1

PHYS 705: Classical Mechanics

Canonical Transformation
 2

Canonical Variables and Hamiltonian Formalism

As we have seen, in the Hamiltonian Formulation of Mechanics,

 q j , p j are independent variables in phase space on equal footing

 The Hamilton’s Equation for q j , p j are “symmetric” (symplectic, later)

H H
q j  and p j  
p j q j
 This elegant formal structure of mechanics affords us the freedom in
selecting other (may be better) canonical variables as our phase space
“coordinates” and “momenta”

- As long as the new variables formally satisfy this abstract structure (the
form of the Hamilton’s Equations.
 3

Canonical Transformation
Recall (from hw) that the Euler-Lagrange Equation is invariant for a point
transformation: Q j  Q j ( q, t )
L d  L 
i.e., if we have,     0,
q j dt  q j 
L d  L 
then,     0,
Q j dt  Q j 
Now, the idea is to find a generalized (canonical) transformation in phase
space (not config. space) such that the Hamilton’s Equations are invariant !

Q j  Q j (q, p, t ) (In general, we look for

transformations which
Pj  Pj ( q, p, t ) are invertible.)
 4

Invariance of EL equation for Point Transformation

First look at the situation in config. space first:
d  L  L
Given:     0, and a point transformation: Q j  Q j ( q, t )
dt  q j
 q j
d  L  L
 Need to show:     0
dt  Q j  Q j
L L qi L qi
Formally, calculate:   (chain rule)
Q j i qi Q j i Q j
i q

L L qi L qi
 

Q j 
i qi Q j i Q j
i q

From the inverse point transformation equation qi  qi (Q, t ) , we have,

qi qi qi qi  qi

 0 and  qi   Qk 
Q j Q j Q j k Qk t
 5

Invariance of EL equation for Point Transformation

Q d  L  L
Forming the LHS of EL equation with j : LHS    
dt  Q j  Q j
d  L qi  L qi L qi
LHS    
dt  i qi Q j  i qi Q j i Q j
i q

 L d  qi  d  L  qi  L qi L qi

        
  

i  qi dt  Q j  dt  qi Q
 j  i qi Q j
i Q j
i q

 d  L  qi  L  qi L d  qi  L qi 

          
i  dt  qi  Q j  qi  Q j qi dt  Q j  q i Q j 

 6

Invariance of EL equation for Point Transformation

(exchange order of diff)

  d  L   L   qi L  d  qi  qi  

LHS             
i   dt  qi   qi   Q j qi  dt  Q j  Q j  

 dqi qi
 
0 Q j dt Q j
(Since that’s what given !) qi qi
  0
Q j Q j
d  L  L
LHS  0     0
dt  Q j  Q j
 7

Canonical Transformation
Now, back to phase space with q’s & p’s, we need to find the appropriate
(canonical) transformation Q j  Q j ( q, p, t ) and Pj  Pj (q, p, t )
such that there exist a transformed Hamiltonian K (Q, P, t )
with which the Hamilton’s Equations are satisfied:

K K
Q j  and Pj  
Pj Q j

(The form of the EOM must be invariant in the new coordinates.)

** It is important to further stated that the transformation considered

must also be problem-independent meaning that (Q, P ) must be canonical
coordinates for all system with the same number of dofs.
 8

Canonical Transformation
To see what this condition might say about our canonical transformation,
we need to go back to the Hamilton’s Principle:

Hamilton’s Principle: The motion of the system in configuration space is

such that the action I has a stationary value for the actual path, .i.e.,
t2

 I    Ldt  0
t1

Now, we need to extend this to the 2n-dimensional phase space

1. The integrant in the action integral must now be a function of the

independent conjugate variable q j , p j and their derivatives q j , p j
2. We will consider variations in all 2n phase space coordinates
 9

Hamilton’s Principle in Phase Space

1. To rewrite the integrant in terms of q j , p j , q j , p j , we will utilize the
definition for the Hamiltonian (or the inverse Legendre Transform):
H  p j q j  L  L  p j q j  H ( q, p, t ) (Einstein’s sum rule)

Substituting this into our variation equation, we have

t2 t2

 I    L dt     p j q j  H (q, p, t )  dt  0
t1 t1

2. The variations are now for n q j ' s and n p j ' s : (all q’s and p’s are independent)

The rewritten integrant ( q, q , p )  p j q j  H (q, p, t ) is formally a

function of q j , p j , q j , p j but in fact it does not depend on p j , i.e.  p j  0
This fact will proved to be useful later on.
again, we will required the variations for the q j to be zero at ends
 10

q j  q0 j   j
Hamilton’s Principle in Phase Space
p j  p0 j   j
Affecting the variations on all 2n variables q j , p j , we have,
t2

I     q j  q j   qj 's
d      d  d 
   q j  
  j  q j  
t1

  p j  p j  
 pj 's   d  d   dt  0

j  p j  p j  
 
As in previous discussion, the second term in the sum for  q j ' s can be
rewritten using integration by parts:
t2 t2
t2
  q j  q j  q j d   
 q  dt  q  
  dt  q dt
 j j t1   j 
t1 t1
 11

Hamilton’s Principle in Phase Space

 Previously, we have required the variations for the q j to be zero at end pts
t2 t2
t2
q j   q j  q j  q j d   
so that, 0  q  dt  q  
  dt  q dt
 t  t1 ,t2  j j t1   j 
t1 t1

So, the first sum with  q j ' s can be written as:

t2
   q j q j d    
  
 j  q j 
d  
 

 dt  q j 
d  dt

  
t1
t2
   d    q j
      q j dt where  q j  d
 j  q j dt  q j 
    
t1
 12

Hamilton’s Principle in Phase Space

Now, perform the same integration by parts to the corresponding term for  p j ' s
t2 t2
t2
we have,   p j  p j  p j d   
 p  dt  p  
  dt  p dt
 j j t1   j 
t1 t1
t2
 p j
 Note, since  p j  0 ,  0 without enforcing the
p j 
t1
variations for p j to be zero at end points.

This gives the result for the 2nd sum in the variation equation for p j ' s :
t2
   d    p j
   p j dt
    where  p j  d
j  p j dt  p 
   j  
t1
 13

Hamilton’s Principle in Phase Space

Putting both terms back together, we have:
t2

I     d      d    
d          q j        p j  dt  0
   j  q j dt  q j   j  p j dt  p
 
    j 
t1
1 2
Since both variations are independent, 1 and 2 must vanish independently !

d     (q, q , p)  p j q j  H (q, p, t )
1    0
dt  q j  q j
  H
  pj and 
q j q j q j
 H 
p j    0
 q j  H
p j   (one of the Hamilton’s
q j equations)
 14

Hamilton’s Principle in Phase Space

t2

I     d      d    
d          q j        p j  dt  0
   j  q j dt  q j   j  p j dt  p
 
    j 
t1
1 2

d     (q, q , p)  p j q j  H (q, p, t )
2    0
dt  p j  p j
  H
  0 and  q j 
p j p j p j
 H 
0   q j  0
p j  H
 q j  (2nd Hamilton’s
p j equations)
 15

Hamilton’s Principle in Phase Space

So, we have just shown that applying the Hamilton’s Principle in Phase
Space, the resulting dynamical equation is the Hamilton’s Equations.

H
p j  
q j

H
q j 
p j
 16

Hamilton’s Principle in Phase Space

Notice that a full time derivative of an arbitrary function F of ( q, p, t ) can
be put into the integrand of the action integral without affecting the
variations: t2
 dF 
  p j q j  H (q, p, t )   dt
 dt 
t1
t2
t2
dF
   p j q j  H (q, p, t )  dt  
 dt
t1
 dt
t1

t2

  dF  F t2  const
t
1
t1

Thus, when variation is taken, this constant term will not contribute !
 17

Canonical Transformation
Now , we come back to the question: When is a transformation to Q, P
canonical?

 We need Hamilton’s Equations to hold in both systems

This means that we need to have the following variational conditions:

   p j q j  H (q, p, t )  dt  0 AND    Pj Q j  K (Q, P, t )  dt  0

 For this to be true simultaneously, the integrands must equal

 And, from our previous slide, this is also true if they are differed by
a full time derivative of a function of any of the phase space variables
involved + time:
dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   q , p , Q , P, t 
dt
 18

Canonical Transformation
dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   q, p, Q, P, t  (*) (G9.11)
dt
F is called the Generating Function for the canonical
transformation:
Q j  Q j ( q, p, t )
(q j , p j )   Q j , Pj  : 
 Pj  Pj (q, p, t )

 As the name implies, different choice of F give us the ability to

generate different Canonical Transformation to get to different Q j , Pj  
 F is useful in specifying the exact form of the transformation if it
contains half of the old variables and half of the new variables. It,
then, acts as a bridge between the two sets of canonical variables.
 19

Canonical Transformation
dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   q, p, Q, P, t  (*) (G9.11)
dt
F is called the Generating Function for the canonical
transformation:
Q j  Q j ( q, p, t )
(q j , p j )   Q j , Pj  : 
 Pj  Pj (q, p, t )

 Depending on the form of the generating functions (which pair of

canonical variables being considered as the independent variables for
the Generating Function), we can classify canonical transformations
into four basic types.
 20

Canonical Transformation: 4 Types

dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   old , new, t 
dt

Type 1: F1 F1 F1

pj   q, Q, t  Pj    q, Q, t  K  H 
F  F1 ( q, Q, t ) q j Q j t

Type 2: F2 F2 F2

F  F2 (q, P, t )  Qi Pi
pj   q, P, t  j P  q, P, t 
Q  K H
t
q j j

Type 3: F3 F3 F3

F  F3 ( p, Q, t )  qi pi
qj    p, Q, t  Pj    p, Q, t  KH
t
p j Q j

Type 4: F F F4
q j   4  p, P, t  Q j  4  p, P, t  K H
F  F4 ( p, P, t )  qi pi  Qi Pi p j Pj t
 21

Canonical Transformation: Type 1

(old) (new)
Type 1: F  F1 (q, Q, t ) | F is a function of q and Q + time

Writing out the full time derivative for F, Eq (1) becomes:

F F F
p j q j  H  Pj Q j  K  1  1 q j  1 Q j (again E’s
t q j Q j sum rule)

Or, we can write the equation in differential form:

dq j  dQ j
(write out q j  , and Q j  and multiply the equation by dt )
dt dt

 F1   F   F1 
 p j   dq j   Pj  1
  j 
dQ  K  H   dt  0
 q j   Q j   t 
 22

Canonical Transformation: Type 1

Since all the q j , Q j are independent, their coefficients must vanish
independently. This gives the following set of equations:

F1 F F1
pj   q, Q, t  (C1) Pj   1  q, Q, t  (C 2) K H (C 3)
q j Q j t
These are the equations in the Table 9.1 in the book.

For a given specific expression for F1  q, Q, t  , e.g. F1  q, Q, t   q j Q j

 Eq. (C1) are n relations defining p j in terms of q j , Q j , t and they can

be inverted to get the 1st set of the canonical transformation:

In the specific example, we have:

F1
pj   q , Q, t   Q j  Qj  p j
q j
 23

Canonical Transformation: Type 1

 Eq. (C 2) are n relations defining Pj in terms of q j , Q j , t . Together with
our results for the Q j , the 2nd set of the canonical transformation
can be obtained.
Again, in the specific example, we have:

F1
Pj    q , Q, t    q j  Pj   q j
Q j
 Eq. (C 3) gives the connection between K and H:
F1
KH  KH
t
(note: K (Q, P, t ) is a function of the new variables so that the RHS needs
to be re-express in terms of Q j , Pj using the canonical transformation.)
 24

Canonical Transformation: Type 1

In summary, for the specific example of a Type 1 generating function:

F1  q, Q, t   q j Q j
We have the following:
Qj  p j
Canonical Transformation
Pj   q j

and K H Transformed Hamiltonian

Note: this example results in basically swapping the generalized coordinates with
their conjugate momenta in their dynamical role and this exercise demonstrates
that swapping them basically results in the same situation !

 Emphasizing the equal role for q and p in Hamiltonian Formalism !

 25

Canonical Transformation: Type 2

(old) (new)
Type 2: F  F2 ( q, P, t )  Q j Pj , where F2 is a function of q and P + time

(One can think of F2 as the Legendre transform of F (q, Q, t ) in

exchanging the variables Q and P.)

Substituting into our defining equation for canonical transformation, Eq. (1):

F2 F2
 j KKdF F2 
p j q j  H  PPjjQ
Q  
q j  Pj  Pj Q j  Pj Q j
dtt q j Pj
j

Again, writing the equation in differential form:

 F2   F2   F2 

 j
p  
 j  j
dq Q  
 j 
dP K  H   dt  0
 q j   Pj   t 
 26

Canonical Transformation: Type 2

Since all the q j , Pj are independent, their coefficients must vanish
independently. This gives the following set of equations:

F2 F F2
pj   q, P, t  Q j  2  q, P, t  K H
q j Pj t

For a given specific expression for F2  q, P, t  , e.g. F2  q, P, t   q j Pj

F2 F2
pj   q, P, t   Pj Qj   q, P, t   q j
q j Pj
 Pj  p j  Qj  q j K H

Thus, the identity transformation is also a canonical transformation !

 27

Canonical Transformation: Type 2

Let consider a slightly more general example for type 2: F  F2  q, P, t   Q j Pj

with F2  q, P, t   f  q1 , , qn , t  Pj  g  q1 , , qn , t 

where f and g are function of q’s only + time

Going through the same procedure, we will get:

F2 F2
pj  Qj 
q j Pj f g
KH Pj 
f g t t
 pj  Pj   Q j  f  q1 , qn , t 
q j q j

Notice that the Q equation is the general point transformation in the

configuration space. In order for this to be canonical, the P and H
transformations must be handled carefully (not necessary simple functions).
 28

Canonical Transformation: Summary

The remaining two basic types are Legendre transformation of the remaining
two variables:
F  F3 ( p, Q, t )  p j q j q p
F  F4 ( p, P, t )  p j q j  Q j Pj q  p&Q  P
(Results are summarized in Table 9.1 on p. 373 in Goldstein .)

Canonical Transformations form a group with the following properties:

1. The identity transformation is canonical (type 2 example)

2. If a transformation is canonical, so is its inverse
3. Two successive canonical transformations (“product”) is canonical
4. The product operation is associative
 29

Q j  Q j ( q, p, t )
Canonical Transformation: 4 Types Pj  Pj (q, p, t )

dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   old , new, t 
dt
dep var
Type 1: F1 F1 F1
pj   q, Q, t  Pj    q, Q, t  K  H 
F  F1 ( q, Q, t ) q j Q j t
ind var
Type 2: F2 F2 F2
F  F2 (q, P, t )  Qi Pi
pj   q, P, t  j P  q, P, t 
Q  K H
t
q j j

Type 3: F3 F3 F3

F  F3 ( p, Q, t )  qi pi
qj    p, Q, t  Pj    p, Q, t  KH
t
p j Q j

Type 4: F F F4
q j   4  p, P, t  Q j  4  p, P , t  K H
F  F4 ( p, P, t )  qi pi  Qi Pi p j Pj t
 30

Canonical Transformation (more)

If we are given a canonical transformation
Q j  Q j ( q , p, t )
(*)
Pj  Pj (q, p, t )
How do we find the appropriate generating function F ?

- Let say, we wish to find a generating function of the 1st type, i.e., F  F1 ( q, Q, t )

(Note: generating function of the other types can be obtain through an

appropriate Legendre transformation.)

- Since our chosen generating function (1st type) depends on q, Q, and t

explicitly, we will rewrite our p and P in terms of q and Q using Eq. (*):

p j  p j ( q, Q, t ) Pj  Pj (q, Q, t )
 31

Canonical Transformation (more)

Now, from the pair of equations for the Generating Function Derivatives
(Table 9.1), we form the following diff eqs,
F1  q, Q, t 
p j  p j ( q , Q, t ) 
q j
F  q, Q, t 
Pj  Pj (q, Q, t )   1
Q j
F1 (q, Q, t ) can then be obtained by directly integrating the above equations
and combining the resulting expressions.

Note: Taking the respective partials of q and Q of the above equations,

Pj   F1  q, Q, t     F1  q, Q, t   pi

       
qi qi  Q j  Q j  q j  Q j
 32

Canonical Transformation (more)

Now, from the pair of equations for the Generating Function Derivatives
(Table 9.1), we form the following diff eqs,
F1  q, Q, t 
p j  p j ( q , Q, t ) 
q j
F  q, Q, t 
Pj  Pj (q, Q, t )   1
Q j
F1 (q, Q, t ) can then be obtained by directly integrating the above equations
and combining the resulting expressions.

Note: Since dF1 is an exact differential wrt q and Q, so the two exps are equal,

Pj  2 F1  2 F1 pi (We will give the full

   
qi qi Q j Q j qi Q j list of relations later.)
 33

Canonical Transformation (more)

Example (G8.2): We are given the following canonical transformation for a
system with 1 dof:

Q  Q (q, p )  q cos   p sin  (HW: showing this

P  P( q, p)  q sin   p cos  trans. is canonical)

(Q and P is being rotated in phase space from q and p by an angle )

We seek a generating function of the 1st kind: F1 ( q, Q )

First, notice that the cross-second derivatives for F1 are equal as required for a
canonical transformation:
  F1       Q  1
   p     q cot    
Q  q   Q  Q  si n   sin 
  F1       q  1
   P     Q cot    
q  Q   q  q  sin   sin 
 34

Canonical Transformation (more)

Rewrite the transformation in terms of q and Q (indep. vars of F1 ):
F1 F1
 p  p ( q, Q )   P  P ( q, Q )
q Q
Q  Q cos  
  q cot   q sin     q  cos 
sin   sin  sin  
q
  Q cot 
sin 
Now, integrating the two partial differential equations:
Qq q 2 qQ Q 2
F1    cot   h  Q  F1    cot   g  q 
sin  2 sin  2
Comparing these two expression, one possible solution for F1 is,

Qq 1 2
F1  q, Q      q  Q 2  cot 
sin  2
 35

Canonical Transformation (more)

Now, let say we want to fine a Type-2 Generating function F2 (q, P, t ) for this
problem…
As we have discussed previously, we can directly use the fact that F2 is the
Legendre transform of F1,
F1  F2 (q, P, t )  Q j Pj

F2 (q, P, t )  F1 ( q, Q, t )  QP
Qq 1 2
F2  q, P      q  Q 2  cot   QP
sin  2
Now, from the CT, we can write Q by q and P (F2 should be in q & P):
q
Q  q cos   p sin  Q  P tan 
P  q sin   p cos  cos 
 36

Canonical Transformation (more)

As we have discussed previously, we can directly use the fact that F2 is the
Legendre transform of F1,
F1  F2 ( q, P, t )  Q j Pj

F2 (q, P, t )  F1 (q, Q, t )  QP
Qq 1 2
F2  q, P      q  Q 2  cot   QP
sin  2
This then gives:

 1 2  q  
2
 q  q
F2  q, P    P    P tan     q    P tan    cot 
 sin    cos   2  cos   
Q
 37

Canonical Transformation (more)

As we have discussed previously, we can directly use the fact that F2 is the
Legendre transform of F1,
F1  F2 (q, P, t )  Q j Pj
 q  q  1   q  
2

F2  q, P    P    P tan     q  
2
 P tan    cot 
 sin    cos   2   cos   

2qP q2
  P 2 tan 
cos  sin  cos 
1 2  q2 2qP 
 q cot      P tan   
2

2  cos  sin  cos  

 38

Canonical Transformation (more)

As we have discussed previously, we can directly use the fact that F2 is the
Legendre transform of F1,
F1  F2 ( q, P, t )  Q j Pj
 1 2  q  
2
 q  q
F2  q, P    P    P tan     q    P tan    cot 
 sin    cos   2  cos   

2 qP q2
  P 2 tan 
cos  sin  cos 
1 2  q2 2qP 
 q cot      P tan   
2
  
2  cos  sin  cos  
 39

Canonical Transformation (more)

As we have discussed previously, we can directly use the fact that F2 is the
Legendre transform of F1,
F1  F2 ( q, P, t )  Q j Pj
 q  q 
 1 2  q  
2

F2  q, P    P    P tan     q    P tan    cot 

 sin    cos   2  cos   

qP 1
Finally, F2  q, P     q 2  P 2  tan 
cos  2
 40

Canonical Transformation (more)

Alternatively, we can substitute F into Eq. (*)G9.11 , results in replacing the Pj Q j
term by Q j Pj in our condition for a canonical transformation,

F  F2 (q, P, t )  Q j Pj
dF
p j q j  H  Q j Pj  K 
dt

Recall, this procedure gives us the two partial derivatives relations for F2:

F2  q, P, t  F2  q, P, t  [Or, use the Table]

pj  Qj 
q j Pj
 41

Q  Q( q, p )  q cos   p sin 
Canonical Transformation (more) P  P ( q, p )  q sin   p cos 

To solve for F2 ( q, P, t ) in our example, again, we rewrite our given canonical

transformation in q and P explicitly.

F2 F2
 p  p (q, P )  Q  Q ( q, P )
q P sin  
P  P
  q tan   q cos    q  sin 
cos   cos  cos  
q
  P tan 
cos 
Integrating and combining give,

qP 1
F2  q, P     q 2  P 2  tan 
cos  2
 42

Q  Q( q, p )  q cos   p sin 
Canonical Transformation (more) P  P (q, p )  q sin   p cos 
Notice that when   0 , sin   0
so that our coordinate transformation is just the identity
transformation: Q  q and P  p

p, P CANNOT be written explicitly in terms of q and Q !

so our assumption for using the type 1 generating function

(with q and Q as indp var) cannot be fulfilled.

Consequently, F1  q, Q  blow up and cannot be used to derive the

canonical transformation:
Qq 1 2
F1  q, Q      q  Q 2  cot    as   0
sin  2
But, using a Type 2 generating function will work.
 43

Q  Q(q, p)  q cos   p sin 

Canonical Transformation (more) P  P (q, p )  q sin   p cos 

Similarly, we can see that when   , cos   0
2
our coordinate transformation is a coordinate switch Q   p , P  q

p, Q CANNOT be written explicitly in terms of q and P !

so the assumption for using the type 2 generating function

(with q and P as indp var) cannot be fulfilled.

Consequently, F2  q, P  blow up and cannot be used to derive the

canonical transformation:

qP 1
F2  q, P     q 2  P 2  tan    as   0
cos  2

But, using a Type 1 generating function will work in this case.

 44

Canonical Transformation (more)

- A suitable generating function doesn’t have to conform to only one of
the four types for all the degrees of freedom in a given problem !
- There can also be more than one solution for a given CT
- First , we need to choose a suitable set of independent variables for the
generating function.
For a generating function to be useful, it should depends on half of
the old and half of the new variables
As we have done in the previous example, the procedure in solving
for F involves integrating the partial derivative relations resulted from
“consistence” considerations using the main condition for a canonical
transformation, i.e., dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )  (G9.11)
dt
 45

Canonical Transformation (more)

For these partial derivative relations to be solvable, one must be
able to feed-in 2n independent coordinate relations (from the given
CT) in terms of a chosen set of ½ new + ½ old variables.

- In general, one can use ANY one of the four types of generating
functions for the canonical transformation as long as the RHS of the
transformation can be written in terms of the associated pairs of phase
space coordinates: (q, Q, t), (q, P, t), (q, Q, t), or (p, P, t).
T1 T2 T3 T4
- On the other hand, if the transformation is such that the RHS cannot
written in term of a particular pair: (q, Q, t), (q, P, t), (q, Q, t), or (p, P, t),
then that associated type of generating functions cannot be used.
 46

Canonical Transformation: an example with two

dofs

- To see in practice how this might work… Let say, we have the following
transformation involving 2 dofs:  q1 , p1 , q2 , p2    Q1 , P1 , Q2 , P2 

Q1  q1 (1a) Q2  p2 (2a )
P1  p1 (1b) P2   q2 (2b)

- As we will see, this will involve a mixture of two different basic types.
 47

Canonical Transformation: an example with two

dofs
Q1  q1 (1a ) Q2  p2 (2a )
P1  p1 (1b) P2   q2 (2b)

- First, let see if we can use the simplest type (type 1) for both dofs, i.e., F
will depend only on the q-Q’s:
F (q1 , q2 , Q1 , Q2 , t )
Notice that Eq (1a) is a relation linking q1 , Q1 only , they CANNOT
both be independent variables  Type 1 (only) WON’T work !

- As an alternative, we can try to use the set p1 , q2 , Q1 , Q2 as our independent

variables. This will give an F which is a mixture of Type 3 and 1.
(In Goldstein (p. 377), another alternative was using q1 , q2 , P1 , Q2 resulted in a
different generating function which is a mixture of Type 2 and 1.)
 48

Canonical Transformation: an example with two

dofs
From our CT, we can write down the following relations:
Dependent variables Independent variables
q1 , p2 , P1 , P2 p1 , q2 , Q1 , Q2

q1  Q1
p2  Q2
(*)
P1  p1
P2   q2

Now, with this set of ½ new + ½ old independent variable chosen, we need to
derive the set of partial derivative conditions by substituting F ( p1 , q2 , Q1 , Q2 , t )
into Eq. 9.11 (or look them up from the Table).
 49

Canonical Transformation: an example with two

dofs
The explicit independent variables (those appear in the differentials) in Eq.
9.11 are the q-Q’s. To do the conversion:

q1 , q2 , Q1 , Q2 (Eq. 9.11’s explicit ind vars)


p1 , q2 , Q1 , Q2 (our preferred ind vars)
we will use the following Legendre transformation: F  F '( p1 , q2 , Q1 , Q2 , t )  q1 p1

Substituting this into Eq. 9.11, we have:

p1q1  p2 q2  H  PQ  P Q  K  dF

1 1 2 2
dt
F ' F ' F '  F '  F '
 P1Q1  P2Q 2  K  p1  q2  Q1  Q2  q1 p1  p1q1 
p1 q2 Q1 Q2 t
 50

Canonical Transformation: an example with two

dofs
p2 q2  H  PQ  P Q  K  F ' p  F ' q  F ' Q  F ' Q  q p  F '
1 1 2 2 1 2 1 2 1 1
p1 q2 Q1 Q2 t
Comparing terms, we have the following conditions:

F ' F '
q1   P1  
p1 Q1 F '
K H
F ' F ' t
p2  P2  
q2 Q2

As advertised, this is a mixture of Type 3 and 1 of the basic CT.

 51

Canonical Transformation: an example with two

dofs q 1  Q1
p2  Q2
Substituting our coordinates transformation [Eq. (*)] into the partial
P1  p1
derivative relations, we have : P2   q2
F '
  q1  Q1 F '  Q1 p1  f (q2 , Q1 , Q2 )
p1
F '
  P1   p1 F '   p1Q1  g ( p1 , q2 , Q2 )
Q1 F '   p1Q1  q2Q2
F '
  P2    q2  F '  q2Q2  h( p1 , q2 , Q1 )
Q2
F ' F '  Q2 q2  k ( p1 , Q1 , Q2 )
 p2  Q2
q2
(Note: Choosing q1 , q2 , P1 , Q2 instead, Goldstein has F ''  q1 P1  q2Q2 . Both
of these are valid generating functions.)
 52

Q j  Q j ( q, p, t )
Canonical Transformation: Review Pj  Pj (q, p, t )
dF
p j q j  H (q, p, t )  Pj Q j  K (Q, P, t )   old , new, t 
dt

Type 1: F1 F1 F1

pj   q, Q, t  Pj    q, Q, t  K  H 
F  F1 (q, Q, t ) q j Q j t

Type 2: F2 F2 F2

F  F2 (q, P, t )  Qi Pi
pj   q, P, t  j P  q, P, t 
Q  K H
t
q j j

Type 3: F3 F3 F3

F  F3 ( p, Q, t )  qi pi
qj    p, Q, t  Pj    p, Q, t  KH
t
p j Q j

Type 4: F F F4
q j   4  p, P, t  Q j  4  p, P, t  K H
F  F4 ( p, P, t )  qi pi  Qi Pi p j Pj t
 53

Canonical Transformation: Review

- Generating function is useful as a bridge to link half of the original set

of coordinates (either q or p) to another half of the new set (either Q or P).

- In general, one can use ANY one of the four types of generating
functions for the canonical transformation as long as the
transformation can be written in terms of the associated pairs of phase
space coordinates: (q, Q, t), (q, P, t), (q, Q, t), or (p, P, t).

- On the other hand, if the transformation is such that it cannot be written

in term of a particular pair: (q, Q, t), (q, P, t), (q, Q, t), or (p, P, t), then
that associated type of generating functions cannot be used.

- the procedure in solving for F involves integrating the resulting partial

derivative relations from the CT condition