HAL Id: jpa-00212407
https://hal.archives-ouvertes.fr/jpa-00212407
Submitted on 1 Jan 1990
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of
sci-entific research documents, whether they are
pub-lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Relaxation of entangled polymers at the classical gel
point
M. Rubinstein, S. Zurek, T.C.B. Mcleish, R.C. Ball
To cite this version:
Relaxation of
entangled polymers
at the
classical
gel point
M. Rubinstein
(1),
S. Zurek(*),
T. C. B. McLeish(2, *)
and R. C. Ball(2)
(1)
Corporate
Research Laboratories, Eastman KodakCompany,
Rochester, NY 14650-2115,U.S.A.
(2)
CavendishLaboratory, Madingley
Road,Cambridge,
CB3 0HE, G.B.(Reçu
le 6 novembre 1989,accepté
le 18 décembre1989)
Résumé. 2014
Nous considérons la relaxation d’un
système
de chaînes enchevêtrées etpontées
auvoisinage
de la transitionsol-gel,
dans le cadre de la théorie dechamp-moyen
(Flory
etStockmayer).
Le calcul utilise le modèle de « tube » des enchevêtrementstopologiques :
lacontrainte diminue à cause de la fluctuation
hiérarchique
des chaînes entre les ponts. Le temps de relaxation est calculé par une relation ayant une solutionanalytique
auvoisinage
de la transition. Une conclusion surprenantes’impose :
tous lespolymères
relaxent avant le temps finiT~
et donnent une courbe de relaxationG(t)
=G0
03B3-2[a-1 ln (T~/t)]4,
où 03B1 dénombre lesenchevêtrements entre les ponts, et 03B3 est une constante de l’ordre de l’unité. Pour des temps
beaucoup plus
courts queT~,
on retrouve une loi d’échelle faible. La relaxationrapide
despolymères
lesplus grands
(qui
n’a pas de sensphysique)
estgênée
par une transition vers lastatistique
de lapercolation.
Les deux domaines sontséparés
par un temps de l’ordre deT~.
Abstract. 2014
We examine the relaxation behaviour of an
entangled
cross-linkedpolymer gel
as itapproaches
thegel point
in mean field(Flory-Stockmayer)
percolation.
The calculation is basedon a tube model for the
topological
interactions in which stress is lost via hierarchical fluctuation of theprimitive paths
between cross-links. Thedecay
time of a segment is calculated via arecursion relation which has an
analytic
solution near thegel point.
Thestartling
conclusion is that all clusters relax in a finite timeT~ giving
a relaxation modulusG (t )
=Go
03B3-2[03B1-1 ln
(T~/t)]4,
where 03B1 counts the number of
entanglements
between cross-links and 03B3 is a constant of orderunity.
For timescales much shorter thanT~
this may resemble a weak power law. Theunphysically rapid
relaxation of thelargest
clusters isprevented by
a transition topercolation
statistics at
long
length
scales. The timescaleseparating
the tworegimes
is close toT~.
Classification
Physics
Abstracts 47.50 - 61.25 - 46.30Introduction.
The
dynamical properties
ofpolymer gelation
haveenjoyed
considerable recent interest bothexperimentally
andtheoretically
[1-7].
Near thegel
point,
at which an infinite cluster first appears, staticproperties
of thesystem
take on ascaling
form. The molecularweight
(*)
Permanent address :Dept.
ofPhysics, University
of Sheffield, Sheffield, S3 7RH, G.B.distribution as
specified
by
the numberdensity
of clusters of massM,
n(M),
is apower-law
[8-10] :
where
f
is a cut-offfunction,
constant for molecularweights
betweenM.,,
which is a low-masscut-off,
andMchar’
The latter is the molecularweight
of thelargest
clusterpresent
anddiverges
at thegel point :
Here p
is the numberdensity
of bonds in thesystem,
or an extent ofreaction,
and Pc is the criticaldensity
atpercolation.
The zero shear rateviscosity
qo and recoverablecompliance
J2
diverge
at pc and the zero shear rate modulusGo
appearsbeyond
it :The
frequency-dependent
modulus also exhibitsscaling
behaviour at thegel point
[3-7] :
There is no direct way of
determining
thedynamic
exponents s, t
and u from the static oneswithout further
assumptions
about the nature of thedynamic
interaction[2,
6,
7].
One of the most
important
quantities
affecting
thedynamics
ofgelation
clusters is theirdegree
ofoverlap.
Even without thecomplication
oftopological
interactions between clusters theoverlap
determines whether thehydrodynamic
interaction between monomers is screened or not ; i.e. whether thedynamical
behaviour is Rouse-like or Zimm-like[10].
This is seen in solutions of linearpolymers.
The motion of monomers onreally
dilute chains isstrongly
coupled
via the backflow of solvent to distant monomers on the same chain. When chainsbegin
tooverlap,
however,
hydrodynamic
contributions from other chains in theneighbour-hood interfere and
conspire
to screen the self interaction onlength
scaleslonger
than thecorrelation
length
fordensity
fluctuations. This «screening length
» is thereforedependent
onconcentration. It divides the
dynamical
modes of apolymer
chain into those oflarge length
scales which see local friction
(Rouse-like)
and those ofsmall length
scales which arenon-local
(Zimm-like).
Strong
overlap
willadditionally
mean thatentanglements
maydominate,
though
thesebecome
important
at ratherhigher
concentrations than the onset ofhydrodynamic
screening.
This is seen in melts of linear chains where there is a critical molecularweight
(which
isusually
several
thousand)
above whichdynamical
behaviourdiverges
from thephantom-chain
predictions. Entanglements
are also affectedby
polydispersity
so that in a melt ofpolymer
bimodally
distributed(say
that a fewlong
chains arepresent
in a bath of much shorterones)
for
example,
it ispossible
for thelong
chains to beeffectively
unentangled (at
timescales characteristic of their orientational relaxationthey only entangle
with otherlong
chains)
eventhough
the short chains are wellentangled.
The richness of thisproblem
has beeninvestigated
by
Doi et al.[11].
For
gelation
clusters we thereforeexpect
entanglements
to beimportant only
when aof the
degree
ofoverlap
for a cluster of molecularweight
M was introducedby
Cates[12].
We count the number ofpieces
of mass M from alllarger
clusters that sit within the volumespanned by
asingle
cluster of mass M. The radius ofgyration
of a cluster scales asR(M) ’" M1/df
where df
is the fractal dimension of the clusters. Thedegree
ofoverlap
4S (M)
[12]
is thengiven by
anintegral
over the mass distributionlarger
than M :Here d the dimension of space. We notice that an entire distribution of clusters may exist at
marginal overlap
(e (M) - const. )
if theexponents df and T obey
the «hyperscaling »
[8]
relation :
« Blobs » of mass M
just
manage to fill spaceindependently
of thescale,
M,
that we choose-the
system
is self-similar. This is satisfiedby
3-dimensionalpercolation
clusters which haved f ---
2.5 and T = 2.2[1],
so are notstrongly overlapped.
A moregeneral
result due to Cates[12]
is that branchedpolymers
with excluded volume which is screened atlarge overlap
willobey
(1.6)
for a range of values of T anddf.
This was usedby
Rubinstein et al.[6]
to calculate thedynamical scaling
ofpolymeric percolation
clusters on theassumption
thatmarginal
overlap
was effective inhydrodynamic screening
but did not cause the clusters toentangle
strongly,
results which are confirmedby experiment
[3,
6,
7].
Near thegel
point
we find that thedynamic
modulus satisfies(1.4)
with u = 0.69compared
with the theoretical value of 0.67.A
contrasting
case is found in thecross-linking,
orvulcanisation,
of densepolymer
chains when excluded volume is screenedthroughout
the distributionexcept
for the verylargest
clusters[13].
Now the classicaltheory
is valid andgives
theexponents df
= 4 andT = 5/2. From
(1.5)
we see that the the clusters arestrongly overlapped
and that over somerange of timescales at
least,
entangled dynamics
must dominate. It becomescompelling
to ask what viscoelastic behaviour this newassumption implies.
The
viscoelasticity
ofentangled
linearpolymers
is now understoodusing
the tube model oftopological
interactions[14]
and its extension to branchedpolymers
has been an active area of interestrecently. Applications
haveprogressed
from treatments of star molecules[15]
withoutaccounting
for « constraint release » to models for morecomplex
molecules[16, 17]
which treat thetopological
constraintsself-consistently
[18].
The main results forentangled
tree-like branchedpolymers
(containing
no closedloops)
may be summarised as follows :1)
as for linearpolymers,
stress(after
astep
strain)
comes from theanisotropy
of thepolymer
segments.
In fixed networks this is isproportional
to the fraction oforiginally
occupied
tube that has not beenpassed by
a free chainend ;
2)
stress relaxesby
arm retraction. Mostsegments
have an end whose chemical distance to the outside of the tree is shorter(by
the molecularweight
of thesegment)
than the distance from the other end. The former fluctuates on a characteristic timescale Tri in theentropic
potential
forpath length
that isapproximately quadratic
about the meanlength,
as for starFig.
1. - Aregion
of the Bethe lattice on which we grow the clusters. Bond i has aseniority
of 3, whichis the
longest
chemicalpath
to the outside of the cluster of the tree shown in(b).
The other tree(c)
has alonger
LCP of 4.Here
Mx
is the molecularweight
between the cross-links andMei
is theentanglement
molecularweight appropriate
to the timescale at which the i-thsegment
relaxes.v is a constant of order one which may be calculated from the tube model’s
prediction
for theplateau
modulus of linearpolymers
to be 15/8[11].
Thosesegments
whose ends share thesame chemical distance to the ouside of the tree are the last in their cluster to
relax,
and do soby reptation
(see
later) ;
3)
relaxation is hierarchical. Ti +1 becomes
thetimestep
for thepath length
fluctuations of theneigbouring
segment
nearer in chemical distance to the centre of the tree ;4)
thelarge separation
of timescales on which differentparts
of a molecule relax meansthat all relaxed
segments
act as effective solvent for unrelaxedparts.
IfCi
is the concentration of unrelaxedsegments
at the timescale on whichsegment
i relaxes then the effective dilutionmeans that the stress is
proportional
toC 2 rather
thanCi [19].
This may be recast in terms of adynamic
value of theentanglement
molecularweight
when it becomes clear that the dilution has a
dynamical
effect aswell,
accelerating
the relaxation ofneighbouring
segments
via(1.7
and1.8).
This acceleration can have verydramatic effects indeed even in the case of
simple
starpolymers
[18].
In this paper weapply
these ideas to
Flory-Stockmayer
gelation.
Statistics of
entangled paths
atgelation.
Consider a Bethe lattice with co-ordination number z
(such
as shown inFig.
la forz =
3).
Each bond isoccupied
withprobability
p andempty
withprobability 1 - p
independent
of any other bonds of the lattice. This defines the mean field(Flory-Stockmayer)
percolation
model. Itproduces
an ensemble of branchedpolymers
which may be embedded inconcentration is to be
unity
as it must if we wish to model amelt).
We note that a certainamount of linear
polymer
ispresent
at anystage
of reaction(or
probability
p).
In an accurate treatment this fraction must be handledseparately
from the remainder becausethey
will relax their stressby reptation
rather than fluctuation.However,
wesuspend
this consideration forthe moment
(but
see the discussionbelow),
observing only
that the fraction of linear chains becomesexponentially
small athigh
molecularweight,
and consider the result offluctuating
dynamics
alone on the modelsystem.
We now define the
concept
ofseniority
of a bond. If the bond isempty,
we say that it hasseniority
0. If it isoccupied,
then itbelongs
to a cluster - either finiteor infinite. Consider
bond i
belonging
to a cluster as shown infigure
la. This bond divides the cluster into two trees with bond i as the trunk of each of them(as
shown inFigs.lb
and1c).
Letm/l)
andmi(2)be
thelongest
consecutive sequence ofoccupied
bonds which we call thelongest
chemical
path
(LCP)
in each of these trees(in
Fig.
1mi(1)
= 3 andmi(2)
=4).
We say that bondi is of
seniority
This careful
prescription
above is unnecessary when thepolymer
itself is aregular Cayley
tree
[17]
in which every branchpoint i
is connected to z - 1equivalent
branchpoints
with characteristic relaxation timescales Ti _ 1, and one with a timescale of Ti + 1. In a tree formedby
randomgelation,
however,
the relaxation times ofneighbouring
branchpoints
aretypically
very different. To allow the branch
point
to move, z - 1 arms must retracecompletely
within their tubes(Fig. 2),
e.g. in theCayley
treepolymer
werequire
consecutivecomplete
Fig.
2. - Branchpoint
i is mobilisedby
thesequential
retraction of z - 1 arms. In this casefluctuation of z - 1 arms. In the case where the fluctuation rates of these arms are very
different,
thewaiting
time for theentanglement
loss offigure
2 is dominatedby
the slowest of the arms(excepting
the arm that remainsunrelaxed).
This isbecause,
given
an event ofwaiting
time Tthe probability
of an(uncorrelated)
event ofwaiting
time T2occurring
before another of the first kind isp
(event
2 before event1 ) =
this tends to
unity
whenTi >
r2 which musttypically
be the case when thewaiting
times areexponentially
separated
as in(1.7).
Henceforth we assume that the relaxation of a bond isdominated
by
theneighbouring
tree with the fastest relaxation time. This will be thelongest
of the relaxation times of the firsthierarchy
ofsegments
in the tree. Thissegment
in turn will be dominatedby
the relaxation of the faster of its twoneighbouring
trees(the
slowerbegins
with the the first bond weconsidered)
and so onfollowing
the LCP for bond i out to theedge
of the cluster. So theseniority
of a bond determines itsdynamics.
Theseniority
of a bond canbe infinite if it is
part
of the backbone of the infinite cluster. Bonds of infiniteseniority
do not relaxby
retraction. Bonds that are not in the backbone of the infinite cluster may relaxby
retraction when their turn comes. Perimeter bonds have
seniority
one.They
are the first to relax in anentangled
melt of clusters. Theirneighbours
which are notperimeter
bonds themselves haveseniority
two and are next to relax after theperimeter
bonds,
and so on until either the backbone of the infinite cluster is reached or thetopological
constraints become ineffective(see below).
We now need to calculate the concentration cm of bonds of
seniority
m. Let us first definetwo
auxiliary quantities :
qm is theprobability
that arandomly
chosen bond is the base of atree with the LCP
equal
to m andQm
is theprobability
that arandomly
chosen bond is thebase of a tree with the LCP not
larger
than m soWe can calculate qm
recursively
in terms of theQm
noting
that the chosen bond must beoccupied
(with
probability
p)
and that at least one of the z - 1 branches must have a LCPequal
to m - 1(we
exclude the case wherethey
all have branches of LCP of m - 2 orsmaller) :
using
the fact that qm =Q m - Q m - 1.
These recurrence relationtogether
with the initial valuesThis
probability
is nonzeroonly
above thepercolation
threshold,
i.e.for p
> Pc whereFor
large
m the finite differenceequation (2.4)
can beapproximated by
a nonlinear second order differentialequation
in terms of a continuous functionQ (m ) :
(From
henceforth we will usesubscripted
variables to denote thequantities
with discretearguments
and standard functional notation to denote their continuousapproximants).
Asm becomes
large,
we can find theasymptotic
behaviour ofQ(m)
by writing
with
when to second order in
Q(m),
Away
from thegel point
the first term dominatesgiving exponential decay
ofq(m) =
dQ /dm.
At thegel point,
however(p = pc = 1 / (z - 1 »
thisleading
order term isidentically
zero and the next term becomes dominant. This has apower-law
solution and wehave to first order in
1 /m
and for the
leading
behaviour ofq(m)
This can be checked
against
exact evaluation of(2.4)
and(2.5).
We find that the errorbecomes very small for values of m
larger
than10,
and thatasymptotic
forms are reached sooner forlarge
values offunctionality
z.We are now in a
position
to calculate the lattice concentrationcL (m )
of bonds withseniority
m in the continuous
approximation.
It isproportional
to theprobability
that one of its two trees has the LCPequal
to m, while the other has a LCPequal
to orlarger
than m. So the exact discrete result iswith the continuous
approximation :
(once
in eachtree).
The resultgives
the lattice concentration of clusters. In the melt the total concentration(volume fraction)
of all clusters isunity independent
of thedegree
ofpercolation
so for thereal-space
concentration we divideby
another factorofp(= l/(z -1)
atpercolation).
Theleading
behaviour ofc (m )
forlarge m
is determined from(2.11), (2.12)
and(2.13)
to beWe now
apply
this to the hierarchical relaxation of the ensemble of branchedpolymers
at thegel point.
Relaxation
dynamics.
The characteristic times for the relaxation of stress from bonds of
seniority
m isgiven
nowquite directly by
(1.7)
and(1.8)
with the effective concentration C(m)
given by
the fraction of bonds ofseniority
greater
than m :and in the continuous
approximation,
where this is a definition of the
0 (1)
prefactor
y (z).
Between the two characteristic timescalesdelimiting complete
relaxation of onebond,
stress relaxationproceeds
inexactly
the same way as for asingle
star arm, and follows the functiong (t)
calculatedby
Pearson and Helfand[15].
The finalexpression
for thedynamic
modulus,
whereGo
is thehigh-frequency
plateau
modulus is(Note
that this is the correctexpression,
not the onegiven by
McLeish in[17],
which overcounts contributions fromhigher
generations).
We needonly apply
thedynamical
rules summarised in the first section to calculate the series of timescales Tm. We can extract the essential behaviour at thegel
point by taking
the continuum limit of the discrete seriesTm
just
as we did for the statistics of bondseniority.
This was done for a melt of starsby
Ball and McLeish[16]
who showed thatMe
needs to be allowed to vary over the relaxation of the arms as the effective concentrationchanges.
In our case, we havewhere
The
dynamical equation
for T(m )
can bethought
of as a series of activated processes in which the rateT - 1 (m + dm)
isgiven by
anattempt
frequency
T - 1 (m)
and anentropic
barrierenergy dU =
kTvMx/Me(m)
dm(this
is exact for dm =1).
Thepre-exponential
factorgives
only
small corrections to theresult,
as may be checkedby
comparing
theanalytic
solution of(3.3b)
with evaluation of(3.3a).
We have to use the current effective value forM,
at each level in thehierarchy
so that all relaxed bonds act as solvent for unrelaxedbonds,
while all unrelaxed bonds are effective in
creating
theentanglement
network. Thisrequires
the continuous differential
equation
forr (m)
to be of the form of(3.3b)
rather thand(log T) = d(U(m)/kT)
in which the functionMe(m)
is differentiated. ForFlory-Stockmayer gelation
(3.3)
becomesso that
Here a =
v M xl M eO
and we havearranged
the units of time so thatT (1 )
= 1. The«
accelerating »
effect of thedynamic
dilution is verystrong :
we recover a finite time T(1 ) exp {a
y(z)}
in the limit m - oo which we callT 00.
This isexponentially longer
than thetime to relax the
perimeter
bonds. We may now calculate the relaxation modulus under theassumption
that bonds of agiven seniority
relaxindependently,
but that thedensity
ishigh
enough
topermet
a well-definedpre-averaged
effective concentration :This is a universal function of
a -1 log
(T oo/t)
up to thefunctionality-dependent prefactor,
asmay be seen from
(3.6).
Just as in the case ofsimple
stars, however for times shorter thanT 00
theexponential
term in(3.7)
becomeseffectively
astep
function in m and the relaxationmodulus becomes from
(3.1)
and(3.6) :
A
graph
of[G (t) ]1/4
against log t will give straight
line for timescales dominatedby
entanglements.
For times t «T 00
the bracket in(3.8)
can beexpanded
togive
aregime
overwhich
G (t )
has anapproximate power-law
form t-" :In const with
though
the full functiondrops steeply
away from anypower-law approximation
at timesAN EXACT RESULT. - We noted that the
asymptotic
forms for the concentrationsq (m )
andc (m )
become betterapproximations
at any finite m as we increase thefunctionality
z. It is therefore no smallsurprise
thatthe
final form for the relaxation modulus(3.8)
becomesexact for all
integer
values of m in the case of z =3,
the lowestpossible functionality,
provided
we choose a renormalised value forToo.
This isfairly
straightforward
todemonstrate,
and is a consequence of thesimple
form of the recursion relation(2.4)
in thiscase. The derivation is
given
inappendix
A. The distinction is thatTOC)
is renormalised tot (1) exp {3 a}
rather thant (1) exp {16 a }.
Such achange
is to beexpected
from the observation that theasymptotic
calculation overestimatesC(m)
for small values ofm
(Eq. (3.1)).
An exact solution mustproduce
the faster relaxation that results fromgreater
dilution. The result for z = 3 is useful because itgives
a bound on the renormalisation ofTOC)
(the
asymptotic
results become more accurate forlarge
z)
and because it indicates that the form of the relaxationspectrum
at lowseniority
is notstrongly dependent
onfunctionality
atexperimentally
relevant values. Theprediction
of a lineardependence
of[G (t) ]1/4
onlog
(t )
is notchanged.
Discussion.
The model described above is
idealised,
andrequires
carefulqualification
if it is to beapplied
toexperiment.
We saw earlier that there are stillapproximations
in our treatment which may beimportant,
but even as it stands we know that atlong
times ourpicture
of ahighly
entangled
network must fail. We consider thisregime
first.TRANSITION AT LONG TIMES. - The existence of
a finite time for the relaxation of
indefinitely
large
finite clusters isunphysical,
and weexpect
the model to break down at times(and
corresponding length
scales)
shorter thanToo
There are apriori
two routesby
which the model may fail : on the one hand theassumption
of classical statistics for the static distribution of the clusters is wrong atlarge length
scales because clusterswith df
= 4eventually pack
in d = 3[13] ;
on the other we know thatdynamic
dilutioneventually
renders thelargest
clusterseffectively
dilute,
after whichunentangled dynamics
areappropriate.
Inpractice,
whichevertransition,
dynamic
orstatic,
occurs first will define the crossover to thenew
regime.
First we consider
possible
dynamic
transitions. At the level of the small modificationsintroduced
by
the term inlog
(Mx/Me)
we mustexpect
achange
in behaviour when theeffective
entanglement
molecularweight
reachesM,,
[17].
This will occur at ahigh
seniority
m,,
given by
giving
This cannot affect the main features of the
entangled
relaxation,
however,
becausesegments
there is no
topological
constraint from the tube[22].
We arrive at aseniority
for thedynamic
transition
Mdy.
given by
This
seniority
varies as the square of mx, so in the limit ofincreasing
distance between branchpoints, entangled dynamics
maypersist
wellbeyond
mx.This
potential dynamic
transition may not be active if our model fails toobey
classical statistics at molecularweights corresponding
to smaller seniorities thanmdyn.
We must check whether the staticconfiguration
affects ourassumptions
at an earlier level in thehierarchy.
Atlarge length
scales and cluster masses fluctuations in the number densitiesdestroy
theapplicability
of the mean-fieldresults,
aspointed
outby
de Gennes[13].
Beyond
thispoint
percolation
statisticshold,
which as we have seen leads toonly marginal overlap
andunentangled dynamics.
De Gennes casts the result in the form of an intervalOp
around thepercolation
threshold inside which classical statistics break down and findswhere
Mk
is the molecularweight
of a Kuhnlength.
We may use therelationship
between thelargest
mass of the distribution and crosslinkdensity Mchar -
Mx(J1p /Pc)-l/u to
convert this into a critical cluster massMstat -
MX5/3 Mk 213
using
the classical value of u = 1/2.Finally
weuse the fact that bonds of
seniority m typically belong
to a cluster of LCP - m and of massM - Mx m 2
(the
massscaling
of clusters withpath length
must be half thespatial
fractal dimension because thesegments
areGaussian).
Thisproduces
When
Mx
becomeslarge,
this transition dominates because itgives
the smallest value for atransition
seniority.
In this case, thedynamics
become « slaved » to the static distribution and must beunentangled
at timeslonger
thanT (mstat) (which
can be calculated from(3.6)).
Which
type
of crossover wins inpractice
may be affectedby
the rather differentprefactors
toMx
in(4.3)
and(4.5).
The static transition dominatesonly
forwhich may be
large
sincepolymer
chains are flexible well below theentanglement
molecularweight
in the melt.However,
the differences ofseniority
between these twopossible
transitions(and
even theseniority
mx)
are masked in anexperiment
which measures relaxation as a function of time because all m » 1correspond
to times close toT 00
from(3.6).
The form ofG (t )
observed isexpected
to follow(3.8)
for 0 tT 00
and then cross over to thepercolation power-law
relaxationpredicted
and observedby
Rubinstein et al.[6],
G(t) -- t-0.67.
Wepresent
acalculation of
G (t )
for awell-entangled
ensemble near thegel
point
infigure 3.
Thisillustrates three
regimes :
theentangled
for 0 tT 00’
theunentangled
forT 00
tT max
and the terminalregime
for t >T max
whereT max
is the Rouse relaxation time of thelargest
Fig.
3. - Thedynamic
modulusG (t )
of theentangled
ensemblejust
belowgel point.
The molecularweight
between cross-links is such thatT 00
exp ={60}.
The crossover tounentangled
dynamics
occursjust
beforeT 00 , Tmax, corresponding
to the Rouse relaxation of thelargest
cluster isexp {70} .
Alsoshown
superimposed
is theresulting compliance
J (t) (the
rising
curve)
with a vertical shift factor toallow use of the same axes.
FORM OF THE COMPLIANCE. - Because of the very
long
terminal times associated withentangled
branchedpolymers,
creepexperiments
on thecompliance
J(t)
will be moreaccurate in
assessing
the linearrheology
than constant-strainexperiments
such asoscillatory
orstep-strain rheometry.
The characteristic behaviour of the different relaxationregimes
discussed above is reflected in thecompliance
just
asstrongly
as in the modulus. In the tworegimes
whereG (t )
isapproximately
apower-law
(0 - t T 00
whenG(t) = t-4/ay,
and t >T 00
whenG (t) = t - 0.67)
J (t)
also has apower-law
form. This may be seen from theintegral relationship
betweenG (t )
andJ (t) [20] :
Taking Laplace
transforms andputting
G (t )
=Go t-U
gives
for 0 u 1 :So we
expect
to seeJ(t )
following
the inverse power toG (t ).
Numerical evaluation ofJ(t)
issuperimposed
infigure 3.
Below thegel point
theviscosity
is finite(the
largest
characteristic cluster has a terminal time ofT max
ofFig.
3)
so that the eventualslope
of a4/a y.
BetweenToo
andTmax
theslope
is 0.67. The existence of these tworegimes
ofentangled
andunentangled
relaxation dividedby
a transition nearT 00
is the mainexperimental signature
that wepredict.
EFFECT OF LINEAR AND LINEAR-LIKE SECTIONS. - The
gelation
process introduced in section 2 above does not exclude the presence of a fraction ofpurely
linearpolymer
at thegel
point.
By
randomly occupying
bonds on theCayley
tree we may achieve linear sectionsby
occupying
just
two of the z bonds at each nodealong
asimply
connectedpath
between nodes at whichonly
one bond isoccupied
(terminating
thechain).
Now when weembed
the ensemble intoreal space, the fraction of
segments
whichbelong
to linear chains will not relax their stressby
fluctuation,
butby
reptation.
As the molecularweight
between cross-linksincreases,
so does the ratio of the characteristic times for fluctuation and forreptation,
so forwell-entangled
ensembles,
the main effect of these linearcomponents
is to aprovide
a way to relax a fraction of theoriginal
stressquickly,
and to renormalise the concentration for the branchedpolymers
whose relaxation follows thepattern
describedabove,
butinitially
dilutedby
the fraction of linear material.The fraction of linear material of molecular
weight
kM. (consisting
of k consecutivebonds)
we denote
by
L (k, 0 ) (we
explain
the notationbelow)
The term in p comes from the necessary
occupied
bonds,
in1 - p
from theunoccupied
onesboth
along
thelength
and at the ends of thechain,
and the term in(z - 1 )
is a combinatorialfactor
arising
from the choice ofpaths
on the lattice. The total fraction of bonds in linearmaterial is
given
by
the sum ofL (k, 0)
over k. This isdependent
of thefunctionality
z and is
greatest
for small z. At thegel
point
p = 1 / (z - 1 )
giving,
The total concentration of linear material is the ratio of this sum to the total fraction of
occupied
bonds(which
fill space in themelt).
This is 1/2 for z = 3 so at thisfunctionality
1/4 ofthe total mass is in the form of linear chains and 1 -
[1 -
(1/4 ) ]2
= 7/16 of the total initialstress is relaxed
by
these linearsegments.
The relaxationspectrum
of thesereptating
chains will begiven by
a sum over theirexponential
molecularweight
distribution :The
reptation
time of k-chains isdependent
on theirlength
cubed,
the well-known result fromlinear melts
[14],
but is also modifiedby
the effective concentrationC,
which isdefined,
as forIt is clear that this dilution effect reduces the
reptation
timesby
9/16 over thereptation
régime.
The full relaxation modulus of(4.11 )
iscomplicated,
but itsapproximate
form may be extractedby
representing
the sum over molecularweights
as anintegral
which is then evaluatedby
steepest
descents,
giving
a « stretchedexponential
» :After this extra short-time response, the branched
polymers
relaxby
fluctuation.However,
polymers
which are notlinear,
andbegin
to relax their stressby
fluctuation,
may becomeeffectively
linear atlong
times whenonly
high
seniorities remain unrelaxed. Anexample
is shown infigure
4,
which behaves as a diluted linearpolymer
after bonds ofseniority
1 have relaxed. The backbone bonds which have seniorities of two and threeactually
relax via thereptation
of the diluted backbone aftert (1 ),
sogiving
an error to the relaxationspectrum
calculated in section3,
which assumes that all relaxationsproceed
via fluctuation. Ingeneral
we must calculate the number of
objects
which become linear-like at somestage
of therelaxation. We denote
by
L (n, m )
the lattice-concentration ofobjects
whichbeyond
aseniority
m consist ofjust
a linear chain of nsegments
(L (n, 0 )
counts thesimple
linear chainsas
above).
Wefind,
after a littlethought,
thatThe first term,
q;" + l’
is theprobability
that the two end bonds of the linear subsection are ofseniority
m + 1. Then we have theprobability
that theremaining n -
2 bonds areoccupied,
pn- 2multiplied
by
the number of ways this can occur(z -
1 )n -1.
Wemultiply
thisby
theprobability
that all(z - 2) (n - 1)
side branches haveseniority
less than orequal
tom,
QJ: - 2)(n -1).
Weshould, however,
subtract theprobability
that the linear subsection ispart
of alonger
linear subsection that was able toreptate
at an earlierstage.
In this casejust
one ofthe z - 1 bonds at each end of the section must have
seniority
m -giving
the factor(z -
1)2 q 2,
and theremaining
z - 2 must all have seniorities less than mgiving
the factorFig.
4. - The occurrence of linear-like sections atlong
times. After bonds ofseniority
one have relaxed,this cluster follows
reptative dynamics
within the effectiveentanglement
network because no treeQ2(,- 2
All n bonds of the subsection must beoccupied
asbefore,
but now all the side branches must have senioritiesstrictly
less than m. Thisexplains
the second term of(4.14).
Beyond
seniority
m, theseobjects
relaxby reptation.
We need an estimate of the error thatthey
introduce into our calculation of the relaxationspectrum.
Agood
measure of this is theconcentration
F (m )
ofsegments
of nominalseniority
m that have their actual relaxationspeeded
upby
belonging
to a backbone that becomeseffectively
linear,
i. e. that relaxby
reptation
at an earliertimestep.
We observe thatobjects belonging
to the fractionL (n, m’ )
affect bonds withseniority
m such that m’ m m’ +n /2.
The total fraction ofbonds of
seniority m
affected is the sum over relevant linear-likeobjects :
The case of z = 3 is
especially
instructive here because weexpect
it togive
thelargest
effect oflinear-like sections
(only
one side branch per node need relaxalong
asubsection),
and alsobecause it is
possible
to calculateF (m )
exactly
in terms ofC (m ) (see
appendix
B).
We find thatwhich means that 1/4 of all relevant bonds are affected. This
implies
aninteresting
form ofdynamic
scale-invariance at thegel point :
the linear-like chains aresignificant
at all scales.However,
the form of the relaxation modulus is stillunchanged,
because thesystem
behavesas if it were at the renormalised concentration
Ceff (m) = C (m) - F (m) = 3 C (m)14.
Repeating
the calculation ofappendix
Ausing
the effective concentrationsyields
The
only
differenceis,
as before in the case ofaccounting
fornon-asymptotic
behaviour,
arenormalisation of
TOC)
which now has a value oft1
exp{9 a /4}.
A smaller size of error is
expected
to arise from effectivedangling
ends oflengths longer
than onesegment.
Imagine
the linear-likeobject
offigure
4 connected to alarge
clusterby
one of its ends. Its stress is still relaxedby
fluctuation,
but now theentropic potential
is thatfor a
dangling
arm foursegments
long
rather than one, and allows the intermediate bonds to relax faster thanthey
would in astrictly
consecutivehierarchy.
Such structures can blur thetime-orderedness of
seniority,
but occur with afrequency
given
by
anexpression
similar to(4.14).
Moreover,
there-ordering only
occurs within the hierarchies of theeffectively
lineardangling
end.Large experimental
deviations from the modelspectrum
are notexpected
from this source.AN ALTERNATIVE ENSEMBLE. - Instead of the mechanism of
creating
clusters on aCayley
tree and then
embedding
in real space, consider thefollowing
procédure :
Webegin
with abath of
monodisperse
linear flexiblepolymers
with reactive end groups offunctionality
z
(that
is,
with z - 1 sparebonds).
We allow each end group to attach to z - 1 other chains(so
that itsvalancy
issaturated)
withprobability
p, and to becompletely
unreacted withprobability
1 - p.
The reactions areindependent
ormean-field,
and so lead toFlory-Stockmayer
statistics,
when the chains starthighly
overlapped
(loop
formation issuppressed).
If we consider the calculation ofprobabilities
of LCPs in treesstarting
on a chosenbond,
wetoo,
except
thatq (0)
now describes the fraction of linear sections in the finalbath,
not theempty
bonds on aCayley
tree. Such asystem
may arisechemically,
and is a little « cleaner »than the
Cayley
tree model in that alllong
linear chains aresuppressed, though
we note thatlong-time
linear-like material is stillpresent.
BEHAVIOUR AWAY FROM GEL POINT.
- Finally
we may commentbriefly
on the behaviouraway from the
gel point.
Belowgel point
the terminal timeTmax
of thelargest
cluster may be less thanT 00
(rigorously
it must be less than T(mstat)
in which case there will be nounentangled
regime
where thepercolation
relaxation appears. The relaxationspectrum
is cut off atTmax.
Above thegel point
the situation is morecomplicated.
There is a finite fraction ofbonds which contribute to the backbone of infinite clusters. These never
relax,
andgive
rise toa finite modulus. Their existence affects the relaxation of other bonds because now the
effective
entanglement
molecularweight
does notdiverge
but is bounded aboveby
theentanglement
of thepermanent
network. As we move abovegel point
thepermanent
network becomes denser. There will be a new contribution to thedynamic
moduluscoming
from clustersdangling
from the infinitebackbone,
which have anexponential
distribution ofdangling
LCPs. The relaxation of such an ensemble was examinedby
Curro and Pincus[21]
in the case ofstrictly
linearsegments
and led to apower-law
stressdecay.
We defer a detailedcalculation of this
complex regime
to a further paper, but awaitexperiments
onentangled
gelling
systems
with interest.Appendix
A.Here we derive the exact discrete form of the
seniority
statistics and relaxationspectrum
for the case of z = 3. We start with the recursion relation for theQ., equation (2.4).
Forz = 3 this becomes
which we can use to prove a result
special
to this case for the summedquantities
Q,,,:
The result may be
proved by
induction as follows. First we observe that it is true form =
0,
forQo
= qo = 1/2 and ql = 1/8. Now suppose it to be true for ageneral
m -
1,
thenwhich
completes
theproof.
We also have a relation between the lattice concentration ofseniorities cLm and the one-sided LCP
probabilities :
This is true for m > 0
only.
For m = 0 we have CU = 1/2.The modulus
depends
on the effective lattice concentration of unrelaxedsegments
which,
as we saw
above,
is thecomplement
of the sum of these cm. However(A6)
makes the sumvery easy to
do,
soThis can be
directly
related to the time at which bonds ofseniority m relax tm
via their recursion relation(we
are still not at the level of theprefactor) :
The factor of z - 1 accounts as before for the renormalisation of concentration when we
embed the
Cayley
tree lattice into real space.Finally
we may sum this relation witht1 = 1
to show thatlog
{tm}
=8 a [Qm - q0 - q1]
and use(A5)
to relate the effectiveconcentration to the relaxation time :
So the form of the relaxation modulus derived from the
asymptotic
results becomes exact in the case of z = 3(for
the discrete sequence oftimes tm
withinteger
m),
but with arenormalised value for
Too
of tl
exp(3 a ).
Appendix
B.In this
appendix
we calculate the fraction of linear-like chains at eachseniority
m,which we must use in the double sum for
F (m ) :
We may
perform
the sum over nusing
the two standard resultsThe first term
contributing
toF(m)
from(Bl)
gives
onsumming
over n :using
(A5)
A similar summation can be
performed
on the second term togive,
oncombination,
The second sum starts from 1 rather than zero because the
expression
forL (n, 0 )
does not contain the subtractedpart
of(B1) .
Terms cancel between the two sums inpairs
except
for thelargest
element in the first sum, sowhich proves the result.
Acknowledgments.
We thank M. E.
Cates,
R.Colby
and L. Leibler for useful discussions. SZ and TCBM weresupported by
ICIplc.
References
[1]
STAUFFER D., CONIGLIO A. and ADAM M., Adv.Polym.
Sci. 44(1982)
103.[2]
CATES M. E., J.Phys.
France 46(1985)
1059.[3]
DURAND D. , DELSANTI M. , ADAM M. and LUCK J. M. ,Europhys.
Lett. 3(1987)
297.[4]
DAOUD M., J.Phys. A
21(1988)
L237.[6]
RUBINSTEIN M., COLBY R. H. and GILLMOR J. R.,Space-Time Organisation
in MacromolecularFluids, Eds. F.
Tanaka,
T. Ohta and M. Doi(Springer-Verlag, Berlin)
1989 ;Polymer
Preprints
30(1989)
81.[7]
MARTIN J. E., ADOLF D. and WILCOXON J. P.,Phys.
Rev. Lett. 61(1988)
2620.[8]
STAUFFER D., Introduction to PercolationTheory (Taylor
and Francis,London)
1985.[9]
STOCKMAYER W. H., J. Chem.Phys.
11(1943)
45.[10]
FLORY P. J., J. Am. Chem. Soc. 63(1941)
3083,
3091, 3096.[11]
DOI M., GRAESSLEY W. W., HELFAND E. and PEARSON D. S., Macromolecules 20(1987)
1900.[12]
CATES M. E., J.Phys.
Lett. France 46(1985)
L837.[13]
DE GENNES P.G.,
J.Phys.
Lett. France 38(1977)
L355.[14]
DOI M. and EDWARDS S. F. E., TheTheory
ofPolymer Dynamics
(Oxford
University
Press,Oxford)
1986.[15]
DE GENNES P.G.,
J.Phys.
France 36(1975)
1199 ;DOI M. and Kuzuu N., J.
Polym.
Sci.Polym.
Lett. Edn.18(1980)
775 ;PEARSON D. S. and HELFAND E., Macromolecules 17