14 Auguq Il)<)5

2) 291
PHYSICS LETTERS A
's. Rev.
ELSEVIER Physics Letters A 204 (1995) 133-135

l) 759.
7.
. to be Ten1perature, least action, and Lagrangian mechanics
l.
843. William G. Hoover
Department ofApplied Science, Ulliuersity of California at Davis / Livermore alld Lawrence Licermore Nalional Laboratory,
LiueTlrJOre, CA 94551-7808, USA

Received 28 March 1995; revised manuscript received 2 June 1995; accepted for publication 2 June J995
Communicated by A.R. Bishop

Abstract

Temperature is considered from two different mechanical standpoints. The more approach uses Hamilton's least
action principle. From it we derive thermostatting forces identical to those found using Gauss' Principle. This approach
applies to equilibrium and nonequilibrium systems. The Lagrangian approach is different, and less useful, applying only at
equilibrium.

Keywords: Least action principle; Nonequilibrium; Lagrangian mechanics

1. Introduction When the context is clear we use just a single q or q

to represent whole sets.
Feynman repeatedly emphasized the generality of Lagrangian mechanics is particularly well-suited
Hamilton's principle of least action [1], the physical to the treatment of constrained systems. Constraints,
law from which both kinds of mechanics, classical used first to represent mechanical linkages and cou­
and quantum, follow. The principle states that the plings, have more recently been used in molecular
observed trajectory {qed} is that which minimizes simulations, where they maintain the shape of a
the action integral, polyatomic molecule and reduce the number of dy­
namical equations to be integrated. The usual ap­
ofL(q,q,t)dt of(K <P)dt=O, proach is to add a "Lagrange multiplier" to the
Lagrangian in such a way as to impose the constraint
where K is the kinetic energy, <P is the potential without affecting energy conservation [2,3]. For in­
energy, and L( q, q) K q) is the Lagrangian. stance, a particle with unit mass, constrained to
The integration time interval is fixed, as are also the circle the origin, at unit radius, can be described with
coordinate sets at both end points. Here we restrict the Lagrangian
our attention to equilibrium or nonequilibrium
many-body systems with kinetic energies quadratic L(x,y,x,y)
in the velocities and potential energies which depend
only on the coordinates {q} In this case, the Lagrangian equations of motion,

K= Lmq2j2, <P <P(q). (djdt)(aLjaq)q (aLjaq)q,

0375-960] /95/$09.50 1995 Elsevier Science B.Y. All rights reserved

SSDI 0375-960 1(95)00464-S
1:\4 WG. /lOOl'I'f / {,hysics LeU!'fS A 204 (iCj')5) /.!3-/35

multiplier? to insure that the solution is consistent

with the variation of the isokinetic constraint,
x= 2Ax, y 2Ay.
oQo(a/aqo)(m/2 d(2)[ (qo q _)2 - (q.! qu)2]
The value of the Lagrange multiplier A follows from
the second derivative of the constraint O.
Adding the two variational equations, and then tak­
ing the limit dt -7 0 gives the isokinetic differential
It is unnecessary to specify whether the Lagrange equation of motion,
multiplier depends upon time, upon space, or even (la)
space, velocity, and time simultaneously, because the
corresponding derivatives in the equations of motion The Lagrange multiplier ? can be identified, as in
would all be multiplied by the constraint 1, and the simple rigid-rotor example, by evaluating the
therefore vanish automatically. time derivative of the constraint. The result for the
In the next two sections we apply the two ideas Lagrange multiplier!: (which in this case has the
just described, Hamilton's principle and Lagrangian form of a "friction coefficient") is
constraints, to the more complicated case of con­ (lb)
straining temperature(s) in many-body systems [3,4].
As is usual in classical nonequilibrium statistical and provides familiar time-reversible constraint
mechanics, we define temperature [5] in terms of the forces. This result was already known to follow from
mean-squared kinetic energy of a typical Cartesian Gauss' principle of least constraint [8,9], but Hamil­
degree of freedom, ton's least action principle is more general, because
it promises also the possibility of extending the
kT=. < ). thermostat idea to nonequilibrium quantum systems.
In the following section we will see that a straight­
forward application of classical Lagrangian mechan­
ics to the same problem yields a less useful result.
2. Temperature and Hamilton's principle

Let us apply the least action principle to trajecto­

3. Temperature and Lagrangian mechanics
ries defined in the infinitesimal time interval be­
tween dt and + dl, considering all the coordinates
If we choose to use a Lagrange multiplier, A,
to be fixed at the endpoints, and with the set of
which varies with time, and can also depend upon
coordinates at the central time, {qo}, able to vary [6].
the coordinates and velocities, and use it to impose
The centered-difference approximation to Hamilton's
the constant-kinetic-energy constraint,
principle,

of L( q, q) d t = 0, the resulting equation of motion is

is (d/dt)[mq(l+A)] F(q)

oqo(a/aqo) {( m/2 dt 2 ) [( qo q)2 + (q + - qO)2] = mij(l + A) + Amq F(q).

Multiplying by it, and summing over degrees of
H <P _ + 2<1\ + <P +) } dt freedom, gives a simple differential equation for .A,

A.'[mq2 = 2A. K o F'q~ -$

oqo[(m/dt 2 )(2qo-q_-q+) +FoJ dt=O,
giving the usual StOrmer-Verlet equation of motion = .A=( E)/2K o ,
[6,7]. To add temperature we introduce a Lagrange where Eo is the energy at the initial time, t o.
 l\' NOlII'('/' / Physics LCI/ers:\ 2(N (191)5) /33-/35 135

istent Notice that the {illle deril'atil'<' of the Lagrange coefficients, {{I}' These same ideas apply to isobaric
multiplier, A, is <:xactly equal to the friction coeffi­ [12,13] or isoenergetic nonequilibrium algorithms.
cient ( found previously using the principle of leasl Because the least action principle applies to quantum
action. To make this identification explicit, we write systems too, there is potential for developing new
the present corresponding time-reversible Lagrangian methods for treating nonequilibrium quantum sys­
equa~ion of motion in terms of both these multipliers, tems [14]. Our specific isokinetic example illustrates
tak­
( A and A, the fundamental limitation of Lagrangian mechanics
;nlia1
mq=[F(q) (mq]/(1 A), to equilibrium.

(1a) (= LJ·q/E2K o.
as in Acknowledgement
The equation of 1110tion has an interesting form. For
:., the
small fluctuations in the energy, as for instance occur
I~ the in a large equilibrium system, the Lagrange multi­
I thank Brad Holian (Los Alamos), Dave Boer­
is the cker, Carol Hoover and Oyeon Kum (Livermore),
plier A is small, and the motion reduces to the
and Harald Posch (Vienna) for many useful conver­
Gaussian form (1) found above. On the other hand,
sations concerning the development of algorithms for
in a nonequilibrium steady-state system, the inte­
nonequilibrium simulations. Support of the staff at
grated energy change (which is mainly heat extracted
the Habitat (Netherlands Antilles), where a part of
by the thermos tatting constraint forces), and the La­
'[raint this work was completed, is gratefully acknowl­
from grange multiplier A, grow without bound, so that the
edged, as is also support at the Lawrence Livermore
ami!­ equations have no long-time solution.
National Laboratory under United States Department
cause Thus the Lagrange-multiplier approach, unlike the
of Energy Contract W-7405-Eng-48.
" the principle of least action, is not applicable to
~ems. nonequilibrium systems. This finding confirms Feyn­
ight­ man's emphasis of the fundamental nature of least
References
han­ action. Evidently classical Lagrangian mechanics is
ult. restricted to equilibrium problems in a way which [1] R. Feynrnan, The character of physical law (Modern Library,
the principle of least action is not. New York, 1994).
[2] G. Ciccotti and J.P. Ryckaert, Comput. Phys. Rep. 4 (1986)
345.
4. Summary [3] W.G. Hoover, Computational statistical mechanics (Elsevier,
Amsterdam, 1991).
r, A, [4] S. Nose, Prog. Theo. Phys. SUpD!. 103 (1991) 1.
,upon We have here generalized the basis of Gaussian [5] W.G. Hoover, RL. Holian and HA Posch, Phys. Rev. A 48
lpose isothermal mechanics by deriving the equations of (1993) 3196.
motion directly from Hamilton's least action princi­ [6] R.E. Gillilan and KR. Wilson, J. Chern. Phys. 97 (1992)
ple. This approach yields algorithms useful in simu­ 1757.
[7] D. Levesque and L. Verict, 1. Stat. Phys. 72 (1993) 519.
lating nonequilibrium systems too, by restricting the [8] W.G. Hoover, AJ.e. Ladd and R Moran, Phys. Rev. Lett.
number of degrees of freedom to be constrained. For 48 (J 982) 1818.
very recent summaries of nonequilibrium work see [9] DJ. Evans. W.G. Hoover, RH. Failor, B. Moran and A.J.e.
Refs. [lO,ltJ. In the simple case of nonequilibrium Ladd, Phys. Rev. A 28 (1983) 1016.
[10] W.G. Hoover, Physica A 194 (1993) 450.
many-body heat flow, a set of bulk particles obeys
[11] E.G.D. Cohen, Physica A 213 (1995) 293.
the usual Newton equations, while two reservoirs, [12] S. Nose, J. Chern. Phys. 81 (1984) 511.
es of one hot and one cold, obey the constrained Gauss [13] W.G. Hoover, Phys. Rev. A 34 (1986) 2499.
lr A, equations of motion (1) with two separate friction [14] D. Kusnezov, Phys. Lett. A 184 (1993) 50.

).