Pressure coupling / barostats

Victor Rühle
February 19, 2008

Abstract
This is a brief overview in terms of the journal club about pressure
coupling in molecular dynamics (MD). It is far from being complete, also
some formulas might be erroneous. For more details have a look the
references.

1 Pressure
In some cases, it is important to simulate a system at constant pressure. As
it was already a common practice to do that in Monte Carlo simulations, also
algorithms for MD were developed. The first thing to start of is an expression
for the pressure of a simulated system.
The pressure tensor P can be calculated using Clausius virial theorem as
2
P= (Ekin − Ξ) (1)
V
with the box volume V , the kinetic energy Ek in and the inner virial tensor
1X
Ξ=− rij · Fij . (2)
2 i<j

See e.g. Ref [1]. The isotropic pressure is calculated from the trace of the
pressure tensor

P = Tr (P) /3 . (3)

Details of the derivation are described in the appendix.

2 Isotropic coupling
Correcting the pressure in a simulation can be achieved through a change in
the inner virial Ξ by scaling the inter particle distances. This is the common
procedure in all barostats described below.

1
2.1 Berendsen barostat
In the Berendsen method[2], the system is weakly coupled to an external bath
using the principle of least local perturbation. Similar to the temperature cou-
pling, an extra term is added to the equations of motion that effects a pressure
change
µ ¶
dp p0 − p
= (4)
dt bath τp

where τp is the time constant for the coupling. A simple proportional coordinate
scaling, concomitant with volume scaling, minimizes local disturbances. An
extra term is added to the equations of motion:

ẋ = v + αx , (5)

while the volume changes accoringly:

V̇ = 3αV . (6)

The pressure change is related to the isothermal compressibility β

dP 1 dV 3α
=− =− . (7)
dt βV dt β
With Eq. 4 α evaluates to

β (p0 − p)
α=− . (8)
3τp

Thus the modified equation of motion is

β (p0 − p)
ẋ = v − x (9)
3τp

and represents a proportional scaling of coordinates.

The compressibility, that may not be accurately known, occurs in the ex-
pression for the scaling, but from Eq. 9 one can easily see, that only that ratio
β/τp enters in the equations of motion.

2.2 Andersen barostat

The Andersen method[3] was developed to adjust the pressure in a simulation of
interacting particles. In the following description, only systems of pairwise inter-
acting particles are treated. The method was later first extended to anisotropic
coupling by Parrinello et al[4] and later also to molecular systems by Nosé et
al[5].
Andersen proposed to replace the coordinates ri by scaled coordinates ρi
defined as

ρi = ri /V 1/3 (10)

2
Consider the new Lagrangian, in which a new variable Q appears:
³ ´ 1 XN XN ³ ´ 1
L ρN , ρ̇N , Q, Q̇ = Q2/3 mi ρ̇i 2 − U Q1/3 ρij + M Q̇2 − p0 Q .
2 i=1 i<j=1
2
(11)
If we interpret Q as the volume V , the first two terms on the right are just the
Lagrangian of the unscaled system. The third term is a kinetic energy for the
the motion of Q, and the fourth represent a potential energy associated with Q.
Here p0 and M are constants. A physical interpretation of the additional terms
would be: Assume the system is simulated in a container and can be compressed
by a piston. Thus, Q, whose value is the volume V , is the coordinate of the
piston. p0 V is the potential derived from an external pressure p0 acting on the
pisten and M is the mass of the piston.
In the original paper by Andersen, now follows some math, where the Hamil-
tonian is derived, out of that the equations of motion for the scaled system and
finally mapping back to the real variables. Here I skip all the details and just
give the equations of motion
dri pi 1 d ln V
= + ri , (12)
dt mi 3 dt
dpi X 1 d ln V
=− r̂ij U 0 (rij ) − pi , (13)
dt µ 3 dt ¶
M d2 V 2 X p2i 1X 0
= p0 + − r ij U (r ij ) /V . (14)
dt2 3 2mi 3
Add note from andersen paper on how to choose M!!

3 Anisotropic coupling
3.1 Anisotropic Berendsen pressure coupling
There is also an anisotropic version of the Berendsen barostat, but this will not
be treated in this summary.

3.2 Parrinello-Rahman barostat

When simulating crystal structures, it is not sufficient only to scale the volume.
Parinello and Rahman extended the method proposed by Andersen the let the
simulation box also change it’s shape.
Let’s start with some notation: The cell can have an arbitrary shape, it’s
volume completely described by three vectors a , , b , c. The vectors can have
different lengths and arbitrary mutual orientations. An alternative description
is obtained by arranging the vectors as {a, b, c} to form a 3 × 3 matrix h whose
columns are the latter vectors. The volume is given by
V = det h = a · (b × c) . (15)
The position ri of a particle can be written in terms of h and a column
vector si , with components ξi , ηi andζi as
ri = hsi = ξi a + ηi b + ζi c (16)

3
with 0 ≤ ξi , ηi , ζi ≤ 1. The square of the distance between particle i and j is
given by
2
rij = sTij Gsij (17)

where the metric tensor G is

G = hT h . (18)

Using the latter notation, the Lagrangian can be written as

1X XX 1 ³ ´
L= mi ṡTi Gṡi − U (rij ) + M Tr ḣT ḣ − pV (19)
2 2
Deriving the equations of motion is similar to the isotropic case from Andersen.

4 Appendix
A Pressure and the Clausius virial theorem
Starting from the equations of motion for an N-particle system

mi r̈i = Fi j = 1, 2..N (20)

with particle coordinates ri , mi and force Fi acting on particle i. Multiplication

with ri and using the relation
d
rr̈ = (rṙ) − ṙ2 , (21)
dt
leads to
d
(mi ri ṙi ) − mi ṙi = Fi ri (22)
dt
Averaging over all particles and time, the first term due to the derivation in
time becomes zero. The second term is the total energy
X
mi ri 2 = 2Ekin . (23)

P
N
The last term Fi ri is called the virial. Finally, Eq. 22 leads to
i=1
X
−2Ekin = Fi ri . (24)

Now we have a more detailed look at the virial term. We assume that the
particles are trapped in a box with volume V and excert a pressure p onto the
walls. In other terms, that means, that a wall element excerts a force p df onto
nearby particles and is directed inside the box. This part of the virial we call
Wa . The part due to particle-particle we call ”inner virial” denoted as Ξ.
In Fig. 1, the force p df acting on particles arising from a wall element df
is depicted. The acts on all nearby particles an is directed into the box. Since
df is small, the positions ri have approximately the same value r which denotes

4
Figure 1: A wall element df excerts a force p df on the nearby particles. r
denotes the vector from the origin to the wall element.

the vector from the origin to the wall element df . The part of df to the outer
virial can be written as
N
X XN
ri Fp,i = r Fp,i = −rp df . (25)
i=1 i=1

The outer virial us calculated by integration over the whole surface

Z Z
Wa = −p ∂V r df = −p divrdV = −3pV . (26)
V

Doing the splitting into outer and inner virial in Eq. 24 and solving for the
pressure leads to
2
p= (Ξ − Ekin ) (27)
3V

References
[1] Gromacs manual
[2] H.J.C. Berendsen et al., J. Chem. Phys. 81, 3684 (1984)
[3] H.C. Andersen, J. Chem. Phys. 72, 2384 (1980)
[4] M. Parinello and A. Rahman, J. Appl. Phys 52, 7182 (1981)
[5] S. Nosé and M.L. Klein, Molecular Physics 50, 1055 (1983)
[6] R. Becker, Theorie der Wärme, Berlin, Springer (1985)