E.

WOLF, PROGRESS IN OPTICS XXI

0 ELSEVIER SCIENCE PUBLISHERS B.V., 1984

RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS

BY

D. MAYSTRE
Laboratoire d'Optique Electromagnktique, Equipe de Recherche Associke au C.N.R.S. no 597,
Facultk des Sciences et Techniques,
Centre de Sainr-Jkr6me, 13397 MARSEILLE CEDEX 13, France.
 CONTENTS

PAGE

$ 1 . INTRODUCTION . . . . . . . . . . . . . . . . . . . . 3

Q 2. PRESENTATION OF THE PROBLEM AND PRELIMINARY

RESULTS . . . . . . . . . . . . . . . . . . . . . . . 5

Q 3. THE RAYLEIGH METHOD . . . . . . . . . . . . . . . 19

Q 4 . THE WATERMAN METHOD . . . . . . . . . . . . . . 36
Q 5 . THE INTEGRAL METHOD . . . . . . . . . . . . . . . 37
$ 6 . OTHER METHODS . . . . . . . . . . . . . . . . . . . 43
$ 7. METALLIC AND DIELECTRIC GRATINGS . . . . . . . 46
Q 8. THE GRATING IN CONICAL DIFFRACTION . . . . . . 56
Q 9. CONCLUSION . . . . . . . . . . . . . . . . . . . . . 57

APPENDIX A . . . . . . . . . . . . . . . . . . . . . . . . 58

APPENDIX B . . . . . . . . . . . . . . . . . . . . . . . . 59

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . 64
 8 1. Introduction

Diffraction grating was born almost two centuries ago. Its father, an
American astronomer, saw a spectrum produced by hairs placed in the threads
of two parallel screws (RITTENHOUSE [ 17861). However, it was not until 1821
that the first metallic grating was ruled by FRAUNHOFER who found the famous
grating formulagiving the directions of diffraction (FRAUNHOFER [ 18211). One
century after the discovery of RITTENHOUSE, ROWLAND [ 18821 initiated the
production of high quality ruled gratings. Recently there appeared a new type
of diffraction grating, constructed by recording on a photoresist interference
fringes produced by a laser source (RUDOLPHand SCHMAHL[1967],
LABEYRIE and FLAMAND [ 19691).
Indisputably, the diffraction grating is a valuable instrument for scientific
research. Mainly, it is used for spectroscopy and filtering. Even though the
holographic grating has not eliminated the classical, ruled grating which
remains the most suitable for important applications, it has permitted a
considerable extension of the use of gratings for industrial or scientific purposes :
wavelength selectors for tunable lasers, selective surfaces for solar energy,
masks for photolithography, beam sampling mirrors for high power lasers,
spectrometers in extreme UV or X-ray regions for Space Optics (MAYSTRE,
NEVIEREand PETIT[ 19801).
Before the second world war, the pitch of the standard gratings was large
with respect to the wavelengths of visible radiations, due to the difficulty of
manufacturing high quality gratings having high space frequency. Under these
circumstances, the properties of gratings do not depend, in practice, upon the
polarization of the incident light. This explains why by far the major part of the
theories dealing with these instruments are “scalar” and do not take the
polarization into account (see for example BORNand WOLF [ 19651). On the
other hand, there exists experimental evidence that the distribution of energy
among the various diffracted waves depends on polarization as soon as the
wavelength to pitch ratio exceeds a number of the order of 0.2 (MADDENand
STRONG[ 19581). Today, gratings having more than 1000 grooves per mm, i.e.
with groove spacings lower than 1 pm, are commonplace. When these gratings
are employed in the visible region, the wavelength of the light is of the order
4 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 1

of the groove spacing. To our knowledge, there do not exist non-rigorous

theories able to predict accurately the properties of gratings in this “resonance
domain”. As a consequence, it is necessary to achieve rigorous vector theories.
Here, we call rigorous theory any theory able to reduce the physical problem
to an equivalent classical problem of mathematics, using the elementary laws
of electromagnetism and rigorous mathematical theorems. By classical problem
of mathematics, we mean a problem whose solution can be directly obtained
on a computer using numerical analysis. Approximations are made only in this
numerical implementation of the classical mathematical problem, never before.
To point out the difference between rigorous theories and the others, a
comparison will be made between the numerical results obtained from a rigorous
theory and from a famous non-rigorous theory (BECKMANN[ 19671,
BECKMANN and SPIZZICHINO [ 19631).
The first attempt at solving the electromagnetic problem of gratings was
made at the beginning of the century (RAYLEIGH [ 19071). The Rayleigh theory
is based on an assumption, the so-called Rayleigh hypothesis, which remained
unquestioned for almost 50 years, but provoked considerable controversies
thereafter. Though the controversial aspect of the Rayleigh hypothesis has not
yet died down, we have thought it useful to present a synthesis of the theoretical
achievements on this topic in order to conclude whether or not the Rayleigh
method is a rigorous vector theory.
It is impossible to describe in detail all the rigorous vector theories of
gratings. Since achoice was necessary, we have selected the theories numerically
implemented on a computer and giving accurate results for arbitrary gratings
in a large domain ofwavelengths. We have also taken into account the possibility
of implementing these theories in other problems of optics or electromagnetism.
Practically all the methods presented here can be generalized to other problems
of scattering or propagation : diffraction by non-periodic objects, propagation
of radio waves, etc. . . . The reader will also find many references to studies in
acoustics: from a mathematical point of view, the problems of periodic
structures in acoustics and optics are very much akin. Mainly, three formalisms
will be described in detail after the Rayleigh method: the integral and differential
formalisms, the most widely used today, and the WATERMAN [ 19751formalism,
well known in electromagnetism.
It seems useful to point out the main advantages and shortcomings of each
method for numerical application. Although the opinions expressed in this
chapter are personal, they take into account the experience of many other
researchers in the field of gratings, first of all my colleagues in the laboratory
at Marseille.
1. § 21 PRESENTATION A N D PRELIMINARY RESULTS 5

8 2. Presentation of the problem and preliminary results

The physical problem is presented and the fundamental laws of electro-

magnetism show that it reduces to a boundary value problem of mathematical
physics. Some elementary properties of the diffracted field are deduced.

2.1. THE PHYSICAL PROBLEM (Fig. 2.1)

In the rectangular coordinate system Oxyz, the grating has a profile 3 of

equation y = f ( x ) , invariant on the Oz axis and periodic in x , which separates
a region R ( y > f ( x ) ) from a region R - ( y < f ( x ) ) . The period d = 2 4 K of
+

f ( x ) will be called groove spacing or pitch or period of the grating. The

maximum and minimum values of f ( x ) will be denoted by yM and ym,respec-
tively, H = yM - y , being the groove depth. The unit normal to B is denoted
by 6. Region R mostly contains air and thus its index will be assumed as being
+

equal to 1. Region R - is filled with a metal or a dielectric. An incident

monochromatic homogeneous plane wave of wave-vector k (with
k = I k I = 2 ./A, A being the wavelength), lying in the Oxy plane, illuminates the
grating with the angle of incidence 0, algebraic number measured clockwise.
Assuming a time dependence of the fields in exp( - iot), we denote by v the
complex index of region R - . In fact, for historical reasons, and also for the sake

T'
R
region
+

>
X

region R-
Fig. 2.1. The physical problem: OD = d, yM - ym = H, the angle of incidence 0 and the angle of
diffraction 0,, are both positive in this figure; n^ is the unit vector of the normal to 9 oriented
upward.
6 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 2

of simplicity, it is assumed, until 0 7, that region R - is perfectly conducting,

i.e. that the electromagnetic field vanishes in this region.
The complex vector amplitude of the incident electric field, perpendicular to
k, is denoted by a. So, this incident electric field can be written in the form:
E' = a exp(iax - ipy), (2.1)
where a and - /3 are the two components of the incident wave-vector k :
a = ksin8, p = kcos8.

The incident magnetic field can be deduced easily from (2.1) and Maxwell's
equations :
H i= b exp(iax - ipy), (2.2)
with b = k x a/wp,.
The physical problem is to find the total fields E and Hi at any point.
The diffracted fields are defined in region R by E d = E - E' and Hd =
+

= H - Hi.

2.2. THE MATHEMATICAL BOUNDARY VALUE PROBLEM FOR THE PERFECTLY

CONDUCTING GRATING

First, it is worth noting that the physical problem remains unchanged after
translation of the grating along the Oz axis; in other words, neither the grating
profile, nor the incident wave are dependent on z. As a consequence, if
E ( x , y , z) and H(x, y, z ) are the total fields corresponding to a given grating and
a given incident field, E(x, y , z + zo) and H(x, y , z + zo) will be solutions too,
regardless of the value of z,,. If, from the physical meaning, we believe that the
total field is unique, we derive that E and H are independent of z.
Now, the mathematical problem is stated using the harmonic Maxwell
equations for the diffracted fields in region R : +

V x Ed = iopoHd, V x Hd = - i o k E d ,
V Ed = 0,
0 V.Hd= 0.
We note that the last two equations, consequences of the first two
(V V x J = 0), can be eliminated, and introducing the components of the
diffracted fields on the three axes, we then have:
(2.3a)
1 9 8 21 PRESENTATION A N D PRELIMINARY RESULTS 7

a,E," = -iwpoH,d, (2.3b)

axE," - a,E," = iopoH,d, (2.3~)

a,H,d = -ioqE,", (2.4a)
&H,d = iwcoE,d, (2.4b)
&H," - a,H,d = - i m 0 E:, (2.4~)
where c,, and po are the electric permittivity and the magnetic permeability of
vacuum, the symbol a,E," denoting the partial derivative aE,"/ay. In the same
way, the boundary conditions for the fields on the grating surface
fi x (Ed + E ' ) = 0, fi.(Hd + Hi)= 0,
are linked to each other, and it suffices to write the first:
E," = -Ej, (2.5a)
nxE," - n,E,d = - (n,Ei - n,,EL). (2.5b)
It is worth noting that the linkage between these two boundary conditions is
a typical example of an elementary property which is very difficult to establish,
at least for those who are not acquainted with the theory of distributions.
The fundamental feature of equations (2.3a) to (2.5b) is that they can be
separated into two independent sets. The first set, viz. (2.3a), (2.3b), ( 2 . 4 ~and
)
(2.5a), only contains the transverse component E," of the electric field and the
components H," and H," of the magnetic field. The same remark applies to the
complementary set, but this time with the transverse component H," of the
magnetic field and the components E," and E," of the electric field. We deduce
that the general problem can be decomposed into two elementary mathematical
problems, called TE (transverse electric) or TM (transverse magnetic)
according to whether they deal with E," or H,". Equations (2.3a) and (2.3b) for
the TE problem and (2.4a) and (2.4b) for the TM problem provide a simple
means of deducing the non-transverse components from E," and H,". So,
introducing these values in the remaining equations enables us to derive two
independent sets of equations, each set containing only one function.
In the TE problem, (2.3a), (2.3b) and (2.4~)show that E,d satisfies a
Helmholtz equation:
V2E,d +- k2E,d = 0 in R . + (2.6a)
In addition, (2.5a) gives the boundary condition:
E," = - Ej = - a, exp(icrx - ipJ(x)) on 9. (2.7a)
8 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, $ 2

Similarly, for the TM problem, (2.3c), (2.4a) and (2.4b) yield the Helmholtz
equation
Q2H,"+ k2H," = 0 in R , , (2.6b)
and the boundary condition on 9 is derived from (2.4a), (2.4b) and (2.5b):
dH,"/dn = - dHi/dn
= ib,(pn,, - an,) exp(iax - ipf(x)) on 9, (2.7b)
where d/dn denotes the normal derivative on 9 Now, the question which arises
is whether the two couples (2.6a), (2.7a) or (2.6b), (2.7b) are sufficient to
determine the transverse components of the fields. To this question, the
mathematician gives a negative answer since these equations do not include
conditions at y = co. Therefore, to complete these equations, we must add the
so-called radiation condition (or outgoing wave condition), which seems to be
obvious for the physicist: when y tends to infinity, E,d and H," must remain
bounded and can be described as a superposition of outgoing plane waves
propagating upward. Since (2.6a)-(2.7b) have been obtained using the
elementary laws of electromagnetism, the physicist is tempted to deduce that
the solution of these equations satisfying the radiation condition exists and is
unique. However, it appears that its existence has never been established by
mathematicians. Yet, this question cannot be called useless: physical evidence
is not a proof of mathematical existence! On the other hand, the uniqueness
of the TE problem ((2.6a), (2.7a) and the radiation condition) can be proved
assuming the local square integrability of the field (CADILHAC [ 1980]), but the
generalization of this theorem to the TM problem is not straighforward.
Perhaps, the specialist of Helmholtz equations and boundary value problems
will be surprised to read that existence and uniqueness theorems, which have
been demonstrated for diffracting structures apparently very close to the grating,
have not been generalized to gratings (WILCOX[ 19751). In fact, the difficulty
for the grating problem lies in the fact that both region R and boundary 9
+

are unbounded. In spite of this lack of demonstration, the existence and

uniqueness of the TE and TM problems will be assumed in the following. More
generally, we assume these properties of existence and uniqueness when the
right-hand member of (2.7a) or (2.7b) is replaced by a locally square integrable
function of x.
I , § 21 PRESENTATION A N D PRELIMINARY RESULTS 9

2.3. THE TWO FUNDAMENTAL (TM and TE) PROBLEMS

Let us define the fundamental TE case of polarization (FTE) where a plane

wave of unit amplitude strikes the grating (a, = 1, a, = a,, = b, = 0). For this
particular case, the TE problem is described by (2.6a) and (2.7a) with a, = 1.
On the other hand, the TM problem of this FTE case is described by (2.6b)
and (2.7b) and is homogeneous, since b, = 0. Thanks to the uniqueness of the
solution, we can deduce that the solution H," of the TM problem for the FTE
case vanishes. So, we conclude that, in the FTE case, the dfracted electric Jield
remains parallel to Oz and is given by (2.6a), (2.7a) and the radiation condition.
The same considerations hold for a TM polarized incident wave where the
fundamental TM case of polarization (FTM) can be defined (b, = 1,
b, = by = a, = 0). Thus,for each of these two fundamental cases ofpolarization,
the problem is scalar. Let us denote by F' and F the amplitude of Ei and E," (for
the FTE case) or the amplitude of Hi and H," (for the FTM case). From the
equations of TE and TM problems, we derive that F is the solution of Dirichlet
(FTE case) or Neumann (FTM case) boundary value problems:
V2F+k2F=0 inR,, (2.8)
F = - F' = - exp(iax - ipf (x)) for the FTE case, (2.9a)
on P {
dF/dn = - dF'/dn = i(pn, - an,) exp(iax - ipf(x))
for the FTM case, (2.9b)
F satisfies a radiation condition when y + cx), (2.10)
the incident field being given by:
F' = exp(iax - ipy). (2.11)
When the FTE and the FTMproblems have been solved, the solution of a general
problem corresponding to an arbitrary polarization can be derived very simply: it
sufficesto multiply by a, the solution of the FTE problem and by b, the solution
of the FTM problem to obtain E," and H,". So, it will suffice to study these two
fundamental cases and first, we show that (2.8)-(2.10) allow us to derive well
known elementary properties of the diffracted field F.

2.4. THE PSEUDO-PERIODICITY OF THE FIELD

.In this section, it is shown that F(x, y ) exp( - iax) is a periodic function. To
this aim, it is useful to consider the function G defined by:
G(x, y ) = F ( x + d, y ) exp( - iad).
10 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, 8 2

To prove the periodicity of F(x, y ) exp( - iax) it suffices to show that

C(x, y ) = F(x, y). Owing to the uniqueness of the solution, this equality will be
satisfied if G(x, y ) obeys eqs. (2.8)-(2.10). Bearing in mind that F(x, y ) satisfies
(2.8) and (2. lo), it is clear that G(x,y ) satisfies these equations also. It is not
difficult to see that G satisfies (2.9a) or (2.9b) if the periodicity d of f(x) has
been remembered.
In conclusion, it has been proved that F(x, y ) can be expressed in the form:
F(x, y ) = exp(iax) T(x, y ) ,
where T(x,y) has a period d in x. Such a function will be called a pseudo-
periodic function, with a coefficient of pseudo-periodicity a and pseudo-period
d.
For a given grating, we shall say that two functions F and F' have the same
pseudo-periodicity when a - a' = pK, p being an integer, the pseudo-periods
d and d' being identical and equal to 2 4 K .

2.5. THE RAYLEIGH EXPANSION ABOVE THE GROOVES

Let us develop T(x,y) in a Fourier series:

T(x, Y ) = C
n
ln(y) exp(inKx),

denoting a sum from - CO to + 00. The two equations above imply:

I1

with
So, a pseudo-periodic function can be developed in a pseudo-Fourier series
of exp (i a, x).
Now, we have to find the form of tnO)). Introducing the Fourier expansion
of F in (2.8), then multiplying by exp( - iax) yields:

If y > y,, the above equation is valid for any value of x and each term of the
Fourier series must be null:
1, $21 PRESENTATION A N D PRELIMINARY RESULTS 11

Of course, the physicist is very well acquainted with the above equation. To
express t,(y), we denote by U the finite set of integers for which kZ - a: is
positive and we define:
p, = (k2- a:)'/' ifnE U
= i(a: - k2)'IZ i f n $ U. (2.12b)
So, the expression of t,(y) is given by:
tn(v) = An exp ( - ipny) + Bn exp (iflny) if y '
YM.

The first term, of coefficient A,, becomes infinite when y -+ co if n 4 U,or

represents an incoming plane wave propagating toward the grating surface if
n E U.Thus, the radiation condition means that the A , vanish and the diffracted
field F(x, y ) can be expressed as a series R ( x , y ) of plane waves R,(x, y ) :
F(x,y)= R(x,y)= c R,(x,y)
n
= 1 B,@,(X,Y)
n
ifY>Yh4 (2.13)

with @,(x, y ) = exp(ia,x + ip,y). (2.14)

The above result is very simple: above the top of the grooves, the dffractedjield
can be represented by the so-called "Rayleigh expansion"(RE)used by RAYLEIGH
[ 19071, i.e. a series of outgoing plane waves having the same pseudo-periodicity.
It should be borne in mind that these plane waves can be divided into two
categories. If n 4 U,p, is imaginary and R,(x, y ) represents an evanescent wave
propagating in the x direction. On the other hand, for the finite number of n E U,
R,(x, y ) is an homogeneous plane wave propagating upward with a wave-vector
k,,(a,, fin). Denoting by 6, the angle of k, with the Oy axis (measured anticlock-
wise this time), it can be stated:
sin6, = a,/(a," + p:)'" = a,/k, (2.12c)
which, after replacing a, by its value given by (2.12a) yields the famous grating
formula:
sin On = sin 8 + nA/d. (2.15)
The complicated convention for 6 and 6, is justified by the fact that the above
formula takes exactly the same form as the one classically demonstrated for
transmission gratings, using optical path considerations. For the opticist, the
evanescent waves have no interest (except in integrated optics) and the on'ly
interestingquantities are the finite number of B,, for n E U.Indeed, the efficiency
e,, in the nth diffracted order, viz. the nth order diffracted energy to incident
energy ratio, is very easily deduced from the B,. It suffices to calculate the flux
12 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, $ 2

of Poynting’s vector of the incident wave and of the nth order diffracted wave
through a plane surface parallel to the x z plane at y > y,, of unit length on the
O z axis and of length d on the O x axis, to deduce:
e, = B , K cos B,/cos 8, (2.16)
-
B, denoting the conjugate of B,.
The validity of the RE above the top of the grooves, which is rigorous, should
not be confused with the so-called “Rayleigh hypothesis” (RH) which assumes
that the plane wave expansion remains valid inside the groove, and whose
theoretical validity will be studied further.

2.6. A LEMMA AND ITS APPLICATIONS: ENERGY BALANCE CRITERION,

RECIPROCITY AND EXPRESSION OF THE B ,

Here, we give a mathematical lemma, inspired by some previous papers

(URETSKY [ 19651, PETIT[ 1967,19801, MAYSTRE and M c PHEDRAN [ 19741,
CADILHAC [ 1980]), which enables us to establish some important properties
of the coefficients B,. To this end, we consider a given grating and we define
a certain class Fuof functions which contains in particular the two functions
F(x, y ) corresponding to the two fundamental cases of polarization for a given
incidence 8. More precisely, any function u(x, y ) of Fuwill satisfy the following
conditions:
- It is pseudo-periodic with coefficient of pseudo-periodicity a.
- It satisfies (2.8) (Helmholtz equation).
- It is bounded for y + co.
It will be assumed that u(x, y ) also satisfies a local square integrability in x and
y and that the boundary values on 9for u and duldn are square-integrable too.
Obviously, Fucontains F(x, y), whatever the fundamental case of polarization,
but we do not impose a boundary condition on the grating, or a radiation
condition. As a consequence, Yucontains the incident field F’ too. Using the
same arguments as in $ 2.5, it can be shown that such an arbitrary function
u(x,y) can be represented by a sum of plane waves above the grooves:
u(x,y) = 1U A , exp(ia,x
ne
- i&y) t 1 B,, exp(ia,x
n
t i&y). (2.17)

The above expansion differs slightly from the expansion of F (see (2.13)) since
it contains incident homogeneous plane waves of amplitudes A,, but it cannot
contain incident terms corresponding to n 4 U since it is bounded at infinity.
1, § 21 PRESENTATION A N D PRELIMINARY RESULTS 13

Considering a second function U(X,y ) of the same set, with amplitudes Al,
and Bl,, we will establish the following lemma:

where j 9 denotes a curvinilear integral on a period of the profile ds being

the differential of the curvilinear abscissa s on 9 The demonstration of this
lemma lies in the fact that u and u (thus 5) satisfy the same Helmholtz equation:
V2u + k2u = 0, V2V + k2E = 0.
Multiplying the first equation by V , the second equation by u and subtracting
the first from the second yield:
uV2U - VV2u = 0 in R, ,

By integrating the above equation in the hatched area of Fig. 2.2 and applying
the second Green identity, it can be deduced:

where %' is the boundary of the hatched area and I its curvilinear abscissa.
Since ud$dn and vduldn are periodic, the contributions of QR and SP are
opposite and cancel each other. Moreover the normal to RS is parallel to Oy,
thus the above equation yields:

d; -du di -du

0
. D X

Fig. 2.2. Demonstration of a lemma. %'isthe boundary PQRS ofthe hatched area and L denotes
the curvilinear abscissa on '6.
14 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 2

It is a cumbersome, but not difficult, exercise to transform the right-hand

member of the above equation in order to get (2.18). The reader may reach this
goal by introducing expansions (2.17) for u and u in this right-hand member,
separating the terms n $ U from the others, and bearing in mind that
S,dexp(inKx)dx = d.bn, where 6, = 0 if n # 0 and 6, = 1.
The energy balance criterion is obtained by taking u = u = F + F' in (2.18).
Since F + F' or d ( F + F')/dn vanishes on the left-hand member is null and
we deduce:
1 B n B , P n = Po,
nsU

or in other terms, using (2.12b), (2.12~)and (2.16):

1 en=l.
nsU
(2.19)

It is important to notice that the evanescent orders are not involved in this
relation, called the energy balance criterion, which expresses a result with an
obvious physical meaning: the diffracted energy is equal to the incident energy.
The lemma will enable us to establish another result, familiar to opticists: the
reciprocity relation. Here we take u = F + F', as before (case l), for a given
fundamental case of polarization. Now, we consider thepth order of diffraction
( p E U ) of diffraction angle 0,. Let us call 0:, a:, &' and B; the parameters
of the Rayleigh expansion of the field u" obtained for the same fundamental
case of polarization, but this time with the angle of incidence 0" = - 6, (case 2).
Bearing in mind our conventions for incident and diffracted angles, the incident
wave in this second case has a direction of propagation which is just the
opposite of that of the pth diffracted order in the first case. From the grating
formula, a straightforward calculation shows that 0; = - 0, a; = - a, and
P,,' = Po, which means that the pth diffracted order in case 2 and the incident
wave in case 1 have opposite directions, a fact which will probably seem
obvious to opticists familiar with the reversion theorem (Fig. 2.3). Function u
and u = u" wholly satisfy the conditions of the lemma since the grating formula
shows that they have the same pseudo-periodicity. On the other hand, u and
u are very different since u includes all the possible incident waves (n E U ) but
only one propagating diffracted wave, due to the conjugation which transforms
an incident wave into a diffracted wave propagating in the opposite direction,
and conversely. Proceeding carefully to calculate the right-hand member of
(2.18) and remarking that the left-hand member of this equation vanishes, it can
be found:
I. § 21 PRESENTATION A N D PRELIMINARY RESULTS 15

0
case 1
X
. case 2

Fig. 2.3. The reciprocity theorem: the efficiency of the pth order is the same in the two cases.

Multiplying each member by its conjugate, and dividing by &&’ yield, after
simplification:
B;qP;lP; = B,B,P,/Po, (2.21)
a result which can be easily outlined: The eficiency in the pth order is the same
in the two cases of Fig. 2.3.
After the famous reciprocity relation, let us obtain a last result, which generally
appears as the consequence of much more complicated considerations : the
coefficients B, may be given by a curvilinear integral on 9 Once more, the value
of u will be the function I: + F’, corresponding to a fundamental case of
polarization. The function v will be very simple:
u = @Jx, y) = exp(ia,x + iP,y), n E U.
In other words, u will be a single diffracted wave corresponding to the nth order.
So, the A,; and all the BA except BI, = 1 are nil. It follows that the right-hand
member of (2.18) will be equal to - &B,. Evaluating the left-hand member
requires the separation of FTE and FTM problems. Since either u, or duldn

s
vanishes on <%it follows immediately that:

B, = __ *S,(X, f(x))ds for the FTE case,

2idP, iy dn

(x, f ( x ) )ds for the FTM case.

By replacing ds by (1 + f ’ ( x ) ~ ) ”d~x and using the value of @,(x, y), it can be

sod
derived that

B, = exp( - ia,x - i&f ( x ) ) $ ( x ) d x (2.22a)

2id&
~
16 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, $ 2

t s: (
for the FTE case,

Bn = p,, )
1 - 5 f’(x) exp(-ia,x - ifl,,f(x))ll/(x)dx (2.22b)

for the FTM case. Thus

$(x) = (1 + f’(x)2)’’2 d(F + F’)/dn, (2.22c)
$(x) = JTx, f (XI) + FYx, f (XI). (2.22d)
This result is very interesting. The coefficients B,, and thus the eficiencies, are
knownfrom a simple integral as soon as the normal derivative of the totalfild (for
the FTE case) or thefild on the grating (for the FTM case) is given. A simple
physical interpretation of this result may be given by noticing that functions $J

and ll/ are closely linked to the surface current density j , on the grating surface.
Indeed, j ’p = n^ x H and, from (2.3a) and (2.3b), we derive:
$(x) = - iopo(l + f’(x)2)’/2j,(x) for the FTE case, (2.23a)
=J&) for the FTM case, (2.23b)
the scalar*functionj, being the z component ofj, for the FTE case and the
projection j rp * t of j of the unit tangent vector f = n^ x 2 for the FTM case.
It is not surprising to see that the B, can be derived from the surface current
density since j rp actually generates the diffracted field. It is worth noting that
(2.22a) and (2.22b), demonstrated for n~ U, hold for n $ U, as we shall see
further. Similar formulae can be obtained for the incident amplitudes A,, by
setting u = exp(ia,,x - ip,,y). Since A , = 6, for FTE and FTM cases:

for the FTE case, (2.24a)

bn=’ 2d jod(l
+;/’(x))exp(-ia,x+ i&f(x))ll/(x)dx
for the FTM case, (2.24b)
and these relations hold when n $ U, as we shall see later.

2.7. WHY RIGOROUS VECTOR METHODS?

For a long time, it has been difficult to convince some grating users of the
usefulness of rigorous grating theories and, from my own experience, it seems
1. § 21 PRESENTATION A N D PRELIMINARY RESULTS 17

that even today, some opticists are not aware of the troubles they may encounter
using approximate theories. So, we have chosen a theory, one of the most
familiar to opticists, based on the Kirchhoff approximation (BECKMANNand
SPIZZICHINO [ 19631, BECKMANN[ 1967]), in order to compare its numerical
results with exact values. Though the basic approximation is the same, the
demonstration developed here to obtain the approximate formula will be rather
different from the demonstration made by BECKMANN,and the resultant
formula will appear to be slightly different.
Let us consider the TE fundamental case of polarization where, according
to (2.22a), the B, are known as soon as the function $ (closely linked with the
normal derivative of the total field) is known. The approximate value h of $
will be obtained in the following manner: we shall assume that the normal
derivative of the field at a certain point of 9’is the same as the normal derivative
of the field which would be obtained by replacing 9 by a plane mirror tangent
to 9 at the same point. So the “local” value of F’ + F i n the vicinity of 9’ will
be the sum of the incident wave and the wave reflected by the fictitious mirror.
An effortless calculation shows that:
$ B ( ~=) - 2i(aJ’(x) + fl) exp(iax - iflf(x)). (2.25)

Introducing $I~(x)in (2.22a) and integrating by parts to remove the term f’(x)
gives the following formula:

1 1 + cos(e+ 0,)
B = --
d cos O,, (cos 6’ + cos 0,)
1; exp [ - inKx - i(fl + fl,)f(x)] dx,
(2.26)
a formula which differs from the BECKMANNformula by the presence of the
factor cos O,, (instead of cos 0) in the denominator, but we have to bear in mind
that the quantity given by BECKMANNis not B,. After calculation, the value
of B,, obtained in the FTM case is identical, except for the sign ( - ) in front
of the right-hand member, which must be eliminated. Though the expression
of B,, given by (2.26) obviously satisfies the reciprocity theorem stated in (2.20),
we must notice that it finally gives a scalar result, which does not predict any
polarization effect, at least for the efficiency of perfectly conducting gratings.
In the particular case of the sinusoidal grating:

H
J(X) = - cos Kx, (2.27)
2
the integral can be evaluated in closed form using classical formulae
18 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 2

and STEGUN[ 1970]), in such a way that, for the FTE case:
(ABRAMOWITZ

B , = -(-i),
1 + cos(e + en)
cos o,,(cOs 8 + cos 8,) 1
(cos8 + C O S ~ , ,.)
(2.28)
,
The corresponding efficiency e - has been drawn in Fig. 2.4 for H / d = 0.2 in
Littrow mount. In this mounting, very familiar to grating users, the - 1 order
diffracted wave and the incident wave are propagating in opposite directions,
so 0- = - 8, and from the grating formula we deduce that in this mounting,
the wavelength, groove spacing and incidence are linked by a very simple
relation:
I / d = 2 sin8. (2.29)
Furthermore, from (2.16) and (2.28) the efficiency e - is given by:
e- = J:(kH cos O)/cos48.
Figure 2.4 also shows the efficiency obtained from a computer code developed
by the present author, using a rigorous integral theory, which will be described
in Q 5. It will be shown further that the accuracy of the “rigorous” results is
better than 1 per cent and it is to be hoped that the reader will accept this

0 05 10 15 20
wavelength/ period

Fig. 2.4. Comparison ofthe numerical results obtained from a rigorous theory and a non-rigorous
one. Solid line: rigorous result from the integral theory; dashed line: result from the BECKMANN
theory.
 T H E RAYLEIGH METHOD 19

assertion at this point. Obviously, the agreement between the two theories is
very good as long as I / d is lower than 0.4 approximately: the results are nearly
identical. Unfortunately, the agreement does not hold any longer in the comple-
mentary region since the efficiency predicted by BECKMANN’S formula tends
to infinity when I / d tends to 2 (this corresponds to a grazing incidence, as
shown by (2.29)). The limit of the region of validity of BECKMANN’S theory
varies with H , and decreases when H increases. Since many commercial
gratings have groove depth to groove spacing ratios higher than 0.2, we must
conclude that it is dangerous to trust in such a theory in the resonance region.
However, it must be emphasized that such a theory is very valuable when Ild is
low: it gives accurately the properties of gratings without the help of a big
computer and moreover, the rigorous computation becomes more and more
difficult and expansive when Aid decreases. So, there does not exist any
challenge between rigorous and non-rigorous theories, but rather a complemen-
tarity. We hope only to convince the reader that using non-rigorous theories has
a prerequisite: the thorough knowledge of its domain of validity, in order to
avoid disasters. Of course, this conclusion still holds for non-periodic rough
surfaces.

4 3. The Rayleigh method

We have seen in the previous section that the RE (Rayleigh Expansion, see
(2.13)) is valid above the top of the grooves. Rayleigh made the assumption that
this expansion was a valid representation of the diffracted field F not only for
y > y M but throughout the region R , , including the grooves (RAYLEIGH
[ 1907, 19451). The first objections about the RH (Rayleigh Hypothesis) were
formulated thirty years ago (DERIUGIN [ 19521, LIPPMANN [ 19531) and until
now, this topic has raised considerable controversy. At present, there is no
doubt that the RH is neither always valid, nor always invalid. The main
theoretical achievements in this topic will be presented in a synthetic manner.
Due to the difficulty of some demonstrations based on the properties of complex
functions of complex variables, we thought it useful to give at the beginning a
non-rigorous but much more simple demonstration founded on BECKMANN’S
approximation, which nevertheless leads to an exact result. The problem we are
dealing with in the last part of this section is the main subject of the actual
controversies: why and in what conditions can the RH be used to obtain accurate
numerical results, even in conditions where this RH theoretically fails?
Now, to begin this discussion, we intend to convince the reader that the
20 RlGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 3

question of the validity of the RH is not academic: the RH simplifies

considerably the work of the theoretician and provides a very simple means of
calculating the grating efficiencies.

3.1. BASIC EQUATION AND NUMERICAL IMPLEMENTATION OF THE

RAYLEIGH METHOD

After Rayleigh, the Rayleigh hypothesis has been used by many authors
(FANO [ 19411, MEECHAM[ 19561, STROKE[ 19601, BOUSQUET[ 1963a,
1963b], PETIT [1963], DELEUIL[1963], JANOT and HADNI [1963],
YAKOVLEV [ 19651, PETITand CADILHAC [ 19661, WIRGIN[ 19671, JIRACEK
[ 19731). Some of them implemented this theory numerically when sufficiently
powerful computers became available. Here, we describe three ways of imple-
menting this method. They have in common the basic equation, obtained by
writing that the RE described by the series R ( x , y ) (see 2.13)), which obviously
satisfies the Helmholtz equation and radiation condition stated in (2.8) and
(2. lo), must also satisfy the boundary condition on the grating surface expressed
by (2.9a) or (2.9b). For example, in the FTE case, we obtain:

c B,, Qn(& f(x)) + FYx,

n
f(XN = 0, vx. (3.la)

In this equation, the only unknown is the set of complex coefficients Bn which
appear to be the coefficients of the expansion of - F'(x, f ( x ) ) into a series of
functions Qn(x,f ( x ) ) . All the numerical methods described here reduce (3. la)
to a system of (2N+ 1) linear equations with (2N+ 1) unknowns: the
coefficients B, for - N Q n Q + N . So, these methods neglect all the B,
corresponding to la I > N and replace (3.la) by:

sN(X) = c
+ N

n= -N
BiN)@,l(x,f ( x ) ) + F'(x, f ( x ) ) = 0. (3. Ib)

Of course, it will be necessary to check the convergence of the B',", approxi-

mations ofB,, obtained for a size 2N + 1 ofthe linear system, when Nincreases.
In the first method, called the point-matching method ( P M M ) , (3.lb) is
written at 2N + 1 points of abscissae x , placed on one period of the grating:

c
+N
B',N)Qn(~,,f ( x , ) ) = - Fi(x,, f ( x , ) ) , Vm E (1,2N + 1). (3.2)
n= -N

The linear system of 2N + 1 equations with 2N + 1 unknowns can be solved

1 , s 31 THE RAYLEIGH METHOD 21

on the computer using classical methods. Of course, the choice of the x, is

fundamental. In practice, the x, are chosen equidistant.
In the second numerical method, called the Fourier Series Method (FSM),
(3.lb) is projected on the 2N + 1 functions exp(ia,x) ( - N < m < + N ) in
order to obtain a linear system of equations:

n= -N
VmE(-N, +N), (3.3)

To justify this method, it suffices to notice that the left-hand member of (3.lb)
is the product of a periodic function and exp(iax). In practice, two cases are
observed: either the scalar products can be calculated in closed form or they
require a numerical integration.
In the last method, described in this section, the functional

is minimized. To obtain a linear system, it must be noticed that

TN = (SN(x), sN(x) ) 1

and that TN,considered as a function of the B‘,”, must be stationary:

+N
6TN = 2 1
n= -N
Re{6B‘,N) ( Dn,s,(x))} = 0,

Re {p(x)} denoting the real part of p(x). Therefore, this Fourier variational
method (FVM) leads to the following equations, obtained by equating to zero
all the coefficients of GBsr):
+ N
1
m= - N
B!” ( @Ax,f(x)), @m(x,f(x))> + ( @ h ,f(x)), F ’ h f ( x ) ) > = 0,
VnE(-N, +N). (3.4)
This time, (3. lb) is not projected on the Fourier basis but on the set of functions
@,f(x)>.
k
When they were implemented on a computer, between 1960 and 1970, these
methods were applied, in the first instance, to ruled echelette gratings, viz.
gratings having triangular grooves, and also to sinusoidal gratings. In these two
cases, the coefficients of the linear system can be evaluated in closed form for
22 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 3

the FSM and for the FVM: exponentials for echelette gratings and Bessel
functions for sinusoidal gratings. For shallow gratings the efficiencies obtained
from these methods satisfied the energy balance criterion and the reciprocity
theorem with very high accuracy. They also converged when N was increased.
On the other hand, the Rayleigh methods completely failed for deeper grooves.
The region of convergence of the FSM and the FVM was significantly larger
than that of the PMM. For instance, calculations using the FSM for sym-
metrical echelette gratings (Fig. 3.1) of groove spacing d = 1.25 pm used in
normal incidence with 1 = 0.546 pm gave very good results until tanb = 0.15.
These results were less precise but acceptable for deeper grooves until
tanb = 0.3, for which the energy balance criterion was satisfied to within 2%.
For greater values of tanb, the results were unprecise and for tanb = 1, a
divergence was observed (PETIT[ 19631). For the same grating, the numerical
domain of convergence is dependent on the wavelength to groove spacing ratio
(Mc PHEDRAN [ 19731). Due to this numerical failure, many authors questioned
the validity of the Rayleigh hypothesis and remembered the former warnings
of DERIUGIN and LIPPMANN. For these authors, the RE is not valid at a
point P inside the grooves since it does not contain waves propagating towards
y = - co, which are generated by the surface currents at the points of the
grating surface situated above the point of observation. We shall see that the
serious objections to the RH apparently have no connection with the remarks
of DERIUGIN and LIPPMANN.

3.2. CONCERNING THE VALIDITY OF THE RAYLEIGH HYPOTHESIS

3.2.1. Preliminary results

From a theoretical point of view, it is interesting to consider the behavior of
the Rayleigh expansion R ( x , y ) inside the grooves, i.e. in the region where it

T:
_ d_ 0 d X
2 +2
Fig. 3.1. The symmetrical echelette grating used in normal incidence.
1. I 31 THE RAYLEIGH METHOD 23

cannot be used a priori to describe the field. First, let us demonstrate that, if
the RE, which represents the diffracted field above the top of the grooves,
converges at any point inside the grooves, it actually represents the diffracted
field F at this point. To this end, let us study the asymptotic behavior of the
nth term R,(x, y ) of the RE when n -+ f 00. We denote byp the positive integer
p = In 1 and we call afp, b,,(x, y), i?&, y ) asymptotic expressions of P k p ,
@,..(x, y ) and R ,,,(x, y), respectively. From (2.12a), (2.12b) and (2.14), we
derive : a,,
= ipK icr, (3.5)
b” = u(x, Y)W(X,Y)”, (3.6a)
9-, = U(x, y)- ‘W(x,y)”, (3.6b)
with
u ( x , y ) = exp(iax - cry), (3.7)
w(x, y ) = exp(iKx - Ky). (3.8)
The first step of the demonstration is described in Appendix A. Let us state the
conclusion:
Ifthe RE converges at a certain point of ordinate yo, it converges absolutely and
uniformly to an analytic function of the two real variables x and y in the halfplane
y 2 y1, whatever the value y1> yo.
An obvious consequence of this fundamental property is that the RE diverges
in the half plane y < y3 if it diverges for a certain point of ordinate y 2 , whatever
the value y , < y,. So, we are led to define an “ordinate of convergence”y, such
that:
- the RE converge for y 2 y , , whatever y , > y,,
- the RE diverges for y 6 y,, whatever y , < y,.
Of course, since the RE represents the field F(x, y ) and converges for y 2 y,,
it can be inferred that y, 6 y , and that F(x, y ) is an analytic function of the
real variables x and y when y > y,. In fact, F(x, y ) is an analytic function of
x and y in the entire region R + ,because of the following mathematical property
of the Helmholtz equation:
Since the Helmholtz equation V2u + k2(x,y ) u = 0 is a linear ellqtic partial
differential equation, i f k is constant (or analytical in x and y ) in a certain domain
GI of R2, every continuous solution u(x, y ) is an analytic function of the two real
variables x andy in this domain (SOMMERFELD [ 19491, COURANT and HILBERT
[ 19621, MIRANDA [ 19701). We deduce that F(x, y ) is analytic in R . +

Now, let us summarize the above theoretical results:

R(x, y ) is an analytic function of x andy fory > y , > y,; F(x, y ) is an analytic
function of x and y in R ; F(x, y ) 3 R(x, y ) for y > yM.
+
24 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 3

So, in the domain 62, of R situated above y , > y,, u(x, y ) = F(x, y ) - R ( x ,y )
+

is an analytic function of x and y. Since u(x, y ) vanishes in the subdomain Q2

of 62, defined by y > yM,it vanishes in the entire domain 62,. So, the Rayleigh
expansion R(x, y ) well represents the diflractedfield F(x, y ) in a,, i.e. above the
ordinate ofconvergence y,. Now, if y, < ym,the RE converges and identifies to
the diffracted field throughout R ,therefore the RH is valid. This conclusion
+

does not hold any longer ify, > y,, since the RE, which diverges in a certain
part of R , cannot be used to express the field at any point of the profile .9,
+

in order to write the boundary conditions (2.9a) or (2.9b). Finally, the study
of the validity of the RH reduces to the study of the ordinate of convergence
y,, which will be compared with the ordinate of the bottom of the grooves y,.
In the following, the value of yc will be deduced from the asymptotic behavior
of R,(x, y ) when n + + co. It is worth noting that the value of R ( x , y ) in the
region of R - situated above y = yc is the analytic continuation of the diffracted
field F ( x , y ) below 9?

3.2.2. A simple but non-rigorous demonstration

The expression of B, obtained from Beckmann's theory (cf. (2.26)) may be

used to evaluate the asymptotic behavior of the RE, provided the integral can
be calculated in closed form. Though the Beckmann approximation becomes
more and more precise when A + 0, this obviously cannot be considered as a
rigorous investigation but rather an attempt at suggesting a fundamental result
from elementary mathematics. For the sinusoidal grating whose profile is
described by (2.27), introducing the expression of B",, given by (2.28) and
remarking that k cos O,, = /I,,+ ipK whenp + co we find, for the FTE case:
B+, = c * Z,(Ph)/P,
with h = nH/d, c + being two complex coefficients and I,, the modified Bessel
function (ABRAMOWITZ and STEGUN[ 19701). From the uniform asymptotic
behavior of Z,(pz) which can be found in the same reference, we derive:
4 , = c;p-3/2
I

exp{p[(l + h2)ll2 - arcsinh(l/h)]},

c'+
- being independent ofp, which yields, taking into account (3.6a) and (3.6b),

with Ky, = (1 + h2)lI2 - arcsinh(l/h). (3.9)

 THE RAYLEIGH METHOD 25

So, the general condition of validity of the RH, i.e. y, < y,,, = - H/2, becomes
h < h,, h, being given by the transcendental equation:
h, + ( 1 + h:)’” - arcsinh (l/hc) = 0, (3.10)
which finally gives:
h, = 0.447143, (3.11)
i.e. the value first given by PETITand CADILHAC
[ 19661from a counterexample
described in the next section.

3.2.3. A counterexample

Now, we briefly outline the fist rigorous theoretical demonstration of the

non-validity of the RH (PETITand CADILHAC [ 19663). It applies to sinusoidal
gratings, whose profile is given by (2.27).
Let us assume that the RE is valid at any point of the profile 9 Thus, it
satisfies the boundary condition (3. la) which, for sinusoidal gratings, becomes:

= 0, V x .
n

(3.12)
It will be shown that the above equation cannot be satisfied, at least above the
critical value H, of H deduced from (3.11).
First, we must envisage mathematical considerations which perhaps will
appear strange to the physicist: the behavior of (3.12) for complex values of x.
In fact, we have already investigated in Appendix A some properties of the
Rayleigh expansion R ( x , y ) for complex values of x and y . These properties
apply as well to the first term of (3.12) which is simply R(x, 3H cos K x ) , a
function of a single complex variable x . If (3.12) is satisfied for any value of
x , it applies in particular to K x = - n,i.e. at the bottom of a groove. Of course,
this requires the convergence of R( - n / K , - H/2). A direct consequence of
Appendix A is that R(x, $H cos K x ) will converge to an analytic function of the
complex variable x in the region axE C defined by:

I.( , :cosKx)l< I.( , -2c o s K x ,

>I ,

IF(., ~cosKx)<
( (~(x,,
26 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 3

- x / K (and therefore whose ordinate on

x, being a real number different from
S is greater than y,) but which can be very close to it.
We have drawn in Fig. 3.2 the contour lines of I w(x, $HcosKx)I for
H/d = 0.175, i.e. h = 0.55. The dashed line shows the contour line passing
through a point of abscissa 5, = K x , close to - x. Taking into account the
contour lines of W(x,$HcosKx), which are symmetrical to those of Fig. 3.2
with respect to the 05 axis, 0,is the hatched region of Fig. 3.3. So, the left hand
member of (3.12), which is analytic in QX and vanishes on the real axis AE, must
vanish throughout ax,and in particular on the line 9 of positive abscissa
4, < n/2. Now, a careful examination of (3.12) shows that, on this line, the
series converges and even tends to 0 when q + - co,whereas the second term
F' tends to infinity. Thus, (3.12) cannot be verified on 9 E OX, a fact which is
in contradiction with the definition of GIX. So, our hypothesis (the convergence
of the RE on the entire profile) is false. The demonstration holds as long as the
countour line passing through the saddle point S of Fig. 3.2 crosses the vertical
5 = - x above point A. Otherwise, the two dashed lines meet and cut the O q
axis, the domain sl,is bounded and the demonstration cannot be achieved. The
limit value h: is obtained by writing that the contour line passing through A

Fig. 3.2. Contour lines of Iw(x,~HcosKx)I= exp( - M ) for H/d = KH/2n = 0.175, in the com-
<
plex plane of Kx = + iq. S and S' are the saddle points of w or I/w. The values ofM are given
on the curves.
1.8 31 THE RAYLEIGH METHOD 21

Fig.3.3. Had the Rayleigh expansion been convergent at the point A, it would have been
convergent throughout the finite hatched region ax.

contains S. To find the value of x, corresponding to the saddle point, we must

look for the solution of dw(x, f(x))/dx = 0, i.e.
f'(x,) = i, (3.13)
and for the sinusoidal grating
Kx, = - i arcsinh(l/h), (3.14)
the corresponding value M , of M being
M, = (1 + h2)'j2 - arcsinh(l/h). (3.15)
Remarking that the value of M at the point A is equal to - h, we derive h;,
satisfying (3.10), and thus hb = h,. So, this rigorous demonstration shows that
the RH is untenable if h > h,. MILLAR[ 19711 showed that the RH is valid
below h,.

3.2.4. A general investigation

Now, our aim is to investigate the validity of the RH for gratings whose
profiles are given by analytic functions f(x), for both the FTE and the FTM
28 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS s
[I, 3

cases. The asymptotic values HfP will be derived from (2.22a) and (2.22b). Let
us consider the FTE case. By using (3.5) and (3.6a), it can be shown that:

(3.16a)

l d
= -~ exp[p( -iKx + Kf(x))]$(x)u(x, f(x))-’ dx. (3.16b)
4np 0
The asymptotic behavior of this integral can be obtained using a saddle point
method (DENNERYand KRZYWICKI [1967]), as made by HILLand CELLI
[ 19781. Since f ( x ) is analytic and may be extended to complex values of x, and
assuming that $(x) satisfies the same property (cf. MILLAR[ 19691 in the case
of sinusoidal gratings), the integration path may be modified in the x complex
plane according to Cauchy’s theorem and the asymptotic behavior is deduced
from a path passing through the saddle point. For instance, for sinusoidal
gratings, the saddle point S of l / w is shown in Fig. 3.2. The initial path AE is
replaced by ABSCE, and finally by BSC since the contributions of AB and CE

-
cancel each other. It turns out finally:
Bp = T+pP3/’exp[pKf(x,) - ipKx,], (3.17)
T , being a complex coefficient.
For the FTM case, though the coefficient in 1/& is removed in (2.22b), (3.17)
still holds if we notice that (1 - f’(x)ocp/fiP)asymptotically tends to 0 at the
saddle point. From (3.6a) and (3.17), we derive an asymptotic value Rpof the
RE :
RP(xyy)= T + u(x,y)~-~/’exp{i~K[(x - x,) + i(y - f(x,))l}. (3.18)

So, for real or complex values of x and y, a necessary condition for the
convergence of the RE is:
Im{x - x, + i(y - f(x,))} > 0. (3.19a)
A similar calculation for the terms R --p leads to a second necessary condition:
Im{x - x, + i(7 - f ( x , ) ) } > 0. (3.19b)
It emerges that the domain of convergence R E C2of the RE is the intersection
of the two regions defined by (3.19a) and (3.19b).
For real values of x andy, (3.19a) and (3.19b) show that the RE converges
above the horizontal line passing through the point J of real coordinates (xJ,yJ)
such that:
xJ + iyJ = x, + if(x,), (3.20)
 THE RAYLEIGH METHOD 29

a fundamental result previously found by other authors using steepest descent

methods (HILL and CELLI [ 19781, DE SANTO [ 19811) or an elementary
conformal mapping (VANDEN BERGand FOKKEMA [ 19791).
To interpret (3.20), it is necessary to bear in mind that x,, given by (3.13)
is complex, whilst x J and y , are real. In its region of convergence, the RE
identifies with F(x, y), viz. the diffracted field or its analytic continuation below
9!Therefore, it may be shown that the point J is a singularity of F(x, y). Since
F is analytic in R , we deduce that J E R - . *
+

From the fundamental result stated in (3.20) it can be expected that, in

general:
YJ = Y c * (3.21)
However, this is not a general rule. For instance, CADILHACrecently suggested
the case of the following grating (private communication):

1
f (x)= arcsinh (a cosKx),
K

which is close to the sinusoidal grating for low values of a. For this grating, there
is no saddle -point since (3.13) has no solution and yJ = - co. However, a
branch point limits the domain of convergence of the RE, and yc is finite.
From another remark of Cadilhac, let us state an original result which could
permit the opticists to find a physical meaning for the point J. The equation of
the tangent to 9 at a point (xo, f ( x o ) ) is:
Y - f ( x J - f’(x,)(x - X d = 0 9

and at the saddle point x = x , of 3 it becomes, according to (3.13):

Y - f(x,) = i(x - x,),
which is the equation of the first isotropic line tangent to 9 It can be shown
easily that the point J, which obviously belongs to this first isotropic line
(IYANAGA and KAWADA [ 19801) tangent to 3 according to (3.20), belongs to
the second isotropic line as well. Thus, J is the intersection of the two isotropic
lines tangent to 9 For conics, this is one of the definitions of the focus. So, we are
led to extend this definition of the mathematical focus to an arbitrary curve given
by an analyticfunction f (x). Thefocus Jof theprojle is a singularity of the analytic
continuation of the dflractedfild in region R - and the RE does not converge below
it. So, the RH is not valid when thefocus is situated above the bottom of the grooves.
It may be expected in general that the RH is valid when the focus is situated below
the bottom of the grooves.
30 RIGOROUS VECTOR THEORIES OF DIFFRACI‘ION GRATINGS 119 §3

This fundamental remark leads us to generalize the result obtained in this

section. For instance, if the profile B is given by parametric equations:
x = x(t), Y = YtO,

(2.22a) and (2.22b) can be used as well to investigate the asymptotic behavior
of B,,, using the saddle point method in the t complex plane. The result is the
same: the singularity of the RE is the focus of the curve. It will be found by
writing that the tangent to B at the point to,
-y’(to)(x - xo) + X’(tO)(Y - Y o ) = 0 , (3.22)
is an isotropic line which satisfies
Y’(t0)= iX’(lO), (3.23)
and by looking for the real coordinates X , and x,, such that
x, + iy, = x(to) + iy(to).
The same remark applies to a grating for which B is given by ~ ( xy)
, = 0 where
the focus will be found by solving:

(3.24)

X(X09Yo) = 0, (3.25)
and by writing that X , + iy, = xo + iy,.
For the opticist, it is not surprising to find a singularity of the diffracted field
at a focus. Unfortunately, this is not the optical focus of geometrical optics,
which cannot be defined for an arbitrary profile 9 It is the mathematical focus,
which may be considered as the “asymptotic optical focus of the evanescent
waves” and which is independent of the wavelength and of the incidence. This
author verified an interesting property in the case of sinusoidal gratings: the
mathematical focus J is always close to the optical focus obtained from
geometrical optics by replacing the top of 9 by circles having the same radius
of curvature at this point. Thanks to the definition of the focus, all the above
conclusions can be generalized to any two-dimensional diffracting subject: the
focus is a singularity of the solution of the Dirichlet and Neumann problems of
diffraction.
The reader interested in calculating the, limit of validity of the RH for a
particular grating can use the above equations. Many examples of calculation
of y , have been given by NEVIERE and CADILHAC [ 19701, HILLand CELL]
1, I 31 THE RAYLEIGH METHOD 31

[ 19781, VAN DEN BERG and FOKKEMA [ 19791, DE SANTO[ 19811. For non-
analytic profiles, it has been shown by MILLAR [1969], NEVIEREand
CADILHAC [ 19701, that in general, the RH fails as soon as the upper point of
B is an edge. This phenomenon can be easily explained by the fact that the field
Fis singular on the edge (a singularity is a point where the field is not analytic),
a fact which implies the divergence of the RE on the edge. This rule is not
general since counterexamples have been given for echelette gratings in the
FTM case (MARECHAL and STROKE[ 1959]), echelette and lamellar gratings
for the FTE and FTM cases (MAYSTRE [ l974,1980b]), in which the diffracted
field is described by a small number of plane waves and has no singularities.
It is amazing to notice that one of the main results of this section was
implicitly stated by SOMMERFELD [ 19491 in his theory of “characteristics”. He
showed that the necessary condition for finding the analytic continuation of the
field at the vicinity of the profile was that the characteristic lines (which are
nothing but the isotropic lines for the Helmholtz equations) must not be tangent
to the profile. A simple generalization of this remark to complex values of x
and y shows that singularities of F(x, y ) may be expected in the vicinity of the
point (x, ,f ( x s )).

3.3. A PARADOXICAL RESULT: THE SYSTEM OF RAYLEIGH FUNCTIONS IS A

TOPOLOGICAL BASIS

Apparently, the theoretical results about the validity of the RH are the
explanation for the numerical difficulties observed in its first implementations.
This explains why rigorous theories have been developed further. Nevertheless,
since 1971, a series of theoretical and numerical studies have raised a new
interest in the Rayleigh method. Exploiting an analysis of VEKUA[ 19531 for
other diffracting structures, some authors (YASUURA and IKUNO[ 19711,
IKUNOand YASUURA[1973]) claimed the completeness of the set of
@,,(x, f ( x ) ) in the Hilbert space of square integrable functions (cf. MILLAR
[ 19731 or CADILHAC[1980]). Let us outline this demonstration for the
interested reader. Assuming that the pseudo-periodic, locally square integrable
function T(x) is orthogonal to the @,,(x, f ( x ) ) , i.e. that
( T ( x ) ,@,,(x, f ( x ) ) ) = 0, Vn, we must prove that T ( x ) 3 0. With this aim we
use some results established in Appendix B. The functions f ( x ) and T ( x )being
given, we consider the profile 8’ given by y = f( - x ) , separating R and R
and a continuous function U(x,y ) satisfying (B.8) and (B.9), i.e. Helmholtz
equations in both R ‘+ and R’- ,and radiation conditions for y .+ & 00. If we give
32 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, I 3

the jump p( - x ) ( 1 + (df( - x)/dx)2)- of its normal derivative, then U ( x ,y )

is given by (B. 10) and this expression reduces, for y < y,, to a sum of plane
waves (see (B.16)) whose coefficients D; are given by (B.17). Here, since U
is continuous, z = 0 and moreover, q = T( - x), according to (B. 13) and (B.14).
We deduce that

So, U ( x ,y ) vanishes for y < y , and since U is analytic in R'_, it vanishes

throughout R'_. By continuity, its boundary value above 9 is null and the
uniqueness of the solution for the FTE case allows us to state that U vanishes
throughout R'+ . Finally, the jump of the normal derivative of U is null and
r(x)= 0.
Considering (3. la), the basic equation of the Rayleigh method, the comple-
teness of the Rayleigh basis seems to justify the RH and to be in contradiction
with our former theoretical study since the incident field F'(x, f ( x ) )is a locally
square integrable function. To understand the paradox, we must recall that the
@,,(x,f ( x ) ) are not orthogonal. In fact, they are members of a topological basis.
Physicists are very well acquainted with this notion for orthogonal systems but
the situation is much more difficult in our problem. Indeed, due to the non-
orthogonality of the @,, the completeness of the topological basis means that
a square integrable function, for instance - F'(x, f ( x ) ) , can be represented as
a limit of linear combinations of the @,(x, f ( x ) ) :

(3.26)

with

and we shall prove that:

(3.28)

where B, is the actual Rayleigh coefficient. Two fundamental remarks must be

made about these equations. First, (3.26) must be understood in the sense of
 T H E RAYLEIGH METHOD 33

the least squares approximation. In other words, the BLN)must be chosen in

order to minimize I, such that:

IN = I s N ( x ) + F’(x, f 1 * dx*
Equation (3.26) only means that I N - + 0 when N -+ a,and does not mean that
the two members are equal for any value of x. The second fundamental remark
is that the BLN)of(3.27) depend on N . So, the convergence of S,(x) when N -+ 00
does not imply the convergence of the RE, even though BLN)-+B,. To cast light
on this rather subtle fact, let us give a simple numerical example of such an
apparent paradox. S, is now, in this simple example, defined by:

2
N
S, = riN) with rLN)= 1 - 2n/N.
n=O

Obviously, S, = 0 for any value of N since r(“1 ,= - rLN)and, of course, S,

converges to 0 when N + 00. On the other hand, remarking that
limN*a rLN)= 1 when n is fixed, the series

diverges. This is exactly what happens in our case: the convergence of the S,(x)
does not imply the convergence of the RE, R ( x , f (x)), since in this series, the
BLN) are replaced by their asymptotic values B,.
Of course, these new theoretical considerations have raised a new interest
in the RVM described previously, which appears to be justified from a
theoretical point of view. However, recent numerical experiments have proved
that, even though the RVM can be used for some non-shallow gratings, it
converges slowly and does not converge for groove depths higher than
H, = 0.54d in the case of sinusoidal gratings (WIRGIN[ 1979a1, HUGONIN,
PETITand CADILHAC [ 19811).
It has been shown (see for instance the paper by MILLAR[ 19731) that the
set of S,(x) uniformly converges to the diffracted field F i n any closed subset
above % assuming the existence of a certain Green function vanishing on the
grating surface 9 Here, we use a more general demonstration to show that S,
uniformly converges to F on 9provided the profile is analytic.
Let us define, for a given profile 3 a second periodic curve 9*, given by
y = f *(x) 6 f ( x ) . We denote by R*, the region y > f *(x). We assume that the
analytic continuation of F (also denoted by F ) has no singularities in R*, and
is locally square integrable on 9*, properties always satisfied if P = 9
34 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 3

If we denote by G Jp* ( x , y, x' ,y' ) the pseudo-periodic Green function

vanishing when ( x ,y ) E P* and satisfying a radiation condition at y = 00, we
can write:

F(x, y ) = 1 iF
F(x, f * ( x ) ) dC,( x , y, x ' , f * ( x ' )) ds *,
dn'
(3.29)

s* being the curvilinear abscissa of the point of abscissa x' on ?*, and x , y the
coordinates of a point P of R*, (cf., for example, MORSEand FESHBACH
[ 19531, for the definition and properties of a similar Green's function obtained
when 9*is replaced by a closed surface). Now, let us call S$(x) the set given
by the sum described in (3.27), the B r being chosen, this time, in order to
minimize 1%:

I,$ = s,* 1 S,$(X)+ F(x, f*(x))I dx.

The completeness of the an(x,f * ( x ) ) implies the convergence of I$ to 0 when
N 00. Since each term of the sum representing S$(x) can be represented as
--f

in (3.29), we derive:

(x,y,x', f*(x')) ds*

(2 I"'
.

Since the first term of the right-hand member tends to zero and the second is
finite:
lim 1 F(x, y ) - T$(x, y ) 1 = 0
N - cc,
(3.30)

in any closed subset of R*, .

The interpretation of this result differs according to whether 9 is analytic
or not. In the second case, we can expect a singularity of F on 9 So, P* and
Pnecessarily have some points in common and (3.30)is not valid on the entire
profile 9 Nevertheless, taking 9*= 9 shows that the "RayZeigh set" T$(x, y )
unformly converges to F(x, y ) above the profile.Since this property is true in
 T H E RAYLEIGH METHOD 35

particular for y > y,, where F is given by the RE, it is easy to deduce that
B::) 4 Bn when N -,CQ.If we notice that the B‘,” so defined are nothing but
the coefficients determined by using the RVM (cf. $ 3.1), we are led to the
conclusion, stated above without a complete demonstration, that this method
is theoretically valid for any kind of grating, even when the RH fails. Of course,
all these conclusions hold when B is analytic, but in addition, we can conjecture
that 9*can have no point in common with because of the absence of
singularity of F on 9 So, R*, includes B and from (3.30) we derive that:
+N
lim
N-CC
n = -N
BLN)4 ( x , f (x)) + F i ( x ,f (x)) = 0, (3.3 1)

in the sense of uniform convergence. The existence of such B‘,” is the proof
of the theoretical validity of the PMM and FSM (cf. $ 3.1) for such profiles.
In this case, the FSM seems to be better and simpler than the FVM, as shown
numerically by WIRGIN[ 1979bJ: for sinusoidal gratings, the first method
converges below H / d = 0.7, agroove depth five times greater than the theoretical
limit of the RH. This success of the FSM probably has to be qualified for
non-sinusoidal analytic profiles, where the coefficients of the linear system of
equations are not analytic. This remark is not without substance: it can be
conjectured that the progress of the numerical implementation of the FSM in
the last fifteen years is also linked to the progress in the computation of the
Bessel functions on computers.
So, we are led to an unexpected conclusion: the numerical limits of the
Rayleigh methods are not directly linked to the theoretical failure of the RH, but
to purely numerical reasons at least for analytic profiles. For instance, in the
PMM, these reasons are obvious. According to (3.2), (3.6a), (3.6b) and (3.8),
the n, m coefficients Qn(x,, f ( x , ) ) of the linear system of equations to be
solved are very different in magnitude. If, for simplicity, we assume that the
ordinate of the top of the grooves y M = 0, the coefficient at the corresponding
abscissa will tend to 1 in the modulus, while at the other points x,, it will tend
to 0 in the modulus when n -,CQ. So, it suffices to adopt a sufficiently large
value of N to obtain identical lines for high values of n, at least for computer
purposes, and the matrix becomes singular. This phenomenon arises for any
value of H, including those inferior to H,, and this proves that the PMM is
ill-conditioned. The same kind of limitation appears for FSM and for FVM but
for higher groove depths. The interested reader can find theoretical investi-
gations, thorough comparisons of these methods and comparison with other
rigorous methods in the papers by BATES[ 19751, KALHOR[ 19761, NAMIOKA,
HARADA and YASUURA [ 19791, WIRGIN[ 1979a, 1979b, 1979c, 1980a, 1980b,
36 RIGOROUS VECTOR THEORIES O F DIFFRACTION GRATINGS [I, J 4

19811, VAN DEN BERG [1980, 19811, HUGONIN,PETIT and CADILHAC

[ 19811.
In conclusion, the possibilities of the Rayleigh theory were underestimated
fifteen years ago, as stressed by WIRGIN,and the Rayleigh methods may be
valuable in the study of some types of gratings, in particular the FSM. However,
because of the theoretical and numerical limitations, these methods are far from
being a near universal tool such as the integral method described in Q 5, at least
for the present.

4 4. The Waterman method

The basic idea of the WATERMAN method, for the FTE case, is to use the
completeness of the set of functions @,(x, -f(x)) to deduce the function $(x)
from (2.24a), then to calculate the amplitudes B, by introducing $(x) in (2.22a).
As regards the FTM case, the same procedure may be applied using (2.24b)
and (2.22b). It is worth noting that the fundamental equations (2.24a) and
(2.24b) are sometimes presented as consequences of the so-called “extinction
theorem”, well known in Optics and Electromagnetism. As for Rayleigh’s
method, there exist various ways of implementing (2.24a). For instance, a point
matching method can be used by representing $(x) by its values at 2N + 1
points x, of the interval (0, d ) and by writing (2.24a) for n E ( - N, + N ) . If a
trapezoidal rule is used, the coefficients of the linear system of equations to be
inverted are proportional to @,(x,, f(x,)). Obviously, the remarks of Q 3.3
about the ill-conditioning of the system for the PMM in the Rayleigh theory
apply to the PMM in the WATERMAN theory as well. Therefore, the numerical
limitations may be expected to be close to each other. A similar remark can be
made about the FSM where $(x) is represented as a series of exp(ia,x). The
coefficients,proportional to (exp - ia,x, @,(x, -f(x))) ,will be closely linked
with those obtained in the Rayleigh method, according to (3.3), and the
numerical limitation may be expected to be of the same order (cf. for instance
WIRGIN[ 1979a, 1979b1).
This method, or methods relying on the same basic ideas, has been used by
various authors (WATERMAN [ 19751, URETSKY[ 19651, DE SANTO [ 19751,
WHITMAN and SCHWERING [ 1977],WIRGIN[ l978,1979a, 1979b],WHITMAN,
LESKIVand SCHWERING [ 19801. An interesting feature of the DE SANTO
method, which has been numerically implemented (JORDANand LANG[ 19791,
LANGand JORDAN[ 19801) lies in the representation of $(x) by the product
of a Fourier series by the physical optics approximation (viz. the function $B
 THE INTEGRAL METHOD 31

given in (2.25)). As may be expected from Fig. 2.4, this feature enhances the
accuracy of the numerical results (at least when I / d is low), as announced by
WHITMAN,LESKIVand SCHWERING[ 19801 who developed the same
formalism independently. In conclusion, the main quality of the Waterman
methods rests on its simplicity, equivalent to that of the Rayleigh method.
Unfortunately, the numerical limitations of these two methods are also very
close to each other.

6 5. The integral method

5.1. HISTORICAL SURVEY AND INTUITIVE APPROACH

To our knowledge, the integral method has been the most widely used of the
rigorous theories of gratings. Until now, it has been the most versatile and the
most accurate tool for investigating grating properties. On the other hand, its
main shortcoming lies in its mathematical complexity and in the toughness of
numerical implementation.
After the pioneering studies of PETITand CADILHAC [ 19641, WIRGIN[ 1964,
19671, URETSKY[ 19651, PETIT [ 1965, 1966a1, PAVAGEAU,EIDO and
KOBEISSE [ 19671, the various ways of implementing this theory have been
widely investigated, at least for perfectly conducting gratings. Thus, many
integral treatments of the grating problem are now available (GREEN[ 19701,
DUMERY and FILIPPI [ 19701, MAYSTRE and PETIT[ 1970a, 1970b1,MAYSTRE
[ 1974, 1980a1, VAN DEN BERG[ 1971a, 1971b1, KALHORand NEUREUTHER
[ 19711, M c PHEDRAN[ 19731, BOTTEN [ 1978a, 1978b], GARCIAand
CABRERA [ 19781). Many of these methods have been inspired by the ideas of
MAUE[ 19491. M c CLELLAN and STROKE[ 19661 also published an integral
equation for the FTM case but no numerical application supported the theory,
which seems to be questionable since it requires the continuity of the tangential
component of the magnetic field on the grating surface.
In spite of its complexity, the integral formalism lies on a very simple and
intuitive background as we shall briefly describe.
Due to the incident field, there exists on 9 a surface current density j9(x).
Unfortunately, finding j,(x) is not straightforward: the existence of j , ( x ) at
a given point M of 9is not only the effect of the reflection of the incident wave
at this point, as assumed in the BECKMANN theory; in fact, the surface current
densitiesj,(x') at the other points M ' of 9 act like secondary sources which
radiate a field in all directions, and in particular toward M. So, j,(x) depends
38 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 5

not only on the characteristics of the incident jield at M, but also on the value of
j y ( x ‘ ) at the otherpoints M ’ of 9
Now, it is felt intuitively that the diffracted field is nothing but the field
generated by this surface current; therefore it may be expected that the
knowledge of the function j,(x) will enable us to express the diffracted field
F at any point of region R . This intuitive fact is fully confirmed by mathe-
+

matics: after defining the field generated at the point P of region R by a single
+

elementary current placed at M ’ E B (the so-called Green function), we obtain

F(P) by integrating on 9 the effects of all these elementary currents. So, we
obtain an integral expression of F(P) containing the function j9(x). Even
though j!,(x) is unknown, the first step of the integral formalism is achieved:
the problem ofjinding a function F(x, y ) of two variables reduces now to thejinding
of a function j,(x) of one variable.
It is very interesting that the integral expression of F(P) in terms ofj,(x)
satisfies both the Helmholtz equation (2.8) and the radiation condition (2. lo),
even though an arbitraryfunction a ( x ) is introduced within it, instead ofj,(x). Of
course, the actual value o f j , 9 ( x ) will be found by writing that the integral
expression satisfies the boundary conditions (2.9a) or (2.9b) on B:the condition
so obtained is an integral equation. The remarkable mathematical feature of this
equation is that it concentrates all the conditions ofthe boundary value problem.

5.2. THE BASIC MATHEMATICAL FORMULAE

The basic mathematical formulae are demonstrated in Appendix B. For the

reader not interested in the mathematical demonstrations, let us summarize the
results in the following theorem, called below FTIM (Fundamental Theorem
of Integral Method).
If a pseudo-periodic function U(x,y ) , continuous everywhere except on 2
satisfis (B.8) and (B.9), i.e. Helmholtz equations in R and R - , and radiation
+

conditionsfor y -+ & 00, it can be expressed by (B. 10) at anypoint of space in terms
of thejumps of U and d Uldn on 9 Furthermore, the limits U * of U on both sides
of 9 are given in terms of these jumps by (B.21) (or (B.24) in an operatorform)
and (B.25) provides the values d U It ldn of the normal derivative in the particular
case where U is continuous on 9? Outside the grooves, U(x,y ) can be represented
as series ofplane waves (B. 15) and (B. 16) whose coeficients are given by (B. 17),
(B. 18) and (B. 19).
It is worth noting that Appendix B uses “Dirac functions” 6(x), easy to
handle for the physicist but incorrect for the mathematician. The interested
1, B 51 THE INTEGRAL METHOD 39

reader may find rigorous demonstrations of the above theorem using the
[ 1966a, 1966bI) in a paper by the
Schwartz theory of distributions (SCHWARTZ
present author (MAYSTRE [ 1980al).

5.3. THE FTE CASE

The FTE case is described by the Dirichlet boundary value problem stated
by (2.8), (2.9a), (2.10). According to 0 5.1, the first step is to express the
diffracted field I: in terms of the surface current density. We define the
“generalized diffracted field” U(x, y ) at any point of space by the difference
between the total field (which vanishes in region R - ) and the incident field:
UX, .Y) = F(X, v) in R , ,
= -F’(x,y) in R - . (5.1)
Obviously, U(x,y) fulfills the conditions of the FTIM and can be expres-
sed in terms of q = (1 + f’(~)~)’/’(dU+/dn(x) - dU-/dn(x)) and
z = U+(x) - U-(x). Bearing in mind the definition of U stated in (5.1), the
boundary condition (2.9a) clearly shows that z = 0, while (2.22~)and (2.23a)
indicate that q = $, i.e. a function closely linked with the surface current density
j,?. So, the FTIM enables us to show that the intuitive remark stated above
is true: the diffracted field can be expressed in terms ofj,. From (B.lO) we
deduce:
W ,Y)= jod@’b, y , x’ 1 $ 4 ~) dx’
’ , (5.2)

I ( x , y , x‘) being given by (B.11).

The FTIM also allows us to state that above the grooves, the diffracted field
can be represented by a Rayleigh expansion, as previously shown in 0 2.5. This
enables us to find again the grating formula (cf. (2.15)) and to express the
coefficients Bn of (2.13) using (B. 17):

s,’ being given by (B.18). This formula is identical to (2.22a) but has been
demonstrated in Appendix B for any value of n. Equation (2.24a) may be
derived from (B. 17) as well, by writing that Uidentifies with - E’ in region R - .
To find $(x), we remember that the right-hand member of (5.2) satisfies both
(2.8) and (2. lo), even though an arbitrary function is introduced in the integral
instead of $(XI). To fulfill the last condition (2.9a), we write:
U+(x) = u- (x) = - F’(x, f(x)). (5.4)
40 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I. I 5

Replacing the left-hand member by its value given by (B.21) yields:

with
which is an integral equation of the first kind (PETITand CADILHAC[ 19641)
since the unknown function $ is not present outside the integral contained in
the left-hand member. The kernel G(x, x’) is given by (B.22) and (B.11). It is
straightforward to see that introducing the solution of (5.5) in the right-hand
member of (5.2) provides an expression for U (thus an expression of F in R ) +

which satisfies all the conditions of the FTE case boundary value problem.
Another equivalent way to write the boundary condition for F is:

Indeed, U given by (5.2) satisfies the Helmholtz equation in region R - and

radiation condition for y -+ - 00. So, if this expression satisfies (5.7), it must
identify with - F’(x, y ) in region R - , since - F’ is an obvious solution of this
Neumann boundary value problem whose solution has been assumed to be
unique in $ 2.2. Now, since Ugiven by (5.2) is continuous, we derive that (5.4)
is satisfied. Thus, (5.7) implies (5.4). Finally, introducing the value dU-ldn
given by (B.25) in (5.7), then multiplying the two members by (1 + f ‘ ( ~ ) ~ ) ’ ’ ~
yields:

with $o(x) = - i(p + af’(x)) exp(iax - ipf(x)), (5.9)

N ( x , x’ ) being given by (B.26), which is an integral equation of the second kind
(PAVAGEAU, EIDO and KOBEISSE[ 19671, PAVAGEAU and BOUSQUET
[ 19701). Either (5.5) or (5.8) may be used to determine 4.

5.4. THE FTM CASE

In the FTM case, we have to solve the Neumann boundary value problem
stated by (2.8), (2.9b) and (2.10). After using (5.1) to define a “generalized
diffracted field” which satisfies the conditions of the FTIM, U(x,y) can be
expressed by (B.10) as a function of its jump, owing to the continuity of its
normal derivative stated in (2.9b):

u(x, Y ) = jod
B(x, Y , x’ 1 $(x’ 1 dx‘ , (5.10)
1, J 51 THE INTEGRAL METHOD 41

B(x, y , x’ )being defined by (B. 11) and $(XI)being the surface current density,
according to (2.22d) and (2.23b). As in the preceding section, U(x, y ) has been
expressed as a function of the surface current density, which satisfies (2.8) and
(2. lo), whatever the function introduced in (5.10) instead of $(x’ ).
Due to the limitation of the FTIM for non-continuous functions U(x, y), it
is not possible to express directly the boundary condition (2.9b). Let us show
that the very simple condition:
W ( X ) = - F’(x, f ( x ) ) (5.11)
is equivalent to this boundary condition.
Indeed, the uniqueness of the solution of the Dirichlet boundary value
problem allows us to assert that Ugiven by (5. lo), which satisfiesthe Helmholtz
equation in region R - and the radiation condition for y -,coy is equal to the
obvious solution - F’(x,y) in region R - if (5.11) is satisfied. So
dU-/dn(x) = - dF’/dn(x, f(x)) and since the expression (5.10) has a con-
tinuous normal derivative on 9 d U + / d n = -dF‘/dn and (2.9b) is satisfied.
Using (B.24) to express the left-hand member of (5.11) yields:
*/2 = $0 + N’ { $1, (5.12)
N’ being given by (B.23) and (B.12).
After solving this integral equation of the second kind, the amplitudes En of
the Rayleigh expansion of F above the grooves are deduced from (B.17):
(5.13)
t,‘ being given by (B.19). This formula extends (2.22b) to any value of F,,.
It is interesting to notice that other integral equations may be obtained for
FTE and for FTM cases, using the FTIM (MAYSTRE [ 1980al).

5.5. NUMERICAL CONSIDERATIONS

In the first attemps at solving the above integral equations, a Fourier Series
Method was used: the unknown function was represented as a series of
exp(ia,x). Truncating this series from m = - N to m = + Nleads to a system
of 2N + 1linear equations with 2N + 1unknowns: the coefficients of the series.
This method is interesting for echelette or trapezoidal gratings since the
coefficients of the linear system can be calculated in closed form (PETITand
CADILHAC [ 19641, MAYSTREand PETIT[ 1970a, 1970bl). Unfortunately, it
does not supply great precision and is difficult to generalize to other profiles.
42 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 5

This explains why, to our knowledge, this method has been abandoned and
replaced by a point matching method.
In ( 5 . 9 , (5.8) and (5.12), we have to solve an equation of the form:

4x) = yo(x) + Jod L(x, x’) Y(X‘) dx‘, (5.14)

where yo is a known function, L being the kernel, given by a series, c a coefficient

(equal to 0 for (5.5)) and y the unknown. Equation (5.14) is written at M points
xi of the period from x = 0 to xM < d:

oy(xi) = yo(xi) + Jod L(xi, x ’ ) y(x’) dx‘, i~ (1,M). (5.15)

It is now sufficient to express the integral in terms of the M unknowns ?(xi).

With this aim, we set:
M
JodL(xi, x’) y(x’) dx’ = C L, y(x,). (5.16)
j = 1

Introducing the right-hand member of (5.16) in (5.15) yields a linear system

of M equations with M unknowns. So, the main work is to find “good values”
of L,. Indeed, these values cannot be exact from a mathematical point of view,
at least ifMis finite: we know that the integral of the left-hand member of (5.16)
cannot be rigorously achieved with a finite number of sample points. So, the
problem is to choose the best method of integration. When the kernel is regular,
this is no great undertaking: because of the periodicity of the integrand, the
trapezoidal rule is very precise (MAYSTRE [ 1980a]), and the value of L, will
be given by:
L, = (xi + - xj - * L(xi, xj)/2, ifj # 1 or M ,
, ,
xi - being replaced by xM - d for 2 = 1, and xi + being replaced by d for
j M.
=
The kernels N and N‘ of (5.8) and (5.12) are regular, even when x’ + x ,
contrary to what is expected at first glance (PAVAGEAU, EIDOand KOBEISSE
[ 19671, MAYSTRE [ 1974, 1980al). This fortunate feature no longer holds good
for ( 5 . 9 , where the kernel has a logarithmic singularity when x’ + x. Here,
efficient methods may be employed to remove the singularity and perform the
computations of L, (MAYSTRE [ 1980al). The number M of sample points on
9 may be estimated using the following empirical rule: about six points per
wavelength must be placed on 9 if a precision of about 1% is required. In all
cases, great attention must be paid to the summation of the kernels: a rough
1 9 8 61 OTHER METHODS 43

truncation of the series leads to imprecise numerical results (MAYSTRE

[ 1980al). The vital importance of the care which must be devoted to these
numerical problems of summation and integration cannot be over-emphasized.
From our own experience, we are able to warn the non-specialist, that a rough
numerical application of the integral formalism cannot furnish reliable numerical
results.
Finally, because of the presence of f ’ ( x ) or f ’ ( x ’ )in the expression of the
kernels N and N ‘ (cf. (B.26) and (B.23)), the question which arises is: what
happens when the profile has edges? In fact, the problem of edges is a serious
and general problem in electromagnetism, which cannot be treated in this
chapter, and we must refer to the studies of MEIXNER[ 19721. For our specific
problem, it will be noticed that the numerical difficulty disappears as soon as
the function f ( x ) is replaced by its truncated Fourier series, whose correspond-
ing profile has no edges. Numerical experience shows that in general the
efficiencies of the new gratings so obtained tend rapidly toward those of the
grating having edges, when the number of Fourier coefficients increases. When
the profile cannot be given by a function y = f ( x ) (for instance when a vertical
x = xo cuts 9 at several points), the integral formalism applies as well,
provided s is used instead of x in the integrals.

5 6. Other methods
6.1. DIFFERENTIAL METHODS

Differential methods have played a valuable role in the field of grating

theories, but mainly for dielectric and metallic gratings, and that is why one of
these methods will be described in detail in the next paragraph. For perfectly
conducting gratings, two kinds of differential methods have been used. In the
first method, the Maxwell equations are projected onto the x , y and z axes and
directly written at a finite number of points placed inside the groove between
y = y, and y = f ( x ) .This is a striking example of a very simple theory, which
unfortunately can lead to strong numerical instabilities and long computer time
(MOAVENI, KALHORand AFRASHTEH[ 19751).
More sophisticated are methods of the second kind, whose basic idea is to
replace the x , y coordinates by new coordinates X,Y which map the grating
surface 9 onto the plane Y = 0. The expression F(X,Y) of the diffracted field,
in the new coordinate system, no longer satisfies the initial Helmholtz equation.
However, the more complicated equation so obtained leads, after projection of
44 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 6

F on the exp(icr,X), to a system of coupled differential equations which can

be solved on the computer. Thus, in outline, the basic idea is to replace the
complicated boundary condition of 9 by a simpler condition, on Y = 0. Unfor-
tunately, there exists a disadvantage: the Helmholtz equation is replaced by a more
complicated one.
The earliest study in this domain employs a conformal mapping technique
(NEVIERE, CERUTTI-MAORI and CADILHAC[ 19711, NEVIERE,CADILHAC
and PETIT[ 19731, VINCENT[ 1980bl). It leads to short computer times and
good accuracy provided the adequate conformal mapping is found. A second
study, using the simpler transformation Y = y - f (x) (CHANDEZON,
MAYSTRE and RAOULT[ 19801) appeared more recently. This work has been
extended to finite conductivity grating covered with dielectric layers (CHANDE-
ZON,DUPUIS,CORNETand MAYSTRE[ 19821) and the resulting computer
code can deal with very deep gratings.

6.2. MODAL METHODS

Modal methods are of great interest and have in common three fundamental
features: they are based on very simple mathematics; they are very eflective and
can deal with highly modulated gratings with low computer time; they are restricted
to special groove geometries. More precisely, they can apply as soon as the field
inside the grooves can be expressed in the form of a modal expansion, similar
to the modal expansions encountered in the problems of wave-guides. In this
case, the grating problem may be solved by matching the modal expansion to
the Rayleigh expansion on the line y = y,. This leads to the inversion of a linear
system of equations whose unknowns are the coefficients of the modal
expansion. This method is particularly useful for lamellar gratings, i.e. gratings
whose grooves are rectangles, the sides of which are parallel to O x and O y
(MAYSTREand PETIT [1972b], HESSEL,SCHMOYSand TSENG[1975],
ANDREWARTHA, Fox and WILSON[1979a, 1979b1). Though the theory is
presented in quite a different manner, we can class in this group also the
methods used for lamellar gratings by WIRGIN[ 1967, 19691 and WIRGINand
DELEUIL [ 19691. The modal theory has been generalized by ROUMIGUIERES,
MAYSTRE and PETIT[ 1974, 19751 to perfectly conducting rectangular rods
lying on a dielectric stack. Recently, a new modal theory has been applied by
BOTTEN,CRAIG,M c PHEDRAN, ADAMSand ANDREWARTHA [ 1981a, 1981bl
to rectangular groove dielectric or metallic gratings.
JOVICEVIC and SESNIC[1972] used a modal theory to investigate the
1, § 61 OTHER METHODS 45

properties of echelette gratings, but their method relied on a questionable

assumption (PETIT[ 1980]), contrary to the formalism of ITOHand MITTRA
[ 19691 for the same gratings, which is quite rigorous. This kind of formalism
is also well adapted to the study of gratings with semi-circular grooves
(ANDREWARTHA, DERRICK and M c PHEDRAN[ 1981al). A generalization of
the modal theory to gratings with arbitrary profiles has been proposed by Fox
[ 19801, and also by ANDREWARTHA, DERRICK and M c PHEDRAN[ 1981bl
but, in our view, the modal method loses its main advantages over other
classical methods for such arbitrary profiles.

6.3. MISCELLANEOUS

MARCUSE[1976] proposed for dielectric gratings an original method in

which the field inside the grooves is represented by a double Fourier series, with
a period d in x and a period d’ > H in y. He applied this method to echelette
gratings, though some numerical difficulties appeared for large groove depth.
The specialist of boundary value problems would be astonished by the
absence of a variational method for the grating problem. A recent study by
BREIDNE and MAYSTRE [ 19821 investigated the possibilities of this method,
very well known in electromagnetism (VANBLADEL[ 19641) and in many other
domains of physics. Comparisons with results obtained from the integral
formalism show that this method gives accurate results for medium-depth
gratings (H/d 5 0.6) and is particularly well adapted to the study of ghosts, viz.
directions of diffractions not predicted by the grating formula and due to
irregularities of period nd on the grating profile.
Finally, let us point out the original work of FAcQ [ 1976, 19771who investi-
gated some properties of finite gratings using an integral method. Though it is
not easy to renounce the periodicity of the grating problem, FACQ was able to
investigate the properties of gratings having a small number of grooves (less
than fifteen) using the inversion algorithm of AKAIKEand ROBINfor large
systems of linear equations (AKAIKE[ 19731).
It is difficult to give some idea of the computation time required by all the
methods described in this paragraph and in previous ones. They are generally
very low, less than one second on a CDC 7600 computer. The reader interested
in more precise details may refer to a recent work for comparison of compu-
tation times of the Rayleigh methods with that of a computer code written by
the present author using an integral formalism (HUGONIN,PETIT and
CADILHAC [ 19811). As regards the precision of the results, the experience of
46 RIGOROUS VECTOR THEORIES O F DIFFRACTION GRATINGS [I, § 7

our laboratory, where practically all these methods have been numerically
implemented, allows the statement that the results of the rigorous methods
agree to within 1% in the resonance region, provided the numerical implemen-
tation is carefully carried out. As far as the integral, differential and modal
methods are concerned, an agreement better than 10 - in relative value on the
efficienciesis reached in general. Perhaps, the reader will judge such a precision
to be quite impractical. This opinion was justified until recent years but at
present, this precision is sometimes needed for important investigations, for
example the use of gratings as beam samplers for high power lasers (MAYSTRE,
NEVIERE and PETIT[ 19801).

8 7. Metallic and dielectric gratings

Nowadays, the computer codes devoted to real metallic gratings are in most
widespread use, even though they are more complicated and time-consuming
than those devoted to perfectly conducting gratings. This is due to theoretical
(PETIT,MAYSTRE and NEWERE[ 19721, MAYSTRE [ 1973, 19741) and experi-
mental (HUTLEYand BIRD [ 19731, LOEWEN,MAYSTRE,M c PHEDRAN and
WILSON [ 19741) evidence that the model of perfectly conducting gratings fails
in the visible and near infrared regions, at least for the FTM case. This
phenomenon is not intuitive since the reflectivity of metals like aluminium,
silver or gold employed for the construction of commercial gratings exceeds
90% in these regions.

7.1. BOUNDARY VALUE PROBLEM AND ELEMENTARY PROPERTIES

In this section, R - is filled with a dielectric or a metal of complex index v.

Proceeding as in 2.2, but this time with the Maxwell equations in both regions
R and R ~,it can be shown that the total field does not depend on z and
+

remains TE (or TM) polarized when the incident field is TE (or TM). So, for
both polarizations, we shall deal with the projections of the field on Oz. The
diffracted field F(x, y ) is defined as the difference between the total field and
the incident field in R and identifies with the total field in R - . So, it is possible
+

to define the FTE (or FTM) cases, where the incident electric (or magnetic)
field F' is given by (2.11). In these fundamental cases, the diffracted field F
obeys the following boundary value problem:
V2F + k2F = 0 in R +,
1, Q 71 METALLIC AND DIELECTRIC GRATINGS 47

V2F + k2v2F= 0 in R -,
F (x) + F'(x, f(x))
+ = F - (XI,

dF-
d F +/dn(x) + dF'/dn(x, f(x)) = C, - (x), (7.4)
dn
F satisfies a radiation condition for y-, k 00, (7.5)
with C, = 1 for FTE, (7.6)
C, = l/v2 for FTM.
It is worth noting that, for the FTE case, (7.3), (7.4) and (7.6) mean that the
total field and its normal derivative are continuous on 9?This continuity holds
for the total field but not for its normal derivative for the FTM case since
C, # 1. The uniqueness of the solution of the FTE problem has been shown
by CADILHAC [ 19801, at least when v is real. As in 5 2.4 and 2.5, it can be
established from the above equation that F is pseudo-periodic in x, and is
described above the top of the grooves by the Rayleigh expansion (2.13) and
below the bottom of the grooves by another Rayleigh expansion:
F(x,y) = EnB P ) exp(ia,x - i/?(ny)y) ify <y,, (7.7)

with p'," = (k2v2 - a:)'/2. (7.8)

The above definition of is ambiguous since v may be complex. Its
determination will be fixed by the following inequality:
Re{P',".') + Im{P:)} > 0, (7.9)
which generalizes the convention for p, stated in (2.12b). So, the classical
grating formula for reflected orders is unchanged but must be completed for
dielectric gratings (v real) by the grating formula for transmitted orders, which
deduces easily from (2.12a) and (7.8):
v sin e(;) = sin 0 + nA/d, (7.10)
where the diffraction angle of the nth transmitted order is measured
anticlockwiselike the incidence angle. The set U(') of propagating orders in R -
may be defined easily when v is real but, more generally, U(") will be defined
for an arbitrary value of v by:
Re{/?:)} - Irn{p?)} >0 ifIm{v} > 0,
U'"' = U(V) ifIm{v}<O.
48 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 7

In order to state some fundamental properties of suchgratings, it is interesting

to generalize the lemma of $ 2.6. We consider two pseudo-periodic functions
u and u containing incoming and outgoing waves on both sides and satisfying:

V2u + k2u = 0, V2u + k2u = 0, inR+,

V2u + k2v2u = 0, V2v + k2T2u = 0, inR - ,
and the boundary conditions (7.3) and (7.4) on 9,
v having a positive imaginary
part. So, u may be written in the form:
u= c,lsu
A,, exp(ia,x - ip,y) + 1,B, exp(ia,x + $,y) in R ,
+

+ c B',")exp(ia,x - ifi,")) in R - . (7.11)

n

By definition, the expression for u will be derived from the above expressions
for u by replacing A,,B,, A',") and B',") by A:, B,!,,A',")' and B',")', p'," being
replaced by flp)= by),U(') being equal to U(").Proceeding as in $ 2.6 for R , +

then R - yields:

dv' -du+
u+
dn
-) ds = x
nE U
p,,(A,K - B,K), (7.12)

and from the boundary conditions (7.3) and (7.4),we can eliminate the left-hand
members of the above equations:
p,,(A,K - B , K ) = C, 1 p',")(BP)BT- A',")AT). (7.14)
ns U nE U'"'

Using the same functions u and u as in $ 2.6 for the reciprocity theorem, the
"reciprocity by reflection" stated in (2.20) remains unchanged, whatever the
index v of R - (MAYSTRE and M c PHEDRAN [ 19741).
For the "reciprocity by transmission" (NEVIERE and VINCENT [ 1976]), the
function u is defined as the conjugate of the total field v" of Rayleigh coefficients
B,: above 2 which exists when an incident wave exp(ia"x + i v y ) propagates
upward in R - , with 01" = - 01, and pl, = p!,'". The result is that, when n E U :
BB," = CP p(")B',").
" (7.15)
When v is real, the energy balance criterion can be derived from (7.14) by
1,s 71 METALLIC AND DIELECTRIC GRATINGS 49

setting u = v, which yields:

1 P n I 4 I 2+ C, 1
neU flcU'"'
b\y)lA?)12 = 1U flflIInI2+ C, 1u(")fl?)IE?)12.
ns nE

(7.16)
Here, also, the use of Poynting's theorem shows that the quantities in (7.16)
are no other than the flux of energy of incident waves in R + (first term), or
in R - (second term), and the flux of energy of diffracted waves in R (third
+

term) or R - (last term). So, the physical meaning is very simple: the incident
energy is equal to the diffracted energy. It is interesting to note that, when v is
real, (7.15) reveals that the efficiency is the same in the two cases: there exists
a "reciprocity by transmission".

7.2. THE INTEGRAL FORMALISM

The first rigorous theoretical studies on this topic proposed the solving of two
coupled integral equations (WIRGIN[ 19681, NEUREUTHER and ZAKI[ 19691,
VAN DEN BERG [ 1971a, 1971bl). Indeed, contrary to what happens with
perfectIy conducting gratings, where j , is a relevant function to express the
diffracted fieid, here the physical meaning cannot suggest a comparable
function defined on the profile 9 However, it turns out that the field F ( x ) on
+

9 and its normal derivative d F /dn(x) are sufficient to express the diffracted
+

field in R and R - using (B. 10). Indeed, (B. 10) may be used in the first
+

instance to express a function of U and d U/dn being equal to F and d F /dn.

+ -+

Likewise, the field F i n R - may be expressed by applying (B. 10) in region R -

(viz. replacing d and B by d(") and B(")deduced from (B.11) and ( B . 1 2 ) by
replacing Pn by P$"))to a function equal to 0 in R and to F i n R - . Of course,
+

two integral equations are necessary to find the two unknown functions: they
are obtained by using the boundary conditions (7.3) and (7.4). Unfortunately,
these two integral equations are coupled, a fact which leads to serious numerical
limitations.
Now, we shall present the theory developed by the present author, which
leads to a single integral equation (MAYSTRE [ 1972, 1973, 1974, 1980al). From
an intuitive point of view, this formalism relies upon the following idea: a
fictitious su$ace current density j,(x) exists which, placed on would generate
in free space (the materialfilling region R - being replaced by air) a di@actedjield
F - identical in R to the actual dfracted jield F. Of course, the field F'
+

diffracted by this fictitious surface current density in R - has no physical

meaning. If we assume the existence of this functionj,(x), we must notice that
50 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, § 1

it enables us to express F’ at any point in space and thus, to express Fin region
R + . So, Appendix B can be used to express the limits F and d F /dn above
+ +

9 and then the limits F - and dF-/dn will be deduced from (7.3) and (7.4).
Now, the field in R - may be derived from F- and dF-/dn using (B.lO) for
a material of index v. Finally, we have expressed the field at any point of space
in terms ofj,v(x) and obviously, a single integral equation is enough to determine
this single unknown function. It is worth noting that, from this point of view,
the theory leading to a pair of integral equations is redundant: in fact, the two
unknown functions F and d F /dn may be deduced from a single relevant
+ +

function j:p(x).
More precisely, the mathematical steps of this method are outlined in
Fig. 7.1. We first define a function U(x, y ) by:
U=F inR+, (7.17)
V2U+ k2U= 0 inR-, (7.18)
u-(x) = U+(X), (7.19)
U satisfies a radiation condition for y + - co. (7.20)
The function U,which is defined simply by (7.17) in R , is also well defined
+

in R - by (7.18)-(7.20), due to the existence and uniqueness of the Dirichlet

problem assumed in 8 2 (since U’ = F’ is fixed, U - is fixed too, according
to (7.19)). Moreover, U obviously satisfies the conditions of the FTIM stated
in J 5.2. So, using (B. lo), U(x, y ) which is continuous on S (cf. (7.19)) may
be expressed at any point of space from the jump $(x)/(l + f’(x)’)’/’ of its
normal derivative:

(7.21)

Fig. 7.1. An integral method for finite conductivity gratings in outline.

 METALLIC AND DIELECTRIC GRATINGS 51

It is interesting to notice that the above formula, quite similar to (5.2), gives the
expression of the field U generated in free space by a surface current density
j.(x) = i$(x)(l + f ' ( ~ ) ~ ) - ~ / So, ~ / we
~ pnow
~ . have the expression of F in
R (arrow a of Fig. 7.1) and (B.17) allows us to express the coefficients B, of
+

the diffracted field in R : +

Bn = [s,', $1. (7.22)

Furthermore, (B.24) and (B.25) give the limits F and d F /dn of U above 9+ +

(arrow b):
F+ = G { $ } , (7.23)
(1 + S ' ( X ) ~ ) d' /F~+ / d n = $/2 + N { $}. (7.24)
The limits F- and dF-/dn can be easily deduced from (7.23) and (7.24) by
adding the incident field to F (arrow c) and using the boundary conditions (7.3)
and (7.4) (arrow a):
F- = G{$} + $0, (7.25)
(1 + f ' ( ~ ) ~ )dF-/dn
'/~ = [$/2 + N { $ } + $o]/C,,, (7.26)
$, and q0 being given by (5.6) and (5.9).
Now, to express the total field Fin R - in terms of $, it suffices to use (B. 10)
for a function U'(x, y ) satisfying the conditions of the FTIM (but this time with
a Helmholtz equation of wave-number k v ) :
U'(x,y) = 0 inR+,
U'(x, y ) = F(x, y ) in R - ,
and (B.lO) yields (arrow e):

u'(x,y) = - IOd d ' " ) ( x , y , x')

dF-
~

dn
(x')(l + f'(x')2)'/2 dx'

(7.27)

Equation (B.17) permits us to evaluate the amplitude B',") of the Rayleigh

expansion of F (i.e. U ' ) in R - from (7.25) and (7.26):

1
B Y ) = _ _ [&"'-, $42 + N { $ } + $01 - [t',"'-, G{$} + $01, (7.28)
CP

s2)- and t2)- being obtained from s; and t,- by replacing b,, by fir).Similarly,
52 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, 8 7

putting (7.25) and (7.26) in (B.24) and bearing in mind that U' = 0 above 9
allows us to express the limit F - of U' below 9 (arrow f):
F- = (G{$} + $OY2 - G'"'{$/2 + N{$l + $ O W P - N(")'{G{$1 + $01.
(7.29)
The integral equation is obtained by writing that the above value is linked to
F + + F', by (7.3). Using F' given by (7.23) yields:

= - ($0/2 t N ( " ) ' {$o} t G'"'{ $o}/Cp). (7.30)

It can be proved that the diffracted field F derived from (7.21) and (7.27) well
satisfies the boundary value problem stated in 0 7.1, provided (7.30) is verified
(MAYSTRE[ 1980al).
This integral equation is very well adapted to highly conducting gratings:
when I vI --t co, it tends toward integral equations devoted to perfectly
conducting gratings. The above formalism has been generalized to more com-
plicated gratings of great practical interest, for instance metallic or dielectric
gratings covered with a stack of dielectric layers (MAYSTRE[ 1978a, 1978b],
BOTTEN[ 1978a, 1978b1) or bimetallic gratings, i.e. gratings whose surface is
composed of two alternating metals (BOTTEN[ 19801). Another interest of this
formalism lies in the fact that the integral equation is practically the same for
the FTE and FTM cases: it is very interesting to treat these two cases
simultaneously. The computation time is of the order of a second on a
CDC 7600, if one is content with a precision of 12.

7.3. DIFFERENTIAL METHOD

We have seen in 5 6.1 that differential methods permit us to solve the

problem of perfectly conducting gratings. In this section, we shall describe a
particular differential formalism, which had great importance in the develop-
ment of grating theories, not only because of its particular possibilities, but also
since it permitted fruitful comparisons with the results obtained from the
integral formalism of finite conductivity gratings described in the previous
paragraph.
A feature of this particular differential formalism is that it cannot be applied
to perfectly conducting gratings. On the other hand, it can deal with periodic
structures like arrays of identical dielectric (or metallic) cylinders, or even
'generalized gratings' described by a non piecewise continuous index v(x, y )
1, I 71 METALLIC A N D DIELECTRIC GRATINGS 53

periodic in x, for which domains R and R - cannot be defined. For the sake
+

of simplicity, the description of the formalism will be restricted to the simplest

case of Fig. 2.1, where v(x, y ) is equal to 1 in R and to v in R - .
+

The method was first proposed by PETIT [1966b] and implemented by

CERUTTI-MAORI, PETIT and CADILHAC [ 19691. Unfortunately, due to the
algorithm used in the computer code for solving the system of coupled
differential equations (finite difference method), strong difficulties for con-
ducting gratings emerged. The breakthrough in the differential method was
realized by NEVIERE[ 19751 and NEVIERE, VINCENT and PETIT[ 19741, who
not only improved the formalism but also used reliable and accurate numerical
methods which revolutionized the domain of application of this formalism (see
also CHANG,SHAHand TAMIR[ 19801). However, the curse of this method,
to which no solution has so far been found, lies in its inability to provide reliable
results for highly conducting gratings, such as aluminum gratings in the visible
region, even though some progress has been made recently (VINCENT[ 1980al).
We deal with the FTE case, and denoting by ~ ( xy ,) a function equal to k2
in R and k2 v2 in R -, we deduce that the total field F'(x, y) satisfies a
+

Helmholtz equation obtained by putting (7.1) and (7.2) together and bearing
in mind that F' satisfies (7.1) in R : +

V2F'(x,y) + x(x,Y) F'(X,Y)= 0 . (7.3 1)

Of course, the diffracted field F satisfies (7.31), except on 9 where F is
discontinuous. On the other hand, (7.31) is valid at any point of space, even
on 9 due to the continuity of F' and its normal derivative dF'/dn on 9 stated
in (7.3) and (7.4). One can say that (7.31) is valid in the sense of distributions,
which means, in this particular case, that the definition of the left-hand member
of (7.31) does not cause any problem for those who are not acquainted with
the theory of distributions. Now, we can transform (7.31) by expanding F' and
x in series:
Ft(x, Y ) = C Fn(y) exp (ianx), (7.32)
n

x(x, A = c x , b ) exp(inKx),
n
(7.33)
and we derive easily:

c (Fi
n
- a:F, + c4 xn_,F4) exp(ia,x) = 0. (7.34)
Since (7.34) is valid at any point of space, including 9 (7.34) is satisfied for
any value of x and y, and so any term of the series must vanish:
(7.35)
54 RIGOROUS VECTOR THEORIES OF DIFFRACtlON GRATINGS [I, § 7

with ~,,(y) = 46, - ,- X, - ,(A. (7.36)

Furthermore, (7.31) and the continuity of F' and dF'/dn (and thus of grad F')
on 9 imply the continuity of F,(y) and FA(y) for any vahe ofy. Above y , and
below y,, J,,(y) is constant since x remains constant; thus, we find again that
F' is described by plane wave expansions which can be deduced from (2.13)
and (7.7) by adding the incident field to F i n R . From (7.32), we derive the
+

expressions for F,,(y) imposed by the radiation conditions:

F,(y) = B Y ) exp( -iPp)y) for y 6 y,. (7.38)

n

Obviously, the values of the Rayleigh coefficients B, and B '," may be deduced
fromF,(y,) andF,(y,). Now, to determine the two sets ofF,(yM) andF,(y,)
we must recall that F,(y) satisfies between y,,, and y, a system of coupled
differential equations (7.35). From a mathematical point of view, this is not
sufficient since boundary conditions for F,(y) at y , and y , are necessary.
These boundary conditions may be inferred from the special form of F,(y) at
the limits y , and y,, given by (7.37) and (7.38). Indeed, it is easy to verify on
these equations that:
FA(ym) + iP',"'Fn(yrn) = 0, (7.39)
F A ( h ) - iPnFn(.YM) = -2iP exp(-iPyM)Gn. (7.40)

So, the grating problem has been reduced to the solvingof the system of coupled
differential equations (7.35) with boundary conditions (7.39) and (7.40) on the
boundaries y , and y,.
The numerical implementation is based on the "shooting method". First, we
have to truncate the infinite differential system. A system of 2N + 1 coupled
differential equations with 2N + 1 unknowns F,(y) is obtained by eliminating
the F, having a number n outside ( - N, + N). The first step of the "shooting
method" is to construct a computer code numerically able to provide the
F,(y,) as soon as the F,(y,) are known from - Nto + N. Let us show briefly
that this is not a difficult problem. Indeed, as soon as the F,(y,) are known,
the Fi(y,,J can be derived from (7.39) in a linear manner. Now, standard
algorithmslike Numerov, Runge-Kutta or Adams-Moulton algorithms, enable
the calculation of F,(y) and FA(y) up to y , (VINCENT[ 1980bl). Intuitively, it
is easy to see that as soon as the F,(y) and FA(y) are known all the derivatives
FLm)(y)may be deduced from (7.35) or by differentiating (7.35). So, we can
1, I 71 METALLIC AND DIELECTRIC GRATINGS 55

compute with the desired precision the values F,(y + dy) and FA(y + dy) at a
slightly higher ordinate using, for instance, a Taylor expansion for F,(y).
Proceeding in like manner at M points between y , and y , finally gives F,(y,)
and FA(y,). In fact, the numerical algorithms used in practice are much simpler
and very efficient. From F,(y,) and F:(JJ,), we may compute the set of
un = FA(yM) - iflnFn(yM)* (7.41)
Of course, if the exact values of F,(y,) corresponding to the physical
problem have been employed at the starting point of the integration, U, must
be equal to - 2iB exp ( - iBy,)S,, according to (7.40), ifthe integration has been
done with infinite precision. But the exact F,(y,) are unknown. To understand
the “shooting method”, it must be envisaged that the F,(y,) are replaced by
arbitrary values c, in (7.39), and it must be noted that the corresponding U,
numerically obtained by using (7.39), (7.35) and (7.41) are linearly linked to
these c, arbitrarily chosen at the starting point of integration:

(7.42)
q= -N

By adopting a set of cq = Sq - r , we shall find out the set of U, = T,,, which

permits us to reconstitute the rth column of the matrix T of size 2N + 1. Thus,
2N + 1 successive integrations corresponding to 2N + 1 different values of r
permit the calculation of the entire matrix T.
Now, it suffices to use (7.42) to derive the actual coefficients F,(y,) by
solving a system of 2N + 1 linear equations with 2N + 1 unknowns:

(7.43)

The field between y , and y , and the Rayleigh coefficients B, and B‘,“)may be
deduced easily from F,(y,), as seen before.
The FTM case of polarization may be treated by the differential method but
is more complicated because (7.31) is not true in the sense of distributions.
One of the origins of the instabilities observed using the differential formalism
for highly conducting metals may be understood from an intuitive point of view.
When the conductivity increases, the skin depth decreases. So, for a given
grating of groove depth H,the function F‘(x, y,) tends to a “delta function”
when the skin depth to groove depth ratio becomes very weak. So, the
truncation of (7.35) will entail a large error, whatever the value of N . It seems
that problems of stability of algorithms have at least the same importance in
this failure.
56 RIGOROUS VECTOR THEORIES O F DIFFRACTION GRATINGS [I, B 8

It is interesting to notice that the integration of the system of differential

equations can be performed in closed form in the particular case of the lamellar
grating, where the x,, are independent of y. This feature permits the represen-
tation of the field in the form of “modes” and the numerical application leads
to very efficient computer codes (KNOP[ 19781).

# 8 . The grating in conical diffraction

8.1. DEFINITIONS

The term conical diffraction is used as soon as the incident wave-vector k

does not lie in the x O y plane. All the problems posed by this new kind of
mounting are solved by a theorem of invariance (PETITand MAYSTRE [ 19721,
MAYSTRE [ 1980a]), at least for perfectly conducting gratings. We shall give this
theorem without demonstration. The FTE (or FTM) cases of polarization may
be defined in this more complicated case as well. In the FTE (or FTM) case,
the incident magnetic field (or electric field) is perpendicular to Oz and it can
be shown that the corresponding diffracted fields satisfy the same property
provided the grating is perfectly conducting. To state the “invariance theorem”,
let us associate an equivalent classical mounting (i.e. the usual mounting
described in Fig. 2.1) to an arbitrary diffraction mounting by projecting the
incident wave-vector k and the diffracted wave-vectors k, on the O z axis and
the x O y plane:
k = kxY + PE,
k,, = kf;Y + kzE,
E being the unit vector along Oz.
The equivalent classical mounting is defined by an incident wave-vector
equal to k”y. This not only gives the incidence P but also the wavelength
,lxY = 2.11 kxyl = ,l/cosA, A being the angle between k and the x O y plane.

8.2. THEOREM OF INVARIANCE

When a perfectly conducting grating is used in conical diffraction, the projection

k; of the wave-vector of the nth order difracted wave on the Oz axis is equal to
‘k and the direction of k:y in the xOyplane is identical to the direction of the nth
diffracted order in the equivalent classical mounting. The eficiency in the nth
1. s 91 CONCLUSION 51

difracted order is equal to the efJiciency in the nth difracted orderfor the equivalent
classical mounting, in the same fundamental case of polarization.
So, the directions of diffraction k , lie on a cone containing k (this is the origin
of the name “conical diffraction”) and the projections of the directions of
diffraction on the x O y plane are deduced from Py by the classical grating
formula, using a fictitious wavelength ,Ixy.The computer codes devoted to the
usual mounting can be used to determine the efficiencies from the parameters
B” and AxY. It is worth noting that the invariance theorem enables the
realization of a mounting with a constant efficiency (MAYSTREand PETIT
[ 1972a1): it suffices to keep Pyconstant and to vary ,Iand A in order to keep
A/cos ( A ) constant.
The part of the theorem devoted to the directions of diffraction still holds for
metallic and dielectric gratings. On the other hand, the efficiency of such
gratings does not obey the theorem since the vector boundary value problem
cannot be separated into two scalar boundary value problems: a TE (or TM)
incident wave in general does not generate a TE (or TM) diffracted wave.
However, rigorous integral (MAYSTREand PETIT[ 19741) and differential
(VINCENT, NEVIERE and MAYSTRE [ 19781)methods show that the invariance
theorem can be a good empirical rule for metallic gratings, even in the extreme
UV region, provided the variation of reflectivity of the metal with angle A is
taken into account (MAYSTRE, NEVIERE and PETIT[ 19801).

8 9. Conclusion

Some recent developments in the field of rigorous grating theories are not
described in this chapter. The reader interested in the very complicated problem
of diffraction by “crossed gratings” (viz. gratings which are not invariant in z
but periodic in x and z with groove spacings d and d ’ ) may refer to the review
of Mc PHEDRAN, DERRICK and BOTTEN [ 19801. Though the present computer
codes devoted to these crossed gratings are time-consuming and do not yield
exceptional performance, they have the merit of showing the high interest, as
filters in infrared or solar selective elements, of these new gratings which can
be constructed holographically.
The fascinating field of inverse scattering for gratings has been investigated
in the last few years. Here, the unknown is the profile of the grating, which must
be deduced from an efficiency curve of the grating, for instance, in Littrow
mount. Though the mathematical problems encountered in the first investiga-
tions are tremendously difficult to handle, computer codes have been con-
58 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, A m . A

structed (ROGERand MAYSTRE [ 19811) and the first theoretical reconstitution

of a grating profile from actual experimental data has been achieved by ROGER
and BREIDNE [1980].
At present, one can consider that rigorous vector theories of gratings are able
to solve almost all the classical problems of gratings encountered by manufac-
turers and users. Moreover, the numerous computer codes provide for the
theoretician the opportunity to evaluate the advantages and shortcomings of
each rigorous theory, as well as the precise limits of non-rigorous theories. This
information is invaluable for investigating the somewhat tougher and very
important problems of optics, such as diffraction by non-periodic rough sur-
faces. Finally, these accurate computer codes have often played a role compar-
able to that of accurate experimental devices: numerical exploitation of them
has given valuable clues in phenomenological studies and has enabled the
precise verification of purely theoretical predictions. This explains why some
specialists of grating theories have dubbed themselves “numerical experi-
menters” to show the importance they attach to this research.

Appendix A

Let us assume, by hypothesis, that the Rayleigh expansion R(x, y) converges

for x = xo and y = yo. We derive that the modulus of R,(x, y ) tends to 0 when
n + +a:
~ B n ~ @ n ( x o , y o ) ~i f~no+ *a. (A.1)
Now, it is useful to consider the behavior ofR(x, y ) in C2,i.e. for complex values
of x and y. What we intend to show in the 6rst place is that the RE converges
uniformly and absolutely in the region R E C2 defined by:
IW(XYY)I G lw(xl9Yl)lY (A4
I@(x,r)l G lw(xI9Yl)lY (A.3)
provided x, and y , are real, with y , > yo, which implies from (3.8):
= 4 < 1.
Iw(~lY~l)Iw(xoY~o)l (A.4)
To clarify (A.3), it must be recalled that W(x, y ) denotes the complex function
w(X, 7) which is analytic in C2(contrary to w(Z, 7) or w(x, y)). The exact limits
of a may be derived easily from (3.8) but it suffices to note from this equation
and (A.2) that the subset of R, corresponding to real values of x and y, is the
half plane defined by y > y , .
I, APP. Bl APPENDIX B 59

Assuming, for simplicity, that the incidence angle 8 = 0 (which implies a = 0

and u(x,y) = 1, according to (3.7)), it is interesting to express
I R,(x, y ) I = I B, @,(x, y) I in the form:

Let us examine the four terrns of the right-hand member of (AS) when
n - i +co:
- the first term tends to 0, according to (A.1);
- the second and third terms tend to 1, according to (3.6a);
- the last term is bounded by q", according to (A.2) and (A.4).
So, I R, I is bounded, for sufficiently large values of n, by the nth term q" of a
convergent geometrical series independent of x and y. The same conclusion
may be drawn, for n -i - 00, by replacing w(x, y) by E(x, y) in (AS). This shows
that R(x, y ) converges absolutely and uniformly in 62. Since each term R,(x, y )
is an analytic function of the two complex variables x and y , the series R(x, y)
uniformly converges in 62 to an analytic function of 6'. Of course, the restriction
of R(x, y ) to real values of x and y is an analytic function of the two real
variables x and y in the part of D which belongs to R2, i.e. the half plane y 2 yl.

Appendix B

Let U be a function satisfying in R + the following conditions:

vzU(X, y) + kz U(X, y) = 0,
a radiation condition for y -i coy (B.1)
k 2 being a real positive number or eventually a complex number with a positive
imaginary part. Let us obtain an integral expression of U in R from the values
+

of U and its normal derivative dU/dn on 9

To this end, we first define the elementary solution B(x, y ) of the Helmholtz
equation (B. l), pseudo-periodic in x and satisfying a radiation condition for
y - r +co:
VzY + k 2 Y = 6(y) n exp(inad) 6(x - nd), (B.2)

where 6 denotes the Dirac function. Since B is pseudo-periodic, it can be

expressed as a series of exp(ia,x):
%¶ Y)= c 4 ( r )exP(i%x).
n
60 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, APP. B

Introducing the above expression in (B.2), multiplying by exp( - iax) and

identifying the nth terms of the Fourier series of both members yield:

d2 Bn
-+ 3
/: Bn = S(y)/d.
dY2

Remarking that the right-hand member vanishes for y # 0, it turns out that ’3”
can be expressed in the following form:

Bn(y) = a exp(i&y) + a‘ exp( - i&y) ify > 0,

= b exp($,,y) + b’ exp( - i&y) ify -= 0.

So, 9(xyy ) is expressed for both regions y > 0 and y < 0 in the form of plane
waves, and the radiation condition allows us to eliminate the incoming waves
in such a way that a’ = b = 0. The determination of a and b’ is achieved by
observing that (B.3) implies the continuity of gnat y = 0 and a jump of l/d of
its derivative in y , which entails a = b’ = 1/2id&. Finally, we obtain:

%(x,y) = - C -exp(ia,x
1 1
+ iP,IyI).
2id n p,,

This elementary solution of the Helmholtz equation is no other than the field
diffracted at the point P of coordinates x and y by an infinite set of elementary
currentsj, = exp(inad) in wires placed at the points x = nd of the Ox axis.
To express U(x,y), we consider Green’s function, so called,
$(x - x’,y - y ’ ) which satisfies:

= 6 ( y - y ’ ) C exp(inad) 6(x - x’ - nd), (B.5)

n

the differentiation being made either in x and y or in x’ and y’.

Now, let us outline a classical calculation, which enables us to express U in
an integral form. A first equation is obtained from (B.5) by multiplying the two
members by U(x’, y’ ). The second one is obtained by replacing x and y by x’
andy’ in (B.l), then multiplying the left-hand member by B(x - x’,y - y ’ ) .
Subtracting the second equation from the first, integrating in the hatched
domain of Fig. 2.2 with contour % (the segment SR being at y = co) and using
I, App. Bl APPENDIX B 61

the second Green identity yield:

dU’ d l ‘ -
dn
1 ?!
Y dn’
(x - xf,y - y ‘ ) U +dl’

= U(x,y ) in R ,
+

=o ipR-,
d Bldn’ being the normal derivative of B(x - x’,y - y ’ ) considered as a
function of x’ and y ‘ , U and d U /dn the limit values of U and d U/dn on
+ +

and I‘ the curvilinear abscissa on % It is worth noting that the signs in the
left-hand member of (B.6), unusual in the Green theorem, are imposed by the
sense of the normal 8, which is oriented toward the interior of R + . Since the
integrand of (B.6) is periodic in x’, the contributions of QR and SP cancel each
other. The contribution of SR vanishes, as shown by the lemma of 3 2.6, by
setting u = U(x’,y’), u = B(x - x’,y - y ‘ ) , and remarking that u only
contains outgoing waves whereas u is incoming. So,

U(x,y) 1, B(x - x‘, y - f ( x ‘ ) ) d U + (x’) ds‘

1*
=
dn
- (x - x ’ , y - f(xf)) U+(x‘)ds‘,
!P dn’ (B.7a)
the right-hand member vanishing in R - .
An equation similar to (B.7a) can be written for a function U(x, y ) satisfying
a Helmholtz equation in R - and a radiation condition for y + - co :

U(X, Y)= ~ (X - x’, y - f ( x ’ ) ) U - ( X ’ ) ds‘

.q dn’

- 1, dU-
B(x - x’, y - f ( x ’ ) )-(x’) d d ,
dn (B.7b)

the right-hand member vanishing in R . +

This time, the sign of the right-hand member is the usual one since the normal
is oriented toward the exterior of R - , to keep 8 unchanged on 9
Finally, let us generalize (B.7a) and (B.7b) by considering a function U
satisfying:
V2U+k2U=0 inR+ andR-, (B.8)
a radiation condition for y + f co. (B.9)
62 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I, A m . B

Such a function can be considered as the sum of two functions U1and U,,U,
being identical to U in R and null in R - , U, being null in R and equal to
+ +

U in R - . Using (B.7a) and (B.7b) to express U,and U, in all the space, then
adding the two expressions to derive U = U,+ U, and finally replacing ds‘ by
(1 + f ( ~ ’ ” ) ’ ’dx‘
~ yield:

d(x,y, x’) ~ ( x ’dx’

) + B(x,Y,x’) ~ ( x ’dx’
) (B.lO)

with:
1 1
d(x,y,x’) =
21d
Cn -exp[ia,(x
8,
- x’) + i&ly - f ( x ’ ) l ] , (B.ll)

x exp[ia,(x - x’) + ifinly - f(x’)l], (B.12)

~ ( x ’=) (1 + f ’ ( ~ ’ ” ) ’ ’[dU+/dn(x’)
~ - dU-/dn(x’)], (B.13)

T(X‘) = U+(x’) - u-(x’). (B.14)

When y > y , o r y < y m , the sign ofy - f ( x ’ ) is equal to + 1 or - 1 and

remains constant, so, the symbol I I in (B.11) and (B. 12) may be replaced by
+ ( ), the terms in x and y may be removed from the integrand of (B. 10) and
-
we derive:

with

(B.18)
 APPENDIX B 63

1 an
t,’ = -(* 1 - -f ’ ( x ) ) exp( - ianx T i&f(x)), (B.19)
2d Pn

[u,U ] = (ii, U ) =
Ld U ( X ) U ( X )dx

So, (B.lO)-(B.14) enable us to express U satisfying (B.8) and (B.9) at any

point of space, once the limit values of U and d U/dn are known on both sides
of 9 These equations are generally considered as the fundamental equations
of the integral theory. Nevertheless, we believe that it is very important to
supply an answer to a second problem, which will cast light on the integral
approach. Let us imagine that the functions ~ ( x ’ and ) z ( x ’ ) of (B.lO) are not
derived from (B. 13) and (B.14), i.e. from an actual function U,but are pseudo-
periodic functions arbitrarily chosen. Under these circumstances, what is the
function U ( x , y ) obtained after integration of (B.10)? It is not difficult to show
that such a function satisfies (B.8) and (B.9) and thus the unknown limit values
of U and d U/dn on both sides of 9 satisfy (B. 13) and (B. 14). Unfortunately,
these two equations cannot provide the four limit values U , U - , d U /dn, + +

d U - /dn, and two supplementary equations must be found for this purpose. To
this end, it can be noticed that, for given y , the right-hand member of (B. 10)
is the product of a continuous function and a Fourier series of x. Hence, the
value U ( x , f ( x ) ) of the right-hand member of (B. 10) on 9 is the mean value
of the two limits on both sides:

U+(X)+ U-(x) = 2U(X, f ( x ) ) . (B.20)

Eliminating U + or U - from (B.14) and (B.20) and replacing U(x,f ( x ) ) by its

expression yield:

G ( x , x ’ ) ~ ( x ’dx’
) + ” ( x , x ’ ) z ( x ’ ) dx’,(B.21)
2

with (B.22)

” ( x , x ‘ ) = a ( x ,f ( x ) ,x’). (B.23)
64 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS [I

For simplicity, (B.21) will be written in a symbolic operator form:

U* = 5: + G { q } + N’{T}. (B.24)
2
The same considerations about discontinuities of Fourier series, applied to
grad U,allow us to write, in the particular case where ~ ( x =) 0:

dU+ q
[ 1 + f’(X’)2]’’* -- - ?-+N{?}, (B.25)
dn 2
with

Equations (B.21) and (B.25), which express U * and d U + / d n as functions of

q and T, appear as being reciprocal of (B. 13) and (B. 14). They can be considered
as the two fundamental equations of the integral formalism.

References

ABRAMOWITZ, M. and A. STEGUN,1970, Handbook of Mathematical functions (Dover Publ.,

New York).
AKAIKE, H., 1973, SIAM J. Appl. Math. 24, 234.
ANDREWARTHA, J. R., G. H. DERRICK and R.C. MC PHEDRAN, 1981a, Opt. Acta 28, 1177.
ANDREWARTHA, J. R., G. H. DERRICKand R. C. Mc PHEDRAN, 1981b, Opt. Acta 28, 1501.
ANDREWARTHA, J. R., J. R. FOXand I. J. WILSON,1979a, Opt. Acta 26,69.
ANDREWARTHA, J. R.,J. R. Fox and I. J. WILSON,1979b, Opt. Acta 26, 197.
BATES,R. H., 1975, IEEE Trans. Microwave Theory & Tech. 23, 605.
BECKMANN, P., 1967, in: Progress in Optics, ed. E. Wolf (North-Holland, Amsterdam) vol. VI,
p. 53.
BECKMANN, P. and A. SPIZZICHINO, 1963, The Scatteringof Electromagnetic Waves from Rough
Surfaces (Pergamon, Macmillan, London, New York).
BORN,M. and E. WOLF, 1965, Principles of Optics (Pergamon Press, Oxford) p. 401.
BOTTEN,L. C., 1978a, Opt. Acta 25, 481.
BOTTEN,L. C., 1978b, Ph. D. Thesis (Univ. of Tasmania, Hobart).
BOTTEN,L. C., 1980, J. Opt. 11, 161.
BOTTEN, L. C., M.S. CRAIG,R. C. MC PHEDRAN, J. L. ADAMSand J. R.ANDREWARTHA, 1981a,
Opt. Acta 28, 413.
BOTTEN,L. C., M.S. CRAIG,R.C. MC PHEDRAN, J. L. ADAMSand J. R. ANDREWARTHA, 1981b,
Opt. Acta 28, 1087.
BOUSQUET, P., 1963a, C. R. Acad. Sci. Paris 256, 3422.
BOUSQUET, P., 1963b, C. R. Acad. Sci. Paris 257, 80.
BREIDNE, M. and D. MAYSTRE.1982, J. Opt. SOC.Amer. 72,499.
11 REFERENCES 65

CADILHAC, M., 1980, in: Electromagnetic Theory of Gratings, ed. R. Petit (Springer, Berlin) p.
53.
CERUTTI-MAORI, G., R. PETITand M. CADILHAC, 1969, C. R. Acad. Sci. Paris B8, 1060.
CHANDEZON, J., M. T. DUPUIS,G. CORNETand D. MAYSTRE,1982, J. Opt. SOC.Amer. 72,839.
CHANDEZON, J., D. MAYSTREand G. RAOULT,1980, J. Opt. 11, 235.
CHANG,K. C., V. SHAHand T. TAMIR,1980, J. Opt. SOC.Amer. 70, 804.
COURANT, R. and D. HILBERT,1962, Methods of Mathematical Physics, vol. 2 (Interscience,
New York).
DELEUIL, R., 1963, C. R. Acad. Sci. Paris 258, 506.
DENNERY, P. and A. KRZYWICKI, 1967, Mathematics for Physicists (Harper & Row, New York,
Evanston, London, and John Weatherhill, Tokyo) p. 87.
DERIUGIN, L. N., 1952, Dokl. Akad. Nauk. SSSR 87(6), 913.
DE SANTO,J. A,, 1975, J. Acoust. SOC.Amer. 57, 1195.
DE SANTO,J. A., 1981, Radio Sci. 16, 1315.
DUMERY, G. and P. FILIPPI,1970, C. R. Acad. Sci. Paris 270, 137.
FACQ,P., 1976, Ann. Telecommun. 31, 99.
FACQ,P., 1977, Thesis (Univ. of Limoges, France).
FANO,U., 1941, J. Opt. SOC.Amer. 31, 213.
Fox, J. R., 1980, Opt. Acta 27, 289.
FRAUNHOFER, J., 1821, Denkschr. Akad. Wiss. Miinchen 8, 1.
GARCIA,N. and N. CABRERA, 1978, Phys. Rev. B18, 576.
GREEN,R., 1970, IEEE Trans. Microwave Theory & Tech. 18, 313.
HESSEL,A., J. SCHMOYS and D. Y. TSENG,1975, J. Opt. SOC.Amer. 65, 380.
HILL,N. R. and V. CELLI,1978, Phys. Rev. B17, 2478.
HUGONIN, J. P., R. PETITand M. CADILHAC,1981, J. Opt. SOC.Amer. 71, 593.
HUTLEY,M. C. and V. M. BIRD, 1973, Opt. Acta 20, 771.
IKUNO,H. and K. YASUURA,1973, IEEE Trans. Antennas & Propag. 2, 657.
ITOH, T. and R. MITTRA, 1969, IEEE Trans. Microwave Theory & Tech. 17, 319.
IYANAGA, S. and Y. KAWADA,1980, Encyclopedic Dictionary of Mathematics (MIT Press,
Cambridge, MA) p. 344E.
JANOT,C. and A. HADNI,1963, J. Phys. 24, 1073.
JIRACEK, G. R., 1973, IEEE Trans. Antennas & Propag. 21, 393.
JORDAN, A. K. and R. H. LANG,1979, Radio Sci. 14, 1077.
JOVICEVIC, S. and S. SESNIC,1972. J. Opt. SOC.Amer. 65 865.
KALHOR,H., 1976, IEEE Trans. Antennas & Propag. 24, 884.
KALHOR,H. and A. NEUREUTHER, 1971, J. Opt. SOC.Amer. 61,43.
KNOP,K., 1978, J. Opt. SOC.Amer. 68, 1206.
LABEYRIE, A. and J. FLAMAND, 1969, Opt. Commun. I, 9.
LANG,R.H. and A.K. JORDAN,1980, NRL Memorandum Rept. 4188, March 17, Naval
Research Laboratory, Washington, DC 20375.
LIPPMANN, B. A., 1953, J. Opt. SOC.Amer. 43, 408.
LOEWEN, E. G., D. MAYSTRE, R.C. M c PHEDRAN and I. J. WILSON,1974, Jpn. J. Appl. Phys.,
suppl. 14-1, 143.
MADDEN,R.P. and J. STRONG,1958, Concepts of Classical Optics (Freeman & Co,
San Francisco).
MARCUSE, D., 1976, Bell Syst. Tech. J. 55, 1295.
MARECHAL, A. and G. W. STROKE,1959, C. R. Acad. Sci. 249,2042.
MAUE,A. W., 1949, Z. Phys. 126, 601.
MAYSTRE, D., 1972, Opt. Commun. 6, 50.
MAYSTRE, D., 1973, Opt. Commun. 8, 216.
66 RIGOROUS VECTOR THEORIES OF DIFFRACTION GRATINGS

MAYSTRE, D., 1974, Thesis (Univ. of Aix-Marseille 3, France).

MAYSTRE, D., 1978a, J. Opt. SOC.Amer. 68, 490.
MAYSTRE, D., 1978b. Opt. Commun. 26, 127.
MAYSTRE, D., 1980a, in: Electromagnetic Theory ofGratings, ed. R. Petit (Springer, Berlin) p. 63.
MAYSTRE, D., 1980b, Proc. Int. U.R.S.I. Symp. 1980 (Munich, August 26-29) p. 144 B/1.
MAYSTRE, D. and R. C. MCPHEDRAN, 1974, Opt. Commun. 12, 164.
MAYSTRE, D., M. NEVIEREand R. PETIT,1980, in: Electromagnetic Theory of Gratings, ed. R.
Petit (Springer, Berlin) p. 159.
MAYSTRE, D. and R. PETIT,1970a, C. R. Acad. Sci. Paris 271, 400.
MAYSTRE, D. and R. PETIT,1970b, Opt. Commun. 2, 309.
MAYSTRE, D. and R. PETIT,1972a, Opt. Commun. 5, 35.
MAYSTRE, D. and R. PETIT,1972b, Opt. Commun. 5, 90.
MAYSTRE, D. and R. PETIT, 1974, J. Spectr. SOC. Jpn, 23 Suppl., 61.
Mc CLELLAN, R. C. and G. W. STROKE,1966, J. Math. Phys. 45, 383.
M c PHEDRAN, R. C., 1973, Ph. D. Dissertation (Univ. of Tasmania, Hobart).
MC PHEDRAN, R. C., G. H. DERRICKand L. C. BOTTEN, 1980, in: Electromagnetic Theory of
Gratings, ed. R. Petit (Springer, Berlin) p. 227.
MEECHAM, W. C., 1956, J. Appl. Phys. 27, 361.
MEIXNER, J., 1972, IEEE Trans. Antennas & Propag. 20,442.
MILLAR, R. F., 1969, Proc. Camb. Phil. SOC.65, 773.
MILLAR, R. F., 1971, Proc. Camb. Phil. SOC.69, 217.
MILLAR, R. F., 1973, Radio Sci. 8, 785.
MIRANDA, c., 1970, Partial Differential Equations (Springer, Berlin).
MOAVENI, M. K., H. A. KALHORand A. AFRASHTEH, 1975, Comput. & Electr. Eng. 2, 265.
MORSE,P. M. and H. FESHBACH, 1953, Methods ofTheoretical Physics, Part I (Mc Craw-Hill,
New York) p. 804.
NAMIOKA, T., T. HARADA and K. YASUURA, 1979, Opt. Acta 26, 1009.
NEUREUTHER, A. and K. ZAKI,1969, Aka Freq. 38, 282.
NEVIERE, M., 1975, Thesis (Univ. of Aix Marseille 3, France).
NEVIERE, M. and M. CADILHAC, 1970, Opt. Commun. 2,235.
NEVIERE, M., M. CADILHAC and R. PETIT, 1973, IEEE Trans. Antennas & Propag. 21, 37.
NEVIERE, M., G. CERUTTI-MAORI and M. CADILHAC, 1971, Opt. Commun. 3, 48.
NEVIERE, M. and P. VINCENT, 1976, Opt. Acta 23, 557.
NEVIERE, M., P. VINCENT and R. PETIT, 1974, Nouv. Rev. Opt. 5, 65.
PAVAGEAU, J. and J. BOUSQUET, 1970, Opt. Acta 17, 469.
PAVAGEAU, J., R. EIDOand H. KOBEISSE,1967, C. R. Acad. Sci. Paris 264, 424.
PETIT,R., 1963, Rev. Opt. 42, 263.
PETIT,R., 1965, C. R. Acad. Sci. Paris 260, 4454.
PETIT,R., 1966a, Rev. Opt. 45, 249.
PETIT, R., 1966b, Rev. Opt. 8, 353.
PETIT, R., 1967, Opt. Acta 14, 301.
PETIT,R., 1980, in: Electromagnetic Theory of Gratings, ed. R. Petit (Springer, Berlin) p. 1.
PETIT,R. and M. CADILHAC, 1964, c. R. Acad. sci. Paris 259, 2077.
PETIT,R. and M. CADILHAC, 1966, C. R. Acad. Sci. AB 262, 468.
PETIT,R. and D. MAYSTRE,1972, Rev. Phys. Appl. 7, 427.
PETIT, R., D. MAYSTRE and M. NEVIERE,1972, Communication to the Ninth International
Congress of Optics, Santa-Monica, CA (see Space Opt., 1974, Proc. of the ninth ICO, p. 667).
RAYLEIGH, Lord, 1907, Proc. Roy. SOC.A79, 399.
RAYLEIGH, Lord, 1945, The Theory of Sound, vol. 2 (Dover, New York) p. 89.
RITTENHOUSE,D.. 1786, Trans. Amer. Phil. SOC.2, 201.
I1 REFERENCES 67

ROGER,A. and M. BREIDNE,1980, Opt. Commun. 35, 299.

ROGER,A. and D. MAYSTRE,1981, J. Opt. SOC.Amer. 70, 1483.
ROUMIGUIERES, J. L., D. MAYSTRE and R. PETIT,1974, Proc. Fifth Colloquium on Microwave
Communications, Budapest, ET 305.
ROUMIGUIERES, J. L., D. MAYSTRE and R. PETIT, 1975, Opt. Commun. 7,402.
ROWLAND, H. A., 1882, Phil. Mag. 13,469.
RUDOLPH,D. and G. SCHMAHL,1967, Umschau Wiss. Tech. 67, 225.
SCHWARTZ, L., 1966a, Mathematics for the Physical Sciences (Addison-Wesley, Reading, MA).
SCHWARTZ, L., 1966b, ThCorie des Distributions (Hermann, Paris).
SOMMERFELD, A,, 1949, Partial Differential Equations (Academic, New York).
STROKE,G. W., 1960, Rev. Opt. 39, 350.
URETSKY,J. L., 1965, Ann. Phys. 33, 400.
VAN BLADEL,J., 1964, Electromagnetic Fields (Mc Graw-Hill, New York).
VAN DEN BERG, P. M., 1971a, Appl. Sci. Res. 24, 261.
VAN DEN BERG,P. M., 1971b. Thesis (Delft, The Netherlands).
VAN DEN BERG,P. M., 1980, Proc. Int. URSI Symp. 1980 (Munich, August 26-29) p. 21 1 B/1.
VAN DEN BERG,P. M., 1981, J. Opt. SOC.Amer. 71, 1224.
VAN DEN BERG,P. M. and J. T. FOKKEMA, 1979, J. Opt. SOC.Amer. 69, 27.
VEKUA, 1. N., 1953, Dokl. Akad. Nauk. SSSR 90, 715.
VINCENT, P., 1980a, SPIE Proc., 240.
VINCENT,P., 1980b,in:ElectromagneticTheoryofGratings,ed. R. Petit (Springer, Berlin)p. 101.
VINCENT, P., M. NEVIEREand D. MAYSTRE,1978, Nucl. Instrum. and Meth. 152, 123.
WATERMAN, P. C., 1975, J. Acoust. SOC.Amer. 57, 791.
WHITMAN, G., D. M. LESKIVand F. SCHWERING, 1980, J. Opt. Soc. Amer. 70, 1495.
WHITMAN, G. and F. SCHWERING, 1977, IEEE Trans. Antennas & Propag. 25, 869.
WILCOX,C. H., 1975, Scattering Theory for the d'Alembert Equation in Exterior Domains
(Springer, Berlin).
WIRGIN,A., 1964, Rev. Opt. 9, 449.
WIRGIN,A., 1967, Thesis (Univ. of Paris, France).
WIRGIN,A., 1968, Rev. Cethedec 5, 131.
WIRGIN,A., 1969, Aka Frequenza, Special-Issue from URSI Symposium 38, 327.
WIRGIN,A., 1978, Opt. Commun., 27, 189.
WIRGIN,A., 1979a, C. R. Acad. Sci. Paris, A289, 259.
WIRGIN,A., 1979b, C. R. Acad. Sci. Paris, B289, 273.
WIRGIN, A., 1979~2,C. R. Acad. Sci. Paris, B288, 179.
WIRGIN,A., 1980a, J. Acoust. Soc. Amer. 68, 692.
WIRGIN, A., 1980b, Opt. Acta 27, 1671.
WIRGIN,A., 1981, Opt. Acta 28, 1377.
WIRGIN, A. and R. DELEUIL,1969, J. Opt. SOC.Amer. 59, 1348.
YAKOVLEV,E. A., 1965, Opt. Spektrosk. 19, 417.
YASUURA, K. and H. IKUNO,1971, Proc. Int. Symp. on Antennas and Propagation, Sendai,
Japan, 173.