Sie sind auf Seite 1von 12

M ODULE 3

Approximations

Weeks 6 & 7
For many electromagnetic problems, analytical solutions are either unknown or
require a great deal of computational resources to convert to numerical results.
We can also find that general computational methods are far beyond available
computational resources in terms of time and computer memory requirements.
For these reasons, approximate methods are often used.
These can directly yield a simple result, or can form the basis of a computa-
tional method. In either case, it is useful to understand the nature and limitations
of the approximations used. In this module, we will look at some widely used
approximations for electromagnetics and optics.

Reading
Main reading:
• Chapter 3, Max Born and Emil Wolf, Principles of Optics, Cambridge Univer-
sity Press, Cambridge, 1997.
• Ovidio M. Bucci and Giuseppe Pelosi, “From wave theory to ray optics”,
IEEE Antennas and Propagation Magazine 36(4), 35–42 (1994).
• Bruce A. Sherwood and Ruth W. Chabay, “A unified treatment of electro-
statics and circuits”
Other:
• Jonathan Molcho and Dan Censor, “A simple derivation and an example
of Hamiltonian ray propagation”, American Journal of Physics 54(4), 351–353
(1986)
• J. M. H. Peters, “The deviation and curvature of a light ray”, Physics Educa-
tion 19, 200–203 (1984)
• Orestes N. Stavroudis, “Simpler derivation of the formulas for generalized
ray tracing”, Journal of the Optical Society of America 66(12), 1330–1333 (1976).
• J. A. Arnaud
“Analogy between optical rays and nonrelativistic particle trajectories: A
comment”
American Journal of Physics 44(11), 1067–1069 (1976)

57
58 PHYS3051—Semester 1, 2011

3.1 Long wavelength approximations of radiating


sources
We can write the fields produced by a source as a superposition of outgoing TE
and TM spherical waves:
∞ n
E= ∑ ∑ anm Mnm + bnm Nnm . (3.1)
n=1 m=−n

If the source is small compared to the wavelength, we expect the dipole mo-
ments to be the most significant multipole moments. Therefore, we can approxi-
mate the field of a small source by
1
E= ∑ a1,m M1,m + b1,m N1,m . (3.2)
m=−1

We can further reduce the number of required terms, since we can often ignore
either the TE or TM mode. If we know that the source is rotationally symmetric,
we can also restrict ourselves to m = 0. Together, these will result in a single term
remaining:
E = a1,0 M1,0 (3.3)
or
E = b1,0 N1,0 . (3.4)
These provide simple expressions for dealing with radiation by small wire an-
tennas (dipole antennas) and scattering by small particles (Rayleigh scattering).
• short straight wire → electric dipole → TM
• dielectric particle → electric dipole → TM
• small wire loop → magnetic dipole → TE

3.2 Diffraction theory


We will now consider the classic problem of the diffraction of a wave incident on
an aperture in a screen. We will consider specifically the case where the aperture
is much larger than the wavelength, and the incident field (at least in the region
of the aperture) is such that we can make a scalar approximation.
Clearly, a simple approximate method is geometric optics, which results sim-
ply in a sharp-edged shadow. The only real choices to be made are whether
to assume that the incident field is a plane wave (parallel rays), due to a point
source (spherical wave), or use a possibly more realistic extended source. But
this completely fails to reproduce the effects of diffraction. Thus, the problem of
diffraction by an aperture is an important problem in practice where it is neces-
sary to account for wave effects in propagation. Diffraction theory is often (usu-
ally?) considered as corrections to this simple ray picture to account for wave
behaviour, with successive approximate methods, such as Fraunhofer diffraction
and Fresnel diffraction approaching reality more and more closely, until the “rig-
orous” Kirchhoff theory is reached.
MODULE 3. APPROXIMATIONS 59

3.2.1 Fraunhofer diffraction


If one assumes that the incident wave is a uniform plane wave over the aperture,
the screen containing the aperture is a flat planar screen, and the point of obser-
vation is sufficiently far, the difference in optical path length from the aperture to
the observation point is a linear function of position in the aperture. This leads
to the simple result that the diffraction pattern is the Fourier transform of the
aperture.
Since analytical expressions of Fourier transforms of many functions are
known, this is a Very Useful Result.
We can note that assuming that the transmitted field is equal to the incident
field over the entire aperture is an unphysical approximation. We can investigate
the limits imposed on Fraunhofer theory by this approximation by considering a
transmitted field that is the convolution of the aperture and a smooting function.

3.2.2 Fresnel diffraction


If we remove the requirement that the observation point is sufficiently far so as to
give a linear dependence of the optical path length, we obtain Fresnel diffraction
theory. The path length has a quadratic dependence on position in the aperture
plane, and the resulting integrals are known as Fresnel integrals, often graphi-
cally represented as the Cornu spiral.
Note that the change in distance to the observation point will affect the mag-
nitude as well as the phase of the field. This is not usually taken into account in
Fresnel theory.
Another thing that is ignored is the dependence of the contribution on direc-
tion to the observation point. Consider the problem of backwards propagation
resulting from a basic simple formulation of Huygen’s principle. Fresnel’s (in-
correct) assumption was to assume that there is no propagation in any backward
direction. The correct variation is 1 + cos θ.

3.3 Kirchhoff diffraction


With the scalar approximation, we have

(∇2 + k2 )ψ = 0 (3.5)

as our starting point. Noting that the Green function G (r, r " ) is the solution of

(∇2 + k2 ) G = δ (r − vr" ), (3.6)

we can substitute ψ and G into Green’s theorem


! !
2 2 ∂ψ ∂G
ψ∇ G − G ∇ ψdV = G − ψ dS (3.7)
V S ∂n ∂n
which yields
! "
∂ψ ∂G
ψ(r) = − ψ dS" .
G (3.8)
S ∂n ∂n
We can consider the geometry of diffraction, where the wave is incoming over
only a portion of the closed surface S " , and is purely outgoing over the remainder
60 PHYS3051—Semester 1, 2011

of the surface. If the outgoing portion of the surface is chosen to be hemispherical


and distant, then
ψα exp(ikr) (3.9)
and
Gα exp(ikr) (3.10)
and the contribution to the integral vanishes.
If we use the free-space Green function
exp(ikR)
G= (3.11)
4π R
where R = |r − r" |, we obtain the Kirchhoff diffraction integral.
Note that we need to know ψ and ∂ψ/∂n over the screen and aperture. We will
typically assume that these are equal to the incident field in the aperture, and zero
elsewhere. This is a little troubling mathematically, since we know that if these
are both zero over a finite region (like the screen), the field is zero everywhere.
Thus, we don’t choose both to be zero. In practice, this means we don’t use the
free-space Green function. Depending on whether we choose Dirichlet (ψ = 0
outside aperture) or Neumann boundary conditions (∂ψ/∂n = 0), we use
! "
1 exp(ikR) exp(ikR" )
Gm athrmD, N = ∓ (3.12)
4π R R"
(which is the Method of Images).

3.4 Vector diffraction


If our scalar approximation fails, we can use a vector diffraction theory. This will
either be the vector version of Kirchhoff diffraction, or electromagnetic theory
without approximations. Vector Kirchhoff diffraction theory is sometimes said
to be the most general diffraction theory, but it isn’t. But it is the most general
useful approximate diffraction theory. (See, for example, the Stratton-Chu formula
for exact diffraction theory.)
One important failure of vector diffraction theory is due to edge effects. The
simple approximation that our fields in the aperture are equal to the incident
field, and zero elsewhere, doesn’t give the correct fields. The error is largest at
the edges of the aperture, and therefore is most important for small apertures
where the edge contributes more to the total result.
For larger apertures, we will probably use scalar diffraction theory, so in prac-
tice, this is a more important problem for vector diffraction.

3.5 The need for short-wavelength approximations


Problems in propagation and scattering involving structures much larger than
the wavelength or fields with variations in amplitude over distances much larger
than the wavelegth can be difficult to treat in terms of solutions to the vector
Helmholtz equation. The difficulties are generally
• using a discrete basis: too many modes are required
• using a continuous basis: the discretisation of the set of continuous modes
must be very fine—thus too many modes are required
• other methods: integrating rapidly oscillating functions
MODULE 3. APPROXIMATIONS 61

3.6 Laser beams and paraxial waves


For example, consider an electromagnetic beam of minimum radius w 0 . Such a
beam can readily have w 0 > 1000λ. A multipole expansion of the beam requires
Nmax ≈ kw0 leading to calculations requiring 10,000 modes or more.
This is an important point! To have a beam that will spread only slowly, it
needs to be wide (think: diffraction). So this is needed, e.g., for laser beams,
microwave beams (typically from a parabolic reflector), etc.
While we might be tempted to resort to ray optics—that is, geometric optics
based on rectilinear propagation of rays, θ reflection = θincidence , and Snell’s law,
assuming a uniform medium—this fails for many cases of interest. For example,
at a focal point, all of the rays will pass through the same point, giving infinite
irradiance and zero-width beam.

3.7 Paraxial wave equation and solutions


Consider the vector Helmholtz equation:

(∇2 + k2 )E = 0. (3.13)

If we wish to consider a beam of width & λ, the cross-sectional shape of such a


beam will only slowly change with z. Thus, the amplitude E can be written as

E = U(e1 , e2 , z) exp(ik z z), (3.14)

where U is a slowly varying function of z. This also requires that transverse


energy flows are small, so | S x |, | S y | ' | S z |. Therefore, both E z and Hz must be
small. This also means that we have k z ≈ k, so we write

E = U(e1 , e2 , z) exp(ikz). (3.15)

Substituting this into the Helmholtz equation we obtain


! "
2 ∂
∇ + i2k U = 0. (3.16)
∂z
This result is exact. Also not very useful.
However, since E z and Hz are small, we have

Hy ≈ some stuff × E x
Hx ≈ some stuff × E y
Ey ≈ some stuff × Hx
Ex ≈ some stuff × H y .

Thus, there is no coupling between the two orthogonal linear polarisations 1 , and
we can use the scalar equation
! "
2 ∂
∇ + i2k U=0 (3.17)
∂z
1 Actually,
there is a sneaky little hidden assumption at this point—we are assuming that the
beam can be characterised by a global polarisation. In practice, this will often be the case, but
not always. An interesting case to consider is that of radially and azimuthally polarised beams,
wherein we arive at a similar expression, but with E r , Eφ , and so on.
62 PHYS3051—Semester 1, 2011

and put polarisation information back in afterwards using

U = ( A x x̂ + A y ŷ)U. (3.18)

Next, we note that


∂2 ∂
2
U'k U (3.19)
∂z ∂z
so we can simplify our wave equation to
! "
2 ∂
∇t + i2k U=0 (3.20)
∂z
2 2
where we can call ∇2t the transverse Laplacian, equal to ∂x ∂
2 + ∂y2 in Cartesian

coordinates. This is the scalar paraxial wave equation.


We might hope to find useful solutions of this wave equation by separation of
variables. However, we would like our solutions to have some useful character-
istics:
• discrete set of modes
• orthogonality such that Ptotal = ∑ Pn
• finite power
• some kind of shape invariance.
Note that the Cartesian and cylindrical solutions to the vector Helmholtz equa-
tion satisfy the last of these, but none of the first three. In general, we can’t have
strict cross-section invariance of the modes under propagation with finite power
modes—diffraction must result in a spread of the energy content over a larger
cross-sectional area. So, rather than looking for strict invariance, we will be sat-
isfied with an invariance of cross-sectional shape only, not size. This means that
the cross-sectional shape will be a function of z, and we cannot simply solve our
wave equation by separation of variables and obtain such modes.
However, we can verify that
# $
1 −r 2
U00 = exp 2
) (3.21)
1 + iz/ z R w0 (1 + iz/ z R

is a finite-power solution, where

z R = kw20 /2 (3.22)

is called the Rayleigh range. The beam width varies with z as


1
w = w0 [1 + z2 / z2R ] 2 . (3.23)

This suggests
√ that the
√ dimensional variables R = 2r/w and Z = z/ z R are useful
(X = 2x/w, Y = 2y/w if using Cartesian coordinates). With this change of
variables, we can look for solutions of the form

U = U00 f ( R) g(φ)h( Z ) (3.24)

or
U = U00 f ( X ) g(Y )h( Z ). (3.25)
With the benefit of hindsight, we can choose h( Z ) = exp(−iΦ( Z )).
MODULE 3. APPROXIMATIONS 63

The Cartesian coordinate solutions are of the form


Unm = U00 Hn ( X ) Hm (Y ) exp(−i(n + m) arctan z/ z R ) (3.26)
where Hn are Hermite polynomials. These are the familiar(?) Hermite-Gauss
laser modes, and are usually written as TEMnm or HGnm .
The cylindrical coordinate solutions are
U pl = U00 Rl Llp ( R2 ) exp(ilφ) exp(−i(2p + l + 1) arctan z/ z R ) (3.27)
where Llp is the Laguerre polynomial of degree p and order l. These modes are
usually written as LG p l.
Note that the beam waist size w 0 is arbitrary.

3.8 Going the other way ...


It isn’t easy to reverse this procedure, and go from an expression for a paraxial
beam to a suitable expression for nonparaxial beams.
There are (at least) three different methods of dealing with this:
• Higher order corrections. For example, higher order terms can be retained
when going from the Maxwell equations to the paraxial wave solutions.
– Melvin Lax, William H. Louisell, and William B. McKnight, “From
Maxwell to paraxial wave optics”, Physical Review A 11, 1365–1370
(1975)
– J. P. Barton and D. R. Alexander, “Fifth-order corrected electromagnetic
field components for a fundamental Gaussian beam”, Journal of Applied
Physics 66, 2800–2803 (1989)
– L. W. Davis, “Theory of electromagnetic beams”, Physical Review A 19,
1177–1179 (1979)
• Find a solution to the Helmholtz equation (which could be either the scalar
or or vector Helmholtz equation, depending on whether we want scalar
or vector waves). For calculating scattering by a sphere, we would want
the multipole (spherical wavefunction) expansion/spectrum for the beam.
Traditionally this was done using an integral transform, integrating over
the surface of the sphere. This introduces an artificial dependence on the
size of the sphere. There are other, and I believe better, choices:
– T. A. Nieminen, H. Rubinsztein-Dunlop, and N. R. Heckenberg, “Mul-
tipole expansion of strongly focussed laser beams”, Journal of Quanti-
tative Spectroscopy and Radiative Transfer 79–80, 1005–1017 (2003)
– G. Gouesbet, “Exact description of arbitrary-shaped beams for use in
light-scattering theories”, Journal of the Optical Society of America A 13,
2434–2440 (1996)
• Just use the paraxial field as the input into the computational window of an
FDTD simulation.

3.9 On the arbitrariness of w 0


While, in principle, any (positive real) value can be chosen for w 0 , some choices
give a more convenient set of mode coefficients. This is particularly useful in
practice, since you will want to truncate the infinite series.
64 PHYS3051—Semester 1, 2011

• R. Borghi, F. Gori, and M. Santarsiero, “Optimization of Laguerre-Gauss


truncated series”, Optics Communications 125, 197–203 (1996)

3.10 A few properties of laser beams


Consider our lowest-order solution (in either Cartesian or cylindrical coordi-
nates), the Gaussian beam:
# $
1 −r 2
U00 = exp 2
) (3.28)
1 + iz/ z R w0 (1 + iz/ z R

Note that this maintains a Gaussian profile for all z. We can write the width
as
2( z2R + z2 )
w2 ( z) = (3.29)
kz R
or %
w( z) = w0 1 + z2 / z2R . (3.30)
Different modes will not stay in phase with each other as the beam propagates
from the focal plane to the far-field, so a beam in general will not maintain its
shape—the far field beam profile will be the Fourier transform of what is in the
focal plane. Each individual mode, however, is self-Fourier, its own Fourier trans-
form, and a single-mode beam will maintain its shape.

3.11 Ray optics as a short wavelength approximation


... or geometric optics or geometrical optics.
We can summarise the essential features of the ray optics approximation very
briefly:

1. Energy in a plane wave in transported normal to the wavefronts, in the


direction of the wavevector.

2. If any wave is approximately plane in some volume of space, this will be


approximately true.
a) A wave will be approximately plane if variation along the wavefronts
in negligible compared to variation along the wavevector (including
the phase variation).
b) This is satisfied if the wave varies only over distances large compared
to the wavelength.

3. Since the boundary conditions for the wave tell us what happens at inter-
faces, we immediately have the laws of reflection and refraction for rays
(i.e., Euclid’s law and Snell’s law).

What have we left out here?


1. The above is not always true in anisotropic media (e.g., birefringent mate-
rials) or some other complex materials.
2. How to deal with non-uniform media, where rays will not always travel in
straight lines.
MODULE 3. APPROXIMATIONS 65

So, it’s useful to have a mathematical derivation, so we can go ahead and


mathematically calculate the trajectories of rays. The classic treatment is to be
found in
• Max Born and Emil Wolf
Principles of Optics
Cambridge University Press, Cambridge, 1997.
However, this is not the easiest or simplest derivation. There have been many
attempts at simple understandable derivations. Some examples are:
• Jonathan Molcho and Dan Censor
“A simple derivation and an example of Hamiltonian ray propagation”
American Journal of Physics 54(4), 351–353 (1986)
• J. M. H. Peters
“The deviation and curvature of a light ray”
Physics Education 19, 200–203 (1984)
• Orestes N. Stavroudis
“Simpler derivation of the formulas for generalized ray tracing”
Journal of the Optical Society of America 66(12), 1330–1333 (1976).
These are available on the course website (along with the rest of Born and Wolf
chapter 3).
These give the basic equations for the paths of rays, including in non-uniform
media. The basic procedure in making the approximation is essentially the same
approximation as made in obtaining classical mechanics from (non-relativistic)
quantum mechanics (in which case, the wave is the wavefunction, and the “ray”
is the trajectory of the classical particle).
• J. A. Arnaud
“Analogy between optical rays and nonrelativistic particle trajectories: A
comment”
American Journal of Physics 44(11), 1067–1069 (1976)

3.12 Usefulness of ray optics


1. Numerical solution of wave problems typically scales as ( L/λ ) N for an N-
dimensional problem. As the size of the problem grows and the problem
becomes less and less computationally feasible, we often find that the ray
optics approximation becomes more and more accurate.

2. If analytical solutions are available, they are often quicker, easier, and more
informative than ray optics. There are exceptions, where “rays” are a natu-
ral way to describe a system.

3. Where the main point of interest is some wave effect, such as diffraction
or interference, then ray optics won’t give this to us. As a result, there has
been interest in “improved” ray optics, where some of the wave behaviour
has been put back in. This is usually some consideration of interference or
diffraction effects.
• Ovidio M. Bucci and Giuseppe Pelosi
“From wave theory to ray optics”
IEEE Antennas and Propagation Magazine 36(4), 35–42 (1994).
66 PHYS3051—Semester 1, 2011

3.13 Circuit theory


3.13.1 Circuits, antennas, and radiation
Consider the relationship between electric circuits and antennas. In particular, a
magnetic dipole antenna, a small loop of wire with a current, is a simple circuit.

• So, why don’t circuits radiate?


– They do, but usually not very much.
– If a circuit is small compared to the wavelength, it will radiate very
little.
– DC has an infinite wavelength, so DC circuits will radiate little.
– If the current changes, what frequencies are present? Find the Fourier
transform! If the current changes suddenly, there is more power
present in higher frequencies, and more power will be radiated. Com-
pare the spectrum of a sine wave and a square wave. What is the fre-
quency spectrum of noise?

3.13.2 Current in a wire, power, and the Poynting vector


If we consider a straight conducting wire, with a potential difference producing
a current, we can find the most important quantities using circuit theory: the
relationship between the potential difference and the current, and the power dis-
sipated.
We can also find the electric and magnetic fields, and the Poynting vector. (See
problem set.)

E (along wire; note that V=EL)

H (around wire)

current (in wire)

H (around wire)

3.13.3 Curved circuit paths and surface charges


Noting that the current in a conducting wire flows along the conductor, the elec-
tric field in the conductor must be directed along the conductor. This means that
the electric field in the conductor isn’t directly produced by the power supply! Where are
the charges that produce this electric field? They must lie on the surface of the
conductors.
This is covered in detail by Sherwood and Chabay, “A unified treatment of
electrostatics and circuits”.
MODULE 3. APPROXIMATIONS 67

Questions
Q3.1 How is Fraunhofer diffraction affected if the incident wave is not normally
incident?

Q3.2 Verify that (3.7) results in (3.8).

Q3.3 Higher order paraxial laser modes are composed of discrete bright spots or
rings. These are separated by dark regions. Why aren’t these dark regions
filled in by diffraction as the beam propagates?

Q3.4 Derive the eikonal equation.

Q3.5 If a complete treatment of circuits requires us to find the fields, why can
we, in practice, just use potential differences and currents?

Discussion board question


• If we consider a weak gravitational field, we can write a simple expres-
sion for gravitational time dilation. Looking from a region distant from the
gravitating body, light will appear to travel more slowly due to this time di-
lation. We can model this as a refractive index, with n > 1. What refractive
index distribution in a spherical lens mimics gravitational deflection by an
object? If we replace the lens by a thin plate of varying thickness, what is
the required thickness?

Problem set—due Thursday 21sth April


Required
The grade of 1–5 awarded for these exercises is based on the required exercises. If all of
the required exercises are completed correctly, a grade of 5 will be obtained.

R3.1 Find the relationship between the current in a small dipole antenna and the
radiated power.

R3.2 Find the far-field and near-field limits for a Gaussian beam. That is, for
z & z R and z ' z R .

R3.3 Find the relationship between the beam divergence and the beam waist w 0 .
If we were to illuminate a retro-reflector on th surface of the Moon with
a laser beam, what beam waist would we need to obtain a beam width of
50 m at the Moon?

R3.4 Consider a straight section of cylindrical wire, of length L, radius a, and


conductivity σ . What is the resistance of the wire? If the wire is connected
across a potential difference of V, what is the current in the wire? (You
can consider the wire to connect two infinite perfectly-conducting parallel
plates.) What are the electric and magnetic fields inside and outside the
wire? What is the Poynting vector? How much power is lost due to the
resistance of the wire?
68 PHYS3051—Semester 1, 2011

Additional
Attempts at these exercises can earn additional marks, but will not count towards the
grade of 1–5 for the exercises. Completing all of these exercises does not mean that 4
marks will be obtained—the marks depend on the quality of the answers. It is possible to
earn all 4 marks without completing all of these additional exercises.

A3.1 Find the radiation resistance of a small dipole antenna. (As for R3.1, but we
want a quantitative relationship, with real units.) From this, find the scatter-
ing cross-sections of a short conducting wire scatterer and a small spherical
dielectric scatterer. Compare. (The scattering cross-section is defined as the
total scattered power for an incident plane wave of unit irradiance, so that
Pscat = σscat I0 , where I0 is the incident irradiance.)

A3.2 In the far field of a Gaussian beam, the wavefronts are a wavelength apart.
What about in the near field? In-between? Where the wavefronts are not
a wavelength apart, we can describe this is terms of a phase shift, which is
the Gouy shift (or the Guoy shift, given a common mis-spelling).
What effect does this have on a multi-mode beam? Compare with a di-
rect calculation of the fields from the paraxial solution. Discuss in terms of
diffraction theory.

A3.3 Are the derivations of ray optics in the papers listed in the notes equivalent?
Are they equivalent to the one given by Born and Wolf? Which one do you
prefer?

A3.4 Calculate the surface charges and fields for a toroidal conductor carrying a
current I.

Das könnte Ihnen auch gefallen