2.2. Microstructure and Dynamics of the ILs
In order to understand the microscopic properties of the newly synthesized ILs, we performed force field molecular dynamics (MD) simulations of the liquids at 80 °C. For details on the simulations, see
Section 3.2. It should be noted that the computations are not based on any results measured in this study; the force field parameters are either taken from the literature or obtained from ab initio calculations.
It is well known that hydrogen bonds play a significant role in imidazolium- and triazolium-based ILs [
29,
30,
31,
32,
33,
34,
35,
38].
Figure 3 presents the radial distribution functions (RDFs) between several hydrogen bond donors and acceptors in the investigated systems (for the atom labels, see
Figure 1). The RDFs are normalized to uniform density: a value below one indicates a depletion of the observed atom type at the given distance, while values above one stand for an enrichment. Hydrogen bonds are characterized by a distinct peak at O⋯H distances of around 200 to 300 pm. In the upper left panel of
Figure 3, the RDFs corresponding to the hydrogen bonds between the imidazolium and triazolium ring protons and the carboxylate oxygen atoms of the anions are depicted. It is clearly visible that all four ILs possess a distinct peak at typical hydrogen bond distances here, and we therefore can deduce that hydrogen bonding between these atoms is strong and important in the liquids. The intensity of the hydrogen bond is stronger for triazolium cations when compared to imidazolium and is slightly stronger for benzoate anions in comparison to salicylate. While the former observation can be understood from the fact that there are less ring protons in triazolium, and therefore each single interaction is more intense, the latter finding is due to the competition for the hydrogen bond acceptors because of intramolecular hydrogen bonding in salicylate. The competition between hydrogen bond donors and acceptors will be further discussed in
Section 2.4. The position of the peak is around 230 pm for all systems, as is expected for aromatic C–H⋯O hydrogen bonds.
The upper-right panel of
Figure 3 shows the RDFs for the hydrogen bond between the cations’ ring protons and the hydroxyl oxygen atom in the salicylate anion. There is a small peak visible at a distance of around 230 pm. The height of the peak is barely above one, which indicates a rather weak hydrogen bond between these atoms. The RDFs in the lower two panels of the Figure correspond to hydrogen bonds originating from the hydroxyl proton in salicylate as a donor. In both cases, there is no peak visible at typical hydrogen bond distances, and we can conclude that these types of hydrogen bonds are non-existent in our simulations. This is due to the fact that the hydroxyl proton forms an intramolecular hydrogen bond to one of the carboxylate oxygen atoms in salicylate, which is very strong and exists almost all the time [
48,
49]. Therefore, this proton is not available for intermolecular hydrogen bonds.
After having discussed the structure of the hydrogen bonds in the ILs, we will now investigate their dynamics. A reliable and well-established way to compute average hydrogen bond lifetimes is the autocorrelation approach [
50,
51] as implemented in the TRAVIS program package [
52,
53]. This approach is based on a geometric criterion to define if a hydrogen bond exists at a given time. We use a simple O⋯H distance criterion of 300 pm here. The resulting intermittent hydrogen bond lifetimes are given in
Table 2. For the strong hydrogen bond between the cations’ ring protons and the carboxylate oxygen atoms in the anions (second column), we find lifetimes in the order of 1000 ps, which is rather long when compared to other results from simulation studies of ILs [
31,
32,
35]. This indicates a permanent and only slowly varying hydrogen bond network between the anions and cations of the IL, which is in line with the experimentally observed high viscosities (see
Table 1). We find the hydrogen bonds from triazolium to live longer than those originating from imidazolium, while the hydrogen bonds to benzoate live longer than those to salicylate. These observations are very well in line with the hydrogen bond intensities in the RDFs discussed above.
The hydrogen bond between the cations’ ring protons and the hydroxyl oxygen atom in salicylate, which was already found to be rather weak above, possesses lifetimes of around 200 ps (third column of the table), which is around a factor of five smaller than in the previous case. For completeness, we also computed the average lifetimes of the hydrogen bonds originating from the hydroxyl proton in salicylate, which were found to be almost non-existent above (last two columns of the table). They are all in a range of 20 to 50 ps, and therefore, as expected, significantly shorter lived than the previously discussed hydrogen bonds.
Another interesting property that is directly accessible from MD simulations is the self-diffusion coefficient of the system’s ingredients. These are computed from the slope of the mean squared displacement of the ions’ centers of mass. The results for the anions and cations in the four studied ILs are shown in
Table 3. We find that the anions possess a slightly larger self-diffusion coefficient than the cations, as was observed in simulation studies of imidazolium-based ILs before [
35]. Imidazolium cations diffuse faster than triazolium cations (in agreement with the longer hydrogen bond lifetimes for the latter ones), while benzoate anions diffuse faster than salicylate anions. It should be noted that the values are all rather low when compared to molecular liquids: water possesses a self-diffusion coefficient of around 6600
m
s
at the same temperature [
54]. This is in line with the experimentally observed high viscosities of the ILs (see above).
Finally, we turn our focus to
–
stacking of the aromatic ions. This effect has already been discussed much in the literature for liquids containing aromatic rings [
55,
56,
57,
58]. Despite the repulsive interaction due to the positive charge, imidazolium cations have been found to adopt configurations with
–
stacking in ILs. In our case, both the anions and cations are aromatic, so a very strong stacking of these rings can be expected (supported by the attractive Coulomb interaction). To investigate this claim, we computed spatial distribution functions (SDFs) with the TRAVIS program package [
52,
53]. In
Figure 4, the probability of finding the ring center of an anion is presented as an isosurface in a local coordinate system fixed at the cation, averaged over all cations and the complete trajectory length. As can be seen from the plots, there is a very distinct localization of the anions’ ring centroids on top and below the cation’s aromatic ring. This confirms the presumption of very strong
–
stacking between the anions and cations in the four ILs. For the imidazolium-based cations, there is a slight offshoot of the distribution towards the isolated ring proton, while for the triazolium-based cations, the distribution of the anions’ ring centroids is symmetrically located above and below the triazolium ring. The distribution is slightly wider (and therefore less localized) for salicylate anions when compared to benzoate.
2.3. Cellulose Solubility
To visualize the dissolution of cellulose, we dissolved microcrystalline cellulose in [EMTr][OBz] at
T = 60 °C and monitored the process by polarized optical microscopy (see
Figure 5a–f). It is well visible how the cellulose crystals dissolve over time. The right-hand panel (g) of
Figure 5 depicts the measured
C NMR spectrum of the obtained solution after three weeks at
T = 80 °C. The spectrum shows all characteristic NMR resonances of cellulose, indicating that the polymer strands remain intact in solution, and no significant amount of derivatization or decomposition occurs even after a relatively long time.
The measured cellulose solubility of the four ILs is presented in
Table 4. For [EMTr][OBz] and [EMTr][OSc], solubility limits of 8.5 wt.-% and 4.8 wt.-% were obtained at
T = 80 °C, respectively. In the case of [EMIm][OBz] and [EMIm][OSc], the spread is slightly larger, and the solubility limits were found to be 7.4 wt.-% and 2.9 wt.-% of cellulose at the same temperature, respectively. All measurements were performed with 200 mg of the corresponding IL.
Based on our force field MD simulations, we predicted the solvation enthalpy
of cellulose in the four ILs, normalized to one glucose unit of the cellulose strand; see Column 4 in
Table 4. The solvation enthalpy is the energy that is released when one single strand of cellulose is transferred from vacuum into the IL, normalized to one glucose unit of the cellulose strand. It does not include the energy required to break up the cellulose crystal structure; however, this value is a constant for cellulose, and we are only interested in the differences in solvation enthalpy here. Details on the protocol used for the computation can be found in
Section 3.2. It is already visible that there exists a common trend in experimental cellulose solubility and predicted solvation enthalpy. While [EMIm][OBz] possesses both the highest cellulose solubility and the largest solvation enthalpy, [EMIm][OSc] is found to have both the smallest cellulose solubility and the smallest solvation enthalpy, with the two triazolium-based ILs found between these two cases. In Columns 5 and 6 of
Table 4, the contributions of the cellulose–anion and cellulose–cation interactions to the total potential energy are given. We find that the cellulose–anion interaction is significantly stronger than the cellulose–cation interaction, as has already been discussed before in the literature [
38,
59]. Therefore, the choice of the anion is of primary importance for the cellulose solubility of an IL. The details of the interactions between cellulose and the ILs will be discussed in
Section 2.4. Note that the sum of the cellulose–anion and cellulose–cation interaction energies does not equal the solvation enthalpy, because the latter additionally contains the (positive) energetic contribution to create a void of suitable shape for the cellulose strand in the solvent, which involves the breaking of many hydrogen bonds in the IL.
To further investigate the correlation between experimental cellulose solubility and computationally predicted solvation enthalpy, we present a correlation plot for these two properties in
Figure 6 (note the logarithmic horizontal axis for the cellulose solubility). While the four red points correspond to the ILs from this study, the seven black points have been taken from our previous article [
38]. Already at first sight, a clear correlation between the quantities can be seen. A least-squares regression of the points is shown by the blue curve, and the corresponding equation is given in the upper-right corner of the plot (note that a logarithmic function appears as a straight line with a logarithmic horizontal axis). The correlation coefficient of the regression is
, indicating a rather good fit. The two light blue lines indicate the range of a factor of two deviation between predicted and experimental cellulose solubility. All points are located within this range. Therefore, we can conclude that our simple model can give a semi-quantitative prediction of cellulose solubility from molecular dynamics simulations, with a deviation between prediction and actual value less than a factor of two in all cases.
The reason for this correlation can be understood from the following consideration, which has already been briefly discussed in our previous article [
38]. Dissolving cellulose in some solvent is an equilibrium process with equilibrium constant
K. For such processes, it holds
. Furthermore, we have
. For solvents with a similar molecular structure (as is the case for the ILs investigated here), it can be approximated that
is constant for all the solvents, which we will call
here. Therefore, we find the approximate relationship
with another constant
. This is exactly the relationship that is obtained from the regression above.
2.4. Microscopic Picture of Cellulose Solvation
To obtain a microscopic understanding of the interactions that cause the differences in cellulose solubility, we further analyze the trajectories obtained from our MD simulations of cellulose in the respective ILs in the following section. As already mentioned in the Introduction, a non-derivatizing cellulose solvent needs to break up the hydrogen bonds between the cellulose strands by forming strong hydrogen bonds to cellulose. In other words, dissolving cellulose is all about competition between different hydrogen bond donors and acceptors in the system. For a cellulose solution in [EMIm][OSc] or [EMTr][OSc], there are three different hydrogen bond donors—H(Ring), H(Hyd), and H(Cel)—and three different acceptors—O(Car), O(Hyd), and O(Cel)—present (see
Figure 1 for atom labels). In the case of the two benzoate-based ILs, there are only two donors and acceptors, because H(Hyd) and O(Hyd) are missing. It is important to understand the topology of the hydrogen bond network in these systems, which is characterized by the competition between the donor and acceptor sites.
To do so, we performed our recently developed hydrogen bond network topology analysis [
53], which is based on Sankey diagrams [
60]. The results are presented in
Figure 7. Each panel corresponds to one of the four ILs investigated here. The hydrogen bond donors are placed on the left-hand side of each diagram and the acceptors on the right-hand site. The numbers depict the average count of hydrogen bonds formed by the donor/acceptor atom of the specified kind. The width of the bars is proportional to the corresponding numbers. Please note that the average numbers count the hydrogen bonds per atom, not per molecule. For example, an average number of 0.83 hydrogen bonds for the [EMIm]
ring protons correspond to
hydrogen bonds on average donated by each [EMIm]
ion. The different width of the bars on the left-hand and right-hand side in each diagram is a consequence of the different molecule count.
First, we will discuss the two diagrams on the left-hand side, corresponding to the two ILs with the benzoate anion. A significant amount of hydrogen bonding takes place between the cations’ ring protons and the anions’ carboxylate oxygen atoms, as already observed above for the pure ILs. In both [EMIm][OBz] and [EMTr][OBz], each carboxylate oxygen atom accepts around 1.2 hydrogen bonds on average. However, in the former liquid, each ring proton of the cation donates around 0.8 hydrogen bonds, while this number is around 1.2 in the latter system. This is a simple consequence of [EMIm]
having one more ring proton, as
. Concerning the cellulose, slightly less than half of the donated and accepted hydrogen bonds are intramolecular (as the simulation contains only one cellulose strand, no intermolecular cellulose–cellulose hydrogen bonds can exist). These are the hydrogen bonds shown in blue color in the top panel of
Figure 1. The other half of the hydrogen bonds formed by the cellulose donors and acceptors are formed to anions’ carboxylate oxygen atoms and cations’ ring protons, respectively. In the case of [EMTr][OBz], the cellulose hydrogen bond donors form significantly more hydrogen bonds in total, because more of the carboxylate oxygen atoms are free, as the cation only possesses two instead of three ring protons, which could occupy the oxygen atoms.
When considering the two ILs based on salicylate anions on the right-hand side of
Figure 7, a similar picture emerges. Again, a significant amount of hydrogen bonding takes place between the cation’ ring protons and the anions’ carboxylate oxygen atoms. The hydroxyl proton of salicylate is donating only negligible amounts of intermolecular hydrogen bonds (0.06 and 0.08 on average), because it is occupied virtually all the time by a strong intramolecular hydrogen bond to one of the carboxylate oxygen atoms. This intramolecular hydrogen bond is deliberately excluded from the diagrams, because it would shadow the intermolecular hydrogen bonds of this proton. The hydroxyl oxygen atom in salicylate, on the other hand, accepts more hydrogen bonds on average, which are mostly donated by the cations’ ring protons. For the cellulose as well, a similar picture emerges. Around half of the hydrogen bonds donated and accepted by cellulose are intramolecular, while the other half is formed with anions’ carboxylate oxygen atoms and cations’ ring protons, respectively. Only a tiny amount of hydrogen bonding between cellulose and the hydroxyl group of salicylate is observed.
From this thorough investigation of the hydrogen bond network topology, we conclude that only the cations’ ring protons and the anions’ carboxylate oxygen atoms form relevant amounts of hydrogen bonds to cellulose, while the hydroxyl group of salicylate is negligible here. In a next step, we would like to find out if there are differences in the hydrogen bond strength for the different oxygen and hydrogen atoms in cellulose. To do so, we computed the relevant RDFs between the different sites in cellulose and the hydrogen bond donors and acceptors in the anions and cations. The results are presented in
Figure 8. For the atom labels, please see
Figure 1.
As a first observation, we note that the hydrogen bonds donated by cellulose and accepted by the anions’ carboxylate oxygen atoms—shown in blue and orange—are much shorter on average when compared to those donated by the cations’ ring protons and accepted by cellulose (peak position in the RDF of 200 pm compared to 260 pm). This is expected, because O–H⋯O hydrogen bonds are known to be significantly shorter than aromatic C–H⋯O hydrogen bonds, as the proton is much more polar in the former case. Furthermore, the height of the peak, and therefore the strength of the hydrogen bond, is much larger in the former case. All three hydrogen bond donor sites of cellulose behave similarly, with H exhibiting slightly weaker hydrogen bonds on average. The hydrogen bonds donated by cellulose are generally stronger in the benzoate-based ILs (left-hand side) when compared to the salicylate-based liquids (right-hand side), which is presumably due to the fact that the hydrogen bond acceptor abilities of salicylate are weakened due to the strong intramolecular hydrogen bond.
For the oxygen atoms of cellulose, larger differences are visible. The hydrogen bond acceptors O, O, and O form relatively strong hydrogen bonds to the cations’ ring protons, which are stronger for the triazolium-based ILs (lower two panels) when compared to the imidazolium-based liquids (upper two panels). The two acetal oxygen atoms O and O, on the other hand, form significantly weaker hydrogen bonds to the cations’ ring protons, which is most probably due to the steric hindrance in the cellulose strand. Here also, the hydrogen bonds to [EMIm] are weaker than those to [EMTr].
After having discussed the structure of the individual hydrogen bonds of cellulose, the focus will now be on their dynamics and lifetimes. For the sake of clarity, we grouped together the donor and acceptor sites of cellulose that exhibit a similar behavior in
Figure 8, as already indicated by identical coloring in that figure. To compute the average lifetimes of hydrogen bonds, the autocorrelation formalism was applied, as already described in
Section 2.2. Again, a simple geometric distance criterion of 300 pm was used to define hydrogen bonds. The resulting intermittent lifetimes are given in
Table 5.
In the table, several trends in the lifetimes can be observed. First, the hydrogen bonds donated by cellulose and accepted by the anions’ carboxylate oxygen atoms have significantly longer lifetimes when compared to those donated by the cations’ ring protons and accepted by cellulose. This was already expected due to the larger intensity and shorter average distances of the former group of hydrogen bonds in
Figure 8. Among the cellulose hydroxyl protons, H
and H
form longer lived hydrogen bonds than H
. In the case of [EMIm][OBz], the average lifetime of the former hydrogen bond even exceeds 10 ns, which is an exceptionally long-living hydrogen bond (when, e.g., compared to those in common pure ILs [
17,
31,
32,
35,
58]). It should be noted that such large lifetimes possess a very large statistical uncertainty, because the total production run of the simulation only amounted to 20 ns. The hydrogen bonds between cellulose and benzoate live longer on average than those between cellulose and salicylate, which is in line with the slightly lower peak height for the latter case in
Figure 8. When considering the hydrogen bonds from the cations’ ring protons to the oxygen atoms of cellulose, the average lifetimes are significantly shorter, similar to those observed for the pure ILs above (see
Table 2). Apart from small statistical deviations, the hydrogen bonds accepted by cellulose live longer in the triazolium-based ILs when compared to the imidazolium-based ILs, which is in line with the results in
Figure 8. The lifetimes of the hydrogen bonds accepted by O
, O
, O
, and O
are similar, while those accepted by O
live slightly shorter. This is an interesting finding, as it is contrary to the structural results in
Figure 8, where O
was found to behave very similar to O
and O
, while O
and O
were found to form significantly weaker hydrogen bonds. We consider this another example of the importance of investigating hydrogen bond structure and dynamics separately.
Finally, we want to take a closer look at the spatial distribution of the cations’ hydrogen bond donor sites and the anions’ hydrogen bond acceptor sites around cellulose. To do so, SDFs of these sites were computed in the local coordinate system of one glucose unit of the cellulose strand, averaged over all units in the strand. The results are presented in
Figure 9, where each panel corresponds to one of the four ILs studied here. The red isosurfaces depict the distribution of the cations’ ring protons around cellulose, while the green isosurfaces indicate the distribution of the anions’ carboxylate oxygen atoms. It is visible that the localization of the ring protons is much more distinct in the case of the triazolium-based ILs, which is in line with the higher hydrogen bond intensity observed in these systems above. When comparing the benzoate-based and salicylate-based systems, as well as the distribution of the carboxylate oxygen atoms, there are no significant differences visible in the spatial distribution.