Purification, separation and extraction of inner tubes from double-walled carbon nanotubes by tailoring density gradient ultracentrifugation using optical probes.

We studied the effect of varying sonication and centrifugation parameters on double-walled carbon nanotubes (DWCNT) by measuring optical absorption and photoluminescence (PL) of the samples. We found that by using a low sonication intensity before applying density gradient ultracentrifugation (DGU), only inner tube species with a diameter [Formula: see text]0.8 nm can be identified in absorption measurements. This is in stark contrast to the result after sonicating at higher intensities, where also bigger inner tubes can be found. Furthermore, by comparing PL properties of samples centrifugated either with or without a gradient medium, we found that applying DGU greatly enhances the PL intensity, whereas centrifugation at even higher speeds but without a gradient medium results in lower intensities. This can be explained by extraction of inner tubes from their host outer tubes in a two-stage process: the different shearing forces from the sonication treatments result in some DWCNT to be opened, whereas others stay uncut. A subsequent application of DGU leads to the extraction of the inner tubes or not if the host nanotube stayed uncut or no gradient medium was used. This work shows a pathway to avoid this phenomenon to unravel the intrinsic PL from inner tubes of DWCNT.

Introduction

Since its discovery in 2002 by Bachilo et al. [1], photoluminescence (PL) of single-walled carbon nanotubes (SWCNT) has triggered worldwide investigations in this field due to their possible applications, e.g. for usage as biomarkers [2]. In recent years, DWCNT have come into focus, since they exhibit some superior properties compared to SWCNT, especially mechanical and chemical stability. The optical properties of double-walled carbon nanotubes (DWCNT) however, regarding their capability of exhibiting PL, remain rather elusive due to a true cornucopia of different or even (apparently) contradictory works in the past. Due to inevitable deviations during the production process, there’s always the possibility of SWCNT byproducts what is often seen as the (main) cause of PL from DWCNT samples [3]. Furthermore, there is an ongoing debate concerning the basic physical ability of DWCNT to exhibit PL due to possible similarities of PL quenching in DWCNT caused by the small intertube-distance compared to PL quenching in SWCNT bundles due to small inter-nanotube distances [4]. Additionally, Koyama et al. [5] found that the relative intensity of steady-state luminescence from inner walls in DWCNT is about 700 times weaker than that from SWCNT by comparing the PL decay rates of emitters within their samples. Contradictory to that, Yang et al. [6] state that PL is possible from inner tubes, but only if they have the right diameter. On the other hand, Hertel et al. [7] showed evidence that a majority of inner tubes exhibit PL with a slight red-shift in the exciton transition compared to SWCNT. Jung et al. [8] also observe a red-shift in their PL spectra on samples which have been covered with a mussel protein that is known to eliminate the possibility of emission from SWCNT impurities. Also, by rendering the optical response of the outer tube inactive by fluorination, Hayashi et al. [9] conclude, that their PL signals come from inner tubes.

An obvious problem in comparing these different studies lies within the deviating sample preparations. First of all, the nature of the pristine sample itself is of uttermost importance: different synthesis processes such as Chemical Vapor Deposition (CVD) [5,10], the High-Pressure CO conversion process (HiPco) [11] or transformation of C60 [12] or metalorganic compounds such as Ferrocene [13] within an SWCNT to form a DWCNT result in different nanotube properties, be it different diameter and/or length distributions, chirality preferences, number of defects produced etc. Secondly, the importance of the concrete parameters of the solubilization process, i.e. the path to receiving a homogeneous solution of individualized nanotubes with (almost) identical properties as free-standing nanotubes, can not be underestimated; strong sonication is known to cut nanotubes (as seen in the work of Heller et al. [14] for SWNT or the work from Green et al. [15] for DWNT) and therefore having an impact on the sample quality by introducing defects and as for example short nanotubes exhibit different optical properties than long nanotubes [16]. And finally, the purification process is evenly crucial: Ultracentrifugation has been shown to extract the inner tube from the DWCNT [17], what in return means, that the source of a measured near-IR PL is from simple SWCNT [18]. On the contrary, Green et al. showed that DGU can be applied to separate nanotube samples with respect to their wall numbers [15]. From another point of view, by nanomanipulation in a scanning electron microscopy (SEM) system equipped with a cantilever, Zhang et al. have recently successfully shown that inner tubes can be pulled out from their host outer tube, if a directed pulling force is applied [19]. The application of DGU on a tip-sonicated DWCNT sample seems to be a crucial factor in the discussion, whether DWCNT are capable of exhibiting PL or not. Interestingly, most works claiming to observe PL from inner tubes did not apply DGU to the sample ([7-10,12,20-22]) whereas those claiming that PL from inner tubes does not occur or is severely quenched [3,15,17]), applied DGU to their samples. The work of Yang et al. [6] represents a special case, i.e. that the selective observation of PL from inner tubes could be caused by the sample preparation.

In this work, we overcome this problem by studying the differences in the optical properties of DWCNT samples from the same sample batch with respect to varying solubilization techniques. In order to elucidate these differences, samples produced via high vacuum chemical vapour deposition of alcohol (HV-CVD) were either tip sonicated under different conditions but purified with the same technique of DGU or tip sonicated under the same conditions but purified with or without a gradient medium.

Experimental

Five different kinds of samples were used in this study: CVD-DWCNT (same precursor as used by Endo et al. [23] but adapted for HV-CVD [24]) with an average inner tube diameter of 0.8 nm plus commercially available HiPco SWCNT (Unidym) with an average diameter of 1 nm. The carbon source for HV-CVD growth was ethanol, the growth temperature was 875 °C for 10 min. For purification, the DWCNT samples were first exposed to hydrochloric acid to remove remaining catalysts followed by annealing in air flow at 400 °C for 2 h to remove amorphous carbon and after another acid treatment annealed at 500 °C in air for 2 h to remove SWCNT impurities (pristine sample). The effectiveness of the latter step to remove SWNT has been shown by Li-Pook-Than et al. [25], where the intensity of the integrated Raman signal for SWNT is much less than 1%. After this, four different individualization techniques were applied:

Procedure A: The DWCNT sample was individualized via ultrasonication in a 2% w/v sodium deoxycholate (DOC) solution, using a steel tip with a diameter of 1/4 inch, 60 W of power, for 4 h. The sample was subsequently centrifugated at 10,000g for 30 min to remove metallic particles, with a speed low enough to ensure inner tube containment. After that, the supernatant (top 50% of the centrifugated solution) rich with isolated DWCNT was collected for further investigation.

Procedure B: The DWCNT sample was individualized via ultrasonication similar as in Procedure A, but afterwards a first purification treatment by centrifugating at 220,000g for 2 h was applied to remove non-nanotube impurities followed by the main DGU treatment in a gradient medium containing varying concentrations of OptiPrep solution (60% Iodixanol in H2O, Sigma Aldrich) in a 2% w/v sodium dodecyl sulfate (SDS) solution: 40%, 35%, 32,5%, 30% and 27% iodixanol content for 9 h and 240,000g. This is a slightly modified procedure from the method used by Yanagi et al. [26].

Procedure C: The DWCNT sample was individualized via ultrasonication in a 2% w/v sodium deoxycholate (DOC) solution, using a steel tip with a diameter of 1/2 inch, 17 W of power, for 5 h, followed by the same first purification by centrifugation as well as DGU treatment as in Procedure B.

Procedure D: Another part of the same sample batch sonicated in Procedure A was centrifugated at 1,000,000g for 30 min and the top 50% of the centrifugated solution rich with isolated DWCNT was collected afterwards; this is the identical procedure as for the HiPco SWCNT sample for SWCNT control used in this study and as in [27].

After separation of the layers (be it the supernatant (top 50%) from Procedure A and D or the individual layers from B and C), Optical Absorption Spectroscopy (OAS) was performed. For checking the abundance of inner tube species in the pristine sample, photoluminescence spectroscopy was performed using a tuneable Dye-Laser coupled into a Nanolog spectrometer.

Results and discussion

Before analyzing the samples after the DGU treatment, we compare the absorption spectra of the DWCNT sample not treated by DGU and using a low centrifugation speed (Procedure A) with the HiPco-sample that mainly contains SWCNT (for ease of comparison, the spectra were normalized to their optical density at 900 nm): In Fig. 1 we can see that the DWCNT absorption shows only weak features of the excitonic transitions between corresponding van Hove singularities in the density of states of inner tubes with diameters between 0.6 and 1 nm. The reason for this lies within the DWCNT structure: For DWCNT, the space between inner and outer tubes is given by the van der Waals radius which can vary depending on the synthesis procedure [28]. DWCNT produced by similar CVD processes with comparable diameter distributions as in this work (e.g. [29,30]) showed values in between 0.33 and 0.41 nm, this means that the outer tube diameters in our samples are between 1.3 and 1.8 nm. Early works assigned the absorption signal in the interval between 900 and 1200 nm to an overlap of the inner tube E11 and outer tube E22 transition for such a diameter distribution [31,32]. In the work of Iakoubovskii et al. [10], where a similar diameter distribution is used like in this work, the outer walls of DWCNT have been exposed to ozone etching to decompose the absorption spectra of DWCNT to their inner and outer shell contributions. After applying this method it can be seen that the E11 transition wavelengths of the inner tubes are in the same interval (between 900 and 1200 nm) as the E22 transition wavelengths of the outer tubes. Therefore we can safely assume that the small size of the peaks in the absorption spectrum of our samples is caused by the same mentioned overlap of inner tube and outer tube contributions.

Fig. 1

Absorption spectra of the pristine DWCNT sample not treated by DGU and using a low centrifugation speed (Procedure A) and the HiPco SWCNT control. Spectra normalized to their optical density at 900 nm, with the green graph multiplied by 2 for better differentiation. (A colour version of this figure can be viewed online.)

For checking the abundance of different species in the DWCNT sample, photoluminescence spectroscopy was performed. Fig. 2 shows the line scans of the DWCNT sample from Procedure A at two different excitation wavelengths, namely 569 nm to excite inner tubes with smaller diameter and 660 nm for bigger diameter inner tubes. The PL response shows that the line scans cover all non-zigzag semiconducting nanotube species with a diameter between 0.65 and 1 nm as also seen by Bachilo et al. [33] (PL emission of zigzag tubes could not be resolved which is most likely due to their lower PL quantum yield compared to semiconducting species with bigger chiral angles [34-36]). We also measured the Raman signal of this solution that gave us the same results as Kim et al. obtained while investigating on dispersed DWCNT [22] and additionally, the sample from Procedure A showed a much lower PL intensity when compared to the SWCNT reference sample (not shown). Combined with the result from the optical absorption measurement this highlights that the mild centrifugation from Procedure A is not affecting the diameter-distribution of inner and outer tubes within the sample and we can see no possible extraction of inner tubes.

Fig. 2

PL Line scans at two different excitation wavelengths for the DWCNT sample from Procedure A. Spectra normalized to their optical density at 900 nm. (A colour version of this figure can be viewed online.)

With this background we can analyze the DGU experiments: As it can be seen in Fig. 3a and b, the liquid columns resulting after applying the same DGU procedure to identical samples look very different from each other. This difference becomes even more remarkable when analyzing their optical absorption spectra (Fig. 3c). The green curve shows the absorption spectrum of the sample prepared by Procedure B. The peaks correspond to the well known E22 and E11 transition energies of various nanotubes that can be assigned to the different species with the widely accepted assignement by Bachilo et al. [33]. Basically all (semiconducting) species with a diameter in the same range as the inner tubes seen in Raman measurements from [24] can be seen; the strength and sharpness of these peaks resemble the SWCNT spectrum as in Fig. 1, as well as the spectrum of separated DWCNT reported by Miyata et al. [17]. In this work, this behaviour was associated with the extraction of inner tubes of DWCNT due to the sample preparation by tip-sonication and applying DGU.

Fig. 3

Outcome of the DGU process after different sample treatments for the pristine DWCNT samples by varying sonication parameters. A thicker area at the top with a thin layer containing all extracted inner tubes whilst the part at the bottom seems to be made out of the remaining outer tubes (a). A single, violet layer and a black, seemingly homogeneous distribution of the rest of the material become apparent. The violet layer contains nanotubes with a diameter 0.8 nm, whereas the other, bigger part contains outer tubes selected by their metallicity as well as remaining DWCNT (not shown) (b). Optical absorption spectra of the extracted layers from both procedures after applying the same DGU treatment (c). The shape and size of the peaks from both procedures look similar to the peaks from the SWCNT sample in Fig. 1. The biggest difference however is that some of the bigger inner tubes (dt > 0.8 nm) are missing in Procedure C (marked red) although traces of the (10,2) tubes (dt = 0.88 nm) can be seen. The small hump at 800 nm in the black curve is caused by the detector change during the measurement, whereas the non-marked peaks in the green curve are associated with other big inner tubes or small outer tubes. Spectra normalized to their optical density at 900 nm, with the green graph being offset by a constant factor for better differentiation. (A colour version of this figure can be viewed online.)

In stark contrast to the latter absorption measurement stands the result of the DGU step from Procedure C, where a lower sonication intensity was applied in the solubilization process than in Procedure A and B. As it can already be assumed from the violet layer seen in the bottom left picture in Fig. 3b which is usually a sign of enrichment of only a fraction of nanotubes [37], the investigation of this layer by checking its absorption spectrum confirms this assumption. Reminding the abundance of different inner tube species confirmed by PL measurements (Fig. 2), several different species are not to be seen in this graph (or at least too less abundant to be noticed): (7,5) with a diameter dt = 0.83 nm, (8,4) with dt = 0.84 nm, and (7,6) with dt = 0.89 nm. This gives rise to the question, where in the sample these tubes are: according to the PL measurement from Procedure A in Fig. 2 and since all the procedures have been performed on samples from the same batch, these bigger tubes must be in a lower layer of the DGU column. Surprisingly, these tubes are not present in a less peculiar layer directly underneath the violet one as one would expect from the diameter distribution of the former (ranging from dt = 0.62 nm for the (5,4) tubes to dt =  0.80 nm for the (9,2) species) but they are abundant in the more distant thick black layer underneath. For example the (8,4) species with a very prominent E11 peak at 1124 nm in the green curve in Fig. 3c but without a similar peak in the black curve, can be found in PL measurements by exciting the top of the black layer with their E22 transition wavelength of 596 nm, as it is seen in Fig. 4. Interestingly, these tubes show also a much weaker luminescence intensity, indicating that these species are abundant as DWCNT which are expected to have a much lower PL quantum yield than SWCNT from the layers above (this point is examined in more detail later). On the other hand, smaller diameter inner tubes are barely to be noticed in this lower layer (e.g. the (6,4), (6,5) or (8,3)), which is a sign of extraction of these inner tubes from their host tubes and therefore having a lower buoyant density than the DWCNT in this layer, so that only very few of these species are left in this area. The latter can be caused by insufficient debundling and therefore protection from cutting through the sonication process.

Fig. 4

Line scans of the PL signal from different layers of the sample treated under Procedure B, excited at 596 nm, the E22 transition wavelength of 596 nm of the (8,4) species. This ’missing’ species can be seen, although being considerably weaker than other present species. This can be a hint that in the lower layers of the as-centrifugated sample the nanotubes are abundant as DWCNT, therefore showing PL of inner tubes. Another interesting part of this figure is, that the lower layers show only very weak features of smaller tubes like (6,5), (6,4) or (8,3). Spectra normalized to their optical density at 900 nm. (A colour version of this figure can be viewed online.)

This also applies to other species like the (9,4) dt = 0.91 nm and the (8,6) dt =  0.96 nm tubes with an E22 transition wavelength near 725 nm; this transition can also only be seen for Process B. The other layers from the DGU process that where not mentioned in both procedures contain smaller amounts of the species discussed earlier or bundles of nanotubes that are not showing luminescence. These findings support the argument of inner tube extraction due to the tip-sonication treatment, but it also implies a diameter-dependant threshold for the cutting and therefore opening of the DWCNT what greatly influences the outcome after the centrifugation treatment.

An interesting part in the discussion on DWCNT PL is the emission intensity of the inner tubes. Factors range in the literature from being weaker in comparison to SWCNT by a factor of at least 10.000 [3], or about a factor of 700 weaker [5] up to almost equal signal strength [8]. Tsyboulski et al. [3] compare the emission of two different layers of a DGU column to determine the difference in the PL quantum yield of DWCNT compared to SWCNT. The SWCNT origin is assigned to be only from residual SWCNT that survived the oxidation process and not extracted inner tubes from the centrifugation process. Extraction of inner tubes though can either enrich the number of emitters in a DGU layer (by moving to top layers) or deplete it (seen from a lower layers perspective) so this process can severely alter the abundance of discrete species within different layers. Koyama et al. [5] take a different path by analyzing the PL relaxation times of DWCNT compared to SWCNT. For their calculations, they equalized the relaxation rates of radiative and non-radiative decay processes from SWCNT to their DWCNT counterparts. However, this has to be questioned since the environment for a SWCNT is rather different than for a DWCNT, and environmental effects were shown to be of vital importance for the determination of luminescence decay rates [38,39].

As mentioned in Fig. 4, the PL signals in the black curve with smaller emission wavelengths than the (7,5) species (i.e. with dt < 0.8 nm) are barely distinguishable against the background although being already amplified by a factor of 50. The very faint signals of the (8,3) (dt = 0.78 nm) and the (6,5) (dt = 0.76 nm) species can be understood as signals from tubes that have not been extracted and/or isolated completely. However, in our case, we chose a very mild form of purification for the pristine DWCNT sample to preserve the original DWCNT structure in Procedure A. For the normalization of the optical density that is crucial for a valid comparison between different samples as it gives an estimate on the number of emitters in the sample, we took the value of the optical density at 900 nm. To finally compare the various PL intensities, we took the intensity of the (6,5) species which has a excitation wavelength of 569 nm, because it is one of the most prominent species within the samples as well as it is abundant in both of the marked layers from Procedure B and C as seen in Fig. 3. The difference in the intensity of the (6,5) inner tube species within the pristine DWCNT to these layers from each DGU experiment is approximately a factor of 50, as seen in Fig. 5. It is unambiguous that for the PL process in Procedure A far more nonradiative decay channels after E22 excitation are available than for Procedure B and C.

Fig. 5

Line scans of the PL signal from the different centrifugated samples with an excitation wavelength of 569 nm, the E22 transition wavelength of the (6,5) species (Emission Wavelength 985 nm). Spectra normalized to their optical density at 900 nm. (A colour version of this figure can be viewed online.)

The possible physical processes involved in PL quenching and their eventual impact on the PL intensity are not yet fully understood. The work of Shen et al. [40] gives an overview of different mechanisms that could explain this effect. The overall electronic structure of a DWCNT resulting from the inner and outer tube contributions can be rather complicated and vary heavily for each DWCNT, since it depends on several different factors like interval distance, curvature, coupling strength, commensurability, metallicity of the outer tube, etc.

However, in Fig. 5 it is clearly seen that PL from Procedure A is severely quenched in comparison to PL from Procedure B and C, where DGU was applied. This is a strong indication for the extraction of inner tubes from their outer tube hosts in the DGU process, whereafter quenching mechanisms induced by the presence of the outer tubes are eliminated.

The follow-up question is, whether applying DGU is essential for the extraction process or if inner and outer tubes are already separated before the centrifugation process. It has been mentioned in previous works that sonication alone can lead to extraction of inner tubes from DWCNT due to simultaneous opening of the host nanotube and ’shaking’ of the DWCNT, where the inner tube is extracted since the frictional force between inner and outer tube is insignificant small [17,41]. In that case, centrifugation at high speeds without a gradient medium as in the DGU process should be sufficient to remove remaining DWCNT, bundles, etc. due to their highly different buoyancies, leaving a sample consisting only of SWCNT. These, in return, should show (at least) a similar PL intensity than the extracted tubes from the DGU process. However, this is not the case. Applying the same centrifugation technique from the HiPco SWCNT sample to the DWCNT sample (Procedure D) leads only to a minor amplification of the PL intensity when compared to the pristine DWCNT sample (Procedure A), as seen in Fig. 6.

Fig. 6

Line scans of the PL signal from the different centrifugated samples with an excitation wavelength of 569 nm, comparing the PL intensity from samples being treated by DGU (Procedure B) or not (Procedure A and D). Spectra normalized to their optical density at 900 nm. (A colour version of this figure can be viewed online.)

This can be explained by the lack of a selective gradient surrounding the opened DWCNT in the centrifugation process. The surrounding medium of Procedure A and D is a homogeneous DOC solution, so no density differences that could act as a pulling force to the inner tubes are present, thus leaving either individualized DWCNT in the sample and/or SWCNT that were extracted by chance. However the case is, the resulting PL intensity differs strongly from the DGU processed nanotubes, with at least an order of magnitude difference in the PL intensity of those procedures. These findings, along with the differences seen in the DGU procedures as well as the difference from using a gradient medium or not, suggest, that strong ultrasonication and density-gradient ultracentrifugation are two essential complementary steps in achieving inner tube extraction of DWCNT samples.

Concommitant to previous studies by Miyata et al. [17] we found strong indications for inner tube extraction from DWCNT caused by the aforementioned solubilization techniques. We additionally confine that this process is strongly dependant on the parameters for sonication and centrifugation as well as on the tube chirality and separation of inner and outer tube. Our results indicate that by applying a lower sonication power to the sample prior to the DGU process, bigger DWCNT, that also have bigger inner tubes, are not opened in this process and therefore its inner tubes are not susceptible to extraction in a subsequent application of DGU. This can be explained by a higher mechanical stability of bigger tubes due to their lower curvature compared to smaller tubes. These results in return confine a pathway on how to avoid inner tube extraction in order to study the intrinsic PL response of inner tubes of DWCNT. We found strong evidence for a final proof that inner tubes of DWCNT are capable of exhibiting PL.

Conclusion

Optical absorption as well as photoluminescence spectroscopy were performed on DWCNT samples produced in the same CVD synthesis step. Although being purified by the same DGU technique, the samples treated with different sonication intensities show clearly unsimilar absorption spectra due to the difference in the sonication procedure. While all the inner tube species that have been seen in the PL measurement were also abundant after applying DGU when being sonicated under a stronger input power, some tubes (dt >  0.8 nm) can not be seen if a lower input power has been used. This can be attributed to the well known fact that nanotubes can be cut in the sonication process ([14,15]) and smaller tubes are more susceptible to damage than bigger tubes, implying a diameter-dependant threshold for the cutting (i.e. opening) of the DWCNT. This effect is the basic requirement of the extraction process in the subsequent DGU procedure, as it can be seen by the developement of the abundance of different nanotube species in the optical absorption spectra after applying the same DGU technique to samples treated by different sonication intensities. However, tiny fractions of bigger diameter inner tubes like the (10,2) (dt = 0.88 nm) can be observed, which can be explained by possible defect-assisted cutting of bigger tubes. Furthermore, the difference in the PL intensity of the extracted tubes compared to the pristine DWCNT sample was found to be a factor of 50 stronger; this is associated to the outer tube shielding of the DWCNT lowering the PL quantum yield of the inner tube. In the final graph it has been shown that the factor of the purification technique (DGU or no DGU) is not negligible, since the resulting PL intensities vary by at least one order of magnitude.

For the investigation on inner tube extraction, the details of the solubilization technique with respect to sonication intensity and centrifugation parameters as well as nanotube parameters such as tube type and chirality are of vital importance. Our results show strong indications that inner tube extraction can be avoided and the resulting PL signal originates from the inner tubes of DWCNT.