J Mol Model (2017) 23: 305
DOI 10.1007/s00894-017-3472-z
REVIEW
Molecular modeling studies of structural properties of polyvinyl
alcohol: a comparative study using INTERFACE force field
Lukasz Radosinski 1 & Karolina Labus 1
Received: 19 April 2017 / Accepted: 10 September 2017 / Published online: 5 October 2017
# Springer-Verlag GmbH Germany 2017
Abstract Polyvinyl alcohol (PVA) is a material with a variety
of applications in separation, biotechnology, and biomedicine.
Using combined Monte Carlo and molecular dynamics tech-
niques, we present an extensive comparative study of second-
and third-generation force fields Universal, COMPASS,
COMPASS II, PCFF, and the newly developed INTERFACE,
as applied to this system. In particular, we show that an
INTERFACE force field provides a possibility of composing
a reliable atomistic model to reproduce density change of PVA
matrix in a narrow temperature range (298–348 K) and calcu-
late a thermal expansion coefficient with reasonable accuracy.
Thus, the INTERFACE force field may be used to predict me-
chanical properties of the PVA system, being a scaffold for
hydrogels, with much greater accuracy than latter approaches.
Keywords Polyvinyl alcohol . Structural properties .
Molecular modeling . Hydrogel
Introduction
Polyvinyl alcohol (PVA), developed in 1924 by Hermann and
Haehnel [1], is one of the oldest synthetic compounds used for
hydrogels manufacturing [2, 3]. Due to several useful proper-
ties, such as hydrophilicity, permeability, biodegradability, and
excellent biocompatibility, PVA in a gel form is extensively
* Lukasz Radosinski
lukasz.radosinski@pwr.edu.pl
1
Division of Bioprocess and Biomedical Engineering, Faculty of
Chemistry, Wroclaw University of Science and Technology, Norwida
4/6, 50-370 Wroclaw, Poland
studied for a wide variety of medical and pharmaceutical ap-
plications [4–6]. A further relevant application of this polymer
is the formation of PVA-based materials intended for use in
different separation processes including ultrafiltration [7, 8],
nanofiltration [9], pervaporation [10, 11], and reverse osmosis
[12].
Mechanical, transport, and adsorption properties of PVA
materials depend on many factors such as water swelling de-
gree, crosslinking degree, or temperature. The understanding
of the functional dependence between the aforementioned var-
iables has a crucial impact on the design of a given polymeric
matrix for strictly specified applications. So far, most studies
have been focused on an experimental approach, with only a
little theoretical research [13, 14]. Therefore it would be de-
sired if, instead of long-lasting and laborious experimental
studies, we would be able to use a reliable computational tool
to design polymer and hydrogel networks of demanded prop-
erties and to simulate the potential behavior of these materials
under real process conditions. This could contribute to the
substantial reduction of the experimental time to the required
minimum as well as to significant decrease in the production
costs of such materials. Such approach, namely numerical
simulations, has already wide range of applications in medical
and chemical industry [15, 16].
Atomistic modeling is a very powerful tool for studying
structural and transport properties of polymer matrices.
There are numerous computational studies of polymeric sys-
tems including PVA, where the temperature dependence of
various mechanical and structural parameters was of a princi-
pal interest, like glass transition temperature [13, 14].
However, most of them are focused on a wide temperature
range extending from room temperature to nearly 600 K [13,
14]. Our overriding goal is to study the properties of PVA-
based hydrogels as semipermeable matrix for enzymes immo-
bilization. In order to study such complicated material, the
305 Page 2 of 9
major concern is a proper force field model reproducing fun-
damental properties of the dry PVA matrix. Furthermore, since
most of the enzymes operate efficiently under the rather nar-
row temperature interval (298–333 K), we have to verify
whether it is possible to track the changes of chosen observ-
ables within a narrow temperature interval using relatively
small temperature increments. This is a complex task since
small change in the value of the observable is challenged by
large fluctuations due to relatively small number of atoms in
the system. Furthermore, the possibility of adsorption of PVA
matrix on a solid state, addition of different solutes, requires a
proper transferability of the chosen force field model. The
appropriate choice of the force field is of special importance
in the process of creating a reliable molecular model of
crosslinked PVA matrix. In the following studies we will be
concerned with more complex systems including water mol-
ecules and thus hydrogel structure based on the PVA.
The aim of this work is to test molecular dynamics [17] and
Monte Carlo [18] methodologies using the newly introduced
third-generation INTERFACE force field [19] against other
popular second- and third-generation force fields like
Universal [20], COMPASS [21], COMPASS II [21], and
PCFF [20] to predict structural and transport properties of a
PVA amorphous system. The INTERFACE force field [19], to
the best of the author’s knowledge, has never been applied to
study this material. The paper is divided into two parts: in the
first part, we present results of atomistic simulations of PVA
matrix using the aforementioned force fields concerning de-
pendence of the density of PVA on its temperature and its
thermal expansion coefficient: in the second part, we analyze
the resolution of the method.
Calculations
Materials and methods
In this study, we employ classical molecular dynamics and
Monte Carlo algorithms as implemented in BIOVIA
Materials Studio 8.0 Forcite package. Due to the fact that
PVA matrix is a highly amorphous system, the initial config-
uration was created in multiple steps. First, an atomistic model
of a single PVA unit was created and then appropriately mul-
tiplied in order to create PVA atactic polymer consisting of
200 PVA units (Fig. 1).
Then, using a specific force field and the Amorphous
Cell module in Materials Studio 8.0, we constructed an
amorphous PVA matrix with periodic boundary condi-
tions. The protocol involves Monte Carlo search of the
torsion angles between PVA single units and subsequent
minimization in a given force field. This procedure en-
sures that resultant structure is in energetic minimum and
thus provides reliable initial system for further dynamical
J Mol Model (2017) 23: 305
calculations. The final density of the matrix was origi-
nally set to 1.3 g.cm−3. The detailed parameters of the
construct are presented in Tables 1, 2, and 3.
The equations of motion are being integrated using the
Velocity Verlet algorithm using a 1-fs time step. The system is
highly amorphous and has a large number of degrees of free-
dom thus it requires an equilibration protocol that prevents the
system to occupy a high-energy metastable state. Thus we have
derived the equilibration protocol that includes heating up and
careful cooling down. First, the system is heated up to 600 K in
the NPT canonical ensemble using the Nose–Hoover thermo-
stat and Andersen barostat. The system is then equilibrated for
300 ps and then gradually cooled down by 50 K in each step. In
each run, the temperature and pressure are equilibrated for 50 ps
until the next step begins. Upon reaching the desired final tem-
perature, i.e., 298 K, the system is equilibrated for another
500 ps in NPT ensemble and then for another 500 ps in NVT
ensemble. After preparation of the initial system, the polymer is
then heated up from 298 K to 348 K by a 10-K increment.
During the equilibration process, the fluctuations of the
temperature, density, and pressure are measured. We state
that the system is equilibrated once the mean of standard
deviation around the mean of the density of the last 20 ps
σ20 is constant. The standard deviation of the density is
defined as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n  2 .
σ20 ¼ ∑ ρ*i −ρ* ðn−1Þ
i¼1
where n – is the number of frames during the last 20 ps (20), ρ*i
– current density, ρ* – mean density in the 20 ps run.
This criterion is fulfilled if the standard deviation of the
σ20ðσÞ in the time interval Δt is zero. Hence, we verify the
following hypothesis:
H0 : σ ¼ 0
Ha : σ ≠0
with two-tail z-test and Δt = 20 ps. This model is based on
the assumption that the system in thermodynamics equilibrium
should be described by stationary thermodynamic distribution.
It is the fundamental assumption behind the ergodic hypothesis
allowing to relate the thermodynamical averages over ensem-
bles with averages over time. Using this procedure, the system
was considered equilibrated in temperature 298 K after 68 ps,
308 K 69 ps, 318 K 71 ps, 328 K 71 ps, 338 K 73 ps, and 348 K
after 75 ps. The production phase lasted 50 ps for each temper-
ature in the NVT ensemble.
Force fields
In this study, we use five force fields (FF): Universal,
COMPASS, COMPASS II, PCFF, and INTERFACE.
J Mol Model (2017) 23: 305 Page 3 of 9 305

Fig. 1 Single PVA unit (left) and

atomistic model of the amorphous
PVA matrix (right) with periodic
boundary conditions (denoted by
white box). The red indicates the
oxygen, gray indicates carbon,
and white indicates hydrogen
atoms

Universal force field COMPASS force field

The Universal force field has purely diagonal terms based on the The Condensed-phase Optimized Molecular Potential for
hybridization of the given atom. It includes a hybridization- Atomistic Simulation Studies (COMPASS) force field is
derived atomic radii term, the bond term expressed by harmonic an ab initio-derived force field optimized for both gas
2 k ðr−r0 Þ term where r0 is the bond equilibrium length and k is
1 2 phase, liquid phase, and condensed phase properties.
an appropriate force constant; bond angle contribution includes The functional form of the force field is of the form:
triple term Fourier cosine series of the form
3
E θ ¼ K ijk ∑ C n cosðnθÞ; the dihedral angle term is of the form
i¼0
n   E ¼ E bond þ E angle þ E oop þ E torsion þ E cross þ Eelec þ E LJ
E ϕ ¼ K ijkl ∑ C n cos nϕijkl ; the inversion term is defined as
i¼0
3  
E ω ¼ ∑ C n cos nωijkl . The nonbonded interactions include
   including contributions of respectively of bond
σ 12 
i¼0
σ 6 stretching, angle bending, out-of-plane angle contribu-
van der Waals term E vdW ¼ ϵij −2 rij þ rij and electro-
  tion, torsion, cross-coupling, electrostatic and van der
qi q j
static coupling Eel ¼ 332:0637 ϵrij . Waals of the form:

4 4 3  
E bond ¼ ∑ ∑ k i ðb−b0 Þi Eangle ¼ ∑ ∑ Ai ðθ−θ0 Þi E torsion ¼ ∑ ∑ V i 1−cosðiϕ−ϕ0 Þ2 E oop
b i¼2 θ i¼2 ϕ i¼1
 0 0  0 0
¼ ∑ k χ χ E cross ¼ ∑ ∑ F bb ðb−b0 Þ b −b0 þ ∑ ∑ F θθ0 ðθ−θ0 Þ θ −θ0 þ ∑ ∑ F bθ ðb−b0 Þðθ−θ0 Þ
2
χ b b0 θ θ0 b b

3 3 3  0 
þ ∑ ∑ ∑ F bϕ V i cosðiϕÞ þ ∑ ∑ ∑ F bϕ ðb−b0 ÞV i cosðiϕÞ þ ∑ ∑ ∑ F b0 ϕ b −b0 0 V i cosðiϕÞ
b ϕ i¼1 b ϕ i¼1 b ϕ i¼1

3 0
 0 0
þ ∑ ∑ ∑ F θϕ ðθ−θ0 ÞV i cosðiϕÞ þ ∑ ∑ ∑ k ϕθθ ðθ−θ0 Þ θ −θ0 E el
θ ϕ i¼1 ϕ θ θ0
 9  6 !
qi q j σij σij
¼∑ E LJ ¼ ∑ ∈ij 2 −3
i; j ∈r ij i; j r ij rij
305 Page 4 of 9 J Mol Model (2017) 23: 305

Table 1 Parameter values used during design of the amorphous PVA Table 3 Parameter values used during design of the amorphous PVA
matrix using Amorphous Cell as implemented in BIOVIA Materials matrix using Forcite module as implemented in BIOVIA Materials Studio
Studio 8.0 8.0 in optimization procedures

Amorphous cell Forcite optimization

Parameter name Value Parameter name Value

Total energy difference 0.001 kcal.mol−1 Total energy difference 0.001 kcal.mol−1
Force difference 0.5 kcal.mol−1.Å−1 Force difference 0.5 kcal.mol−1.Å−1
Temperature 298 K Displacement difference 5.00E-05 Å
Loading steps 1000 Max iteration 1.00E04
Electrostatic Atom based Electrostatic Atom based
Van der Waals Atom based Van der Waals Atom based
Cutoff distance 15.5 Å Cutoff distance 15.5 Å
Spline width 1Å Spline width 1Å

functional form with harmonic and triple Fourier terms for

where b , θ , ϕ and χ denote bond lengths, valence angles, bonded interactions and 9–6 Lennard–Jones potential:
torsion angles, and out-of-plane angles, respectively.
4  i 4
E ¼ ∑ ∑ K ib rb −r0;b þ ∑ ∑ Aθ;i ðθ−θ0 Þi
COMPASS II force field b i¼1 θ i¼2

3  
The COMPASS II force field is a significant enhancement of the þ ∑ ∑ V i 1−cosðiϕ−ϕ0 Þ2
ϕ i¼1
original COMPASS force field by extending the number of force
field types by over twice (from 3856 to 8420) and revision of 1 qi q j
þ ∑ k χ χ2 þ E cross þ ∑
existing parametrization by combination with the B3LYP/6- χ 4π∈0 ∈r ij rij
31G(d,p) level of theory [18]. In this way, COMPASS II covers  9  6 !
σij σij
parametrization for number of polymers, ionic liquids, and drugs. þ ∑ ∈ij 2 −3
ij rij rij
PCFF force field

The Polymer Consistent Force field (PCFF) is a CFF91-derived INTERFACE force field
force field created to reproduce experimental results of poly-
mers and organic molecules. It is useful for polycarbonates, The INTERFACE is an extension of the PCFF force field
melamine resins, polysaccharides, other polymers, organic using same functional form, however the parametrization
and inorganic materials, about 20 inorganic metals, as well as has been majorly improved. The interesting property of the
for carbohydrates, lipids, and nucleic acids. It uses common INTERFACE FF is its transferability giving ability to study
organic–inorganic systems. Furthermore, it was optimized to
Table 2 Parameter values used during design of the amorphous PVA study phenomena in various pH. Thus, it is of special interest
matrix using Forcite module as implemented in BIOVIA Materials Studio in the case of study of hydrogel matrices due to possibility
8.0 in molecular dynamics calculation
analyzing not only hydrogels but also complex solid-state hy-
Forcite dynamics drogel pH-dependent interfaces. The detailed discussion about
parametrization and properties of the INTERFACE force field
Parameter name Value may be found in [19].
Andersen barostat cell time constant 1.0 ps
Andersen barostat decay constant 0.1 ps
Integration tolerance 5.00E + 04 kcal.mol−1 Results
Nose–Hoover thermostat Q ration 0.01
Electrostatic Atom based Thermal expansion coefficient
Van der Waals Atom based
Cutoff distance 15.5 Å In our considerations, we have chosen density of the dry mass
Spline width 1Å and isobaric thermal coefficient of the PVA as the principal indi-
cator of a force field validity. This is a crucial step towards
J Mol Model (2017) 23: 305 Page 5 of 9 305

Table 4 Density and relative error between results of the simulation and experimental value obtained for simulations using different force fields

Force field Universal COMPASS COMPASS II PCFF INTERFACE Experimental result [22]

Density [g.cm−3] 0.887 1.248 1.245 1.235 1.246 1.293

Relative error [%] 31.4 3.49 3.73 4.47 3.62 0

developing more detailed theories relating structural and chemi- In the next step of our analysis, we investigated the changes
cal properties of this complicated amorphous matrix. of the density with the temperature of the system. Since in our
In the first part of our calculations, we have compared how study the crucial factor is the temperature resolution of the
different force fields reproduce the density of the dry mass of model, we studied the change of temperature in 298–348 K
PVA system. According to experimental results, the density of regime with a 10-K temperature increment. The mean isobaric
the PVA dry mass is 1.293 g.cm−3 [22]. The comparison of thermal coefficient of the PVA in the temperature range 273–
obtained equilibrium densities are given in Table 4. 318 K is 7.10−5 1/K [20]. The isobaric thermal coefficient is
As one can observe, all force fields except the Universal are defined as:
in acceptable agreement with the experiment however the
COMPASS and INTERFACE force fields have the smallest  
1 ∂V
relative error. This can be explained if we analyze the contribu- αp ¼
V ∂T p
tion of non-bonded terms predicted by each force field as pre-
sented in Fig. 2. It can be noted that the major contribution to the
overall non-bonded energy is due to electrostatic interactions. In our simulation, we calculate the density of the system.
The Universal force field largely overestimates this term in com- Using the fact that ρ = M/V, where M is the total mass of the
parison with other force fields. The PCFF force field on the other system and V its volume, the equation can be easily trans-
hand, although the electrostatic energy contribution to the total formed to:
energy is almost exactly the same as predicted by the  
∂lnρ
COMPASS, COMPASS II, and INTERFACE FF, the van der αp ¼ − :
∂T p
Waals coupling is almost twice as in other FF. This results in a
slight mismatch in the density prediction. All third-generation
Thus, by measuring the slope of the lnρ(T) function, one
force fields give reasonable predictions of the PVA density,
can calculate αP. The density ρ in the current temperature is
however COMPASS and INTERFACE are in the best agree-
being calculated as the mean value of the densities ρ*i in the
ment in terms relative error with experimental result.
single run for the current temperature.
In order to confirm the equilibration procedure, we also
performed longer 1.6-ns runs (Fig. 3). The average density 1 N *
value over 1 ns is 1.2454 g/cm3 and the difference from the ρ¼ ∑ρ ;
N i¼1 i
shorter run is neglectable.
where N is the number of frames in the current run evaluated
using aforementioned procedure. In our numerical experi-
ment, each frame was 1 ps long.

Fig. 2 Non-bonded energy contributions terms of the PVA matrix Fig. 3 A 1.6-ns trajectory of the density of the PVA system in 298 K
(standard conditions) as predicted by different force fields using the INTERFACE force field
305 Page 6 of 9 J Mol Model (2017) 23: 305

Table 5 Regression parameters for PVA density dependence presented where

in Figs. 3 and 4 S lnρ
S β1 ¼ pffiffiffiffiffiffiffiffi
Force field Slope of regression R2 SSE  n−1
 ST 
line ln훒(T) ∑i¼1 T i −T lnρi −lnρ
n

β1 ¼  2
Universal 7.281e-06 0.1871 2.8165–04 ∑ni¼1 T i −T
COMPASS −7.002E-06 0.4569 1.999–04
COMPASS II 1.2913e-05 0.608 2.6959E-04
PCFF 5.449e-06 0.3079 2.1111E-04 And T ; lnρ are mean values of T and lnρ respectively and S T
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
INTERFACE −2.96E-05 0.7385 4.5945–04
∑ni¼1 ðT i −T Þ
2

¼ n−1 is the standard deviation of the temperature

and t(0.95, df) is the value of t distribution function. In our case,
n = 6 and number of degrees of freedom for the t distribution
Although the method is quite simple, it is limited by ther- function is df = n − 2 = 4.
mal fluctuations of the given observable, in this case the den- The results are given in Figs. 4, 5, and 6, and Table 5. Due
sity. The problem in general is quite complex to investigate. to small sample size (6) the 95% confidence interval for lnρ(T)
From fundamental thermodynamics, one knows that standard has been calculated using t distribution formula:
deviation of any observable of the system being in thermody-
namic equilibrium is proportional to the inverse of the square vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 ffi
u 2 1
root of number of particles, namely: u
u 1 B T −T C
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi lnρðT Þ  t ð0:95;df Þ * S lnρ u
tn þ @ A;
< A2 > − < A>2 2 ðn−1ÞS 2T
σA ¼ ∼ pffiffiffiffi
< A2 > N
where Slnρ(T) is given by
In atomistic simulations, where N is relatively small, the σA
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
is quite significant, thus reducing the resolution of the method. u  2
u n
In order to investigate this problem, we have used statistical t ∑i¼1 lnρi −lnρ
inference in order to verify whether the statistical data provid- S lnρ ¼
n−1
ed by the molecular simulations are sufficient to compute
isobaric thermal expansion coefficient. Thus, as a frame of
reference, we have chosen a relatively small system (1402 In order to indicate variability of the collected data to cal-
atoms) and relatively small temperature increment (10 K) to culate each density point, the error bars have been defined as:
compare the resolution of the force fields. In the first part, we lnρ  S *lnρ .
calculated αp using a simple linear regression model. The The S *lnρ is the standard deviation of density sample col-
predictions of αp for a given force field are given in Tables 5 lected in the numerical experiment during single run and its
and 6. The 95% confidence margin for the αp is calculated defined as:
using t distribution as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
t ð0:95;df Þ ∑ni¼1 lnρ*i −lnρ
αp  S *lnρ ¼
S β1 N −1

Table 6 Thermal expansion coefficient (αp) obtained from simulation using different force field compared with experimental value

Force field Universal COMPASS COMPASS II PCFF INTERFACE Experimental value [22]

αp[K−1] −7.281E-06 7.002E-06 −1.2913E-05 −5.449E-06 2.96E-05 7.00E-05

95% Confidence margin ±3.87E-07 ±1.95E-07 ±2.64E-07 ±2.08E-07 ±4.49E-07 –
J Mol Model (2017) 23: 305 Page 7 of 9 305

Fig. 4 Dependence of the natural

logarithm of the density of PVA
matrix with error bars as a
function of temperature together
with linear regression (solid red
line) and confidence intervals
(dashed black lines) obtained in
the simulation using a Universal,
b PCFF, and c COMPASS II force
field

Fig. 5 Dependence of the natural logarithm of the density of PVA matrix with error bars as a function of temperature together with linear regression
(solid red line) and confidence intervals (dashed black lines) obtained in the simulation using a COMPASS and b INTERFACE force fields

where N – is the sample size as given as in Materials and
methods paragraph.
As one can observe, only the COMPASS and INTERFACE
force fields provide positive thermal expansion coefficients.
However, αp predicted by COMPASS force field is almost an
order smaller than the experimental value. Although the results
given by the INTERFACE force fields miss the experimental
value by nearly 56%, it successfully reproduces both signs and
order of αp. Considerable error may be due to relatively small
number of data points and small temperature increment.
Thermal resolution of the model
The negative thermal expansion coefficient value does not
necessarily indicate that the current force field model is invalid
but simply the statistics is too small. To verify this, we have
performed statistical testing of the slope of the regression of
ρ(T) = β1T + β0. In our t test, we formulate the null hypothesis
as follows:

H 0 : β1 ¼ 0
H a : β 1 ≠0:

In other words, the test verifies if data acquired in a given

force field has enough statistical quality to claim the linear
Fig. 6 Total NVT trajectory of PVA density calculations using dependence and thus whether it is possible to calculate the
INTERFACE force field slope (thermal expansion coefficient). The t statistics has the
305 Page 8 of 9
Table 7 Results of t test of zero slope hypothesis for simulations of density vs. temperature using different force fields
Force field Universal
T-value 0.9596
p value 0.3916
Can we reject the null hypothesis? No
following form:
β1
tc ¼
S β1
Using Student’s t cumulative distribution function (CDF)
as implemented as in MATLAB 2015, one can calculate the
p value for the two-tailed t test defined as:
 
p−value ¼ 2* Pðt > t c Þ ¼ 2* 1−CDF df ðt c Þ :
Hence the p value is the probability that the null hypothesis
is true. The null hypothesis is thus rejected when the p value of
the t test is less than α.
The results clearly suggest that the INTERFACE force field
provides valid results within 10-K temperature increment re-
gime whereas the other force fields fail (Table 7).
Conclusions
The knowledge of the structural and mechanical properties as
function of temperature of the PVA system is crucial in order to
design a valuable semipermeable matrix for enzyme immobili-
zation. In this work, we have successfully applied molecular
dynamics and Monte Carlo methodology to study this kind of
phenomena. Particularly, we have tested second- and third-
generation force fields. We have shown that among them, the
INTERFACE force field may be the best solution since it pre-
dicts the density of the PVA system with excellent accuracy.
Furthermore, it reproduces the order of thermal expansion co-
efficient even though a small density data set has been used
together with 10-K temperature increment. Thus we validate
the INTERFACE force field as a proper tool to model proper-
ties of the PVA to PVA, a polymer scaffold for hydrogel sys-
tems. The INTERFACE force field is a major improvement of
PCFF force field, and due to its robustness, it can be success-
fully applied to study more complicated systems. The appropri-
ate choice of the force field model on the initial stage of study-
ing given system is a crucial and very often the most volatile
step. With the use of simple statistical analysis the equilibration
times in this system may be shortened and still provide valid
qualitative results. More extensive study of mechanical and
transport properties of PVA hydrogel and its blends using the
INTERFACE force field will be presented in following papers.
J Mol Model (2017) 23: 305
COMPASS COMPASS II PCFF INTERFACE
−1.8343 2.4951 1.334 −3.361
0.1404 0.0672 0.253 0.0282
No No No Yes
Acknowledgements This study was supported by the project
BDetermination of an influence of composition and spatial structure of
hydrogels on the properties of entrapped biocatalysts^ financed by NCN
2015/19/D/ST8/01899 (Poland). The computations have been carried out
in Wroclaw Networking and Supercomputing Centre WCSS under com-
putational grant no. 172 using BIOVIA Materials Studio 8.0 software.
References
1. Hermann W.O., Haehnel W. (1924) Verfahren zur herstellung von
polymeren vinylalkohol, German Patent DE490041
2. Peppas NA, Merrill EW (1977) Crosslinked poly(vinyl alcohol)
hydrogels as swollen elastic networks. J Appl Polym Sci 21:
1763–1770
3. Hyon SH, Cha WI, Ikada Y (1989) Preparation of transparent
poly(vinyl alcohol) hydrogel. Polym Bull 22:119–122
4. Hassan CM, Peppas NA (2000) Structure and applications of poly(-
vinyl alcohol) hydrogels produced by conventional crosslinking or
by freezing/thawing methods. Adv Polym Sci 153:37–65
5. Kenawy ER, Kamoun EA, Eldin MSM, El-Meligy MA (2014)
Physically crosslinked poly(vinyl alcohol)-hydroxyethyl starch
blend hydrogel membranes: synthesis and characterization for bio-
medical applications. Arab J Chem 7:372–380
6. Muppalaneni S, Omidian H (2013) Polyvinyl alcohol in medicine
and pharmacy: a perspective. J Dev Drugs 2:112 (1-5)
7. Wang X, Fang D, Yoon K, Hsiao BS, Chu B (2006) High-
performance ultrafiltration composite membranes based on poly(-
vinyl alcohol) hydrogel coating on crosslinked nanofibrous poly(-
vinyl alcohol) scaffold. J Membr Sci 278:261–268
8. Zhang Y, Li H, Li H, Li R, Xiao C (2006) Preparation and charac-
terization of modified polyvinyl alcohol ultrafiltration membranes.
Desalination 192:214–223
9. Jegal J, Lee KH (1999) Nanofiltration membranes based on poly(-
vinyl alcohol) and ionic polymers. J Appl Polym Sci 72:1755–1762
10. Gimenes ML, Liu L, Feng X (2007) Sericin/poly(vinyl alcohol)
blend membranes for pervaporation separation of ethanol/water
mixtures. J Membr Sci 295:71–79
11. Premakshi HG, Kariduraganavar MY, Mitchell GR (2016)
Development of composite anion-exchange membranes using
poly(vinyl alcohol) and silica precursor for pervaporation separa-
tion of water–isopropanol mixtures. RCS Adv 6:11802–11814
12. Lang K, Sourirajan S, Matsuura T, Chowdhury G (1996) A study
on the preparation of polyvinyl alcohol thin-film composite mem-
branes and reverse osmosis testing. Desalination 104:185–196
13. Bermejo JS, Ugarte CM (2009) Chemical Crosslinking of PVA and
prediction of material properties by means of fully atomistic MD
simulations. Macromol Theory Simul 18:259–267
14. Bermejo JS, Ugarte CM (2009) Influence of cross-linking density
on the glass transition and structure of chemically cross-linked
PVA: a molecular dynamics study. Macromol Theory Simul 18:
317–327
15. Havryliv R, Maystruk V, Biliak V (2015) Development of a numer-
ical model for gas-solid flow in the industrial cyclone-calciner fur-
nace. EEJET 3/8(75):14–21
J Mol Model (2017) 23: 305
16. Havryliv R, Maystruk V (2017) Development of a combustion
model in the industrial cyclone-calciner furnace using CFD-model-
ing. Chem Chem Technol 11(1):71–80
17. Chang K-S, Yoshioka T, Kanezashi M, Tsuru T, Tung K-L (2010) A
molecular dynamics simulation of a homogeneous organic-
inorganic hybrid silica membrane. Chem Commun (Camb)
46(48):9140–9142
18. Lithoxoos GP, Labropoulos A, Peristeras LD, Kanellopoulos N,
Samios J, Economou IG (2010) Adsorption of N2, CH4, CO and
CO2 gases in single walled carbon nanotubes: a combined experi-
mental and Monte Carlo molecular simulation study. J Supercrit
Fluids 55:510–523
Page 9 of 9 305
19. Tzu-Jen, Heinz H (2016) Accurate force field parameters and pH
resolved surface models for hydroxyapatite to understand structure,
mechanics, hydration, and biological interfaces. J Phys Chem C
120(9):4975–4992
20. Chakraborty T, Hens A, Kulashrestha S, Chandra MN, Banerjee P
(2015) Calculation of diffusion coefficient of long chain molecules
using molecular dynamics. Physica E 69:371–377
21. Bazooyar F, Momany FA, Bolton K (2012) Validating empirical
force fields for molecular-level simulation of cellulose dissolution.
Comput Theor Chem 984:119–127
22. Technical references of SEKISUI taken from http://www.sekisui-sc.
com/wp-content/uploads/SelvolEPVOH305_EN_TechData.pdf