PHYSICAL REVIEW E 67, 046608 共2003兲

Analysis of mode structure in hollow dielectric waveguide fibers

Mihai Ibanescu, Steven G. Johnson, Marin Soljačić, J. D. Joannopoulos, and Yoel Fink
Center for Materials Science and Engineering and Research Laboratory of Electronics, Massachusetts Institute of Technology,
77 Massachusetts Avenue, Cambridge, Massachusetts 02139

Ori Weisberg, Torkel D. Engeness, Steven A. Jacobs, and M. Skorobogatiy

OmniGuide Communications, One Kendall Square, Building 100 #3, Cambridge, Massachusetts 02139
共Received 10 December 2001; revised manuscript received 22 April 2002; published 17 April 2003兲
In this paper, we analyze the electromagnetic mode structure of an OmniGuide fiber—a hollow dielectric
waveguide in which light is confined by a large index-contrast omnidirectional dielectric mirror. In particular,
we find that the modes in an OmniGuide fiber are similar to those in a hollow metallic waveguide in their
symmetries, cutoff frequencies, and dispersion relations. We show that the differences can be predicted by a
model based on a single parameter—the phase shift upon reflection from the dielectric mirror. The analogy to
the metal waveguide extends to the transmission properties, resulting in the identification of the TE01 mode as
the lowest-loss mode of the OmniGuide fiber.

DOI: 10.1103/PhysRevE.67.046608 PACS number共s兲: [Link]

I. INTRODUCTION of the large penetration of the field into the low index-
contrast multilayer structure. Recently, with the discovery of
The advent of high-purity ultra-low-loss silica fiber as a the omnidirectional high-reflectivity properties of multilayer
transmission medium in the late 1970s provided a basis for films 关10–12兴, researchers have focused instead on high
the modern optical communication infrastructure. Although index-contrast dielectric waveguides 关4,13–18兴. Fink et al.
highly successful, silica waveguides have fundamental limi- fabricated a large-core high index-contrast hollow dielectric
tations in their attenuation and nonlinearities that result from waveguide and demonstrated that light is transmitted even
the interaction of light with a dense, material-filled core. A when tight bends are introduced in the waveguide, a result of
different approach to waveguiding circumvents these prob- the omnidirectionality of the dielectric mirror 关13兴. Xu et al.
lems by confining light in a hollow core using highly reflec- developed an efficient analytical method for calculating ap-
tive walls. This approach is exemplified by hollow metallic proximate dispersion relations of the modes of such
waveguides that are very efficient in the millimeter wave- waveguides, making use of the radial periodicity of the clad-
length range. Prior to the emergence of silica fiber, these ding layers 关14兴. Using this method, they calculated the num-
waveguides were seriously considered as candidate media ber of layers required to suppress radiation as a function of
for long-distance telecommunications 关1兴. An impairment of the index contrast and substantiated that a large index con-
metallic waveguides is that they become lossy at high fre- trast is necessary in order to limit the number of layers to a
quencies due to the finite conductivity of metals. Thus, their reasonable value 关15兴. Ibanescu et al. exploited the similari-
use is restricted to low frequencies. This severely limits the ties between a dielectric multilayer mirror and a metal mirror
ultimate bandwidth that they can transmit. By adding a di- and showed that it is possible to design an all-dielectric co-
electric coating on the inside of the metallic waveguide one axial waveguide that supports a TEM-like mode 关17兴.
can improve its properties. Such metallodielectric In this paper, we will study the properties of a hollow
waveguides have been developed for infrared wavelengths, dielectric waveguide in which light is confined by a large
in particular, for laser power delivery 关2,3兴. index-contrast omnidirectional dielectric mirror. The large
Here we analyze an OmniGuide fiber—a hollow all- index contrast produces a high degree of optical confinement
dielectric waveguide in which light is confined by an omni- in the core and results in a waveguide mode structure that is
directional mirror 关4兴. The limitations that exist in both the very similar to the mode structure of a hollow metallic wave-
silica fiber and the hollow metal waveguides can be system- guide. We present the similarities and differences between
atically reduced in this waveguide due to its hollow core and the hollow dielectric and metallic waveguides, and explain
the use of dielectric materials transparent at high frequencies. these results using a single-parameter model based on the
Yeh et al. showed in 1978 that it is possible to transmit a phase shift upon reflection from the dielectric mirror. In ad-
non-index-guided mode in a multilayer cylindrical dielectric dition, we show that the similarities between these two types
waveguide 关5兴. The radiative decay of the mode is sup- of waveguides extend to their transmission properties, in
pressed using a Bragg reflector, i.e., a multilayered structure which they have the same lowest-loss mode, the TE01 mode.
that is periodic in the radial direction. Initial work in this
area has been focused on structures based on doped-silica II. HOLLOW WAVEGUIDES
technology that have a low index contrast between the layers
关6 –9兴. These structures are useful, for example, to control The waveguides that we compare are shown in Fig. 1. On
the dispersion parameter of a fiber in ways that are not the left is depicted an OmniGuide fiber. A hollow core 共index
achievable in a standard silica fiber 关9兴. However, such struc- of refraction n 0 ⫽1) is surrounded by an omnidirectional
tures cannot confine light primarily in a hollow core because mirror that consists of alternating layers having high and low

1063-651X/2003/67共4兲/046608共8兲/$20.00 67 046608-1 ©2003 The American Physical Society

IBANESCU et al. PHYSICAL REVIEW E 67, 046608 共2003兲

FIG. 1. 共Color兲 共Left兲 OmniGuide fiber of radius R. Light is

confined in the hollow core by a multilayer dielectric mirror made
of alternating layers with high 共blue兲 and low 共green兲 indices of
refraction. 共Right兲 Hollow metallic waveguide of radius R. Light is
confined in the hollow core by a metallic cylinder.
FIG. 2. 共Color兲 共Left兲 Projected band structure associated with
the dielectric mirror. The blue regions correspond to ( ␤ , ␻ ) pairs for
indices of refraction 关10兴. The high and low indices of re-
which light can propagate through the mirror. The meaning of the
fraction layers are shown in blue and green, respectively. The parallel wave vector ␤ is shown in the inset. k T is the transverse
indices of refraction are 4.6 and 1.6 with corresponding component of the wave vector. The diagonal black line represents
thickness 0.33a and 0.67a, where a is the total thickness of the light line ( ␻ ⫽c ␤ ). Shown in gray are two frequency ranges
a pair of layers. These values for the indices of refraction corresponding to omnidirectional reflectivity of the mirror. 共Right兲
have been used in previous works 关4,10,13–15,17兴, including Frequency cutoffs of the lowest 11 modes supported by a hollow
a fabricated multilayered waveguide operating at 10.6 ␮ m. metallic waveguide are plotted versus the radius of the waveguide.
The large index contrast between the two types of layers TE polarized modes are shown in red, while TM modes are shown
allows us to better illustrate the comparison between the hol- in blue. The number m in the labels is the angular momentum index
low dielectric and metallic waveguides; the conclusions are of each mode. Note the degeneracy of the TE01 and the TM11
still valid for a smaller index contrast that one may have to modes. The two thick vertical dashed lines correspond to radii R
use in practice. In the frequency range of omnidirectionality ⫽2.0a and R⫽3.8a, which are the values used for the two exem-
of the dielectric mirror, the similarities between the two plary waveguides in this paper.
types of waveguides are enhanced because, in both cases, the
mirror that confines light in the core is a very good reflector thicknesses as those of the layers present in the waveguide. If
for all angles of incidence and polarizations. Moreover, the ␻ is the frequency of an incoming plane wave and ␤ is the
omnidirectionality of the dielectric mirror is important for component of the wave vector parallel to the mirror, then
practical considerations: in an imperfect waveguide, the cou- regions in the ( ␤ , ␻ ) plane shown in blue correspond to
pling between the operating core mode and cladding modes plane waves that can couple to propagating modes in the
above the light line is eliminated if the frequency of opera- mirror and are transmitted through the mirror. The two gray
tion lies in the omnidirectional frequency range of the mirror. areas in the plot correspond to the omnidirectional frequency
The exact value of black a in microns is determined by ranges of the dielectric mirror. For these frequencies, light
the wavelength of operation of the waveguide. For example, cannot be transmitted through the mirror for any angle of
if one wants to use ␭⫽1.55 ␮ m, then the thickness of a incidence of a plane wave incoming from the outside air
bilayer should be chosen somewhere in the range a region. Both the TE and TM polarizations are overlapped in
⫽0.3–0.4 ␮ m. The radius R of the waveguide will vary in the plot.
the different examples presented in this paper from a mini- In the right panel of Fig. 2, the frequency cutoffs of the
mum of 2a to a maximum of 20a. lowest 11 modes supported by a hollow metallic waveguide
On the right side of Fig. 1, we show a hollow metal wave- are plotted versus the radius of the waveguide. The disper-
guide. The core is the same as in the OmniGuide fiber 共index sion relation of a mode with cutoff frequency ␻ c is given by
of refraction 1, radius R), but a perfectly conducting metal ␻ 2 ⫽c 2 ␤ 2 ⫹ ␻ 2c . TE polarized modes are shown in red, while
cylinder now replaces the multilayer cladding. TM modes are shown in blue. The number m in the labels is
We expect the multilayer structure to confine light in the the angular momentum of each mode. Note the degeneracy
core of the waveguide in a frequency range that corresponds of the TE01 and the TM11 modes. The two vertical dashed
to the band gap of a planar multilayer mirror having the same lines correspond to radii R⫽2.0a and R⫽3.8a, which are
parameters as those of the multilayer structure in the wave- the values used for the OmniGuide fibers that we analyze.
guide. Indeed, it is precisely in this band-gap region that we
expect the mode structure of the OmniGuide fiber to re- III. CALCULATION OF MODES OF THE OMNIGUIDE
semble that of a hollow metal waveguide. FIBER
In the left panel of Fig. 2, we plot the projected band
structure for a planar multilayer dielectric mirror, for which We calculate the modes supported by the OmniGuide fi-
the layers have the same indices of refraction and the same ber using the transfer-matrix method. As in any waveguide

046608-2
ANALYSIS OF MODE STRUCTURE IN HOLLOW . . . PHYSICAL REVIEW E 67, 046608 共2003兲

with cylindrical symmetry, there are three useful conserved

quantities: ␻ , the frequency; ␤ , the wave vector in the axial
direction; and m, the angular momentum (m⫽0,1,2, . . . ).
The general form of the z component of the electric field,
E z , in the ith layer of the structure for a mode with wave
vector ␤ and angular momentum m is

E z 共 z,r, ␾ 兲 ⫽ 关 A i J m 共 k T,i r 兲 ⫹B i Y m 共 k T,i r 兲兴 e i( ␤ z⫺ ␻ t) e im ␾ ,

共1兲

where A i and B i are coefficients that vary from layer to layer,

J and Y are Bessel functions, and k T,i is the transverse wave
vector, k T,i ⫽ 冑(n i ␻ /c) 2 ⫺ ␤ 2 . H z is given by a similar ex-
pression, but with different coefficients C i and D i . From the
E z and H z components one can calculate all the other com-
ponents of the electromagnetic field components 关19兴. The
coefficients in the (i⫹1)th layer are related to those in the
ith layer by a 4⫻4 transfer matrix such that the boundary
conditions are satisfied 关5兴. The last step before finding the
modes is to establish boundary conditions at the outside sur- FIG. 3. 共Color兲 Modes supported by an OmniGuide fiber of
face of the outermost layer. Here we require that the incom- radius R⫽2.0a are shown with solid lines. Red is for TE and HE
ing radial electromagnetic flux be zero, which corresponds to modes, while blue is for TM and EH modes. Only the first three
the physical situation in which the sources of electromag- modes are labeled in each band gap. The lowest three modes in a
netic field are only inside the waveguide 关4兴. The solutions hollow metallic waveguide with the same radius are plotted as dots.
that we obtain with this boundary condition are guided
modes 共with real wave vectors兲 and leaky modes 共with com- between the core and the multilayer cladding. 共They can also
plex wave vectors兲. be found deeper in the multilayer structure if a defect is
The OmniGuide fiber supports modes that are analogous introduced by altering the thickness or the index of refraction
to the guided modes of a hollow metallic waveguide. These of one or more layers兲. In Fig. 2, these modes are situated
modes are resonant modes that have most of their energy below the light line in the band gap of the dielectric mirror.
traveling in the hollow core. Their field decays exponentially
in the radial direction in the dielectric layers. These modes IV. COMPARISON OF OMNIGUIDE FIBER MODES
have ( ␤ , ␻ ) pairs situated at the band gap of the dielectric AND METAL WAVEGUIDE MODES
mirror and above the light line. The wave vector ␤ has an
imaginary part that is proportional to the radiative loss of the In Fig. 3, we show the lowest-frequency resonant modes
modes. This imaginary part decreases exponentially with the of the OmniGuide fiber for a radius R⫽2.0a. TE modes and
number of layers in the dielectric mirror; in the limit of an TE-like 共HE兲 modes are shown as solid red lines, while TM
infinite number of layers these resonant modes become truly and TM-like 共EH兲 modes are shown as solid blue lines. The
guided modes. For the dielectric mirror with only five layers modes of a hollow metallic waveguide with the same radius
shown in Fig. 1, the imaginary part is of the order of 10⫺4 are shown as dots. The figure shows that, indeed, the disper-
⫻2 ␲ /a. This means that light can be confined in the hollow sion relations of the OmniGuide fiber are very similar to
core for a distance equal to several hundred wavelengths those of the metallic waveguide, but the modes can now only
before it radiates away. For a waveguide intended for trans- exist in the TE or TM band gaps of the multilayer mirror. For
mission of light over longer distances, more layers need to be example, the TM01 mode in the metallic waveguide 共blue
added to the cladding. Fortunately, the imaginary part of the dots兲 is now split into two submodes, TM01 and TM01 ⬘ . Thus,
wave vector decreases exponentially with the increase in the it is useful, in a zeroth-order approximation, to think that the
number of layers. Thus, for pure TE modes, only 20 layers dispersion relations for the different modes of the OmniGu-
are enough to give radiative decay lengths of more than a ide fibers can be obtained by overlapping the band structure
100 km while for TM modes and mixed modes, 40 layers are of the dielectric mirror on top of the dispersion relations for
enough. For a detailed analysis of radiation suppression, see the metallic waveguide. Finally, we note that it is possible for
Refs. 关4,15兴. resonant modes to exist outside the omnidirectional fre-
The OmniGuide fiber also supports other categories of quency range 关such as the HE11 mode for ␤ ⬍0.05(2 ␲ /a) in
modes. One category includes modes that are index guided Fig. 3兴 as long as the modes fall within the band gap.
in the dielectric mirror. In Fig. 2, these modes are situated In Fig. 4, we show the dispersion relations for a larger
below the light line, in the light-blue regions of the ␻ ( ␤ ) radius of the fiber, R⫽3.8a. More resonant modes can be
diagram. Their field decays exponentially in the air regions, found now in the first band gap as the cutoff frequencies vary
but extends throughout the dielectric mirror. Another cat- inversely with the radius R. We focus on the first band gap of
egory includes modes that decay both in the air regions and the dielectric mirror, and we consider only the lowest five
in the dielectric layers. These only exist as surface states resonant modes of the waveguide.

046608-3
IBANESCU et al. PHYSICAL REVIEW E 67, 046608 共2003兲

FIG. 5. 共Color兲 The electric-field time-average energy density

for the five modes in Fig. 4 is shown in a transverse cross section of
the waveguide. The color scheme is such that the energy density
goes from zero 共black兲 to maximum 共yellow兲. The thin blue con-
tours represent the contours between the different layers of the Om-
niGuide fiber. For all five modes, the energy density is plotted for a
FIG. 4. 共Color兲 The lowest five modes supported by an Om- frequency ␻ ⫽0.230(2 ␲ /a).
niGuide fiber of radius R⫽3.8a are shown as solid red 共TE or
TE-like modes兲 and blue 共TM or TM-like modes兲 curves. Higher metallic waveguide. The first three modes are taken from
modes exist, but are not shown. Fig. 3 for a radius R⫽2.0a, and the next two are from Fig. 4,
corresponding to R⫽3.8a. The cutoff frequencies for the
We will compare the resonant modes shown in Figs. 3 and two types of waveguides are quite similar for all the five
4 with the modes in metal waveguides of corresponding radii modes. The differences vary from ⫺8% for the HE21 mode
(R⫽2.0a and R⫽3.8a). In particular, we will compare the to ⫹6% for the TM01 mode. Also note that the modes in the
resonant modes to the metal waveguide modes in terms of metal waveguide having low cutoff frequencies 关below
their cutoff frequencies, group velocities, group velocity dis- 0.18(2 ␲ /a)] are shifted up in the OmniGuide fiber, while
persions, degeneracies, and mode symmetries. The lowest metallic modes with higher cutoff frequencies are shifted
resonant mode in Fig. 3 is HE11 with a cutoff frequency ␻ down. This effect will be explained later in terms of the
⫽0.155(2 ␲ /a). This corresponds to TE11 , the fundamental phase shift of the dielectric mirror.
mode in a hollow metal waveguide, which has a cutoff at The group velocity of a resonant mode is zero at ␤ ⫽0
␻ ⫽0.146(2 ␲ /a). The second mode in Fig. 3 is TM01 with and approaches c as the frequency is increased, as is the case
␻ ⫽0.189(2 ␲ /a), corresponding to TM01 in the metal wave- in a metal waveguide. However, as the mode nears the upper
guide with ␻ ⫽0.191(2 ␲ /a). Note that for m⫽0 modes, the band edge, the group velocity starts decreasing as a result of
OmniGuide fiber modes are exactly TE or TM polarized, as the gradual loss of confinement in the core.
is the case with all-metal waveguide modes. For nonzero m, The group velocity dispersion of a resonant mode is posi-
TE modes become HE modes, and TM modes become EH tive for low frequencies 共closer to the lower edge of the band
modes. Also, in both waveguides, the nonzero m modes are gap兲, negative for high frequencies and transitions through
doubly degenerate, while the m⫽0 are nondegenerate. zero. This is in contrast to the group velocity dispersion of
In Table I, we summarize the similarities between the modes in a metal waveguide, which is always positive.
equivalent modes in the OmniGuide fiber and the hollow To provide a better understanding of the field patterns of
OmniGuide fiber modes, in Fig. 5 the electric-field time-
TABLE I. Summary of comparison between OmniGuide fiber ជ 兩 2 is plotted for the five modes
average energy density 21 ␧ 兩 E
共OGF兲 modes and their hollow metallic waveguide 共HMW兲 coun-
shown in Fig. 4 at a frequency ␻ ⫽0.230(2 ␲ /a). All modes
terparts. The first three modes are taken from Fig. 3 for a radius
are normalized such that the power flowing in the z direction
R⫽2.0a, and the next two are from Fig. 4, corresponding to R
⫽3.8a.
is the same. Also, in order to capture the angular dependence
of a mode, we no longer use the e im ␾ complex form of the
Mode label Cutoff frequency Degeneracy field. Instead, we use linear combinations of the degenerate
OGF HMW OGF HMW Difference OGF HMW m and ⫺m modes to obtain real fields.
All modes appear to be very well confined to the hollow
HE11 TE11 0.155 0.146 ⫹6% 2 2 core. The energy density in the cladding layers is much
TM01 TM01 0.189 0.191 ⫺1% 1 1 smaller than the energy density in the core, which is why the
HE21 TE21 0.223 0.243 ⫺8% 2 2 cladding layers appear dark in the figure. The TE01 mode is
EH11 TM11 0.163 0.160 ⫹2% 2 2 different from the other four modes, in which its electric-
TE01 TE01 0.163 0.160 ⫹2% 1 1 field energy density has a node near the core-mirror inter-
face. This is true, in general, for all modes of the same sym-

046608-4
ANALYSIS OF MODE STRUCTURE IN HOLLOW . . . PHYSICAL REVIEW E 67, 046608 共2003兲

metry, i.e., TE0ᐉ with ᐉ⫽1,2, . . . . Note that the number of

angular oscillations of the energy density is twice the angular
momentum of a mode.
In order to quantify the comparison between the field dis-
tribution of modes in the OmniGuide fiber and in the metal
waveguide, we use a correlation function ⌫ defined by

冏冕 ជ 1 共 r, ␾ 兲 •Eជ 2 共 r, ␾ 兲
rdrd ␾ E 冏
冋冕 册
⌫⫽ 共2兲
冕
1/2 .
rdrd ␾ 兩 Eជ 1 共 r, ␾ 兲 兩 2 ជ 2 共 r, ␾ 兲 兩 2
rdrd ␾ 兩 E

⌫ can take values in the interval 关0,1兴, a value of 1 corre-

sponding to maximum correlation (⌫⫽1 can only happen if FIG. 6. 共Color兲 Combined phase-shift and dispersion relation
diagrams. Left panel: contour lines for the phase shift ⌬ ␾ are plot-
the two modes are the same up to a constant amplitude scal-
ted together with the band gap of the dielectric mirror 共white re-
ing factor兲. For all modes, the correlation is largest, close to
gion兲, and the light line 共black oblique line兲. The modes of the
the middle of the band gap and decreases as the mode ap-
OmniGuide fiber with R⫽2.0a are shown as solid red and blue
proaches the band edges. The maximum value of the corre- lines for TE and TM-like modes, while the modes in the metal
lation as a function of frequency is above 95% for all modes, waveguide are shown as dotted lines. The vertical arrows show the
and the largest correlation is found for the TE01 mode, i.e., frequency difference between equivalent modes in the waveguides.
99.1%. The left and right panels are for modes with TE-like and TM-like
polarization, respectively. The crossing between the OmniGuide
V. PHASE-SHIFT MODEL HE11 mode and the metallic TE11 mode is shown as a black dot.

In both the hollow metallic waveguide and the OmniGu- of the band gap 共shown with a thicker black line兲. This is
ide fiber, light is confined in the air region because it is where we expect the dielectric mirror to be most similar to a
reflected back to the core by a cladding: a metal wall in the metal mirror. The other contour lines correspond to phase
first case and a dielectric multilayer film in the second. Thus, shifts of 1.1␲ , 1.2␲ , . . . for increasing frequencies, and
it is convenient to analyze the differences between the modes 0.9␲ , 0.8␲ , . . . for decreasing frequencies. The qualitative
in the two waveguides in terms of the different reflection difference between the two polarizations is that for the TE
properties of the two mirrors. Since the reflectivity is very polarization the phase shift at a given frequency approaches
close to 100% in both cases, the important difference be- ␲ near the light line, whereas for the TM polarization it
tween the two mirrors is found in the phase shift ⌬ ␾ ac- varies substantially near the light line.
quired by an electromagnetic wave upon reflection. For the Also shown in Fig. 6 are the confined modes for the me-
TE polarization, this phase shift is defined by tallic waveguide 共dashed lines兲 and for the OmniGuide fiber
共solid lines兲. On the left we present TE-like modes, while on
E 兩兩r the right the pure TM mode is shown.
⫽e i⌬ ␾ , 共3兲 As seen in Fig. 6, the difference in the dispersion curves
E 兩兩i for the confined modes in the metallic waveguide and the
OmniGuide fiber are correlated with the difference in phase
where E 兩兩i and E 兩兩r are the tangential components of the elec- shifts between the dielectric and metal mirrors. In particular,
tric field for the incident and reflected waves, respectively. In the following observations can be made.
the above formula, we assume that the reflectivity is 100%, 共i兲 The point of crossing between the mode and the me-
which is true for a perfectly conducting wall and is a very tallic TE11 mode lies very close to the ⌬ ␾ ⫽ ␲ line.
good approximation for a dielectric mirror with a large 共ii兲 The frequency distance between the equivalent modes
enough number of layers. For a metallic mirror, ⌬ ␾ is al- in the metal and dielectric waveguides increases as the phase
ways equal to ␲ because the tangential electric fields of the shift of the dielectric mirror moves away from ⌬ ␾ ⫽ ␲ .
incident and reflected waves have to cancel. For the TM 共iii兲 The confined modes in the dielectric waveguide are
polarization, we calculate ⌬ ␾ in a similar manner by using drawn towards the ⌬ ␾ ⫽ ␲ line when compared with the
the tangential components of the magnetic field this time: modes confined in the metal waveguide.
H 兩兩r /H 兩兩i ⫽⫺e i⌬ ␾ . With this definition, ⌬ ␾ for a metal mirror Next we present a simple model that explains qualita-
is equal to ␲ for all polarizations and angles of incidence. tively the above observations by relating the phase shift to
In Fig. 6, we plot the phase shift of the dielectric mirror as the dispersion curves. We approximate the transverse field
contours on a plot, which includes the band-gap edges, the profile of a confined mode of the OmniGuide fiber by a
light line, and dispersion relations for the fiber with R standing wave between two planar walls separated by the
⫽2.0a. The phase shift is a function of ␻ and ␤ , or, equiva- diameter of the waveguide. The total phase acquired by a
lently, of the frequency and the angle of incidence. Also, it is wave that traverses the waveguide diameter twice has con-
a function of the polarization of the incident wave. For both tributions from propagation over a distance 2(2R) and from
polarizations, we can identify a ⌬ ␾ ⫽ ␲ line near the center two reflections upon the mirror. Thus,

046608-5
IBANESCU et al. PHYSICAL REVIEW E 67, 046608 共2003兲

⌬ ␾ total ⫽4k T R⫹2⌬ ␾ , 共4兲

where k T is the transverse wave vector of the mode in core.

The condition for a standing wave is that this phase must
equal a multiple of 2 ␲ : ⌬ ␾ total ⫽2ᐉ ␲ . This in turn leads to
the approximate dependence of the transverse wave vector
on the phase shift,

k T ⫽ 共 ᐉ ␲ ⫺⌬ ␾ 兲 / 共 2R 兲 . 共5兲

Calculating k T from this equation should be done self-

consistently because ⌬ ␾ is a function of ␻ and ␤ , and there-
fore a function of k T .
From Eq. 共5兲 we see that why the phase shift ⌬ ␾ has
important effects on the dispersion relations of modes. Since
k T is a decreasing function of ⌬ ␾ , and because ␻ 2 /c 2 ⫽ ␤ 2
⫹k T2 , it follows that an increase in the phase shift ⌬ ␾ will
cause a decrease in the frequency ␻ of a mode at a fixed FIG. 7. 共Color兲 共Left兲 The confinement of electromagnetic en-
wave vector ␤ . This result explains the three qualitative ob- ergy in the hollow core of the OmniGuide fiber is plotted for three
servations made above about the OmniGuide modes when modes as a function of core radius. Actual data is shown as open
compared to the metallic waveguide modes. circles. The solid lines represent fits as discussed in the text, and the
The most important and visible difference between the dashed line is simply connecting the data points. 共Right兲 The attenu-
two sets of modes is that the OmniGuide modes are pulled ation coefficient of the three modes is plotted as a function of ra-
inside the band gap when compared to their metallic coun- dius. Also shown are the attenuation coefficient in silica fibers 共0.2
terparts. This is entirely due to the fact that the phase shift dB/km兲 and a much larger attenuation coefficient 共100 dB/km兲 as-
increases with increasing frequency. One consequence is that sumed for the materials in the multilayer mirror. The circles and the
lines have the same significance as in the left panel.
the group velocity of the OmniGuide modes is always
slightly smaller than that of the corresponding metallic
modes. It is interesting to imagine what would happen if one TE01 mode has the largest confinement for all values of R.
could design a multilayer dielectric mirror for which the Furthermore, its confinement in the core approaches 100% at
phase shift decreased with increasing frequency: in such a a faster rate than the other modes. The penetration of the
waveguide, the modes would be pulled out towards the band- TE01 field into the multilayer structure is very well approxi-
gap edges, leading to modes with larger group velocities that mated by a 1/R 3 dependence. We present a justification for
could approach the upper bound of c. However, it is an em- the 1/R 3 dependence in the Appendix. A fit to the actual data
pirical law that the phase shift always has the general behav- points 共open circles兲 is shown in Fig. 7 as a solid red line.
ior shown in Fig. 6. Finally, we note that the point of maxi- The TM01 confinement approaches 100% at a much slower
mum correlation between OmniGuide fiber modes and rate. A fit with a 1/R 3 dependence is shown as a solid blue
metallic modes, as defined above, occurs very close to the line. The HE11 mode has a mixed polarization and has a
⌬ ␾ ⫽ ␲ line. more complex variation with the core radius. However, the
rate at which its confinement approaches 100% is less than
VI. SPECIAL CHARACTERISTICS OF THE TE01 MODE
that of the TE01 mode.
In the right panel of Fig. 7, we plot the attenuation coef-
For practical applications, it is of interest to identify the ficient caused by dissipation losses for the three modes. 共We
losses associated with the different modes of the waveguide. assume that the dielectric mirror has enough layers such that
Metallic waveguides were the subject of intensive research the radiation losses are negligible兲. We assume the core of
aimed at utilizing them for long-distance optical communi- the waveguide is lossless, while the materials in the dielec-
cations. It was found that the TE01 mode is the lowest-loss tric mirror have a dissipation coefficient of 100 dB/km 共cho-
mode in a metal waveguide due to the presence of a node in sen to much larger than the loss of silica兲. Even with such
the electric field at the metallic wall. This mode was the lossy materials in the dielectric mirror, the loss of the TE01
operating mode for the long-distance communication lines mode is smaller than the loss of silica fibers 共0.2 dB/km兲 if
designed by the Bell Laboratories prior to the advent of silica the core radius R is larger than ⬇8a. This is because the
fibers 关1兴. In the remaining portion of this paper, we will dissipation loss for TE0ᐉ (ᐉ⫽1,2, . . . ) modes decreases as
examine some of the properties of the TE01 mode in the 1/R 3 . This general result is a consequence of the presence of
OmniGuide fiber; these properties were studied in more de- a node in the electric field near the core-mirror interface. We
tail in Ref. 关4兴. present a more detailed justification in the Appendix. Note
In the left panel of Fig. 7, we plot the confinement in that while the silica fiber has a very low loss only for wave-
percent of the electromagnetic energy in the core for the lengths near ␭⫽1.55 ␮ m, the OmniGuide fiber can be de-
TE01 , TM01 , and HE11 modes as a function of the core ra- signed to have a very low loss in the vicinity of any desired
dius for a fixed frequency ␻ ⫽0.256(2 ␲ /a). As shown, the wavelength, from infrared to visible wavelengths.

046608-6
ANALYSIS OF MODE STRUCTURE IN HOLLOW . . . PHYSICAL REVIEW E 67, 046608 共2003兲

TE01 is the lowest-loss mode within the TE0ᐉ family of tional Science Foundation under Grant No. DMR-9400334
modes. From the derivation presented in the Appendix one and by the Defense Advanced Research Projects Agency and
can show that the loss of a TE0ᐉ mode is proportional to the United States Army Research Office under Grant No.
square of the transverse wave vector in the core, k T,core . DAAD19-01-1-0647. M.I. would like to thank the Donald
This implies a monotonic increase of the loss with the order Stockbarger Fund for financial support.
ᐉ. In particular, the loss of TE02 is about 3.3 times larger
than that of TE01 for almost all frequencies and core radii. APPENDIX: DEPENDENCE OF TE01 ABSORPTION
For modes other than the TE0ᐉ modes, the absence of the LOSSES ON CORE RADIUS
node in the electric field near the core-layer interface results
in more penetration of the field into the cladding, and also The 1/R 3 dependence of the TE01 dissipation loss is due
different scaling laws for the R dependence. Turning back to to the presence of a node in the electric field of this mode
Fig. 7, we see that the TM01 dissipation loss follows exactly near the interface between the hollow core and the first layer
a 1/R dependence 共solid blue line兲, while the loss of HE11 of the dielectric mirror. The only nonzero component of the
has some intermediate dependence with a slope that ap- electric field for the TE01 mode is E ␾ . In the case of a
proaches 1/R for large R. metallic waveguide, this component would be exactly zero at
Another reason for the low losses of the TE01 mode is the the interface between the hollow core and the metal. In the
fact that this mode is confined in the core only by the TE case of the dielectric waveguide, the node in E ␾ is found
gap. TM modes and mixed modes have a component that is very close to the interface.
confined by the weaker TM gap, and thus they penetrate To get an estimate of the dissipation losses, we take the
deeper into the multilayer cladding. ratio of the power that travels in the layers to the total power
The fact that the TE01 loss is smaller than that of the other carried by the mode 关20兴. Let A core be the amplitude of the
modes is very important for the transmission properties of electric field in the core, and A layers the amplitude in the
OmniGuide fibers. Even if several modes are excited at the layers. The area of the core is ␲ R 2 . Due to the exponential
input of a waveguide, after a certain distance only the TE01 radial decay induced by the band gap, the power in the layers
mode will remain in the waveguide. This loss-induced modal is found mostly in the first few layers, in an area approxi-
discrimination results in an effectively single-mode operation mately equal to 2 ␲ Rd p (d p is the penetration depth of the
in a waveguide that supports many modes. In our analysis, field into the layers and is independent of R). With these
we assumed that the number of layers in the dielectric mirror notations, the estimate for the loss coefficient L is

冉 冊
is large enough such that radiation losses can be neglected.
By choosing the number of layers appropriately, the loss-
2
A layers 2 ␲ Rd p 1 A layers 2
L⬃ ⬃ , 共A1兲
induced modal discrimination can be further enhanced since 2
A core ␲R2 R A core
the radiation losses are also the smallest for TE01 关4,5,8兴.
While the radiation and absorption losses of the TE01 de- where we kept only terms that influence the R dependence.
crease substantially when the core radius is increased, the Due to the small amplitude of E ␾ at the core-layer inter-
analysis of attenuation in a practical waveguide necessitates face, it is the derivative of E ␾ with respect to r that connects
the consideration of other loss mechanisms. In particular, the the amplitudes A layers and A core . If we approximate the ra-
loss of the TE01 mode is increased by coupling to other reso- dial field oscillations by sinusoidal oscillations, the deriva-
nant modes or to cladding modes due to deviations from a tive of E ␾ close to a node is given by the transverse wave
straight and perfect waveguide. These effects are analyzed in vector times the amplitude of E ␾ . Thus, the amplitudes in
detail in Ref. 关4兴. the core and in the layers are connected by k T,core A core
⯝k T,layers A T,layers . The transverse wave vector k T,core is
VII. CONCLUSIONS inversely proportional to R because the TE01 mode always
has half an oscillation of E ␾ between the origin and r⫽R. In
In conclusion, we have shown that the mode structure of a layer with index of refraction n, we have k T
an OmniGuide fiber has many similarities with that of a hol- ⫽ 冑(n ␻ /c) 2 ⫺ ␤ 2 . As R becomes larger, the TE01 mode will
low metallic waveguide. We explained why these similarities get closer to the light line, and therefore ␤ approaches ␻ /c.
exist, and we presented a simple model that accounts for the If n is not equal to 1, k T will be almost independent of R.
differences. We identified the TE01 mode as the lowest-loss Thus k T,layers is almost independent of R. Thus, we have
mode in the dielectric waveguide, as it was the case for the
metal waveguide. It is hoped that the analogy developed in A layers k T,core 1
this paper has provided an intuitive understanding of the ⯝ ⬃ . 共A2兲
A core k T,layers R
modal structure of the OmniGuide fiber and could be applied
to the design of transmission lines or optical devices based
The expression for the loss becomes
on this type of fiber.

ACKNOWLEDGMENTS L⬃ 冉
1 A layers
R A core 冊 2
⬃
1
R3
, 共A3兲
This work was supported in part by the Materials Re-
search Science and Engineering Center program of the Na- which is the result we wanted to justify.

046608-7
IBANESCU et al. PHYSICAL REVIEW E 67, 046608 共2003兲

关1兴 W.D. Warters, Bell Syst. Tech. J. 56, 1825 共1977兲. 关11兴 J.N. Winn, Y. Fink, S. Fan, and J.D. Joannopoulos, Opt. Lett.
关2兴 M. Miyagi and S. Kawakami, J. Lightwave Technol. LT-2, 116 23, 1573 共1998兲.
共1984兲. 关12兴 D.N. Chigrin, A.V. Lavrinenko, D.A. Yarotsky, and S.V.
关3兴 Y. Matsuura and J.A. Harrington, J. Opt. Soc. Am. A 14, 1255 Gaponenko, Appl. Phys. A: Mater. Sci. Process. 68, 25 共1999兲.
共1997兲. 关13兴 Y. Fink, D.J. Ripin, S. Fan, C. Chen, J.D. Joannopoulos, and
关4兴 S.G. Johnson, M. Ibanescu, M. Skorobogatiy, O. Weisberg, E.L. Thomas, J. Lightwave Technol. 17, 2039 共1999兲.
T.D. Engeness, M. Soljacic, S.A. Jacobs, J.D. Joannopoulos, 关14兴 Y. Xu, R.K. Lee, and A. Yariv, Opt. Lett. 25, 1756 共2000兲.
and Y. Fink, Opt. Express 9, 748 共2001兲. 关15兴 Y. Xu and A. Yariv, in Proceedings of the SPIE, edited by A.
关5兴 P. Yeh, A. Yariv, and E. Marom, J. Opt. Soc. Am. 68, 1196 Dutta, A.A.S. Awwal, N. K. Dutta, and K. Okamoto 共SPIE-Int.
共1978兲. Soc. Opt. Eng., Washington, 2001兲, Vol. 4532, pp. 191–205.
关6兴 N.J. Doran and K.J. Bulow, J. Lightwave Technol. LT-1, 588 关16兴 T. Kawanishi and M. Izutsu, Opt. Express 7, 10 共2000兲.
共1983兲. 关17兴 M. Ibanescu, Y. Fink, S. Fan, E.L. Thomas, and J.D. Joannopo-
关7兴 A.N. Lazarchik, Radiotekh. Elektron. 共Moscow兲 1, 36 共1988兲. ulos, Science 289, 415 共2000兲.
关8兴 C.M. de Sterke and I.M. Bassett, J. Appl. Phys. 76, 680 共1994兲. 关18兴 A. Argyros, Opt. Express 10, 1411 共2002兲.
关9兴 F. Brechet, P. Roy, J. Marcou, and D. Pagnoux, Electron. Lett. 关19兴 J.D. Jackson, Classical Electrodynamics, 3rd ed. 共Wiley, New
36, 514 共2000兲. York, 1998兲.
关10兴 Y. Fink, J.N. Winn, S. Fan, C. Chen, J. Michel, J.D. Joannopo- 关20兴 A more precise expression is derived in Ref. 关4兴, but the proof
ulos, and E.L. Thomas, Science 282, 1679 共1998兲. of the 1/R 3 dependence is the same.

046608-8