Fundamentals of Nonlinear Optics: ECED 6400 Lecture Notes
Fundamentals of Nonlinear Optics: ECED 6400 Lecture Notes
1 Introduction 3
1
4 Nonlinear optics 60
4.1 Introduction. Qualitative description of nonlinear optical processes . . 60
4.2 Nonlinear processes generated by arbitrary fields: Spatial and temporal
dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3 Formal properties of nonlinear optical susceptibilities . . . . . . . . . 67
4.4 Nonlinear wave equation approach: Classical coupled-wave equations 71
4.5 Second-harmonic generation . . . . . . . . . . . . . . . . . . . . . . 74
4.5.1 Coupled wave equations and phase matching considerations . 74
4.5.2 Second-harmonic generation: Beyond the undepleted pump
approximation . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.6 Sum-frequency generation . . . . . . . . . . . . . . . . . . . . . . . 82
4.6.1 Coupled wave equations and their solution in the undepleted
pump approximation . . . . . . . . . . . . . . . . . . . . . . 82
4.6.2 Manley-Rowe relations . . . . . . . . . . . . . . . . . . . . . 84
4.7 Difference-frequency generation (parametric down-conversion) . . . . 86
4.8 Four-wave mixing: General considerations . . . . . . . . . . . . . . . 90
4.9 Third harmonic generation . . . . . . . . . . . . . . . . . . . . . . . 91
4.10 Self-focusing, nonlinear absorption, and spatial solitons . . . . . . . . 93
4.11 Z-scan measurement of nonlinear refractive index . . . . . . . . . . . 101
4.12 Polarization dynamics of third-order processes . . . . . . . . . . . . . 106
4.13 Electro-optical Kerr effect . . . . . . . . . . . . . . . . . . . . . . . 111
4.14 Spontaneous and stimulated Raman scattering: CW case . . . . . . . 114
4.15 Transient stimulated Raman scattering . . . . . . . . . . . . . . . . . 123
4.16 Spontaneous Brillouin scattering . . . . . . . . . . . . . . . . . . . . 129
4.17 Brillouin phonon propagation . . . . . . . . . . . . . . . . . . . . . . 131
4.18 Stimulated Brillouin scattering . . . . . . . . . . . . . . . . . . . . . 133
2
Chapter 1
Introduction
In this course, we will be describing all optical phenomena classically within the
framework of macroscopic Maxwell’s equations written in terms of macroscopic elec-
tromagnetic fields. The latter are obtained by averaging rapidly varying microscopic
fields over spatial scales much larger than characteristic material microstructure scales
(atomic size, lattice scale, etc). The averaging procedure is examined in detail in stan-
dard electrodynamics textbooks1 . Within the framework of such a phenomenological
approach, which circumvents a detailed microscopic light-matter interaction descrip-
tion, external or driving volume charge and current densities, ρex and Jex , give rise to
the electromagnetic fields obeying the Maxwell equations in the form
∇ · D = ρex , (1.1)
∇ · B = 0, (1.2)
∇ × E = −∂t B (1.3)
and
∇ × H = Jex + ∂t D. (1.4)
The set of equations (1.1) through (1.4) is not closed, however, until we provide any
information about the material media. Such information is furnished by supplying phe-
nomenological constitutive relations among the four fields, E, D, B, and H. Without
much loss of generality we will assume hereafter that all material media are nonmag-
netic, which holds true for virtually all natural media at optical frequencies 2 . We can
then represent the magnetic constitutive relation in its simplest form as
B = µ0 H, (1.5)
with µ0 being the free space permeability in the SI units we will be employing hereafter.
A general electric constitutive relation states
D = 0 E + P; P = P(E), (1.6)
1 J.
D. Jackson, Classical Electrodynamics (Wiley, New York, NY, 1999) 3rd edition.
2 Thiscriterion, however, breaks down for some artificial materials, the so-called metamaterials, which
we will not consider in this course.
3
where 0 is the free space permittivity and P is a macroscopic polarization field. The
latter is in turn a function of the applied electric field. For sufficiently weak applied
fields, P is a linear function of E; this is a regime of linear optics. However, even in
the linear optics regime, the dependence of the polarization on the applied electric field
can be rather complicated to account for possible medium anisotropy and–temporal
and sometimes even spatial–dispersion. While the former implies that the medium
response in a particular direction can be affected by the electric field components or-
thogonal to this direction, the latter acknowledges the fact that the medium response
at a given space-time point can depend on the applied electric field in the past (tem-
poral dispersion) and/or on the fields in the neighborhood of the spatial point (spatial
dispersion). We will study all these cases in detail in the subsequent chapters.
As the magnitude of the applied electric field increases, the linear relationship be-
tween P and E breaks down and we enter the realm of nonlinear optics. If the electric
field intensity is far below a critical value, Ecr ∼ 109 V/cm needed to ionize a material
atom, the resulting polarization can be expressed as a series in increasing powers of the
electric field. Schematically, such a series can be expressed as
where we ignored the vector nature of the fields as well as dispersion, for simplicity.
The expansion coefficients, χ(1) and χ(2) , etc., are identified as linear and nonlinear
susceptibilities, respectively. The linear and nonlinear susceptibilities should be treated
as phenomenological constants in our classical description. The condition E Ecr is
typically met with a vast majority of laser sources which rarely generate fields in excess
of 106 V/cm. However, even if the applied field does not exceed Ecr , the power series
expansion can fail, provided the carrier frequency of the field lies close to any internal
resonance of the medium. In the latter case, the material response tends to saturate
at high enough field intensities. The proper quantitative description of such nonlinear
saturation phenomena calls for a quantum mechanical treatment of the medium. When-
ever, the power expansion of P is valid, though, we shall refer to the lowest-order term
in the expansion as a linear contribution and designate the rest to be nonlinear polar-
ization such that
P = PL + PNL . (1.8)
Next, the external charge and current densities are not independent from each other.
Rather they are related by another fundamental law, the charge conservation law, which
takes the form of a well-known continuity equation viz.,
The external ρex and Jex drive the electromagnetic fields which, in turn, induce internal
charge and current densities, ρ and J, inside a medium. The induced charges and
currents can be of either free (conduction) or bound (polarized) type and they also
obey the continuity equation,
∂t ρ + ∇ · J = 0. (1.10)
We stress that charge conservation (1.9) amounts to a fundamental law which does
not follow from Maxwell’s equations. The electromagnetic energy conservation law,
4
however, does follow from the Maxwell equations by the same token as the mechanical
energy conservation follows from Newton’s laws.
To derive the electromagnetic energy conservation law or the Poynting theorem, we
take dot products of the both sides of Eqs. (1.3) and (1.4) with H and E, respectively,
and use the constitutive relations (1.5) and (1.6), yielding
H · (∇ × E) = −µ0 H · ∂t H (1.11)
and
E · (∇ × H) = Jex · E + 0 E · ∂t E + E · ∂t P. (1.12)
On subtracting Eq. (1.11) from Eq. (1.12) term by term, we obtain
0 2 µ0 2
2 ∂t E + 2 ∂t H + Jex · E = E · (∇ × H) − H · (∇ × E) − E · ∂t P. (1.13)
∇ · (E × H) = H · (∇ × E) − E · (∇ × H) (1.14)
Here the electromagnetic energy density wem is defined in the same way as in free
space,
wem = 12 0 E 2 + 12 µ0 H 2 , (1.16)
and we introduced the electromagnetic energy flux density, the so-called Poynting vec-
tor, by the expression
S = E × H. (1.17)
Equation (1.15) is often referred to as Poynting’s theorem. In essence, it implies that
the time rate of change of the electromagnetic energy density is determined by the
energy flux density minus losses associated with external as well as internal currents.
The second term on the r.h.s of Eq. (1.15) describes Ohmic losses associated with
external currents and the third one is identified with the energy loss caused by induced
polarization currents, including the ones associated with the generation of nonlinear
polarizations. To reexpress the right-hand side of Eq. (1.15) in a more symmetric from,
we can explicitly define the induced polarization currents as
J = ∂t P, (1.18)
ρ = −∇ · P, (1.19)
such that the continuity equation (1.10) is respected. We emphasize that our separation
of the charges into external (driving) and internal (induced) is much more natural at
optical frequencies than the standard separation into free and bound charges. The latter
5
is a rather arbitrary division3 which can be confusing at optical frequencies, especially
for metals4 . We will then unify free and bound induced charges and currents under the
umbrella of ρ and J.
To better understand Eq. (1.15), we transform it into the integral form
I Z
dWem
= − dσ · S − dv(Jex + J) · E. (1.20)
dt σ v
where Z
Wem = dvwem , (1.21)
v
is the total energy of electromagnetic field inside a given volume v, and we used a
divergence theorem to convert a volume integral on the the r.h.s of (1.20) into the
surface one. Equation (1.20) then implies that the total energy change inside a finite
region of the medium can occur as a result of the energy outflow through the boundary
surface of the region as well as via energy losses inside the region associated with
driving and induced currents. This situation is schematically illustrated in Fig. 1.
J c .E
∂P
E.
∂t
Wem
In many practical situations in nonlinear optics one deals with pulse or beam fields
with their carriers oscillating at optical frequencies. Such fast oscillations can never
be detected by even the fastest modern detectors whose response time is much larger
3 Yu. A. Illinskii, L. V. Keldysh, Electromagnetic response of material media (Plenum Press, New York,
NY, 1994).
4 S. A. Maier, Plasmonics, Fundamentals and Applications (Springer, Berlin, 2007).
6
that an optical period. Consequently, it makes sense to talk about the field quantities
averaged over many optical cycles – it is those quantities that can actually be registered
in optical measurements anyway. In the absence of external currents, Jex = 0, we can
rewrite the time-averaged Poynting theorem as
Here we define time-averaged quantities such as the average Poynting vector by the
expression
1 t+T /2
Z
hSi = hE × Hi ≡ dt (E × H), (1.23)
T t−T /2
where T is an optical period. In case of monochromatic fields, which can be conve-
niently represented via complex amplitudes as
and
H = 12 (He−iωt + c. c.), (1.25)
Eq. (1.23) can be shown to reduce to
7
Chapter 2
By linearity of Maxwell’s equations in free space, we can drop the real part and deal
with complex phasors describing the waves directly. The real part can be taken at the
end of all calculations to yield physical (real) electric and magnetic fields of a plane
wave.
The Maxwell equations in the plane-wave form can be rewritten as
k · E = 0, (2.6)
k · H = 0, (2.7)
k × E = ωµ0 H, (2.8)
and
k × H = −ω0 E. (2.9)
8
In Eqs. (2.6) – (2.9) we dropped plane-wave phasors on both sides.
Next, we can exclude the magnetic field from the fourth Maxwell equation leading
to
k × (k × E) = −0 µ0 ω 2 E. (2.10)
Rearranging the double cross-product on the left-hand side of Eq. (2.10), we arrive at
(k 2 − µ0 0 ω 2 )E = 0, (2.12)
implying that
√
k = ω 0 µ0 = ω/c (2.13)
where we introduced the speed of light in vacuum
1
c= √ = 3 × 108 m/s. (2.14)
0 µ0
(ek × E)
H= , (2.15)
η0
By the same token, E0 can be inferred from H0 with the help of Eq. (2.9):
It follows at once from Eqs. (2.15) and (2.17) that E, k and H are mutually orthogonal
for a plane wave in free space.
By convention, the wave polarization is associated with the time evolution of the
electric field vector. Let us consider a plane wave propagating along the z-axis in free
space. As, k = kez , and E⊥k, the electric field in the phasor form reads
We will now show that, in general, the tip of the electric field vector moves around
an ellipse as the time evolves. This general polarization is called elliptic. To proceed,
9
K
Η
E
Note that U and V are not orthogonal which makes the situation tricky. We can how-
ever introduce a transformation from U and V to u, v involving an auxiliary parameter
θ such that
U + iV = (u + iv)eiθ , (2.20)
It follows at once from Eq. (2.20) that
U = u cos θ − v sin θ, V = u sin θ + v cos θ. (2.21)
Inverting Eqs. (2.21), we obtain
u = U cos θ + V sin θ, v = U sin θ − V cos θ. (2.22)
We can now use our freedom to choose θ wisely. In particular, choosing it such that
u · v = 0 (orthogonal axes), we obtain by taking the dot product of u and v,
2U · V 1 −1 2U · V
tan 2θ = 2 =⇒ θ = tan . (2.23)
U −V2 2 U2 − V 2
Here we made use of the trigonometric identities, sin 2θ = 2 sin θ cos θ and cos 2θ =
cos2 θ − sin2 θ. By combining Eqs. (2.19) and (2.20), we can rewrite our field as
E(z, t) = Re{(u + iv)ei(kz−ωt+θ) }. (2.24)
Using the orthogonality of u and v, we can write the two orthogonal components of
the field, Eu and Ev as
Eu = u cos(kz − ωt + θ), Ev = v sin(kz − ωt + θ). (2.25)
10
It follows from Eq. (2.25) that
Eu2 Ev2
+ = 1, (2.26)
u2 v2
where u and v are given by Eq. (2.22) and θ by Eq. (2.23). Eq. (2.26) manifestly
represents an ellipse with the semi-major axis making the angle θ with the x-axis as is
shown in Fig. 3.6. The tip of E can move either clockwise or counterclockwise along
Eu
E
Ev
Ey
E
â y
âx Ex
Although, in general, the electric field is elliptically polarized, there are two impor-
tant particular cases. The electric field is said to be linearly polarized if the phases of
11
two orthogonal components of the field in Eq. (2.18) are the same, φ0x = φ0y .
In this case,
E(z, t) = (ex |Ex | + ey |Ey |) cos(kz − ωt + φ0 ), (2.28)
and the electric field is always directed along the line making the angle
In a circularly polarized wave, the E has the same magnitude but is moving along
Ey
E
o Ex
the circle. In the case of “-” sign in Eq. (2.30), E moves counterclockwise around the
circle and the wave is left circularly polarized; for the “+” sign it is right circularly
polarized.
12
and ∞
X Z Z
Ji (r, t) = dt0 dr0 σij (r − r0 , t − t0 )Ej (r0 , t0 ). (2.32)
j=x,y,z −∞
In Eqs. (2.31) and (2.32), the permittivity and conductivity tensors depend only on co-
ordinate and time differences because of homogeneity and stationarity of the medium:
All properties of such media are invariant with respect to translations in time and dis-
placements in space.
The translational invariance of the system prompts the use of plane-wave expan-
sions via Fourier transforms, i. e.,
Z Z
D(r, t) = dω dk D(k, ω)ei(k·r−ωt) , (2.33)
with similar expressions for the other fields. In physical terms, Fourier expansions give
all possible plane waves allowed to propagate in such media; the Fourier coefficients
specify field amplitudes of these plane waves. Introducing also Fourier expansions of
the permittivity and conductivity tensors viz.,
Z Z
ij (r, t) = dω dk ij (k, ω)ei(k·r−ωt) , (2.34)
and Z Z
σij (r, t) = dω dk σij (k, ω)ei(k·r−ωt) , (2.35)
we can use convolution properties of Fourier transforms to cast Eqs. (2.31) and (2.32)
to X
Di (k, ω) = 0 ij (k, ω)Ej (k, ω), (2.36)
j=x,y,z
and X
Ji (k, ω) = σij (k, ω)Ej (k, ω). (2.37)
j=x,y,z
Next, on taking Fourier transforms of Eqs. (1.6), and (1.18) and combining Eqs. (2.36)
as well as (2.37), we can establish a relation between the permittivity and conductivity
tensors in the Fourier space,
i
ij (k, ω) = δij + σij (k, ω). (2.38)
0 ω
Exercise 2.1. Derive Eq. (2.38).
Thus, we conclude that the permittivity and conductivity tensors are actually related
and one can be eliminated in favor of the other. In condensed-matter calculations, it
is the conductivity tensor that is typically employed. On the other hand, optical wave
propagation in the media is more conveniently examined in terms of the permittivity
tensor. In the absence of external charges and currents, the Maxwell equations (1.1)
through (1.4) can be greatly simplified in the Fourier space to read
k · D = 0, (2.39)
13
k · H = 0, (2.40)
k × E = µ0 ωH, (2.41)
and
k × H = −ωD. (2.42)
Next, eliminating the magnetic field from Eqs. (2.39) - (2.42), and using the con-
stitutive relation (2.36), we can express Eqs. (1.1) and (2.41) in the component form
as X
ki ij (k, ω)Ej (k, ω) = 0, (2.43)
i,j=x,y,z
and X h i
ω2
k 2 δij − ki kj − c2 ij (k, ω) Ej (k, ω) = 0. (2.44)
j=x,y,z
Eqs. (2.43) and (2.44) determine all possible plane electromagnetic waves supported
by a given medium.
Exercise 2.2. Show that Eqs. (2.43) and (2.44) are always compatible.
Eq. (2.43 is called a generalized transversality condition, whereas Eq. (2.44) is a dis-
persion relation for the waves. The existence of nontrivial plane-wave solutions to
Eq. (2.44) can be expressed in terms of a determinant condition as
h 2
i
Det k 2 δij − ki kj − ωc2 ij (k, ω) = 0. (2.45)
Let us now consider the important limiting case of an isotropic dielectric. It can be
inferred by inspection that the dielectric permittivity tensor of an isotropic medium can
only be composed of δij and ki kj implying that
where A(k, ω) and B(k, ω) are scalar functions. Instead of using A and B, however,
it will prove convenient to divide ij into a part transverse to the ek = k/k direction,
and that longitudinal to ek . Such a decomposition can be accomplished via
k k k k
ij (k, ω) = ⊥ (k, ω) δij − ki 2 j + k (k, ω) ki 2 j . (2.47)
On substituting from Eq. (2.47) into Eqs. (2.43) and (2.44), the latter can be trans-
formed to
k (k, ω)(k · E) = 0, (2.48)
and
k(k·E )
h ih i
ω2 ω2
k2 − c2 ⊥ (k, ω) E− k2 − k2 c2 k (k, ω)k(k · E) = 0. (2.49)
Eqs. (2.48) and (2.49) then imply the existence of a family of purely transverse plane
waves,
k · E = 0, (2.50)
14
with the dispersion relation,
p
k = ± ωc ⊥ (k, ω), (2.51)
and a family of the waves which have longitudinal component(s) of the electric field,
k · E 6= 0, with the dispersion relation determined by a common solution of Eq. (2.51)
and of the following equation
k (k, ω) = 0. (2.52)
In Eq. (2.51) the two signs on the right-hand side correspond to two plane waves at a
given frequency ω propagating the the opposite directions.
Exercise. 2.3. As we will see in Sec. 2.5., dielectric response of metals at high frequen-
cies can be modeled by the permittivity
!
ωp2
ij (ω) = δij 1 − 2 , (2.53)
ω
where ωp is the so-called plasma frequency. Determine the frequency(s) and dispersion
relation of longitudinal electromagnetic waves propagating in metals at such ultravi-
olet frequencies and interpret your results in physical terms. Show that transverse
electromagnetic waves can only propagate if ω > ωp . What is their dispersion rela-
tion?
Note that the dispersion relation (2.51) is, in general, in the implicit form due to
spatial dispersion of the medium. It is then instructive to examine the limiting case
of local media which lack spatial dispersion. In reality the vast majority of inorganic
media are made of atoms or molecules with the size significantly smaller than the
optical wavelength. Hence, spatial nonlocality of their dielectric response to the applied
field is negligible, resulting in the absence of spatial dispersion in such media. Under
the circumstances, the permittivity tensor can be simplified as
It then follows at once from Eqs. (2.34) and (2.54) that the permittivity tensor in Fourier
space is independent of k, implying that
The dispersion relation for transverse electromagnetic waves can be expressed in the
explicit form as p
k = ± ωc ⊥ (ω), (2.56)
and the generalized transversality condition states
k (ω) = 0. (2.57)
In the following sections, we will explore several commonly occurring types of linear
optical media.
15
2.2.1 Plane waves in homogeneous isotropic media with no spatial
dispersion
Medium isotropy and locality imply a greatly simplified form of the permittivity tensor,
It then follows from Eqs. (2.47) and (2.58) that k (ω) = ⊥ (ω) = (ω). Assuming
further that in the spectral range of interest, (ω) 6= 0, we conclude that in this case, the
only allowed plane waves in such media must be transverse, governed by the dispersion
relation p
k = ± ωc (ω), (2.59)
Representing the dielectric function in terms of its real and imaginary parts,
Here q√
02 +002 +0
β± = ± ωc 2 , (2.62)
and q√
1 02 +002 −0
2 α± = ± ωc 2 . (2.63)
Exercise 2.4. Derive the equations (2.62) and (2.63).
Let us choose the z-axis of our coordinate system along propagation direction of
the wave, k = kez . It then follows from the Maxwell equations (2.39) through (2.42)
that the electric and magnetic field amplitudes are related as
or, alternatively,
(ez × E)
H= , (2.65)
η
where η is a complex impedance of the lossy medium, defined as
r
µ0 η0
η(ω) = =p . (2.66)
0 (ω) (ω)
To illustrate the plane wave propagation in such a medium, let us focus now on a
particular case of a linearly polarized in the x-direction plane wave which propagates
in the positive z-direction. The electric and magnetic fields of the wave can then be
represented as
E(z, t) = 21 ex [Ee−α+ z/2 ei(β+ z−ωt) + c. c., ] (2.67)
and h i
E −α+ z/2 i(β+ z−ωt−θη )
H(z, t) = 12 ey |η| e e + c. c , (2.68)
16
E
e - z
o z
which describe inhomogeneous plane waves thanks to losses. Here we chose β+ and
α+ which describe a plane wave propagating to the right and exponentially decaying
on propagation into the medium; the magnitude and phase of the complex impedance
can be expressed as
η0
|η| = , tan θη = −00 /0 . (2.69)
(02 + 002 )1/4
We can then infer from Eqs. (2.67) and (2.68) that the presence of losses introduces a
phase lag between the magnetic and electric fields in such media as well.
Finally, we can work out the time-averaged energy flux density (Poynting vector),
and hence the optical intensity, associated with the inhomogeneous plane wave. On
substituting from Eqs. (2.67) and (2.68) into Eq. (1.26), we obtain for the optical inten-
sity
|E|2 −z/δ
I = |hSi| = e cos θη . (2.70)
2|η|
Eq. (2.70) is known as Beer’s absorption law, and by measuring the intensity extinction,
one can infer the Beer absorption length, or skin depth
1
δ= . (2.71)
α+
We note that Beer’s absorption length is then a directly measurable quantity. We can
also define a complex refractive index by the expression
p
N (ω) = (ω) = n(ω) + iκ(ω), (2.72)
where n is a real refractive index which can be determined from reflectivity measure-
ments and κ is a so-called extinction coefficient, closely related to Beers’ absorption
length. In fact, it readily follows from Eqs. (2.59), (2.61) and (2.72) that
2κ(ω)ω
δ −1 (ω) = . (2.73)
c
17
The magnitudes of real and imaginary parts of can then be inferred from the knowl-
edge of n and κ, i.e.,
0 = n2 − κ2 , 00 = 2nκ. (2.74)
In particular, in the transparent regions of the spectrum, where 00 0 , 0 ' n2 and
the optical intensity of a plane wave can be expressed as
0 nc 2
I= |E| . (2.75)
2
18
and ita dispersion relation is given by the expression
√
ko = ωc ⊥ . (2.83)
We observe that ordinary waves in uniaxial crystals have all the same properties as
plane waves supported by transparent isotropic media.
kx kx
k0 ke
kz kz
Figure 2.6: Graphical representation of the wave vectors of ordinary (left) and extraor-
dinary (right) waves in a uniaxial crystal.
E = Ex ex + Ez ez , (2.84)
where Ex and Ez are related by Eq. (2.81). We can also derive their dispersion relation
from the determinant condition for Eqs. (2.78) and (2.79). The resulting dispersion
relation reads
kx2 c2 k 2 c2
2
+ 2z = 1. (2.85)
ω k ω ⊥
Using kx = ke sin θ, and kz = ke cos θ, we can cast Eq. (2.85) into the form
ω2 sin2 θ cos2 θ
= + . (2.86)
ke2 c2 k ⊥
Thus the wave vector magnitude of an extraordinary wave depends on its propagation
direction which is a novel propagation feature arising in anisotropic media. The differ-
ence between ordinary and extraordinary waves can be best visualized by comparing
their dispersion relations. It is seen from Eqs. (2.83) and (2.86) that in the k-plane
the dispersion relations of ordinary and extraordinary waves can be represented by a
√ √ √
sphere of radius (ω/c) ⊥ and ellipse with the semi-axes (ω/c) ⊥ and (ω/c) k ,
respectively. The situation is schematically depicted in the figure above.
Exercise 2.5. Using Maxwell’s equations show that the wave vector of the extraordi-
nary wave is not parallel to the Pointing vector, S = E × H. In other words, demon-
strate that the direction of propagation of such a wave does not, in general, coincide
with the direction of the energy flow.
19
2.2.3 Faraday effect and polarization rotation
We will now consider light propagation in an isotropic, weakly dispersive–and hence
lossless–dielectric medium with a weak homogeneous static magnetic field, B, applied
along the z−axis such that B = Bez . We assume that the influence of magnetic
field can be treated as a perturbation and we seek a phenomenological expression for
a dielectric permittivity tensor of an isotropic medium with a small correction due to
the magnetic field. The lowest-order correction is assumed to linear in the magnetic
field. Therefore, the second-order permittivity tensor can only be comprised of δij and
a component linear in Bi . Recall that both D and E are physical vectors that change
their sign upon reflections with respect to the origin of a coordinate system. It then
follows from Eq. (2.31) that ij should be invariant upon reflections. To respect the
reflectional
P invariance of the permittivity tensor, the correction term can only be of the
form k eijk Bk , where
1, clockwise permutation
eijk = (2.87)
−1, counterclockwise permutation
where g(ω) is a phenomenological constant. In the end of this chapter, we will derive
Eq. (2.88) using a simple classical microscopic model of a medium. The permittivity
tensor (2.88) can be written in a matrix form as
(ω) ig(ω)B 0
ij (ω) = −ig(ω)B (ω) 0 . (2.89)
0 0
Let us now assume, for simplicity that the wave propagates along the magnetic
field, k = kez . It then follows from Eqs. (2.44) and (2.43) that
h 2
i 2
k 2 − ωc2 (ω) Ex − ig(ω)B ωc2 Ey = 0, (2.90)
h i
ω2 ω2
ig(ω)B c2 Ex + k 2 − c2 (ω) Ey = 0, (2.91)
and
2
− ωc2 Ez = 0. (2.92)
Exercise 2.6. Derive Eqs. (2.90) – (2.92) from Eqs. (2.44), (2.43), and (2.88).
We can then show that up to the first order in f B/ 1, wave vector magnitude of
any wave existing in such media is given by
k± = k ± ∆k, (2.93)
20
where we introduced the notations
ωp ωg(ω)B
k= (ω); ∆k = p . (2.94)
c 2c (ω)
The plane waves supported by the media must be circularly polarized, i. e.,
Ez = 0, Ey = ±iEx , (2.95)
where the upper (lower) sign on the right-hand side of Eq. (2.95) corresponds to the
upper (lower) subscript on the left-hand side of Eq. (2.93). In other words, the medium
supports left- and right-circularly polarized waves with slightly different wave num-
bers.
We will now explore how a linearly polarized wave evolves in the medium. As-
suming the wave is polarized along the x-axis, say, at the entrance to the medium,
where
ex ± iey
e± = √ , (2.98)
2
are the unit vectors associated with the two circular polarizations. We can now examine
wave propagation in the medium. The electric field in any transverse plane z = const
can be written as
where
ep (z) = ex cos ∆kz − ey sin ∆kz. (2.102)
Exercise 2.7. Derive Eq. (2.101) from Eq. (2.100)
It can be inferred from Eqs. (2.101) and (2.102) that the wave remains linearly po-
larized, but the plane of polarization rotates. Alternatively, one can conclude that the
polarization vector rotates in the transverse plane as the wave propagates along the z-
axis. This phenomenon is called Faraday rotation. The rate of rotation is customary
characterized by the Verdet constant V defined by the expression
∆k = V B, (2.103)
21
It then follows at once from Eqs. (2.94) and (2.103) that for a plane wave propagating
along the magnetic field, the Verdet constant is given by
ωg(ω)
V = p . (2.104)
2c (ω)
Exercise 2.8. Generalize the discussion of this section to the case when a plane wave
propagates at an angle θ to the magnetic field. Derive a generalized dispersion relation
and determine the Verdet constant in this case.
Medium 2 kt
(2 , 2 ) ân
t
r
ki i kr
Medium 1
(1 , 1)
Figure 2.7: Illustrating Snell’s law for oblique incidence of a plane wave.
22
unit normal to the interface pointed along the z-axis. The geometry of the problem is
sketched in Fig. 2.3.1. Note that the incidence, refraction, and transmission angles, θi ,
θr , and θt , respectively, are real angles only in the transparency window of the second
medium, 002 (ω) = 0. Otherwise, all sines and cosines of θt are complex. For this
reason, we will not use the angles hereafter. Rather, we will derive the Fresnel formulas
for transmission and reflection amplitudes in terms of the corresponding projections of
k-vectors which can, in general, be complex. Next, it will prove convenient hereafter
to introduce the notations
k1 = k0 n1 , k2 = k0 N2 , (2.105)
ei(ki ·r−ωi t) |z=0 = ei(kr ·r−ωr t) |z=0 = ei(kt ·r−ωt t) |z=0 . (2.106)
Here the subscripts i, r, and t stand for incident, reflected and transmitted waves, re-
spectively. It follows at once from Eq. (2.106) that
ωi = ωr = ωt = ω, (2.107)
that is the frequencies of the incident, reflected and transmitted waves must match.
Further, it can be inferred from the boundary conditions (2.106) that
In other words, the in-plane components of the wave vectors must match as well. No-
tice that since medium 1 is assumed to be transparent, Eq. (2.108) stipulates that in-
plane components of the wave vectors of all the involved waves be real. We stress that
Eq. (2.108) is a generalized Snell’s law. We shall also introduce the notations
ki = kx ex + k1z ez , (2.110)
kr = kx ex − k1z ez , (2.111)
and
kt = kx ex + k2z ez , (2.112)
where q q
k1z = k12 − kx2 , and k2z = k22 − kx2 . (2.113)
It is easy to see from the geometry of Fig. 2.3.1 that in the transparency window of
medium 2, all angles are real and Eq. (2.108 reduces to
θi = θr ≡ θ1 , θt ≡ θ2 (2.114)
23
and
n1 sin θ1 = n2 sin θ2 . (2.115)
Put another way, the incidence and reflection angles should be equal and the Snell law
should simplify to its familiar form for refraction at the interface of two transparent
media.
Hs = Hs ey , s = i, r, t. (2.116)
Since the magnetic field of a TM-polarized wave has only one component, it is conve-
nient to express the electric field in terms of the magnetic one. It follows at once from
the Maxwell equations (2.41) and (2.42) that
Hi = Hi ey ,
η0 Hi
Ei = (k1z ex − kx ez ), (2.118)
k 0 1
Hr = Hr ey ,
η 0 Hr
Er = (−k1z ex − kx ez ), (2.119)
k 0 1
and
Ht = Ht ey ,
η0 Ht
Et = (k2z ex − kx ez ), (2.120)
k0 2
24
respectively.
The boundary conditions for the tangential components of the fields across the
interface state
Hi + Hr = Ht (2.121)
and
Hi Hr Ht
k1z − k1z = k2z . (2.122)
1 1 2
It then follows from Eqs. (2.121) and (2.122) that
2 k1z − 1 k2z
Hr = Hi , (2.123)
2 k1z + 1 k2z
and
22 k1z
Ht = Hi . (2.124)
2 k1z + 1 k2z
Using (2.117) we arrive at the expressions for the electric fields in the form
Ei = η1 Hi , E r = η 1 Hr , E t = η 2 Ht . (2.125)
Er 2 k1z − 1 k2z
rp ≡ = , (2.126)
Ei 2 k1z + 1 k2z
and r
Et 22 k1z 1
tp ≡ = . (2.127)
Ei 2 k1z + 1 k2z 2
Equations (2.126) and (2.127) are the celebrated Fresnel formulas for the TM case.
Let us now focus on the situation when the wave is incident normally to the inter-
face, such that kx = 0, ksz = ks , s = 1, 2. It then follows from Eqs. (2.105), (2.126)
and (2.127) that
N2 − n1 2n1
r⊥ = , and t⊥ = . (2.128)
N2 + n1 N2 + n1
There are two instructive limiting cases here. First, the second medium is transparent,
N2 = n2 , such that the reflectivity and transmittance are purely real,
n2 − n1 2n1
r⊥ = , and t⊥ = , (2.129)
n2 + n1 n2 + n1
and the latter relations simply quantify the relative amplitudes of the reflected and
transmitted waves. Note that no energy will be lost in transmission in this case.
Exercise 2.9. A plane wave is normally incident at an interface separating two trans-
parent media. Show that the electromagnetic energy fluxes on both sides of the interface
are the same.
25
max(n1 , n2 ) in this case. Thus, the reflectivity and transmittance may be approximated
as
2in2 2in1
r⊥ ' 1 − and t⊥ ' − . (2.130)
κ2 κ2
It follows that most of the incident wave power is reflected from the interface of a good
conductor; only is its tiny fraction transmitted into the conductor.
Exercise 2.10. Consider a plane wave incident normally at the interface separating air
from a good conductor. Determine the portion of the incident wave power absorbed by
the conductor.
x
E
i
H
i âK
E
r
z
H
r
âK 1 2
1 , 1 , 1
Figure 2.8: Normal incidence of a plane wave onto an interface separating a dielectric
and a perfect conductor.
26
and
H1 = Re(Hi + Hr ) = 2ey |Hi | cos k1 z cos ωt. (2.136)
These equations describe standing waves carrying no energy which conforms to our
intuitive picture for reflection from a perfect conductor: The counterpropagating inci-
dent and reflected waves of equal amplitudes interfere to form a standing wave pattern
in medium 1.
Exercise 2.11. A right-hand circularly polarized wave, propagating in the positive
z-direction is normally incident on a perfect conductor wall z = 0. Determine (a) the
polarization of the reflected wave and (b) the induced current on the conducting wall.
Transverse electric (TE) or s-polarization. – In the TE case, the electric field is
normal to the incidence plane,
E i = Ei ey , (2.137)
and it is convenient to work with complex amplitudes of electric fields, expressing the
the magnetic field amplitudes as
(es × E s )
Hs = ; s = i, r, t. (2.138)
ηs
Similarly to the p-polarization case, we can obtain the expressions
E i = Ei ey
Ei
Hi = (−k1z ex + kx ez ), (2.139)
η0 k0
E r = Er ey ,
Er
Hr = (k1z ex + kx ez ), (2.140)
η0 k0
and
E t = Et ey ,
Et
Ht = (−k2z ex + kx ez ), (2.141)
η0 k0
for the complex amplitudes of incident, reflected, and transmitted fields, respectively.
The continuity of tangential components of electric and magnetic fields across the in-
terface leads to
Ei + Er = Et , (2.142)
and
(−Ei + Er )k1z = −Et k2z (2.143)
Solving the last pair of equations, we arrive at the complex reflectivity and transmit-
tance of an s-polarized incident wave in the form
Er k1z − k2z
rs ≡ = , (2.144)
Ei k1z + k2z
and
Et 2k1z
ts ≡ = . (2.145)
Ei k1z + k2z
27
2.3.3 Brewster angle and surface plasmon
polaritons
Z
Ԗ2
߳2
SPP
X
߳1
Let us return to the general case of p-polarized wave reflection form the interface
and study the behavior of reflectivity in more detail. We will assume both media to
be transparent, for simplicity. It can be inferred from Eq. (2.126) that the reflectivity
attains zero under the condition
2 k1z = 1 k2z . (2.146)
Solving Eq. (2.146), together with (2.113), we obtain expressions for the in-plane and
normal components of the wave vectors as
r
ω 1 2
kx = , (2.147)
c 1 + 2
and s
ω 2j
kjz = , j = 1, 2. (2.148)
c 1 + 2
The analysis of Eqs. (2.147) and (2.148) reveals two options. First, if both media
permittivities are positive, j > 0, we may introduce real refractive indices, nj =
√
j . It then follows at once from Eqs. (2.147) and (2.148) that there exists a special
incidence angle θB , given by the expression
tan θB = kx /k1z = n2 /n1 , (2.149)
such that there is no p-polarized reflected wave. This special incidence angle is known
as the Brewster angle. Alternatively, Eqs (2.147) and (2.148) describe a surface wave
propagating along the interface, kx2 > 0 and exponentially decaying in the direction
2
normal to the interface such that kjz is purely imaginary (for transparent media), kjz <
0, Eqs. (2.147) and (2.148) imply that this is possible under the conditions,
1 + 2 < 0, (2.150)
28
and
1 2 < 0. (2.151)
In other words, at least one of the permittivities must be negative. Usually, the wave is
incident form a dielectric medium, 1 > 0, implying that 2 < 0. The latter condition
can be realized for metals as we will see in Sec. 5.
These surface electromagnetic waves are known as surface plasmon polaritons
(SPP). Using Eqs. (2.118), (2.120), and (2.148), the electromagnetic fields of SPPs
on each side of the interface can be expressed as
ey Hi e−|k2z |z ei(kx x−ωt) , z > 0;
H(r, t) = (2.152)
ey Hi e|k1z |z ei(kx x−ωt) , z < 0,
and
(
η0 Hi −|k2z |z i(kx x−ωt)
k0 2 (i|k2z |ex − kx ez )e e , z > 0,
E(r, t) = η0 H i |k1z |z i(kx x−ωt) (2.153)
k0 1 (−i|k 1z |ex − k x ez )e e , z < 0.
Thus, SPP fields propagate along the interface and exponentially decay away from the
interface which is a characteristic signature of surface electromagnetic waves. In case
the second medium is an ideal metal, its permittivity can be successfully modeled by
the expression
ωp2
2 (ω) = 1 − 2 , (2.154)
ω
where ωp is the so-called plasma frequency. It can be seen from Eqs. (2.147) that in
the short wavelength approximation, kx → ∞, the SPP frequency tends to a constant
value, ω∞ given by the expression
ωp
ω∞ = √ . (2.155)
1 + 1
In this case, the SPP approaches its quasi-static limit termed a surface plasmon (SP).
Exercise 2.12. Show that Eq. (2.155) can be derived in the quasi-static limit by solving
Laplace’s equation for the electrostatic potential and matching the appropriate bound-
ary conditions.
So far, we have assumed that the SPPs propagate on the interface of two transpar-
ent media. In reality, of course, all metals are lossy, albeit losses are usually small at
optical frequencies. Realistic metals can then be properly described by complex di-
electric permittivities to account for Joules’ losses. Introducing a complex permittivity
of medium 2 viz.,
2 = 02 + i002 , (2.156)
and assuming that under at optical frequencies of interest |002 | |02 |, we can express
the in-plane component of the SPP wave vector as
kx = kx0 + ikx00 , (2.157)
where s
1 02
kx0 ' k0 , (2.158)
1 + 02
29
and s
1 02 002 1
kx00 ' k0 . (2.159)
1 + 02 202 (1 + 02 )
Here the imaginary part specifies a characteristic inverse damping distance of the SPP,
It follows from Eq. (2.147) that a plane wave in the air with 1 = 1 can never
excite a plasmon because of the wave vector mismatch: the plasmon wave vector com-
ponent along the interface is always greater than that of a plane wave in the air. One
way to generate an SPP then will be to nano-engineer the surface by creating periodic
imperfections such as grooves. The modified surface can serve as a diffraction grating
by shifting the in-plane wave vector component of the incident wave to achieve phase
matching. Introducing the lattice constant of the grooves a and assuming that the light
wave is incident from air, we can write down the matching condition
where θi is the incidence angle. This excitation scheme is sketched in the figure.
It can be readily inferred from Eq. (2.162) that the in-plane component of the wave
vector in medium 2 becomes purely imaginary,
r
n21
k2z = i|k2z | = ik2 n22
sin2 θ1 − 1, (2.163)
It then follows at once from Eqs. (2.126) and (2.163) that the for any wave incident at an
angle grater than the critical angle given by Eq. (2.164), the reflectivity is unimodular,
i. e.,
2 k1z − i1 |k2z |
rp∗ = . (2.165)
2 k1z + i1 |k2z |
30
Alternatively, the reflectivity of a totally internally reflected wave can be expressed as
where the phase can be expressed in terms of the incidence angle and refractive indices
of the media as
1 |k2z |
φp∗ = tan−1 . (2.167)
2 k1z
To better understand the behavior of the transmitted wave, we derive explicit ex-
pressions for its electric and magnetic fields. Using Eq. (2.163) in Eqs. (2.120), we can
cast complex amplitudes of the transmitted magnetic and electric fields into
and
η0 Ht
Et (r, t) = (i|k2z |ex − kx ez )e−|k2z |z ei(kx x−ωt) . (2.169)
2 k0
We can conclude from Eqs. (2.168) and (2.169) that the transmitted wave fields ex-
ponentially decay into medium 2. Next, let us determine the magnitude and direction
of the energy flow specified by the time-averaged Poynting vector. It follows from
Eqs. (1.26) (2.168), and (2.169) after some algebra that
2
42 k1z kx
hSt (z)i = ex Ii e−2|k2z |z , (2.170)
2 2
k0 (2 k1z + 21 |k2z |2 )
where Ii is an optical intensity of the incident wave. It can be concluded from Eq. (2.170)
that the power of the wave incident at an angle greater than the total internal reflection
angle does not flow into the less optically dense medium. Rather, it propagates along
the interface separating the two media, exponentially decaying in the direction normal
to the interface. This is a signature of a surface wave. Such surface waves generated by
total internal reflection are known as evanescent waves. The evanescent waves play a
prominent role in generating surface plasmon polaritons in the laboratory. Indeed, one
of the approaches to SPP generation employs evanescent waves. In practice, one uses
a device referred to as a Kretschmann prism shown in the figure below. The refractive
index of the prism makes it possible to match the in-plane wave vector components
for a plane wave launched through the Kretschmann prism under the conditions of to-
tal internal reflection to that of the SPP. The launch angle is then determined by the
matching condition,
kxpr = kxSPP , (2.171)
implying that r
d m
npr sin θSPP = , (2.172)
d + m
where npr is the refractive index of the prism, typically it is equal to 1.5 for a glass
prism, and d and m are the permittivities of the dielectric and metal on the two sides
of the interface supporting the SPP. In the figure, the SPP is produced at the metal-air
31
Figure 2.10: Illustrating SPP excitation with Kretschmann method. Reproduced from
Novotny& Hecht, Principles of Nanooptics.
interface by an evanescent wave tunneling across the metal film from the glass prism.
Exercise 2.14. Show that the reflectivity of a totally internally reflected TE-wave is
given by the expression
rs∗ = e−2iφs∗ , (2.173)
where
φs∗ = tan−1 (|k2z |/k1z ). (2.174)
Derive an expression for the transmitted energy flux.
32
3
3
2
2
1 1
Figure 2.11: Illustrating the multi-wave reflection and transmission through a film.
We will then find the reflected magnetic field in terms of the incident field by adding
up the contributions from reflected waves of all orders. Let us consider several lowest-
order reflected waves, labeling the reflection order with the corresponding superscript
assigned to Hr . The first-order reflected field is simply the field reflected from the first
interface once. Thus,
H(1)
r = r12 Hi ey . (2.176)
Next, the second-order reflected field is twice transmitted trough the first interface and
once reflected from the second one, i. e.,
H(2)
r = t12 t21 r23 e
i2k2z d
Hi ey , (2.177)
where we also included the accrued phase due to the optical path difference. By the
same token, the third- and fourth-order reflected waves can be represented as
H(3) 2
r = t12 t21 r23 r21 e
i4k2z d
Hi ey (2.178)
and
H(4) 3 2 i6k2z d
r = t12 t21 r23 r21 e Hi ey . (2.179)
Summing up the contributions to all orders, we obtain
∞
!
X
i2k2z d s s i2sk2z d
Hr = ey Hi r12 + r23 t12 t21 e r21 r23 e . (2.180)
s=0
33
Hence, performing the summation on the right-hand side of Eq. (2.180) and employing
Eq. (2.181), yields, after minor algebra, the expression
r ≡ Er /Ei , (2.183)
2k2 d = π, (2.186)
λ
d= . (2.187)
4n2
Eqs. (2.185) and (2.187) are compatible if the refractive index of the slab satisfies the
condition
√
n2 = n1 n3 . (2.188)
The constraints (2.187) and (2.188) establish requirements for reflectionless transmis-
sion of a normally incident plane wave through a dielectric film. In practice, these
conditions are taken advantage of in fabricating antireflection coatings of dielectric
surfaces such as antireflection glass coating to protect against glare or improve night
vision.
The second instance of no reflectivity corresponds to the generation of SPPs on
both surfaces of the film. Under the circumstances, the waves multiply reflected from
the film interfere constructively to transfer their energy into the SPPs. Thus all power
of the incident wave is channeled into the surface waves, resulting in no reflection.
In this case, the normal components of all wave vectors must be purely imaginary, a
34
signature of surface waves. In particular, the normal components of k in media 1 and
3 can be represented as q
k1z = −iq1 = −i kx2 − k12 , (2.189)
and q
k3z = iq3 = i kx2 − k32 , (2.190)
to ensure the exponential decay of the waves away from the interfaces. Note that these
definitions imply that q1,3 > 0 since the positive root is taken on the right-hand sides
of Eqs. (2.189) and (2.190). On the other hand, there exist both exponentially growing
and decaying waves inside the slab, implying that
q
k2z = iq2 = ±i kx2 − k22 . (2.191)
The SPP dispersion relation follows at once from Eqs. (2.126), (2.185) and Eqs. (2.189)
through (2.191):
−2q2 d 1 q2 + 2 q1 3 q2 + 2 q3
e = . (2.192)
1 q2 − 2 q1 3 q2 − 2 q3
In general, Eq. (2.192) describes a rather complicated dispersion relation. To gain
a better insight into the SPPs in the film, let us consider a particular case when media
1 and 3 are the same such that 1 = 3 and q1 = q3 . It can then be inferred from
Eq. (2.192) after a minor algebra that two families of SPPs exist in this case with the
dispersion relations governed by the equations
q2 d 1 q2
tanh =− , (2.193)
2 2 q1
and
q2 d 2 q1
tanh =− . (2.194)
2 1 q2
Exercise 2.15. Derive Eqs. (2.193)and (2.194).
It follows at once from Eqs. (2.193) and (2.194) that as the film thickness increases
without limit, d → ∞, both dispersion relations reduce to
1 q2 = −2 q1 . (2.195)
Since in this case, q2 < 0, one of the permittivities ought to be negative, 2 < 0, say.
Comparison of Eq. (2.195) with (2.146) leads to the conclusion that the SPPs on both
sides of a very thick film are uncoupled and have the same dispersion relation as the
SPP at the interface of two unbounded media.
In the other extreme of very thin films, d → 0, particularly simple results can be
obtained under the condition
1
2 q2 d 1. (2.196)
In other words, the characteristic penetration depth in medium 2, δ ' |q2 |−1 is much
smaller than half the film thickness. In physical terms, this condition implies strong
35
coupling between SPPs propagating on both sides of the film. Eqs. (2.193) and (2.196)
then yield an approximate expression
21
q1 ' − , (2.197)
2 d
for the normal component of the wave vector in medium 1. Further, the in-plane com-
ponent of the wave vector is given by
s
42
kx ' k02 1 + 2 12 , (2.198)
2 d
In particular, we apply our results to a thin metal film sandwiched between insulator
media (IMI). Such a thin-film IMI geometry implies the following conditions
Eqs. (2.196) through (2.200) will be consistent for genuinely thin films d λ0 , yield-
ing s
421
|q2 | ' k02 |2 | + . (2.201)
22 d2
such that the light penetration depth into the metal and dielectric are approximately
given by
1 |2 |d
δm ' , δd ' . (2.202)
|q2 | 21
This case would correspond to a 20 nm thin metal film, say, with 2 ∼ −20 illuminated
from glass 1 ' 1.25 by a light beam with λ0 ∼ 500 nm, for example. The SPP con-
finement is still rather tight δd ∼ 200 nm, and δm ∼ 50 nm.
Exercise 2.16. Plot an explicit dispersion relation curve ω = ω(kx ) given by Eq. (2.198).
What happens in the static limit, kx → ∞? You may assume an ideal metal with
(ω) = 1 − ωp2 /ω 2 .
Exercise 2.17. Use the Airy technique to show that the transmittance of the slab exam-
ined in this section is given by the expression
|Et |2
T ≡ , (2.204)
|Ei |2
36
and show that T can be expressed as
T2 1
T = . (2.205)
(1 − R) 1 + F sin2 δ
2
Here
2πnd
δ = δr + cos θt , (2.206)
λ
and we introduced the transmission and reflection coefficients for each interface of the
slab, T and R, respectively, and the interferometer finesse F by the expression
4R
F = . (2.207)
(1 − R)2
The considered system serves as a basis for a Fabry-Perot interferometer used to pre-
cisely measure the wavelength of light. It can be inferred from Eq. (2.205) –(2.207)
that for large enough reflectance, R ' 1, T has very sharp maxima at
δ = πm, m = 0, 1, 2, . . . . (2.208)
In the ideal case, T = 1 at the maxima and T = 1/F at the minima. Thus boosting
the finesse, one can increase the contrast of the interferometer. The distance between
the adjacent maxima can be determined from Eqs. (2.206) and (2.208) to be
λ
dm+1 − dm = . (2.209)
2n cos θ
For sufficiently small angles, θ ' 0, the latter reduces to
λ
dm+1 − dm ' . (2.210)
2n
Eq. (2.210) can be used to infer the value of λ from the measurements of the maxima
positions.
37
2.5 Classical theory of optical dispersion and absorp-
tion
2.5.1 Lorentz-Kramers expression for dielectric permittivity
ݒ
ݔ
0
Figure 2.13: Schematic of a trapping Coulomb potential (solid) for an electron in an
atom and its harmonic approximation (dashed) near the electron equilibrium position;
x is a scalar displacement away from equilibrium.
To drive this point home, we develop a simple classical model of matter response to
an external time-harmonic field. In this model atoms are treated as simple harmonic
oscillators. A linear restoring force proportional to an electron displacement from its
equilibrium position–in the classical sense, of course–is due to a harmonic interaction
potential between an electron and the other electrons in an atom as well as the atomic
nucleus. In reality each atomic electron is trapped by a complicated electrostatic po-
tential which is strongly anharmonic. However, so long as the applied electric field
is sufficiently weak such that the electron displacement from its equilibrium position
is small compared to the characteristic atomic size, the electrostatic Coulomb poten-
tial in the vicinity of the electron equilibrium position can be well approximated by a
harmonic one. The situation is schematically depicted in the figure.
Further, we assume that each atom has Z bound electrons. Assume also that there
are fs electrons per atom having the binding frequency ωs which corresponds to a
particular type of the trapping harmonic potential. The quantities {fs } are referred to
as the oscillator strengths.
38
Whenever an electron having the binding frequency ωs is displaced by the displace-
ment vector rs in response to the external electric field, it experiences three forces: the
restoring force, Fr = −mωs2 rs , the damping force, Fd = −2mγs ṙs –where γs is a
phenomenological damping constant–and the force due to the external electric field,
Fe = −eEω e−iωt .
The equation of electron motion (second law of Newton) is then
Here each “dot” stands for a time derivative. We seek a driven solution to Eq. (2.212)
in the form,
rs (t) = rsω e−iωt . (2.213)
It follows from Eqs. (2.212) and (2.213) that the electron displacement amplitude is
eEω
rsω = − , (2.214)
m(ωs2 − ω 2 − 2iωγs )
implying that
eE(t)
rs (t) = − . (2.215)
m(ωs2 − ω 2 − 2iωγs )
The induced individual dipole moment of the electron of this type will be ps = −ers .
Next, if there are N atoms per unit volume, the induced polarization is
X X N e2 X fs E(t)
P(t) = N fs ps (t) = −N e fs rs (t) = .
s s
m s (ωs − ω 2 − 2iωγs )
2
(2.216)
Note that the oscillator strengths satisfy the so-called sum rule
X
fs = Z. (2.217)
s
N e2 X
(ω) = 1 + fs Ls (ω), (2.218)
0 m s
Eqs. (2.218) and (2.219) give a classical expression for the dielectric permittivity of
materials as a function of frequency of the applied electric field. The real part describes
dispersion while the imaginary part accounts for light absorption by medium atoms.
The latter simply because we identified the imaginary part of with losses as the light
propagates through the medium (c.f. Sec. 2.3). Clearly, the light wave loses its energy
to the medium atoms which is a classical picture of light absorption.
39
Let us now explore what happens if the frequency of the applied electric field is
close to a particular resonant frequency of the material. For the sake of clarity, let that
be the lowest bound frequency of the dielectric, ω0 6= 0, i.e, ω ≈ ω0 . In this case, we
can single out the resonant term in Eq. (2.218) implying that
N e2 f0 1
(ω) = NR (ω) + 2 2
. (2.220)
0 m (ω0 − ω − 2iωγ0 )
−ω 2 + ω02 − 2iγ0 ω ' 2ω(ω0 − ω − iγ0 ) ' 2ω0 (ω0 − ω − iγ0 ). (2.222)
It can be inferred from Eqs. (2.221) and (2.222) that the permittivity near optical reso-
nance can be represented as
where
N e 2 f0
0 (ω − ω0 )
(ω) = NR (ω) + , (2.224)
20 mω0 (ω − ω0 )2 + γ02
and
N e 2 f0
00 γ0
(ω) = . (2.225)
20 mω0 (ω − ω0 )2 + γ02
The real and imaginary parts of the permittivity are sketched as functions of the fre-
quency in Fig. 2.5.
As is seen in Fig. 2.5., the real part of the permittivity sufficiently far below and
above the resonance frequency increases with the frequency. Such a behavior is known
as normal dispersion. In the vicinity of resonance, however, 0 decreases with the fre-
quency which is referred to as anomalous dispersion. Optical absorption is generally
small far from resonance, but is seen to sharply increase as we approach the reso-
nance frequency. Notice also that in regions of weak dispersion are nearly transparent,
whereas strong dispersion is accompanied with pronounced absorption as well. This
connection is not accidental. In fact, we show in the following chapters that there are
fundamental quantitative relations, the Kramers-Kronig relations that link dispersive
and absorptive properties of optical media.
The difference between realistic conductors and dielectrics can be attributed to the
presence of free electrons in the former. Indeed, by looking into the low-frequency
limit, we notice that for pure dielectrics the lowest bound frequency must be nonzero,
while conductors can have a fraction of electrons, f0 , say, that have ω0 = 0; those
40
χ"(ω)
χ ' (ω )
ω0 ω→
Figure 2.14: Imaginary (top) and real (bottom) parts of the electric permittivity as
functions of frequency near resonance.
N f0 e2
c (ω) = b (ω) + i , (2.226)
0 mω(2γ0 − iω)
where b is the overall contribution of the bound electrons with ωs 6= 0. Since free
electrons can conduct currents, we can use Eq. (2.215) to determine the current density
to be
N f0 e2
J = −N ef0 ṙ0 = E. (2.227)
m(2γ0 − iω)
On comparing Eqs (2.37) and (2.227), we infer the expression for the conductivity,
N f0 e2
σ(ω) = . (2.228)
m(2γ0 − iω)
N f0 e2
σ→ = σ0 , (2.229)
2mγ0
the conductivity is real, describing dc currents. In view of Eq. (2.229), the expression
for σ can be cast into the form
σ0
σ(ω) = , (2.230)
1 − iωτ
41
where τ = 1/2γ0 is a characteristic time for current relaxation in conductors.
Next, comparing Eqs. (2.226) and (2.228), we can express the former as
σ
c (ω) = b (ω) + i . (2.231)
0 ω
Eq. (2.231) implies that losses in real conductors/metals come in two guises: absorp-
tion of electromagnetic waves by bound electrons–which is described by the imaginary
part of b –and ohmic losses due to generating electric currents as described by the sec-
ond term on the right-hand side of Eq. (2.231).
Exercise 2.18. Use the limiting case of Eq. (2.38) for isotropic media with no spatial
dispersion and Eq. (2.231) to relate real and imaginary parts of permittivity and con-
ductivity. Thus, you may argue that the distinction between conductors and dielectrics
is rather artificial at optical frequencies.
Next, we note that at the frequencies far exceeding the highest bound frequency,
ω max(ωs ), dielectrics and conductors respond to the applied electric field the same
wave. In this limit, we can neglect all {ωs } and {γs } in the denominator of Eq. (2.218),
leading to
ωp2
c (ω) = 1 − 2 , (2.232)
ω
where we used Eq. (2.217) and introduced the plasma frequency
r
N Ze2
ωp = . (2.233)
m
Incidentally, Eq. (2.232) is a simplified form of the so-called Drude expression for
a dielectric constant of a metal. The Drude model describes well noble metals; it
follows from Eq. (2.232) that becomes negative for the frequencies above the plasma
frequency.
Finally, we note that the polarization caused by a monochromatic applied electric
field in an isotropic linear medium can be represented as
P(r, ω) = 0 χ(ω)E(r, ω), (2.234)
where χ(ω) is a linear susceptibility of the medium. In case of an optical pulse, con-
sisting of many monochromatic components, the electric field of the pulse can be rep-
resented as a Fourier integral viz.,
Z ∞
dω
E(r, t) = E(r, ω)e−iωt , (2.235)
−∞ 2π
where E(r, ω) is the spectral amplitude of the pulse. The polarization field induced by
each spectral component of the pulse is given by
P(r, ω) = 0 χ(ω)E(r, ω). (2.236)
It follows at once from Eqs. (2.235) and (2.236) that the overall polarization field in-
duced by the pulse is given by a time convolution,
Z ∞
P(r, t) = 0 dt0 χ(t − t0 )E(r, t0 ). (2.237)
−∞
42
We will return to Eq. (2.237) in Chap. 4 where we will present a general theory of opti-
cal response of nonlocal noninstantaneous nonlinear media to electromagnetic pulses.
ie2 ω 2
r(1)
sω = − L (ω)[Eω × B0 ]. (2.242)
m2 s
Combining Eqs. (1.6) and (2.36, which furnish a macroscopic description of permittiv-
ity, with the classical microscopic picture of Eqs. (2.216) as well as with Eqs. (2.239)
through (2.242), we finally obtain the following expression for the permittivity tensor,
X
ij (ω) = (ω)δij + ig(ω) eijp B0p . (2.243)
p
Here
N e2 X
(ω) = 1 + fs Ls (ω), (2.244)
0 m s
43
is a dielectric permittivity of an isotropic medium and
N e3 ω X
g(ω) = fs L2s (ω), (2.245)
0 m2 s
44
Chapter 3
where frequency dispersion enters through the dependence of the dielectric permittivity
on the wave frequency. The corresponding wave equation takes the form
2
∇2 Ẽ + (ω) ωc2 Ẽ = 0. (3.2)
Here k0 is a wave number associated with the carrier frequency ω0 , and a slowly-
varying envelope is assumed such that
∂z Ẽ k0 Ẽ, (3.4)
Eqs. (3.3) and (3.4) represent a spectral envelope amplitude of a slowly varying optical
pulse. On substituting from Eqs. (3.3) and (3.4) into Eq. (3.2), we arrive at the paraxial
wave equation in the space-frequency representation,
45
where we introduced the frequency-dependent wave number viz.,
ω2
k 2 (ω) = (ω) . (3.6)
c2
Suppose now the bandwidth of the pulse is small compared to the carrier frequency,
i.e,
∆ω = 2|ωmax − ω0 | ω0 , (3.7)
where ωmax is the frequency of the highest harmonic within the pulse associated with
a finite amplitude. The combined approximations (3.4) and (3.7) constitute the slowly
varying envelope approximation (SVEA) for optical pulses. The SVEA implies that
that is the pulse envelope changes slowly over an optical cycle. To this level of accu-
racy, we can then expand the wave number in a Taylor series as
1
k(ω) ' k0 + k 0 (ω0 )(ω − ω0 ) + k 00 (ω0 )(ω − ω0 )2 . (3.9)
| {z } 2! | {z }
k1 k2
i∂z Ẽ + k1 ω 0 Ẽ + 21 k2 ω 02 Ẽ = 0. (3.10)
The overall electric field can then be factorized into a fast carrier wave and slowly
varying pulse envelope as
Z +∞
0
i(k0 z−ω0 t)
E(t, z) = ex e| {z } dω 0 e−iω t Ẽ(ω 0 , z) . (3.11)
−∞
carrier wave | {z }
slow envelope
we can derive, using Fourier transform properties, a paraxial wave equation for the
temporal envelope
2
2i(∂z E + k1 ∂t E) − k2 ∂tt E = 0. (3.13)
It is now convenient to transfer to a moving reference frame by introducing the coordi-
nate transformation
ζ = z; τ = t − k1 z, (3.14)
One can then re-calculate the derivatives using the chain rules
2
∂t E = ∂τ E; ∂tt E = ∂τ2τ E, (3.15)
46
and
∂z E = ∂ζ E − k1 ∂τ E, (3.16)
to arrive at the final form of the governing pulse propagation equation in linear disper-
sive media
2i∂ζ E − k2 ∂τ2τ E = 0. (3.17)
To elucidate physical meaning of each term in Eq. (3.17), we observe that if one
assumes that at the carrier frequency, k2 (ω0 ) = 0, we arrive at the greatly simplified
equation
∂ζ E = 0, (3.18)
with the solution
E(t, z) = E0 (t − z/vg ), (3.19)
where E0 (t) is a pulse envelope in the source plane, and we introduced
k1 ≡ vg−1 . (3.20)
It can be concluded from Eq. (3.19) that the pulse maintains its shape and its peak
moves inside the medium with the speed vg . This velocity is referred to as the group
velocity of the pulse. To understand the role of k2 , it is sufficient to observe that
Eq. (3.17) is a temporal analog of the paraxial wave equation governing beam diffrac-
tion in free space we have studied before. Hence the second derivative term in Eq. (3.17)
describes pulse spreading in dispersive media. The group velocity dispersion co-
efficient k2 then sets a spatial scale of the problem, the so-called dispersion length,
Ldis = t2p /k2 , where tp is a characteristic duration of the pulse in the source plane
z = 0.
In the preceding development, we ignored spatial distribution of the pulse, which
is justified in a plane wave geometry. Alternatively, pulse propagation in single-mode
dispersive fibers can be of interest. In this case, the spatial distribution of the pulse is
dictated by the fiber mode such that a more appropriate Ansatz for the field,
47
and replace k(ω) and β(ω) in the equation for the fiber mode by their values at the
carrier frequency, i.e,
∇2⊥ φ + [k 2 (ω0 ) − β02 ]φ = 0. (3.25)
The resulting eigenvalue equation, subject to the appropriate boundary conditions at
the fiber boundaries, determines the spatial distribution of the fiber mode and the mode
propagation constant. Further, expanding the frequency dependent propagation con-
stant β(ω) in a Taylor series up to the second order
1
β(ω) ' β0 + β 0 (ω0 )(ω − ω0 ) + β 00 (ω0 )(ω − ω0 )2 , (3.26)
| {z } 2! | {z }
β1 β2
and following exactly the same procedure as before, we can arrive at the paraxial wave
equation for pulse propagation in linear fibers as
where E is the electric field of the pulse in the scalar approximation. In the slowly-
varying envelope approximation (SVEA), the pulse field and atomic dipole moments
can be represented as
48
where ∆ is a detuning of the carrier wave frequency ω from the atomic resonance fre-
quency ω0 . On substituting from Eq. (3.33) into (3.32), we obtain, after some algebra,
the SVEA equation for atomic dipole envelope evolution as
This is called free-induction decay of an individual dipole moment. One can introduce
a characteristic time T0 = 1/γ which is known as a dipole relaxation time.
A Fourier transform of σ can be defined as
Z ∞
σ̃(ω, z) ≡ dt σ(t, z)e−iωt . (3.39)
−∞
The spectral response, S0 (ω, z) ∝ |σ̃(ω, z)|2 , obtained in a typical set of absorption
measurements, is then given by
|σ(0, z)|2
S0 (ω, z) ∝ (3.40)
(ω − ω0 )2 + γ 2
The characteristic absorption spectral width is thus γ = 1/T0 and is referred to as the
width of homogeneous broadening as it is the same for each individual atom.
49
3.2.2 Inhomogeneous broadening
Consider the polarization of a macroscopic sample of atoms. Generally, in solid state
samples, the resonant frequency ω0 of atoms will vary from atom to atom due to local
defects which perturb the atomic transition frequencies. As a result, the polarization is
determined as an average over the resonant frequency fluctuations such that
P (t, z) = 12 [P(t, z)ei(kz−ωt) + c.c], (3.41)
where
P(t, z) = N d0 hσ(t, z, ω0 )i, (3.42)
and the averaging is defined as
Z ∞
hσ(t, z, ω0 )i = dω0 f (ω0 )σ(t, z, ω0 ). (3.43)
0
In reality, the distribution function is often sharply peaked around some value of ω0
which we denote by ω 0 , say, i. e.,
f (ω0 ) ' f (ω0 − ω 0 ) = f (∆).
It then follows by changing the integration variable to ∆ that for any average,
Z ∞ Z ∞ Z ∞
dω0 f (ω0 )(. . .) = d∆ f (∆)(. . .) ' d∆ f (∆)(. . .).
0 −ω 0 −∞
Thus, Z ∞
P(t, z) = N d0 d∆ f (∆)σ(t, z, ∆). (3.44)
−∞
In gases or atomic vapors, Doppler’s effect is at the origin of the frequency detuning
distribution. To make this point clear, suppose a plane wave propagating in a laboratory
frame has the form ei(k·r−ωt) . In the reference frame moving with the atom at the
0 0
velocity v, the plane wave has the form ei(k·r −ω t) , where r0 = r − vt is a position
of the atom at time t. It then follows that the wave form will be the same in the two
frames–which it should as it is the same wave–if the frequencies ω 0 and ω in the moving
and laboratory frames, respectively, are related as ω 0 = ω − k · v. The frequency shift
of the wave in a moving reference frame is known as the Doppler effect. For a plane
wave propagating in the positive z-direction, the Doppler shifted frequency is
ω 0 = ω − kvz . (3.45)
Next, the pulse field and atomic dipole moment distributions in the moving refer-
ence frame are
E(z, t) = 21 [E(z, t)ei[kz−(ω−kvz )t] +c.c]; ex(z, t) = 21 [d0 σ(z, t)ei[kz−(ω−kvz )t] +c.c],
(3.46)
50
and we dropped the prime over z to simplify the notation. The derivation along the
lines outlined in the previous Lecture would yield the dipole evolution equation in the
form
∂t σ = −(γ + i∆)σ + iΩ, (3.47)
where
∆ = ω0 − ω + kvz . (3.48)
Assuming that ω = ω0 –the light is tuned to the atomic transition at rest–we obtain the
dependence of the detuning on the atom velocity,
∆ = kvz . (3.49)
The atom velocities are distributed according to Maxwell’s distribution such that for
the z-component of velocity, we have
mvz2
f (vz ) ∝ exp − , (3.50)
2kB T
where kB is the Boltzmann constant and T is the temperature. It then follows from
Eqs. (3.49) and (3.50) that the detuning distribution is Maxwellian in this case,
m∆2
f (∆) ∝ exp − 2 , (3.51)
2k kB T
Let us now revisit the free-induction decay experiment and examine the polariza-
tion evolution, Z ∞
P(t, z) = N d0 d∆ f (∆)σ(t, z), (3.52)
−∞
which can be rewritten in the free-induction decay as
Z ∞
P (t, z) ∝ N d0 e−t/T0 eiω0 t d∆ f (∆)ei∆t + c.c. (3.53)
−∞
where 1/T∆ characterizes the width of g(∆). Using a Fourier transform table integral,
1
F 2 ∝ e−|t|/T∆ ,
∆2 + 1/T∆
we obtain for t > 0,
P (t, z) ∝ N d0 e−t/Tef f eiω0 t + c.c. (3.55)
Here
1 1 1
= + . (3.56)
Tef f T0 T∆
|{z} |{z}
homogeneous inhomogeneous
51
The second term on the rhs describes inhomogeneous broadening which would occur
in the spectral domain due to fluctuations of atomic detunings; its nature is atom spe-
cific (distribution of resonant frequencies, velocity distributions, etc.) The functional
form of g(∆) and the magnitude of a characteristic damping time T∆ associated with
inhomogeneous broadening depend on a specific broadening mechanism.
P = −N ehxi. (3.58)
In Eq. (3.58), the angle brackets denote averaging over detunings of the pulse from the
resonance frequency ω0 .
In the slowly varying envelope approximation, we can use the representation (3.46)
and assume that
∂z E kE, ∂t E ωE (3.59)
and
∂t σ ωσ. (3.60)
On substituting from Eq. (3.46) into (3.57) and using the SVEA (3.59), we can obtain
the reduced wave equation for the slowly-varying field envelope as
which should be coupled with the derived dipole moment evolution equation (3.47). In
2
Eq. (3.61), we introduced a coupling constant, κ = ωpe /4c, where ωpe = (N e2 /0 m)1/2
is the electron plasma frequency.
Exercise 3.1. Derive Eq. (3.61).
Transforming to the moving reference frame via τ = t − z/c and ζ = z just as we
did in the derivation of nonresonant pulse propagation equation, we finally arrive at the
coupled Maxwell-Lorentz propagation equations
∂ζ Ω = iκhσi, (3.62)
and
∂τ σ = −(γ + i∆)σ + iΩ. (3.63)
To solve Eqs. (3.62) and (3.63) we use the familiar now Fourier transform tech-
nique. First, we introduce temporal Fourier transforms of the field and dipole moment
as Z ∞
Ω(τ, ζ) = dω Ω̃(ω, ζ)e−iωτ , (3.64)
−∞
52
and Z ∞
σ(τ, ζ) = dω σ̃(ω, ζ)e−iωτ , (3.65)
−∞
Substituting those back into our evolution equations, we obtain the algebraic expression
for σ̃ in the form
iΩ̃(ω, ζ)
σ̃(ω, ζ) = . (3.66)
γ + i(∆ − ω)
It then follows from Eq. (3.66) and a Fourier transformed Eq. (3.62) that
where ∞
dt0 iωt0 0
Z
Ẽ(ω) = e E(t , 0). (3.71)
−∞ 2π
On combining Eqs. (3.70) and (3.71), we can express the answer in the original vari-
ables in the form
Z ∞ 0 Z ∞
dt 0
E(t, z) = E(t0 , 0) dω eiω(t −t) exp[iωz/c − κR(ω)z]. (3.72)
−∞ 2π −∞
Integrating Eq. (3.72) over time and using the integral representation of the delta func-
tion, Z ∞
dω −iωt
δ(ω) = e , (3.75)
−∞ 2π
53
we arrive at the area theorem
where A0 = A(0) is the initial area under the pulse profile. In general, the area theorem
can be cast into the form
A(z) = A0 e−αz/2 eiβz/2 , (3.77)
where we introduced a characteristic attenuation decrement α and the phase accumu-
lation factor β by the expressions
2κγ
α= , (3.78)
γ 2 + ∆2
and
2κ∆
β= . (3.79)
γ + ∆2
2
Thus, regardless of the incident pulse shape, the area under the pulse will exponentially
decay on pulse propagation in linear resonant absorbers as a consequence of medium
absorption manifested, in general, through homogeneous and inhomogeneous broad-
ening.
Exercise 3.3. Derive Eqs. (3.76) and (3.77).
Finally, we examine the case of very long pulses such that the characteristic pulse
width Tp is much longer than the longer of homogeneous or inhomogeneous damping
times,
Tp max(T0 , T∆ ). (3.80)
It then follows from Eq. (3.63) that the dipole moment can be adiabatically eliminated:
It decays fast to its dynamic equilibrium value determined by the pulse amplitude.
Mathematically, we can formally set ∂τ σ ' 0 in Eq. (3.63) and conclude that
iΩ
σ' , (3.81)
γ + i∆
On substituting back into Eq. (3.62) we arrive at the pulse evolution equation as
1
∂ζ E = −κ E. (3.82)
γ + i∆
54
3.3 Paraxial wave equation and Gaussian beam optics
We consider evolution of a monochromatic electromagnetic field in free space. The
electric and magnetic fields can be represented as
E(r, t) = E(r, ω)e−iωt , H(r, t) = H(r, ω)e−iωt . (3.84)
Thus Maxwell’s equations for the field envelopes read
∇ × E = iµ0 ωH, (3.85)
∇ × H = −i0 ωE, (3.86)
and
∇ · E = 0, ∇ · H = 0. (3.87)
Eliminating the magnetic field in favor of the electric in Eqs. (3.85) – (3.87), we arrive
at the equation for the electric field envelope in the form
∇2 E + k 2 E = 0, (3.88)
where k = ω/c.
We seek a plane polarized beam-like solution to (3.88):
E = ey E(x, z)eikz . (3.89)
Physically, the solution (3.89) represents a beam of light propagating in the z-direction
with an homogeneous electric field in the y-direction and an inhomogeneous inten-
sity distribution in the x-direction. It automatically satisfies the transversality condi-
tions (3.87). Note that in the limiting case when E = const, we have a plane wave.
The beam is different in that its field amplitude should in some sense be a slowly vary-
ing function of coordinates. To make this requirement more quantitative we stipulate
that for the intensity distribution to represent a beam, the complex envelope E change
slowly at the wavelength scale, i. e.,
∂z E kE, (3.90)
The latter condition is referred to as a slowly-varying amplitude approximation (SVEA).
On substituting from Eq. (3.89) and taking the SVEA into account, we arrive at the
paraxial wave equation for the beam envelope in the form
2
2ik∂z E + ∂xx E = 0. (3.91)
Let us now study the evolution of the beam with a Gaussian field profile in the
source plane z = 0,
2 2
E(x, 0) = E0 e−x /2w0 , (3.92)
where w0 characterizes the width of the source intensity profile. We use a Fourier
transform method to address the problem. Consider a Fourier decomposition of the
beam amplitude in the transverse direction,
Z +∞
E(x, z) = dqeiqx Ẽ(q, z), (3.93)
−∞
55
where the Fourier (spectral) amplitude can be determined by the inverse transformation,
Z +∞
dx −iqx
Ẽ(q, z) = e E(x, z). (3.94)
−∞ 2π
iq 2 z
Ẽ(q, z) = Ẽ(q, 0) exp − . (3.98)
2k
Combining Eqs. (3.95) and (3.98) and using the inverse Fourier transform (3.94), we
obtain after some algebra the expression for the Gaussian beam envelope at any z,
x2
E0
E(x, z) = √ exp − 2 . (3.99)
1 + iζ 2w0 (1 + iζ)
Here
ζ = z/zR , zR = kw02 . (3.100)
Exercise 3.5. Derive Eq. (3.99).
To discuss the solution (3.99) it is convenient to represent it in the form where the
complex phase and real amplitude are expressed explicitly as
ikx2 x2
r
w0 iΦ(z)
E(x, z) = E0 e exp exp − 2 . (3.101)
w(z) 2R(z) 2w (z)
56
and the accrued phase Φ(z),
Φ(z) = − 21 arctan(z/zR ). (3.104)
Notice first that although the intensity of a Gaussian beam steadily decreases upon
diffraction in free space, the beam profile remains Gaussian in any transverse plane
z = const. Further, the diffraction length zR sets the characteristic spatial scale for
the problem. It is equal to the distance over which the beam width doubles from its
minimal value w0 at the source. The plane where the beam width is the smallest is
called the beam waist and the diffraction length is often referred to as the Rayleigh
range.
Consider now the wavefront Ψ(x, z) of the beam which is defined as a surface of
constant phase. It follows from Eq . (3.101) that
kx2
Ψ(x, z) = Φ(z) + = const (3.105)
2R(z)
We observe that near the waist of the beam, z zR , the radius of the curvature is very
2
large, R ' zR /z, implying that in the limit z → 0, R → ∞, and the wavefront is flat.
In the opposite limit, z → +∞, the accrued phase is Φ = −π/4. This is the so-called
Gouy phase shift of a Gaussian beam. Finally for large but finite propagation distances,
z zR such that R(z) ' z, the wavefront is parabolic
z ∝ x2 /λ, (3.106)
with the curvature decreasing in the inverse proportion to the propagation distance. The
curvature attains its maximum at the Rayleigh range.
Finally, we mention that a natural generalization of the paraxial equation to two
transverse dimensions is
2ik∂z E + ∇2⊥ E = 0, (3.107)
where ∇2⊥ is a Laplacian operator in the transverse plane defined as
∇2⊥ ≡ ∂xx
2 2
+ ∂yy . (3.108)
The electromagnetic field is supposed to propagate in free space into the half space
z > 0. The representation of the field by Eq. (3.109) is known as the angular spectrum:
The field is composed of plane waves propagating at different angles to the z-axis.
57
Substituting from Eq. (3.109) into the wave equation, we obtain the equation for
the spectral amplitude A as
It follows at once from Eq. (4.184) that A is constrained to lie on the circle in the
k-space, i.e,
Ã(kx , kz ) = A(kx )δ(kx2 + kz2 − k 2 ). (3.111)
The circle in the k-space determines the dispersion relation for the wave vector com-
ponents, p
kx2 + kz2 = k 2 =⇒ kz = k 2 − kx2 . (3.112)
It can then be inferred from Eq. (3.112) that
p
k 2 − kx2 , kx < k
kz = p (3.113)
±i kx2 − k 2 , kx > k
Combining Eqs. (3.109) and (3.113), we arrive at the angular spectrum representation
of any linearly polarized (1 + 1)D electromagnetic field in the half-space z > 0
Z √ 2 2 Z √ 2 2
E(x, z) = ey dkx A(kx )ei(kx x+ k −kx z) +ey dkx A(kx )eikx e− kx −k z .
k <k k >k
| x {z } | x {z }
homogeneous waves evanescent waves
(3.114)
The first and second terms provide contributions of homogeneous and evanescent plane
waves; the latter exponentially decay away from the source plane z = 0. Notice inci-
dentally that we chose “ + ” sign to have the evanescent waves decay into z > 0 as the
exponentially growing solution does not obviously make any sense.
Next, the evanescent waves quickly damp out as the field propagates sufficiently
far from the source and their contribution is negligible outside of the source vicinity.
Thus, we have
Z √ 2 2
E(x, z) = ey dkx A(kx )ei(kx x+ k −kx z) . (3.115)
kx <k
Let us now specialize to the beam case whereupon all the plane waves making up the
field propagate close to the z-axis such that kx k. It then follows upon a Taylor
series expansion in Eq. (3.113) that
p kx2
k 2 − kx2 ' k − ,
2k
Therefore we can rewrite our plane wave decomposition as
Z +∞
ik 2 z
E(x, z) ' ey eikz dkx A(kx ) eikx x exp − x . (3.116)
−∞ 2k
On comparing Eqs. (3.116) and
58
we conclude that we can represent electric fields of optical beams as
Z +∞
ik 2 z
E(x, z) = dkx A(kx ) eikx x exp − x . (3.118)
−∞ 2k
Hence,
ik 2 z
Ẽ(kx , z) = Ẽ(kx , 0) exp − x , (3.120)
2k
which coincides with Eq. (3.98). Thus our angular spectrum representation treatment
is equivalent to the paraxial equation approach. While the latter is usually more con-
venient to solve practical problems and is straightforwardly generalized to nonlinear
situations, the former brings up more insight into the physics of beam propagation in
free space.
Finally, applying the convolution theorem of Fourier transforms to Eq. (3.120) and
using Eq. (3.96) we can derive the Fresnel representation for any (1 + 1)D beam evo-
lution in free space:
r Z +∞
ik(x − x0 )2
k
E(x, z) = dx0 E(x0 , 0) exp . (3.121)
2πiz −∞ 2z
ik(ρ − ρ0 )2
Z
k
E(ρ, z) = dρ E(ρ0 , 0) exp , (3.122)
2πiz 2z
where ρ = xex + yey is a radius vector in the transverse plane of the beam.
59
Chapter 4
Nonlinear optics
P = 0 χ(1) E, (4.1)
where χ(1) is the usual susceptibility of linear optics. In writing Eq. (4.1) we ignored,
for simplicity, the vector nature of both the applied field and the resulting polarization.
As the magnitude of the field increases though, the simple linear relation (4.1)
no longer holds. However, typical electric fields generated by all but most powerful
modern lasers are in the range of 106 to 107 V/cm, whereas the electrons bound to
atoms or molecules experience far greater fields of the order of 109 to 1010 V/cm.
Consequently, one can assume the induced electron displacements in laser fields to be
rather small; the latter circumstance justifies using a power series representation for the
induced polarizations as
where χ(2) and χ(3) are referred to as second- and third-order susceptibilities, respec-
tively.
To estimate orders of magnitude of the nonlinear susceptibilities, we consider non-
linearity of electronic origin. In this case, the nonlinear polarization depends on the
displacements of the electrons from the nuclei. One could expect that the second-order
contribution to the polarization would definitely be of the same order as the first one
if the electrons are displaced a distance as large as the atomic size, which is roughly
of the order of the Bohr radius, a0 = h̄2 /me2 ' 5 × 10−9 cm. The correspond-
ing electric field would be comparable with the field binding electrons to a nucleus,
60
Eat = e/4π0 a20 ' 5 × 1011 V/m. As the linear susceptibility is of the order of unity,
χ(1) ∼ 1, it follows that the second-order susceptibility can be estimated as
−1
χ(2) ∼ Eat ∼ 10−12 , m/V. (4.3)
By the same token, a typical value of the third-order susceptibility for condensed-
matter systems would be
It can be readily inferred from Eqs. (4.3) and (4.4) that (a) one needs very large fields
indeed to probe nonlinear response of dielectric materials and (b) for most laser field
strengths encountered in practice, each higher-order contribution to the polarization
field P is much smaller than the corresponding lower-order one, enabling us to take
into account only the lowest order nonvanishing contribution to P in a given nonlinear
medium.
In the following subsection, we are going to discuss nonlinear optical susceptibil-
ities semi-quantitatively. A note of caution is due before we proceed any further: The
just introduced expansion (4.2) fails in the vicinity of any internal atomic resonance
of the medium, where nonlinear saturation effects start playing a role. Hence, a more
subtle quantum theory has to be developed to describe such resonant light-matter in-
teractions. Hereafter, we assume that frequencies of all electric fields involved are far
away from any material resonance.
We now qualitatively examine second-order processes, starting with the second
harmonic generation (SHG). To this end, consider a monochromatic input field,
P (2) (t) = 0 χ(2) E 2 (t) = 12 0 χ(2) |E|2 + 41 (0 χ(2) E 2 e−i2ωt + c.c).
The first process describes generation of a dc field, optical rectification while the
second is second harmonic generation. It is schematically illustrated in the block-
diagram below.
ω
ω
χ ( 2)
2ω
61
more general processes taking place if two different input frequencies ω1 and ω2 are
present. The input field is then
where the summation is over all possible combinations s of two frequency components
and
PSHG (2ωj ) = 21 0 χ(2) Ej2 ,
ω1 ω1
ω3 =ω1 +ω2
χ (2)
ω2 ω2
ω1 ω1
ω3 = ω1 − ω2
ω2
χ ( 2)
ω2
The fundamental difference between the two processes can be seen from the energy-
level in Figs. 4.4 and 4.5.
In the SFG process two input photons at frequencies ω1 and ω2 annihilate giving
rise to one photon at the sum frequency, ω3 = ω1 + ω2 . In the DFG process, however,
annihilation of a pump photon at frequency ω1 and generation of a difference frequency
photon ω3 = ω1 −ω2 –sometimes referred to as signal–go hand in hand with generation
62
Figure 4.4: Energy-level description of sum-frequency generation.
(a) (b )
ω2 ω3
ω1 ω1
ω3 ω2
63
Figure 4.6: Illustrating stimulated Raman scattering; the subscripts “s” and “as” stand
for the Stokes and anti-Stokes modes, respectively.
PT HG (3ω) = 21 0 χ(3) E 3 ,
The THG process is a third-order analog of the THG process; the THG block diagram
is as follows The SF process is so called because the input field modifies the refractive
ω ω
ω 3ω
ω χ ( 3)
ω
64
index of the medium to
n = n0 + n2 |E|2 ,
leading to self-lensing of a light beam. The self-induced “medium lens” is a posi-
tive one if n2 > 0 and a negative one otherwise. Thus, either self-focusing or self-
defocusing ensues. Another third-order process that, in general, accompanies SF is
two-photon absorption (TPA). In the TPA process, two photons can be absorbed from
a light wave by a medium atom, promoting the latter to an excited state which cannot be
related to the ground state by a dipole transition. The situation is illustrated in Fig. 4.8.
where we have assumed that the polarization is invariant with respect to translations in
space and shifts in time, thanks to stationarity and homogeneity of the medium. By the
same token, the second-order nonlinear polarization can be represented as
Z Z Z ∞ Z ∞
P(2) (r, t) = 0 dr1 dr2 dt1 dt2
−∞ −∞
.
×χ(2) (r − r1 , r − r2 ; t − t1 , t − t2 )..E(r1 , t1 )E(r2 , t2 ), (4.7)
65
The expressions for higher-order nonlinear polarization fields can be expressed in a
similar fashion.
To proceed further, we will assume the medium response to be spatially local; this
is a reasonably good approximation for a vast majority of optical media which we will
rely on hereafter. In these conditions, the susceptibility tensors can be simplified to
(1)
χ(1) (r − r0 , t − t0 ) = δ(r − r0 )χt (t − t0 ), (4.8)
and
(2)
χ(2) (r − r1 , r − r2 ; t − t1 , t − t2 ) = δ(r − r1 )δ(r − r2 )χt (t − t1 , t − t2 ). (4.9)
The corresponding contributions to the polarization field are greatly simplified as well:
Z ∞
.
P(1) (r, t) = 0 dt0 χ(1) (t − t0 )..E(r, t0 ), (4.10)
−∞
Z ∞ Z ∞ .
P(2) (r, t) = 0 dt1 dt2 χ(2) (t − t1 , t − t2 )..E(r, t1 )E(r, t2 ). (4.11)
−∞ −∞
In equations (4.10) and (4.11) we dropped, for brevity, the subscript “t” for the temporal
parts of the linear and nonlinear susceptibilities.
The manifest translational invariance of susceptibilities prompts the introduction of
Fourier transforms Z ∞
χ̃(1) (ω) = dtχ(1) (t)eiωt , (4.12)
−∞
and
2 Z
Y ∞ P2
(2)
χ̃ (ω1 , ω2 ) = dts χ(2) (t1 , t2 )ei s=1 ωs ts
. (4.13)
s=1 −∞
and
XZ ∞
(2) dω1 (2)
P̃i (r, ω3 ) = 0 χ̃ (−ω3 , ω1 , ω2 )Ẽj (r, ω1 )Ẽk (r, ω2 ), (4.16)
−∞ 2π ijk
jk
where ω3 = ω1 + ω2 .
Exercise. 4.1 Derive Eq. (4.16).
66
Generalizing Eq. (4.16) to any order n, we can write down
n−1
YZ ∞
(n)
X dωs (n)
P̃jn (r, ωn ) = 0 χ̃ (−ωn , ω1 , ω2 , . . . ωn−1 )
j1 ...jn−1 s=1 −∞ 2π jn j1 j2 ...jn−1
(2) (2)
Example: χ̃ijk (−ω, ω1 , ω2 ) = χ̃ikj (−ω, ω2 , ω1 ).
The reality of χ(n) in time domain implies the following relation in the Fourier domain
(n)∗ (n)
χ̃jj1 ...jn (−ω, ω1 , . . . , ωn ) = χ̃jj1 ...jn (ω, −ω1 , . . . , −ωn ), (4.20)
• Causality:
67
For the response of a physical medium to be causal, the polarization field must be equal
to zero at any instant before the electric field is applied, which implies, in accord with
Eq.(4.17) that
(n)
χjj1 ...jn (t − τ1 , . . . t − τn ) = 0, for any τs > t. (4.21)
Let us now exhibit very tangible constraints on the functional form of the real and
imaginary parts of the susceptibility functions in the Fourier domain, stemming from
causality.
We begin by considering the linear susceptibility. It follows from Eq. (4.21) that a
causal linear response function must obey
and Z ∞
θ̃(ω) = dτ θ(τ )eiωτ , (4.25)
−∞
where P stands for a principal value, excluding the singularity in the denominator. It
follows from Eqs. (4.26) and (4.27), after simple algebra, that
∞
χ̃(1) (ω 0 )
Z
1
χ̃(1) (ω) = P dω 0 . (4.28)
πi −∞ ω − ω0
Eq. (4.28) implies that real and imaginary parts of the linear susceptibility tensor are
related vie the following Kramers-Kronig relations
∞
Im χ̃(1) (ω 0 )
Z
1
Re χ̃(1) (ω) = P dω 0 , (4.29)
π −∞ ω − ω0
68
and
∞
Re χ̃(1) (ω 0 )
Z
1
Im χ̃ (1)
(ω) = − P dω 0 . (4.30)
π −∞ ω − ω0
Relations (4.29) and (4.30) not only impose a constraint on the functional form of the
real and imaginary parts of the linear susceptibility tensor, but they also enable one to
reconstruct the real part – describing dispersion – from the imaginary one, which is
much easier to measure as it relates to absorption in the medium.
Kramers-Kronig relations can also be derived for some second-order susceptibili-
ties. In particular, starting from the causality condition
69
where the summation over the dummy indices is implied as usual. For instance,
(1) (1)
X
χij = Tik Tjl χkl , (4.36)
kl
or
(2) (2)
X
χijk = Tis Tjl Tkm χslm , (4.37)
lsm
and so on.
Exercise 4.5. A rotation with respect to the z-axis can be described by the matrix
cos θ − sin θ 0
Tij = sin θ cos θ 0
0 0 1
Assume the medium is invariant with respect to rotations by θ = π/2. Determine the
constraints on the components of χ(1) imposed in this case.
One of the most important orthogonal transformations is inversion such that for an
every point in the medium r → −r implying Tij = −δij . It follows at once from
Eq. (4.35) that if the medium is symmetric with respect to inversions – i.e., if it has an
inversion center – then for any susceptibility tensor of odd rank , or for an even n = 2k,
we obtain
(2k) (2k)
χii1 ...i2k = −χii1 ...i2k = 0. (4.38)
In particular, in media with the inversion centers the lowest-order nonlinear response
(3)
is cubic, described by χijkl . Such inversion symmetric media are referred to as cen-
trosymmetric. Most gases and liquids as well as many solids possess such properties.
Another important constraint is imposed by requiring that media be lossless. In
lossless media, equations of motions are symmetric with respect to time reversal –
there are no losses and the microscopic evolution can in principle be reversed. Under
such conditions,
χ(n) (τ1 . . . τn ) = χ(n) (−τ1 . . . − τn ). (4.39)
It can then be readily inferred from Eq. (4.14) that
(n) (n)∗
χ̃jj1 ...jn (−ω, ω1 , . . . ωn ) = χjj1 ...jn (−ω, ω1 , . . . ωn ), (4.40)
70
(2) (2) (2)
Example: χ̃ijk (−ω3 , ω1 , ω2 ) = χ̃jki (−ω3 , ω1 , ω2 ) = χ̃kij (−ω3 , ω1 , ω2 ).
We stress though that Kleinman’s symmetry is only an approximation valid far from
any internal resonances where dispersive properties of nonlinear media are negligible
such that one can virtually neglect frequency dependence of the nonlinear susceptibili-
ties. The Kleinman symmetry breaks down, for instance, if there is an absorption band
sandwiched between a pair of frequencies involved with a nonlinear interaction. In the
latter case, dispersive properties of the medium would be important at those frequencies
near the absorption band.
∇ · D = 0; ∇ · B = 0, (4.41)
∇ × H = ∂t D, ∇ × E = −∂t B. (4.42)
In the optical frequency range, natural materials are nonmagnetic, allowing us close
the set of Eqs. (4.41) and (4.42) with the constitutive relations
B = µ0 H; D = 0 E + P = D L + P N L , (4.43)
where we found it convenient to decompose the polarization field P into linear and
nonlinear components as
P = PL + PN L . (4.44)
DL = 0 E+PL in Eq. (4.43) refers to the linear electric flux density. Using Eq. (4.43)
in Eqs. (4.41) and (4.42) and eliminating H from Maxwell’s equations in favor of E,
we arrive at the set of continuity and wave equations
∇ · (DL + PN L ) = 0, (4.45)
and
2 2
∇ × (∇ × E) = −µ0 ∂tt E − µ0 ∂tt P. (4.46)
We now assume a plane-wave geometry, that is all fields are harmonic and they de-
pend only on one spatial coordinate z, say, along the wave propagation direction, which
incidentally coincides with the optical axis of the system. Under these assumption, the
relevant fields can be expressed as
71
and
PN L (z, t) = P̃N L (z, ωs )e−iωs t . (4.49)
Here ωs is the frequency of the wave we arbitrarily assign as a signal; hence the sub-
script “s”. Henceforth, it will prove convenient to break all fields into longitudinal and
transverse components as
and likewise,
k
D̃L (z, ωs ) = ez D̃L (z, ωs ) + D̃⊥ (z, ωs ), (4.51)
and
k
P̃N L (z, ωs ) = ez P̃N L (z, ωs ) + P̃⊥
N L (z, ωs ). (4.52)
It can be shown–see the exercise–that under the circumstances, ∇ × ∇ × Ẽ(z, ωs ) =
−∂zz2
Ẽ⊥ (z, ωs ).
Exercise 4.8. By expressing the field in the cylindrical coordinates, Ẽ(z, ωs ) = ez Ẽ k (z, ωs )+
2
eρ Ẽρ (z, ωs ) + eφ Ẽφ (z, ωs ), show that ∇ × ∇ × Ẽ(z, ωs ) = −∂zz Ẽ⊥ (z, ωs ).
It then follows that Eqs. (4.45) and (4.46) can be cast into the form
k k
D̃L (z, ωs ) + P̃N L (z, ωs ) = 0, (4.53)
and
2
−∂zz Ẽ⊥ (z, ωs ) = µ0 ωs2 [D̃⊥ ⊥
L (z, ωs ) + P̃N L (z, ωs )]. (4.54)
Let us now assume that a generally anisotropic medium–anisotropy is needed for phase-
matching in some cases–is uniaxial with the optical axis coinciding with the z-axis.
The dielectric tensor of such a medium is known from Sec. 2.2.2. Using the results of
this section, it is easy to see that
k
DL = k (ωs )Ẽ k , D⊥ ⊥
L = ⊥ (ωs )Ẽ ; (4.55)
implying that
k
k (ωs )Ẽ k (z, ωs ) + P̃N L (z, ωs ) = 0, (4.56)
and
ωs2
2
−∂zz Ẽ⊥ (z, ωs ) + ⊥
c2 ⊥ (ωs )Ẽ (z, ωs ) + µ0 ωs2 P̃⊥
N L (z, ωs ) = 0. (4.57)
It can be inferred from Eqs. (4.56) and (4.57) that while the longitudinal field compo-
nent can be determined from a simple algebraic equation, following from Gauss’s law,
the transverse field component is governed by a wave equation. We will now focus on
the transverse fields.
Hereafter, we will restrict ourselves to the case of linearly polarized waves in the
plane transverse to the optical axis. As optical nonlinearities far from internal res-
onances of any natural media are fairly weak, it is reasonable to assume the fields
profiles change very slowly – at the wavelength scale – in the plane, transverse to the
72
propagation direction. Hence, the electric field of a signal wave can be expressed as a
slowly-varying envelope times a fast carrier plane wave,
Ẽ⊥ (z, ωs ) = e(ωs )E(z, ωs )eiks z , (4.58)
which induces the polarization field such that
P̃⊥
N L (z, ωs ) = e(ωs )PN L (z, ωs )e
iks z
. (4.59)
Here, the wave number satisfies the usual linear dispersion relation,
ω2
k 2 (ωs ) = ⊥ (ωs ) c2s . (4.60)
Substituting from Eqs. (4.58) – (4.60), and using the slowly-varying envelope approxi-
mation (SVEA),
2
∂z E ks E; ∂zz E ks2 E, (4.61)
we arrive at the nonlinear wave equation for the signal wave in the form
2iks ∂z E = −µ0 ωs2 PN L . (4.62)
Our treatment has been general so far. We will now specialize to the second-order
processes. Recall that
(2)
X (2)
P̃i (z, ωs ) = 0 c(2) (ω1 , ω2 ) χ̃ijk (−ωs ; ω1 , ω2 )Ẽj (z, ω1 )Ẽk (z, ω2 ), (4.63)
jk
73
4.5 Second-harmonic generation
4.5.1 Coupled wave equations and phase matching considerations
The process of second harmonic generation involves the interaction of two waves at
frequency ω to produce a wave with the frequency 2ω. It is schematically illustrated in
Fig. 1 below.
ω
ω
χ ( 2)
2ω
The coupled wave equations governing the second harmonic generation (SHG) in
lossless media can be obtained directly from the general coupled-mode equations de-
rived in the previous Lecture by specializing to the case of two identical mixing fre-
quencies. The resulting wave equations for the fundamental Eω and the second har-
monic E2ω fields are
iω 2 (2) ∗ −i∆kz
∂z Eω = 2kω c2 χef f (−ω, 2ω, −ω)E2ω Eω e . (4.69)
and
i4ω 2 (2) 2
∂z E2ω = 2k2ω c2 χef f (−2ω, ω, ω) Eω ei∆kz , (4.70)
where the wave number mismatch is now defined as
ωn(ω) 2ωn(2ω)
kω = , k2ω = . (4.72)
c c
It follows from general properties of susceptibilities in the absence of losses that
(2) (2) (2)
χef f (−ω, 2ω, −ω) = 2χef f (−2ω, ω, ω) ≡ χef f . (4.73)
Using (4.73), we can transform the SHG coupled wave equations in the plane wave
geometry to
dEω iω 2 (2)
= χ E2ω Eω∗ e−i∆kz . (4.74)
dz 2kω c2 ef f
and
dE2ω iω 2 (2) 2 i∆kz
= χ E e . (4.75)
dz k2ω c2 ef f ω
74
Let us now study the second harmonic generation in the undepleted pump approx-
imation, which implies that the power of the fundamental wave is high enough and
the efficiency of the second harmonic generation is low enough that we can neglect
the power depletion of the fundamental wave. As the efficiency ηSHG of the second
harmonic generation can be defined as the ratio of the second harmonic intensity at the
output to the input intensity of the fundamental,
I2ω (L)
ηSHG ≡ , (4.76)
Iω (0)
we can define a quantitative criterion for the undepleted pump approximation to hold:
ηSHG 1. (4.77)
In the undepleted pump approximation, Eq. (4.75) can be integrated at once with
the result
(2)
iω 2 (2) 2 ei∆kL − 1 ω 2 Lχef f 2 i∆kL/2 ei∆kL/2 − e−i∆kL/2
E2ω (L) = χ E = E e ,
k2ω c2 ef f ω i∆k k2ω c2 ω 2i(∆kL/2)
(4.78)
where L is the length of the interaction region and Eω = const. Further, equation (4.78)
can be simplified as
(2)
ω 2 Lχef f Eω2 sin(∆kL/2)
E2ω (L) = ei∆kL/2 . (4.79)
k2ω c2 ∆kL/2
It can be readily inferred from Eq. (4.79) that the intensity of the second harmonic is
given by
(2)2
ω 2 L2 χef f Iω2
2 ∆kL
I2ω (L) = sinc , (4.80)
20 n2ω n2ω c3 2
where we defined
sin x
sinc(x) ≡ . (4.81)
x
The analysis of Eq. (4.80) reveals that if the phases of the fundamental and second
harmonic waves are matched, the intensity of the second harmonic is proportional to
the square of the interaction length, I2ω (L) ∝ L2 . Physically, it can be interpreted
by observing that if all N polarized atomic dipoles in the interaction volume – whose
total number is proportional to L – radiate in phase, their resulting fields interfere
constructively; consequently the total intensity of the second harmonic is such that
I2ω (L) ∝ N 2 ∝ L2 . On the other hand, if the phase matching condition (4.82) is
not met, the efficiency of the second harmonic generation decreases dramatically, as is
shown in Fig. 2.
Let us now discuss the efficiency of the SHG process. It follows from Eqs. (4.77)
and (4.80) that under the best possible condition of the perfect phase matching
∆k = 0, (4.82)
75
Figure 4.10: Second harmonic output as a function of the interaction length in the
undepleted pump approximation
Ld L, (4.85)
for the plane wave approximation to hold. Using typical values, for moderate-to-high
power lasers P ∼ 1 W, and χ2ef f ∼ 5 × 10−23 m2 /V2 , for LiNbO3 , say; with the other
parameters being chosen as follows: L ∼ 1 cm, nω ∼ n2ω ∼ 2, λ ∼ 5 × 10−5 cm, and
the spot size w0 ∼ 100 µm, such that Ld ∼ 10 cm, we obtain the order-of-magnitude
estimate as ηSHG ∼ 10−3 1. Clearly, the undepleted pump approximation is a
good one even for relatively high power laser sources in the plane wave geometry. To
increase the SHG conversion efficiency, it is advised that (a) pulsed lasers be employed
to augment the input power and (b) source light beam be tightly focused into the inter-
action volume to significantly increase the intensity of the fundamental input wave. In
general, the analysis of the SHG with such tightly focused laser beams requires a more
careful consideration of diffraction effects. With this in mind, however, we could still
make a rough order-of-magnitude estimate of the efficiency using Eq. (4.83) by taking
76
the spot size of a focused beam to be w0 ∼ 10 µm, even though Ld L. The resulting
efficiency is of the order of 10%, which is already quite an improvement.
Due to the importance of phase matching, we briefly discuss the ways of realizing
the condition (4.82), which, when translated in terms of the refractive indices, implies
First, we note that the requirement (4.86) cannot be satisfied in an isotropic medium
because of frequency dispersion: Typically, the refractive index of a nonlinear medium
far below absorption resonances is a monotonically increasing function of frequency, a
phenomenon referred to as normal dispersion. Thus isotropic media are in general not
phase matchable.
Phase matching can be realized in anisotropic media, which is referred to as bire-
fringence phase matching. As we previously mentioned, the distribution of the ordi-
nary wave vectors is spherically symmetric–which is graphically illustrated in Fig. 3–
where we assumed, for simplicity, the wave vector lies in the xz-plane–and one can
introduce the corresponding frequency-dependent refractive index no (ω) by the ex-
pression
ko c p
no (ω) ≡ = ⊥ (ω). (4.87)
ω
The extraordinary wave vector, on the other hand, does depend on the propagation
direction, and the associated extraordinary refractive index is given by
−1/2
sin2 θ cos2 θ
ke c
ne (θ, ω) ≡ = + . (4.88)
ω ⊥ (ω) k (ω)
The surface ne (θ, ω) = const is, in general, an ellipsoid, but it reduces to an ellipse if
we restrict the extraordinary wave vector to lie in the xz− plane, see Fig. 3.
kx kx
k0 ke
kz kz
Figure 4.11: Graphical representation of the wave vectors of ordinary (left) and ex-
traordinary (right) waves in a uniaxial crystal.
Assume now that the fundamental is an ordinary wave and the second harmonic is
an extraordinary one. It can then be inferred from Fig. 4 that provided the extraordinary
77
ne (2ω)
n0 (ω)
θ*
OPTICAL AXIS
Figure 4.12: Illustrating phase matching for the SHG in uniaxial crystals.
refractive index for the SH along the crystal axis is smaller than the ordinary refractive
index of the fundamental, which can be mathematically expressed by the inequality
the phase matching is possible at the angle θ∗ which can be determined from Eqs. (4.86),
(4.87) and (4.88) to be v
u 1
u k (2ω) − ⊥1(ω)
tan θ∗ = t 1 1 . (4.90)
⊥ (ω) − ⊥ (2ω)
Unfortunately, whenever the angle between the ordinary and extraordinary wave
vectors is other than 90deg , a spatial walkoff accrues on propagation of the two waves
as a consequence of directional mismatch between the Poynting vector and propagation
direction of an extraordinary wave. The walkoff reduces spatial overlap between the
polarization modes, thereby drastically reducing the SHG efficiency. Fortunately, some
nonlinear crystals, such as lithium niobate, have a pronounced dependence of their
birefringence on the temperature. Thus, one can achieve phase matching by keeping
the angle between the modes fixed at 90deg and varying the temperature of the crystal.
This is called temperature phase matching.
In the situations when neither birefringence nor temperature phase matching is pos-
sible, the most powerful phase matching technique is used, the so-called quasi-phase-
matching. The technique involves periodically polling χ(2) samples to modulate the
second-order susceptibility. The latter can then be expanded in a Fourier series
∞
X
χ(2) (z) = χ(2)
m e
i2πmz/Λ
,
m=−∞
where Λ is a spatial period of the structure. The phase mismatch is then modified to
(2)
∆kef f = ∆k − 2πm/Λ. As χm decreases with m, reducing the SH intensity, it is
preferable to work with m = 1 harmonic and choose the period Λ to phase match the
78
interaction, i.e.,
Λ = 2π/∆k.
If ∆k is so large, ∆k ∼ k that it is impossible to attain perfect phase matching, quasi-
phase-matching allows to extend, at least, the effective interaction length to
Lef f = L(1 + 2π/Λ∆k),
where the smallest available Λ should be used.
79
Similarly, the equations for the phases take the form
dφω A2ω
= cos θ, (4.100)
dz l
dφ2ω A2ω
= cos θ. (4.101)
dz lA2ω
Introducing ζ = z/l, we can cast our equations into the following dimensionless form
dAω
= Aω A2ω sin θ, (4.102)
dζ
dA2ω
= −A2ω sin θ, (4.103)
dζ
dφω
= A2ω cos θ, (4.104)
dζ
dφ2ω A2
= ω cos θ. (4.105)
dζ A2ω
It can be inferred at once from Eqs. (4.104) and (4.105) as well as from Eq. (4.98) that
θ obeys the equation
A2
dθ
= ∆s + 2A2ω − ω cos θ, (4.106)
dζ A2ω
∆s = ∆kl. (4.107)
We can easily see from Eqs. (4.102) and (4.103) that the set possesses the integral
of motion
A2ω + A22ω = 1, (4.108)
which implies the power conservation in the SHG process in a lossless medium. It then
follows from Eqs. (4.102) and (4.103) that
1 d
A2ω = ln Aω , (4.109)
sin θ dζ
and
A2ω 1 d
=− ln A2ω . (4.110)
A2ω sin θ dζ
Substituting from the last two equations into Eq. (4.98), we obtain the equation for the
phase difference in the form
dθ d
= ∆s + cot θ ln(A2ω A2ω ). (4.111)
dζ dζ
80
Hereafter we focus on the perfect phase matching situation, ∆s = 0. In this case,
we can transform Eq. (4.111), with the aid of Eq. (4.103) to
d ln cos θ d
= − ln(A2ω A2ω ), (4.112)
dζ dζ
which can be integrated at once yielding the second integral of motion as
A2ω A2ω cos θ = Γ. (4.113)
Normalized Intensity
ߞ = ݖൗ݈
Figure 4.13: Intensity of the fundamental and second harmonic as functions of the
interaction distance in the case of perfect phase matching.
Suppose now that Γ = 0 implying a fixed phase difference between the FW and
SH, θ = −π/2. It then follows that the equations of motion for the mode amplitudes
simplify to
dAω
= Aω A2ω , (4.114)
dζ
dA2ω
= −A2ω , (4.115)
dζ
Using Eq. (4.108), we can eliminate the fundamental from Eq. (4.115) resulting in
dA2ω
= −(1 − A22ω ), (4.116)
dζ
which can be integrated at once yielding
A2ω = tanh ζ; Aω = sechζ. (4.117)
The intensities of the fundamental and second harmonic are displayed in the Fig. 5.
81
4.6 Sum-frequency generation
4.6.1 Coupled wave equations and their solution in the undepleted
pump approximation
In this Lecture, we examine the sum-frequency generation (SFG), which involves mix-
ing a signal wave at frequency ω1 with a pump wave at frequency ω2 to yield a har-
monic oscillating at ω3 = ω1 + ω2 , to be referred to as the sum-frequency (SF) wave.
The SFG process is schematically illustrated in Fig. 1.
ω1 ω1
ω3 =ω1 +ω2
χ (2)
ω2 ω2
The wave equations governing the SFG can be readily obtained from the general
coupled wave equations, yielding the following set
iω12 (2) ∗ −i∆kz
∂z E1 = 2k1 c2 χef f (−ω1 ; ω3 , −ω2 )E3 E2 e . (4.118)
∆k = k1 + k2 − k3 . (4.121)
82
dE3 iω32 (2)
= χ E1 E2 ei∆kz . (4.126)
dz 2k3 c2 ef f
Although Eqs. (4.124) – (4.126) can be solved in general, the solution is very compli-
cated and not too instructive. Instead, we will study the SFG process in the so-called
undepleted pump approximation, i.e, when the amplitude of the pump wave E2 is so
much larger than those of the other waves that we can neglect the pump depletion –
that is we will assume E2 = const – which enables us to rewrite Eqs. (4.124) – (4.126)
as
dE1
= κ1 E3 e−i∆kz , (4.127)
dz
and
dE3
= κ3 E1 ei∆kz . (4.128)
dz
Here we introduced the notations
(2)∗ (2)
iω12 χef f iω32 χef f
κ1 = E2∗ , κ3 = E2 . (4.129)
2k1 c2 2k3 c2
Let us then assume perfect phase matching, ∆k = 0. In this case, we can eliminate
one of the fields from Eqs. (4.127) and (4.128) in favor of the other, reducing the set to
a second-order ODE; for instance,
d2 E1
+ κ2ef f E1 = 0, (4.130)
dz 2
with
(2)
ω12 ω32 |χef f |2 |E2 |2
κ2ef f = −κ1 κ3 = . (4.131)
4k1 k3 c4
A general solution to (4.130) is
where C1 and C2 are arbitrary constants. It then follows from Eqs. (4.127), (4.128)
and (4.132) that
κef f C1 κef f C2
E3 = − sin κef f z + cos κef f z. (4.133)
κ1 κ1
Specifying the initial conditions, E1 (z = 0) = E1 (0) and E3 (z = 0) = 0 – there is
no SF at the entrance to the medium – we obtain the expressions for the signal and the
SF waves as
E1 = E1 (0) cos κef f z, (4.134)
and
κef f
E3 = −E1 (0) sin κef f z, (4.135)
κ1
In physical terms, the SFG in the undepleted pump approximation describes periodic
power exchange between the signal and the SF waves. The periodic character of the
83
power exchange between the signal and the SF can be explained by observing that to
create an SF photon, a signal photon has to be annihilated, ω3 = ω1 + ω2 , such that the
more the power residing with the SF, the less the power of the signal and vice versa.
Exercise 4.10. Solve Eqs. (4.127) and (4.128) for ∆k 6= 0 in the case when initially
all power resides with ω1 harmonic. Determine the SF intensity and show that its
maximum reduces precipitously as ∆k increases. Comment on the importance of phase
matching for efficient SFG. Hint: look for solutions in the form
The wave equations in the plane wave geometry, (4.124) – (4.126), can then be cast
into the form
dE1 iω12 (2)
= χ E3 E2∗ e−i∆kz , (4.140)
dz 2k1 c2 ef f
dE2 iω22 (2)
= χ E3 E1∗ e−i∆kz , (4.141)
dz 2k2 c2 ef f
and
dE3 iω32 (2)
= χ E1 E2 ei∆kz . (4.142)
dz 2k3 c2 ef f
Let us now study relations among the energy fluxes associated with the mixing waves.
To this end, we derive the following equations for the wave intensities
d|E1 |2 ω1 (2)
= χ Im(E1∗ E2∗ E3 ei∆kz ), (4.143)
dz n1 c ef f
d|E2 |2 ω2 (2)
= χ Im(E1∗ E2∗ E3 ei∆kz ), (4.144)
dz n2 c ef f
d|E3 |2 ω3 (2)
=− χ Im(E1∗ E2∗ E3 ei∆kz ), (4.145)
dz n3 c2 ef f
where we introduced kj = nj ωj /c.
84
Further, we introduce the optical intensities of the signal, pump and the SF waves
as
0 n j c
Ij = |Ej |2 , (4.146)
2
with j = 1, 2, 3. It can then be inferred from Eqs. (4.143) – (4.146) that
dI1 0 ω1 (2)
= χ Im(E1 E2 E3∗ ei∆kz ), (4.147)
dz 2 ef f
and
dI2 0 ω2 (2)
= χ Im(E1 E2 E3∗ ei∆kz ), (4.148)
dz 2 ef f
as well as
dI3 0 ω3 (2)
=− χ Im(E1 E2 E3∗ ei∆kz ). (4.149)
dz 2 ef f
It follows at once by adding Eqs. (4.147), (4.148) and (4.149) that
3
X
Ij = const, (4.150)
j=1
which is tantamount to energy conservation for the SFG in lossless media. We can also
infer from Eqs. (4.147) – (4.149) that
d I1 I2
− = 0, (4.151)
dz ω1 ω2
d I1 I3
+ = 0, (4.152)
dz ω1 ω3
and
d I2 I3
+ = 0. (4.153)
dz ω2 ω3
The preceding differential laws are equivalent to the three new invariants for the SFG
process, which are known as the Manley-Rowe relations; the latter take the form
I1 I2
− = M1 = const, (4.154)
ω1 ω2
I1 I3
+ = M2 = const, (4.155)
ω1 ω3
I2 I3
+ = M3 = const. (4.156)
ω2 ω3
The physical interpretation of Eqs. (4.154) – (4.156) can be best furnished using the
photon picture. To this end, one can introduce the photon number fluxes – the number
of photons at frequency ωj created or annihilated per second – by the expression, Nj =
Ij /h̄ωj . It then follows from Eq. (4.154) – (4.156) that the numbers of signal and idler
photons generated per unit time in any SFG process must be separately equal to the
number of pump photons destroyed per unit time. Summarizing, we can say that to
generate one SF photon, a signal and a pump photon must be destroyed. The qualitative
photon picture of the SFG is exhibited in the form of a simple three-photon diagram in
Fig. 2.
85
Figure 4.15: Illustrating Manley-Rowe relations with a photon diagram.
ω1 ω1
ω3 = ω1 − ω2
ω2
χ ( 2)
ω2
The paraxial wave equations governing DFG can be shown to take the form
iω12 (2) i∆kz
∂z E1 = 2k1 c2 χef f (−ω1 ; ω2 , ω3 )E2 E3 e , (4.157)
and
iω22 (2) ∗ −i∆kz
∂z E2 = 2k2 c2 χef f (−ω2 ; ω1 , −ω3 )E1 E3 e , (4.158)
as well as
iω32 (2) ∗ −i∆kz
∂z E3 = 2k3 c2 χef f (−ω3 ; ω1 , −ω2 ) E1 E2 e , (4.159)
where the wave number mismatch is now defined as
∆k = k1 − k2 − k3 , (4.160)
86
and the signal frequency is given by
ω3 = ω1 − ω2 . (4.161)
It follows from Eqs. (4.157) – (4.159) and (4.162) that in the plane wave geometry, one
can obtain the following set of DFG wave equations
iωj2 (2)∗
ζj = χ E1 , j = 2, 3. (4.168)
2kj c2 ef f
Stipulating that initially all power reside with the idler, E3 (z = 0) = E3 (0) and E2 (z =
0) = 0, yields the solution
87
(a) (b )
ω2 ω3
ω1 ω1
ω3 ω2
and
ζef f E3∗ (0)
E2 (z) = sinh ζef f z. (4.173)
ζ3∗
It can be easily inferred from Eq. (4.172) and (4.173) that both the signal and the
idler monotonically grow with the distance z. Such a behavior – which is in sharp
contrast with the SFG – is graphically presented in the diagram in Fig. 2. To explain
the diagram, it is sufficient to notice that in the DFG process, the signal and idler
photons are created and annihilated in pairs, ω1 = ω3 + ω2 . In other words, the greater
the power of one wave – be it the signal or the idler – the greater the power of the other.
The two possibilities are illustrated in Figs. 2(a) and 2(b), respectively.
We can show that the monotonic character of the signal and idler wave growth
depends on the pump power level in case of finite mismatch ∆k 6= 0. To this end, we
transform Eqs. (4.166) and (4.167) to
dE2∗
= ζ2∗ E3 e−i∆kz . (4.174)
dz
dE3
= ζ3 E2∗ e−i∆kz . (4.175)
dz
Introducing the new variables viz.,
E2∗ = E 2 ei∆kz/2 , E3 = E 3 e−i∆kz/2 , (4.176)
we arrive at the equations
0
E 2 + 21 i∆kE 2 = ζ2∗ E 3 , (4.177)
and
0
E 3 − 12 i∆kE 2 = ζ3 E 2 . (4.178)
Seeking solutions to Eqs. (4.177) and (4.178) in the form
(0) (0)
E 2 (z) = E 2 eΩef f z ; E 3 (z) = E 3 eΩef f z , (4.179)
we obtain from the determinant condition, the expression for Ωef f :
q
2 2
Ωef f = ζef f − (∆k/2) . (4.180)
88
The latter implies that in the presence of phase mismatch, there exists a power threshold
for simultaneous amplification of the signal and idler modes,
2
20 n1 n2 n3 c3
∆k
Ith = (2)
. (4.181)
ω2 ω3 |χef f |2 2
Thus, for a given phase mismatch, the pump intensity must be greater than a certain
critical value, I1 ≥ Ith , for parametric amplification to take place.
Next, general solutions for the idler and signal modes can be expressed above
threshold as
E 2 (z) = E2∗ (0) cosh Ωef f z + A sinh Ωef f z, (4.182)
and
E 3 (z) = E3 (0) cosh Ωef f z + B sinh Ωef f z. (4.183)
Substituting from Eqs. (4.182) and (4.183) into Eqs. (4.177) and (4.178), we determine
the coefficients A and B:
ζ2∗ E3 (0) − 12 ∆kE2∗ (0)
A= , (4.184)
Ωef f
and
ζ3 E2∗ (0) + 12 ∆kE3 (0)
B= , (4.185)
Ωef f
Finally, on substituting from Eqs. (4.184) and (4.185) into (4.182) and (4.183) we ob-
tain, upon a slight rearrangement, the signal and idler fields in the form
ζ3 E2∗ (0)
i∆k
E3 (z) = E3 (0) cosh Ωef f z + sinh Ωef f z + sinh Ωef f z e−i∆kz/2 ,
2Ωef f Ωef f
(4.186)
and
ζ2 E3∗ (0)
i∆k
E2 (z) = E2 (0) cosh Ωef f z + sinh Ωef f z + sinh Ωef f z e−i∆kz/2 .
2Ωef f Ωef f
(4.187)
Exercise 4.11. Show that below threshold, the solutions can be obtained with the
substitutions,
Ωef f → iΩef f ; cosh iΩef f z → cos Ωef f z, sinh iΩef f z → i sin Ωef f z,
yielding
ζ3 E2∗ (0)
i∆k
E3 (z) = E3 (0) cos Ωef f z + sin Ωef f z + sin Ωef f z e−i∆kz/2 ,
2Ωef f Ωef f
and
ζ2 E3∗ (0)
i∆k
E2 (z) = E2 (0) cos Ωef f z + sin Ωef f z + sin Ωef f z e−i∆kz/2 .
2Ωef f Ωef f
89
How can you reconcile the periodic power exchange between the signal and idler
modes with the photon diagram of Fig. 2 demanding that signal and idler photons
be created or annihilated in pairs?
The DFG process is also known as parametric down-conversion: A high-frequency
pump photon generates a signal-idler photon pair at lower frequencies. It is the key
process to generate a pair of entangled photons from a single pump photon in χ(2)
nonlinear media; the latter finds numerous applications in quantum optics.
Exercise 4.12. Show that the Manley-Rowe relations for the DFG without the unde-
pleted pump approximation take the form
I1 I2
+ = M1 = const,
ω1 ω2
I1 I3
+ = M2 = const,
ω1 ω3
I2 I3
− = M3 = const.
ω2 ω3
and interpret your results using the photon diagram of Fig. 2.
Since a linearly –or circularly – polarized field maintains its state of polarization in an
isotropic medium, the induced polarization field is then given by the expression
where
ω2
ks2 = (ωs ) c2s . (4.190)
90
The coupled nonlinear wave equations governing the field evolution can be expressed
as
2iks ∂z Es = −µ0 ωs2 PN L . (4.191)
Here we introduced the notation
Es ≡ E(z, ωs ). (4.192)
with ωs = ω1 + ω2 + ω3 . Using Eqs. (4.193) and (4.189), we obtain for the slowly-
varying third-order polarization field the expression
(3)
X (3)
Pi (z, ωs ) = 0 c(3) (ω1 , ω2 , ω3 ) χ̃ijkl (−ωs ; ω1 , ω2 , ω3 )ei (ωs )
jkl
×ej (ω1 )ek (ω2 )el (ω3 )E(z, ω1 )E(z, ω2 )E(z, ω3 )ei∆kz ,(4.194)
we finally arrive at the paraxial wave equation governing the four-wave mixing pro-
cesses
iω 2 (3)
∂z Es = 2k(ωss)c2 χef f (−ωs ; ω1 , ω2 , ω3 )E1 E2 E3 ei∆kz . (4.197)
The family of third-order processes is very large; each particular process is specified
by a choice of four mixing frequencies. In the following, we will only consider two
commonly encountered processes: third-harmonic generation and self-focusing.
91
ω ω
ω 3ω
ω χ ( 3)
ω
and
9iω 2 (3) 3
∂z E3ω = 2k3ω c2 χef f (−3ω; ω, ω, ω) Eω ei∆kz , (4.199)
where the phase mismatch is given by
The analysis reveals that the degeneracy factors associated with the corresponding mix-
ing processes, (3ω = ω + ω + ω) and (ω = 3ω − ω − ω) are related as
Using Eq. (4.202) and assuming a plane wave geometry, we can transform the
governing coupled wave equations, Eqs. (4.198) and (4.199), into the form
dEω 3iω (3)
= χ E3ω Eω∗2 e−i∆kz . (4.203)
dz 2nω c ef f
and
dE3ω 3iω (3) 3 i∆kz
= χ E e . (4.204)
dz 2n3ω c ef f ω
The last equations are very similar to those describing second harmonic generation.
Unfortunately, though, third harmonic generation is a rather weak process. There-
fore relatively high optical intensities are required to generate THG in a crystal with a
reasonable efficiency. To estimate the THG efficiency, we consider the THG process
in the undepleted pump approximation, Eω = const. Under these conditions, equa-
tions (4.203) and (4.204) can be easily integrated to give an expression for the third
harmonic field in the form
i3ω (3) 3 i∆kL/2
E3ω (L) = χ E e sinc(∆kL/2). (4.205)
2n3ω c ef f ω
In complete analogy with the SHG theory, we introduce the THG efficiency by the
expression
I3ω (L)
ηT HG = . (4.206)
Iω (0)
92
It follows from Eqs. (4.205) and (4.206), assuming perfect phase matching that in the
undepleted pump approximation,
2 (3)2
L χef f Iω2
ηT HG = 36π 2 . (4.207)
λ n3ω n3ω 20 c2
(a) (b ) (c)
ω ω ω
ω 3ω ω 3ω ω 3ω
ω ω ω
Figure 4.19: Illustrating the third-harmonic generation in gases under resonant excita-
tion conditions. The laser is tuned to either two- or one- or else three-photon resonance
in parts (a), (b) and (c), respectively.
93
toward the beam center. Such a behavior is termed self-focusing of light in a nonlinear
medium, and the medium with a positive nonlinear refractive index forming a focusing
lens, self-focusing. As a consequence of self-focusing, the beam narrows and its peak
intensity is enhanced with the propagation distance. On the other hand, every beam
tends to spread due to diffraction which tends to decrease light intensity at the center.
The two opposing trends are characterized by different spatial scales. We can easily es-
timate such scales – referred to as nonlinear and diffraction lengths, respectively – from
elementary considerations. The characteristic diffraction length was defined before in
the studies of Gaussian beam diffraction in free space:
k∆nN L LN L ∼ 1, (4.210)
where the nonlinear change in the refractive index ∆nN L can be estimated using the
peak intensity of the beam as
∆nN L ∼ n2 I0 . (4.211)
Here I0 is the peak intensity and n2 > 0 is a nonlinear refractive index coefficient to
be discussed in greater detail below. It follows from Eqs. (4.210) and (4.211) that
1
LN L ∼ . (4.212)
kn2 I0
The beam evolution scenario entirely depends on the relative sizes of the two char-
acteristic lengths. In particular, if LD < LN L , diffraction dominates, and the beam
spreads. However, if the diffraction and nonlinearity operate at the characteristic scales
of the same order, exact balance of the two opposing trends is possible, leading to the
formation of spatial solitons, i. e., the beams whose spatial profiles and widths do not
change upon propagation in self-focusing nonlinear media. A soliton can be formed
if the optical power of the beam is exactly equal to a certain critical power such that
the nonlinearity can arrest diffraction-induced spreading. We can estimate the critical
power necessary for soliton formation by imposing the balance condition
LD ' LN L . (4.213)
It follows at once from Eqs. (4.209), (4.212) and (4.213) that the critical power, Pcr =
Icr πw02 is given by
λ20
Pcr ' , (4.214)
4πn0 n2
94
where λ0 = 2π/k0 = 2πc/ω.
Exercise 4.13. The magnitude of the nonlinear refractive index for carbon disulfide
(CS2 ) is n2 ' 3 × 10−14 cm2 /W, the linear refractive index is equal to 1.63. Estimate
the critical power for spatial soliton formation at λ0 ' 1 µm. Compare your results
with Pcr for silica glass for which n2 ' 5 × 10−16 cm2 /W, and n0 ' 1.4.
w0
θ st
zst
If the characteristic nonlinear length is smaller than the diffraction length, the non-
linearity prevails, causing self-focusing of the beam. One can estimate a characteristic
self-focusing distance in the limit LN L LD . In this case, diffraction is negligible,
and geometrical optics approach would suffice for a rough estimate. According to Fer-
mat’s principle, any ray traveling
R from the wavefront up to the focusing point must
traverse the same optical path, dsn(s) = const. As a result, we obtain for the paths
exhibited in Fig. 3,
!
2
δn q 2 δn w 0
(n0 + δn)zf = n0 + zf + w02 ' n0 zf 1 + 1 + 2 , (4.215)
2 2n0 2zf
where we have assumed that the refractive index along the central ray is n0 + δn,
whereas the peripheral ray experiences the refractive index strength of roughly n0 +
δn/2. It then follows from (4.215) after simple algebra that the self-focusing distance
is r r
n0 n0
zf ' w0 ' w0 . (4.216)
δn n2 I0
Finally, using the expressions for the beam power and the critical power as
95
power of Pcr , long before the self-focusing distance is reached. The filamentation is
caused by a transverse instability resulting from the growth of tiny imperfections of the
beam wave front.
To quantitatively describe self-focusing, nonlinear absorption, and and soliton for-
mation, we can derive the nonlinear wave equation corresponding to the self-action
process by a fundamental wave of frequency ω. Mathematically, the corresponding
susceptibility tensor is χ(3) (−ω; ω, −ω, ω). The resulting equation takes the form
i 2 iω 2 (3)
∂z Eω − 2kω ∇⊥ Eω = 2kω c2 χ (ω)|Eω |2 Eω , (4.219)
Recall that we assumed the medium to be transparent. We now lift that restriction by
allowing for linear and nonlinear losses in the medium. In physical terms, linear losses
lead to exponential decay of the field amplitude as we saw in Sec. 2. 2 with a decrement
α/2 determined by the imaginary part of the complex refractive index. Mathematically,
linear losses can then be easily incorporated into Eq. (4.222) introducing the transfor-
mation
Eω = E˜ω e−α(ω)z/2 , (4.223)
implying that
iω 2
∂z Ẽω = 2kω c2 χ
(3)
(ω)e−α(ω)z |Ẽω |2 Ẽω . (4.224)
Nonlinear losses are accounted for by assuming a complex χ(3) such that
(3)
χ(3) (ω) = χ(3)
r (ω) + iχi (ω). (4.225)
Next, introducing the amplitude and phase of Eω viz., Eω = |Eω |eiΦω , and separating
real and imaginary parts in Eq. (4.224), we arrive at
3k0 χ(3) 2 −αz
∂z Φ = 8n0 |Ẽ| e , (4.226)
r
96
and (3)
3k0 χi
∂z |Ẽ| = − 8n0 |Ẽ|3 e−αz . (4.227)
Here we dropped frequency subscripts, for notational simplicity, introduced k0 = ω/c,
kω ' k0 n0 , where n0 is the real part of the linear refractive index, and assumed linear
losses to be sufficiently weak, α(ω) kω which is a reasonably good assumption at
optical frequencies far from any internal resonances of the medium. It is customary to
introduce the optical intensity,
0 n 0 c 2
I˜ = |Ẽ| . (4.228)
2
It then follows that
˜ −αz ,
∂z Φ = k0 n2 Ie (4.229)
and
∂z I˜ = −β2 I˜2 e−αz , (4.230)
where we introduced the nonlinear refractive index n2 and the two-photon absorption
(TPA) coefficient β2 by the expressions
(3)
3χr
n2 = , (4.231)
40 n20 c
and
(3)
3k0 χi
β2 = . (4.232)
20 n20 c
The two-photon coefficient is so called because the absorption rate on the r.h.s of
Eq. (4.230) is proportional to the square of intensity, implying, in the photon picture,
that two photons are absorbed in each elementary nonlinear absorption act as is de-
picted in Fig. 4.8. TPA processes play an important role in the optical excitation of
semiconductor materials whenever the energy of a photon pair is greater than a semi-
conductor energy gap as is sketched in Fig. 4.21.
Integrating Eq. (4.230) with the initial condition, I(ρ, 0) = I0 (ρ) at the source, we
obtain the beam intensity at any z = const > 0 as
I0 (ρ)e−αz
I(ρ, z) = , (4.233)
1 + β2 I0 (ρ)Leff (z)
97
Figure 4.21: Illustrating two-photon absorption in direct-gap semiconductor materials.
Photo-excitation is possible whenever 2h̄ω ≥ Eg , where Eg is the gap energy.
To gain a better qualitative understanding of beam phase behavior, let us restrict our-
selves again to the transparent case, α = β2 = 0 and assume, for simplicity, the beam
has a Gaussian intensity profile at the source,
ρ2
I0 (ρ) = I0 exp − 2 . (4.236)
2w0
In most practical situations the nonlinear refractive index is quite small. Assuming fur-
ther the medium sample thickness to be small as well–this is referred to as a thin sample
approximation–the resulting nonlinear phase shift can be evaluated by expanding the
Gaussian envelope in Eq. (4.236) in a Taylor series to the first order. On substituting
the resulting expansion into Eq. (4.235) and integrating, we obtain the nonlinear shift
k0 n2 I0 ρ2
∆Φ(ρ) ' − ∆L, (4.237)
2w02
where ∆L is a medium sample thickness. On comparing this expression with a quadratic
phase acquired by a diffracting Gaussian beam, we can conclude that the transparent
nonlinear medium works as a thin lens, imparting a quadratic phase shift on a beam
wavefront. The latter corresponds to a converging spherical wave for n2 > 0 causing
self-focusing of the beam, or a diverging spherical wave in the self-defocusing case,
n2 < 0. The self-focusing case is sketched in Fig. 4.22.
To explain the identification of n2 with a nonlinear refractive index, we shall con-
sider the polarization field. To this end, we assume, for simplicity, a transparent
medium and linear polarization of the beam – such that any polarization effects can
be ignored – and introduce the scalar polarization field Ptot ≡ Pi ei , by the expression
(1) 3 (3) 2
Ptot = 0 χ E + χ |E| E = 0 χtot E, (4.238)
4
98
Figure 4.22: Self-phase modulation in self-focusing nonlinear media resulting in the
extra beam focusing, the so-called nonlinear lens effect.
n2 = 1 + χtot , (4.240)
n = n0 + n2 |E|2 . (4.241)
It follows from Eq. (4.239) and (4.240) and the fact that the nonlinear refraction is
always a small effect as compared with the linear one that
3χ(3)
n2 = , (4.243)
8n0
which provides a relation between the third-order susceptibility and the nonlinear re-
fractive index.
The nonlinear refractive index associated with the optical intensity is introduced
viz.,
n = n0 + n2 I, (4.244)
The nonlinear refractive index n2 has the units of m2 /V2 whereas the other one, n2 , is
measured in m2 /W2 . The two indices are related as
2
n2 = n2 . (4.245)
0 n 0 c
It follows from Eqs. (4.243), (4.244) and (4.245) that n2 is given by Eq. (4.231) with
(3)
χr = χ(3) in the transparent case, and hence its identification with the nonlinear
refractive index.
The nonlinear wave equation for self-focusing in a transparent medium can be
rewritten as
1
i∂z E + 2k ∇2⊥ E + kn 2
n0 |E| E = 0.
2
(4.246)
99
Equation (4.246) is referred to as the nonlinear Schrödinger equation (NLSE) because
of its formal similarity with the Schrödinger equation in quantum mechanics. We can
now introduce dimensionless variables, Z = z/LD , U = E/E0 , R⊥ = r⊥ /w0 , E0 =
(2I0 /0 cn0 )1/2 , and transform the NLSE to the dimensionless form
i∂Z U + 12 ∇2⊥ U + N 2 |U |2 U = 0. (4.247)
Here we have introduced the only dimensionless parameter – the soliton parameter N ,
governing the dynamics of the system. It is defined as follows
LD
N2 ≡ , (4.248)
LN L
where the diffraction and nonlinear lengths, LD and LN L are given by the expressions
1
LD = kw02 , LN L = . (4.249)
kn2 I0
A numerical analysis of Eq. (4.247) confirms formation of a spatial soliton for
the beam power such that N = 1. However, the soliton turns out to be unstable with
respect to small perturbations. Stable solitons can be formed in two-spatial dimensions,
provided the saturation of nonlinear refractive index is allowed. Stable spatial solitons
can be generated in Kerr-like nonlinear media in a planar waveguide geometry where
trapping in one spatial dimension is realized by the nonlinear medium whereas the
other spatial dimension is trapped by the waveguide. The dimensionless NLSE in the
planar waveguide geometry takes the form
i∂Z U + 21 ∂XX
2
U + N 2 |U |2 U = 0. (4.250)
The lowest order soliton corresponds to the exact balance between the nonlinearity and
diffraction, N = 1, and its spatial profile is given by
U (Z, X) = sech X e−iZ/2 . (4.251)
Higher-order solitons also exist. They correspond to more intense beams, N > 1. In
such cases, the nonlinearity dominates at first, causing self-focusing of the beam. How-
ever, in (1 + 1)D geometry – indicating one transverse dimension plus one dimension
along the waveguide unaffected by the waveguide trapping – the initial self-focusing
can be slowed down and eventually reversed by increased diffraction of a more tightly
focused beam. As a result, the periodic pattern of contraction and expansion of the
soliton manifests itself, with the soliton returning to its initial shape and transverse size
every half-period. Such solitons are called optical breathers. An example of a breather
is displayed in Fig. 4(b) for N = 3.
Exercise 4.14. Show that the 1D NLSE is invariant with respect to the Galilean
transformation,
X 0 = X − vZ; Z 0 = Z,
for an arbitrary speed v. In other words, demonstrate that Eq. (4.250) has the same
form in the “primed” variables, provided the fields in the two coordinate systems are
related by a gauge transformation,
0
,X 0 )
U (Z, X) = V (Z 0 , X 0 )eif (Z .
100
Figure 4.23: Intensity of the fundamental (a) and the third-order (b) soliton as function
of the propagation distance.
Determine the phase f . Draw conclusions about the functional form of a moving soli-
ton field.
Figure 4.24: Schematic illustration of the z-scan experimental arrangement. The self-
focusing sample is placed behind beam focus.
focusing medium causes additional focusing of the beam and hence a stronger beam
divergence past the focal plane. Therefore, the fraction of the beam power passing
through a detector pinhole on the axis reduces. Thus, the on-axis beam intensity at the
detector decreases. The situation is depicted in Fig. 4.24. On the other hand, if the
sample is placed in front of the focal plane, the additional focusing due to the nonlinear
medium results in the increased power fraction captured by the on-axis detector and
hence the on-axis intensity increase as is seen in Fig. 4.25. In practice, the on-axis
intensity is always normalized to its magnitude in the absence of the sample, yielding
a transmittance. The latter is greater than one for the in-front-of-the-focus sample
position and less than one for the behind-the-focus position. The situation is reversed
for a self-defocusing nonlinear sample.
101
Figure 4.25: Schematic illustration of the z-scan experimental arrangement. The self-
focusing sample is placed in front of the beam focus.
102
where we assumed that ∆L min(zs , zR ) (thin sample). The accrued nonlinear phase
can be expressed through its value on the axis as
ρ2
∆ΦN L (ρ, ∆L) = ∆Φ0 exp − 2 . (4.260)
ws
where
∆φ = k0 n2 I0 ∆L. (4.262)
is the on-axis phase shift for the sample placed in the focal plane, z = 0.
The Gaussian beam envelope emerging from the sample can then be written as
ikρ2 ρ2
E(ρ, zs ) = Es exp exp − 2 exp[i∆ΦN L (ρ, ∆L)], (4.263)
2Rs 2ws
It follows from Eqs. (4.263) and (4.264) that the Gaussian beam envelope at the exit to
the sample is then
∞
ikρ2 (i∆Φ0 )m ρ2
X
E(ρ, zs ) = Es exp exp − 2 (1 + 2m) . (4.265)
2Rs m=0
m! 2ws
Employing the Fourier transform technique discussed in Sec. 3.1, we can determine
the beam envelope in the detector aperture plane, located a distance La away from the
sample, in the form
∞
(i∆Φ0 )m ρ2
X
E(ρ, La ) = E s exp − 2
, (4.266)
m=0
qm m! 2qm σm
where
1 1 + 2m ik
2
= 2
− , (4.267)
σm ws Rs
and
La i(1 + 2m)La
qm = 1 + + . (4.268)
Rs zR
103
Exercise. 4. 15. Derive Eqs. (4.266) through (4.268).
It follows from Eq. (4.266) that the beam envelope on the axis at the aperture location
is then
∞
X (i∆Φ0 )m
E(0, La ) = E s . (4.269)
m=0
qm m!
The actually measured quantity is the on-axis detector transmittance defined as
I(zs + La , ρ = 0, ∆φ)
T (La , ∆φ) ≡ . (4.270)
I(zs + La , ρ = 0, ∆φ = 0)
We assume that the detector aperture is placed in the far zone of the sample such
that La zR and the sample is placed sufficiently close to the focal plane of the beam,
zs zR to ensure the beam divergence is not too large and there is enough power
captured by the detector aperture to guarantee a reasonable signal-to-noise ratio. Next,
the nonlinearity is typically small, implying the accumulated phase shifts |∆φ| 1.
Therefore one can expand the r.h.s. of Eq. (4.269) in a Taylor series and keeping the
lowest nontrivial contribution in Eq. (4.270), we arrive, after minor algebra, at the
expression
4x∆φ
T (x, ∆φ) ' 1 + 2 , (4.271)
(x + 9)(x2 + 1)
where x = zs /zR . The analysis of Eq. (4.267) reveals that it has a maximum (peak)
and a minimum (valley) as is sketched in Fig. 4.26. The difference between the two is
given by
Tmax − Tmin ≈ 0.406 ∆φ. (4.272)
By measuring the curve T as a function of x and determining the peak-value difference,
one can determine the nonlinear phase shift on the beam axis and hence infer the non-
linear refractive index using Eq. (4.262). This is the essence of the z−scan technique.
Exercise. 4. 16. Derive Eq. (4.271).
104
The necessity to determine the whole T -curve to infer the peak-valley difference is
a disadvantage of the traditional z−scan approach. In practice, an experimentalist has
to translate a sample by translating a stage which entails some technical limitations.
Exercise 4. 17. Assuming your a translation stage allows for a 20 cm travel, how tight
a focus is required such that you could capture a typical z−scan signature (peak-to-
valley) of the transmittance curve? What is a maximum sample thickness such that the
thin sample approximation, ∆L zR , is satisfied for a 632 nm laser beam?
As an alternative, one can tightly focus a Gaussian beam and apply the so-called
quadratic phase approximation (QPA) for the accumulated nonlinear phase shift
2
/ws2
exp[i∆ΦN L (ρ, ∆L)] ' ei∆Φ0 e−i∆Φ0 ρ , (4.273)
which is obtained by expanding the exponent in the last term on the r.h.s. of Eq. (4.263)
into a Taylor series and keeping only quadratic term in ρ/ws . The QPA can be justi-
fied by observing that it is very accurate near the beam axis, ρ ws . As one moves
away from the axis, the QPA accuracy decreases. However, it is largely irrelevant as
the diffraction pattern contribution from the beam tails is practically negligible due to a
very fast (Gaussian) fall off of the field intensity towards the beam periphery. The QPA
is expected to be particularly accurate for tightly focused Gaussian beams, typically
employed in the z-scan measurements in most practical situations.
It then follows at once from Eqs (4.263) and (4.273) that
ikρ2 ρ2
E(ρ, zs ) = Es ei∆Φ0 exp exp − 2 , (4.274)
2Reff 2ws
where the effective radius of wavefront curvature Reff is defined as
1 1 2∆Φ0 zR
= 1− . (4.275)
Reff Rs zs
It can be inferred from Eq. (4.275) that for n2 > 0, Reff > Rs , resulting in partial
diffraction suppression by self-focusing in the sample. Conversely, if n2 < 0, Reff <
Rs causing additional beam spreading due to self-defocusing in the sample. Thus, the
sample works as a thin lens.
Having elucidated the thin sample effect on the beam in the QPA approximation
in physical terms, we can proceed to study beam propagation from the sample to the
aperture. The beam envelope at the aperture can be derived, given its expression at the
sample exit (4.274). The resulting on-axis aperture transmittance takes the form
2
z2 2
zR
1 + z2s 1 + zs L a
R
TQPA (∆Φ0 ) = 2
2 . (4.276)
zs2 zR
1 + z2 1 − 2∆Φ0 zzRs + zs L a
R
Eq. (4.276) can be easily inverted, yielding the expression for the nonlinear phase shift
as
v
2
u 1 − TQPA + z2s2 1 + zR
u
2
zs z zR t zR zs La
∆Φ0 = 1 + R − . (4.277)
2zR zs La zs TQPA
105
Expressions (4.261), (4.262), and (4.277) make it possible to extract both the sign–
determined by the sign of ∆Φ0 in Eq. (4.277)–and magnitude of n2 from a single
measurement of the relative transmittance in the Fresnel zone, a distance La away from
the sample placed at the position zs . There is no need for sample position scanning in
this approach.
Let us discuss key restrictions on relevant distances under the QPA. First, the sam-
ple must be thin, implying that ∆L zs . Second, the beam emerging from the sample
must still be tightly focused for the QPA to hold well, leading to the restriction on the
sample position relative to the beam Rayleigh range, zs zR . Third, the beam should
not have significantly diverged on arrival at the aperture to avoid substantial power loss,
implying that La ≤ zR . Combining these criteria, we arrive at
∆L zs La ≤ zR . (4.278)
Note also that if Eq. (4.278) is met, ∆Φ0 ' ∆φ, yielding
v
2
u 1 − TQPA + z2s2 1 + zR
u
2
zs z zR t zR zs La
∆φ = 1 + R − . (4.279)
2zR zs La zs TQPA
106
while we will then explore the influence of linear anisotropy of the nonlinear medium
– whose nonlinear properties can still be assumed isotropic – on light polarization dy-
namics in such media.
We begin by observing that since there is no privileged direction in such a medium,
the third-order susceptibility tensor cannot have an index – corresponding to a given
(3)
Cartesian coordinate – repeat an odd number of times: In other words, χijjj = 0 for
any j = x, y, z. To demonstrate this property, consider a polarization component, Px ,
(3) (3)
say. If χxyyy 6= 0, it follows that Px = 0 χxyyy Ey Ey Ey 6= 0. On the other hand, po-
larization along the x−axis in an isotropic medium should not be affected by reflections
with respect to the xz−plane. The latter affect the y-component of the field, though,
(3) (3)
Ey → −Ey . Consequently, Px = 0 χxyyy Ey Ey Ey = (−1)3 0 χxyyy Ey Ey Ey , im-
(3)
plying that χxyyy = 0. By the same token, all the other tensor components containing
three repeated indices can be shown to be zero.
Further, we conclude by inspection that there are four kinds of nonzero tensor ele-
ments which are mutually related by the symmetry relations as
(3) (3)
with similar ones for χyyyy and χzzzz . We can then infer from Eqs. (4.282) – (4.285)
as well as Eq. (4.286) that the third-order susceptibility tensor in isotropic media with
inversion symmetry takes a general form
(3)
χijkl = χ(3) (3) (3)
xxyy δij δkl + χxyxy δik δjl + χxyyx δil δjk . (4.287)
The expression (4.287) can be simplified even further for particular nonlinear pro-
cesses if one recalls intrinsic symmetries of χ(3) with respect to frequency permuta-
tions. We will focus here on the self-focusing (SF) process, ω1 = ω2 = −ω3 =
−ω4 = ω. The intrinsic permutation symmetry then implies
(3) (3)
χijkl (−ω, ω, −ω, ω) = χilkj (−ω, ω, −ω, ω). (4.288)
χ(3) (3)
xxyy (−ω, ω, −ω, ω) = χxyyx (−ω, ω, −ω, ω). (4.289)
107
We can then arrive at the final form for the third-order susceptibility for SF in isotropic
media,
(3)
χijkl (−ω, ω, −ω, ω) = χ(3)
xxyy (−ω, ω, −ω, ω)(δij δkl + δil δjk )
+ χ(3)
xyxy (−ω, ω, −ω, ω)δik δjl . (4.290)
(3)
Exercise 4.19. Use intrinsic permutation symmetries of χijkl to determine a general
(3)
form of the susceptibility tensor χijkl (−3ω, ω, ω, ω) for the third harmonic generation
in isotropic media.
where
Ek (−ω) = Ek∗ (ω). (4.292)
Substituting from Eqs. (4.290) into (4.291) we obtain, after some algebra, the expres-
sion
30 (3) X 30 (3) ∗ X
Pi = χxxyy Ei Ek Ek∗ + χ E El El . (4.293)
2 4 xyxy i
k l
e± · e± = 0, e± · e∓ = 1. (4.298)
108
The dot product of the two electric field vectors can then be expressed as
Let us define the nonlinear polarization field in the circular polarization basis by
the expression
(+) (−)
P N L = PN L e+ + PN L e− , (4.300)
It can be inferred from (4.294), (4.299) and (4.300) that
(±)
PN L = 0 [A|E± |2 + (A + B)|E∓ |2 ]E± . (4.301)
Further, we can represent the total polarization field as a linear superposition of the
circular polarization components as
P = P+ e+ + P− e− , (4.302)
where P+ and P− are effectively decoupled – there is an indirect coupling, though, via
the nonlinear susceptibility – such that each polarization component is proportional to
the corresponding electric field viz.,
h i
(±)
P± = 0 χL + χN L E± . (4.303)
The corresponding effective refractive index, including linear as well as nonlinear parts,
can be defined as
(±)
n2± = 1 + χL + χN L . (4.305)
Since in practice, χN L χL , we can make the approximation
(±)
χN L
n± ' nL + . (4.306)
2nL
The analysis of Eqs. (4.302) – (4.306) reveals that in the circular polarization basis,
the nonlinear wave equation in isotropic media can be effectively decoupled into the
two as
∂ 2 E± n2 ∂ 2 E ±
2
− ± = 0, (4.307)
∂t c2 ∂z 2
where we have neglected any spatial dependence in the transverse directions. Equa-
tions (4.307) are satisfied by the plane wave solutions
where
n± ω
k± = . (4.309)
c
109
On substituting from Eq. (4.306) into (4.308), and using the identities
n± = n ± ∆n/2, (4.310)
where
n+ + n−
n≡ , ∆n = n+ − n− ; (4.311)
2
we obtain the expression for the total field as
(2A + B)
n = nL + (|E+ |2 + |E− |2 ), (4.313)
4nL
and the refractive index difference,
B
∆n = n+ − n− = (|E− |2 − |E+ |2 ), (4.314)
2nL
respectively. The electric field can be represented as
Analyzing Eqs. (4.316) – (4.318), we can conclude that the electric field is ellipti-
cally polarized at any position z, according to Eq. (4.315); yet the polarization ellipse
rotates in the xy− plane at the rate proportional to the differences of refractive indices
along the two principal axes. The latter is referred to as birefringence; it is the nonlin-
ear birefringence of the medium that gives rise to polarization rotation even in isotropic
optical media. Notice also that the rate of polarization rotation depends only on the co-
efficient B as is evidenced by Eqs. (4.314) and (4.317), (4.318). Hence the second
term on the r.h.s. of Eq. (4.294) is wholly responsible for nonlinear birefringence ef-
fects. The first term on the r.h.s of (4.294) contributes to the overall phase accretion
factor which is proportional to n, but it does not affect polarization rotation.
110
Exercise 4.21. In the fiber optical case, the nonlinear response of the medium is
of electronic type such that A = B. Silica-glass optical fibers can serve as an im-
portant particular example. Linear birefringence of the fiber is typically introduced –
either intentionally or inadvertently – at the fabrication stage. On account of linear
birefringence, the most general field propagating in such a fiber can be represented as
1
ex Ex eiβx z + ey Ey eiβy z e−iωt + c. c,
E=
2
where βx,y is the propagation constant of the corresponding linear polarization com-
ponent; the field components are assumed to be polarized along the principal axes of
the fiber. Show that the polarization field at the frequency ω is then given by
1
ex Px eiβx z + ey Py eiβy z e−iωt + c. c.,
PN L = (4.319)
2
where
30 (3) 2 2 2 1 ∗ 2 −2i∆βz
Px = χ (|Ex | + |Ey | )Ex + Ex Ey e , (4.320)
4 xxxx 3 3
30 (3) 2 1
Py = χxxxx (|Ey |2 + |Ex |2 )Ex + Ey∗ Ex2 e2i∆βz . (4.321)
4 3 3
Here ∆β = βx − βy .
(3) (3)
The intrinsic permutation symmetry, χijkl (−ω, ω, 0, 0) = χijlk (−ω, ω, 0, 0), implies
that
χ(3) (3)
xyxy = χxyyx , (4.324)
111
Using Eq. (4.324), Eq. (4.323) can be written in the vector form as
Equation (4.325) represents the general form of the polarization associated with the
electro-optical Kerr effect for the electrostatic and optical fields of any polarizations.
Let us focus on the case of linear polarization of the dc field such that
E0 = E0 ex , E = Ex ex + Ey ey . (4.327)
and
PN Ly = 30 χ(3) 2
xxyy E0 Ey . (4.329)
It can be inferred from Eqs. (4.328) and (4.329) that the components of the total polar-
ization field can be represented as
χx = χ(1) + 3χ(3) 2
xxxx E0 , (4.331)
and
χy = χ(1) + 3χ(3) 2
xxyy E0 . (4.332)
The corresponding components of the total refractive index are given by
(3)
3χxxxx 2
nx ' n + E0 , (4.333)
2n
and
(3)
3χxxyy 2
ny ' n + E0 . (4.334)
2n
It follows from Eqs. (4.330) – (4.332) that in electro-optical Kerr effect with the
dc field breaking the azimuthal symmetry, it is the Cartesian components of the optical
field that are decoupled; each satisfies the wave equation – neglecting spatial depen-
dence in the transverse plane – of the form
112
The plane-wave solutions to Eq. (4.335) are
where
nx,y ω
kx,y = . (4.337)
c
The evolution of the optical field can then be represented as
∆n = KλE02 . (4.341)
The Kerr constant is related to relevant components of the susceptibility tensor by the
expression
(3)
3χxyxy
K= . (4.342)
nλ
Exercise 4.22. Determine the dc field strength needed to produce a circular polariza-
tion by a 10 cm long Kerr cell filled with carbon disulfide, CS2 . For carbon disulfide,
K = 3.6 × 10−14 m/V2 .
113
4.14 Spontaneous and stimulated Raman scattering: CW
case
114
Figure 4.28: Illustrating the concept of scattering cross-section.
N ≥ 1 of the medium molecules within the interaction region and the collection solid
angle ∆Ω is typically rather small such that the overall power scattered into ∆Ω can
be evaluated approximately as
Z
dP dσ
P∆ = ' N I0 ∆Ω . (4.345)
dΩ dΩ dΩ
It follows at once from Eq. (4.345) that the scattered power can be calculated from the
knowledge of the differential cross-section which is tabulated in the table below1 for
common gases and liquids, relative to its value for nitrogen.
Molecule Raman shift (cm−1 ) dΩ dσ dσ
/ dΩ N2
N2 2330 1
H2 4156 3.2
O2 1556 1.23
CO2 1388 1.51
CO 2143 0.98
NO 1876 0.49
H2 O 3657 2.5
SO2 1151 4.9
O3 1103 3.0
CH4 2917 7.3
The nitrogen differential cross-section at λp = 488 nm is
dσ
= 5.5 × 10−31 cm2 /Sr/molecule. (4.346)
dΩ N2
Note that as is customary in spectroscopy, Raman shifts are measured in cm−1 , i. e.,
in the inverse vacuum wavelengths in cm. One can convert these units to Raman shifts
1 Source: S. A. Danchkin et.al., “Raman scattering parameters for gas molecules (survey),” Journal of
115
∆λ in wavelength units of µm through
where 104 comes from the difference between nm and cm. The corresponding Raman
frequency shift is then
∆ν = c/∆λ. (4.348)
It can be inferred from the table that due to tiny magnitudes of dσ/dΩ spontaneous
Raman scattering is an extremely weak process with the scattered power being propor-
tional to the incident intensity.
W = − 12 pE, (4.349)
where the induced dipole moment p can be expressed in terms of the molecular polar-
izability α,
p = 0 α(Q)E. (4.350)
116
For sufficiently small vibrations, the polarizability can be expanded into a Taylor series
in terms of the generalized coordinate as
dQ
α ' α0 + Q. (4.351)
dα 0
On combining Eqs. (4.350) and (4.351), we arrive at
dQ
p ' 0 α0 + Q E. (4.352)
dα 0
The force can be determined as a gradient of the energy. It then follows from Eqs. (4.349)
and (4.352) that
0 dQ
Fd = −∇Q W = E2, (4.353)
2 dα 0
Within the oscillator model framework, the vibrational equation of motion can be writ-
ten as
∂t2 Q + 2γ∂t Q + ω02 Q = Fd /m, (4.354)
where ω0 and γ are the resonant frequency and damping rate, and m is a molecule
mass.
The external electric field is due to traveling pump and Stokes waves,
where “NR” stands for non-resonant terms we are not interested in. We seek a driven
solution to the molecular vibration in the form
ω∆ = ωp − ωs . (4.359)
The Raman interaction is only efficient near resonance such that ω∆ ≈ ω0 . Assuming
the pump and Stokes frequency difference falls in the vicinity to resonance, it follows
that
ω02 − ω∆2
− 2iγω∆ ' 2ω0 (ω0 − ω∆ − iγ). (4.360)
On substituting from Eq. (4.360) into (4.358), we obtain
0 (dα/dQ)0 Ep Es∗
Qω = − . (4.361)
4mω0 [(ω∆ − ω0 ) + iγ]
117
We can now determine the induced polarization,
dQ
PN L = N pN L = 0 N QE, (4.362)
dα 0
which can be expressed in terms of the Stokes, pump, and non-resonant frequency
components as
We can infer from Eqs. (4.355), (4.357), (4.361) through (4.363) that the polarization
component oscillating at the Stokes frequency is given by the expression
By the same token, the pump frequency component to polarization takes the form
0 N (dα/dQ)20
χ(3)
s (−ωs ; ωp , −ωp , ωs ) = − . (4.367)
12mω0 [(ω∆ − ω0 ) − iγ]
0 N (dα/dQ)20
χ(3)
p (−ωp ; ωs , −ωs , ωp ) = − , (4.369)
12mω0 [(ω∆ − ω0 ) + iγ]
Thus,
χ(3) (3)∗
p = χs = χ∗s . (4.370)
Let us now write down coupled-wave equations for the stimulated Raman scattering
process,
2ikj ∂z Ej = −µ0 ωj2 PN L (ωj )e−ikj z , j = p, s. (4.371)
118
Explicitly, we obtain for the pump and Stokes field amplitudes the equations
dEp 3ωp2 ∗
i =− χ |Es |2 Ep , (4.372)
dz 4kp c2 s
and
dEs 3ωs2
i =− χs |Ep |2 Es . (4.373)
dz 4ks c2
Assume that the amplitude of the pump field is very large and can be treated as unde-
pleted during the SRS process. Under the undepleted pump approximation, the Stokes
field is governed by the equation
dEs 3ω 2 χs |Ep |2
i =− s Es . (4.374)
dz 4ks c2
After a simple algebra, Eq. (4.374) can be cast into the Stokes intensity evolution equa-
tion as
dIs 3ωs Ip Im{χs }
= Is . (4.375)
dz 20 np ns c2
Eq. (4.375) can be integrated, yielding an exponential growth of the Stokes mode,
where
ωs N (dα/dQ)20
gR0 = , (4.380)
8mω0 γc2 np ns
is a gain factor at the line center and the gain spectrum is Lorentzian,
γ2
L(ω∆ ) = . (4.381)
[(ω∆ − ω0 )2 + γ 2 ]
Typical values of center-line gain factors for gases are around a few cm/GW. For in-
stance, gR0 ≈ 1.5 cm/GW for molecular hydrogen H2 .
Whenever pump depletion cannot be neglected, coupled pump and Stokes wave
dynamics must be studied using the full coupled-wave equations
dIs 3ωs Im{χs }
= Ip Is , (4.382)
dz 20 np ns c2
119
Figure 4.30: Photon number fractions in the Stokes (red) and pump (blue) pulses as
functions of Gz beyond the undepleted pump approximation with x = 0.1.
and
dIp 3ωs Im{χs }
=− Ip Is . (4.383)
dz 20 np ns c2
It follows from Eqs. (4.382) and (4.383) after some algebra that
1 dIp 1 dIs
+ = 0. (4.384)
ωp dz ωs dz
where Nj = Ij /h̄ωj is a photon flux of the pump (Raman) beam. In other words, the
total photon flux is conserved because in an elementary act of Raman scattering a death
of a pump photon corresponds to birth of a Stokes photon and vice versa.
Eq. (4.385) can be used to eliminate the Stokes intensity, say, in favor of the pump
intensity viz.,
Is = N0 h̄ωs − Ip ωs /ωp . (4.386)
On substituting from Eq. (4.386) into (4.382) and (4.383), we can integrate the latter,
yielding
(1 − x)e−Gz
Np (z)/N0 = , (4.387)
x + (1 − x)e−Gz
and
x
Ns (z)/N0 = . (4.388)
x + (1 − x)e−Gz
120
Here we introduced the initial photon number fraction x = Ns0 /N0 in the Stokes
pulse that indicates the percentage of photons carried by the incident Stokes pulse of
the overall input photon number. The total gain is given by
G = G0 L(ω∆ ), (4.389)
where
N0 h̄ωs ωp N (dα/dQ)20
G0 = , (4.390)
8mω0 γc2 np ns
is the gain at the center-line of the gain spectrum. The behavior of the Stokes and
pump intensities is sketched in the Fig. 5.4. It is seen either in the figure or from
Eqs. (4.387) and (4.388) that the Stokes intensity starts off growing exponentially at
short interaction distances. As the interaction distance increases, though, the Stokes
mode growth saturates and the pump mode becomes depleted, transferring energy into
the Stokes mode and medium molecule vibrations.
We have so far focused on the Stokes mode. Let us include the anti-Stokes mode
into our framework as well. To this end, we consider an input field comprised of the
there components, the pump, Stokes and anti-Stokes modes such that
E(t, z) = 21 [Ep ei(kp z−ωp t) + Es ei(ks z−ωs t) + Eas ei(kas z−ωas t) + c. c.]. (4.391)
Let us also introduce the Raman frequency change
ω∆ = ωp − ωs = ωas − ωp . (4.392)
It follows at once from Eq. (4.391) that
E 2 (t, z) = 41 {[Ep Es∗ ei(kp −ks )z + Eas Ep∗ ei(kas −kp )z ]e−iω∆ t + N R + c. c.]. (4.393)
Therefore, we seek the driven solution for the generalized molecule vibration coordi-
nate in the form
Q(t, z) = 12 {[Qωs ei(kp −ks )z + Qωas ei(kas −kp )z ]e−iω∆ t + c. c.}. (4.394)
On substituting from Eq. (4.394) into (4.354), we obtain for the spectral amplitude of
the Stokes mode
0 (dα/dQ)0 Ep Es∗
Qωs = − , (4.395)
4mω0 [(ω∆ − ω0 ) + iγ]
and for the anti-Stokes one
0 (dα/dQ)0 Eas Ep∗
Qωas = − . (4.396)
4mω0 [(ω∆ − ω0 ) + iγ]
The nonlinear polarization then can be expressed as
PN L = 21 [PN L (ωs )e−iωs + PN L (ωas )e−iωas + PN L (ωp )e−iωp + c. c.], (4.397)
where the Stokes and anti-Stokes components take the form
20 N (dα/dQ)20 eiks z ∗ i∆kz
|Ep |2 Es + Ep2 Eas
PN L (ωs ) = e , (4.398)
8mω0 [(ω∆ − ω0 ) − iγ]
121
and
20 N (dα/dQ)20 eikas z
|Ep |2 Eas + Ep2 Es∗ ei∆kz .
PN L (ωas ) = (4.399)
8mω0 [(ω∆ − ω0 ) + iγ]
The coupled-mode equations can then be written as
2ikj ∂z Ej = −µ0 ωj2 PN L (ωj )e−ikj z , j = p, s, as, (4.400)
or explicitly for the Stokes mode:
∗ i∆kz
i∂z Es = −ξs |Ep |2 Es + Ep2 Eas
e , (4.401)
and for the ant-Stokes mode:
i∂z Eas = −ξas |Ep |2 Eas + Ep2 Es∗ ei∆kz .
(4.402)
Here
0 ωs N (dα/dQ)20
ξs = , (4.403)
16mω0 ns c[(ω∆ − ω0 ) − iγ]
0 ωas N (dα/dQ)20
ξas = , (4.404)
16mω0 nas c[(ω∆ − ω0 ) + iγ]
are the corresponding coupling constants and the wave-vector mismatch is defined as
∆k = 2kp − ks − kas . (4.405)
In the undepleted pump approximation and with very large wave-vector mismatch,
the fast oscillating second term on the r. h. s of Eqs. (4.401) and (4.402) can be dropped
and the resulting decoupled wave equations for the Stokes and anti-Stokes modes can
be written as
i∂z Es = −κs |Ep |2 Es , (4.406)
and
i∂z Eas = −κas |Ep |2 Eas . (4.407)
The equation for the Stokes mode is equivalent to Eq. (4.375) describing exponential
gain, and the anti-Stokes mode evolution is governed by
dIas
= −γas Ip Ias , (4.408)
dz
where the anti-Stokes loss factor is defined as
γas = γas0 L(ω∆ ), (4.409)
with the center-line loss factor given by
ωas N (dα/dQ)20
γas0 = . (4.410)
8mω0 γc2 nas np
It follows from Eqs. (4.408) through (4.410) that for sufficiently large phase mismatch,
the anti-Stokes mode decays exponentially with the decrement γas ,
Ias = Ias0 exp(−γas Ip z). (4.411)
Thus unless one takes care to phase-match the anti-Stokes mode, it exponentially de-
cays on propagation. This is the reason one usually focuses on the Stokes mode.
122
4.15 Transient stimulated Raman scattering
We present a semiclassical theory2 of transient SRS in a hollow-core photonic crystal
fiber (HCPCF) geometry. We consider a single-mode HCPCF filled with a molecular
gas such as hydrogen. The single-mode HCPCF use guarantees a very large interaction
length as light diffraction is arrested and its energy is squeezed into a tight fiber mode.
Further, a properly engineered HCPCF can ensure all higher-order Stokes mode sup-
pression such that the SRS excitation process engages only the first-order Stokes mode,
thereby allowing to focus on the basic two-mode, pump and Stokes, situation. Quan-
tum mechanically, a pump pulse photon absorption promotes a molecule from its initial
state, labelled as “1”, to one of the intermediate excited states, labelled as “i”, that are
far off resonance with both laser and Stokes pulses. As the excited molecule emits a
photon at the Stokes frequency, it descends to a final state “3” which differs from the
ground state. The released energy h̄(ωi1 − ωi3 ) excites molecular vibration/rotation;
ωi1 and ωi3 are transition frequencies characterizing dipole allowed transitions from
the initial and final levels, respectively, to any intermediate level. The appropriate en-
ergy level diagram is sketched in Fig. 4.31
We will make two assumptions. First, the pump and Stokes pulse frequencies are
tuned to exact Raman resonance such that ω31 ≡ ω3 − ω1 = ωp − ωs , where ω1,3 =
E1,3 /h̄ describe initial/final energies, E1,3 in frequency units, and ωp as well as ωs
are pump and Stokes pulse frequencies, respectively. The exact resonance implies that
ωi1 −ωi3 = ωp −ωs . The exact Raman resonance is desirable to increase SRS efficiency
and it can be attained in HCPCFs thanks to homogeneous line broadening there. This
is because molecule collisions with the fiber core walls are the dominant mechanism of
2 Some quantum mechanics background is required to understand the material of this section–see, for
instance, Chaps. 2, 3, 4 & 14 of “Quantum Mechanics for Scientists & Engineers,” by D. A. B. Miller,
Cambridge University Press, 2008.
123
Figure 4.32: Forward-and backward-propagating SRS geometries
spectral line broadening in the system for sufficiently high gas pressures. The second
assumption, corresponding to realistic experimental conditions, is weak excitation such
that all excited level populations–including that of the final level–are negligible. The
lack of population transfer implies that
and consequently, the off-diagonal density matrix elements (coherences) are small as
well,
ρi1 , ρi3 , ρ31 1. (4.413)
The density matrix evolution is governed by the Schrödinger equation that reads
Here +(−) sign in the first term on the r.h.s. corresponds to the co-propagating
(counter-propagating) geometry of Fig. 4.32; Ep and Es are the pump and Stokes
pulse amplitudes, respectively, and di1 and di3 are the corresponding dipole matrix
124
Figure 4.33: Forward-and backward-propagating SRS excitation geometries.
elements. Finally “h.c.” is a shorthand for Hermitian conjugate. It follows at once from
Eq. (4.415) that the only nonzero matrix elements of V̂ are
and
Vi3 = 21 di3 Es ei(ks z−ωs t) , V3i = Vi3∗ , (4.417)
The corresponding density operator matrix elements obey the evolution equations
and
i ∗
P
∂t ρ̃31 = −γ31 ρ̃31 − 2h̄ i (d3i Es ρ̃i1 − di1 Ep ρ̃3i ). (4.425)
Here we introduced ∆ = ωi3 − ωs = ωi1 − ωp .
Let us now recall that all intermediate levels are far off resonance with the two
dipole-allowed transitions. Mathematically, this implies large frequency detuning such
that
γi1 ∆, γi3 ∆. (4.426)
125
We can then adiabatically eliminate the intermediate levels. This term implies that
the far-off resonance density matrix elements ρ̃i1 and ρ̃i3 rapidly oscillate around their
local equilibrium values, determined by the pulse field amplitudes and the molecular
coherence between the initial and final states, ρ31 . We can then formally drop the time
derivatives on the l.h.s of Eqs. (4.423) and (4.424) to obtain
di1 Ep di3 Es
ρ̃i1 = − − ρ̃31 , (4.427)
2h̄∆ 2h̄∆
and
di1 Ep ∗
ρ̃i3 = − ρ̃ . (4.428)
2h̄∆ 31
It follows from Eq. (4.413) that to the leading order in the small parameter, ρ̃31 , the
off-diagonal matrix elements read
di1 Ep
ρ̃i1 ' − , ρ̃i3 ' 0. (4.429)
2h̄∆
On substituting from Eq. (4.429) into (4.425), we arrive at
i X d3i di1
∂t ρ̃31 = −γ31 ρ̃31 − 2 Ep Es∗ . (4.430)
4h̄ i ∆
and
2ikp (±∂z Ep + β1p ∂t Ep ) = −µ0 ωp2 PN L (ωp )e∓ikp z , (4.434)
where β1p,s are group velocities of the pump and Stokes pulses. The nonlinear polar-
ization field can then be expressed as
ˆ = PN L (ωp )e−iωp t + PN L (ωs )e−iωs t + c.c..
PN L = N hdi (4.435)
126
Here the average dipole moment is determined by
ˆ = Tr{ρd},
hdi ˆ (4.436)
It then follows from Eqs. (4.435) through (4.437) after some straightforward albegra
that X
PN L (ωp ) = N di1 ρ̃i1 eikp z , (4.438)
i
and X
PN L (ωs ) = N di3 ρ̃i3 eiks z . (4.439)
i
On substituting the approximate expressions (4.427) and (4.428) for the density
matrix elements into Eqs. (4.433) and (4.434) and using Eqs. (4.438) and (4.439), we
arrive, after some algebra, at the following set of coupled-wave equations for the pump
and Stokes pulse amplitudes,
iωp N reff
±∂z Ep + β1p ∂t Ep = σEs , (4.440)
20 cnp
and
iωs N reff
∂z Es + β1s ∂t Es = σ ∗ Ep . (4.441)
20 cns
Note that in deriving the last two equations we dropped a term on the l.h.s of Eq. (4.440)
linear with respect to Ep . This term arises when the first term on the r.h.s. of Eq. (4.427)
is substituted to the r.h.s. of Eq. (4.433). One can show the dropped linear term leads to
a common global phase of the pump and Stokes fields which does not affect the Raman
dynamics.
Exercise 4.23. Re-derive the coupled-wave equations for transient SRS by keeping the
term linear in the pump pulse amplitude. Show that by a gauge transformation, Ep =
Ẽp eiαz and Es = Ẽs e−iαz with the appropriate α to be determined, the linear term can
be eliminated from your equations and Eqs. (4.440), (4.441) can be recovered. Notice
also that the Raman density matrix evolution, Eq. (4.432), is unaffected by the gauge
transformation.
Let us now consider co- and counter-propagating cases separately.
Co-propagating geometry. – We assume the fiber is engineered to be dispersion-flat,
implying that β1p = β1s = β1 . We can then transform to a reference frame co-moving
with the pulses and introduce new variable, ζ = z and τ = t − β1 z. The resulting SRS
equations read
iω N r
∂ζ Ep = 2p0 cneff p
σEs , (4.442)
∂ζ Es = iω2s0Ncnreff
s
σ ∗ Ep , (4.443)
127
and
ireff
Ep Es∗ .
∂τ σ = −γσ + 4h̄ (4.444)
We can then introduce the pulse peak optical intensities viz.,
0 ns,p c|Es0,p0 |2
Is0,p0 = , (4.445)
2
and proceed normalizing the pulse fields to the peak pump intensity at the source,
Ep = (2Ip0 /0 cnp )1/2 E p and Es = (2Ip0 /0 cnp )1/2 E s and introducing dimensionless
distance and time, Z = ζ/LSRS and T = τ /TSRS . We introduced here characteristic
SRS interaction distance and time viz.,
−1 −1
N reff ωp ωs reff Ip0
r
LSRS = , TSRS = . (4.446)
20 c np ns 2h̄0 cnp
∂Z E p = iκσE s , (4.447)
∂Z E s = iκ−1 σ ∗ E p , (4.448)
and
∗
∂T σ = −Γσ + iE p E s . (4.449)
p
Here κ = ωp ns /ωs np and Γ = γTSRS is a key dimensionless parameter governing
the SRS process.
Counter-propagating geometry. – Assuming a dispersion-flat fiber, we can arrive at the
dimensionless SRS field equations as
and
∂Z E s + δ ∂T E s = iκ−1 σ ∗ E p , (4.451)
where δ = β1 LSRS /TSRS . Note that whereas in the co-propagating case one can
transform away the drift term, the latter is not possible in the counter-propagating case
even if the two group velocities are identical. Physically, this drift term indicates that
any fixed points in pulse profiles, the peak intensity positions, say, shift relative to their
positions at the fiber inputs as the Stokes pulse zaps by the pump one.
Let us now briefly discuss order-of-magnitude values of relevant parameters for
SRS in HCPCF. For simplicity, we assume the fiber is filled with molecular hydrogen.
Typically nanosecond pump pulses, 1 ≤ tp ≤ 10 ns carrying from 10 to 100 µJ are
employed. The Raman dipole moment matrix element can be estimated as reff '
1.4 × 10−41 C m2 /V. Typical gas densities and relaxation time at 1 bar of pressure are3
N ' 2 × 1020 cm−3 and TR = γ −1 ' 5 ns.
Finally, we can derive an approximate expression for the threshold gain required
to jumpstart the SRS process from noise. To this end, we need an expression for a
3 Source: F. Flora and L. Giudicotti, Appl. Opt. 26, 4001-4008 (1987).
128
linear gain in the cw case. The latter can be easily obtained by looking int a cw limit of
Eqs. (4.442) through (4.444) in the undepleted pump approximation. Namely, assum-
ing, Ep = const and letting ∂τ σ = 0, we can derive an expression for the amplified
Stokes pulse intensity and comparing with Eq. (4.376), we can infer conclude that
ωs N |reff |2
gR = . (4.452)
420 c2 ns np h̄γ
Exercise 4.24. Derive Eq. (4.452) and compare with Eq. (4.380).
We can finally estimate a critical power required for SRS generation with a cw laser
source in an HCPCF. The input intensity can be estimated as Ip0 ' P/Aeff , where P is
the pump power and Aeff is a an effective HCPCF core area which takes into account
the spatial pump mode distribution in the fiber. At the threshold, Raman gain is
The threshold gain is determined empirically to fall in the range 20 ≤ Gth ≤ 30,
yielding the power estimate
Gth Aeff
Pth ' . (4.454)
gR L
We assume that the medium is isotropic which is an adequate model for gases and
liquids. The first (large) term on the r.h.s. of Eq. (4.455) describes the contribution due
to electrostriction, i. e, the tendency of the dielectric constant to vary with the medium
density, whereas the second–smaller–term is due to temperature variations caused by
absorption in the medium. In practice, the second term on the r.h.s of Eq (4.455) is
negligible for transparent or nearly transparent media which we assume to be the case
hereafter. Thus, to good accuracy, we can write
0 ∂ γe
= ρ0 = ρ0 , (4.456)
∂ρ ρ
where we dropped the subscript T for notational simplicity and introduced the elec-
trostriction constant γe , defined as
∂
γe ≡ ρ . (4.457)
∂ρ
129
Figure 4.34: Illustrating momentum conservation in a single photon-phonon scattering
act.
PN L = 0 0 E, (4.458)
where in the absence of the input Stokes mode, the electric field is represented by the
pump wave such that
E = 21 Eei(kp ·r−ωp t) + c.c., (4.459)
and the density fluctuations constitute a propagating acoustic wave,
Here q and Ω are the wave vector and frequency of an acoustic (Brillouin) phonon
which are related by the dispersion relation
Ω = |q|cs , (4.461)
where
ωs = ωp − Ω, ωas = ωp + Ω, (4.463)
are the generated Stokes and anti-Stokes frequencies. The Stokes polarization compo-
nent, for example, can be written explicitly as
P(ωs ) = 02ργe E ρ̃∗ eiks ·r , (4.464)
130
where
ks = kp − q. (4.465)
As Ω ωp , ωs due to phonon sluggishness, cs c, it is reasonable to assume that the
magnitude of photon momentum does not appreciably change at each scattering event,
kp ≈ ks . It then follows from Eqs. (4.461), (4.463) and (4.465) and the geometry of
Fig. 5. 6 that
Ω = 2kp sin θ/2, (4.466)
where θ is a photon scattering angle. It follows at once from Eq. (4.466) that the most
energetically efficient Stokes wave generation–because ΩB is maximized–takes place
in the backward direction, θ = π. Moreover, the energy-momentum conservation
laws in each scattering event explicitly prohibit Stokes wave generation in the forward
direction, θ = 0.
In the absence of heat transfer in transparent media, we can assume that acoustic waves
are adiabatic, that is the entropy is conserved, S = const. It then follows that the
pressure gradient can be expressed as
∂p ∂ρ
= c2s , (4.471)
∂z ∂z
131
where s
∂p
cs = , (4.472)
∂ρ S
ρ = ρ0 + ρ0 , ρ0 ρ0 , (4.473)
into Eqs. (4.467) and (4.468) and keep only terms linear with respect to either ρ0 or v,
leading to the acoustic approximation:
∂t ρ0 + ρ0 ∂z v = 0, (4.474)
and
c2s
∂t v = − ρ0 ∂z ρ0 + 0 γe
2ρ0 ∂z E 2 + (ζ + 43 ν)∂zz
2
v (4.475)
Taking the time and space derivatives on both sides of Eqs. (4.474) and (4.475), respec-
tively, we can transform our acoustic equations to
2 2 0
∂zt v = − ρ10 ∂tt ρ, (4.476)
and
4
c2s (ζ+ 3 ν) 3
2 2 0 0 γe 2
∂tz v=− ρ0 ∂zz ρ + 2ρ0 ∂zz E2 + ρ0 ∂tzz ρ0 . (4.477)
In deriving Eq. (4.477), we eliminated the fluid velocity on the r. h. s., with the help of
Eq. (4.474). Equating the mixed velocity derivatives from Eq. (4.476) and (4.477), we
eliminate the velocity field and arrive at the wave equation for the density variations
alone in the form
2 0 2 0
∂tt ρ − c2s ∂zz 3
ρ − Γ∂tzz ρ0 = − 21 0 γe ∂zz
2
E2. (4.478)
ρ0 ∝ ei(qz−Ωt) . (4.480)
On substituting from Eq. (4.480) into (4.478) with E=0, we arrive at the dispersion
relation
−Ω2 + c2s q 2 − iΓΩq 2 = 0. (4.481)
132
Or,
Ω2 /c2s Ω2
2 iΓΩ
q = iΓΩ
' 2 1+ 2 , (4.482)
1 − c2 cs cs
s
where we assumed that ΓΩc2s 1. At the same level of approximation, we can obtain
the wave number as
q ' Ω/cs + iα/2, (4.483)
where we introduced the inverse phonon damping length α as
α = Γq 2 /cs . (4.484)
We can now estimate characteristic parameters of Brillouin acoustic waves noticing that
as follows from Eqs. (4.465) for backward propagating Stokes modes, q ' 2kp ∼ 4π ×
106 m−1 for the pumping wavelength of the order of 1µm. Assuming further typical
values: Γ ∼ ζ ∼ ν = η/ρ0 with ρ0 ∼ 103 kg/m3 and the dynamic sheer viscosity
η ∼ 10−3 N s/m2 as well as cs ∼ 103 m/s, we estimate the Brillouin frequency to
be Ω/2π ∼ qcs /2π ∼ 2 GHz, and the inverse damping length α ∼ 1.6 × 105 m−1 ,
implying that phonons are damped over a characteristic length α−1 of just 5 µm. Thus,
generated ultrasound phonons are strongly damped over a characteristic length over
which the fluid density or the applied electric field amplitude changes.
where Ep and Es are slowly varying pump and Stokes pulse amplitudes, and we as-
sume that the Stokes mode propagates in the backward direction to maximize the SBS
efficiency. The electrostriction coupling is specified by the term
where NR, as before, stands for non-resonant terms. The generated acoustic wave can
be described in terms of the medium density variations as
Here,
q = kp + ks ' 2k, Ω = ωp − ωs , (4.488)
where k = kp ' ks .
133
On substituting from Eq. (4.485) through (4.487) into the acoustic wave equa-
tion (4.478), we obtain
∆ = ΩB − Ω. (4.492)
Eq. (4.491) is the material evolution equation for the SBS process.
The coupled-wave equations for the pump and Stokes pulses can be derived in strict
analogy with the SRS wave equations. Specifically, starting from Maxwell’s equations
an applying SVEA, we obtain
and
2ikp (∂z Ep + βp ∂t Ep )eikp z = −µ0 ωp2 PN L (ωp ). (4.494)
Here βs and βp are the inverse group velocities of the Stokes and pump pulses. The
induced polarization field can be expressed as
where using Eqs. (4.456), (4.458) and (4.487), we obtain, by analogy with Eq. (4.464),
PN L (ωs ) = 2ρ0 γe
0
Ep ρ̃∗ e−iks z , (4.496)
and
0 γe
PN L (ωp ) = 2ρ0 Es ρ̃eikp z . (4.497)
On substituting from Eqs. (4.496) and (4.497) into Eqs. (4.494) and (4.495), we obtain,
after simple algebra, the following equations
ωs2 γe
2iks (−∂z Es + βs ∂t Es ) = − 2 Ep ρ̃∗ , (4.498)
c 2ρ0
and
ωp2
γe
2ikp (∂z Ep + βp ∂t Ep ) = − 2 Es ρ̃. (4.499)
c 2ρ0
134
Introducing ω = ωp ' ωs , n = np ' ns , we can cast Eqs. (4.498) and (4.499) to
iωγe
−∂z Es + βs ∂t Es = 4cnρ 0
Ep ρ̃∗ , (4.500)
and
iωγe
∂z Ep + βp ∂t Ep = 4cnρ0 Es ρ̃. (4.501)
Eqs. (4.500) and (4.501), together with Eq. (4.491) form the basis for quantitative de-
scription of SBS.
In the cw limit, the stationary solution to Eq. (4.491) can be easily obtained drop-
ping the time derivative of ρ̃, yielding
i0 γe k 2 Ep Es∗
ρ̃ss = . (4.502)
ΩB (ΓB /2 + i∆)
On substituting from Eq, (4.502) into (4.500) and (4.501), and dropping the time
derivatives of the field amplitude, we obtain
0 ωγe2 k 2
∂z Es = − |Ep |2 Es , (4.503)
4ΩB cnρ0 (ΓB /2 − i∆)
and
0 ωγe2 k 2
∂z Ep = − |Es |2 Ep , (4.504)
4ΩB cnρ0 (ΓB /2 + i∆)
The latter can be transformed to the evolution equations for the optical intensities, by
analogy with the SRS case, yielding
dz Is = −gB Is Ip , (4.505)
and
dz Ip = −gB Is Ip , (4.506)
where we introduced the SBS gain coefficient as
Γ2B /4
gB = gB0 2 . (4.507)
ΓB /4 + ∆2
ω 2 γe2
gB0 = . (4.508)
nc3 cs ρ0 ΓB
Let us estimate a typical value of gB0 . Consider CS2 at 1 µ as a characteristic medium
for SBS. Taking typical values ω/2π = 3 × 1014 Hz, γe = 2.4, n ' 1.7, cs ' 103 m/s,
ρ0 ' 1300 kg/m3 , and Γ−1 B ' 4 × 10
−9
s, we obtain an estimate gB0 ' 0.2 cm/MW.
In the undepleted pump approximation, Ip = const and a straightforward integra-
tion of Eq. (4.505) yields,
135
It follows at once from Eq. (4.509) that the Stokes mode intensity grows exponen-
tially on its propagation backward from the exit to the interaction volume towards its
entrance.
Whenever pump depletion is no longer negligible, Eqs. (4.505) and (4.506) must
be considered together. It can be easily inferred that
dz Is = dz Ip , (4.510)
implying that
Ip (z) = Is (z) + C, (4.511)
where C = Ip0 −Is0 is determined by initial conditions. Eliminating the pump intensity
Figure 4.35: Intensity transfer characteristics of SRS beyond the undepleted pump
approximation.
However, as the Stokes mode propagates backwards, we must express the Stokes inten-
sity at the input Is0 in terms of its value at the output Is (L) which serves as the proper
initial condition. It follows from Eq. (4.513) that
136
which is a transcendental equation giving the unknown reflection coefficient at the
input, Is0 /Ip0 in terms of Is (L)/Ip0 . Instead of solving it however, we can simply plot
gB Ip0 L as a function of Is0 /Ip0 and swap the axes. The result is displayed in Fig 5.7.
It is seen from the figure that significant pump depletion takes place when the overall
gain G = gB Ip0 L exceeds the value of 20.
In general, there are two possible SBS setups illustrated in Fig. 5.8. If both pump
and Stokes fields are present–as is seen in Fig. 5.8 on the left–the process is referred to
as SBS amplification whereby the initially Stokes mode is amplified while propagating
in the backward direction. In the second instance, shown in Fig. 5.7 on the right,
Figure 4.36: Illustrating two SBS modalities: SBS amplification (left) and SBS gener-
ation (right).
no Stokes component is initially present and SBS generation starts off from thermal
noise near distributed throughout the interaction region. The process then starts as
spontaneous Brillouin scattering triggered by thermal noise. The key metric for such
SBS generation efficiency is the reflection coefficient defined as
Is0
R= . (4.515)
Ip0
In the initial stage of the process pump depletion is negligible. Hence, the Stokes
intensity at the output face z = 0 is related to its magnitude at the input z = L
according to Eq. (4.509):
Is0 = Is (L)eG , (4.516)
where the overall gain at the Brillouin resonance is defined as
137
The critical power is determined by G = Gth such that
Gth λp
Pth ' . (4.519)
2gB0
For CS2 , for example, gB0 ' 0.2 cm/MW and using Gth = 28 and λp = 1 µm, we
arrive at Pth ' 7 kW.
138