CATALYST PREPARATION METHODS TO REDUCE CONTAMINANTS IN A HIGH-YIELD PURIFICATION PROCESS OF MULTIWALLED CARBON NANOTUBES

- CoMo/MgO catalysts with low loading of cobalt and molybdenum were prepared by incipient wetness impregnation and the sol-gel method for high yield production of carbon nanotubes (CNT). A detailed characterization of the catalysts, as-produced CNT and purified CNT, indicated a relationship between carbon nanotube properties, synthesis and purification steps. Excess of molybdenum in the catalyst caused a decrease in carbon yield and produced CNT with a lower degree of graphitization. Catalysts prepared by the sol-gel method and with higher molybdenum loading showed an improved interaction between cobalt and molybdenum, increasing the formation of CoMoO 4 , which was not completely removed during the purification step. The purification removed all MgO and part of the cobalt and molybdenum present in samples without significant damage to the CNT structure. An increase in sample purity could be achieved by increasing reaction time.


INTRODUCTION
The production of carbon nanotubes has become interesting due to their unique properties since first reported by Ijima (1991). Then, the development and optimization of nanotube synthesis methods have been the subject of research, seeking higher carbon yields and selective processes towards materials with superior purity and quality (Xu et al., 2011). Their remarkable properties, such as high elastic modulus and tensile strength, high flexibility and elasticity, high electrical and thermal conductivity, allow the application of CNT in different fields. Studies involving the use of these materials cover areas of energy storage, biotechnology, microelectronics and catalysis (Liu et al., 2014;Wang et al., 2018).
The most commonly cited methods for carbon nanotube synthesis are laser ablation, arc-discharge and chemical vapor deposition (CVD). Among them, the CVD technique is a viable option for large-scale production, since it allows a more effective control of reaction conditions, favoring the formation of nanotubes with high quality structure and morphology. Besides, it may be carried out in milder temperatures, thereby lowering production costs (Liu et al., 2014).
Cobalt, iron and nickel catalysts, supported on silica or alumina, are frequently used in the CVD method (Tran et al., 2007). Additionally, bimetallic catalysts have demonstrated higher efficiency than monometallic catalysts in nanotube production. The addition of a promoter, such as molybdenum, vanadium or tungsten, originates a stabilizing effect, preventing the agglomeration of the catalyst active phase and improving yield and graphitization of the nanotubes (Jourdain and Bichara, 2013).
Cobalt and molybdenum supported on SiO 2 and Al 2 O 3 catalysts were extensively studied by several groups (Kitiyanan et al., 2000;Alvarez et al., 2001;Flahaut et al., 2003;Serquis et al., 2003;Chai et al., 2011) and catalyst parameters that influence the production and morphology of the carbon nanotubes are well known. However, the purification of CNT synthetized by catalysts supported on SiO 2 and Al 2 O 3 demands the use of strong acids, such as nitric acid, sulfuric acid and hydrofluoric acid. These acids may damage the carbon nanotube structure. Magnesium oxide is a good choice as catalyst support since it is easily removed through a mild hydrochloric acid treatment. The use of easily soluble catalysts in CNT production allows the application of efficient and less destructive purification processes (Yeoh et al., 2009). Moreover, it minimizes the formation of amorphous carbon and the agglomeration of metal particles during the nanotube growth process (Pudukudy et al., 2017).
CoMo/MgO catalysts with low content of cobalt and molybdenum exhibit high yield of SWNT per gram of catalyst (Ni et al., 2006;Yardimci et al., 2015). High yield of MWNT is achieved with catalysts with high cobalt and molybdenum content, usually higher than 25 wt.% (Ni et al., 2006;Yeoh et al., 2009Yeoh et al., , 2013Núñez et al., 2011;Milone et al., 2014;Awadallah et al., 2014;Yardimci et al., 2015). The yield increases and the use of low-cost feedstocks, such as methane, in CNT production have decreased MWNT prices. However, for incorporation of carbon nanotubes into commonly used materials, they need to be produced on a large-scale and according to predetermined specifications and essential requirements for their industrial use.
Large-scale processes yield contaminants that can change CNT properties and usually require costly chemical treatment and thermal annealing for their removal. Thus, the purification of CNT synthesized by CVD is an important step to achieve the requirements for their applications. For example, the hydrogen uptake capacity of SWNT requires a low amount of metal particles in order to obtain high surface area (Mohanapriya and Lakshminarayanan, 2007). Also, defects within MWNT would reduce their electrical properties, which are very important in lithium ion batteries where a small amount of MWNT provides an increased electrical connectivity and mechanical integrity (De Volder et al., 2013).
Thus, this paper discusses the production and purification of high yield MWNT synthesized through methane CVD in a fluidized bed reactor over CoMo/ MgO catalysts. The catalysts were prepared by the two most usual preparation methods: incipient wetness impregnation and sol-gel techniques. Two different loadings of cobalt and molybdenum were used for each catalyst preparation method: a typical loading of 5 wt.% of Co and 20 wt.% of Mo and another with a lower molybdenum loading, 5 wt.% of Co and 10 wt.% of Mo. A lower molybdenum loading was used with the aim of reducing the concentration of contaminants in the final product. Then, a detailed characterization of the catalysts, as-produced MWNT and, especially, of the purified MWNT are related to the carbon nanotube synthesis and purification process.

EXPERIMENTAL
CoMo/MgO catalysts were prepared by incipient wetness impregnation and the sol-gel method. The amount of precursors used in both methods was suitable to obtain catalysts with weight ratios of 5 6H 2 O (98%, Sigma-Aldrich) along with 10 g of citric acid. Then, the mixtures were stirred at 90 °C for 1 h. The resulting viscous mixtures were dried at 120 °C and calcined at 550 °C. The catalysts were denoted as 5Co10Mo Imp , 5Co20Mo Imp , 5Co10Mo SG and 5Co20Mo SG .
The CNT production was performed in a vertical fluidized bed reactor, using 1 g of catalyst, by the methane CVD method. The process was performed by continuous heating from room temperature to 800 °C at a rate of 10 °C/min. Initially, the catalyst was heated under a H 2 flow (150 mL/min) until 500 °C. Then, the flow was switched to N 2 (150 mL/min) until the synthesis temperature of 800 °C was reached. At the synthesis temperature, the N 2 flow was replaced by CH 4 gas (150 mL/min) for 30 min. At the end of the reaction, CH 4 gas was turned off and the reactor was cooled to room temperature under a N 2 flow. The as-produced CNT were denoted as CNT.5Co10Mo Imp , CNT.5Co10Mo SG , CNT.5Co20Mo Imp and CNT.5Co20Mo SG .
For purification of the as-produced CNT, 0.4 g of as-produced CNT were macerated and then mixed with 100 mL of a 0.3 M HCl solution. Each solution was agitated for 2 h, sonicated for 30 min and agitated again for 2 h. The solution was aged for 8 h and then it was filtrated and rinsed with deionized water until pH 7. Finally, the product was dried at 120 °C for 12 h. The purified samples were designed as CNTp.5Co10Mo Imp , CNTp.5Co10Mo SG , CNTp.5Co20Mo Imp and CNTp.5Co20Mo SG .
X-ray diffraction (XRD) analysis was performed with a Rigaku Miniflex II X-ray diffractometer using monochromatic Cu-Kα radiation (λ = 1.540 Å). The XRD patterns of the catalysts, as-produced and purified CNT were recorded in a 2θ range of 10°-100°, with a step of 0.02° and measuring time of 1 s per point.
Surface area, average pore diameter and pore volume of the calcined catalysts were determined by N 2 adsorption in a Micromeritics ASAP 2020 analyzer, with sample weights around 0.2-1.0 g. The samples were dried at 150 °C and degassed at 150 °C for 2 h under vacuum before analysis. The surface area was obtained by the BET (Brunauer, Emmett and Teller) method and the average pore diameter and pore volume were obtained by the BJH (Barrett, Joyner and Halenda) method.
Energy dispersive X-ray spectrometry (EDX) analysis of the calcined catalyst and purified samples were performed with a Shimadzu EDX-720 spectrometer in vacuum (< 30 Pa) to determine the chemical compositions of the materials.
Temperature programmed reduction (TPR) experiments were carried out in an U-shaped quartz reactor using 200-300 mg of calcined catalyst. The samples were dried at 300 °C under a He flow of 30 mL/ min during 50 min. After cooling, the reduction was performed with a mixture of 5% H 2 /Ar at 30 mL/min from room temperature to 1000 °C with a heating rate of 10 °C/min and the temperature was held constant at 1000 °C for 1 h. The H 2 consumption was monitored with a Pfeiffer Prisma mass spectrometer connected to the equipment.
X-ray photoelectron spectroscopy (XPS) analysis for the catalysts was performed with a Thermo Scientific ESCALAB 250Xi spectrometer using a monochromatic Al-Kα X-ray source (1486,7 eV). High-resolution spectra were collected with a pass energy of 25.0 eV, energy step of 0.05 eV and dwell time of 50 ms. Charging effects were minimized with a low energy electron flood gun. During analysis, the pressure was kept around 10 -8 mbar inside the analysis chamber. The peak positions in the spectra were corrected using as reference the C 1s peak (284.5 eV).
Thermogravimetric analysis (TGA) was performed with a Shimadzu DTG-60H differential thermal gravimetric analyzer to quantify the weight loss of the as-produced and purified CNT. The analysis was carried out in an atmosphere of 5% O 2 /He with a flow rate of 30 mL/min from room temperature to 1000 °C with a heating rate of 5 °C/min. The carbon yield was calculated using Eq. (1): where m inpuruties is the weight of cobalt, molybdenum and magnesium at the end of the TGA.
Temperature programmed oxidation (TPO) experiments were carried out in an U-shaped quartz reactor using 20 mg of the as-produced and purified CNT. The samples were dried at 200 °C for 30 min under He with a flow rate of 30 mL/min. After cooling, the oxidation was performed with a mixture of 5% O 2 /He with a flow rate of 30 mL/min from room temperature to 1000 °C at a rate of 5 °C/min. The CO 2 production was monitored with a Pfeiffer Prisma mass spectrometer connected to the equipment.
Raman spectra of the catalysts, as-produced and purified CNT were obtained in a Witec alpha 300 Confocal Raman Microscope, using a 50x objective lens and a 532 nm excitation laser wavelength. The diameter of the single walled carbon nanotubes was estimated by equation 3 (Rols et al. 2000): where m carbon is the carbon weight loss during TGA analysis and m catalyst is the weight of the catalyst after sample oxidation obtained at the end of the TGA. The sample purity was estimated by Eq. (2) where ν RBM (cm -1 ) is the wavelength and d is the nanotube diameter (nm).
The as-produced and purified CNT were also characterized by scanning electron microscopy (SEM) in a JEOL JSM-7100F microscope and by transmission electron microscopy (TEM) with a JEOL JEM-1011 microscope. The samples were prepared by sonication of the as-produced CNT in isopropyl alcohol. A few drops of the resulting suspension were deposited on a copper support grid and dried at room temperature.

Characterization of the catalysts
The chemical compositions of catalysts and purified CNT were analyzed by EDX. The catalyst composition is consistent with the nominal values used in the catalyst preparation (Table 1). The EDX spectrometer used in the experiments is equipped with a beryllium window, so the detectors cannot detect X-rays from elements lighter than sodium, including carbon and oxygen X-rays. Thus, MgO weight percentage was estimated by stoichiometry.
The values of surface area (S BET ), pore volume (V p ) and the average pore diameter of catalysts (D p ) prepared by the impregnation method were slightly higher than the values found for the support. This behavior is due to the destruction of the MgO original texture with face centered cubic structure in function of its hydration during the impregnation step, which leads to the production of Mg(OH) 2 with hexagonal close-packed structure (Chang et al., 1992). H 2 O evolution during calcination, along with the structure changes, caused a surface area enhancement (Zdražil, 2003). The low surface area observed for the catalysts prepared by the sol-gel method may be related to the phenomenon of pore collapse. As already reported by Hench and West (1990), solvent evaporation during the drying process of the catalyst can cause the collapse of its structure, in particular, the collapse of pores smaller than 20 nm, reducing its surface area due to the formation of a xerogel.
XRD analyses were performed to identify and analyze the crystallite phases in the CoMo/MgO catalysts, as shown in Figure 1. The most intense peaks of the catalysts were identified as diffraction lines of MgO (Wang et al. 2012) in 2θ = 36.9°; 42.8°; 62.3°; which match with the standard data (JCPDS no. 45-0946).
Intense peaks were also observed at 2θ = 19.1°, 23.0° and 26.3°, which can be assigned to the most intense peaks of MgMoO 4 or CoMoO 4 patterns (JCPDS no. 72-2153 and 21-868 respectively). These species have approximately the same diffraction pattern. The formation of these species has been observed in the case of CoMo/MgO catalysts calcined at temperatures above 400 °C (Radwan et al., 2003). In XRD patterns of the catalysts prepared with the sol-gel method, the intensity of MgMoO 4 or CoMoO 4 peaks in relation to MgO diffraction peaks were higher when compared with catalysts prepared by impregnation. Also, an increase in molybdenum loading increased the intensity of those peaks. The value of the ratio between the most intense peak of MgMoO 4 or CoMoO 4 at 26.3 º and the most intense peak of MgO at 42.9 º can be observed in Table 1 The characteristic patterns of cobalt species, such as Co 3 O 4 which has the most intense peaks in 2θ = 31.3°; 36.8° and 65.2° (JCPDS no. 43-1003), were not observed. The interaction between magnesium and cobalt oxides may lead to formation of a CoO-MgO solid solution. In this case, Co +2 ions enter into the MgO lattice, as discussed by Ulla et al. (2001), and it causes similarities between the patterns of the solid solution and MgO. It is important to highlight that complete dissolution of Co 2+ into the MgO lattice only occurs at temperatures above 800 °C (Radwan et al. 2003). Also, the three most intense peaks of CoO (JCPDS no. 65-0902) at 2θ = 36.5°; 42.4° and 61.5° are very close to the three most intense peaks of MgO. As a result, diffraction patterns of MgO and CoO may overlap (Wang et al., 2012).
H 2 TPR profiles of the support and calcined catalysts are presented in Figure 2. The reduction profile of the pure MgO did not exhibit hydrogen consumption as reported before by Parmaliana et al. (1990). All catalysts exhibited three reduction peaks: the first small peak, at 380 °C, may be assigned to the reduction of cobalt particles with weak metalsupport interaction, possibly in the form of Co 3 O 4 . It has been reported that reduction of Co 3 O 4 happens in a stepwise process around 250-380 °C for bulk Co 3 O 4 and around 470-520 °C for Co/MgO (Kaluža et al., 2007). As the hydrogen consumption was very small in the temperature range of typical Co 3 O 4 reduction, Table 1. Sample chemical composition obtained by EDX. Surface area (S BET ), pore volume (Vp) and average pore diameter analyzed by N 2 physisorption. Ratio between the most intense peak at 26.3 º (related to MgMoO 4 or CoMoO 4 ) and the most intense peak of MgO at 42.9 º. * Ratio between the most intense peak of MgMoO 4 or CoMoO 4 at 26.3 º and the most intense peak of MgO at 42.9 º. the major reduction of the cobalt species has shifted to higher temperatures.
The second peak around 660-720 °C may be attributed to reduction of CoMoO 4 and stepwise reduction of molybdenum species, such as MoO 3 (Wang et al., 2012). There is a 60 °C shift towards higher temperatures of the second peak for catalysts with 20 wt.% of Mo as compared to the catalysts with 10 wt. % of Mo.
The third peak, obtained above 800° C, and during the isothermal period, may be attributed to the reduction of three different compounds: CoO-MgO solid solution (Wang and Ruckenstein 2002), reduction of MgMoO 4 (Awadallah et al., 2014) and reduction of MoO 2 species that correspond to a stepwise reduction of Mo 4+ to Mo 0 (Medema et al., 1978).
XRD and TPR results showed that the calcined catalysts have many cobalt, molybdenum and magnesium oxide species and some of them are reduced at the same temperature range, which made it difficult to identify and quantify these oxides. The TPR analysis could not help in identification of the diffraction lines at 2θ = 19.1°, 23.0° and 26.3°, which can be assigned to MgMoO 4 or CoMoO 4 patterns. Both compounds may be present in catalysts as pointed by TPR profile.
As the catalysts have several cobalt, molybdenum and magnesium species, the degree of reduction was calculated by the integration all along the TPR curve and considering Co 3 O 4 and MoO 3 as precursors of Co 0 and Mo 0 respectively. All the studied catalysts presented similar degrees of reduction, approximately 70%. Kaluža et al. (2007) studied the effect of the support in the reduction of Co/Mo catalysts and they found that the 12.4%Co20.6%Mo/MgO catalyst has a degree of reduction close to value found in our experiment, about 75 %.
XPS was performed to evaluate the surface chemical composition of the catalysts. Co 2p photoelectron spectra of the calcined catalysts are shown in Figure  S1. Co 2p 3/2 peak centers were identified with binding energies between 780.4-781.0 eV and there was a spin-orbit-splitting varying between 15.6-15.8 eV. Besides, it was possible to observe characteristic shakeup satellites in the recorded spectra. These features are characteristic of CoO and CoMoO 4 compounds (Mcintyre et al., 1990) . The surface and the bulk phase have, approximately, the same cobalt compounds.
The Mo 3d photoelectron spectra of the calcined catalysts are also shown in Figure S1. The 3d 5/2 line of Mo is centered in binding energies between 231.9-232.7 eV. These values are within the range of the binding energies of Mo +6 species, such as MoO 3 and CoMoO 4 (Mcintyre et al., 1990). Figure 3 shows the Raman spectra of the support and calcined catalysts. The Raman spectrum of the MgO support did not exhibit any Raman active band, as also reported in the literature (Ulla et al., 2001). All Raman spectra of the catalysts showed the presence of characteristic vibrational modes of Co 3 O 4 . The spectrum of the catalysts prepared by the impregnation method, Figure 3  There were not any bands at 455 and 675 cm -1 , which would be assigned to the vibrational modes of CoO (Tang et al., 2008). However, the presence of this species may not be ruled out, since it might be oxidized  to Co 3 O 4 due to the laser intensity used during the Raman analysis. A broad band placed at 575 cm -1 was attributed to CoO-MgO solid solution, where Co 2+ ions replace Mg 2+ atoms in the MgO lattice. Ulla et al. (2001) reported the characterization of Co/MgO catalysts and the CoO-MgO solid solution. They observed a broad band at 580 cm -1 , which was attributed to the partial formation of the solid solution in the Co/MgO catalyst. They also found the possibility of the simultaneous observation of this band and the band of Co 3 O 4 .
Bands attributed to MgMoO 4 were observed at 335, 910 and 960 cm -1 (Chang et al., 1992). Increases in the intensity of these bands were also observed for the catalysts with higher molybdenum loading. The catalyst 5Co20Mo Imp showed the presence of MoO 3 through the attribution of the band located at 820 cm -1 (Chang et al., 1992). The formation of CoMoO 4 was also observed through the identification of the bands at 371, 698, 878 and 943 cm -1 and it was more intense in the case of the 5Co20Mo SG catalyst (Moura et al., 2012).
The catalyst characterization techniques confirmed the presence of different cobalt and molybdenum compounds such as CoMoO 4 , MgMoO 4 , Co 3 O 4 , CoO-MgO and MoO 3 . The catalyst method preparation and the molybdenum loading only affected the relative amount of each compound in the catalysts. The catalysts with 20 wt.% of molybdenum have a higher fraction of cobalt interacting with molybdenum, such as in CoMoO 4 species, especially in the 5Co20Mo SG catalyst. However, their quantification could not be estimated due to signal overlaps of these compounds in many analyses.
The catalyst properties are very important in carbon nanotube synthesis. Under the reaction conditions, only small metallic cobalt particles are able to produce carbon nanotubes. So, cobalt oxides, precursors of metallic cobalt, should be well dispersed. XRD analyses could not detect CoO or Co 3 O 4 crystallites. This indicates that these oxide species are well dispersed on the catalyst support. TPR, Raman spectroscopy and XPS analysis showed that a fraction of the cobalt is interacting with molybdenum. Cobalt-molybdenum interaction is an important catalyst property because molybdenum is responsible for stabilization of cobalt particle size under the reaction conditions, avoiding cobalt agglomeration. Moreover, molybdenum can also decompose methane, modifying the carbon feed for CNT growth.
For CNT production, catalysts were previously reduced in a hydrogen flow for catalyst activation. The reduction temperature of the catalysts for CNT production was chosen from TPR profiles, i.e., 500 °C. At this temperature, Co 3 O 4 and CoMoO 4 species initiate their reduction. Higher temperatures during catalyst reduction may lead to sintering of the cobalt particles, which inhibit the nanotube growth, but lower temperatures may not allow enough activation of catalysts.

Characterization of the as-produced and purified CNT
Carbon nanotubes were characterized after the reaction step in the presence of catalysts, named "as-produced CNT", and after catalyst removal by purification with mild hydrochloric acid treatment, named "purified CNT".
The SEM images of as-produced CNT showed a surface almost completely covered by carbon nanotubes. Only a few areas of the images show the catalyst surface. After the purification step, no trace of catalysts could be observed. Some of the SEM images of the as-produced and purified CNT are in Figure S2. TEM images of as-produced CNT show a diameter distribution from 3 nm to 20 nm for all samples, the histograms are in Figure 4 (a). The diameter distributions of CNT.5Co10Mo Imp and CNT.5Co10Mo SG samples were similar with an average diameter of 8.1 and 9.0 nm, respectively, and they show the same standard deviation, 3.5 nm. Also, the carbon nanotubes produced by 5Co10Mo Imp and 5Co10Mo SG catalysts are well-structured. Figure  4 (b) shows a well-structured CNT from sample CNT.5Co10Mo SG .
The values of the standard deviation of the carbon nanotube diameter produced by catalysts with 20 wt.% of Mo were higher than those produced by catalysts with 10 wt% of Mo, 4.1 nm for CNT.5Co20Mo Imp and 4.7 nm for the CNT.5Co20Mo SG sample. Moreover, some defective filaments were found in those samples, such as bamboo-like MWNT, described as nanotubes with spaced hollow compartments divided by graphite layers perpendicularly oriented (Lin et al., 2007). Figure 4 (c) is an example of these structures. The formation of a bamboo-like structure may happen when the rate of carbon decomposition is higher than the CNT growth rate as found by Son et al., (2008). The CNT growth rate depends on parameters such as reaction temperature, carbon source and catalyst properties. These parameters are constant in our experiments except for catalyst properties. However, cobalt is the active phase in CNT production and all catalysts have the same cobalt loading (5 wt.%). Therefore, all catalysts should have, approximately, the same CNT growth rate. But, the rate of carbon decomposition is also dependent on the molybdenum loading. Molybdenum is active for methane decomposition (Ni et al., 2009). This may explain the formation of bamboo-like structures in the catalyst with higher molybdenum loading.
The TEM images exhibited encapsulated metal particles inside the CNT, indicating an incomplete XRD, TPO and Raman spectroscopy. The images also showed a few CNT with structural defects in the external walls. This indicates that the purification step may damage the CNT surface. The degree of defects was analyzed by Raman spectroscopy and will be discussed further.
The catalyst selectivity toward MWNT was determined by TPO analysis. Usually, SWNT and amorphous carbon are produced with MWNT. However, these carbon species oxidize at different temperatures. Amorphous carbon is the first species to be oxidized followed by SWNT and MWNT (Wang et al., 2012). TPO profiles of as-produced CNT showed a single oxidation peak in the range of 550-590 °C, as shown in Figure S4. This peak is within the oxidation temperature range of MWNT. The absence of peaks between 300-500 °C indicates a low production of amorphous carbon and SWNT, as indicated by TEM images.
In the methane decomposition reaction, a fraction of MoO x is also transformed into Mo 2 C, as discussed later and in TPO analyses, it can be oxidized to MoO x and CO x . However, the amount of carbon in the form of Mo 2 C after carbon nanotube production is very small, even for catalyst with 20 Wt.% of molybdenum. Then, these other minor forms of carbon such as graphene, amorphous carbon and Mo 2 C were not taken into account in the carbon nanotube purity expression, Equation (2).
After the purification step, the TPO profiles shifted to high oxidation temperatures. This behavior could be attributed to the partial removal of the catalyst during purification. The catalyst can speed up the process of oxidation of the carbon deposits, leading to lower oxidation temperatures (Abdullahi et al., 2013) CNT structure was studied by Raman spectroscopy, for which results are depicted in Figure 5. More than ten different spots were analyzed for each sample. The graphitization of the carbon species was evaluated through the average values of the ratio of D and G band intensities (I D /I G ). The G-band, in the range of 1500-1600 cm -1 , is related to the tangential stretching of the C-C bond in the graphene layer. The structural defects of the produced CNT and the presence of disordered carbon materials were observed through the identification of the D-band in the range of 1250-1350 cm -1 . (Wang et al., 2012;Abdullahi et al., 2013).
The CNT produced from catalysts with 10 wt.% of molybdenum exhibited higher degrees of graphitization than the catalysts with 20 wt.% as observed in TEM images. Flahaut et al. (2004) also observed an increase in I D /I G ratio with the increase of molybdenum loading. Figure 5 (a-d) shows the Raman spectra of the as-produced CNT and their average I D /I G ratio. The I D /I G ratio average values of the as-produced removal of the catalyst in the purification step, as shown in Figure S3. These particles are usually active particles for CNT growth. However, there are no large gray areas as found in as-produced samples. This points to an efficient removal of the catalyst support. The purified CNT were also analyzed by TEM, EDX, CNT are comparable with I D /I G ratio values observed in high yield of MWNT obtained by other authors in similar experimental conditions (Son et al., 2008;Awadallah et al., 2014).
In some Raman spectra of each sample, there was a low intensity band in the range of 100-400 cm -1 . Most of these bands are not in the range of catalyst species such as Co 3 O 4 , CoMoO 4 , MoO 3 , MgMoO 4 . Thus, these bands were related to the radial breathing mode (RBM) vibration of SWNT (Abdullahi et al., 2013) .
Raman results showed that the purification process increased the degree of disordering of the CNT samples by the increase of I D /I G ratio values (Ling et al. 2013). The I D /I G ratios of purified samples Figure 5 (e-h) were slightly higher than the I D /I G ratios of the asproduced CNT and showed average values lower than 0.8 ( Figure 5 (a-d). The most affected as-produced CNT were those with low I D /I G. , i.e. catalysts with 10 wt.% of molybdenum. Huang et al. (2003) studied the purification of MWNT with I D /I G ratio values higher than 0.8 and they found that I D /I G values of the purified MWNT were not significantly affected by the acid treatment.
The characteristic vibrational modes of Co 3 O 4 , MgMoO 4 , CoMoO 4 and MoO 3 observed in calcined catalyst, see Figure 3, were not detected after CVD reaction, see Figure S5, probably due to the high carbon content in the samples.
Weight loss curves of as-produced and purified CNT are shown in Figure 6 (a) and (b), respectively. The large weight loss between 450 and 700 °C is associated with oxidation of MWNT as demonstrated by TPO results. Carbon yield was calculated from Equation (1) using the weight loss curves of the as-produced CNT and the sample purity was calculated from Equation (2) by means of the weight loss profile of the purified samples. Equations (1) and (2) consider that the weight loss from TGA cannot be attributed only to the  . Weight loss curves: (a) as-produced and purified CNT produced by 5Co10Mo Imp and 5Co10Mo SG catalysts; (b) as-produced and purified CNT produced by 5Co20Mo Imp and 5Co20Mo SG catalysts.
oxidation of carbon species. Cobalt and molybdenum also get oxidized during the experiment. This means that the fraction of carbon is higher than the resulting weight loss due to the weight gain compensation related to the simultaneous oxidation of metals.
The 5Co10Mo Imp and 5Co10Mo SG catalysts showed the highest carbon yield, 88 and 85 %, respectively, while the catalysts with 20 wt.% of molybdenum showed lower carbon yield, 68 % for 5Co20Mo Imp and 81 % for 5Co20Mo SG . Also, the purified samples produced by catalysts with 20 wt.% of molybdenum showed a significant weight loss between 600-1000 °C. This weight loss may be attributed to sublimation of MoO 3 formed from the oxidation of Mo 2 C also detected in DRX results (Figure 7). Mo 2 C oxidation has already been studied and it is known that it starts around 450 °C and may be completed with the formation of MoO 3 at 550 °C. Then, increasing the temperature, MoO 3 sublimation occurs around 700 °C (Chen et al., 2011).
After the purification process, there was a complete removal of the catalyst support. No traces of Mg X-ray lines were detected in EDX analyses. However, only a fraction of the cobalt and molybdenum were removed. The composition of the purified samples is reported in Table 2. The degree of catalyst elements extraction from as-produced CNT was calculated from EDX and TG results. The fraction of cobalt and molybdenum removed from CNT.5Co10Mo Imp was similar (60 % Co and 62 % Mo) and the same trend was observed for CNT.5Co10Mo SG (46 % Co and 49 % Mo). However, for catalysts with high molybdenum loading, the proportion of cobalt removed was higher than molybdenum.
After the purification step, MgO diffraction peaks were eliminated in all samples (Figure 7 (b), (d), (f) and (h)) which is in agreement with the EDX results. The position of peaks characteristic of MgMoO 4 or CoMoO 4 did not disappear after the purification step. Moreover, no traces of Mg X-ray lines were detected in EDX analyses of the purified samples. Thus, these diffraction lines were ascribed to CoMoO 4 . The ratio between the area of the most intense peak of MWNT at 26.5 º and the area of the peak of CoMoO 4 at 34.4º increased after the purification step due to the complete removal of the MgMoO 4 or due to the partial removal of CoMoO 4 or both. The Mo 2 C species were formed during the CVD method as found by Xu et al. (2011). However, it is not clear if Mo 2 C come from MoO 3 , MgMoO 4 or CoMoO 4 .
As observed in Table 2, the purification process was more efficient for as-produced CNT synthetized with impregnated catalysts than those prepared by the sol-gel method. The catalysts prepared by the sol-gel method have a higher fraction of cobalt in the form of CoMoO 4 , which has low solubility in acidic solutions. The mass of residual catalyst particles in purified samples was determined by TG and their composition by EDX.
5Co10Mo Imp catalyst showed the best performance for CNT production. It produced more uniform and well-structured nanotubes with good production performance. Also, the purification process was able to completely remove the catalyst support and most of the cobalt and molybdenum from the CNT_5Co10Mo_ Imp sample. Therefore, 5Co10Mo Imp catalyst was chosen for new catalytic tests. The reaction time was increased (60 min, 120 min. and 180 min) to evaluate the evolution of the carbon yield and the catalyst deactivation. The as-produced CNT were named CNT_60ʹ, CNT_120ʹ and CNT_180ʹ.
The carbon yield in function of the reaction time is linear, indicating that no significant catalyst deactivation was observed, as can be seen in Figure 8. Ni et al. (2006) studied the kinetics of carbon nanotube growth over Mo/Co/MgO catalysts and, for most of the catalytic tests, the authors observed a decrease in reaction rate only after 180 min.
Usually, long reactions time are essential to achieve high yield of MWNT. Also, the empty volume of the reactor must be large to allow an efficient growth of the MWNT. For example, at the beginning of the reaction of the CNT_180ʹ sample, the catalyst occupies only a small fraction of the reactor volume, smaller than 5 %. Then, as the carbon nanotubes grow, they fill the reactor. After 180 min of reaction, the carbon yield was 360 % and the reactor was almost full. Thus, no further reaction time was tested.
The purification of the CNT_30ʹ, CNT_60ʹ, CNT_120ʹ and CNT_180ʹ samples showed a complete removal of MgO and a partial removal of Co and Mo from CNT. The sample purity was estimated by Eq.
(2). The residual weight of CNT_30ʹ is 10.0%, which implies that approximately 92.9% of the mass of the sample is carbon, considering the weight fraction of cobalt and molybdenum in the sample and its conversion to oxide in the process of TGA analysis. The degree of purification of these samples increased with increasing carbon yield, from 92.9% for the CNT_30ʹ sample to 95.8% for the CNT_180ʹ sample, as can be seen in Figure 8.

CONCLUSIONS
The catalysts prepared by sol-gel method improved the interaction between cobalt and molybdenum, increasing the formation of CoMoO 4 and decreasing the formation of Co 3 O 4 and CoO. This interaction was also higher for the catalysts with 20 wt.% of molybdenum than those with 10 wt.%. The excess of molybdenum decreased the carbon yield and produced CNT with lower degree of graphitization.
The purification step by mild acid treatment was effective to remove all MgO and a major fraction of the cobalt and molybdenum present in CNT samples without significant damage to the carbon nanotube structure. The purification process was more efficient for as-produced CNT synthetized with impregnated catalysts than those synthetized with catalysts prepared by the sol-gel method. The catalysts prepared by the sol-gel method presented a higher fraction of cobalt in the form of CoMoO 4 , which has low solubility in acidic solutions.
High yields of MWNT produced by methane chemical vapor deposition on CoMo/MgO catalysts cannot be free of impurities after purification by mild acid treatment. The cobalt and molybdenum particles inside the carbon nanotubes could not be completely removed. However, an increase in sample purity could be achieved by increasing reaction time. There was no evidence of catalyst deactivation in 180 min of reaction.

ACKNOWLEDGEMENTS
This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) -Finance Code 001 and Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro (FAPERJ). Table 2. Degree of catalyst element extraction from as-produced CNT and composition of purified CNT.