You are on page 1of 14

Electrostatic interactions in dissipative particle dynamics—simulation of

polyelectrolytes and anionic surfactants


R. D. Groot

Citation: The Journal of Chemical Physics 118, 11265 (2003); doi: 10.1063/1.1574800
View online: http://dx.doi.org/10.1063/1.1574800
View Table of Contents: http://aip.scitation.org/toc/jcp/118/24
Published by the American Institute of Physics

Articles you may be interested in


Dissipative particle dynamics: Bridging the gap between atomistic and mesoscopic simulation
The Journal of Chemical Physics 107, 4423 (1998); 10.1063/1.474784

Electrostatic interactions in dissipative particle dynamics using the Ewald sums


The Journal of Chemical Physics 125, 224107 (2006); 10.1063/1.2400223

Erratum: “Electrostatic interactions in dissipative particle dynamics—simulation of polyelectrolytes and anionic


surfactants” [J. Chem. Phys. 118, 11265 (2003)]
The Journal of Chemical Physics 119, 10454 (2003); 10.1063/1.1621380

Dynamic simulation of diblock copolymer microphase separation


The Journal of Chemical Physics 108, 8713 (1998); 10.1063/1.476300

Bead–bead interaction parameters in dissipative particle dynamics: Relation to bead-size, solubility parameter,
and surface tension
The Journal of Chemical Physics 120, 1594 (2004); 10.1063/1.1630294

Many-body dissipative particle dynamics simulation of liquid/vapor and liquid/solid interactions


The Journal of Chemical Physics 134, 204114 (2011); 10.1063/1.3590376
JOURNAL OF CHEMICAL PHYSICS VOLUME 118, NUMBER 24 22 JUNE 2003

Electrostatic interactions in dissipative particle dynamics—simulation


of polyelectrolytes and anionic surfactants
R. D. Groota)
Unilever Research Vlaardingen, P. O. Box 114, 3130 AC Vlaardingen, The Netherlands
共Received 6 December 2002; accepted 25 March 2003兲
Electrostatic interactions have been incorporated in dissipative particle dynamics 共DPD兲 simulation.
The electrostatic field is solved locally on a grid. Within this formalism, local inhomogeneities in the
electrostatic permittivity can be treated without any problem. Key issues like the screening of the
potential near a charged surface and the Stillinger–Lovett moment conditions are satisfied. This
implies that the method captures the essential features of electrostatic interaction. For the direct
simulation of mixed surfactants near oil–water interfaces, or for the simulation of Coulombic
polymer–surfactant interactions, this method has all the advantages of DPD over full atomistic
molecular dynamics 共MD兲. DPD has proven to be faster than MD by many orders of magnitude,
depending on the precise scaling factor chosen for the simulation. This brings phenomena of
microseconds in reach of routine simulation, while maintaining a fairly accurate representation of
the structure of the molecules. As an example of this simulation tool, the interaction between a
cationic polyelectrolyte and anionic surfactant is discussed. Without a surfactant, the polyelectrolyte
shows a fractal dimensionality that is in line with the theoretical and experimental values cited in
literature, it behaves as a fairly stiff rod, d f ⬃1.1. When salt is replaced by anionic surfactant, the
polymer wraps around one or more discrete surfactant micelles, in line with the current
understanding of these systems, and scaling invariance in the correlation function is broken.
© 2003 American Institute of Physics. 关DOI: 10.1063/1.1574800兴

I. INTRODUCTION dral angles within the same molecule. On a time scale of a


Mesoscopic simulation is a tool that is aimed at model- few tens of picoseconds, trans-gauche isomerizations of the
ing physical and chemical phenomena at length and time dihedrals occur.9 On a time scale of a few nanoseconds, the
scales, which are intermediate between the atomic length phospholipids rotate around their axis, and on the time scale
scale and the macroscopic length scale. The atomic length of tens of nanoseconds, two lipids switch place within one
scale here is defined as the length scale to which full mo- bilayer, giving rise to lateral diffusion. Within this time scale,
lecular dynamic 共MD兲 simulations are limited, and ‘‘macro- the individual lipids orient, and lipid membranes show
scopic’’ starts at a length scale where the materials are suffi- protrusions.10 Finally, on a time scale of 100 ns peristaltic
ciently homogeneous to justify a continuum description. In motions and undulations occur.11
between these length scales, there is a vast range where nei- By virtue of parallelization over several processors or
ther continuum methods nor atomic simulation can be ap- PC clusters, hardware developments have now pushed the
plied. This runs from some 10 nm to about 100 ␮m. limit of molecular simulation to 100 ns.11 Nevertheless, there
Phenomena that occur at these length scales are polymer is a limit beyond which hardware developments cannot help
adsorption,1 polymer–surfactant interaction,2 mesophase for- us. For instance, phenomena, like the cooperative motions in
mation of surface-active materials and block copolymers,3 phase transitions and membrane fusion, occur on much
and the breakup and coalescence of droplets in an emulsion larger time scales and are well outside the range of current
during processing.4 Furthermore, many biologically relevant simulation power. This requires simulation of the microsec-
processes occur at this length scale, like transport phenom- ond range, while a different set of phenomena could be stud-
ena through living cells,5 the structuring of cell membranes, ied if we could address the millisecond time scale.
and their perforation under the influence of surfactants or The question thus arises as to how these phenomena can
adsorbing polypeptides.6 Finally, the formation of polymer be modeled. One approach is the dissipative particle dynam-
networks takes place at this intermediate length scale.7 ics 共DPD兲 method. In this method, a number of atoms are
Apart from this length scale gap, there is also a time taken together into one simulation bead, that is used as a
scale gap between the macroscopic world and what can be simulation element. The reliability of the result obviously
reached by conventional MD. If we take the example of lipid depends on how the underlying atoms translate into the in-
bilayers, we find other phenomena to occur on every next teraction parameters between the DPD beads. The method
time scale studied.8 On the shortest time scale of a few pi- has been used with considerable success to model a wide
coseconds, lipids show bond and angle fluctuations of dihe- range of physical–chemical problems as discussed next.
A brief overview of the DPD method is now given. In
a兲
Electronic mail: rob.groot@unilever.com this simulation method, all particles are soft beads that inter-

0021-9606/2003/118(24)/11265/13/$20.00 11265 © 2003 American Institute of Physics


11266 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

act with the pair-wise force f i ⫽⌺ j (F Ci j ⫹F Ri j ⫹F D


i j ) where the limit k→0, as this determines the free-energy change asso-
sum runs over all neighboring particles within a certain dis- ciated with density fluctuations. This, in turn, is, related to
tance R c . All forces depend on coordinate differences. The the compressibility and solubilities of the components. Note
conservative force is chosen as12 that the pressure itself drops out in an NVT ensemble, as this


is a linear variation of the free energy. This condition implies
⫺a i j 共 1⫺ 兩 ri j 兩 /R c 兲 r̂i j if 兩 ri j 兩 ⬍R c
F Ci j ⫽ 共1兲 that the system should satisfy
if 兩 ri j 兩 ⬎R c ,
冉 冊 冉 冊冉 冊
0
1 ⳵p 1 ⳵n ⳵p
where a i j is a maximum repulsion between particle i and ⫽ •
kT ⳵␳ kT ⳵␳ ⳵n
particle j, ri j ⫽r j ⫺ri and r̂i j ⫽ri j / 兩 ri j 兩 . Between neighboring simulation experiment
particles on a chain, an extra spring force is defined that
binds the particles together, given by ⫽ 冉 冊
Nm ⳵p
kT ⳵ n experiment
, 共5兲
F Si j ⫽4ri j if i is connected to j. 共2兲
where N m is the number of water molecules, e.g., that each
The drag force F D R
i j and the random force F i j act as heat sink DPD bead represents. N m can been chosen freely in general,
and source, respectively, so their combined effect is a ther- it is a real-space renormalization factor of the simulation. For
mostat. They are given by F Ri j ⫽ ␴␻ (r i j )r̂i j ␨ / 冑␦ t and F D ij the choices N m ⫽1 and N m ⫽3, the compressibility of water
⫽⫺1/2␴ 2 ␻ (r i j ) 2 /kTr̂i j (vi j "r̂i j ), where ␨ is a random vari- at room temperature is matched if the repulsion parameter in
able with zero mean and variance 1, and ␻ (r)⫽(1⫺r) for Eq. 共1兲 between particles of the same type is given by a ii
r⬍1 and ␻⫽0 for r⬎1. ⫽25 and by a ii ⫽78, respectively.13,18
In this article, a fixed noise amplitude ␴⫽3 is used, fol- Groot and Rabone recently showed18 that the effective
lowing Groot and Warren.13 The precise value of the noise time scale ␶ of the simulation can be found by matching the
amplitude and friction in this range has little influence on the bead diffusion constant in the simulation to that of pure wa-
efficiency of the simulation algorithm, it merely provides an ter. Following their analysis for a fixed repulsion parameter
efficient thermostat. This particular thermostat is special in a ii ⫽25, we have
that it conserves 共angular兲 momentum, which leads to a cor-
rect description of hydrodynamics.14 The particle mass, tem- N m D simR 2c
perature, and interaction range are chosen as units of mass, ␶⫽ ⫽25.7⫾0.1N m
5/3
关 ps兴 . 共6兲
D water
energy, and length, hence m⫽kT⫽R c ⫽1 so that the simu-
lated time is expressed in the natural unit of time In Eq. 共6兲, it is implicitly assumed that the repulsion param-
eter between equal beads is fixed at the value of a⫽25, and
␶ ⫽R c 冑m/kT. 共3兲 that the bead density is fixed at ␳⫽3. In practical terms, this
The DPD method, in general, has been shown to produce a means that time steps of 9.6 ps are taken when N m ⫽3 and
correct 共NVT兲 ensemble as the fluctuation–dissipation rela- a ii ⫽25. A comprehensive table of all parameter values used
tion is satisfied.13,15 At every time step, the set of positions is given in Sec. IV.
and velocities, 兵 ri ,vi 其 is updated from the positions and ve- At this point, we can understand why the DPD method is
locities at earlier time using a modified version of the so much faster than straightforward MD. There are two com-
velocity-Verlet algorithm: bined effects that lead to speedup. In an ordinary liquid, like
water, the interaction potential contains a hard core repul-
r i 共 t⫹ ␦ t 兲 ⫽r i 共 t 兲 ⫹ ␦ t v i 共 t 兲 ⫹ 21 ␦ t 2 f i 共 t 兲 , sion. This hard core leads to a caging effect, so that an atom
undergoes many collisions before it is actually transported.
ṽ i 共 t⫹␭ ␦ t 兲 ⫽ ṽ i 共 t 兲 ⫹␭ ␦ t f i 共 t 兲 , The soft potential used here removes this caging effect, so
共4兲 that the diffusivity of particles is increased by a factor of
f i 共 t⫹ ␦ t 兲 ⫽ f i 共 r i 共 t⫹ ␦ t 兲 , ṽ i 共 t⫹␭ ␦ t 兲兲 ,
1000. The second factor leading to fast simulation is the
scaling of the physical time with the renormalization factor
v i 共 t⫹ ␦ t 兲 ⫽ v i 共 t 兲 ⫹ 21 ␦ t 共 f i 共 t 兲 ⫹ f i 共 t⫹ ␦ t 兲兲 .
N m as in Eq. 共6兲. On top of the power 5/3 by which the
As the masses of the particles are put at 1, the force acting on physical time scale increases, the amount of central process-
a particle equals its acceleration. The force is updated once ing unit 共CPU兲 time will decrease inversely proportional to
per iteration. Because the force depends on the velocities, the N m if we want to simulate a given volume, simply because
velocity in the next time step has to be estimated by a pre- we have to update the position of fewer objects. Thus, for a
dictor method. This is done in the second step of the algo- given system volume, DPD can be expected to be faster than
rithm. The velocity is corrected in the last step. If parameter MD by a factor of roughly 1000 N m 8/3
⬇2⫻104 for N m ⫽3
␭ is put at ␭⫽0.5, this scheme equals the velocity-Verlet and about 10 for N m ⫽6, independent of hardware and dis-
5

algorithm.16 However, here we use ␭⫽0.65, where we find a regarding the CPU time spent on evaluating the 共relatively
very accurate temperature control even at the time-step ␦ t long-ranged兲 Lennard-Jones potential.
⫽0.06␶ that is used. A more systematic study into the influ- After Hoogerbrugge and Koelman12 introduced this
ence of parameter ␭ was presented by Den Otter and simulation method and a physical basis was given by Es-
Clarke.17 pañol and Warren,15 and by Groot and Warren,13 it has been
To find, in practice, the interaction parameters for this applied to a range of mesoscopic problems. Some examples
model, one needs to match the liquid structure function in the are polymer–polymer surface tension and block copolymer
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11267

mesophase formation,19,20 and polymer–surfactant interac- In Eq. 共7兲, express the charge density ␳ e (r)⫽⌺ i e i ␦ (r⫺r i )
tions in bulk solution.21 The scaling laws of the polymer was substituted for the sum of individual charges, and e i
endpoint distribution have been checked by Kong et al.22 and ⫽eq i is the charge of ion i. Some care must be taken to
by Spenley.23 The method has been applied to describe sur- subtract the self-energy of the charges, but this issue will not
factant mesophases by Jury et al.24 Clark et al.25 further ap- be addressed here.
plied this method to the breakup of liquid droplets by hydro- The idea behind a field formulation is that all charges
dynamic interactions, while Novik and Coveney26 used this interact with a local field, ␾ (r), which propagates from one
method to study spinodal decomposition. The rheology of a charge to another. To find this field, we first re-express the
colloidal suspension was simulated by Boek et al.,27 the charge density by its Fourier transform ␳ e (k)
structure and rheology of biomembranes was studied by ⫽ 兰 exp(ik"r) ␳ e (r)d 3 r. The Fourier transform of the interac-
Groot and Rabone18 and, finally, the spontaneous formation tion 1/(4 ␲ ⑀ 0 ⑀ r r) is obtained as 1/⑀ 0 ⑀ r k2 and, hence, the en-
of vesicles was studied recently by Yamamoto et al.28 ergy of the system follows as
DPD is thus demonstrated to be a versatile method that
can be applied to modeling on various length and time
scales. When the problem is defined on a large length scale,
U⫽
1
2
冕␳ e 共 k 兲 ␳ e 共 ⫺k 兲

⑀ 0⑀ rk 2
d 3 k. 共8兲
e.g., the breakup of droplets in viscous flow, the method can
be used effectively by matching surface tensions and liquid Now, the electrostatic field is introduced as the function that
viscosities. However, when modeling specific molecules, we maximizes the following functional:

冕冋 册
still lack a quantitative relation that links atoms and mol- 1
ecules to the DPD beads. An initial link to molecules was ⍀⫽max ⑀ ⑀ ␾ 共 r 兲 ⌬ ␾ 共 r 兲 ⫹ ␾ 共 r 兲 ␳ e共 r 兲 d 3r
␾ 2 0 r
given by Groot and Warren,13 who showed that the thermo-

冕冋
dynamics of the simulation model is very close to that of the 1
Flory–Huggins model, and a method to determine the rel- ⫽max ⫺ ⑀ 0 ⑀ r k 2 ␾ 共 k 兲 ␾ 共 ⫺k 兲
2
evant Flory–Huggins ␹-parameters from experimental data ␾


has been described.18 An alternative way to coarse-grain the
1 1
potential is to start with MD simulated pair-correlation func- ⫹ ␾ 共 k 兲 ␳ e 共 ⫺k 兲 ⫹ ␾ 共 ⫺k 兲 ␳ e 共 k 兲 d 3 k
tions, and extract an effective potential for a coarse-grained 2 2


model from these, by requiring that the DPD simulated pair- 1
correlation matches the MD result.29–31 However, so far, ⫽ 关 ␳ e 共 k 兲 ␳ e 共 ⫺k 兲 / ⑀ 0 ⑀ r k 2 兴 d 3 k. 共9兲
2
electrostatic interactions have not been considered in combi-
nation with the soft DPD potential. One practical reason for This functional is maximized by the field
this is that the soft particles are allowed to overlap, which
may give rise to strong ion pairing. k 2␾ 共 k 兲 ⫽ ␳ e共 k 兲 / ⑀ 0⑀ r or ⌬ ␾ 共 r 兲 ⫽⫺ ␳ e 共 r 兲 / ⑀ 0 ⑀ r ,
For many molecular problems, however, electrostatics 共10兲
are of paramount importance, e.g., for the description of which is, in fact, Poisson’s equation for the electrostatic field
biomembranes. Therefore, a method to tackle this problem is E⫽⫺ⵜ ␾ . At this point, the correspondence between the
described here. In Sec. II, a review is given of the interrela- charge–charge interaction in Eq. 共7兲 and the field formula-
tion between 1/r interactions and a field theoretic descrip- tion in Eq. 共9兲 is made.
tion. How to apply this theory to mesoscopic simulations, In many practical cases, the dielectric permittivity that
particularly the DPD model, is described in Sec. III and vali- appears in Eq. 共10兲 is not constant throughout space, for
dated in Sec. IV, and a brief example of an application is instance, when a water/oil interface is considered. Therefore,
given in Sec. V to demonstrate the potential of the method. the dielectric displacement D was introduced by Maxwell,
which he defined by D⫽ ⑀ r E, and which satisfies the field
equation
II. FIELD FORMALISM
ⵜ•D 共 r 兲 ⫽ ␳ e 共 r 兲 / ⑀ 0 . 共11兲
In this section, the basic electrostatic field theory is
The field D incorporates the local polarization of the me-
briefly reviewed. The aim of this work is to describe the
dium, P⫽( ⑀ r ⫺1)E. Variations in the local value of ⑀ r sig-
behavior of an ensemble of ions, polyelectrolytes, and
nify a change in the local polarization that, in itself, become
charged particles in a mesoscopic simulation. For this pur-
a source for the electrostatic E field, but not for the D field. If
pose, two options are open: Either all interactions between
we substitute the relation between D, ⑀ r , and ␾ into Eq. 共11兲,
all charges are summed in real space or, instead, all particles
we find
are coupled to a local electrostatic field, for which the field
equations are solved separately. The first option comes down ⵜ• 共 ⑀ r 共 r 兲 ⵜ ␾ 共 r 兲兲 ⫽⫺ ␳ e 共 r 兲 / ⑀ 0 . 共12兲
to the following expression for the internal 共electrostatic兲 en-
ergy: One advantage of using a field formulation is that it is quite
straightforward to do simulations for liquids with different
⫽ 冕 冕
ee 1 ␳ 共 r 兲␳ 共 r⬘兲
U⫽ 兺 i j
i⬍ j 4 ␲ ⑀ 0 ⑀ r 兩 r i ⫺r j 兩 2
e e
4 ␲ ⑀ 0 ⑀ r 兩 r⫺r ⬘ 兩
d 3 rd 3 r ⬘ .
dielectric permittivity, whereas a summation over explicit
charge interactions involves summations over mirror image
共7兲 charges as well 共Sec. IV D兲.
11268 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

III. HOW TO APPLY FIELD THEORY TO MESOSCOPIC ␶ ⫽25.7N m5/3 ps⫽160 ps for N m ⫽3, a⫽25, ␴ ⫽3.
SIMULATION 共20兲
A. Scaling the parameters The pressure near a charged wall follows from the contact
The dimensionless electrostatic field that will be used in theorem by Hendersom, Blum, and Lebowitz,32 which, in
simulations is turn, follows from a force balance argument near a hard wall.
This contact theorem states that
␺⫽␤e␾. 共13兲
␴2
The Poisson equation for this field is p⫽kT 兺 ␳ contact⫺
2⑀
ⵜ• 共 ⑀ ⵜ ␺ 兲 ⫽⫺ ␤ e ␳ . 2
共14兲
1
In DPD, liquid elements are represented by beads that ⇒p * ⫽ pR 3c /kT⫽ 兺 ␳ contact
* ⫺ ⌫ ␴ *2,
2
共21兲
interact with each other up to a cutoff distance R c . It is
convenient to take this distance as the unit of length where ␴ is the surface charge in Cm⫺2 and ␴* is the surface
in the simulations, and to take kT as unit of energy. To make charge in unit charges per R 2c . For a bulk system, the pres-
Eq. 共14兲 dimensionless, we will extract the dielectric sure follows from the virial theorem, which can be expressed
permittivity ⑀ ⫽ ⑀ 0 ⑀ r (r) by defining a permittivity rela- as16
tive to the value in pure water: ⑀⫽⑀ 0 ⑀ r p(r). Here, ⑀ 0 1
⫽8.854 187 82 10⫺12 C2 J⫺1 m⫺1 , the dielectric constant of p⫽kT 兺␮ ␳ ␮ ⫹ 3V 兺i 共 x i f x,i ⫹y i f y,i ⫹z i f z,i 兲 , 共22兲
vacuum, ⑀ r ⫽78.3 is the relative permittivity of water at
25 °C, and p⫽1 for water and p⬇0.025 for hydrocarbons. where ␮ runs over all particle types, and i runs over all
Multiplying Eq. 共14兲 by R 2c and dividing by ⑀ 0 ⑀ r , we find particles in the system, (x i ,y i ,z i ) is the position of particle i,
the dimensionless equation and ( f x,i , f y,i , f z,i ) is the conservative force acting on par-
ticle i. This is the sum of the conservative forces given in
e 2 ␳ R 3c
R 2c ⵜ• 共 p 共 r 兲 ⵜ ␺ 兲 ⫽⫺ Eqs. 共1兲 and 共2兲, and the electrostatic forces acting on the
kT ⑀ 0 ⑀ r R c particles that will be discussed in Sec. III C. Even when the
⇒ⵜ * • 共 p 共 r 兲 ⵜ * ␺ 兲 ⫽⫺⌫ ␳ * . 共15兲 polarizability is a function of (x,y,z), the same expression
applies, as the effect of polarizability and therefore mirror
Here, ␳* is the concentration of cations minus that of anions charge effects are included in the field and therefore also
or per R 3c , ⵜ* is the gradient in DPD units, and p(r) is the included in the forces. This expression also contains all
local polarizability relative to pure water: forces caused by image charges that result from the periodic
p 共 r 兲 ⫽ 具 p i 典 i苸cell . 共16兲 boundary conditions used. Note that the virial equation here
does not contain coordinate differences but the absolute co-
The average runs over all particles in a cell near r, and p i is ordinates. However, the center of mass of the system drops
the polarizability of particle i in this cell. The precise details out of the equation because the sum of all forces vanishes.
are given in Appendix II.
The unit of length R c can be related to the volume of a
DPD bead. Let N m be the number of water molecules that B. Smearing out the charges
each DPD bead represents. Then, by matching the volume of Before we can solve the field equations, we need to
a bead to that of N m water molecules,18 we find at bead choose a model for the density distribution that corresponds
density ␳⫽3 with a charged bead. Point charges would collapse on top of
R c ⫽0.448 14N m
1/3
关 nm兴 ⫽0.64 633 关 nm兴 for N m ⫽3. 共17兲 each other, forming infinitely strongly bound ion pairs. This
is a consequence of the soft core interaction presently used in
This relation implies that the coupling constant at room tem- DPDs. Therefore, we need to spread out the charge. For each
perature is given by unit charge in the simulation, we choose the charge distribu-
e2 tion
⫺1/3
⌫⫽ ⫽20.00N m ⫽13.87 for N m ⫽3, 共18兲
kT ⑀ 0 ⑀ r R c 3
f 共 r 兲⫽ 共 1⫺r/R e 兲 for r⬍R e , 共23兲
and, likewise, the unit of concentration is ␲ R 3e
R ⫺3 ⫺1
c ⫽11.11N m
⫺1
关 nm⫺3 兴 ⫽18.45N m 关 Mol/l兴 where R e is the electrostatic smearing radius, and f (r)⫽0
for r⬎R e . The interaction potential between two of these
⫽6.150 关 Mol/l兴 for N m ⫽3. 共19兲
charge clouds is worked out in Appendix I. A plot of the
⫺2
Hence, if a salt concentration of 10 M is to be simulated, exact potential between two smeared-out charges and its first
the concentration in DPD units should be 10⫺2 /6.15⫽1.63 derivative is given in Fig. 1.
⫻10⫺3 , leading to a screening length in DPD units of ␬ ⫺1 The depth of the potential well between two unlike
⫽1/冑 (2⌫c)⫽4.71 R c ⫽3.0 nm. As discussed by Groot and charges is 52/35 ⌫/4␲ R e ⬇1.64/R e when we take the ex-
Rabone,18 the time scale is fixed by matching the diffusion ample of N m ⫽3 for which in water ⌫⫽13.87. Now if the
constant of water. For the repulsion parameter a⫽25, we interaction potential between two ions at contact is much
have lower than ⫺kT they may form an ion pair, whereas they are
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11269

the coupling constant ⌫ does not appear in this expression


for the energy, as the coupling constant is already accounted
for in the solution of the field. The force that is exerted on an
ion in this field is given by

i 共 r 兲 ⫽⫺q i
F el 冕 f 共 r⫺r ⬘ 兲 ⵜ ␺ 共 r ⬘ 兲 d 3 r ⬘ . 共26兲

The field equations, in turn, are solved by maximizing the


functional 关see Eq. 共9兲兴

⍀⫽ 冕 ⫺ p共 r 兲
共 ⵜ * ␺ 共 r 兲兲 2
2⌫
⫹ ␺¯␳ d 3 r. 共27兲

This free energy also implies a force on the neutral particles,


FIG. 1. Electrostatic potential between linearly smeared-out charges. The which is given by

冕 ␦␦ ␳
lower curve gives the force; the dashed curve gives the 1/r potential be-
tween point charges. ⳵ ␦⍀ ⳵ p 共 r ⬘ 兲 共 ⵜ * ␺ 共 r ⬘ 兲兲 2 3
f m ⫽⫺ ⫽ d r ⬘.
⳵ r i ␦ ␳ i共 r 兲 ⳵ r i i共 r 兲 2⌫
共28兲
unlikely to pair if the potential well is shallower. This im-
This force is the Maxwell force on the dielectric medium. Its
plies that ion pairing might occur if R e ⬍1.64, whereas the
effect is that a medium of low dielectric permittivity 共e.g.,
electrolyte will completely dissociate if R e ⬎1.64. Since, in
oil兲 is displaced by a medium of high dielectric permittivity
practice, there is also a short-range ion–ion repulsion in the
共e.g., water兲 if a strong electric field is applied. Since this is
DPD potential that counteracts ion pairing, a reasonable
a second-order effect in the field, it has been neglected in the
choice for the smearing radius is R e ⫽1.6.
simulations described next but, in principle, this effect can be
Following the method by Beckers et al.,33 the electro-
taken into account in the present method.
static field is solved on a lattice. In their algorithm, the
Rather than using a Fourier transform method to solve
charges are spread out over the lattice nodes, and the long-
the field equations, a real-space successive overdamped re-
range part of the interaction potential is calculated by solving
laxations method is used here. This method has been applied
the Poisson equation on the grid. Details on how the charges
successful by Beckers et al.33 to solve the far field on a lat-
are spread out and how derivatives on the lattice are taken
tice. The basic idea behind this is that the field is updated via
are given in Appendix II. Beckers et al.33 refer to this
method as the particle–particle particle–mesh 共PPPM兲 algo-
rithm, though in the original PPPM algorithm the far field
d␺
dt
⫽␨
D⍀
␦␺
⫽␨
␦⍀
␦␺
⫺ⵜ•
␦⍀
␦ⵜ␺冋 册
⫽ ␨ 关 ⌫¯␳ e ⫹ⵜ• 共 pⵜ ␺ 兲兴 .
was solved using a fast Fourier transform.34 The accuracy of 共29兲
this method was analyzed critically by Deserno and Holm.35
This evolution equation is analogous to that of a particle in a
The present method to solve the field equation is close to the
potential well of energy ⫺⍀, where ␨ is the analog of a
multigrid methods described by Sagui and Darden.36 Another
friction factor. However, the sign is different, because the
alternative is not to solve the field equation, but to make the
solution is not a minimum of the grand potential ⍀, but a
field a dynamic variable such that its correlation matches the
maximum. The maximum is reached by iterating
electrostatic potential,37 and to update these variables during
the simulation using the Monte Carlo algorithm. d␺
␺ new⫽ ␺ old⫹ . 共30兲
dt
C. Solving the field equations
Since the smallest wave that fits the lattice is sin(␲x), ␨
To solve the field equations for an ensemble of smeared should be of the order 1/␲ 2 ⫽0.1 for a one-step convergence
charges, we first consider the energy of a charge density ␳ e of the highest-frequency modes. Larger values of ␨ lead to a
in a field ␺. This is given by faster convergence of the longer waves, but if ␨⬎0.2, the

冕冕 ␺
smallest wave does not converge anymore. Therefore,
U⫽ 共 r 兲 f 共 r⫺r ⬘ 兲 ␳ e 共 r ⬘ 兲 d 3 rd 3 r ⬘ ␨⫽0.15 is a good compromise.
To monitor how accurate the field is solved, the root-

⫽ 冕␺ 共 r 兲¯␳ e 共 r 兲 d 3 r, 共24兲
mean-square value of the right-hand side of Eq. 共29兲, which
should converge to zero, is compared to the root-mean-
square value of the divergence of the field. This gives the
where the locally averaged density is defined by relative error

¯␳ e 共 r 兲 ⫽ 冕 f 共 r⫺r ⬘ 兲 ␳ e 共 r ⬘ 兲 d 3 r ⬘ , 共25兲
err⫽ 冑 兺 cells关 ⌫¯␳ e ⫹ⵜ• 共 pⵜ ␺ 兲兴 2
. 共31兲
兺 cells关 ⵜ• 共 pⵜ ␺ 兲兴 2
and where ␳ e (r) is the nonsmeared charge density, ␳ e (r)
⫽⌺ i q i ␦ (r⫺r i ). Here, i runs over all charges in the system, If the previous solution to the field, or a time extrapolation
and q i ⫽⫾1 is the number of unit charges of ion i. Note that thereof, is taken as initial guess for the field, some six to
11270 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

TABLE I. Summary of parameters used, unless stated otherwise.

Parameter Value Meaning

a ii 25 Repulsion between identical particle types


err 0.03 Relative error to which field equation is solved, Eq. 共31兲
kT 1 Temperature
m 1 Particle mass
Nm 3 Number of water molecules represented by one bead
Rc 0.646 33 nm Range of soft repulsive interaction, Eq. 共17兲
Re 1.6 R c Range over which charges are smeared, Eq. 共23兲
Rn 2 Rc Range over which neighboring grid points are searched
⌫ 13.87 Electrostatic coupling, Eqs. 共15兲 and 共18兲
␦t 0.06␶ Time step
␭ 0.65 Velocity prediction parameter in update algorithm, Eq. 共4兲
␴ 3 Noise amplitude
␶ 160 ps Characteristic time scale, Eqs. 共3兲 and 共6兲
␨ 0.15 Friction factor in solving field equations, Eq. 共29兲

seven iterations are sufficient to solve the equation up to ⫹a exp(␬(x⫺20))⫹ ␺ ⬁ , where a, ␬, and ␺ ⬁ are free fit pa-
relative error 0.03 in a box of size 20⫻20⫻20 containing rameters. Over the range 2⬍x⬍18, this led to a screening
1600 ions. The total CPU time necessary to distribute the parameter ␬⫽0.572⫾0.009, i.e., an exact match with the
charges, solve the field equation, and calculate the forces on value it should have 共see Fig. 2兲.
the particles is small compared to the rest of the CPU time To check this further, the density profiles and field were
used per time step. In theory, the sum of all forces that act on checked for consistency with the Poisson–Boltzmann equa-
all charges should vanish. However, in practice, a small re- tion. If this equation is valid, the concentration of positive
sidual force per ion still appears to be left over that fluctuates and negative ions should follow ␳ ⫹ ⫽c exp(⫺␺) and ␳ ⫺
with an amplitude of order 0.000 05, even if the charge dis- ⫽c exp(␺). Therefore, ⫺ln(␳⫹ /c) and ln(␳⫺ /c) are plotted
tribution around each particle is balanced as described in together with the field in Fig. 2. It is clear from Fig. 2 that
Appendix II. To conserve total momentum, this residual the match is very good. The only deviations occur near the
force is, therefore, subtracted again from all charges. This is charged plates as a result of the smearing of the charges. As
a minor correction of the order 10⫺5 relative to the electro- a technical point, it should be noted that to arrive at this
static force. comparison, the field has first been shifted upward by sub-
tracting the constant ␺ ⬁ so that the field extrapolates to ␺⫽0
IV. VALIDATION far away from the charged wall. Such a shift is known as a
The 共implicit兲 parameters used in the validation runs de- gauge transformation. The right gauge follows from the re-
scribed next are summarized in Table I. When other param- quirement that ␳ ⫹ ⫽ ␳ ⫺ ⫽c when ␺⫽0.
eters are used this is stated explicitly.
B. Bulk electrolyte
A. Electric double layers
Bulk electrolytes satisfy two exact relations for the pair
As a first test to validate the method, a salt solution near
correlation function, known as the Stillinger–Lovett moment
a charged plate was studied. In this particular simulation, the
conditions.38 For a symmetric electrolyte, these can be sum-
plate was completely permeable for all particles in the sys-
marized by
tem, and also the electrostatic field was calculated with full
periodic boundary conditions. Thus, the charged plate is rep-
resented by a two-dimensional 共2D兲 array of fixed charges at
x⫽0. The box size was 20⫻10⫻10 in DPD units, which is
12.9⫻6.5⫻6.5 nm3 in physical units. Taking N m ⫽3, the cou-
pling constant ⌫⫽13.87 was used. The plate was charged by
ten unit charges, i.e., ␴*⫽0.1 which corresponds to ␴⫽0.038
Cm⫺2. The box contained 30 counterions, 20 co-ions and
5950 neutral 共water兲 particles. The system was evolved over
5⫻105 time steps, corresponding to 5 ␮s.
To analyze the data, the mean salt concentration c
⫽( ␳ ⫹ ⫹ ␳ ⫺ )/2 was averaged between 4⬍x⬍16. Over this
interval, the concentration was c⫽0.011 836⫾0.000 017, in-
dependent of x. According to Eq. 共19兲, this corresponds to a
concentration of 0.0728 Mol/l. At this concentration, the in-
verse screening length should be ␬ ⫽ 冑 2⌫c⫽0.5730
FIG. 2. Density profiles of positive charges 共⫹兲 and negative charges 共䊊兲,
⫾0.0004. To test this prediction, the field ␺ (x) was fitted to and the mean electrostatic field near a charged plate 共diamond symbols兲. The
a three-parameter function of the form ␺ (x)⫽a exp(⫺␬x) full curve is a fit to a cosh function.
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11271

FIG. 3. Pair-correlation functions between equal and unequal ion pairs in


aqueous solution at 0.6 M, and between neutral particles, in a box of 6.5 FIG. 4. Structure function simulated at ⌫⫽13.87 for salt concentrations c
⫻6.5⫻6.5 nm3, averaged over 1 ␮s physical time. ⫽0.025 共䉱兲, c⫽0.05 共䊊兲, c⫽0.1 共䉭兲 and c⫽0.2 共䊉兲. For comparison, the
dashed curves have the exact limiting slopes 0.18, 0.36, 0.72, and 1.44,
respectively.

k2
lim ch D 共 k 兲 ⫽h 0 ⫹h 2 k 2 ⫹O 共 k 4 兲 ⫽⫺1⫹ ⫹O 共 k 4 兲 ,
2⌫c
k→0
共32兲 ␳ e 共 k兲 ⫽ 兺j q j exp共 ik"rj 兲 ⫽ 兺j q j exp共 2 ␲ in"rj /L 兲 ,
where c is the salt concentration and h D (r)⫽g ⫹⫹ (r) 共34兲
⫺g ⫹⫺ (r). The physical meaning of the first Stillinger– was determined first. Here, n is an integer vector to which
Lovett relation (h 0 ⫽⫺1) is that each charge is exactly the wave vector is related as k⫽2 ␲ n/L, and q j ⫽⫾1 is the
screened by a counter charge. The implication of the second charge of ion j expressed in unit charges. The charge struc-
condition (h 2 ⫽1/关 2⌫c 兴 ) is that the direct correlation be- ture function, in turn, is obtained as the ensemble average of
tween two charges at large separation is exactly given by 共see e.g., Hansen and McDonald兲39
minus the electrostatic potential divided by kT. The pair-
correlation functions g ⫹⫹ (r) and g ⫹⫺ (r) are shown in Fig. S 共 k兲 ⫽ 具 ␳ e 共 k兲 ␳ e 共 ⫺k兲 典 /N⫽1⫹ch D 共 k兲 , 共35兲
3 for salt concentration c⫽0.1 共i.e., 0.6 M兲 and ⌫⫽13.87. where N is the total number of ions in the system. All 924
The box size is 10⫻10⫻10, and the pair-correlation func- modes for which n2 ⭐36 have been monitored from which
tions are averaged over 105 simulation steps, corresponding the angle-average structure function was averaged over 105
to 1 ␮s physical time. The liquid density is ␳⫽3 and the time steps for each system.
repulsion parameter between all particles is a⫽25. Figure 3 For an ideal ion gas, the charge structure function takes
also gives the pair-correlation function for uncharged par- the form39 S(k)⫽k 2 /(2⌫c⫹k 2 ). However, since the soft
ticles. It is found that for the present parameters the correla- short-range DPD interaction and the extended charge per ion
tion function between charged particles to a high degree of effectively excludes ion overlap, the structure function must
accuracy satisfies g ⫹⫹ (r)g ⫹⫺ (r)⫽g(r) 2 , where g(r) is the contain k 4 terms. Therefore, the structure functions are fitted
pair correlation function between neutral particles. The im- to a function of the form
plication is that g ⫹⫹ (r) and g ⫹⫺ (r) satisfy
a 0 k 2 ⫹a 2 k 4
g ⫹⫹ 共 r 兲 ⫽g 共 r 兲 exp关 ⫺ ␸ 共 r 兲兴 S共 k 兲⫽ . 共36兲
2⌫c⫹a 1 k 2 ⫹a 2 k 4
and 共33兲
These fits are shown in Fig. 4, together with the exact limit-
g ⫹⫺ 共 r 兲 ⫽g 共 r 兲 exp关 ⫹ ␸ 共 r 兲兴 , ing laws given by k 2 /2⌫c. Parameter a 0 in Eq. 共36兲 should
take on the value of 1. The fit values actually are a 0 ⫽1.03
where ␸ (r) is an effective electrostatic interparticle poten-
⫾0.02, 1.04⫾0.02, 1.04⫾0.03, and 1.01⫾0.03 for c
tial. For this reason, the perturbation theory may be applied,
⫽0.025 up to c⫽0.2, respectively. Hence, the second mo-
e.g., reference hypernetted chain theory. This is outside the
ment and, therefore, the long-range direct correlation is cor-
scope of this article.
rect up to about 3%, which is also the accuracy by which the
To check the Stillinger–Lovett moment conditions, the
slope was determined.
charge structure function has been determined at a range of k
values for four different salt concentrations, c⫽0.025, 0.05,
C. Electrostatic force
0.1, and 0.2, which correspond to a range from 0.15 to 1.2
M. The simulation box size was L 3 ⫽20⫻20⫻20, so the In Appendix I, the force between two smeared-out
simulations contained, respectively, 400, 800, 1600, and charges is calculated. In this section, the actual force, as
3200 ions. For these simulations, the electrostatic field was obtained from the lattice method, will be compared with the
solved up to a relative accuracy err⫽0.005. To obtain the theoretical curve. Toward this end, a simulation with just two
structure function, the Fourier transform of the charge den- ions immersed in water was run in a box of size 10⫻10⫻10.
sity distribution The electrostatic force exerted on the ions as a function of
11272 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

mogeneous media is described correctly in the simulation


method, therefore, comes down to the question: Is the Pois-
son equation solved correctly if a dielectric discontinuity is
present? Two examples are now described to demonstrate
this. The first problem is that of the field between two fixed
walls, and the second is the interaction between a wall and an
ion in the presence of a dielectric discontinuity. In the former
problem, a box is considered of sizes 80⫻10⫻10, with full
periodic boundary conditions. Two charged walls perpen-
dicular to the x axis are defined, of charge density ␴⫽0.1
共⫽0.038 cm⫺2兲 at x⫽0, and ␴⫽⫺0.1 at x⫽40. Two slabs of
low dielectric oil are inserted at 0⬍x⬍20 and at 60⬍x
⬍80, and one slab of water is inserted at 20⬍x⬍60. The
FIG. 5. Force between two ions as function of distance. The dashed curve polarizability of the water phase is defined as p⫽1 共or ⑀ r
gives the theoretical force for R e ⫽1.6, the full curve is the force for R e
⫽1.725. ⫽78.3) and the polarizability of the oil is p⫽0.5 共or ⑀ r
⫽39.15). The phases are represented by soft beads with re-
pulsion a ii ⫽25 between particles of the same type, and with
radial distance was averaged over the simulation run. This repulsion a i j ⫽57.7 between water and oil, which corre-
leads to the angle-averaged force which, for an infinitely sponds to ␹⫽10.
large system, should be given by the derivative of the poten- The electric field was solved to an error of err⫽0.001
tial given in Appendix I. The charge smearing radius R e was 关see Eq. 共31兲兴, and averaged over 650 time steps after an
taken at R e ⫽1.6. equilibration of 350 steps. The mean electrostatic field thus
In a short run (104 steps兲 some fluctuations in the force contains an average over the fluctuations of the water–oil
are seen to occur at short separation (1⬍r⬍2). Some points interface, and it incorporates the width of the interface. The
fall on the theoretical curve and others fall below. Appar- average field gradient is shown in Fig. 6. Indicated in Fig. 6
ently, there is a different force depending on the orientation. are the locations of the charge planes and the areas where the
One possible cause of this is the nonspherical charge distri-
dielectric permittivity is inhomogeneous. For this geometry,
bution, another cause could be the breaking of spherical
the field can easily be solved analytically, in the low dielec-
and translational symmetry of the field by the lattice.
tric medium, the field should be ⵜ␺ ⫽⫾⌫ ␴ /2p⫽⫾1.387,
An average over a longer run (105 steps兲 gives the mean
and in the high dielectric medium, it should be ⵜ␺⫽⫾⌫␴/2
共spherically averaged兲 force as function of distance, which is
shown in Fig. 5. Because the force in some directions is ⫽⫾0.6935. The actual values found in the simulation are
below the theoretical curve, the overall average is below the ⵜ␺⫽1.385⫾0.001 and ⵜ␺⫽0.6941⫾0.0004, respectively,
expected force. By taking the derivative of the potential in hence, the field in simulation deviates from the exact value
Appendix A as a fit function, the simulated force–distance by 0.1%, which is indeed the error tolerance used in solving
curve can be reinterpreted as the force of charge clouds with the field.
R e ⫽1.725. In the second example studied, a single ion of charge ⫺1
Hence, even though the force does not exactly fit is enclosed in a box of sizes 40⫻10⫻10, with full periodic
the force of the distribution that we aimed to describe, boundary conditions for the electrostatic field, but all par-
this analysis shows that the electrostatic force is reasonably ticles interact with soft impenetrable walls at x⫽0 and at x
well reproduced by the lattice method. The deviations ⫽40. At x⫽0, a surface charge ␴⫽0.01 is positioned, so that
from the force profile given in the Appendix show up at overall charge neutrality is maintained. Again a slab of lower
short distance and are due to lattice artifacts. This difference dielectric oil (p⫽0.5) surrounds this wall, which ranges over
will fall away when the force field is parameterized 0⬍x⬍10 and over 30⬍x⬍40. The mean electrostatic force
properly. on the ion was monitored over 105 time steps as the ion was
moved randomly through the system, and is shown in Fig. 7.
D. Mirror image effects The natural system to compare these simulations with, is that
of a point ion immersed in a box of size L⫻R⫻R, interact-
In general, the electrostatic interaction can be treated in
two ways. Either all charges are coupled to a local field that ing with a charged wall at x⫽0, in the presence of dielectric
is the solution of the Poisson equation, or an explicit sum- medium that has infinitely sharp jumps in the permittivity at
mation over all other charges is done using the appropriate x⫽L/4 and at x⫽3L/4.
Green’s function. This Green’s function is the solution of the To solve the Poisson equation for this system, the field ␺
Poisson equation for a point charge. When the Green’s func- can be expanded in Fourier modes parallel to the yz plane,
tion method is used, mirror image charges need to be con- hence ␺ (x,y,z)⫽⌺ k ␺ k (x)exp(i2␲k"r/R), where R⫽10 is
sidered explicitly, because in the presence of inhomogeneous the width of the simulation box, k is a 2D integer vector, and
dielectric media, the solution of the Poisson equation takes r is the (y,z) coordinate relative to the particle. The Poisson
the form of a sum over image charges. To check if, indeed, equation 关Eq. 共15兲兴 has two cases, one for k⫽0, and one for
the effect of mirror image charges in the presence of inho- k⫽0:
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11273

FIG. 6. The electrostatic field in a periodic 80⫻10⫻10 box with charged FIG. 7. Force on an ion in a system with a dielectric discontinuity at x
plates at x⫽0 and x⫽40, in the presence of a dielectric discontinuity at x ⫽10, and a charged wall at x⫽0. The dashed curve is the exact k⫽0
⫽20 and at x⫽60. The charged plane at x⫽40 is indicated by the dashed contribution to the force, the full curve is a one parameter fit. Deviations
line. near x⫽10 result from the smearing of the ion charge.

共 1⫺p 兲关 ␦ 共 x⫺ 41 L 兲 ⫺ ␦ 共 x⫺ 43 L 兲兴 ␺ ⬘0 ⫹p 共 x 兲 ␺ ⬙0 exp共 ⫺2 ␲ 兩 x⫺L/4兩 /R 兲


f el共 x 兲 ⫽ f el
0 共 x 兲 ⫹A , 共39兲
p 共 x 兲共 x⫺L/4兲 2
⫽⫺⌫ ␦ 共 x 兲 /R ⫹⌫ ␦ 共 x⫺x c 兲 /R ,
2 2

共37兲 where the numerical value of A is the same on both sides of


共 1⫺p 兲关共 ␦ 共 x⫺ 41 L 兲 ⫺ ␦ 共 x⫺ 43 L 兲兴 ␺ k⬘ ⫹p 共 x 兲 ␺ ⬙k x⫽L/4. The full curves in Fig. 7 are one-parameter fits to the
function in Eq. 共39兲. On the left- and right-hand sides of the
⫹p 共 x 兲 ␺ k 4 ␲ 2 k 2 /R 2 ⫽⌫ ␦ 共 x⫺x c 兲 /R 2 . discontinuity, the fit values A⫽0.24⫾0.01 and A⫽0.245
Here, p is the ratio of the dielectric permittivity in the oil ⫾0.002 are obtained. Deviations from this curve occur
phase relative to that of water, and x c is the x coordinate of within 2R c from the discontinuity and for x⬍1.6R c , which
the particle. The k⫽0 case is equal to the case of two can be attributed to the smearing of the charge over a dis-
charged planes as just described, but now the planes are situ- tance of R e ⫽1.6R c , and to the width of the interface.
ated at x⫽0 and at x⫽x c . This equation can be solved
straightforwardly, giving the following expression for the V. APPLICATION: POLYELECTROLYTE–SURFACTANT


force on the particle: INTERACTIONS
p 共 1⫺2x c /L 兲 This simulation method described has all capabilities of
⫺ •⌫/R 2 for L/4⬍x c ⬍3L/4
共 1⫹p 兲 ordinary DPD, but includes applications where electrostatics
0⫽
f el is essential and that were previously inaccessible. One key
共 1⫹p⫺4x c /L 兲
⫺ •⌫/R 2 for 0⬍x c ⬍L/4. example is the interaction between surfactants and polymers.
2p 共 1⫹ p 兲 Often polymer–surfactant interactions occur via electrostatic
共38兲 interactions,2 and to predict the phase behavior of mixtures
This k⫽0 force is shown in Fig. 7 by the dashed curve, and of cationic, anionic, and nonionic surfactants, and their inter-
gives an excellent match with the simulated force far away action with polyelectrolytes, electrostatic interaction cannot
from x⫽10, where the discontinuity in the permittivity is be neglected.40 To illustrate the potential of this method,
located. Closer to this point, mirror image charge effects oc- some qualitative aspects are discussed here.
cur, which arise from the k⫽0 modes. For a typical anionic surfactant, like sodium dodecylsul-
For an infinitely large system (R,L→⬁), ions to the phonate, the surfactant is modeled by a dimer, and the inter-
right-hand side of L/4 interact with a mirror image charge of action parameter between hydrocarbon tails 共c兲 and water 共w兲
size (1⫺p)/(1⫹ p)q⫽1/3 q, located at the opposite of the is taken as ␹ cw ⫽ 18, following earlier work.21 This means that
plane of dielectric discontinuity. Charges located to the left- the tail size is roughly N m ⫽9 carbon atoms. Hence, we have
hand side of this plane interact with the opposite charge R c ⫽0.93 nm, ⌫⫽9.615, and ␶⫽1.00 ns. The head groups 共h兲
⫺1/3 q. This leads to the mirror image force ⌫/(48␲ p(x c carry a charge ⫺1, and are parameterized by ␹ ch⫽6 and
⫺L/4) 2 ), where p⫽1 to the right-hand side and p⫽1/2 to ␹ wh⫽0. One polymer of 50 beads and 75 surfactant mol-
the left-hand side of the discontinuity. In a finite system, ecules are simulated in a box of size 15⫻15⫻15. In physical
however, the lowest k⫽0 mode is screened, see Eq. 共37兲. The units, this is a volume of some 2700 nm3. Each polymer bead
solution takes on the form ␺ k (x)⫽A exp(⫺2␲kx/R) 共p兲 carries a charge ⫹1/2, and has a solubility parameter
⫹B exp(2␲kx/R), with discontinuities in the first derivative ␹ pw⫽0.65, so that it would be in a ␪-solution if it had no
of ␺ k at x⫽L/4, at x⫽3L/4 and at x⫽x c . Since the lowest charge. This corresponds to a polymer of molecular weight21
lateral mode is screened on a distance of R/2␲ , the total M w ⬃8000. The other parameters for the polymer are: ␹ cp
electrostatic force near the dielectric discontinuity takes on ⫽ ␹ ch⫽6 and ␹ ph⫽0. Counterions of a charge of ⫺1 are
the form added to preserve charge neutrality, and have the
11274 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

␹-parameters of water. The polarizabilities are chosen as


p w ⫽p h ⫽1, p p ⫽0.35, and p c ⫽0.025. In this context, the
Flory–Huggins ␹-parameter equals the excess repulsion be-
tween unlike bead types, up to a constant. Following Groot
and Warren,13 the expression for the repulsion parameter,
a i j ⫽25⫹ ␹ i j /0.306, is used.
A typical conformation is shown in Fig. 8. In this con-
formation, the polymer wraps around two discrete micelles.
This is the accepted state of an electrostatically interacting
flexible polymer.41 In other conformations 共not shown兲, the
polymer wraps around a single micelle. Already from this
picture, one can see that a relatively high fraction of surfac-
tant molecules is in solution as a monomer or dimer. When
similar interaction parameters are used without a surfactant,
the monomer concentration is much lower.21 This reflects the
general trend that the critical micelle concentration of an-
ionic surfactants is higher than that of nonionic surfactants.
For the present amount of surfactant, the polymer end-
point distribution is close to a Gaussian, i.e., the distribution
that the polymer would have without charge and without
surfactant. However, if we study the pair-correlation func-
tions, a different picture emerges. In Fig. 9, the radial distri-
bution functions between head groups 共h兲 and polymer beads FIG. 8. 共Color兲 Typical conformation of cationic polyelectrolyte interacting
with anionic surfactant. Green dots are counterions.
共p兲 are shown. The polymer–polymer pair-correlation func-
tion shows a distinct shoulder at a length scale of 3 nm, i.e.,
the size of a micelle, when compared to the pair-correlation
obtained by scaling theory,42,43 which predicts fully extended
function for a noninteracting neutral polymer. Also, the sur-
chains, but experimental results44 indicate a similar fractal
factant head group–head group and the head group–polymer
dimension of d f ⬃1.2. MD simulations at a vanishing salt
correlation function show structures on the length scale of a
concentration45 show a continuous crossover from d f ⫽1.1
micelle.
up to d f ⫽2, depending on polymer concentration. These re-
For comparison, the dots in Fig. 9 give the polymer–
sults are in line with the present simulation results for poly-
polymer radial distribution function for the case where all
electrolytes.
surfactant molecules are replaced by salt ions. This is a poly-
To put this result into perspective a comparison should
electrolyte solution at a salt concentration of 0.045 M. A
be made with uncharged polymers. Polymers in a melt gen-
typical conformation is shown in Fig. 10. A quantitative mea-
erally follow Gaussian statistics, r⬀M ␯ ⫽M 1/2 or d f ⫽2.
sure of the structure is found by analyzing the pair-
Polymers in a bad solvent coil up to form a globule, hence,
correlation functions that are shown in Fig. 9. For the poly-
␯⫽1/3 or d⫽3, and polymers in a good solvent swell to ␯
mer in the presence of anionic surfactant, we find an initial
⫽0.588 (d f ⫽1.7) according to the renormalization theory,46
zone r⬍1 nm where the correlation is dominated by bead–
bead repulsion. For 1⬍r⬍3 nm, we have a small scaling
regime, and for r⬎3 we see a terminal correlation range. In
contrast, the polyelectrolyte without a surfactant shows scal-
ing behavior over the entire range from r⬎1 nm. The mea-
surement of the correlation function was truncated at r
⫽7.5, half of the box size.
In general, when scaling applies, the mass M within a
sphere of radius r is M ⬀r d f , where d f is the fractal dimen-
sion, which is related to the swelling exponent by d f ⫽1/␯ .
The correlation function scales with the local density, hence,
g(r)⬀r d f ⫺3 . By fitting the correlation function to a power
law, for the polyelectrolyte in salt solution a fractal dimen-
sion d f ⫽1.10 was found, which holds over about a decade in
length scale. On the other hand, the same polyelectrolyte in
anionic surfactant solution gives a slope corresponding to a
fractal dimension d f ⫽1.82 over a limited range but, in real- FIG. 9. Intramolecular polymer–polymer, polymer–head group, and head
ity, one should say that scaling invariance is broken. These group–head group pair-correlation functions 共full curves兲 for a polyelectro-
lyte interacting with surfactant micelles, see Fig. 8. The dots give the in-
differences are very significant. Whereas in surfactant solu- tramolecular polyelectrolyte pair-correlation function without surfactant
tion the polymer is folded around the surfactant micelle, it is present, see Fig. 10. The drawn curves are lines of slope ⫺1.18 and ⫺1.90,
strongly stretched in salt solution. The stretching is less than corresponding to fractal dimensions d f ⫽1.82 and d f ⫽1.10, respectively.
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11275

tant is, therefore, intricately linked to the presence of a long-


range electrostatic interaction.

VI. CONCLUSIONS

Electrostatic interactions can be incorporated in mesos-


copic DPD simulation. The grid method used here proved to
work efficiently if the grid size is taken equal to the particle
size. Within this formalism, local inhomogeneities in the
electrostatic permittivity can be treated without any problem.
The rather large grid size used here causes some slight arti-
facts in the ion–ion force at short distances. However, the
short-range DPD repulsion dominates the pair-correlation at
short length scales, therefore, this effect is not very impor-
tant. Moreover, it can be compensated by a short-range pair
interaction if required. When the grid size is reduced, the
accuracy of the electrostatic interaction at short distance
would be improved, but this will be achieved at the cost of
increased CPU time. Key issues like the screening of the
potential near a charged surface and the Stillinger–Lovett
moment conditions are satisfied. Since the electrostatic field
FIG. 10. 共Color兲 Typical conformation of a cationic polyelectrolyte in salt is solved explicitly, effects caused by dielectric inhomogene-
solution of 0.045 M. ities, like image charge effects, are included in the method.
Also, the Maxwell force on dielectric media can be readily
incorporated. This implies that the method captures the es-
or to ␯ ⫽3/5 (d f ⫽5/3) according to the Flory theory. The sential features of electrostatic interaction.
transition from globule to swollen chain has been observed For the direct simulation of mixed surfactants near oil–
in DPD simulation,22 which shows that indeed ␯ increases water interfaces, or for the simulation of Coulombic
continuously from ␯⫽1/3 to ␯⫽3/5 as the quality of the sol- polymer–surfactant interactions, this method has all the ad-
vent increases. From a simple argument, it can be seen that a vantages of DPD over full atomistic MD. DPD has proved to
short-range repulsive interaction can never produce a swell- be faster than MD by many orders of magnitude,18 depend-
ing exponent larger than ␯⫽3/5. The free energy of a chain ing on the precise scaling factor chosen for the simulation.
of M segments in mean field is given by F⬃R 2 /M This brings phenomena of microseconds in reach of simula-
⫹bM 2 /R 3 , where R is the polymer endpoint separation and tions that still contain a fairly accurate representation of the
b is the second virial coefficient. Minimizing F with respect structure of the molecules.
to R gives R⬃b 1/5M 3/5. So, irrespective of how large the As an example of the method, a cationic polyelectrolyte
second virial coefficient is, in this simple mean-field analy- solution was studied. The polyelectrolyte shows a fractal di-
sis, the swelling exponent is always ␯⫽3/5 as long as b is mensionality that is in line with theoretical and experimental
positive. values cited in literature, it behaves as a fairly stiff rod of
Finding a swelling exponent ␯⫽0.9 as in these simula- fractal dimension d f ⫽1.1. When salt is replaced by an an-
tions is, therefore, very special. It effectively means that the ionic surfactant, the polymer wraps around one or more dis-
chains are stiff, with a persistence length of at least ten crete surfactant micelles, in line with current understanding
beads, as the scaling is observed over a decade in length of these systems. In short, the present simulation method,
scale. To generate such polymer conformations without long- which includes long-range electrostatic interactions, pro-
range interactions would require a strong bond–angle inter- duces realistic results for polymer–surfactant systems, with-
action, i.e., a three-body interaction. Such a three-body inter- out the need of introducing effective short-range potentials to
action produces stiff polymers or wormlike chains. Over force the simulation to a desired result. This makes the
distances smaller than the persistence length, the correlation method a powerful, predictive simulation tool, which can be
function of these polymers shows an apparent swelling ex- applied to polyelectrolytes, surfactant systems, biological
ponent close to ␯⫽1, but beyond a distance of a persistence membranes, and interactions between these.
length, the polymers behave as self-avoiding walk again, APPENDIX A: THE CHARGE–CHARGE INTERACTION
␯⫽3/5. Such a bond–angle interaction was not included
here, however, and it would exclude a realistic interaction of For charge clouds centralized around the points 0 and r,
the polymer with surfactant micelles. Stiff polymers with a the potential is given by

冕冕
persistence length as seen in the present simulations cannot ⌫
bend around the micelle for realistic interactions between the U共 r 兲⫽ f 共r1兲 f 共 r⫺r 2 兲 d 3 r 1 d 3 r 2 , 共A1兲
4 ␲ 共 r 1 ⫺r 2 兲
polymer and surfactant, as this is excluded by the stiffness of
the polymer. The transition from ␯⫽0.9 without a surfactant, when we take R c as unit of length (R e ⫽R phys e /R c ).
to a breaking of scale invariance in the presence of a surfac- The double convolution is easiest evaluated by first consid-
11276 J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 R. D. Groot

ering the Fourier transform, which is given by U(k)


⫽⌫ f (k) 2 /k 2 . Substituting the Fourier transform of f (r)
⫽3(1⫺r/R e )/ ␲ R 3e the interaction is obtained as

U 共 r 兲 /⌫⫽
1
2␲ r 2 冕0

k sin共 kr 兲 f 共 k 兲 2 /k 2 dk


52 4 r

35 5 R e 冉 冊 冉 冊 2

2 r
5 Re
4
⫺..
, 共A2兲
4␲Re
where the series was found by expanding the sine in a Taylor
series, and evaluating the resulting moment integrals. The
numerically calculated back transform was fitted to the
asymptotic expansion plus a power law, which gives a very
good description at r⬍R e :

4 ␲ R e U 共 r 兲 /⌫⫽
52 4 r

35 5 R e 冉 冊 冉 冊 2

2 r
5 Re
4
FIG. 11. Grid points within neighbor distance R n are checked for their
distance from the particle, those within a distance R e are actually used for

冉 冊 5.145 smearing out the ion charge. The arrows point at the lattice sites over which
r the charge of the ion is spread.
⫺0.135 87 . 共A3兲
Re
In the range R e ⬍r⬍2R e , the potential is empirically very
well described by cides with the particle center. Toward this end, a dipole and a

冉 冊 6
quadrupole moment are added such that the center of lattice
Re r charges coincides with the particle position. The correction
4 ␲ R e U 共 r 兲 /⌫⫽ ⫺3.2100 1⫺ , 共A4兲
r 2R e to the charge distribution has the form
and finally for r⬎2R e , the potential is exactly given by ␦ f i ⫽ f i • 关 1⫹␭ x 共 x i ⫺x c 兲 ⫹ ␮ x 共 x i ⫺x c 兲 2 ⫹ 共 x→y 兲
U(r)⫽⌫/4␲ r. Over the entire range 0⬍r⬍2R e , the error
in the fit is of the order 10⫺4 . Assuming power laws as in ⫹ 共 x→z 兲兴 , 共B2兲
Eqs. 共A3兲 and 共A4兲, the coefficients follow analytically from where x i is the x coordinate of a lattice position, x c is the x
the power 5.145, and from the conditions that the potential coordinate of the particle, and the last two terms are analo-
and its first derivative must be continuous in r⫽R e . gous expressions for the y and z direction. The redistributed
charge is calculated from the conditions
APPENDIX B: CHARGE REDISTRIBUTION
AND DIFFERENTIATION ON THE LATTICE 兺i 共 x i ⫺x c 兲共 f i ⫹ ␦ f i 兲 ⫽0 and 兺i ␦ f i ⫽0. 共B3兲
In principle, the lattice grid size can be chosen at any
value, but as we want to have a fast implementation and To solve Eq. 共B3兲, the following moments of the charge dis-
describe local polarizability effects, it is prudent to choose tribution are calculated first:
the lattice cells equal to the cells that are used for the linked
cell administration. The charges are smeared out over all S xn ⫽ 兺i 共 x i ⫺x c 兲 n ␦ f i , 共B4兲
lattice nodes that are closer by than a distance R e . Therefore,
at forehand, a neighbor list is made of all nodes within a for n⫽1, 2, and 3, and analogous expressions for the y and z
distance R n of the central cell 共see Fig. 11兲. Then during the directions are determined. If the charge redistribution in the x
simulation, all nodes within the outer circle are checked for direction would not influence the moments in the other di-
their distance from the ion. The nodes that are found within rection, ␭ x and ␮ x appearing in Eq. 共B2兲 could directly be
a distance R e from the ion are used for smearing. solved as
The actual smearing is done in two steps. First, a charge ␭ x ⫽⫺S x1 S x2 / 关共 S x2 兲 2 ⫺S x1 S x3 兲 ]
proportional to f (r)⫽1⫺r/R e is assigned to each of the
nodes i, normalized such that the sum of all these charges and 共B5兲
equals the charge of the ion, hence,
␮ x ⫽ 共 S x1 兲 2 / 关共 S x2 兲 2 ⫺S x1 S x3 兲 ].
1⫺ 兩 ri ⫺rc 兩 /R e
f i 共 rc 兲 ⫽ , 共B1兲 However, since the moments in the y direction also change
兺 i ⬘ 1⫺ 兩 ri ⬘ ⫺rc 兩 /R e
when the distribution in the x direction is balanced, Eqs.
where ri is the position of node i, rc is the position of the ion, 共B2兲, 共B4兲, and 共B5兲 have to be iterated three to four times to
and the sum over i ⬘ runs over all nodes within a distance R e satisfy Eq. 共B3兲 up to six decimal places.
from rc . To enable the simulation of a surface charge of a wall at
In the second step of the smearing operation, the distri- x⫽0, the grid points at which the field and charge density are
bution is balanced such that its center of mass exactly coin- defined, are taken at integer multiples of the cell size used for
J. Chem. Phys., Vol. 118, No. 24, 22 June 2003 Electrostatics in dissipative particle dynamics 11277

4
H. A. Stone, Annu. Rev. Fluid Mech. 26, 65 共1994兲.
5
D. Toomre, P. Keller, J. White, J. C. Olivo, and K. Simons, J. Cell. Sci.
112, 21 共1999兲.
6
L. R. Forrest and M. S. P. Sansom, Curr. Opin. Struct. Biol. 10, 174
共2000兲.
7
A. H. Clark and S. B. Rossmurphy, Adv. Polym. Sci. 83, 57 共1987兲.
8
D. P. Tieleman, S. J. Marrink, and H. J. C. Berendsen, BBA-Rev.
Biomembr. 1331, 235 共1997兲.
9
H. Heller, M. Schaefer, and K. Schulten, J. Phys. Chem. 97, 8343 共1993兲.
10
R. Lipowsky and S. Grotehans, Europhys. Lett. 23, 599 共1993兲.
11
E. Lindahl and O. Edholm, Biophys. J. 79, 426 共2000兲.
12
P. J. Hoogerbrugge and J. M. V. A. Koelman, Europhys. Lett. 19, 155
共1992兲.
13
R. D. Groot and P. B. Warren, J. Chem. Phys. 107, 4423 共1997兲.
FIG. 12. Grid cells used to define the polarizability between points i and j. 14
P. Español, Phys. Rev. E 52, 1734 共1995兲.
15
P. Español and P. Warren, Europhys. Lett. 30, 191 共1995兲.
16
M. P. Allen and D. J. Tildesley, Computer Simulation of Liquids 共Claren-
the linked cell method (⬇R c ). To calculate the local polar- don, Oxford, 1987兲.
izability, we need to find the mean dielectric permeability on 17
W. K. Den Otter and J. H. R. Clarke, J. Mod. Phys. C 11, 1179 共2000兲.
each link between grid points. Therefore, the polarizability is
18
R. D. Groot and K. L. Rabone, Biophys. J. 81, 725 共2001兲.
19
R. D. Groot and T. J. Madden, J. Chem. Phys. 108, 8713 共1998兲.
averaged over all particles in the four cells adjacent to each 20
R. D. Groot, T. J. Madden, and D. J. Tildesley, J. Chem. Phys. 110, 9739
link. In Fig. 12, this is illustrated. For the link between points 共1999兲.
i and j, the polarizability is averaged over the shaded area. If 21
R. D. Groot, Langmuir 16, 7493 共2000兲.
22
we denote the polarizability on the link between ri and Y. Kong, C. W. Manke, W. G. Madden, and A. G. Schlijper, J. Chem.
Phys. 107, 592 共1997兲.
ri ⫹e ␮ by p ␮ (ri ), where e ␮ is a lattice vector in the positive 23
N. A. Spenley, Europhys. Lett. 49, 534 共2000兲.
␮-direction, then the mean dielectric displacement is 24
S. Jury, P. Bladon, M. Cates, S. Krishna, M. Hagen, N. Ruddock, and P. B.
Warren, Phys. Chem. Chem. Phys. 1, 2051 共1999兲.
D ␮ 共 ri ⫹ 12 e ␮ 兲 ⫽⫺p ␮ 共 r i 兲关 ␺ 共 ri ⫹e ␮ 兲 ⫺ ␺ 共 ri 兲兴 /h, 共B6兲 25
A. T. Clark, M. Lal, J. N. Ruddock, and P. B. Warren, Langmuir 16, 6342
where h is the lattice size. Consequently, the divergence of 共2000兲.
26
K. E. Novik and P. V. Coveney, Phys. Rev. E 61, 435 共2000兲.
the field is 27
E. S. Boek, P. V. Coveney, H. N. W. Lekkerkerker, and P. vanderSchoot,

ⵜ•D 共 ri 兲 ⫽ 兺␮ 冋 冉 1 1
冊 冉
D ␮ ri ⫹ e ␮ ⫺D ␮ ri ⫺ e ␮
2 2 冊册 冒 h,
28
Phys. Rev. E 55, 3124 共1997兲.
S. Yamamoto, Y. Maruyama, and S. Hyodo, J. Chem. Phys. 116, 5842
共2002兲.
共B7兲
29
D. Reith, H. Meyer, and F. Muller-Plathe, Macromolecules 34, 2335
共2001兲.
which can be worked out by substitution of Eq. 共B6兲. 30
W. Tschop, K. Kremer, J. Batoulis, T. Burger, and O. Hahn, Acta Polym.
The same smearing distribution that was used to solve 49, 61 共1998兲.
the electrostatic field, is consistently used to evaluate the
31
A. P. Lyubartsev and A. Laaksonen, Phys. Rev. E 52, 3730 共1995兲.
32
D. Henderson, L. Blum, and J. L. Lebowitz, J. Electroanal. Chem. 102,
force on an ion, see Eq. 共26兲. For this purpose, the field 315 共1979兲.
gradient at the lattice positions is determined as 33
J. V. L. Beckers, C. P. Lowe, and S. W. de Leeuw, Mol. Simul. 20, 369
共1998兲.
ⵜ␮ ␺ 共 ri 兲 ⫽ 关 ␺ 共 ri ⫹e ␮ 兲 ⫺ ␺ 共 ri ⫺e ␮ 兲兴 /2h, 共B8兲 34
J. W. Eastwood, R. W. Hockney, and D. Lawrence, Comput. Phys. Com-
mun. 19, 215 共1980兲.
and the electrostatic force on ion c is calculated from 35
M. Deserno and C. Holm, J. Chem. Phys. 109, 7678 共1998兲.
36
C. Sagui and T. Darden, J. Chem. Phys. 114, 6578 共2001兲.
F el共 rc 兲 ⫽⫺q c 兺i f i 共 rc 兲 ⵜ ␺ 共 ri 兲 . 共B9兲 37
38
A. C. Maggs and V. Rossetto, Phys. Rev. Lett. 88, 196402 共2002兲.
F. H. Stillinger and R. Lovett, J. Chem. Phys. 94, 7353 共1991兲.
39
J. P. Hansen and I. R. McDonald, Theory of Simple Liquids 共Academic,
Again the sum over i is limited to lattice nodes within the London, 1986兲.
distance R e from rc , see Eq. 共B1兲, and f i (rc ) is the distribu- 40
T. Wallin and P. Linse, Langmuir 14, 2940 共1998兲.
tion obtained from Eqs. 共B1兲–共B5兲.
41
T. Wallin and P. Linse, Langmuir 12, 305 共1996兲.
42
P. G. de Gennes, P. Pincus, R. M. Valesco, and F. Brochard, J. Phys.
共France兲 37, 1461 共1976兲.
1
P. G. de Gennes, Macromolecules 14, 1637 共1981兲. 43
T. Odijk, Macromolecules 12, 688 共1979兲.
2 44
P. Hansson and B. Lindman, Curr. Opin. Colloid Interface Sci. 1, 604 S. Sasaki, H. Ojima, K. Yataki, and H. Maeda, J. Chem. Phys. 102, 9694
共1996兲. 共1995兲.
3
F. S. Bates and G. H. Fredrickson, Annu. Rev. Phys. Chem. 41, 525 45
M. J. Stevens and K. Kremer, J. Chem. Phys. 103, 1669 共1995兲.
共1990兲. 46
J. C. le Guillou and J. Zinn-Justin, Phys. Rev. Lett. 39, 95 共1977兲.

You might also like