Lecture 3. Classical description of nonlinear cubic interactions.
Hamiltonians leading to
the generation of nonclassical light
Cubic interactions: Kerr effect, four-wave mixing, third harmonic generation. Hamiltonians producing
photon pairs.
1. Third-order nonlinear effects.
(3)
These effects are related to the third-order polarization, P EEE . In the tensor form, it
(3)
is written as
Pi ( 3) 0 ijkl
( 3)
E j E k El . (1)
The third-order susceptibility ijkl
( 3)
is a tensor of rank 4. It is present in all materials, although
of course it can be higher or weaker. We will now consider several effects originating from it.
Kerr effect. Taking the first and third terms in the general equation
P 0 (1) E 0 ( 2 ) EE 0 ( 3) EEE ..., (2)
we see that the linear susceptibility gets an additional term quadratic in the field:
(1) ~ (1) (1) ( 3) EE. This means that the displacement can be represented as
D 0 E P 0 (1 ~ (1) ) E 0 n~ 2 E ,
and one can introduce the ‘corrected’ refractive index n~ quadratic in the field.
n~ n n2 I . (3)
Four-wave mixing. Let two fields be present at the input of a
nonlinear medium (crystal, glass, fibre, liquid, atoms, …): the
pump E p and the signal E s . Then, there will be nonlinear p s
polarization
P3 ( 3) ( E p( ) E s( ) c.c.) 3 .
There will be a lot of terms here, for instance, third-harmonic p 2 p s
generation, but only some of them will be such that phase
matching will be satisfied. Now we are interested in the term
Fig.1
( 3) [ E p( ) ]2 E s( ) . It will oscillate at frequency 2 p s and
have the wavevector 2k ( p ) k (s ) . It will correspond to the four-photon diagram shown in
Fig.1. Similar to the difference-frequency generation, this effect is described classically. The
pump can be a single one or two different beams at two different frequencies as well.
However, if the signal beam is absent, there will still be
radiation at the output (Fig.2). This effect, spontaneous four-
wave mixing, can be only considered quantum mechanically. p s
Third harmonic generation. This is similar to second
harmonic generation, perfectly classical effect. More p 2 p s
interesting is its inverse, photon triplet generation, a purely
quantum effect that has never been observed before.
Fig.2
2. Pair-creating Hamiltonians.
A Hamiltonian in quantum mechanics is just the energy, but as any observable in quantum
mechanics, it is an operator. So the strategy will be to write the energy of the nonlinear
interaction but to keep quantum what should be kept quantum.
In the dipole approximation, the energy of interaction between the matter and light is
H dE , (4)
where E is the electric field, and the dipole moment can be written as the volume integral of
the polarization,
d d 3 rP ( r ) . (5)
Parametric down-conversion. Consider first quadratic nonlinear interaction and the second-
order nonlinear polarization,
P ( 2 ) 0 ( 2 ) EE . (6)
After substituting (5) and (6) into (4), we see that H 0 d 3 r ( 2 ) ( r ) EEE.
We omit the vectors, and assume first that there are three real fields: the pump (labeled with
0), signal (1), idler (2). Then,
E 3 ( E0 ei0t ik0r E1ei1t ik1r E2 ei2t ik2r c.c.) 3 (7)
We are interestedonly in the term
~ E0 e i0t ik0r E1 e i1t ik1r E 2 e i2t ik2 r c.c.
* *
Now we have to take into account that the fields are actually quantum fields. We have to
recall quantization of the field, which means that each field mode (for simplicity, a plane-
wave mode given by the wavevector k ) is similar to a harmonic oscillator. It can be populated
by a certain (or, rather, uncertain) number of quanta. There are photon creation and
annihilation operators a k , a k , which increase and decrease the number of quanta by one. And
then the positive- and negative-frequency fields are also operators; they can be written as a
superposition of mode contributions, each of them containing a photon annihilation or
creation operator:
()
E ( ) (r , t )
c a e
k k
i ( k ) t ik r
, E ( r , t )
c i ( k ) t ik r
a e
k k
. (8)
k k
Therefore, we have an operator a instead of ()
Re{ E ((tt )) }
the analytic signal. Recall that the operators
a k , a k satisfy the commutation relations
[a k , a k ] 1 . E ( ) (t )
It is also useful to recall the quadratures, or
()
position and momentum operators, which are Re{ E (t ) }
defined as the Hermitian and anti-Hermitian
parts of a k , a k :
1
qˆ k ( a k a k ) / 2, pˆ k ( a k a k ) / 2i, [qˆ k , pˆ k ] .
2i
But if we try to show the operator, as the field,
as a point on the phase plane (Fig.3), the
quadratures being its real and imaginary parts,
we face a problem. Quantum mechanics
teaches us that such a single point makes no Fig.3
sense: position and momentum cannot be
1
defined simultaneously, there is the uncertainty relation qˆ k pˆ k . Therefore, any state
2
will look like a set of points, or an area. For instance, the vacuum will look not like a point at
the center but as a blurred ‘circle’. For this reason, in Eq. (7) we substitute fields 1,2 with
expressions (8). For simplicity, we will only take into account one term in the sum over the
modes. At the same time, the pump is strong and can be considered as classical light. In Fig.3,
it will be a blurred circle displaced VERY far from the origin. (We will discuss later what all
this means, in connection with the Wigner function.) Then the Hamiltonian is proportional to
Hˆ ~ d 3 r ( 2 ) ( r ) E0 e i (0 1 2 ) t i ( k0 k1 k 2 ) r a1 a 2 h.c. d 3 r ( 2 ) ( r ) E0 e it ik r a1 a 2 h.c.
Here we introduced frequency and wavevector mismatches,
0 1 2 , k k0 k1 k 2 . The volume integral is over the space in the crystal
occupied by the pump. The transverse size of the pump, for simplicity, can be assumed to be
infinite. This leads to a factor ( k x ) ( k y ) in the Hamiltonian. Along the z axis (Fig.4),
there is the length of the crystal, and the integral is
0
e ik z z 2 sin( k z L / 2)e ik z L / 2
0
dze
ik z z
ik z
L sin c( k z L / 2)e ik z L / 2 LF ( k z )
L L
k z
With the mismatch zero, the Hamiltonian becomes simple:
Hˆ ~ ( 2 ) E0 La1 a2 eit h.c.
The Hamiltonian becomes also time-independent (why do we need this?) if 0 .
Then, the Hamiltonian can be written as
Hˆ ~ ( 2 ) E0 La1 a 2 h.c. ia1 a 2 h.c. , (9)
where we deliberately dragged out the Planck constant and
included all relevant parameters (crystal length, quadratic L
nonlinearity, field of the pump, etc.) into (coupling y
parameter).
We have obtained an interesting operator: it creates photons z
only in pairs. And the probability of pair creation depends on x
the quadratic susceptibility, pump field, and the length of the
crystal. In the next lecture, we will calculate the mean number Fig.4
of these pairs and their various properties.
To satisfy the phase matching, one can simply use a birefringent crystal and choose different
polarizations for the pump and for the signal/idler photons. For instance, in type-I PDC the
pump is polarized as an extraordinary beam while the down-converted photons are ordinary.
For degenerate phase matching, one should then satisfy k eff ( 2 ) 2k o ( ) , or
neff ( 2 ) no ( ) , where neff is the effective refractive index (was considered in the problem
class).
One can also distinguish type-II phase matching, for which signal and idler photons are
polarized orthogonally.
Quasi phase matching. It is also possible to phase match PDC with all photons (pump, signal,
idler) polarized the same way. It can be useful, for instance, because one can then take
advantage of the strongest quadratic susceptibility component zzz
(2)
, which is the largest in
most crystals. In a periodically poled crystal, is periodically modulated. In reality it just
(2)
changes sign, but for simplicity we will assume that it varies sinusoidally:
0
( 2 ) ( z ) 0 e iKz c.c. Then, in the Hamiltonian, instead of the integral dze ik z we will z
L
0
dze
ik z z iKz
obtain . This immediately means that the mismatch will be
L
~ 2
k z k 0 k1 k 2 K . By choosing an appropriate poling period , one can satisfy
K
the phase matching. Of course the ( 2 ) dependence on z is not sinusoidal but rather meander-
type, but this simply means that only a single harmonic will play a role.
Four-wave mixing/modulation instability. Consider now cubic nonlinear interaction and the
third-order nonlinear polarization,
P ( 3 ) 0 ( 3 ) EEE . (10)
Similarly to the previous case, we see that H 0 d r ( r ) EEEE.
3 ( 3)
We omit the vectors again, and again assume only three fields: the pump (labeled with 0),
signal (1), idler (2). Then, the Hamiltonian will contain
E 4 ( E0 e i0t ik0r E1e i1t ik1r E2 e i2t ik2r c.c.) 4 , (11)
And we will be interested in the term
~ E 2 e 2i0t 2ik0r E1 e i1t ik1r E2 e i2t ik2r c.c.
* *
We proceed the same way; the only difference from the case of PDC is that here, two pump
fields enter the Hamiltonian. The latter then takes the form (9) with ~ ( 3) E02 L and the
frequency and wavevector mismatches being 20 1 2 , k 2k0 k1 k 2 .
Phase matching in four-wave mixing and
modulation instability. Because k
these effects are usually observed in
( 3)
optical fibers, one cannot use ANOMALOUS GVD
birefringence to provide the phase
matching. Periodic poling also does not
work, since poling does not change ( 3) .
In this case, it is important whether the NORMAL GVD
pump wavelength is below or above the
‘zero-dispersion wavelength’, ZDW
0 2c / 0 (Fig.5). At 0 , the group
velocity dispersion is zero and the Fig.5
dispersion dependence is flat.
Above 0 (normal group-velocity dispersion range), it is possible to satisfy the relation
k (1 ) k (1 ) 2k (0 ) , where the indices 0,1,2 still mean the pump, signal, and idler. This is
shown by the blue line. The pump wavelength should not be far below the ZDW, though. This
regime is called four-wave mixing.
But below 0 (anomalous dispersion range) the pump wavevector cannot be the mean
arithmetic of the signal and idler wavevectors. And here the Kerr effect helps. Indeed, if the
pump is strong, its refractive index changes due to the Kerr effect (red vertical bar in the
figure). And the stronger the pump, the larger the nonlinear change in the refractive index,
and hence the further apart the signal and idler wavelengths. This regime is called modulation
instability.
Fig.6 shows the signal and idler
wavelengths as functions of the
pump wavelength for different pump
powers. We see that as the pump
power increases, in the anomalous
group-velocity dispersion range, the
signal and idler wavelengths get
separated due to the Kerr effect. In
the normal dispersion range, their
wavelengths are very different and
do not depend on the pump power.
Home task: What will happen if the
pump beam is not infinitely broad, Fig.6 (from Wadsworth et
but has a Gaussian shape with a waist al., Op. Ex. 12, 299 (2004)).
w?
Books:
1. Boyd, Nonlinear optics
2. Klyshko, Physical foundations of quantum electronics