Lecture 13: Polymer networks: the percolation theory.

If the concentration of cross-linkers is too small, the continuous network is not formed. Such
a network is formed by chains which are connected to the rest at least at two different points
(so that the part of the chain between the connection points can be mechanically loaded),
which, for too small concentration of cross-linkers, will not be the case. There exists some
critical concentration 𝑐! of cross-linkers at which such an infinite network (infinite cluster) is
formed. For 𝑐 < 𝑐! the system stays fluid (although it may contain large but finite clusters of
connected chains), above 𝑐! it acquires the properties of a (very soft) solid. From the chemical
point of view the infinite cluster is a single molecule.

There is a large class of systems showing a similar sol-gel transition (i.e. the transition between
fluid and solid state under building of intermolecular links). All these models share a
considerable degree of universality; some of their properties are described by percolation
theory (Broadbent & Hammersley, 1957).

Bernoulli percolation
A simple situation (at least a one which is much simpler than cross-linking the long molecules)
corresponds to chemical reaction of functional units (stars) or to reactions in mixtures of stars
and linear units (this is exactly what happens in a two-component glue). On a lattice these
situations could correspond to something like this:

A two-component glue:
- multifunctional unit
- linear chain

The simplest situations, however, correspond to the two classical models of percolation
theory: to the site model or to the bond model. In these models the sites or bonds can be
present (intact) or absent (broken), and the state of each site or bond (intact or broken) is
absolutely independent on what the state of any other site or bond is. In the site model, the
two present neighboring sites are always connected by a bond. In a bond model all sites are
present, but the bonds between them can be intact or broken (see Fig. below). Let p be the
concentration of intact bonds (sites). For 𝑝 → 0 only small clusters of connected sites (single
sites, pairs, etc.) are present, there is no infinite cluster. For 𝑝 → 1, on the contrary, there
exists an infinite cluster comprised of practically all system’s sites, with only few small holes.
There exists a well-defined critical concentration 𝑝! at which such an infinite cluster appears.
The critical concentration 𝑝! is non-universal: it depends on the lattice and on the particular
problem (site or bond) discussed. The examples of critical concentrations on different lattices
are given in a table below.

Bond (left) and site (right) problems of percolation theory.

Table: Critical concentrations on some lattices:

lattice pc , site pc , bond

honeycomb 0.6962 0.6527
2d square 0.5928 0.5
triangular 0.5 0.3463
diamond 0.43 0.38
3d sc 0.321 0.249
bcc 0.246 0.1803
fcc 0.198 0.119

The critical concentrations for the bond problem in 2d for the square, triangular and
" $ $
honeycomb lattices are known exactly (these are #, 2 sin "%, and 1 − 2 sin "%, respectively) due
to the fact that the square lattice is self-dual and that the honeycomb and triangular lattices
are dual to each other, and in addition connected by a triangle-star transformation (for more
careful explanation you should better have attended the lecture!). The critical concentration
"
for the site percolation on a triangular lattice is also equal to #. All other critical concentrations
are known from numerical simulations. In general, as one can see from the table, 𝑝! is higher
in the site than in the bond problem on the same lattice and in the same dimension, and in
any case 𝐶𝑝! ≥ 1 (this is a mathematical theorem), with C being the coordination number of
the lattice. For a bond problem an approximate relation holds
𝑑
𝐶𝑝! ≈ ,
𝑑−1
where d is the dimension of space.

Many other properties, especially the ones related to the behavior of the system close to the
critical concentration, are universal, i.e., independent on the type of the lattice. They only
depend on the dimension of space.
Some facts about percolation

Looking at the percolation system at 𝑝 < 𝑝! we see that it consists only of finite clusters of
different size. Looking at it at 𝑝 > 𝑝! , we see that it consists of the infinite cluster (a large
connected part taking the finite portion of the overall volume) with holes in which the smaller
finite clusters sit.

It is proved that under typical conditions there is a unique infinite cluster in a system, if any.
The typical size of the holes (for 𝑝 > 𝑝! ) defines the characteristic scale in a system, which we
will call the correlation length 𝜉 (essentially, the correlation length 𝜉 is defined as the mean
gyration radius of finite clusters. It is finite both below and above the critical concentration
but diverges at 𝑝! ).

When removing all finite clusters which do not contribute to elasticity of the system, and
looking at the structure of the infinite one at scales larger than 𝜉, this will appear
homogeneous, with the density 𝑆& < 𝑝. The density 𝑆& of the infinite cluster is nothing else
than the fraction of sites belonging to it. Below percolation threshold the infinite cluster
doesn’t exist, and 𝑆& = 0. Close to the percolation threshold the density 𝑆& behaves as a
power law
𝑆& ∼ (𝑝 − 𝑝! )' ,

where the values of critical exponent 𝛽 are known (at least numerically) in all relevant
dimensions. While approaching the critical point from above, the density of the percolation
cluster decays, which means that the holes in it get larger, so that 𝜉 diverges. This divergence
also follows a power law, 𝜉 ∝ (𝑝 − 𝑝! )() (here 𝜈 is not the Flory exponent). On the opposite
side of the transition, at 𝑝 < 𝑝! , the correlation length 𝜉 (defined as a typical size of largest
finite clusters) shows a similar behavior. Thus, on both sides of the transition

𝜉 ∝ |𝑝 − 𝑝! |() .

Since above percolation transition the holes in the infinite cluster are filled by the finite ones,
both definitions coincide. The properties of a percolation system close to the transition (from
both sides) are defined by the properties of large finite clusters.

The “mean polymerization degree”, i.e. the typical number of sites in a finite cluster (both
below an above the percolation transition) scales as

𝑁 ∝ |𝑝 − 𝑝! |(* .
The values of the corresponding exponents in two and three dimensions are as follows:

Exponent d=2 d=3

𝛽 5/36 0.41
𝛾 43/18 1.80
𝜈 4/3 0.88

Turning to the elastic properties of the clusters we note that the Young’s modulus and the
Lamé coefficients are nonzero only on the “solid side” of the transition, and
 𝐸 ∝ (𝑝 − 𝑝! )+ .
It is easy to see that 𝑡 > 𝛽 (elastic moduli grow slower then the density of the infinite cluster)
since close to the transition the major part of the cluster consists of dangling ends which do
not contribute to elasticity. A theoretical model of an infinite cluster is presented in the figure
below.

node

a dangling a blob of multiply

end shunted bonds

„red“ (non-
duplicated)
bonds

The experimental value of t (3d) is 𝑡 ≈ 1.7.

Percolation on a tree
The oldest percolation model would correspond to a classical model of a polymer network due
to Flory and Stockmayer (1943) and corresponds to percolation on a regular tree.

Since trees (being infinitely dimensional lattices) do not grow infinite in any finite dimension
(you know what happens with dendrimers!), this is not a very good model for percolation in a
three-dimensional world. There are also some differences in the behavior of the percolation
system on a tree and in lower dimension. For example, the number of infinite clusters (if any)
is infinite on a tree (it is enough to cut one bond to separate the system into two independent
parts) and is one in a three-dimensional lattice.

However, the percolation model on a tree is an exactly solvable one. The fact that it is exactly
solvable is connected with the absence of the loops which permits us, after defining some
(arbitrary) intact site of the tree as its root, to give each bond a direction (from the root
outwards) and to classify all sites according to their generation (distance from the root). If C is
the coordination number of the tree, the root possesses exactly C ourgoing bonds and each
site other than the root has one incoming and 𝐶 − 1 outgoing bonds.
 bond of second
generation

root

cluster intact bond

broken bond

Note also that for a tree the bond and the site problems are equivalent: the absence of the
site is the same as the absence of the incoming bond (the last one on the way from the root
to this site). For the sake of concreteness in what follows we concentrate of the bond
formulation of the problem.

Let S¥ be the probability that the root belongs to an infinite cluster (i.e. that there is an infinite
way starting at a root). Such a way only exists if there is an infinite way from one of C neighbors
of the root. Denoting the probability that a given site other then the root possesses an infinite
way starting on it by Q we get:
(1 − 𝑆& ) = (1 − 𝑝𝑄),
(the probability that there is no infinite way from the root is the probability that there is no
infinite way starting with each of its outgoing bonds. The outgoing bond starts an infinite way
if it is intact, and there is an infinite way starting from the site it is showing to). If 𝑄 = 0 than
𝑆& vanishes as well, and if 𝑄 > 0 than the value of 𝑆& is nonzero and positive.

Similar discussion also holds for any site other then the root, with the only difference that
there are not 𝐶 but 𝐶 − 1 outgoing bonds. Therefore
(1 − 𝑄) = (1 − 𝑝𝑄),(" .
Expanding the binomial in the right-hand side of this equation we get
(𝐶 − 1)(𝐶 − 2) #
𝑄 = 𝑝(𝐶 − 1)𝑄 + 𝑝# 𝑄 + ⋯.
2
This equation always possesses a trivial solution 𝑄 = 0, but may also possess a nontrivial one.
To look for this we note that Q is always non-negative and that close to the critical
concentration Q is small (it is enough to keep the terms up to the second order, just as written
above). We thus get
2
𝑄= # [𝑝(𝐶 − 1) − 1],
𝑝 (𝐶 − 1)(𝐶 − 2)
and the nonnegative solution only exists for 𝑝(𝐶 − 1) − 1 > 0, i.e. for
1
𝑝 > 𝑝! = .
𝐶−1
Note that close to 𝑝! (but above 𝑝! ) the value of Q behaves approximately as 𝑄 ≃ const ⋅ (𝑝 −
𝑝! ) and thus 𝑆& ∝ (𝑝 − 𝑝! ) (to see this just expand the binomial in the equation for 𝑆& ). Thus,
on a tree we have 𝛽 = 1. We can also find the values of other critical exponents, but this
needs for some additional mathematical techniques.