Professional Documents
Culture Documents
We present an elementary derivation of the equation for the infinitesimal generators of variational symmetries of a Lagrangian for a system
with a finite number of degrees of freedom. We also give a simple proof of the existence of an infinite number of Lagrangians for a given
second-order ordinary differential equation.
Presentamos una derivacion elemental de la ecuacion para los generadores infinitesimales de simetras variacionales de una lagrangiana
para un sistema con un numero finito de grados de libertad. Damos tambien una prueba simple de la existencia de un numero infinito de
lagrangianas para una ecuacion diferencial ordinaria de segundo orden dada.
where g is a constant, where t00 , t01 are the values of t0 corresponding to the points
(x0 , t0 ), (x1 , t1 ), respectively, according to the transforma-
x gt3 s t
x0 = , t0 = , (3) tion (1). The last term on the right-hand side of Eq. (7) is
1 ts 2(1 ts)2 1 ts equal to the difference of the values of F at the endpoints,
and and it is therefore a constant when one considers curves with
x0 = xes , t0 = te2s . (4) the same endpoints. Hence, a curve that minimizes (or max-
imizes) the first term on the right-hand side of Eq. (7) is
It may be noticed that the transformations (2) and (4) are de- mapped into a curve that minimizes (or maximizes) the in-
fined for all s R, but (3) is defined only for s 6= 1/t. tegral on the left-hand side.
Actually, Eqs. (2)(4) are examples of local one- For instance, the family of transformations (2) is a varia-
parameter groups of transformations, which means that, if tional symmetry of the Lagrangian
we define s (x, t) (x0 , t0 ), then
1 3 1 2
L(x, x, t) = x + g x t g 2 xt, (8)
s (u (x, t)) = s+u (x, t), (5) 6 2
for all values of s and u such that both sides of the equation where g is a constant [the constant appearing in Eqs. (2)]. In
are defined. For instance, in the case of the transforma- fact, treating the derivative as a quotient of differentials (or,
tions (4), s (x, t) = (xes , te2s ); hence, s (u (x, t)) = equivalently, using the chain rule), from Eqs. (2) we have
s (xeu , te2u ) = ((xeu ) es , (te2u ) e2s ) = dx0 es dx (e3s es )gtdt s/2
(xe su
, te 2s+2u
) = s+u (x, t). These one-parameter = =e x (e3s/2 es/2 )gt,
dt0 e3s/2 dt
groups of transformations arise in a natural way in the solu-
tion of systems of first-order ODEs (see the examples below). hence,
We shall say that the one-parameter family of transfor- 0
0 dx
0
0 dt 1 s/2 3
mations (1) is a variational symmetry of a given Lagrangian L x , 0 ,t = e x e3s/2 es/2 gt
L(x, x, t) if dt dt 6
1
0 2
dx0 dt dx d + g es/2 x e3s/2 es/2 gt te3s/2
L x0 , 0 , t0 = L x, , t + F (x, t, s), (6) 2
dt dt dt dt
1
g 2 xes gt2 e3s es te3s/2 e3s/2
for all values of s for which the transformation is defined, 2
where F (x, t, s) is some function. Some authors reserve the
dx d 1 2 2
name variational symmetry for the point transformations sat- = L x, ,t + g t x 1 e4s
isfying Eq. (6) without the last term of the right-hand side dt dt 2
(e.g., Refs. [58]), and the point transformations satisfying 1 3 4 4s 6s
(6) with dF/dt 6= 0 are sometimes called Noether symme- + g t 1 6e + 5e .
24
tries [5] or divergence symmetries [6]. As we shall show be-
low, each one-parameter family of point transformations sat- (Note that s is a parameter, that does not depend on t.)
isfying Eq. (6) yields a constant of motion for the ODE given On the other hand, the transformations (4) are variational
by the Lagrangian L. symmetries for the Lagrangian
More precisely, a transformation satisfying Eq. (6) maps
t2 x6
any solution of the EulerLagrange equation corresponding L(x, x, t) = x2 . (9)
to L, into another solution, which follows from the fact that 2 3
the EulerLagrange equation determines the local extrema of Indeed, from Eqs. (4) we have
the integral
Zt1 dx0 es dx
dx = 2s = e3s x,
L x, , t dt, dt 0 e dt
dt
t0
and, therefore,
with fixed endpoints (x0 , t0 ), (x1 , t1 ) [13, 8]. Condition (6) 0
0
amounts to 0 dx 0 dt (te2s )2 3s 2 (xes )6 2s
L x , 0 ,t = (e x) e
0
dt dt 2 3
Zt1 0
Zt1
0 dx 0 0 dx t2 x6
L x , 0 , t dt = L x, , t dt = x2 ,
dt dt 2 3
t00 t0
showing that Eq. (6) holds with F = 0.
Zt1
d Finding the variational symmetries of a given Lagrangian,
+ F (x, t, s) dt, (7) making use of the definition (6), is not an easy task. How-
dt
t0 ever, as we shall see, this problem is simplified if we start
looking for the infinitesimal generators of such symmetries. two functions and , which depend on (x, t) (recall that,
Indeed, differentiating both sides of Eq. (6) with respect to s, e.g., d/dt = /t + x /x). The left-hand side of
at s = 0, making use of the chain rule and the definitions Eq. (12) is a linear operator acting on the functions and
and, therefore, any linear combination, with constant co-
x0 (x, t, s) efficients, of solutions of Eq. (12) is also a solution of this
(x, t) ,
s s=0 equation (see, e.g., Eqs. (22), below).
t0 (x, t, s) As pointed out above, the interest in the variational sym-
(x, t) , (10) metries of L comes from the fact that, each pair of functions
s s=0
, , that satisfies Eq. (12) gives rise to a constant of motion;
we obtain that is, to a function with a total derivative with respect to the
time equal to zero, if the EulerLagrange equations hold. In
L L dx0 L
+ + fact, from the EulerLagrange equation
x x s dt0 s=0 t
d L L
dt0 dF = , (13)
+ L(x, x, t) = . (11) dt x x
s dt s=0 s dt s=0
and the chain rule, we have
Treating the derivative dx0 /dt0 as a quotient of differen-
tials, using the elementary rules of differentiation and the def- dL L L L
= x + x +
initions (10), we see that dt x x t
d L L
0 0 0 0 = x +
dx0 (dt ) dx (dx ) dt dt x t
= s s
s dt0 s=0 (dt0 )2 therefore, Eq. (12) can be written as
s=0
0 0 d L L d d
x t + x
(dt0 ) d (dx0 ) d dt x x dt dt
s s d d
= = x . d L d dG
(dt0 )2 dt dt + L x + L = ,
dt x dt dt
s=0
equating the coefficients of x3 , x2 , x, and x0 on both sides of It may be remarked that, if one assumes that G is equal to
the equation we obtain the four equations zero, or a trivial constant, then c1 = c2 = c4 = c5 = 0,
and, instead of the five-parameter family of symmetries ob-
= 0, (16) tained above, one is left with just a one-parameter group of
x
variational symmetries of the Lagrangian (15), which is the
1 obvious one (x0 = x, t0 = t + s), which is related to the fact
= 0, (17)
x 2 t that L does not depend explicitly on the time.
G The constant of motion associated with the infinitesimal
m mgx = , (18)
t x x generator (22) is [see Eq. (14)]
G
mg mgx = . 1 1 1 1 1
t t
(19) c1 m xxt g xt3 x2 t2 x2 + gxt2 g 2 t4
2 2 2 2 8
Equation (16) implies that is some function of t only,
1 3 2 1 2 1 1 2 3
+ c2 m xx g xt x t + gxt g t
= A(t) (20) 2 4 2 2 4
and from Eq. (17) it follows that 1 2 1 2
+ c3 m x gx + c4 m xt x + gt
2 2
1 dA
= x + B(t), (21) + c5 m(x + gt). (24)
2 dt
where B(t) is another function of t only. Using now Eqs. (18) Since the constants c1 , . . . , c5 are arbitrary, each of the func-
and (19), the equality of the partial derivatives 2 G/tx tions inside the parentheses is a constant of motion, though
and 2 G/xt gives they cannot be functionally independent; for a second-order
2 ODE, there are only two functionally independent first in-
m = mg mg , tegrals. In this case, any constant of motion must be some
t2 x t
function of, e.g.,
that is,
1 d3 A d2 B 3 dA 1
x + 2 = g . 1 x + gt, 2 x xt gt2 . (25)
2 dt3 dt 2 dt 2
Since A and B are functions of t only, we have The values of 1 and 2 are the values of x and x at t = 0,
d A3 2
d B 3 dA respectively. It may be noticed that the constant of motion
= 0, and = g . multiplying c3 is minus the total energy.
dt3 dt2 2 dt
The functions and can be conveniently combined in
These equations imply that the linear partial differential operator
A(t) = c1 t2 + c2 t + c3 ,
+ . (26)
where c1 , c2 , c3 are arbitrary constants, and t x
This combination is invariant under coordinate transforma-
1 3
B(t) = c1 gt3 c2 gt2 + c4 t + c5 , tions and constitutes a vector field, in the terminology of the
2 4 theory of differentiable manifolds (see, e.g., Refs. [6, 11]).
where c4 and c5 are two additional arbitrary constants. Sub- Substituting Eqs. (22) into Eq. (26) we obtain the vector field
stituting these expressions into the previous results we obtain
2 1 3
= c1 t2 + c2 t + c3 , X = c1 t + xt gt
t 2 x
1 3 1 3 2 1 3
= c1 xt gt + c2 x gt + c4 t + c5 (22) + c2 t + x gt2
2 2 4 t 2 4 x
showing that in this case the solution of Eq. (12) contains
five arbitrary constants. Making use Eqs. (18) and (19) we + c3 + c4 t + c5 ,
t x x
find that, up to an irrelevant constant term,
which is a linear combination of the five vector fields
1 3 1
G = c1 mx2 mgt2 x + mg 2 t4 2 1 3
2 2 8 X1 t + xt gt ,
t 2 x
3 1 1 3 2
+ c2 mgtx + mg 2 t3 X2 t + x gt ,
2 4 t 2 4 x
1
+ c4 mx mgt2 + c5 (mgt). (23) X3 , X4 t , X5 . (27)
2 t x x
These vector fields form a basis for the infinitesimal genera- which is the local group of point transformations given in
tors of the variational symmetries of the Lagrangian (15) and Eqs. (3).
also form a basis of a Lie algebra, with the bracket given by In a similar way, one finds that the one-parameter group
the commutator, but this fact will not be elaborated here. of transformations generated by X4 is that of the Galilean
Once we have obtained the infinitesimal generator of a transformations t0 = t, x0 = x + st; X3 generates the time
variational symmetry of a Lagrangian, we can use it di- displacements t0 = t + s, x0 = x; X5 generates the transla-
rectly to find a first integral [by means of Eq. (14)], with- tions x0 = x + s, t0 = t; and X2 is the infinitesimal generator
out the need to know a one-parameter family of transfor- of the group of point transformations
mations corresponding to that infinitesimal generator. How-
1
ever, it is always possible, in principle, to find a unique local x0 = xes/2 + gt2 (es/2 e2s ), t0 = tes .
one-parameter group of transformations whose infinitesimal 2
generator is defined by a given pair of functions (x, t) and 2.2. The existence of an infinite number of Lagrangians
(x, t). All we have to do is to find the solution of the system
of first-order ODEs Given a second-order ODE
0 0
dx dt x = f (x, x, t),
= (x0 , t0 ), = (x0 , t0 ), (28) (30)
ds ds
with the initial condition (x, t). Equation (28) follows from which may correspond to a mechanical system or may have
the definition (10) written as some other origin, we want to find some function, L(x, x, t),
such that the EulerLagrange equation (13) be equivalent to
x0 (x, t, s)
Eq. (30). To this end, we note that Eq. (13) amounts to
(x0 (x, t, s), t0 (x, t, s)) = ,
s=0 s s=0
L L L L
+ x + x =
t0 (x, t, s) t x x x x x x
(x0 (x, t, s), t0 (x, t, s)) = ,
s=0 s s=0 and therefore we are looking for a function L(x, x, t) such
demanding that the equalities hold for all values of s (not that
only for s = 0), treating x and t as parameters (that specify L L L L
the initial conditions). For example, in the case of the vector + x + f (x, x, t) = (31)
t x x x x x x
field X1 [see Eq. (27)], the system of equations (28) takes the
form holds for all values of x, x, and t. Taking the partial deriva-
dx0 1 dt0 tive with respect to x on both sides of Eq. (31), assuming that
= x0 t0 gt03 , = t02 . (29)
ds 2 ds the partial derivatives of L commute and letting
From the second of these last equations (separating variables)
we obtain 2L
1 M , (32)
0 = s + const. x2
t
one obtains
and the integration constant is determined by the condition
that t0 = t at s = 0; hence, 1/t0 = s 1/t, that is M M f M
+ x +M + f (x, x, t) = 0, (33)
t x x x
t
t0 = . which is a first-order linear partial differential equation for
1 ts
M . Making use of Eq. (30), Eq. (33) can also be written as
Inserting this expression into the first equation in (29) we ob-
tain the linear equation dM f
= M . (34)
3 dt x
dx0 t 0 1 t
x = g This last equation shows that M is defined up to a multi-
ds 1 ts 2 1 ts
plicative constant of motion; that is, if M1 and M2 are two
(recall that here t is a parameter determining the initial con- solutions of Eq. (34), then d(M1 /M2 )/dt = 0, which means
dition). Thus that there is an infinite number of Lagrangians for Eq. (30)
since M1 /M2 can be a trivial constant (i.e., a real number) or
gt2 a function of x, x, ant t with a total derivative with respect to
(1 ts)x0 = + const.
2(1 ts) the time equal to zero as a consequence of Eq. (30) (see the
example in Sec. 2.2.3, below).
The integration constant has to be chosen in such a way that
Once we have a solution of Eq. (34), from Eq. (32) we
x0 = x for s = 0. Hence,
can find an expression for L, containing two indeterminate
x gt3 s functions of x and t. Substituting the expression for L thus
x0 = , obtained into Eq. (31), the Lagrangian is determined up to
1 ts 2(1 ts)2
the total derivative with respect to the time of an arbitrary where k is a constant, or, equivalently,
function of (x, t). r
du 1 4
As pointed out in Ref. [4], Eq. (33) is the equation for the = u2 u4 + 4ku,
Jacobi last multiplier of the system of equations dt t 3
3. Systems with an arbitrary number of de- with the initial condition qi0 (0) = qi , t0 (0) = t.
grees of freedom By contrast with the case of a single second-order ODE,
considered in Sec. 2.2, not every system of two or more
Since the derivations are almost identical to those presented second-order ODEs can be derived from a Lagrangian (see,
in Sec. 2, in this section we only give the definitions and main e.g., Ref. [8] and the references cited therein).
results applicable to the case of a mechanical system with an
arbitrary number, n, of degrees of freedom or, more gener-
ally, we have a system of n second-order ODEs derivable 3.1. Example
from a Lagrangian L(qi , qi , t).
The one-parameter family of point transformations A simple example is given by the Lagrangian
x = g, where g is a constant, possesses an eight-parameter systems of EDOs, without making use of a Lagrangian, with
group of point symmetries [7], while the Lagrangians (15) the aid of the so-called adoint symmetries of the system of
and (8), which lead to this equation, admit five-parameter equations. A similar result is presented in Ref. [17], making
and three-parameter groups of variational symmetries, re- use of Lie group analysis.
spectively. The review paper [18] presents various generalizations of
In spite of the fact that, for a given ODE or system of the basic results given here, making use of the language of
ODEs, the variational symmetries may not be the more gen- differentiable manifolds, vector fields and differential forms.
eral point symmetries of the equation or system of equations,
the variational symmetries are very useful because there ex- Acknowledgment
ists a first integral associated with each of them, which can be
readily calculated [Eqs. (14) and (46)], though, as we have The authors wish to thank the referee for helpful comments.
seen, the first integrals obtained in this manner need not be Two of the authors (C.A.M. and R.I.B.R.) wish to thank
functionally independent. On the other hand, not all first inte- the Vicerrectora de Investigacion y Estudios de Posgrado of
grals are associated with variational symmetries. In Ref. [16], the Universidad Autonoma de Puebla for financial support
it is shown that it is possible to find first integrals of EDOs or through the program Jovenes Investigadores.
1. V.I. Arnold, Mathematical Methods of Classical Mechanics, 9. G.W. Bluman and S. Kumei, Symmetries and Differential Equa-
2nd ed. (Springer-Verlag, New York, 1989). tions (Springer-Verlag, New York, 1989).
2. M.G. Calkin, Lagrangian and Hamiltonian Mechanics (World 10. H. Rund, The HamiltonJacobi Theory in the Calculus of Vari-
Scientific, Singapore, 1996). Chap. VII. ations (Van Nostrand, London, 1966). Chap. 2.
3. H. Goldstein, C.P. Poole, Jr. and J.L. Safko, Classical Mechan- 11. G.F. Torres del Castillo, Differentiable Manifolds: A Theoreti-
ics, 3rd ed. (Addison-Wesley, San Francisco, 2002). cal Physics Approach (Birkhauser Science, New York, 2012).
4. E.T. Whittaker, A Treatise on the Analytical Dynamics of Par- 12. M.C. Nucci and P.G.L. Leach, J. Nonlinear Math. Phys. 16
ticles and Rigid Bodies, 4th ed. (Cambridge University Press, (2009) 431.
Cambridge, 1993). Chap. X.
13. P. Guha and A. Ghose Choudhury, Acta Appl. Math. 116 (2011)
5. H. Stephani, Differential Equations: Their Solution Using Sym- 179.
metries (Cambridge University Press, Cambridge, 1990).
14. M.C. Nucci, J. Phys.: Conf. Ser. 380 (2012) 012008.
6. P.J. Olver, Applications of Lie Groups to Differential Equations,
15. O.P. Bhutani and K. Vijayakumar, J. Austral. Math. Soc. B 32
2nd ed. (Springer-Verlag, New York, 2000).
(1991) 457.
7. P.E. Hydon, Symmetry Methods for Differential Equations: A
16. S.C. Anco and G. Bluman, Euro. Jnl of Applied Mathematics 9
Beginners Guide (Cambridge University Press, Cambridge,
(1998) 245.
2000).
8. B. van Brunt, The Calculus of Variations (Springer-Verlag, 17. M. Marcelli and M.C. Nucci, J. Math. Phys. 44 (2003) 2111.
New York, 2004). 18. W. Sarlet and F. Cantrijn, SIAM Rev. 23 (1981) 467.