Nanocarbon oxidation in the environmental transmission electron microscope - Disentangling the role of the electron beam

Environmental transmission electron microscopy (ETEM) can provide unique insights into nanocarbon oxidation processes through atomic resolution and real time imaging of materials at high temperatures in reactive atmo-spheres. However, the electron beam can also influence the reaction rates, and even alter the processes entirely, complicating the interpretation of the in situ observations. Many mechanisms have been proposed to account for the impact of the electron beam, predominantly involving ionization of the oxidative gases to form more reactive species. However, these mechanisms have not been critically evaluated and compared to predictions from theory. Here, we evaluate the impact of the electron beam both qualitatively (oxidation mode and spatial extent) and quantitatively (oxidation rates), using high resolution imaging and electron energy loss spectroscopy, at different electron energies and dose rates. We demonstrate that transient defects generated by elastic scattering, forming highly active sites for carbon abstraction by oxygen, is the main mechanism for the enhanced oxidation rates observed in situ. This is evident from an insensitivity to electron energy and saturation of the effects at high electron dose rates. To avoid undue influence of the electron beam in future ETEM studies, we therefore recommend conditions where the intrinsic oxidation dominates over the beam-enhanced oxidation (note that no conditions are completely “ safe ” ) and extensive comparisons with other methods.


Introduction
Oxidation of carbon nanostructures is integral to both the synthesis of industrially important materialse.g., carbon black (CB) and carbon nanotubes (CNTs) -and to the gasification of carbonaceous materials where soots can be formed as a byproduct [1,2].Environmental Transmission Electron Microscopy (ETEM) is increasingly used to study these oxidative processes through real-time, atomic-scale resolution imaging; this method offers unique possibilities of correlating reaction mechanisms and rates with the evolving structure and composition at the nanometer scale.ETEM has thus allowed for direct observation, and extension, of nanocarbon oxidation mechanisms previously based on kinetic data and ex situ characterization of partially oxidized materials [1,3].For uncatalyzed CB oxidation, examples include observations of surface and internal oxidation [4], as well as pore formation [4,5], of individual CB particles; both CB microstructure and oxidation conditions were found to determine which mechanism was dominant.For CB agglomerates, it has also become possible to directly distinguish oxidation of the primary particles from the initial neck-region oxidation, where material between the primary particles is selectively removed [6].Catalytic soot and CB oxidation has also been extensively studied using ETEM [7][8][9][10].For CNTs [11][12][13][14], ETEM studies indicate that CNTs, under mild conditions, react via surface oxidation unless gas diffuses through pores, in which case the oxidation starts from the inside layer [11].For few-layer graphene, different oxidation modes have been observed depending on O 2 pressures; at low pressures, and consequently low reaction rates, the edges form stable armchair reconstructions which are lost or disrupted as the pressure is increased [15].
However, in ETEM experiments the electron beam has a dramatic impact on the observed events, complicating their interpretation and rendering their connection to real-world oxidation processes tenuous.For carbon nanostructures the oxidation rate is greatly increased under electron irradiation, which has been connected to the accumulated electron dose [13] and the electron dose rate [15,16].Based on these findings, different mechanisms for how the electrons mediate this beam-enhanced oxidation rate have been hypothesized; the most commonly suggested mechanism is ionization of O 2 by the high-energy primary electrons (PE) in the beam, resulting in reactive species that more readily react with, and abstract carbon atoms [4,5,[11][12][13].Ionization of O 2 by low-energy secondary electrons (SE) emitted from the sample upon irradiation has been suggested to add to this effect [14].Atomic displacements from interactions between the sample and PE is known from ETEM studies of oxide catalysts [17], and could also occur during nanocarbon oxidation.Here, such displacement effects have previously been suggested to play only an indirect role by graphitizing the sample, thereby reducing reactivity [18].
Determining which beam damage mechanism(s) actually occurs is crucial for (1) understanding and translating ETEM findings to realworld nanocarbon oxidation, and (2) for choosing experimental conditions to minimize beam effects.In terms of (1), gas ionization is for instance proposed to lead to similar chemistry and oxidation mechanisms as is present in a flame [4].In contrast, other proposed mechanisms completely lack real-world analogues and present greater challenges for interpreting the results.In terms of (2), gas ionization can for instance be controlled by PE energy since ionization cross-sections have a strong energy dependence.While all proposed mechanisms have unique implications for how to interpret the results and select experimental parameters, this has, to our knowledge, not been comprehensively studied.
Herein, we describe the response of nanocarbon oxidation processes in the ETEM with respect to different primary electron energies and electron dose rates through both imaging and electron energy loss (EEL) spectroscopy.CB is used as a model material as it provides a range of possible oxidation mechanisms that are sensitive to the reaction conditions [6].To provide a framework for interpreting the results, we first present an overview of the theory behind the potential mechanisms underlying the electron irradiation effects.The predictions of these mechanisms are then compared to the experimental results using both qualitative and quantitative measures, involving both the reaction mechanism and rate.

Nanocarbon model material
CB was oxidized at 600 • C in two different ETEM setups to characterize the dose rate dependence of CB oxidation at different electron energies and O 2 pressures.Carbon Black N220, sourced from Orion Engineered Carbons GmbH, was selected for its potential to exhibit several different modes of oxidation, both internal-, surface-, and neckregion oxidation.The CB particle diameters were typically in the range of 15-40 nm, as can be seen from the images presented in later figures.Additionally, 600 • C was chosen as the reaction temperature due to the variety of oxidative modes possible at this temperature [6].

ETEM imaging
Low-pressure oxidation studies of individual particles were conducted in order to determine the nanoscale dynamics of the oxidation using a Hitachi HF3300S ETEM.This microscope is an open cell ETEM with gas supplied via the holder and controlled through mass flow controllers.Further specifics are discussed elsewhere [19].The microscope was operated at either 300 kV or 100 kV acceleration voltages.CB was dry dispersed on Micro Electromechanical System-(MEMS)-chips from Norcada Inc., resistively heated to 600 • C using a tungsten heating coil embedded in the MEMS-chip, with a factory-provided temperature-resistivity calibration.The oxidation was carried out using 1 Pa of oxygen gas (Linde HiQ oxygen 6.0, ≥99.9999% O 2 ).The pressure was measured using transducer near the sample with a correction factor, as detailed in reference [20].The sample region of the MEMS-chips consists of amorphous SiN x , with several electron transparent regions with holes to allow for imaging without the SiN x support present.Time-series  of images at 20 frames per second were acquired on an AMT-XR401 camera for several different dose rate conditions.The dose rate was quantified by recording the current from a Faraday cup located on a Gatan Double Tilt 646 Analytical holder and then, without adjusting microscope settings, imaging the entire beam on a Gatan Orius SC1000B camera.This was used to calibrate the dose rates in the image series.The dose rate was controlled by setting the illumination area with the condenser lenses for a fixed beam current; typically, a 10 μA extraction current was used, however to reach the highest dose rates, the extraction current was increased to 20 μA (note that the actual beam current on the sample is several orders of magnitude smaller than the extraction current).
The size of individual CB particles was measured at multiple points in each image series by measuring the projected area of the particles.To calculate the radii of the particles, the square root of the area was calculated.Subsequently, the rate of oxidation was calculated through the construction of a linear regression of the measured radii, where the slope is the oxidation rate.The particles were oxidized until fully consumed.The projected area measurements are orientation dependent, which can result in more imprecise individual measurements of the oxidation rate, although this effect is accounted for by averaging the oxidation rate over many individual particles.Some images (stated in the figure captions) have had their contrast enhanced using the CLAHE algorithm as implemented into ImageJ [21].

In situ EEL spectroscopy
To more precisely measure average oxidation rates in agglomerates of CB particles, EEL spectroscopy measurements were performed using a FEI Titan ETEM with a Tridem 866 GIF.This microscope is also an open cell where the gas was injected directly into the sample cell.It was operated at either 80 kV or 300 kV acceleration voltages, further details are described elsewhere [22].EEL spectra were acquired in selected area mode with a dispersion of 0.2 eV, a collection angle of 2.9 mrad, a full-width at half-maximum of the zero-loss peak of ca 1.1 eV, and an exposure time of either 0.05 or 0.1 s depending on the dose rate used.The spectrum time series were recorded at approximately 3 spectra per second with a custom acquisition script in Gatan Microscopy Suite (Digital Micrograph), with acquisition being halted after 2000 spectra, ca 10 min, or when the carbon signal was no longer discernible.Similar to the image series at low oxygen pressure, these spectra were acquired at various dose rates set by changing the gun lens setting while keeping the illumination area roughly constant.The dose rates were determined by acquiring images of the beam and using the calibration of a Gatan OneView camera.The sample was dry dispersed on Wildfire Nano-Chip MEMS-chips from DensSolutions with continuous SiN x films.The oxidation was carried out at 600 • C and 260 Pa O 2 , and at 600 • C and 670 Pa O 2 .

Analysis of EEL spectra
EEL spectroscopy analysis of the carbon K-edge was carried out by fitting a background from 250 eV to 270 eV.The background was modelled as an exponential function which has been found to give better fits to data at the tail of a strong plasmonic peak [23,24].For background fitting of the entire time series, a moving average of 21 (corresponding to 7 s) was applied to reduce the influence of noise in the fitting region on the extrapolation of the background.Subsequently, a wide signal integration window, 282 eV-350 eV, with the onset of the C-π* edge being set to 284 eVwas used to reduce variation induced from multiple scattering.
As EEL spectroscopy directly probes the volume of the material, the cube root of the signal, normalized to the initial intensity, was calculated to obtain a radius equivalent.Subsequently, a linear regression was used to determine the slope of the quantified EEL signal.This linear regression was fit to the mid-region of the oxidation process avoiding the nonlinear onset and tail, and the slope of this fit was determined as the oxidation rate.
Note that the observed reaction rate is described in terms of an intrinsic oxidation rate and an additional extrinsic, beam-enhanced, oxidation rate.The intrinsic oxidation rate is determined by the reaction conditions, e.g., temperature, gas pressure, and carbon nanostructure; whereas the beam-enhanced oxidation rate depends on the dose rate and energy of the PE in concert with the intrinsic factors.Depending on the method used to calculate the beam-enhanced oxidation rate, the units are different; nm s − 1 when the projected area method is used, and s − 1 when EEL spectroscopy is used.Both methods provide a signal that is proportional to the amount of carbon oxidation per second, however they are not directly comparable.

Theoretical evaluation of electron-matter interactions
The likelihood of oxygen ionization (labelled reactions (1-3) in the theory section) by electrons was evaluated using theoretical ionization cross-sections calculated using the Relativistic Binary Encounter Bethe model [25] with values for oxygen from Hwang et al. [26].Homolytic cleaving, i.e., producing neutral oxygen radicals (labelled reaction (4) in the theory section), was modelled using a relativistic Bethe asymptote, which was fit to data from reference [27].The sputtering cross-section was calculated using Mott's theoretical framework reworked for sputtering [28] as well as bond strength data from Atkins et al. [29], both are discussed in more detail in supplementary information S1.

Theory
The nature of the electron beam effects for in situ oxidation experiments are disparate from the extensively studied "ex situ" damage mechanisms in vacuum, such as those discussed by Egerton et al. [30].In oxygen atmospheres with elevated sample temperatures, drastic increases in oxidation rates (compared to intrinsic rates without an electron beam at otherwise identical conditions) are typically observed [16].In contrast, in vacuum conditions carbonaceous materials tend to graphitize [31,32], bend and form so-called "carbon onions" [32], and slowly lose mass [32].These transformations have been ascribed to atomic displacements: elastic interactions between the primary electrons (PE) from the beam and the sample atoms' nuclei, resulting in sample atoms being displaced from their lattice sites through momentum transfer [30,33].Inelastic interactions with the electrons in the sampleresulting in electronic excitations, ionization, and charging are considered less important for graphitic nanocarbons [32]; here, the resulting excitations are quenched by conduction band electrons from outside the irradiated area.Finally, electron beam irradiation can result in heating of the sample (as a byproduct of inelastic interactions).While the extent of heating is difficult to estimate and under debate [30], it is unlikely to play a major role during nanocarbon oxidation; drastically increased oxidation rates linked to the electron dose rate are also observed for graphene [15], a material well known for its high heat conductivity [34].
To understand how high-energy electrons can affect the oxidation of CBhaving ruled out the ex situ mechanisms abovea brief overview of the intrinsic oxidation reaction mechanism is helpful.CB oxidation is idealized as a surface process where the individual particles oxidize from the outside in.The oxidation rate then becomes proportional to the surface area of the particle, resulting in a linear reduction of the diameter over time, as shown in reference [16].However, CB is a highly heterogenous material on the nanoscale consisting of turbostratic carbon [35]: a mixture of graphitic and disordered carbon.Nevertheless, CB tends to consist of a shell with a higher degree of graphitic carbon and core with a higher degree of disordered carbon, each with different reactivities.The graphitic shell can be likened, in terms of reactivity, with the basal sites of graphene with strong oxidative resistance [7,22,23].In graphene, oxygen has a strong preference for reacting with edge sites rather than the with the basal planes [3,36] due to unbonded sp 2 -orbitals [37]; this difference in reactivity has been measured to the order of 10 12 [38].As a result, oxidation will be concentrated to parts of the CB shell with edge sites, present where a carbon atom has been removed, i. e., at vacancies and defects; and to the more disordered core if oxygen can diffuse through the shell.Reactive defects are inherent in the graphitic basal planes in CB [10], but can also be formed through interactions with activated oxygen species [13,39], or by atomic displacement, where carbon atoms are knocked out of the structure.The defects in turn act as active sites where adsorbed oxygen species can directly abstract carbon [36,40,41].
With this intrinsic reaction mechanism in mind, PE have three main routes for enhancing the oxidation rate as illustrated in Fig. 1: ionization of molecular oxygen (1) by PE [11,12] or (2) by SE [14] resulting in atomic and ionized oxygen species which can attack the basal graphitic planes directlyor (3) by atomic displacement resulting in defects or vacancies where molecular oxygen can abstract carbon.While atomic displacement effects have been discussed for ETEM experiments in for example heterogenous catalysis in thin oxide films [42], this mechanism has received little attention for CB oxidation experiments.Finally, we note that the electron beam will not only influence the induced oxidation rate but may also change the oxidation mechanism; a large amount of induced defect sites in the outer graphitic shell, or formation of highly reactive oxygen ions and radicals, can be expected to shift the reaction towards more surface oxidation.
In the following theory sections, the three different routes outlined here will be described in more detail to show their distinguishing features and how they can be controlled by experimental parameters.

Ionization of gas by primary electrons
The interaction between PE and gas molecules produces reactive oxygen species (through reactions 1-4) able to directly attack the basal planes of the graphitic shell of CB, as schematically illustrated in Fig. 1 (1).Note that these reactions and the resultant oxidative species are also present in fuel-oxygen flames [4,43].Of these reactions, three (1-3) are classical ionization reactions [27,44].The remaining reaction (4) is the homolytic cleaving of oxygen into two radicals; note that while homolytic cleaving can occur as a result of an elastic process [30], it primarily arises from optically allowed transitions excited through inelastic processes [45].While oxygen is of primary interest for CB oxidation, previous reports have suggested that similarly ionized species from inert gases, such as N 2 , are sufficiently reactive to affect carbon nanostructures [12].The framework presented here is also applicable to such diatomic gases.
The likelihood of these reactions occurring is described by the ionization cross-section, for reaction (1-3), and by the cross-section for nonionizing electron transitions, for reaction (4) [45].The larger the cross-section, the higher the probability of a PE interacting with an oxygen molecule, resulting in a higher generation rate of reactive species, and consequently larger impact on CB oxidation.For the ionization processes (1-3), the cross-sections show a similar trend: they all have a threshold of the order of 10 eV below which they cannot occur [27,46], then rapidly increase to maxima at around 100 eV, after which they slowly decrease with increasing electron energies.For oxygen gas, the total ionization cross-section (summing the cross-sections of (1-3)) maximum occurs already at 112 eV [27], and drops to only ca.1/100 of the maximum magnitude at 100 keV (a more TEM-relevant energy).Increasing the electron energy to 300 keV further decreases the cross-section by a factor of 2. When modelled to data from reference [27] and using a relativistic Bethe asymptote as per reference [45,47], the cross-section of reaction (4) varies similarly to that of reactions ( 2) and (3) at TEM relevant energies, see supplementary information S1 for more details.
The small cross-sections for ionization imply that only a small fraction of the O 2 present in the microscope will be ionized, see supplementary information S2.This means that the fraction of ionized species will scale linearly with the beam current (i.e., double ionization is negligible).Note that the small fraction of ionized species does not in itself rule out ionization by PE as a relevant effect since these might have an outsized contribution to the oxidation.
The influence of the ionized oxygen species will not be confined to the area illuminated by the electron beam.For typical pressures (1 Pa-1500 Pa) used in ETEM studies, gas molecules have mean free paths of 10 − 2 to 10 − 5 m, resulting in ionized species being able to travel far from their point of origin before interacting with other gas molecules.Furthermore, there is no mechanism inherent in ionization to guide the species towards the sample; the momentum imparted on the gas in the direction of the sample is negligible.As such, the ionized species should impinge on the sample in an area much larger than that directly illuminated by the electron beam.Specifically, the spreading of ionized species can be modelled as either diffusion in the high-pressure limit or ballistic transport in the low-pressure limit, with the distribution at ETEM-relevant pressures lying in between, see supplementary information S3.Both these models indicate that the impingement area of ionized species should greatly exceed that of the electron beam, and consequently that there should be appreciable beam-enhanced oxidation occurring outside the irradiated area.
The charge on the species formed by PE ionization further implies that the illuminated region is somewhat shielded from the ionized species.The sample will have a slight positive charge due to inelastic interactions with the electron beam [30], and the ionized species will be predominantly positively charged as a result of the relative magnitudes of the cross-sections for reactions (1-3), see supplementary information S1 and S4.Therefore, some repulsion between the positively charged oxygen species and the positively charged CB sample is to be expected [48], further decreasing the localization of the effect.

Ionization of gas by secondary electrons
In addition to interactions with PE in the beam, gas molecules can be ionized by secondary electrons (SE) emitted from the sample and sample support, as illustrated in Fig. 1(2).Because of their low energy, sub-50 eV, SE have much larger interaction cross-sections than PE for reactions (1-4) as described in the previous section.SE therefore generate a larger number of ionized oxygen species per electron than PE.Note that while SE are also generated from the gas in reactions ( 2) and ( 3), these will provide a negligible addition to the total ionization and will be subject to the same localization limitations as ionization by PE.However, SE emission from the sample represents a truly additional process, which needs to be accounted for on its own.
As SE are a byproduct of inelastic interactions, the SE emission rate depends on the PE energy (in addition to the dose rate) [49,50].At PE energies relevant for TEM imaging, the PE interaction cross-section of the sample and sample support decreases with increasing energy, just as described for gas ionization in the previous section; an increase from 100 keV to 300 keV results in the ionization cross-section, coupled to SE generation, decreasing by a factor of 2, see supplementary information S1.Consequently, the beam-enhanced oxidation from gas ionization by SE is expected to diminish with increased electron energy in a similar manner to gas ionization by PE.
In contrast to the SE emission rate, the SE energy distribution is independent of PE energy and peaks at low energies [51][52][53].The low energy of SE results in a larger generation of negatively charged species compared to ionization by PE.Due to the low energy, only around half of the SE will have sufficient energy to partake in reaction (1-3) [51,54], see supplementary information S4.Therefore, of the few SE generated, fewer still are able to influence the reaction.Despite the low energy, the small interaction cross-section of (1) ensures a predominance of positively charged and neutral oxygen radicals for this process; nevertheless, some negatively charged species will be present.
Ionization by SE is a more local effect than ionization by PE.In contrast to ionization by PE, ionization by SE predominantly occurs close to the irradiated region on the sample where the SE are generated.Given that the ionization occurs closer to the sample, a larger fraction of the ionized oxygen species will impinge in the irradiated area.This local nature is compounded by the much larger interaction probability of SE (reducing their mean free path in the gas).Additionally, as the sample emitting the SE is slightly positively charged [30], there is a potential for attracting the (few) negatively charged oxygen species to the sample.

Atomic displacement in sample
Elastic scattering, whereby PE transfer momentum to atomic nuclei in the sample, can create active sites for oxidation by displacing atoms from their lattice positions, in a process called atomic displacement, Fig. 1(3).Above a certain PE energy threshold, related to the energy required for breaking the bonds between the target atoms, this process is largely unaffected by changes in PE energy (in contrast to ionization).For example, in carbon the atomic displacement cross-section is reduced only by approximately 15% as the PE energy is increased from 80 keV to 300 keV (assuming a displacement barrier of 15 eV) [32], whereas the ionization cross-section is halved, see supplementary information S1.
The PE energy threshold for permanent atomic displacement in terms of sputtering can be high (above 100 keV, in contrast to only tens of eV for ionization of oxygen gas) for perfect crystals.However, this threshold is drastically reduced by imperfections in the material.As a baseline, the displacement of a carbon atom in bulk graphite from its lattice position to an interstitial site requires PE energies greater than 140 keV [30].For a single layer of graphene, the threshold is lowered to around 110 keV [55,56] due to the lack of confinement from surrounding layers.The displacement threshold is furthermore reduced around defects, as illustrated by the 95 keV threshold for removing a carbon atom from 5-membered carbon rings in 5-7 dislocations in D. Wahlqvist et al. graphene [57].Consequently, nanocarbons display even lower atomic displacement thresholds: CNTs being essentially unidirectionally curved (strained) graphene sheets have an ideal displacement threshold of 86 keV.However due to defects in the lattice structure, they have been shown to be damaged even by 80 keV PE [12].Since CB consists of turbostratic carbon, there should be a large number of sites susceptible to re-arrangement even by low energy PE.Finally, we note that PE energies below the threshold value does not entirely prevent atomic displacements; due to coupling of displacements with thermal lattice vibration, there is a gradual onset of damage at least 20 keV below the threshold (for graphene) [56].
For in situ oxidation, nanocarbons are even more susceptible to atomic displacements effects than suggested by the (already low) sputtering thresholds described above.Smaller displacements, with lower energy barriers (and consequently lower PE energy thresholds), can result in transient defects with increased reactivity towards oxygen.This transient damage mode will not be observed during ex situ observations where such small defects will anneal over time (or constitute a first step towards rearrangement into carbon onions).Such transient displacement has been experimentally confirmed; electron irradiation for instance prevents graphene from attaining stable edge configurations [58].It has furthermore been observed that linear carbon-carbon bonds break rapidly during observation but reform faster than atoms can be removed [59].
These atomic displacement effects should scale linearly with dose rate [15].However, a saturation of active sites may be reached due to merging at high concentrations [60].Furthermore, displaced atoms (unless sputtered away completely) can reform broken bonds at nearby sites [59,61].The increased reaction rate can therefore also level off if the impingement rate of gas molecules is insufficient to react with all active sites before they anneal.

Results and discussion
As is evident from the theoretical overview, the different mechanisms through which high energy PE can affect the CB oxidation process have different implications for whereand under what conditionseach mechanism should be most prominent.To differentiate between the suggested mechanisms, we focus on qualitative comparisons such as the effect of the reactive gas and the localization of the effect on the CBagglomerates.Additionally, we provide quantitative comparisons through measurements of the oxidation rates at various PE energies and dose rates for both individual CB particles via images and agglomerates via EELS.

Qualitative measurements
Electron irradiation effects predominantly serve to enhance the rate of the intrinsic oxidation process, i.e., the drastic effects only occur in conditions where oxidation also occurs on its own to some extent.In inert gases, as illustrated for N 2 at 600 • C in Fig. 2, little to no carbon is abstracted over the course of the experiment.Instead, the sample is observed to gradually graphitize, in line with the expected ex situ damage mechanisms [32].Similarly, in O 2 at 200 • C the sample is not significantly affected over 5 min of exposure to the electron beam.In contrast, during in situ oxidation at 600 • C, the beam-enhanced oxidation is typically completed within a few minutes as detailed extensively in the latter sections.Thus, the dramatic beam-enhancement of the oxidation rate requires conditions where oxidation would occur also without electron irradiation.Note that while damage to CNTs have been reported for N 2 in combination with electron irradiation [12], the small defects generated here are of a different nature than the orders-of-magnitude enhanced reaction rates observed in oxygen.
The beam-enhanced oxidation rate is highly localized to the irradiated area, as illustrated in Fig. 3. Here, overview images of a CB agglomerate, acquired before and after condensing the beam to a smaller (intense) spot in the center of the field-of-view, are compared; the conditions (670 Pa and 600 • C) are conducive to slow oxidation, with the irradiation (80 keV, 7 × 10 4 e − nm − 2 s − 1 ) providing a significant enhancement.An even lower-magnification overview of the same area is shown in supplementary information S5.The CB located just outside the electron beam is largely unaffected (Fig. 3(3a and b)); the change in the size and shape of the pore is mainly due to movement of the individual CB particles as discussed in supplementary information S5.However, there is clear oxidative damage in the area of the sample irradiated by the condensed electron beam (Fig. 3(4a and b)).The CB located around the edge of the electron beam is unevenly affected as some areas are highly oxidized (Fig. 3(2a and b)), whereas others are essentially unaffected (Fig. 3(1a and b)), likely dependent on the initial nanostructure of the particles and the unsharp cut-off of the electron beam.The electron irradiation extends 15-20 nm from the apparent edge, illustrated by the dashed oval in Fig. 3, before the intensity drops below 10% of the average in the center of the beam, see supplementary information S6.This indicates a predominantly local nature of the beamenhanced oxidation confined to the carbon within the electron beam path.
The local nature of the effect indicates that gas-ionization by PE is not the primary mechanism responsible for the increased oxidation rate.As detailed in the theory section, ionized species can travel far outside the beam region and there is no plausible mechanism by which the ionized species should be attracted to the irradiated area to enhance the localization of the effect.Rough calculations indicate that the number of ionized oxygen species impinging within 100 nm (which would include the whole region in Fig. 3(3)) is comparable to the directly illuminated area for all experimental conditions examined herein, see supplementary information S3.In contrast, atomic displacement-induced oxidation will only occur within the irradiated area where the incoming electrons displace atoms to form active sites.Ionization by SE is also expected to be highly concentrated near, but can extend slightly outside, the illuminated area (possibly contributing to the small changes observed in Fig. 3(3)).

Quantitative analysis of oxidation behavior
Quantitative measurements of the oxidation rate provide further information for distinguishing between gas ionization and atomic displacement through comparisons of different dose rates and PE energies.Two complementary methods of quantifying the oxidation rate of CB are used.For individual particles, high resolution image series are collected during oxidation.In each frame, the projected area of the CB particle is extracted and converted to the radius of the equivalent circle, as illustrated in Fig. 4a-b.As expected for these conditions, the radius decreases linearly with time, as exemplified for the particle in Fig. 4c, due to the surface area-limited nature of the oxidation reaction [16].This radius decrease is directly linked to the oxidation rate of the CB.To get a more accurate reaction rate averaged over many CB particles, time series of EEL spectra are collected during oxidation of agglomerates of CB.The oxidation kinetics of the agglomerates is tracked in EEL spectroscopy from the gradual decrease of the C-K edge over time.The signal in EEL spectroscopy arises from the entire volume of the sample.To determine a radius-equivalent oxidation rate, a linear regression is calculated for the cube root of the C-K edge signal, normalized by its initial intensity.The oxidation rate is calculated as the slope of the linear region of the EEL-time series, as illustrated in Fig. 4f.During the initial oxidation, individual CB particles, initially covered by layers of haphazardly ordered carbon [6], are gradually exposed to O 2 as the less ordered carbon is preferentially oxidized.Thus, oxygen gas becomes increasingly able to permeate the entire agglomerate, leading to a gradual increase in the total oxidation rate as the particles oxidize independently and a linear oxidation regime is established.Note the similarity of the oxidation kinetics evaluated for single particles by HR images and for agglomerates by EEL spectroscopy, as shown in Fig. 4c D. Wahlqvist et al. and f, with both having a linear region where the oxidation rate is constant.Many individual particles also show a delayed start to their oxidation, see supplementary information S7, which provides connections to the delayed start observed for agglomerates in the EEL measurements.
The quantitative EEL spectroscopy measurements demonstrate a saturation effect of the beam-enhanced oxidation at high dose rates, which is difficult to explain in the gas ionization framework.The agglomerate data in Fig. 5, and supplementary information S8, show that the oxidation rate increases with the dose rate.Initially, the observed beam induced enhancement is consistent with a linear effect of the electron dose rate.However, the enhancement becomes sub-linear at the highest dose rates.This non-linearity of the dose rate response is an indication that ionization is not the dominant process; the number of ionized species should scale linearly with the dose rate since the interaction probability should be a simple product of the cross-section of ionization and the beam current (when the fraction of ionized species is low and double-ionization events are negligible).In contrast, atomic displacement displays characteristics consistent with non-linearity, e.g., if the oxidation process becomes limited by the impingement of oxygen atoms and the defects have time to anneal (as detailed in section 3).
The beam-enhanced oxidation rate is furthermore independent of PE energy; no significant difference can be discerned between 80 keV and 300 keV, as shown in Fig. 5.This is inconsistent with the ionizationbased mechanisms, both by PE and SE; a halving of the oxidation rate is expected as per supplementary information S1 when increasing the PE energy from 80 keV to 300 keV.However, it is consistent with atomic displacement where a small decrease of only a few percent is expected, less than the accuracy of the experiments given the heterogeneous nature of CB.Note that it is imperative to quantify the beam-enhanced oxidation per electron in order to understand the mechanisms by which the electrons affect the reaction; often, changes in acceleration voltage also lead to large changes in electron beam current for otherwise similar illumination settings.
High resolution TEM imaging provides an alternative measurement of the beam-enhanced oxidation rate for individual particles.While EEL spectroscopy is an excellent method for measuring the collective beamenhanced oxidation of agglomerates, the signal-to-noise ratio was found to be insufficient for analysis of single particles.Here, high resolution imaging provides complementary information and allows for direct studies of the structural evolution during the oxidation process.Using high resolution imaging, we confirm the same saturation phenomenon as was found for the agglomerate data.This saturation is illustrated by the black line connecting the mean of the oxidation rate for each dose rate at 300 keV in Fig. 6.The saturation is independent of PE energy and oxygen pressure and is therefore linked to the electron interaction mechanism.
There is no clear dose rate threshold below which the electron beam does not increase the oxidation rate from its intrinsic value, as shown in Fig. 6.However, it is possible to determine dose rate limits where the intrinsic processes dominate the oxidation to some (arbitrary) degree.To illustrate the principle, we here define such a limit as where the total oxidation rate is less than twice the intrinsic (zero dose rate) oxidation rate.The intrinsic oxidation rate is (approximately) determined by sparse intermittent imaging at low dose rates.With this definition, electron dose rates must be kept below ca.150,000 e -nm − 2 s − 1 for the present conditions (1 Pa O 2 and 600 • C).Note that where such limits are used in future studies, they must be determined for the sample and oxidation condition in question and be subjected to proper statistical tests.
The electron beam does not only alter the oxidation rate, but also affects the oxidation mode.This is exemplified via a qualitative determination, see supplementary information S9, of whether particles react primarily via internal or surface oxidation.Fig. 6b shows that internal oxidation tends to be more common at lower electron dose rates.We set a broad range of 0.4-0.7 (due to the large variation between CB particles, and hard-to-discern features in the high-resolution images) for the intrinsic internal oxidation fraction based on an observation of 90 Fig. 5. Oxidation rates measured from in situ EEL spectra for CB at 600 • C and 670 Pa O 2 for 300 keV (➕ ➕) and 80 keV (⬟) electrons.The dashed black line is a guide for the eye to show the saturation effect at high electron dose rates.Note that there is an intrinsic oxidation rate even at 600 • C, discussed in further detail in Fig. 6.EELS data for oxidation at 600 • C and 260 Pa can be found in supplementary information S8.individual particles.An exact dose rate limit for when the primary oxidation mode is changed cannot be determined, but it clearly shifts towards surface oxidation when comparing e.g., 10 5 e -nm − 2 s − 1 to 10 6 e -nm − 2 s − 1 .This effect is mainly dependent on the dose rate, not PE energy, as the trends for the 100 keV and 300 keV series are similar.The increased prevalence of surface oxidation at higher electron dose rates can be explained by a higher concentration of active sites on the surface of the CB particles, resulting in a preference for surface oxidation rather than oxygen diffusion into the disordered core.
Even though we show that PE energy has little effect on the beamenhanced oxidation, it may still be the case that lower energies are preferable [4].Even though the effect per electron does not change, as per Figs. 5 and 6, fewer low-energy electrons are required to acquire images and EEL spectra with sufficient signal-to-noise ratios.In aggregate, the effect of the beam during data acquisition can be smaller for lower energy electrons at the most commonly used acceleration voltages for TEM.However, note that very low PE energies (well below 80 keV) may end up completely shifting the balance by greatly increasing the effects from gas ionization (by either PEs or SEs).
Having established atomic displacement generating reactive sites as the primary mechanism behind the beam-enhanced oxidation, we turn to some previously reported phenomena.Dadsetan et al. found that certain CB types showed a reduced oxidation rate when the sample is irradiated [18] in scanning TEM (STEM) mode.The use of STEM complicates the dose rate calculations due to the intermittent flow of electrons over the sample: the dose rate is extremely high for a short time (the dwell time of the electron probe) followed by a longer time when it is zero [62].This high dose rate means that STEM imaging, intermittently, will be well into the saturation regime as noted in Figs. 5 and 6.This gives time for other beam damage mechanisms to occur, for instance graphitization and removal of defects and heteroatoms, resulting in the CB becoming more resilient to oxidation.An excess of active sites has also been shown to reduce the oxidation rate in pure graphite due to a possible reformation to a diamond-like structure [60].In contrast, Yoshida et al. show that gas ionization by SE can play an important role in enhancing the etch (oxidation) rate of CNTs [14].This is based on supported CNTs showing greater oxidation rates than the suspended CNTs, particularly at 80 kV acceleration voltage, due to the additional SEs generated from the underlying SiNx film.However, they also identify an inherent difference between 200 keV and 80 keV electron energies, with the former always showing the highest oxidation rates.This cannot be attributed to gas ionization by SE, and Yoshida et al. identifies this to be atomic displacement, in line with our findings.Note that CNTs with few defects will have a PE energy threshold for atomic displacement right around 80 keV, which explains the comparatively large contribution from SEs in this case.Nonetheless, SEs might provide a small oxidation enhancement also for CB under the conditions reported here, e.g., potentially contributing to the small alterations seen outside the irradiated area in Fig. 3(3).
Finally, we note that internal burning is not only applicable to the low-pressure regime but is also relevant at higher pressures.Toth et al. observed CB oxidation at different experimental conditions; 1 Pa, 10 Pa and at 600 • C and 900 • C [6].From these results, they constructed a table over which oxidation processes occur at different reaction conditions and different times in the oxidation process [6].From Fig. 7, it is evident that internal oxidation dominates the reaction over long time scales at higher pressures, extending the trends previously described at 600 • C for 1 and 10 Pa [6] to 260 Pa and 670 Pa.

Conclusions
The beam-enhanced oxidation rate of carbon black in the ETEM is strongly dependent on the incident electron dose rate and the primary electron energy in a way that indicates that transient damage in the form of atomic displacements, followed by fast oxidation of the generated defect, is the primary underlying mechanism.This is supported by several lines of experimental evidence: the local nature of the oxidation, the saturation of the beam-enhanced oxidation rate at higher dose rates, and the lack of change in the beam-enhanced oxidation rate as the PE energy is changed.These data preclude ionization by PE and SE as the primary mechanisms and are better described by atomic displacement.Note that SE can contribute, in a manner similar to that shown by Yoshida et al. [14].
As atomic displacement effects on the sample are the main cause of the increased oxidation rate, there is no completely safe conditions for observation in the ETEM; there is no threshold for either electron dose rate or PE energy (currently feasible for ETEM) below which beamenhancement of the oxidation rate will not occur.However, it is always possible to select conditions where the beam-induced oxidation will be (arbitrarily) smaller than the intrinsic oxidation.As an illustration of this principle, 150,000 e − nm − 2 s − 1 is suggested to be the upper limit for CB oxidation in 1 Pa O 2 at 600 • C, below which the intrinsic process accounts for more than half of the total observed oxidation.In conclusion, ETEM can be an invaluable tool for carbon oxidation studies, especially for comparing trends between different oxidation conditions or nanocarbon structures.However, care must be taken to account for electron beam effects and validate the results with external methods.D. Wahlqvist et al.

Fig. 1 .
Fig. 1.The three proposed modes of beam-induced oxidation; (1) gas ionization by primary electrons (PE), where a high-energy PE interacts with an oxygen molecule (red spheres) knocking out an electron, resulting in either (1a) direct ionization or (1b) dissociative ionization; (2) gas ionization by secondary electrons (SE), where PE interact with the sample in such a way that a lowenergy SE is emitted, which in turn interacts with an oxygen molecule leading to either (2a) direct or dissociative ionizationsame as for (1) -or (2b) electron capture; (3) atomic displacement, where a temporary or permanent displacement of a carbon atom (black spheres) by the PE gives rise to a defect site where molecular oxygen can oxidize the sample directly.

Fig. 2 .
Fig. 2. The effect of the electron beam in non-reactive conditions.Contrast enhanced TEM micrographs showing the effects of electron beam irradiation (dose rate 5 × 10 5 e -nm − 2 s − 1 ) on CB in (a) 1 Pa N 2 at 600 • C and (b) O 2 and 200 • C. In N 2 , only graphitization of the CB is observed after 34 min of continuous irradiation.In O 2 at 200 • C the reaction is slow (but detectable) even under intense electron irradiation.The scale bars are 10 nm.

Fig. 3 .
Fig. 3.The localization of the electron beam effect.The overview image shows an area before converging the beam to the blue circle (the solid circle corresponds to the approximate edge of the beam and the dashed circle corresponds to where the beam intensity is more than 10% of the average of the central area, see supplementary information S6).The insets show areas of interest selected based on the fine features recognizable (a) before and (b) after oxidation.Note the lack of apparent damage to the insets located outside the electron beam.The area was irradiated for 8 min with a dose rate of 7 × 10 4 e -nm − 2 s − 1 .

Fig. 4 .
Fig. 4. Overview of TEM imaging and EEL spectroscopy methods for quantifying CB oxidation.(a-b) Sample HR micrograph time series illustrating how the oxidation of an individual CB particle is followed and quantified as the corresponding radius based on the shrinking of the projected area.(c) The radius decreases linearly over time, giving an oxidation rate of 0.06 nm s − 1 .(d-e) Selections from an EEL spectra time series showing the reduction in the carbon K-edge, between the red vertical lines, over time.(f) The integrated background-subtrated carbon signal from the entire time series is shown with the points corresponding to the individual spectra in (d) and (e) marked, the radius equivalent decreases linearly with an oxidation rate of 0.002 s − 1 .

Fig. 6 .
Fig. 6.(a) Oxidation rates and (b) mechanisms determined from highresolution ETEM images of CB acquired at 600 • C and 1 Pa O 2 .The effects of increasing electron dose rates are shown for both 300 keV (➕ ➕) and 100 keV (⬟) PE energies.The black line directly connects the mean of each 300 keV data to illustrate the saturation effect.The green shaded area represents a range of intrinsic oxidation behavior, determined from low-dose intermittent imaging.In (b), note the dramatically increasing fraction of surface oxidation at high electron dose rates.

Fig. 7 .
Fig. 7. Internal burning as observed with EELS and imaging at the higher pressure ranges.(a) An example of undamaged CB before oxidation and (b) the same area after oxidation at 600 • C and 260 Pa.Note the loss of contrast inside the particles which indicates internal burning.(c) Quantification of the C-K edge during the oxidation process, with arrows indicating when the micrographs in (a) and (b) were acquired.The sample was oxidized for 12 min between (a) and (b) with a dose rate of 3 × 10 4 e − nm − 2 s − 1 at a PE energy of 80 keV.The insets in (c) emphasizes an area where internal oxidation is occurring and the original particles are recognizable (the scale bar is 20 nm).