ENSENANZA Revista Mexicana de Fsica E 57 158163 DICIEMBRE 2011

The generating function of a canonical transformation

G.F. Torres del Castillo
Departamento de Fsica Matematica, Instituto de Ciencias, Universidad Autonoma de Puebla,
72570 Puebla, Pue., Mexico.
Recibido el 27 de junio de 2011; aceptado el 3 de octubre de 2011

An elementary proof of the existence of the generating function of a canonical transformation is given. A shorter proof, making use of the
formalism of differential forms is also given.

Keywords: Canonical transformations; generating function.

Se da una prueba elemental de la existencia de una funcion generatriz de una transformacion canonica. Se da tambien una prueba mas corta,
usando el formalismo de formas diferenciales.

Descriptores: Transformaciones canonicas; funcion generatriz.

PACS: 45.20.Jj

1. Introduction 2. Canonical transformations

In order to present the ideas in a simple way, it is convenient
One of the main reasons why the Hamiltonian formalism is
to consider firstly the case where there is only one degree of
more useful than the Lagrangian formalism is that the set of
freedom, which greatly simplifies the derivations.
coordinate transformations that leave invariant the form of
the Hamilton equations is much wider than the set of coordi- 2.1. Systems with one degree of freedom
nate transformations that leave invariant the form of the La-
grange equations. Furthermore, each of the so-called canoni- We shall consider a system with one degree of freedom, de-
cal transformations leaves invariant the form of the Hamilton scribed by a Hamiltonian function H(q, p, t). This means
equations and can be obtained from a single real-valued func- that the time evolution of the phase space coordinates q and
tion of 2n + 1 variables, where n is the number of degrees of p is determined by the Hamilton equations
freedom of the system, which is therefore called the generat-
dq H dp H
ing function of the transformation. = , = . (1)
dt p dt q
The proof of the existence of a generating function for We want to find the coordinate transformations,
an arbitrary canonical transformation given in most standard Q=Q(q, p, t), P =P (q, p, t), that maintain the form of the
textbooks is usually based on the calculus of variations (see, Hamilton equations (1). That is, we want that Eqs. (1) be
e.g., Refs. 1 to 6), which allows one to obtain the basic rela- equivalent to
tions quickly. The aim of this paper is to give a straightfor-
dQ K dP K
ward, elementary derivation of the existence of the generating = , = , (2)
function of a canonical transformation, not based on the cal- dt P dt Q
culus of variations. One of the advantages of the proof given where K may be the original Hamiltonian H expressed in
here is that it allows one to see clearly the assumptions in- terms of the new coordinates or another function. (The last
volved, by contrast with the more diffuse proof usually given possibility is relevant since it turns out that the new Hamilto-
in the textbooks, and to realize that the canonical transfor- nian can be made equal to zero by means of a suitable trans-
mations are not the most general transformations that leave formation.)
invariant the form of the Hamilton equations. In Sec. 2, the Assuming that the transformation Q = Q(q, p, t),
definition of a canonical transformation is briefly reviewed in P = P (q, p, t) is differentiable and can be inverted (that is, it
order to derive the basic equations that lead to the existence is possible to find q and p in terms of Q, P , and t and, there-
of the generating function of the transformation. In Sec. 3 we fore, H can be viewed also as a function of Q, P , and t), mak-
point out some of the frequent errors contained in the proofs ing use repeatedly of the chain rule and of Eqs. (1) and (2) we
given in some of the standard textbooks. For those readers ac- find that
quainted with the formalism of (exterior) differential forms, K dQ Q H Q H Q
a considerably shorter proof is given in the appendix. The = = + (3)
P dt q p p q t
simplicity of this latter proof may serve as an invitation to
learn the language of differential forms for those not already Q H Q H P
= + (4)
familiar with it. q Q p P p
 THE GENERATING FUNCTION OF A CANONICAL TRANSFORMATION 159

for any pair of functions f, g, if and only if Eq. (9) holds. In

fact, making use of the chain rule, one can readily show that
QH Q H P Q
+ + (5)
pQ q P q t f g f g f g f g
= {Q, P } . (11)
q p p q Q P P Q
H Q P Q P Q
= +
P q p p q t Thus, restricting ourselves to coordinate transformations
satisfying Eq. (9), but allowing them to involve the time ex-
H Q
= {Q, P } + , (6) plicitly, Eqs. (6) and (8) yield
P t
Q (H K) P (H K)
where we have made use of the definition of the Poisson = , = . (12)
brackets t P t Q
f g f g Now, it turns out that Eqs. (9) and (12) are necessary and suf-
{f, g} . (7)
q p p q ficient conditions for the local existence of a function F such
In a similar way, we obtain that
P dQ Kdt pdq + Hdt = dF, (13)
K H P
= {Q, P } . (8) as can be seen writing the left-hand side of the last equation
Q Q t
as
Now we have two possibilities: either the Hamilto-
Q Q Q
nian K is essentially the original Hamiltonian P p dq + P dp + P + (H K) dt
H, ex- q p t
pressed in terms
of the new variables [that is, K Q(q, p, t),
P (q, p, t), t =H(q, p, t)], or K differs from H. In the first and applying the standard criterion for a linear (or Pfaffian)
case, Eqs. (6) and (8) will hold, independent of the Hamilto- differential form to be exact. For instance, by considering the
nian H, if and only if coefficients of dq and dt (recalling that q, p, and t are treated
as three independent variables), we have
{Q, P } = 1 (9)
Q Q
P + (H K) P p
and the coordinate transformation does not involve the time, q t t q
Q = Q(q, p), P = P (q, p). Then, Eq. (9) is the necessary P Q P Q (H K)
and sufficient condition for the local existence of a function = +
q t t q q
F such that
P dQ pdq = dF. (10) P (H K) Q (H K) (H K)
= + =0.
q P q Q q
(That is, the function F may not be defined in all the phase
If q and Q are functionally independent, then the function
space, we can only ensure its existence in some neighbor-
F appearing in Eq. (13) can be expressed in terms of q, Q,
hood of each point of the phase space.) In fact, writing the
and t (in a unique way), and from Eq. (13) it follows that
left-hand side of Eq. (10) in the equivalent form
F F F
Q Q P = , p= , H K = , (14)
P p dq + P dp, Q q t
q p
and, necessarily, 2 F/qQ 6= 0 (otherwise q and p would
one finds that the condition not be independent). Conversely, given a function F (q, Q, t)
such that 2 F/qQ 6= 0, Eqs. (14) can be locally inverted
Q Q
P = P p to find Q and P in terms of q, p, and t. In this way, F is a
q p p q
generating function of a canonical transformation.
is equivalent to Eq. (9) [1]. If q and Q are functionally dependent (that is, Q can be
Even though more general transformations are also possi- expressed as a function of q and t only, or q can be expressed
ble (see below), attention is restricted to the transformations as a function of Q and t only), the function F appearing in
satisfying Eq. (9), also when the coordinate transformation Eq. (13) can be written in infinitely many ways in terms of q,
involves the time explicitly. The coordinate transformations Q, and t, and the first two equations in (14) make no sense
satisfying Eq. (9) are called canonical transformations. One (since, e.g., keeping q and t constant in the partial differentia-
good reason to consider only canonical transformations is tion with respect to Q, would make Q also constant). In such
that the Poisson brackets (7) are invariant under these trans- a case, the variables p and Q (as well as P and q) are neces-
formations, in the sense that sarily functionally independent (otherwise q and p would be
dependent). Then, writing Eq. (14) in the equivalent form
f g f g f g f g
= , P dQ Kdt + qdp + Hdt = dF 0 , (15)
q p p q Q P P Q

Rev. Mex. Fis. E 57 (2011) 158163

160 G.F. TORRES DEL CASTILLO

where F 0 F + pq, it follows that the generating function In place of an equation of the form (13), in this case one finds
F 0 can be expressed in a unique way as a function of Q, p, the relation
and t, and the canonical transformation is determined by
F 0 F 0 F 0 P dQKdt = 2(p2 +q 2 )1/2 pdqHdtd(pq/2) . (17)
P = , q= , H K = (16)
Q p t
A second example, related to the previous one, is given
and, necessarily, 2 F 0 /pQ 6= 0. Conversely, a given func-
by the coordinate transformation
tion F 0 (p, Q, t) such that 2 F 0 /pQ 6= 0, defines a canoni-
cal transformation by means of the first two equations in (16). 2
q 1 2
In a similar way, one can consider generating functions de- Q= t arctan , P = (p + q 2 ).
pending on (q, P, t), or (p, P, t) (see, e.g., Refs. 1 to 6). p 2
It should be clear, from the derivation above, that the co-
Now {Q, P } = 2(t arctan q/p), which is also a con-
ordinate transformations satisfying condition (9) are not the
stant of motion if H = (1/2)(p2 + q 2 ), as above. Further-
most general coordinate transformations that leave invariant
more, dQ/dt = 0, dP/dt = 0, which can be written in the
the form of the Hamilton equations and, by contrast to what
form (2) with a new Hamiltonian K = 0. This is not strange,
is claimed in some textbooks (e.g., Refs. 3 and 4), the Pois-
since in the HamiltonJacobi method one finds a transforma-
son bracket {Q, P } needs not be a (trivial) constant. (By a
tion leading to a new Hamiltonian equal to zero, but this is
trivial constant we mean a function whose value is the same
usually done with the aid of canonical transformations (the
at all points of its domain or, equivalently, a function whose
solution of the HamiltonJacobi equation is the generating
partial derivatives are all identically equal to zero.) A simple
function of a canonical transformation to a new set of vari-
example is given by the transformation
q p ables corresponding to a Hamiltonian equal to zero). For this
Q = arctan , P = p2 + q 2 . transformation we obtain the relation
p

One readily finds that the Poisson bracket {Q, P } is equal q
P dQ = 2 t arctan pdq Hdt d(pq/2)
to (p2 + q 2 )1/2 , which is not a trivial constant, but is a p
constant of the motion if the Hamiltonian is, for instance,
H = (1/2)(p2 + q 2 ) (corresponding to a harmonic oscil- [cf. Eqs. (13) and (17)].
lator). Then, the Hamilton equations (1) yield dq/dt = p, The most general coordinate transformation that pre-
and dp/dt = q; therefore, we have, dQ/dt = 1 and serves the form of the Hamilton equations (1) corresponds
dP/dt = 0, which can be expressed as the Hamilton equa- to {Q, P } being a constant of the motion. Indeed, making
tions (2) if the transformed Hamiltonian is chosen as K = P . use of the definition of the Poisson bracket (7), Eqs. (6), (8),
the chain rule, and Eqs. (1)

Q P P Q Q P P Q Q H K
{Q, P } = + = {Q, P }
t q t p p t q p t q q t p q p Q Q

P H K Q H K P H K
+ {Q, P } + {Q, P } {Q, P } +
p q P P p q Q Q q p P P

{Q, P } H Q H P {Q, P } H P H Q
= + +
p Q q P q q P p Q p

Q H P H Q H P H
+ {Q, P } +
q p Q p q P p q Q q p P
Q K P K Q K P K {Q, P } dp {Q, P } dq
+ + =
q p Q p q P p q Q q p P p dt q dt

H H K K
+ {Q, P } Q, + P, Q, P, .
Q P Q P
Now, according to Eq. (11) we have, for instance,

H Q H Q H H
Q, = {Q, P } = {Q, P }
Q Q P Q P Q Q P Q
and
H P H P H H
P, = {Q, P } = {Q, P } ;
P Q P P P Q P Q P

Rev. Mex. Fis. E 57 (2011) 158163

 THE GENERATING FUNCTION OF A CANONICAL TRANSFORMATION 161

therefore By analogy with the case where the number of degrees

H H of freedom is 1, the canonical transformations are defined by
Q, + P, =0
Q P the conditions
and, similarly,
{Qi , Qk } = 0, {Pi , Pk } = 0, {Qi , Pk } = ki . (23)

K K
Q, + P, = 0, Then, Eqs. (20) and (22) yield
Q P
Qi (H K) Pi (H K)
thus showing that {Q, P } is a constant of motion (cf. Ref. 1). = , = (24)
t Pi t Qi
(A shorter proof is given in the appendix.)
[cf. Eqs. (12)]. As is well known, Eqs. (23) imply that
2.2. Systems with an arbitrary number of degrees of Qk Pk Pk Qk
freedom = 0,
q m q j q m q j
When the number of degrees of freedom is greater than 1, the Qk Pk Pk Qk j
existence of a generating function of any canonical transfor- = m , (25)
q m pj q m pj
mation can be demonstrated following essentially the same
steps as in the preceding subsection. We start assuming that Qk Pk Pk Qk
=0
the set of Hamilton equations pm pj pm pj

dq i H dpi H (as a matter of fact, Eqs. (25) are equivalent to Eqs. (23)
= , = i (18) [1, 4]). Indeed, assuming that Eqs. (23) hold, we have
dt pi dt q
Qi Qk i Pk
(i = 1, 2, . . . , n), is equivalent to the set = {Q , Pk } m {Qi , Qk }
q m q m q
dQi K dPi K
= , = i, (19) Qk Qi Pk Qi Pk
dt Pi dt Q = m
q q j pj pj q j
where the new coordinates Qi and Pi are functions of q i , pi ,
Pk Qi Qk Qi Qk
and possibly also of the time. Then, by virtue of the chain m
q q j pj pj q j
rule and Eqs. (18) and (19) we obtain (here and in what fol-
lows there is summation over repeated indices) Qi Qk Pk Pk Qk
=
q j q m pj q m pj
K dQi Qi H Qi H Qi
= j
j
+ Qi Qk Pk Pk Qk
Pi dt q pj pj q t
pj q m q j q m q j
Qi H Qk H Pk
= + and, in a similar manner,
q j Qk pj Pk pj

Qi H Qk H Pk Qi Qi Qi Qk Pk Pk Qk
+ + =
pj Q q k j Pk q j t pm q j pm pj pm pj
i
H Q Qk Qi Qk Qi Qk Pk Pk Qk
= ,
Q k j
q pj pj q j pj pm q j pm q j

H Qi Pk Qi Pk Qi Pi Pi Pk Qk Qk Pk
+ + =
Pk q pjj pj q j t q m pj q m q j q m q j

H H Qi Pi Pk Qk Qk Pk
= {Qi , Qk } + {Qi , Pk } + , (20) j ,
Q k Pk t q q m pj q m pj

with the Poisson brackets being now defined by Pi Pi Pk Qk Qk Pk
=
pm pj pm q j pm q j
f g f g
{f, g} i
, (21) Pi Pk Qk Qk Pk
q pi pi q i j ,
q pm pj pm pj
and, similarly,
and this set of relations implies Eqs. (25).
K H H Pi Equations (24) and (25) are necessary and sufficient con-
i = {Pi , Qk } + {Pi , Pk } + (22)
Q Qk Pk t ditions for the local existence of a function F such that
[cf. Eqs. (6) and (8)]. Pi dQi Kdt pi dq i + Hdt = dF, (26)

Rev. Mex. Fis. E 57 (2011) 158163

162 G.F. TORRES DEL CASTILLO

as can be readily verified writing the left-hand side of the last Since Eq. (26) does not necessarily hold [see, e.g.,
equation in terms of the original variables Eq. (17)], in the case of a non-canonical transformation that
preserves the form of the Hamilton equations, the integrals
Qj Qj Qi
Pj i pi dq i +Pj dpi + Pi +(H K) dt Zt2
q pi t
(pi dq i Hdt)
and applying again the standard criterion for the local exact-
t1
ness of a linear differential form.
If the 2n variables q i , Qi are functionally independent and
(which is not necessarily the case), Eq. (26) implies that F Zt2
can be expressed as a function of q i , Qi , and t, in a unique (Pi dQi Kdt)
way, and t1

F F F do not coincide nor are simply related. However, the actual

Pi = , pi = , H K = . (27) path followed by the system in phase space corresponds to
Qi q i t
stationary values of both functionals (this is analogous, for
The independence of the 2n variables q i , pi requires instance, to the fact that the point x = 0 is a local minimum
that det( 2 F/q i Qj )6=0. Conversely, given a function for the functions f (x) = x4 and g(x) = 1 cos x, despite
F (q i , Qi , t) satisfying this condition, Eqs. (27) define a lo- the fact that these functions are not the same).
cal canonical transformation.
For the canonical transformations such that the set q i , Qi
is functionally dependent, one can employ generating func-
Acknowledgements
tions that depend on other combinations of old and new vari- The author would like to thank the referees for helpful com-
ables; some or all of the q i can be replaced by their conju- ments.
gates pi and, similarly, some or all of the Qi can be replaced
by their conjugates Pi , giving a total of 22n possibilities (not
only the four cases considered, e.g., in Ref. 3). Appendix
A. Derivation using exterior forms
3. Comparison with other treatments
Making use of the properties of the contraction (or interior
The presence of the combinations pi dq i Hdt and product) of a vector field with a differential form (see, e.g.,
Pi dQi Kdt in Eq. (26) is not accidental. It is related to Refs. 7 to 11), one finds that there is only one vector field of
the fact that one obtains the Hamilton equations looking for the form
the path in phase space, q i = q i (t), pi = pi (t), along which
X= + Ai i + B i (A.1)
the integral t q pi
Zt2 whose contraction with the 2-form
(pi dq i Hdt) (28)
dpi dq i dH dt (A.2)
t1
has a stationary value (usually a minimum) when compared is equal to zero (that is, X = 0). In fact, making use of
with neighboring paths with the same end points in phase the expressions (A.1) and (A.2), one finds that X = 0 is
space for t = t1 and t = t2 . Since the addition of the dif- equivalent to
ferential of any differentiable function F (q i , pi , t) to the inte-
H H
grand in (28) changes the value of the integral by a term that Ai = , Bi = .
is the same for all the paths with the same end points in phase pi q i
space for t = t1 and t = t2 , it is right to say that if Hence, the integral curves of X correspond to the solutions
of the Hamilton equations (18).
Pi dQi Kdt = pi dq i Hdt + dF, Equations (19) are then equivalent to the condition
[which is Eq. (26)] then the Hamilton equations (18) will be X = 0, where
equivalent to Eqs. (19). What is wrong to say is that the dPi dQi dK dt. (A.3)
converse is also true (see, e.g., Refs. 2, 5, and 6), or that
Pi dQi Kdt and pi dq i Hdt can only differ by a triv- If we restrict ourselves to canonical transformations, then
ial constant factor and the differential of a function (see, e.g., = , or, equivalently, d(Pi dQi Kdtpi dq i +Hdt) = 0,
Refs. 3 and 4) if Eqs. (18) are equivalent to (19). which implies the local existence of a function F such that
Even though Eq. (26) implies that there exists a functional Eq. (26) holds. However, there is an infinite number of
relation among F , Qi , q i , and t, another frequent error is to 2-forms of the form (A.3), that do not differ by a triv-
conclude that this implies that the 2n variables q i , Qi , are ial multiplicative constant from such that, simultaneously,
functionally independent (see, e.g., Refs. 4 to 6). X = 0 and X = 0 [11].

Rev. Mex. Fis. E 57 (2011) 158163

 THE GENERATING FUNCTION OF A CANONICAL TRANSFORMATION 163

Only in the case of systems with one degree of freedom, any two such 2-forms must be related by = f , where f is
some, nowhere vanishing, real-valued function [11]. Among other things, from = f it follows that {Q, P } = f [with
the Poisson brackets defined by Eq. (7)]. Since and are both closed (that is, their exterior derivatives are equal to zero),
equation = f implies that f must obey the condition

df = 0, (A.4)

that is

f f f H H f H f H f
0= dq + dp + dt dp dq dq dt dp dt = + dp dq dt.
q p t q p q p p q t
By virtue of the Hamilton equations (1), this equation holds if and only if f is a constant of the motion, that is Xf = 0 (see the
examples at the end of Sec. 2.1).

1. M.G. Calkin, Lagrangian and Hamiltonian Mechanics (World
Scientific, Singapore, 1996). Chap. VII.
2. H.C. Corben and P. Stehle, Classical Mechanics, 2nd ed. (Wi-
ley, New York, 1960). Sec. 58.
3. H. Goldstein, C. Poole and J. Safko, Classical Mechanics, 3rd
ed. (Addison-Wesley, San Francisco, 2002). Chap. 9.
4. D.T. Greenwood, Classical Dynamics (Prentice-Hall, Engle-
wood Cliffs, NJ, 1977). Chap. 6.
5. C. Lanczos, The Variational Principles of Mechanics, 4th ed.
(University of Toronto Press, Toronto, 1970). Chap. VII.
6. D. ter Haar, Elements of Hamiltonian Mechanics, 2nd ed. (Perg-
amon, Oxford, 1971). Chap. 5 2.
7. V.I. Arnold, Mathematical Methods of Classical Mechanics
2nd ed. (Springer, New York, 2010).
8. M. Crampin and F.A.E. Pirani, Applicable Differential Geome-
try (Cambridge University Press, Cambridge, 1986).
9. L.H. Loomis and S. Sternberg, Advanced Calculus (Addison-
Wesley, Reading, MA, 1968).
10. S. Sternberg, Lectures on Differential Geometry (Chelsea, New
York, 1983).
11. G.F. Torres del Castillo, Differentiable Manifolds: A Theoreti-
cal Physics Approach (Birkhauser Science, New York, 2012).

Rev. Mex. Fis. E 57 (2011) 158163