Accurate
Calculation of Far Field Diffraction Pattern (Chuck
Titus)
OUTLINE:
1.Diffraction Theory.
There are a number of different diffraction models.
These models can be divided into two classes, vector and scalar.
Scalar treatments include Huygens’ Principle[1,2,3], Rayleigh-Sommerfeld
theory[1,2,3,4], the Kirchhoff formulation[1,2,3,4], and a model referred
to as the angular spectrum of plane waves[3,5]. The well-known Fraunhofer
and Fresnel approximations, as they appear in most introductory optics
texts, are derived from the Kirchhoff model. Vector models include
a treatment by Stratton and Chu[6], another based on the surface equivalence
principle[7], and a rigorous electromagnetic boundary value model which
appears in Jackson’s text[7] (among other places). Each of these
theoretical models has strengths and weaknesses, and each can be satisfactorily
employed for some range of problems. The choice of an appropriate
model is based on what is known about a specific problem.
Application of any diffraction model can be divided
into two separate tasks[9]. First, one must obtain the fields exiting
a diffracting object (i.e. the near fields, or the boundary field values),
or a reasonable approximation thereof. The second step involves propagating
those fields to the desired observation point. These are distinct
and separate parts of the diffraction models listed above. Most texts
do not make this separation clear. Instead, the boundary value assumptions
and subsequent propagation into the far field are lumped together into
one theoretical treatment. If the resulting diffraction pattern is
at all inaccurate, it is difficult to determine how much of that error
is due to incorrect boundary fields and how much is the result of the propagation
calculation. Because of this, it is often difficult to know which
model is appropriate for a particular problem.
We are not searching for an accurate complete diffraction
model. What is sought is only that portion of one of the diffraction
theories which can accurately calculate far zone diffraction patterns given
accurate near fields. Our purpose here is not to provide rigorous
proof of accuracy. Instead, what follows is an attempt at sorting
through the various theories in order to provide rationale for the choice
of one such method.
It is often said that scalar diffraction models are not applicable for diffracting apertures possessing features with sizes on the order of a wavelength. In addition, some models, notably the Fraunhofer and Fresnel approximations, are of questionable value for calculating nonparaxial diffracted fields.
In general, scalar theories assume the exact boundary
field distribution (i.e. the near field) is not known, and make an intelligent
guess about its values. For instance, Rayleigh and Kirchhoff treatments
of diffraction by a slit in an opaque screen often assume the fields inside
the opening are the same as the undisturbed (i.e. incident) values, and
are zero elsewhere on the screen. That assumption produces boundary
fields which are discontinuous at the edges of the slit. Approximate
boundary fields obtained in such a manner are not an exact representation
of the true boundary fields. The approximation neglects to apply
Maxwell’s equations to obtain exact boundary conditions at the edges of
the aperture. As a result, subsequent calculation of the diffracted
fields will contain some error. The magnitude of error increases
as the width of the slit is reduced to and below the wavelength of incident
light.
In any case, scalar theory calculations based on
simple assumptions about boundary fields generally depart somewhat from
experimental data. This is particularly true for cases when the diffracting
aperture or features therein are not large compared to a wavelength[10,11].
Approximation of the boundary field distribution is at least partly to
blame.
There have been attempts to obtain more accurate
representations of boundary fields for some simple problems such as an
infinite slit[12], an infinite half-plane[13], or a circular aperture[2].
However, those boundary field computations are difficult and not generally
applicable to arbitrary diffracting apertures.
Vector models are also capable of obtaining more
accurate near fields. Such treatments apply electromagnetic boundary
conditions, derived from Maxwell’s equations, to the diffracting aperture.
The accuracy of boundary fields obtained in such a manner depends on the
ability to solve the boundary condition equations. Again, for complicated
diffracting objects, the calculation is often extremely difficult if not
impossible.
So it is imperative that one obtain the most accurate
possible representation of the near fields. The FDTD
method is capable of computing accurate near fields. But if one
can obtain accurate boundary fields by some method such as FDTD, which
propagation method should be employed? Are there restrictions which
accompany any of these propagation methods?
First, we should make note of one feature of the
FDTD computation method. The core algorithm employed by the FDTD
method is derived directly from the differential forms of Ampere’s and
Faraday’s laws, also known as the two Maxwell curl equations. The
two divergence laws are not explicitly enforced in the FDTD method, but
it has been shown to possess a divergence-free nature[14]. Consequentially,
it can be said that fields computed by the FDTD method are solutions to
Maxwell’s equations, to within any error incurred by the numerical nature
of the computation. FDTD can provide significantly more accurate
near fields than the approximations often encountered in scalar diffraction
theory.
Given that the near fields produced by the FDTD
computational method can be close approximations of Maxwell fields, all
that should be required of a propagation model is that it retain the Maxwellian
nature of those fields as they propagate away from the near zone.
Any such propagation model should produce an accurate diffraction pattern
from accurate fields on the surface of a diffracting aperture.
With the exception of the original Huygens’ Principle,
all of the diffraction theories listed above contain propagation models
derived from Maxwell’s equations. Propagation models in the scalar
Rayleigh and Kirchhoff theories are based on the Helmholtz wave equation,
which is obtained directly from Maxwell’s curl equations. After application
of Green’s Theorem, the propagation models are derived. The scalar
propagation models are then applied to each vector component of the boundary
field. If the vector near fields adhere to Maxwell’s equations, the
component-by-component propagation of those fields employing a method derived
from the wave equation should, in theory, produce Maxwell far fields.
It should be noted that the differences between
these scalar theories result from different aspects of the boundary fields
employed in their respective propagation models. As a result, incorrectly
specified boundary fields can lead to different diffraction patterns from
these scalar theories. Conversely, given the same accurate boundary
values, these propagation models should produce the same diffraction pattern[9].
The preceding discussion provides reasonable expectation
that we can safely use the propagation models contained in Rayleigh and
Kirchhoff theories. The major problem with those theories is the
possibility of incorrectly specified boundary fields. Use of FDTD-computed
near fields significantly reduces that concern.
We have not discussed the angular spectrum representation (Dr. E. C. Gartland here in Kent State's Math and CS Dept. knows a lot about that) or any of the vector theories. Each of these methods has its proponents. We can say that there are doubts about the validity of the Stratton-Chu theory because its propagation model is not divergence-free. The Surface Equivalence Principle is popular in the electrical engineering community, but is difficult to implement in its non-paraxial form. In theory, it should produce Maxwell far fields from Maxwell near fields. The angular spectrum of plane wave representation can be connected to the wave equation and, like the Rayleigh and Kirchhoff propagation models, is theoretically correct.
2. Reduction of Two-D Kirchhoff Surface
Integral to One-D Line Integral.
In theory, we could have employed either Rayleigh integral. However, the Kirchhoff integral is a simple average of the two Rayleigh integrals. One of the Rayleigh integrals makes use of near field magnitudes, and the other utilizes the derivatives of near fields. It is possible that numerical errors may be more pronounced in one or the other Rayleigh integral, and that may change from one problem to another. If so, an average of the two can reduce that concern. For that reason, and for reasons of coding ease, we chose to implement the primitive form of the Kirchhoff propagation integral.

FIGURE
1. Geometry of diffraction problem. The diffracting object
lies in the y’=0 plane between x’=(-d) and x’=d.
Diffracted light is observed at (x,y,z).
The problem at hand is
shown in Figure 1. The drawing represents a planar cross-section
from a three-dimensional diffraction geometry. Points on the "exit"
surface of the diffracting object are described by (x',y',z').
The diffraction pattern is observed in the half-space y>0, at points
(x,y,z). For this general geometry, the primitive
form of the Kirchhoff surface integral takes the form:
|
|
The Green’s function G(r,r’) is defined as:
|
|
|
|
Now suppose the diffracting
object and the fields emerging from it do not vary in the z direction,
and that the exit surface of the diffracting object is flat, existing on
the y'=0 plane. Such cases may include flat liquid crystal
layers of finite thickness whose director varies internally in the x and
y directions but not in the z direction. Then the above surface integral
must be reduced to a line integral by carrying out the z' integration (over
an infinite range).
Before we reduce the surface
integral to a line integral, a few more aspects of this restricted problem:
Clearly the diffraction pattern produced by this one-dimensionally varying
(along x') aperture will not vary with z. Because
of the z-invariance of this problem, we can arbitrarily choose this problem
to exist in the plane z=0. Then the distance R becomes:
|
|
|
|
If the near fields emerging
from the diffracting object (e.g. the flat liquid crystal layer) are non-zero
only over a limited range from x'=-d to x'=d, the
Kirchhoff integral may also be so restricted. This may occur for
a liquid crystal grating which has been illuminated by a gaussian (laser)
beam, or a single transmitting pixel in an otherwise dark liquid crystal
display.
Now we can reduce the surface integral to a line integral. Making
use of the Hankel function of the first kind to evaluate the integrals
along z’:
|
|
|
|
![]() |
![]() |
|
|
3. References (Bibliography).
1) B. Baker, E.Copson; The Mathematical Theory of Huygens’ Principle; 2nd ed.; Oxford Univ. Press; London; 1950.
2) C. Bouwkamp; Diffraction Theory; Rep. Prog. Phys.; 17; pp. 35-100; 1954.
3) M. Born, E. Wolf; Principles of Optics; 6th Edition; Cambridge University Press; Cambridge; 1997.
4) J. W. Goodman; Introduction to Fourier Optics; McGraw-Hill; New York; 1968.
5) P. Clemmow; The Plane Wave Spectrum Representation of Electromagnetic Fields; Pergammon Press; London; 1966.
6) J. A. Stratton; Electromagnetic Theory; 1st Edition; McGraw-Hill; New York; 1941.
7) S. Schelkunoff; Some Equivalence Theorems of Electromagnetics and Their Application to Radiation Problems; Bell Sys. Tech. J.; 15; pp. 92-112; 1936.
8) J. D. Jackson; Classical Electrodynamics; Wiley; New York; 1962.
9) M. Kowarz; Diffraction Effects in the Near Field; PhD Thesis; University of Rochester; 1995.
10) M. Totzeck; Validity of the Scalar Kirchhoff and Rayleigh-Sommerfeld Diffraction Theories in the Near Field of Small Phase Objects; J. Opt. Soc. Am. A; 8; pp. 27-32; 1991.
11) M. Totzeck; Test of Various Diffraction Theories in the Near Field of Phase Objects; Ultramicroscopy; 57; pp. 160-164; 1995.
12) P. M. Morse, P. J. Rubenstein; The Diffraction of Waves by Ribbons and Slits; Phys. Rev.; 54; pp. 895-898; 1938.
13) A. Sommerfeld, Lectures on Theoretical Physics; Volume VI; Academic Press; New York; 1954.
14) Taflove; Computational Electrodynamics, The Finite-Difference Time-Domain Method; Artech House; Boston; 1995.