Polymer 53 (2012) 3594e3601

Contents lists available at SciVerse ScienceDirect

Polymer
journal homepage: www.elsevier.com/locate/polymer

Modelling tie chains and trapped entanglements in polyethylene

F. Nilsson a, X. Lan a, T. Gkourmpis b, M.S. Hedenqvist a, U.W. Gedde a, *
a
KTH Royal Institute of Technology, School of Chemical Science and Engineering, Fibre and Polymer Technology, SE-100 44 Stockholm, Sweden
b
Borealis AB, SE-444 86 Stenungsund, Sweden

a r t i c l e i n f o a b s t r a c t

Article history: A Monte Carlo random walk model was developed to simulate the chain structure of amorphous layers in
Received 1 April 2012 polyethylene. The chains emerging from the orthorhombic crystal lamellae were either folding back
Received in revised form tightly (adjacent re-entry) or performing a random walk (obeying phantom chain statistics) forming
20 May 2012
statistical loops or tie chains. A correct amorphous density (ca. 85% of the crystalline density) was
Accepted 21 May 2012
Available online 8 June 2012
obtained by controlling the probability of tight folding. Important properties like fracture toughness
depend on the number of chains covalently linking together the crystalline regions. The model structure
was analysed with a novel numerical topology algorithm for calculating the concentration of tie chains
Keywords:
Polyethylene
and trapped entanglements. The numerical efficiency of the algorithm allowed molecular cubic systems
Tie chains with a side length of 100 nm to be readily analysed on a modern personal computer. Simulations showed
Trapped entanglements that the concentration of trapped entanglements was larger than the concentration of tie chains and that
the thickness of the amorphous layer (La) had a greater impact than the crystal thickness (Lc) on the tie-
chain concentration. In several other commonly used models, such as the HuangeBrown model, the
influence of trapped entanglements and the effect of the La/Lc ratio are neglected. Simulations using as
input the morphology data from Patel generated results in agreement with experimental rubber
modulus data.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction amorphous polyethylene [7], crystallization of polyethylene [8] and

The structure of a semi-crystalline polymer such as polyethylene
can on a limited length scale be viewed as a sandwich with alter-
nating layers of crystalline and amorphous regions. As sketched in
Fig. 1, the chains in the amorphous regions can be either tight folds,
statistical loops, loose chain ends (cilia), fully amorphous chains, tie
chains or chains connecting via permanently entangled loops to the
adjacent crystal, the latter being referred to as trapped entangle-
ments [1,2].
Early recognition of covalent bridges (tie chains) between
crystal lamellae was due to electron microscopy studies by Keith
et al. [3]. The importance of tie chains for fracture toughness and
slow crack growth is generally recognized. Notable studies on this
theme are due to Brown et al. [4,5] and Seguela [6]. Theoretical
models to predict properties of the amorphous interlayer are
usually based on molecular dynamics simulation, statistical models
or Monte Carlo random walk models.
Molecular dynamics models and similar models that include the
dynamics of the atoms or groups of atoms have been used to study
* Corresponding author. Tel.: þ46 8 7907640; fax: þ46 8 208856.
E-mail address: gedde@kth.se (U.W. Gedde).
the inter-crystalline layers of polyethylene [9e11]. Atomistic
processes involving relatively small molecules can be mimicked by
molecular dynamics simulation. Unfortunately, due to the high
computational cost, molecular dynamics models are not yet
capable of predicting the concentrations of tie chains and trapped
entanglements. Although molecular dynamics simulation can
estimate atomistic processes containing a relatively small number
of molecules, the statistics required to compute the concentrations
of tie chains and other entanglements by this simulation method is
computationally very expensive.
Among the computationally inexpensive statistical tie chain
models, the HuangeBrown model [5] is the most widely used. This
model predicts the fraction of tie chains (P) according to:
0 1
ZL  
1@ 4b3
P ¼ 1  pffiffiffiffi r 2 exp b2 r 2 drA (1)
3 p
0
where b2 ¼ 1.5/(6.8 nl2) for linear polyethylene [12] with n back-
bone carbonecarbon bonds each with a bond length (l) of 0.153 nm
[12]. The critical length L is usually chosen as twice the crystal
thickness (Lc) plus the amorphous layer thickness (La). Later
improvements of the HuangeBrown model include attempts to use

0032-3861/$ e see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.polymer.2012.05.045
 F. Nilsson et al. / Polymer 53 (2012) 3594e3601
Fig. 1. Schematic description of the possible shapes of polymer chains entering the
amorphous region between two crystalline layers: (a) tight fold, (b) statistical loop, (c)
loose chain end, (d) free chain, (e) tie chain, (f) trapped entanglement involving two
statistical loops.
finite integration limits [13], allowing a distribution in molar mass
and short-chain branching [14], as well as chain entanglements
[6,15]. Drawbacks of these mentioned statistical methods include
the use of an arbitrary L, the assumption that L is equally influenced
by a change in La and in 2Lc, and the assumption that trapped
entanglements are always formed if two chains are sufficiently
close to each other.
Monte Carlo simulations for predicting tie chain formations
include the Guttman model [16], which uses gamblers ruin statis-
tics to predict the shape of random walks of polymer chains in the
amorphous interlayer. This enables both the average length of
statistical loops and the average length of the tie chains for infinite
chains to be estimated. Early on-lattice simulations were per-
formed by Bryant and Lacher [17] to assess entanglements of
simplified random walk paths using knot theory [18]. These
simulations ignored the influence of crystal thickness and
approximated each random walk path with a simple triangle or
square. Nilsson et al. [19] developed a more realistic off-lattice
Monte Carlo model for predicting tie chain concentration using
a phantom chain random walk model to simulate the chains in the
amorphous interlayer.
Several entanglement models have been previously suggested.
Padding and Briels [20] developed a model for handling entangle-
ments in course-grained simulations of polymer melts. They were
modelling uncrosslinked polymers where no trapped entangle-
ments could ever be formed; their definition of an entanglement
was thus simply a temporary crossing of two polymer chains.
Lacher et al. [21] performed early on-lattice random walk simula-
tions of trapped entanglements between crystal layers and several
authors including Michalke et al. [22] analysed on-lattice simula-
tions with knot theory. An excellent review on trapped entangle-
ment models applicable to polymer networks was presented by
Lang et al. [23]. So far, no off-lattice method for simulating and
analysing trapped entanglements in semi-crystalline polymers
with realistic molar weights has been presented.
Monte Carlo random walk models can potentially provide
a good compromise between physical realism and computational
3595
speed. However, none of the currently presented models is
sufficiently realistic. They either oversimplify the amorphous
chain structure, ignore the effect of crystal thickness, simplify or
ignore the entanglements, assume non-branched chains of
constant or even infinite length, use small computational
domains, ignore the need to control the polymer density and
crystallization temperature or do not use realistic morphology
data. The aim of the present study was to improve the model
developed by Nilsson et al. [19] and to attempt to eliminate the
main shortcomings of the afore-mentioned Monte Carlo models.
The new model includes a method to control the amorphous
density and an advanced, realistic and efficient novel numerical
topology method for calculating the fraction of trapped chain
entanglements.
2. Models and methods
2.1. Model overview
In the model, the procedure for estimating the concentrations of
tie chains and trapped entanglements was divided in two consec-
utive steps. In the first step, parallel semi-crystalline layers were
built using a Monte Carlo method [19] allowing the positions of the
carbon atoms in the inter-crystalline layers to be stored for further
analysis. In the second step, a novel numerical algorithm for
calculating the degree of entanglement between pairs of loops was
used to determine the total fraction of entanglements as well as tie
chains. The algorithm was based on the observation that the only
information needed to validate the existence of a knot is that
describing the properties of the intersection points of the curves
when mapped on a plane.
2.2. The semi-crystalline layer model
In this study we assumed that the crystal thickness (Lc), the
amorphous thickness (La), the crystallization temperature (T), the
amorphous density (ra), the crystal density (rc) and the weight
average molar mass ðM W Þ as input parameters either already
known or determined by experimental methods. Molar mass
distributions could be used instead of M W and branches could be
included, but these features were not incorporated in this study.
The dimensions of the simulation box were chosen, as
Dx ¼ Dy ¼ 100 nm and Dz ¼ k(La þ Lc) where k is a positive integer
constant. The axis z is the normal to the crystal fold surface, y is the
direction of crystal growth (b-axis) and x the direction along the
crystal front orthogonal to the other two directions. Once the
periodical computational domain was set, the model calculated an
orthorhombic grid, which described the positions on the crystal
(001) surfaces where the simulated polymer chains could enter or
exit.
An atomistic model was generated as a sequence of chains
starting in the amorphous region at random. Each atom was
assigned a set of internal coordinates defining the chain confor-
mation in terms of bond lengths (0.153 nm; [12];), bond angles
(109.47 ) and torsion rotations. Each chain in the model was built
using a stochastic Monte Carlo method, leading to a number of
different initial configurations, without taking into account the
excluded volume. Ignoring the excluded volume is based on the
Flory principle that chains in the melt are unperturbed and their
structural characteristics can be considered to be the sum of all
possible configurations of a single chain [24]. The three torsional
conformations used were trans (T), gauge (G) and anti-gauge (G0 )
with the dihedral angles 180 , 60 and 60 , respectively. The
statistical weight of T was set to 1 and the statistical weights (s) of
the two gauche conformers (G and G0 ) are:
3596 F. Nilsson et al. / Polymer 53 (2012) 3594e3601

  otherwise it continues its progression in the amorphous region.

E
s ¼ exp  (2)
RT
where E is the potential energy difference between the gauche and
the trans states, equal to 2.1 kJ mol1 [24], and R is the gas constant.
The first conformer was simulated according to these statistical
weight factors. The probability for the next carbonecarbon bond in
a polyethylene chain to have a specific dihedral angular state was
determined according to the following probability matrix (statis-
tical weight matrix with each row being normalized, i.e. the sum of
the elements in a row is equal to unity):
2 3
1 s s initiated in the same way until the maximum filling degree x was
6 1 þ 2s 1 þ 2s 1 þ 2s 7
T
6 7
6 1 s 7
P ¼ 6
6 1þs 0 7 7G
6 1þs 7 (3)
4 1 s 5 0
0 G
1þs 1þs
T G G0
where the rows correspond to the last decided bond state and the
columns correspond to the adjacent not yet decided bond. As an
example, if the previous state was G then the probability for
entering T is given by 1/(1 þ s).
The random walk routine generated results relatively closely
resembling a polyethylene phantom chain; the characteristic ratio
for long polymer chains amounted to 6.5 at 413 K, which is close to
the experimental value at this temperature, 6.8 [25]. It should be
noted that the method used included two simplifications: (i) the
dihedral angles for G and G0 were set to þ60 and 60 . More
accurate values are 67.5 [26]. (ii) the statistical weight factors of
GG0 and G0 G were set to 0. In practice the chain will adapt to
a distorted GG0 with the dihedral angles þ100 and 67 [26]. The
two simplifications are counteractive and the final result is
reasonably close to the real phantom chain in terms of character-
istic ratio. It is clearly possible to use alternative random walk
routines such as course-grained models [27].
If the simulated chain ‘touched’ one of the crystal surfaces, i.e.
came closer to the surface than one bond length in the z direction,
and the remaining chain length was longer than the crystal thick-
ness, then the chain attempted to enter the crystal. If there were
still empty crystal grid positions on the surface sufficiently close to
the entry, then the chain would enter the crystal at the closest node,
Typically, a search radius of 1 nm was used as a limit, but the model
was not very sensitive to the exact value once it was larger than
about 0.5 nm.
When the chain exits a crystal surface, it can either make a tight
fold by immediately reentering the crystal at any adjacent empty
position or start a new random walk in the amorphous region. The
probability of tight folds (pi) was controlled by the model through
pi þ 1 ¼ pi þ (rA/rC)i  (rA/rC)AIM, so that the desired ratio of 0.852
between amorphous and crystalline densities was obtained [12].
The procedure continued until the positions of all carbon atoms in
the entire simulated chain were determined. New chains were
reached. The filling degree was defined as the number of occupied
grid positions on the crystal surfaces over total number of crystal
positions. Since the numerical errors of the model were largest at
either very low or very high filling fractions, x ¼ 0.5 was found to be
the optimal compromise. The positions of all carbon atoms were
stored in a matrix and analysed by the entanglement algorithm. To
achieve good statistics and to allow for a good representation of all
relevant conformational characteristics each individual setting
corresponded to an average of J simulated systems, where typi-
cally J ¼ 100 was used.
2.3. The entanglement algorithm
The concentrations of trapped entanglements and tie chains
were determined by analysing the carbon atom coordinate matrix
with our novel knot algorithm. The stored amorphous chain
segments were distinctly separated in the matrix and the tie chain
concentration was calculated as the ratio of tie chains over the
number of filled grid positions. All free-chains, loose ends, tight
folds and tie chains were removed at this stage because only
statistical loops can form trapped entanglements. All loops with
crystal entries on opposite crystal surfaces were compared in pairs
with each other.
Fig. 2a and b demonstrates the principle for the first step in the
identification of the trapped entanglements. Each loop was pro-
jected onto the xy-plane and mapped onto a grid with a spacing
chosen so that one step in the grid corresponded to a step slightly
larger than 0.153 nm, i.e. the distance between two adjacent
covalently bonded carbon atoms. All grid positions containing at
least one carbon atom were saved as the number 1 in a sparse

Fig. 2. a,b. Sketches of pairwise comparison of two loops anchored in adjacent crystals. (a) 3D graph of the two loops. (b) projection of the loops onto a grid in the xy-plane. Chain A
is coloured turquoise and chain B is violet. The chains can only collide inside the red bricks, corresponding to positions where the added sparse matrix representations of the loops
are summed to ‘2’. Inside the red bricks, all chain segments on opposite chains must be tested in pairs for collision in order to find the coordinates of all possible intersection points
of the projected chains. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
 F. Nilsson et al. / Polymer 53 (2012) 3594e3601
matrix of integers. In order to create continuous lines, the nearest
grid positions adjacent to diagonal grid movements were also filled.
When two loop projections were compared, their sparse matrix
representations were added. The only regions that could poten-
tially contain intersections were those with a sparse matrix sum
equal to 2, which reduced the computational cost of the method
remarkably.
Inside the areas with potential crossovers, all active line
segments from the first chain were tested for crossovers against all
the line segments from the second chains. This was made with
a standard lineeline collision detection method [28]. Inside the red
square in the example shown in Fig. 2b, two line segments of loop A
must be compared with four segments of loop B, resulting in 8
lineeline comparisons and two crossovers. For each chain and
intersection point, the z-value and the contour length from the loop
starting point to the intersection were computed. For both chains,
the sequence of intersection points was stored as a vector with (þ)
if the chain passed above the other and () otherwise, resulting to
a sign vector of the form: v1 ¼ [þ   þ þ þ ]. If all the signs were
equal, no entanglement was present. Trapped entanglements were
present if the number of elements in v1 was odd, or if it was even
and the number of (þ) was odd. Otherwise the self-crossings had to
be computed and a procedure for reducing the two sign vectors was
applied.
The first reduction step (Fig. 3a) was that two adjacent (þþ) or
( ) signs on curve A could be removed if there were no self
intersections between p1 and p2 on A and if all intersection points
between p1 and p2 on B had the same sign. The second step (Fig. 3b)
was that two adjacent (þþ) or ( ) signs on curve A could be
removed if all self intersections on line A between p1 and p2 had the
same signs as p1 and p2 while all encircled intersections on line B
had the opposite sign. The third and also final step (Fig. 3c) was that
two adjacent (þþ) or ( ) signs could be removed if all the con-
tained self intersection points had the property that their signs and
the signs of their encircled intersection points contained only one
sign change, e.g. [þ þ  ] while the intersections on B had the
same sign. Additional reduction steps might be needed for
capturing every possible knot configuration, but these three steps
definitely captured the vast majority of the configurations obtained
with random walks.
When the first sign vector was traversed and all removable sign
doublets including their corresponding intersections in the second
Fig. 3. aec. Simplification steps 1e3 used in the entanglement algorithm.
3597
vector had been removed, the same concept was applied on the
second vector. This procedure was iterated until no changes
occurred over a whole cycle. Only if the sign vectors could be
reduced to empty vectors were no knots present.
The concentration of entanglements was obtained by first taking
the lower of the calculated number of entanglements and the
double number of loops on the side with fewest loops involved in
entanglements. This number was divided by the product of the
number of filled positions and the filling fraction in order to take
into account the fact that the number of entanglements increases in
proportion to the square of the filling fraction.
3. Results and discussion
3.1. Parameter study: effect of molar mass, crystal thickness, and
amorphous layer thickness
In this section the effect of some important parameters on the
concentrations of tie chains (xtc) and trapped entanglements (xte) is
presented. Some of the values given to these parameters were not
realistic, e.g. the ratio of the amorphous density to the crystal
density received clearly unrealistic values in some of the demon-
strations. The key parameters were given the following standard
values unless otherwise stated: Nx ¼ Ny ¼ 100 nm, Lc ¼ La ¼ 10 nm,
T ¼ 300 K, x ¼ 0.5, rA ¼ 1 nm, ra/rc ¼ 0.852, J ¼ 100 and
M ¼ 140 kDa. All the parameters are defined in Section 2.2.
Fig. 4a presents the results of simulations using different values
for the molar mass and ra/rc (note that the realistic value is 0.852).
As expected, the tie chain concentration increased with increasing
molar mass (using the standard values for the other parameters)
according to the expression:
xtc ¼ a  b  expðc  log MÞ (4)
where a, b and c are constants. The concentration of tie chains
increased in a linear fashion with increasing ra/rc in the range from
0.8 to 0.975. Note that the other key parameters were set at their
standard values. The strict demand that ra/rc ¼ 0.852 was met by
having a high concentration of tight chain folds. This is relaxed by
allowing higher ra/rc values leading to a decrease in the tight folds
and increase in the concentration of all statistical amorphous
chains, including tie chains.
3598 F. Nilsson et al. / Polymer 53 (2012) 3594e3601

Fig. 4. a). Tie-chain fraction as a function of Mw for different values of rA/rC. The tie-chain fraction increases with increasing values of both variables. (b). Tie-chain fraction as
a function of Mw and crystal thickness Lc. The tie-chain fraction increases with increasing Mw but decreases approximately linearly with increasing Lc. (c). Tie-chain fraction as
function of Mw and amorphous thickness La. The tie-chain fraction increases with increasing Mw but decreases approximately as 1/La with increasing La. (d). Tie-chain fraction (filled
symbols) and entanglement fraction (open symbols) as a function of Lc/La when the critical distance L ¼ 2Lc þ La ¼ 25 nm. Note that the influence of the entanglements is not
negligible and that the functions are not constants, as is assumed in Brown type models.

Fig. 4b shows that the tie chain concentration decreased in an
approximately linear fashion with increasing crystal thickness for
different constant molar masses from 14 to 1400 kDa (keeping the
other key parameters at the standard values, including the amor-
phous layer thickness). Fig. 4c shows that the tie chain concentra-
tion was inversely proportional to the thickness of the amorphous
layer thickness (keeping the other key parameters at the standard
values, including the crystal thickness). These data are not in
accordance with the HuangeBrown model, because this model
does not predict that the ratio La/Lc has any impact on the tie chain
concentration.
Fig. 4d presents data for the concentrations of tie chains and
trapped entanglements from simulations using a constant value
(25 nm) for the critical length L ¼ 2Lc þ La. The molar mass was varied
between 14 and 1400 kDa and the temperature was constant at
417 K. The crystallinity was varied between 14 and 67% by changing
the Lc/La ratio (but always keeping L constant at 25 nm). The tie chain
concentration increased in an approximately linear fashion with
increasing Lc/La. The concentration of trapped entanglements
showed a more complex dependence on Lc/La. For the longest chains
(M ¼ 1400 kDa), the entanglement fraction decreased almost line-
arly with increasing Lc/La. The systems built with molecules of
medium molar mass (M ¼ 140 kDa) showed a parabolic function and
the system built with molecules of a molar mass of 14 kDa displayed
a linear increase in the concentration of trapped entanglements with
increasing Lc/La. The simulations indicated that the trapped
entanglements were more numerous than the tie chains in the
systems of low crystallinity and sufficiently high molar mass,
whereas systems with higher crystallinity were more dominated by
the tie chains. These simulations again indicated that the molecular
continuity (expressed in the concentrations of tie chains and trap-
ped entanglements) was sensitive to the Lc/La ratio.
3.2. Application of the model to real polyethylenes
Experimental data for 11 linear polyethylenes with different
molar masses (Mw from 20 kDa to 750 kDa and densities at room
temperature between 938 and 976 kg m3) reported by DesLauriers
and Rohlfing [14] were used as input in the simulations. The input
data used in the simulations were obtained directly from these
authors or by further calculation based on the following equations:
 
r kg m3 ¼ 1074:8  24:1  log Mw (5)
Tm ðKÞ ¼ 0:006423  r2 þ 12:72  r  5878 (6)
0
0:624  Tm
Lc ðnmÞ ¼   (7)
0 T
Tm m
 F. Nilsson et al. / Polymer 53 (2012) 3594e3601 3599

   
rc r  ra
wc ¼  (8)
r rc  ra

rc  Lc  ð1  wc Þ
La ðnmÞ ¼ (9)
ra  wc
DesLauriers and Rohlfing [14] provided molar mass data as well
as the following input data used: Tm 0 ¼ 418.5 K, r ¼ 852 kg m3 (at
a
20  C) and rc ¼ 1006 kg m3 (at Tm 0 ). Eq. (5) was obtained by linear

fitting to data reported by DesLauriers and Rohlfing [14]. Eq. (6) was
obtained by fitting a second order polynomial function to the data
of Patel et al. [12] and Mirabella and Bafna [29]. Eq. (7) is the
GibbseThomson equation. Eq. (8) relates mass crystallinity (wc) to
density (r) for polyethylene. Eq. (9) relates the thicknesses of the
crystal and amorphous layers to the mass crystallinity.
Fig. 5 shows xtc and xte as functions of molar mass using the
input data obtained from DesLauriers and Rohlfing [14]. Very low
molar mass polymers had essentially no tie chains or trapped
entanglements. Both xtc and xte increased strongly between 100
and 1000 kDa, above which the increase was only moderate (Fig. 5).
Both curves have a similar shape; the only difference being that, at
any given molar mass the concentration of trapped entanglements
was approximately twice the concentration of the tie chains.
Fig. 6 shows the effect of crystallisation temperature in the case
of a polymer with a relatively low molar mass (32.1 kDa). Input data
for Lc were obtained from experimental data provided by Barham
et al. [30], who showed that the initial crystal thickness (Lc*) had the
following dependence on the degree of supercooling (DT):
k1
L*c ðnmÞ ¼ þ k2
DT
where k1 and k2 are constants, and DT was calculated taking
0 ¼ 418.5 K for the equilibrium melting point [14]. The other
Tm
variables used in the simulations reported in Fig. 6 have the same
values as those used in Fig. 5. Fig. 6 shows that xtc showed
a pronounced decrease with increasing crystallisation temperature
at high crystallisation temperatures (>380 K), whereas after crys-
tallisation at temperatures below 380 K the polymers showed
essentially a constant xtc. The concentration of trapped entangle-
ments showed, interestingly, a gradual (but retarding) increase
with decreasing crystallisation temperature (Fig. 6).
Fig. 6. Tie-chain and entanglement fractions as functions of crystallization tempera-
ture. Data from Barham et al. [26] are used as input to the model. The entanglements
are effected more than the tie chains by crystallization temperature.
Fig. 7 shows the result of simulations based on experimental
structural data provided by Patel et al. [12] for 13 polyethylenes
including both homopolymers (linear polyethylenes) and copoly-
mers produced by single-site technology. Patel et al. [12] provided
data for Lc, La and volume crystallinity (12e69%). In the same study,
data for the rubber modulus for the same set of polymers were
presented. Our simulation results were normalized and plotted
together with the normalized rubber modulus data and corre-
(10) sponding HuangeBrown simulation data. At low vc-values, our
model data agreed very well with the rubber modulus data, but at
high vc it gave at least a better prediction than the HuangeBrown
model. The main reason for this is that the HuangeBrown type
model erroneously neglects the effect of changes in Lc/La. The
experimental data show, as depicted in Fig. 8, that even though the
critical length L diminished rapidly with vc, this was due only to
a decrease in Lc. The more important factor La decreased instead,
and this partly counteracted the increase in L. The effect would
probably be even more pronounced if the simulations could be
performed with correct molar weight distributions instead of using
only the weight average molar mass value.

Fig. 5. Tie-chain and entanglement fractions as functions of Mw. Experimental Lc and Fig. 7. Normalized tie chain fraction as a function of polymer density plotted together
La data from DesLauriers and Rohlfing [14] are used. The influence of entanglements is with the Brown model [4,5] (;), the present model (-), and the normalized exper-
approximately twice as large as the influence of tie chains. imental rubber modulus data (B) [12].
3600 F. Nilsson et al. / Polymer 53 (2012) 3594e3601
Fig. 8. The experimental in-data from Patel et al. [12] used in Fig. 7. The Lc, La and L-
values are plotted as a function of crystallinity; La decreases but Lc increases with
increasing wc.
The mechanical properties are however not only affected by the
fraction of tie chains and trapped entanglements, but also on the
degree to which the chains are stretched. For a tie chain with end
points pA and pB, n links, link length l and jjvABjj ¼ jjpB  pAjj, the
“stretch potential factor” is stc ¼ nl6/jjvABjj where the constant
6 ¼ cos(90  109.47/2) ¼ 0.8165 is chosen so that s ¼ 1 for all-trans
chains. A corresponding expression can be derived for two loops on
opposite crystal faces by first calculating the minimum distance
corresponding to two triangles formed by chains in all-trans. The
triangles will have one common point pE. The end points of the
loop-forming triangle 1 are pA and pB while the end points of
triangle 2 are pC and pD. The midpoints are denoted
pMAB ¼ (pA þ pB)/2 and pMCD ¼ (pC þ pD)/2 and the vector between
the midpoints vMM ¼ pMCD  pMAB. The distances between the
points were calculated as jjvAEjj ¼ jjpE  pAjj, jjvBEjj ¼ jjpE  pBjj,
jjvCEjj ¼ jjpE  pCjj and jjvDEjj ¼ jjpE  pDjj. The numbers of links in the
chains are n1 and n2. If the proportions between the chain lengths
are conserved, pE can be found by solving Eqs. (11) and (12)
numerically with respect to d.
pE ¼ pMAB þ d  vMM ð0  d  1Þ (11)
kvAE k þ kvBE k n
¼ 1 (12)
kvCE k þ kvDE k n2
Once the distances between pE and the loop end points have
been determined, the ‘stretch potential factor’ for entanglements
can be determined as:
6  ðn1 þ n2 Þ  l
ste ¼ (13)
ðkvAE k þ kvBE k þ kvCE k þ kvDE kÞ
Stretch simulation results for one of the polymers reported by
Patel et al. [12] are shown in Fig. 9. The following polymer
parameter values were used: Lc ¼ 9.2 nm, La ¼ 10.8 nm, M ¼ 99 kDa,
ra/rc ¼ 0.852 and T ¼ 414 K [12]. The following simulation
parameters were used: Dx ¼ Dy ¼ 100 and J ¼ 1000. A total of over
one million tie chains and entanglements were analysed. The
stretch potentials for the tie chains were in general smaller than for
the trapped entanglements, indicating that the average tie chain
has a greater impact on the mechanical properties of the polymer
than the average trapped entanglement. In this simulated polymer
system, the mean stretch potential was 6.92  0.19 for tie chains
and 8.38  0.36 for trapped entanglements.
Fig. 9. The stretch potential for tie chains and entanglements for one realistic polymer
system with polymer in-data from Patel et al. [12]. The stretch potential for trapped
entanglements is clearly larger than that for tie chains, indicating that one of the latter
has a greater impact on the mechanical properties of the polymer than one of the
former.
In the simulations reported in Fig. 9, the probability for tight
folds obtained with our model became 0.5912  0.0095. This can be
compared with values reported by other authors, e.g. 0.744
according to Kumar and Yoon [31], 0.73 by Marqusee [32],
0.50e0.80 according to Mansfield [33], 0.65 according to Hoffman
et al. [34] and 0.2 according to Flory and Yoon [35]. Note that our
value includes only the tight folds that are not generated by random
walks; a value including all tight folds would be higher, close to 0.7.
4. Conclusions
A Monte Carlo simulation method has been developed for
modelling tie chains and trapped entanglements in lamellar semi-
crystalline polymers such as polyethylene. Particularly important,
a novel numerical topology algorithm for finding entanglements
between pairs of loops was included. The simulations revealed that
the concentration of trapped entanglements is substantial and that
they must have a major impact on the stress transfer between crys-
tals. The model predicted that the impact of the amorphous thickness
is greater than that of the crystal thickness on the molecular conti-
nuity between adjacent crystals expressed by the concentrations of
tie chains and trapped entanglements. This explains why our model
gives better predictions than the HuangeBrown model in the case of
experimental rubber modulus data.
Acknowledgements
The Swedish Research Council (grant 621-2005-6138) and Bor-
ealis AB, Sweden, are acknowledged for their financial support.
References
[1] Gedde UW, Mattozzi A. Adv Polym Sci 2004;169:29.
[2] Nilsson F, Gedde UW, Hedenqvist MS. Comp Sci Tech 2011;71:216.
[3] Keith HD, Padden FJ, Vadimsky RG. J Polym Sci Polym Phys 1966;4:267.
[4] Brown N, Lu X. Polymer 1995;36:543.
[5] Huang Y-L, Brown N. J Polym Sci Polym Phys 1991;29:129.
[6] Seguela R. J Polym Sci Polym Phys 2005;43:1729.
[7] Mattozzi A, Hedenqvist MS, Gedde UW. Polymer 2007;48:5174.
[8] Meyer H, Müller-Plathe F. Macromolecules 2002;35:1241.
[9] Balijepalli S, Rutledge GC. J Chem Phys 1998;109:6523.
[10] Balijepalli S, Rutledge GC. Comp Theo Polym Sci 2000;10:103.
[11] Wilhelmi JL, Rutledge GC. J Phys Chem 1996;100:10689.
[12] Patel RM, Sehanobish K, Jain P, Chum SP, Knight GW. J Appl Polym Sci 1996;
60:749.
 F. Nilsson et al. / Polymer 53 (2012) 3594e3601 3601

[13] Strebel JJ, Moet A. J Polym Sci Polym Phys 1995;33:1969. [26] Boyd RH, Phillips PJ. The science of polymer molecules. Cambridge: Cam-
[14] DesLauriers PJ, Rohlfing DC. Macromol Symp 2009;282:136. bridge University Press; 1993.
[15] Yeh JT, Runt J. J Polym Sci Polym Phys 1991;29:371. [27] Binder K. In: Colbourn EA, editor. Computer simulations of polymers. Harlow,
[16] Guttman CM, DiMarzio EA, Hoffman JD. Polymer 1981;22:1466. UK: Longman Scientific and Technical; 1994. p. 91e129.
[17] Sumners D. J Math Chem 1987;1:1. [28] Schneider P, Eberly D. Geometric tools for computer graphics. San Francisco:
[18] Bryant J, Lacher R. Topology Proc 1988;13:1. Elsevier; 2003.
[19] Nilsson F, Hedenqvist MS, Gedde UW. Macromol Symp 2010;298:108. [29] Mirabella FM, Bafna A. J Polym Sci Polym Phys 2002;40:1637.
[20] Padding JT, Briels WJ. J Chem Phys 2001;115:2846. [30] Barham PJ, Chivers RA, Keller A, Martinez-Salazar J, Organ SJ. J Mater Sci 1985;20:
[21] Lacher RS, Bryant JL, Howard LN. J Chem Phys 1986;85:6147. 1625.
[22] Michalke W, Lang M, Kreitmeier S, Göritz D. Phys Rev E 2001;64:12801. [31] Kumar SK, Yoon DY. Macromolecules 1989;22:3458.
[23] Lang M, Michalke W, Kreitmeier S. J Comput Phys 2003;185:549. [32] Marqusee JA. Macromolecules 1986;19:2420.
[24] Flory PJ, Yoon DY, Dill KA. Macromolecules 1984;17:862. [33] Mansfield ML. Macromolecules 1983;16:914.
[25] Flory PJ. Statistical mechanics of chain molecules. Munich, Vienna, New York: [34] Hoffman JD, Guttman CM, DiMarzio EA. Faraday Disc Chem Soc 1979;68:177.
Hanser Publishers; 1969. [35] Flory PJ, Yoon DY. Macromolecules 1984;17:862.