Revealing the Adsorption Mechanisms of Nitroxides on Ultrapure, Metallicity-Sorted Carbon Nanotubes

Carbon nanotubes are a natural choice as gas sensor components given their high surface to volume ratio, electronic properties, and capability to mediate chemical reactions. However, a realistic assessment of the interaction of the tube wall and the adsorption processes during gas phase reactions has always been elusive. Making use of ultraclean single-walled carbon nanotubes, we have followed the adsorption kinetics of NO2 and found a physisorption mechanism. Additionally, the adsorption reaction directly depends on the metallic character of the samples. Franck–Condon satellites, hitherto undetected in nanotube–NOx systems, were resolved in the N 1s X-ray absorption signal, revealing a weak chemisorption, which is intrinsically related to NO dimer molecules. This has allowed us to identify that an additional signal observed in the higher binding energy region of the core level C 1s photoemission signal is due to the C=O species of ketene groups formed as reaction byproducts . This has been supported by density functional theory calculations. These results pave the way toward the optimization of nanotube-based sensors with tailored sensitivity and selectivity to different species at room temperature.


Samples
The single-walled carbon nanotubes (SWCNTs) used in our experiments are material in which semiconducting and metallic separation has been achieved, allowing the formation of high-purity networks. S1 The overall morphology of these films can be observed in the scanning electron microscopy (SEM) micrograph shown in Fig. S1. This image corresponds to the semiconducting sample but there is no observable difference between both types of films with this imaging technique. Both types of samples were purified and separated into metallic and semiconducting tubes. They were then deposited on sapphire substrates. S2,S3 In our previous studies on these kinds of samples, we have shown that the SWCNT buckypapers have less than 1 wt% of processing residuals with X-ray photoelectron spectroscopy (XPS), with negligible residual magnetic metal impurity content. Typically, in these samples the remaining amount of surfactants, such as deoxycholate sodium salt, is below the detection limit of XPS and Fourier transform infrared spectroscopy (FT-IR). The G/D ratio in the Raman spectra of these SWCNT samples is typically around 20 from previous studies. S2,S3 As seen in the optical absorption spectra of these high purity samples (metal and semiconducting) shown in Fig. S2, no bands associated with the other conduction type are identified in either spectra.
Other spectroscopy studies that prove the high purity of the samples can be done with photoemission and X-ray absorption spectroscopy, and in particular valence band photoemission. These studies would be useless in the presence of impurities using a macroscopic sample, as can be understood from some previous publications. S4-S7  Additionally, our previous reports on these type of nanotube networks have confirmed their ultra-high purity with other techniques. For instance, it is now understood how the presence of semiconducting species affects the conduction mechanisms in SWCNTnetworks. The transport mechanisms systematically change as the S1 relative content of metallic to semiconducting SWCNTs is varied and quantum transport was achieved only in macroscopic networks of pure metallic SWCNTs. S2 Core-level shifts provide an electronic "fingerprint" of the local chemical environment of an atomic species. This is due to the high degree of localization of these nearly-atomic levels, such as the C 1s. For this reason, simulated C 1s spectra for molecular species may be used to obtain a "blueprint" of the expected core-level shifts for various atomic species (C≡O, C=O, C-O, C-N) within representative functional groups (ketene, carbonyl, acetyl, hydroxyl, epoxide, nitro, amine oxide, pyridyl, etc.). Thus, to obtain further insight into the possible functional groups that may be present in the experimental SWCNT samples we have simulated the C 1s spectra for the representative molecules listed in Table S1 in gas phase.

Simulated C 1s Spectra
To this end, we have employed more than 6 Å of vacuum and non-periodic boundary conditions in all directions. In this way, we ensure a common vacuum-level reference between the all-electron C 1s eigenenergies calculated for the various molecular species in Table S1. The C 1s spectra is then simulated by employing a Lorentzian broadening with an inverse lifetime of 0.27 eV, and aligning the calculated and experimental C 1s levels of the C-C species in acetone, as shown in Fig. S3.
Comparison of the simulated C 1s spectra in Fig. S3 with the high binding energy feature in the XPS C 1s spectra for SWCNTs dosed with NO 2 clearly indicates this feature is due to a C=O species, as both C-O and C-N shifts are too small. Further, the core-level shift arising from a ketene group yields the largest shift, and is well separated from the carbonyl and acetyl groups. This provides further verification that this feature oberved in the XPS C 1s spectra is due to a C=O species in a ketene group on a SWCNT.Furthermore, the good matching shown in Fig. S3 between calculated and experimental XPS spectra for acetone (red squares) S8 validates our computational approach.

Oxygen on SWCNTs.
To further justify our assignment of the feature observed in the XPS C 1s spectra at high binding energy to the C=O species of a ketene group, we have performed further calculations for the C=O species of a carbonyl group and C-O species of epoxide and chemisorbed O 2 . Since core states are highly localized, we can investigate the effect of several functional groups within the same calculation. In this way, we can ensure a common reference between calculations. The optimized geometries, C 1s levels, and simulated spectra for a carbonyl group adsorbed on a carbene site of a monovacancy, an epoxide group  Figure S3. Simulated C 1s core-level shifts in eV relative to the C 1s level of the C-C species in acetone for various molecular species. Schematics of each molecule and isosurfaces (0.11 e Å 3 2 ) of the most strongly bound C 1s level are shown as insets. H, C, N, and O atoms are depicted as white, gray, blue, and red spheres, respectively. For acetone, simulated C 1s core-level shifts for the experimental geometry (---) and XPS C 1s measurements ( ∎ ) are also provided.
adsorbed on a pentagon adjacent to a monovacancy, and an O 2 molecule adsorbed on a semiconducting (10,0) and metallic (6,6) SWCNT are shown in Fig. S4. For both types of SWCNT, we find the C=O species of a carbonyl group and C-O species are at too weak binding energy to give rise to the experimentally observed feature. However, it should be noted that the presence of carboxyl and epoxide groups in the experimental spectra cannot be ruled out as their signals may appear in the region between the main and ketene peaks.  Figure S4. Simulated C 1s spectra for defective semiconducting (10,0) (top) and metallic (6,6) (bottom) SWCNTs with adsorbed molecular oxygen (orange), epoxide (blue and green), and carbonyl (red) groups. Schematics of each system and isosurfaces (0.11 e Å 3 2 ) of the four most strongly bound C 1s levels are shown as insets. C and O atoms are depicted as gray and red spheres, respectively.

Calculated Geometries and Total Energies
All geometries were optimized within the local density approximation (LDA) S9 until a maximum force less than 0.05 eV/Å was obtained. The total energy for the relaxed structure is given in Hartrees. All coordinates (provided in xyz format below) and cell dimensions are given in Å. For the acetone molecule, we also provide the experimental geometry and its corresponding LDA energy in Table S8.       12.976722 11.850618 11.252406 C 12.960192 11.793675 9.865016 C 12.208698 12.779135 9.135638 C 12.523917 13.151997 11.789493 C 11.157788 13.339562 11.265154 C 11.113446 13.341983 9.864008 C 9.882596 13.457912 9.190187 C 10.018458 13.203401