1: \documentclass[byrevtex,pre,preprint,showpacs]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{amsmath}
4: \bibliographystyle{apsrev}
5: \newcommand{\fig}[1]{Fig.~\ref{#1}}
6: \newcommand{\eq}[1]{Eq.~(\ref{#1})}
7: \begin{document}
8: \title{Transport properties of incipient gels}
9:
10: \author{Sune N\o rh\o j Jespersen}
11: \email[]{sjespers@sfu.ca}
12: \affiliation{Department of physics, Simon Fraser University, Burnaby,
13: British Columbia, Canada V5A 1S6}
14: \author{Michael Plischke}
15: \email[]{plischke@sfu.ca}
16: \affiliation{Department of physics, Simon Fraser University, Burnaby,
17: British Columbia, Canada V5A 1S6}
18:
19: \date{\today}
20:
21: \begin{abstract}
22: We investigate the behavior of the shear viscosity $\eta(p)$ and the
23: mass-dependent diffusion coefficient $D(m,p)$ in the context of a simple
24: model that, as the crosslink density $p$ is increased, undergoes a
25: continuous transition from a fluid to a gel. The shear viscosity
26: diverges at the gel point according to $\eta(p)\sim (p_c-p)^{-s}$ with
27: $s\approx 0.65$. The diffusion constant shows a remarkable dependence
28: on the mass of the clusters: $D(m,p)\sim m^{-0.69}$, not only at $p_c$
29: but well into the liquid phase. We also find that the Stokes-Einstein
30: relation $D\eta\propto k_BT$ breaks down already quite far from the
31: gel point.
32: \end{abstract}
33:
34: \pacs{82.70.Gg, 61.43.Hv, 36.40.Sx}
35: \maketitle
36:
37: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
38: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
39: \section{\label{sec.intro}Introduction}
40: When a system of polyfunctional molecules is crosslinked, the transport
41: properties such as the shear viscosity and the diffusivity can be
42: dramatically affected. In particular, the diffusivity decreases as the
43: number of crosslinks is increased and the shear modulus increases, diverging
44: at the critical point at which a gel is formed. The diffusivity remains
45: finite as the system gels since monomers and small clusters can diffuse
46: through the tenuous structure that characterizes the amorphous solid close to
47: the critical point. Although gels have been studied for many years
48: \cite{adam96}, their critical behavior remains poorly understood. In
49: particular the question of whether or not there exist universality classes
50: into which different materials can be grouped remains largely unanswered.\par
51: In this article, we report on extensive molecular dynamics simulations of a
52: simple model for a gel. We study the system on the fluid side of the gel
53: point from the simple liquid limit into the critical region. We investigate
54: the structural properties of clusters and calculate both the shear viscosity
55: $\eta(p)$ and the mass-dependent diffusion constant $D(m,p)$ as function of
56: the crosslink density $p$. We find that as $p\to p_c$, $\eta(p)\sim
57: (p_c-p)^{-s}$ with $s\approx 0.65$, a value somewhat smaller than that
58: conjectured by de Gennes \cite{deGennes79} on the basis of an analogy with
59: a random superconductor network and also predicted recently by Broderix {\it
60: et. al.} \cite{broderix99} for a Rouse-like model network. The mass-dependent
61: diffusion constant $D(m,p)\sim m^{-0.69}$ for a range of $p$ near the
62: critical point and $3\leq m\leq 50$. This behavior is consistent with earlier
63: results for $p=p_c$ \cite{sj02} and rather close to a prediction
64: \cite{deGennes79} made on the basis of a simple scaling argument. On the other
65: hand, our value for $s$ is somewhat lower than the one found by K\"untzel
66: \textit{ et. al.} in a recent article \cite{kuntzel03} in which $s=0.8$ is
67: found by a theoretical analysis of the Zimm model. The
68: diffusion coefficient $D(m,p)\to {\rm const.}$ as $p\to p_c$ for $m$ at
69: least as large as 10 but displays critical behavior in the next leading term.
70: It is also worth noting that, in contrast to simple liquids, the product
71: $D(p)\eta(p)$ is not a constant but rather reflects the divergence of $\eta$
72: at the gel point.\par
73: The structure of this article is as follows. In section
74: \ref{sec.model} we describe our model and the computational details. Section
75: \ref{sec.static} contains a discussion of the geometric properties of the
76: clusters and the nature of the percolation transition. The shear viscosity
77: calculation and results are described in section \ref{sec.visc} and results
78: for the diffusion constants are found in \ref{sec.dif}. We conclude with a
79: short discussion in section \ref{sec.con}.
80:
81:
82: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
83: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
84:
85:
86: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
87: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
88: \section{\label{sec.model} Model}
89: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
90: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
91: The model is similar to the one employed in \cite{sj02}, but we
92: include the details below for completeness.
93: Our system is composed of $N=L^3$ ($L=10,13,15,20$ and $30$) particles
94: interacting pairwise through the shifted
95: Lennard-Jones potential
96: \begin{equation}
97: U({\mathbf r})=\begin{cases}
98: U_{LJ}(r)-U_{LJ}(2.5\sigma), & r \leq 2.5\sigma\\
99: 0 & \text{otherwise,}\end{cases}
100: \end{equation}
101: where
102: $U_{LJ}(r)=4\epsilon((\sigma/r)^{12}-(\sigma/r)^{6})$. All of our
103: simulations are $3D$ constant energy molecular dynamics (MD) simulations
104: corresponding to an average temperature of $k_BT/\epsilon\approx 1$ and
105: density
106: $\Phi=0.8\sigma^{-3}$. These
107: choices ensure that the system is in the liquid-phase region of the
108: phase-diagram \cite{smit92,felicity97}. We use periodic boundary conditions
109: and a
110: time step of magnitude $dt=0.005\tau$, where
111: $\tau=\sqrt{m\sigma^2/\epsilon}$ is the reduced Lennard-Jones time. From a
112: typical equilibrium state of this liquid we let the particles form a
113: specified number $n$ of permanent chemical bonds if they come closer than
114: $r_c=2^{1/6}\sigma\approx 1.12$, coinciding with the minimum of $U(r)$.
115: The bond interaction is a harmonic oscillator
116: potential
117: $U_{\text{harm}}(r)=1/2 kr^2$: in our simulations we take
118: $k\sigma^2/\epsilon=2.0$ (different from \cite{sj02}). Note that this way
119: of adding bonds violates energy
120: conservation; indeed we actually pump energy into the system when
121: adding bonds. To compensate we cool down the system again after having
122: established the required number of bonds.
123: %Sune: You should reconcile this with the statement
124: %above that the simulations are at a given average energy. Do you cool the
125: %system down after the bonding?} %
126: With
127: this bonding procedure cross linking is very fast --- the average distance
128: between the particles is comparable to $r_c$, so a large number of
129: particles are available for bonding at any given instant.
130: Each particle can bond to a maximum of $f=6$ other
131: particles (excluding itself), and the cross link density $p$ is then given in
132: terms of the number of bonds $n$ as $p=2n/fN$. Any number of particles, if
133: fulfilling
134: the conditions above, can be cross linked per time step, but we halt
135: the bond formation when $p$ reaches a predetermined value.
136:
137: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
138: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
139: \section{\label{sec.static}Geometric properties}
140: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
141: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
142: Before discussing the dynamic properties of this model, we need some basic
143: information about the static properties. In this section we determine the
144: geometrical percolation point $p_c$ as well as the two critical
145: exponents $\nu$, the correlation length exponent, and $\gamma$ the
146: exponent characterising the divergence of the weight average cluster mass
147: (of finite clusters).
148: We follow a similar procedure to the one used and outlined in
149: \cite{sj02,plischke02,vernon01}. In
150: order to find $p_c$ we calculate numerically the fraction $W(L,p)$ of
151: percolating systems of size $L^3$ with a bond density $p$. This
152: function is plotted in \fig{fig.W} for all five system sizes. The
153: crossing points of the different curves seem to coincide, and the
154: corresponding value of $p$ is thus a good estimate of $p_c$ \cite{sj02,ziff02}:
155: From the
156: figure we determine $p_c=0.2565$ as in \cite{sj02}.
157: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
158: \begin{figure}
159: \includegraphics{W}
160: \caption{\label{fig.W} Fraction of systems $W(L,p)$ percolating in the
161: $x$-direction as a function of $p$ and for five different system sizes
162: as indicated on the plot. The lines are guides for the eye,
163: except in the case $L=30$ for which the data are fitted to a stretched
164: exponential \cite{newman01}. We estimate $p_c=0.2565$.}
165: \end{figure}
166: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
167: Finite size scaling theory predicts that $W(L,p)$ does not depend on
168: $L$ and $p$ separately but only on the combination $L/\xi$ (and the
169: sign of $p-p_c$) where
170: $\xi=|p-p_c|^{-\nu}$ is the correlation length
171: and $\nu$ the correlation length exponent \cite{stauffer92}. Thus we
172: may write
173: \begin{equation}
174: \label{sizescale}
175: W(L,p)=f(L^{1/\nu}(p-p_c)),
176: \end{equation}
177: where $f(x)$ is a scaling function.To test this hypothesis we
178: replot the data for $W(L,p)$ from \fig{fig.W} in \fig{fig.Wscaling} as
179: a function of $L^{1/\nu}(p-p_c)$ with $p_c=0.2565$ as determined above and
180: $\nu=0.9$.
181: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
182: \begin{figure}
183: \includegraphics{Wscaling}
184: \caption{\label{fig.Wscaling} Same as \fig{fig.W}, except here
185: $W(L,p)$ is plotted as a function of $L^{1/\nu}(p-p_c)$ with
186: $p_c=0.2565$ and $\nu=0.9$. The data collapse very nicely in agreement
187: with finite size scaling theory.}
188: \end{figure}
189: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
190: The collapse is very good confirming the correctness of the values for
191: $p_c$ and $\nu$.
192:
193: To compare with percolation theory we need one more exponent, and here
194: we consider the behavior of the weight average cluster mass $M_w$. In
195: the thermodynamic limit the expected behavior is $M_w(p)\sim
196: |p-p_c|^{-\gamma}$ \cite{stauffer92}: Therefore we compute $M_w$ as a
197: function of $p$ for different system sizes, and in \fig{fig.mwscal} we plot
198: the results in the form $M_w/L^{\gamma/\nu}$ versus
199: $L^{1/\nu}(p_c-p)$ with $\gamma=1.8$ being the expected $3D$
200: percolation value and $\nu$ and $p_c$ as determined previously. Again
201: there is a very nice data collapse, and we therefore conclude that
202: here as in \cite{sj02,plischke02,vernon01} our system is consistent
203: with the $3D$ percolation universality class in so far as static
204: properties are concerned.
205: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
206: \begin{figure}
207: \includegraphics{mwscal}
208: \caption{\label{fig.mwscal} Scaling plot of the weight average molecular
209: weight $M_w$. The quality of the data collapse confirms $\gamma=1.8$
210: in accordance with the $3D$ percolation value.}
211: \end{figure}
212: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
213:
214: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
215: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
216: \section{\label{sec.visc}Viscosity}
217: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
218: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
219: We measure the shear viscosity $\eta(p)$ by using the appropriate
220: Green-Kubo formula \cite{Allen, Hansen}:
221: \begin{equation}
222: \label{eta}
223: \eta=\frac{1}{V k_B T}\int_0^\infty \, dt \langle
224: \sigma_{xy}(t)\sigma_{xy}(0)\rangle,
225: \end{equation}
226: where $V$ is the volume and $\sigma_{xy}(t)$ the $xy$ component of the
227: stress tensor:
228: \begin{equation}
229: \sigma_{xy}(t)=\sum_{i=1}^N m v_{x,i}v_{y,i}+
230: \sum_{i=1}^N \sum_{j<i}(y_i-y_j)f_{x,ij}.
231: \end{equation}
232: In this equation $f_{x,ij}$ is the $x$ component of the force from
233: particle $j$ on particle $i$, and the meaning of the remaining terms
234: is self explanatory. In the simulations we average over several
235: hundred samples for each $p$ and over three
236: off-diagonal components ($xy,yz$ and $zx$) of the stress tensor to obtain
237: slightly better
238: statistics. It is important to note that we have discarded any sample
239: containing a spanning cluster since for such a system the viscosity is not
240: defined, i.e.\ the right hand side of \eq{eta} diverges. Although we have
241: simulated very long runs (up to
242: $t=750\tau$) the stress correlator $ C_{\sigma\sigma}(t)\equiv \langle
243: \sigma_{xy}(t)\sigma_{xy}(0)\rangle$ has still not decayed completely,
244: and it is necessary to add by hand an additional contribution, in
245: particular for $p$ close to $p_c$. A stretched exponential
246: $C_{\sigma\sigma}(t)= a \exp(-bt^c)$ with $0.1<c<0.3$ seems to fit the data
247: well for
248: long times, and there are also theoretical reasons \cite{broderix01} to
249: believe that this is
250: the appropriate form. See \cite{vernon01} for a thorough discussion of this
251: point.
252:
253: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
254: \begin{figure}
255: \includegraphics{visc}
256: \caption{\label{fig.visc} The dimensionless shear viscosity as a function of
257: $p_c-p$ for
258: different $L$.}
259: \end{figure}
260: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
261: In \fig{fig.visc} we have plotted the resulting values for the
262: viscosity for different systems sizes and at different stages of the
263: cross linking. We note the clear power law behavior outside the critical
264: region, and a fit to the $L=10$ data in this region yields $s=0.65$. The line $\eta\propto
265: (p_c-p)^{-0.65}$ has also been drawn on the plot, and it is apparent that the
266: data are consistent with this exponent. For large $p$, $p>0.23$, there are
267: larger error bars and this will also affect the scaling plot.
268: Since the viscosity diverges at the critical
269: point with an exponent $s>0$, the finite size scaling form is
270: \begin{equation}
271: \eta(p,L)=L^{-s/\nu}g(\xi/L) \quad p<p_c,
272: \end{equation}
273: where $g$ is a scaling function with the limits
274: \begin{equation}
275: g(x)\propto
276: \begin{cases}
277: x^{-s/\nu} & x\to 0\\
278: \text{const.} & x\to \infty,
279: \end{cases}
280: \end{equation}
281: and $\xi\sim(p_c-p)^{-\nu}$ is the
282: correlation length, c.f. Sec.~\ref{sec.static}. Therefore we plot in
283: \fig{fig.visc_scal} $\eta L^{s/\nu}$ versus $L^{1/\nu}(p_c-p)$, and the
284: collapse is quite good outside the critical region with $s=0.65$, whereas
285: there is a larger scattering of the points for $p$ closer to $p_c$.
286:
287: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
288: \begin{figure}
289: \includegraphics{visc_scal}
290: \caption{\label{fig.visc_scal} Same as \fig{fig.visc}, but here plotted in a scaling form with $s=0.65$.}
291: \end{figure}
292: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
293:
294: de Gennes has suggested \cite{deGennes79} a value of $s\approx 0.7$ based
295: on an analogy
296: between gelation and conductance in a random mixture of normal and
297: superconducting elements, and nice agreement with this was found in a
298: related model in \cite{vernon01}. Here we have observed a slightly smaller
299: value for $s$.
300:
301:
302: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
303: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
304: \section{\label{sec.dif}Diffusion}
305: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
306: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
307: In this section we extend our earlier study \cite{sj02} on the diffusion
308: of clusters. Previously we were concerned mainly with the behaviour of
309: the diffusion constant $D(m,p)$ as a function of cluster mass $m$ at
310: the gelation point $p=p_c$. Here we address the $p$
311: dependence of $D$ for different clusters and the validity of the
312: Stokes-Einstein relation $D(p)\propto k_BT/\eta(p)$ for a
313: given cluster mass. We restrict our attention to the $L=20$ system.
314:
315: To determine the diffusion constant we use the Einstein relation:
316: \begin{equation}
317: \label{einstein}
318: \frac{1}{6t} \langle (r_m(t)-r_m(0))^2 \rangle \xrightarrow{t\rightarrow
319: \infty} D(m,p),
320: \end{equation}
321: where $r_m(t)$ is the center-of-mass position of a cluster of mass $m$
322: at time $t$, for a given value of $p$ (for clarity of the presentation we omit
323: the explicit dependence on $p$ in the notation). When calculating the
324: diffusion constant numerically we have averaged over all clusters of a
325: given mass $m$ and over several hundred crosslinkings, and we have
326: discarded any percolating samples. This has been done mainly for
327: consistency when comparing with $\eta$, but in any event we do not
328: expect this to affect the diffusion of any but the very largest clusters.
329:
330: First we examine the convergence of \eq{einstein} by plotting in
331: \fig{fig.tail1} the behavior of $\langle (r_m(t)-r_m(0))^2\rangle /6t$
332: for $m=1$ (monomers) as a function of time and for three different values of
333: $p$.
334: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
335: \begin{figure}
336: \includegraphics{tail1}
337: \caption{\label{fig.tail1} $\langle (r_1(t)-r_1(0))^2\rangle /6t$ a
338: function of time for monomers, and for three different values
339: of $p$:
340: $0.125$, $0.2$ and $0.25$ from top to bottom. The long time
341: tails are clearly visible, and the solid lines are fits to
342: the same functional form (see text).}
343: \end{figure}
344: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
345: From these curves we clearly see the existence of long-time tails in the
346: velocity auto-correlation function. Consider the ``Green-Kubo'' formula
347: corresponding to \eq{einstein}:
348: \begin{equation}
349: \label{difgk}
350: \frac{\langle (r_m(t)-r_m(0))^2\rangle}{t}= \int_0^t ds\,
351: \langle \mathbf{v}_m(s)\cdot \mathbf{v}_m(0) \rangle (1-s/t).
352: \end{equation}
353: The dominant contribution to $\langle (r_m(t)-r_m(0))^2 \rangle /t$
354: at large times is
355: \begin{equation}
356: \label{difasymp}
357: \frac{\langle (r_m(t)-r_m(0))^2\rangle}{t} = D(m,p)-\int_t^\infty ds\,
358: C^{(m)}_{vv}(s).
359: \end{equation}
360: where $C^{(m)}_{vv}(s)=\langle \mathbf{v}_m(s)\cdot \mathbf{v}_m(0) \rangle$ is
361: the velocity auto-correlator. Therefore, a power law tail
362: $C^{(m)}_{vv}(s)\sim t^{-\alpha}$ in the velocity auto-correlation function
363: will translate into a corresponding power law tail $\langle
364: (r_m(t)-r_m(0))^2 \rangle /t \sim D(m,p)+\text{const.}\,t^{1-\alpha}$ in the
365: Einstein relation. In simple liquids a value of $\alpha=3/2$ is
366: ubiquitous \cite{Hansen}, and has also been observed for gelating systems in
367: \cite{vernon01}: here we find that the same power-law provides a very good
368: fit to the data for all $m$, but in particular for the small
369: clusters. In \fig{fig.tail1} we have also plotted these fits to the
370: power-law $a+bt^{-1/2}$, and the deviation from the simulation results
371: at early times
372: is barely visible. In figure \fig{fig.tail2} we have done the same for
373: clusters of mass $10$, and we see the same behavior. The agreement is
374: slightly worse, presumably due to poorer statistics of larger
375: clusters.
376: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
377: \begin{figure}
378: \includegraphics{tail2}
379: \caption{\label{fig.tail2} Same as \fig{fig.tail1} but this time for
380: clusters of size $10$.}
381: \end{figure}
382: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
383: We note the existence of a maximum in all of the curves (though not
384: visible on \fig{fig.tail1} for $m=1$) for $\langle (r_m(t)-r_m(0))^2 \rangle
385: /t$. By differentiating \eq{difgk} this can be shown to occur at $t_m$, where
386: $t_m$ is the solution to
387: \begin{equation}
388: \int_0^{t_m}ds\, C^{(m)}_{vv}(s)s =0.
389: \end{equation}
390: An obvious consequence of the fact that for $t>t_m$ (using \eq{difasymp})
391: \begin{equation}
392: \frac{d}{dt}\frac{\langle (r_m(t)-r_m(0))^2\rangle}{t} \approx
393: C^{(m)}_{vv}(t) < 0
394: \end{equation}
395: is that $C^{(m)}_{vv}(t)$ becomes negative
396: (anti-correlation) for large $t$ and stays negative thereafter. This means that
397: the $1/\sqrt{t}$ tails in Figs.~\ref{fig.tail1}--\ref{fig.tail2} correspond to
398: a {\em negative} $t^{-3/2}$ tail in $C^{(m)}_{vv}(t)$. We also note that
399: $t_m$ is an increasing function of $m$ and a decreasing function of $p$.
400:
401: The error made by taking $D(m,p)$ to be the value of
402: $\langle(r_m(t)-r_m(0))^2 \rangle /6t $ at the end of the simulation
403: time $t=120\tau$ is neglibly small: for $p=0.2$ we have compared with
404: simulations that are twice as long, and at least for the $20$ lightest clusters
405: for which we had good enough statistics, the error was less than
406: $5\%$. For the smallest $m$ where the statistics are very good, one
407: can also obtain $D(m,p)$ from a power-law fit as mentioned above, and
408: the outcome is still consistent with the statement just made (the
409: error here is even much smaller than $5\%$).
410:
411: In \cite{sj02} we studied $D(m,p_c)$ and we found the power-law
412: $D(m,p_c)\sim m^{-0.69}$. We have repeated this study up to clusters
413: of size $50$, and we observe the same behavior over the entire
414: range. For $p< p_c$, we see the same power law as a function of
415: $m$, at least for small cluster sizes. The quality of the statistics for larger
416: cluster sizes is insufficient to determine whether there is a
417: cross-over or cut-off as $m\to m^*(p)$, where $m^*(p)\sim
418: (p_c-p)^{-1/\sigma}$ is the mass of the largest cluster, but it seems
419: likely that there is. de Gennes has argued that for masses $1<m<m^*(p)$,
420: $D(m)\sim m^{-(\nu+s)/(\beta+\gamma)}$ on the basis of a Stokes-Einstein
421: relation with a mass dependent viscosity \cite{deGennes79}. Here $\beta$ is
422: the exponent that describes the decrease of the order parameter near
423: percolation:
424: $x_{\text{gel}}\sim (p-p_c)^{\beta}$, where $x_{\text{gel}}$ the
425: fraction of particles on the spanning cluster and $p\to p_c+$. The other
426: exponents have been
427: introduced already. By using the appropriate scaling relations for $3D$
428: percolation
429: \cite{stauffer92}, the exponent can be rewritten so the prediction is
430: $D(m)\sim m^{-2(\nu+s)/(d\nu+\gamma)}$, where $d=3$ is the Euclidian
431: dimension. With our values for the remaining exponents we get:
432: \begin{equation}
433: D(m,p)\sim m^{-0.69}\quad\text{for}\quad 1<m<m^*(p).
434: \end{equation}
435: This is in very good agreement with our simulation results within
436: the observed power law regime. The theoretical prediction can be
437: rewritten as $D(R_g)\sim R_g^{-(1+s/\nu)}$ where $R_g\sim m^{1/D_f}$ is
438: the radius of gyration and $D_f$ the fractal dimension. This form of the
439: relation has sometimes (see for example \cite{gado00}) been used to infer
440: $s$ from the scaling of $D$
441: with $R_g$, but to the best of our knowledge the present study
442: presents the first direct verification of such a link.
443:
444: However, even in this regime one would expect some additional $p$
445: dependence of the diffusion coefficient, a point not addressed in
446: \cite{deGennes79}. To this end, we plot in \fig{fig.Dvsp} $D(m,p)$
447: as a function of $p$ for monomers, dimers and trimers, and we see
448: that the diffusion constants decrease (almost linearly) as a function of
449: $p$. Moreover the curves seem to fit nicely to the functional form
450: $D(p)=a(p_c-p)^{b}+D_c$, with a value of the exponent $b=1.1$.
451: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
452: \begin{figure}
453: \includegraphics{Dvsp}
454: \caption{\label{fig.Dvsp} Diffusion constant as a function of $p$ for
455: clusters of three different sizes: $m=1$, $m=2$ and $m=3$ from top
456: to bottom. The solid line is a fit to the function
457: $D(p)=a(p_c-p)^{b}+D_c$ and $D_c=0.0398$, $a=0.131$ and $b=1.103$
458: (see text).}
459: \end{figure}
460: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
461: In \fig{fig.Dvsp2} we have made a similar plot for masses $m=2\ldots
462: 10$, and the trends observed above appear to carry over to larger
463: masses. The curves are roughly parallel, and therefore it is not
464: unlikely that the value of $b$ is independent of $m$, but we are unable to
465: confirm this from a fit to the data: the exact value of the
466: exponent appears to be very sensitive to noise in the data.
467: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
468: \begin{figure}
469: \includegraphics{Dvsp2}
470: \caption{\label{fig.Dvsp2} Same as \fig{fig.Dvsp}, but for clusters of
471: sizes $m=2\ldots 10$ from top to bottom. The solid line is a fit,
472: and here $D_c=0.0064$, $a=0.0324$ and $b=1.029$.}
473: \end{figure}
474: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
475:
476: Finally we demonstrate a striking violation of the Stokes-Einstein
477: relation when approaching the gelation transition. The idea that
478: $D\propto 1/\eta$ is used so widely that one may sometimes forget its
479: lack of universal validity. In \fig{fig.Dvsv} however it is clear that
480: $D(m,p)\eta$ increases significantly when $p\to p_c$.
481: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
482: \begin{figure}
483: \includegraphics{Dvsv}
484: \caption{\label{fig.Dvsv} Plot of $D(m,p)$ times $\eta$: obviously
485: this is only approximately a constant, as predicted by the
486: Stokes-Einstein relation, far away from $p_c$.}
487: \end{figure}
488: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
489: This is consistent with our previous observations that whereas
490: $\eta$ diverges at the gelation point, $D(m,p)$ approaches a
491: non-vanishing constant even for large masses $m$. Further away from
492: the gelation point $p\lesssim 0.20$ there does however seem to be an
493: approximate proportionality between $D(m,p)$ and $\eta$. However, in
494: Figs.~\ref{fig.Dvsp}--\ref{fig.Dvsp2} we saw indications that $D(p)\sim
495: a(p_c-p)^b+D_c$ with $b>1$ whereas $\eta \sim (p_c-p)^{-0.65}$, and so
496: this apparent proportionality is at best only approximate.
497:
498: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
499: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
500: \section{\label{sec.con}Conclusions}
501: The main results of this study are the power law behavior of the mass
502: dependent diffusion coefficient which seems to hold well away from the
503: critical point, the failure of the Stokes-Einstein relation and
504: the result $s\approx 0.65$ for the critical exponent of the shear viscosity.
505: This last result, taken together with other recent results
506: \cite{vernon01,plischke02} seems to support the conjecture that the gelation
507: transition is not classifiable in terms of a single universality class:
508: exponents in the range $0.3\leq s\leq 0.7$ have been found for models that
509: seem, on the surface, to be very similar. The experimental situation also
510: does not provide much evidence for universality: both exponents
511: near $s=0.7$ \cite{adam81,adam85,durand87} and in the range $1.1\leq s\leq 1.3$
512: \cite{lusignan95,martin88a,adolf90,martin91,martin88b} have been reported.
513: We sound a note of caution here: The
514: determination of exponents through finite size scaling is not very
515: precise, especially when quantities that are as difficult to calculate
516: as the shear viscosity form the data set. However, it seems very unlikely
517: that the errors are large enough that a factor of more than 2 in the
518: exponent could be explained that way.\par
519: The mass-dependent diffusion coefficient in this model displays a power
520: law behavior $D(m,p)\sim m^{-0.69}$ consistent with a scaling argument
521: of de Gennes \cite{deGennes79}. Reexpressing this in terms of the radius
522: of gyration of clusters through $m=R_g^{D_f}$ where $D_f=2.5$ is the fractal
523: dimension of the percolating cluster, the scaling prediction is $D(R_g,p)\sim
524: R_g^{-(1+s/\nu)}$. This yields an estimate $s=0.65$ for the viscosity exponent,
525: in good agreement with the direct calculation from the Green-Kubo
526: formula. Whether this connection between diffusion and viscosity is general
527: or specific to the present model and whether there exist a similar
528: relationship between diffusion and the elastic shear modulus in the
529: solid phase remains a subject for further study.
530: Using a quite different model, del Gado {\it et al.}
531: \cite{gado00} have studied the self diffusion of crosslinked polymer
532: clusters on a lattice by bond fluctuation dynamics. They have also used
533: this scaling ansatz to infer the critical exponent of the shear viscosity
534: and found $s\approx 1.3$. Their result translates to a mass dependence of
535: the diffusion constant $D(m,p)\sim m^{-1}$, very different from that of the
536: present model.\par
537: Finally, we have shown that as the fluid becomes more viscous there is a
538: breakdown of the Stokes-Einstein law $D\eta\propto k_BT$ that generally
539: holds for simple liquids. For relatively small concentrations of crosslinks,
540: this product varies only very little but for $p\approx p_c$ the divergence of the
541: viscosity begins to dominate the diffusion constant which seems to saturate
542: for all cluster sizes studied at $p_c$.
543: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
544: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
545:
546:
547:
548: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
549: \begin{acknowledgments}
550: The authors wishes to thank B. J\'oos and D. Vernon for helpful
551: discussions. Financial support from the Danish National Research
552: Council grant $21$-$01$-$0335$ and from the NSERC of Canada is gratefully
553: acknowledged.
554: \end{acknowledgments}
555:
556:
557:
558: \bibliography{sfu}
559: %\bibliography{sfu}
560: %\BibTeX{sfu.bib}
561: %\bibitem{zipp99} K. Broderix, H. L\"{o}we, P. M\"{u}ller and A. Zippelius,
562: %Europhys. Lett., {\bf 48}, 421 (1999); Phys. Rev. E {\bf 63}, 011510 (1001).
563: \end{document}
564:
565:
566:
567: