Statistical

thermodynamics

in the classical

moteculak dynamics

ensemble.

Rolf Lustig
Institut ftir Physikalische Chemie, Rheinisch- Westftilische Technische Hochschule Aachen, 52056 Aachen, Germany

(Received 10 June 1993; accepted 26 October 1993) The statistical thermodynamics of a classical system composed of rigid molecules is considered in the molecular dynamics ensemble. Accepting Boltzmann S= k, In W as the basic assumps tion of statistical mechanics, exact formalisms for two classical choices of Ware derived. Since there are no restrictions on the order of thermodynamic derivatives, any measurable quantity is directly accessible in this ensemble. Explicit statistical analogs are given for the derivatives of the Helmholtz energy including an approximation for the chemical potential. Basic phase space functions are identified and their properties are explored. It is shown that complete therthe modynamics is governed by small perturbations of these functions from universal behavior.

1. INTRODUCTION Molecular dynamics computer simulation (MD) is a powerful tool in modern molecular physics. In principle, static, dynamical, and thermophysical properties of matter, made up of appropriate model molecules, can be studied in detail. Moreover, if assumptions about the force laws governing the behavior of the systems are made, MD is a predominant method to test their validity by comparing theory to experimental data for real systems. Here thermodynamic properties play an outstanding role, This is because of the many possibilities to check consistencies, to consider phase transitions or other state regions that exhibit peculiar behavior in certain thermodynamic properties. Thus, it is not surprising that MD is used to mimic various statistical mechanical ensembles with fixed macroscopic parameters that correspond to the indspendent variables of the thermodynamic fundamental equations. An excellent review on these methods is given in Refs. 1 and 2. Despite the fact that the mechanical aspects of MD are highly developed, the statistical thermodynamics is worked out only to first order derivatives of the fundamental equations and some second order derivatives such as the specific heat or the compressibility. The classical molecular dynamics ensemble (NV&P) with fixed numb& of particles N, volume V,total energy E, and total linear momentum P is closest to the microcanonical ensemble (NVE) , the natural starting point of statistical mechanics. However, even though the microcanonical ensemble has the most simple distribution function, a constant, the analytical analysis of this ensemble is considered the most difiicult.3 This is 4 probably why the NVEP ensemble is treated through an NVE ensemble in either an intuitive way,5 an indirect way,6 or in an approximate way using the statistical thermodynamics of another ensemble, such as the canonical ensemble, as a reference. However, it was shown recently by Pearson, Halicioglu, and Tiller* that there exists a straightforward way to set up the complete and exact statistical thermodynamics

of structureless particles in the microcanonical ensemble through the use of statistical mechanical formulas which have been known for a long time.g This method allows the calculation of thermodynamic derivatives to arbitrary order and does not involve any assumption about the thermodynamic limit (N-t 03, V/N=const). An extension to the NVEP ensemble of structureless particles was then worked out by Ca&n and Ray. These works were completely ignored by the simulation community. A vohimeby-volume search of the Citation Index up to February 1993 reveals that the results of the two papers8 were never used but by the authors themselves. The only exception is the very recent book of Haile, who describes these methods in an appendix. Some plausibility arguments were used independently by Litniewski to derive a phase space integral for a molecular system which under a certain condition is identical to Eq. ( 13) of this work. In those papers some selected first and second order thermodynamic derivatives are given and numerical tests were performed on (quite large) systems of about 500 atoms. This work aims at a systematic development of the statistical thermodynamics of rigid molecules in the classical NVEP ensemble building on the methods described by Pearson, Halicioglu, and Tiler* and by Cagin and Ray. The paper is organized as follows. In Sec. II the classical phase space integrals of a molecular system are transformed into a tractable form. It is emphasized in Sec. III that there is no concensus about the use of Boltzmann s entropy relation. In Sec. IV the basic thermodynamic properties are identified. The basic phase space functions from which the exact thermodynamics can be derived are outlined in Sec. V. Residual and ideal properties are dealt with in Sec. VI. Statistical analogs of special thermodynamic properties are derived in Sec. VII. Section VIII discusses the chemical potential. Some particular features of the basic phase space functions and their consequences are discussed in Sec. IX. Section X contains the conclusions.

J. Chem. Phys. 100 (4), 15 February 1994 3048 0021-9606/94/100(4)/3048/12/$6.00 @ 1994 American Institute of Physics Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

3049

II. CLASSICAL STATISTICAL

MECHANICS

the phase space density

Consider an isolated system of N rigid nonspherical particles in a volume V with total energy E, constrained to a subspace of the microcanonical ensemble by a total linear momentum Z~flpi=P. The classical phase space integrals arei3***io the phase space volume
Q[NV..)

u(NVEP)=

an(NVEP)
aE

= const

NE--H&J 1 HW <E
(2)

=const
HCpd <E

@[E--H(pq)

XS(P- jl Pi)+ d q,

x6

1 1
Pii
i-l

pi dpd q,

(1)
I

and the ensemble average of any dynamical property

constSSH(pg)<~X(pq)~[E--H(pq)lG[P--X~v=lpildPd q
(---(Pq)
)NVEP=

u(NVEP)
I

(3)

where the unit step function 0 and the delta function S pick out the desired E, P, and H(pq) is the Hamiltonian of the system. Note that the notation is such that p stands for all momenta conjugate to q and that the volume element of configuration space is given as drq=drNdrtiN with d o =dadpdy (phase element) as opposed to do ==$nPdadfldy (solid angle element, usually used in the theory of molecular fluids), so that dq= (I-IL 1 sin &)d q. These distinctions are essential in what follows. r is the position vector of a molecule center of mass and (a&) s are the Euler angles of-a principal axes frame with respect to some space-fixed frame. The normalization of fi and w to dimensionless numbers is absorbed into the constants const which, however, are of no interest for the rest of what follows. As they stand, the relations (l)-(3) are of little use because the integrals, are not factorized into momentum and configurational parts. In the following the Hamiltonian will be assumed to be of the classical form H(pq) -K(p) + U(q), with U the potential energy of the system. The kinetic energy is expressed by

COS

cos yj, sin fii , JikPy* * For linear molecules the symmetry axis is along the bodyfixed z-axis and thus the third Euler angle y is egelic so that there is no contribution to the rotational kinetic energy through py, . In that case Eq. (5) holds if one sets pri, yirO* Transformation phase space integrals. The method for partitioning the phase space integrals for a system of spherical particles of Pearson, Halicioglu, and Tiller and Gagin and Ray consists of (i) Laplace transformation with respect to E; (ii) solution of the momentum integrals; and (iii) inverse Laplace transformation, and was thus termed Laplace-transformation-technique (LTT) . Similar results had been obtained earlier by Minister9 using the inverse Fourier transform representation of the delta function. An equivalent way is the use of the inverse Laplace transform representation of the delta functions for energy and momentum
(5)

pi

of

(4)

S[E--H&J I=&
where i labels the individual molecules, m stands for mass, lik is the eigenvalue of the inertia tensor along the principal axis k, Jik is the kth component of the angular momentum vector Ji in the principal axis frame and f is the number of degrees of freedom per molecule. Since the rotational parts of the integrals ( l)-( 3) are over the conjugate (canonical) momenta p. , one needs a relation between pw and the angular momentum in the principal axis frame. By geometrical considerations one obtains for the general case of nonlmear molecules 4

I_ expCdE-H(pq) 13d~, ,
(6)

+-

g Pj=(&)3Jjj

--m

mpi k~y,z+~--.$~~)]

where the arbitrary constant (T is inwith SE,k=c+irE,k, troduced to ensure that no poles occur along the path of integration. Using Eqs. (4)-(6) in Eq. (3) one obtains

J. Chem. Phys., Vol. 100, No. 4, 15 license 1994 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP February or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

3050

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

C O IIS~(&)~JJJJJ~k~~,=~k~k)I(sE,sk)d~~~~~~~= exp( , W (Pd )NVEP= w(NVEP)

with

(7)
J(SE,Sk) =

X(pqbq4dE-

U(q)Ikd

-c&(p) lexp

To factor the integral I, the dynamical quantities have to be restricted to ones of the form X(pq) =X (p)X(q). Fortunately this is all one needs for the complete statistical thermodynamics, as will be shown later, where X (p) is associated with the kinetic and X(q) with the potential energy. Furthermore, due to the constant total energy constraint and the use of the classical Hamiltonian, the kinetic energy Eq. (4) is actually a function of the configurations. Thus, it is completely sufficient to consider X=X(q) only, so that I(sE,.%) =Ikin(SE,Sk) Wl 4) X(q)

After inserting Eq. (5) into Eq. (8) the rotational part of Ikin contains terms that are formally identical to the ones that occur in the rotational partition function of the canonical ensemble. Details of these kinds of integrations are outlined in Ref. 16. After considerable algebra one finally obtains sin pi.

(10)

With Eqs. (9) and ( 10) at hand oneecan insert Eq. (8) back into Eq. (7) and do. the integrations over the r s;.f?rst rk and then rE. Using Ref. 15 the final result is (Nq) )Nvm=a for [EU(q).-P2/(2M)] <O,

XexpCsdE-W I) IW q,
with IkinbE,Sk) = JJ exp[ ---s.,&(p) I
/ iv

(8)
(au PI

(x(d =

)NYEP

cJJcall

q~X(q)[E-U[q)-P2/(2M)](F-5) 2dq

F-3 w(NVEP) --T( 1

for [E-u(q)-P2/(2M)] >O,

7 (11)

d p.

Using Eq. (4)) the kinetic part of Eq. ( 8) further factorizes into a translational and a rotational part. For the translation part one obtains

with C=const(2~M) 2 - 3 ii
i=l

(2~mi) 3 ~ 2
i=l

~ (2?rIik) I *.
k=l

I trans=

b,

&?a=

I+m , -co

W?(-~~$.-sip,)dPik

= jj

kgyz (~jl"exp($$)% I

(9)
I

The constant C is of no interest for the rest of this work. Note also that the prime has been dropped from dq, due to the sines in Eq. ( 10). The above two cases can be put together by using the properties of the unit step function,

(X(q))NVm=

CJ-J(,~ ,,xcq) [E- u(q) -P*/(~M>

l(P-5) 2@[E-u(q) -P*/(=f) I&

(12)

w(NVEF+)
r

with

F= 2 fip
i=l

M= fil mi.

For labels the individual molecules, F is the total number of degrees of freedom, and M is the total mass of the system. Note that Eq. (12) also holds for systems composed of different species. From Eq. (12) one gets the phase space density by setting X= 1,

J. Chem. Phys., Vol. 100, No. 4, 15 February 1994 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

3051

According to Eq. (2) the phase space volume is obtained through integration of w. This is most easily done by writing Eq. ( 13) in a form similar to Eq. ( 11) and choosing a lower integration bound such that fi(E,) +O. Then the result is

R(NVEP)=
\

r(F-l l
L

JJ [E-W) -bl1 9)

- p*/( &$f) ] (F--3) *

x@[E- u(q) -P*/(+W I&

(14)

space volume are given. However, there remains the fact that Eq. (15a) implies the counting of phase space points that are a priori prohibited due to the constraint of fixed total energy. This was already outlined by Miinster who also argues that from a physical point of view the entropy definition through the phase space density Eq. (15b) should be considered the primary one. Litniewski12 also uses Eq. (15b) but without discussing the point further. Obviously, there is no agreement in the literature about which phase space integral ought to be used for Win Boltzmann basic assumption of statistical mechanics s S=k, In W in the case of small systems. In this work the thermodynamics. will be constructed. from both entropy definitions. It will turn out that there exists a general connection between the resulting formalisms. Then, it. is perfectly appropriate to try to explore by numerical work which entropy definition extrapolates correctly to smaller numbers of. particles (cf. part III of this series). IV. BASIC THERMODYNAMIC PROPERTIES

where the property of the gamma function xl?(x) = I (x -+l) was used. As can be seen, the transformed phase space integrals (12)-( 14), here derived for a multicomponent system of arbitrary spatial rigid particles, are no more complicated than the corresponding ones of a system of spherical particles, lo which of course is included as a special case. The step function in ( 12)-( 14) occurs for formal reasons. If the trajectory of the system is generated physically through, for example, a MD computer simulation, it is guaranteed that 0 never vanishes. This is easy to see for B- LJ=K is a quadratic function of the momenta and the kinetic energy of a physical system cannot be less than the translational energy of the total center of mass. Thus, the step function can be ignored for real systems. However, in Sec. VIII a nonphysical situation is met, where the step function plays a crucial role. III. CONNECTION TO THERMODYNAMKS In classical statistical mechanics it is shown that in the thermodynamic limit both the phase space volume ( 1) and the phase space density (2) are proper analogs to the entropy13 through
S&7,= SrS,=kB kB ln i-I, In co, (154 (15b)

To set up the thermodynamics systematically it is convenient to start from the fundamental equation E=E(S,V), or equivalently, S=S(E,V) for the NVEP ensemble. Then, the complete thermodynamic information about the system is contained in a?S/(aEaV> for
m,n>O.

For practical purposes, however, one does not need the energy or entropy (caloric) dependencies of thermodynamic properties, but the temperature or pressure (thermal) dependencies. Thus, for thermodynamic considerations it is appropriate to represent the results with T and V as independent variables, so that the Helmholtz energy A =A ( T, V) is the fundamental quantity. Since A is related to the entropy it cannot be obtained directly from MD without additional concepts such as thermodynamic integration dr a proper particle insertion method. However, all T- and V-derivatives of A are directly accessible as ensemble averages, that is am+n+lA am+p
-aTma~+T=Z%F
am+n+lA -TaTm+layn=armdyn+(m-l) am+nE

3 a + A
-9

(16)

for m,n>O. For m =II =0 the right of Eq. ( 16) yields the pressure and the internal energy so that in the following these quantities are considered as the basic thermodynamic functions from which the rest of thermodynamics will be constructed by differentiation.19 Of course, not all derivatives of E and p are independent. From Eq. (16) follows am+++ 1 am+n+lE am+np P
aTmaV+ =TaTm+lavn+(m-- )

with the property that17

Thus, from the classical point of view it was argued that temperature can only be assigned to systems large enough to make Eqs. (15a) and (15b) identical, so that the question about what entropy definition is more correct does not make sense.17However, since MD systems are small, the * point is different here. In fact, using the two entropy definitions will result in formalisms that differ in all thermodynamic quantities. In the previous work of Pearson et al. and Ca&in and Ray some arguments for using the phase

aTmavn

for m,n>O. Thus, for a complete thermodynamic characterization (except for A itself) one has to derive the statistical analogs for the independent derivatives

Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to 4, 15license or 1994 J. Chem. phys., Vol. 100, No. AIP February copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

3052

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

am+%
Em+~,o~apn+~

am+np
9 Pmn ai;mavn

7 for m,n>O,

(17) where in the following the shorthand notation a fx + r sXif for thermodynamic quantities,
aPavJ

o where (=,- w ) indicates whether Eqs. (Ha) or (15b) used as the entropy definition. To obtain other derivativ in Eq. (17) the recursive relations

1 a +jfi 0-j

(18)
=:--62, for phase space quantities,

afb -=%?l+l,n-~zo%m~ aE an, ---=~m,n+~-~~~~mn,m,n>O for

av

(2(

will be used. Then, for A,, with m,n>O the phase space functions ti2, for at most i,j<m& l,n will be required. The procedure to be-followed now consists of (1) expressing E,,,, and pmn as functions of ai+iE,~(&lVj) and d +*p/ a jE,,-J( LIE??Vi) * into (8Y?a Vj) . (2) Transforming ak+k$/(aEkav9 as th will simplify the rest of the cal1s +~E; /(&%hVj) and diffp/ culation. (3) Expressing a (&?W~) as functions of the phase space integrals fl and w. As an example, from the thermodynamic relations

are useful. Explicit results are listed in Appendix A. Equation (19) suggests what eventually proves rigc ously true through mathematical induction. The two e tropy deli&ions lead to statistical thermodynamical fc malisms that are related by the simple transformation

&(uLl3)

=&l(cfb+l,n&.o3~.

(2

k~/Elo=k~(~~),=k~[~(~)G i.
- P=(36 ($ .;
and the entropy definitions JZq. (15) one obtains for the tirst few properties

In other words, any thermodynamic property coming frc S, Eq. (Hb), can be generated from the same proper coming from So Eq. (15a), by replacing each member i of the set {C&J with R i+r,j/&e. This is an importa result which shows that there are certain relations amo. the (a,,] in the thermodynamic limit (cf. Sec. IX).

V. BASIC PHASE SPACE FUNCTIONS Next, the phase space integrals firnn will be identifi with an ensemble average of a dynamical property of t system. The configuration space in Eqs. (12)-( 14) consi of center of mass positions and orientations, that is =drNdoN, where the position space extends over the VI ume of the system. For the volume derivatives the positi space is scaled byi rj= Vq-f ,
dri= Vdr[ ,
(F

a. kB/EIO= 0) kB/EIO= 1 - &,&,,, 1 -Q&o,

for i= 1,2,..., N,

to make the upper integration bounds in Eqs. ( 12)-( 1 constants. Then the mixed E, V-derivatives of E!q. ( 14) : suit in .-

m F-3
2

[E-U(r%N)--B2/(2it4)]~F-3) 2-m

X@[E-u(r;N)-P2/(2M)]dr NdWN+(l-60n)
. -

i
i-1

n--i (_ y-f(
w-

--Njnei

IE-

u(+VmN)

-p2/(2~>

IF-3)/2-m-l

x [ kmfo
k=l

c&S@-~~~]@[E-- U(rNmN)

-P2/(2M)

]drtNdtiN
I

Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp J. Chem. Phys., Vol. 100, No. 4, 15 February 1994

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

3053

where (x), is short for the Pochhammer polynomial x(x+ 1) (x+2) **,*(x+n1) with (x)crl symbol. Dividing Bq. (23) by Eq. (13), using E!q. ( 12) yields after some rearrangements

and a,, is the Kronecker

am,=

N y

(--YY-N),
N

I
(->- (-N)n-i
; pl-i ix I=1

E-U(rNmN)--P2/(2A4)
F-3 2

-(m-l)

>

NVEP

-l-(1--6on)

i=l

c.

n i

--... _I

I,+m

1
r

prilk
_ F-3 2
I ciik , N \ i

(24)

where the angular brackets denote an ensemble average. The combinatorial number ciN( and the product of various volume derivatives of the potential energy rilk (for instance Y5 6s2= (-JU/av) ( --L?U/C~V~) ( --LJ3U/8v3) are given in Appendix B. The phase space functions am,,, and thus the complete thermodynamics, are given as ensemble averages of some power of the kinetic energy times certain products of various volume derivatives of the potential energy. The first two factors in the curly brackets go to unity as N becomes large and so do the corresponding factors in the second summand. Equation (24) suggests that the natural intensitivity parameter of the kinetic energy is not N but (F-3)/2cN[l-H(N- )I. Another interesting property follows from the occurence of the Pochhammer polynomials. With ( --I)o= 1 and ( --i)i>I=O one has
Omn=

tions to the thermodynamic properties can thus be removed, and full advantage can be taken of the theorem of corresponding states when comparing different substances. However, care should be taken not to confuse the configurational (usually used in the statistical thermodynamics of the canonical ensemble) and the residual properties. Both measure,, the effect of the intermolecular forces, but they differ by the configurational properties of the ideal gas.21 The residual properties can be determined experimentally, and they are used in the formulations of the most accurate empirical equations of state in terms of the Helmholtz energy.2 On the microscopic level the ideal parts of the thermodynamic functions are obtained by setting -thepotential energy equal to Y&o, m-.

0: i 0:

for m > (F- 3)/2, arbitrary for n> N, arbitrary m.

n,

X%Tv) +-W, v>ty(rN&=o:

The phase space functions Eq. (24) become

This means that well delined thermodynamic derivatives of a l&rite classical system are limited to a finite order. Note that up to this point no approximations have been made, except the requirement that the container holding the system be cubic, which enters through the scaling Eq. (22). VI. RESIDUAL AND IDEAL PROPERTIES It is common practice in thermodynamics ideal gas as a reference state according to
X( T, Y) =Xtd( T, V) + [X( T, v) -Xfd(

F-3 ,2 (7)

m . .=

E-.p2/(2&f) I.. .~ X
:-

-(m-1)
I I

to use the (25)

I---

F-3 2

(26)

T,V) 1,

where the term in parentheses is the so-called residual part Xres of X. Instead of T, V as the independent variables T,p may also be used. The separation Bq. (25) is advantageous in the statistical theory of fluids when considering, systems with and without intermolecular forces at the very same state point. Density independent intramolecular contribu-

Applying the procedure described in Sec. IV, any Xid can be evaluated from Eq. (26). In particular one can derive for the basic thermodynamic properties in Eq. (17) the general expressions
{E, (Eff,,)n,o= I F---l, 2

for m=n=O
0

kg,

for m=l,n=O (27)

I 0,

otherwise,

J. Chem. Phys., Vol. 100, No. 4,15 Februav lQQ4 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

3054

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

(P%h,*= I
n! (-YN 0,

..
N + kg, 7

with In==3,

1,=5.

for m=l,n>O

0 otherwise,

I The subscripts 0,w indicate whether Eqs. (15a) or ( 15b) is used as the entropy definition. A representation of the residual functions through the independent variables T,V in the MD ensemble requires care, for
xreS( T,V)#X(E,V) -Xid(E,V).

(28)

The natural independent variables are after all still E, V so that the temperature is of course a dependent variable. The first term on the right does correspond to X( T, V), but the second term corresponds to the temperature Tid. Here it is important to realize that E-P24
F---l&3

2jW) 2

i2$=

(kBT)g

(294 (29b)

derivatives and will be given in terms of the phase space functions using the entropy definition Eq. ( 15a) which, if neccessary, will be indicated by the symbol 2. The corresponding relations for the entropy definition Eq. ( 15b) can immediately be obtained through the transformation Eq. (21). The thermodynamic internal energy is the total energy of the system and wassdenoted by E. From the last section it follows that the residual internal energy is the average potential energy of the system. At this point one encounters notational problems, for in phenomenological thermodynamics the total internal energy is denoted by U. To avoid confusion, the following definitions are necessary:

= I (n, )=(k,T)~

E=U=( U(q? + UGpN, > >, E==P=(U(by)).

The thermal pressure coefficient is given by ~11-~01~20 , 1 - QoofJ20

To make Eq. (28) an identity, one has to substitute Tid for T. Comparing Eq. (29) with Eq. (19) shows that this corresponds to the exchange
E-P2/(2M)

fkoo for S=kB 4 I f&g

In n
.

(304
(3Ob)

(324

F-4-+ 2

for S=kBhw

If this exchange is performed in each X,. as in Eq. (27), then the residual properties Xs are properly represented by T,V. As indicated by Eq. (27) there is a difference between caloric and thermal properties. The ideal gas part of the caloric properties (such as the internal energy) may depend on the individuality of the molecules (vibrational, electronic contributions). Consequently, a rigid model, as used in .this work, cannot account for the details of the temperature dependences of a real system. In that case, the left-hand side .of Eq. (25) is known a priori to be different. The ideal gas part of the thermal properties (such as the pressure and its T, V derivatives) does not depend on the individuality of the molecules. Consequently, all terms of Eq. (25), for the model and for the real substance, do correspond directly to one another. VII. SPECIAL THERMODYNAMIC PROPERTIES

and the isothermal coefficient of bulk compressibility given by

is

ap -v (av T 1
c-z

1 o PT

~olc2fhl-~ol~2o>

-Go%

1 - %0~20

--,no2 7

1(33 (33
(344

For the adiabatic pressure coefficient one gets

In principle any thermodynamic property (except the fundamental functions) can be obtained from the T,Vderivatives of the internal energy and the pressure, as given in the preceding sections. However, some properties that are of special practical interest21 will be given explicitly below. They are obtained by transforming to E,V-

Since the caloric quantity 5is involved, the ideal part does depend on the individuality of the molecules. The adiabatic coefficient of bulk compressibility is given by

J. Chem. Phys., Vol. 100, No. 4, 15 February 1994

Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

Fioif Lustig: Statistical thermodynamics

in the MD ensemble. I

3055

(354

The ideal part of C~JT vanishes, confirming the well-known fact that thisquantity is entirely determined by the intermolecular forces. As a third order thermodynamic derivative the temperature variation of the isochoric heat capacity is given by

As for ys, the ideal part does depend on the individuality of the molecules. Note that Eq. (35a) gives the s eed of sound through the thermodynamic relation w- P= (M/ V)& where M is the total mass of the system. The isochoric heat capacity is given by
n
v==Cv=kB[ 1 --o&o] -1 ,

( 1
dT

acv

n Jel

~20~~--2(~--~~20~1-~00~30

Vf-

Cl-%0~20>

3 (39)

(36d

According to Eq. (27)) the ideal part of this quantity vanishes for a rigid model. To express the residual properties of this section with T, V the substitution Eq. (30) *must be performed.

(36b)

VIII. THE CHEMICAL POTENTIAL where the ideal part, of course, depends on the individual_~ ity of the molecules. The isobaric heat capacity is not a new quantity if CV, &, and fis are known, since they are interrelated by Cd Cv=&/&. The statistical analog is given by An approximation to the chemical potential in the MD ensemble was given by Frenkel.23 Here the method of Sec. II is used to derive expressions that differ between the two entropy definitions E!qs. ( 15a) and ( 15b). Accepting the usual difference scheme,24

(i$C,3kB[ l+=k,( l+$)=NkB(

00(Ro2 20--n:1) , ] -
~01(~01~20-~~,1> +a02 (374

-S(N+
as the only approximation

l,V,E) --S(N,V,E)
involved, one has

I+%).

(37b)

Again, the ideal part depends on the individuality of the molecules. The Joule-Thomson coefficient is also not a new quantity because of the thermodynamic relation pcLJT = V( T&yv1 )/C The statistical analog is explicitly . given by
aT

I
exp

n(N+

1, VEP)

~(NVEF )

for S=kBlnfI

(@aI
4NVEI)
, for

w(N+l,VEP)

-S=kB

h W.

( apJf1 =PJT
-2 I l+ ~~~Q11~~+~~11~-~20~~01+~~02~1 ~r~02--Sl01~~5111-~01~20~

With the difference in the potential energy between an (N +~l >- and an N-body system

UT(q? = u(p+ - U(q9, > )

1 1.

(38) I

the ratios of the phase space integrals become

mN+l

E-u@)

-p2/(2i5f)

U(fl) 1 [E(F+fiv+l-h,o)/2

--P2/(2hf)]fN+1 2

NVEP

&iv+

1,

J. Chem. Phys., Vol. 100, No. AIP February copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to4, 15 license or1994

3056

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

where superscripts label collective variables and subscripts label particles. F,M are the system total degrees of frees dom and mass, f, m those of one particle and in,,= 3 or 5. In other words, insert the test particle (N+ 1) uniformly into the phase space trajectory of the N-particle system and determine the average of the integrand. The formal occurrence of the step function is crucial, for it takes care of the unphysical nature of random insertion of the test particle. Any random insertion that results in @(flf )
> E-

Let Y be a homogeneous function of degree k in the variables N,V,E. Then, for an arbitrary parameter ;1 one has
Y(AN,AV,ilE) =AkY,

which, upon differentiation of both sides with respect to d, results for /1=1 and k=O in
g N+f& V+g E=O.

(43)

-P2/(2M)
W(fl)

My+&

-P2/(uM)

Taking the limit N+ 00 with V/N, E/N=const rigorously [and not as a substitute for N-d(NA)], Bq. (24) shows that fima becomes a truely intensive state function, lim fl,,(dN,AV,A.E)
N+m =Cimn.

must be counted but is assigned a value of zero. This makes perfect sense, because without this restriction the averages in Bq. (41) diverge towards f CO Besides this point and a i hybrid of the two entropy definitions Eqs. (40a) and (40b) which results in deviations of the order d (N-l), the expressions in Ref. 23 are in agreement with Bq. (41) for P=O. The ideal part of the chemical potential is exp[ ~PvJ = Cn,,V(1,4a,8~}

This makes fimn a homogeneous function of degree zero, so that Eq.. (43) applies. Next it is stated, that in the thermodynamic limit j-mm T

~%?ln

E=i+rnW 7

~fLzn

V=f-mw

a~

6Qm,

N=o

(44)

holds. This, in connection with Bq. (20), leads to lima

N-co lim %,n+l=Q~~%,, N- m (45b) m+ 1,n=~20%n 9 (454

X [E-P2/(2M)]fN+ ,

I
l-

(F+f~+1--kt,di2

-P2/(uM)
E--P2;I(~)

miV+I
M+mN+l

and to the additional explicit relation

I !2 --i-k,,i-l,, lim N-m fTn+l m+1,n-~20%m= --EN
N V

(42)

(46)

where the constants in the curly brackets are for spherical, linear, and nonlinear molecules,. respectively. Often variables other than E, V are needed to represent the residual chemical potential. As outlined in Sec. VI, a representation with T,V is given by substituting in Bq. (42),. Most important, however, is the representation pres =,o( T,p) since this .gives a fundamental equation and determines thermodynamic phase equilibrium. A T,prepresentation of pres, however,, is the same as a T,Vrepresentation, because the statistical analog of the chemical potential Bq. (41) does not depend upon a generalized virial as they occur in Bq. (24). All constants in Bqs. (41) and (42) are of no importance for the residual part of the chemical potential. Since the integral in Bq. (41) contains the full volume element of the configuration space, it is easy to see that both the volume and the geometry factors drop from Eq. (42) when taking the residual part.

The vanishing of the first term in Eq. (44) follows from the transformation relation Eq. (21) in connection with Bq. (20). A successive application of Bqs. (45) to the righthand sides of the relations shown in Appendix A 2 makes all inverse extensive quantities disappear. This proves the vanishing of the second term in Bq. (44). The vanishing of the third term then follows trivially from Eq. (43). Bquations (45) are recursion relations that can be generalized to sin
N-02

m+k,n+l=~~owLn.

(47)

IX. SPECIAL FEATURES Some general properties of the basic phase space functions <I,, will now be derived.

If amn is considered as a continuous function of the variables m, n, then Bq. (47) implies an exponential surface in both m and n with the fixed point Rio= 1. If one were to be in the true limit N+ M), then this surface and thus the complete thermodynamics would be given by only two points, namely a, i and fi2,. This is in contradiction to the outline around Bq. (17) of Sec. IV and shows the supposedly trivial fact that a thermodynamic system ought to be flnite to be meaningful in the classical sense. More important than this is the fact that obviously the thermodynamics in the N VEP ensemble is completely governed by small departures of the surface arnn from exponential behavior in either variable. This suggests a perturbational treatment.

J. Chem. Phys., Vol. 100, No. 4, 15 February 1994 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

3057

X. CONCLUSION It has been shown that the complete thermodynamics of a system of rigid particles can be set up rigorously in the classical molecular dynamics ensemble. The only assumptions are Boltzmann s relation for the entropy, S=k, ln W, and a cubic container holding the system. Using for W the two classical choices, phase space volume a2, and phase space density w, exact formalisms are derived that are shown to be interrelated through a general transformation. Statistical analogs for a variety of thermodynamic properties are given. Building on the usual difference scheme for constructing the thermodynamic fundamental equations, the chemical potential is derived for both entropy definitions without further assumptions. It is shown that in general the thermodynamics of this ensemble is governed by a small departure of the basic phase space functions cn,,=o- a+~/(aEaY) from known behavior.

p11=E,o(&l-foP2o-f24o), ~12=E,0V 12--2~21+f~304f2P11+fd 20 + (fof4+f$-f3>p1019 p2o=~o(p2o-f7ploh

P21=~0[p21-fop30-f7p11-f8p20 - (f4-f2f7)&019

(A3)

P30=~0(p30-3f7p20-f9e0)1

with
fo=EloTol, ~~=E~o[EIoToI(~~oI--I~~oTo~) f2=&oUol-E1d10T01), ~~=EIoVO~-~~O~~~~ +11oTo2
To113, --To219

APPENDIX A: STbTISTICAL

ANALOGS

+ (~11--E1d1do1)

In Sec. IV the principal steps towards the complete thermodynamics are outlined. Here all statistical analogs are given to determine the derivatives of the Helmholtz energy up to total order four.25 The notation is as in Eq. ( 18). The basic properties for all following computations are chosen to be those of IQ. (19) using the entropy definition Eq. (15a). To avoid notational confusion one should bear in mind the definition of the internal energy through Eq. (31). 1. Basic quantities and auxiliary quantities From the general transformation

f4=E,oC~11-E1oI:2~1do1f(~20-~1d~o)To1l)t

(A41 fs=EloCTo3-EloWo1To2+ f6=E1o[EloTol(41ol--&dloTol)--o21, f7=4dlot fs=E1oWo1--3Eld,oTo,), fs=EloU2o-Wd;o).

(~02--%?~1)To1lh

These formulas may be checked through the observation that after insertion of ECq.(A4) into Eq. (A3 ) the sum of the coefficients vanishes upon expansion, except for plo.

(Al) (~),=(;;V),--13io(qo)v
and the shorthand notation
P fp LlaEl?z&,M am+E$ aEnay 2

2. Auxiliary, quantities

and phase space functions

Through a systematic application of the relations (20) the following results are obtained for Pm,, I,, , and T,, :
g=~ll--RolLL20, (A54

Irnn=

C-42)

aT b~V=~o,-W-h, JP
~v=~02-~01 n111

(A5b)

drnfT -Tmn=apapm

(A@

one can derive successively the following formulas:

P01=P01--fd30~ poz=po2--2

~=n,,-~o,n,,-zn,,~,o+2~0,~:,

(A74

foPll+fao+f

IPI
kB

Po3=P03-3foe2+3f~2l-f~3o+3flP1l--foflP20
+ (2

aE,l -=n20-- n00~30-2~*0Q20+2~n00~~, aE (A7b) d2T

=~02-4&12-~ n01~11+~J-&&:1,

fof3-f&-f

1f2-fdPl0,

~lo=E&o,

kBiw

(A7c)

Cher+ Phys., Vol. 100, No. 4, 15 February 1994 J; Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

3058

Rolf Lustig: Statistical thermodynamics

in the MD ensemble. I

Wd

(A9)

The form of these equations is suggestive. There is a lineby-line correspondence between the derivatives of p, EG , and T, which may serve for checks. Notice that for this reason the identity alo= 1 was not explicitly used. Setting all CL,= 1, the result is always zero. Setting all C.Ijj arbitrarily equal, the results for equal total order of derivatives are always equal. Let k be the total order, and I the number of .C+factors of a summand, then the SUIII of the coefficients of these summands is always a roduct of Stirling numbers of the first and second kind,2 ? that is, Sj1)Yit2 for derivatives of E, and S~l c(p~z~I for derivatives of p and T. APPENDIX. @: MULTINOMIALS The terms of type (c;lkFilk) in Eq. (24) can be obtained from combinatorial analysis. Consider a general function y(x) which may stand for U( V). Then .WiR is the kth product made up of 1 factors, where each factor is a derivative (Py/axm), so that the total order of derivatives is i. In other words, i is decomposed into I distinct summands of positive integers, without -regard to order. Then, ciN( is the number of possjbilities to realize this decomposition and thus a multinomial coefficient,26 .11
cilk=

a~!(1!)= a~!~2!>~~~~a~(il)

with at+2a2+***+ia~=i. (Bl) This is to be read, for a given decomposition (ilk) the integer n (n= 1,2,..., i) occurs a, times (see also Ref. 26).
M. P. Allen and D. J. Tilde&y, Computer Simulation ofLiquids (Clarendon, Oxford, 1987). W. G. Hoover, Computational Statistical Mechanics, Studies in Modern Thermodynamics II (Elsevier, Amsterdam, 1991). L. Hill, Statistical Mechanics (McGraw-Hill, T. New York, 1956), p. 71. 4K. Huang, Statistical Mechanics, 2nd ed. (Wiley, New York, 1987), p. 140. Reference 1, p. 47. 6Reference 2, Chap. 3.4. L. Lebowitx, 3. K. Percus, and L. Verlet, Phys. Rev. 153,250 (1967). 5. M. Pearson, T. Halicioglu, and W. A. Tiller, Phys. Rev. A 32, 3030 E (1985). 9A. Miinster, Statistfsche Thermodynamik (Springer, Berlin, 1956), Chaps. 5.11 and 5.12. T. Cagin and J. R. Ray, Phys. Rev. A 37, 247 (1988). 5 M. Haile, Molecular Dynamics Simulation Elementary Methods (hey, New York, 1992). M. Litniewski, J. Phys. Chem. 94, 6472 (1990). l3 Reference 4, Chap. 6. 14C. G. Gray and K. E. Gubbins, Theory of Molecular Fluids Vol. I, Fundamentals (Clarendon, Oxford, 1984), Appendix 3A. 8. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and s1. Products (Academic, Orlando, 1980), Chap. 3.3-3.4. 16J. E. Mayer and M. Goeppert Mayer, Statistical Mechanics, 2nd ed. (Wiley, New York, 1977), p. 200. R. Becker, Theorie der Wiirme, 3rd ed. (Springer, Berlin, 1985), Chap. 36. More precisely, since, even though periodic boundary conditions are usually employed, the total number of degrees of freedom is small. lgThe choice of these state functions as basic quantities is also of importance for devising proper cutoff corrections (cf. part II of this series).

+ 2@t - 6~o&r~20~

(A12a)

~=no4-no,R,,-3n,n,,-3R,,n,, +6rIzo1~lln12+6~02 n:1.-6~01~~1. (A13)

J. Chem. Phys., Vol. 100, No. 4, 15 February 1994 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

Rolf Lustig: Statistical thermodynamics 2oReferknce 14, Appendix 3C. J. S. Rowlinson and I? L. Swinton, Liquids and Liquid Mixtures, 3rd ed. (Buttenvorths, London, 1982), Chap. 2. =R. Schmidt and W. Wagner, Fluid Phase Equilibria 19, 175 (1985). D. Frenkel, in Proceedings of the 97th International %nrico Fermi School of Physfcs, Varenna, 1985, edited by G. Ciccotti and W. G. Hoover (North-Holland, Amsterdam, 1985).

in the MD ensemble. I

3059

(a) B. Widom, J. Chem. Phys. 39,2808 (1963); (b) J. Phys. Chem. 86, 869 ( 1982). Since the formulas become increasingly lengthy, the development given here is restricted to order four. An extended list of results can be found in R. Lustig, Habilitation-thesis, RWTH, 1993. 26A Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (Dover, New York, 1970), Chap. 24.

J. Chem. Phys., Vol. 100, No. 4, 15 license 1994 Downloaded 12 Feb 2003 to 137.148.55.158. Redistribution subject to AIP February or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp