5

Fourier Series and

Fourier Transforms

In Chapter 3, we touched upon the analogy between the diffraction of x-rays and
that of visible light. Here, we extend that discussion and consider some aspects of
Fourier series and Fourier transforms.

5.1 Image Formation and Focusing

The formation of the optical image of an object involves first, a scattering

of light from it, and then a recombination of the scattered light rays. We could,
for example, prepare a transparency of one half of Figure 1.1 and project it on to
a screen, focusing correctly, thus revealing an enlarged image of the model of the
unit cell of sodium chloride.
If we now take the lens out of the projector, there will be just a patch of light
on the screen, even though the object, the transparency, is still in the same position.
All the information provided by the transparency is still present in the patch of
light, but it is not immediately decipherable. The lens has no information about the
transparency, but once it is in position in the projector the image becomes clear.
The lens rearranges the scattered light so as to be understandable to us.
The process of focusing is complex, but we perform it by adjusting the
position of the lens until the image assumes our expectation of the object. Evidently,
some foreknowledge of the object is needed, and we assume that the appearance of
sharply defined boundaries in the image is a condition of being in focus. If we do
not have the necessary foreknowledge, we must determine the relative positions of
transparency, lens and screen by the methods of geometrical optics. It is possible to
show that the scattered radiation is everywhere within the scattered light by moving
through it with a hand lens and a piece of white card, although the resulting image
will not be as satisfactory as that obtained with the projector lens system. If we
cannot obtain sharp boundaries in the image, we need an aid to focusing. One
simple practical method, sometime used in microscopy, would be to have a minute
speck of dust adhering to the surface of the transparency. When the speck is in
focus, it can be assumed that the whole image is also in focus.

289
M. Ladd et.al., Structure Determination by X-ray Crystallography
© Kluwer Academic/Plenum Publishers, New York 2003
290 5 FOURIER SERIES AND TRANSFORMS

Visible light can resolve separations in an object down to approximately

2500 A, if a high-quality microscope is employed. Resolution and radiation wave-
length go hand in hand. The human eye can observe two objects as separate entities
provided that they are no less than approximately 0.15 mm apart. Rayleigh's
formula for the limit of resolution, or resolving power, R of a microscope is
given as

R =0.6lA.j(nsina) =0.6lA.jN (5.1)

where 2a is the angle of scatter, n is the refractive index of the medium and N is
the numerical aperture of the objective. For a microscope working dry N can reach
0.95, and in oil immersion up to 1.45. Hence, R is approximately 3850A (dry) or
2520A (oil), corresponding to magnifications of 400 and 600, respectively. More
recently, higher resolution has been demonstrated by scanning near-field optical
microscopy (SNOM),a and resolution ranges between O.D. and O.OD. have been
reported.
The refractive index of materials for x-rays is approximately unity, so that
from (5.1) wavelengths in the range 110.61, or approximately 1.6 A, are required
in order to resolve an atomic separations of 1 A; such wavelengths are obtained
with x-rays or neutrons. X-rays and neutrons of wavelength 1-2 A can be used to
resolve atomic detail, but they can be focused only by special systems of curved
mirrors that lead to impracticably low magnification. With x-rays or neutrons, the
scattered radiation must be recombined by calculation.
The electron microscope can provide resolution of atomic detail, provided
that the structure is not too complex, but the resolution is limited by spherical
aberration of the focusing system.
Certain simplifications exist in the applications of these radiations com-
pared to visible light. The regular packing of atoms and molecules in crystals and
a restriction of the radiation to a monochromatic source together give rise to a spot
diffraction pattern rather than a diffuse patch.
A close analogy to x-ray diffraction, albeit in two dimensions, can be seen
by viewing a sodium street-lamp through a fine, stretched gauze, such as a net
curtain or a handkerchief. The spot pattern that is obtained is invariant under trans-
lation of the object, but rotates as the object is rotated; we shall return to these
two properties later. To form an image from the diffraction pattern, the scattered
radiation must be recombined in both amplitude and phase. The lens system used
with visible light enables this process to be carried out directly. With x-rays or
neutrons, however, not only can the focusing not be done directly, but the impor-
tant phase information required is not obtained explicitly from the experimental

a V. Sandoghdar, in Proceedings of the International School of Physics, Enrico fermi, lOS Press,
Amsterdam (2001).
5.2 FOURIER SERIES 291

procedure: we record I(hkl), but from Figure 3.28, we see that IF(hkl)1 2 =
A'(hkl)2 + B'(hkl)2 = IF(hkl)1 2 cos 2 ¢(hkl) + IF(hkl)1 2 sin 2 ¢(hkl) = I(hkl),
so that information about ¢(hkl) is not given directly.
In crystal structure analysis, we use Fourier series to carry out the focusing
process. In the next section, we shall assume that we have the necessary phase
information; and the acquisition of this phase information will be the subject of
much of the remainder of this book.

5.2 Fourier Series

We consider the function 1/1 (X) in Figure 5.1: it is continuous, single-valued,

and periodic in the repeat distance a. According to Fourier's theorem, it can be
represented by a series of cosine and sine terms that may be written conveniently as

00

1/I(X) = L {C(h) cos(2rrhX/a) + S(h) sin(2rrhX/a)}

h=-oo

which can also be expressed as

00

1/I(X) = C(O) + 2 L{C(h) cos(2rrhX/a) + S(h) sin(2rrhX/a)} (5.2)

h=!

The index representing the hth term in this series is a frequency, or wavenumber,
that is, the number of times its own wavelength fits into the repeat period. In order
to find the coefficients C(h) and S(h), we form the integral

I= faa 1/I(X)cos(2rrHX/a)dx (5.3)

for a general value H of the frequency variable h. Substituting (5.2) into (5.3), and
noting that the integral of a sum is equal to the sum of the integrals of the separate

of (xi aXi,'"
30
20
10
o
x aXIs

FIGURE 5.1. One-dimensional periodic function 1/f(X) of repeat a.

292 5 FOURIER SERIES AND TRANSFORMS

parts, we have

I = C(O) faa cos(2lT HX/a) dx

+ E {C(h) faa 2[cos(2lThX/a) cos(2lT HX/a)] dx

+S(h) faa 2[sin(2lThX/a) cos(2lT HX/a)] dX} (5.4)

Equation (5.4) contains a number of integrals in x of the forms

faa 2cos(2lThX/a) cos(2lT HX/a) dx (5.5)

and
faa 2 sin(2lThX /a) cos(2lT H X/a) dx (5.6)

It is straightforward to show, using identities given in Appendix A6, that these

integrals are orthogonal, that is, (5.5) = 0 except for H = h, when it is equal to
a, whereas (5.6) = 0 for all values of H. Thus, C(h) and S(h) may be expressed
by the equations

C(h) = -1 loa 1/1 (X) cos(2lThX/a) dx (5.7)

a 0

and
S(h) = -1 loa 1/1 (X) sin(2lThX/a) dx (5.8)
a 0
If the fonn of the function 1/I(X) is known, C(h) and S(h) can be evaluated. We
shall carry out this process for the square wave shown in Figure 5.2.

5.2.1 Analysis of the Square Wave

Let the square-wave function 1/1 (X) be defined in the range -IT ::::; X ::::; IT,
with a repeat of 2lT. For X < 0, 1/I(X) = 0, and for 0 < X ::::; IT, 1/I(X) = IT (we
shall see from the analysis that 1/I(X) = IT/2 at X = 0). Hence,

C(h) = -1 Ion: IT cos(hX) dx

2lT 0

Integration gives the result that for h i= 0, C(h) = 0, whereas for h = 0,

C (h) = ~ 1(; dx = IT /2. In a similar manner, we find that for h = 0, S (h) = 0,
5.2 FOURIER SERIES 293

~(x) axis

n+---------------~

-n o Tt
---
x axis

FIG URE 5.2. Square wave 1jJ(X), defined in the range -Jf to Jf and with a repeat of 2Jf.

but for h "# 0, S(h) = (l/2h)[1 - cos(hn)], Substituting these results in (S.2),
we find
00

1/I(X) = n 12 + 2 ~)'1/2h)[1 - cos(nh)l sin(hX)

h=]

or

L
00

1/I(X) = nl2 + 2 (Ilh) sin(hX) (S.9)

h~1
(h~211+11

since the term [1 .- cos(nh)] is zero for even values of h. Hence, finite values of
1/I(X) arise for h := 2n+ 1 (n = 0, 1,2, ... ).
Range of X
In (S.9), the variable X defines a sampling point in any repeat interval
between X = ±n. For convenience, we will choose a zero arbitrarily at X = and
sample the function at intervals of 2n ISO, that is, we shall calculate the function
°
°
from to SO(2n 150). The results will show that we could, and in general would,
make use of reflection symmetry in the function at X = (2m + l)n 12 (m =
0, I, 2, ... , 00) to decrease the amount of calculation.

Range of h
The summations of a Fourier series extend, theoretically, from -00 to 00.
In practice, however, the range becomes hmin :::: h :::: h max , where the limits of h
are preset, normally by experimental conditions. In this example hmin is unity, and
h max values of 3 and 7 are used in the results presented in Table S.1 and Figure S.3.
It is notable, even with these few numbers of terms, that increasing h max has a
dramatic effect on the series. As h max increases so the series (5.9) approaches more
294 5 FOURIER SERIES AND TRANSFORMS

TABLES.I. Values of the function (nI2) + 2 L~:"l(llh) sin(hX)

(h odd) for h max = 3 and 7. The true value 1/!(X), for all h, appears in the
extreme right -hand column

1f-r(X) ::0: ~n
1f-r(X)::o: ~n + 2sinX +2sinX + ~ sin3X
+ ~ sin3X +~ sin5X + .~ sin 7X
X/2n h max = 3 h max = 7 1f-r(X)

0/50 1.571 1.571 3.142

1/50 2.067 2.522 3.142
2/50 2.525 3.186 3.142
3/50 2.910 3.428 3.142
4/50 3.200 3.330 3.142
5/50 3.380 3.109 3.142
6/50 3.454 2.977 3.142
7/50 3.433 3.017 3.142
8/50 3.343 3.158 3.142
9/50 3.215 3.265 3.142
10/50 3.081 3.249 3.142
11/50 2.972 3.137 3.142
12/50 2.912 3.034 3.142
m- -m
13/50 2.912 3.034 3.142
14/50 2.972 3.\37 3.142
15/50 3.081 3.249 3.142
16/50 3.215 3.26S 3.142
17/50 3.343 3.1S8 3.142
18/S0 3.433 3.017 3.142
19/50 3.4S4 2.977 3.142
20/50 3.380 3.109 3.142
21/S0 3.200 3.330 3.142
22/50 2.910 3.428 3.142
23/S0 2.S25 3.186 3.142
24/50 2.067 2.522 3.142
25/S0 1.571 1.571 3.142
26/50 1.075 0.619 0
27/50 0.6\7 -0.044 0
28/50 0.231 -0.287 0
29/50 -0.058 -0.188 0
30/S0 -0.239 0.033 0
31/S0 -0.312 0.164 0
32/S0 -0.291 0.125 0
33/S0 -0.201 -0.017 0
34/50 -0.073 -0.123 0
35/50 0.061 -0.107 0
36/50 0.169 0.005 0
37/50 0.230 0.108 0
m- -m
38/50 0.230 0.108 0
39/50 0.169 0.005 0
5.2 FOURIER SERIES 295

TABLE 5.1. (Continued)

1j!(X) :::: ~n
V!(X):::: ~n+ 2sinX +2sinX + sin3X i
+~sin3X +~ sin5X + .~ sin 7X
X/2n h max =3 h max = 7 1/I(X)

40/50 0.061 -0.107 0

41150 -0.073 -0.123 0
42/50 -0.201 -0.017 0
43/50 -0.291 0.125 0
44/50 -0.312 0.164 0
45/50 -0.239 0.033 0
46/50 -0.058 -0.188 0
47/50 0.231 -0.287 0
48/50 0.617 -0.044 0
49/50 1.075 0.619 0
50/50 1.571 1.571 3.142

a The corresponding curves are shown in Figure 5.3.

(a) ljJ(x) axis

.... ....
"

x axis
-~ ..... . '. o "
'.

(b) ljJ(x) axis

"

. x axis

'.' 0
"
-7t . . '

FIGURE 5.3. Square waves calculated from (5.9): (a) h max = 3, (b) h max = 7. The positive and
negative fluctuations of the calculated function arise because there are insufficient terms to provide
good convergence of the Fourier series; they are known as series termination errors.
296 5 FOURIER SERIES AND TRANSFORMS

closely the square-wave function in Figure 5.2. In general, the more independent
terms that can be included in a Fourier series, the better it represents the periodic
function under investigation, from which the terms have been derived.
The process of determining the coefficients of a Fourier series is called
Fourier analysis, and the process of reconstructing the function by the summation
of a series such as (5.9) is Fourier synthesis. A microscope, in forming an image of
an object, effectively performs a Fourier synthesis of the scattered light, a process
that we shall have occasion to refer to again later.

5.2.2 Exponential Forms of Fourier Series

Let a function G(h) be represented as

G(h) = C(h) + is(h) (5.10)

Using de Moivre's theorem, exp(iX) = cos(X) + i sin(X), it follows that

G(h) exp( -i2nhX/a) + G( -h) exp(i2nhX/a)

= [C(h) + is(h)][cos(2nhX/a) - i sin(2nhX/a)]
+ [C(h) - is(h)][cos(2nhX/a) + i sin(2nhX/a)]
= 2C(h) cos(2nhX/a) + 2S(h) sin(2nhX/a) (5.11)

Comparison with (5.2) shows that 1jJ(X) may be cast as

1
1jJ(X) = "2 L G(h) exp( -i2nhX/a) + G( -h) exp(i2nhX/a)
CXl

h=O

which reduces to

L
CXl

1jJ(X) = G(h)exp(-i2nhX/a) (5.12)

h=-CXl

The multipliers 1/2 in (5.12a) disappear in (5.12) because the summation limits
are now -00 to 00, whereas in (5.11) the positive and negative values of h appear
explicitly; we shall consider this point again later. Following Section 5.2, we can
show
G(h) = -1 Ina 1jJ(X) exp(i2nhX/a) dx (5.13)
a 0
Equations (5.12) and (5.13) are Fourier transforms of each other, a topic that we
shall consider in more detail shortly; the signs of the exponents should be noted.
5.3 FOURIER SERIES IN X-RAY CRYSTALLOGRAPHY 297

5.3 Fourier Series in X-ray Crystallography

The lattice basis of crystal structures introduces a three-dimensional peri-

odicity which pervades the properties of crystals, including the electron density
distribution. The square wave that we have just analyzed may be likened to a one-
dimensional crystal, or the projection of a crystal structure on to a single axis. The
first applications of Fourier series in crystallography were with one-dimensional
series, and for good reason. We must consider first the significance of the functions
1/f(X) and G(h).
We have shown in Chapter 3 how x-rays are scattered by the electrons
associated with atoms in a crystal. The concentration of electrons and its distribu-
tion around an atom is called the electron density, and it is measured in electrons
per unit volume (usually A-3 or nm- 3 ). At any point X, Y, Z the electron density
is written normally as p(XYZ), and we may identify p(XYZ) with the function
1/f (X) in (5.12), so that we must now determine the meaning of G(h).
We consider electrons in an atom as though they were concentrated at a point,
but specify their distribution by a shape factor, the atomic scattering factor f, which
is equivalent to an amplitude. The exponential term in (5.12) represents the phase
of the wave scattered by an atom at X with respect to the origin of the unit cell.

5.3.1 One-Dimensional Function

The one-dimensional electron density function p (X) (e A-I) shown in
Figure 5.4 was calculated from a small number of experimental terms. In a small
interval dX along the X axis, the electron density may be regarded as being con-
stant, so that the associated electron count is p(X) dX. Its contribution to the hth

pIx) axis (e .l-',

120

100

80

60

40
pix)
20

o 40 50 60 80 90 100 -
x axis(l 00 ths of B'
FIGURE 5.4. One-dimensional electron density projection p(X) for pyrite. FeS2.
298 5 FOURIER SERIES AND TRANSFORMS

structure factor F(h) is given, following Section 3.S.3, by p(X) exp(i21l'hX) dX,
where exp(i21l'hX) is the phase associated with p(X) dX with respect to the ori-
gin. The contribution from the whole repeat period, the structure factor F(h), is
now given by
F(h) = foa p(X) exp(i21l'hX/a) dX (S.14)

where we choose the integration limits 0 to a instead of -a /2 to a /2.

Equation (S.14) is a generalized one-dimensional structure factor; analogous
expressions exist in two and three dimensions. Substituting for p(X), equivalent
to 1/r(X), from (S.12), we have

F(h) = 10
r L 00
G(h') exp( -i21l'h'X/a) exp(i21l'hX/a) dX (S.1S)
h'=-oo

where h' indicates the range of values of h under the summation sign. Since the
integral of a sum is equal to the sum of the integrals of the separate terms, we write

F(h) = L00
G(h') 10
r exp[i21l'(h - h')X/a] dX (S.16)
h'=-oo 0

The integral evaluates to exp[i21l'(h - h')X/a]/(i21l'(h - h')/a). Since both hand

h' are integers, the numerator of this expression is zero, except when h' = h. In this
special case, we can see from (S.16) that the integral becomes J;
dx which has the
value a. It follows immediately that G(h) = F(h)/a, so that from (S.12) we write

p(X) = -
a
1
L
00
F(h)exp(-i21l'hX/a) (S.17)
h=-oo

which is the Fourier transform of (S.14).

From Figure 3.28, we see that we can write

F(h) = A(h) + iB(h)

(S.18)
F(-h) = A(h) - iB(h)

so that (S.17) may be written as

piX) = II F(O) + ~[A(h) + iB(h)][cos(21l'hX/a) - i sin(21l'hX/a)]

+ ~[A(h) - iB(h )][ca,(2n-hX fa) + i ,inC -2n-hXfa )]\ (5.19)

5.3 FOURIER SERIES IN X-RAY CRYSTALLOGRAPHY 299

I
which reduces to

p(X) =~ F(O)+ 2 E[A(h) cos(2"hXfa)+ B(h) sin(2"hX fa)l\ (5.20)

which may be compared with (5.2).

5.3.2 Two- and Three-Dimensional Functions

Analogous expressions can be formulated for two and three dimensions; we
will consider the two-dimensional case in detail as the program system XRAY*
(see Chapter 11) uses this form of the Fourier series. First, we state, by analogy
with (5.17) and using fractional coordinates, the three-dimensional electron density
equation as
00 00 00

p(xyz)=~
Vc
L L L F(hkl)exp[-i27T(hx+ky+lz)] (5.21)
h=-oo k=-oo 1=-00

Since, from Figure 3.28, A(hkl) = IF(hkl)1 cos¢(hkl) and B(hkl) = IF(hkl)1
sin ¢(hkl), it follows from the form of (5.20) that p(xyz) may be written as
1
L L L
00 00 00

p(xyz) =- IF(hkl) ICOS[27T(hx + ky + lz) - ¢(hkl)]

Vc h=o-oo k=-oo 1=-00
(5.22)
which serves to show how the electron density depends upon the phase angles: only
IF(hkl)1 is measured by experiment, ¢(hkl) must be determined before (5.22)
can be summed. This situation constitutes the phase problem in crystallography,
of which we shall hear more.
Equation (5.21) can be rewritten in the form of (5.20), which is most
convenient for calculation by again making use of Friedel's law, namely that
IF(hkl) I = IF(hkl)l, or A(hkl) = A(hkl) and B(hkl) = -B(hkl). Thus, we
obtain

p(xyz) = -
1{ F(OOO) +2L
00
L L
00 00
A(hkl) COS[27T(hx + ky + lz)]
Vc h=] k=-oo 1=-00

+ B(hkl) sin[27T(hx + ky + lz)] } (5.23)

Again, this equation could be put in the form of (5.22) if desired, and variations
can be developed according to space-group symmetry, as discussed in Chapter 3
for structure factors.
We consider the two-dimensional equation. The generalized structure factor
F(hkl) is given by

F(hkl) = Vc 10 1 10 1 10 1 p(xyz) exp[i27T(hx + ky + lz)] dx dy dz (5.24)

300 5 FOURIER SERIES AND TRANSFORMS

we are using the limits 0 to 1 because we are now using fractional coordinates x,
y, and z. For a projection along the z axis, we need the F(hkO) reflections, where

F(hkO) = Vc fa 1 fa 1 { fa 1 p(xyz) dZ} exp[i2n(hx + ky + lz)] dx dy (5.25)

To interpret the integral over z, consider an element of structure of cross-sectional

area dx dy and length c along the z axis. In an element of length dz, the electron
content is p(xyz)cdz, so that the total electron content in the element of length
cis c dx dy fol p(xyz) dz. Hence, the projected electron density at a point x, y is
given by

p(xy) = Cfal p(xyz) dzdx dy

from which F(hkO), or F(hk) becomes

F(hk) = A fa 1 fa 1 p(xy) exp[i2n(hx + ky)] (5.26)

since fd
exp( -i2n /z) dz is zero unless I = O.
where A is the area of the a, b face of the unit cell. It follows that
1
ALL
00 00

p(xy) = F(hk) exp[ -i2n(hx + ky)] (5.27)

h=-oo k=-oo

which may be written more conveniently as

I {.
+2L L
00 00
p(xy) =A F(OOO) [A(hk) cos(2n(hx + ky)
h=1 k=-oo

+ B(hk) sin(2n(hx + kY»]} (5.28)

Unless one is concerned with absolute values of electron density, the F(OOO) term
can be omitted and p(xy) scaled to a convenient maximum value.
We conclude this section by a consideration of the one-dimensional electron
function which we discussed in Section 5.2. We draw Figure 5.1 again as Figure 5.5,
now with the amplitudes and phases of the waves shown. Following the arguments
above, we can write the electron density function in terms of amplitude and phase as

p(x) = ~{20 + 5 cos[2n(x) - 3.023] + 2.5 cos[2n(2x) - 2.400]

+ 4cos[2n(3x) - 4.060]}

This equation would then be evaluated at a suitable interval x, 20th of a would

be appropriate, and the function plotted. Practice with Fourier series follows from
the Problems section.
5A HOLES AND ATOMS 301

p(x) axis 1
'4300 .
I
1
~ A
'10 r
0-- I - - - ' 1 " - - ------¥-------'!''----~ axis J

IF(O)I t= - 2 0 - - 1 - -
I --t-.~-~-+------+-------
Wavelength 00 (1 )=173,2'
Wave number 0 -.-
21F(1 )11- 5
wavelen--=g':-:th-'a~-+-----'f'-'------'~--+------;",c----------'~-+--~---------,~-+--~
Wave number 1
B

Wavelength a/3
Wave number 3

FIG URE 5.5. The periodic function of Figure 5.1. with amplitudes and phases for waves of indices
O. I. 2. and 3: the amplitudes for the indices I. 2. and 3 have been given twice their weight according
to (5.20) for piX). The portion A may be regarded as the electron density piX) with a repeat distance
a of 5 A. The portion B corresponds to the structure factors, or harmonics, with their phases. X-ray
diffraction may he regarded as the path A -+ B. Structure determination is the path B -+ A, and the
need for correct phases is clear. The values for the phases given are based on their own wavelengths.
0, a. 2a. and 3a in order to comply with the definition of phase (Section 3.2.3).

5.4 Holes and Atoms

When visible light is incident upon a circular hole in an otherwise opaque

card, its diffraction pattern is somewhat diffuse. The theory of this scattering is
complex, but it can be simplified by considering the pattern at an ideally infinite
distance-the Fraunhofer diffraction pattern of the hole. Figure 5.6 shows a simple
schematic experimental arrangement for viewing a Fraunhofer diffraction pattern.
Parallel light from a laser source S is incident upon the object hole at 0, which
must be smaller than the diameter of the laser beam. The diffraction pattern can
be viewed or photographed at F, the back focal plane of the lens L. The whole
of the radiation is not completely in phase at F, and its intensity falls off with
the distance from F. The larger the hole, the more rapid is the fall off, since the
addition of waves at a given distance from F is then less complete.
The diffraction pattern of the hole has circular symmetry, and it may be
represented by a radial distribution function of the form

(5.29)
302 ) FOURIER SERIES AND TRANSFORMS

o
s

FIGURE 5.6. Simple experimental arrangement for producing Fraunhofer diffraction: S,

helium-neon laser source; 0, object; L, lens (focal length ca 1m); F, back focal plane of lens
L, 'Where the diffraction pattern may be recorded. [Reproduced from Diffraction, by C. A. Taylor, with
the permission of the Institute of Physics.]

where A (e) is the amplitude of the scattered radiation at an angle e to the plane of
the hole, Jl is a first-order Bessel function, and X is a function that is proportional to
the radius of the hole and to (sin e) IA. Figure 5.7a illustrates the diffraction pattern
of a hole; Figure 5.7b is a plot of (5.29), normalized to A(e) = 6 at (sin e)/A = 0,
together with the atomic scattering factor curve for carbon, atomic number 6.
Whereas A(e) alternates in sign, the atomic scattering factor f does not, but at
low values of (sin e) IA the two curves are closely similar. The diffraction pattern
of the hole is its Fourier transform, and (5.29) is a mathematical representation of
the transform. Similarly, f is the transform of an atom, and the f -curve shows its
variation with (sine)/A.

5.5 Generalized Fourier Transform

We refer back to Figure 3.8, in Chapter 3, and some development thereof.

Let the three-dimensional electron density for the body at the point A be per) with
respect to a single electron at the origin O. An element of volume 8 V around 0
has an electron content of p(r)8V, and its phase with respect to 0 is 2JTr . S.
Hence, the contribution of the quantity p(r)8V to scattering in the direction is e
per) exp[i2JT(r· S)]8V (5.30)

Then, the total scattering for the body is

Iv per) exp[i2JT(r. S)] dV (5.31)

where the integral extends over the volume V of the body. This expression is the
Fourier transform of the body, and may be written

G(S) = IvP(r)eXP[i2JT(r'S)]dV (5.32)

5.5 GENERALIZED FOURIER TRANSFORM 303

(a)

(b)
---X-ray scaltuing "y cuf,on atom
---Light seaHering hycircular hole

Sea tfcred

t
amplitude

A(81

FIGURE 5.7. Scattering from a circular hole. (a) Diffraction pattern for a hole of 5 mm diameter.
[Reproduced from Diffraction, by C. A. Taylor, with the permission of the Institute of Physics.] (b)
The scattered amplitude function A(e) for a given hole radius, normalized to equal the value of f
(carbon) at (sin ellA ,= O. [Reproduced from Optical Transforms, by C. A. Taylor and H. Lipson, with
the permission of Routledge .]

Following Section 3.6.1, we have

/(S) = G(S)G*(S) (5.33)

So that / (S) is the intensity of the transform. For an atom, we may usually assume
spherical symmetry, so that the Fourier transform for an atom is just j(S), where
304 5 FOURIER SERIES AND TRANSFORMS

1 is the atomic scattering factor. Since centro symmetry is a subgroup of spherical

symmetry
I(S) = 1*(S) (5.34)

so that
(5.35)

In practice, 1 (S) is defined to be positive for all values of sin e. Strictly, there
is a phase change of n when x-rays are scattered, as we discussed in Chapter 3,
but 'since it is true for all atoms it is usually ignored, and 1 (S) is given by the
Fourier transform of (5.32), where per) = 4nr21jf(r), 1jf(r) being an appropriate
radial wave function, or a combinations of wave functions, for the atom. Evi-
dently, we evaluated the Fourier transform 1 (S) surreptitiously in Section 3.2.5
and Problem 3.3.

5.5.1 Fourier Transform of a Molecule

Let the atoms of a molecule be characterized by the coordinates X j , Yj, and
Zj (j = 1,2,3, ... , n), each with a scattering factor I(S). The vector rj from the
origin to the jth atom is
(5.36)

From (5.31), the wave scattered by the jth atom at a distance r from it, with respect
to the origin, is given by

(5.37)

or
Iv per) exp[i2n(r. S)] dV exp[i2n(rj . S)] (5.38)

which becomes
fJ(S) exp[i2n(rj . S)] (5.39)

Thus, the total wave from all n discrete atoms in the molecule is its Fourier
transform G(S), given by

n
G(S) = L fJ exp[i2n(rj • S)] (5.40)
j=i

We write fJ for fJ (S), because we have already decided that fJ is spherically

symmetrical, and its variation with e is inherent in f itself.
5.6 PRACTICE WITH TRANSFORMS 305

5.5.2 Fourier Transform of a Unit Cell

Let x j, Yj, Zj now be the fractional coordinates of the jth atom in a unit cell,
so that
(S.41)

From the Bragg equation, (2 sin 8)/'A = lid, and the fact that

S = d* = ha* + kb* + lc* (S.42)

we have
(S.43)

Then, the total transform for the unit cell is

n
G(S) = L fj exp[i21T(hxj +kYj +lzj)] (S.44)
j=]

which is identical to the structure factor equation (3.107) for the reflection hkl,
where hkl is related to S through (S.42). The Fourier transform (S.44) is valid for all
values of h, k, and I. In a crystal, however, because of interference, the transform
can be observed only at those points where scattering is reinforced, that is, at the
reciprocal lattice points hkl, where h, k, and I are necessarily integral. Thus, the
structure factor equation for a crystal is its Fourier transform sampled at the recip-
rocallattice points. We may imagine the Fourier transform for a molecule overlaid
by the reciprocal lattice in the correct orientation: only those points that satisfy the
limiting conditions for the space group could give rise to x-ray reflections.

5.6 Practice with Transforms

We can calculate transforms by (S.44), and we can prepare them experimen-

tally by means of the optical diffractometer. Both techniques can provide useful
results in developing Fourier transform theory.

5.6.1 Optical Diffractometer

The optical diffractometer permits the preparation of diffraction patterns in
a relatively straightforward manner. In Figure S.8, S is a helium-neon laser of
ca SOmW intensity, M] and M2 are mirrors, E is an expander that extends the
laser beam without loss of spatial coherence, L] and L2 are lenses, and 0 is the
object. The diffraction pattern is brought to a focus at F, the back focal plane of
lens L2, where it may be viewed with an eyepiece, photographed or input to a
television camera; Figures S.7a and S.9 were produced with such an instrument.
306 5 FOURIER SERIES AND TRANSFORMS

o
FIGURE 5.8. Schematic arrangement of an optical diffractometer: S, helium-neon laser source; M[
and !rh , mirrors; E, beam expander; 0 , object; LJ and L2, lenses; F, back focal plane of lens L2,
where the diffraction pattern may be recorded. [Reproduced from Diffraction, by C. A. Taylor, with
the permission of the Institute of Physics.]

(a) (b)

FIGURE 5.9. Diffraction patterns from circular holes: (a) one hole I mm diameter, (b) two holes
each of I mm diameter, set 3 mm apart in the horizontal direction. [Reproduced from Diffraction, by
C. A. Taylor, with the permission of the Institute of Physics.]

5.6.2 Single Hole

The diffraction pattern for a single hole has been illustrated in Figure 5.7a,
and equation (5.29), which represents the diffraction function, or transform, is
plotted in Figure 5.7b. The diffraction pattern of a circular hole is often called the
Airy disk, after its discoverer.

5.6.3 Two or More Holes

The Fourier transform of a single hole may be represented generally by
(5.32), as well as by (5.29). A second hole, displaced by a vector distance a from
the first, has an identical transform function but with a vector distance r + a
5.6 PRACTICE WITH TRANSFORMS 307

from the origin. The total transform is the sum of the displaced and undisturbed
transforms:

GT(S) = i P (r)eXP[i2:rr(r'S»)dV + i P (r)eXP[i2:rr(r+a) ·S)dV

= i per) exp[i2:rr(r· S») dV {I + exp[i2:rr(a· S»)}

= Go(S){l + exp[i2:rr(a . S»)} (5.45)

where Go(S) is the transform of the undisturbed function. The term {I +

exp[i2:rr (a· S)]} is a fringe function modifying Go(S).lthas the value 2 when a·S is
integral, and is zero halfway between, as can be shown from de Moivre's theorem.
Thus, the total transform is that of a single hole (Figures 5.7a or 5.9a) crossed by
a system of planar fringes (Young's fringes), as shown by Figure 5.9b. The fringe
system lies perpendicular to the direction joining the two holes. As the distance
a is increased, the distance between the individual fringes becomes smaller, an
example of the reciprocal nature of the diffraction process. The amplitudes of the
fringes vary sinusoidally, alternate fringes having a relative phase difference of :rr .
The addition of further pairs of holes of different spacings and differing
orientations gives rise to more fringe systems, all with the same reciprocal property
discussed above. The complete diffraction pattern of a molecule may be thought
of as a superposition of many sets of fringes. The sequence of optical transforms
in Figure 5.10 shows not only an increase in the number of pairs of holes, as with
the benzene ring itself, but also the effect of increasing the numbers of benzene
ring entities in both one and two dimensions.

5.6.4 Change of Origin

If the origin to which a scattering species is referred be changed by the
addition of a fixed vector p to all rj vectors, then from (5.40)

=L
II

Gp(S) fJ exp[i2:rr(rj + p) . S)
j=l
n
= L fJ exp[i2:rr(rj . S») exp[i2:rr(p. S»)
j=l

= Go(S) exp[i2:rr(p • S)) (5.46)

In (5.46), Go(S) is modified by the fringe function exp[i2:rr(p . S»). Since, in

practice p • S is integral at points where the transform can be observed, it follows
that exp[i2:rr(p.S)] has a magnitude of unity. Thus, the amplitude (and intensity) of
308 5 FOURIER SERIES AND TRANSFORMS

FIGURE 5.10. Optical diffraction patterns illustrating scattering: (a) two-dimensional lattice (portion
only); (b) single molecule (simulated benzene ring); (c) two such molecules; (d) four molecules; (e) a
row (portion only) of six molecules; (f) a net (portion only) of molecules. [Reproduced from Optical
Trans/orms , by C. A. Taylor and H. Lipson. with the permission of Routledge.l
5.6 PRACTICE WITH TRANSFORMS 309

the transfonn is invariant under translation, as can be demonstrated as described

in Section 5.1. The phase, however, is dependent upon position, from which
it is apparent that a true position cannot be detennined without a knowledge of
relative phases-the central problem in crystal structure detennination, as we have
remarked already.

5.6.5 Systematic Absences

At this point, we can show how the Fourier transform of a crystal can predict
systematic absences. Suppose that the vector displacement a in (5.45) is replaced
by (a/2 + b/2), consistent with a C-centered unit cell. Then, we have

GdS) = Go(S) {I + exp[i2n ~(a + b) . S]} (5.47)

which, from (5.42), becomes

GdS) = Go(S){l + exp[in(h + k)]} (5.48)

For (h + k) odd, GdS) is identically zero, but it equals 2Go(S) for (h + k) even,
which we recognize as characteristic for a C -centered unit cell (see Section 3.7).
Equations like (5.48) can be developed for all translational symmetries. Such
results show clearly those reciprocal lattice points that cannot be sampled, whatever
the nature of the contents of the unit cell.

5.6.6 Reconstruction of the Image

Consider again the diffraction pattern of the hypothetical two-dimensional
crystal in Figure 5.10f. The lattice in this example is defined by the separations
of the simulated benzene molecules. The weighted reciprocal lattice is the trans-
fonn of a single molecule (Figure 5 . lOb ) crossed by fringes, the principal sets
of which are governed by the basic translations of the lattice (Figure 5.10a).
Again, we see the continuous transfonn decomposed into, or sampled as, a spot
pattern.
In reconstructing an image from its diffraction pattern, our object is to attain,
say, the structure of Figure 5.10f given only its diffraction pattern, or transform.
The fringes themselves are rather like a diffraction grating. A set of true sinusoidal
fringes used as a mask would produce a diffraction pattern that is the original two-
hole object. Thus, in reconstructing the object from its diffraction image, we are
really seeking the diffraction pattern of the diffraction pattern, or the transform of
the transfonn. We may express this result mathematically in the following manner.
The transfonn of (5.32) will, for a crystal, give an expression for the dis-
tribution of the electron density per) in the fonn (note the change of sign of the

1
exponent)
per) = G(S) exp[ -i2n(r . S)] dv (5.49)
310 5 FOURIER SERIES AND TRANSFORMS

where the integral extends over a volume v in reciprocal space. Using (5.40) in the
context of the unit cell of a crystal, we write

per) = Iv F(hkl) exp[ -i27r(r . S)] dv (5.50)

Equation (5.49), or the equivalent (5.50), is the Fourier transform of (5.32). Thus,
the transform in Figure 5.lOf corresponds to the calculation in (5.32), whereas the
structure in Figure 5.lOf is the result of evaluating (5.49) or (5.50).
From the F (hOI) weighted reciprocal lattice section of the crystalline plat-
inum derivative of phthaIocyanine, a mask (Figure 5.11 a) was prepared in which the
relative amplitudes of the structure factors were indicated by the sizes of the holes.
When this mask was used in the optical diffractometer, the transform obtained
was that shown by Figure 5.11b. Comparison with the electron density map from
the fully solved structure (Figure 5.11h) shows that the recombination obtained
furnishes a good reconstruction of the molecule that, in the crystal, gave the spot
pattern from which the mask was prepared. One will ask immediately how the
phase problem has been overcome in this reconstruction, as it appears not to have
been even considered.
We have, in this example, a rather special case. In the crystal of platinum
phthalocyanine, the space group is P2] / c and there ~e two molecules in the
unit cell. Thus, the platinum atom occupies the origin (l) in the asymmetric unit,
and makes a positive contribution to all structure factors. It is such a heavy atom
(high atomic number) that it dominates the contributions from all other atoms in
the structure so causing all F(hOl) structure factors to have a positive sign (zero
phase angle). Thus, all hOt structure factors have the same relative phase, and
the transform of the diffraction pattern gives a true representation of the structure
without further reference to phase.
Figure 5.11c-g and c'-g' shows different portions of the hOI reciprocal lattice
section and the corresponding transforms. The effect of the cut-off on the resolution
is well illustrated.

Representation of Fourier Transforms

If an object is centrosymmetric and the origin is taken at a center of symmetry,
then the Fourier transform is real (see Section 3.6.2) and only one diagram is needed
to represent it. This situation exists for the hOI section of the transform of platinum
phthaIocyanine (Figure 5.11a). If the object is not centro symmetric, then it is
necessary to use two diagrams to display the transform, either amplitude and phase
or real and imaginary parts. The phase, or the real and imaginary components, will
vary according to the choice of origin, but the amplitudes remain invariant under
change of origin. (See Section 5.6.4.)
5.6 PRACTICE WITH TRANSFORMS 311

(a) (b)

(c) (d) (e)

. .
•••

(f) (g)

FIGURE 5.11. Platinum phthalocyanine. (a) Mask of the hOI diffraction pattern; (b) Optical
transform of (a), showing a complete molecule and portions of neighboring molecules in the (projected)
crystal structure. [After Dunkerley and Lipson, Nature, 176, 81 (1955).] (c)-(g) Increasingly large
portions of the diffraction pattern; (c')-(g') Corresponding transforms-the effect of the cut-off of
the pattern on the resolution in the transform is evident; (h) Electron density contour map. All the
diffraction patterns relate to the hOI data, so that the corresponding transforms are x, z projections in
real space [after Taylor and Lipson, loco cit.].
312 5 FOURIER SERIES AND TRANSFORMS

FIGURE 5.11. (Continued)

5.6.7 Transforms and Inverse Transforms

Consider a one-dimensional square-wave function defined by

f(x) = [2 for -a/2 .:s x .:s a/2

o for -a /2 > x > a /2

which has the form shown in Figure 5.12. The Fourier transform in the one
dimension x is, from (5.32),

G(S) = j a/ 2
2exp(i2nSx)dx (5.51)
-a/2

which is solved readily to give

G(S) = 2sin(nSa)/(nS) = 2asin(nSa)/(nSa) (5.52)

This transform, in which the function sin(n Sa) / (n Sa) is typical for the transform
of a pulse waveform, has the form shown in Figure 5.13, where G(S) is plotted as a
function of S in units of 1/a. It may be noted that, characteristically, the length 2/ a
5.6 PRACTICE WITH TRANSFORMS 313

fix)

,.- 2.0 -+----,

I I x
-a -ta o a

FIGURE 5.12. Square wave of amplitude 2.0, defined for the period -a/2 to a/2; f(x) = 2.0 for
-a/2 :: x:: a/2; f(x) = 0 for Ixl > a/2.

G(S)

a a a a

FIGURE 5.13. Fourier transform of the square wave in Figure 5.12. Subsidiary, decreasing maxima
arise at intervals (2n + \)a/2 (n = 1,2, ... ).

between the first two nodes on each side of the central maximum is the reciprocal
of the width of the function f (x). We would have obtained the same result if we
had used the real part of exp(i2JTSx) in (5.51), that is, 21;/2 2cos(2JTSx) dx,
because Figure 5.12 indicates an even function: f( -x) = f(x).
Consider next the function

for _12 -< x -< 12

for Ixl > 1
which has a saw-tooth wave form, Figure 5.14. Because of the symmetry of this
function about the point x = 0, the Fourier transform of this function has the
general form

G(S) =2 f
-1/2
o
(1 + 2x) cos(2JT Sx) dx + 2 101/2 (1 -
0
2x) cos(2JT Sx) dx

(5.53)
314 5 FOURIER SERIES AND TRANSFORMS

fix)

s
FIGURE 5.14. Saw-tooth wave form defined forthe period - ~ to ~; f(x) = 1 - 21x I for Ixl :::: 1/2;
f(x) = 0 for Ixl > ~.

which may be simplified to 2 Jo1/ 2 cos(2n Sx) dx - 4 J01/ 2 X cos(2n Sx) dx. The
first of these two integrals solves to sin(n S) I (n S). Integrating by parts, the second
integral gives (l/n 2 S2)[1 - cos(n S)] - (lIn S) sin(n S), so that the total result is

(5.54)

It is left as an exercise to the reader to plot this transform, as a function of S,

(S = 0,0.05,0.10, ... ), and to show that (5.54) is obtained also if exp(i2n Sx)
is used in place of cos(2n Sx) in (5.53). The plot of the transform should have
a maximum at S = 0, zeroes at S = 2n (n = 1,2,3, ... ), and small decreasing
maxima at S = 2n + 1 (n = 1,2,3, ... ).
A function and its Fourier transform are reciprocally related; we noted this
feature in studying Fourier series in Section 5.3. We can illustrate this property by
means of two programs, TRANS 1* and FOURID*, that are part of the suite of
programs that is included with this book.
Consider the one-dimensional, periodic function f (X) listed in Table 5.2
at 30ths of the repeat distance along the X axis, from n = 0 to 30/30. It can be
represented by a series of sine and cosine terms with coefficients determined by
the program TRANS 1* (see Chapter 11), which calculates the Fourier transform
of the function. When this program is executed on the data shown in Table 5.2,
with h max = 10, the coefficients listed in Table 5.3 are obtained. Note that only
the values of the function f(x) are entered as data. If these coefficients are used
with the program FOURID*, which calculates a one-dimensional Fourier series,
the initial function is regained (Table 5.4): the transform of the transform is the
5.6 PRACTICE WITH TRANSFORMS 315

TABLE 5.2. Periodic function p(X), listed at 30ths (n) of the repeat
distance along the X axis
n f(x) n f(x) n f(x) n f(x)

0 50 52 2 57 3 63
4 69 5 76 6 81 7 84
8 85 9 83 10 79 II 73
12 66 13 60 14 50 15 42
16 37 17 32 18 41 19 62
20 90 21 117 22 126 23 123
24 113 25 95 26 78 27 65
28 56 29 51 30 50

TABLE 5.3. Coefficients A(h) and B(h), for h = 0 to 10, transformed from p(X) data
in Table 5.2 using TRANS 1*
h A(h) B(h) h A(h) B(h) h A(h) B(h) h A(h) B(h)

0 71.87 0.00 I 2.22 -4.61 2 -15.08 -2.53 3 0.55 5.84

4 2.14 -1.54 5 -0.99 -0.21 6 0.19 0.36 7 0.15 -0.25
8 -0.26 -0.04 9 -0.07 0.13 10 0.43 0.06

original function. Further details on the manipulation of these programs are given
in Chapter 11. However, we mention here that the value of h should be below the
critical value of ~ /2, where ~ is the sampling interval, 30 in this example. The
sampling theorem states that if the Fourier transform of a function is zero for all
frequencies greater than a critical frequency fe, that is, its bandwidth is limited
to frequencies smaller than fe, then the continuous function can be determined
from a knowledge of its sampled values. The function in Table 5.2 is not of this
*1
character, so that the maximum frequency h used should be to of the sampling
interval, 30. The reader may care to investigate this property by using the programs
with different values of h. Further reading about sampling may be found in the
literature. a This problem is rarely manifested in the normal Fourier syntheses of
x-ray crystallography. The experimental IFo I data fall towards zero at the higher
values of h, k, and I because of the attenuation arising from the temperature factor
effects on the atomic scattering factors (see Section 3.9.8). In working with the
sharpened Patterson function and the E-map (q.v.), the sharpening inherent in the
coefficients of the former reduces the tendency to zero of the higher order spectra,
whereas in the latter the coefficients in the Fourier transformation to an E-map

aSee Bibliography (Brigham).

316 5 FOURIER SERIES AND TRANSFORMS

TABLE 5.4. Transformation of the Coefficients A(h) and B(h) from

Table 5.3, using FOURID*, to give the Original Function I(x) of Table 5.2

n f(x) n f(x) n f(x) n f(x)

0 50.43 51.95 2 56.47 3 63.52

4 69.15 5 75.50 6 81.72 7 83.67
8 84.48 9 83.91 10 78.39 II 72.55
12 67.43 13 59.02 14 50.57 15 43.00
16 34.66 17 32.62 18 41.49 19 60.90
20 90.76 21 117.64 22 126.27 23 122.34
24 112.70 25 95.35 26 77.50 27 65.48
28 55.99 29 50.54 30 50.43

are both sharpened and significantly decreased in number. We consider in later

chapters how these cases are treated in practice.

5.6.8 Delta Function

Another important function, considered in Chapter 3, is the Gaussian
distribution
(5.55)
where k may be regarded as the width of the function. In finding the Fourier
transform of this function, we can make use of the fact that it is an even function,
so that we use the cosine part of exp(i2n Sx) and determine twice the sum from
zero to infinity, that is,

If
-
k
00
-00
exp( -71' x 2/k2) exp(i2n Sx) dx

=-21
k 0
00
exp(-nx 2 /k 2 ) cos(2nSx) dx

10
From tables of standard integrals, or otherwise, a 00 exp( _a 2x 2 cos(bx) dx
(71' /4a 2 ) 1/2 exp( _b 2 /4a 2 ), so that the required Fourier transform becomes

(5.56)

Thus, the transform of a Gaussian function is another Gaussian, in reciprocal space,

of a width 1/k, the reciprocal of the width k in real space.
The integral

a Handbook of Chemistry and Physics, edited by R. C. Weast, Chemical Rubber Co. Cleveland, Ohio
(1974-1975).
5.6 PRACTICE WITH TRANSFORMS 317

evaluates to (1/n)I!2r(~), which is unity (normalized) for all values of k.

Consider next the function

(5.57)

This function has the following properties:

= 10 for x =F 0

i:
8(x)
00 for x =0

8(x)dx = 1

and is known as the (Dirac) 8-function; it corresponds to an infinitely sharp line of

unit weight at the origin. As the width k in (5.57) tends to zero, so the transform
of the 8-function tends to unity, and in the limit where k = 0 reaches it.

i:
This result follows, since

f(x)8(x) dx = f(O)

i:
and if we let f(x) = exp(i2nSx), then
exp(i2nSx)8(x) dx = exp(O) = 1
If the 8-function is located at x = Xo, we have f(x) = 8(x - xo). The Fourier

i:
transform of f (x) is then

i:
C(S) = 8(x -xo)exp(i2nSx)dx

= 8(x) exp [i2nS(x + xo) dx

= exp (i2n Sxo)

Thus, the Fourier transform of a 8-function at a point xo is equal to exp(i2n Sxo);

when the 8-function is at the origin, its Fourier transform is at unity for all
values of S.
Of particular interest is a set of 8-functions that define a one-dimensional
lattice of spacing a. Its Fourier transform will be another set of 8-functions of spac-
ing Va which define the corresponding one-dimensional reciprocal lattice. While
the above discussions have been confined, for convenience, to one-dimensional
space, the results are equally true in higher dimensions, and we shall consider such
applications in later sections.
318 5 FOURIER SERIES AND TRANSFORMS

5.6.9 Weighted Reciprocal Lattice

As a final, practical example, we illustrate the power of the Fourier transform
with the crystal structure of euphenyl iodoacetate, Figures 1.2 and 1.3. The crystal
is monoclinic, with unit-cell dimensions a = 7.260, b = 11.547, c = 19.217 A,
f3 = 94.10°. There are two molecules in the unit cell, and systematic absences
indicated space group P21 or P21 1m. The latter is not possible, because the two
molecules in the unit cell would have to lie on special positions of symmetry either
m or 1. The chiral nature of the molecule, a tetracyclic triterpene, precludes both
of these symmetries, so that the two molecules lie in general positions in space
group P21.
Figure 5.15 is a reconstruction of the x-ray photograph, or weighted
reciprocal lattice of the hOl section for this crystal; this section is centric, symmetry
p2. Since the iodine atom is very much heavier that the other atoms present in the

h (XO) 1lXl~'s_--x_------1

'-
" .." ".. .....
'"" " "' "- ,
'"

....•......." .........., ". ..

"
. ,••
" ~ ."
. •• - '1ft'.••

..... .••....
" ~o.. ,
~~ ,6':~
,/ ....""

'." ...
. ~ ~

" "" ..
'-

~,' ,
.~,
~
"" "-
~
~, -'" ~.
o /45 0 \ - - 1(z") axis

0· •• •· •••••.. • • • • • • •
•• •••• ••• •••
•
••

•• ••• ••••••
•• ••• • • ••
•
•

•• • •• • • • •••• oL.-----:. _ T...,

FIGURE 5.15. The hOI section of the weighted reciprocal lattice of euphenyl iodoacetate; the
I-I fringe spacing (three lines of zero amplitude are indicated) is 0.18 reciprocal lattice units, and
the I-·J vector makes an angle of ca 45° with the z* axis (It may help to inspect the diagram edgewise).
5.6 PRACTICE WITH TRANSFORMS 319

)( axis
X
..aXIs

(11- 90
X~
x ------.,.
,,' :
" I ..
1~,'~4~~~~I======~::::==============~z~a~x~is~ aXIs
, I
Z

,,/ 0 z (11- 90)

,,,
,
"
-I

FIGURE 5.16. Geometry of the reconstruction of the coordinates of the iodine atoms in euphenyl
iodoacetate. The f3 angle is 94.1 0; X, X', and Z are measured along the x, x*, and z axes, respectively.
Note that since the space group of euphenyl iodoacetate is P21, the y coordinates of the iodine atoms
may be set at ± ~, so as to fix the position of the origin with respect to the symmetry elements.

crystal, the unit cell may be regarded, to a first approximation, as a pair of iodine
atoms distant d apart.
From Section 5.6.3, we would expect to find a set of fringe systems running
through the transform, arising from the heavy iodine atoms. In Figure 5.15, the
two most obvious fringe systems at an angle of 85.90° to each other arise from the
unit-cell translations a and c. In addition, the dashed lines indicate another promi-
nent fringe system, that of the iodine atoms. Note that the system is more prominent
at higher angles, because the scattering from heavy atoms falls of relatively much
less than that from the lighter atoms present in the structure.
The perpendicular spacing of the fringes is 0.18 reciprocal units which,
because eu Ka x-radiation (A = 1.5418 A) was used in the experiment, corre-
sponds to an I-I spacing of 8.565 A (in projection) lying in a direction normal to
the fringe system, that is, at approximately 45° to z*, Figure 5.16. Measuring Xl
and z along the x* and z axes respectively, we have

Xl = (dj2) sin(45 + f3 - 90)0 = 3.237 A

Z = (dj2) cos(45 + f3 - 90t = 2.804 A

320 5 FOURIER SERIES AND TRANSFORMS

From the figure, X = X' coseB -90)°, whence X = 3.229 A. Hence, thefractional
coordinates are x = 0.445 and z = 0.146. After the structure was fully solved, the
refined values of these coordinates were x = 0.4274 and z = 0.1431.

5.7 Some General Properties of Transforms

We summarize here some of the more important properties of Fourier trans-

forms in their applications to x-ray crystallography; for each property the spaces
may be interchanged.

Operation in Bravais space Result in reciprocal space

Rotation about an axis Rotation about a parallel axis at the same speed
Change of scale in a given direction Reciprocal change of scale in the same direction
Translation Modulus unchanged; phase modified by a fringe function
Addition of n units Vector summation of n transforms referred to a common origin
(a) Two parallel units Transform for one unit crossed by parallel, planar fringes;
maximum amplitude doubled
(b) Two units related by a center Transform for one unit crossed by wavy fringes which may
of symmetry be approximately planar in limited regions
Convolution of two functions Transform is the product of the individual transforms

The discussion of the convolution of two functions is our next task.

5.8 Convolution

We consider here the last of the properties of transforms listed in the previous
section. The convolution integral is often called the "folding" integral, for a reason
that will become clear as we continue the discussion.

5.8.1 Convolution and Diffraction

The Fourier transform of a slit diffraction grating of a given width can be
considered in terms of the diffraction at a single slit of that width, together with that
of an ideal, infinite grating. We have, first, the diffraction pattern, or transform,
of the single slit, shown by Figure 5.17a, and then the transform of the ideal
grating, Figure 5.17b. These two transforms, in reciprocal space, are multiplied,
point by point, to give the product in real space (Figure 5.17c), which is called
the convolution of the two functions. In a crystal, we have the contents of the
unit cell and a point function, the lattice, in Bravais space. The convolution of
these two functions is the diffraction pattern in reciprocal space. The transform
of a lattice is another set of points, the reciprocal lattice points, and it has unit
5.8 CONVOLUTION 321

P
II
II
II
II
II
II
II

FIGURE 5.17. Relationship between amplitudes of diffraction. (a) Single slit of finite width (thin
line); (b) Ideal slit grating (dashed line); (c) Grating of finite-width slits (bold line). The transform
of the grating of finite-width slits is the convolution of the transforms of the single slit and the ideal
grating. [Reproduced from Optical Transforms, by C. A. Taylor and H. Lipson, with the pennission of
Routledge.]

value at each point and zero value elsewhere. The product of the two transforms
is the transform of the contents of the unit cell sampled at the reciprocal lattice
points.
The image of a diffraction spectrum is influenced also by the divergence
of the incident x-ray beam, the crystal shape and the mosaic structure of the
crystal, because all of these factors have a bearing on the Fourier transform of the
crystal.

5.8.2 Convolution Integral

The convolution of two functions is a physical concept of significance in
divers scientific fields. We consider the process of convolution now in a little
more detail. Consider two functions fer) and g(r) in Bravais (real) space, and let
their transforms in reciprocal space be F(S) and G(S), respectively. The Fourier
transform of the product of these two transforms may be written as

T(r) = f F(S)G(S) exp( -i2nr· S) dVs (5.58)

and is a function in Bravais space. From earlier sections we know that F(S) is
given by

F(S) = f fer') exp(i2nr' . S) dVr (5.59)

where r' is a vector different from r in general, but ranging in the same Bravais
space, and the change of sign arises because we are considering the reverse
322 5 FOURIER SERIES AND TRANSFORMS

transformation. Thus, we have

T(r) = f G(S) {f fer') exp(i2nr' . S) dV r ,} exp( -i2nr· S) dVs

= f f(r'){f G(s)eXP[-i2n(r-r').S] dVs} dV r,

= f f(r')g(r - r') dVr , (5.60)

which is the convolution fer) * g(r), or c(r); g(r - r') is the function g(r) with
its origin moved from r = 0 to r = r'.
We can give a physical interpretation to this process. In Figure 5.1Sa, an
element or' of the function fer') is defined. According to (5.60) each such ele-
ment of fer') must be multiplied by the sharp function g(r - r') in Figure 5.lSb
before integration. In Figure 5.lS(c), this process is shown for three elements,
with the result in Figure 5.lS(d). The completed convolution is shown by
Figure 5.1Se; it is evident that the function fer')
has been repeated at each value

t
(a) (b)

Lfi~ oCr') r'

rJ
r
ft

fu
(c) (d)
rr'J' g(rJ

r'
fi~ fi n~ r
n
(e)

~ r
FIGURE 5.IS. Convolution of two functions in the one-dimension space of r. (a) Function f(r'),
showing an element of width or'; (b) Function g(r), showing two sharp peaks; (c) Function fer')
with three selected elements or'; (d) Convolution of (b) and (c); the three elements are reproduced by
g(r). modified by the values of that function; (e) the completed convolution fer') * g(r); because the
function g(r) does not have infinitesimal width. fer') is reproduced in a slightly modified form.
5.8 CONVOLUTION 323

of g(r). Note that g(r - r') is g(r') mirrored, orfolded, across the ordinate axis
and shifted by an amount r.
The converse of (5.60) is equally true, that is,

T[c(r)] = F(S)G(S) = f {f f(r')g(r - r') dV r, } exp( -i2nr· S) dVr

= f fer') {f g(r - r') dV r } exp( -i2nr· S) dVr, (5.61)

If we now let r - r' = r", where r" is another independent variable in the same
space as rand r". Hence,

T[c(r)] = f f(r')exp(-i2nr'. S)dVr' f g(r")exp(-i2nr". S)dVr,

= F(S)G(S) (5.62)
so that the transform of the convolution is the product of the individual transforms.

5.8.3 Crystal Structure and Convolution

We have shown in the previous section, and illustrated in Figure 5.18, that
the convolution, in one-dimensional space, of a function fer) with a function
g(r) that consists of two sharp peaks, results in fer) being repeated, modified
slightly in accordance with the width of g(r), at the two locations in g(r). A pair
of 8-functions can be derived from g(r) by allowing the width of the sharp peaks,
normalized to unit area, to be reduced while maintaining the area at unity. In the
limit as the width becomes infinitely small, the function becomes infinitely high
at each location. We now have a pair of 8-functions that have the value unity at
each location but zero otherwise. If these 8-functions were to be convolved with
fer), then fer) would be repeated exactly at the locations of the r vectors. We
may write this process in mathematical terms in the following way. Initially, we
have the convolution expressed as

c(r) = f fer' - r)g(r) dVr

where g(r) is normalized to unit area. As the width tends to infinitesimal size, so
we have

c(r) = f fer' - r)8(r) dVr = fer') (5.63)

from Section 5.6.8. If we translate the 8-function by a vector r" along the positive
axis, so that the 8-function is defined as 8(r - r"), then the convolution becomes

c(r) = f fer' - r)8(r - r") dVr (5.64)

324 5 FOURIER SERIES AND TRANSFORMS

so that the function f (r') is reproduced at r" from the origin, that is, it is the function
fer - r"). It is straightforward to extend the argument to a one-dimensional
lattice. The convolution of f (r') with an infinite array of 8-functions, g(r)
Lj~-oo 8(r - jr"), is given by

L
00

c(r) = fer - jr") (5.65)

j=-oo

which is periodic in r", Figure 5.19.

Thus, a crystal structure may be regarded as the convolution of a unit cell,
defined by vectors a, b, and c, and a three-dimensional set of 8-functions that have
the value unity at each Bravais lattice points but zero elsewhere.
The transform of a unit cell contents may involve one or more molecules,
and is a continuous function in reciprocal space. However, the conditions for inter-
ference lead to the fact that this transform may be sampled, experimentally, only at
the reciprocal lattice points, so that the convolution of the unit cell and the lattice
of 8··functions is the product of the transform of the unit cell contents and the
reciprocal lattice, and is manifested in the experimental diffraction pattern.
The essence of this process may be grasped from Figure 5.20, in which
the transform of the contents of the centrosymmetrical unit cell of naphthalene,
projected on to the ac plane, has been overlaid with a drawing of the x*, z*
reciprocal lattice in the correct orientation. The weights of the transform at the
reciprocal lattice points hkO correspond to the values of IF(hkO)I.

(a)
i
t
f(r)
(b)

1~
i~ I I I I I I I I
r r

(e)
c(r)

r
FIGURE 5.19. Convolution: (a) Structural entity fer); (b) The function g(r) is an infinite set of
o-functions oCr); (c) Convolution c(r) = fer) * oCr).
5.9 STRUCTURE SOLUTION IN BRIEF 325

FIGURE 5.20. Fourier transform of the hOI projection of the centrosymmetric structure of naphtha-
lene, showing the a* and c* directions. The transform is overlaid with a drawing of the reciprocal lattice
in the correct orientation. The amplitudes IF (hOI) I are the amplitudes of the transform at the reciprocal
lattice points. Reflections such as 201, 202, and 801 are sensitive to small changes in the orientation
of the reciprocal lattice, whereas reflections such as 203, 603, and 802 are relatively insensitive under
rotation.

5.9 Structure Solution in Brief

An x-ray diffraction pattern can be recombined to form an image of the

object from which it was derived provided that we know the indices h, k, l, the
amplitude IF I and the relative phase ¢ for each reflection. The absence of direct
measurements of the phases means that they must be obtained indirectly. If it can be
done satisfactorily, a Fourier transform (Fourier series) can be used to convert the
reciprocal-space diffraction pattern into a real-space object, the crystal structure.
This section is just a brief overview of some of the problems in crystal structure
analysis, and how they may be overcome.

5.9.1 Use of Heavy Atoms

We considered a special case of this method in Section 5.6.6, where all
F(hOl) data for platinum phthalocyanine were positive in sign. In a more gen-
eral case, the heavy atom (which acts like the speck of dust on the transparency,
Section 5.1) will lie at a general position in the unit cell, probably related to one or
more atoms by symmetry. Assuming for the moment we can determine the coor-
dinates of the heavy atoms in the unit cell (see, for example, Section 5.6.9), we
can use them to calculate approximate phases for the reflections. Then, the Fourier
transform (5.50) becomes a summation, or series, with coefficients of the form
326 5 FOURIER SERIES AND TRANSFORMS

!Fol exp( -i¢e), where !Fol is an experimentally determined structure amplitude

and ¢e is the relative phase calculated from the positions of the heavy atoms alone.
We may write these coefficients in the form

(5.66)

where IFe I is the amplitude produced by the heavy atoms alone. The right-hand
side of (5.66) is the product of the transform of the heavy-atom portion of the
structure and a function (IFol/!Fel) which is of zero phase for every reflection.
The result in real space is the convolution of the heavy-atom portion of structure
with the transform of (!Fo I/!Fe I). This transform, because of its zero phase, will
have a large peak at the origin, and the convolution will tend to make the heavy-
atom part of the structure dominant. However, because !Fol i- !Fel, there will
be a background effect imposed upon the heavy-atom part of the structure that
will modify it in the direction of IFo I. If the heavy-atom positions are correct, or
nearly so, the convolution will lead to an improved model for the structure. It is
important to realize, however, that because of (5.66) such a convolution is always
biased toward the heavy-atom part of the structure, and success with this method
depends strongly on the degree of correctness of the heavy-atom positions.

Series Termination Effect

If a Fourier series is calculated with data only up to a certain value sin 8max , it
is equivalent to multiplying the transform of the structure by a function that is unity
up to sin 8max and zero above it. The result is the convolution of the complete trans-
form with that of the exclusion function. In two dimensions, the exclusion function
simulates a circular hole, and its transform is a Bessel function (Figure 5.7b) which
has a central maximum surrounded by maxima of alternating sign and decreasing
magnitude. The transform of the limited data set is, thus, the complete transform
convoluted with the Bessel function. The result can be seen in electron density
maps as contours around the atomic positions that decrease to zero from the central
maximum, and then are surrounded by ripples, or rings of decreasing magnitude
and alternating sign. The effect is most noticeable in the regions around the heavy
atoms in a structure (see, for example, Figure 6.47a).

5.9.2 General Phase-Free Transform

Although a diffraction pattern is invariant under the operation of translation,
a representation of translation is contained within the pattern, because when the
pattern is recombined the translational property of the object crystal is revealed.
The relative phases at different parts of a transform do change under translation,
but the changes are not evident because we observe the intensity of the transform

(5.67)
5.9 STRUCTURE SOLUTION IN BRIEF 327

(a)

r><J-'a<"
lx axis

1x axis
FI G U RE 5.21. Convolution with a rectangle of scattering centers (atoms). (a) H ypothetical four-atom
structure of coordinates ±(O. 1,0.2; 0.1, -0.2); interatomic vectors are shown by arrows. (b) Schematic
convolution of the four-atom structure with its inversion in the origin. Multiweight peaks arise from
a superposition of identical vectors.

Equation (5.67) is the counterpart of (5.33) for observations at reciprocal lattice

points.
Recombination based on intensity alone contains no information about the
lateral position of the object. All possible pairs of scattering points in the object
will be reproduced with the correct orientation and separation, but symmetrically
disposed about the center of the image. Hence, the image from the transform of
intensities is that of vector positions between pairs of scattering points (atoms)
in the object, all taken to a common origin. It is the Patterson function (see
Section 6.4.2), and may be thought of as the result of the superposition of numer-
ous fringe systems, all of which have a positive sign at the origin of the unit cell.
Mathematically, the transform of intensity, G(S)G*(S), is the convolution of the
transform of G(S), which is the electron density, with the transform of G*(S),
which is the electron density inverted in the origin.
In the hypothetical four-atom rectangle structure (Figure 5.21 a), there are 16
(4 2) interatomic vectors. Four of them are of zero length and coincide at the origin
(Figure 5.21b), and 12 are arranged in centrosymmetric pairs. Those formed by the
m symmetries of the rectangle (Figure 5.21a) are of double weight, whereas those
formed by symmetry 2, are of single weight. Figure 5.21b can be considered in
terms of Figure 5.21 a by transferring all interatomic vectors (arrowed) to a common
origin.
Alternatively,. we can think of Figure 5.21b as Figure 5.21a convoluted with
its inversion in the origin (which is the same as itself in this centrosymmetric
328 5 FOURIER SERIES AND TRANSFORMS

h~

~~, '1~.19
19~ 6~15
:I 7 \

: 5
\?:!) /7
\.::.J
\

6 \ ·15
0
·19 ·19

.15: 5 ~
.
~7 \ ~_kaxis

7~\-6~-6/ \
--~ --,-

~~;1,
·19 -6 I

-_ ....
·7 ·2 J 6 6".2
/ - - - ............ _-_.-,
72,-5.5\272/.5
·2 I 6
,,----,
6

·5\27
\.2
...... _-- ...
~5:~\5 ~6!~\5 ~
'i4\ I 7 I /.19 .191 7 I (z;\ \ 7 /.19 ,191, 7 I "
I I

!
h axis

FIGURE 5.22. Calculated transform for the four-atom structure (Figure 5.2Ia) convolved with its
inversion in the origin; contours are shown at -20, -10, 10, and 20. Fringe systems are apparent,
corresponding to the normals to the directions between all pairs of atoms in the four-atom structure.
(The labelled vectors h, k, and h - k refer to the discussion Section 5.9.3.)

arrangement). Thus, the structure in Figure 5.21a is drawn four times, with its
center of inversion at each of the atoms of the inverted structure in tum, and in the
same orientation to give Figure 5.21b.
We can arrive at the same results by calculating the appropriate transforms.
Using (5.44) with h = 6 (carbon), and neglecting the change in h with 8,
Figure 5.22 was plotted for h = 0 to 10 and k = 0 to 10. The contours (disregard the
three vectors highlighted for this application) show clearly the fringe systems that
are perpendicular to each of the six pairs of points in Figure 5.21a. In Figure 5.23,
the phase-free transform of the intensities (the squares of the values shown in
Figure 5.22, suitably scaled), we see the 16 peaks and 4 rectangles, just as in
Figure 5.21b. The orientations and separations are correct, but there is a four-fold
ambiguity with respect to the lateral positions. If, by some stratagem, we were
able to determine the correct relative phases, that those shown in Figure 5.22 were
in fact correct, then the transform of these amplitudes, with those phases, would
lead to the one correct result (Figure 5.24), from which we determine the atomic
positions ±(0.1,0.2;0.1, -0.2). We shall investigate the required cunning in the
next two chapters.
5.9 STRUCTURE SOLUTION IN BRIEF 329

11 15 2 5 3 11 2

13 3 13 3
4 4

5
6 6 5_yaxis

4 13 5 3 14 4
13 3 13
2 11 15 2 3 11 2

1
x axis

FIGURE 5.23. Calculated transform from Figure 5.22 for hand k = 0 to 10, using the squares of
the given amplitudes (phase-free) suitably scaled; contours are shown at 20,30,40, and 50, with the
origin peak shaded. The rectangles are correct in size and orientation, but there is a 4-fold ambiguity in
the lateral positions in the x,y plane. The ambiguity is related to the symmetry and not to the number
of atoms in the structure.

-1 0 0 0 0 0 4 0 0 -1
-1 0 0 0 0 0 0 -1
0 0 0 0 0 0 0 0

0 0 0 0 0 0

0 0 0 0 0 0 0 0 0

0 0 0 0 o _ y axis
0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0 0

-1 0 0 0 0 0 0 0 -1
-1 0 0 0 0 0 4 0 0 -1

1
x axis

FIGURE 5.24. Calculated transform from Figure 5.22 for hand k = 0 to 10, using the given
amplitudes and signs; contours are shown at 6, 14, 22, and 30. A single, correct structure is now
revealed.
330 5 FOURIER SERIES AND TRANSFORMS

5.9.3 Sign Relationships

Sign relationships (phases in the most general case) form a basis of the direct
method of structure determination (see Chapter 7). At this stage, it is interesting
to show how an important sign relationship may be deduced from the transform
of a centrosymmetric arrangement of atoms (Figure 5.22) that will be considered
to exist within a crystal unit cell.
Three vectors of the type h, k, and h - k, where we use h to represent the
triplet hkl and k another triplet hi k'l', form a triangle in reciprocal space, with one
vertex of the triangle at the origin; these vector terminations correspond to x-ray
reflections.
If, as in this chosen example, the vectors terminate in regions of relatively
high magnitude in the transform (strong reflections), then by counting the number
of times that the zero-boundary contour is traversed we can arrive at the signs of
the transform at the reciprocal lattice points h, k, and h - k, since the transform
must be positive at the origin:

Times zero-boundary
Vector hkl crossed Sign

h 450 3
k 420 2 +
h- k 030

Hence,
s(h) s(k) s(h - k) = +1 (5.68)
which is an expression of the triple product (~2) sign relationship, or Sayre's
equation (7.7) for a centro symmetric crystal.
If a vector terminates in a region of low magnitude in the transform (weak
reflection), then no certain conclusion can be drawn about its phase, because of the
difficulty in locating exactly the zero boundary contour in practice. In this context,
consider reflections such as 840 or 810 in Figure 5.22.

Bibliography

HARBURN. G., TAYLOR, C. A., and WELBERRY, T. R.. An Atlas of Optical Transforms, Bell (1975).
JAMES, R. w., The Crystalline State, Vol. II: The Optical Principles of the Diffraction of X-rays,
Bell ([ 950).
LIpSON, H., Optical Transforms, Academic Press (1972).
ORAN BRIGHAM, E., The Fast Fourier Transj(Jrm and its Applications, Prentice Hall ([ 988).
TAYLOR, C. A., Diffraction, Institute of Physics (1975 ).
TAYLOR, C. A., and LIpSON, H., Optical Transformations, Bell (1964).
WOOLFSON, M. M., An Introduction to Crystallography, 2nd ed., Cambridge University Press (1997).
PROBLEMS 331

Problems

5.1. Show that f~~~2 sin(2nmx/c) cos(2nnx/c) dx is zero for all x, except
when m = ±n; m and n are both integers. Hence, state the single finite
value of the integral. One or more of the identities in Appendix A6 may be
useful.

5.2. Magnesium fluoride is tetragonal, with space P42/m nm, a = 4.625,

c = 3.052 A and two species MgF2 per unit cell at positions

2Mg O,O,O;~,~,~
4F ±(x,x,O;~+x,~-x,~)

The F(hOO) data are listed below:

h 2 4 6 8 10 12
F(hOO) -2.7 12.0 7.2 0.1 3.2 0.1

Calculate p (x) using the program FOURID* with a subdivision of 40,

and plot the function. Determine x. What length of the repeat is sufficient
to define the complete function. Investigate the effects of (a) using terms
up to h = 6, and (b) changing the sign of F(600).

5.3. Find the Fourier transform of the function

f(x) = j a
a/2
o
for Ixl < p
for x = ±p
forlxl>p

Then, transform the transform, and show that it equals the original function.
5.4. Find the Fourier transform and its inversion for the periodic function

f(t) = A cos(2nfot).

5.5. Using Equation (5.40), deduce the Fourier transform for two asymmet-
ric scattering units related by a center of symmetry; the center may be
taken as the origin. What general feature might be expected in the resulting
transform?

5.6. Show, from Fourier transform theory, the nature of the systematic absences
that would arise from a 21 screw axis along the line [0, y, ~] in an
orthorhombic unit cell.
332 5 FOURIER SERIES AND TRANSFORMS

5.7. Tetraethyldiphosphine disul!ide, (C2HS)4P2S2, crystallizes in the triclinic

system, with space group PI and one molecule in the unit cell. Figure P5.l
is the hkO section of the weighted reciprocal lattice for this crystal. Make
a photocopy of the diagram, identify, and draw lines to indicate, the fringe
system for the (double weight) P-S vector. Hence, allocate signs to the more
intense reflections (about 25) in the asymmetric portion of the reciprocal
space diagram.

•••.
..•• • ....••••••
.:·.:!e
....

...
• •• ••
• ••
"
•
• •
~, . . : .!, :••
• •

.. ,.: :..
t· .
. .... ..
.
.

• •
Ii
.
... •••..
...•
.. :;
•• :
•••
••
~

:
!!. .
•
-.
k

FIGCRE PS.1. The hkO section of the weighted reciprocal lattice of tetraethyldiphosphine
disulphide. [Reproduced from Optical Transforms, by C. A. Taylor and H. Lipson, with the permission
of Routledge.]

5.8. A hypothetical, two-dimensional, three-atom structure has the atomic coor-

dinatesO.l, 0.1; 0.1, -0.1; 0.2,0.3, withrespectto the origin of rectangular
axes. Draw a diagram to represent the convolution of this structure with its
inversion in the origin.

5.9. The figure field of Figure P5.2 represents the phase-free Fourier transform
of the intensities of the diffraction pattern of the three-atom structure in
Problem 5.8. How many images of the structure are present, and what are
the sets of possible atomic coordinates?

5.10. Refer to Figure 5.20. What are the signs of IFI for the reflections 001,
1= 1-6?
PROBLEMS 333

5 10 7 14 6 21 6 14 7 10 5

6 12 8 20 7 22 7 16 7 10 6

7 14 9 68 7 22 9 22 8 10 7

7 20 9 24 8 31 10 90 9 15 7

9 82 9 17 9 32 11 23 9 16 9

11 23 9 23 11 322 11 23 9 23 11 _ y axis

9 16 23 11 32 9 17 9 82 9

7 15 90 10 31 8 24 9 20 7

7 10 8 22 9 22 7 68 9 14 7

6 10 7 16 7 22 7 20 8 12 6

5 10 7 14 6 21 6 14 7 10 5

x axis
~
FIGURE PS.2. Figure field of the phase-free Fourier transform of the intensities of the hypothetical
structure in Problem S.8.

5.11. If J(x) = 1/ J(21T) exp( _x 2 /2) andg(x) = 8(x - 2), find the convolution
c(x) = J(x)*g(x) by multiplying the transforms of J(x) and g(x).