CHINESE JOURNAL OF PHYSICS VOL. 45 , NO.

4 AUGUST 2007

Invariance of the Canonical Quantization Prescription Under Classical

Canonical Transformations

Gin-Yih Tsaur1 and Jyhpyng Wang2

1
Department of Mathematics, Tunghai University, Taichung 407, Taiwan
2
Institute of Atomic and Molecular Sciences, Academia Sinica,
P. O. Box 23-166, Taipei 106, Taiwan
(Received January 7, 2007)
Dirac’s postulate of canonical quantization, [p̂i , q̂j ] = −ih̄δij for conjugate canonical vari-
ables, has been the most concise and general prescription on how to quantize a classical
system. Since classical systems described by variables connected with canonical transforma-
tions are equivalent, [p̂i , q̂j ] = −ih̄δij must remain invariant under classical canonical trans-
formations. This invariance has not been proved except for the limited class of cascaded
infinitesimal transformations. In this paper it is shown that if (P̂i , Q̂j ) are related to (p̂i , q̂j )
by a classical canonical transformation, then [p̂i , q̂j ] = −ih̄δij implies [P̂i , Q̂j ] = −ih̄δij . In
other words, the canonical quantization prescription is invariant for variables connected with
classical canonical transformations.
PACS numbers: 03.65.-w, 04.60.Ds

I. INTRODUCTION

In classical mechanics, canonical transformations are the transformations (pi , qj ) →

(Pi , Qj ) which preserve the Poisson brackets; {pi , qj } = −δij ⇒ {Pi , Qj } = −δij . By not-
ing the analogy between Poisson bracket and commutator, Dirac postulated that classical
systems can be quantized by replacing the Poisson brackets {pi , qj } = −δij with the com-
mutation relations [p̂i , q̂j ] = −ih̄δij . This procedure, known as the canonical quantization,
has been the most concise and general prescription for quantizing a classical system [1].
Since classical systems described by variables connected with canonical transformations
are equivalent, it would be interesting to ask: Does the canonical quantization procedure
preserves this equivalence?
It has been shown for classical canonical transformations that are infinitesimally close
to the identity transformation, the corresponding transformations in quantum mechanics are
unitary [2]. These transformations can be cascaded to form transformation groups. Because
unitary transformations P̂i = Û p̂i Û −1 , Q̂j = Û q̂j Û −1 preserve the commutation relations,
for such transformation groups the invariance of the canonical quantization procedure is
obvious. However, not all classical canonical transformations have corresponding unitary
transformations. Many important transformations, such as transformation to the polar or
spherical coordinates, or transformation to the action-angle variables, do not belong to this
category [2]. Therefore, in the discussion of canonical quantization, one may not wish to
be limited to canonical transformations that have corresponding unitary transformations

http://PSROC.phys.ntu.edu.tw/cjp 425 c 2007 THE PHYSICAL SOCIETY

OF THE REPUBLIC OF CHINA
426 INVARIANCE OF THE CANONICAL QUANTIZATION . . . VOL. 45

in quantum mechanics.
The Stone-von Neumann uniqueness theorem [3] shows all quantum canonical trans-
formations that yield self-adjoint variables are unitary transformations. For this reason,
unitary transformations have been used synonymously as quantum canonical transforma-
tions. However, in the past decade applications of a broader class in quantum mechanical
calculations have been explored [4–6]. Non-unitary transformations are used with a corre-
sponding change of measure density to conserve probability. Even though non-self-adjoint
operators do not represent physical variables, they can be useful in the mid-steps of quantum
mechanical calculations [7–9], much like complex analysis can be useful in the evaluation
of real integrals.
In this paper, we show that the canonical quantization prescription is invariant for
variables connected with all classical canonical transformations that can be written in op-
erator forms with well defined ordering. Aside from its importance to the foundation of
quantum mechanics, the invariance of the canonical quantization prescription can be use-
ful in quantum dynamical analysis. For example, by changing to an appropriate set of
variables, the equation of motion could be solved systematically [7–9], as is the case for
classical mechanics. It is also interesting to note that in the past few decades important
advancements in classical nonlinear dynamics, such as the superconvergent perturbation
theory [10, 11] and the soliton theories [12], are based on nonlinear transformation of vari-
ables. Establishing the quantization conditions for general variables may help bringing
these advancements to the quantum domain.

II. INVARIANCE OF THE CANONICAL QUANTIZATION PRESCRIPTION

We shall present a proof to show that the canonical quantization prescription is

invariant under transformations that satisfy the symplectic condition, which is a necessary
and sufficient condition for classical canonical transformation [13]. Before we proceed,
the definition of operator differentiation should be clarified. A general theory of operator
differentiation can be found in Ref. [14], where the derivative of an operator function is
defined as
∂ F̂ d ˆ · · · , p̂N , q̂1 , · · · , q̂N )
= F (p̂1 , · · · , p̂k + tI, ,
∂ p̂k dt t=0

∂ F̂ d ˆ · · · , q̂N )
= F (p̂1 , · · · , p̂N , q̂1 , · · · , q̂k + tI, ,
∂ q̂k dt t=0

where Iˆ is the identity operator. The definition leads to the same rules of c-number differ-
entiation. For example,
∂ d h ˆ 2 q̂ 3
i
p̂2 q̂ 3 = = 2p̂ q̂ 3 ,


(p̂ + tI)
∂ p̂ dt t=0
∂ d h
ˆ3
i
p̂2 q̂ 3 = p̂2 (q̂ + tI) = 3p̂2 q̂ 2 ,


∂ q̂ dt t=0
VOL. 45 GIN-YIH TSAUR AND JYHPYNG WANG 427

and the order of differentiation can be interchanged,

∂ ∂ ∂ ∂
p̂2 q̂ 3 = p̂2 q̂ 3 = 6p̂q̂ 2 .



∂ q̂ ∂ p̂ ∂ p̂ ∂ q̂
The rule of multiplication is also satisfied,
∂   d h ˆ q̂)G(p̂ + tI,
ˆ q̂)
i
F̂ Ĝ = F (p̂ + tI,
∂ p̂ dt t=0
d ˆ q̂) d ˆ q̂)
= F (p̂ + tI, Ĝ + F̂ G(p̂ + tI,
dt t=0 dt t=0
∂ F̂ ∂ Ĝ
= Ĝ + F̂ ,
∂ p̂ ∂ p̂
and similarly

∂   ∂ F̂ ∂ Ĝ
F̂ Ĝ = Ĝ + F̂ .
∂ q̂ ∂ q̂ ∂ q̂
In classical mechanics, all canonical transformations (pi , qj ) → (Pi , Qj ) satisfy the
symplectic condition
∂p ∂q ∂Pi ∂Pi
! !
− ∂Qji − ∂Qji ∂qj − ∂pj
∂pj ∂qj = ∂Qi ∂Qi .
∂P i ∂P i ∂qj − ∂pj

This condition has been proved to be a necessary and sufficient condition for preserving the
Poisson brackets [13], i.e.,

{pi , qj } = −δij , {pi , pj } = {qi , qj } = 0

⇒ {Pi , Qj } = −δij , {Pi , Pj } = {Qi , Qj } = 0.

In quantum mechanics, the corresponding transformations are (p̂i , q̂j ) → (P̂i , Q̂j ) that sat-
isfy the symplectic condition
 
∂ p̂ ∂ q̂ ! ∂ P̂i ∂ P̂i
− j − j −
∂ Q̂i
∂ p̂j
∂ Q̂i
∂ q̂j =  ∂∂Q̂
q̂j ∂ p̂j 
∂ Q̂i
. (1)
∂ P̂ ∂ P̂ ∂ q̂j
i
− ∂ p̂j
i i

We shall prove that Eq. (1) is a necessary and sufficient condition for preserving the com-
mutation relations, i.e.,

[p̂i , q̂j ] = cδij , [p̂i , p̂j ] = [q̂i , q̂j ] = 0, (2)

⇒ [P̂i , Q̂j ] = cδij , [P̂i , P̂j ] = [Q̂i , Q̂j ] = 0, (3)

where the constant c can be arbitrary, even though in reality c = −ih̄.

428 INVARIANCE OF THE CANONICAL QUANTIZATION . . . VOL. 45

We use the following notation and lemma to construct the proof.

r̂ = (r̂1 , · · · , r̂N , r̂N +1 , · · · , r̂2N ) ≡ (p̂1 , · · · , p̂N , q̂1 , · · · , q̂N ),

R̂ = (R̂1 , · · · , R̂N , R̂N +1 , · · · , R̂2N ) ≡ (P̂1 , · · · , P̂N , Q̂1 , · · · , Q̂N ).

Lemma: Let r̂ represent an array of canonical variables (r̂1 , r̂2 , · · · , r̂2N ). If [r̂i , r̂k ] = cki
are constants for all i, k, then for any differentiable operator function F̂ = F (r̂),
!
  ∂ F̂
[F̂ , r̂k ] = (cki ) . (4)
∂r̂i

Proof: In the same spirit as substituting p̂ with −ih̄(d/dq) in the Schrödinger equa-
tion,
P2Nthe condition [r̂i , r̂k ] = cki can be satisfied by substituting the operator r̂k with
− j=1 ckj (∂/∂r̂j ). Then for any differentiable operator functions F̂ and Ĝ
2N 2N 2N
!
X ∂ Ĝ X ∂(F̂ Ĝ) X ∂ F̂
[F̂ , r̂k ]Ĝ = −F̂ cki + cki = cki Ĝ.
∂r̂i ∂r̂i ∂r̂i
i=1 i=1 i=1
P2N
Because Ĝ is arbitrary, one has [F̂ , r̂k ] = i=1 cki (∂ F̂ /∂r̂i ). Eq. (4) is its matrix form.
First we prove the symplectic condition is the necessary condition, namely Eqs. (2)
and (3) imply Eq. (1). As the lemma above, for any differentiable operator functions
F̂ = F (r̂) and Ĝ = G(R̂), Eq. (2) implies

∂ F̂ 1 ∂ F̂ 1
= − [F̂ , p̂j ], = [F̂ , q̂j ], (5)
∂ q̂j c ∂ p̂j c
and Eq. (3) implies

∂ Ĝ 1 ∂ Ĝ 1
= − [Ĝ, P̂i ], = [Ĝ, Q̂i ]. (6)
∂ Q̂i c ∂ P̂i c

Let F̂ = P̂i or Q̂i , the right hand side of Eq. (1) becomes
 
∂ P̂i ∂ P̂i
−
 
 ∂ q̂ j ∂ p̂ j  = − 1 [P̂i , p̂ j ] [P̂i , q̂ j ]
∂ Q̂i ∂ Q̂i c [Q̂i , p̂j ] [Q̂i , q̂j ]
∂ q̂j − ∂ p̂j
 
1 [p̂j , P̂i ] [q̂j , P̂i ]
= . (7)
c [p̂j , Q̂i ] [q̂j , Q̂i ]

Let Ĝ = p̂j or q̂j , the left hand side of Eq. (1) becomes
∂ p̂ ∂ q̂ !
− ∂ Q̂j − ∂ Q̂j
 
1 [p̂j , P̂i ] [q̂j , P̂i ]
∂ p̂j
i
∂ q̂j
i
= . (8)
c [p̂j , Q̂i ] [q̂j , Q̂i ]
∂ P̂i ∂ P̂i

The symplectic condition is satisfied.

VOL. 45 GIN-YIH TSAUR AND JYHPYNG WANG 429

Next we prove the symplectic condition is the sufficient condition, namely Eqs. (1)
and (2) imply Eq. (3). As mentioned, Eq. (2) implies Eq. (5) and hence Eq. (7). Namely,
the right hand side of Eq. (1) is
 
∂ P̂i ∂ P̂i
−
 
 ∂ q̂j ∂ p̂j  1 [p̂j , P̂i ] [q̂j , P̂i ] 1 
∂ Q̂i
= = [r̂ j , R̂i ] , (9)
∂ q̂j − ∂ Q̂i
∂ p̂j
c [p̂j , Q̂i ] [q̂j , Q̂i ] c

where the subscripts of r̂ and R̂ run from 1 to 2N . The left hand side of Eq. (1) can also
be written in r̂ and R̂ form as
∂ p̂ ∂ q̂ !
− ∂ Q̂j − ∂ Q̂j
  
0 −1 ∂r̂j
∂ p̂j
i
∂ q̂j
i = , (10)
1 0 ∂ R̂i
∂ P̂i ∂ P̂i

where 1 is the N × N identity matrix and 0 is the N × N null matrix. Therefore Eq. (1)
implies
  
0 −1 ∂r̂j 1 
= [r̂j , R̂i ] . (11)
1 0 ∂ R̂i c

What we want to prove is [R̂i , R̂k ] = cki with

 
0 −1
(cki ) = c , (12)
1 0

which is equivalent to Eq. (3). First we show that all cki are constants by proving
(∂/∂r̂l )[R̂i , R̂k ] = 0 for all i, k, l. For simplicity, we only show (∂/∂ q̂l )[P̂i , Q̂k ] = 0. Proofs
for (∂/∂ p̂l )[P̂i , Q̂k ] = 0, (∂/∂r̂l )[P̂i , P̂k ] = 0, and (∂/∂r̂l )[Q̂i , Q̂k ] = 0 are similar.
" # " #
∂ ∂ P̂i ∂ Q̂k
[P̂i , Q̂k ] = , Q̂k + P̂i ,
∂ q̂l ∂ q̂l ∂ q̂l
   
∂ p̂l ∂ p̂l
= − , Q̂k + P̂i , (by (1))
∂ Q̂i ∂ P̂k
∂ ∂
= − [p̂l , Q̂k ] + [P̂i , p̂l ]
∂ Q̂i ∂ P̂k
∂ ∂ Q̂k ∂ ∂ P̂i
= −c −c (by (9))
∂ Q̂i ∂ q̂l ∂ P̂k ∂ q̂l
∂ ∂ p̂l ∂ ∂ p̂l
= −c +c (by (1))
∂ Q̂i ∂ P̂k ∂ P̂k ∂ Q̂i
= 0.

The third equality above is because ∂ Q̂k /∂ Q̂i = ∂ P̂i /∂ P̂k = 0 (for i 6= k) or 1 (for i = k) and
[p̂l , 0] = [p̂l , 1] = 0. Now that [R̂i , R̂k ] = cki are constants, according to the lemma above,
430 INVARIANCE OF THE CANONICAL QUANTIZATION . . . VOL. 45

for any differentiable operator function F̂ = F (R̂), one has ([F̂ , R̂k ]) = (cki )(∂ F̂ /∂ R̂i ). Let
F̂ = r̂j , it is
 
  ∂r̂j
[r̂j , R̂k ] = (cki ) . (13)
∂ R̂i
Substituting this into the right hand side of Eq. (11), one obtains Eq. (12). That is,
[P̂i , Q̂j ] = cδij , [P̂i , P̂j ] = 0, [Q̂i , Q̂j ] = 0.

III. DISCUSSION

A potential application of the invariance of the quantization procedure is quanti-

zation in the action-angle variables. Action-angle variables play a special role in classi-
cal mechanics. For integrable systems the Hamiltonians H = H(p, q) can be reduced by
canonical transformations to H = H(J), where J is the action variable and θ is its con-
jugate angle variable [13]. As a result, the equation of motion becomes greatly simplified:
J˙ = −∂H/∂θ = 0 implies J is a constant of motion, and θ̇ = ∂H/∂J =constant implies
that θ evolves linearly in time. Moreover, if the Hamiltonian changes with a slowly vary-
ing parameter, it can be shown that the cycle-averaged action variables remain invariant
(adiabatic invariance theorem [15]). This is why action-angle variables are important in
the analysis of dynamic systems. In quantum mechanics, the equation of motion in the
action-angle space becomes
∂
ih̄ ψ(θ, t) = H(Jˆ)ψ(θ, t), (14)
∂t

where Jˆ is the action operator and its eigenvalue will be represented by J. From the invari-
ance of the quantization procedure, one may replace Jˆ by −ih̄(∂/∂θ) and its eigenfunction
is simply exp(iJθ/h̄). Since H(J) ˆ commutes with J, ˆ they share the same eigenfunction
and the corresponding eigenenergy is E = H(J). In the real space, the choice of E is de-
termined by the boundary conditions or the normalization condition of the eigenfunctions,
however, in phase spaces made of general abstract canonical variables the normalization
condition may not be obvious, particularly because in such phase spaces the measure den-
sity may not be the same as the real space [6]. A remedy is to transform the wavefunction
exp(iJθ/h̄) back to the real space, then impose the boundary condition, as demonstrated
in Ref. [8, 9]. Although it is not guaranteed that the transformations from the real space to
the action-angle space and their inverse transformations can always be found in analytical
forms, in Ref. [8] it was shown that the method can be applied successfully to all known
one-dimensional solvable models.
In summary, by presenting a proof for the invariance of the canonical quantization
prescription under classical canonical transformations, we may have filled a missing link in
the theory of correspondence between classical and quantum mechanics. The proof provides
a more solid basis for extending the quantum theory to general abstract variables.
VOL. 45 GIN-YIH TSAUR AND JYHPYNG WANG 431

References

[1] P. M. Dirac, The Principles of Quantum Mechanics, 4th ed. (Clarendon Press, Oxford, 1958),
Sec. 21.
[2] E. Merzbacher, Quantum Mechanics, 2nd ed. (John Wiley & Sons Inc., 1970), pp. 342-345.
[3] M. Reed and B. Simon, Methods of modern mathematical physics, I: Functional analysis, revised
and enlarged edition (Academic Press, New York, 1972), Chap. 8.
[4] F. Leyvraz and T. H. Seligman, J. Math. Phys. 30, 2512 (1989).
[5] J. Deenen, J. Phys. A 24, 3851 (1991).
[6] A. Anderson, Phys. Lett. B 305, 67 (1993).
[7] A. Anderson, Ann. Phys. 232, 292 (1994).
[8] G. Tsaur and J. Wang, Phys. Rev. A 65, 012104 (2001).
[9] G. Tsaur and J. Wang, Am. J. Phys. 74, 606 (2006).
[10] A. N. Kolmogorov, Dokl. Akad. Nauk. SSSR 98, 527 (1954).
[11] B. McNamara, J. Math. Phys. 19, 2154 (1978).
[12] P. G. Drazin and R. S. Johnson, Solitons: an intrduction (Cambridge University Press, New
York, 1996).
[13] H. Goldstein, Classical Mechanics, 2nd ed. (Addison-Wesley Publishing Company Inc., 1980),
Chapter 9.
[14] R. Bhatia, D. Singh, and K. B. Sinha, Commun. Math. Phys. 191, 603 (1998).
[15] L. D. Landau and E. M. Lifshitz, Mechanics, 3rd ed. (Pergamon Press Ltd., 1976), Sec. 49.