THE JOURNAL OF CHEMICAL PHYSICS 137, 064116 (2012)

Computer simulation of the role of torsional flexibility on mass

and momentum transport for a series of linear alkanes
Carlos Bragaa) and Karl P. Travisb)
Department of Materials Science and Engineering, The University of Sheffield, Sir Robert Hadfield Building,
Mappin Street, Sheffield S1 3JD, United Kingdom

(Received 16 April 2012; accepted 13 July 2012; published online 13 August 2012)
We present the results obtained from a systematic equilibrium molecular dynamics study of the effect
of torsional flexibility on the diffusion and viscosity of a series of linear alkanes. To make unambiguous comparisons between molecules with torsional flexibility and those without, we use the frozen
distribution sampling (FDS) method introduced by Travis et al. [J. Chem. Phys. 98, 1524 (1993);
J. Chem. Phys. 102, 2174 (1995)] but modified and updated for increased efficiency. We first demonstrate comprehensively that FDS guarantees corresponding thermodynamic states. We then show that
removal of torsional flexibility results in a significant lowering of the diffusion coefficient (and corresponding increase in shear viscosity) and furthermore that this effect increases with increasing chain
length. The results are discussed in terms of the possible mechanism giving rise to this dynamic
coupling phenomenon. 2012 American Institute of Physics. [http://dx.doi.org/10.1063/1.4742187]
I. INTRODUCTION

One of the fundamental aims of liquid state chemical

physics is the ability to understand and predict the behaviour
of a liquid based on a knowledge of the properties of its constituent molecules. Properties such as the shape of a molecule
are known to play a role in determining the rheological behaviour of a liquid. In addition, intramolecular flexibility is
another important property which affects the behaviour of the
liquid phase. Neither of the aforementioned properties can be
defined precisely enough to be treated in a satisfactory manner
by standard statistical mechanical methods; molecular shape
for instance, depends in general on molecular flexibility and
thus changes with time, while the many different degrees of
freedom that contribute to flexibility are intrinsically coupled.
Transport properties, e.g., diffusion and viscosity coefficients, are state point dependent and thus any experiment
aimed at establishing the effect of a given molecular property
on the transport behaviour of the material must ensure comparisons are made between materials in equivalent thermodynamic states. Many simulations have been conducted with
the aim of determining the influence of molecular flexibility
and static shape on fluid transport properties. Unfortunately,
comparable thermodynamic states were not obtained in most
of these studies.311 The exceptions to these studies include
the work of Brown and Clarke, in which they established the
effect of mass distribution upon the transport behaviour of a
triatomic fluid12 and the recent work of Walser et al.,13, 14 who
built upon this result and studied the dependence of viscosity
and diffusion on the mass distribution of the molecules in a
liquid with the aim of checking the validity of Stokes law at
the molecular scale.
a) Present address: Department of Chemical Engineering, Centre for Process

Systems Engineering, Imperial College London, South Kensington Campus, London SW7 2AZ, United Kingdom.
b) k.travis@sheffield.ac.uk.

0021-9606/2012/137(6)/064116/11/$30.00

The success of these two studies is predicated on the fact

that two model fluids with distinct mass distributions can be
prepared in identical thermodynamic states because the statistical mechanical expressions for their configurational properties do not depend on mass. Scaling the total mass of the system by a factor is equivalent to scaling the corresponding
time by 1/2 .14 Changing the mass distribution of a molecular
fluid while keeping its total molecular value constant has the
end effect of scaling the associated time scales with the individual degrees of freedom by different factors. Building upon
this result, the same authors compared two models of hexane,
one with torsional motions restricted by a realistic torsional
potential, and the other in which torsional motions were unrestricted (the so-called flexane model).15 Brown and Clarke
suggested that hexane and flexane were in closely corresponding states but were unable to give conclusive proof of this fact.
Nevertheless, their work highlighted the important role played
by torsional oscillations in enhancing self-diffusion.
The outstanding problem of ensuring that comparisons
between hexane and their rigid, torsion-free, counterparts, are
performed with the fluids in identical thermodynamic states
was later solved by Travis, Brown, and Clarke.1 They called
their method frozen distribution sampling (FDS) due to the
manner in which a subset of the equilibrium torsional distribution is frozen in by application of a set of torsional constraints. The authors went on to show that for n-butane, with a
single torsional angle, any torsional coupling with mass transport is negligible,1 while in n-hexane, the effect is small, but
significant, and increases with temperature.2
In this paper, we use the FDS method of Travis et al.1 to
explore the torsional coupling effect across a series of straight
chain alkanes ranging from C6 to C30 in order to determine
how chain length modifies the phenomenon. We have also
improved the method to enable long alkanes to be treated efficiently. The remainder of this paper is organised in the following way: In Sec. II, we describe the FDS method from the

137, 064116-1

2012 American Institute of Physics

064116-2

C. Braga and K. P. Travis

point of view of statistical mechanics, and also how it is implemented in a simulation. In Sec. III, we describe the models
used for both rigid and flexible alkanes along with the algorithms used to simulate them. In Sec. IV, we give the results
of our simulations and discuss them in detail, while in Sec. V
we draw some conclusions.
II. FROZEN DISTRIBUTION SAMPLING METHOD
A. Theoretical background

In this work, we are interested solely in the effect of

torsional motions on molecular transport properties. To determine their effect, we introduce an experimental control
in which there are no intramolecular degrees of freedom
the molecules are fully rigid. Application of holonomic constraints has some subtle (though well known) consequences
for the statistical mechanics of the corresponding system, resulting from changes in the dimensionality of the phase space.
These changes can lead to differences in the configurational
properties, including the pressure. The way in which the control experiment is set up is thus crucial if we are to make unambiguous comparisons with the fluid comprising of flexible molecules such that they are in identical thermodynamic
states.
The method of FDS was introduced by Travis, Brown,
and Clarke1, 2 to allow unambiguous comparisons to be made
in the manner described above. The FDS method has been
previously expounded in detail for n-butane simulations.1
Here, we give only a brief outline of the theoretical justification of the FDS method. The underlying idea is that the
value of any configurational property can always be obtained
by averaging separately over independent sets of coordinates.
We use this to compare the dynamical properties of a liquid of
flexible molecules, A, with those of an identical liquid, B, in
which the molecules have no intramolecular degrees of freedom but at the same time have the torsional distribution of
system A frozen-in. In system B, this is equivalent to averaging separately over the intramolecular coordinates and corresponds to a molecular dynamics (MD) simulation where a
configuration of system A at any time t is taken as input to
a simulation of system B in which either extra constraints are
added to render the molecules rigid or the molecules are modelled as rigid bodies in the classical mechanical treatment.
Consider a system of molecules which may be subject
to an arbitrary set of holonomic constraints. The configurational phase space average of some property of the coordinates, (r), is given by

(r) PA (r) dr
,
(1)
 A = 
PA (r) dr
where PA (r) is the canonical configurational phase space
probability density which, in general, may include a Jacobian
factor due to any constraints applied.1618 Consider this same
system subject to a further set of constraints to yield a second
system B. The configurational properties for the new system
are formally given by

(r) PB (r) dr
,
(2)
 B = 
PB (r) dr

J. Chem. Phys. 137, 064116 (2012)

where PB (r) is the canonical configurational phase space

probability density of system B. This probability density is
related to that of system A by
PB (r) |Y|1/2 PA (r) ,

(3)

where |Y| is the determinant of the corresponding metric tensor, which arises from the additional constraints imposed to
create system B. The canonical ensemble average of a configurational property (r) in system B can be written in terms of
PA as

(r) |Y|1/2 PA (r) dr
 B = 
.
(4)
|Y|1/2 PA (r) dr
If the additional set of constraints applied to A to yield B
removes all the intramolecular degrees of freedom, i.e., it
renders the molecules fully rigid, then the determinant |Y|
becomes a constant independent of the remaining degrees of
freedom. Equation (4) therefore simplifies to

(r) PA (r) dr
=  A ,
(5)
 B = 
PA (r) dr
which is identical to the configurational average for system A.
The end result of the FDS method is to effectively decouple
the frozen degrees of freedom from the rest of the system
in a thermodynamically consistent manner. It is important
to emphasise that the FDS method works well for flexible
systems only if their intramolecular degrees of freedom are
completely frozen. Although it is certainly of interest to be
able to discriminate the influence of different degrees of
freedom, the resulting determinant of the Jacobian of the
transformation |Y| will not be a constant in this case and the
evaluation of the integral in (4) is not so straightforward.
B. Methodology

In the original formulation of the FDS algorithm,

molecules were rendered completely rigid through the use
of holonomic constraints; nearest and second nearest neighbour distance constraints were used to keep the bond lengths
and valence angles at their specified values, while vectorial constraints were employed to fix the torsion angles in
the Ryckaert-Bellemans models of n-butane and n-hexane.
The constraint dynamics were then solved using the SHAKE
algorithm.19 The use of constraint dynamics is a case of
overkill; the molecules, being completely rigid, require only
3 degrees of freedom to describe their position in cartesian
space, and 4 quaternions to define their orientations in a computationally stable fashion. For example, the n-triacontane
molecular model contains 30 united atom sites and needs a
total 84 constraints to render the molecule completely rigid.
Whether SHAKE of Gaussian constraints are used, solving
the constrained equations of motion would become unnecessarily expensive. In the present formulation, we simply solve
the resulting rigid body equations of motion for the periods of
the simulation which involve fully rigid systems.
What is perhaps unclear from the statistical mechanical
argument given above is that in practice, due to the finite system size used in a typical MD simulation, one cannot hope to

064116-3

C. Braga and K. P. Travis

TFM Equilibration

Crystalline
Configuration

J. Chem. Phys. 137, 064116 (2012)

Molecular Dynamics Simulation

Equilibrated
Configuration
Rigid Body
Equilibration

Sample 1

Sample 2

Sample n

Rigid Body
Molecular
Dynamics

Frozen Distribution Sample Set

FIG. 1. Schematic outline of a FDS experiment.

lock-in the true equilibrium distribution of torsion angles in

a single freezing experiment. What happens in practice is
that a subset of the equilibrium distribution of torsion angles
is frozen. In order to sample the true equilibrium distribution, one must freeze a set of independent intramolecular configurations and average over the resulting set. Travis et al.2
found that a total of 10 independent frozen segments were
sufficient to sample the equilibrium torsional distribution of
n-butane. This number depends on the system being studied
and upon the thermodynamic state. Superimposition of the
segment-averaged mean torsional angle distribution upon the
corresponding equilibrium distribution for flexible molecules
is but one criterion to decide how many rigid body segments
to use. A more complete criterion is to ensure that the intermolecular properties are identical. The simplest quantity to
compare in this case is the pair distribution function since
it describes the liquid structure and to which configurational
properties are related.
A typical FDS scheme is set out in Fig. 1. Each FDS experiment begins with a solid crystalline configuration which
is relaxed using the truncated force method.20 After the system has relaxed from the initial configuration, an equilibrium
MD simulation is performed at constant density and temperature. From the resulting trajectory a series of snapshots is
taken, the configurations of which are used as input configurations to the rigid body molecular simulations. Before conducting the rigid body simulations, short equilibration runs
are performed to allow the rotational and translational temperatures to come to equilibrium following the lowering of the
phase space dimensionality (by removal of the intramolecular
degrees of freedom). At the end of the FDS experiment, an
average is computed over the set of independent rigid body
molecular simulations yielding a single set of thermodynamic
data. The configurational properties of this set should then be
equivalent to the corresponding flexible system averages obtained from the equilibrium MD simulation.
A FDS experiment requires the specification of two quantities: (i) the number of snapshots to take, and (ii) the sampling
interval between snapshots. The first quantity is determined
by the equivalence between the static thermodynamic properties of the rigid and flexible systems. The second parameter
depends on the rate constant for the interconversion between
different conformation states in the molecules. This quantity
is not usually known, although it is possible to compute it via

MD simulations.2124 A fast interconversion rate means the

instantaneous torsional distribution is rapidly changing and,
therefore, a shorter sampling interval may be chosen. In lieu
of any rate constant data for a given system, a sensible strategy
is to undertake a series of short pilot simulations to ascertain
appropriate values of sampling intervals.
We have conducted a series of FDS experiments on
a series of linear alkanes of increasing chain length. All
the simulations were conducted at the same reduced site
number density *(= 3 ) = 1.537 and reduced translational
temperature TM (= T /kB ) = 5. This state point corresponds
to the one adopted in previous works3, 25 thus providing a
crosscheck on some of the parameters obtained.
III. MODEL AND SIMULATION DETAILS
A. Model details

The same alkane model was used in all molecular

simulations of rigid and flexible alkanes. Following the
model introduced by Ryckaert and Bellemans (RB) for the
simulation of n-butane,26 an alkane of mass M and containing
ns carbon atoms consists of ns identical Lennard-Jones sites.
These represent both methyl CH3 and methylene CH2 groups,
each with mass M/ns . The groups are connected together such
that the nearest neighbour distances are fixed throughout the
simulations at 1.54 and valence angles are held at the value
of 109.47 . Sites separated by three bonds interact through a
RB torsional angle potential in the form of a truncated power
series
Utors /kB =

5


Cn cosn () ,

(6)

n=0

where is the dihedral angle defined between sets of four

sites, and the expansion coefficients are given by
Cn = {1.116, 1.462, 1.578, 0.368, 3.156, 3.788} 103 K.
(7)
For molecules with several torsional degrees of freedom, the
total torsional potential is obtained by summing up the individual contributions from each dihedral angle. No explicit
coupling between torsional degrees of freedom is taken into
account. Sites on distinct molecules and sites belonging to the
same molecule but separated by more than three bonds, interact through a Weeks-Chandler-Andersen (WCA) potential
  
 6 
12
4 r
+  r < 2(1/6)
r
(8)
UW CA =
0
r 2(1/6) ,
where the LJ coefficients are = 3.923 and /kB = 72 K.
The WCA potential has been chosen over a more standard
Lennard-Jones potential for reasons of computational efficiency and to aid comparisons with simulations already in the
literature. Since we make no attempt to compare directly with
experiment, this is not a serious issue. Liquid state transport
properties are in any case largely determined by the repulsive
part of the potential.25, 27
The use of holonomic constraints in simulations involving molecules is usually justified in terms of computational

064116-4

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

efficiency.16, 28 This is certainly true for the fast, high frequency bond stretching vibrations, which, when eliminated
allow the use of a larger time step. However, the consequences
of this degree of freedom removal upon configurational averages (and hence the thermodynamic state) are not always insignificant (a position dependent Jacobian appears in the partition function18, 28 ). It is now well known that constraining
the valence angles as well as the bond lengths changes significantly the mean pressure of a simulation as well as affecting the equilibrium distribution of conformers. In the present
study, we have chosen to use the RB model of alkanes despite
the foregoing discussion because this model has been well
studied and, furthermore, we are making a comparison with
fully constrained models rather than fully flexible versions of
these alkanes. We could just as easily have taken fully flexible
modes for comparison with the fully rigid ones.
Since the introduction of the RB model, there have been
many improvements made to models of linear alkanes, particularly with respect to the intermolecular potential. These
improvements are meant to yield a better agreement between
the model thermodynamic and transport properties and those
of the real material. Nevertheless, the original RB model is
able to capture the essential physics of the dynamical system
from which the main features and trends can be obtained.

simple algebra,

Fi pi /Mi
i
,

p2i /Mi

where Fi is the total force on molecule i. A continuous proportional feedback scheme is used to cancel any numerical
drift in the value of TM .32 The feedback term takes the form
(TM T0 ) pi /ns , where = 10. Holonomic constraint
forces were also derived from the application of Gauss principle of least constraint. For a given molecule, fixed bond
lengths and bond bending angles were achieved by formulating a set of distance constraints of the form
gn = R2n dn2 = 0,

n = 1, nc .

1. Flexible model simulations

Molecular dynamics simulations were conducted using a

set of coupled 1st order equations of motion that were integrated using a 4th order Gear predictor-corrector29 scheme
r i =

pi
,
mi

p i = Fi + FCi

mi
pi ,
Mi

(9)

(10)

where ri , pi , Fi , and FCi are the position, momentum, total

Newtonian force, and constraint force on site of molecule i,
respectively. A closed expression for the Gaussian multiplier
is obtained from Gauss principle of least constraint;30, 31
beginning with an expression for the instantaneous translational temperature, TM ,
 p2
i
i Mi
=
,
(3N 4) kB

(11)

the total time derivative is obtained and then the equations of

motion are substituted into that expression to yield, after some

(13)

Here, R is a vector connecting the pair of sites involved in the

constraint n, d is the constraint length value. Following the
notation of Edberg et al.,31 the expression for the Gaussian
constraint forces on atom in a given molecule (where the
subscript i is dropped for clarity) is

FC =
Sn n Rn ,
(14)
n

where S is a selector matrix and is the column vector of

constraint multipliers. The vector is obtained from the acceleration dependent form of the constraint Eq. (13),
n + (R
n )2 = 0,
g n = Rn R

Two different types of simulations were conducted; one

employing cartesian coordinates for all molecular sites together with holonomic constraints the flexible system, and
one employing generalised coordinates for the simulation of
the rigid or frozen molecules. We shall henceforth refer to
these as flexible and rigid models, respectively.

(12)

B. Simulation details

TM

(15)

n is
where the equation of motion for the constraint vector R
given by

n = Fn +
R
Lnm m Rm ,
(16)
m

and the nc nc matrix L is obtained by taking the difference

between the corresponding rows from the selector matrix S.
Substitution of Eq. (16) into (15) for all constraints results in
the following set of algebraic equations:

n )2 ).
(Lnm Rm Rn ) m = (Rn Fn + (R
(17)
m

In addition to the holonomic constraint force terms, continuous proportional feedback needs to be used to counteract
any numerical drift in the values of the constrained distances
and their first derivatives.32 This was achieved by adding
the following terms to the equations of motion for the positions and momenta of the alkane sites for a given molecule,
respectively,

n,
Sn (| Rn | dn ) R
(18)
C
n

R
n,
Sn (Rn R)

(19)

where the circumflexes denote unit vectors. The following

constants were used: C = 10 and D = 15. Although continuous proportional feedback introduces irreversibility into otherwise time reversible equations motion, it is only used here
to cancel the numerical drift resulting from the finite precision of the numerical implementation. The time step (defined

064116-5

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

in reduced units by t* = (/m 2 )1/2 t) used to integrate the

equations of motion ranged from t* = 0.0001 for C6 to t* =
0.00005 for C30 . Throughout the remainder of this paper, all
quantities will be given in reduced unit form and for the sake
of brevity we shall henceforth drop the asterisk notation.
The flexible model simulations were initiated by first
constructing simulation boxes containing the appropriate
number of molecules consistent with a reduced site density
= 1.537. Each molecular system was relaxed using the
truncated force method of McKechnie et al.20 from a corresponding fcc lattice with initial velocities sampled from a
Maxwell-Boltzmann distribution at the appropriate temperature. Following the initial relaxation phase, a longer period of
equilibration was undertaken. The resulting equilibrated configuration was then used for the FDS experiment.

employed to ensure that the translational kinetic energy was a

constant of the motion. The implementation of this thermostat
is identical to that for the corresponding flexible system. As
explained above, the FDS method requires configurations to
be periodically sampled from a time evolving flexible system
trajectory. The sampled configurations are in turn cast with
their conformational states frozen-in. The conversion of the
site based Cartesian variables to the variables needed for the
integration of the rigid body equations of motion is accomplished through the following scheme. The molecular positions and momenta ri , pi are computed from the site based
quantities ri , pi ,
ri =

2. Rigid model simulations

To investigate the role of torsional motion, we have removed the torsional degree of freedom by simulating rigid
molecules as described previously. The rigid molecule simulations were carried out by solving the following set of first
order coupled equations of motion33 using a 4th order Gear
predictor-corrector29 algorithm
pi
,
(20)
r i =
M
p i = Fi pi ,

b
xi
=

b
yi

b
zi
=

q0i =

b
xi
b b
+ yi
zi
Ixxi
b
yi

Iyyi

b b
zi
xi

zib
b b
+ xi
yi
Izzi

Iyyi Izzi
Ixxi
Izzi Ixxi
Iyyi

Ixxi Iyyi
Izzi

(21)

(22)

(23)

(24)


1
b
b
b
q1i xi
,
q2i yi
q3i zi
2


1
b
b
b
q0i xi
,
q3i yi
+ q2i zi
qxi =
2

1
b
b
b
q3i xi
,
+ q0i yi
q1i zi
qyi =
2

pi =

ns


pi .

(29)

(30)

=1

The molecular velocities follow from the molecular momenta.

All other derivatives of these variables are set initially to zero.
The inertia tensor of each molecule is calculated in the laboratory frame from


2
Ilab
1ri
(31)
= Mi
ri ri ,
i

where the primes denote positions relative to the molecule

centre of mass and 1 is the isotropic second rank tensor. The
inertia tensor relative to the laboratory frame of reference is
then diagonalised to yield the three principal components of
inertia (computed from the eigenvalues) and the rotation matrix A (computed from the corresponding eigenvectors). The
set of quaternions is then calculated from the elements of the
rotation matrix33 via

q0 = (A11 + A22 + A33 ) /4 + 1,
(32)

qx =
(25)
qy =
(26)
qz =

(A22 + A33 ) /2 + q02 ,

(33)

(A11 + A33 ) /2 + q02 ,

(34)

(A11 + A22 ) /2 + q02 .

(35)

(27)


1
b
b
b
q2i xi
+ q1i yi
+ q0i zi
,
(28)
2
where, for a given molecule i, r i is the linear velocity, Fi is
the total force acting on molecule, bi is the angular velocity
in the body fixed frame of reference, I are the 3 principal
moments of inertia,  i is the torque about the centre of mass
(i.e., spin torque), and qxi , qyi , qzi are the four components of
the quaternion for the molecule. A Gaussian thermostat was
qzi =

ns
1 
ri ,
ns =1

Angular velocities (relative to the principal frame of reference) were assigned to each molecule by sampling their values from a Maxwell-Boltzmann distribution at the appropriate rotational temperature, equal to the molecule translational
temperature. Spin torques in the principal frame were obtained from spin torques evaluated relative to the laboratory
reference frame

=
ri Fi ,
(36)
 lab
i

064116-6

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

TABLE I. Main results for a series of n-alkanes at constant volume and temperature. The flexible models are represented here as Ryckaert-Bellemans on
left half of the table and their frozen counterparts are represented here as rigid body on the right half of the table. The symbols in the first column refer,
respectively, to: molecular translational temperature, rotational temperature, scalar pressure, torsion angle energy, Lennard-Jones intermolecular energy, rootmean-square end-to-end distance, root-mean-square radius of gyration, percentage of trans conformations, self-diffusivity, and viscosity, respectively. The
numbers in parentheses are the statistical uncertainty in the last digits computed from the standard deviations divided by the square root of the number of
segments.
Ryckaert-Bellemans

TM 
TR 
p
Utors 
ULJ 
2 1/2
1n
Rg2 1/2
%trans
Ds

Rigid body

C6

C10

C16

C20

C30

C6

C10

C16

C20

C30

5
4.998(3)
17.185(6)
15.50(1)
6.026(3)
1.3951(3)
0.51341(5)
63.15(5)
0.2421(1)
2.1905

5
4.999(3)
11.818(4)
35.96(4)
6.147(3)
2.2936(7)
0.7998(1)
64.81(4)
0.2562(1)
1.5

5
4.999(6)
9.274(6)
66.91(5)
7.110(7)
3.404(3)
1.1799(5)
65.12(4)
0.2245(2)
1.5466

5
4.999(4)
8.489(4)
87.4(1)
7.969(4)
4.024(4)
1.4061(6)
65.31(4)
0.1923(5)
1.2258

5
5.004(9)
7.498(6)
139.1(1)
9.886(4)
5.314(9)
1.904(2)
65.31(3)
0.1622(3)
1.192

5
4.998(4)
17.18(2)
0
6.026(8)
1.394(6)
0.5136(9)
63.3(7)
0.2413(4)
1.9029

5
5.000(4)
11.82(1)
0
6.147(6)
2.29(1)
0.799(2)
64.6(6)
0.2153(2)
2

5
5.000(7)
9.27(2)
0
7.10(2)
3.41(3)
1.181(6)
65.1(6)
0.1251(1)
2.3961

5
5.004(3)
8.51(1)
0
7.93(3)
4.01(4)
1.412(8)
65.4(1)
0.0762(3)
3.7177

5
5.000(9)
7.50(5)
0
9.91(8)
5.3(1)
1.90(2)
65.3(5)
0.0106(5)
10.5564

which were in turn calculated from

 i = A  lab
i .

(37)

IV. RESULTS AND DISCUSSION

A. FDS at constant volume and temperature

Simulations were carried out in the canonical ensemble

on five n-alkane systems with different chain lengths. These
are, respectively: n-hexane C6 , n-decane C10 , n-hexadecane
C16 , n-eicosane C20 , and n-triacontane C30 . As the atomic reduced number density = 1.537 is the same for all systems,
the corresponding molecular number density will decrease
with increasing chain length. For each n-alkane system, the
same FDS procedure outlined in Figure 1 has been applied.
The thermodynamic properties obtained from both the
Ryckaert-Bellemans and the corresponding rigid body models are collected in Table I. The quantities are described as follows: TM corresponds to the molecular temperature computed
from the centre-of-mass velocities of each molecule which
is fixed by the isokinetic thermostat and TR is the rotational
temperature. The average pressure of each system is represented by p, Utors is the average torsion angle energy, ULJ
2 1/2
is the average intermolecular Lennard Jones energy, 1n

is the root mean square end-to-end distance of the n-alkane
molecule, where 1n = rn r1 , %trans, is the average percentage of dihedral angles in the trans conformation, Ds is the
self-diffusivity computed from Einsteins mean square displacement equation
Ds =

N
1 
d
lim
[ri (t) ri (0)]2 ,
dt t 6N i=1

(38)

and the shear viscosity was obtained via the corresponding

Green-Kubo expression

V
(39)
dtPos (t) : Pos (0),
=
10 kB T 0

where Pos is the traceless, symmetric part of the molecular

pressure tensor. This form of the correlation function uses
the maximum information available for an isotropic fluid and
gives a significant reduction in the statistical noise inherent in
this quantity.
From Table I, it can be seen that the FDS method has
succeeded in generating liquids of flexible and rigid models
of all n-alkane models with identical canonical configurational properties. The agreement between the configurational
properties of flexible models and their rigid body counterparts
is excellent. Average molecular structure is quantified by the
mean end-to-end distance and fraction of dihedral angles in
the trans conformation is also preserved. Further evidence
of the success of FDS method is illustrated by Figures 2 and
3 where the dihedral angle and end-to-end distributions are
shown for both flexible and rigid models of C10 and C20 ,
respectively.
The effect of freezing the torsional motions on the transport properties of n-hexane molecules is assessed through
computation of centre of mass mean square displacements
(38) and stress autocorrelation function (39). The resulting
coefficients Ds and are similar within the statistical error
as can be observed in Table I, thus showing that no coupling
effects can be observed for the n-hexane system. Travis et al.2
found evidence for the existence of a coupling of torsional to
translational modes in a similar model of n-hexane. However,
the state point conditions at which n-hexane was studied
were not the same. The atomic density was higher to a value
of s = 2.025 and the temperatures studied covered a range
between T
= 2.025 and T
= 13.9. Since the range of studied
temperatures includes the temperature chosen for the present
study, it would seem that such coupling is more sensitive to
the system density. Changes in density affect not only the
length of the mean free path between molecular collisions,
but also the energy of the interaction.
The equilibrium averages obtained from the FDS experiment on n-decane and its frozen counterparts are also
collected in Table I. Unlike n-hexane, the n-decane system

064116-7

C. Braga and K. P. Travis

FIG. 2. Comparison between the dihedral angle distributions f() obtained

from the average of all rigid body segments (in red) and the corresponding
flexible system (in blue) at constant volume and temperature. Top distribution corresponds to n-decane (a) and bottom distribution corresponds to neicosane (b). Similar correspondence exists for n-hexane, n-hexadecane, and
n-triacontante models.

exhibits a clear difference in the resulting self-diffusion

coefficients Ds . Figures 4 and 5 illustrate the short time
dynamics of both liquids, flexible and rigid body, for different
n-alkane systems.
With the exception of n-hexane, the rigid body molecules
have a velocity autocorrelation function that decays at a faster
rate than the flexible ones. For chains larger than n-decane,
a cage effect starts to appear. The back scattering increases
with chain length and, apart from n-hexane, these differences may provide an insight into the coupling mechanism
between torsion and translational motions. The presence of
constraints impedes the molecules ability to change its shape
in response to the interactions with other neighbours. Rigid
body molecules are unable to transfer energy into their internal dihedral modes and therefore loose memory of their translational motion at a faster rate. Note comparisons are being
made between systems that have, within the statistical uncertainties, identical configurational properties which mean that
this behaviour is dynamic in nature. From Figures 4 and 5, it
can be observed that this effect increases with chain length.
The self-diffusion coefficient of flexible n-hexadecane model
is almost twice that of the corresponding value for the frozen

J. Chem. Phys. 137, 064116 (2012)

FIG. 3. Comparison between the end-to-end distance distributions f (1n )

obtained from the average of all rigid body segments (in red) and the corresponding flexible system (in blue) at constant volume and temperature. Top
distribution corresponds to n-decane (a) and bottom distribution corresponds
to n-eicosane (b). Similar correspondence exists for n-hexane, n-hexadecane,
and n-triacontante models.

n-hexadecane model up to n-triacontane which has a diffusion

coefficient that is more than one order of magnitude larger that
the value of its frozen counterpart. The effect of coupling of
torsional motions to diffusion for such a large molecule seems
to be the most important mechanism of translational motion.
In Figure 6, a qualitative comparison between the
self-diffusion coefficients for the flexible and rigid body
n-alkane systems is presented as function of chain length,
measured here by the number of carbon atoms present in the
chain. There is clear evidence for the enhancement of the
self-diffusion coefficient due to coupling with torsional motions. Although no quantitative comparison between different
chain lengths can be made as the systems do not share the
same equilibrium configurational averages, e.g., pressure and
intermolecular energy, the increment in the self-diffusion coefficient augments with chain length. This supports previous
conclusions1, 2 which suggested that the difference between
the diffusion coefficients of flexible systems and their frozen
counterparts is predominantly a dynamical effect. For example, at the simulated state point conditions of T = 5.0 and
= 1.537, differences between the flexible and rigid body

064116-8

C. Braga and K. P. Travis

FIG. 4. Comparison between the short time velocity autocorrelation functions v(t) v(0) obtained from the average of all rigid body segments (in
red) and the corresponding flexible system (in blue) at constant volume and
temperature. The top distribution corresponds to n-hexane (a) and bottom
distribution corresponds to n-decane (b), respectively.

J. Chem. Phys. 137, 064116 (2012)

FIG. 5. Comparison between the short time velocity autocorrelation functionstions v(t) v(0) obtained from the average of all rigid body segments
(in red) and the corresponding flexible system (in blue) at constant volume
and temperature. The top distribution corresponds to n-hexadecane (a) and
bottom distribution corresponds to n-eicosane (b), respectively.

n-hexane models are negligible, which means that, as far

as diffusion is concerned, n-hexane behaves like a rigid
molecule.
B. FDS at constant pressure and temperature

Normally, one would expect the rate of mass transfer,

measured here by the self-diffusion coefficient, to decrease
with chain length. However, the self-diffusivity of n-hexane
is smaller than n-decane at the same temperature. This attribute merits some discussion. The n-decane system is at
a lower molecular density and different pressure conditions
that those of n-hexane. Mondello et al.11 have studied the
crossover to Rouse-type behaviour for a series of n-alkanes
with varying chain lengths. By modelling the dynamics of the
chain through a corresponding Langevin equation, one may
define a monomeric friction coefficient which is related to
the diffusion coefficient through Ds = kB T/n , where n is the
number of united atom sites in the chain and kB is the Boltzmann constant.34 Mondello et al.11 have observed a minimum
in the friction coefficient for smaller chain lengths at around
n = 24 where n is the number of carbons in the chain. It was
further realized that, for small n, although such behaviour

FIG. 6. Comparison of the self-diffusion coefficients for the flexible and

frozen systems as function of chain length, measured here by the number
of carbon atoms present in the alkane chain.

064116-9

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

2  divided
FIG. 8. Comparison of the mean square end-to-end distance 1n
by the radius of gyration Rg2  as a function of n. For a Gaussian chain
2 /6R 2  = 1.
1n
g

FIG. 7. Comparison between the mean square displacements obtained from

simulations at constant volume (a) and from simulations at constant pressure (b), respectively. Note that n-hexane (full line style) has a smaller slope
value than n-octane and n-decane when simulated under constant volume
conditions. This effect disappears when the systems are simulated at constant
pressure.

was sensitive to the choice of the Lennard-Jones parameter  (known to decrease the molecular mobility4 ), the local
maximum in the diffusion coefficient remained present albeit
shifted to a different chain length thus suggesting that the origin of the maximum in Ds was related to the chain flexibility.
In order to assess whether this behaviour is, as suggested, a dynamical effect related to the molecular flexibility,
isothermal-isobaric simulations were performed on model nhexane, n-octane, and n-decane systems at T = 5.0 and at a
pressure equal to the one obtained by n-hexadecane system p
= 9.274. The isothermal-isobaric dynamics were achieved by
using a Gaussian isokinetic thermostat30 to control the temperature and Nose-Hoover barostat35 to control the pressure.
The equations of motion of each n-alkane model were solved
using a Gear predictor-corrector method29 simulated for
4 106 steps with a timestep size of 104 .
The dynamic properties of each n-alkane system are illustrated in Figure 7 which compares the mean square displacements obtained from simulations at constant volume and
from simulations at constant pressure. At constant volume and
temperature, the mean square displacements of both n-hexane
and n-hexadecane systems possess a lower slope value than n-

octane and n-decane. A local maximum in the corresponding

self-diffusion coefficients therefore occurs for n-octane and ndecane. When the systems are simulated at constant temperature and pressure, this behaviour changes markedly where
the diffusion coefficient decreases monotonically with chain
length.
The structural properties of each n-alkane molecule as
2
/6Rg2  are illustrated in Figure 8.
expressed by 1n
2
 = 6Rg2 . From Figure 8,
For a Gaussian chain, 1n
it can be observed that the n-alkane chains are clearly non2
 > 6Rg2 . There is a maximum in the strucGaussian 1n
tural ratio for n-hexadecane in agreement with the results of
Mondello et al.11 who also observed a maximum for n = 16
and with results obtained by Bashnagel et al.36 and by Brown
et al.10 For smaller n-alkanes n  16, the molecule is relatively linear and the change in the chain dimensions is similar
to the one obtained for a linear sequence of uniform beads
2
/6Rg2 goes from 1 for small n to 2 when n . When
1n
n  16, the flexibility of the molecule plays an increasingly
2
/6Rg2  decreases with n.
significant role in its shape and 1n
From Figure 8, we can observe that, although the maximum in Ds is removed when simulating at constant pressure
conditions, the molecule shape, measured here by the ratio
of the end-to-end distance to the radius of gyration, does
not change significantly. This effectively suggests that such
behaviour is a dynamic effect associated with the way the
molecule responds to its environment. To further investigate
the mechanism behind this effect, the same FDS procedure
outlined in Figure 1 has been applied to C6 , C8 , and C10 at
the same state point described by the temperature and
pressure corresponding to C16 .
Simulations were carried out in the isothermal-isobaric
ensemble for each of the aforementioned systems. From
the resulting trajectories, a set of configurations were sampled and used as a starting point for the FDS rigid body
simulations.
The thermodynamic properties obtained from both
Ryckaert-Bellemans and the corresponding rigid body
models are collected in Table II. Figure 9 compares the

064116-10

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

TABLE II. Main results for a series of n-alkanes at constant pressure and temperature. The flexible models are represented here as Ryckaert-Bellemans on
left half of the table and their frozen counterparts are represented here as rigid body of the right half of the table. The symbols in the first column refer,
respectively, to: molecular temperature, rotational temperature, scalar pressure, torsion angle energy, Lennard-Jones intermolecular energy, root-mean-square
end-to-end distance, root-mean-square radius of gyration, percentage of trans conformations, and self-diffusivity, respectively. The numbers in parentheses are
the statistical uncertainty in the last digits computed from the standard deviations divided by the square root of the number of segments.
Ryckaert-Bellemans

TM 
TR 
p
Utors 
ULJ 
2 1/2
1n
Rg2 1/2
%trans
Ds

Rigid body

C6

C8

C10

C16,NV T

C6

C8

C10

C16,NV T

5
5.003(3)
9.2759(2)
15.49(1)
3.5954(7)
1.3981(4)
0.51403(5)
63.69(5)
0.4149(6)

5
5.000(3)
9.2760(1)
25.71(1)
4.3079(7)
1.8657(3)
0.66092(6)
64.63(4)
0.37296(7)

5
5.003(3)
9.2760(1)
35.95(3)
5.0151(8)
2.2946(4)
0.80008(8)
64.90(3)
0.32114(5)

5
4.999(6)
9.274(6)
66.91(5)
7.110(7)
3.404(3)
1.1799(5)
65.12(4)
0.2245(2)

5
5.01(1)
9.2758(2)
0
3.597(2)
1.399(5)
0.5142(7)
63.8(9)
0.428(9)

5
5.009(9)
9.2760(1)
0
4.311(3)
1.867(9)
0.661(1)
64.6(7)
0.357(6)

5
5.01(1)
9.2761(2)
0
5.020(3)
2.294(9)
0.800(2)
65.0(7)
0.284(9)

5
5.000(7)
9.27(2)
0
7.10(2)
3.41(3)
1.181(6)
65.1(6)
0.1251(1)

self-diffusion coefficients for the flexible and rigid body nalkane systems as a function of chain length at both constant
density and constant pressure conditions. For larger chain
lengths, the coupling between torsional and translational motions has an increasingly significant role with regards to the
mechanism of diffusion. For smaller chain lengths, the selfdiffusion coefficient decreases with increasing chain length as
expected.
Further insight into such coupling may be gained from
short time angular velocity correlation functions measured in
the space-fixed frame. Figures 10 and 11 compare the angular
velocity autocorrelation functions  (t) (0) of n-hexane
and n-decane (flexible and rigid body) at constant density
and constant pressure, respectively. The observed behaviour
of all flexible n-alkane systems (blue curve) is typical of a
low torque fluid. The correlation function of n-hexane at constant pressure decays at a lower rate than the corresponding
correlation function of n-hexane at constant volume. At constant pressure, n-hexane is at a lower density than the same
system at constant volume and the larger mean intermolecular distance results in a weaker interaction with the neigh-

FIG. 9. Comparison of the self-diffusion coefficients for the flexible and

frozen systems as function of chain length, measured here by the number
of carbon atoms present in the alkane chain.

bouring molecules. This has the physical effect that n-hexane

molecules at constant pressure retain memory of their previous angular velocity for longer times than n-hexane molecules
at constant volume. The difference between constant volume
and constant pressure systems is larger for n-hexane than for

FIG. 10. Comparison between the angular velocity autocorrelation functions

(t) (0) obtained from the average of all rigid body segments (in red)
and the corresponding flexible system (in blue). Top distribution corresponds
to n-hexane at constant volume (a) and bottom distribution corresponds to
n-hexane at constant pressure (b), respectively.

064116-11

C. Braga and K. P. Travis

J. Chem. Phys. 137, 064116 (2012)

clearly the principal mechanism of diffusion. In the other

limit, for the state point discussed, a small molecule such as
n-hexane may be indistinguishable from its frozen counterpart. This means that torsional motions do not play a significant role in the diffusion mechanism at the studied density
and temperature. The comparison of this coupling mechanism
for different molecular systems is not straightforward as these
do not share the same equilibrium configurational properties.
Performing simulations at constant temperature and constant
pressure is a way to remove the volume dependence of such
behaviour.
ACKNOWLEDGMENTS

The authors thank Dr. Peter Daivis for the helpful discussions during the course of this work. This work has been
funded by Engineering and Physical Sciences Research Council (UK) (EPSRC) Grant No. GR/R13265/01.
1 K. P. Travis, D. Brown, and J. H. R. Clarke, J. Chem. Phys. 98, 1524 (1993).
2 K.

FIG. 11. Comparison between the angular velocity autocorrelation functions

(t) (0) obtained from the average of all rigid body segments (in red)
and the corresponding flexible system (in blue). Top distribution corresponds
to n-decane at constant volume (a) and bottom distribution corresponds to
n-decane at constant pressure (b), respectively.

n-decane. All correlation functions are weakly damped in

comparison with their corresponding rigid body ones. Rigid
body molecules are not able to change their shape in response to their environment and hence experience significant
torques when compared with their flexible counterparts. This
results in a faster decorrelation of the angular velocity of the
molecule.
V. CONCLUSIONS

The application of the FDS method as a tool to study

the role played by intramolecular degrees of freedom (represented here by the torsional motions) has been demonstrated
for a series of linear n-alkanes. The FDS method was successful in generating flexible and frozen liquids of molecules
which share the same equilibrium configuration properties
and molecular structure. Any difference between the transport properties of these systems can therefore be ascribed to
the coupling of torsional to translational degrees of freedom
thus establishing a direct relation between the internal degrees
of freedom and the diffusion properties of molecules.
The degree of coupling increases with chain length and
for a molecule, such as n-triacontane, these motions are

P. Travis, D. Brown, and J. H. R. Clarke, J. Chem. Phys. 102, 2174

(1995).
3 G. P. Morriss, P. J. Daivis, and D. J. Evans, J. Chem. Phys. 94, 7420 (1991).
4 M. Mondello and G. S. G. Grest, J. Chem. Phys. 103, 7156 (1995).
5 S. Karaborni and J. P. OConnell, J. Chem. Phys. 92, 6190 (1990).
6 V. A. Harmandaris, M. Doxastakis, V. G. Mavrantzas, and D. N.
Theodorou, J. Chem. Phys. 116, 436 (2002).
7 P. J. Daivis, D. J. Evans, and G. P. Morriss, J. Chem. Phys. 97, 616 (1992).
8 J. H. R. Clarke and D. Brown, J. Chem. Phys. 86, 1542 (1987).
9 S. T. S. Cui, S. A. S. Gupta, P. T. P. Cummings, and H. D. Cochran,
J. Chem. Phys. 105, 1214 (1996).
10 D. Brown, J. H. R. Clarke, M. Okuda, and T. Yamazaki, J. Chem. Phys.
100, 1684 (1994).
11 M. Mondello, G. S. G. Grest, E. B. Webb, P. Peczak, and E. Webb III,
J. Chem. Phys. 109, 798 (1998).
12 D. Brown and J. H. R. Clarke, J. Chem. Phys. 86, 6446 (1987).
13 R. Walser, A. Mark, and W. van Gunsteren, Chem. Phys. Lett. 303, 583
(1999).
14 R. Walser, B. Hess, A. Mark, and W. van Gunsteren, Chem. Phys. Lett.
334, 337 (2001).
15 J. H. R. Clarke and D. Brown, Mol. Phys. 58, 815 (1986).
16 G. Ciccotti and M. Ferrario, Mol. Phys. 47, 1253 (1982).
17 M. Fixman, Proc. Natl. Acad. Sci. U.S.A. 71, 3050 (1974).
18 E. Helfand, J. Chem. Phys. 71, 5000 (1979).
19 J. J.-P. Ryckaert, G. Ciccotti, and H. J. C. H. Berendsen, J. Comput. Phys.
23, 327 (1977).
20 J. McKechnie, D. Brown, and J. Clarke, Macromolecules 25, 1562 (1992).
21 D. Brown and J. H. R. Clarke, J. Chem. Phys. 93, 4117 (1990).
22 D. Brown and J. H. R. Clarke, J. Chem. Phys. 92, 3062 (1990).
23 K. P. Travis and D. J. Searles, J. Chem. Phys. 125, 164501 (2006).
24 S. J. Brookes, D. J. Searles, and K. P. Travis, Mol. Simul. 35, 172 (2009).
25 P. J. Daivis and D. J. Evans, J. Chem. Phys. 100, 541 (1994).
26 J. Ryckaert and A. Bellemans, Chem. Phys. Lett. 30, 123 (1975).
27 H. C. Andersen, J. D. Weeks, and D. Chandler, Phys. Rev. A 4, 1597
(1971).
28 G. Ciccotti and J. Ryckaert, Comput. Phys. Rep. 4, 346 (1986).
29 M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids, 2nd ed.
(Oxford University Press, New York, 1989).
30 D. J. Evans, W. G. Hoover, B. H. Failor, B. Moran, and A. J. C. Ladd, Phys.
Rev. A 28, 1016 (1983).
31 R. Edberg, D. J. Evans, and G. P. Morriss, J. Chem. Phys. 84, 6933 (1986).
32 A. Baranyai and D. J. Evans, Mol. Phys. 70, 53 (1990).
33 H. Goldstein, C. P. Poole, and J. L. Safko, Classical Mechanics (AddisonWesley, 2002).
34 R. B. Bird, Dynamics of Polymeric Liquids: Kinetic Theory, Dynamics of
Polymeric Liquids Vol. 2 (Wiley, 1987).
35 W. G. W. Hoover et al., Phys. Rev. A 34, 2499 (1986).
36 J. Baschnagel, K. Qin, W. Paul, and K. Binder, Macromolecules 25, 3117
(1992).

The Journal of Chemical Physics is copyrighted by the American Institute of Physics (AIP). Redistribution of
journal material is subject to the AIP online journal license and/or AIP copyright. For more information, see
http://ojps.aip.org/jcpo/jcpcr/jsp