1

38. Quantization of the Free Electromagnetic Field

We recall that quantum mechanics was initiated by Planck's suggestion that the energy, E, of the electromagnetic field was quantized in terms of the frequency, , in discrete units called photons according to E = . Yet up to this point we have generally regarded the electromagnetic field as a classical entity; we must therefore construct a method to quantize it. The method we will use is to rewrite the electromagnetic field so that it has the formal structure of a superposition of harmonic oscillators. We will then quantize these oscillators as if they were ordinary mechanical oscillators (Dirac 1929) In the Hamiltonian formulation of quantum mechanics the electromagnetic field enters the wave equations for particles only through the vector and scalar potentials, A, and . Since we will later want to couple matter fields to quantized electromagnetic fields we will formulate the quantized theory of the electromagnetic field in terms of A. We start by recalling the four Maxwell equations
! " E = 4 #$; 1 'B !%E=& ; c 't ! " B = 0; !%B= 4# 1 'D j+ . c c 't

(38.1a-d)

Here we will assume that there are no permeable media, resulting in D = E and H = B, and that any charge or current densities are produced by explicitly accounted for electrons. Making the usual substitutions
E= 1 !A " #$; c !t B=#%A

(38.2a,b)

we obtain the equations

1 #2 A 1 #% ) 4, & !2A " 2 2 " ! ( ! $ A + =" j + ' c #t* c c #t

(38.3a)

and
! 2" + 1 # (! $ A ) = %4 &' . c #t 1 #$ =0 c #t

(38.3b)

If we choose a gauge, the so-called Lorentz gauge, in which

!"A +

(38.4)

these equations uncouple and we are left with

QM59: Revised: 6/21/95

1 # A 4$ ! A" 2 2 =" j c c #t
2

(38.5a)

and
1 $" ! 2" # 2 = #4 %& . c $t

(38.5b)

For the remainder of this section we confine ourselves to free fields; Eq.s (38.5a.b) and (38.4) then have a solution where = 0 and A satisfies
1 !2 A c
2

!t

"2A = 0

(38.6)

where instead of (38.4) we have the gauge condition

!"A = 0 .

(38.7)

38.1. Expansion of the vector potential in plane waves Since (38.6) is a linear equation we may represent a general solution as a superposition of "normal modes" which we choose to be plane electromagnetic waves; imposing the condition that A(r,t) must be real our superposition will have the form
ik"r % A(r, t) = ! # A k (t)e ik"r + A* k (t)e $ &. k

(38.8)

To satisfy (38.7) we require

k ! A k (t) = 0 .

(38. 9)

Although A(r, t) is real, A k (t) and A* k (t) are complex. Using (38.2a,b) and assuming a time dependence of the form e
i! k t

with a frequency ! k = ck (to satisfy (38.6)), the

corresponding electric and magnetic fields are then

ik#r E(r, t) = ! Ek," (t)e ik#r + E* ke k

) )
(38.10a)

=
and

i ik#r $ k A k (t)& k," e ik#r A* % k (t)e c k

i k#r H(r, t) = ! H k (t)" k e ik#r + H* k,$ (t)e k

) )

i k#r = i! !k % A k,$ (t)e ik#r A* . k,$ (t)e k,

(38.10b)

In order to quantize the electromagnetic field, which we do in the next subsection, it is convenient to introduce two new variables, which play the role of cannonical displacements and momenta, as originally done by Dirac. The equations of motion of these quantities will mimic those of a harmonic oscillator for each value of k and ! in our expansion of the field. This will allow us to immediately quantize the photon frequencies and with it the Planck photons will emerge. These new variables are chosen as

Qk,! =
and

1 A k,! + A# k,! 4"

) )

(38.11a)

Pk,! =

" i# k ! A k,! " A% k,! = Qk,! ; 4$

(38.11b)

from these definitions it is clear that Qk,! and Pk,! are real, as are their mechanical counterparts. The inverse relations are
' 1$ i A k,! = (4 " )1/2 & Qk,! + Pk,! ) 2% #k (

(38.12a)

and
' i 1/2 1 $ . A* Q P k,! = (4 " ) k , ! k , ! ) 2& #k % (

(38.12b)

In terms of the new variables we may rewrite (38.8) as

A(r, t) = 4 !
k,"

#% &Qk," (t)cos(k $ r) + Pk," (t)sin(k $ r) ' (.

(38.13)

We interpret the energy density associated with the electromagnetic field as the Hamiltonian density

H!(r, t) = E2 (r, t) + H2 (r, t) ;

we may then write the total Hamiltonian as
H(t) = 1 8! "

(38.14a)

(E2 (r, t) + H2 (r, t)) d3r .

(38.14b)

Inserting (38.13) into (38.10a,b) and the resulting forms into (38.14b) and carrying out the integral over the spatial coordinates we obtain

QM59: Revised: 6/21/95

4
1 " 2 k,!
2 3 (Pk2,! (t) + # 2 k Qk,! (t)) d r

H(t) =

(38.15)

Due to the condition (38.9) the vectors Qk,! (t) and Pk,! (t) lie in the plane perpendicular to k; we may therefore resolve the vectors into two components, Q k,! and

Pk,! , along arbitrary but mutually perpendicular directions in this plane where the
associated unit vectors are referred to as polarizations, denoted ! " , and ! = 1,! 2 ; (38.15) then becomes

H=

1 2 Pk + #2 Q2 " , ! k k,! . 2 k,!

(38.16)

38.2. Quantization of the electromagnetic field Eq. (38.16) has the form of a sum over independent harmonic oscillators, provided we identify Pk,! as a canonical momentum and Q k,! as a canonical coordinate. It is then natural to make the transition to a quantum electromagnetic field

, and Q theory by assuming P k,! k,! are operators and that they obey the same
commutation relations as for a mechanical system; i.e.,

Q P k! k "! " Q k "! " Pk! = i! # kk " # !! " .

The Hamiltonian (38.13) then becomes an operator
= H
k,!

(38.17)

"H k!

(38.18a)

where the operators

1 2 2 H Pk! + " 2 k! = k Q k! 2

(38.18b)

can be viewed as a set, (k, ), of one-dimensional quantum oscillator Hamiltonians (with

unity mass) where, as noted above, we identify P k,! and Q k,! as the canonical
momentum and the coordinate operators respectively. (38.18b) may be written immediately as The energy eigenvalues of

E=
or

!, k

" E k!

5
1$ " E k! = # N k! + % !& k 2

(38.19a)

where N k! is the number of photons (Planck light quanta) of wave vector k and polarization . It therefore follows that the total energy is given by
E= k, !

" $N k! + 2& !' k .

1%

(38.19b)

Since P k,! and Q k,! are now operators, it follows from (38.13c,d) that the

* corresponding quantities A k! and A k! are also operators (as are the related Fourier , !E * and H , !H * ); being coefficients of the electric and magnetic fields, E k! k! k! k!
operators, the Fourier components of the various electromagnetic fields clearly cannot be thought of as wavefunctions in the usual quantum mechanical sense. Given that our field quantities are now operators we must identify the space on (or in) which they operate. We have, in fact, already implicitly identified that space in writing equation (38.19): the Hamiltonian (38.18) operates in the occupation number space of photons. In our earlier discussion in Sec. 26, the occupation number representation was introduced as a (possible) mathematical convenience. Particles are conserved in non-relativistic quantum mechanics and it is a matter of personal choice whether one wants to visualize a particle as: i) decreasing its "occupation number" by one in level En and increasing its "occupation number" by one in level Em or ii) making a "transition" from En to Em. To complete our interpretation of the electromagnetic field operators as entities acting in the space of photon occupation numbers, we must identify the corresponding creation and annihilation operators. To do this we substitute the operator form of (38.13) into (38.18) obtaining
= H #2 k A A " 4$ k! k! + A k! A k! k, !

(38.20)

where, as noted above, we have interpreted the classical quantities Ak! and A* k! as the
operators A k! and A k! in making the transition to the quantum theory. We rewrite

(38.12a,b) in operator form as

QM59: Revised: 6/21/95

6
(4 " )1/2 A = #k Q k! k! + iPk! 2# k

)
(38.21a)

!!!!!!!! $

(4 " )1/2 (2# k / ! )1/2

k! a

and
1/2 = (4 " ) A #k Q k! $ iPk! k! 2# k

)
(38.21b)

!!!!!!!! %

(4 " )1/2 (2# k / ! )

a 1/2 k!

k! and a where, as defined, a k! are dimensionless. By manipulating the commutation relations (38.17) we find that A k! and A k! obey the commutation relations

, A [A k! k "! " ] = 0 , A [A k! k "! " ] = 0

(38.22a) (38.22b)

and

,A % = 2 '! ) ) #A k ! kk " !! " . k "! " & $ (k

k! and a The corresponding commutation relations for a k! are

(38.22c)

a k! , [

a k "! " ] = 0 ,

(38.23a) (38.23b)

[ ak! , ak"!" ] = 0 ,
and

[ ak! , ak"!" ] = #kk"#!!" ,

We may then write the Hamiltonian (38.20) as

(38.23c)

in common with the boson commutation relations (26.21) (26.22a-c) obtained earlier.

= H

# 1& N + k! 2 ( !) k "% $ ' k ,!

(38.24)

where, in analogy with (26.38), we introduced the photon number operator

k! . N k! = a k! a

(38.25)

7 A new feature arising here (which we did not encounter in Sec. 26) is the zero point energy of the vacuum

1 E zero point = " !# k ; 2 k,!

(38.26)

since there is no limit on k, the zero point energy sum is divergent. One often sees the argument that the zero of energy is arbitrary and proceeds to ignore this zero point energy (or equivalently shift the zero of energy by an infinite amount to compensate for it). This is a controversial point since zero-point energy might contribute to the energymomentum tensor which enters the gravitational field equations in general relativity.
k! , In terms of the operators a a k ! we may write the operator for the vector

potential as

(r, t) = 4 ! A

! ' (" ) ik&r ) k" (t)e ik&r + % (" )*a % a k" (t)e ( * k," 2$ k

(38.27)

where () denotes the two independent polarization vectors introduced earlier and

!( " ) # k = 0 . We may likewise write the operators for the electric and magnetic field as

(r, t) = 4 ! E

!$ k ' (" ) ik&r ) k" (t)e ik&r + % (" )*a n % a k" (t)e ( * 2 k,"

#i
#

(38.28a)

(r, t) = 4 ! H

i ik'r * k" (t)e ik'r & (" )*a . k % ( & (" )a k" (t)e ) + 2 $ k k,"

(38.28b)

We may rewrite the classical expression for the momentum density

P= 1 " d3r E # H ; 4!

(38.29)

by substituting the above expressions and integrating (using orthonormality of the modes) to obtain for the total momentum operator

= P

k,!

" !k$ N k! + 2& .

1%

(38.30)

(Note the vacuum carries no momentum since the sum symmetry).

QM59: Revised: 6/21/95

k,!

"2k

over k vanishes by

8 It must be pointed out here that the derivation carried out here has assumed that
i! k t the operators A ). In Schroedinger k! are functions of time (through the factor e

wave mechanics operators are time independent; hence the operators in our previous formalism must correspond to the Heisenberg representation. A Heisenberg operator,

(S) , as (H) (t) , is related to a Schroedinger operator, O O

(S) i Ht (H) Ok! (t) = e iHt O . k! e

(38.31)

k! (t) , and the Let us consider the relation between the Heisenberg operator, a k! . For this case we can commute the last two corresponding Schroedinger operator, a k! reduces the number of terms in (38.31) provided we take account of the fact that a
particles in the system by one (in the mode k, ) and the corresponding energy by ! !" k ; thus

k! (t) = e iHta k ! e i Ht = e iHte iHt i"k t a k! a

or
k! (t) = e i" k t a k! a

(38.32)

which is precisely the time dependence we have earlier assumed. 38.3 Explicit forms for the occupation number representation To complete this section we examine an explicit representation for the creation and destruction operators and the associated number eigenstates, !k" (N) , where N designates the number of photons. To simplify the discussion we temporarily limit the discussion to a single value of k, . We may then write the number eigenstates as
" 1% $ 0' !(0) = $ ' ; $ 0' $ # !' & " 0% " 0% $ 1' $ 0' $ ' $ ' !!!!!!!! !(1) = $ 0 ' ; !!!!!!!!!!! !(2) = $ 1 ' ; $ 0' $ 0' $ ' $ ' $ $ # !' & # !' &

etc.

(38.33)

and a take the form In this representation the operators a

!0 1 # 0 0 =# a # #0 0 # "! !

0 2 0 !

!$ & 0 !& ; & 3 !& ! !& % 0

! 0 # 1 = # !!!!!!! a # 0 # "!

!$ & 0 0 0 !& 2 0 0 !& & ! ! ! !% 0 0 0

(38.34b)

(38.34a) The number operator is then

" 0 0 0 0 !% $ ' $ 0 1 0 0 !' a a= . $ 0 0 2 0 !' $ ' #! ! ! ! !&

The above forms satisfy the Bose behaviors

(38.35)

!(N) = N !(N 1) a
!(N) = N + 1 !(N + 1) a a !(N) = N !(N) a

(38.36a) (38.36b) (38.36c)

as well as the commutation relations (38.24) (as is easily verified by performing the required matrix multiplications). It is interesting to examine the behavior of (38.38) - (38.36) if we (artificially at first) limit the number of components of the occupation number eigenfunctions to two. We then have

"1 % !(0) = $ ' # 0&

(38.37a)

" 0% ! (1) = $ ' . #1 &

(38.37b)

The corresponding destruction and creation operators, which for this we denote as
and c , then have to be c

! $ =# 0 1 &, c " 0 0 %
(38.38a) and

! $ = # 0 0 & c " 1 0 %
(38.38b)

QM59: Revised: 6/21/95

10

! $ c =# 0 0 &, c " 0 1 %
(38.39a) It then turns out that

! $ c = # 1 0 & . c " 0 0 %
(38.39b)

c +c c = 1 ; c

(38.40)

i.e., the operators anticommute, as expected for Fermi-Dirac operators. More generally we write the commutation relation as

k "! " + c k "! " c c k! c k! = # kk " # !! "

which may be denoted as

(38.41)

{c

k "! " k! , c

}=#

kk " # !! " .

(38.42a)

We also have the relations

{c

k! , c k "! "

} = {c

k "! " k! , c

} = 0.

(38.42b)

Finally we note that the last form implies

c k! c k! " = 0 ;
principle. 38.4. The Casimir effect

(38.43)

i.e., we cannot create two particles in the same state, as expected by the Pauli exclusion

In the discussion of the vacuum zero-point energy following Eq. (38.26) we ignored, with reservations, the over-all contribution of this term, since it was divergent. Suppose instead we consider the zero-point energy the vacuum would have if it were confined within a box with perfectly conducting (reflecting) walls and dimensions
L x , L y , L z ; this energy would also be infinite. However it turns out that the difference,
!E , between this energy and the zero-point energy of an identical volume of space

without walls is finite and depends on the dimensions of the box. As a result forces, named after Casimir (1948) who first predicted them, act on the walls; e.g. the force on a wall perpendicular to the z direction would be F = ! ! "#E / "L z . Formally the

11 phenomena involves an interaction with the walls of the box, since the presence of free electrons is required to produce the conductivity that results in the reflection of the incoming waves.1 The phenomena may be regarded as one of the many phenomena in quantum electrodynamics in which finite contributions result from the careful subtraction of infinite quantities. The components of the standing wave electric field satisfying the condition that the tangential component of E vanish on the surface of the box (required in the presence of infinite conductivity) are given by2

E x = E 0x cos k x x sin k y y sin k z ze ! i" k t E y = E 0y sin k x x cos k y y sin k z ze ! i" k t E z = E 0x sin k z x sin k y y cos k z ze ! i" k t
where the components of the vector k satisfy
ny! n ! n ! k x = x ,!!!!! k y = ,!!!!! k z = z ,!!!!! Lx Ly Lz

(38.44)

(38.45)

and the n i are integers. The frequencies follow from

2 2 !2 k =c k 2 2 !!!!!! = c 2 (k 2 x + k y + k z ).

(38.46)

In addition, we must have

k ! E0 = 0

(38.47)

in order to satisfy the free-space Maxwell equation ! " E = 0 . If two of the n i vanish (38.44) yields only a single field component, and it then follows from (38.47) that this component must vanish; i.e. no fields can be present. If a single n i vanishes, (38.47) relates the two remaining fields; i.e. we have a single mode. Finally if all three n i are non zero, E0 has three non-zero components and is perpendicular to k; it can be resolved into two independent orthogonal components; i.e., we have two modes.

More generally, forces act on objects made from both dielectrics and conductors and result from the socalled Lifshitz van der Waals interaction, which may be thought of as a generalization of the ordinary van der Waals interaction. The theory is somewhat more complex than that given here and is discussed in E. M. Lifshitz and L. P. Pitaevskii, Statistical Physics Part II, Pergamon Press (1980), Pg. 331. 2 See, e.g., W. K. H. Panofksy and M. B Phillips, Classical Electricity and Magnetism, Addison Wesley.

QM59: Revised: 6/21/95

12 We will perform the calculation at absolute zero where only the zero-point contribution to the energy of the electromagnetic oscillators is present. To calculate the total energy in the box we must sum over all the modes under the assumption that each has an energy (1 / 2)!! k ; this involves sums over the doubly and singly degenerate modes. We will assume the dimensions L x ! and ! L y are very large compared to L z and treat the quantum numbers n x ,! n y , or equivalently k x ,! k y , as continuous. In this limit we may ignore the contributions when either k x ! or ! k y vanishes and write the doubly and the remaining singly degenerate contributions to the energy as

E=

$ (2 ! )2 "#
LxLyLz (2 ! )
3

LxLy #

dk x $

& # !c 2 2 ) !c 2 2 ! nz 2+ dk y ( 2 % k2 + k + + k + k x y x y . (38.48) 2 "# 2 ( n =1 2 + L z ' z *

#

On the other hand, in the absence of the walls we would write the energy as
E= 2$
# "#

dk x $

"#

dk y $

"#

dk z

!c 2 2 kx + k2 y + kz . 2

(38.49)

Eq.s (34.48) and (34.49) differ in that the behavior as a function of k z is a sum in the first case and an integral in the second. We can then write the difference between (34.48) and (34.49) (in polar coordinates) as

!E =

( $ !c + LxLy $ " 2n2 2 z + 1 k ' 2 L z 2 $ dk !c k 2 + k 2 - . * k dk 2 k + & # # # # z # z 2 " %0 2 2 " %0 2 * n =1 2 L2 z ) z ,

(38.50)

It turns out that even with this subtraction procedure !E is still divergent. In order to extract a finite result from this expression we multiply the integrand by a regularizing function F(k) that rolls off high momentum values,3 which we may take to have the form F(k) = exp(! k 2 / k 2 c ) where k c is some (large) cutoff wave vector (that ultimately approaches infinity) and k = k 2 !+
" 2n2 z L2 z

; Eq. (38.51) then takes the form

It turns out the form of this function is not that important so long as it goes to zero sufficiently rapidly for large k and is essentially unity for small k.

13
- $ & ) !cL x L y $ " 2n2 " 2n2 2 2 / z F z !E ! = k dk k + k + ( + # #/ , # # 2 2 + 2 " %0 ( L L z z * ' / . n z =1 ) 1 & $ 1 " 2n2 " 2n2 2 z 3F z . !!!!!!!!!!! + F(k # )k # 0 % dn z k 2 + k + ( + # # 2 + 0 2 ( 3 L2 L z 2 ' z *

(38.51)

where we changed to an integration over the dimensionless variable n z . Eq. (38.51) can then be written
# * !cL x L y ' 1 # ) !E ! = f(0) + $ f(n z ) % & dn z f(n z ) , 0 2" ) 2 , n z =1 ( +

(38.52)

where we defined
f(n z ) = !!!!!!! = % ( $ 2n2 $ 2n2 2 2 zF z k dk k + k + * ! #0 ! ! ! L2 ' ' L2 z z * & ).
"

(38.53)

1 " 1/2 F(x) 2 2 2 dx x # 2 $ n z /L z

To evaluate this quantity we use the Euler-Maclaurin formula4 that formally expresses the difference between an integral and an N-dependent trapezoidal approximation as a series of the form
N !1 n =1

" f(n) = #0

dn ! f(n)

(38.54)

1 1 1 !!!!!!!!! + (f(N) ! f(0)) + (f $(N) ! f $(0)) ! (f $$$(N) ! f $$$(0)) + % % %!. 2 12 720

Due to the factor F(k) the terms involving f(N) and all its derivatives vanish for large N. The term f(0) is divergent (in the absence of the cutoff) but cancels against a corresponding term in (38.52). We can evaluate f !(n z ) analytically with the result

n2# 3 f !(n z ) = " z 3 Lz

from which we obtain f !(0) = 0 ; differentiating (38.55) twice we have

(38.55)

M. Abrimowitz and I. Stegun, Handbook of Mathematical Functions, Dover (1972). Pg. 19.

QM59: Revised: 6/21/95

14

f !!!(0) = "

2# 3 L3 z

All higher derivatives vanish. The energy shift !E then follows immediately from (38.52) as

!E = "

!cL x L y # 2 720L3 z

(38.56)

corresponding to an attractive force5

Fz = !

!cL x L y " 2 240L4 z

(38.57)

The existence of this force has been confirmed experimentally.

Problems 38.1. Verify Eq.s (38.22a,b) and (38.23a,b,c).

Note the difference between the force obtained here and that obtained in problem 31.2. When the Van der Waals force theory is generalized to allow for the finite propagation velocity light, it takes on the same position dependence for larger values of the seperation as the Casimir force.