8162 Macromolecules 2010, 43, 8162–8169

DOI: 10.1021/ma1004925

Polymer Dynamics in Nanochannels of Porous Silicon:

A Neutron Spin Echo Study
Andr
e Kusmin,*,† Simon Gruener,‡ Anke Henschel,‡ Nicolas de Souza,§ J€urgen Allgaier,†
Dieter Richter,† and Patrick Huber*,‡
†
Institut f€
ur Festk€
orperforschung, Forschungszentrum J€ ulich, Germany, ‡Experimental Physics,
ulich, 52425 J€
Saarland University, 66041 Saarbr€ ucken, Germany, and §J€
ulich Centre for Neutron Science, c/o TU M€
unchen,
Lichtenberg Strasse 1, 85747 Garching, Germany
Received March 3, 2010; Revised Manuscript Received May 19, 2010

ABSTRACT: Neutron spin echo spectra of poly(ethylene oxide) (PEO) melts confined in porous silicon
(the mean pore diameter: 13 nm) were recorded at Q = 0.05, 0.08, and 0.11 Å-1 and successfully analyzed in
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

terms of a two-state model, where chains adsorbed to the pore walls exhibit much slower internal dynamics
than in the bulk and their centers-of-mass do not move, while free chains have bulklike internal dynamics and
diffuse within an infinite cylinder in the center of the pore. The radius of this cylinder was found to be 1.4 nm
for PEO 3 kg/mol (3k); this corresponds to the thickness of the adsorbed layer to be approximately equal to the
Downloaded via BOGAZICI UNIV on July 17, 2022 at 15:38:55 (UTC).

Flory radius. As opposed to PEO 10k, for which details on the center of mass diffusion could not be discerned,
for PEO 3k the diffusion rate along the pore was found to be smaller than that in the radial direction.

Introduction capillary force and the hydrodynamic velocity boundary condi-

When a molecular fluid is confined in a porous medium with a tion (slip length) in the nanochannels have to be made. Thus the
pore diameter comparable to the size of its molecules, intuitively, conclusions with respect to the microscopic chain dynamics are
one expects that the diffusion rate or the diffusion mechanism will rather indirect, given the sensitivity of capillary filling dynamics
differ from that in the bulk. In the extreme case, when pores are so on those quantities.6,7
narrow that molecules cannot pass each other, single file diffusion The dynamics of linear polymer chains in the bulk melt has
takes place.1,2 In the intermediate case, when the pore diameter is been extensively studied.8 For short, unentangled chains (and for
several times larger than the molecule, the situation is unclear. In the chains of any length at short times) the dynamics is well
a simulation study3 it was found for the diffusion of water that the described by the Rouse model.9 Here the chains interact solely by
diffusion rate along the pore decreases with decreasing pore local friction with a heat bath representing the neighboring
diameter while the diffusion rate in the radial direction did not chains. For long chains, the entanglement of a chain within the
change and was equal to that in the bulk state. For the smallest matrix of surrounding chains leads to a change in the dynamical
pore diameter studied (1.5 nm) the diffusion along the pore was behavior at intermediate and long times. In the reptation model
≈30% slower. To the best of our knowledge, no experimental the effect of the entanglement is described by a virtual tube that
proof of this kind of anisotropy has been found so far. localizes a given chain and limits its motion to a one-dimensional
The confinement of a complex fluid can be expected to affect Rouse motion inside the tube (local reptation) and a slow diffusive
both the center-of-mass (CM) diffusion and the internal molec- creep motion out of the tube (reptation).10 Neutron spin echo
ular dynamics. For example, an NMR study indicated signifi- (NSE) experiments on bulk polymer melts confirmed the repta-
cantly different dynamics of linear polymer chains confined in meso- tion model and resulted in tube diameter values of about 5 nm.11
pores with diameters ranging from 9 to 58 nm, when compared to While there have been neutron scattering studies on dynamics
the bulk state.4 Specifically, the dynamics of confined chains was of alkanes in confinement (e.g., zeolites12 and mesoporous silica13),
found to be restricted to a snakelike motion (reptation) along the we know of only a few such studies on the dynamics of confined
chain contour within a tube having a diameter of about 0.5 nm, polymer melts.14,15 Here we present an NSE experiment on an
a value that does not depend on the pore diameter D even for unentangled and an entangled melt of polyethylene oxide (PEO)
diameters of D=15
RF, where RF is the Flory radius (rms end- with a molecular weight of 3 kg/mol and ≈10 kg/mol, respec-
to-end distance of a polymer chain in the bulk). Only for much tively, confined in porous silicon with a pore diameter of ≈13 nm
larger pore sizes is a crossover from confined to bulk behavior (a melt becomes entangled above the critical molecular weight
observed. This finding, termed the “corset effect”, is proposed to (Mc), which is 5.9 kg/mol for PEO16). As opposed to zeolites or
arise from system properties like the chain uncrossability, low silica glasses, in porous silicon the pores run strictly perpendicular
compressibility, and impenetrability of the pore walls. A modified to the wafer surface. This allowed us to selectively probe molec-
chain dynamics has also been inferred from an experiment on ular motions along the pore and in the radial direction, and thus
capillary filling dynamics of nanochannels, and thus from rheo- to test for a dynamical anisotropy.
logical properties of confined polymers.5 Note, however, that
in those experiments assumptions regarding both the driving Theory
Basic Model. For a polymer melt, the Flory radius is RF =
*To whom correspondence should be addressed. E-mail: A.K., a.kusmin@ lN1/2, where N is the number of Kuhn (i.e., freely jointed)
fz-juelich.de; P.H., [email protected]. segments in the chain and l is the Kuhn segment length.
pubs.acs.org/Macromolecules Published on Web 09/16/2010 r 2010 American Chemical Society
Article Macromolecules, Vol. 43, No. 19, 2010 8163

the adsorbed state is longer than the observation time of the

experiment; therefore, the exchange between the states has
no effect on measured spectra. For the CM diffusion along
the pore we used the model of the long-range translational
diffusion21
SCM ðQ ;tÞ ¼ expð - DTR Q 2 tÞ ð2Þ

)
)
Figure 1. For PEO 3 kg/mol, free chains (thick lines) diffuse freely in For the CM diffusion in the direction perpendicular to the
the center of the pore. For PEO 10 kg/mol, free chains are entangled.
PEO chains that are close to the walls are adsorbed. pore axis we used the model of diffusion inside a circle of a
radius a22
Taking one segment to consist of one monomer and using
l2 =33.75 Å2,17 RF are 48 and 85 Å for PEO 3k and PEO 10k, X
¥
SCM ðQ^ ;tÞ ¼ B00 ðQ^ Þ þ B0m ðQ^ Þ exp½ - ðx0m Þ2 DTR^ t=a2 
respectively. The pore diameter is ≈13 nm; therefore, it is m¼1
reasonable to assume that (i) confinement does not substan-
tially perturb the chain conformation (as compared to the X
¥;¥
þ2 Bnm ðQ^ Þ exp½ - ðxnm Þ2 DTR^ t=a2  ð3Þ
bulk melt); (ii) the fraction of the chains having a contact to
n ¼ 1;m ¼ 0
the pore walls is not negligible. As the pore surface is covered
with a silica (SiO2) layer, and it is known that PEO adsorbs " #2
on silica through hydrogen bonds between silanol groups 4ðxnm Þ2 Q^ aJn þ 1 ðQ^ aÞ - nJn ðQ^ aÞ
and ether oxygens on the polymer,18 we expect that all PEO Bnm ðQ^ Þ ¼
chains near the wall exhibit a slower dynamics than the chains ðxnm Þ2 - n2 ðQ^ aÞ2 - ðxnm Þ2
in the bulk.  
2J1 ðQ^ aÞ 2
From the above the following basic physical model of our B00 ðQ^ Þ ¼ ð4Þ
samples can be constructed (see Figure 1 for an illustration). Q^ a
For PEO 3k, there is a substantial fraction of free chains; the
center of mass (CM) of these chains will diffuse as in the bulk where Jn are the cylindrical Bessel functions of the first kind
melt, but only within a narrow cylinder in the center of and xnm is the (m þ 1)th root of (d/dx)[Jn(x)] = 0.
the pore. On the contrary, most PEO 10k chains have some The diffusion coefficients along and perpendicularly to the
pore axis, DTR and DTR^, are expected to be equal and given

)
contact to the pore surface, those that are not are likely to be
entangled and will diffuse primarily along the pore. by the Rouse model:
We assume that the dynamics of free chains follows the DRouse ¼ kB T=Nξ0 ð5Þ
Rouse model. The dynamics of adsorbed chains is difficult to
model because an adsorbed polymer layer is a mixture of where ξ0 is the monomeric friction coefficient. The expres-
trains, loops, and tails.19,20 Since we do not know relative sion for the (presumably isotropic) internal chain dynamics
contributions of these three structural motifs, two further of free chains, Sfree(Q,t), in the frame of the Rouse model
assumptions will be made: (i) the CM of an adsorbed chain reads8
does not move, and (ii) the internal dynamics of such a chain X 
1
can be either neglected entirely or described by the Rouse model S Rouse ðQ;tÞ ¼ exp - Q2 ji - jjl 2
but with a friction coefficient that is larger than in the bulk melt. i;j
6
The choice of the Rouse model for both free and adsorbed     #
chains does not influence the principal results (see Discussion). 2Q2 Nl 2 X 1 pπj pπi 2
- cos cos ½1 - expð- tp =τR Þ
Neutron Spin Echo. As for any other liquid, the coherent 3π2 p ¼ 1 p2 N N
neutron scattering from a polymer melt is low at small Q
values (Q e 0.2 Å-1) because of the absence of a long-range ð6Þ
order. Therefore, at small Q, the scattering by a mixture of
where i and j are indices for the segments. The summation
deuterated and nondeuterated polymer chains consists almost
over p goes over all internal Rouse modes of the polymer
exclusively of the single chain coherent scattering arising
chain. The Rouse time, τR = ξ0N2l2/3π2kBT, is the longest
due to a large difference (a contrast) between the coherent
relaxation time of the chain with N segments. The internal
scattering lengths (b) of H and D nuclei. The intensity of this
dynamics of adsorbed chains was either neglected altogether
scattering can be calculated by considering every hydrogen
(by taking Sads(Q,t)/Sads(Q) = 1) or accounted for by eq 6,
atom in a chain to have the scattering length ΔbHD; (ΔbHD)2 =
with ξ0 either being fixed to the ξ0 value of free chains
Æb2æ - Æbæ2 =F(1 - F)(bH - bD)2, where Æ æ denotes an average
or being an independent free fitting parameter. We realize
over all chains and F is the fraction of deuterated chains. Thus,
that both scenarios are unrealistic: adsorbed chains are not
provided the elastic scattering due to the Si matrix is negligible
immobile, and their dynamics can not, in general, be modeled
(see the next section), our NSE spectra represent the dynamic
using eq 6. However, the structure of an adsorbed layer is
coherent structure factor of a polymer chain (S(Q,t)) normal-
complicated and our approach, although simple, was suffi-
ized to its static coherent structure factor (S(Q)).
cient to satisfactorily describe our NSE spectra.
A general form of the S(Q,t)/S(Q) expression that corre-
Problem of Elastic Scattering. At low Q, the coherent
sponds to the basic model described above reads
scattering from Si wafers filled with a mixture of deuterated
SðQ;tÞ Sfree ðQ;tÞ Sads ðQ;tÞ and nondeuterated polymer chains is a sum of the elastic
¼ fSCM ðQ ;tÞ SCM ðQ^ ;tÞ þ ð1 - f Þ
)

SðQÞ Sfree ðQÞ Sads ðQÞ scattering arising due to a nonuniform distribution of the
ð1Þ coherent scattering length density (SLD), and the (quasielastic)
coherent single chain scattering due to the scattering contrast
where f is the fraction of free chains and Q and Q^ are the Q between H and D nuclei. Thus, a prerequisite for the analysis
)

components directed along and perpendicularly to the pore of neutron diffraction and NSE spectra in terms of the struc-
axis, respectively. We assume that the average time spent in ture and dynamics of polymer chains is either a negligible
8164 Macromolecules, Vol. 43, No. 19, 2010 Kusmin et al.

contribution of the elastic scattering or the correction for this be done by fitting eq 9 to the IExp(Q) curves and comparing
contribution before (or in the course of) the analysis. inc (Q) - Binc(Q)) values with measured ones.
fitted (IExp
The coherent elastic scattering from Si wafers can be pres-
ented as (eq 45 in ref 23) Experimental Section
The polymers hPEO (protonated) and dPEO (deuterated) were
IðQÞ ¼ ðΔrÞ2 I0 ðQÞ þ ΔrI01 ðQÞ þ I1 ðQÞ ð7Þ
prepared by anionic polymerization. Deuterated ethylene oxide
where ΔF is the scattering contrast, i.e., the difference between was purchased from Campro Scientific (deuteration degree
the average SLDs of the Si matrix and the pore content. [This 99%). The triethylene glycol partially metalated with potassium
was used as an initiator, as a result, dPEO chains had a non-
can be shown by writing the SLD of a wafer as F(r)=(ΔF)Fc(r)
deuterated (-C2H4O-)3 fragment in the middle. The molar
þ F1(r) and performing the Fourier transformation of F(r)F*(r). masses (Mn [g/mol]) and polydispersities (Mw/Mn) were respec-
Fc(r) is 1 if r points to a volume element inside of a pore and tively 2940 and 1.02 for hPEO 3k; 3170 and 1.02 for dPEO 3k;
0 otherwise.] If the matrix had no inhomogeneities and pores 9290 and 1.01 for hPEO 10k; and 9560 and 1.01 for dPEO 10k.
were fully filled by a homogeneous medium, the total scattering The porous silicon membranes are produced in-house employ-
would be just (ΔF)2I0(Q). In reality, the terms I01(Q) and I1(Q) ing an electrochemical anodic etching process of Si (100) wafer
are non-negligible due to a nonuniform distribution of the SLD (thickness ∼0.5 mm). The pores are linear, noninterconnected
in the matrix as well as in the pore content; the greater the size of and oriented along the Æ100æ Si crystallographic direction
inhomogeneities, the smaller are Q values at which I01(Q) and (perpendicular to the membrane surface) with a length of ∼0.3
I1(Q) begin to substantially contribute to the total scattering. mm. After a special treatment with hydrofluoric acid and hydro-
Inhomogeneities may have different sources (e.g., the SiO2 gen peroxide the pores have a roughly circular cross-section with
layer on the surface of the pores) and may lead to a significant mean diameter ∼13 nm25 and their walls are covered with silica
elastic scattering contribution even when the contrast is rendering them highly hydrophilic. The porosity of the porous
matched (i.e., when ΔF = 0). layer is about 0.55.
Because of inhomogeneities and a never-perfect contrast The wafers with a polymer powder spread on the top were
matching, one has to make sure that the residual elastic placed on a hot plate and left there for half an hour (PEO 3k) and
scattering is negligible in the Q region of interest. This can be >12 h (PEO 10k). We determined the mass of the polymer taken
done by fitting a theoretical expression for the coherent scatter- up by the wafers; the porosity calculated from this mass indicates
ing by a single chain, SChain(Q), to the experimental coherent that the pores were (nearly) fully filled. As the filling procedure
scattering, IExp(Q). If the fit quality is good and fitted values of was done under the nitrogen atmosphere, we do not expect any
significant degradation of the polymer.
structural parameters are correct/reasonable, any residual elastic
Our PEO samples had 27% dPEO (by mass); such a composi-
scattering with a Q dependence different from that of SChain(Q) tion was chosen to match the coherent scattering length density
is negligible. However, to exclude the possibility that there is (SLD) of the polymer melt to that of the silicon matrix, and thus to
a substantial scattering contribution due to the residual suppress the elastic scattering (SLD of pure Si is 2.07 [1010 cm-2]).
elastic scattering having a SChain(Q)-like Q dependence, The experiment was performed at J-NSE, the NSE instrument
IExp(Q) curves have to be transformed to an absolute scale. at FRM II (Munich, Germany), at two neutron wavelengths (λ),
An alternative approach, an “internal” normalization,24 8 and 12.8 Å, covering the Fourier time range from 0.1 to 30 ns
consists of the normalization of IExp(Q) to the spin-incoherent and from 45 to 110 ns, respectively. All measurements were done
scattering intensity (IExp
inc (Q)). Specifically, one can write for two sample orientations per every scattering angle (θ) and,
thus, per every Q value (Q is the neutron wave vector transfer,
I Exp ðQÞ ¼ nChain S Chain ðQÞFSc þ BðQÞ ð8aÞ Q=4π/λ sin(θ/2)). The experimental setup and the sample orien-
tations are shown in Figure 2.
Exp
Iinc ðQÞ ¼ nChain Sinc
Chain
ðQÞFSc þ Binc ðQÞ ð8bÞ We recorded NSE spectra at Q = 0.05, 0.08, and 0.11 Å-1 for
the six wafers filled with PEO 3k at 150 and 180 C and the six
where nChain is the chain number density, SChain
inc is the wafers with PEO 10k at 100 C. The S(Q,t)/S(Q) data were
theoretical spin-incoherent single chain scattering, FSc is a corrected for resolution and the cryostat scattering using the NSE
scaling factor converting calculated intensities into the spectra of the activated-carbon powder and the cryostat recorded
experimental units, and B(Q) and Binc(Q) are the coherent for the identical instrument setup. (For PEO 3k, at 180 C and
and incoherent scattering from Si wafers, respectively, including Q = 0.11 Å-1 the measurement was done only with λ = 8 Å.)
any residual elastic scattering. From eqs 8a and 8b we get We also recorded diffraction intensities with π flipper turned
off and on, I v (Q) and I V (Q), respectively, for two sets of empty Si
S Chain ðQÞ Exp wafers, and for the same sets after they were filled with polymer.
I Exp ðQÞ ¼ Chain
ðIinc ðQÞ - Binc ðQÞÞ þ BðQÞ ð9Þ The Q range was from 0.05 to 0.12 Å-1. From I v (Q) and I V (Q)
Sinc
we calculated the coherent scattering, IExp(Q), and the spin-
If Binc(Q) and B(Q) are negligible, eq 9 has no free parameters incoherent scattering, IExp
inc (Q) (details on the calculation can be
because both IExp(Q) and IExpinc (Q) can be obtained from a
found elsewhere24). I v (Q) and I V (Q) for the empty cryostat were
polarized neutron diffraction experiment, and SChain(Q) and not measured, and IExp(Q) and IExp inc (Q) were not corrected for the

SChain are given by scattering contributions due to the cryostat.

inc
To account for the contribution of the spin-incoherent scatter-
SChain ðQÞ ¼ N 2 n2 ðΔbHD Þ2 fDebye ðQ2 Rg 2 Þ ing (primarily due to H nuclei) to the NSE spectra (in our case,
this contribution is less than 10% of the total polymer scattering
2 -x
where fDebye ðxÞ ¼ ðe - 1 þ xÞ ð10aÞ for Q < 0.12 Å-1), the S(Q,t)/S(Q) corresponding to the internal
x2 chain dynamics was calculated as
Chain
Sinc ðQÞ ¼ Nn Æσ inc æ=4π ð10bÞ SðQ;tÞ n2 ðΔbHD Þ2 S Rouse ðQ;tÞ - ð1=3ÞðnÆσ inc æ=4πÞSInc
Rouse
ðQ;tÞ
¼
SðQÞ S Chain ðQÞ - ð1=3ÞSInc ðQÞ
Chain

where n is the number of H atoms per segment, Æσincæ is the ð11Þ

incoherent cross-section of H atoms (average over all
Chain
chains), and Rg is the radius of gyration (for a freely jointed where S (Q) and SChain
Inc (Q) are given by eqs 10a and eq 10b,
chain, Rg2 = Nl2/6). A test whether Binc(Q) is negligible can respectively, and SRouse
Inc (Q,t) is calculated from eq 6 but taking
Article Macromolecules, Vol. 43, No. 19, 2010 8165

only the terms for i=j. The factor -1/3 accounts for the change of A satisfactory agreement between the fitted and measured
the polarization of scattered neutrons in case of the spin-incoherent coherent scattering for Q g 0.08 Å-1 indicates that in this
scattering. Note that for t=0, and at large N, SRouse(Q,t) is equal Q region a residual elastic scattering contribution with a
to SChain(Q) given by eq 10a, except for the factor n2(ΔbHD)2 (this Q dependence significantly different from that for the scat-
factor was omitted in eq 6 because it is canceled out by S(Q,t)/ tering by polymer chains is negligible. In the Supporting
S(Q) when the spin-incoherent scattering is negligible). Information we show that a residual elastic scattering con-
When eq 9 was fitted to IExp(Q) curves, the calculated intensity tribution can be neglected altogether.
was also multiplied by the first-order transmission factor, H1(R,θ), NSE Results. All Q=0.05 Å-1 spectra were excluded from
calculated for the model of an infinite plain slab,26 and by the the data analysis because of a very slow S(Q,t)/S(Q) decay
factor accounting for the neutrons missing the sample in case of and significant uncertainties in the data points. As seen from
45 orientation (Scorr).
the analysis of the diffraction data, this slow decay may to
some extent be due to a residual elastic scattering.
Results As shown in Figure 2, at 0 orientation Q is parallel to the
Checking Elastic Scattering Contribution. To test whether wafer surface. Hence, S(Q,t)/S(Q) obtained for this orienta-
the elastic scattering contribution to our NSE spectra is tion reflect the molecular motions in the plane perpendicular
negligible we fitted eq 9 to the coherent scattering from to the pore. At 45 orientation the molecular motions in the
PEO 3k- and PEO 10k-filled wafers. B(Q) (which contains direction along the pore are probed as well. We assumed that
the residual elastic scattering) was fixed at zero, and the only the internal chain dynamics is, on average, isotropic; there-
free fitting parameter was the experimental spin-incoherent fore, Q^ and Q were only used to calculate the contribution

)
scattering from polymer chains, that is, IExp inc (Q) - Binc(Q). due to the CM dynamics.
Here Binc(Q) is the spin-incoherent scattering contribution The following four models were tested: (1) no confine-
from the cryostat and empty Si wafers; it was not negligible ment, free diffusion and Rouse dynamics for all chains; (2)
compared to the spin-incoherent scattering from PEO alone. CM diffusion confined to an infinite cylinder (of a radius
The fit range was Q g 0.08 Å-1. [While the cryostat scatter- RD), Rouse dynamics for all chains; (3) no confinement, free
ing accounts for about 15% of IExp(Q) at Q = 0.05 Å-1, its diffusion and Rouse dynamics for the free chains (f is the
contribution decreases to e5% for Q g 0.08 Å-1.] fraction of the free chains), the CM of an adsorbed chain
As seen from Figure 3, at low Q the relative difference being immobile; (4) CM diffusion confined to an infinite
between the fitted curves and IExp(Q) is smaller for 45 than cylinder, Rouse dynamics for the free chains, the CM of an
for 0 for both PEO 3k and PEO 10k. This could be, to some adsorbed chain being immobile. For models 3 and 4, the
extent, due to a residual elastic scattering that is more intense internal dynamics of adsorbed chains was either neglected or
at 0 orientation. Note that while we assumed that polymer described by the Rouse model. In the latter case the ξ0 value
scattering is isotropic, the fitted curves for 0 and 45 for adsorbed chains was either kept equal to that for free
orientations differ from each other; the difference is nearly chains or was an independent fitting parameter.
Q independent and given by the factor H1(R,θ)Scorr. For both PEO 3k and PEO 10k, whether we fitted the 0
orientation spectra alone, or together with the spectra for the
45 orientation, model 4 provided the best results. Other
models led to a significantly worse fit quality (judging by the
χ2 value and the goodness-of-fit criterion). When we fitted
model 4 to the PEO 3k spectra for both 0 and 45 orienta-
tion simultaneously, the fit quality improved substantially
when the diffusion coefficient along the pore, DTR , was

)
allowed to be a free parameter; its fitted value turned out to
be smaller than the diffusion coefficient in the radial direc-
tion, DTR^ (which is given by the Rouse model, eq 5). For
PEO 10k, making DTR a free fitting parameter did not lead
)

to a significant improvement of the fit quality; therefore, we

continued to assume that DTR = DTR^ = DRouse.
)

Figure 2. Left: top view of the experimental setup. For every scattering
angle θ the sample (with the pores drawn as a series of white cylinders)
When we fitted model 4, the best fit quality was obtained
was rotated by an angle R to adjust the orientation of the neutron wave with ξ0 for the free and adsorbed chains being two separate
vector transfer Q with respect to the pores. Right: sample orientations free fitting parameters; a slightly worse quality resulted when
employed in the experiment. the internal dynamics of adsorbed chains was neglected

Figure 3. Results of the fit of eq 9 to the experimental coherent scattering with the residual elastic scattering contribution set to zero. The fit range was
Q g 0.08 Å-1.
8166 Macromolecules, Vol. 43, No. 19, 2010 Kusmin et al.

Table 1. Results of Fitting of Model 4 to the NSE Spectra of PEO 3k and PEO 10ka
sample T (C) orientation RD ξ (10-20 kg/ns) f DTR /DTR^

)
PEO 3k 150 only 0 14.1 ( 1.9 0.292 ( 0.05 0.59 ( 0.07 1
0 and 45 14.0 ( 2.1 0.270 ( 0.05 0.581 ( 0.08 0.123 ( 0.05
180 only 0 12.2 ( 4.2 0.34 ( 0.18 0.53 ( 0.19 1
0 and 45 13.0 ( 2.9 0.162 ( 0.05 0.41 ( 0.09 0.29 ( 0.10
PEO 10k 100 only 0 2.5 ( 10 0.77 ( 0.9 0.43 ( 0.2 1
0 and 45 0.4 ( 8.4 0.44 ( 0.16 0.34 ( 0.07 1
a
Adsorbed chains were assumed to be completely immobile. f is the fraction of free chains that have a monomeric friction coefficient ξ0; the centers-
of-mass of these chains diffuse inside an infinite cylinder of a radius RD.

Figure 4. NSE spectra (open symbols) and the fit (lines) of model 4. The spectra for Q = 0.08 and 0.11 Å-1 for both 0 and 45 were fitted
simultaneously; the internal dynamics of adsorbed chains was entirely neglected. ξ0 values from these fits were used to calculate the NSE spectra for the
bulk melt using model 1 (closed symbols).

altogether, and significantly worse fits were obtained when a more complicated model to account for this dynamics is
one ξ0 was used for all, free and adsorbed, chains. not justified in our case because we do not know the structure
For the case when adsorbed chains were assumed to be of the adsorbed polymer layer and our data set is not exten-
completely immobile, the fitted values are given in Table 1, sive enough to allow a definite test of such a model.
the spectra together with the fit curves are shown in Figure 4. Compared to the fits of model 4 where the internal
The free fitting parameters were RD, ξ0, and f. For PEO 3k dynamics of adsorbed chains was neglected, the fits with
when both orientations were fitted simultaneously, DTR / the friction coefficient for adsorbed chains being a free fitting
)

DTR^ was a free parameter as well. For the case when ξ0 for parameter provided similar results but only a marginally
adsorbed chains was a free fitting parameter, the fit results better fit quality, and while the resulting ξ0 values were about
are not shown because they are similar to the ones presented 10 times higher than in the bulk, they had large uncertainties.
in Table 1 and Figure 4; fitted ξ0 values for adsorbed chains Apparently, the information content of our data is not
were more than 10 times larger than the ξ0 values for free sufficient to allow one more free fitting parameter, which is
chains. The fit results for the case when ξ0 for adsorbed coupled to the parameters that we already have. Thus, in the
chains was the same as for free chains are substantially worse following, by referring to model 4, we mean model 4 together
and therefore not shown as well. with an assumption that adsorbed chains are completely
immobile.
Discussion From Figure 5 it is clear that experimental S(Q,t)/S(Q) for
General Features. In the models we fitted to our NSE 0 and 45 orientations are rather similar to each other. This,
spectra, the internal dynamics of adsorbed chains was either in principle, can be expected for short times: the molecule
neglected entirely or considered using the Rouse model for does not “feel” the confinement, hence, the scattering does
bulk polymer melts. Both approaches are oversimplifications: not depend on the Q orientation. But for long times, assum-
adsorbed chains have some internal dynamics that certainly ing that the diffusion rate along and perpendicular to the
is different from that in the bulk. However, an attempt to use pore is the same, and that the internal chain dynamics is
Article Macromolecules, Vol. 43, No. 19, 2010 8167

Figure 5. Comparison of NSE spectra recorded for 0 and 45 orientations. The insets contain magnified portions of respective graphs. Note that some
data points shown here are not shown in Figure 4: they were influenced by magnetic disturbance and therefore excluded from fitting.

isotropic, an S(Q,t)/S(Q) curve for the 45 orientation must volume fraction φ(z) of the segments belonging to adsorbed
monotonously decay to a lower value than is the case for the chains is a monotonously decaying (faster than linearly)
0 orientation. The reason for this can be seen from eqs 1-4: function of z.29 Thus, the values of all fitted parameters are
while SCM(Q ,t) decays to zero, SCM(Q^,t) decays to the B00(Q^) reasonable and compatible with the basic model constructed
)

value, which is the elastic incoherent structure factor (EISF). in the Theory section.
Clearly, nearly identical S(Q,t)/S(Q) curves for two orienta- While model 4 provided a good description of the NSE
tions mean that either the diffusion is isotropic, but very slow, spectra for 0 orientation, simultaneous fitting of this model
so the observation time of the experiment is not sufficient for to the spectra for both 0 and 45 orientations suggested a
SCM(Q^,t) to reach its limiting value, or the diffusion along the slower diffusion rate along the pore as compared to that in
pore is slower than the diffusion in the radial direction. As we the radial direction; see Table 1. A similar effect was obser-
will see below, it appears that the first possibility is realized for ved earlier in a molecular dynamics simulation;30 a simple
PEO 10k and the second for PEO 3k. explanation is as follows. For a molecule to move along the
The initial S(Q,t)/S(Q) decay for the confined PEO is pore, it has to push aside its neighbors in the radial direction.
slower than in the bulk (see Figure 4) probably due a sub- To move in the radial direction, it has to push aside its
stantial fraction of adsorbed chains having a much slower neighbors along the pore. Because of the walls, there is a
dynamics than free chains. (At short times the dynamics free resistance to the movement of the neighbors in the radial,
chains should not be influenced by confinement.) but not in the axial direction. Thus, while the diffusion rate in
The maximal thickness of an adsorbed polymer mono- the radial direction is essentially the same as in the bulk state,
layer is limited by the Flory radius of a chain. Different the diffusion rate in the axial direction is slower.
experimental methods probe an adsorbed layer in a different In principle, an exchange between the adsorbed and free
way, hence, measured thicknesses vary from one method to states could lead to an effective diffusion rate slower than
another.19 For example, in the case of the silica particles in the bulk. However, this does not appear to be the case because
dispersed in aqueous solutions of PEO, the thickness of an such an exchange (i) is probably slow and cannot be observed
adsorbed PEO layer was found to be ≈RF by viscometry in the time range of our experiment and (ii) would affect the
while values systematically lower than RF were obtained by diffusion rate in both the radial and axial directions.
dynamic light scattering.27 If the pore radius (D) is known for Assuming that diffusion along the pore and in the radial
chains such that RF < D/2, the thickness of an adsorbed direction are not coupled (see eq 1), NSE spectra for 0
layer can be estimated from RD values found by fitting model orientation are independent of the mechanism of CM diffu-
4 to the NSE spectra; this thickness is just D/2 - RD. sion along the pore. A satisfactory description of these
PEO 3k. The fitted RD values for PEO 3k result in the spectra by model 4 suggests that there is no “corset effect”
thickness of the adsorbed layer equal to 51 Å (D/2 - RD: for PEO 3k or, at the very least, the tube diameter is ≈15 Å,
65 Å - 14 Å), a value comparable to RF = 48 Å. The fitted and not 5 Å, as was reported for PEO 1.7 kg/mol confined in
ξ0 values are reasonable, too: for comparison, for the bulk pores having a diameter of 9 nm.4
PEO 2.1 kg/mol at 140 C ξ0 =0.265
10-20 kg/ns.28 On the To check whether the above results depend on the assump-
other hand, if we estimate the f value as the ratio πRD2/πD2 tion that DTR^ is given by the Rouse model (i.e., eq 5), we
the result is much smaller than the f values given in Table 1. tried to fit models 1-4 to NSE spectra of PEO 3k with DTR^
However, this estimate is wrong because RD is the radius of being a free fitting parameter independent of the friction
the cylinder where the center-of-mass of a free chain, and not coefficient. In fits to the spectra for 0 orientation alone,
the whole free chain, is confined. While in Figure 1 only small when the fit quality was good, the resulting DTR^ were
parts of free chains are outside this cylinder, it is reasonable similar to the value calculated from ξ0 for the bulk PEO.
to expect that fragments of free chains occupy a substantial In fits to the spectra for both 0 and 45 orientation, when
fraction of the adsorbed layer. Consequently, to estimate f DTR^ was a ξ0-independent free parameter, the fit quality
one must use improved significantly when we allowed for DTR 6¼ DTR^.
)

πRD 2 þ πð1 - xÞðD2 - RD 2 Þ Thus, an assumption that DTR^ follows the Rouse model is
f ¼ ð12Þ reasonable, useful, but not necessary.
πD2
PEO 10k. Intuitively, model 4 is not suitable for PEO 10k.
where x is the volume fraction of the adsorbed layer occupied Indeed, even if there are chains that do not have a contact to
by adsorbed chains. For D=65 Å, RD =14 Å, and f=0.59, x the pore wall, these chains are probably entangled, and the
is 0.43. This is a plausible value because it is known that the CM diffusion does not follow the Rouse model. However,
8168 Macromolecules, Vol. 43, No. 19, 2010 Kusmin et al.

Figure 6. NSE spectra calculated for bulk PEO melts (H:D = 73:27) using model 1 (i.e., the Rouse model) with and without the contribution due to the
CM diffusion (filled and open symbols, respectively). DTR is given by eq 5. The spin-incoherent scattering was taken into account.

model 4 satisfactorily describes our PEO 10k spectra (Figure 4), assumption that the Rouse model can be used for both free
and fitted parameter values are reasonable (for comparison, and adsorbed chains.
for the bulk PEO 24 kg/mol at 102 C ξ0 =0.67
10-20 kg/ns17).
For PEO 10k, S(Q,t)/S(Q) spectra depend on the internal Conclusion
chain dynamics to a greater extent than for PEO 3k (see To study the large scale polymer dynamics under confinement
Figure 6). Since the actual diffusion coefficient of PEO 10k is and to test whether this dynamics is isotropic, the NSE spectra of
smaller than the value predicted by the Rouse model (because porous silicon wafers filled with PEO 3 kg/mol and PEO 10 kg/mol
of entanglements), spectra of confined PEO 10k in our time were recorded at two sample orientations. The spectra were
range will depend almost entirely on the internal dynamics. analyzed in the frame of a two-state model where free chains
Thus, model 4 is not “too wrong” in our case because the undergo Rouse dynamics and diffuse within an infinite cylinder in
details of the CM diffusion do not matter anyway, this is the center of the pore and adsorbed chains have immobile center-
reflected in the weak sensitivity of the fit quality to RD; see of-mass and substantially slower (relative to free chains) internal
Table 1. Because of the slow CM diffusion, longer Fourier dynamics. For both PEO 3k and PEO 10k the internal dynamics
times (than for PEO 3k) have to be reached before a diffu- of free chains was apparently unaffected by the confinement, as
sional anisotropy is detectable for PEO 10k. evidenced by the friction coefficient similar to the values reported
Internal Dynamics of Adsorbed Chains. Neglecting the for the bulk PEO melts. For PEO 3k, the radius of the infinite
internal dynamics of adsorbed chains entirely led to only cylinder was found to be 1.4 nm; this corresponds to the thickness
slightly worse fits than in the case when a friction coefficient of an adsorbed layer approximately equal to the Flory radius.
for adsorbed chains was a free fitting parameter. Clearly, this Also for PEO 3k, the diffusion rate along the pore was found to
neglect is not a bad approximation for our samples and the Q be slower than in the radial direction. This anisotropy may be due
region, at least not for PEO 3k. This can be seen from the to an effect predicted by an earlier study:3 for a molecule to move
comparison of the total S(Q,t)/S(Q) decay with a decay due along the pore (or in the radial direction) neighboring molecules
to the internal dynamics alone (Figure 6). must make space by moving in the radial direction (or along the
For the bulk PEO 3k, even for the long times t . τR, the pore); because of the walls, there is a resistance to their movement
S(Q,t)/S(Q) due to the internal dynamics alone is 0.75 at Q= in the radial, but not in the axial, direction. For PEO 10 kg/mol
0.11 Å-1. Given that a fraction of adsorbed chains is 0.4, and the details on the CM diffusion could not be discerned.
assuming that the internal dynamics of adsorbed chains is From a more general perspective our results regarding the self-
several times slower than in the bulk, its contribution to the diffusion dynamics of PEO in aligned mesopores highlight a
total S(Q,t)/S(Q) decay may indeed be approximated by partitioning of the pore-confined polymers in two species with
unity (that is, no internal dynamics at all). For the bulk PEO regard to their microscopic dynamics: A component with reduced
10k melt the relative contribution of the internal dynamics is dynamics in the pore wall proximity and a component with an
more substantial, hence, in this case, neglecting the internal unaltered dynamics in the pore center. The shear viscosity and
dynamics of adsorbed chains is a poorer approximation. thus the flow properties of polymers confined in such channels are
An adequate description of the internal dynamics of intimately related to their microscopic dynamics. Therefore, our
adsorbed chains is increasingly more important not only finding is of importance not only for the equilibrium physics of
with increasing molecular weight but also with increasing Q. this system but also for the nonequilibrium transport behavior of
In a neutron scattering study on the local dynamics of PEO polymers across silicon mesopores. A reduced chain dynamics
43 kg/mol (RF ≈ 18 nm) confined in pores having a diameter in the boundary layer, and thus an increased viscosity of the
≈40 nm it was found that the confinement had little effect molecules adjacent to the pore walls (sticky boundary layer), will
on the segmental dynamics for 0.2 Å-1 < Q < 1 Å-1.14 lead to significantly reduced permeabilities of silicon meso- and
However, in the low Q range, as in our case, it is the large nanopores.6,7 For the future it would be particularly interesting to
amplitude internal motions that contribute substantially; investigate how changes in the pore wall/polymer interaction,
these motions may well be slowed down, and consequently, e.g., introduced by a silanization of the pore walls, affect the
neglecting the internal dynamics of adsorbed chains is more findings on PEO dynamics in porous silicon reported here.
justified.
Finally, we showed that at least for PEO 3k the spectra are Acknowledgment. A.K. thanks O. Holderer and M. Monken-
rather insensitive to the details of the internal dynamics and busch for the help with getting familiar with NSE in general and
are governed by the three major effects: geometrical confine- J-NSE in particular. This work has been supported by the German
ment, PEO-surface interactions, and CM diffusion. Thus, Research Foundation (DFG) within the priority program SP 1164,
the conclusions below are qualitatively independent of our Nano- and Microfluidics, Grant No. Hu 850/2.
Article Macromolecules, Vol. 43, No. 19, 2010 8169

Supporting Information Available: Use of eq 9 in checking a (14) Krutyeva, M.; Martin, J.; Arbe, A.; Colmenero, J.; Mijangos, C.;
residual elastic scattering contribution to the NSE spectra and Schneider, G. J.; Unruh, T.; Su, Y. X.; Richter, D. J. Chem. Phys.
the proof that such a contribution can be neglected. This 2009, 131, 174901.
material is available free of charge via the Internet at http:// (15) Martin, J.; Krutyeva, M.; Monkenbusch, M.; Arbe, A.; Allgaier, J.;
Radulescu, A.; Falus, P.; Maiz, J.; Mijangos, C.; Colmenero, J.;
pubs.acs.org. Richter, D. Phys. Rev. Lett. 2010, 104, 197801.
(16) Fetters, L. J.; Lohse, D. J.; Milner, S. T.; Graessley, W. W.
References and Notes Macromolecules 1999, 32, 6847–6851.
(17) Niedzwiedz, K.; Wischnewski, A.; Pyckhout-Hintzen, W.; Allgaier, J.;
(1) Kukla, V.; Kornatowski, J.; Demuth, D.; Gimus, I.; Pfeifer, H.; Richter, D.; Faraone, A. Macromolecules 2008, 41, 4866–4872.
Rees, L. V. C.; Schunk, S.; Unger, K. K.; Karger, J. Science 1996, (18) Iler, R. K. The Chemistry of Silica; Wiley: New York, 1979.
272, 702–704. (19) De Gennes, P. G. Adv. Colloid Interface Sci. 1987, 27, 189–209.
(2) Burada, P. S.; Hanggi, P.; Marchesoni, F.; Schmid, G.; Talkner, P.
(20) Granick, S. Eur. Phys. J. E. 2002, 9, 421–424.
ChemPhysChem. 2009, 10, 45–54.
(21) Bee, M. Quasielastic Neutron Scattering; Bristol: Philadelphia, PA,
(3) Cui, S. T. J. Chem. Phys. 2005, 123, 054706.
1988.
(4) Fatkullin, N.; Fischer, E.; Mattea, C.; Beginn, U.; Kimmich, R.
ChemPhysChem. 2004, 5, 884–894. (22) Dianoux, A. J.; Pineri, M.; Volino, F. Mol. Phys. 1982, 46, 129–137.
(5) Shin, K.; Obukhov, S.; Chen, J. T.; Huh, J.; Hwang, Y.; Mok, S.; (23) Stuhrmann, H. B.; Miller, A. J. Appl. Crystallogr. 1978, 11, 325–345.
Dobriyal, P.; Thiyagarajan, P.; Russell, T. P. Nat. Mater. 2007, 6, (24) Zajac, W.; Gabrys, B. J.; Scharpf, O.; Andersen, K. H.; Parsonage,
961–965. E. E. Solid State Ionics 2002, 147, 213–223.
(6) Dimitrov, D. I.; Milchev, A.; Binder, K. Phys. Rev. Lett. 2007, 99, (25) Kumar, P.; Hofmann, T.; Knorr, K.; Huber, P.; Scheib, P.; Lemmens,
054501. P. J. Appl. Phys. 2008, 103, 024303.
(7) Gruener, S.; Huber, P. Phys. Rev. Lett. 2009, 103, 174501. (26) Sears, V. F. Adv. Phys. 1975, 24, 1–45.
(8) Doi, M.; Edwards, S. F. The Theory of Polymer Dynamics; Clarendon (27) Lafuma, F.; Wong, K.; Cabane, B. J. Colloid Interface Sci. 1991,
Press: Oxford, U.K., 1986. 143, 9–21.
(9) Rouse, P. E. J. Chem. Phys. 1953, 21, 1272–1280. (28) Reference 17, note that ξ0 = 0.265
10-20 kg/ns follows from the
(10) De Gennes, P. G. J. Phys. (Paris) 1981, 42, 735–740. Rouse rate Wl4 = 21 750 Å4/ns mentioned in the text, while the ξ0
(11) Schleger, P.; Farago, B.; Lartigue, C.; Kollmar, A.; Richter, D. value given in Table 3 (0.225) is apparently a misprint.
Phys. Rev. Lett. 1998, 81, 124–127. (29) Fleer, G. J.; Cohen Stuart, M. A.; Vincent, B.; Cosgrove, T.;
(12) Jobic, H.; Farago, B. J. Chem. Phys. 2008, 129, 171102. Scheutjens, J. M. H. M. Polymers at Interfaces; Chapman and Hall:
(13) Baumert, J.; Asmussen, B.; Gutt, C.; Kahn, R. J. Chem. Phys. 2002, London, 1993.
116, 10869–10876. (30) Reference 3, Figure 4.