The Four Variational Principles of Mechanics

Annals of Physics PH5583
annals of physics 251, 125 (1996)

article no. 0104
The Four Variational Principles of Mechanics

C. G. Gray,* G. Karl, - and V. A. Novikov
GWP 2 and Department of Physics, University of Guelph, Guelph, Ontario N1G 2W1, Canada
Received October 10, 1995; revised March 4, 1996
We argue that there are four basic forms of the variational principles of mechanics:
Hamilton's least action principle (HP), the generalized Maupertuis principle (MP), and their
two reciprocal principles, RHP and RMP. This set is invariant under reciprocity and
Legendre transformations. One of these forms (HP) is in the literature: only special cases of
the other three are known. The generalized MP has a weaker constraint compared to the
traditional formulation: only the mean energy E is kept fixed between virtual paths. This
reformulation of MP alleviates several weaknesses of the old version. The reciprocal Mauper-
tuis principle (RMP) is the classical limit of Schrodinger's variational principle of quantum
mechanics, and this connection emphasizes the importance of the reciprocity transformation
for variational principles. Two unconstrained formulations (UHP and UMP) of these four
principles are also proposed, with completely specified Lagrange multipliers. Percival's varia-
tional principle for invariant tori and variational principles for scattering orbits are derived
from the RMP. The RMP is very convenient for approximate variational solutions to
problems in mechanics using Ritz type methods. Examples are provided. 1996 Academic
Press, Inc.
1. INTRODUCTION
The laws of mechanics have been known for a very long time [1, 2]. These laws
can be expressed either as differential equations (e.g. Newton's, Lagrange's or
Hamilton's) or as integral variational principles (i.e., Maupertuis' and Hamilton's).
Although all formulations of mechanics are mathematically equivalent, variational
principles have certain conceptual, practical, and esthetic merits. Some principles
are ``more equivalent than others.'' Since Maupertuis' and Hamilton's variational
principles are very old, one might suppose that they have already been expressed
in an optimal form. We find that this is not so, and in this paper we propose several
new principles starting with a generalization of Maupertuis' principle. The four
principles we discuss are an invariant set of variational laws of mechanics (under
Legendre and reciprocity transformations). The merit of these four principles is
underlined by a new and direct connection to the Schrodinger variational principle
of quantum mechanics.
* E-mail: cggphysics.uoguelph.ca.
-
E-mail: gkphysics.uoguelph.ca.

E-mail: novikovvxitep.itep.ru. On leave from ITEP, Moscow, Russia.
1
0003-491696 18.00
Copyright 1996 by Academic Press, Inc.
All rights of reproduction in any form reserved.
File: 595J 558301 . By:BV . Date:19:09:96 . Time:13:22 LOP8M. V8.0. Page 01:01
Codes: 4672 Signs: 2833 . Length: 51 pic 3 pts, 216 mm
2 GRAY, KARL, AND NOVIKOV
We start with the well known definitions needed to describe variational prin-
ciples, old and new. For simplicity of notation we assume a single degree of freedom
q, and momentum p=mq* , but the variational principles are true more generally. A
mechanical system can be described by a Lagrangian L(q, q* ) or by a Hamiltonian
H(q, p). The coordinate q(t) and its time derivative q* (t), or alternatively q(t), p(t),
describe the state of the system at time t. These functions q(t), q* (t) satisfy a differen-
tial equation which can be obtained from the Lagrangian or the Hamiltonian
[1, 2]. The functions L, H are in this simple case related to the kinetic energy K
and the potential energy V(q) through L=K&V and H=K+V, where K and V
are expressed in terms of the appropriate variables q, q* (or q, p). The transforma-
tion from q, q* , L to q, p, H is a Legendre transformation (see, e.g., Lanczos [2]),
where
H(q, p)= pq* &L(q, q* ).
Maupertuis [1] proposed in 1744 a global (integral) quantity, which he named

the action and which is ``least'' along the true path and ``greater'' for unphysical
``virtual'' paths. Maupertuis' definition of the action, and therefore his principle, was
a little vague, but this description was improved by Euler and Lagrange who took
the action as
qf t t
W= | qi
p dq= |
0
|
mq* 2 dt$= 2K dt$,
0
(1.0)
where an arbitrary path q(t$) (virtual or real) runs from an initial point q i #q(0)
to a final point q f #q(t). These endpoints q i , q f are always kept fixed, here and
throughout this paper, but the duration t is path dependent and therefore not fixed
in (1.0). We refer to W as the Maupertuis action, which is also known as the time-
independent action. Euler and Lagrange [1, 2] showed that for the true trajectory
W is stationary (not necessarily a minimum), provided the virtual trajectories q(t$)
are all restricted to having the same (fixed) energy E. Here H(q(t$), p(t$))=
K+V=E. We can therefore write Maupertuis' principle of least action in the form
($W) E =0, (1.1)
where $W denotes a first-order variation and the subscript E denotes that the
energy E is held fixed during the variation. Lagrange showed that the solutions
of (1.1) satisfy Newton's equations of motion, and conversely (see e.g. [1]).
Although very attractive as a concise expression of the Laws of Mechanics, the
principle (1.1) has several weaknesses. The weaknesses of (1.1) all stem from the
constraint of fixed energy, and are:
(A) Energy conservation is an assumption of (1.1) rather than a consequence
of it. This is expressed very nicely by Yourgrau and Mandelstam: `Ìn its original
form, as stated by Lagrange, the principle of least action suffers from the limitation
THE FOUR VARIATIONAL PRINCIPLES OF MECHANICS 3
that it applies only to virtual paths having the same energy as the real path.
Accordingly the principle depends in its formulation on the law of conservation of
energy, which must therefore be regarded as logically prior to it'' (ref. [1, p. 46]).
(B) For motion in one dimension, the constraint E = constant leaves very
little room for variation. At a given q, the value of q* is determined (within a minus
sign) by mq* 2 =2(E&V(q)). Therefore the only virtual trajectories allowed by (1.1)
in one dimension differ from the true trajectory by instantaneous velocity reversals.
Such paths are awkward to handle analytically.
(C) Even in higher dimensions the constraint E = constant is hard to imple-
ment in Ritz-type applications, as it is difficult to find virtual paths which satisfy the
constraint.
(D) The constraint of fixed energy E does not fix the duration t among dif-
ferent virtual trajectories in (1.1) and makes it cumbersome to convert (1.1) into a
differential equation for the trajectory.
These problems (AD) motivated Hamilton to propose, in 1832, a different
definition of the action and a better principle of least action. Hamilton's action S,
also called the time-dependent action, has the definition
t
S= |0
L(q(t$), q* (t$)) dt$, (1.2)
where the limits of integration t$=0, t$=t are kept fixed for all virtual paths
between fixed endpoints q i #q(0) and q f #q(t). Hamilton's principle of least action
(HP) can then be written in the form
($S) t =0, (1.3)
to denote that the action S is stationary under variations of path which maintain
the duration t fixed. In (1.3) time is constrained and energy unconstrained, the
reverse of the situation in (1.1). Because of the different constraint, HP does not
suffer from the weaknesses (AD). In particular the conversion of HP to a differen-
tial equation (Lagrange's equation) is rather simple, since there is no contribution
to the variation from the endtime.
It is therefore tempting to forget altogether about Maupertuis' action (1.0), and
his principle (1.1), as an inconvenient quirk of history, riddled with weaknesses. We
will argue in this paper that the problems (A, B, C) can be cured by maintaining
the definition (1.0) of the action W but relaxing the constraint of fixed energy E for
virtual paths. We allow a larger class of trial trajectories (or ``virtual'' paths) which
do not necessarily conserve the energy E. Instead of fixing the energy we keep the
mean energy E fixed. The mean energy is the time average of the Hamiltonian on
the trajectory q(t$):
1 t
E =( H) =
t | 0
H dt$. (1.4)
The old set of trial trajectories with fixed E=E is a subset of this larger set, so that
it is immediate that solutions of (1.1) are also solutions of
($W) E =0. (1.5)
We show in Section 2 that the converse is also true, namely the only solutions of
(1.5) are solutions of (1.1); there are no unphysical solutions of (1.5).
We refer to (1.5) as the Maupertuis Principle (MP) and occasionally, to avoid
ambiguity, as the generalized or reformulated MP. The principle (1.5) does not
have the limitations A, B, and C. But there is more merit in (1.5); it allows a
reciprocity transformation.
A reciprocity transformation converts a variational statement into a different one
with the same solution. Under reciprocity the functional to be varied is inter-
changed with the functional which is fixed (i.e., the constraint). Reciprocity can be
applied to MP (1.5) by interchanging the action W and mean energy E. The
Reciprocal Maupertuis principle (RMP) is a good and new principle of mechanics,
($E ) W =0. (1.6)
The RMP can also be called the principle of stationary mean energy at constant
action. Although reciprocity transformations occur often in thermodynamics we are
not aware of their previous use in mechanics.
The principle (1.6) is very useful not only in applications (see Section 3) but also
to make an immediate connection to the variational principle of quantum
mechanics. The RMP is the classical limit of the Schrodinger variational principle
of wave mechanics ($( | H |) ) n =0, with normalized , and n denoting the
stationary state of interest). This connection is plausible formally. At large n the
quantum expectation value becomes the classical mean energy. Keeping n fixed
corresponds classically to keeping the action W fixed in (1.6). This is described in
more detail and greater generality later. Our derivation of the classical variational
principles hinges on reciprocity, but does not require the use of path integrals, in
contrast to Feynman's well known derivation of the HP (see Ref. [17]).
A reciprocity transformation can also be applied to the HP, Eq. (1.3), to obtain
a Reciprocal Hamilton Principle (RHP), which is also a valid principle of
mechanics,
($t) S =0. (1.7)
Although not in the literature, the RHP reduces in special cases to old variational
principles such as Rayleigh's principle (see Appendix B).
The four variational principles MP (1.5), RMP (1.6), HP (1.3), and RHP (1.7)
form a complete set of principles of mechanics invariant under reciprocity and
Legendre transformations. The importance of reciprocity is underlined by the
connection to quantum mechanics which is now immediate. Formally the four
variational principles of mechanics are entirely analogous to four equilibrium prin-

ciples of thermodynamics which are also connected by reciprocity and Legendre
transformations (see e.g. Gibbs, [7]). Overlooking the reciprocity of the principles
of mechanics is analogous to restricting the equilibrium conditions of thermo-
dynamics to processes at constant energy and ignoring processes at constant
entropy.
The reciprocal pair of variational principles MP and RMP (and similarly
the pair HP and RHP) can also be stated in unconstrained form: one finds a
Maupertuis version (UMP),
$W=t $E, (1.8)
and similarly a Hamilton version (UHP),
$S=&E $t. (1.9)
In (1.8) the time t is the duration of the true trajectory, which is a constant
Lagrange multipler. Similarly, E in (1.9) is a Lagrange multiplier, the energy of the
true trajectory. These equations are derived in the next section.
We have failed to find any mention of the RMP, generalized MP, RHP, UMP,
and UHP in the literature, a situation which this paper is meant to remedy. There
is a near miss, by Schrodinger, in his first paper on wave mechanics [5], where he
tries to derive the variational principle of wave mechanics from something like the
RMP, but in the second paper this first derivation is abandoned as `ìncomprehen-
sible''. For the important special class of periodic and quasi-periodic motions,
the RMP is equivalent to Percival's variational principle for invariant tori [4].
But Percival did not note the greater generality of the RMP (for example, its
applicability to segments of a trajectory or to chaotic trajectories), nor its connec-
tion to the MP. The same criticism can be addressed to the authors of Ref. [3].
It is interesting to note in connection with Percival's principle that Klein and
collaborators have derived this principle of mechanics directly from matrix
mechanics [30].
Finally we give a brief outline of this paper. In Section 2 we give the derivation
of MP, RMP, RHP, UMP, and UHP starting from the very familiar Hamilton
principle. The unconstrained forms are derived first, and the constraints are then
applied. In Section 3 we apply the RMP to solve (approximately) problems in
mechanics using Ritz-type methods: plane pendulum (3.1), scattering by an r &4
potential (3.2), and a two-dimensional, anisotropic quartic potential (3.3). The
reader may wish to read this section first as it is less formal. Section 4 makes con-
nections to other variational principles, i.e., Schrodinger's variational principle of
quantum mechanics (4.1), Percival's variational principle for invariant tori (4.2),
and new variational principles for scattering orbits (4.3). Section 5 contains a
summary. Appendix A gives a direct derivation of the reformulated MP, and
Appendix B discusses relations in the literature connected to the RHP.
2. FOUR EQUIVALENT VARIATIONAL PRINCIPLES
We start by proving the equivalence of the reformulated MP (1.5) and its

reciprocal (RMP) (1.6) to the familiar HP (1.3). This constitutes therefore a proof
of the validity of (1.5), (1.6). We assume again for simplicity one degree of freedom.
The starting point is the relation between the Lagrangian L and Hamiltonian H,
L(q, q* )= pq* &H(q, p), (2.1)
which we integrate over some arbitrary time interval (0, t) along some trial trajec-
tory q(t$), p(t$)=mq* (t$), connecting fixed endpoints, to obtain
t t t
|0
dt$L= | 0
dt$pq* & | 0
dt$H. (2.2)
Using the definitions of S(1.2), W(1.0), and E(1.4), we can rewrite (2.2) as
S=W&Et. (2.3)
For real trajectories, where E =E=const., (2.3) is well known [6]. Taking an
arbitrary variation of our trial trajectory q(t$) q(t$)+$q(t$), with fixed endpoints,
as we assume throughout, and an arbitrary variation of our endtime t t+$t, we
have to first order in $q(t$) and $t:
$S=$W&t $E &E $t, (2.4)
or $S+E $t=$W&t $E. We now show that the two sides of this general kinematic
relation each vanish when we consider variations around a true trajectory, i.e.,
$S=&E $t, $W=t $E, (2.5)
where we have used E =E=const. on a true trajectory in the first of (2.5). Further
discussion of these unconstrained versions of the variational principles is given
later.
To derive the first of the relations (2.5), we recall the HP (1.3)
($S) t =0, (2.6)
which is valid for true and only true trajectories, and where t is fixed. This relation
implies also that for variations around a true trajectory where both S and t vary
we must have
$S=* $t, (2.7)
where * is some (Lagrange) multiplier, a property of the true trajectory with dimen-
sion of energy. If (2.7) were not valid, so that $S=* $t+[terms dependent on
$q(t$)], where we recall that $t and $q(t$) are independent, we could have a situa-
tion where $t=0 but $S{0, violating (2.6). To see that *=&E, we specialize to
the case of variations between two true trajectories with endtimes t and t+dt. In
this case (2.7) reads St=*, and hence *=&E in order to reproduce the well
known relation St=&E. This completes the proof of the first, and hence of the
second, of the relations (2.5).
We now specialize the unconstrained relations (2.5) by applying constraints. If,
in the first of (2.5), we take the constraint of fixed t (i.e., $t=0), we regain the HP
(2.6). If, on the other hand, we fix S, i.e., set $S=0, we get a RHP (1.6) (see also
Appendix B)
($t) S =0. (2.8)
Turning to the second of the unconstrained relations (2.5), if we fix E, i.e., set
$E =0, we get our reformulated MP (1.5),
($W) E =0, (2.9)
whereas for fixed W (i.e., $W=0) we get an RMP (1.6),
($E ) W =0. (2.10)
This constitutes an abstract proof that the MP and RMP are ``good'' variational
principles of classical dynamics. A more detailed, pedestrian proof of these prin-
ciples by traditional variational procedures is given in Appendix A.
We thus have an economical derivation of three other variational principles
starting from the HP. The proof described above is reversible. We may assume the
MP (2.9), and then deduce the second of (2.5), and hence the first, leading to the
RMP, HP, and RHP as special cases. The argument we are using is an adaptation
to (2.3) of Gibbs' familiar argument in thermodynamics [7], when discussing the
Legendre transform relation of free energy, energy, temperature, and entropy, for
example. It is clear that the four principles (1.3), (1.5), (1.6), and (1.7) are equiv-
alent. However, this does not mean that they are equally useful for solving
particular problems. Of the four, the HP (1.3) is most suitable for deriving the
(Lagrange or Hamilton) equations of motion; the constraint $t=0 is very well
suited to convert the definite integral over time into a differential equation. The
RMP (1.5) makes a smooth connection to the Quantum Mechanical Variational
Principle. We also find that the RMP is well suited to solve approximately classical
problems by a procedure which is very similar to that used in the variational
method of solution of quantum problems. Some examples are given in the next sec-
tion. For general potentials, the MP itself (1.5) is useful to determine the shape of
the trajectory, via (1.1) which is a special case, and can be rewritten in Jacobi's
form ([11, p. 142]; Lanczos [2, p. 135]); for central potentials, the new formula-
tions are more useful for the determination of orbit shapes (see Section 4.3),
especially for scattering orbits.
It is also clear that E=const. is a consequence of (1.5) or (1.6), rather than an

input as in (1.1), since these variational principles all imply the equations of
motion. A more direct argument can also be given. We further note that (2.5) can
be written as unconstrained Maupertuis (UMP) and Hamilton (UHP) principles,
$(W&tE )=0 and $(S+Et)=0, respectively, where t and E are constant Lagrange
multipliers in the first and second relation, respectively. For central forces and
some other problems these unconstrained forms are often simplest to apply; see
Sections 4.2 and 4.3.
Finally, we note that the variations in (2.5) can be specialized to describe
the changes between two real trajectories, in which case we obtain the familiar
equations E=&St and t=WE, the first of which was used above. These
relations, and similar ones involving spatial derivatives obtained by considering
endpoint variations, are expressions of what Hamilton termed laws of varying
action [1, 2].
3. EXAMPLES
Because the variational principles are equivalent to equations of motion, they can
be used directly to find approximate trajectories of all types, i.e., periodic, quasi-
periodic, chaotic, scattering, and arbitrary segments of arbitrary trajectories. The
HP has often been used in the direct mode [8] in continuum mechanics and
electromagnetism [9] (e.g., the RayleighRitz method), and occasionally in particle
mechanics [10]. Some simple examples illustrating the various principles follow.
The similarity to the procedure commonly used in quantum mechanics will be
apparent.
The simplest example for a segment of trajectory is the motion of a free particle
between two points along a straight line. A simple trial trajectory has constant
velocity v 1 before the midpoint and constant velocity v 2 after the midpoint. The
mean energy E is proportional to v 1 v 2 and the action W is proportional to
(v 1 +v 2 ). It is trivial to check that the reformulated MP and RMP demand v 1 =v 2 .
Note that the trial trajectory v 1 {v 2 is not allowed by the traditional Maupertuis
principle as it violates the constraint of fixed energy E. This is the case in all
one-dimensional problems, where the constraint of fixed energy allows only trial
trajectories with instantaneous velocity reversals obtained from the true trajectory.
The new formulation is much less restrictive and allows many more virtual paths
even in one dimension, which is an advantage. This very simple example also
provides a good testing ground for the other new variational principles introduced
in Section 2, i.e., the RHP (2.8), and the unconstrained Maupertuis and Hamilton
principles (2.5).
We shall consider examples of periodic motion, scattering, and chaotic motion
below. A nice example of a quasi-periodic motion is that of the spherical pendulum
which we have solved variationally, but the solution is sufficiently intricate that we
do not include it here [13].
3.1. The Plane Pendulum

The Hamiltonian for the plane pendulum is
H= 12 p 23 +| 2o(1&cos 3), (3.1.1)
where | o =( gL) 12 is the frequency in the quadratic approximation to the cosine,
when the period is independent of amplitude 3 max , and p 3 =34 is the (angular)
momentum conjugate to 3. We choose units where m=1 and L=1, where m
and L are the bob mass and pendulum length respectively. We wish to estimate
the period T(3 max ) in the next approximation when the potential energy is
| 2o[(3 22)&(3 424)].
Consider first the anharmonic oscillator with Hamiltonian
H= 12 p 2 + 12 | 20 q 2 + 14 *q 4. (3.1.2)
The pendulum is a special case with *=&| 20 6. Take as the trial trajectory
q(t)=A sin |t, which satisfies the equation of motion for *=0 and the conditions
q=0 at the beginning and end of a cycle. We can compute the action W in terms
of A, |, and then the mean energy E in terms of A, | (or W, |):
T
W= | 0
q* 2 dt=A 2?| (3.1.3)
and
1 T W | 2 3*W
E =
T |0
H dt=
4? \
|+ 0 +
| 8?| 2
.
+ (3.1.4)
Using the RMP, treating | as a variational parameter, we obtain from (E|) W

=0 the condition
3*W 3*
| 2 =| 20 + =| 20 + A 2. (3.1.5)
4?| 4
Note that for *=0, | equals | 0 as it should, and the period T is independent of
the amplitude A, the action W, and the energy E. For nonzero *, which we set
equal to &| 20 6, to return to the pendulum, we have (with A=3 max )
A2 12
A2 A4
|=| 0 1&
_ 8 & =| 0 1&
_ &
16 512 &
+ }}} , (3.1.6)
and the period T is
A2 &12
A 2 3A 4
T(A)=T 0 1&
_ 8 & =T 0 1+
_ +
16 512
+ }}} ,
& (3.1.7)
where T 0 =2?| 0 . The result (3.1.7) is correct to order A 2, as one can compute T
directly through an elliptic integral [11] and the corresponding expansion
A 2 11 3A 4
T exact(A)=T 0 1+
_ +
16 18 512
+ }}} .
& (3.1.8)
The O(A 2 ) accuracy is expected since the trial trajectory is correct to zeroth order
in *, and the variational principle ensures that first-order errors vanish, as in quan-
tum mechanics. The coefficient of A 4 in (3.1.7) is too large compared to that of the
exact result (3.1.8). One could improve the variational result by taking a more
elaborate trial trajectory, with more parameters, and expanding (3.1.1) to higher
order in 3, but that is beyond the scope of this paper (see Ref. 13).
3.2. Scattering by an r &4 Potential

The Maupertuis and Hamilton principles and their reciprocals are put in con-
venient forms for scattering problems in Section 4.3. Here, we consider a simple
repulsive central potential V(r)=Cr 4, C>0, and want to estimate variationally,
using the RMP, the scattering angle 3 as a function of impact parameter b
and incident energy E. The orbit shape is described by r(,) or equivalently
u(,)=1r(,), where , is the polar angle in the scattering plane measured from the
incident asymptotic direction. At ,=0, r takes its largest value, and u its smallest
value. The projectile starts at u=0 (r=) corresponding to ,=0 and returns to
u=0 (r=) at ,=, max , with u0 throughout. The scattering angle 3 is related
to , max by 3=?&, max . The angular momentum l=(2mE) 12b is a constant of
the motion and can be held fixed in all orbit variations. As discussed also in
Section 4.3, the Hamiltonian for fixed l can be written as
l2
H= (u$ 2 +u 2 )+V(u), (3.2.1)
2m
where u$(,)=dud,. We note that, for V=Cu 4, the Hamiltonian (3.2.1) has almost
the same form as that of the anharmonic oscillator (3.1.2). To make the analogy
even closer we introduce the new variable U=(lm 12 ) u, so that (3.2.1) becomes
H= 12 U$ 2 + 12 U 2 + 14 *U 4, (3.2.2)
where *=4Cm 2l 4, and this is exactly of the form (3.1.2) with | 0 being set to unity.
Note that the angle , in (3.2.2) plays the role of the time t in (3.1.2); since u runs
through half a cycle (0 to 0, always positive) as t runs from 0 to T2, it follows that
2, max is the period T and we can rewrite the scattering angle in the form
2?
23=2?&T=2?& . (3.2.3)
|
In particular, with no potential, i.e., *=0 in (3.2.2), |=| 0 =1 and the scattering
angle vanishes, as it should. Therefore to evaluate the scattering angle 3 we need
to estimate |(E ) in (3.2.3). Referring to Section 3.1, we substitute W from
Eq. (3.1.5) into (3.1.4) to obtain
6*E =(| 2 &1)(3| 2 +1) (3.2.4)
with solution
| 2 = 13 [1+2(1+ 92 *E ) 12 ]. (3.2.5)
The scattering angle 3 can be obtained from (3.2.3) and (3.2.5),
12
3 3
?
=1&
_
1+2(1+ 92 *E ) 12 & . (3.2.6)
If we take for the mean energy E the incident energy E (which is true rigorously
for the exact orbit, but only approximately for our optimized trial orbit), the
dimensionless parameter *E will equal CEb 4. With this value of *E the scattering
angle (3.2.6) agrees with exact numerical calculations [14] to within 15 minutes of
arc, over the whole range of the parameter *E >0. The formula (3.2.6) can be con-
tinued to *E <0, as long as the square root is real, i.e., up to |*E | <0.222. For
|*E | >0.222, 3 becomes complex, as an indication of capture of the incident par-
ticle by the potential. In reality, capture should begin for an attractive potential
only when |*E | >0.25. For the attractive case (3.2.6) remains reasonably accurate
for small values of |*E |, corresponding to small scattering angles 3. We find for
C<0 that by 3t30% the error of (3.2.6) has grown to about one degree of arc.
One can obtain more accurate estimates 3 using more flexible trial orbits in (3.2.2)
or (3.1.2), but this is beyond the scope of this paper.
This variational method of estimating the scattering angle 3 as a function of
impact parameter and energy is quite general for central potentials which are even
powers of r (or u); however, the algebraic equations for | 2 need not be quadratic
as in (3.2.5). For potentials which contain odd powers of r (e.g., the Coulomb
potential), the scattering-oscillator analogy becomes more complicated, since the
condition u>0 must be enforced in this case. In place of (3.2.3), we can always
return to the basic relation 3=?&, max and compute , max using the variational
equations of Section 4.3.
3.3. The Anisotropic 2D Quartic Oscillator: A Chaotic System

Numerical evidence [15] suggests that the 2D oscillator with Hamiltonian
2
p 2x p y
H= + +Cx 2y 2, (3.3.1)
2m 2m
is a nonintegrable system and that the vast majority of trajectories is chaotic. To

keep this example simple and illustrative, our limited objective here is to represent
an arbitrary trajectory, of specified action W over a long time T, by an optimized
quasi-periodic trajectory so that quantization semi-classically via the Einstein
BrillouinKeller (EBK) rule becomes a simple procedure.
As a trial trajectory, we use the quasi-periodic 2D anisotropic harmonic
oscillator trajectory
x(t)=A x cos | x t, y(t)=A y cos | y t, (3.3.2)
where | x and | y are in general incommensurate. The action W and mean energy
E over a long time T are found to be
T
W= (| x W x +| y W y ), (3.3.3)
2?
where W k =?m| k A 2k for k=x, y are the one-cycle actions for the x and y motions,
and
m 2 2 C 1 CW x W y
E =
4
(| x A x +| 2y A 2y )+ A 2x A 2y =
4 4? \
| x W x +| y W y + 2
?m | x | y
. (3.3.4)
+
We now apply the RMP, ($E ) W =0, to (3.3.4). As explained in Section 4.2, for
quasi-periodic trajectories, extremizing E at fixed W is equivalent to extremizing at
fixed mean subactions W x and W y . For the simple trial trajectory (3.3.2), we have
W x =W x and W y =W y . Treating | x and | y as variational parameters, and setting
(E| x ) Wx , Wy =0 and (E| y ) Wx , Wy =0, we find
2
C Wy C W 2x
| 3x = , | 3y = , (3.3.5)
?m 2 W x ?m 2 W y
and then combining (3.3.5) and (3.3.4) gives

13
3 C
E(W x , W y )=
4? ?m 2\ + (W x W y ) 23. (3.3.6)
Using the adiabatic approximation method, Martens et al. [16] found an expres-
sion of the same form as (3.3.6), but with a numerical coefficient (316?) 23 which
is smaller than our coefficient (34? 34 ) by a factor of 0.93.
To quantize the energies semi-classically, we simply replace the W k by (n k +12)h
in (3.3.6) following the EBK rule [18], giving
3 Ch 4 13
1 23
1 23
E nx ny =
\ + \
4? ?m 2
nx+
2 + \ ny+
2 + . (3.3.7)
When compared with the available exact results, i.e., the lowest 50 energy levels
obtained by a difficult numerical calculation [16], (3.3.7) is found to be accurate
to within 50. Thus, even for nonintegrable systems, approximate analytic results
can be obtained simply by the RMP.
4. CONNECTIONS TO OTHER VARIATIONAL PRINCIPLES
In earlier sections, we discussed the equivalence of the reformulated MP and

RMP to the original MP and to the HP and RHP. Here we note some connections
to other variational principles, some old and some new.
4.1. Schrodinger's Quantum Variational Principle

Quantum mechanics is a starting point for the derivation of classical mechanics.
One expects therefore to derive the variational principles of classical mechanics
from the variational principles of quantum mechanics. Hamilton's Principle was
derived by Feynman [17] using the path integral formalism of quantum mechanics.
We use the old, Schrodinger version of quantum mechanics to derive the reciprocal
Maupertuis principle of classical mechanics, introduced earlier. Other variational
routes to the classical limit are also known, for example, via matrix mechanics
[30].
The details of reaching the limit of classical mechanics from the quantum domain
are notoriously delicate and difficult. Much of the progress is recent, and for guidance
we rely on various reviews [18]. The simplest case is that of periodic motion in one
dimension, which we discuss first. We start by recalling that Schrodinger's variational
principle of wave mechanics has the form $[( n | H op | n )( n | n ) ]=0 where the
quantum system has a Hamiltonian operator H op which is a function of the
(operators) q op , p op . Although this principle is employed most often to find the
ground state o of the quantum system, the principle applies to all eigenstates
of the operator H op . This point is elaborated in detail in [3], where more references
can be found. We want to demonstrate that at large quantum numbers n, the
Schrodinger principle turns into the RMP, i.e., the following relation holds:
( | H op |) 1 T
0= $
\ ( | ) + n
n>
>1
ww $
\ T | 0
H(q, p) dt
+ W
#($E ) W =0, (4.1.1)
where E and W are defined in the Introduction, and T is the period of the motion.
The connection appears very plausible and a brief argument was presented [3]
which is elaborated here.
On the RHS of (4.1.1) the Hamiltonian H(q, p) is the classical counterpart of the
quantum Hamiltonian H(q op , p op ). On the LHS of (4.1.1) the quantum variation
is made at large n, with n fixed [3]. The state n(q) corresponds to a classical
(periodic) trajectory with precisely the same energy E n . The meaning of (4.1.1) is
that one and the same trajectory q(t), p(t) is obtained from the classical variational
principle, and from the classical limit of the solution n(q) of the quantum variational
principle. Moreover, we restrict even the trial wavefunctions and trial trajectories so
that they also match each other.
When we use a WKB representation for all trial wavefunctions on the LHS of
(4.1.1) we have * dq = const qqmaxmin
dqv=const T0 dt = const T, where v is the
velocity and T is the period of motion. In the same approximation the numerator
of the LHS (4.1.1) becomes const T0 H(q, p) dt, with the same constant as the
denominator. Therefore the quantum expectation on the LHS becomes a classical
time average on the RHS of (4.1.1) if we have WKB wavefunctions for n . This of
course is well known.
All that remains to be discussed is how the constraint of fixed W arises on the
RHS of (4.1.1). We recall that quantization means that only certain classical
energies possible on the RHS match quantum energies E n on the LHS. For periodic
motion in one dimension, the constraint on the energy for an allowed quantum
state n, derived in the WKB approximation, is for large n,
qmax
W(cycle)= p dq=2 | qmin
- 2m(E n &V(q)) dq=nh. (4.1.2)
This is the BohrSommerfeldWilson quantization rule, which for n fixed means

precisely that the action W is to be kept fixed on the RHS of (4.1.1), if n is fixed
on the left.
This completes the demonstration of the transmutation of the Schrodinger
Variational Principle into the RMP for periodic motion in one dimension. We
summarize the two main steps of the argument:
(a) quantum expectation value turns into classical time average,

(b) quantum constraint of fixed n turns into classical constraint of fixed W.
One can generalize the argument without much difficulty for periodic or
quasiperiodic motion with a separable (or more generally integrable) Hamiltonian.
Step (a) is part of the general lore of quantum mechanics [20] and depends again
on finding matching trial wavefunctions and classical trajectories, which for a
separable Hamiltonian, for example, can be carried out separately for each of the
parts. Thus again one can show that the appropriate quantum average can be
rewritten as a classical time average, where T in the quasi-periodic case (see
Section 4.2). As to step (b), the constraint in (4.1.1), the relevant change is from the
BohrSommerfeldWilson quantization condition to the EinsteinBrillouinKeller
quantization rule [18] for large quantum numbers,
W k =n k h, (4.1.3)
where W k is the subaction (4.2.3) for the angle coordinate 3 k of the dynamical torus.
For the varied tori, the mean actions W k are held fixed and equal to the W k (see
Section 4.2). Hence, fixed [n k ] on the LHS of (4.1.1) correspond to fixed [W k ] on the
RHS. As discussed in Section 4.2, variation of E at fixed [W k ] is equivalent to varia-
tion at fixed W. This completes the demonstration of (4.1.1) for the quasi-periodic
case. For this case, Percival [18] noted the analogy between $(| H op |) =0 and
$( H) 3 =0, where ( H) 3 is the classical phase average (4.2.1).
Another relatively straightforward case is 1D scattering (or 3D scattering in
a central potential). On the LHS of (4.1.1) we use stationary state WKB wave
functions over the 1D interval q=(0, L), with L large (eventually L ). On the
RHS we take T large (eventually T ). In the same way as for bound states we
find again, formally, (| H op |)( | ) T0 H dtT. On the LHS, the WKB
phase of at large qtL will be W(orbit), where W(orbit) is the classical action
for the complete scattering orbit. Hence a fixed (and large) W(orbit) on the LHS
corresponds to a fixed W(orbit) on the RHS. Thus both the averages and
constraints again correspond, formally, for the two sides of (4.1.1). (In practice, as
discussed in Section 4.3, because the two sides of (4.1.1) contain divergent quan-
tities in the limits L and T , they are reformulated in terms of manifestly
finite functionals.)
The difficult case is nonintegrable systems, for which the classical motion can be
chaotic and the quantum motion has corresponding signatures [18, 21]. For the
quasiperiodic motions of nonintegrable systems, the argument for integrable
systems will go through. One would expect that (4.1.1) holds also for chaotic
motions, but we are unable to give a direct proof. For ergodic motions, the result
can be made plausible using the adiabatic theorem, or using Berry's [18] conjec-
tured semiclassical wavefunctions. An indirect plausibility argument, using the
Schrodinger equation of motion resulting from the EulerLagrange equation of
the LHS, and Newton's equation of motion resulting from the EulerLagrange
equation of the RHS, can be readily constructed.
By a similar but slightly more involved argument we can derive the classical limit
for the other version [5, 19] of the Schrodinger variational principle, i.e.,
$( | (H op &E) |) =0, where E is here a constant Lagrange multiplier introduced
to relax the constraint of fixed ( | ); it is equal to the energy of an extremizing
, which is an eigenstate of H op . We find the classical limit statement
$[(E &E)T] W =0, which is equivalent to the UHP $(S+ET )=0.
Historically, it is interesting to note that Schrodinger, in his first paper [5] on
wave mechanics, wrote down the equivalent of $( | H op |) =0 using a heuristic
classical argument as motivation, which, with the benefit of hindsight, resembles
$E =0. He proceeded to obtain the Schrodinger equation from the quantum varia-
tional principle in the way which is now standard. In his second paper [5] on wave
mechanics, Schrodinger describes his heuristic argument as `ìncomprehensible'' and
presents a second basis for the Schrodinger equation, used now in standard texts,
based on the analogy between geometric and wave optics on the one hand, and
particle and wave mechanics on the other.
4.2. Percival's Variational Principle for Invariant Tori

We consider a quasi-periodic trajectory of some system (e.g., a 2D harmonic
oscillator with incommensurate frequencies). For integrable systems, all bounded
motions are quasi-periodic or periodic, whereas for nonintegrable systems only a
finite fraction of the bounded motions are quasi-periodic or periodic [22]; the rest
are chaotic. A quasiperiodic motion is confined to a torus in phase space (e.g., a
2D) torus in 4D phase space for the 2D oscillator example above).
Percival [4] has derived a variational principle for these tori. We show that his
principle can be derived as a special case of the RMP. We consider the initial and
final points of the trajectory to be close together, and the trajectory very long
(i.e., the time T eventually). For a true trajectory, the trajectory winds around
the torus with uniform density (for T ), so that time averages can be replaced
by phase averages, e.g.,
1 T d3 N
E = lim
T T
|0
dtH= | (2?) N
H(q (3 k ), p (3 k )), (4.2.1)
where q #(q 1 } } } q N ), p #( p 1 } } } p N ), N is the number of degrees of freedom,

d3 N #d3 1 } } } d3 N , with 3 k the N angle variables parametrizing the torus,
q =q (3 k ) and p = p (3 k ). The torus is determined by the set of corresponding
actions W k , which are constants of the motion.
The action W for a long true trajectory can be written in the following simple
form (for the terms linear in T )
W=: N k W k , (4.2.2)
k
where N k =& k T is the winding number for angle 3 k , & k =| k 2?=34 k 2?=EW k
is the frequency, and W k is the action conjugate to 3 k 2?, i.e.,
q
Wk= Ck
p } dq = Ck
d3 k p }
3 k
, (4.2.3)
where p } dq = i p i dq i ; and C k is a circuit in 3 k at fixed 3 k${k . (For an invariant

torus W k does not depend on the value of 3 k${k .)
Consider now an arbitrary small deformation of a given invariant torus and a
trial trajectory on this trial torus, which is close to the true trajectory for all times
t and has the same end points as the true one. (The total time can be different,
$T{0, but $TTR1.) If we take the angle variables 3 k as coordinates, we can
express the usual condition for small variations $q i (t) as
$3 k(t)R1. (4.2.4)
Since the endpoints for the real and trial trajectories are the same, we have equal
winding numbers N k for the two trajectories, since for all k we have
(3 k ) T =(3 k +$3 k ) T+$T =2?N k , (4.2.5)
where (3 k ) T #3 k(t=T). The real trajectory covers the invariant torus with
uniform density for T . The condition (4.2.4) allows us to choose the trial
trajectory to cover the trial torus uniformly as well. In this case the action W for
the trial trajectory is
W=: N k W k , (4.2.6)
k
where W k is the mean action W k on the trial torus,
d3 N&1 d3 N q
W k = | (2?) N&1
W k =2? | p }
(2?) N 3 k
, (4.2.7)
where d3 N&1 denotes integration over all 3 k$ except 3 k . Note that for a trial torus
the W k do depend on 3 k${k and only the mean value W k can be compared with
W k on the dynamical torus (where E =E and W k =W k ).
We are now in a position to apply the unconstrained version of the Maupertuis
principles, the second of Eqs. (2.5), for the considered varied paths. It reads as
1
$E & $W=0, (4.2.8)
T
where T is the Lagrange multiplier, equal to the total time for the true trajectory.
Since the N k are the same for varied trajectories we have from (4.2.6)
$W=: N k $W k ,
k
and (4.2.8) can then be rewritten as
Nk
0=$E &:
k
\ T + $W =$E&: & $W
k
k
k k
or
\
$ E &: & k W k =0,
k
+ (4.2.9)
where & k are the frequencies for the invariant torus. Equation (4.2.9) is Percival's
[4] variational principle for invariant tori; the mean energy E is extremized, with
the set of mean actions W k held fixed. The constant Lagrange multipliers & k are the
invariant torus frequencies. Equation (4.2.9) is the unconstrained form. The explicit
constrained form is ($E ) [Wk ] =0.
If we note that, with the & k not held fixed in the product term k & k W k so that
$ : & k W k =: $& k W k +: & k $W k ,

k k k
and if we recall WT= k & k W k (cf. (4.2.6)) and E =WT&ST=WT&L

(cf.(2.3)), where L is the mean Lagrangian, we can Legendre transform (4.2.9) to
the form $L & k W k $& k =0 or
\
$ L &: W k & k =0,
k
+ (4.2.10)
where we have used the fact that W k =W k on the invariant torus. Percival [23]
(and independently Klein and Li [24]) has also derived the fixed-frequency form
($L ) [&k ] =0; i.e., L is extremized with the set of frequencies & k held fixed. In the
unconstrained form (4.2.10), the constant Lagrange multipliers W k are the invariant
torus actions. As discussed in Appendix B, less complete mean Lagrangian varia-
tional principles have also been given in the literature, and we state there a general
mean Lagrangian principle, not restricted to periodic or quasi-periodic motions.
4.3. Variational Principles for Scattering Orbits

The Maupertuis and Hamilton principles, their reciprocals, and their uncon-
strained versions apply to arbitrary segments of any trajectory, and hence in
particular to a complete scattering trajectory. Application to scattering does require
some care, however, because in the limit t , W and S diverge and E approaches
a trivial value, i.e., the initial (kinetic) energy, for all trial orbits. Since scattering
orbits are dominated by the asymptotic free particle parts, we reformulate the varia-
tional principles in forms which deal only with the ``sub-asymptotic,'' or interaction
parts of the orbits, which lead to manifestly finite functionals.
We restrict ourselves to scattering of one particle by a central potential V(r). In
terms of the planar polar coordinates (r, ,) of the particle in the orbit plane, the
Hamiltonian H=K+V is
p 2r l2
H= + +V(r), (4.3.1)
2m 2mr 2
where l=mr 2,4 is the angular momentum. Rewriting (4.3.1) in terms of the variable
u=1r often introduced for central force problems, we have
l2
H= (u$ 2 +u 2 )+V(u), (4.3.2)
2m
where u$=dud,. The orbit shape is described by r(,) or u(,).
We first apply a variational principle to a finite segment of an orbit connecting

points (r 1 , , 1 ) and (r 2 , , 2 ). Later we shall take the limits r 1 , r 2 . It is easiest
to use one of the general relations (2.5), which allow unconstrained variations
around a true orbit. Applying the first of (2.5) and converting t-integrals to
,-integrals using dt=(ml ) d,u 2, we have
,2 l2 d,
$ |
,1 _ 2m
(u$ 2 +u 2 )&V+E 2 =0
u & (4.3.3)
where the end points (u 1 , , 1 ) and (u 2 , , 2 ) are fixed as always, and the Lagrange
multiplier E is the energy of the true orbit. As is well known, because the potential
is central, l is a constant of the motion, which can be derived from the variational
principles from the rotational invariance of H or L (Noether's theorem [25]). We
can therefore restrict the variations in (4.3.3) to those with fixed l.
The difficulty discussed above is apparent in (4.3.3); for scattering orbits, the
regions where u 0 (i.e., r ) cause the integral to diverge. We can, however,
transform (4.3.3) to the equivalent divergence-free form
,2 l2
$ | ,1 _ 2m &
(u$ 2 &u 2 )&V d,=0 (4.3.4)
A straightforward way to show the equivalence of (4.3.3) and (4.3.4) is to show that
both yield the same EulerLagrange differential equation for the orbit, i.e.,
(l 2m)(u"+u)+dVdu=0. A more elegant approach is to recognize that, for con-
stant l, the effective radial potential V l in (4.3.2) is V l =(l 22m)u 2 +V(u), so that
Hamilton's principle, with , playing the role of t, can be applied to the functional
Sl [u(,)], where S l = L l d,, with L l =(l 22m)u$ 2 &V l , which is (4.3.4).
Equation (4.3.4) can be written in the abbreviated notation used earlier,
($S l ) 2, =0, (4.3.5)
where 2,=, 2 &, 1 , which corresponds to the HP (1.4). Similarly, in corre-

spondence with (1.5), (1.2), and (1.3), we can write three more variational
principles for orbits in a central potential,
($2,) Sl =0, ($W l ) El =0, ($E l ) Wl =0, (4.3.6)
where W l =(l 2m) u$ 2 d, and E l =( H l ) #(2,) &1 [(l 22m) u$ 2 +V l ] d,. More
general relations involving unconstrained variations, corresponding to (2.5), can
also be written, i.e.,
$S l =&E l $2,, $W l =2, $E l (4.3.7)
where E l =E l =E for a true path.

Equations (4.3.4)(4.3.7) are general, and permit variational estimates of both
bound and scattering orbit shapes u(,) in a central potential. If we now consider
a complete scattering orbit, with , measured from the direction of incidence, (4.3.4)
becomes
,max l2
$ | 0 _ 2m &
(u$ 2 &u 2 )&V d,=0, (4.3.8)
where , max is the maximum value of ,, corresponding to r 2 or u 2 0, and

which is related to the scattering angle 3, e.g., by 3=?&, max for a repulsive
potential. The variations in (4.3.8) are constrained to those with fixed , max and l,
but unconstrained regarding energy. Five more variational equations for scattering
can be derived from (4.3.6) and (4.3.7). Equation (4.3.8) has appeared in the
literature [14], but the other five variational principles appear to be new. Note that
there are no divergence difficulties associated with any of these variational prin-
ciples for scattering. As an example, in Section 3.2 we apply the third equation in
(4.3.6) (i.e., the RMP version) to the scattering by a repulsive r &4 potential.
For quantum scattering in a central potential, variational principles for the phase
shift and radial wave function have been derived [26] and are quantum analogues
of (4.3.5)(4.3.7).
5. SUMMARY AND DISCUSSION
The main point of this paper is to suggest that the traditional variational
principles of mechanics are not optimally formulated. A set of four principles is
proposed, of which one is well known (HP), but the others (MP, RHP, RMP) are
known only in special cases. The set is invariant under reciprocity and Legendre
transformations. There are numerous advantages of the new formulations:
v energy conservation is a consequence of the reformulated MP (1.5), rather
than an assumption, as in (1.1);
v the reciprocal MP (1.6) is also a new and useful variational principle of
classical mechanics;
v the reformulated MP and RMP are nontrivial and useful even for one-
dimensional motion;
v the RMP is the classical limit of the quantum variational principle;
v analogous to the standard technique in quantum mechanics, the RMP
provides a convenient method of approximation, via the direct or Ritz-type method
of variational calculus, for the complete range of dynamics problems, including
periodic and quasi-periodic motions, scattering, and nonintegrable systems;
v again analogous to techniques in quantum mechanics (e.g., the Schwinger
procedure), for scattering problems, the new MP and RMP can be reformulated in
manifestly divergence-free forms;
v for quasi-periodic motions the RMP is equivalent to Percival's principle for

invariant tori;
v the reformulated MP (1.5) contains the traditional MP (1.1) as a special
case;
v unconstrained versions (2.5), i.e., the UHP and UMP, are also given, with
Lagrange multipliers completely specified.
The new principles (1.5) and (1.6) are statements of the laws of classical
mechanics equivalent to the traditional ones, i.e., Newton's equations, Hamilton's
principle, etc., and as such they must have the same consequences. Whether for
problems of current interest Eqs. (1.5) and (1.6) offer particular advantages remains
to be seen. We believe that the direct link to quantum mechanics may be useful in
problems related to quantum chaos. It seems likely also that there are applications
to n-body and other nonintegrable systems which are beyond the scope of this
paper.
The general variational principles of classical mechanics were formulated in the
period 17401840, and the subject appeared to be complete. It therefore came as a
great surprise to the authors to find that the general principles (1.5), (1.6), (1.7),
and the unconstrained combinations (2.5) presented here are unknown in the
literature, especially as there is an important link to quantum mechanics. Such
references as the authors know have been given in the Introduction and Appendix
B. It is possible however, that we have missed relevant references, and we would
appreciate information in this regard.
APPENDIX A: DIRECT DERIVATION OF REFORMULATED

MAUPERTUIS PRINCIPLE
For notational simplicity we assume one degree of freedom (x), but the argument
is the same for any number of degrees of freedom. The kinetic energy K and
Hamiltonian H are given by
K= 12 mx* 2, H= 12 mx* 2 +V(x), (A1)
where x* =dxdt. We shall apply the general first variation theorem [27] from the
calculus of variations, which includes boundary variations, to functionals J[x(t$)],
t
J[x(t$)]= | 0
F(x, x* ) dt$, (A2)
where F is an arbitrary function. Taking an arbitrary variation of some virtual

trajectory, x(t$) x(t$)+$x(t$), assuming no endpoint variations (i.e., the initial x
has the same value for the original and varied paths, and so also for the final x),
but allowing an arbitrary final endtime variation t t+$t, we have for the first
variation $J
t F d F F
$J= | 0
$x
\ x& dt$ x* + dt$+ _F(t)&x* (t) \x* + & $t t
(A3)
where F(t)and (Fx* ) t are final time values.

We apply (A3) to the two functionals J= to 2K dt$ and J= to H dt$, with K and
H given by (A1). Using
1 t $t 1 t
$
\ t | 0 t+
H dt$ =& E + $
t | 0
Hdt$,
we get
t t
$ | 0
2K dt$= |0
$x(&2mx ) dt$&2K(t) $t, (A4)
1 t 1 t 1
$
t | 0
H dt$=
t | 0
$x(V$&mx ) dt$+ [H(t)&E &mx* (t) 2 ] $t,
t
(A5)
where V$=dV(x)dx. Recalling the definitions W= t0 2K dt$ and E =t &1 to H dt$,
and subtracting (A4) from (A5) multiplied by t gives
t
t $E &$W= | 0
$x(V$+mx ) dt$+[H(t)&E ] $t. (A6)
From (A6) we see that: (a) if the trajectory satisfies the Newton equation of motion
V$+mx =0, from which energy conservation follows, so that H(t)=const=E, we
have t $E &$W=0; and (b) conversely, if t $E &$W=0 for all variations $x(t$)
and $t (which are independent) of the trajectory, we must have V$+mx =0 and
H(t)=E =const. Thus a necessary and sufficient condition for the trajectory to be
a true dynamical one is t $E &$W=0.
This completes the direct derivation of the second of Eqs. (2.5). We can derive
the first directly in a similar fashion. As discussed in Section 2, the MP and RMP
follow immediately from t $E &$W=0. This unconstrained MP is therefore some-
what stronger than the separate Maupertuis principles ($W) E =0 and ($E ) W =0,
and also has practical value in its own right as a variational principle (see, e.g.,
Sections 4.2 and 4.3).
APPENDIX B: WORK RELATED TO THE RECIPROCAL

HAMILTON PRINCIPLE
The RHP (1.7) does not appear to be known, but for the case of periodic motion
Blatt and Lyness [10] derived an equivalent variational principle for the period T,
($T ) L =0, where L # T0 L dtT#ST is the mean Lagrangian. The reciprocal of the
BlattLyness principle, which corresponds to the HP itself (1.3), i.e., ($L ) T =0, has
been derived for 1D periodic motion by Gilmartin et al. [10]. (One can in fact
show that ($L ) t =0 and its reciprocal ($t) L =0 are equivalent to (1.3) and (1.7) not
only for periodic motion, but in general.) For harmonic systems, for which
L #K &V =0 for true trajectories (virial theorem), the BlattLyness principle can
be written as ($&) L =0 =0, where &=1T is the frequency. This is Rayleigh's
Principle (Gerjuoy et al. [2]), the origins of which go back to the work of
Lagrange (Temple and Bickley [29, p. 147]); in any harmonic or linear system, the
frequencies are stationary under the constraint K =V. This reflects the fact that
frequencies are independent of (varied) amplitudes in a linear system.
As discussed in Section 2, the RHP and the HP itself can be combined into a
more general variational principle involving unconstrained variations of S and t,
i.e., $(S+Et)=0, where here E is a constant Lagrange multiplier equal to the true
trajectory energy. Although apparently also unknown for the general case, it too
has been derived for a special case, i.e., 1D periodic motion, by Luttinger and
Thomas [10], who find $(S+ET )=0, where S is the action over one cycle and T
is the period. An unconstrained mean Lagrangian principle for periodic motion has
been given by Blatt and Lyness [10] in the form $(T 2 +*L )=0, where
*=T 2(E&V ) is a constant Lagrange multiplier, with V the mean potential over
one period. The BlattLyness relation can be simplified to $(L &W&)=0, where
&=1T is the frequency, with constant Lagrange multiplier W the action over one
cycle. The last relation generalizes for quasi-periodic motions to the form (see
Eq. (4.2.10) of Section 4.2)
\
$ L &: W k & k =0,
k
+ (B1)
where & k are the frequencies and W k are the (constant Lagrange multiplier) actions.
Equation (B1) has had a checkered history. It was first proposed by Trkal [28] in
1922 for separable systems, and in 1923 by Van Vleck [12] for integrable systems
generally; these authors were interested in reformulating the old quantum theory as
a variational principle. However, Trkal's derivation was criticized by Van Vleck,
and a half-century later, Van Vleck's was criticized in turn by Percival [18].
Correct derivations have also been given by Percival [23] and by Klein and Li
[24] for the constrained form, ($L ) [&k ] =0. For general motions, the corresponding
unconstrained mean Lagrangian principle can be written as $(L +*t)=0, where the
constant Lagrange multiplier here is *=Wt 2, with W the action and t the time for
the true trajectory. For fixed L or fixed t, this relation reduces to the constrained
relations given above in the first paragraph.
One other special case of the unconstrained HP $(S+Et)=0 has been noted in
the literature, by Hamilton himself, i.e., when the virtual trajectories are restricted
to be dynamical ones. For this special case he also noted the unconstrained MP
derived in Section 2, $(W&tE )=0, where now E =E=const on each virtual
trajectory but varies with trajectory, and t is a constant Lagrange multiplier equal
to the trajectory time. These two relations are contained in a group of relations
termed by Hamilton the laws of varying action [1, 2].
ACKNOWLEDGMENTS
CGG and GK gratefully acknowledge the financial support of NSERC (Canada). GK thanks the
Alexander von Humboldt Foundation for support in Germany when part of this paper was written,
Professor Harald Fritzsch for the hospitality of his department, and Professors L. B. Okun and Larry
Spruch for helpful comments. VN is grateful to GK and to the Department of Physics, University of
Guelph, for their hospitality. The authors are very grateful to Mrs. Brenda McLeod for her ceaseless
efforts in processing the manuscript to make it as accurate, misprint-free, and attractive as possible.
REFERENCES
1. W. Yourgrau and S. Mandelstam, ``Variational Principles in Dynamics and Quantum Theory,''

3rd ed., Pitman, London, W. B. Saunders, Philadelphia, 1968; reprinted, Dover, New York, 1979.
The notations S and W are interchanged in this reference compared to the present text, which
follows current usage.
2. C. Lanczos, ``The Variational Principles of Mechanics,'' 4th ed., University of Toronto Press,
Toronto, 1970; reprinted, Dover, New York, 1986; E. G. Jourdain, ``The Principle of Least Action,''
Open Court Publishing, Chicago, 1913; W. Thomson (Lord Kelvin) and P. G. Tait, ``Treatise on
Natural Philosophy,'' Part I, Cambridge Univ. Press, Cambridge, 1879 and 1912; reprinted as
``Principles of Mechanics and Dynamics,'' Dover, New York, 1962; A. W. Conway and A. J.
McConnell (Eds.), ``The Mathematical Papers of Sir William Rowan Hamilton,'' Vol. II,
``Dynamics,'' Cambridge Univ. Press, Cambridge, 1940; E. Gerjuoy, A. R. P. Rau, and L. Spruch,
Rev. Mod. Phys. 55 (1983), 725.
3. G. Karl and V. A. Novikov, Phys. Rev. D 51 (1995), 5069; J. Exp. Theor. Phys. 80 (1995), 783.
4. I. C. Percival, J. Phys. A: Math. Nucl. Gen. 7 (1974), 794, and Ref. [18].
5. E. Schrodinger, Ann. Phys. 79 (1926), 361, 489; reprinted and translated, in ``Collected Papers on
Wave Mechanics,'' Blackie, Glasgow, 1928; reprinted by Chelsea, New York, 1982.
6. See, e.g., Ref. [19, p. 165].
7. J. W. Gibbs, ``The Scientific Papers of J. Willard Gibbs,'' Vol. I, ``Thermodynamics,'' p. 56,
Longmans, Green, London, 1906; reprinted by Dover, New York, 1961.
8. For a general discussion of direct methods of variational calculus, see, e.g., I. M. Gelfand and
S. V. Fomin, ``Calculus of Variations,'' Chap. 8, PrenticeHall, Englewood Cliffs, NJ, 1963.
9. T. Mura and T. Koya, ``Variational Methods in Mechanics,'' Oxford Univ. Press, New York, 1992;
P. M. Morse and H. Feshbach, ``Methods of Theoretical Physics,'' Part II, p. 1106, McGrawHill,
New York, 1953; L. Cairo and T. Kahan, ``Variational Techniques in Electromagnetism,'' Blackie,
London, 1965.
10. The few references of which we are aware which apply the direct (Ritz-type) method [8] of varia-
tional calculus to Hamilton's principle to obtain particle trajectories are listed here. Not surprisingly,
Hamilton himself was aware of this method, see Hamilton Papers, Ref. [2], especially the
correspondence with J. W. Lubbock (1837), pp. 250263. A. G. Basile and C. G. Gray, J. Comp.
Phys. 101 (1992), 80; T. L. Beck, J. D. Doll, and D. L. Freeman, J. Chem. Phys. 90 (1989), 3181;
D. L. Hitzl, J. Comp. Phys. 38 (1980), 185; H. Gilmartin, A. Klein, and C-t Li, Amer. J. Phys. 47
(1979), 636; R. H. G. Helleman, in ``Topics in Nonlinear Dynamics'' (S. Jorna, Ed.), A.I.P.
Conference Proceedings, Vol. 46, p. 264, Amer. Ins. of Phys., New York, 1978; D. W. Schlitt, Amer.
J. Phys. 45 (1977), 205; C. D. Bailey, Found. Phys. 5 (1975), 433; J. M. Blatt and J. N. Lyness,
J. Austral. Math. Soc. 2 (1962), 257; J. M. Luttinger and R. B. Thomas, J. Math. Phys. 1 (1960), 121;
P. H. Miller, Amer. J. Phys. 25 (1957), 30; J. H. Van Vleck, Phys. Rev. 22 (1923), 547; G. Duffing,
`Èrzwungene Schwingungen bei Veranderlicher Eigenfrequenz und ihre Technische Bedeutung,''
p. 130, F. Vieweg 6 Sohn, Braunschweig, 1918.
11. See, e.g., L. D. Landau and E. M. Lifshitz, ``Mechanics,'' 2nd ed., p. 26, AddisonWesley, Reading,
MA, 1969.
12. J. H. Van Vleck, Ref. [10].
13. C. G. Gray, G. Karl, and V. A. Novikov, Amer. J. Phys., in press.
14. M. M. Gordon, Amer. J. Phys. 25 (1957), 32.
15. P. Dahlqvist and G. Russberg, Phys. Rev. Lett. 65 (1990), 28; and references herein to the extensive
literature on the x 2y 2 potential.
16. C. C. Martens, R. L. Waterland, and W. P. Reinhardt, J. Chem. Phys. 90 (1989), 2328.
17. R. P. Feynman, Rev. Mod. Phys. 20 (1948), 267; see also P. A. M. Dirac, Physik. Z. Sowjetunion 3
(1933), 64.
18. M. V. Berry and K. E. Mount, Rep. Progr. Phys. 35 (1972), 315; I. C. Percival, Adv. Chem. Phys.
36 (1977), 1; M. V. Berry, in ``Chaotic Behaviour of Deterministic Systems'' (G. Iooss, R. H. G.
Helleman, and R. Stora, Ed.), p. 171, North-Holland, Amsterdam, 1983; B. Eckhardt, Phys. Rep. 163
(1988), 205; M. C. Gutzwiller; ``Chaos in Classical and Quantum Mechanics,'' Springer-Verlag,
New York 1990; M. V. Berry, in ``Chaos and Quantum Physics'' (M. J. Giannoni, A. Voros, and
J. Zinn-Justin, Ed.), p. 252; Elsevier, Amsterdam, 1991; M. S. Child, ``Semiclassical Mechanics
with Molecular Applications,'' Oxford Univ. Press, Oxford, 1991; O. Bohigas, S. Tomsovic, and
D. Ullmo, Phys. Rep. 223 (1993), 43.
19. See, e.g.; L. D. Landau and I. M. Lifshitz, ``Quantum Mechanics (Nonrelativistic Theory),'' 3rd ed.,
Pergamon, New York, 1977.
20. See, e.g., H. A. Kramers, ``Quantum Mechanics,'' p. 207, North-Holland, Amsterdam, 1957;
reprinted, Dover, New York, 1964.
21. F. Haake, ``Quantum Signatures of Chaos,'' Springer-Verlag, New York, 1991.
22. This is part of the KolmogorovArnoldMoser (KAM) theorem; see, e.g., Gutzwiller, [18, p. 132].
23. I. C. Percival, J. Phys. A 12 (1979), L57.
24. A. Klein and C-t Li, J. Math. Phys. 20 (1979), 572.
25. See, e.g., Gelfand and Fomin, [8, p. 79, 87].
26. L. Hulthen, Fysiogr. Sallsk. Lund Forh. 14 (1944), 2; J. Schwinger, Phys. Rev. 72 (1947), 742; Phys.
Rev. 78 (1950), 135; W. Kohn, Phys. Rev. 74 (1948), 1763; I. E. Tamm, Zh. Eksp. Teor. Fiz. 18
(1948), 337; B. A. Lippman and J. Schwinger, Phys. Rev. 79 (1950), 469; H. Feshbach and
S. I. Rubinow, Phys. Rev. 88 (1952), 484; for a review, see, e.g., Gerjuoy et al., [2].
27. See, e.g., Gelfand and Fomin, [8, Chap. 3].
28. V. Trkal, Proc. Cambridge Philos. Soc. 21 (1922), 80.
29. G. Temple and W. G. Bickley, ``Rayleigh's Principle,'' Oxford Univ. Press, Oxford, 1933; reprinted,
Dover, New York, 1956.
30. W. R. Greenberg, A. Klein, and C-t Li, Phys. Rev. Lett. 75 (1995), 1244.

The Four Variational Principles of Mechanics

Uploaded by

Copyright:

Available Formats

You might also like

The Four Variational Principles of Mechanics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Four Variational Principles of Mechanics

Uploaded by

Copyright:

Available Formats

Annals of Physics  PH5583

annals of physics 251, 125 (1996)

The Four Variational Principles of Mechanics

Received October 10, 1995; revised March 4, 1996

H(q, p)= pq* &L(q, q* ).

Maupertuis [1] proposed in 1744 a global (integral) quantity, which he named

($W) E =0, (1.1)

($S) t =0, (1.3)

($W) E =0. (1.5)

($E ) W =0. (1.6)

($t) S =0. (1.7)

variational principles of mechanics are entirely analogous to four equilibrium prin-

$W=t $E, (1.8)

and similarly a Hamilton version (UHP),

$S=&E $t. (1.9)

2. FOUR EQUIVALENT VARIATIONAL PRINCIPLES

We start by proving the equivalence of the reformulated MP (1.5) and its

L(q, q* )= pq* &H(q, p), (2.1)

$S=$W&t $E &E $t, (2.4)

$S=&E $t, $W=t $E, (2.5)

($S) t =0, (2.6)

$S=* $t, (2.7)

($t) S =0. (2.8)

($W) E =0, (2.9)

whereas for fixed W (i.e., $W=0) we get an RMP (1.6),

($E ) W =0. (2.10)

It is also clear that E=const. is a consequence of (1.5) or (1.6), rather than an

3.1. The Plane Pendulum

H= 12 p 23 +| 2o(1&cos 3), (3.1.1)

Using the RMP, treating | as a variational parameter, we obtain from (E|) W

and the period T is

3.2. Scattering by an r &4 Potential

6*E =(| 2 &1)(3| 2 +1) (3.2.4)

| 2 = 13 [1+2(1+ 92 *E ) 12 ]. (3.2.5)

The scattering angle 3 can be obtained from (3.2.3) and (3.2.5),

3.3. The Anisotropic 2D Quartic Oscillator: A Chaotic System

is a nonintegrable system and that the vast majority of trajectories is chaotic. To

x(t)=A x cos | x t, y(t)=A y cos | y t, (3.3.2)

and then combining (3.3.5) and (3.3.4) gives

4. CONNECTIONS TO OTHER VARIATIONAL PRINCIPLES

In earlier sections, we discussed the equivalence of the reformulated MP and

4.1. Schrodinger's Quantum Variational Principle

This is the BohrSommerfeldWilson quantization rule, which for n fixed means

(a) quantum expectation value turns into classical time average,

4.2. Percival's Variational Principle for Invariant Tori

where q #(q 1 } } } q N ), p #( p 1 } } } p N ), N is the number of degrees of freedom,

where p } dq = i p i dq i ; and C k is a circuit in 3 k at fixed 3 k${k . (For an invariant

(3 k ) T =(3 k +$3 k ) T+$T =2?N k , (4.2.5)

where W k is the mean action W k on the trial torus,

and (4.2.8) can then be rewritten as

$ : & k W k =: $& k W k +: & k $W k ,

and if we recall WT= k & k W k (cf. (4.2.6)) and E =WT&ST=WT&L

4.3. Variational Principles for Scattering Orbits

where u$=dud,. The orbit shape is described by r(,) or u(,).

We first apply a variational principle to a finite segment of an orbit connecting

($S l ) 2, =0, (4.3.5)

where 2,=, 2 &, 1 , which corresponds to the HP (1.4). Similarly, in corre-

($2,) Sl =0, ($W l ) El =0, ($E l ) Wl =0, (4.3.6)

$S l =&E l $2,, $W l =2, $E l (4.3.7)

Annals of Physics PH5583

annals of physics 251, 125 (1996)

($W) E =0. (1.5)

($E ) W =0. (1.6)

$W=t $E, (1.8)

$S=$W&t $E &E $t, (2.4)

$S=&E $t, $W=t $E, (2.5)

($W) E =0, (2.9)

($E ) W =0. (2.10)

Using the RMP, treating | as a variational parameter, we obtain from (E|) W

6*E =(| 2 &1)(3| 2 +1) (3.2.4)

| 2 = 13 [1+2(1+ 92 *E ) 12 ]. (3.2.5)

4.1. Schrodinger's Quantum Variational Principle

This is the BohrSommerfeldWilson quantization rule, which for n fixed means

where q #(q 1 } } } q N ), p #( p 1 } } } p N ), N is the number of degrees of freedom,

where p } dq = i p i dq i ; and C k is a circuit in 3 k at fixed 3 k${k . (For an invariant

where W k is the mean action W k on the trial torus,

$ : & k W k =: $& k W k +: & k $W k ,

and if we recall WT= k & k W k (cf. (4.2.6)) and E =WT&ST=WT&L

where u$=dud,. The orbit shape is described by r(,) or u(,).

($2,) Sl =0, ($W l ) El =0, ($E l ) Wl =0, (4.3.6)

$S l =&E l $2,, $W l =2, $E l (4.3.7)

where E l =E l =E for a true path.

where , max is the maximum value of ,, corresponding to r 2 or u 2 0, and

where F(t)and (Fx* ) t are final time values.