Diffraction of Electromagnetic Waves

Ulrich Brosa∗
arXiv:0911.3663v2 [physics.optics] 2 Feb 2010

Brosa GmbH, Am Brücker Tor 4, D-35287 Amöneburg, Germany

and Philipps-Universität, Renthof 6, D-35032 Marburg
February 2, 2010

Abstract
The general method to obtain solutions of the Maxwellian equa-
tions from scalar representatives is developed and applied to the diffrac-
tion of electromagnetic waves. Kirchhoff’s integral is modified to pro-
vide explicit expressions for these representatives. The respective in-
tegrals are then evaluated using the method of stationary phase in two
dimensions. Hitherto unknown formulae for the polarization appear
as well as for imaging by diffraction. Ready-to-use formulae describing
Fresnel diffraction behind a round stop are presented.
Key words: Electromagnetism; Optics; Diffraction; Polarization
PACS numbers: 41.; 42.; 42.25.Fx; 42.25.Ja

(The final version appeared in Z. Naturforsch. 65a, 1-24 (2010).)

1 Sad state of theory

The first impetus to this work came when your author read an altogether
reasonable preprint on the Casimir effect. Its author, though obviously being
a skilled physicist, wrestled heftily to establish the normal modes of the
electromagnetic field between two metallic plates. The modified preprint is
now published [1].
The second impetus came when your author lectured on theoretical op-
tics. Diffraction of light belongs to the canon. He explained to the students
what he had found in the treatise of Nobelist Max Born [2] and realized only
∗
brosa-gmbh@t-online.de

1
after a few weeks how bad it is. Born’s book and its enhancement [3] are sub-
titled “Electromagnetic Theory of Propagation, Interference and Diffraction
of Light”. A truthful subtitle should rather be “A Wrong Theory of Sound”,
at least with respect to diffraction.
Nobody, neither physicist nor mathematician, seems to be able to system-
atically solve partial differential equations for vector fields. But the Maxwell
equations, foundation of optics and quantum electrodynamics, are just equa-
tions of that kin. As a consequence, the theory of diffraction is still a grave-
yard of intellectual atrocities.
“Graveyard” in German means “Kirchhof”. It is Gustav Kirchhoff who
is held responsible for the theory stolid till now [4],[5],[6]. Even in most
recent textbooks Kirchhoff’s mistakes are recited as the theory of diffraction
[7],[8],[9],[10], see [11] for a historic review.
Kirchhoff was aware of Maxwell’s work, but did not esteem it too much.
He preferred a product made in Germany, namely the Helmholtz equation
∇2p ψ(rp ) + k 2 ψ(rp ) = 0 (1)
wherein rp is to denote the place of the probe, ∇p the nabla operator acting
on rp and k the wave number. ψ is the world-reputed symbol of parapsychol-
ogy. Nobody knows what it means. Kirchhoff baptized it “Verrückung eines
Aethertheilchens” translatable as “insanity of an etheral piece” [6, p.9].
Today Maxwell’s equations are not despised that much. It is generally
accepted that light is an electromagnetic wave. Nevertheless scientists, even
highly qualified ones, stick to Kirchhoff’s doctrines. They observe sagaciously
that in Cartesian coordinates one may extract from Maxwell’s equations one
Helmholtz equation for each and every Cartesian component of the electric
field E and the magnetic one B [7, Chap.3]. The value of this observation is
zero. The components of the electromagnetic vectors are coupled through the
Maxwellian equations for curls and divergences. Solving the six Helmholtz
equations for the components of E and B independently generates grossly
wrong results. Even worse: In problems where Cartesian coordinates do not
suit, as diffraction by spheres, there is, for the components of E and B, no
Helmholtz equation at all [12].
The next row of atrocities was digged when Kirchhoff, following Hermann
Helmholtz, considered an integral which, seemingly, solves the Helmholtz
equation [6, p.82]
1 ZZ exp(ik|rp − r|) exp(ik|rp − r|)
 
ψ(rp ) = ∂n ψ(r) − ψ(r) ∂n df (2)
4π F |rp − r| |rp − r|
wherein the surface F is to divide the entire three-dimensional space in an
inner and an outer part; the inner part is where the probe rp resides. r points

2
to a point on F ; it incorporates the variables of integration as components.
df is the respective element of the surface. ∂n symbolizes a differentiation in
the direction of the outer normal on F . The integral is interpreted in a way
that primary waves ψ(r) travel through outer space until they strike F ; there
they excite secondary spherical waves exp(ik|rp − r|)/|rp − r| interfering to
produce the wanted ψ(rp ).
Arnold Sommerfeld has criticized that simultaneous fixing of boundary
values for the function and its derivative as required in Kirchhoff’s integral
(2) causes contradictions [13],[14],[15]. But scientists answered and still an-
swer, they would take for ψ(r) a plane wave and differentiate it consistently,
without contradiction. These scientists do not understand that diffraction
is possible only if part of the primary wave is screened. How should one
treat these parts of F ? Kirchhoff and his followers demand simultaneously
ψ(r) = 0 and ∂n ψ(r) = 0 on the screened part of F and call it a black screen
[6, p.40]. Yet the Helmholtz equation is elliptic of second degree. It has no
real characteristics. Hence it follows from the Cauchy Kovalevskaya Theo-
rem that the only solution compatible with a black screen is a global zero.
Kirchhoff’s integral produces no better than order-of-magnitude results.
The third row of atrocites was graven when Kirchhoff wanted to evaluate
the integral (2), but could not do it exactly. Especially the function in the
exponents seemed to be invincible. Kirchhoff did then what all physicists do
when they lack ideas. He substituted for the invincible function its crippled
Taylor expansion. The evaluation with linear terms in r only was called
Fraunhofer diffraction, whereas truncation after quadratic terms in r was
said to yield Fresnel diffraction [6, p.86].
One should assume that this kind of Fresnel diffraction includes Fraun-
hofer diffraction as the special case in which quadratic terms are negligible.
But this is not true. The formulae derived in this way describe fundamentally
different pattern.
All these problems will be solved in the present article.
The general method to decouple Maxwell’s equations will be developed in
section 2. The electromagnetic field is represented by three scalar functions a,
b and c which obey separated differential equations. One may solve the sepa-
rated equations for these mathematical auxiliaries one by one and afterwards
calculate the physical fields E and B from the auxiliaries by straightforward
differentiation. c is the familiar scalar potential, a and b are common factors
in the components of the simple and the double vector potential, respec-
tively. This will be demonstrated for all electrodynamics in homogeneous
and isotropic materials, in particular for those with finite conductivity. The
General Representation Theorem in section 2 constitutes the first main result
in this article.

3
 In electrodynamics, diffraction is just a simple special case for which the
representatives a and b suffice.
In all problems with partial differential equations, also in the theory of
diffraction, suitable boundary values must be fixed. Instead of the inconsis-
tent black screen we will put up a screen of perfect conductivity. This entails
boundary conditions for the electrical field E. When these are converted for
a and b, it turns out that the one representative obeys a homogeneous bound-
ary condition of the first kind (aka Dirichlet), whereas the other is subject
to a homogenous boundary condition of the second kind (aka Neumann), see
section 3.
To complete the mathematical definition of diffraction, initial values must
be given. In section 4 we will account for them using a Laplace transform.
Then the transformed representatives a and b fulfill separated Helmholtz
equations. This explains why Kirchhoff’s formulas are not entirely wrong.
For the representation theorem the way the representatives are found
does not matter. Expansions in terms of partial waves are, for example,
feasible, but the aim of this article is to mend Kirchhoff’s advances. Since
the representatives a and b fulfill Helmholtz equations, one may derive for
them integrals similar to (2). As Sommerfeld’s criticism must be attended,
the spherical wave will be replaced with genuine Green functions for boundary
problems of the first and the second kind in section 5.
Hence it might seem that, for the true theory of electromagnetic diffrac-
tion, expense is double as two integrals must be computed. Yet if for the
primary wave a spherical wave is appointed that proceeds from a source at
rq , both a and b are determined by the same function of rp and rq , the only
difference being that probe and source are interchanged. The equations (57)
through (60) display the second main result in this article.
The invincible function in the integral, mentioned above, can be rewritten
to have an intuitive meaning: It is the difference of lengths when the points
rq and rp are connected either directly or via a point r on the screen. These
very terms occur in the familiar triangle inequation. It follows from the
properties of the triangle function that simple geometry determines the basics
of diffraction. For example, the border of shadow can be derived from the
zero of the triangle function. The Criterion of Light will be established in
section 6.
Augustin Fresnel introduced, for the description of diffraction, certain in-
tegrals. For a realistic theory of diffraction these integrals are too clumpsy.
It is more convienient to use instead the error function known to all statis-
ticians. Although one needs, for diffraction, the error function of a complex
argument, it behaves in a similar way as that of a real argument. Aside
for greater simplicity, we get from the error function more results, namely

4
diffraction in dissipative materials, which is of paramount importance for
practical purposes. The salient properties of the complex error function will
be described in section 7.
There are several methods to evaluate the integral (58), e.g. for great
distances of the probe from the screen. In this article we will apply the
method of stationary phase to find the integral in the limit of short waves. It
is this what Kirchhoff wanted when he expanded his functions up to second
powers. We will avoid Taylor’s expansion and introduce instead new variables
that map the triangle function exactly. How this has to be done for two-
and higher-dimensional integrals with finite limits, was hitherto unknown.
It requires a novel deliberation. The Principle of Utter Exhaust will be
introduced in section 8. It constitutes the third main result in this article.
Utter exhaust (90) will be applied to the corrected Kirchhoff integral
(58) in section 9. It yields the fourth main result in this article, the Uni-
versal Formula of Diffraction (105): The diffracted wave equals the primary
wave times the complementary error function which accounts through its ar-
gument for the specific shape of the edge; the argument is found by a purely
algebraic calculation. The formula is universal in so far as it holds for all
single diffracting edges. The diffraction by screens with several edges can be
derived from it by mere superposition. The universal formula also holds for
probes arbitrarily close to the screen. Therefore it can describe the transition
from Fresnel to Fraunhofer diffraction. It also holds for dissipative materi-
als; the argument of the error function also accounts for damping. Yet the
formula (105) does not describe diffraction of very long waves. Moreover we
will notice, under certain conditions, its weakness for great distances of the
probe from the screen.
In this article, no graph of calculated fields will be shown. Detailed com-
parison to or prediction of measurements is here out of scope. Nevertheless
some impressive applications will be outlined. The first is diffraction by a
straight edge in section 10. Diffraction means breaking beams asunder. Nev-
ertheless a metallic half screen creates via diffraction an image of the source,
that is focusing of beams. Astonishing as this result might appear, the same
formula contains as a limiting case Sommerfeld’s stringent solution describ-
ing the diffraction of a plain wave by a half plane, see section 11. Thus
the asymptotic methods developed here have more power than Sommerfeld’s
stringent integration of Maxwell’s equations.
From the diffraction by a single straight edge it is just a small step to
the diffraction by a slit, namely a superposition. Nevertheless the ensuing
formula describes both Fresnel and Fraunhofer diffraction and the transition
between these regimes, see section 12.
In section 13 the universal formula is applied to diffraction by a circular

5
aperture. It results in simple formulae describing Fresnel diffraction behind
a round stop, a device that is used in almost all optical instruments.
In the concluding section 14 the possibly novel results of this work will
be listed and a program what is to be done next be given. The article ends
with a surprise.
This work is a hommage to the natural mind. All calculations were done
using a pencil and sheets of paper. Aid from computers was rejected.

2 The representatives of electrodynamics

The purpose of the representation theorem is to replace Maxwell’s vector
equations for the magnetic and electric fields B and E with separated differ-
ential equations for scalar representatives a, b and c. As soon as the latter
equations are solved, successively, one may determine the vector fields from
the scalar representatives by straightforward differentiation. To understand
and to prove the general representation theorem of electrodynamics, three
requisites are needed. First, the
Lemma of Triple Curl. The vector differential equation

∇ × ∇ × ∇ × va(r, t) = −∇ × Dt va(r, t) (3)

can be reduced to the scalar differential equation

f (v0 r, t) if v1 = 0

2
∇ a(r, t) = Dt a(r, t) + (4)
f (|v|, t) otherwise
if and only if the supporting vector field v is preformed as

v = v0 + v1 r (5)

with arbitrary constants v0 and v1 . Dt symbolizes an operator possibly in-

cluding differentiations with respect to time t, but definitely no differentiation
with respect to space r. f (·, t) denotes a free function. 1
Dt is, for example, εµ∂t2 +µσ∂t , see equation (13) below, with constants ε,
µ and σ denoting dielectric constant, magnetic permeability and conductivity,
respectively. f (·, t) is a so-called gauge, a function which can be chosen
according to convenience. For present purposes f (·, t) = 0 suffices.
The lemma was proven in [16]. It was published in [17] and is available
now in a textbook [18]. It is useful for uncoupling all vector equations which
1
In this section and the ensuing two, the index p at the position of the probe rp will
be omitted.

6
describe physical phenomena in a homogeneous and isotropic space, e.g. in
the theory of elasticity and liquidity.
Second, it is assumed that the electromagetic field propagates in an
isotropic and at least piecewise homogeneous medium. The constitutive re-
lations D = εE and H = B/µ, which relate the force-exerting fields E and
B with the source-caused fields D and H, as well as Ohm’s law

j = σE , (6)

which relates the electric field E with the electric current density j, will be
mounted in Maxwell’s equations from the very start.
Third, as we want to solve initial- and boundary-value problems in time t
and three-dimensional space r for the electric field E(r, t) and the magnetic
field B(r, t), it is necessary to discriminate between charges and currents
that are enforced from outside and those that come about through the free
play between inside fields. The charge density ρo (r) at the initial time t =
to is determined by reasons outside the considered system. In conducting
materials it decays exponentially. The remainder ρ(r, t) is zero at t = to and
thus determined by the density of the current because of continuity
σ
 
∂t (ρ(r, t) + ρo (r) exp (to − t) ) = −∇(je (r, t) + j(r, t) + jo (r, t)). (7)
ε
The current density, in its turn, is partly generated by external sources
je (r, t). In conducting materials, moreover, the electric field drives inner
currents j(r, t) + jo (r, t) according to Ohm’s law (6). The latter is the current
caused by the decay of ρo (r).
General Representation Theorem of Electrodynamics. Solutions of
the Maxwellian equations

∇ × B(r, t) = εµ∂t E(r, t) + µσE(r, t) + µje (r, t) (8)

∇ × E(r, t) = −∂t B(r, t) (9)

1 σ
 
∇E(r, t) = (ρ(r, t) + ρo (r) exp (to − t) ) (10)
ε ε
∇B(r, t) = 0 (11)
are provided by the solutions of the differential equations for the scalar rep-
resentatives a, b and c
t σ
Z  
2
∇ a(r, t) = εµ∂t2 a(r, t) + µσ∂t a(r, t) − µ je (r, τ ) exp (τ − t) dτ (12)
to ε

7
 ∇2 b(r, t) = εµ∂t2 b(r, t) + µσ∂t b(r, t) (13)
1
∇2 c(r) = − ρo (r) (14)
ε
if the magnetic and electric fields B and E are computed from a, b and c
according to
σ
 
B(r, t) = ∇ × v + ∂t a(r, t) − ∇ × ∇ × vb(r, t) (15)
ε
1
E(r, t) = ∇ × ∇ × va(r, t) + ∇ × v∂t b(r, t)
εµ
σ 1Z t σ
   
−∇c(r) exp (to − t) − je (r, τ ) exp (τ − t) dτ . (16)
ε ε to ε
The supporting vector field v must be parallel to the density of the enforced
current
je (r, t) = vje (r, t) . (17)
For the general three-dimensional density, this amounts to choosing three
linearly independent supporting fields according to (5), introducing three
representatives a and solving three equations (12).
The equations (12) through (17) constitute the first major item of this
article.
The proof proceeds in three steps since Maxwell’s equations constitute a
linear system. We compose the general solution from particular ones. Let us
begin with the extraction of nonhomogeneities.
First step: The ansatz
σ
 
E(r, t) = −∇c(r) exp (to − t) B(r, t) = 0 (18)
ε
is to extract the nonhomogeneity of (10) with ρo (r). The ansatz satisfies all
Maxwellian equations except the third (10). The third yields the Poisson
equation (14) for the scalar potential c(r). Please find ansatz (18) linearly
enclosed in the general representation formulae (15) and (16).
The Ohmian current (6) caused by this electric field
σ
 
jo (r, t) = −σ∇c(r) exp (to − t) (19)
ε
balances in (7) the term with the charge density ρo (r). We discard them
both to go on with a simplified equation of continuity and observe that it
still facilitates the elimination of ρ(r, t).

8
 Hence, for the remaining nonhomogeneities of Maxwell’s equations, we do
not miss anything when we differentiate the remainder of (10) with respect
to time
σ 1
∇(∂t E(r, t) + E(r, t) + je (r, t)) = 0 . (20)
ε ε
The equation of continuity (7) was used to get rid of the charge density
ρ(r, t). Ohm’s law (6) was applied to eliminate the current density j(r, t).
Equation (20), however, is guaranteed if the first Maxwellian equation (8) is
fulfilled. To see this, one just has to take its divergence.
We can completely forget about the third and fourth Maxwellians (10-
11) when we represent the magnetic field B by a vector potential A(r, t), i.e.
B(r, t) = ∇ × A(r, t). This shall be done henceforth. One should, however,
keep in mind that mere introduction of a vector potential does not help much.
The differential equations for the vector potential are coupled in a similiar
way as Maxwell’s equations for the electric and magnetic fields. We must
construct special vector potentials to obtain uncoupled differential equations.
Second step: The ansatz
B(r, t) = ∇ × vα(r, t) (21)
introduces, as announced, a special vector potential, A(r, t) = vα(r, t), viz.
a predetermined vector field times a free scalar function.
It shall be used to extract the nonhomogeneity je (r, t). The first Maxwellian
(8) can be written as
σ 1 1
∂t E(r, t) + E(r, t) + je (r, t) = ∇ × ∇ × vα(r, t) . (22)
ε ε εµ
Integrating this equation with respect to time yields a representation of the
electric field
1 σ
Z t  
E(r, t) = ∇ × ∇ × v α(r, τ ) exp (τ − t) dτ
εµ to ε
1 t σ
Z  
− je (r, τ ) exp (τ − t) dτ . (23)
ε to ε
Most people prefer differentiations over integrations. Therefore we redefine
the representative
t σ
Z  
a(r, t) = α(r, τ ) exp (τ − t) dτ . (24)
to ε
This transforms (21) to
σ
 
B(r, t) = ∇ × v + ∂t a(r, t) (25)
ε

9
and (23) to
1 1 t σ
Z  
E(r, t) = ∇ × ∇ × va(r, t) − je (r, τ ) exp (τ − t) dτ . (26)
εµ ε to ε
Please find these terms enclosed in the representation formulae (15) and (16).
The only Maxwellian equation (9) not yet fulfilled produces after insertion
of (25) and (26) the nonhomogeneous telegraph equation (12). To see this,
one has to apply the lemma of triple curl (3). This is possible only if the
current density is parallel to the supporting field, i.e. if condition (17) is
fulfilled.
Now that we have taken care of the enforced current je (r, t), we may
assume for the rest of the proof that only the current caused by the inner
electric field via Ohm’s law (6) remains. This is not a nonhomogeneity. The
first Maxwellian (8) can be simplified

∇ × B(r, t) = µσE(r, t) + εµ∂t E(r, t) . (27)

Third step: The ansatz

E(r, t) = ∇ × vβ(r, t) (28)

inserted into the second Maxwellian (9) produces the representation

Z t
B(r, t) = −∇ × ∇ × v β(r, τ )dτ . (29)
to

Again, for calculational convenience we redefine

Z t
b(r, t) = β(r, τ )dτ (30)
to

to obtain from (28) and (29) the still missing terms in the general repre-
sentation formulae (15) and (16). The only Maxwellian which is yet not
satisfied, the first one (27), causes, after application of the lemma of triple
curl (3), the condition (13), again a telegraph equation, this time, however,
a homogeneous one. Q.E.D.
The general representation theorem copes with almost everything dis-
cussed in most monographies on electrodynamics [19]: electrostatics, mag-
netostatics, also electric discarge, skin effects in conductors, propagation of
electromagnetic waves in space, guides and resonating cavities, optics, metal-
lic optics too, radiation from antennae and all kinds of electromagnetic scat-
tering, especially Mie scattering, which appears in thoses monographies as an
extremely difficult case. For the theory of supraconductivity, Ohm’s law (6)

10
must be replaced with London’s equation, but this even simplifies the deriva-
tion of a slightly modified representation theorem. Only the electrodynamics
of nonisotropic materials cannot be tackled in this way.
Vector potential and double vector potential, dubbed Hertzian vector, are
known since long. The furthest reaching representation of electromagnetic
fields in terms of these potentials was probably found by Max von Laue [20].
For instance, the term in (16) with the integral over the current density,
which bewilders saplings, was given by Laue, but without consideration of
conductivity. However, Laue did not disentangle Maxwell’s equations. Many
scientists subject their vector potentials to so-called Coulomb or Lorentz
gauges. These gauges are related to invariances, but initial and boundary
data break them. Thus in initial- and boundary-value problems, these vector
potentials mislead. What one must use instead are adaptive scalars times
fixed vectors. One can imagine suitable vector potentials as scalars riding on
prepared vector fields like trains ride on rails. Inspiring in this direction was
Peter Debye’s simplified solution of Mie’s problem using a special Hertzian
vector [21]. A remarkably complete list of scalar functions that are useful for
the disentanglement of Maxwell’s equations was presented by Meixner and
Schäfke [22]. Yet this list is valid only for free propagation of harmonic elec-
tromagnetic waves. The general principle of representation, i.e. the lemma
of triple curl, was obviously unknown to all these scientists. Later on elec-
tronic computers spread and absorbed interest. So this gap in mathematical
physics was filled only in 1985 [16].
The brightest indication that scientists do not understand the general
principle of representation is the lack of a reasonable theory of electromag-
netic diffraction. There were attempts, for example [23],[24],[25],[26], to ac-
count for the vector fields, but they produced after longish explications only
approximations - if at all. Also, partial wave expansions are not helpful since
they converge well only for long waves [22]. By contrast, the theory that will
be developed in the following sections is straightforward, yields exact equa-
tions and is easily applied to practical problems. Kirchhoff or his scholars
had done this if they had only known the approach.
The situation was similar in hydrodynamics. With the same methods
as explained here it was possible to derive, for the first time, turbulence in
pipes from the Navier-Stokes equation [27]. In 1989 your author published
the prediction that pipe turbulence consists of transients [28]. It was verified
experimentally in 2006 [29].

11
3 Boundary values on perfect conductors
While the representatives a, b and c obey separated differential equations,
they are usually tied together in the conditions on the boundary- and initial
values of B and E. Consider, for simplicity, the diffraction of electromagnetic
waves where explicit consideration of sources is not necessary. We do not
need a scalar potential c. a and b must solve only homogeneous telegraph
equations, cf. (12) and (13).
Waves are diffracted when impeded by a screen. The only way to get on
with boundary conditions rather than with conditions of transition is to have
the screen made of perfectly conducting material. Then, because of Ohm’s
law (6) for σ → ∞, the electric field E cannot maintain, on the screen, any
tangent component. The waves do not penetrate.
Let t denote any vector tangent to the screen S not to be confounded
with the time t. The boundary conditions follow from the representation
(16)
1
(∇ × ∇ × va(r, t))t + ∂t (∇ × vb(r, t))t = 0 for r ǫ S . (31)
εµ
These are two equations because there are two linearly independent tangen-
tial vectors t on a two-dimensional boundary. All the more it is surprising
that the two unknowns a and b can satisfy the next four equations
(∇ × ∇ × va(r, t))t = 0 and (∇ × vb(r, t))t = 0 for r ǫ S . (32)
This is possible if the supporting field v is parallel or perpendicular to the
diffracting screen.
To prove this, define a local Cartesian coordinate system
r = ex x + ey y + ez z t = ex tx + ey ty (33)
such that it unit vectors ex and ey be parallel to the screen, whereas ez
pierce it normally. Just the components tx and ty of the tangential vector
t are arbitrary though constant. Consequently only x and y components of
curl and double curl need to be considered if the boundary conditions (32)
are to be satified.
v = ez is according to (5) an admissible choice for the supporting vector
field. The equations (32) become
(∇ × ez b)t = ex t∂y b − ey t∂x b = 0
(∇ × ∇ × ez a)t = ex t∂x ∂z a + ey t∂y ∂z a = 0 . (34)
The differentiations with respect to x and y are inner ones since the tangent
vector t of equation (32) is spanned by ex and ey . Both ∂x b = 0 and ∂y b = 0

12
are satisfied on the screen if b = 0. Equally, both ∂x ∂z a = 0 and ∂y ∂z a = 0
are satisfied on the screen if ∂z a = 0.
Generally the supporting vector field v (5) is not constant. However,
if it is normal to the surface, we can construe its length as a factor of the
representative and repeat the preceding calculation. Hence the following
theorem:
Theorem on Boundary Conditions. A supporting vector field v normal
to the surface S of a perfect conductor induces homogeneous boundary con-
ditions of the first kind for the representative b, whereas the representative
a must fulfill homogeneous boundary conditions of the second kind

b(r, t) = 0 and ∂n |v|a(r, t) = 0 for r ǫ S . (35)

Oppositely, a supporting vector field tangent to the surface induces homoge-

neous boundary conditions of the first kind for the representative a, whereas
the representative b must fulfill homogeneous boundary conditions of the
second kind
a(r, t) = 0 and ∂n b(r, t) = 0 for r ǫ S . (36)
∂n denotes differentiation along the normal on S.
For convenience of reference, your author bundled the essentials of this
section in one theorem. Its second part still has to be proven. With the local
coordinate system introduced above, v = ex is according to (5) an admissible
choice for the supporting vector field, too. The equations (32) are now

(∇ × ex b)t = ey t∂z b = 0
(∇ × ∇ × ex a)t = ex t(∂x2 − εµ∂t2 − µσ∂t )a + ey t∂x ∂y a = 0 . (37)

For the first component of the latter equation the homogenous telegraph
equation (12) was exerted. Again, the differentiations with respect to x and
y are inner ones. The same is true for the differentiations with respect to
t because boundary conditions must hold for all times. Hence ∂z b = 0 and
a = 0 on the screen. When this is written without Cartesian coordinates, the
second part of the above theorem emerges. A correction with the length of
the supporting vector is not necessary here because v as defined in (5) does
not vary in the direction of the normal if it is perpendicular to that normal.
Q.E.D.
Screens do not enclose radiation. Much of it spreads in open space. Thus
the boundary conditions need to be completed, namely by retardation
√
fa or b (|r| − t/ εµ)
 
a or b(r, t) ∼ + O(σ) e−σt/2ε for |r| → ∞ (38)
|r|

13
meaning that waves trail away in nirvana and never return. The functions
fa and fb may depend on the direction of the radiation, but they depend
√
on the distance |r| only through the compound argument |r| − t/ εµ. O(σ)
is E.Landau’s order symbol to appraise neglected terms on the right-hand
side, see e.g. [30, Section 1.1]. Notice: σ denotes here the conductivity of the
propagating medium, for example air.
Hence we may have boundary conditions that do not couple the represen-
tatives. Yet to profit from the theorem we need supporting vector fields that
are everywhere either normal or tangent to the screen. According to (5) this
can be achieved for four types of screens: for plane ones, for parts of spheres,
for parts of cones and for parts of cylinders, i.e. for cylinders with arbitrary
cross sections. In this article we will be busy enough to cope with diffraction
by plane screens and will use the theorem with the tangent supporting vector
field (36).

4 Initial values transformed

We account for arbitrary initial values using a Fourier or rather a Laplace
transform. All fields are proportional to exp(−iωt).

B(r, t) = Bk (r) e−iωt E(r, t) = Ek (r) e−iωt (39)

a(r, t) = ak (r) e−iωt b(r, t) = bk (r) e−iωt (40)

The telegraph equations (12-13) become thus Helmholtz equations

∇2 ak (r) + k 2 ak (r) = 0 ∇2 bk (r) + k 2 bk (r) = 0 . (41)

The wave number k depends on the frequency ω according to

k 2 = εµω 2 + iµσω . (42)

One of the two parameters, ω or k, can be chosen as real. In the latter case,
a negative imaginary part of ω describes fading with time t → ∞. This is
expressed through
k σ
ω = √ − i + O(σ 2) (43)
εµ 2ε
and corresponds to equation (38). In the first case, a positive imaginary part
of k describes attenuation in space as |r| → ∞. Modulus and phase of k can
be read from
i σ
q  
k= 4
ε2 µ2 ω 4 + µ2 σ 2 ω 2 exp arctan . (44)
2 εω

14
The point to be made is that the phase of k varies only between 0 and π/4.
This will matter in the discussions of section 7.
Finally we must translate retardation (38) into the language of Fourier
transforms. Both representatives ak and bk behave as leaving spherical waves
exp(ik|r|)
a or bk (r) = Fa or b + O(|r|−2) for |r| → ∞ (45)
|r|
at which the real parts of k and ω are supposed to carry the same sign. The
scattering amplitudes Fa and Fb may depend on the wave number k and the
direction of the radiation, but not on |r|. Though Sommerfeld preferred to
write this as ∂|r| ak (r) = ikak (r)+O(|r|−2) etc., your author persists in calling
(45) Sommerfeld’s radiation condition.

5 Kirchhoff ’s theory corrected

Altogether we found that, for a plane screen, the representatives ak (r) and
bk (r) can be computed separately. Both satisfy Helmholtz equations (41)
and exhibit the same behavior in infinity (45). Yet on the screen ak (r) must
solve a boundary-value problem of the first kind, whereas bk (r) is subject to
boundary conditions of the second kind, see equations (36). 2
Modifications of Kirchhoff’s integral (2) inaugurated by Sommerfeld are
useful to solve both boundary-value problems. Instead of the spherical wave
exp(ik|rp − r|)/|rp − r| flexible Green functions G(rp , r)
1
ZZ  
ψ(rp ) = G(rp , r) ∂n ψ(r) − ψ(r) ∂n G(rp , r) df (46)
4π F
are introduced. Kirchhoff’s formula remains valid if these Green functions
fulfill the Helmholtz equation (1), if they have the same singularity as the
spherical wave
∂|rp −r| G(rp , r) = −(rp − r)−2 + O(|rp − r|−1 ) for rp → r (47)
and respect the radiation condition as in equation (45).
In a boundary-value problem of the first kind, the values of the function
ψ(r) itself are known on the plane, but no information on the values of the
derivative ∂n ψ(r) is available beforehand. Hence we need a Green function
that is zero on the plane lest the unknown values matter. We find from (46)
−1
ZZ
ψ(rp ) = ∂n G1 (rp , r) ψ(r) df if G1 (rp , r) = 0 for r ǫ F . (48)
4π F
2
From now on, the index p at the position of the probe rp is indispensable again as
there are other locations which must be discriminated, viz. the position of the source rq
and arbitrary points on the screen r.

15
On the other hand, if the values of the derivative are known, we must do
without the function itself. For the solution of the boundary-value problem
of the second kind we need a second Green function
1
ZZ
ψ(rp ) = G2 (rp , r) ∂n ψ(r) df if ∂n G2 (rp , r) = 0 for r ǫ F . (49)
4π F

For F being a plane characterized in Cartesian coordinates (33) by, say, z=0,
both Green functions are easily found as a spherical wave around the point
of measurement rp plus its mirrored image on the other side of the plane

exp(ik|rp − r|) exp(ik|rm − r|)

G1 or 2 (rp , r) = − or + (50)
|rp − r| |rm − r|

with
rp = ex xp + ey yp + ez zp rm = ex xp + ey yp − ez zp . (51)
Both functions own the correct singularity as in (47) and respect the radiation
condition (45).
Furthermore it shall be assumed that the primary wave is created by a
point-like source at rq behind the plane

exp(ik|r − rq |)
ψ(r) = . (52)
|r − rq |

This kind of primary wave is to be preferred over the usual plane wave
because it respects a radiation condition of type (45) and ensures thus the
validity of the generalized Kirchhoff formula (46). Also it is closer to exper-
iments as it is simpler to produce an approximate spherical wave than an
approximate plain wave. The beloved plain wave can be obtained from the
spherical wave by a straightforward limiting transition |rq | → ∞. Radiation
from higher multipoles can be derived by differentiation with respect to rq
and the general primary wave by superposition.
Writing down the integral for the boundary conditions of the first kind
(48) in Cartesian coordinates ends up with

exp(ikRpq ) −zp Rpq Z +∞ Z +∞ ik − 1/Rp (x, y) ik∆(x,y)

ψ(rp ) = 2 (x, y)R (x, y)
e dxdy (53)
Rpq 2π −∞ −∞ Rp q

and the integral for the boundary conditions of the second kind (49) becomes

exp(ikRpq ) zq Rpq +∞ +∞ ik − 1/Rq (x, y) ik∆(x,y)

Z Z
ψ(rp ) = e dxdy (54)
Rpq 2π −∞ −∞ Rq2 (x, y)Rp (x, y)

16
with
q
Rp (x, y) = |rp − r(x, y)| = (x − xp )2 + (y − yp )2 + zp2
q
Rq (x, y) = |r(x, y) − rq | = (x − xq )2 + (y − yq )2 + zq2 (55)
q
Rpq = |rp − rq | = (xp − xq )2 + (yp − yq )2 + (zp − zq )2

and the triangle function

∆(x, y) = Rp (x, y) + Rq (x, y) − Rpq (56)

which measures the difference of lengths when the points rp and rq are either
directy connected or via an arbitrary point r = ex x+ey y on the intermediate
plane.
It is not by accident that the ψ’s in the equations (53) and (54) are
identical. The factors behind the spherical wave exp(ikRpq )/Rpq including
the double integrals extending from −∞ to +∞ have both the value 1; a
proof of this fact will be given in section 11. The formulae (53) and (54)
are just different mathematical realizations of Huygens’ principle: A wave
propagating from a source point that strikes a plane excites there secondary
waves interfering to reproduce the original wave.
Diffraction happens only if a screen covers parts of the plane. Using the
boundary conditions in the theorem with the tangent supporting field (36)
yields explicit expressions for the representatives
exp(ikRpq ) exp(ikRpq )
ak (rp ) = II(rp , rq ) bk (rp ) = II(rq , rp ) (57)
Rpq Rpq
with the double integral
−|zp |Rpq Z Z ik − 1/Rp (ξ, η) ∂(x, y) ik∆(ξ,η)
II(rp , rq ) = 2 (ξ, η)R (ξ, η) ∂(ξ, η)
e dξdη . (58)
2π D Rp q

To be general enough for all applications, arbitrary variables ξ and η shall

substitute the Cartesian ones x(ξ, η) and y(ξ, η). The functional determinant
∂(x, y)/∂(ξ, η) enters to transform the element of the surface df = dxdy. The
integral II extends only over the aperture or diaphragm D in the surface F .
The arbitrariness of ξ and η shall be used for a simple description of the
aperture in the way that freely varying ξ with a fixed η depicts an edge. The
aperture is described, for example, by −∞ < ξ < ∞ and −η− < η < η+ .
The enthusing result is that, for the full electromagnetic theory of diffrac-
tion with all possible polarizations, one needs to evaluate only one integral.
There is scarcely more work to be done than in the theory for scalar waves.

17
The addititional work consists in some elementary differentiations as pre-
scribed in the representation formulae (15-16) simplified for Fourier trans-
forms as
σ
 
Bk (rp ) = − iω ∇p × tak (rp ) − ∇p × ∇p × tbk (rp ) (59)
ε
1
Ek (rp ) = ∇p × ∇p × tak (rp ) − iω∇p × tbk (rp ) (60)
εµ
with a supporting vector field v=t that consists of a constant tangent as in
(33).
The equations (57) through (60) constitute the second major item of
this article. The integral (58) can be evaluated using various techniques.
Your author will discuss in the following sections only one: the method of
stationary phase applicable for short waves with ℜk → ∞ and ℑk > 0 as
conditions on the real and imaginary parts of the wave number k, respectively.
While equations (57-58) solve mathematically well-posed boundary-value
problems without any error, they do not describe physical diffraction exactly.
When a conducting screen diffracts a wave, reflection cannot be avoided.
Most of the reflected wave stays in outer space, zp < 0, but it is also diffracted.
A tiny fraction of reflection invades inner space zp > 0. We will learn to
handle this in section 11.

6 Light and shadow

The triangle function rules the diffraction of short waves. In its definition (56)
interest was focused on the point of the screen r. Yet the triangle function
also depends on the points of probe rp and source rq

∆(r, rp , rq ) = |rp − r| + |r − rq | − |rp − rq | . (61)

Your author decided to locate

screen r = ex x + ey y + ez z at z = 0 ,
probe rp = ex xp + ey yp + ez zp at zp > 0 , (62)
source rq = ex xq + ey yq + ez zq at zq < 0 .

The triangle function is positive except at that point r = rs = ex xs + ey ys

where the straight line between probe and source pierces the plane

zp (xq + iyq ) − zq (xp + iyp )

xs + iys = . (63)
zp − zq

18
Complex notation is preferred because it eases transformation to other co-
ordinate systems, see below. The point rs is the location of the absolute
minimum of the triangle function, and all the more it is a stationary point.
The consideration holds for fixed points of source and probe. However,
if only rq is fixed, whereas rp varies while r = rs slides on the edge of the
screen, then
∆(r, rp , rq ) = 0 if r on the edge (64)
determines as function of rp a surface, namely the border of shadow.
For calculating diffraction, we will need the root of the triangle function.
Your author utilizes the ambiguity of the root to demand
 q
q  +| ∆(r, rp , rq )| if rp in the shadow,
∆(r, rp , rq ) = q (65)
 −| ∆(r, r , r )| if rp in the light.
p q

It is cogent to assign different signs to the dark and the bright if the function
is to be differentiable. The triangle function is an analytic function which
depends, around its minimum defined by (64), quadratically on its variables.
Therefore omitting√the signs in (65) would induce a similar discontinuity as
in the assignment x2 = |x|. The absolute assignment of the sign, on the
contrary, is arbitrary since diffraction either by a screen or its complement
is equal; remember Babinet’s principle [2, § 47].
To decide where there is light or shadow, a handy criterion is needed.
There is light on the probe if the screen does not impede the straight con-
nection between source rq and probe rp . Thus the positive sign in (65) is to
be taken if the stationary point xs + iys in (63) misses the aperture.
Criterion of Light. When one uses transformed coordinates ξ, η adapted
to the screen such that the aperture is described by η− < η < η+ while ξ
varies freely, the negative sign of triangle function’s root (65) has to be taken
if
η− < ηs < η+ . (66)
ηs is calculated from the stationary point (63) via coordinate transformation.

For example on a screen with a circular aperture, cylindrical coordinates

ρ, ϕ, z = 0 suit. The aperture is defined by ρ < ρ0 with ρ0 as the radius of
the stop, while ϕ varies freely. The transformation to cylindrical coordinates
is facilitated through

xs + iys = ρs eiϕs xp + iyp = ρp eiϕp xq + iyq = ρq eiϕq (67)

19
with the ρ’s as axial distances and the ϕ’s as azimuthal angles. Insertion into
(63) produces two equivalent formulas for the axial distance
q
(zp ρq − zq ρp )2 + 4zp zq ρp ρq S 2
ρs =
zp − zq
q
(zp ρq + zq ρp )2 − 4zp zq ρp ρq C 2
= (68)
zp − zq

with the abbreviations

ϕp − ϕq ϕp − ϕq
C = cos S = sin . (69)
2 2
ρs > ρ0 is thus the criterion for the domain of shadow, i.e. for the positive
sign in (65).
Instead of applying elementary geometry, as was done in this section,
one may calculate the stationary point ξs , ηs by simultaneous solution of the
equations

∂ξ ∆(ξ, η) = 0 ∂η ∆(ξ, η) = 0 ↔ ξ = ξs η = ηs . (70)

The result is, of course, the same as that given in (63) with subsequent trans-
formation of coordinates, but the computational effort is larger. Your author
displays the equations (70) only to ease comprehension of the astonishing
equation (88) which will appear in section 8.

7 Using the error function for diffraction

In the theory of diffraction, Fresnel integrals
s s
2 z 2 z
Z Z
2
C1 (z) = cos w dw S1 (z) = sin w 2 dw (71)
π o π o

are still custom, but the error function or rather the complementary error
function is handier [31],[32]
2 ∞
Z
2
erfc(z) = √ e−w dw . (72)
π z

All statisticians become perfect opticians when they are willing to handle
their favorite function with complex argument.

20
 The error function comprises the Fresnel integrals in a similar fashion as
the exponential function contains sine and cosine
√ √ √
erfc( −iz) = 1 − −i 2 ((C1 (z) + iS1 (z)) . (73)
√
If z is assumed as real,
√ −iz varies
√ on the second main diagonal of the
complex plane since −i = (1 − i)/ 2 = exp(−iπ/4). The features known
for real argument remain if the complex argument is enclosed between the
first and the second main diagonals of the complex plane; expressed by a
relation between imaginary and real parts: |ℑz| ≤ |ℜz|. This is exactly what
we need for optics, see (44). For large negative real parts ℜz → −∞ of the
argument the complementary error function starts at the value 2, assumes
the value 1 at the origin z = 0, and attains the value 0 for large positive
real parts ℜz → +∞. In the crudest approximation, one may think of the
complementary error function as 2 for negative arguments and 0 for positive
ones. It is a switch.
The asymptotic expansion of the error function familiar on the real axis
remains valid in the wedge between the main diagonals which contains the
real axis
1 2
erfc(z) = √ e−z (1 + O(|z|−2 ) for ℜz → +∞ . (74)
πz
The asymptotic expansion on the other side of the complex plane ℜz < 0
follows from
erfc(z) = 2 − erfc(−z) . (75)
The only refinement due to complexity is that the complementary error
function decreases monotonously on the real axis, whereas it takes complex
values and both real and imaginary parts oscillate when the argument be-
comes complex.

8 The method of stationary phase for two-

dimensional integrals
Let us recall the asymptotic calculation of one-dimensional integrals
Z η+
I(k, η− , η+ ) = A(η) eik∆(η) dη (76)
η−

for ℜk → ∞ with ℑk > 0. We assume that the real function ∆(η) ≥ 0 is

stationary for some η = ηs , i.e.
∆ηη
∆(η) = (η − ηs )2 + O((η − ηs )3 ) . (77)
2
21
The indices at ∆ indicate that the function be differentiated twice and the
result evaluated at η = ηs .
The familiar approach in the method of stationary phase is to introduce
a new variable s
∆ηη
v= (η − ηs ) (78)
2
and to forget the higher-order terms O((η − ηs )3 ) in equation (77). The
function s
k ikv2
δ(v) ∼ e for ℜk → ∞, ℑk > 0 (79)
πi
may be construed as a representation of Dirac’s delta function. Thus the
integral (76) yields
s
2πi
I(k, −∞, +∞) ∼ A(ηs ) . (80)
k∆ηη
The amplitude A(η) appears as a constant.
If the limits of integration η± are ±∞, this is correct, but for finite limits
the local approximation (78) induces systematic errors. What we have to use
instead is the global map η → v
q
v= ∆(η) . (81)
To rewrite the integral from the variable η to the variable v, we must calculate
the differential dη = (dη/dv)dv = (dv/dη)−1dv. Because of (79) it is sufficient
to know the value of the differential for v = 0 corresponding to η = ηs . Thus
the value of the differential of the global map (81) to be used in the integral is
the same as the differential of the local approximation (78). The peculiarity
of the global map appears only in the limits:
q q
erfc −ik∆(η− ) − erfc −ik∆(η+ )
s
2πi
I(k, η− , η+ ) ∼ A(ηs ) . (82)
k∆ηη 2
While the preceding is not familiar, it is known [30, Section 2.9]. Yet in
the theory of diffraction one needs to evaluate two-dimensional integrals
Z η+ Z ∞
II(k, η− , η+ ) = A(ξ, η) eik∆(ξ,η) dξdη . (83)
η− −∞

Again it is assumed that the real function ∆(ξ, η) ≥ 0 is stationary at ξs , ηs

∆ξξ ∆ηη
∆(ξ, η) = (ξ − ξs )2 + ∆ξη (ξ − ξs )(η − ηs ) + (η − ηs )2
2 2
+O(|ξ − ξs |3 + (|η − ηs |3 ) . (84)

22
It seems to be a hitherto unsolved problem to find a suitable two-dimensional
global map ξ, η → u, v such that

u2 + v 2 = ∆(ξ, η) . (85)

Here is the solution: The map is

q
u= ∆(ξ, η) − ∆s (η) (86)
q
v= ∆s (η) . (87)

The function ∆s (η) is determined by the

Principle of Utter Exhaust. Eliminate ξ from the derivative

∂ξ ∆(ξ, η) = 0 ↔ ξ = Ξs (η) (88)

to find the exhausting dependence ξ = Ξs (η). Insert the exhausting depen-

dence into the function ∆(ξ, η) to obtain the exhausting function

∆s (η) = ∆(Ξs (η), η) . (89)

The integral (83) is, for ℜk → ∞ and ℑk > 0, asymptotically equal to

q q
2πi A(ξs , ηs ) erfc −ik∆s (η− ) − erfc −ik∆s (η+ )
II(k, η− , η+ ) ∼ q .
k ∆ξξ ∆ηη − ∆2ξη 2
(90)

Utter exhaust follows from the following indispensable requirements:

ξ = ξs , η = ηs be mapped to u = 0, v = 0 , (91)
u, v be real for all ξ, η , (92)
2
v = f (η) be a function of η only . (93)

We need the third requirement (93) to retain the simplicity of the limits in
the integral II (83) when mapping ξ, η to u, v. Because of equation (85) and
the requirement (92) the function f (η) must never exceed ∆(ξ, η) whatever
value ξ takes. Nevertheless for every η there must exist ξ =q Ξs (η) such that
equality is reached: f (η) = ∆(Ξs (η), η). Otherwise u = ∆(ξ, η) − f (η)
cannot take 0, a contradiction to the requirement (91). In other words, f (η)
and ∆(ξ, η) coincide at that value of ξ = Ξs (η) where ∆(ξ, η) gets stationary
with respect to ξ. Hence Ξs (η) is determined by elimination of ξ in the
condition (88).

23
 The final formula (90) can be understood when compared to formula q
(82). The double integration generates the factor πi/k, the square of πi/k.
q
The determinant in 2/ ∆ξξ ∆ηη − ∆2ξη of the map ξ, η → u, v replaces the
q
differential in 2/∆ηη of the map η → v. Q.E.D.
The principle was named as of utter exhaust because the function ∆s (η)
is the largest possible function of η only that fulfills ∆s (η) ≤ ∆(ξ, η); it takes
from the two-variable function as much as a one-variable function can afford,
cf. equation (86).
A further intuitive interpretation follows from a comparison of equation
(88) with the equations (70). The stationary point is where the function
∆(ξ, η) takes its absolute extremum. On the exhausting dependence, by con-
trast, ∆(ξ, η) is extremized only with respect to the one variable ξ, whereas
the other variable η is fixed. In optics, ∆(ξ, η) essentially measures the
distance between points. The absolutely shortest connection is of course a
straight line. But the exhausting function ∆s (η) measures a conditionally
shortest distance, namely if the connecting line is forced to touch the edge
of the diffracting screen. One can determine the exhausting function experi-
mentally using a ribbon of rubber and a lubricated model of the edge.
The difficult part of utter exhaust is the elimination according to equation
(88). It is therefore gratifying to possess linear approximations of equations
(86-87), similar to the one-dimensional case (78). These approximations are
s
∆ξξ ∆ξη
u≈ (ξ − ξs ) + q (η − ηs ) (94)
2 2∆ξξ
v
u ∆ξξ ∆ηη − ∆2ξη
u
v≈ t (η − ηs ) . (95)
2∆ξξ

They were found by utter exhaust (88) applied to the quadratic terms on the
right-hand side of (84); searching principal axes of the ellipse is not a good
idea. For the integral II (83), equation (90) can be used when the exhausting
function is approximated as
∆ξξ ∆ηη − ∆2ξη
∆s (η+or− ) ≈ (η+or− − ηs )2 . (96)
2∆ξξ
The relation to the border of shadow discussed in section 6 appears here at
first sight.
The method of stationary phase is sometimes blamed as not being math-
ematical stringent. It is argued that certain integrals do not converge and
thus certain errors cannot be estimated. The criticism does not apply here.

24
Guided by mother nature, we made a theory where the parameter k of asymp-
toticity has a positive imaginary part ℑk > 0. This guarantees the conver-
gence of those integrals. We can even write the complete asymptotic series.
Augustin Fresnel invented the zone construction in 1816 to prove that
light travels as a wave, but this was just a semi-quantitive idea [33, p.247],
[3, § 8.2]. Its mathematical solution is the principle of utter exhaust presented
only now.

9 The universal formula of diffraction

To make use of stationary phase for diffraction, the amplitude in the integral
(58)
−zp Rpq ik ∂(x, y)
A(ξ, η) ∼ 2
for ℜk → ∞, ℑk > 0 (97)
2π Rp (ξs , ηs )Rq (ξs , ηs ) ∂(ξ, η)
is to be evaluated
q at the point of stationarity ξs , ηs and multiplied by the
factor 2πi/k ∆ξξ ∆ηη − ∆2ξη of equation (90).
Trivially, Rp and Rq are just the two fractions of Rpq
zp −zq
Rp (ξs , ηs ) = Rpq Rq (ξs , ηs ) = Rpq . (98)
zp − zq zp − zq
Distances as they are, they do not depend on the coordinate system. This
is different for the determinant ∆ξξ ∆ηη − ∆2ξη . Let us calculate it first in
Cartesian coordinates. Differentiating the triangle function (56) twice and
evaluating it at the stationary point gives
(zp − zq )2 ((yp − yq )2 + (zp − zq )2 )
∆xx = 3
−zp zq Rpq
(zp − zq )2 ((xp − xq )2 + (zp − zq )2 )
∆yy = 3
(99)
−zp zq Rpq
−(zp − zq )2 (xp − xq )(yp − yq )
∆xy = 3
.
−zp zq Rpq
Consequently
(zp − zq )6
∆xx ∆yy − ∆2xy = 2 2 4 . (100)
zp zq Rpq
The root of this determinant is transformed multiplying it by the functional
determinant taken also at the point of stationarity
q q ∂(x, y)
∆ξξ ∆ηη − ∆2ξη = ∆xx ∆yy − ∆2xy . (101)
∂(ξ, η)

25
Additional terms do not occur. They would contain first derivatives of the
triangle function which are zero at the stationary point, see conditions (70).
The result is astonishing:

2πi A(ξs , ηs )
q = 1 + O(k −1 ) (102)
k ∆ξξ ∆ηη − ∆2ξη

though it should be noted that it is -1 if the functional determinant is nega-

tive, i.e. if the coordinates ξ,η form a left-handed system. The final formulae
for the representatives derived from (90) are thus simple and identical
q q
exp(ikRpq ) erfc −ik∆s (η− ) − erfc −ik∆s (η+ )
ak (rp ) = + O(k −1) (103)
Rpq 2
q q
exp(ikRpq ) erfc −ik∆s (η− ) − erfc −ik∆s (η+ )
bk (rp ) = + O(k −1) . (104)
Rpq 2

Further simplifaction is possible. If the upper edge is removed, i.e. η+ →

+∞, the exhausting function is supposed to increase beyond measure ∆s (η+ ) →
+∞ and, according to the asymptotics (74), the formulae (103-104) are re-
duced to the
Universal Formula of Diffraction. The diffracted wave is the primary
wave times a universal function describing the change from light to shadow.
√
exp(ikRpq ) erfc −ik∆s
a or bk (rp ) = + O(k −1 ) . (105)
Rpq 2

The geometry of the diffracting edge enters just through the argument of
that function. This argument is calculated from the exhausting function
(89) evaluated at the position of the edge: ∆s = ∆s (η− ).
From here one may return to the apparently more general formulae above
due to the linearity of Maxwell’s equations and a superposition of their so-
lutions. So ∆s may be understood as an abbreviation for ∆s (η− ) or ∆s (η+ )
as required. By the way linearity: Factors on the right-hand sides that do
not depend on rp are always free. Your author omitted them. This is all the
more tolerable since the representatives get effective only if taken together
with the supporting vector field t in equations (59) and (60). According to
(33) the tangent vector t contains the components tx and ty . They can be
chosen to adjust strength and polarization of the primary wave according to
experimental givens.
Formula (105) constitutes the main achievement of this article.

26
 One may check the accuracy of (105) by insertion into the Helmholtz
equations (41). One finds that (105) satisfies these equations in the order of
k 2 identically, but the terms proportional to k 3/2 cancel only if the exhausting
function ∆s satisfies an equation of eikonal type.
rp − rq
(∇p ∆s )2 = −2 ∇p ∆s . (106)
Rpq

Indeed, the fulfillment of this equation follows from the definition of the
triangle function (61) and the condition of utter exhaust (88). Hence the
largest erroneous term is O(k). But since the order of the Helmholtz operator
is k 2 , the relative error is O(k −1 ), as expected.
Checking the fulfillment of the boundary conditions yields at first sight a
worse result, namely a relative error of O(k −1/2 ). We see this from the asymp-
totic expansion of the complementary error function (74). This asymptotic
expression applies because, directly behind the √ screen, the probe resides in
the shadow where there √ is, according to (65), ∆s > 0. Thus the argument
of the error function −ik∆s lies in the right-hand side of the complex plane.
Hence
exp(ikRpq ) exp(ik∆s )
ak (rp ) = √ + O(k −3/2 ) (107)
Rpq 2 −iπk∆s
instead of an exact zero as required in the boundary conditions (36). A
similar result holds for the representative bk (rp ). To verify its boundary
condition, one must calculate its normal derivative, i.e. its derivative with
respect to z. While this derivative generally is O(k), its value on the screen
is O(k 1/2 ) such that the relative error is O(k −1/2 ).
At the second view, both results are better than expected. First, in
Kirchhoff’s scalar theory of diffraction, the probe must not approach the
screen since Kirchhoff’s functions become singular [6, p.9]. In the present
theory, all expressions stay regular. This facilitates, as we will in section
11, a simple trick to suppress errors on the boundaries. Second, formula
(105), even as it is now, satisfies the boundary conditions not exactly, but
accurately. The reason is the product k∆s in the denominator of (107). The
exhausting function ∆s measures, in a slightly unusual way, the distance of
the probe from the border of shadow. The distance between that border and
the screen is the largest possible distance available in a diffracting system.
Hence for short waves, both the wave number k and the distance ∆s are big
making (107) negligibly small.
For applications it is worthwile to calculate from (105) the field strengths
according to (59) and (60). For simplicity we let ak (rp ) = 0. Then the
electric field Ek (rp ) stems only from the simple curl in (60) and the magnetic

27
field Bk (rp ) only from the double curl in (59).
√
Rpq (rp − rq ) × t erfc −ik∆s
Ek (rp ) = ωk +
exp(ikRpq ) Rpq 2
s
∇p ∆s × t iω 2 k
√ exp(ik∆s ) + O(k) (108)
∆s 4π

2
√
Rpq (rp − rq )((rp − rq )t) − tRpq erfc −ik∆s 2
Bk (rp ) = 2
k +
exp(ikRpq ) Rpq 2
s
∇p ∆s ((rp − rq )t) + (rp − rq + Rpq ∇p ∆s )(∇p ∆s t) ik 3
√ exp(ik∆s )
Rpq ∆s 4π
+O(k) . (109)

The ubiquitous primary wave was drawn to the left-hand sides to direct
attention to the nontrivial terms on the right-hand sides.
We learn from these equations that diffraction of electromagnetic waves
without polarization does not exist. Yet the contributions to the right-hand
sides with the complementary error function describe just the vector proper-
ties of the primary wave. The changes of polarization effected by the screen
are described by the contributions with the exponential function. The con-
tributions with the error function are proportional to k 2 since ω = O(k),
cf. equation (43), whereas those with the exponential function are propor-
tional only to k 3/2 . Nevertheless one should not be deluded that polarization
by diffraction is always a lower-order effect. In the shadow the error func-
tion decreases dramatically. It ceases to be O(1) and weakens to be only
O(k −1/2 ) as indicated in equation (107). In the shadow both contributions
to the right-hand sides of (108) and (109) reach comparable sizes. We must
expect hitherto unseen effects.
All terms in (108) and (109) remain finite in the entire domain of solution
except, of course, immediately at the edge where Bremmer’s √ effect takes
place. The singularities on the border of shadow caused by ∆s in the
denominators are cancelled by the derivatives ∇p ∆s in the nominators since
∆s varies quadratically in the vicinity of the border.
For radiation with low frequencies, e.g. microwaves, the electromagnetic
fields can be measured directly. However, the Maxwell equations are real
equations for real observables. Thus we must extract from the fields (108)-
(109) their real parts prior to comparison with experimental data. In optics
with its much higher frequencies, the classic measurements are all calorimet-
ric ones. One collects the energy flux impinging on a given surface for times

28
much longer than inverses of these frequencies. The observable is here the
time average of the Pointing vector S
1 τ dt 1
Z
S̄(rp ) = lim ℜE(rp, t) × ℜB(rp , t) = ℜ(Ek (rp ) × B∗k (rp )) . (110)
τ →∞ τ o µ 2µ
The relation between time-dependent and time-independent fields is as
defined in equations (39). The asterisk denotes complex conjugation; one
can affix it either to the magnetic or to the electric field without affecting
the result. The a priori decomposition of the complex fields (108-109) is not
needed here.

10 Diffraction by a straight edge

The first application is, of course, diffraction by a straight edge. For its
description, the Cartesian coordinate system is best. Hence ξ = x, η = y
and
− ∞ < x < +∞ − ∞ < y < y− = const z=0 (111)
defines the screen. y− is the location of the edge.
The exhausting dependence is found from equations (88) and (56)

∂x Rp (x, y) = −∂x Rq (x, y) . (112)

Squaring this equation and cancelling identical terms on both sides results
in a quadratic equation for x

(xp − x)2 ((yq − y)2 + zq2 )) = (xq − x)2 ((yp − y)2 + zp2 )) . (113)

y must be considered as given. One solution of this equation is a mathemat-

ical ghost. It appears because the original equation (112) was squared to get
rid of the roots. The other solution, however, does solve the original. It is the
searched-for exhausting dependence ξ = Ξs (η) (88) in Cartesian coordinates:
q q
xp (yq − y)2 + zq2 + xq (yp − y)2 + zp2
x = Xs (y) = q q . (114)
(yq − y)2 + zq2 + (yp − y)2 + zp2

Inserting this dependence in the triangle function (56) gives the exhausting
function of the straight edge
r q q
∆s (y) = (xp − xq )2 + ( (yp − y)2 + zp2 + (yq − y)2 + zq2 )2
q
− (xp − xq )2 + (yp − yq )2 + (zp − zq )2 . (115)

29
We take this
q function at the edge y = y− and need to determine the sign
√
of ∆s = ∆s (y− ), i.e. where there is shadow or light. Shadow prevails,
according to the criterion (66), when ys < y− . With ys from equation (63)
we find
zp
yp < y− + (y− − yq ) . (116)
−zq
Hence equation (65) reads here
( √
+|√∆s | if yp < y− − (y− − yq )zp /zq with zp > 0,
q
∆s = (117)
−| ∆s | elsewhere.

What remains to be done is to insert this in the universal formula of diffrac-

tion (105).

11 Imaging by diffraction
We have now enough experience to perform the simple trick that was an-
nounced in section 9. To improve the fulfillment of boundary conditions, one
inserts a mirrored picture of the source in the same way as Sommerfeld did
when he calculated suitable Green functions, see section 5. If rq is the posi-
tion of the source, rm has the same coordinates xq and yq , but the opposite
value of zq as explained in (51). Let the distances Rpq and Rpm be defined
according to (55):
q
Rpq = (xp − xq )2 + (yp − yq )2 + (zp − zq )2
q
Rpm = (xp − xq )2 + (yp − yq )2 + (zp + zq )2 . (118)

Hence original and mirrored distance become equal, lim Rpq = lim Rpm , if
the probe approaches the screen, zp → 0; it does not matter if the approach
takes place in inner or outer space, zp > 0 or zp < 0.
We must, however, be careful when we introduce the mirrored exhausting
function. In the definition of the original exhausting function taken at the
edge y = y−
r q q
∆sq = (xp − xq )2 + ( (yp − y− )2 + zp2 + (yq − y− )2 + zq2 )2
q
− (xp − xq )2 + (yp − yq )2 + (zp − zq )2
 q
q  +| ∆sq | if yp < y− − (y− − yq )zp /zq with zp > 0,
∆sq = q
 −| ∆ |
(119)
sq elsewhere.

30
the assignment of signs of the root belongs to the definition, cf. equations
(115) and (117). In about three quarters of space, the root of the original
exhausting function is negative, viz. for all negative values of zp and, if yp is
sufficiently large, for positive values of zp too. The source lights the major
part of space.
By contrast, the mirrored exhausting function must be defined as
r q q
∆sm = (xp − xq )2 + ( (yp − y− )2 + zp2 + (yq − y− )2 + zq2 )2
q
− (xp − xq )2 + (yp − yq )2 + (zp + zq )2
( √
−|√∆sm | if yp < y− + (y− − yq )zp /zq with zp < 0,
q
∆sm = (120)
+| ∆sm | elsewhere.
Almost everything follows from (119) replacing zq with −zq . The only excep-
tion is the assignment of signs of the root. Using the freedom mentioned in
the discussion behind equation (65), your author reversed it. In about three
quarters of space, the root of the mirrored exhausting function is positive,
viz. for all positive values of zp and, if yp is sufficiently large, for negative
values of zp too. Shadow prevails.
Yet immediately over and under the screen, the signs of original and
mirrored functions are the same, lim ∆sq = lim ∆sm , if the probe approaches
the screen, zp → 0; it does not matter if the approach takes place in inner or
outer space, zp > 0 or zp < 0.
With these functions it is straightforward to construct solutions that sat-
isfy the boundary conditions (36) exactly. Instead of (105) we obtain
q √
exp(ikRpq ) erfc −ik∆sq exp(ikRpm ) erfc −ik∆sm
ak (rp ) = − + O(k −1)
Rpq 2 Rpm 2
(121)
q √
exp(ikRpq ) erfc −ik∆sq exp(ikRpm ) erfc −ik∆sm
bk (rp ) = + + O(k −1 ) .
Rpq 2 Rpm 2
(122)
The symbol O(k −1 ) was retained to remember that the differential equations
(41) are not exactly fulfilled.
The second terms on the right-hand sides of equations (121) and (122)
considered isolated appear fantastic. They describe ghostly radiation from
a source at rm that becomes bright only when it permeates the screen. Yet
when they are considered in cooperation with the first terms, it is recognized
that they describe the unavoidable reflection that is diffracted in a similar
way as the primary wave. The solutions (121) and (122) are valid in entire
space.

31
 From these solutions it follows that radiation emitted at rq and diffracted
by a screen causes a second singularity at rm . In other words, diffraction
creates a sharp image. Of course, the image is as weak as the multiplicative
error function indicates, but it should be observable because the singularity
sticks out.
Folks might feel this prediction as daring. They might not appreciate the
immense power of the methods developed here. The approximate solution
(121-122) contains, as a limiting case, the only exact solution of diffraction
problems known so far, namely Sommerfeld’s celebrated stringent solution
[13].
Sommerfeld’s solution deals with the diffraction of a plane wave. Using
the definitions in (55) and (118), Rpq = |rp − rq | and Rpm = |rp − rm |, the
spherical waves in front of the error functions in (121) and (122) can be
expanded as
exp(ik|rp − rq |) exp(ik|rq |) −rq
 
= exp ik rp + O(|rq |−2 ) (123)
|rp − rq | |rq | |rq |
exp(ik|rp − rm |) exp(ik|rq |) −rm
 
= exp ik rp + O(|rq |−2 ) . (124)
|rp − rm | |rq | |rq |
The first factors on the right-hand sides do not depend on the position of
the probe rp . They are absorbed in the normalizations of (121) and (122).
When the source is moved to infinity, |rm | = |rq | → ∞, only the plane waves
remain. Their fronts are defined by the normal vectors
−rq −rm
cq = lim cm = lim (125)
|rq |→∞ |rq | |rq |→∞ |rq |

with directional cosines as components

cq = ex cx + ey cy + ez cz cm = ex cx + ey cy − ez cz . (126)
Likewise the exhausting function for the plane-wave problem is defined
as the limit δs (y) = lim ∆s (y) for |rq | → ∞. The straightforward calculation
transforms (115) to
q q
δs (y) = c2y + c2z (yp − y)2 + zp2 − cy (yp − y) − cz zp . (127)

Taking the value on the edge y = y− yields the analog of (119)

q q
δsq = c2y + c2z (yp − y− )2 + zp2 − cy (yp − y− ) − cz zp
 q
q  +| δsq | if yp < y− + zp cy /cz with zp > 0,
δsq = q (128)
 −| δ | elsewhere.
sq

32
and the mirrored exhausting function as analog of (120)
q q
δsm = c2y + c2z (yp − y− )2 + zp2 − cy (yp − y− ) + cz zp
( √
−|√δsm | if yp < y− − zp cy /cz with zp < 0,
q
δsm = (129)
+| δsm | elsewhere.
Thus the analogs of (121) and (122) are
q √
erfc −ikδsq erfc −ikδsm
ak (rp ) = exp(ikcq rp ) − exp(ikcm rp ) (130)
2 2
q √
erfc −ikδsq erfc −ikδsm
bk (rp ) = exp(ikcq rp ) + exp(ikcm rp ) . (131)
2 2
These functions form an exact description of diffraction. Not only the bound-
ary conditions are satisfied perfectly, also the Helmholtz equations (41) as
direct checking shows. One may be surprised at this, but it becomes com-
prehensible when it is noticed that in this problem with the infinite edge
and the primary plane wave, the only length is provided by the position of
the probe rp . One can introduce new variables of position instead of rp by
scaling the latter with with the inverse of the wave number k. Hence if the
solution is known for one wave number, here for k → ∞, it is known for all
wave numbers.
Based on this argument, one supplements the proof of the fact that the
two integrals in (53) and (54) are equal to ∓2π/zp Rpq , respectively. There
is no edge at all. The only length is the distance between source and probe.
One evaluates the integrals for k → ∞ using stationary phase according to
section 8 and introduces thereafter scaled variables to show that the result
does not depend on k.
Sommerfeld’s q solution is a special case of (130-131) for flat incidence,
i.e. cx = 0 thus c2y + c2z = 1. In this case, bk (rp ) is proportional to the
magnetic field Bk (rp ) = −ex k 2 bk (rp ). The factor −k 2 is swallowed by nor-
malization. In the problem with the other polarization, the electric field
Ek (rp ) is proportional to ex ak (rp ). Therefore Sommerfeld got along with-
out the representation theorem of section 2. Moreover, Sommerfeld neither
knew the method of stationary phase for two-dimensional integrals nor did
he use the standardized complementary error function although it does not
take much work to show that Sommerfeld’s function is equivalent. Sommer-
feld found his function via an ingenious contour integration, an approach
which does not seem to admit generalization. Finally, it played a role that
Sommerfeld was Felix Klein’s pupil. Klein was the most influential advo-
cate of uniformization meaning that roots had to be avoided because of their

33
ambiguities and to be replaced with suitable parametrizations necessitating
transcendental functions. So Sommerfeld introduced, instead of the roots in
(128) and (129), artifical angles, sines and cosines. Probably this rootopho-
bia is the reason why Sommerfelds exceedingly important solution - it is the
‘harmonic oscillator’ of diffraction - was never sufficiently appreciated and is
not presented in most modern textbooks.
Though Sommerfeld’s solution holds only for flat incidence cx = 0, it is,
due to translational invariance, easily generalized for skew incidence. This
was done long before this article was drafted [3, § 11.6]. Moreover, the solu-
tions (130) and (131) can be conceived as Fourier components and used to
build the most general diffraction at the straight edge. For example, diffrac-
tion of a spherical wave, which is described by equations (121) and (122) only
asymptotically, was calculated by Macdonald [35]. Yet Macdonald’s formu-
lae are so difficult to survey that imaging by diffraction might have escaped
notice. The present approach is simpler and generalizable.

12 Diffraction by a slit
Let us clarify next one of most lugubrious offences of traditional theory: the
difference between Fresnel and Fraunhofer diffraction. To anticipate the an-
swer, Fraunhofer diffraction is a misnamer. In the proper sense of diffraction,
Fraunhofer diffraction does not exist. There is at best interference. It occurs
only when a primary wave strikes an aperture with at least two opposite
edges. The thus excited secondary waves interfere to generate a pattern now
named Fraunhofer diffraction. The formulae derived in this article are pow-
erful enough to describe the gradual transition from Fresnel diffraction to
Fraunhofer interference. All is included in the equations (103) and (104).
To understand how the transition proceeds, the simplest example suffices:
(i) Damping negligible; imaginary part ℑk → 0; the wave number k is a
positive number. (ii) Diffraction by a straight slit with edges at y± = ±y0 ;
2y0 is a positive number meaning the width of the slit. (iii) Normal incidence
of a plane wave; thus cx = cy = 0, cz = 1, see (125-127); in equation (104), the
plane wave replaces the spherical wave; consequently the exhausting function
∆s gives way to its descendant δs (127)
q q
erfc −ikδs (y− ) − erfc −ikδs (y+ )
bk (rp ) = exp(ikzp ) + O(k −1 ) (132)
2
with
q
δs (y± ) = (yp ∓ y0 )2 + zp2 − zp

34
  q
q  +| δs (y± )| if yp < ±y0 ,
δs (y± ) = q (133)
 −| δ (y )| elsewhere .
s ±

(iv) The representative ak (rp ) is disregarded. (v) Effects of the reflected wave
are neglected; zp > 0 is implied.
Equations (132) and (133) are still powerfull as the probe can be moved
freely within the inner space zp > 0. Especially the equations remain valid
when the probe approaches the screen zp → 0. In most classical experiments,
however, the probe is far away from the screen, zp → ∞. Therefore the
exhausting function (133) can be expanded in terms of inverse powers of zp
yp2 ∓ 2yp y0 + y02  yp ∓ y0 2

δs (y± ) = 1 + O(( )) . (134)
2zp zp
It is worthwhile to mention that using the approximated exhausting function
(96) yields the same result. The linearization (94-95) spoils the approach
to the screen. In the error functions of (132), which oscillate quickly, the
first two terms on the right-hand side of (134) must be retained because only
the second term will yield specific results. But in algebraic expressions, the
leading term will be enough, for example
√ 
1 2zp y0 yp 2

q = 1 + O( ) + O(( ) ) . (135)
| δs (±y0 )| |yp| yp zp

To pacify simple-minded scientists’ dismay at the error functions in (132),

let us return to elementary functions using the asymptotic expansion (74).
Because of (133) we must do it differently for yp → −∞ and yp → +∞. For
the latter case, equation (75) has to be considered before (74) can be applied
q exp(ikδ(±y0 ))
erfc −ikδs (±y0 ) ∼ 2 − √ q for yp → +∞ . (136)
−iπk | δ(±y0 )|
The minus signs from the last line of (133) and of the argument in (75)
cancel. Opticians interprete this formula as representing the primary wave
by the 2 and a secondary wave radiated from the edge by the exponential
function. Therefore it is a formula for the domain of light. This seems to
be a contradiction to physics since the domain of light is limited to |yp | < y0
while we suppose here yp → +∞. The contradiction is quashed since the two
leading 2’s cancel in (132).
Calculation based on (134-136) yields
v
√ y 2 u 2ky02 sin(ky0 yp /zp )
 u
bk (rp ) ∼ −i exp ik(zp + p ) t . (137)
2zp πzp ky0 yp /zp

35
Notice the sine is brought about by the interference of the two terms in (132).
The formula is symmetric with respect to yp . It holds both for positive and
negative values. A simple calculation for yp → −∞ along the lines just
sketched confirms this.
For a comparison with experiments, we need the time-averaged Pointing
vector (110). The probe usually is an absorbing layer spanned perpendic-
ularly to the energy flux of the primary wave. The representative in (137)
produces via (59) and (60) with t = ex the electromagnetic fields

Bk (rp ) = −ex k 2 bk (rp ) Ek (rp ) = ey ωkbk (rp )(1 + O(zp−1 )) + ez ... . (138)

The dots behind ez substitute a function we presently do not need to know.

The measured component of the energy flux is thus

ωk 3
ez S̄(rp ) ∼ |bk (rp )|2 . (139)
2µ

S̄0 = ez ωk 3 /(2µ) is, because of the normalization chosen in (132), the energy
flux of the primary wave. Straightforward evaluation of (139) with (137)
gives
ky0 sin(ky0 yp /zp ) 2
 
ez S̄(rp ) ∼ |S̄0 | 2y0 . (140)
πzp ky0 yp /zp
This is the most fabulous formula of traditional diffraction theory - Kirch-
hoff’s formula of Fraunhofer diffraction. However, Kirchhoff found only the
factor with the squared sine. Historically, various normalizing factors were
adjusted a posteriori, but the fabulous formula was never perceived as an
equation among vectors.
Here, by contrast, everything arises from first principles. The last two
factors in (140) when integrated over yp from −∞ to +∞ yield 1. The first
two factors describe the primary energy flux squeezed through the slit of
width 2y0 . Equation (140) expresses energy conservation as warranted by
Maxwell’s equations in a medium without damping.
The fabulous formula is worse than admitted in textbooks. From the
estimate in (135) we see it holds only for

y0 << |yp | << zp . (141)

y0 << zp means that the probe is far away from the screen. This is all right
because zp → ∞ was announced at the start of the calculation. But the
extendend condition (141) entails that Kirchhoff’s formula provides neither
a valid description of the central peak around |yp | ≤ y0 nor of large-angle
diffraction |yp | > zp . Only some intermediate wiggles are correctly seized.

36
That the condition y0 << |yp | matters, is corroborated through the use of the
asymptotic expansion (136). It is utterly wrong at values of the exhausting
function close to zero. The triangle function and thus the exhausting function
of optics is, according to (64), zero on the border of shadow defined here by
y0 = |yp|.
The positive outcome of the preceding discussion is better understanding
of the working of formula (132) and its prototypes (103) and (104). Let us
position the probe first close to an edge. To describe this measurement, one
of the terms in (132) is enough. We must, however, utilize the error function
full-fledged. When the probe is removed from the edge, but kept close to the
screen, we are still in the realm of pure diffraction. Still one term of (132)
suffices, but we may use now its asymptotic expansion (74). Only in that part
of space where distances to the edges are about equal, interferences show up.
Both terms in (132) must be kept. When the distances become large, we may
use asymptotic expansions, but not at the borders of shadow. With increasing
distances from the edges, the diffracted beams spread. In particular this is
true close to the borders. The central beam is enclosed between these borders.
Therefore it is always questionable to decribe diffraction and interference in
the vicinity of the central beam using the asymptotic expansion (74).

13 Diffraction by a circular aperture

After it is confirmed that the new theory encloses correct parts of the tradi-
tional theory as special cases, we are free to step into the entirely unknown.
The round stop is the most often installed component in optical instru-
ments. We want to calculate the diffraction it causes, more precisely its
Fresnel diffraction, since some information on Fraunhofer interference is al-
ready known - Airy’s formula [3, § 8.5.2]. The edge of a circular aperture is
not straight. Nevertheless the universal formula of diffraction (105) applies
again. All we have to do is to find the suited exhausting function.
The suitable coordinate system is the cylindrical one introduced in section
6, see (67).
− π < ϕ ≤ +π 0 ≤ ρ < ρ0 = const z=0 (142)
defines a circular aperture in the screen. The exhausting dependence is found
from equations (88) and (56)
∂ϕ Rp (ρ, ϕ) = −∂ϕ Rq (ρ, ϕ) (143)
with
q
Rp (ρ, ϕ) = ρ2p + ρ2 − 2ρp ρ cos(ϕp − ϕ) + zp2

37
 q
Rq (ρ, ϕ) = ρ2q + ρ2 − 2ρq ρ cos(ϕq − ϕ) + zq2 . (144)

For normal incidence, ρq = 0, one can observe the solutions using the rubber-
ribbon experiment explained in section 8. Actually, there are two:

ϕ = ϕp and ϕ = ϕp + π . (145)

These exhausting dependencies ϕ = Φs (ρ), analogues of ξ = Ξs (η) in (88),

surprise as they do not depend on ρ. Inserting them into the triangle function
(56) gives the exhausting function of the circular stop. We need it at the edge
ρ = ρ0 :
q q q
∆s = (ρp − ρ0 )2 + zp2 + ρ20 + zq2 − ρ2p + (zp − zq )2 . (146)

The definition has to be completed with the disposal of the root

( √
+|√∆s | if ρp > ρ0 (zq − zp )/zq with zp > 0,
q
∆s = (147)
−| ∆s | elsewhere,

according to equations (65) and (68). The exhausting function of the second
solution in (145) is derived from this replacing ρ0 with −ρ0 . Your author
names the first solution with the positive ρ0 near because its point on the
edge is closer to the probe than in the opposite case. The second solution
with the negative ρ0 is thus called far.
For mathematical reasons, only the near solution may dominate. The
definition (142) of cylindrical coordinates restricts the range of the angle ϕ.
Yet unrestricted variation of the variable ξ, analog of ϕ, is a premise of the
principle of utter exhaust as declared in section 8. The violation of this
premise is not disastrous as long as the exhausting function ∆(ξ, η) takes a
sharp extremum on the edge. In this case, the finiteness of integration causes
negligible corrections. But for the round stop ∆(ξ, η) becomes equal for all
points on the edge when source and probe both approach the optical axis;
there is no extremum of ∆(ξ, η) at all. Consequently the universal formula
of diffraction (105) applied to circular apertures needs ρp > zp > 0 to secure
its applicability.
Of course, the primary wave is also diffracted at the far side of the circu-
lar stop. There will be similar interferences as those caused by the slit, see
section 12, but the contributions from the far side are here, in the range of va-
lidity just declared, small. Fraunhofer interferences cannot be described well.
Nevertheless, within the restricted range, there seems to be no rival of the
formulas (146) and (147). We can calculate diffraction immediately behind
a circular aperture and predict what a curved edge does to polarization.

38
 For general parameters, the exhausting equation (143) cannot be solved
in terms of roots. When the transcendental functions sine and cosine are
rationalized introducing, for instance, the variables c and s
ϕp + ϕq ϕp + ϕq
c = cos (ϕ − ) s = sin (ϕ − ) (148)
2 2
where c2 + s2 = 1, and when the roots are removed by squaring, an algebraic
equation of the sixth degree is obtained.
Nevertheless we can find more physically relevant solutions, even for
oblique incidence, performing the rubber-ribbon experiment as above. Namely
the one-sided flat situation where source and probe sit on the same side of
the optical axis, and the two-sided flat situation. In both situations, there
are near and far solutions:

one-sided flat: ϕp = ϕq near: ϕ = ϕp far: ϕ = ϕp + π (149)

two-sided flat: ϕp = ϕq + π near: ϕ = ϕp far: ϕ = ϕp + π . (150)

But this is not enough when vector fields as in (108) and (109) are to be
calculated. For them we must differentiate the exhausting function with
respect to zp , ρp and ϕp . The last differentiation is not feasible when the
function is only known for a fixed ϕp .
Mending the shortcoming is easy. The sine S and the cosine C were
defined in (69). In the context of (143), they must be considered as given
and fixed. The sine s and the cosine c were defined in (148). They contain
the angle which is sought as the exhausting dependence ϕ = Φs (ρ). With
the abbreviations
q q
P (ρ) = (ρp − ρ)2 + zp2 /ρp Q(ρ) = (ρq − ρ)2 + zq2 /ρq (151)

the exhausting equation (143) can be written in four equivalent variants

Cs − Sc Cs + Sc
q = −q .
P (±ρ)2 − 2ρ(Cc + Ss ∓ 1)/ρp Q(±ρ)2 − 2ρ(Cc − Ss ∓ 1)/ρq
(152)
Differently to all other equations in this article, signs can be chosen indepen-
dently. Just under the roots, the signs must be altered coherently.
These variants permit to find solutions in the neighborhood of (149) and
(150) such that they are exact in linear terms. From (69) we have
one-sided: S = O(ϕp − ϕq ) C = 1 + O((ϕp − ϕq )2 )
two-sided: S = 1 + O((ϕp − ϕq − π)2 ) C = O(ϕp − ϕq − π) .
Inserting the estimates according to (149-150)

39
 one-sided near: ϕ = ϕp + O(ϕp − ϕq )
one-sided far: ϕ = ϕp + π + O(ϕp − ϕq )
two-sided near: ϕ = ϕp + O(ϕp − ϕq − π)
two-sided far: ϕ = ϕp + π + O(ϕp − ϕq − π)
in the definitions (148) yields
one-sided near: s = O(ϕp − ϕq ) c = 1 + O((ϕp − ϕq )2 )
one-sided far: s = O(ϕp − ϕq ) c = −1 + O((ϕp − ϕq )2 )
2
two-sided near: s = 1 + O((ϕp − ϕq − π) ) c = O(ϕp − ϕq − π)
2
two-sided far: s = −1 + O((ϕp − ϕq − π) ) c = O(ϕp − ϕq − π) .
Hence the nominators in (152) are always small of first order. In the de-
nominators, P (±ρ) and Q(±ρ) are positive constants. But the expressions
Cc + Ss ∓ 1 and Cc − Ss ∓ 1 are always small of second order if signs are suit-
ably selected. Thus these expressions can be neglected. Without extertion
one finds now the exhausting dependencies for the one-sided flat situation

P (ρ) − Q(ρ) P (−ρ) − Q(−ρ)

near: s = −S far: s = S (153)
P (ρ) + Q(ρ) P (−ρ) + Q(−ρ)

and for the two-sided flat situation

P (ρ) + Q(−ρ) P (−ρ) + Q(ρ)
near: c = −C far: c = C . (154)
P (ρ) − Q(−ρ) P (−ρ) − Q(ρ)

The exhausting function of the one-sided flat situation, based on the near
solution and taken at the edge ρ = ρ0 , is, when abbreviations are restored,
q q
∆s = (ρp − ρ0 )2 + zp2 + (ρq − ρ0 )2 + zq2
q
− (ρp − ρq )2 + (zp − zq )2
q q
 (ρp − ρ0 )2 + zp2 (ρq − ρ0 )2 + zq2
+2 ( + )−1
ρp ρ0 ρq ρ0
q
(ρp − ρq )2 + (zp − zp )2 −1

ϕp − ϕq
−( ) sin2 (155)
ρp ρq 2
√ √
∆s = +| ∆s | if

ρ0 (zq − zp ) + ρq zp ρq zp ϕp − ϕq
ρp > (1 + 2 sin2 ) with zp > 0, (156)
zq ρ0 (zp − zq ) 2
√ √
∆s = −| ∆s | elsewhere.
The equations (68-69) were used to determine the border of shadow.

40
 The respective expressions the two-sided flat situation are
q q
∆s = (ρp − ρ0 )2 + zp2 + (ρq + ρ0 )2 + zq2
q
− (ρp + ρq )2 + (zp − zq )2
q q
 (ρp − ρ0 )2 + zp2 (ρq + ρ0 )2 + zq2
+2 ( − )−1
ρp ρ0 ρq ρ0
q
(ρp + ρq )2 + (zp − zp )2 −1

ϕp − ϕq
+( ) cos2 (157)
ρp ρq 2
√ √
∆s = +| ∆s | if
ρ0 (zq − zp ) − ρq zp ρq zp ϕp − ϕq
ρp > (1 − 2 cos2 ) with zp > 0, (158)
zq ρ0 (zp − zq ) 2
√ √
∆s = −| ∆s | elsewhere.
All exhausting functions based on the far solutions can be obtained from
these formulas inverting the sign of ρ0 . The errors are O((ϕp − ϕq )4 ) or
O((ϕp − ϕq − π)4 ), respectively.
Generally, one may solve equation (152) numerically. The effort for this
is orders of magnitudes less than for a direct numerical solution of Maxwell’s
equations.

14 Outlook
As we all know, everthing was foretold by Aristotle or Jesus Christ. So it is
only with timidly throbbing heart that your authors dares to ask the deciding
question: What might be novel in this article?
All monographies and textbooks pretending to teach the solution of Max-
well’s equations should be committed to paper recycling. The same applies
to those treatises wherein optics is allegedly derived from electrodynamics.
The general representation theorem in section 2 beats them all. That the
boundary conditions on metallic surfaces are as simple for the representa-
tives as deduced in section 3, this was never anticipated. These discoveries
are synthesized with Sommerfeld’s criticisms and proposals in the sections 4
and 5. The thus derived solution of electromagnetic diffraction polarization
inclusive, described in section 5, is a new finding. Compared to this, the
discussion of the triangle function in section 6 appears at first sight of minor
importance. But the root of the triangle function and a consistent assignment
of its signs are of prime importance for applications. The big thing, however,

41
is the principle of utter exhaust declared in section 8. For the first time,
stationary phase is applied where it is really needed. Utter exhaust seems to
have no predecessors at all. It is so powerful, it would be explained in dozens
of textbooks if it was known. Utter exhaust shows its power in the universal
formula of diffraction, see section 9. Its application to diffraction by an edge
in section 10 yields a comprehensive formula which encloses, as a limiting
case, Sommerfeld’s stringent solution, see section 11. The great wonder is
the perfect agreement though utter exhaust is just an asymptotic evaluation
for short wavelengths. For diffraction by a slit, as discussed in section 12,
utter exhaust copes with Fresnel diffraction and Fraunhofer interference in
a unified way. The gradual transition between these fundamentally different
patterns can be calculated. The derivation of these formulae and their no-
tation is new. Everything is now easier to survey than in previous theories
The formulas in section 13 describing diffraction by a round stop, derived via
utter exhaust, are completely new.
Hundreds of physicists will find rewarding foundations for future busi-
ness. The physical discussion of the formulas derived here has only begun.
Important systems, e.g. diffraction by a hook or a cusp, were not yet an-
alyzed though they are now accessible. Moreover, the circular aperture is
an especially rich system which deserves further studies. The foundation of
its theory is the integral (58). Here it was evaluated for short wavelengths
only, but more asymptotics must be considered. Evaluation for large dis-
tances from the screen will yield Fourier integrals and thus comfort all those
who understand imaging as a sequence of Fourier and inverse Fourier trans-
forms. The asymptotics for small angles are scarcely understood though they
are most important for applications in optical systems. Quantitative com-
parisons with experiments are altogether rare. The best measurements are
probably those made with microwaves [34].
At last we have a reasonable theory of electrodynamic diffraction, but
also part of acoustics will change. The methods developed from section 3 on
apply to the velocity field of sound if the representatives ak (rp ) and bk (rp ) are
construed as scalar potentials. The boundary conditions derived in section
3 are just the most often used ones in acoustics, namely on the soft or the
hard wall, i.e. acoustic impedance zero or infinite. Almost everything remains
the same, in particular the universal formula of diffraction (105). Only the
dispersion relation of sound is different from (42) since compressible fluids
refuse the telegraph equation [17].
It is time to reconcile with Gustav’s fans. Your author knows the pit-
falls of pioneering. He reveres Gustav Kirchhoff affectionately and repents
bitterly not to have read his books first. Almost all epigone scientists copied
Kirchhoff’s ideas, and it is sometimes felt that the copies were made without

42
understanding. In other words, modern scientists do not attain Kirchhoff’s
intellectual level. It is, however, remarkable that just Max Born copied Kirch-
hoff’s. Born is reputed as the founder of quantum mechanics in so far as he
invented the probabilistic interpretation. A careful comparison of Born’s
monography [2] with Kirchhoff’s textbook on optics [6] revealed that Born
copied the probabilistic interpretation, too. Kirchhoff, an adherent of Ger-
man idealistic philosophy according to which material objects descend from
ideas, made a theory for Verrückungen, i.e. for nonobservable items. The
reader might remark the contrast to Maxwell who based his theory on mea-
surable quantities, namely on the electromagnetic fields. Nevertheless Kirch-
hoff, being also a physicist, desired to compare his ideas to measurements.
So he stipulated, without any reason, that the squares of those Verrückungen
were intensities, i.e. nonnormalized probabilities: “Von dieser hängt die In-
tensität J des Lichtes ab. Als Maass derselben führen wir den Mittelwerth
des Quadrates der Verrückung eines Aetherteilches ein” [6, p.9]. Kirchhoff
is the true inventor of the probabilistic interpretation, but the invention was
not a sign of mental strength, but lack of ability to solve the correct differ-
ential equations correctly. As shown in this article, in particular in section
12, electrodynamics do not need a probabilistic interpretation.
The Dirac equation, actually a system of coupled partial differential equa-
tions, does the same for electronic motion as Maxwell’s equations do for pho-
tonic propagation. A probabilistic interpretation is not necessary because the
Dirac equation describes observables, namely the external electrical charge
and current densities which were discussed in section 2. The Dirac equa-
tion even defies that interpretation. Eighty years after Dirac’s discovery, no
compatible probability formalism for the Dirac equation is known.
Your author meanwhile derived representation theorems for the Dirac
equation [36]. They simplify its solution in the same way as the represen-
tation theorem of section 2 simplifies the solution of Maxwell’s equations.
Schrödinger’s ψ turns out to be a representative similar to the functions a
and b used here. The Schrödinger equation appears as an auxiliary simi-
lar the telegraph equation. Interpreting Schrödinger’s mathematical tools
physically appears as Verrückung again.
Moreover it is already known that several effects seeming to prove prob-
abilistic behavior in quantum electrodynamics can be derived instead from
the spatial distributions of charge and current as determined by the Dirac
equation [37]. Nevertheless, a certain group of experiments exhibits random
behavior of photons and electrons, viz. absorption by small absorbers [38].
Yet the respective dynamical problems were never solved - as pipe turbulence
was not solved before 1988 [27]. Random behavior in quantum electrodynam-
ics might stem from deterministic chaos similarly as in other areas of nature.

43
It might happen that Heisenberg’s mysticism regarding quantum “particles”
and “uncertainty” will be gone soon.

References
[1] F.Schuler, Z.Naturforsch. 63a, 571 (2008).

[2] M.Born, Optik, 2.Auflage, Springer, Berlin 1965.

[3] M.Born and E.Wolf, Principles of Optics, Cambridge University

Press, Cambridge 1997.

[4] G.Kirchhoff, Berlin.Ber. XX, 641 (1882).

[5] G.Kirchhoff, Ann.Phys. 18, 663 (1883).

[6] G.Kirchhoff, Mathematische Optik, Teubner, Leipzig 1891.

[7] E.Hecht, Optics, Addison Wesley, Upper Saddle River 2002.

[8] D.Meschede, Optics, Light and Lasers, Wiley V-CH, Weinheim

2004.

[9] K.K.Sharma, Optics, Academic Press, Amsterdam 2006.

[10] J.A.Kong, Electromagnetic Wave Theory, EMW Publishing, Cam-

brigde Mass. 2008.

[11] K.Hönl, A.W.Maue and K.Westphal, Theorie der Beugung in

Handbuch der Physik Bd.XXV/1 (ed. S.Flügge), Springer, Berlin
1961.

[12] G.Mie, Ann.Physik 25, 377 (1908).

[13] A.Sommerfeld, Math.Ann. 47, 317 (1896).

[14] A.Sommerfeld, Optik, Geest & Portig, Leipzig 1964.

[15] A.Sommerfeld, Optics, Academic Press, New York 1964.

[16] U.Brosa, Zur Lösung von Randwertproblemen mit Vektorfeldern,

Habilitationsschrift, Universität Marburg 1985.

[17] U.Brosa, Z.Naturforsch. 41a, 1141 (1986).

44
[18] S.Grossmann, Mathematischer Einführungskurs für die Physik,
7th edition, B.G.Teubner, Stuttgart 1993.

[19] J.D.Jackson, Classical Electrodynamics, 2nd edition, John Wiley

& Sons, New York 1975.

[20] M.von Laue, Röntgenstrahl-Interferenzen, Geest & Portig, Leipzig

1964, 2.Kapitel.

[21] P.Debye, Ann.Physik 30, 57, 1909.

[22] J.Meixner and F.W.Schäfke, Mathieusche Funktionen und

Sphäroidfunktionen, Springer, Berlin 1954, 357-358.

[23] L.A.Stratton and L.J.Chu, Phys.Rev. 56, 99 (1939).

[24] W.R.Smythe, Phys.Rev. 72, 1066 (1947).

[25] W.R.Smythe, Static and Dynamic Electricity, Benjamin, New

York 1969.

[26] O.Theimer, G.D.Wassermann and E.Wolf, Proc.Roy.Soc. A212,

426 (1952).

[27] L.Boberg and U.Brosa, Z.Naturforsch. 43a, 697 (1988).

[28] U.Brosa, J.Stat.Phys. 55, 1303 (1989).

[29] B.Hof, J.Westweel, T.M.Schneider and B.Eckardt, Nature 443, 59

(2006).

[30] A.Erdélyi, Asymptotic Expansions, Dover, New York 1956.

[31] M.Abramowitz and I.Stegun, Handbook of Mathematical Func-

tions, Dover, New York 1970.

[32] E.Jahnke, F.Emde, F.Lösch, Tables of Higher Functions, Teubner,

Stuttgart 1966.

[33] A.Fresnel, Œuvres complètes, Imprimerie impériale, Paris 1866.

[34] R.W.P.King and T.T.Wu, The Scattering and Diffraction of

Waves, Havard University Press, Cambridge Mass. 1959.

[35] H.M.Macdonald, Proc.Lond.Math.Soc. 14, 410 (1915).

[36] U.Brosa, to be published.

45
[37] A.O.Barut and J.Kraus, Found.Phys. 13, 189 (1983).

[38] A.Tonomura, J.Endo, T.Matsuda, T.Kawasaki and H.Ezawa,

Am.J.Phys. 57, 117 (1989).

46