Ana R.
Brás
*a,
Sebastian
Gooßen
a,
Margarita
Krutyeva
a,
Aurel
Radulescu
b,
Bela
Farago
c,
Jürgen
Allgaier
a,
Wim
Pyckhout-Hintzen
a,
Andreas
Wischnewski
a and
Dieter
Richter
a
aJülich Centre for Neutron Science JCNS (JCNS-1) and Institute for Complex Systems (ICS), Forschungszentrum Jülich GmbH, 52425 Jülich, Germany. E-mail: a.bras@fz-juelich.de; Fax: +49 2461 61-2610; Tel: +49 2461 61-5093
bJülich Centre for Neutron Science JCNS, Forschungszentrum Jülich GmbH, Outstation at FRM II, Lichtenbergstraße 1, 85747 Garching, Germany
cInstitute Laue-Langevin (ILL), F-38042 Grenoble Cedex 9, France
First published on 4th February 2014
We present a neutron scattering analysis of the structure and dynamics of PEO polymer rings with a molecular weight 2.5 times higher than the entanglement mass. The melt structure was found to be more compact than a Gaussian model would suggest. With increasing time the center of mass (c.o.m.) diffusion undergoes a transition from sub-diffusive to diffusive behavior. The transition time agrees well with the decorrelation time predicted by a mode coupling approach. As a novel feature well pronounced non-Gaussian behavior of the c.o.m. diffusion was found that shows surprising analogies to the cage effect known from glassy systems. Finally, the longest wavelength Rouse modes are suppressed possibly as a consequence of an onset of lattice animal features as hypothesized in theoretical approaches.
Therefore one of the remaining important challenges in polymer science is summarized in one question: what happens when polymers have no ends at all? An elaborate study of the chain dynamics within a ring melt is a promising mission to address this fundamental aspect of polymer physics. Beyond that, rings are fundamental structures in biology. Mitochondrial and plasmic DNA are cyclic and often have a knotted structure; therefore ring polymers may serve as models for fundamental bio-physical problems as well.
While theoretical predictions and computer simulation studies are available,5–10 high quality experimental data on the microscopic structure and dynamics of rings are rare. One important reason for this deficiency is the impact of linear contaminants present in the samples.11 Only recently it has been possible to produce highly pure and monodisperse rings in quantities sufficient for elaborate studies.12
In this work we present a small angle neutron scattering (SANS) and neutron spin echo (NSE) study combined with pulsed-field-gradient (PFG)-NMR experiments on a 5 kg mol−1 monodisperse and pure PEO ring (360 bonds per ring) melt. The following results stand out:
(i) The structural data demonstrate that the rings display a non-Gaussian compact structure. (ii) The chain center of mass motion (c.o.m.) crosses from a subdiffusive regime at short times to Fickian diffusion at longer times. The decorrelation time agrees with mode coupling predictions. (iii) In analogy to the cage escape in glassy dynamics, the c.o.m. diffusion displays non-Gaussian behavior, a phenomenon not observed before in polymer diffusion studies. (iv) Finally, long wavelength internal modes are suppressed, an observation that might indicate the onset of lattice animal features.
Size exclusion chromatography (SEC) in THF/N,N-dimethylacetamide (85/15) with PEG calibration yielded molecular weights Mn = 5200 g mol−1 for the d-polymer and Mn = 4980 g mol−1 for the h-polymer, respectively. From nuclear magnetic resonance (1H-NMR) Mn = 5470 g mol−1 for the d-polymer and Mn = 5280 g mol−1 for the h-polymer were obtained. The polydispersity indices Mw/Mn were 1.01 for the hydrogenous PEG and 1.03 for the deuterated PEG, respectively.
Blends of hydrogenous and deuterated polymers (h/d) were prepared in solution and freeze-dried from benzene. The volume fraction of the hydrogenous polymers in h/d blends was 17% in all experiments.
The NMR measurements were carried out using a magnetic resonance analyser Bruker Minispec (mq20) that operates at an 1H frequency of 20 MHz and is equipped with a permanent magnet. Diffusion coefficients were measured using a standard stimulated echo pulsed-field-gradient sequence.14 The echo signal intensity is given by:
(1) |
The diffusion attenuation curves were found to be single-exponential, demonstrating a simple self-diffusion process to be expected for a monodisperse polymer. Additionally, no dependence on the observation time Δ was observed in the interval from 15 to 500 ms. Thus the polymers undergo Fickian diffusion in the melt state (see Fig. 1). The diffusion coefficients were evaluated from the slope of the attenuation curves and were found to be (2.44 ± 0.34) Å2 ns−1 for the ring PEO and (0.87 ± 0.07) Å2 ns−1 for the equivalent linear chain.
Assuming Gaussian statistics, the form factor for the ring polymer in the discrete representation reads
(2) |
The finite length of the considered chains and its implication for the asymptotic Q dependencies are herewith considered explicitly. The ring closure is taken into account by F2ν with F2ν =(1 − k/N)2ν, k being the segment number (1 < k < N), ν the Flory excluded volume exponent and l the segment length. Various suggestions for F2ν have been recently proposed.15 All approaches practically yield equally good descriptions of the data with variations in ν between 0.43 and 0.45. This result already indicates a significant deviation from Gaussian statistics of these polymers in the bulk, for which ν = 1/2 is expected. We used this empirical modification to describe not only the SANS data but also the structural part of the Rouse model applied for the NSE data (vide infra). The linear chain is retrieved with F = 1 and ν = 1/2.
To account for interchain contributions due to the small differences in the h- and d-chain lengths, together with eqn (2), we employed the random phase approximation, fixing thereby the chain lengths to the chemically determined values. A fit to the ring or linear chain SANS data led to radii of gyration RG,c = (15.64 ± 0.08) Å resp. RG,l = (24.58 ± 0.24) Å. The linear-to-ring ratio of 1.57 significantly exceeds the value of that is predicted for Gaussian rings and signifies an increased compactness of the ring. In the intermediate Q-region the intensity decays with S(Q) ∝ Q−P with P = 2.3 leading to the peak observed in a 2nd moment representation of the intensity (Fig. 2a). The best fit to eqn (2) required ν = (0.44 ± 0.01), i.e. P = 1/ν = 2.3. The data in the intermediate Q range satisfactorily agree within the statistical uncertainty with the presented model, although the experimental peak is slightly more pronounced than predicted by the model. This may be due to an even more enhanced compactness of the ring. This is in best agreement with recent computer simulations predicting a limiting behavior of P = 3 (i.e. ν = 1/3) for high N.5 As shown in Fig. 2b our data approach the predicted trend. According to ref. 5, the investigated chain length is about one order of magnitude below the region where ν = 1/3 should be found and well above the Gaussian regime.
(3) |
Fig. 3 S(Q, t)/S(Q, 0) vs. t for the ring and the linear chain at the lowest Q value (Q = 0.05 Å−1). Error bars are smaller than the symbol size. The lines are a fit to the data with eqn (3). |
The fitting procedure arrived at D = (2.05 ± 0.02) Å2 ns−1 for the (more compact) pure rings, and D = (0.94 ± 0.01) Å2 ns−1 for the linear chains. Note: especially in the case of the linear chain the description of the lowest Q data shown by the solid lines is clearly not perfect – the dynamic structure factor appears to be more stretched than indicated by the theory.
There are two origins for this different behavior: (i) the chains have a molecular weight corresponding to about 2.5 times the entanglement distance. Therefore for the linear chain we expect the onset of topological constraints which will slow down the diffusion coefficient and modify the mode spectrum, while (ii) entanglements should not play a role for the ring polymers. Therefore deviations from the diffusion or internal dynamics as predicted by eqn (3) are observed. In the following we analyze in more detail the effects on the mode spectrum and its contribution to the structure factor.
In fact, the absence of every second mode has an important consequence: at the smallest investigated Q = 0.05 Å−1 the ring S(Q, t) is determined by c.o.m. diffusion only, which is not the case for the linear analog. To clarify this point, a mode contribution analysis using the unmodified or modified Rouse mode spectrum respectively was performed for both systems. The contribution of the different normal modes to the dynamic structure factor depends strongly on the momentum transfer Q. Since S(Q, t) does not simply decompose into a sum of contributions from the different modes, the mode contribution factors were defined as18
(4) |
R p(Q) describes to what extent a mode “p” may relax S(Q, t) in the limit of long times (t → ∞) and under the premise that all other modes are not active. Fig. 4 displays these mode contribution factors for the linear (p = 1 and p = 2) and for the ring chains (p = 2 and p = 4). In the linear case it is obvious that segmental relaxation already leads to a decay of S(Q, t) in the entire Q and time range probed by the NSE experiment. In contrast to the linear analog, for ring polymers we expect only translational diffusion contributions except for Q values higher than 0.1 Å−1. Therefore the faster c.o.m. diffusion of the rings compared to the linear analog, as well as the problem with the fit of the latter at the lowest Q (see Fig. 3), is obviously only due to the onset of entanglements in the case of the linear polymer.
Fig. 4 Contribution factors for the first modes of both the ring and the linear chains to S(Q, t)/S(Q, 0) (see legend in the figure). Note that the first mode for the rings is p = 2. |
The absence of mode contributions at low Q for the ring allows to extract the c.o.m. mean square displacement (m.s.d.) directly from the measured S(Q = 0.05, t). Fig. 5 shows such obtained m.s.d.'s vs. Fourier times for Q = 0.05 Å−1 from the NSE ring data. These can easily be extracted from the scattering function by 〈r2(t)〉 = (−6/Q2)ln{S(Q, t)}. As found for melts of linear chains and as confirmed by computer simulations at short times the m.s.d. displays a sub-diffusive regime followed by a transition to Fickian diffusion (t1) at longer times.19 As presented in ref. 20 this sub-diffusive behaviour can be understood in terms of an inter-chain potential. The influence of this potential becomes negligible as soon as the polymer has travelled a distance comparable to its radius of gyration. This model leads to a decorrelation time by τdecorr = RG2/D. Using the RG as determined by SANS and the “true diffusion” as extracted from the Fickian regime in Fig. 5, we find τdecorr = 125 ns, in very good agreement with the experiment.
The diffusion coefficient as measured by NSE and by PFG-NMR is shown in Fig. 5 as lines. They correspond to the mean squared displacement constructed as, 〈r2(t)〉 = 6Dt, where D is the diffusion coefficient obtained by NSE and PFG-NMR, respectively. The PFG-NMR value is in very good agreement with the NSE result.
With this approach, i.e. by just considering the Gaussian form e[−r2(t)Q2/6] we now calculate the diffusive contribution to S(Q, t) for all Q. We obtain the dashed lines in Fig. 6.
Now, S(Q, t) at Q = 0.05 Å−1 naturally is described perfectly. However, solely based on c.o.m. diffusion the relaxation at intermediate Q is already over-predicted. This over-prediction demonstrates that the Gaussian approximation, i.e. the Q2 dependence of the diffusion relaxation function, appears to be violated – the c.o.m. diffusion displays non-Gaussian behavior. Note that this non-Gaussian c.o.m. diffusion has already been discussed for linear melts.21,22 On the other hand for Q = 0.2 Å−1 the calculated decay under-predicts the experimental data – here obviously segmental relaxation contributes to S(Q, t).
Correcting for the Gaussian approximation of S(Q, t) to a first order the non-Gaussian effects displayed in Fig. 6 may be described in terms of a positive fourth order term in Q:23,24
(5) |
The data points in Fig. 7b show the result of this procedure. In particular at longer times α(t) is low, however at intermediate times values of 0.2 to 0.3 are reached, a clear indication of non-Gaussian center of mass diffusion of the rings.
Fig. 7 Dynamic structure factor for the 5 kg mol−1 PEO ring melt at different Q-values. (a) Coloured lines depict the Rouse model fit taking into account the subdiffusivity and including the non-Gaussianity parameter function α(t). Long Rouse modes are suppressed (see text). (b) The plot shows the non-Gaussianity parameter α(t) from eqn (5) (see text) compared with the α(t) obtained from the fit to the S(Q, t) data (red full line). |
Finally, an alternative way of extracting α(t) from the S(Q, t) data has been taken. All S(Q, t) data were fitted with a Gaussian shape for α(t) (see later) additionally allowing for mode contributions. However, as already may be suspected from Fig. 4 and other than predicted by theory the fitting procedure evidences that the longer wavelength Rouse modes do not seem to relax the rings. To take this into account we limited the contributing modes by a minimum value pmin that removed long wavelength modes with p < pmin. The result is shown in Fig. 7a. Now the fit describes all data very well, the obtained α(t) compares reasonably well with the values determined for each single time separately (see Fig. 7b). The resulting pmin = 4 reveals that the first allowed Rouse mode (p = 2) for ring topology is suppressed and only modes with wavelengths shorter than half the ring size are allowed. It should be mentioned that allowing mode p = 2 to contribute cannot be compensated by different α(t), i.e. the obtained combination of α(t) and pmin is the only possible way to get a good description of the S(Q, t) data. While for a linear chain the first mode exhibits one knot or the wavelength is twice the full chain length, for the first ring active mode the amplitude pattern is characterized by 4 knots. Lower Fourier components are absent. This pattern, connected with the closed shape of the ring polymer, might be a first indication of a lattice animal formation.
For a better quantitative determination of non-Gaussian behavior, S(Q, t) data over a wider Q-range that are solely influenced by c.o.m. diffusion are needed. This implies – at least if the single chain dynamic structure factor is concerned – shorter chains. An approach to lower Q would not help, since then the Q4 term in eqn (5) becomes negligible. In ref. 12 we investigated 2 kg mol−1 PEO rings. Since the NSE data for that sample were not influenced by segmental modes at all, these small rings are ideal candidates to verify the existence and to explore the shape of the time dependent non-Gaussianity parameter for the c.o.m. diffusion of polymer rings. α(t) values for the 2 kg mol−1 rings are obtained by the same procedure as applied initially to the large rings. The result is displayed in Fig. 8. Note, that here S(Q, t) data from five different Q values were used to determine α(t) over the entire time range. In spite of the still modest statistics we may draw two conclusions from this result: (i) the non-Gaussianity parameter with values between 0.1 and 0.25 is not negligible (typical values in glasses are around 0.4) corroborating the observations made with the 5 kg mol−1 sample and (ii) the peak structure in the α(t) is evident, a phenomenon that has not been observed before for the c.o.m. diffusion in a polymer melt. Note that typical values for the non-Gaussianity parameter in glasses are 0.4 and it was found for the α-relaxation that values below 0.2 are negligible.23,24 This is obviously not the case for the c.o.m. diffusion on polymer ring melts.
Fig. 8 Non-Gaussianity parameter α(t) for PEO rings with Mw = 2 kg mol−1. The line is a guide for the eye. |
Going back to original ideas of non-Gaussianity and to earlier investigations mainly on glassy systems, Fig. 8 reminds on what was found for molecular glasses and referred to as the “cage effect”.21,24 In a glass forming system one molecule is captured in a cage built by the adjacent molecules. There it rattles around, testing possible escape routes from the cage. This leads to an increase of α(t).24 When the molecule eventually escapes the cage confinement the non-Gaussianity parameter goes through a maximum, and for longer times decreases again. Translating this picture to our polymer ring system and having in mind that these rings are very compact objects, the peak time τpeak of α(t) would refer to an escape time of a ring from the cage formed by the adjacent rings. In fact τpeak is roughly 12 ns which is in a surprising agreement with τdecorr = 9 ns published in ref. 12. As mentioned above τdecorr marks the time at which the polymer chain has escaped from an inter-chain potential by traveling roughly the distance of its own size (RG). For the larger ring αmax is found at about 30 ns, a factor of 4 smaller than τdecorr = 125 ns. Seemingly the cage effect is now restricted to portions of the ring, possibly to the pattern indicated by the internal mode structure.
This journal is © The Royal Society of Chemistry 2014 |