This article was downloaded by: [Duke University Libraries]

On: 03 July 2012, At: 02:56

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Molecular Physics: An International

Journal at the Interface Between
Chemistry and Physics
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tmph20

A molecular dynamics method

for simulations in the canonical
ensemble
a b
Shūichi Nosé
a
Division of Chemistry, National Research Council Canada,
Ottawa, Canada
b
Department of Physics, Faculty of Science and Technology,
Keio University, 3-14-1 Hiyoshi, Kohoku-ku, Yokohama, 223,
Japan

Version of record first published: 22 Aug 2006

To cite this article: Shūichi Nosé (1984): A molecular dynamics method for simulations in the
canonical ensemble, Molecular Physics: An International Journal at the Interface Between
Chemistry and Physics, 52:2, 255-268

To link to this article: http://dx.doi.org/10.1080/00268978400101201

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-

conditions

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any
representation that the contents will be complete or accurate or up to date. The
accuracy of any instructions, formulae, and drug doses should be independently
verified with primary sources. The publisher shall not be liable for any loss, actions,
claims, proceedings, demand, or costs or damages whatsoever or howsoever
caused arising directly or indirectly in connection with or arising out of the use of
this material.
 MOLECULAR PHYSICS,1984, VOL. 52, No. 2, 255-268

A molecular dynamics method for simulations in the

canonical e n s e m b l e t
by S H U I C H I N O S I ~
Division of Chemistry, National Research Council Canada,
Ottawa, Ontario, Canada K I A 0R6

(Received 3 October 1983 ; accepted 28 November 1983)

A molecular dynamics simulation method which can generate configura-

tions belonging to the canonical (T, V, N) ensemble or the constant
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

temperature constant pressure (T, P, N) ensemble, is proposed. The

physical system of interest consists of N particles (f degrees of freedom), to
which an external, macroscopic variable and its conjugate momentum are
added. This device allows the total energy of the physical system to
fluctuate. The equilibrium distribution of the energy coincides with the
canonical distribution both in momentum and in coordinate space. The
method is tested for an atomic fluid (At) and works well.

1. INTRODUCTION
Th e molecular dynamics ( M D ) method has become an important technique
for the study of fluids and solids. In the standard MD method, the newtonian
equations of motion of the particles in a fixed MD cell of volume V are solved
numerically. T h e total energy E is conserved, and thus the ensemble generated
by the simulation is the microcanonical or (E, V, N) ensemble.
With the MD method, not only the static quantities but also the dynamic
quantities can be obtained. This is one advantage over the Monte Carlo (MC)
method. However, a disadvantage of the M D method is that the conditions of
the simulations are not the same as those normally encountered in experiments
(constant temperature, constant pressure or (T, P, N) conditions).
In this regard, Andersen's introduction of the constant pressure M D method
represented a significant breakthrough [1]. In his modified MD method, the
volume becomes a variable and is allowed to fluctuate. T h e average volume is
determined by the balance between the internal pressure and the externally set
pressure P~x. T h e enthalpy of the system is approximately conserved, so this
method generates the constant enthalpy, constant pressure (H, P, N ) ensemble.
Parrinello and Rahman subsequently extended the method to allow for changes
of the M D cell shape [2, 3 ]. T he usefulness of this latter method has been demon-
strated by numerous applications to structural changes in the solid state [2-7].
To perform M D simulations at constant temperature, one can simply keep
the kinetic energy constant by scaling the velocities at each time step and this
approach is now widely employed [8, 9]. However, there seems to be no
rigorous proof that the latter approach produces configurations belonging to the
canonical ensemble.
t Issued as N.R.C.C. No. 23045.
$ Present address : Department of Physics, Faculty of Science and Technology, Keio
University, 3-14-1 Hiyoshi, Kohoku-ku, Yokohama 223, Japan.
M.P. 1
 256 Shfiichi Nos~

Hoover et al. proposed a constraint method in which an additional velocity

dependent term is added to the forces to keep the total kinetic energy constant
[10]. Their method produces the canonical distribution for the potential energy
term. The fluctuations of the kinetic energy are suppressed.
A method for constant temperature MD simulations was also proposed by
Andersen [1]. This is effectively a hybrid of MD and MC methods since the
velocities of the particles are changed stochastically to produce the Boltzmann
distribution. It also lacks a well defined conserved quantity. Tanaka et al.
applied Andersen's method to a Lennard-Jones system and also to a water
system [11]. They found that if the probability of the stochastic Collision
exceeds a certain value, the diffusion coefficient decreases appreciably. To
prevent the decrease of the diffusion coefficient and at the same time to maintain
the temperature constant, it seems to be necessary to select the collision
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

probability in a certain range.

In the present article, a new molecular dynamics method at constant tempera-
ture is proposed. By introduction of an additional degree of freedom s, the total
energy of the physical system is allowed to fluctuate. A special choice of the
potential for the variable s guarantees that the averages of static quantities in this
method are equal to those in the canonical ensemble. This method is purely
dynamical. In the extended system of the particles and the coordinate s, the
total hamiltonian is conserved and all the equations of motion are solved without
introducing any stochastic process.
The formulation of the method is presented in w2. As an example, an
application to a system of argon atoms is given in w3. If the present method is
combined with the constant pressure MD method, then simulations at conditions
of a constant temperature and constant pressure are also possible. The formu-
lation for the (T, P, N) ensemble is given in the Appendix.

2. A CANONICAL ENSEMBLE MOLECULAR DYNAMICS METHOD

2.1. Equations of motion
We formulate the system which produces configurations following the
canonical distribution. We limit the discussion to the case of atoms, but the
extension to molecular systems is straightforward. First, consider a physical
system ; N particles with coordinates rl, r 2. . . . , r v in a fixed volume V, and
potential energy ~(r). An additional degree of freedom s is introduced, which
acts as an external system. The interaction between the physical system and s
is expressed via the scaling of the velocities of the particles,

v~:= si"i , (2.1)

and v~ is considered as the real velocity of particle i. We can interpret this as an

exchange of heat between the Fhysical system and the external system (heat
reservoir).
We associate a potential energy (f+ 1)kTeq i n s with the variable s, where [
is the number of degrees of freedom in the physical system, k Boltzmann's
constant, and Toq the externally set temperature value. As we shall see, this
@oice of potential energy ensures that canonical ensemble averages are recovered.
 Canonical ensembleMD method 257

The lagrangian of the extended system of particles and s is thus postulated to be

s ~i mi
-~- s 3"r~2 - r + Q i~-(I+l)kTo,~lns. (2.2)

The kinetic energy term, 89 2 is introduced in order to be able to construct a

dynamic equation for s. The parameter Q has the dimensions of energy 9 (time) 2
and determines the time scale of the temperature fluctuation. The equations of
motion for r and s are derived from the lagrangian equation

= ~A'
where A stands for one of the variables mentioned above.
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

T h e equations for the particles are

d ~r
d--t (m~s~ ~'~)= ~ri' (2.4)

or

1 ~r 2i
i~ = f'i. (2.5)
m i s2 ~r i $
The equation for s is
Qk= ~ misi.i2- ([+ 1)kTeq (2.6)
i $
If we denote the average in the extended system by ( . . . ) , the relation

( ~i9 mis2
s
f'i~) = ( f + l ) h T e q I! 5 (2.7)

is obtained from (2.6) because the time average of a time derivative (e.g. Q+')
vanishes. This suggests that the average of the kinetic energy coincides with the
externally set temperature Teq.
The momenta are given by

p i = ,,.
(7 r i
=m is i"i (2.8)

and

p,=--~- =Qi. (2.9)

The conserved quantities in this extended system are the hamiltonian

Pfl +r 1)kTeq In s, (2.10)

the total m o m e n t u m
P = ~i p i = ~i (m,d2i'), (2.11)

and the total angular m o m e n t u m

M = ~., r i x pi = ~. r, • (m,s2 f'i). (2.12)

12
 258 Shfiichi Nos~

2.2. Proo[ of the equivalence of the present method and the canonical
ensemble
Now we prove that the equations of motion derived from (2.2) produce
configurations in the canonical ensemble at temperature Teq. Our extended
system produces a microcanonical ensemble of (f + 1) degree of freedom. The
partition function of this ensemble is defined by

Z -_ ~
1l dpsI dsl dpl dr ~ ( ~2--~+r
P~ .Ps ~ . ) . (2.13)
Here, ~(x) denotes the Dirac ~ function, and the shortened forms d p =
dpl dp2 . . . dpN , dr = dr I dr 2 . . . dr N are used.
T h e m o m e n t u m Pi is transformed as
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

PJ=p'
$
~, (2.14)

and the volume element becomes d p = s ! dp' ([ is the number of degrees of

freedom in the physical system). There is no upper limit in momentum space,
so we can change the order of integration of dp' and ds ;

Z = -~.v. l dps l dp' I dr l ds st 3 ~" /.mi

P2--~i~+ ~( r ) + 2---Q+ (l + l )k T eq ln s - E

Using the equivalence relation for ~ function ~(g(s))= ~(s-so)/g'(s), where

so is the zero of g(s)= 0, and the shortened form
oUf(p', r) = 2 P t2~/2mi + r
i
we get
1
Z= i. ldP Idp'Idrlds(t+l)kTe
,,+1 ( [(ozztO(p,r)+(ps2/2Q)_E)l~
,-exp (}--~i )-~eq .]1

1 1
-q+l)kre~NtldPsldp'Idrexp
[(- p2)/]
a~(p',r)+~QQ-E kr~ .
The integration with respect to p~ can be carried out immediately, so we get
the final result
1(2 e)1,'
Z=([+I ) \kTe J exp ( E / k T e q ) Z e . (2.15)

Z e is the partition function of the canonical ensemble

Z~ = 1 1 d P ' I dr exp [ - 34~(p', r)/hT~q]. (2.16)

The average of some static quantity which is an arbitrary function of pJs and
r~ in the extended system is exactly the same as that in the canonical ensemble,

<A(-Ps,r)>= <A(p',r))o, (2.17)

where ( . . . ) e means the average in the canonical ensemble.
 Canonical ensemble MD method 259

The above result is derived from the assumption that the only conserved
quantity is the total hamiltonian YF1. In fact, the system has other conserved
quantities: total m o m e n t u m and angular momentum. These deviate by an
order O(1/N) from the canonical ensemble averages. A similar result was
already pointed out by Hoover and Alder for the microcanonical ensemble [12].
The correction for total momentum conservation in this method can be
achieved by using f - 3 in place of f in the definition of both the lagrangian and
the instantaneous temperature.

2.3. Some average quantities in the canonical ensemble MD method

If the instantaneous temperature T is defined as
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

p2 f kT,
~i 2mis
--2 - 2
the average and the fluctuation of T are, respectively

<T> = <T>o= T~q (2.18)

and

< ( I ' - T~q)2>

, = < ( T - Teq)2>c= Teq z 2[ " (2.19)

The formula for the heat capacity cv is derived from the fluctuation of the total
energy
,r~ P i 2
ET= .~, ~ + r ;
i Lmis

<ET2>c--<ET>2 1 [~(kT(,q) e < r 1 6 2

c~ - Nk T~qe k Teq2 ~ q N
fh
-- F
2N NkT~q 2 (2.20)

We can also obtain averages of quantities depending on s,

I @~ I alp' I dr B(p', r) I d s . st+1+"

• [~(P"r)+(P~2/2Q)-E])(f+I)kT~q

• -[#f(p''r)+(p~e/2Q)-(f+l)kTeqE]])

mE f + 1 ~1/2
=exp [if + i)-kTcq] ([ ~- l +m ]

x B(p',.)exp q+1) kTo A/o" (2.21)

 260 Shfiichi Nos6

Some special case of (2.21) are

mE
( s ~ ) = e x p [([+l~_Teq ] \Vi--~m,
] (1+1 )(1+1)1~

x@xp [---
m r ,
(2.22)
1+1" kTeq])c

( P: \ = l P ' : \ = l l \ f
misa/ \mis/ \'/
+ l kreo.
I
This last equation is identical to (2.7).
The fluctuation of s is
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

(exp { - [2/( I + 1)](r

• . (2.23)
(exp { - [1/(l + 1 )](,;b/k Teq)})2
For large f, (2.23) tends to

((s-<s>?\=55
\ <s> ] / 12 k
(2.24)

2.4. An interpretation of the variable s

To obtain the dynamical quantities, the variable s can be interpreted as a
scaling factor for the time step At in the simulations. The real time step At'
is obtained by the relation
At
At'- . (2.25)
$

The length of each time step At' is now unequal.

Equation (2.25) is deduced as follows. In simultaneously transforming
s'=s/a, and t'= t/a by a constant a, the coordinates and their time derivatives
change as
r ' = r , f"=aF, F'=a2F .....

p'=p/a, t'=t/a, (2.26)

s'=s/a, ~'=~, #'=ag, ... ;

the hamiltonian ~ 1 is invariant except for a constant term
dC't= ~ 1 - - (f+ 1)kTeq In a. (2.27)
So only the ratio t/s=t'/s' has any real meaning, and in the case of s ' = 1, the
hamiltonian recovers its normal, unscaled form. The time in this case is
considered to correspond to the real time. The velocity in (2.1) is re-expressed
as

dri dr i
 Canonical ensemble M D method 261

The length of the time step is unequal in the canonical ensemble M D method.
For calculation of time dependent quantities, it is convenient to sample at intervals
that are integer multiples of a unit time step. This can be done very easily by
interpolation. At equilibrium, the fluctuation of s is of order N -112 (see 2.24)),
so in practice, the differences in the lengths of individual time steps can some-
times be ignored. An averaged real time can be obtained by multiplying the
simulation time with (s-l).
The detailed nature of the dynamics depends upon the value of Q chosen.
However, some dynamic quantities (especially, one body quantities) of the system
are, we believe, less sensitive to the value of Q. If this is the case, we can get
information on both static and dynamic quantities.
The frequency of the s oscillation can be estimated from (2.5),
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

Q~= S" pi2 (f+ 1) kTeq

mi$3 $

We assume the system is in equilibrium, and s fluctuates around the averaged

value <s), s = (s) + ~s. Then the above equation can be simplified as

Q(~)=fhT~q ((s)2 ~) 2fhTeq

\ s3 = (s)2 68. (2.28)

This is the equation for a harmonic oscillator, with frequency

[2fhT~q~ lI~.
w = ~,Q - - - ~ j (2.29)

The period of the oscillation is

t0=--=2w
(Q<s)2 1 2 (2.30)
o~ \2fkr~q/
A typical value of Q for a 108 particle system, at Teq = 150 K and for t 0= 1 ps
is 20 (kJ mol-X)(ps) 2.
The transformation (2.26) can be used to change the length of time step.
Without this Control, the real time step in the simulations may become too small
or too large. For a small time step, the calculation is inefficient ; for a large
one, the precision of the calculation cannot be maintained.

3. AN APPLICATION
The canonical ensemble M D method was tested on a system of 108 argon
atoms (mass 39.9 g tool-l), interacting with a Lennard-Jones 12-6 potential
(E = 1.039 kJ mo1-1, a = 3.446 A) which was truncated at 8-5 A. The M D cell
was a cube of edge length 17.5 A (or volume 29.88 cm 3 mo1-1) and as usual, the
periodic boundary condition was adopted. A 5th order predictor-corrector
algorithm was employed for integration of the equations of motion. A time step
At=2.5 x 10 .25 s was used, except for the runs with Q = 100 (kJ mol-1)(ps) 2 for
which At=5.0 x 10 -is s. Each run consisted of the calculation of 2500 time
steps, the first 500 steps being discarded from the averaging.
In all applications, differences in the length of individual time steps were
ignored and the real time was calculated by multiplying the time of simulation
by the factor <s-l>.
 Downloaded by [Duke University Libraries] at 02:56 03 July 2012

262

r
7

9
9

oo

o~
o~

II

~J
~L
7

~J
1
1
1
1
1
1
Shfiichi Nos6

1
1
1
 Canonical ensemble MD method 263

The simulations were carried out by the following scheme. First, starting
from an initial configuration (f.c.c. structure) a standard MD run (No. 1) was
carried out at about 150 K. This was followed by a constant temperature
simulation at Teq=100 K (No. 2). Then, the temperature was reset to
Teq = 150 K (No. 3) and finally the effects of using different Q values 1, 10, and
100 (kJ mol-1)(ps) 2 were compared at 100 K and 150 K (No. 4 9). The results
are listed in the table. The deviationsof the temperature from Teq a r e less than
1 K. The runs with Q = 1 show especially good agreement. The averages of
the potential energy and pressure give almost the same results for different
values. The run No. 9 clearly did not reach equilibrium because the tempera-
ture fluctuation is too large. The sudden increase in the value of ~ at the
beginning of the run gave a huge amount of kinetic energy to the variable s,
which could not relax to equilibrium in the 2500 timesteps.
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

250

Q = l.O ( k J / m o l ) ( p s ) 2
I
200 1. . . . . . . . . . . . .
L
L

t~
lO0 . . . . . . . . . . . . .

50
1250 2500 3750 5000 6250 7500
step

Figure 1. Evolution of the t e m p e r a t u r e . T h e first 1250 steps s h o w n (1250-2500 step)

are carried out with the s t a n d a r d M D m e t h o d . A t step 2500, the simulation is
c h a n g e d to the c o n s t a n t t e m p e r a t u r e m e t h o d with Teq = 100 K. At step 5000, Teq
is c h a n g e d to 150 K.

Figure 1 shows the evolution of the instantaneous temperature through the

first 3 runs. In the constant temperature method, the temperature reaches
Teq quickly and fluctuates around this value. The approximate formula for the
fluctuation of temperature in the microcanonical ensemble [13, 14].

((3T)2)E,V,X= Teq2~--~ 1 - (3.1)

(c,~, heat capacity) shows that the temperature fluctuations in the canonical
ensemble are larger than those in the microcanonical ensemble. We can readily
recognize this in figure 1. The values of AT=(((3T)2)) 1/~, 8-3 K at 100 K
and 12.4 K at 150 K are comparable with those given by (2.19), 7.9 K at 100 K
and 11.8 K at 150 K. As expected, the value of AT, and the heat capacity
which is calculated from AT and the fluctuation of the potential energy, are less
accurate than the energy or pressure results.
 264 Shfiichi Nos~

140
Q = 1.o (k#~o0(p,) ~
120

100

80

60

q : too.o (k'J/moO(ps)2
120 ]
,,% /%, ~ 6t
,~, ~,
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

100 ........ ~'~ ',. L~

i~ = , !
/, 'w~,

60 ............. '~ . . . . . . . . . .

60
600 1600 2400 stpe 3200 4000

Figure 2. Comparison of the temperature fluctuations in runs with different Q values at

100 K. Above : Q = 1 (kJ m o l - : ) ( p s ) 2, Below : Q = 100 (kJ m o l - : ) ( p s ) 2. T h e
total real time is about 7 ps.

20 /

./~
15
.~
/

dx ..... ~
/..,~ ..........
vo I0
I
v
.~176 /
....~176
V ~....~176 .~j
5 .~176176176176176176
..~

0 I t [ I
0 2 4 6 8 I0
tips

Figure 3. Comparison of the mean square displacements for different Q values at 150 K.
Solid-line Q = 1, dashed line Q = 10, and dotted line Q = 100 (kJ m o l - : ) ( p s ) 2.
 Canonical ensemble M D method 265

The comparison of temperature fluctuations in the run with Q = 1 and

100 (kJ mol-1)(ps) ~ is given in figure 2. With Q = I , the time scale of the
fluctuation of s and that of the temperature are ah'nost the same. On the other
hand, with ~ = 100, the temperature fluctuation can be decomposed into two
components : the intrinsic temperature fluctuation of the physical system and a
slow modulation due to fluctuation in s.
The time periods of fluctuations in s are t o = 0.34 ps for ~ = 1 and t o = 2.99 ps
for Q=100. The estimates given by (2.30) are to=0.38 ps for Q = I and
t o=3"8 ps for Q = 100.
As a check on some dynamic quantity, the diffusion coefficient D was
calculated from the mean square displacement.

D= lim ([ri(t)-ri(0)[2) (3.2)

Downloaded by [Duke University Libraries] at 02:56 03 July 2012

t~o 6t

The plot of the time evolution of the mean square displacement for different Q
values is given in figure 3. All values for D for a given temperature seem to
agree reasonably well. Even run No. 9 gives almost the same result as the
other Teq = 150 K runs, in spite of the large deviation from a straight line.

4. CONCLUSION
An MD method for constant temperature simulations has been presented.
One can see from the derivation in w2 that different kinds of constant tempera-
ture simulation methods can be constructed by changing the functional form of
the potential for the variable s. However, the equilibrium distribution function
thus obtained is related to the inverse function of the potential for s, so that any
other choice but a logarithmic form, results in an ensemble different from the
canonical one. Therefore, a constant temperature method does not always
imply that the calculation samples from the canonical ensemble.
The present method gives static quantities in the canonical ensemble. If
w e employ the interpretation that s is a scaling factor of time and that the real
unit time At' is related to the simulation unit time At by At'= At~s, calculation
of dynamical quantities also seems to be possible. Of course, we have rigorously
proved nothing concerning with the dynamic quantities, but it will be worth
trying to determine under what conditions can one obtain the dynamical
quantities reasonably well.
Main features of the present method are
(1) The extended system of the particles and the variable s conserves the total
hamiltonian. This gives a powerful error checking method in the process
of programming and a useful precision control method in the simulations
[7].
(2) Since the present method employs a similar technique as the constant
pressure MD method, it can readily be extended to the constant tempera-
ture constant pressure ensemble.
(3) The scaling of the velocities can be interpreted as a scaling of the time
variable. At the same time, the constant pressure MD method is
derived by scaling the coordinates. The extension of the MD method
 266 Shfiichi Nos6

to ensembles other than the microcanonical ensemble is formulated in a

unified fashion.
(4) This method gives the rigorous canonical distribution both in momentum
and in coordinate space.
If a structural change occurs during the simulation, the temperature neces-
sarily changes and the standard M D method cannot produce the data at the
required external conditions. The constant temperature method, especially the
(T, P, N) ensemble method given in the Appendix, will be particularly useful in
this situation.
During the preparation of the present paper, other constant temperature M D
methods were proposed by Evans [15], by Evans and Morriss [16], and by Haile
and Gupta [17]. The present method is the only one which gives the rigorous
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

canonical distribution both in momentum and in coordinate space. The

comparison with other methods will be presented in a forthcoming paper.

The author thanks Mike Klein, Ian McDonald, Bart de Raedt, Ray Somorjai,
and Michiel Sprik for their interest and helpful discussions.

APPENDIX

Formulation [or the constant temperature constant pressure ( T, P, N)

ensemble
The constant temperature M D method is readily extended to the (T, P, N)
ensemble.
In this Appendix, we use the formulation for uniform dilation by Andersen
[1], but the extension to the generalized form of the constant pressure simulation
method by Parrinello and Rahman can be done in a similar way [2, 3, 7].
In the case of the (T, P, N) ensemble, the M D cell is considered to be a cube
of edge length V li3, the coordinate r~ being expressed as
r i = V ll3 xi, (A 1)
where x i is a scaled coordinate and the values of its components are limited to the
range of 0 to 1. The velocity is also expressed by a scaled form
~,:= V 113 i i. (A 2)
The lagrangian is

m's2VZl3"xi~- q~(Va,3x) + 7 g2- ( f + l )kTeq ln s + W f l 2 - P e x V . ( A 3 )

~ = ~ --~

Pox is the externally set pressure. The kinetic energy term for the volume is
89 The equations of motion are

d (mis 2 V2!3 X i ) - ~C~ V lla f--~-~ (A 4)

dt ~x i ~r~
or
 Canonical ensembleMD method 267

Qs= E misV2j3~ti2- (f + 1____kT~q

__) (A 6)
S
and

(A 7)
With the momenta
Pi=miss V2/3x~:, pv= WV, and P s = Q i ,
the hamiltonian becomes
pv 2
~2= ~i 2m,zspi~V
2 2/a~-(~(V1/ax)+Ps2+(t+ l)hTeqlns+2-W . (AS)
Downloaded by [Duke University Libraries] at 02:56 03 July 2012

The partition function of this system is

Z ---~v.
1 S dpsS ds~ dpv~ dV~ dp ~ dx (~aff2-E) ( a 9)

By scaling p and x as p/sVa/a=p', V1/a x = r, we get

Z _- 1~ I dp~ I dp~ I dV I dp' I dr I a s s ~ ( ~ - e)

Following the steps in w2, one derives the result

1 ((27r)2QW)l/2exp
Z - ( I + 1)
[ ~ E ] ~.l1 ~dV~dp'~dr

9 p, 2+ P~xV)/kT~q ]

- ( f + 1) ((2rr)2QW)a/2 exp ZTpN. ( a 10)

ZTpx is the partition function of the (T, P, N) ensemble. The averages of

static quantities which are functions of p', r, V are the same as those in the
(T, P, N) ensemble :

s~' x, V = (A(p', r, V))TpN. (A 11)

REFERENCES
[1] ANDERSEN,H. C., 1980, ft. chem. Phys., 72, 2384.
[2] PARmNELLO,M., and RAHMAN,A., 1980, Phys. Rev. Lett., 45, 1196.
[3] PARRINELLO,M., and RAHMAN,A., 1981,ff. appl. Phys., 52, 7182.
[4] PARRINELLO,M., RAHMAN, A., and VASHZSHTA,P., 1983, Phys. Rev. Lett., 50, 1073.
[5] Nos~, S., and KLEIN, M. L., 1983, Phys. Rev. Lett., 50, 1207.
[6] Nos~, S., and KLEIN, M. L., 1983, ft. chem. Phys., 78, 6928.
[7] Nos~, S., and KLEIN, M. L., 1983, Molec. Phys., 50, 1055.
[8] WOODCOCK,L. V., 1971, Chem. Phys. Lett., 10, 257.
[9] ABRAHAM,F. F., KOCH, S. W., and DESAI, R. C., 1982, Phys. Rev. Lett., 49, 923.
[10] HOOVER,W. G., LADD, A. J. C., and MORAN, B., 1982, Phys. Rev. Lett., 48, 1818.
[11] TANAKA,H., NAKANISHI, K., and WATANABE,N., 1983, J. chem. Phys., 78, 2626.
 268 Shfiichi Nos6

[12] HOOVER,W. G., and ALDER, B. J., 1967, ft. chem. Phys., 46, 686.
[13] LEBOWZTZ,J. L., PERCtSS,J. K., and VERLET,L., 1967, Phys. Rev., 153, 250.
[14] CHgUNG,P. S. Y., 1977, Molec. Phys., 33, 519.
[15] EVANS,D. J., 1983, ft. chem. Phys., 78, 3297.
[16] EVANS,D. J., and MORRZSS,G. P., 1983, Chem. Phys., 77, 63.
[17] HAZLE,J. M., and GUPTA, S., 1983, J. chem. Phys., 79, 3067.
Downloaded by [Duke University Libraries] at 02:56 03 July 2012