonyl oxygen atoms on both sides.

The shift in the onset temperatures of desorptionpeaks observed in TPD can be explained in terms ofthe thermal stability of the functionalities. In GP-H2-800C samples, we expect some model functionalities(phenols, anhydrides, quinones, isolated carbonyls onzigzag edges, and ethers) and a reduced concentra-tion of dicarbonyls on the zigzag edge exist on thesamples. In GP-H2-900 and GP-H2-1000C, only qui-nones and other stable functionalities at tempera-tures greater than 1000 8C (lactones, ethers, and iso-lated carbonyls on zigzag edges) seem to exist.Although observed during our CO2 desorption measurements,our calculations do not show any peaks between 600 and700 8C. Possible reasons for this could be that we have notconsidered peroxides (Td�600 8C) and defects on basal planesand edges. Also, lactones and anhydrides can decomposethrough multiple pathways.[20]

The nature of the surface functionalities was also verified byXPS.[7] Both O 1 s and C 1 s X-ray photoelectron spectroscopy(XPS; see Figure S3 in the Supporting Information) similarlyreveal a complex C�O bonding configuration. XPS results forthe O 1 s region for the GP-H2-500C sample have been decon-voluted at 533.5, 532.2, and 530.6 eV; this indicates a mixtureof C�O (binding energy�533 eV) and C=O bonds (bindingenergy�531 eV).[21] In addition, a third O feature at approxi-mately 535 eV is observed and is attributed to adsorbed water.As the reduction temperature is increased to 1000 8C, there isa sharp decrease in intensity for the GP-H2-1000C sample overthe entire O 1 s range and the peak at 530.6 eV disappears.The C 1 s spectra are dominated by the peak of graphiticcarbon at approximately 284.5 eV and a small shoulder be-tween 285 and 290 eV corresponding to the C�O functionali-ties.[7]

Catalytic performance of GP catalysts

Each sample was tested for its catalytic performance in theODH of isobutane and an apparent activation energy was mea-sured by carrying out the reaction between 380 and 425 8C.This temperature interval was chosen based on prior stabilitywork,[7] which indicated that the functionalized few-layer GPsamples were stable up to temperatures of around 400 8C.In this temperature range, conversions of isobutane varied be-tween 1.1 and 6.7 % and selectivities to isobutene varied be-tween 20 and 60 %. Figure 3 shows the rate of isobutane con-

sumption as a function of temperature in an Arrhenius formfor a single space velocity and O2/iC4H10 ratio of 1:2.

Calculated reaction pathways for ODH of isobutane

We used DFT to calculate the reaction profiles for the ODHcycle and compared the activation barriers with measuredvalues to identify the functional group responsible for the cat-alytic activity. We chose two functionalities, dicarbonyls andquinones because two oxygen sites would be needed for de-hydrogenation of isobutane to isobutene; these functionalitiesare the majority among the various functionalities;[20] and datafrom our calculations suggest that dicarbonyls decomposecompletely in samples reduced above 800 8C, which wouldprovide us with two sets of samples reduced above and below800 8C to probe the catalytic activity.

A schematic for the ODH of isobutane by dicarbonyl, similarto the proposed mechanism,[2a] is shown in Figure 4. The firsthydrogen abstraction from isobutane starts with the dissocia-tion of either the tertiary or primary C�H bond and hydrogena-tion of the functional oxygen on GP. Following the first step,hydrogen abstraction from the butyl radical (tertiary or iso) by

Figure 3. ODH isobutane reaction activity of all carbon catalysts measured for 0.04 g ofcatalyst and a space velocity of 32 700 mL of iC4H10 per gram of catalyst per hour. Ratefor isobutane consumption is expressed per total amount of catalyst.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 486

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

the adjacent oxygen in the same functional group results inisobutene leaving the hydrogenated dicarbonyl group. Catalyt-ic site regeneration was modeled as O2(g) reacting with the hy-drogenated functionality to form H2O2. Further decompositionof H2O2 was not modeled in this paper; however, decomposi-tion of H2O2 to H2O and O2 is probable under the reactionconditions.

The calculated reaction profiles for ODH by dicarbonyls andquinones are shown in Figure 5. We first discuss the reactionprofiles in the case of dicarbonyls. The initial state correspondsto gas-phase isobutane and a dicarbonyl group on the zigzagedge. The reaction can proceed via an isobutyl radical or a terti-ary butyl radical. For hydrogen abstraction by the functionaloxygen, the reaction coordinate is along C�H bond stretching,while the formation of an O�H bond and the transition statealong the reaction coordinate is the activated configurationwith a bridging hydrogen between tertiary (or primary) carbonand the functional oxygen. Between the two pathways, the ac-tivation barrier for first hydrogen abstraction, with respect toisobutane in the gas phase, by tertiary C�H bond dissociation,is in the order of 80 kJ mol�1 and the alternate pathway via anisobutyl radical is higher in energy by 24 kJ mol�1 (based ontotal energies). Although the alternate pathway is not favoredover the first path, the probability of traversing it increases asa function of temperature and, at the reaction temperature(400 8C), the ratio of the Boltzmann probability is in the orderof 1.34 � 10�2. The C�H and O�H bond lengths at the transitionstate are 1.41 and 1.20 �, respectively, which are comparativelylonger than the C�H bond in isobutane (�1.10 �) and theO�H bond (�1.09 �) in the product. The final state after firsthydrogen abstraction corresponds to hydrogen attached toone of the two oxygen atoms and connected to the second

oxygen through a hydrogenbond (bond length�1.37 �),with an adsorbed tertiary (or pri-mary) butyl radical. Followingthe first hydrogen abstraction,the tertiary (or primary) butylradical loses one of its hydrogenatoms from one of the three pri-mary C�H bonds, resulting inisobutene and hydrogenation ofthe second oxygen on the func-tionality. The activation barrierrequired for dehydrogenation isin the order of 53 kJ mol�1, al-though tertiary and primarybutyl radicals favor losing hydro-gen to form a closed-shell spe-cies (isobutene). The transitionstate is similar to the first stepwith bridging hydrogen be-tween primary (or tertiary)carbon and the oxygen (withoutattached hydrogen) on the func-tionality. The final state corre-sponds to the configuration with

hydrogen atoms attached to both oxygen atoms on the dicar-bonyl group.

Regeneration of catalytic sites is critical to sustain the ODHcatalytic reaction and the calculated reaction energies for thesteps are shown in Figure 5. This regeneration proceeds by re-action between O2 in the gas phase with the hydrogenated di-

Figure 4. Schematic representation of the ODH of isobutane by dicarbonyl groups on the zigzag edge, similar tothe proposed mechanism given in Ref. [2] .

Figure 5. Calculated reaction pathways for the ODH of isobutane on dicar-bonyls at zigzag edges (upper) and quinones (lower) at armchair edges.Transition-state energies are shown in red. All energies are given in kJ mol�1.Carbon, oxygen, and hydrogen atoms are shown as gray, red, and whitespheres, respectively.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 487

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

carbonyl group. The initial state is O2 weakly adsorbed on theedge and the final state corresponds to the adsorbed OOHradical formed upon one hydrogen abstraction by O2. The re-action is exothermic by 68 kJ mol�1 and has a negligible activa-tion barrier. Further dehydrogenation to form H2O2 and regen-erate the active sites is endothermic with energy in the orderof 74 kJ mol�1. Interestingly, the activation barrier for first hy-drogen abstraction and the reaction energy for the final stepresulting in H2O2 are almost equal (80 kJ mol�1 activation barri-er for the first dehydrogenation step and 74 kJ mol�1 for regen-eration of active sites).

The reaction pathways are similar in quinones; the barrierfor first hydrogen abstraction is higher than the rest and, con-sequently, all steps are accessible once that barrier is over-come. The first hydrogen abstraction via a tertiary butyl radicalhas an activation barrier of 97 kJ mol�1 and, similar to dicarbon-yls, this pathway is favored over the alternate pathway via anisobutyl radical by 37 kJ mol�1. The C�H and O�H bondlengths at the transition state are 1.47 and 1.18 �, respectively.The final state after first hydrogen abstraction corresponds tohydrogen attached to one of the two oxygen atoms and con-nected to the second oxygen through a hydrogen bond (bondlength�1.67 �), with an adsorbed butyl radical. There is no ac-tivation barrier for the second hydrogen abstraction from bothtertiary and isobutyl radicals and the fully hydrogenated state(hydroquinone) is stable by 44 kJ mol�1 compared to the initialconfiguration. Both steps involved in the regeneration of qui-nodic sites, the formation of OOH and H2O2, are uphill inenergy, as opposed to regeneration of dicarbonyls. Hydrogenabstraction by O2(g) to oxidize hydroquinone and form OOH isendothermic with an activation barrier and energy differencein the order of 44 and 20 kJ mol�1, respectively. The conversionof OOH into H2O2 is also endothermic with an energy differ-ence in the order of 20 kJ mol�1.

To rationalize the behavior of hydrogen abstraction by O2

during regeneration, we calculated the binding energy of hy-drogen to the dicarbonyls and quinones. The binding energieswith respect to H2(g) are in the order of approximately190 kJ mol�1 for partial hydrogenation (adding one H to dicar-bonyl) and less than zero for adding a second hydrogen. In thepresence of a strong reductant such as tertiary and isobutylradicals, the second hydrogen binding to the dicarbonyl groupis weakly favored; hence, the dicarbonyl group likes to losethe second hydrogen, even in presence of a mild oxidant, andthe reaction with O2(g) to form an adsorbed OOH radical is fa-vored. Further dehydrogenation to form H2O2 is energeticallyunfavorable and highly exothermic because of strongly boundhydrogen in the partially hydrogenated state of the dicarbonylgroup. In quinones, hydrogen binding is stronger than that ofdicarbonyls; partial hydrogenation is exothermic by117 kJ mol�1 and full hydrogenation to hydroquinone is alsoexothermic by 71 kJ mol�1 with respect to H2(g). Hence, bothsteps during regeneration of quinodic sites are endothermic.

There is also the possibility of an alternative reaction pathduring hydrogen abstraction from isobutyl radicals (not shownin Figure 5), in which the probability of the radical binding tothe hydrogen-free functional oxygen or the basal plane is

slightly higher than that of losing hydrogen. The energy differ-ences between the isobutyl radical bound to the oxygen func-tionality and isobutene adsorbed on the hydrogenated func-tionality are in the order of 12 kJ mol�1 and close to zero at di-carbonyls and quinones, respectively. Insignificant energy dif-ferences in the final states indicate competition betweenlosing hydrogen and forming a C�O bond with the functionaloxygen. No further reaction pathways were constructed in thisstudy to conclude the mechanism of further oxidation ofbound isobutyl radicals. The chemisorbed isobutyl radicalscould result in isobutene upon loss of the beta hydride anddissociation of the C�O bond. Otherwise, without dissociationof the C�O bond, it could be the onset of total oxidation toCO/CO2 because there are no evident closed-shell species thatcould result upon losing one or more hydrogen or carbonatoms.[22] The ratio of isobutyl radicals that undergo total oxi-dation to CO/CO2 should contribute to the lower selectivity toisobutene. Our calculations of isobutyl radicals binding to cata-lysts agree with recent computational studies using clustermodels of carbon catalysts and hydrocarbons, which reportedthe formation of C�OR species without a barrier, based on thespin state of the catalysts, if the nascent radicals reacted withneighboring quinone groups.[9] As opposed to isobutyl radicals,tertiary butyl radicals prefer to adsorb and not form a chemicalbond with the functional oxygen; hence losing hydrogenseems to be the only pathway that results in isobutene.

DFT-GGA (GGA = generalized-gradient approximation) failsto include van der Waals (vdW) forces, and hence, underesti-mates the binding of adsorbates that do not have a significantoverlap in charge density with surfaces.[23] For example, in theinitial state in which isobutane is adsorbed on the dicarbonylgroup at the zigzag edge, GGA results in an adsorption energyless than zero, but GGA + vdW-DFT results in an adsorptionenergy of 13 kJ mol�1. Hence, we calculated the vdW correc-tions to the total energies to estimate the contribution fromdispersion interactions to the relevant energetics. The activa-tion barriers for hydrogen abstraction by dicarbonyls and qui-nones with respect to gas-phase isobutane were lowered by2–10 kJ mol�1. The formation of H2O2 from the adsorbed OOHradical was also lowered by 4 kJ mol�1. In addition to vdW cor-rections, zero-point energies were also calculated; this loweredthe activation barriers by 13–15 kJ mol�1. The entropic contri-butions that increase as a function of temperature (not includ-ed herein) contribute to an increase in the activation barriersin the same order as zero-point energies.

Proposed catalytic mechanism of ODH of isobutane

Oxygen functionalities on GP with their varying nucleophiliccharacter introduce a variety of catalytic sites. We consolidatethe data from theory and experiments to explain the role offunctionalities on GP in the ODH of isobutane. Two observa-tions can be drawn from the TPD and activation energy meas-urements on samples reduced at increasing temperatures:1) the CO desorption peak shifts to higher temperatures, and2) the activation energy increases. The activation energies forGP-H2-XC (X = 500 and 600 8C) are in the range between 76

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 488

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

and 78 kJ mol�1, whereas in the case of GP-H2-1000C the acti-vation energy is as high as 96 kJ mol�1. To rationalize the shiftin the CO desorption peak, we compared the thermal stabilityof model functionalities and concluded that the functionalitiespresent on GP-H2-XC sample that became thermally unstableat the reduction temperature (X 8C) desorbed during pretreat-ment and did not contribute to the desorption peaks fromTPD of that sample. Based on the calculated desorption tem-peratures, carbonyls at armchair edges and dicarbonyls atzigzag edges decompose below 800 8C, phenols and anhy-drides decompose between 800 and 900 8C, and quinones de-compose at around 1000 8C. XPS O 1 s results for GP-H2-500Cand GP-H2-1000C were consistent with the TPD observationsthat groups containing C�O single bonds were mostly re-moved with increased reduction temperatures and that qui-nones and carbonyls were present on both samples. In thecase of GP-H2-1000C, it is very likely that quinones would bepresent as hydroquinones, causing a shift in the bindingenergy from the typical value of 531 to 532.2 eV.[24] However, itis not possible to differentiate dicarbonyls from quinonegroups by means of XPS.

To rationalize the difference in the measured activation ener-gies, we compared the calculated activation barriers from di-carbonyls and quinones. The calculated lowest activation barri-ers (including zero-point energies and vdW corrections) forODH of isobutane are in the order of 55 and 72 kJ mol�1 for di-carbonyls and quinones, respectively. From the calculated ther-mal stability and relative activation barriers, both dicarbonylsand quinones are responsible for the activity in samples re-duced below 800 8C with a greater contribution from dicarbon-yls compared with quinones, and quinones play a major role insamples reduced above 800 8C.

The activation barriers for the ODH of isobutane by otherfunctionalities were not calculated herein; however, taking theadvantage of the universal relationship between the activationenergy and binding energy,[25] we could estimate the activationenergy required for C�H bond dissociation at those functional-ities when compared with those of dicarbonyls and quinones.Zboray et al. reported the role of C�H bonds in the selectiveoxidation of alkanes on alumina-supported VOx catalysts, forwhich the activation energy of C�H bond dissociation was pro-portional to its bond energy and the apparent activationenergy depended on the weighted-average number of suchbonds to be activated; this suggested that the ODH of alkanessatisfied the Brønsted–Evans–Polanyi (BEP) relationship.[26]

Pallassana also reported that, during hydrogenation of ethyl-ene to an ethyl functionality, reverse C�H bond breaking wasfacile on metal surfaces that strongly bind decomposed prod-ucts (ethylene and H); this suggested the dependence of anactivation barrier on metal–adsorbate strength that satisfiedthe BEP relationship.[27] We have also seen a similar correlationbetween the binding energy of hydrogen and the regenera-tion of dicarbonyls groups and quinones, as explained in theprevious section, to understand the behavior of hydrogen ab-straction by O2(g) to regenerate the active sites. Hence, basedon the BEP relationship, for a given C�H bond (primary or terti-ary) in isobutane, we can expect a linear dependence of the

activation energy on the binding energy of hydrogen to a func-tional group. We calculated binding energies of hydrogen tothe stable functionalities that remained at various reductiontemperatures. We found that the hydrogen binding strength(one H per functional group) decreased in the order dicarbon-yls>quinones> lactones>anhydrides. Therefore, the majorcontribution to the rate is from dicarbonyls and quinones, andthe other groups, if present, do not actively participate inODH.

The catalytic mechanism involves weakly adsorbed isobu-tane reducing dicarbonyls/quinones and leaving as isobutenethen O2 in the feed weakly adsorbs and reacts with the hydro-genated functionality, leaving as H2O2 and regenerating thecatalytic sites. This mechanism is similar to that proposed bySchlçgl and co-workers,[10] which they classified as a dual-siteLangmuir–Hinshelwood-type mechanism. The calculated acti-vation barriers for the removal of the O functionality as H2Owere high (>150 kJ mol�1) ; this excludes the possibility ofa Mars–van Krevelen-type mechanism seen in metal oxidecatalysts.

Conclusions

Our combined experimental and theoretical study enabled theidentification of the oxygen functionalities that should behighly active for the selective ODH of isobutane to isobuteneand provided a proposed mechanism for the reaction byoxygen-functionalized, few-layered GP sheets. We concludethat dicarbonyls at the zigzag edges and quinones at armchairedges are appropriately balanced for high activity comparedwith other model functionalities considered herein. In the ODHof isobutane, both dehydrogenation and regeneration of cata-lytic sites are relevant at the dicarbonyls, whereas regenerationis facile compared with dehydrogenation at quinones. Our cal-culations exclude the possibility of H2O formation througha Mars–van Krevelen mechanism believed to occur on metaloxide surfaces. Exploiting the thermal stability of the function-alities gives the opportunity to tailor the oxygen functionalitieson carbon catalysts, and, consequently, of tuning the catalyticactivity for ODH reactions of alkanes.

Experimental Section

Synthesis of materials: A series of controlled experiments generat-ing TPD profiles and measuring activation energies and atomisticmodeling by using DFT were performed to identify the activeoxygen functionalities on GP for the ODH of isobutane. The precur-sor, GP oxide, was prepared by the modified Hummers methodand few-layered GP sheets were exfoliated from the precursor byusing microwave radiation for 30–120 s. Reductive treatment atambient pressure for 10 h in 4 % H2/He at variable temperaturesbetween 500 and 1000 8C was performed to tailor the concentra-tion and type of the oxygenated functional groups on the surfaceand edges of the few-layered GPs. The GP samples prepared bythis systematic approach were labeled GP-H2-XC, (X = 500, 600, 800,900, and 1000 8C). For further details on syntheses, characterization,and experimental methods, please refer to our previous publica-tion on few-layered GP catalysts for ODH.[7]

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 489

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

Temperature-programmed desorption: TPD was used to quantifyand investigate the nature and stability of the surface oxygen. Theexperiments were done in a U-tube reactor (Altamira AMI-200) ina flow of helium (20 mL min�1) with a heating rate of 10 C min�1

from room temperature to 1000 8C. During TPD, surface oxygengroups on carbon materials decompose upon heating in He by re-leasing CO, CO2, and H2O, which were probed by a quadrupolemass spectrometer (Pfeiffer-Balzer Omnistar).

Catalytic performance: The catalytic performance test for the iso-butane ODH reaction was performed at atmospheric pressure ina packed-bed stainless-steel autoclave reactor (PID Eng&Tech,Spain).[11b] Functionalized GP catalyst (0.04 g) mixed with quartzsand (2 g) was placed in the reactor for the reaction test. The feedcomposition for the reaction was 4 % iC4H10 and 2 % O2 in He foran isobutane/oxygen ratio of 2/1 and a total flow rate of the feedof 20 sccm. The reaction temperature was varied between 380 and425 8C to obtain activation energy values. The reactants and prod-ucts were analyzed by online gas chromatography conducted onan Agilent 6890N gas chromatography system with a Hayesep Ncolumn and a molecular sieve column. CO and CO2 were convertedinto CH4 through a methanizer before they passed througha flame ionization detector for analysis. Isobutene, CO, and CO2

were the only products detected. In all tests, carbon mass balanceswere within (100�0.5) %.Computational methods: DFT calculations were performed withinthe generalized-gradient approximation (GGA-PBE),[28] as imple-mented in the Vienna ab initio simulation package (VASP).[29] Thevalence electrons were described by Kohn–Sham one-electron or-bitals and core electrons with projector-augmented wave-basedpseudo-potentials.[30] All calculations were spin-polarized by usinga plane-wave cutoff of 400 eV and a 5 � 1 � 1 k-point mesh. Geome-try optimizations were considered to be converged when the forceon each atom was below 0.02 eV ��1. Activation barriers were cal-culated by using the climbing image nudged elastic band (CI-NEB)method.[31] Vibrational contributions to free energy were calculatedby using density functional perturbation theory (DFPT) and vdWcorrections to total energy were calculated by using the vdW-DFfunctional[32] implemented in VASP. GP models were built as nano-ribbons (1D periodic) with four terminal carbon atoms at one edgein a unit cell. Two types of edge terminations were considered:armchair and zigzag. The ribbons are periodic along the x axis andstacked along the z axis with a vacuum spacing of 10 � betweenthe layers. Along the y axis, hydrogen atoms capped carbon atomsat one edge, representing bulk-terminated GP, and the other edgewas allowed to relax if oxygen functionalities (Figure 6) were graft-ed onto selected carbon atoms with the rest of carbon atoms onthe free edge capped by hydrogen atoms. A vacuum spacing of25 � along the y axis was used to separate two edges and thedipole correction term[33] was applied along the y axis to avoid in-teractions between the periodic images of GP ribbons.

Acknowledgements

This work was conducted at the Center for Nanophase MaterialsSciences, Oak Ridge National Laboratory (ORNL), which is spon-sored by the Scientific User Facilities Division, Office of BasicEnergy Sciences, US Department of Energy (US-DOE). Computingresources provided by the National Energy Research ScientificComputing Center, which is supported by US-DOE Office of Sci-ence under Contract DE-AC02-05CH11231; by the Oak RidgeLeadership Computing Facility, which is supported by US-DOE

Office of Science under Contract DE-AC05-00OR22725 ; and bythe Oak Ridge National Laboratory, which is managed by UT-Bat-telle, LLC, for US-DOE, were used and are gratefully acknowl-edged. We thank Dr. Zili Wu for helpful discussions related tocharacterization of graphene samples and the catalytic mecha-nism of ODH.

Keywords: density functional calculations · carbon ·graphene · nanostructures · oxidative dehydrogenation

[1] a) N. Hamilton, T. Wolfram, G. Tzolova M�ller, M. H�vecker, J. Krçhnert,C. Carrero, R. Schom�cker, A. Trunschke, R. Schlçgl, Catal. Sci. Technol.2012, 2, 1346 – 1359; b) S. D. Jackson, P. C. Stair, L. F. Gladden, J. McGre-gor in Metal Oxide Catalysis (Eds. : S. D. Jackson, J. S. J. Hargreaves),Wiley-VCH, Weinheim, 2008, pp. 595 – 612; c) X. Rozanska, R. Fortrie, J.Sauer, J. Phys. Chem. C 2007, 111, 6041 – 6050; d) M. V. Mart�nez-Huerta,X. Gao, H. Tian, I. E. Wachs, J. L. G. Fierro, M. A. BaÇares, Catal. Today2006, 118, 279 – 287; e) S. Yang, E. Iglesia, A. T. Bell, J. Phys. Chem. B2005, 109, 8987 – 9000; f) T. V. M. Rao, G. Deo, J.-M. Jehng, I. E. Wachs,Langmuir 2004, 20, 7159 – 7165; g) K. Chen, A. T. Bell, E. Iglesia, J. Catal.2002, 209, 35 – 42.

[2] a) D. S. Su, J. Zhang, B. Frank, A. Thomas, X. Wang, J. Paraknowitsch, R.Schlçgl, ChemSusChem 2010, 3, 169 – 180; b) D. S. Su, S. Perathoner, G.Centi, Chem. Rev. 2013, 113, 5782 – 5816.

[3] a) X. Liu, B. Frank, W. Zhang, T. P. Cotter, R. Schlçgl, D. S. Su, Angew.Chem. 2011, 123, 3376 – 3380; Angew. Chem. Int. Ed. 2011, 50, 3318 –3322; b) J. Zhang, X. Liu, R. Blume, A. Zhang, R. Schlçgl, D. S. Su, Science2008, 322, 73 – 77.

[4] C. Liang, H. Xie, V. Schwartz, J. Howe, S. Dai, S. H. Overbury, J. Am.Chem. Soc. 2009, 131, 7735 – 7741.

[5] T.-J. Zhao, W.-Z. Sun, X.-Y. Gu, M. Rønning, D. Chen, Y.-C. Dai, W.-K. Yuan,A. Holmen, Appl. Catal. A 2007, 323, 135 – 146.

[6] D. Su, N. I. Maksimova, G. Mestl, V. L. Kuznetsov, V. Keller, R. Schlçgl, N.Keller, Carbon 2007, 45, 2145 – 2151.

[7] V. Schwartz, W. Fu, Y.-T. Tsai, H. M. Meyer, A. J. Rondinone, J. Chen, Z.Wu, S. H. Overbury, C. Liang, ChemSusChem 2013, 6, 840 – 846.

[8] M. P. Woods, E. J. Biddinger, P. H. Matter, B. Mirkelamoglu, U. S. Ozkan,Catal. Lett. 2010, 136, 1 – 8.

Figure 6. DFT-optimized configurations of model functionalities on GPedges. Gray, red, and white colored balls represent carbon, oxygen, and hy-drogen, respectively.

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 490

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org

[9] O. V. Khavryuchenko, B. Frank, A. Trunschke, K. Hermann, R. Schlçgl, J.Phys. Chem. C 2013, 117, 6225 – 6234.

[10] J. Zhang, D. Su, A. Zhang, D. Wang, R. Schlçgl, C. H�bert, Angew. Chem.2007, 119, 7460 – 7464.

[11] a) V. Schwartz, H. Xie, H. M. Meyer, S. H. Overbury, C. Liang, Carbon2011, 49, 659 – 668; b) H. Xie, Z. Wu, S. H. Overbury, C. Liang, V.Schwartz, J. Catal. 2009, 267, 158 – 166.

[12] a) N. P. Subramanian, S. P. Kumaraguru, H. Colon-Mercado, H. Kim, B. N.Popov, T. Black, D. a. Chen, J. Power Sources 2006, 157, 56 – 63; b) Y.Nabae, I. Yamanaka, K. Otsuka, Appl. Catal. A 2005, 280, 149 – 155;c) J. A. Maci-Agull, D. Cazorla-Amors, A. Linares-Solano, U. Wild, D. S.Su, R. Schlçgl, Catal. Today 2005, 102 – 103, 248 – 253; d) M. F. R. Pereira,J. J. M. �rf¼o, J. L. Figueiredo, Appl. Catal. A 2000, 196, 43 – 54; e) M. F. R.Pereira, J. J. M. �rf¼o, J. L. Figueiredo, Appl. Catal. A 1999, 184, 153 –160.

[13] a) D. R. Dreyer, S. Park, C. W. Bielawski, R. S. Ruoff, Chem. Soc. Rev. 2010,39, 228 – 240; b) A. Bagri, C. Mattevi, M. Acik, Y. J. Chabal, M. Chhowalla,V. B. Shenoy, Nat. Chem. 2010, 2, 581 – 587.

[14] Z. Karpinski, G. S. Szymanski, Carbon 2002, 40, 2627 – 2639.[15] a) S. Zhou, A. Bongiorno, Sci. Rep. 2013, 3, 2484; b) S. Kim, S. Zhou, Y.

Hu, M. Acik, Y. J. Chabal, C. Berger, W. de Heer, A. Bongiorno, E. Riedo,Nat. Mater. 2012, 11, 544 – 549.

[16] Y. Xu, W. a. Shelton, J. Electrochem. Soc. 2011, 158, A1177 – A1177.[17] J. Carlsson, M. Scheffler, Phys. Rev. Lett. 2006, 96, 046806.[18] NIST Computational Chemistry Comparison and Benchmark Database,

NIST Standard Reference Database Number 101, Release 16a, August2013 (Ed. : R. D. Johnson III): http://cccbdb.nist.gov.

[19] a) J. L. Figueiredo, M. F. R. Pereira, M. M. A. Freitas, J. J. M. �rf¼o, Carbon1999, 37, 1379 – 1389; b) U. Zielke, K. J. H�ttinger, W. P. Hoffman, Carbon1996, 34, 983 – 998.

[20] M. Acik, G. Lee, C. Mattevi, A. Pirkle, R. M. Wallace, M. Chhowalla, K.Cho, Y. Chabal, J. Phys. Chem. C 2011, 115, 19761 – 19781.

[21] a) B. Frank, A. Rinaldi, R. Blume, R. Schlçgl, D. S. Su, Chem. Mater. 2010,22, 4462 – 4470; b) T. I. T. Okpalugo, P. Papakonstantinou, H. Murphy, J.McLaughlin, N. M. D. Brown, Carbon 2005, 43, 153 – 161; c) D. Rosenthal,M. Ruta, R. Schlçgl, L. Kiwi-Minsker, Carbon 2010, 48, 1835 – 1843.

[22] a) R. Schlçgl in Modern Heterogeneous Oxidation Catalysis (Ed. : N.Mizuno), Wiley-VCH, Weinheim, 2009, pp. 1 – 42; b) A. Costine, B. K.Hodnett, Appl. Catal. A 2005, 290, 9 – 16.

[23] J. Klimes, A. Michaelides, J. Chem. Phys. 2012, 137, 120901 – 120901.[24] S. Sundaram, S. K. Annamalai, Electrochim. Acta 2012, 62, 207 – 217.[25] a) R. A. van Santen, M. Neurock, S. G. Shetty, Chem. Rev. 2010, 110,

2005 – 2048; b) J. K. Nørskov, T. Bligaard, A. Logadottir, S. Bahn, L. B.Hansen, M. Bollinger, H. Bengaard, B. Hammer, Z. Sljivancanin, M. Mavri-kakis, Y. Xu, S. Dahl, C. J. H. Jacobsen, J. Catal. 2002, 209, 275 – 278.

[26] M. Zboray, A. T. Bell, E. Iglesia, J. Phys. Chem. C 2009, 113, 12380 – 12386.[27] V. Pallassana, J. Catal. 2000, 191, 301 – 317.[28] J. P. Perdew, Y. Wang, Phys. Rev. B 1992, 45, 13244 – 13249.[29] a) G. Kresse, J. Furthm�ller, Comput. Mater. Sci. 1996, 6, 15; b) G. Kresse,

J. Furthm�ller, Phys. Rev. B 1996, 54, 11169.[30] P. E. Blçchl, Phys. Rev. B 1994, 50, 17953.[31] G. Henkelman, B. P. Uberuaga, H. Jonsson, J. Chem. Phys. 2000, 113,

9901 – 9904.[32] J. Klimes, D. R. Bowler, A. Michaelides, Phys. Rev. B 2011, 83, 195131 –

195131.[33] J. Neugebauer, M. Scheffler, Phys. Rev. B 1992, 46, 16067 – 16080.

Received: September 20, 2013Revised: November 12, 2013Published online on January 24, 2014

� 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemSusChem 2014, 7, 483 – 491 491

CHEMSUSCHEMFULL PAPERS www.chemsuschem.org