1 Introduction
In classical mechanics, there is no unique prescription for one to choose the generalized
coordinates for a problem. As long as the coordinates and the corresponding momenta
span the entire phase space, it becomes an acceptable set. However, it turns out in prac-
tice that some choices are better than some others as they make a given problem simpler
while still preserving the form of Hamilton’s equations. Going over from one set of chosen
coordinates and momenta to another set which satisfy Hamilton’s equations is done by
canonical transformation.
For instance, if we consider the central force problem in two dimensions and choose Carte-
k
sian coordinate, the potential is − p . However, if we choose (r, θ) coordinates,
x2 + y 2
k
the potential is − and θ is cyclic, which greatly simplifies the problem. The number of
r
cyclic coordinates in a problem may depend on the choice of generalized coordinates. A
cyclic coordinate results in a constant of motion which is its conjugate momentum. The
example above where we replace one set of coordinates by another set is known as point
transformation.
In the Hamiltonian formalism, the coordinates and momenta are given equal status and
the dynamics occurs in what we know as the phase space. In dealing with phase space
dynamics, we need point transformations in the phase space where the new coordinates
(Qi ) and the new momenta (Pi ) are functions of old coordinates (pi ) and old momenta
(pi ):
Qi = Qi ({qj }, {pj }, t)
Pi = Pi ({qj }, {pj }, t)
D.
c K. Ghosh, IIT Bombay 2
Does a Hamiltonian K(P, Q) exist for which P and Q satisfy the Hamilton’s equations:
∂K
Ṗ = −
∂Q
∂K
Q̇ =
∂P
∂ Ṗ ∂ Q̇
Existence of such a Hamiltonian would imply = − as each of the expressions
∂P ∂Q
∂ 2K ∂ Ṗ ∂ Q̇ 1
equals − . However, Since = = , this condition is not satisfied and the
∂P ∂Q ∂P ∂Q t
modified Hamiltonian K does not exist. Thus the contact transformations (1) are non-
canonical, they do not preserve the volume of the phase space.
It may be noted that this does not imply that one cannot find an equation of motion
using these variables. Indeed, since p is constant, dP/dt = p = P/t, which gives P = kt,
where k is a constant. Likewise, we have,
d Q dq p Pt k
= = = =
dt t dt m m m
D.
c K. Ghosh, IIT Bombay 3
k
so that Q = t2 + C, where C is a constant.
m
To illustrate a canonical transformation, consider the free particle Hamiltonian again for
which ṗ = 0 and q̇ = p/m, as before. Consider a linear transformation of (p, q) to a new
set (P, Q), which is given by
P = αp + βq
Q = γp + δq (2)
α β
If the matrix of transformation is non-singular, i.e. if ∆ =
=6 0, then the above
γ δ
transformation is invertible and we have
1
p= (δP − βQ)
∆
1
q = (−γP + αQ) (3)
∆
It can be checked that
p β
Ṗ = αṗ + β q̇ = β q̇ = β = (δP − βQ)
m m∆
p δ
Q̇ = γ ṗ + δ q̇ = δ q̇ = δ = (δP − βQ) (4)
m m∆
Is there a Hamiltonian K for which P and Q are canonical? It is seen that
∂ Ṗ βδ
=
∂P m∆
∂ Q̇ βδ
=−
∂Q m∆
∂K
so that K exists. To determine K note that Ṗ = − which gives
∂Q
β 2 Q2
Z
βδ
K=− Ṗ dQ = − PQ + + f (P )
m∆ m∆ 2
∂K
and Q̇ = gives
∂P
δ2 P 2
Z
βδ
K=− Q̇dP = − PQ + + f (Q)
m∆ m∆ 2
These two expressions give
1
K= (δP − βQ)2
2m∆
Except for the factor ∆, this is just the old Hamiltonian written in terms of the new vari-
ables. For ∆ = 1, the two Hamiltonians are identical and the transformation corresponds
D.
c K. Ghosh, IIT Bombay 4
Consider a second example where, once again we take the free particle Hamiltonian of
the previous example. Consider a transformation
and
∂ Ṗ ∂ Q̇
One can check 6= − , so that the transformation is not canonical with respect to
∂P ∂Q
the given Hamiltonian. This shows that (5) is not a valid canonical transformation. This
is because, in order that a transformation may be canonical, it must satisfy Hamilton’s
equation for all systems having the same number of degrees of freedom.
d ∂H ∂H
Since H = H(p, q), we get from above, p+ = 0, i.e. ṗ = − and likewise, since
dt ∂q ∂q
∂H ∂H
pq̇ − H no dependence on ṗ, −q̇ + = 0, i.e., q̇ = . The variational porinciple
∂p ∂p
satisfied by the old pair of variables (q, p) is
Z
δS = δ (pq̇ − H)dt = 0
with no variation at the end points. With respect to the new variables, one then must
have Z
0
δS = δ (P Q̇ − H)dt = 0
These two forms of S and S 0 are completely equivalent if the two integrands differ either
by a scale factor or by total time differential of a function F , i.e. if
X X
Pi Q̇i − K = λ (pi q̇i − H) (7)
i i
we notice that the difference between the two differential forms must be an exact dif-
ferential. Analogous statement can be made regarding (9) as well. It turns out that
this condition of exact differentiability is both necessary and sufficient condition for a
transformation to be canonical.
F = fi (q1 , q2 , . . . , qn ; t)Pi − Qi Pi
∂fi
The set of equations above can be satisfied by taking Qi = fi , K = H + Pi and
∂t
∂fi
requiring pi q̇i = q̇j Pi which can be rearranged as follows:
∂qj
∂fi ∂fj
pi q̇i = q̇j Pi = q̇i Pj
∂qj ∂qi
where in the last term we have interchanged the sum over i and j. This gives
X ∂fj
pi = Pj
j
∂qi
where we have explicitly reintroduced the sum for clarity. These are n equations which
must be inverted to get Pi . It may be noted that the choice of F is not unique corre-
sponding to a particular canonical transformation. For instance, if we add any g(t) to
dF dF dg
F it dopes not affect the action integral, because → + does not change the
dt dt dt
Lagrangian equations.
Let us illustrate how to find generator. Let K(Q, P, t) = H(q, p, t). Plugging this into
(8), we get
dF X
= (pi q̇i − Pi Q̇i )
dt i
This is not very interesting because there is no time dependence. The simplest way to
satisfy it is
F = F (q, Q)
We have
dF X ∂F ∂F
= q̇i + Q̇i
dt i
∂q i ∂Q i
X
= (pi q̇i − Pi Q̇i )
i
∂F ∂F
This gives with = pi ; = −Pi . A trivial way to satisfy the equations is to take
P ∂qi ∂Qi
F (qi , Qi ) = i qi Qi so that
∂F
= Qi = pi
∂qi
∂F
= qi = −Pi
∂Qi
which simply requires us to swap the coordinates and the momenta in the Hamiltonian
formalism (but for a sign).
D.
c K. Ghosh, IIT Bombay 9
F = F1 (qi , Qi , t) (11)
dF1
where the last expression is from (8). We can compare the last two expressions for
dt
to get
∂F1
pi =
∂qi
∂F1
Pi = −
∂Qi
∂F1
K =H+ (13)
∂t
Note that (12) can be rewritten as
If the Hamiltonian has no explicit time dependence, the generating function will also
not have an explicit time dependence and in such a case K = H. In such a situation (8)
can be written as (summation convention used)
dF
= pi q̇i − Pi Q̇i
dt
For instantaneous transformation, we can write the above as
dF = pi δqi − Pi δQi
D.
c K. Ghosh, IIT Bombay 10
Since the left hand side is a perfect differential, so is the right hand side. In order that
this may be so, we require
∂pi ∂Pi
= (15)
∂Qi ∂qi
When the functional dependence are given, one can always use (15) to test whether the
transformation is canonical.
As shown earlier, one of the simplest examples of a F1 type transformation is F1 (qi , Qi ) =
qi Qi , which gives the swapping transformation pi = Qi and Pi = −qi . Let us see what it
p2 1
does to the Hamiltonian of the Harmonic oscillator for which H = + kq 2 . Since the
2m 2
Hamiltonian has no explicit time dependence, we have K = H. In such a case
Q2 1
K= + kP 2
2m 2
The new equations of motion are
∂K
Q̇ = = kP
∂P
∂K Q
Ṗ = − =−
∂Q m
Thus we get
Q
Q̈ = k P̈ = −
m
Q̇
which has the solution Q = Q0 cos(ωt + δ) . We can obtain P from the equation P =
k
Q0 ω
which gives P = − sin(ωt + δ). Since this is equal to −q, we can identify q0 = Q0 ω/k.
k
We then can write
q = q0 cos(ωt + δ 0 )
where δ 0 − δ = π/2.
∂F1
Pi = −
∂Qi
and we wish to replace the variable Qi by the new variable Pi , we define
X
F2 (qi , Pi , t) = F1 (qi , Qi , t) + Pi Qi (16)
i
D.
c K. Ghosh, IIT Bombay 11
Note that starting with the equation (14) for F1 can be written (use summation conven-
tion) by observing
dF1
= pi q̇i − Pi Q̇i + K − H (17)
dt
Using (16), we have
dF2 dF1
= + Pi Q̇i + Ṗi Qi
dt dt
= pi q̇i + Ṗi Qi + (K − H) (18)
We then have
dF3 dF1
= − pi q̇i − qi ṗi
dt dt
= −qi ṗi − Pi Q̇i + (K − H) (20)
What we need to show is that Poisson’s bracket calculated using either basis is the same,
i.e.,
{Qi , Pj }QP = {Qi , Pj }qp (24)
Clearly, since {Qi }, {Pi } are canonically conjugate pairs, they satisfy
5 Illustrative Problems
Example 1: Harmonic Oscillator The Hamiltonian for the Harmonic oscillator is
p2 1
H(p, q) = + mω 2 q 2 . Let us carry out a transformation
2m 2
p = f (P ) cos Q
f (P
q= sin Q (28)
mω
f (P )2
This converts the Hamiltonian to a function of P alone with K = H = . Note that
2m
the Hamiltonian is cyclic in Q so that P is constant. The problem is to find f (P ) such
D.
c K. Ghosh, IIT Bombay 14
∂F1
p=
∂q
∂F1
P =−
∂Q
∂F1
K =H+ =H (29)
∂t
the last equality because the Hamiltonian is time independent. From (28), we get,
∂F1
p = mωq cot Q =
∂q
Integrating, we get
mωq 2
F1 = cot Q
2
The momenta conjugate to Q is then given by
∂F1
P =−
∂Q
mωq 2
= cosec2 Q
2
p2 1
=
2mω cos2 Q
which gives √
p= 2P mω cos Q
√ f (P )2
so that f (P ) = 2P mω. This in turn gives the new Hamiltonian to be K = = ωP .
2m
Thus the transformation (28) is equivalent to
√
p = 2P mω cos Q
r
2
q= sin Q
mω
Inverting, we get
1
P = (q 2 m2 ω 2 + p2 )
2mω
−1 mωq
Q = tan
p
D.
c K. Ghosh, IIT Bombay 15
Note that
∂Q mωp
=
∂q p2 + m2 ω 2 q 2
∂P p
=
∂p mω
∂Q mωq
=−
∂p p2 + m2 ω 2 q 2
∂P
= qmω
∂q
which shows that
{Q, P }qp = 1
i.e. as expected, the transformation preserves Poisson’s bracket.
E
As the energy is constant we have P = . The Hamilton’s equations for the new
ω
variables then gives
∂K
Q̇ = =ω
∂P
yielding Q = ωt + α The original problem is now solved,
√ √
p = 2mE cos Q = 2mE cos(ωt + α)
r r
2P 2E
q= sin Q = sin(ωt + α)
mω mω 2
Example 2:
1 1
Consider the Hamiltonian H = (p2 +q 2 ). Determine if the transformation Q = (q 2 +p2 )
2 2
and P = − tan−1 (q/p) is canonical. If so, find a generating function of type F1 .
p2
It can be easily seen that both sides of the above condition equals so that the
p2 + q 2
transformation is canonical.
D.
c K. Ghosh, IIT Bombay 16
We can obtain F by integrating either of the factors. For instance, a bit of exercise in
integration gives
Z
−1 q
F = p + q tan dq
p
pq q 2 + p2 q
= + tan−1
2 2 p
However, the above form cannot be a generating function as it has no dependence on the
new variables. Sincep we wish it to be of form F1 , we need to eliminate p. This is done
q q q
by substitution p = 2Q − q 2 so that tan−1 = tan−1 p = sin−1 √ . With
p 2Q − q 2 2Q
these substitutions, we get
q 1 p
F = Q sin−1 √ + q 2Q − q 2
2Q 2
1 ∂K
The Hamiltonian K = (q 2 + p2 ) = Q, so that Q̇ = = 0, i.e. Q is constant, and
2 ∂P
∂K
Ṗ = − = −1. Since Q is constant, so is the Hamiltonian and P decreases with time.
∂Q
Example 3:
sin p
Prove that the transformation Q = ln and P = q cot p is canonical and derive all the
q
different forms of generating functions.
We have, for the given teansformation
1
δQ = cot pδp − δq
q
Thus
1
pδq − P δQ = pδq − q cot p cot p δp − δq
q
= (p + cot p) δq − q cot2 p δp
It can be seen that both sides equal − cot2 p and hence the transformation is canonical.
It is also easy to see that Poisson’s bracket is preserved.
(We can also integrate the other factor over p and show that both the forms are the
same if the constant integration is taken to be zero). We will now express this in terms
of different forms of the generating
p function. Note that for p the given transformation,
Q
sin p = qe which gives cos p = 1 − q e and q cot p = e−2Q − q 2 . Substituting
2 2Q
these, we get p p
F1 = q cos−1 1 − q 2 e2Q + e−2Q − q 2
Suppose we wish to find the generator form F2 . In such a case, we need to replace p by
P , which is quite straightforward. We can find F2 by a Legendre transformation. Since
∂F1
P =− , we have F2 (q, P ) = F1 (q, Q) + P Q. Thus
∂Q
1p
F2 (q, P ) = F1 (q, Q) = q sin−1 (qeQ ) + 1 − q 2 e2Q + P Q
eQ
However, as F2 must be expressed as a function
p of q and P , we must accordingly express
1 − q e2Q
2 1
the variables. We had shown that P = Q
, which gives eQ = p . Thus
e P 2 + q2
!
q P
F2 (q, P ) = q sin−1 p +P − ln(P 2 + q 2 )
P 2 + q2 2