Carbon-Involved Near-Surface Evolution of Cobalt Catalyst during Reaction: An in-situ Study

When carbon-containing species are involved in reactions catalyzed by transition metals at high temperature, the diffusion of carbon on/in catalysts dramatically inuence the catalytic performance. Acquiring information on the carbon-diffusion-involved evolution of catalysts at atomic level is crucial for understanding the reaction mechanism yet also challenging. For the chemical vapor deposition process of single-walled carbon nanotubes (SWCNTs), we developed methodologies to record in-situ the near-surface structural and chemical evolution of Co catalysts with carbon permeation using an aberration-corrected environmental transmission electron microscope and the synchrotron X-ray absorption spectroscopy. The nucleation and growth of SWCNTs were linked with the partial carbonization of catalysts and the alternating dissolvement-precipitation of carbon in catalysts. The dynamics of carbon atoms in catalysts brings deeper insight into the growth mechanism of SWCNTs and also sheds light on inferring mechanisms of more reactions. The methodologies developed here will nd broad applications in studying catalytic and other processes.


Background
Transition metals (Fe, Co, Ni, Cu, etc.) are in widespread use as heterogeneous catalysts for many reactions of both scienti c and industrial interest. For example, such catalysts play a leading role in petrochemical engineering processes, including Fisher-Tropsch synthesis [1][2][3] , selective hydrogenation of alkyne 4,5 , reforming and pyrolysis of methane 6 . Carbon atoms in transitional-metal catalysts may act as either promotors or poisoners. At low carbon content, C atoms can induce the formation of metal carbide intermediates, which then serve as catalytically active sites 5,[7][8][9][10][11] . However, at high carbon content, graphite or coke can emerge on top of the metal surface, and even surround and encapsulate metal nanoparticles, fully blocking the active sites 1 .
The transition metals are also generally applied as high e cient catalysts for the synthesis of carbon nanomaterials, including SWCNTs 12-20 and graphene 21,22 . The structure and composition of catalysts are crucially important and determinative for achieving high selectivity in diameter and chirality for SWCNT growth 13,14,23,24 . During the CVD process, carbon atoms may diffuse on surface or into the body of transitional-metal nanoparticles, resulting in a diverse and heterogeneous composition and structure of the catalysts [23][24][25][26][27][28][29] . Hence, there are a long-time controversy on the active catalytic species. The carbon enrichment in metal nanoparticles also modify their electronic property and wetting property 27 , which in turn leads to different growth modes of SWCNTs on catalysts 30 . In the controlled growth of graphene, the selection of metal substrate with suitable a nity to carbon is also a key issue. A precise tuning of carbon solubility in catalysts by combining different transition metals precisely controlled the production of mono-or bi-layer graphene 22,[31][32][33] . These reports highlight the importance of the interaction between metal and carbon, which needs further investigation. The key challenge is to gure out the behavior of carbon atoms in metal catalysts, especially to obtain atomic scale and dynamic information under reactive environment.
Techniques based on aberration-corrected environmental transmission electron microscope (AC-ETEM) [34][35][36][37][38][39][40] have shown to be a powerful tool for the investigation of such process at atomic spatial and high temporal resolution. With the help of in-situ imaging, the nucleation process of carbon nanotubes on both liquid and solid catalysts were recorded 25,29,35,[41][42][43] . Nonetheless, much less is known about structural and compositional information at atomic scale on carbon involved catalyst surface and how nanotube nucleation is aroused from carbon-metal interaction.
Herein, taking the catalytic synthesis of SWCNTs as a model process, we investigate the diffusion behavior of carbon atoms in near-surface of cobalt catalysts under carbon feeding conditions by using AC-ETEM. The insu cient carbon supply is a general problem in ETEM study due to the low pressure in ETEM chamber and the tiny heating area of the in-situ chip-heating system. Here we used cobalt acetate (Co(OAc) 2 ) as both the catalyst precursor and the solid carbon source. The in-situ decomposition of acetate groups upon heating brings about a localized carbon feeding high enough to facilitate the growth of SWCNTs. Through atomic-scale visualization and elemental electron energy loss spectrum (EELS) analysis, we found that the cobalt catalyst was activated through carbon dissolving in near-surface and composed of Co-Co 3 C. The dynamic structure evolution of catalyst revealed that carbon dissolving and precipitation in cobalt facilitated the growth of SWCNTs. These ndings were also proved by in-situ ambient-pressure synchrotron X-ray absorption spectroscopy (XAS).

Results
In-situ observation of the growth of SWCNTs from cobalt catalysts. The in-situ experiments of SWCNT growth were conducted on an aberration-corrected FEI Titan G2 80-300 ETEM microscope (operated at 80 kV) with a resolution of 1.0 Å. The ETEM chip with amorphous SiO 2 /SiN x membrane was mounted onto a Micro Electron-Mechanical System based micro heater (ThermoFisher Scienti c). The cobalt acetate dispersed on SiO 2 /SiN x membrane of ETEM chip was annealed in vacuum (~10 -5 Pa) at 600-750 ○ C for 10 min. The cobalt acetate was decomposed into CoO and further transformed into Co as revealed by the in-situ selected-area electron diffraction patterns shown in Supplementary Fig. 1. In this process, the carbon species or other carbon-hydrogen species generated from the decomposition of acetate groups took the roles of reductants. Meanwhile, the carbon species decomposed from acetate groups could also act as carbon source 44 to feed the growth of SWCNTs.  Table 1). Indeed, they could not be assigned to either fcc/hexagonal-close-packing (hcp) Co or Co 2 C. Then the graphite shell lifted off at 59 s to form a cap with a size of ~1 nm ( Fig. 1c, d, g, h). The active catalyst nanoparticle kept structurally inhomogeneous with co-existence of Co and Co 3 C phases. To avoid the effect of e-beam irradiation, we also performed the nanotube growth with e-beam off and then captured the ETEM images with e-beam on. Figure 1i, j show typical single-shot ETEM images of individual SWCNTs nucleating from cobalt catalysts at 600 ○ C. The FFT derived from the interface of catalyst-nanotube also showed the diffraction spots of metallic Co and Co 3 C (Fig. 1k).
More active catalysts containing Co and Co 3 C were shown in Supplementary Fig. 2 and 3. Our observations in ETEM demonstrated generally that inhomogeneous Co-Co 3 C was active species for nucleating SWCNTs.
To gain deep insights into the growth dynamics of SWCNTs, we linked the catalyst evolution to the nucleation rate of SWCNTs (Fig. 1l, m). The nucleation rate was plotted as a function of the SWCNT length to growth time (Fig. 1m). The catalyst structure of each snap point was determined from FFT of time-sequenced ETEM images ( Supplementary Fig. 2). It is obvious that the rate-limiting step was the incubation of graphitic shell on metallic Co nanoparticle, which lasted near 120 s. The catalyst maintained the metallic state without Co 3 C species during the incubation process. Then in the follow-up stages, the formation of structurally inhomogeneous Co-Co 3 C at ~120 s resulted in rapid elongation of carbon cap (Fig. 1m). After that, the active catalyst nanoparticle kept inhomogeneous containing Co and Co 3 C. These observations indicate that the presence of Co 3 C might be essential in promoting the nucleation of cap and its elongation.
To further evaluate the general role of catalysts, we performed the X-ray photoelectron spectroscopy (XPS) to probe the composition of catalysts after growth of SWCNTs ( Supplementary Fig. 4). The spectra were recorded at around Co 2p core levels, where the background component was subtracted by the Shirley method. The two peaks with low intensity appearing at 778.3 and 793.7 eV could be ascribed to Co 2p 3/2 and Co 2p 1/2 of metallic Co. Co 2p 3/2 peak at 781.3 eV with satellite peak at 786.2 eV were attributed to Co 3 C (Fig. 2a). Spin-orbital splitting between Co 2p 3/2 and 2p 1/2 is 15.8 eV, which further indicates the existence of Co (II) species 45 . It is obvious that Co 3 C and Co co-exist in catalysts. XAS was also used to investigate the local coordination structure of cobalt catalysts. In-situ X-ray absorption nearedge spectroscopy of MgO supported Co catalysts recorded at Co K-edge was collected under the atmospheric pressure H 2 and CH 4 at 700 °C, respectively. The extended X-ray absorption ne structure (EXAFS) oscillation of Co under H 2 showed a rst nearest-coordination peak of 2.1 Å observed from R space, indicating the reduction of cobalt oxide. The rst three nearest-coordination peaks of 1.5, 2.1, and 2.6 Å appeared when switching to CH 4 atmosphere, which could be attributed to Co-C, Co-Co, and Co-C based on tting result, respectively (Fig. 2b). All above in/ex-situ X-ray spectroscopic results were consistent with ETEM observations that the Co catalysts were partially carbonized to form Co-Co 3 C during the growth of SWCNTs. Local carbonization of Co by carbon spatial permeation. To gure out the distribution of Co 3 C species in active catalyst nanoparticles during the growth of SWCNTs, we carried out further investigation on the interface of catalyst-nanotube. Figure 3a shows a single shot AC-ETEM image of a nanotube growing from a cobalt nanocrystal at 600 ○ C. FFT of this seed nanoparticle was identi ed as inhomogeneous Co-Co 3 C (Fig. 3b), which coincides with previous observations. By applying masks to the series of spots of Co marked with circles, the resulting inversed FFT (IFFT) showed the region of fcc-Co, which covers the entire catalyst (Fig. 3c). However, IFFT images derived from the series of spots of Co 3 C marked with boxes showed the uneven distribution of Co 3 C in catalyst (Fig. 3d). Additionally, we found from IFFT images that Co 3 C region was close to the nucleation site of nanotube (Fig. 3d, marked by circles). More in-situ data were shown in Supplementary Fig. 3, which all delivered similar results that Co 3 C species tended to aggregate around the nucleation site of nanotubes.
To quantitatively evaluate the carbon dissolved in catalysts, we made a statistical analysis of the lattice spacings of the outer layers of catalysts. Figure 4a shows a high-resolution AC-ETEM image of a graphitic cap on Co catalyst at 600 ○ C. The corresponding FFT marks a resolution information of 1.1 Å. We assigned atomic columns to hcp-Co [0001] and Co 3 C phases accordingly (Fig. 4b), which con rms again the co-existence of Co and Co 3 C. A t of two dimensional (2D) Gaussians to the intensity of distribution of the atomic columns was used to acquire the precise column coordinates, which improved the resolution to sub-angstrom 38,46 . The perpendicular layer spacings were determined locally from those tted coordinates and then averaged for each shell (d 12 , d 23 , d 34 ...) ( Supplementary Fig. 5). Figure 4c summarizes the statistical analysis of the lattice distances of Co catalysts. The (11) fcc-Co lattice spacings of the active catalyst nanoparticle with a cap gradually increased from 2.15(±0.06) Å to 2.58(±0.17) Å from core region to the outermost surface. Such a ∼37% lattice expansion of the surface layer could be induced by the carbon atoms dissolving into Co nanoparticle. We also analyzed the near surface lattice spacings of other active catalysts of either fcc or hcp-Co phase. A general lattice expansion of 14~37% from the outmost surface toward the core was observed in all the active catalysts (Fig. 4d). For comparison, we also measured the near-surface layer spacings of inactive catalyst without CNT nucleation. It was found that only the outermost surface (d 12 =2.17 ± 0.05 Å) displayed a larger spacing and the inner lattice spacings (d 23 , d 34 , d 45 …) were close to theoretical value (d (11)Co =2.06 Å) (Fig. 4c,   Supplementary Fig. 6). This nding indicates that carbon atoms did not diffuse into inner part of inactive Co nanoparticles.
We also compared the lattice expansion from the different directions of active catalysts with SWCNTs ( Fig. 4e-h). The near-surface crystal lattice was signi cantly expanded near the nucleation site of SWCNT (Fig. 4e, f). This nding could originate from the diffusion of carbon atoms near the region where the SWCNT grows, which was consistent with above IFFT results (Fig. 3d). However, the spacings away from the nucleation site were not expanded, which is similar to the inactive catalysts (Fig. 4g, h).
To direct visualize the carbon dissovled in catalysts, in-situ scanning TEM (STEM)-EELS was collected on an acitve catalyst with a CNT grown at 600 °C (Fig. 5a, b). The nanoparticle slightly rotated after EELS mapping due to the e-beam irradiation. STEM-EELS elemental mapping showed that carbon element distributed in both CNT and the near-surface of catalyst nanoparticle, whereas cobalt element homogeneously distributed in the whole catalyst (Fig. 5c-e). The line scan pro le of EELS across the CNT-catalyst interface clearly showed co-existence of C and Co in the near-surface of catalyst ( Fig. 5f, g, layer 3-6). The intensity of energy loss peaks of C K-edge decreased from catalyst surface to body, whereas the Co L 2,3 -edge showed the inverse tendency (Fig. 5g, layer 3−12 and Fig. 5h). For comparison, we also collected the in-situ EELS on inactive catalyst without CNT nucleation, showing the absence of carbon signal in Co catalyst, i.e., no carbon diffusion into catalyst (Supplementary Fig. 7).
All the above in-situ atomic scale and elemental characterizations as well as quantitative analysis supported that carbon atoms heavily dissolved in near-surface (thickness: 1~3 nm) of active Co nanoparticles and consequently forming Co 3 C during the growth of SWCNTs.
Dynamics of carbon dissolving and precipitation in catalyst during the nucleation of carbon cap. Figure   6a-c showed the time-sequenced ETEM images of carbon cap growing on a cobalt nanoparticle at 700 ○ C. At the relative time de ned as 0 s, the carbon cap started to nucleate on surface of Co nanoparticle with a height of 2.6 Å. Then the cap gradually grew larger and lifted off from the catalyst surface at 975 s. All the corresponding FFT derived from the near-surface of the catalyst evidenced the co-existence of Co and Co 3 C (Fig. 6d-e). The IFFT images derived from the diffraction spots of Co 3 C showed that the Co 3 C species rotated during growth of carbon cap. The orientation of Co 3 C changed from [21] to [12] by 41.5 ○ , and further changed to [10] by 33.8 ○ (Fig. 6g-i). However, the orientation of metallic Co species maintained stable (Fig. 6j-l). Figure 6m summarized the (11) Co spacings at different exposure time. At the time of 0 s, we observed a ∼37% lattice expansion from outermost surface to core region of Co nanoparticle. This data has been shown in Fig. 4c. However, this tendency reversed at 975 s that the lattice spacings decreased from core region to outermost surface (Fig. 6m). Such a structural uctuation induced by the lattice expansion and shrink in near-surface could be related to carbon atoms' dissolving and precipitation in Co nanoparticle. The precipitation of carbon atoms was responsible for the nucleation and growth of the carbon cap.

Discussion
To further understand the role of Co catalysts in nucleating the SWCNTs, we also analyzed the poisoned catalysts, which are inactive for the growth of nanotubes. We observed that these poisoned catalyst nanoparticles were either entirely encapsulated by the graphitic shell or coated by amorphous carbon. For those nanoparticles entirely encapsulated by the graphitic shells, they still maintained the composition of metallic Co and did not form cobalt carbides (Co 2 C or Co 3 C) ( Supplementary Fig. 8a-f). This could be attributed to that the metallic Co nanoparticles were wrapped by the graphitic shells, thus carbon atoms could not further diffuse into Co to form Co 3 C. The pure Co 3 C carbide nanoparticles were coated by amorphous carbon instead of graphitic shells ( Supplementary Fig. 8g-j). This indicated that pure Co 3 C was favorable for the precipitation of amorphous carbon but not effective in catalyzing the formation of graphitic sp 2 carbon. These observations about inactive catalysts further veri ed that the growth of SWCNTs was realized only by the cooperation of Co and Co 3 C.
To summarize, aberration-corrected ETEM has been used in combination with quantitative 2D Gaussian tting, in-situ EELS, and XAS analysis, respectively, to systematically investigate at atomic scale the behavior of Co nanoparticles under carbon feeding and the nucleation and growth of SWCNTs. The inhomogeneous Co-Co 3 C was evidenced as catalytically active species for the growth of SWCNTs. The Co 3 C species were structurally uctuated in near-surface of Co catalysts induced by carbon diffusion and precipitation, acting as carbon carriers in the growth of SWCNTs. These ndings elucidate the dynamic behavior of carbon in transitional-metal nanoparticles at the atomic scale and reveal the general importance of carbon permeation in metal in hydrocarbon or other carbon-containing species involved heterogeneous reactions. Our in-situ methodology based on atomic-scale ETEM imaging and elemental spectroscopy is shown to be a powerful tool for the investigation of light elements in solids and their dynamics in real time and direct space. Such in-situ methodologies should be valid for the mechanistic investigation of a wide variety of metal catalyzed reactions and other processes such as crystallization and alloying.
Methods ETEM experiments. The in-situ study was conducted by using an aberration-corrected Titan G2 80-300 ETEM equipped with a MEMS based micro heater. The catalyst precursor of cobalt acetate (Co(OAC) 2 ) dissolved in ethanol (0.1 mmol•L -1 ) was dropped onto the ETEM chips. We used two methods to prepare Co catalyst and perform in-situ SWCNT growth. (i) The as-prepared specimen was rst heated to 600 ○ C for 30 s and stabilized in high vacuum (~10 -5 Pa) at 600 ○ C for 10 min with e-beam off. After that, TEM images and videos were captured with e-beam on. During annealing at 600 ○ C, the cobalt acetate decomposed to metallic Co nanoparticles, carbon, and other reductive species (CO, C 2 H 6 , H 2 ). The carbon containing species could act as carbon source to feed the growth of carbon nanotubes. (ii) The asprepared specimen was rst heated to 500 ○ C in pure O 2 (50 Pa) for 30 min to form Co 3 O 4 and reduced in H 2 (50 Pa) at 700 ○ C for 30 min to form metallic Co catalysts. Then CH 4 (99.99%), or CH 4 /H 2 (v/v=7/3), or CO (99.999%) was introduced to TEM chamber to grow SWCNTs at 600-750 ○ C in a pressure of 28-503 Pa.
STEM-EELS experiments. In-situ STEM EELS data were collected in another, but otherwise the same aberration-corrected Titan 80-300 ETEM. It was equipped a Gatan image lter (Quantum 936) with an energy dispersion of 0.25 eV operated at an acceleration voltage of 300 kV. The zero-loss EELS was acquired immediately after obtaining the core-loss EELS on individual nanoparticle. The obtained EELS data were further analyzed in Digital Micrograph software. The position of core-loss EELS was corrected with the corresponding zero-loss peak and the extrapolated background was then subtracted from the edge of interest. A Fourier-ratio deconvolution was further performed to remove the effect of plural scattering.
2D Gaussian tting of TEM image. The 2D Gaussian t to the intensity distribution of the atomic columns from HRTEM image was conducted using the StatSTEM V3.0 code run on MATLAB (R2015b) 46 . More details were showed in Supplementary Fig. 5 and corresponding discussion.
Ex-situ XPS. The XPS experiments were performed using an AXIS Supra/Ultra Imaging X-ray Photoelectron Spectrometer. The binding energy of the XPS spectra was calibrated to that of C 1s (284.6 eV). It was performed and exposed a monochromated Al Kα (1486.6 eV). We prepared Co catalysts and perform low-pressure (1.79 kPa) ethanol-CVD growth of SWCNTs at 900 ○ C. SEM image of as-grown SWCNTs can be seen in Supplementary Fig. 4.
In-situ synchrotron X-ray absorption spectroscopy. The XAFS spectra at Co K-edge were obtained at BL14W-1 beamline of Shanghai Synchrotron Radiation Facility operated at 3.5 GeV. For in-situ XANES measurements, a homemade CVD apparatus was equipped in BL14W-1. The quartz tube was strictly parallel to beam direction. The MgO supported Co catalyst (Co 5 wt%) thin ake was inserted in quartz tube, facing the beam direction at 90 ○ . The in-situ Co K-edge XANES spectra were collected in a transmission mode at 700 ○ C with the energy range of 7513-8509 eV, under the atmospheric pressure of H 2 /He (10% H 2 ) (200 cm 3 ·min -1 ) and CH 4 (20 cm 3 ·min -1 ), respectively. The data were interrupted using Athena and Artemis codes to remove background, normalize edge-step, and t the EXAFS data using standard procedures. The absorption edge of Co foil (E 0 =7709 eV) was used to calibrate the energy edge of all samples.
Data availability. The data that support the plots within this paper and other ndings of this study are available from the corresponding author upon reasonable request.    The sequence of layer spacings is labeled as d12, d23, d34 ..., from surface to core. b Corresponding FFT of near-surface of this nanoparticle. c The representative (1 1)fcc-Co lattice spacings of active and inactive catalysts as a function of sequencial atomic numbers from the outermost surface to core region (d12, d23, d34…d89). d Summary of lattice expansion measured from both fcc and hcp Co active catalysts. The degree of lattice expansion was calculated by (dmeasureddtheoretical)/dtheoretical×100%. e-h Near-surface lattice expansion measured from different directions of the same catalyst as shown in Fig. 3a. The arrows indicated the measured directions from outmost surface to core region (from d12 to d89). All error bars were calculated from the standard deviation of the mean lattice parameters among 6-12 individual measurements for a given atomic shell.

Figure 5
In-situ STEM-EELS characterization of catalyst during the growth of CNT. a, b High-angle annular dark led (HAADF)-STEM image of a semi-suspended cobalt catalyst with a CNT grown on it (a) and the enlarged image (b). c-e Corresponding EELS-STEM elemental mapping of C K-edge (c) and Co L2,3-edge (d) on catalyst-CNT, and their combinatial image (e).
Step size: 1 nm. f, g EELS spectrum image (f), line scan EELS spectra (g) intergrated from 0-12 layers in panel (f) across the CNT and catalyst. 0-12 marked in (f) indicate the scan direction. Each scan area is 1 nm×10 nm. Background of all spectra were subtracted using the 1st Order Log-Polynomial model with the same rule. h Intensity of C and Co EELS signals across the CNT, catalyst surface, and body derived from (e).

Figure 6
Structural evolution of Co catalyst during the nucleation of carbon cap. a-l Time-sequenced ETEM images of a Co nanoparticle during the growth of carbon cap under CH4/H2 (v/v = 7/3, 49 Pa) at 700 °C (a-c), FFT (d-f) and IFFT (g-i) of the particle region marked with framework in (a-c). The contrast of IFFT images (i, l) at 975 s were inverted due to the defocusing condition. m Co (1 1) lattice spacings from surface to core region at different exposuring times.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryInformation.pdf