Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Molecular Simulation

ISSN: 0892-7022 (Print) 1029-0435 (Online) Journal homepage: http://www.tandfonline.com/loi/gmos20

Molecular dynamics simulation of diffusion


of small atmospheric penetratesin
polydimethylsiloxane

Alexander Sudibjo & Douglas E. Spearot

To cite this article: Alexander Sudibjo & Douglas E. Spearot (2011) Molecular dynamics simulation
of diffusion of small atmospheric penetratesin polydimethylsiloxane, Molecular Simulation, 37:2,
115-122, DOI: 10.1080/08927022.2010.524646

To link to this article: https://doi.org/10.1080/08927022.2010.524646

Published online: 10 Feb 2011.

Submit your article to this journal

Article views: 190

Citing articles: 9 View citing articles

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=gmos20
Molecular Simulation
Vol. 37, No. 2, February 2011, 115–122

Molecular dynamics simulation of diffusion of small atmospheric penetrates


in polydimethylsiloxane
Alexander Sudibjo and Douglas E. Spearot*
Department of Mechanical Engineering, University of Arkansas, Fayetteville, AR 72701, USA
(Received 18 June 2010; final version received 28 August 2010)

Motivated by the development of a microelectromechanical (MEMS) corrosion sensor which utilises a thin polydimethylsiloxane
(PDMS) membrane as a permeable medium, molecular dynamics (MD) simulation is used to study the diffusion of CH4, N2 and O2
penetrates in uncrosslinked PDMS. Both PDMS and penetrates are modelled within the MD framework using a hybrid coarse-
grained interatomic potential; the accuracy of this model and its ability to capture structural aspects of PDMS, including bulk
densities and radius of gyration, is evaluated. Diffusion constants and activation energies are computed after reaching steady state
using simulation models with ‘long’ and ‘short’ PDMS chains (above and below the entanglement molecular weight of PDMS,
respectively). Diffusion coefficients computed using short-chain models are higher than those with long-chain models, due to the
presence of molecular entanglements in the long-chain PDMS systems. A linear correlation is found between the natural logarithm
of the diffusion constant and the activation energy for diffusion in agreement with experimentally determined phenomenological
relationships.
Keywords: molecular dynamics; diffusion; polydimethylsiloxane

1. Introduction model. The role of metallic nanoparticles on the diffusion


A microelectromechanical (MEMS) corrosion sensor is of small atmospheric molecules will be addressed in a
currently under development by Pan and Huang [1] to subsequent work.
provide an inexpensive and reliable means to detect It has been previously established that MD simulations
damage caused by corrosion on civil infrastructure. This are capable of computing accurate diffusion-related data in
sensor is both novel and timely as damage due to corrosion polymers (cf. [4 –12]). For example, Sok et al. [4] analysed
has been estimated to cost over $650 billion annually in the effect of penetrate size on the diffusion coefficient in
PDMS using a hybrid coarse-grained interatomic potential
the USA [2,3]. This expense is attributed primarily to the
and reported very good agreement with experimental
maintenance of civil infrastructure, replacing parts after
results. Tamai et al. [5] used the same force field as Sok
component failure (corrective) or prior to component
et al. to evaluate the effect of free volume distribution of
failure on a conservative maintenance schedule (pre-
various polymers on the diffusion coefficient. More recent
ventative). The proposed sensor by Pan and Huang utilises
studies have examined other aspects of diffusion, such as
a composite film with metallic nanoparticles embedded
motion of penetrate molecules in silicone polymers with
within a polydimethylsiloxane (PDMS) matrix. Small different side groups [7], a wider range of molecular
corrosive agents, such as O2 in the atmosphere or corrosive weights of diffusing species [6] and have provided
ions if the structure is immersed in fluid, will diffuse correlations between diffusion coefficients and the kinetic
through the PDMS membrane and interact with the molecular diameters of the penetrate species [12].
metallic inclusions. This interaction will induce a change Typically, an acceptable agreement is found between the
in the properties of the nanocomposite film, such as diffusion coefficients calculated using MD and exper-
electrical resistivity. Calibration is then necessary to imental data. This work extends beyond the above
correlate changes in electrical resistivity and the level of simulation efforts to examine the role of PDMS chain
corrosion of the metallic nanoparticles. As a first step length (molecular weight) on diffusion and computes the
towards calibration, it is necessary to develop a detailed pre-exponential factors (diffusion constants) and acti-
understanding of the diffusion rate of small atmospheric vation energies explicitly from the diffusion data extracted
penetrates through bulk PDMS and the mechanisms by from the MD simulation. A linear relationship is observed
which diffusion occurs. This understanding will be between the activation energy and the natural logarithm of
achieved in this work using molecular dynamics (MD) the diffusion constant, in good agreement with exper-
simulations with an uncrosslinked PDMS simulation imental observations [13].

*Corresponding author. Email: dspearot@uark.edu


ISSN 0892-7022 print/ISSN 1029-0435 online
q 2011 Taylor & Francis
DOI: 10.1080/08927022.2010.524646
http://www.informaworld.com
116 A. Sudibjo and D.E. Spearot

2. Molecular dynamics Lennard-Jones (LJ) 12-6 potential. In this work, the


MD is a method of modelling atoms as interacting point interatomic potential developed by Frischknecht and
masses in a defined space [14]. The movement of an atom Curro [16] is used to model PDMS. This model combines
or united atom is governed by an augmented version of an explicit atom class II potential for Si and O interactions
Newton’s second law of motion. Augmentation is while keeping the LJ 12-6 potential for other non-bonded
necessary to align the system with a desired temperature interactions. The class II potential allows for weaker non-
and pressure, which in this work is achieved using a bonded interactions by employing a 9-6 term rather than
Nosé – Hoover style thermostat and barostat, so that the the LJ 12-6 term. A cut-off distance of 12 Å is used for
calculations are performed in the NPT ensemble [15]. The both 12-6 and 9-6 non-bonded interactions. Both
resolved force on each atom, F k i , is derived from the interatomic potentials include coulombic contributions.
potential energy of the system, which is defined via an The model by Frischknecht and Curro [16] is used in this
interatomic potential, work because it provides a better representation of PDMS
when pressure is considered. The set of equations for the
›UðkrN Þ Frischknecht and Curro PDMS interatomic potential is
ki ¼ 2
F : ð1Þ
›kri presented in Equations (3) – (8). These equations, in
combination with Equations (1) and (2), are used to
In Equation (1), kri represents the position vector of calculate the forces on each atom or united atom in the
atom i while krN represents the positions of all N atoms in simulation. Model parameters are provided in Ref. [16].
the simulation. For molecular systems, the interatomic
potential, U, is commonly separated into two contri- U bond ðlÞ ¼ kb ðl 2 l0 Þ2 ; ð3Þ
butions,
U angle ðuÞ ¼ ka ðu 2 u0 Þ2 ; ð4Þ
UðkrN Þ ¼ U bonded þ U non-bonded
¼ U bond þ U angle þ U torsion þ U LJ þ U coulombic : U torsion ðfÞ ¼ kt ½1 þ cosðnfÞ; ð5Þ
ð2Þ  
sab 12 sab 6
U LJ ðrÞ ¼ 41ab 2 ;
The first contribution is the intramolecular bonded r r ð6Þ
term, which typically include bond stretch, angle bending
when a or b ¼ CH3 ;
and dihedral angle torsion effects within a polymer chain.
The second contribution is the intramolecular or   
sab 9 sab 6
intermolecular non-bonded term, which account for van U II ðrÞ ¼ 1ab 2 23 ;
der Waals and coulombic interactions. Combined, these r r ð7Þ
two contributions to the potential energy govern the
when a and b ¼ Si and O;
behaviour of the polymer chains in the system. The total
potential energy is taken as the non-weighted sum of qa qb
bonded and non-bonded terms for use in Equation (1). U coulombic ðrÞ ¼ : ð8Þ
4p 10 r
For computational reasons, a simplifying approach is
commonly employed by combining multiple atoms into a A harmonic form is used for bond length and bending
single point mass. This method of modelling is known as angle interactions, with bond spring stiffness kb and angle
the united atom model or coarse-grained model. For spring stiffness ka . The equilibrium bond length between
PDMS, the most common use of this method is to simplify two atoms and angle between a set of three atoms is l0 and
the methyl (CH3) side groups while the Si and O atoms are u0 , respectively. Torsional interactions are accounted for
modelled explicitly, representing a hybrid approach. For using a periodically repeating cosine function with
example, both Sok et al. [4] and Frischknecht and Curro stiffness kt and period n, respectively. Pair interactions
[16] have proposed hybrid coarse-grained interatomic within the siloxane backbone use a soft 9-6 form to better
potentials for PDMS with explicit modelling of Si and O. capture the flexibility of the PDMS chain while all other
The PDMS interatomic potential developed by Sok pair interactions use a more standard LJ 12-6 form to
et al. was used by several authors to study small molecule model non-bonded effects. The parameters sab and 1ab
diffusion [4,5]. Sok et al. [4] showed the validity of define, respectively, the shape and strength of the non-
measuring diffusion properties using this type of united- bonded interaction in each case. Penetrates in this work
atom model; their results were within an acceptable error (N2, O2 and CH4) are also modelled using a coarse-grained
range from the experimental data. They calculated a approach [4]. van der Waals interactions between the
diffusion coefficient of 2.1 £ 1025 cm2/s at 300 K, which penetrates and atoms within the PDMS chain are modelled
is 5% lower than the experimental value [17,18]. In their using the LJ 12-6 form using the Berthelot mixing rules for
model, non-bonded interactions are computed via a cross-interactions.
Molecular Simulation 117

3. Simulation methodology Melchionna et al. [15]. This equilibration step allows the
3.1 Model construction chains to relax into a more appropriate conformation while
the entire system contracts to its equilibrium density. After
The repeating unit of PDMS includes an Si – O segment of
this initial run, the temperature of the system is gradually
the chain backbone with two CH3 groups attached to each
changed to the desired temperature for the diffusion
Si atom, CH3 ½SiðCH3 Þ2 On SiðCH3 Þ3 . Experimentally,
simulation over 100 ps; simulations of diffusion are
chain lengths vary within a given sample; however, for
performed at temperatures between 225 and 400 K in
simplicity in this work, the construction of PDMS will
this work. Once the system reaches the desired
consider only chains of uniform length. To study the effect
temperature, an additional NPT run is done for 100 ps to
of chain entanglement, two different sets of chain lengths
allow the system to reach equilibrium density at the new
will be considered. The short-chain systems will consist of
temperature. Penetrates are then introduced at random in
chains with mass well below the entanglement molecular the system at the current system temperature. Since
weight of 24,500 g/mol [19], while the long-chain systems penetrates are distributed randomly, another NPT run is
will consist of chains with mass above the entanglement done for 500 ps to allow the penetrates to diffuse evenly
molecular weight. The short-chain systems are composed throughout the system, before diffusion-related data are
of n ¼ 10 repeating units with system sizes of 50, 100 and extracted. Finally, an NPT production run is done for
150 chains. The long-chain system are composed of 100 ps and data are recorded during this simulation.
n ¼ 339 repeating units each with system sizes of 10 and
20 chains. Cross-linking between molecular chains is not
considered in this work. For each diffusing species (N2, O2 3.2 Model validation
and CH4), 100 united atoms are used as the penetrate. By
Two methods are used to validate the initial conformation
keeping the number of penetrates constant, changing
of the PDMS chains prior to computing diffusion-related
system size (number and/or length of PDMS chains)
data. First, density is computed using the classical
allows for the analysis of penetrate concentration on
definition of mass per unit volume and compared with
diffusion. Finally, for each system size, three independent
experimental measurements (0.97 g/cm3 [20]). Second, the
simulations are performed with different initial random
radius of gyration of the PDMS chains is computed using
constructions of the molecular chains. Data presented are
the root mean-squared distance from each atom to the
averages of three simulation runs, unless otherwise stated
centre of mass of the system,
in the discussion.
To create the initial configuration for the PDMS D E
chains, a self-avoiding random walk algorithm is used. 1X
R2g ¼ mi ðkri 2 krcm Þ2 : ð9Þ
First, a simulation cell much larger than that necessary for M i
the number of chains to be created is divided into a 3D grid
with 1 Å spacing. A point is randomly selected on the grid In Equation (9), M is the total molecular weight of the
for the initial Si atom in the chain backbone. Adjacent system, mi is the mass of an individual atom or united atom
points to the initial point are stored in an array and one of and krcm is the position of the centre of mass of the system.
those points is chosen randomly. The chosen point is This value is used to determine the size of the molecular
checked for occupancy and if it is empty, that point is chains, normalised by the molecular mass. A high value of
added as the next point in the chain with appropriate radius of gyration indicates linear or ‘stretched-out’
chemical distinction. If the point is occupied, then another chains, while a low value indicates chains that are
point is chosen randomly from the array. This process ‘bunched-up’ in a localised mass. The experimental value
continues until the entire backbone of the chain is for the normalised radius of gyration of PDMS is reported
constructed and is then repeated to add CH3 united-atom to be (7.7 ^ 0.3) £ 1022 Å2·mol/g [21,22].
side groups on the Si atoms within the chain backbone. The densities of each size model considered in this
The entire process is repeated for each chain to be placed work with different initial constructions are evaluated to
in the simulation cell. The resulting initial configuration ascertain the statistical error range of the equilibration
will start in a very high-energy state due to an incorrect method; results are shown in Figure 1. The statistical error
conformational structure with 908 bond angles. Equili- margin of this equilibration method is found to be around
bration procedures are performed to allow the chains to 2%, with respect to the experimental density [20], which is
move into a more appropriate conformational state. an acceptable range. Radius of gyration values are also
After an initial random distribution of molecular evaluated and compared to experimental results as shown
chains is created, the system, undergoes an NPT in Figure 2. The radius of gyration is clearly dependent on
equilibration run at 300 K for 1 ns. During this run and the number of chains included in the simulation (system
all subsequent runs, the system pressure is maintained at size). Regardless, it is found that the calculation of
1 atm using a Nosé –Hoover style barostat as described by the diffusion coefficient for each material system is not
118 A. Sudibjo and D.E. Spearot

1.6 penetrates (diffusing species) [4],


50 short chains
1.4 100 short chains
150 short chains 1 d 2 1d
10 long chains D¼ kr ðtÞl ¼ k½rðtÞ 2 rð0Þ2 l: ð11Þ
1.2 20 long chains 2y dt 6 dt
Experiment
Density (g/cm3)

1.0
Equation (11) is derived by combining Equation (10)
0.8 with a statement of conservation of mass of the diffusing
0.6
species in the system. In Equation (11), kr 2(t)l is the mean-
squared displacement of the diffusing species and y is the
0.4 dimensionality of the system. With the assumption that
0.2
the examined temperature range is sufficiently narrow,
the temperature dependence of the diffusion coefficient
0.0
0 200 400 600 800 1000
can be captured via an Arrhenius relation,
Time (ps)
 
Q
Figure 1. Density evolution during equilibration at 300 K with D ¼ D0 exp 2 ; ð12Þ
RT
comparison to experimental data from [20].
where D0 is the diffusion constant, Q is the activation energy,
R is the gas constant and T is the temperature. As discussed
influenced appreciably by variations in the radius of
above, MD simulations in the work are performed three times
gyration within the range presented in Figure 2.
at different temperatures ranging from 225 to 400 K and thus
multiple values of D are computed. By fitting an exponential
3.3 Computing diffusion coefficients curve to the data (in appropriate format), the values of D0 and
The classical definition of steady-state diffusion is Q can be extracted.
provided by Fick’s first law,
4. Results and discussion
kj ¼ 2D7c; ð10Þ 4.1 Short-chain models
Mean-squared displacement data are plotted along with
where kj, the flux of the diffusing species, is proportional least-squared linear fits for every combination of diffusing
to the negative gradient of the concentration of that species, species, system size and temperature; an example of such
c. The constant of proportionality, D, is known as data is shown in Figure 3 for the case of 100 PDMS chains
the diffusion coefficient. The diffusion coefficient describes with 100 CH4 penetrates. As expected, the observed
the rate of motion of particles in the concentration of that diffusion flux increases as temperature increases, for each
species. Within MD simulations, the diffusion coefficient species studied. Diffusion coefficients are calculated from
can be computed using several methods. The most common MD using Equation (11) and are listed in Table 1 for all
method is to compute the mean-squared displacement of the short-chain simulation models for each penetrate type at

1000
2.0 225K
250K
50 short chains
275K
100 short chains 800 300K
150 short chains 325K
10 long chains 350K
1.5
<[r(t) - r(0)]2> (Å2)

20 long chains 375K


<Rg2> / M (Å -mol / g)

Experiment 600 400K


Least-square fits
2

1.0
400

0.5 200

0
0.0 0 20 40 60 80 100
0 200 400 600 800 1000 Time (ps)
Time (ps)
Figure 3. Mean-squared displacement vs. time for one
Figure 2. Radius of gyration evolution during equilibration at configuration of 100 short PDMS chains with 100 CH4
300 K with comparison to experimental data from [21,22]. penetrates at varying temperatures with least-squared fits.
Molecular Simulation 119

Table 1. Calculated diffusion coefficients at 300 K with the standard deviation obtained from different initial conformations compared
with experimental results at 308 K.

Number of chains (penetrate weight %) Diffusing species Dcalc Standard deviation Dexp [17,18]
50 (3.83%) CH4 5.127 ^0.1986
100 (1.95%) CH4 3.960 ^0.4881 2.2 ^ 0.2
150 (1.31%) CH4 3.509 ^0.2046
50 (6.50%) N2 5.725 ^0.1629
100 (3.36%) N2 5.008 ^0.3675
150 (2.27%) N2 3.507 ^0.3116 3.9
10a (1.13%) N2 2.257 ^0.2424
20a (0.57%) N2 2.187 ^0.2271
50 (7.36%) O2 5.632 ^0.0878
100 (3.82%) O2 4.340 ^0.7604
150 (2.58%) O2 3.689 ^0.3764 4.1
10a (1.29%) O2 2.115 ^0.2815
20a (0.65%) O2 2.337 ^0.2362
a
Units for all diffusion coefficients are 1025 cm2/s. Long-chain systems.

300 K. In Figure 4, the computed diffusion coefficients for stone to understanding the behaviour of PDMS in a
100 PDMS chains with 100 CH4 penetrates are plotted as a rubbery solid state after cross-linking is introduced. For
function of inverse temperature; data from all three the long-chain simulations, system sizes of 10 and 20
simulations are shown at each temperature (labelled as chains are tested. An example of diffusion in a long-chain
1 –3) to illustrate the spread in the data due to statistical system can be seen in Figure 5 for a case of 10 long PDMS
variation in the initial PDMS chain and penetrate
chains with 100 N2 penetrates. Figure 6 shows the
distributions. Equation (12) is fitted to the calculated
temperature dependence of the diffusion coefficient with
diffusion coefficients at each temperature to extract values
for D0 and Q. Note that the gas constant R is included in Arrhenius fits for all three initial distributions; data at
the calculation of D0 and Q, even though data are not 300 K are presented in Table 1. In comparison to Figures 3
plotted as a function of 1/RT in Figure 4. and 4, it is clear that the calculated diffusion coefficients
for the long-chain simulation models are lower than that of
the short-chain models for all temperatures studied. This
4.2 Long-chain models difference can be attributed to the chains in the system
The long-chain system is composed of chains with being entangled due to their length, thereby confining the
individual length greater than the entanglement molecular movement of penetrates and restricting potential diffusion
weight. At this length, the system exhibits significantly jumps in the system. A direct comparison between short-
increased viscosity due to the presence of entanglements chain and long-chain systems is presented in Figures 7 and
and reduced chain self-diffusion. Understanding the 8 for N2 and O2 penetrates for each simulation model.
behaviour in this highly viscous phase provides a stepping
600
2.1e-4
225K
100 short CH41 250K
1.8e-4 100 short CH42 500 275K
100 short CH43 300K
325K
1.5e-4 100 short CH4fit 1
400 350K
<[r(t) - r(0)]2> (Å2)

100 short CH4fit 2 375K


100 short CH4fit 3 400K
D (cm2/s)

1.2e-4 Least-square fits


300
9.0e-5

200
6.0e-5

3.0e-5 100

0.0 0
0.0020 0.0025 0.0030 0.0035 0.0040 0.0045 0.0050 0 20 40 60 80 100
–1
1/T (K ) Time (ps)

Figure 4. Diffusion coefficient as a function of temperature Figure 5. Mean-squared displacement vs. time for one
with Arrhenius fits for three different initial configurations of 100 configuration of 10 long PDMS chains with 100 N2 penetrates
short PDMS chains and 100 CH4 penetrates. at varying temperatures with least-squared fits.
120 A. Sudibjo and D.E. Spearot

1.2e-4 5. Discussion
10 long N21
10 long N22 In comparison with prior MD simulation efforts which
1.0e-4 10 long N23
10 long N2 avg
employed PDMS models with chain lengths below the
8.0e-5
10 long N2 avg fit entanglement molecular weight [4 – 11], the diffusion
coefficients computed at 300 K in this work are slightly
D (cm2/s)

6.0e-5 higher than those previously reported. Discrepancies


between diffusion coefficients can be attributed to
4.0e-5 differences in the molecular weight of the simulation
models, differences in concentration of the penetrate
2.0e-5
species and differences in the way that pressure is (or is not
0.0
in the case of NVT) controlled during the MD simulation.
0.0020 0.0025 0.0030 0.0035 0.0040 0.0045 0.0050 Previous MD studies of diffusion of CH4 in PDMS
–1
1/T (K ) have yielded a diffusion coefficient of 0.57 £ 1025 cm2/s
by Tamai et al. [5] and 2.1(^ 0.8) £ 1025 cm2/s by Sok
Figure 6. Diffusion coefficient as a function of temperature
with Arrhenius fits for three different initial configurations of 10 et al. [4]. Both Sok et al. and Tamai et al. used a system
long PDMS chains and 100 N2 penetrates. size of five PDMS chains with n ¼ 30 and one penetrate
molecule. The longer chain length (than the short-chain
models in this work) reduces the ability of the penetrate
molecule to diffuse through the system, because the local
mobility of the PDMS chains is constrained [13]. Previous
3.0e-4
50 short N2 avg
MD studies of N2 diffusion through PDMS have reported a
2.5e-4
100 short N2 avg diffusion coefficient of 3.74 £ 1025 cm2/s by Jawalkar and
150 short N2 avg
10 long N2 avg
Aminabhavi [9] and 1.2 £ 1025 cm2/s by Charati and Stern
2.0e-4 20 long N2 avg [7]. The studies by Jawalkar and Aminabhavi, and Charati
50 short N2 avg fit
and Stern also reported diffusion coefficient values for O2
D (cm2/s)

100 short N2 avg fit


1.5e-4 150 short N2 avg fit of 3.7 £ 1025 and 1.8 £ 1025 cm2/s, respectively. Here,
10 long N2 avg fit
20 long N2 avg fit discrepancies can be attributed to the system size and
1.0e-4 penetrate concentration. Jawalkar and Aminabhavi used a
unit cell of 15 Å width and Charati and Stern used a unit
5.0e-5
cell of 22 Å width with eight PDMS chains and four
0.0
penetrate molecules. The higher penetrate concentration
0.0020 0.0025 0.0030 0.0035 0.0040 0.0045 0.0050 used in this work may provide a higher driving force for
1/T (K–1) diffusion leading to increased diffusion rates [23]; this is
Figure 7. Comparison between short-chain and long-chain
evident in this work from data presented in Table 1.
systems with 100 N2 penetrates. Finally, most of the previous studies of MD diffusion have
used the canonical NVT ensemble, with the exception of
Sok et al. [4] who used the NPT ensemble. In the NVT
ensemble, the volume and temperature of the system are
held constant and thus there is no guarantee that the system
3.0e-4
pressure is consistent between this work and previous
50 short O2 avg
2.5e-4
100 short O2 avg efforts, making direct quantitative comparison
150 short O2 avg challenging.
10 long O2 avg
2.0e-4 20 long O2 avg In agreement with the experiment, the MD simulation
50 short O2 avg fit
results in this study show that CH4, N2 and O2 exhibit only
D (cm2/s)

100 short O2 avg fit


1.5e-4 150 short O2 avg fit minor differences in diffusion characteristics, as shown in
10 long O2 avg fit
20 long O2 avg fit Table 1. This is true for the diffusion of comparatively
1.0e-4
sized small molecules in highly permeable polymers
5.0e-5
[13,17]. To quantitatively illustrate the accuracy of the MD
simulations in this work in comparison with experiment,
0.0 activation energies and diffusion constants are computed
0.0020 0.0025 0.0030 0.0035 0.0040 0.0045 0.0050
–1
for every combination of system size and diffusing species
1/T (K )
using Equation (12). For amorphous rubbery polymers,
Figure 8. Comparison between short-chain and long-chain a linear relationship has been observed experimentally
systems with 100 O2 penetrates. between the natural logarithm of the diffusion constant and
Molecular Simulation 121

–4 confine the local motion of the PDMS chains and reduce


N2 penetrates
the number of available diffusion jumps for the penetrates.
–6
O2 penetrates In a unique contribution in this work, activation
CH4 penetrates
–8
Stem et al. [24] energies and diffusion constants are calculated from MD
Zheng et al. [13] using an Arrhenius relation to characterise the relationship
–10 Current work
In(Do) (m2/s)

between the diffusion coefficient and temperature. A linear


–12
trend is identified between the natural logarithm of the
–14 diffusion constant and the activation energy, in good
agreement with the proposed phenomenological relation-
–16
ship of Zheng et al. [13]. This provides quantitative
–18 indication of the accuracy of the MD simulations
–20 presented in this work. Future studies will analyse
5000 10000 15000 20000 25000 diffusion characteristics in a nanocomposite system
Q (J/mol)
composed of a PDMS matrix and metallic nanoparticle
Figure 9. Correlation between the natural logarithm of the inclusions to align the simulations with experiments aimed
diffusion constant and the activation energy for diffusion at corrosion sensor development. Cross-linking agents will
showing good agreement with experimentally observed linear also be considered in future studies to evaluate their effect
relationship [13] and experimental data [24]. on diffusion. It is also recognised that thin film properties
of polymers will vary from bulk properties [25 – 28]. As
the activation energy, known as the ‘compensation effect’ such, future studies will also consider surface reconstruc-
[13]. This relationship has been used historically to relate tion and segregation to advance the development of the
experimental data for the diffusion of different penetrates corrosion sensor.
in the same polymer or for the same penetrate with
different polymers (with the same rubbery or glassy
characteristics), implying a universal relationship between
the diffusion constant and the activation energy. Figure 9 Acknowledgements
shows that the MD simulations in this work capture This work was supported by the National Science Foundation
this linear trend and provide the relationship, under Grant No. CMMI#0800718 with additional support
ln D0 ¼ 219:0428 þ 0:3821 £ 1023 Q. By comparison, provided by the University of Arkansas. MD simulations were
performed on the Star of Arkansas provided in part by the
for small molecule diffusion (which includes O2, N2 and National Science Foundation under Grant No. MRI#072265.
CH4) through PDMS Zheng et al. [13] proposed, ln D0 ¼
218:2673 þ 0:2740 £ 1023 Q to describe a range of
amorphous rubbery polymers; this linear relationship
is shown for comparison in Figure 9. Also included in References
Figure 9 are experimental data for diffusion of CO2, CH4 [1] F. Pan and A. Huang, Investigation of electrochemical transduction
and C3H8 through PDMS from Stern et al. [24]. mechanism of metal particle polymer composites for the develop-
ment of MEMS-based corrosion sensor, ASME International
Mechanical Engineering Congress and Exposition, Orlando, FL,
2009.
6. Conclusions [2] P.R. Roberge, Handbook of Corrosion Engineering, McGraw-Hill,
New York, 2000.
In this work, MD simulations are used to compute the [3] W.D. Callister, Material Science and Engineering: An Introduction,
diffusion coefficient, diffusion constant and activation Wiley, New York, 1994.
[4] R.M. Sok, H.J.C. Berendsen, and W.F. van Gunsteren, Molecular
energy for CH4, N2 and O2 penetrates in uncrosslinked dynamics simulation of the transport of small molecules across a
PDMS. The chain length of PDMS is varied to study the polymer membrane, J. Chem. Phys. 96 (1992), pp. 4699–4704.
effect of chain entanglements on diffusion. Diffusion [5] Y. Tamai, H. Tanaka, and K. Nakanishi, Molecular simulation of
permeation of small penetrants through membranes. 1. Diffusion
coefficients are calculated using the mean-squared coefficients, Macromolecules 27 (1994), pp. 4498– 4508.
displacement method. Diffusion coefficients computed in [6] T. Li, D.O. Kildsig, and K. Park, Computer simulation of molecular
this work for short-chain models are higher than previous diffusion in amorphous polymers, J. Control. Release 48 (1997),
pp. 57–66.
MD diffusion studies. The differences are attributed to the [7] S.G. Charati and S.A. Stern, Diffusion of gases in silicone polymers:
molecular weight of the chains, the physical size of the Molecular dynamics simulations, Macromolecules 31 (1998),
system which impacts the concentration of the diffusing pp. 5529– 5535.
[8] P. Padilla and S. Toxvaerd, Self-diffusion in n-alkane fluid models,
species and differences in equilibration procedures. J. Chem. Phys. 94 (1991), pp. 5650–5654.
Comparison of diffusion coefficients between short-chain [9] S.S. Jawalkar and T.M. Aminabhavi, Molecular dynamics
and long-chain systems shows a decrease in the diffusion simulations to compute diffusion coefficients of gases into
polydimethylsiloxane and poly{(1,5-naphtalene)-co-[1,4-durene-
coefficients as the chain length is increased. This trend is 2,2 0 -bis(3,4-dicarboxyl phenyl)hexafluoropropane diimide]},
due to entanglements in the long-chain system which Polym. Int. 56 (2007), pp. 928–934.
122 A. Sudibjo and D.E. Spearot

[10] J. Liu, D. Cao, and L. Zhang, Molecular dynamics study on [20] J. Mark, Polymer Data Handbook, Oxford University Press, New
nanoparticle diffusion in polymer melts: A test of the Stokes– York, 1999.
Einstein law, J. Phys. Chem. C 112 (2008), pp. 6653–6661. [21] H. Ma and J. Xu, Mean-square radius of gyration of poly-(di-
[11] B. Li, F. Pan, Z. Fang, L. Liu, and Z. Jiang, Molecular dynamics methylsiloxane) chain with side groups, Polym. J. 26 (1994),
simulation of diffusion behavior of benzene/water in PDMS- pp. 779–785.
calix[4]arene hybrid pervaporation membranes, Ind. Eng. Chem. [22] J.G. Zilliox, J.E.L. Roovers, and S. Bywater, Preparation and
Res. 47 (2008), pp. 4440– 4447. properties of polydimethylsiloxane and its block copolymers with
[12] B. Prathab and T.M. Aminabhavi, Molecular modeling study on
styrene, Macromolecules 8 (1975), pp. 573–578.
surface, thermal, mechanical and gas diffusion properties of
[23] J. Brandup, E.H. Immergut, and E.A. Grulke, Polymer Handbook,
chitosan, J. Polym. Sci. B: Polym. Phys. 45 (2007), pp. 1260–1270.
[13] J.M. Zheng, J. Qiu, L.M. Madeira, and A. Mendes, Polymer 4th ed., Wiley, New York, 1999.
structure and the compensation effect of the diffusion pre- [24] S.A. Stern, V.M. Shah, and B.J. Hardy, Structure –permeability
exponential factor and activation energy of a permeating solute, relationships in silicone polymers, J. Polym. Sci. B 25 (1987),
J. Phys. Chem. B 111 (2007), pp. 2828–2835. pp. 1263– 1298.
[14] I.M. Torrens, Interatomic Potentials, Academic Press, New York, [25] B. Prathab, T.M. Aminabhavi, R. Parthasarathi, P. Manikandan, and
1972. V. Subramanian, Molecular modeling and atomistic simulation
[15] S. Melchionna, G. Ciccotti, and B.L. Holian, Hoover NPT dynamics strategies to determine surface properties of perfluorinated
for systems varying in shape and size, Mol. Phys. 78 (1993), homopolymers and their random copolymers, Polymer 47 (2006),
pp. 533– 544. pp. 6914– 6924.
[16] A.L. Frischknecht and J.G. Curro, Improved united atom force field [26] B. Prathab, T.M. Aminabhavi, and V. Subramanian, Computation of
for poly(dimethylsiloxane), Macromolecules 36 (2003), surface energy and surface segregation phenomena of perfluori-
pp. 2122–2129. nated copolymers and blends: A molecular modeling approach,
[17] Y. Ichiraku, S.A. Stern, and T. Nakagawa, An investigation of the
Polymer 48 (2007), pp. 417–424.
high gas permeability of poly(1-trimethylsilyl-1-propyne), J. Memb.
[27] B. Prathab and T.M. Aminabhavi, Atomistic simulations to compute
Sci. 34 (1987), pp. 5 –18.
[18] R.D. Raharjo, B.D. Freeman, D.R. Paul, G.C. Sarti, and E.S. surface properties of poly(n-vinyl-2-pyrrolidone) (PVP) and blends
Sanders, Pure and mixed gas CH4 and n-C4H10 permeability and of PVP/chitosan, Langmuir 23 (2007), pp. 5439–5444.
diffusivity in poly(dimethylsiloxane), J. Memb. Sci. 306 (2007), [28] S.S. Jawalkar, S.K. Nataraj, A.V. Raghu, and T.M. Aminabhavi,
pp. 75– 92. Molecular dynamics simulations on the blends of poly(vinyl
[19] L.H. Sperling, Introduction to Physical Polymer Science, Wiley, pyrrolidone) and poly(bisphenol-A-ether sulfone), J. Appl. Polym.
Hoboken, NJ, 2006. Sci. 108 (2008), pp. 3572–3576.

You might also like