Chapter 2

Theoretical fundamentals.
Propagation equations

In this chapter we establish the theoretical basis for the further analytical
and numerical studies.

2.1 The propagation equation

The standard theoretical method in nonlinear and fiber optics is the slowly-
varying envelope approximation. In many cases, the time dependence of the
electric field consists of an envelope which varies on the time scale of the pulse
duration modulated by oscillations on the time scale of the reverse frequency.
~ r, t) can
Therefore, the rapidly varying part (carrier) of the electric field E(~
~ r, t):
be separated from the slowly varying envelope A(~

~ r, t) = 1 A(~
E(~ ~ r, t) exp[iω0 t − k(ω0 )z] + c.c. , (2.1)
2

21
CHAPTER 2. THEORETICAL FUNDAMENTALS 22

where ~r = {x, y, z}, ω0 is the input carrier frequency, k(ω) = n(ω)ω/c the
wavevector and n(ω) the frequency-dependant refractive index. In the SVEA
the slowly varying envelope is assumed to satisfy the condition
 
 ∂A
~   
   ~
   ω0 A  , (2.2)
 ∂t 

which is fulfilled if and only if the spectral width of a pulse ∆ω is much

smaller than the carrier frequency ω0 of the pulse: ∆ω  ω0 . The latter
condition allows to neglect higher-order terms in the Taylor expansion of
k(ω) around ω0 when determining the influence of dispersive effects:

k 00 (ω0 ) k 000 (ω0 )

k(ω) = k(ω0 )+k 0 (ωo )(ω−ω0 )+ (ω−ω0 )2 + (ω−ω0 )3 +. . . (2.3)
2 6

Obviously, these approximations are no longer valid for radiation with ultra-
wide spectra such as pulses with a duration approaching one optical cycle.
Pulse propagation without the special prerequisites of the SVEA can
be studied by the numerical solution of Maxwell’s equations by the finite-
difference time-domain method (see Ref. [39, 40, 14] and references therein).
In this method, two shifted space-time grids in the scale of the wavelength/op-
tical period are used for the description of the electric and magnetic fields
of the pulse. The values of the field for the next time step are found by a
special ”leap-frog” algorithm which ensures the second-order accuracy of this
method. However, due to the high-resolution grid needed in both space and
time the large numerical effort in this approach limits the possible propaga-
tion lengths to a few mm. In several papers various improved equations have
been derived that allow the theoretical description beyond the validity of
the standard approximations [41, 42, 43]. In the following we give a system-
atic derivation of a first-order unidirectional propagation equation without
CHAPTER 2. THEORETICAL FUNDAMENTALS 23

the use of the SVEA and the Taylor expansion of the linear refraction index.
This equation extends previously derived equations into the non-paraxial and
extremely nonlinear region from which the basic equations of Ref. [43] or [41]
can be derived in a physically transparent manner.
We start from the Maxwell equations in the form
~ ~
~ = ∂B , ∇ × H
∇×E ~ = ∂ D + J~f ,
∂t ∂t
~ ~ = 0.
∇ · D = ρf , ∇ · B (2.4)

~ is the magnetic field, J~f and ρf are the current density and the
Here H
~ and D
charge density, B ~ are given by the constitutive relations D ~+
~ = 0 E

P~ , B
~ = µ0 H
~ +M
~ where 0 and µ0 are vacuum permittivity and permeability,

respectively, and P~ and M

~ are induced electric and magnetic polarizations.

The curl ∇× is applied to the first equation of this set, and we can get the
propagation of pulses in nonlinear media in the form
 2 
∂ ~ 1 ∂2 ~ ∂2 ~
+ ∆ ⊥ E − E = µ 0 P , (2.5)
∂z 2 c2 ∂t2 ∂t2

where
∆⊥ = ∂ 2 /∂x2 + ∂ 2 /∂y 2 . (2.6)

Assumptions necessary for rewriting (2.4) in the form (2.5) are

• the density of free charges and currents is zero

• the medium is non-magnetic

• dielectric permittivity is piecewise constant and scalar

which are satisfied with a high accuracy for propagation in fibers. The lat-
ter assumption means that a fiber can be separated into several regions so
CHAPTER 2. THEORETICAL FUNDAMENTALS 24

that within each of them the dielectric permittivity is a constant. On the

boundaries of these regions, the usual boundary conditions are set, i.e. the
~ and a normal component of D.
continuity of a tangential component of E ~

Substituting the Fourier-transformed field

Z ∞
~ ~
E(z, ω, k⊥ ) = exp(iωt − i~k⊥ r~⊥ )E(~
~ r, t)dtd~r⊥ , (2.7)
−∞

where ~r⊥ = {x, y}, ~k⊥ = {kx , ky } into the wave equation (2.5) we obtain
~ ω, ~k⊥)
∂ 2 E(z, 2
+ βNL ~ ω, ~k⊥ ) = 0
(ω)E(z, (2.8)
∂z 2

with
 1/2
ω2
βNL (ω, ~k⊥ ) = [1 + χ(ω)] − ~k⊥
2
+ µ0 ω 2BNL (z, ω, ~k⊥ ) . (2.9)
c2

We assume P~NL ||E,

~ which requires that the

• medium be isotropic,

to get
PNL (z, ω, ~k⊥ )
BNL = . (2.10)
E(z, ω, ~k⊥ )
Here we separate the polarization into a linear and a nonlinear part as

P~ = P~L (~r, ω) + P~NL (~r, ω), (2.11)

where P~NL (~r, ω) is the Fourier transform of the nonlinear part of the polar-
ization and
P~L (~r, ω) = 0 [n2 (ω) − 1]E(~
~ r, ω) (2.12)

its linear part.

In most typical situations the condition µ0 ω 2 BNL  2βNL
2
is satisfied
with high accuracy, because the contribution of nonlinear polarization is
CHAPTER 2. THEORETICAL FUNDAMENTALS 25

much smaller than the field itself for any practical situation in which we can
neglect contribution from ionization as we do here. Therefore, the separation

∂ 2 /∂z 2 + βNL
2
(ω) = [∂/∂z − iβNL (ω)][∂/∂z + iβNL (ω)] (2.13)

is possible. The electric field can be separated into a forward E + and a

backward E − propagating part:

E(z, ~ + (z, ~k⊥ , ω) + E

~ ~k⊥ , ω) = E ~ − (z, ~k⊥ , ω). (2.14)

Let us consider a pulse propagating in the forward direction along the z-

~ + (z, ~k⊥ , ω) ∼ exp[ik(ω)z]E
axis E ~ 0 (z, ~k⊥ , ω) and neglect waves propagating

backward. This requires that the refractive index be constant or a smooth,

slowly changing function of the z coordinate. Therefore we have

∂E~+
(z, ~k⊥ , ω) = iβNL (z, ~k⊥ , ω)E
~ + (z, ~k⊥ , ω) , (2.15)
∂z

when the effect of the backward wave on BNL can be neglected. This equa-
tion is applicable for the description of light propagation, if the following
conditions are satisfied:

• The product of the nonlinear refractive index and intensity is much less
than unity. This is valid for both materials considered here (fused silica
and argon) as well as for almost all other materials for the intensities
below the damage threshold;

• The nonlinear polarization and the electric field have the same direc-
tion;

• The initial condition consists only of the wave propagating in one di-
rection. This requirement is always satisfied after certain propagation
CHAPTER 2. THEORETICAL FUNDAMENTALS 26

length, even if the input field contained both forward and backward
waves initially. Each of the waves can be described by Eq.(2.15);

• The dependence of the refractive index on the propagation length, if

present, must be weak.

For the backward wave we find

∂E~−
(z, ~k⊥ , ω) = −iβNL (z, ~k⊥ , ω)E
~ − (z, ~k⊥ , ω) . (2.16)
∂z

Equation (2.15) represents a more general approach than the standard one
and is even more accurate than previously derived evolution equations with-
out SVEA as presented in [43]. This equation includes in the theoretical
analysis broad bandwidth, sharp temporal features, space-time coupling and
higher-order nonlinear dispersive effects. Note that the SVEA with paraxial
approximation for the transverse momentum fails to describe self-focusing in
dispersive media accurately long before the temporal structure reaches the
time of an optical cycle [44]. This effect is a result of space-time focusing of
short pulses leading to a reduced axially projected group velocity of wide-
angle rays in the angular spectra. Equation (2.15) can be numerically solved
by the second-order split-step Fourier method.
The square root in Eq. (2.9) is expanded as
2
k⊥ µ0 ω 2
β(ω, ~k⊥) ' k(ω) − + BNL (2.17)
2k(ω) 2k(ω)

which additionally requires

• paraxial propagation.

However, space-time coupling is still taken into account. To account for

CHAPTER 2. THEORETICAL FUNDAMENTALS 27

losses, k(ω) can be made complex: k(ω) = k0 (ω) + iα(ω). With the intro-
duction of the moving time coordinates ξ = z, η = t − zn g /c with

∂/∂z = ∂/∂ξ − ng c−1 ∂/∂η (2.18)

we obtain the following basic equation in Fourier presentation which we de-

note in the following as forward Maxwell equation (FME) [43]:

~ r, ω)
∂ E(~ h ωng i ~
= i k(ω) − E(~r, ω)
∂ξ c

i ~ r, ω) + iµ0 ωc P~NL (~r, ω).

+ ∆⊥ E(~ (2.19)
2k(ω) 2n(ω)

Here ng is a group refractive index ng = n(ω) + ω(dn/dω) calculated at

any frequency inside the spectrum. The choice of n g determines only an
arbitrary time shift. Equation (2.19) is a generalization of the so-called
reduced Maxwell equation [41, 49] which is valid only for a

• refractive index close to unity

• and also with neglection of the back propagating wave.

This equation is obtained from (2.5) by the substitution n(ω) − 1 ' [n 2 (ω) −
1]/2 and back transformation into the time domain:
~
∂E 1 ∂ P~
=− . (2.20)
∂ξ 20 c ∂η

This equation is a useful tool for examining the nonlinear effects of ultra-
broadband radiation in gaseous media. It was recently applied for the study
of pulse compression and SC generation by the optical Kerr effect [14, 45] or
by high-order stimulated Raman scattering [46, 47] in hollow waveguides. In
Ref. [48] an extended version of this equation was used with inclusion of the
CHAPTER 2. THEORETICAL FUNDAMENTALS 28

diffraction term. However, as one would expect, for optically dense media
(n − 1 ∼ 1) solutions of this equation differ from those of the exact Maxwell
equation as can be seen by comparison of FME and the reduced Maxwell
equation in Fig. 2.1(b). It can be seen that results obtained by using FME
are in good agreement with exact results obtained by the FDTD method in
Ref. [14]. This shows that Eq. (2.19) is completely valid for description
of ultrabroadband radiation with good accuracy and reasonable numerical
effort.
Equation (2.19) generalizes the standard approximate evolution equation
for the envelope in nonlinear optics. To show this, we introduce the envelope
~ η) as given by Eq. (2.1), choose the same prerequisites (2.2) and expand
A(ξ,
β(ω) in the form (2.17) around ω0 . After substitution of (2.1) into (2.19) the
propagation equation for the linearly polarized field of a plane wave takes
the form which can be found elsewhere [50]
!
~
∂A i ∂2A ~ 1 ~
∂A 2~
+ β 00 2 − β 000 3 = iγ 0 ~ + i ∂|A| A
|A2 |A − αA , (2.21)
∂ξ 2 ∂η 6 ∂η ω0 ∂η

where α is loss and γ 0 = n2 nω0 /2 is the nonlinear coefficient. This equation

can be further simplified if

• higher-order nonlinear and dispersive effects as well as loss can be ne-

glected, which requires still narrower spectrum.

In this case Eq.(2.21) reduces to the nonlinear Schrödinger equation (NSE)

in the form
~
∂A i ∂2A ~  
~ .
+ β 00 2 = iγ 0 |A2 |A (2.22)
∂z 2 ∂η
This equation can be solved analytically by the inverse scattering problem
method.
 CHAPTER 2. THEORETICAL FUNDAMENTALS 29

λ (nm)
1600 800 450

(a)
|E(ω)| (a.u.)

(b)

2 4
ω (fs-1)
Figure 2.1: Pulse spectra calculated by different propagation equations.
Spectra of a 40-TW/cm2 , 15-fs sech-shaped pulse after propagating 0.5 mm
of a standard fiber calculated by the full Maxwell equation, obtained in [14]
are shown by the green curves in (a) and (b). For comparison, by red curves
are shown the solutions of FME (b) and reduced Maxwell equations (a).
In contrast to reduced Maxwell equation, FME shows good agreement with
exact calculation. The initial spectrum (scaled) is shown by the blue curves.
CHAPTER 2. THEORETICAL FUNDAMENTALS 30

The same procedure can be applied if the nonlinear interaction of waves

with different carrier frequencies is considered, for example, in a degenerate
four-wave mixing (FWM). Pump, signal and idler pulses with central fre-
quencies ωP , ωS , ωI = 2ωP − ωS interact through the Kerr nonlinearity P~K .
In this case using SVEA the linearly polarized electric field is presented by
X
E(z, t) = 1/2 Aj (z, t) exp[i(kj z − ωj t)] + c.c. (2.23)
j

(j = S, P, I) and from Eq. (2.19) three propagation equations follow for

AP ,AS , and AI describing the process of FWM:

∂AP

= AP −α + iγP0 [|AP |2 + 2|AI |2 + 2|AS |2 ]
∂z
ωP n2 ∗ ∂ 2 AP
+ 2i AP AS AI exp(i∆k z) + iβ2 (ωP )
c ∂η 2
∂AS

= AS −α + iγS0 [|AS |2 + 2|AP |2 + 2|AI |2 ]
∂z  
ωS n2 ∗ 2 1 1 ∂AS ∂ 2 AS
+ i AI AP exp(−i∆k z) + − + iβ2 (ωS )
c vg,P vg,S ∂η ∂η 2
∂AI

= AI −α + iγI0 [|AI |2 + 2|AS |2 + 2|AP |2 ]
∂z  
ωI n2 ∗ 2 1 1 ∂AI
+ i AS AP exp(−i∆k z) + −
c vg,P vg,I ∂η
∂ 2 AI
+ iβ2 (ωI ) 2 . (2.24)
∂η

Here vg,j = c/ng,j are the group velocities of the corresponding waves, ∆k 0 =
β(ωS ) + β(ωI ) − 2β(ωP ) is the wavevector mismatch. The additional advan-
tage of the evolution equation (2.19) is that all spectral components that can
take part in a given process are included automatically in a single equation.
This is particularly favorable if different processes can affect the propagation.
The cost for this more general approach with a single evolution equation is
the necessarily higher resolution of the temporal grid. The temporal grid
CHAPTER 2. THEORETICAL FUNDAMENTALS 31

step dt should be much less than the period of radiation, while in the SVEA
the resolution is in the scale of the envelope. However, in contrast to FDTD
methods, the numerical step dz in space does not need to be much shorter
than the wavelength, and in fact can be several mm. This explains the ad-
vantage of FME over FDTD methods, where dz cannot be made larger than
cdt.
The differencies between the different propagation equations, including
the propagation equation in fibers which will be derived later, are summarized
in the Table 2.1.
CHAPTER 2. THEORETICAL FUNDAMENTALS 32

Propagation Prerequisites Advantages Disadvantages

equation
Wave equa-  is piecewise Most general Requires a tremen-
tion (2.5) constant dous numerical effort
Forward dn/dz  λ, Can be effectively
Maxwell forward prop- solved numerically
equation agation
(2.19)
FME in fibers Weak selffo- 1+1-dimensional Limits maximum
(2.66) cusing power
Reduced n(ω) − 1  1 None in comparison Not valid for opti-
Maxwell with FME cally dense (n − 1 ∼
equation 1) media
(2.20)
Generalized Narrow spec- Usual equation with Not applicable for
nonlinear trum, SVEA SVEA broad spectra
Schrödinger
equation
(2.21)
Nonlinear Still narrower Can be solved ana- Does not include
Schrödinger spectrum lytically higher-order effects
equation
(2.22)

Table 2.1: The comparison of different propagation equations. Note that

requirements for more general equations apply also for less general, e.g. re-
duced Maxwell equation is valid only when the prerequisites of the wave
equation are fulfilled.
CHAPTER 2. THEORETICAL FUNDAMENTALS 33

2.1.1 Dispersion and nonlinear polarization

Up to this point, the specific form of the polarization P~ as a function of E

~

as well as the physical considerations determining it, have not been speci-
fied. We start from the consideration of the linear polarization. Under the
influence of the electric field, the dipole momentum is induced in the medium
due to the redistribution of charged particles. The general form for the linear
term in the expression for the polarization can be written as
Z ∞
P~L (t) = 0 ~ − τ )dτ.
χ̂(τ )E(t (2.25)
0

Here we neglect effects of optical activity as well as other non-local effects

(dependence of the polarization on the electric field in adjacent points). The
quantity χ̂(τ ) represents a delayed response of a medium to the electric field.
Generally it is a tensor, but since optical fibers are homogeneous, it reduces to
a scalar in our case. The previous convolution expression can be reformulated
more conveniently in the frequency domain:

P~L (ω) = 0 χ(ω)E(ω). (2.26)

For the linear propagation problems, the susceptibility χ is related to the

refractive index n by n2 − 1 = χ. It can be real or complex, with the
imaginary part describing the losses.
The physical origin of the delayed response of the medium can be under-
stood by means of a classical model of electrons moving under an action of
an external field. The binding forces of atoms are described as a potential
ωj2 x2j /2, where ωj characterises the j’th group of the electrons which consist
of fj particles per unit volume and x~j is the displacement for the j’th group.
CHAPTER 2. THEORETICAL FUNDAMENTALS 34

The polarization is given by (see, for example, [51])

X
P~ = −e fj x~j . (2.27)
j

The equation of electron motion is written as

d2 x~j dx~j ~
Ee
2
+ g j + ω x
~
j j = − , (2.28)
dt2 dt m

Here e and m are the electron charge and mass, respectively, and the effects
of the magnetic field are neglected. The effects of loss are also introduced
through gj . The solution for the harmonic driving field E0 exp(−iω0 t) can
be easily found, which yields

eE~
~xj = − (2.29)
m[ωj2 − ω02 − igj ω0 ]

and substituting into (2.27) results in

e2 X fj
χ(ω) = 2 2
. (2.30)
m0 j ωj − ω0 − igj ω0

Thus the susceptibility is determined as a sum of the contributions from sev-

eral absorption lines. The dampings gj determine the widths, and frequencies
ωj the positions of these lines, respectively. For dense media, the effect of the
induced polarization on the driving field should be taken into account, which
~ in the driving field by E
can be done by replacing E ~ + P~ /(30 ) (so-called

Lorentz-Lorentz model). The exact quantum model yields an expression

identical to (2.30), but with the values of fj , ωj and gj calculated from the
wavefunctions of the electronic states. The frequencies ω j are replaced by
the energy level distances (E1 − E2 )/h̄, and the square of the corresponding
dipole moments appear in fj . Nevertheless the basic form of the expression
(2.30) remains, and is used for practical calculations of the dispersion. The
CHAPTER 2. THEORETICAL FUNDAMENTALS 35

frequency dependence in wide spectral range typically can be described by

two or three so-called Sellmeyer terms, coefficients for which can be found in
tables (e.g. [52]). The frequency dependence of χ(ω) = n 2 (ω) − 1 determines
the group velocity vg (ω) = c/[d(ωn(ω))/dω] as well as the dispersion regime.
In the normal-dispersion regime, higher-frequency components travel slower
than lower-frequency ones (dvg /dω < 0), the opposite occurs in the anoma-
lous dispersion regime (dvg /dω > 0). For example, bulk fused silica has
normal dispersion for the wavelength shorter than 1.27 µm and anomalous
dispersion for longer wavelengths.
The expression (2.25) is only the first term in the expansion of P~ in powers
~ because generally the dependence P~ (E)
of E, ~ is not linear. Rather, it can

be represented in the form of series

Z ∞
P~L (t) = 0 ~ − τ )dτ +
χ̂1 (τ )E(t
Z0Z ∞
0 ~ − τ1 )E(t
χ̂2 (τ1 , τ2 )E(t ~ − τ2 )dτ1 dτ2 +
Z Z0Z ∞
0 ~ − τ1 )E(t
χ̂3 (τ1 , τ2 , τ3 )E(t ~ − τ2 )E(t
~ − τ3 )dτ1 dτ2 dτ3 +, . . .
0
(2.31)

where χn is the n-th order susceptibility and we completely exclude nonlo-

cal effects. As in the case of linear susceptibility, transformation into the
frequency domain simplifies the expression to

P~ (ω) = χ̂1 (ω)E(ω)

~ ~ i )E(ω
+ χ̂2 (ω1 , ω = ωi + ωj ) : E(ω ~ i) +

~ i )E(ω
χ̂3 (ω = ωi + ωj + ωk ) : E(ω ~ i )E(ω
~ i) + . . . . (2.32)

Here the susceptibility of n-th order is a tensor of n+1’th order, and the : sign
expresses corresponding index summation. The relations in the argument of
CHAPTER 2. THEORETICAL FUNDAMENTALS 36

χ underline the fact that the sum of frequencies of the photons created and
destroyed in the nonlinear process must be the same as a consequence of the
energy conservation law.
While a full calculation of the nonlinear susceptibilities is possible only by
a quantum model, the origins of the nonlinearity can be understood through
the model of an anharmonic oscillator. Anharmonicity a j x3j /3 is added to
the binding potential of j-th group of atoms, so that equation of motion now
looks like [53]
d2 x~j dx~j ~
Ee
2 2
+ g j + ω j x
~ j + a j x j = − . (2.33)
dt2 dt m
To solve it, a perturbative method can be used for not too strong fields. The
solution of unperturbed equation (2.28) as given in Eq.(2.30) is substituted
in the anharmonical term in (2.33).
The resulting equation can readily be solved. For example, for the com-
ponent χ2 (ω1 + ω2 ) one gets
X −2aj (e/m)2
χ2 (ω1 , ω2 ) =
j
(ωj2 − ω22 − iω2 gj )(ωj2 − ω12 − iω1 gj )
1
× . (2.34)
ωj2 − (ω1 + ω2 )2 − i(ω1 + ω2 )gj

In fact, second-order nonlinearity is absent in fibers due to the symmetry

χ2 = 0. However third-order susceptibilities can be derived in a similar way.
This simple classical model illustrates the mechanism of nonlinearity,
which is most important in the scope of this work, namely Kerr nonlin-
earity. In the quantum consideration, it results from the distortion of the
electron wavefunctions in the field. There exist several other mechanisms of
nonlinearity, such as Raman effect, molecular reorientation, electrostriction,
heating, and so on. Among them, only a Raman effect has sufficiently small
CHAPTER 2. THEORETICAL FUNDAMENTALS 37

response time to play any role in the processes on the femtosecond time scale
considered here. In this effect, incident light is scattered by a molecule in
a certain initial vibrational state. The molecule is excited to a high-energy
electronic state and immediately reemits the photon. However, the final vi-
brational state of the molecule is different from the initial one, therefore the
energy of the photon changes. The stimulated Raman scattering is a third-
order nonlinear process which is responsible for about 1/5 of the nonlinear
response in the silica. It is known that it introduces a shift of the soliton
frequency (see e.g. [54]). However, our numerical simulations show that the
Raman effect does not play a significant role in the processes of SC generation
in PCFs and hollow fibers, therefore it was not included into considerations.
The nonlinear polarization PN L far off the medium resonance for not too
high intensity in a isotropic medium can be treated as a third-order process
and is dominated by the Kerr nonlinearity. Sheik-Bahae et al. [55] have
developed a model for the dispersion of the dominant electronic part of χ 3
for semiconductors and wide-gap optical solids. It gives a universal formula
for χ3 (ω1 , ω2 , ω3 , ω4 ) in quite good agreement with measurements. According
to this model for
Eg
ωj  ω g = , (2.35)
h̄
where j = 1..4 and Eg is the bandgap energy, up to the first correction term
χ3 (ω, ω, −ω, ω) has the form
 
(0) ω2
χ3 = χ3 1 + 2.8 2 + . . . . (2.36)
ωg

For fused silica it holds that Eg = 9 eV and therefore the correction terms
do not play a significant role even for the extremely broad spectra considered
here. With this estimation the nonlinear polarization is instantaneous, i.e.
CHAPTER 2. THEORETICAL FUNDAMENTALS 38

(0)
PN L (z, t) depends only on the E(z, t) as P~N L (z, t) = 0 χ3 E 2 (z, t)E(z,
~ t).

2.1.2 Self-phase modulation

The physical origin of the effect of self-phase modulation is connected with

the distortion of the electronic distribution of the media leading to a refractive
index change by the electric field n = n0 + n2 I. It was first observed in
1967 [56] in gases, and later also in solids [57] and optical fibers [58]. The
description of this process, which is presented here following Ref. [50], can
be done simply in the case when the effects of the linear dispersion can be
ignored; this requirement is expressed as LD  LN L , where LN L = 1/(γ 0 A20 )
and LD = T02 /β 00 are characteristic linear and nonlinear lengths, respectively,
and T0 = τ0 /1.76 is related to the initial full-width half-maximum (FWHM)
pulse duration τ0 . Then the equation (2.21) for the field can be written as

∂U i
= exp(−2αz)|U |2 U (2.37)
∂z LN L

where the variable U = A/A0 is the envelope of the field normalized by its
initial value, and α is the field loss. The solution of this equation is [50]

U (z, t) = U (0, t) exp[iφN L (z, t)] (2.38)

with
zef f
φN L (z, t) = |U (0, t)|2 . (2.39)
LN L
Here zef f is given by
1
zef f = [1 − exp(−2αz)] (2.40)
α
and is smaller than z thus indicating that the loss limits the SPM. As a re-
sult, a time-dependent phase φN L is induced which implies a time-depending
CHAPTER 2. THEORETICAL FUNDAMENTALS 39

instantaneous frequency shift

∂φN L ∂|U (0, t)|2 zef f
δω(t) = − =− . (2.41)
∂t ∂t LN L
The chirp is caused by SPM and increases in magnitude with the propagation.
This chirp leads to negative δω on the leading front of the pulse and positive
δω on the trailing edge. Over a large central region, the chirp is almost linear
and positive. Consequently, new frequencies are generated on both sides on
the input frequency. These SPM-induced components broaden the spectrum
over its input width at the input. An estimate of the magnitude of this
broadening can be done by maximizing δω given by the above equation. The
result for U (0, t) = 1/cosh(t/T0 ) is
φmax
δωmax = 0.86 , (2.42)
T0
where φmax is the maximum value of the phase shift obtained for T = 0,
which is given by
zef f
φmax = . (2.43)
LN L
The resulting spectra have an oscillatory behavior with the number of oscil-
lation approximately equal to φmax /π. The reason for this behavior is that
the same frequency occurs for two values of t in the expression (2.41) for
the frequency shift. Depending on the phase between these components, the
interference between them can be constructive or destructive, thus creating
the oscillatory behavior.
This consideration is valid only if the group velocity dispersion can be
neglected, expressed by the condition LD >> LN L . In the normal disper-
sion regime, pulse broadening happens faster than in the case when GVD
is absent. This is due to the fact that SPM generates new frequency com-
ponents, which are red-shifted near the leading edge and blue-shifted near
CHAPTER 2. THEORETICAL FUNDAMENTALS 40

the trailing edge. Red components move faster than the blue components in
the normal-dispersion regime. Therefore SPM leads to a larger rate of pulse
broadening compared with that expected from GVD alone. The temporal
shape for this case becomes characteristically rectangular-like. The reason
is that any peak in the temporal shape results in the chirp, and newly cre-
ated spectral components are pulled apart by the normal dispersion, thus
leading to smoothing of the initial peak. Quick elongation of the pulse leads
to lowering of the peak intensity and less effective SPM. The ratio of pulse
durations with and without GVD, which is reverse of the ratio of intensities
for unchanged pulse shape, is given by [59]
   2 1/2
τ √ z 4 2 z
= 1 + 2φmax + 1 + √ φmax . (2.44)
τ0 LD 3 3 L2D

Thus for strong normal GVD the spectral broadening becomes less effective
due to decreasing peak intensity.
p
The effects of anomalous GVD are different. For N = LD /LN L ∼ 1 the
spectral broadening is stopped by GVD. This happens as soon as the pulse
reaches a duration which corresponds to that of a fundamental soliton. For a
fundamental soliton, effects of GVD and SPM exactly compensate each other,
and no further broadening occurs. For still larger values of N , higher-order
solitons can be created, as described in the next subsection.

2.1.3 Optical solitons in fibers

Solitons were first observed as steady waves on a water surface in the 19th
century, and since then many systems such as plasma, sound waves, and
optical waves have been shown to exhibit solitonic behavior. Optical solitons
were first observed in fibers by Mollenauer et al. [60].
CHAPTER 2. THEORETICAL FUNDAMENTALS 41

Let us consider the NSE (2.22), which is the propagation equation for
the electromagnetic field in the case when only Kerr–like nonlinearity and
second-order dispersion are included:
∂A i 00 ∂ 2 A
+ β = iγ 0 |A2 |A. (2.45)
∂z 2 ∂η 2
It can be transformed to the normalized form by the variable change (see
[50] for the summary used here)
A z η
U= ,ξ = ,τ = (2.46)
A0 Ld T0
where A0 and τ0 = 1.76T0 are the initial pulse amplitude and duration,
respectively (here sech-shaped pulses are assumed). Then the equation takes
the form
∂U 1 ∂2U
i = sgn(β 00 ) − N 2 |U 2 |U (2.47)
∂ξ 2 ∂τ 2
where the parameter N is given by
ω0 n2 nA20 T02 0 Ld
N2 = 00
= . (2.48)
2β LN L
For β 00 < 0 and N > 0.5, solitons can form from the initial pulse, with
the number of solitons for the sech-shaped pulse being the integer closest
to N . They are a result of the balance between linear dispersion, which
introduces negative (anomalous) chirp, and self-phase modulation, which in-
troduces positive chirp.
This equation can be solved by the inverse scattering method [61, 50],
which allows to analytically find the solution for (2.47) with any localized
initial condition U(0, τ ). The scattering problem associated with Eq. (2.47)
written for u = N U is
∂ν1
+ iζν1 = uν2 (2.49)
∂τ
∂ν2
+ iζν2 = uν1 (2.50)
∂τ
CHAPTER 2. THEORETICAL FUNDAMENTALS 42

for the amplitudes ν1,2 of the waves scattered on the potential u(ξ, τ ), and ζ
is the eigenvalue. First, initial scattering data in the form of the continuous
spectrum r(ζ) and N poles (bound states) with eigenvalues ζ j and residues
cj are obtained from the initial field distribution u(0, τ ). Then, the evolution
with ξ of the scattering data r(ζ); ζj ; cj is determined by simple algebraic
relations found elsewhere [50, 62]. And finally, the solution u(ξ, τ ) is recon-
structed from the scattering data at the point ξ. This last step is especially
simple if r = 0. Then the scattering data in the form of N constant eigen-
values ζj and the residues cj correspond to the N solitons, and the field is
obtained in the form
N
X
u(ξ, τ ) = −2 λ∗j ψ2j
∗
(2.51)
j=1
√
where λj = cj exp(iζj τ + iζj2 ξ) and ψ2j are obtained from the linear set of
equations
Xn
λj λ∗k ∗
ψ1j + ψ =0
∗ 2k
(2.52)
ζ j − ζk
k=1
Xn
λj λ∗k ∗
ψ1j + ψ = 0.
∗ 2k
(2.53)
ζ j − ζk
k=1

Each term in the sum for u is what is called a constituent soliton, a part
of radiation determined by the corresponding eigenvalue ζ j . Generally, N
solitons have different Re(ζj ) and therefore different central frequencies and
different velocities. In the collisions, such solitons preserve their form and
amplitude, and experience only a temporal shift. Far from the collision re-
gion, each of the constituent solitons is the simplest form of a soliton, which
is called a fundamental soliton (N = 1); the field in this case is described by

U (ξ, τ ) = sech(τ ) exp(iξ/2). (2.54)

CHAPTER 2. THEORETICAL FUNDAMENTALS 43

In contrast, if all the central frequencies of the constituent solitons are the
same, a bound higher-order soliton is formed, which is characterized by peri-
odic (with period πLD /2) changes of shape during propagation. The simplest
example for a higher-order soliton for N = 2 is given by

2[cosh(3τ ) + 3 exp(4iξ)cosh(τ )] exp(iξ/2)

U (ξ, τ ) = . (2.55)
cosh(4τ ) + 4cosh(2τ ) + 3 cos(4ξ)

The situation is more complicated if other (higher-order) terms are included

in Eq. (2.47) as e.g. in Eq. (2.21). If third-order dispersion (TOD) or the
Raman effect are taken into account, the stability of the higher-order soli-
ton is broken. It splits into constituent fundamental solitons with different
eigenvalues ζ, which have different frequencies and group velocities. For the
propagation of a pulse at the zero-dispersion wavelength, the initial spectrum
of the pulse splits into the peak in the anomalous region, which is fundamen-
tal soliton, and the peak in the normal region, which constitutes dispersive
non-solitonic radiation [63]. The energy contained in both pulses can be ap-
proximately of the same magnitude. If the initial frequency is far from the
zero-dispersion wavelength, the third-order dispersion can be considered as
a perturbation [64]. The perturbed NSE is written in the form

∂u ∂ 2 u ∂3u
i − 2 − 2u|u|2 = i 3 (2.56)
∂ξ ∂τ ∂τ

which can always be achieved by the corresponding renormalization of vari-

ables. Here  = β 000 /(3τ0 β 00 ) is a dimensionless parameter which describes the
relative impact of third order dispersion. The fundamental soliton solution
(2.54) is modified by the presence of TOD by a small change in the pulse
parameters and the emission of a low-level background radiation field. In
the first order in , only the velocity of the soliton reduces by the amount
CHAPTER 2. THEORETICAL FUNDAMENTALS 44

equal . All other parameters are unchanged in the first order; to determine
them, it is necessary to use the numerical result of Refs. [63, 65]. It is shown
there that the frequency at which the soliton is stabilized differs from the
input frequency by a value determined from the equation A c < 0.04. Here
A and  are the dimensionless amplitude and TOD-factor taken at the final
soliton frequency. If this condition is not initially fulfilled, the soliton emits
non-solitonic radiation and shifts away from the zero-dispersion point until
the stability is reached. To describe the properties of the non-solitonic radi-
ation, a change of variables is introduced on the basis of the fact that Eq.
(2.56) has several conserving quantities, or invariants (for details see [66, 67]).
The new independent function f (ξ, η) is ”associated” with the perturbation
δu = u − u0 of the fundamental-soliton solution u0 , as given by (2.54) [64]:

∂2f ∂f
δu = − 2 + 2tanh(τ ) − tanh2 (τ )f + u20 f ∗ . (2.57)
∂τ ∂τ

This new field satisfies the linear evolution equation

∂f ∂2f ∂f i ∂u0
i = 2
+ i − (2.58)
∂ξ ∂τ ∂τ 2 ∂τ

and its solutions can be found by Fourier-transforming the previous equation

in the form
Z

 ∞
ωsech 12 πω
f (ξ, τ ) = − {exp[i(ω 2 −(ω 3 −ω))]−exp(−iξ)} exp(iωτ )dω.
4 −∞ D(ω)
(2.59)
The dispersion relation D(ω) = ω 2 + 1 − (ω 3 − ω) = 0 determines the spec-
tral position of the generated radiation. This condition is equivalent to the
phase-matching condition ks = k(ωr ), where ks = ∂β/∂ω(ωs ) − (1/2)n2 Iωs/c
is the wavenumber of the fundamental soliton at frequency ω s and ωr is the
frequency of the non-solitonic radiation. The latter condition is also valid if
CHAPTER 2. THEORETICAL FUNDAMENTALS 45

higher (4th and larger) orders of linear dispersion need to be taken into ac-
count, as it is the case for PCF’s. In more detail, this condition is considered
in Chapter 3, as well as a case of higher-order input soliton.

2.2 Propagation equation in fibers

In a fiber, radiation generally can propagate in several transverse modes. In

the absence of birefringence, the energy transfer between the modes can be
caused by nonlinear effects. In a single-mode fiber for a power below the
critical power of selffocusing Pcr = λ20 /(πn2 ) the energy transfer to higher
transverse modes is weak [27]. In Chapter 5 the this limitation on the max-
imum intensity will be considered in more detail. The electric field in the
frequency domain can be separated [50] into the form:

E(x, y, z, ω) = F(x, y, ω)Ẽ(z, ω), (2.60)

where the transverse fundamental mode distribution F(x, y, ω) is the solution

of the Helmholtz equation

∆⊥ F + k(ω)2F = β(ω)2F (2.61)

for a PCF with the eigenvalue β(ω). We substitute (2.60) into (2.5), take
account of (2.61), and assume that the dependence of F~ on ω is much weaker
than that of Ẽ. In this way we obtain
∂ 2 Ẽ(z, ω)
F~ (x, y, ω) 2
+ β 2 (ω)F~ (x, y, ω)Ẽ(z, ω) =
∂z
−µ0 ω F (x, y, ω)F~ (x, y, ω)PN L(z, ω)
2 2
(2.62)

where PN L (ξ, η) = χ3 0 Ẽ 3 (ξ, η). This equation is multiplied by F~ , the inte-

gration is taken over the cross-section, and the result is divided by the mode
CHAPTER 2. THEORETICAL FUNDAMENTALS 46

area
Z
S1 (ω) = F 2 (x, y, ω)dS. (2.63)
S

The equation obtained in this way is

∂ 2 Ẽ(z, ω)

2
+ β 2 (ω) + W (ω)µ0 ω 2 BN L (z, ω) Ẽ(z, ω) = 0, (2.64)
∂z

where PN L (z, t) = χ3 0 Ẽ 3 (z, t) and

R
S
F 4 (x, y, ω)dS
W (ω) = (2.65)
S1

is a nonlinear reduction factor. We can now follow the same argumentation

which allowed to obtain (2.19) from (2.5). With the introduction of the
moving time coordinates ξ = z, η = t − zng /c, the longitudinal distribution
Ẽ(ξ, ω) satisfies the equation

∂ Ẽ(ξ, ω) [n(ω) − ng ]ω ω
=i Ẽ(ξ, ω) + iµ0 c W (ω)PN L (ξ, ω). (2.66)
∂ξ c 2n(ω)

Here PN L (ξ, η) = χ3 0 Ẽ 3 (ξ, η). This equation can be used to model the
propagation of ultrashort light pulses in fibers, and has the same advantages
as the FME equation (2.19). The additional requirement that power be much
less than the power of selffocusing makes it possible to reduce the dimension
of the problem, i.e. the unknown function Ẽ depends on only one spatial
coordinate in (2.66).

2.2.1 Waveguiding model for a step-index fiber

The description of a step-index fiber is done in a standard way found else-

where [68, 69]. Let us consider a step-index fiber, which consists of a core
with radius a and refractive index n0 surrounded by unlimited cladding with
refractive index n1 , n0 > n1 . The problem is to determine the wavenumber
CHAPTER 2. THEORETICAL FUNDAMENTALS 47

β for a mode with frequency ω. The temporal and longitudinal dependence

of the electric and magnetic fields of the mode is given by exp(iωt − iβz).
The Helmholz equation for each of these fields is given by [68, 69]

∂2 1 ∂ 1 ∂2
+ + + n20,1 k02 − β 2 = 0. (2.67)
∂ρ2 ρ ∂ρ ρ2 ∂φ2

Here k0 = ωc, φ is the azimutal coordinate of a cylindrical coordinate system

with the axis directed along the axis of the fiber; the dependence of the field
in this direction is fixed for each mode and is given by exp(ilφ) where l is
the azimuthal order of the mode. This dependence follows from the fact
that any field should be a periodic function of φ. It is useful to introduce
p p p
denotations V = n20 − n21 ak0 , u = a n20 k02 − β 2 , and v = a β 2 − n21 k02 ,
where k0 = ω/c. The Helmholtz equation then transforms to a differential
equation in the ρ direction
 
d2 1 d 2 2 l2
+ + a u − 2 =0 (2.68)
dρ2 ρ dρ ρ

in the core (ρ < a) and

 
d2 1 d 2 2 l2
+ + −a v − 2 = 0 (2.69)
dρ2 ρ dρ ρ

in the cladding (ρ > a). The solutions are given by Bessel functions. If we
account for boundary conditions (all fields should be zero at infinitely large
radius and have finite values at ρ = 0), then the fields inside (outside) the
CHAPTER 2. THEORETICAL FUNDAMENTALS 48

core are given by (modified) Bessel functions of first (second) type:

Ez1 = AE Jl (uρ/a) cos lφ (2.70)

Hz1 = AH Jl (uρ/a) sin lφ (2.71)

a la
Eρ1 = −j [βAE Jl0 (uρ/a) + ωµ0 AH Jl (uρ/a)] cos lφ (2.72)
u uρ
a la
Hρ1 = −j [βAH Jl0 (uρ/a) + ω0 n20 AE Jl (uρ/a)] cos lφ (2.73)
u uρ
a lβa
Eφ1 = j [ AE Jl (uρ/a) + ωµ0 AH Jl0 (uρ/a)] sin lφ (2.74)
u uρ
a lβa
Hφ1 = −j [ AH Jl (uρ/a) + ωn20 0 AH Jl0 (uρ/a)] sin lφ (2.75)
u uρ

in the core and

Ez2 = DE Kl (vρ/a) cos lφ (2.76)

Hz2 = DH Kl (vρ/a) sin lφ (2.77)

a la
Eρ2 = −j [βDE Kl0 (vρ/a) + ωµ0 DH Kl (vρ/a)] cos lφ (2.78)
v vρ
a la
Hρ2 = −j [βDH Kl0 (vρ/a) + ω0 n21 DE Kl (vρ/a)] cos lφ (2.79)
v vρ
a lβa
Eφ2 = j [ DE Kl (vρ/a) + ωµ0 DH Kl0 (vρ/a)] sin lφ (2.80)
v vρ
a lβa
Hφ2 = −j [ DH Jl (vρ/a) + ωn20 0 DH Kl0 (vρ/a)] sin lφ (2.81)
v vρ

in the cladding. The constants A and D have to be chosen so that the fields
satisfy boundary conditions at ρ = a. In the general case of large refractive
index difference, the boundary condition yields a system of homogeneous
linear equations for A and D. Solution of this system exists if the determinant
is equal to zero, which yields the following dispersion relation for the step-
index fibers in the general case:
2
 2
2β 1 1
l 2 2
+ 2 = (Y1 n20 + X1 n21 )(Y1 + X1 ) (2.82)
k0 u v
CHAPTER 2. THEORETICAL FUNDAMENTALS 49

where Xl = Kl0 (v)/(vKl (v)) and Y1 = Jm

0
(u)/(uJm (u)). This equation is to
be solved together with u2 +v 2 = V 2 to obtain unknown variables u and v and
thus β as function of ω. The detailed analysis of this equation, which is out
of the scope of this work, shows that there always exists one solution, which
corresponds to the fundamental mode EH11 . For V larger than the threshold
value of approximately 2.405, other modes can be excited. In this work, only
radiation in the fundamental mode is considered. The simplification of the
dispersion equation is possible for weakly guiding fibers with n 1 ≈ n0 . In
this case for the fundamental mode

uJ1 (u) vK1 (v)

= . (2.83)
J0 (u) K0 (v)

However, for all problems considered here the full equation (2.82) was used
because of a large step between the refractive indexes of core and cladding
in the photonic and tapered fibers.

2.2.2 Effective cladding model for PCF’s

For the tapered and photonic fibers, no closed analytical expression for the
transverse mode distribution exists. There are two common approaches to
the calculation of the dispersive properties of the PCF’s: the full vectorial
model [70] and the approximate effective-cladding model. While for photonic
band gap guiding only the full model can give correct results, for the guiding
by the full internal reflection (as in the standard fibers) the effective-cladding
model [2] gives satisfactory results which coincide with the experimental ones.
It was recently shown that in this case the periodic structure of the PCF does
not play a decisive role in the waveguiding properties [71]. The abovemen-
tioned method to describe the effective cladding is used in this work, as
CHAPTER 2. THEORETICAL FUNDAMENTALS 50

shown in this subsection. To study pulse propagation in PCF’s besides the

evolution equation presented in Section 2.2 we need a model for the PCF
waveguide contribution to dispersion, which is calculated by the following
two-step procedure illustrated in Fig. 2.2. First, an infinite photonic crystal

Figure 2.2: Effective-cladding model for calculation of dispersion in PCF’s

and elementary cell transformation. The hexagonal cell of the effective
cladding with refractive index nef f with a continuity boundary condition
is transformed into circular cell.

fiber without central defect (core) is considered. The fundamental space-

filling mode in such a system can be analyzed in the scalar approximation,
i.e. the transverse dependence of all components of both electric and mag-
netic fields are given by the same function F (ω, x, y). This function satisfies
the Helmholtz equation. Besides that, the fundamental mode has the same
periodical properties as the photonic crystal. We split the photonic crystal
into the elementary hexagonal cells so that the air holes are in the center of
the cells. From the periodicity of F (ω, x, y) and from the condition that the
derivative of F is continuous, it follows that the derivative ∂F/∂~n vanishes at
the cell boundary, where ~n is the vector normal to the boundary. To further
simplify the calculations, we consider a circular cell instead of the hexagonal
CHAPTER 2. THEORETICAL FUNDAMENTALS 51

one. The radius of the inner air cylinder in the circular cell equals d = 2r,
the outer radius is defined by the air fraction in the original crystal as
q√
R=Λ 3/(2π). (2.84)

Now we consider the function F (ω, x, y) in the polar coordinates as F (ω, ρ, ϕ).
The condition
∂F
=0 (2.85)
∂~n
transforms to 
∂F 
= 0, (2.86)
∂ρ R
and the continuity conditions on the inner air-silica interface have to be
satisfied as well. In the inner (0 < ρ < r) area, it holds that

β 2 (ω) − n2Air (ω)ω 2 /c2 = κ2 (ω) > 0 (2.87)

and therefore F ∼ I0 (κ(ω)ρ), where I is the modified Bessel functions of first

type. In the outer area (r < ρ < R) we write

β 2 (ω) − n2Silica (ω)ω 2/c2 = −γ 2 (ω) < 0 (2.88)

and F is equal to the linear combination of J 0 and Bessel function of second

order N0 with coefficients chosen so that the boundary condition at R is
satisfied:
J1 (γ(ω)R)
F (ω, ρ) ∼ J0 (γ(ω)ρ) − N0 (γ(ω)ρ) . (2.89)
N1 (γ(ω)R)
Substituting F (ω, ρ) into the boundary conditions we obtain the dispersion
equation
 
I1 (κ(ω)r) J1 (γ(ω)R)
κ(ω) J0 (γ(ω)r) − N0 (γ(ω)r) =
I0 (κ(ω)r) N1(γ(ω)R)

 
J1 (γ(ω)R)
−γ(ω) J1 (γ(ω)r) − N1 (γ(ω)r) . (2.90)
N1 (γ(ω)R)
CHAPTER 2. THEORETICAL FUNDAMENTALS 52

This equation is solved together with

κ2 (ω) + γ 2 (ω) = [n2Silica (ω) − n2Air (ω)]ω 2 /c2 (2.91)

to find subsequently κ(ω), β(ω) = (κ2 (ω)+ n2Air (ω)ω 2/c2 )1/2 and the effective
refractive index for the fundamental space-filling mode n ef f (ω) = β(ω)c/ω.
Then we consider the omitted hole as a core of a step-index fiber with
diameter 2r = 2Λ − d, where Λ is the center-to-center distance between
the holes (pitch) and d is the hole diameter, and the surrounding photonic
crystal as homogeneous cladding with refractive index n ef f (ω). Note that
for certain PCF’s the air holes are very large and therefore the central core
is supported by the very thin bridges of silica which can be neglected in the
calculation of β(ω) and such a fiber can be described as an isolated strand of
silica surrounded by air. The dispersion of tapered fiber, which is a µm-scale
silica core surrounded by air, can be described in the same way.

2.2.3 Dispersion in hollow fibers

For hollow fibers, the transverse structure and dispersive properties of the
modes can be found analytically, as it was done in Ref. [72] and repro-
duced here. These properties were derived for the conditions λ/a  1 and
√
ν 2 − 1  λ/a. Here a is the radius of the hollow fiber, ν is the ratio of
the refractive indices of the hollow fiber walls and the gas filling. These re-
quirements are satisfied with high accuracy for the typical values λ = 0.8
µm, a = 100 µm, ν=1.45 (for fused silica cladding), which allows sim-
plification of general equations. The expressions for the fields are derived
from the Helmoltz equation and from the continuity boundary conditions
Eφ (a − 0) = Eφ (a + 0), Ez (a − 0) = Ez (a + 0), Dz (a − 0) = Dz (a + 0) set
CHAPTER 2. THEORETICAL FUNDAMENTALS 53

at ρ = a. There exist three types of modes in such a system (as well as in a

step-index fiber): transverse electric, or TE0m (Ez = 0), transverse magnetic,
or TM0m (Hz = 0), and hybrid EHlm (both electric and magnetic field have
nonzero z component). The index l characterizes the azimuthal dependence
of the mode exp(ilφ), m is the order of the mode. The distribution of fields
in the cross-section plane for the case λ/a  1 is
r 
µ0 ρ
Eφ = − Hr = J1 u0m (2.92)
0 a
for TE0m modes,
r 
µ0 ρ
Er = Hφ = J1 u0m (2.93)
0 a
for TM0m modes, and
r 
µ0 ρ
Eφ = − Hr = Jn−1 U0m cos nφ (2.94)
0 a
r  ρ
µ0
Er = Hφ = Jn−1 0m sin nφ (2.95)
0 a
for hybrid EHlm modes. Here Ulm is the m’th root of the equation Jl−1 (Ulm ) =
0. The dispersion relation is given by
q p q
2 k02 − β 2 a
2
Jn−1 ( k0 − β a) = iΛx Jn ( k02 − β 2 a) (2.96)
k0
and for low-order modes, β can be explicitly expressed:
 
ωn(ω) 1
βlm = 1− 2 (2.97)
c 2γg
with the field loss determined by
<(Λx)
αlm = . (2.98)
Rg2 R
Here n(ω) is the refractive index of the fiber filling, and the parameters R g
and Λx are introduced by
2πR
Rg = (2.99)
Ulm λ
CHAPTER 2. THEORETICAL FUNDAMENTALS 54

and 

 √ 1 for TE0m modes

 ν 2 −1
2
Λx = √ν for TM0m modes . (2.100)

 ν 2 −1

 ν 2 +1
√
2 ν 2 −1
for EHlm modes
With propagation, only the mode with lowest loss will remain in the
waveguide. In this case it is the EH11 mode [73], for which the transverse
distribution of intensity is given by J0 (a1 ρ/a). Here r is the transverse coor-
dinate and a1 is the first zero of the Bessel function J1 . The contribution of
the waveguide to the dispersion in this case is given by [73]
 2  
1 2.405c ω 1 + ν2
β(ω) = k(ω) − −i √ . (2.101)
2 ωa c a ν2 − 1

The hollow waveguide is not truly a guiding system in the sense that it
possess no lossless modes. Therefore the loss, which is inversely proportional
to a3 , is introduced. Our calculations show that the values W = 0.567 and
S1 = 0.269πa2 do not depend on frequency in the case of the hollow fibers.

2.2.4 Nonlinear term for propagation in fibers

This procedure allows us to determine the dispersion of PCF or tapered fibers

as determined by the function β(ω). The dependence of the mode area S 1
and frequency can also be determined with this model. A typical case is
presented in Fig. 2.3. Although there exits a slight variation of W in the
infrared, we neglect this effect because dominant part radiation has shorter
wavelength in all spectra considered here. A similar physical picture can be
observed for other fibers considered in this work.
In the literature, for the case of the NSE the definition of envelope which
p
differs from (2.1) is often used. The envelope A∗ (ξ, η) = A(ξ, η) 0 ncS1 /2
CHAPTER 2. THEORETICAL FUNDAMENTALS 55

λ (nm)
1600 800 450

0.8
S1/πr ,W
2

0.4

2 4
ω (fs-1)
Figure 2.3: Nonlinear reduction factor W (solid) and mode area S 1 (dotted)
as functions of frequency for PCF with Λ=1.5 µm and d=1.3 µm.

is normalized so that A2∗ gives the power of the pulse. In this normalization,
the NSE (2.22) reads as

∂A~∗ ~∗
i 00 ∂ 2 A  
2 ~
+ β = iγ |A∗ |A∗ . (2.102)
∂z 2 ∂η 2
CHAPTER 2. THEORETICAL FUNDAMENTALS 56

where γ = ω0 n2 /(cAef f ) with

R 2
2
S
F (x, y, ω)dS S1 (ω)
Aef f (ω) = R 4 = . (2.103)
S
F (x, y, ω)dS W (ω)

The value Aef f is usually called effective mode area. This description gives
an equivalent way to express the nonlinear term, but the quantity S 1 is useful
because it relates power and peak intensity.

2.3 Numerical procedure

For the numerical simulation of Eq. (2.66), the split-step Fourier method was
used. The right-hand side of this equation can be separated into linear and
nonlinear parts. We then define operators ĜL (dz) and ĜN L (dz) for linear and
nonlinear dz-steps, each of which consist of a propagating field Ẽ(ξ, ω) over
dz by linear and nonlinear right-hand terms in (2.66), respectively. It can be
shown [50] that using a certain sequence of such operators one can build a
second-order (in dz) method for the simulation of the field propagation. For
a spatial grid with N+1 points this sequence reads as follows:

ĜL (dz/2)[ĜN L (dz)ĜL (dz)]N −1 ĜN L(dz)ĜL (dz/2). (2.104)

It is possible to construct fourth- and higher-order methods in this way [74],

which however were not implemented in this work, because they significantly
complicate the program without guaranteeing faster calculations.
Note that linear and nonlinear steps in this method are performed in the
different spaces: in the frequency and time space, respectively. For the linear
step, the corresponding equation is written in the frequency space as

∂ Ẽ(ξ, ω) [n(ω) − 1]ω

=i Ẽ(ξ, ω) (2.105)
∂ξ c
CHAPTER 2. THEORETICAL FUNDAMENTALS 57

and can be solved exactly:

 
[<n(ω) − 1]ωdz
ĜL (dz) = exp i exp(α(ω)dz)Ẽ(ξ, ω), (2.106)
c

where α(ω) is the loss. The nonlinear step is expressed by the following
equation in the time domain:

∂ Ẽ(ξ, t) W ∂ 3
= µ0 0 cχ3 Ẽ (ξ, t), (2.107)
∂ξ 2n ∂t

and we assume W (ω)/n(ω) ∼ const. To improve the accuracy of the method

it is necessary to make the nonlinear step by a method with higher order
than the split-step fourier method [75], therefore in this work a fourth-order
Runge-Kutta method was used for solving Eq.(2.107):

∂ Ẽ 3 (z, t)
∆E1 (t) = C (2.108)
∂t
∂(Ẽ(z, t) + ∆E1 (t)/2)3
∆E2 (t) = C (2.109)
∂t
∂(Ẽ(z, t) + ∆E2 (t)/2)3
∆E3 (t) = C (2.110)
∂t
∂(Ẽ(z, t) + ∆E3 (t))3
∆E4 (t) = C (2.111)
∂t
Ẽ(z + dz, t) = Ẽ(z, t) +
1
[∆E1 (t) + 2∆E1 (t) + 2∆E1 (t) + ∆E1 (t)] (2.112)
6

where C = dzW χ3 /(2cn) and ∆E1,2,3,4 (t) are the approximations of the
Runge-Kutta method.
The sequence (2.104) implies that the Fourier transform has to be per-
formed between the linear and nonlinear steps. However this does not signif-
icantly increase the calculation time due to the application of a fast Fourier
transform algorithm. A homogeneous rectangular grid was used for the nu-
merical simulations. The time step dt was chosen to be 0.0276 fs to accu-
CHAPTER 2. THEORETICAL FUNDAMENTALS 58

rately resolve the carrier oscillations even for the high-frequencies compo-
nents which have a period around 1 fs. The number of time points was
2×105 in most calculations, which yields the time window T w of 5.52 ps.
The corresponding resolution in the frequency domain is 0.0227 fs −1 . The
spatial step dz was chosen around 1 µm. The time differential operator in
(2.112) is implemented by the central second-order difference ∂ Ẽ 3 (z, t)/∂t =
(Ẽ 3 (z, t + dt) − Ẽ 3 (z, t − dt))/(2dt). The numerical error introduced by this
algorithm was estimated by repeating the calculations with the halved spa-
tial step; the comparison of the result shows that the error does not exceed
1%.
The initial condition E(0, t) = E0 cos(ω0 t)/cosh(1.76t/τF W HM ) is taken
at ξ = 0. The question of nonreflecting boundary conditions, which have to
be imposed for t = −Tw /2 and t = +Tw /2, is important in many numeri-
cal methods involving modeling of propagation [76, 77]. Without boundary
conditions, the radiation would reflect from the planes t = ±T w /2 and thus
create numerical artifacts. In the case considered here, instead of using non-
reflecting boundary conditions for the extreme points in the temporal grid,
the grid was looped to achieve a stable and simple way to account for waves
which move too slowly or too quickly with respect to the main radiation and
go beyond the time window. Such waves will simply appear on the other
side of the domain. Looping is done automatically in frequency-space steps,
and the corresponding modification of the central difference allows looping
also for the nonlinear step. Of course one has to ensure that waves do not
overlap after making the round-trip over the domain. This method has the
additional advantage that it does not require preliminary knowledge of the
velocities of the waves.
CHAPTER 2. THEORETICAL FUNDAMENTALS 59

The numerical simulations were performed on a workstations with an

Alpha CPU at 667 MHz. The typical computation time was around 24
hours.