This page intentionally left blank

SPECTROGRAPH DESI GN FUNDAMENTALS

In recent years enormous changes have occurred in the eld of optical spectrometry.
The classical spectrometer has become obsolete and the spectrograph, in combin-
ation with the CCD detector, now offers a vastly superior approach. Although
the basic optical principles remain unchanged the design considerations are very
different, and in many cases more demanding. However, developments in computer
ray-tracing and computer-aided design have coped with these extra impositions and
have allowed the construction of a new generation of spectrographs.
The book covers the general principles of spectrographic design, and the prac-
tical and engineering aspects of a broad range of spectrographs and spectrometers.
This allows the reader to make an informed choice of instrument. It will be of
particular use when none of the immense array of manufactured spectrographic
and spectrometric instruments is suitable for a specialised task. The book deals
with materials and methods of construction and includes suggestions for the choice
of optical table, the design of slit mechanisms, and adjustable mirror, grating and
lens mounts, with suggestions for the alignment and calibration of the nished
instrument.
Spectrograph Design Fundamentals describes the design and construction of
optical spectrographs. It will be a valuable resource for academic researchers,
graduate students and professionals in the elds of optics, spectroscopy and optical
engineering.
John James is an Honorary Senior Research Fellowat the University of Glasgow
and a Fellowof the Royal Astronomical Society. He is the author of Students Guide
to Fourier Transforms, also published by Cambridge University Press, now in its
second edition.
SPECTROGRAPH DESIGN
FUNDAMENTALS
J. F. JAMES
Honorary Research Fellow
University of Glasgow
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN-13 978-0-521-86463-3
ISBN-13 978-0-511-27403-9
J. F. James 2007
2007
Information on this title: www.cambridge.org/9780521864633
This publication is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
ISBN-10 0-511-27403-3
ISBN-10 0-521-86463-1
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
hardback
eBook (EBL)
eBook (EBL)
hardback
Contents
Preface page ix
Acknowledgements xi
1 A brief history of spectroscopy 1
2 The relevant regions of the electromagnetic spectrum 6
2.1 The limits of optical spectrography 8
3 Geometrical optics 10
3.1 Rays and wavefronts 10
3.2 Instrumental optics 11
3.3 Centred systems 12
3.4 Gaussian optics 12
3.5 Optical layout 20
3.6 Apertures, stops, elds, irises and pupils 20
3.7 Ray bundles 23
3.8 The HelmholtzLagrange invariants 23
3.9 Surface brightness 25
3.10 Black body radiation 25
4 Optical aberrations 28
4.1 The Seidel aberrations 29
4.2 Zero-order aberration 29
4.3 First-order aberrations 29
4.4 Theorems 35
4.5 Aberration coefcients for mirrors 37
4.6 The achromatic doublet 39
5 Fourier transforms: a brief revision 41
5.1 Fourier transforms 41
5.2 Theorems 42
5.3 Convolutions 42
5.4 The WienerKhinchine theorem 45
v
vi Contents
5.5 Useful functions 45
5.6 More theorems 49
5.7 Aliasing 51
6 Physical optics and diffraction 52
6.1 Fraunhofer diffraction 52
6.2 Two-dimensional apertures and oblique incidence 55
7 The prism spectrograph 57
7.1 Introduction 57
7.2 The traditional prism spectrograph 57
7.3 The focal curve theorem 59
7.4 The Littrow mounting 59
7.5 The PellinBroca prism 60
7.6 Focal isolation 61
8 The plane grating spectrograph 63
8.1 The shape of a monochromatic line spectrum 63
8.2 Blazing of gratings 67
8.3 Apodising 68
8.4 Order overlap and free spectral range 69
8.5 Grating ghosts and periodic errors 70
8.6 The complete grating equation 72
8.7 Differential dispersion 76
8.8 Mounting congurations 76
9 The concave grating spectrograph 89
9.1 The Rowland grating 89
9.2 The concave grating as a spectrograph 92
9.3 The concave grating as a monochromator 96
9.4 The aberrations of the Rowland grating 97
9.5 Practical details of design 98
10 The interference spectrograph 101
10.1 The phase angle 101
10.2 The FabryPerot etalon spectrograph 102
10.3 FabryPerot theory 102
10.4 The FabryPerot monochromator 104
10.5 The FabryPerot CCD spectrograph 108
10.6 Fore-optics 110
10.7 Reference fringes 111
10.8 Extraction of the spectrum 111
10.9 Choice of the resolution and gap 112
10.10 The crossed FabryPerot spectrograph 113
Contents vii
11 The multiplex spectrometer 114
11.1 The principles of Fourier spectrometry 114
11.2 The multiplex advantage 117
12 Detectors 120
12.1 Silver halide photography 121
12.2 Elementary electronic detectors 122
12.3 Detectors with spatial resolution 123
12.4 Exposure limitations 124
12.5 CCD software 125
12.6 CCD calibration 126
12.7 Spectrograph calibration 126
13 Auxiliary optics 128
13.1 Fore-optics 128
13.2 The astronomical telescope as fore-optics 130
13.3 Focal reducers 132
13.4 Schmidt-camera spectrography 134
13.5 Scattered light and bafing 134
13.6 Absorption cells 136
13.7 Fibre optical input 137
14 Optical design 139
14.1 First steps 139
14.2 Initial layout 139
14.3 The drawing board 140
14.4 Computer ray tracing 140
14.5 Renement of the optical design 141
14.6 Requirements of a ray-tracing program 144
15 Mechanical design and construction 150
15.1 The optical layout 150
15.2 Optical materials 162
15.3 Transparent optical materials 163
15.4 Reectors 163
15.5 Metals for construction 165
15.6 Other materials 167
16 Calibration 168
16.1 Sensitivity calibration 168
16.2 Wavelength calibration 169
16.3 Small spectral shifts and radial-velocity
measurement 170
16.4 Absorption measurements 170
viii Contents
17 The alignment of a spectrograph 172
17.1 The optical alignment 172
17.2 The focus 173
Appendix 1 Optical aberrations 175
Appendix 2 Wavelengths of spectral lines for calibration 179
Appendix 3 The evolution of a FabryPerot interference spectrograph 183
Appendix 4 The common calibration curve in silver halide
spectrophotometry 186
Bibliography 187
Index 188
Preface
Thirty-eight years ago, together with my colleague the late Robert Sternberg, I
wrote a book entitled The Design of Optical Spectrometers. It described the state of
the art as it was at that time after the great advances which had come in the previous
ten years, and it was intended for people who wished to build a spectrometer tailored
to a specic purpose, where perhaps one of the commercial designs was inadequate,
unsuitable, unnecessarily cumbersome, or expensive.
When at last the time came to consider a new edition it became clear that the
technology had changed so much that the classical optical spectrometer, in the sense
of monochromator, was more or less obsolete and that later developments such as
the desktop computer and the charge-coupled device had restored the spectrograph
to its former eminence. The restoration in no way annulled the optical improvements
of the previous 30 years but new constraints posed new problems in design. These
problems are now solved and the solutions are presented here.
The fundamentals of optical design have not changed, but the constraints are now
all different, and such properties as at elds are needed where before they could
be largely ignored; and focal ratios matter again when previously we could design
everything so that such trivia as spherical aberration and coma could be neglected.
There is, as always, a gap to be bridged between the elegant theory presented
in the undergraduate textbooks and the practical spectroscopic instrument standing
on the laboratory bench or bolted to the Cassegrain focus of an astronomical tele-
scope. The gap is partly in the limitations imposed by the curse of non-linearity
in geometrical optics and the contumacious aberrations it produces, and partly in
the sometimes obstinate physical properties of the materials of construction. As
always in scientic instrument making, the art is in knowing what must be precise
and exact and what can be left go hang at a crude level. There are tricks in this
trade just as in any other. The traditional engineers description of a physicist is
someone who designs a box that must be screwed together fromthe inside. (There
is a parallel physicists denition of an engineer: someone who, when asked what
ix
x Preface
is three times four? will get out a slide rule and answer about twelve.) There is
an element of truth in this and some of the hints in these pages may help to refute
the calumny.
This work is thus intended for the new generation of researchers who desire
high efciency in an instrument tailored to their own particular purpose, and who
have access to a mechanical workshop of moderate size where an instrument of
their design can be constructed. There has been no attempt to venture into the new
elds of optical resonant scattering spectroscopy, tunable laser spectroscopy or
other specialised techniques and so the book is directed to chemists, astronomers
and aeronomers as much as to physicists.
Manufactured spectrographs are in no way decient, but are necessarily com-
promises, both in performance and cost, and are often intended for teaching or
dedicated to a routine task such as sample analysis. A specially designed instru-
ment has no difculty in excelling them for a particular investigation, particularly
in academic research elds where new areas are being explored and where no
established technology yet exists.
It is to the basic optical design and construction of such individual instruments
that this book is dedicated. I also take the opportunity of acknowledging my late
good friends and colleagues, Dr H. J. J. Braddick, Rob Sternberg and Larry Mertz,
from whom I learned so much.
Acknowledgements
Many individuals and institutions have contributed to the research funds and facil-
ities which yielded the knowledge and experience written down here. Chief among
these are the following:
The Royal Society Paul Instrument Fund for numerous development grants
Royal Society Research grants for eld-work support
The Royal Society and Royal Astronomical Society Joint Permanent Eclipse Com-
mittee for expedition funds
The Japan Society for the Promotion of Science for eld-work support
The British Council, in Spain and Japan, for travel grants for collaborative devel-
opment and eld-work support
The Science and Engineering Research Council, for development grants, expedition
funds and eld-work support
El Instituto Astrosica de Andalucia, for collaboration, hospitality and facilities
during eld-work
The Royal Greenwich Observatory, Herstmonceux, for test facilities and the use of
telescopes while testing prototype instruments
lObservatoire de Haute Provence for observing facilities and the use of telescopes
for observation
The University of Texas McDonald Observatory for technical facilities during
eld-work
The Atenisi Institute, Nukualofa, Tonga, for facilities and hospitality during
eld-work
xi
xii Acknowledgements
The University of Manchester Schuster Physical Laboratory workshop and draw-
ing ofce staff for meticulous design detail and expert workmanship on many
instruments
The University of Glasgow for an Honorary Research Fellowship while this book
was being written
1
A brief history of spectroscopy
The earliest reference to optical spectroscopy that we have in modern times appears
to be the phenomenon of colours in Isaac Newtons Opticks, in which he describes
his famous experiments with prisms and the shaft of sunlight coming through the
hole in his windowshutter. There was much philosophical conjecture at the time but
scientic silence from then on until William Hyde Wollaston (17661828) in 1802,
also in Cambridge, used a lens to focus images of a narrow, sunlight-illuminated
slit through a prism on to a screen. Wollaston appears to have observed the dark
lines across the spectrum transverse to the dispersion direction but ascribed them
to the divisions between the colours. He may be forgiven for this, because with
a single lens the spectral resolution would have been derisory. At about the same
time WilliamHerschel (17381822) discovered the infra-red radiation by the rise in
temperature of the bulb of a thermometer when he held it beyond the red part of the
spectrum in his spectroscope. Joseph von Fraunhofer (17871826) saw more dark
lines but did not guess or deduce their origin. The currently accepted explanation
the absorption of continuous white light by vapours in the atmosphere of the Sun
was given by Gustav Kirchhoff and Robert Bunsen in the University of Heidelberg
who, we may be reasonably certain, passed a collimated beam through their prism
before focusing it, and thereby secured a reasonable resolution. It is probably to
these two but perhaps to Fraunhofer as well that we owe the classical form of the
optical spectroscope: the sequence of collimator, prism and telescope. Little has
changed fundamentally since then.
Developments in geometrical optics kept pace with developments in photogra-
phy. From Lord Rayleigh we have the complete theory of the prism spectroscope
and of the diffraction grating. Comparatively crude but useful diffraction gratings
could be ruled on the bed of any screw-cutting lathe and many individual workers
made gratings for their own use, but without passing on the art. Practical details of
grating construction were given by R. W. Wood of Johns Hopkins University. At
this stage though, they were little more than scientic curiosities emphasising the
1
2 A brief history of spectroscopy
wave nature of light. This was at a time when the corpuscular theory was by no
means dead.
However, diffraction gratings were rst realised at a serious and practical level by
H. A. Rowland, also of Johns Hopkins University, who rst devised ruling engines
capable of the necessary precision.
The prism spectroscope continued in its role as the chief instrument of astro-
nomical botany the taxonomy or classication of stars, that is but the resolution
available then was insufcient to do more than measure wavelengths, identify and
discover elements and to allow wise comments on the mysterious uted spectra
sometimes foundinames as well as incool stars, sodescribedbytheir resemblance,
at low dispersion, to the utings on the columns of classical temples. It was a major
achievement for a spectroscope to be able to resolve the six angstroms separating
the D lines of sodium, and Rayleigh computed that a prismbase length of at least
1 centimetre would be needed for this, that is to say a resolving power of 1000.
Sir Arthur Schuster achieved a resolving power of more than 10 000 at Manch-
ester University with a twelve-prism
1
spectroscope made for himby Cooke of York,
with 30 glassair interfaces (excessive even in a modern coated zoom lens!) and
prism bases of 25.4 mm. With incidences near the Brewster angle there should
have been near perfect transmission for one state of polarisation, but in practice a
transmission coefcient of about 0.05 made it virtually useless except as a sort of
tour de force of the spectroscope-makers art. The way forward in those days was
the manufacture of ever-larger prisms from ever-denser glasses.
The nely ruled diffraction grating by contrast improved the spectral resolution
immeasurably and led to the evolution of the spectrograph at the hands of H.
Ebert, who in 1889 removed chromatic aberration and a great deal more besides by
replacing the lenses of the collimator and the camera with a single concave mirror.
The symmetry was intended to eliminate coma as well as chromatic aberration but
the apparent symmetry is removed by the tilt of the grating and, although reduced,
the coma is still there.
Eberts invention was inexplicably lost until rediscovered by W. G. Fastie
50 years later, by which time the single mirror had been replaced by two sepa-
rate mirrors in the now standard

CzernyTurner conguration.
It was the invention of efcient vacuum pumps which led to the investigation of
gas-discharge phenomena by Crookes and others and the consequent exploration
of gaseous emission line spectra. This led in turn to the empirical discoveries by
Balmer, Paschen and Rydberg which gave rise to the need for an explanation and the
consequent invention of the Bohr model of the atom and the subsequent evolution
of quantum mechanics.
1
In practice six prisms each passed twice.
A brief history of spectroscopy 3
In the twentieth century the spectroscope developed in accordance with the best
detector technology available at the time. Little happened before 1950 in the optical
part of the spectrumalthough infra-red detectors made considerable progress, and at
wavelengths below 0.8 m the photographic emulsion was the universal detector,
the spectroscopic art advancing along with improvements in its sensitivity and
panchromaticity. By c. 1950 the whole electromagnetic spectrum from soft X-
rays to about 1.2 m was available to silver halide emulsions and the short-wave
limit to practical spectrography was set by the reection coefcients of the optical
components, which were and are dismal below about 1200

A. This limitation
was offset by the gradual development of improved sources of XUV light such as
the synchrotron for absorption spectrography and later spectrometry.
Semiconductor detectors improved the accessibility of the infra-red, and the
photomultiplier, which became generally available c. 1950, revolutionised optical
spectroscopy. This device yielded a detective quantum efciency (DQE) in the re-
gion of 0.20.35 as opposed to the DQE of a silver halide crystal of about 0.0005
and, just as importantly, gave a linear response.
2
One of the bugbears of the silver
halide emulsion, especially at the low surface brightness of many astronomical
sources, was reciprocity failure, the inability to respond adequately to long expo-
sures at a low light level. Consequently a monochromator with a photomultiplier
detector, free from the constraints of low focal ratio, could easily out-perform a
spectrograph and moreover could be made to concentrate its efforts on the parts of
the spectrum important to the investigation being made.
Geometrical optical design was then brought to bear, resulting not only in high
efciency but in much greater resolving power, sometimes reaching 8085% of the
theoretical level imposed by the constraints of physical optics.
In parallel with the development of the grating spectrograph, interference spec-
trometry extended the high-resolution end of the art. Rayleigh again observed that
the signal from a Michelson interferometer was the Fourier transform of the spec-
trum of the incoming light but little note was taken of it at the time except as a
scientic curiosity, and it was Charles Fabry and Alfred P erot in the University of
Marseille who in 1899 devised the interferometer which revealed ne and hyperne
structure in spectral lines. Other simultaneous devices, such as the echelon grating,
3
the Lummer plate
4
and the echelle
5
gave equally high resolution but were far less
convenient to use and were not widely copied. In the visible and UV region it is the
2
Based upon the experimental nding that 1000 photoelectrons must be liberated within a microcrystal of
silver bromide, irrespective of its size, to make it developable.
3
A. A. Michelson, Astrophys. J., 8 (1898), 37.
4
O. Lummer, Verh. Deutsch. Phys. Ges., 3 (1901), 85.
5
G. R. Harrison, J. Opt. Soc. Amer., 39 (1949), 522.
4 A brief history of spectroscopy
FabryPerot interferometer in its various guises which still rules the high-efciency
end of the spectrographic art.
Meanwhile high resolution was being achieved by more-or-less heroic means,
with plane-grating instruments of two or three metres focal length and focal ratios of
F,20 or F,30. This ensured at least that the optical resolution would be diffraction
limited and that photographic exposures would take an inordinately long time.
Academic research was simultaneously being carried out using concave gratings of
up to ten metres radius, in spectrographs which lled a whole room in a laboratory
or the whole of the Coud e focus room of a large astronomical telescope. There
appears to have been some confusion at the time between dispersion and resolution,
as though one depended on the other, and dispersion itself was taken as a measure
of quality or performance. With the passage of time has come improved knowledge
and a spectrograph much bigger than one metre or perhaps two in focal length is
now a rarity. Provided that an adequate design is chosen, all the information that
can be extracted from a spectrograph can be revealed with a 1.5-metre instrument.
The period between 1950 and 1970 saw a great surge in activity in the art of
optical spectroscopy. This followed Jacquinots revelation that L-R product or
efciency in a spectrometer, dened as the product of throughput or light-
grasp
6
and resolving power, was a constant and that one quality could be traded
for the other. There was consequently a huge blossoming of ingenious, weird and
wonderful inventions, including the mock-interferometer,
7
the Sisam,
8
the Girard
grille,
9
the M obius-band interferometer,
10
the tilting Michelson interferometer,
11
the lamellar grating interferometer
12
and various chirped and eld-compensated
Michelson Fourier spectrometers.
13
Much time and effort were expended to bring
these ideas to practical use and the two famous Bellevue conferences of 1957 and
1967 brought all these people together for an exchange of ideas.
However, this insubstantial pageant has faded and very few of the devices de-
scribed there have survived the test of timeconveniencecost-benet analysis.
The concept of L-R product is important, indeed fundamental. The idea of re-
solving power, dened as R = , where is the wavelength being observed and
the smallest wavelength separation that can be distinguished as two separate
6
A felicitous phrase from H. A. Gebbie.
7
L. Mertz, N. O. Young &J. Armitage, Optical Instruments and Techniques, K. J. Habell, Ed. (London: Chapman
& Hall, 1962).
8
Spectrom` etre Interf erentiel ` a S election par lAmplitude de Modulation. P. Connes, J. Phys. Rad., 19 (1958),
197.
9
A. Girard, Opt. Acta, 7 (1960), 81.
10
W. H. Steel, Opt. Acta, 11 (1964), 211.
11
R. S. Sternberg & J. F. James, J. Sci. Instrum., 41 (1964), 225.
12
J. D. Strong & G. A. Vanasse, J. Phys. Rad., 19 (1958), 192.
13
L. Mertz, J. Opt. Soc. Am., 49 (1959), iv; P. Bouchareine & P. Connes, J. Phys., 24 (1963), 134; Y. P. Elsworth,
J. F. James & R. S. Sternberg, J. Phys. E, 7 (1974), 813.
A brief history of spectroscopy 5
spectrum lines, is an old, dimensionless quantity and a useful parameter of quality
in a spectroscope. Light grasp on the other hand, also known as luminosit e or
etendue and generally denoted by the letter L, is less familiar and is a measure of
the rate at which an optical instrument can process the power it receives from the
incident electromagnetic eld. It can be dened as the product of detector sensitive
area and the solid angle subtended by the detector at the exit pupil of the system
14
or alternatively as the product of the area of the exit pupil and the solid angle sub-
tended by the detector sensitive area. Both of these denitions are derived from the
HelmholtzLagrange invariants of the system. The constancy of the product can be
envisaged easily by referring to a simple prism spectroscope. If the entry-slit width
is doubled, twice as much light passes through but the resolution is halved.
The concept is much more universal than this and it became apparent that high-
resolution devices such as the FabryPerot interferometer, possessing a much higher
L-R product than a grating spectrometer, could be used at moderate resolution
to examine low-luminosity sources. Soon afterwards the multiplex or Fellgett
15
(so named after its discoverer) advantage made Fourier-transform spectroscopy
with a Michelson interferometer a highly desirable goal despite its formidable
technological problems now happily solved.
For more than 30 years the monochromator in one formor another ruled supreme
until it was rudely overthrown by the arrival of the charge-coupled device, the
CCD. When cooled to the temperature of liquid nitrogen the CCD has a detec-
tive quantum efciency of 0.5 or more and it is, in effect, a photographic plate
with an ISO rating in the region of 3 000 000 and a resolution of anything up to
100 line-pairs/mm.
16
The other improvements which it brought are legion: the lack
of reciprocity failure, the enormous dynamic range in the region of 50 000,
17
the
virtually instant read-out on to a computer screen, and so on. To develop its full
potential in spectrography it has been necessary to return to geometric optics in
spectrograph design, as there is a strict requirement for a at eld, and a distinct
advantage in a low focal ratio; and there are modern advances which would have
beneted the Victorian spectrograph had the problem been tackled at the time.
14
Which for F-numbers 1.5 is related to the focal ratio F by O = ,4F
2
.
15
P. Fellgett, J. de Phys., 19 (1958), 187, 237.
16
As compared with the fastest and consequently coarsest silver halide emulsions with ISO ratings in the region
of 2000 and 50 line-pairs/mm.
17
Silver halide has a dynamic range of about 100: it saturates at 100 times the minimum exposure needed to
produce blackening.
2
The relevant regions of the electromagnetic spectrum
The extent of the electromagnetic spectrumis too well known to require description
here. We shall be chiey concerned with the so-called optical region, of which
the boundaries are determined partly by the methods of detection and partly by the
methods of dispersing and analysing the radiation. What is common to all parts of
the region is the type of optical element and materials of construction of spectro-
scopic instruments. Beyond the region on the long-wave side, coherent detectors,
paraboloidal aerials, waveguides and dipole arrays are used, and on the short-wave
(X-ray) side, optical elements other than diffracting crystals and grazing incidence
reectors are generally unknown. Radiation of 100

A wavelength liberates photo-
electrons with more than 100 eV of energy and the appropriate detection methods
are those of radiography and nuclear physics.
Broadly we can identify six wavelength divisions appropriate to optical design
techniques:
r
5015 m. The far infra-red (FIR), where bolometric, superconductor and semiconduc-
tor detectors are the chief methods of detection and measurement, and only specialised
materials such as selenium, thallium bromide and various polymer resins such as sul-
phones have the necessary transparency to make useful refracting components. Optical
elements may well be polymer plastics of high dielectric constant. Reecting elements
are coated with gold. New FIR-transparent materials with desirable optical properties are
appearing all the time. Nevertheless, the principles of design which make use of them
are unchanging and it is important to pay attention to their refractive index and partial
dispersion or :-value when deciding whether to use them.
r
151.2 m, 0.11 eV. The near infra-red, where there are many suitable refracting ma-
terials, and where semiconductor, superconductor and bolometric detectors prevail. This
region includes the two windows, the 24 m band and the 813 m band where the
atmosphere is transparent. The opacity does not much concern laboratory work, but is im-
portant if aeronomy, telecommunications or astronomical research is in view. The edges
6
The relevant regions of the electromagnetic spectrum 7
of the regions, where absorption is slight but noticeable, are particularly difcult regions
to work in.
r
1.20.35 m, 12 0003500

A, 13.5 eV. The near infra-red, visible and near ultra-violet
(UVOIR)
1
region where both semiconductors, photoconductors, photochemical reactions
and photoemitters can be used, the latter directly as so-called photon-counters (in re-
ality photoelectron counters). This, not unnaturally, is the best understood region of the
spectrum and where the greatest variety of refracting materials are known. It is also the
region between 1 eV and 3 eV of energy, where interesting things happen in chemistry
and physics, where atoms and molecules display their properties, where chemical reac-
tions take place. At wavelengths above 8000

A atmospheric transmission is patchy with
extensive regions of opacity caused by water-vapour and oxygen molecules. Laboratory
work is unaffected but aeronomers, astronomers and surveyors may have problems when
they need long path lengths through the atmosphere.
r
0.350.20 m, 35002000

A, 3.56 eV. The far ultra-violet, where optical components
must be selected carefully for adequate transmission
2
and phosphors are required to
convert the radiation into longer wavelengths for registration by optical detectors. Again
laboratory work is unaffected but the ozone layer of the atmosphere at 45 km effectively
absorbs all extra-terrestrial UV radiation below 2900

A. Laboratory sources emitting
these wavelengths (a quartz-envelope mercury vapour lamp for example, which emits the
2537

A Hg
1
S
0

3
P
1
resonance line) are harmful and must be shielded, together with any
surfaces which may reect that radiation into the surrounding laboratory. Such sources
may also stimulate the production of ozone (O
3
) which has a characteristic electric smell
and has well-known corrosive properties on some materials such as rubber.
r
0.20.12 m, 20001200

A, 610 eV. The vacuum ultra-violet, where atmospheric gases
absorb the radiation (O
2
below 2424

A and N
2
below 1270

A), where most materials
have lost their transparency and where the reecting powers of metals are beginning to
decline. Optical instruments for this region must be in a vacuum chamber or in a non-
absorbing atmosphere such as helium, which is transparent down to 911

A, and other rare
gases similarly down to their respective ionisation potentials. It is an interesting and not
yet thoroughly explored region where autoionisation series spectra can be detected with
their concomitant superallowed transitions and broad line absorption spectra. Filling
the spectrograph with helium does not ease the vacuum problem so far as leak prevention
is concerned but it may ease the stress on thin-lm windows.
r
0.120.01 m, 1200100

A, 10100 eV. The far- or extreme-vacuum ultra-violet, some-
times called the XUV, where radiation is ionising, where there are no transparent ma-
terials except in the form of thin (12 m) lms, where reection coefcients are down
to 0.10.2 and apart from synchrotron radiation there are no reliable, steady sources of
radiation for absorption spectroscopy.
1
Ultra-Violet-Optical-Infra-Red.
2
Most optical glasses are opaque below 3500

A.
8 The relevant regions of the electromagnetic spectrum
Gas molecules have absorption cross-sections measured in megabarns (1 Mb =
10
18
cm
2
) and this, in an optical path of 1 m, implies a need for a vacuum better
than 10
4
mm Hg, which in turn requires a diffusion pump or, if especially pure samples
are to be examined, a turbo-molecular pump. This is probably the most difcult region of
the spectrum for which to design spectrographs and in which to do research. The vacuum
requirements are demanding and not to be undertaken lightly.
At wavelengths below 100

A materials start to become transparent again
3
and this
is the beginning of the soft X-ray region where different techniques supervene.
In each of these regions of the spectrum the optical design requirements have
much in common. There are exceptions.
r
In the infra-red, alternative techniques such as Fourier-transform spectroscopy, usually
with a Michelson interferometer,
4
are possible, are advantageous and are capable of enor-
mous resolving power. Awarning, however: the method lacks spectrometric simultaneity.
A ickering source can have catastrophic effects on the spectrum unless rapid sawtooth
scanning is used at a frequency well above the highest icker frequency or below the
lowest. In commercially available Fourier spectrometers this rapid scanning is unusual.
Fourier spectroscopy has no advantage elsewhere than in the infra-red.
r
In the UVOIR region the plane diffraction grating is the radiation analyser of choice
although there are circumstances where a prism has advantages. The FabryPerot inter-
ference spectrograph, too, has its place where high resolution is needed or where very
low light levels are encountered.
r
In the vacuum ultra-violet between 2000

A and 1200

A prism and lens materials such
as calcite and lithium uoride are available with the required transmission but generally
the concave grating possesses advantages which increase rapidly as the wavelength of
interest decreases. At wavelengths below 2000

A the concave grating is the analyser of
choice. Below 1200

A there is no realistic alternative.
So far as refracting materials are concerned for these various regions, technology
is advancing all the time, especially in the infra-red and the Internet should be
consulted for details of their transmitting and refracting properties and for the
names of their manufacturers.
2.1 The limits of optical spectrography
An optical spectrograph or spectrometer of one sort or another can be made to give
a resolution up to R 10
7
without a struggle. It is difcult to imagine a practical
application for a conventional spectrograph with a greater resolving power than
3
Or the radiation more penetrating!
4
But see Section 7.6.
2.1 The limits of optical spectrography 9
this.
5
A FabryPerot etalon with a 3-metre spacer and a nesse of 50 will give
4.10
8
at optical wavelengths, enoughtoreveal the natural widths of allowedatomic
transitions, providedof course that youcanrst eliminate all the temperature effects.
Extraordinary precautions are needed to ensure stability (in the source as well as the
instrument) since the etendue is tiny and the exposure time long. It has the merit of
being absolute, since the various quantities involved are directly measurable with
a metre-stick. But at this sort of resolving power, more sophisticated spectroscopic
techniques are available and preferable.
5
The measurement of small Doppler shifts, for example, does not require high resolving power, only high spatial
resolution at the focal plane. A wavelength shift of a small fraction of a resolved element is measurable.
3
Geometrical optics
This is the branch of optics which deals with image-forming instruments, includ-
ing of course spectrographs and interferometers. Such instruments employ lenses,
mirrors and prisms and it is the art of combining these elements to make useful
devices which is the subject of the next two chapters.
3.1 Rays and wavefronts
Image-forming instruments are intended to project images of real objects on a
screen or focal surface usually though not always plane, and instrumental optics
is the study of ways of doing this.
The subject comprises two chief parts: instrument design and lens design.
The former is the design of instruments so that light is conveyed through their
optical components from one to another to arrive eventually at the focal surface.
The technique is essentially a graphical one, using a drawing board or a computer
drafting program to lay out the components at their proper places.
The latter involves the accurate tracing of rays through various optical sur-
faces of different types of glass, and reections from mirrors of various shapes
to achieve a correction of the optical aberrations and to ensure that light of all
wavelengths from a point on an object is focused to a corresponding point on its
image.
Traditionally this was done by ray tracing, the accurate computation of ray paths
using seven-gure logarithms and trigonometrical tables, but now is done chiey
by iterative ray tracing in a small computer, using a program specically designed
for that purpose.
A word of warning: an optical system such as a spectrograph cannot be designed
on a ray-tracing program. The proper positions for the various elements must be
found by means of the drawing board or the CADprogram. An iterative ray-tracing
programcan then be used to optimise the performance of the instrument by altering
10
3.2 Instrumental optics 11
the positions of the optical elements, the curvatures and tilts of refracting and
reecting surfaces and their excentricities.
3.2 Instrumental optics
High-school teachingis that light is emittedfroma luminous point, travels instraight
lines outwards called rays, and that these rays can be caught by a lens and bent
to form bundles which then converge to a point. If a screen is placed at the proper
place, a point can be seen, the image of the original luminous object.
Light rays are a convenient ction of course like the lines of force in an electric
or magnetic eld, or the streamlines of uid mechanics. Following Huygens
principle, we postulate that light is emitted as wavefronts (which are also known as
surfaces of constant phase) which spread outwards spherically from each point
of an existing wavefront so that their envelope forms a new wavefront. They are
slowed down when they pass through a refracting transparent mediumsuch as glass
or water. A spherically expanding wavefront from a point source can be converted
by a properly shaped agglomeration of lenses to a spherical converging wavefront,
which converges to form a point image of the point object. The point source is
another convenient ction.
Other necessary concepts are:
r
Refractive index This is the ratio of the speed of light in vacuo divided by the speed
of light in a refracting medium.
1
Refractive index is generally measured indirectly by
observing the deviation of a ray as it passes through a prism for example.
r
Geometrical path This is the actual physical distance between a source of light and
the place where the light falls on a screen and is observed. This is the distance which you
measure with a ruler on a ray diagram on a drawing board.
r
Optical path This is the geometrical path multiplied by the local refractive index.
Usually it is the sum of several sections where the light passes through different media,
and if the refractive index, n, varies continuously it is the integral
_
b
a
n(s) ds
along the path of the ray.
Alternatively and this is useful it is the number of wavelengths of monochromatic
light along the path, multiplied by the vacuum wavelength of the light.
The basic principle of image formation in an optical instrument is this: that the paths
of the rays from a point on an object to its conjugate point on the image should all have
the same optical path length.
1
I use the word speed here, since the speed, unlike the velocity, can be measured directly by ordinary physical
experiments.
12 Geometrical optics

r
u, 0 0, 0 R, 0 v, 0
h
i
Figure 3.1 The basic diagram for ray tracing. It shows the various quantities in-
volved in tracing a ray through a surface, using Snells law to relate the angles i
and r.
r
Reducedpath This is the geometrical path divided by the refractive index. For example,
if you look into a swimming pool it appears to be shallower than it is. What you see or
would measure with a camera rangender for example is the reduced path to the bottom
of the pool.
All these are useful ideas in the eld of optical design.
3.3 Centred systems
Optical image-forming instruments such as telescopes and microscopes generally
have optical trains of spherical surfaces with the centres all lying on a straight
line, the optic axis of the system. The use of spherical surfaces derives from the
physical fact that these are the only surfaces which can be ground and polished
with high accuracy with comparatively simple apparatus. One piece of glass ground
against another with a suitable abrasive automatically produces a common spherical
interface. To achieve other shapes, such as paraboloids, more elaborate methods
are required but with contemporary technology are achieved at small extra cost.
3.4 Gaussian optics
The elementary theory of image formation is due to Carl Friedrich Gauss (1777
1855) and makes the assumption that all the angles between rays and the optic axis
are small, so that the approximations sin and cos 1 are good.
By convention, optical diagrams are drawn with light entering the system from
the left. Suppose (Fig. 3.1) that there are two transparent media of refractive indices
n
1
and n
2
, and a spherical interface between them of radius R with its vertex at the
origin of a Cartesian coordinate system and the centre of the sphere at the point
C(R. 0). Suppose that a ray of light starts from a point on the optic axis A(u. 0)
and reaches the interface at D where it is refracted according to Snells law and
continues in a new direction to meet the optic axis again at B(:. 0). Draw the rays
from C to D to B. Angles are measured clockwise from the optic axis to the ray.
3.4 Gaussian optics 13
Then,
h,u = . h,: = . h,R = .
and by ordinary geometry,
r = . i = +.
and with Snells law,
n
1
_
1
R

h
u
_
= n
2
_
1
R

h
:
_
.
giving
n
1
u

n
2
:
=
n
1
n
2
R
.
or, more usefully,
n
1
_
1
R

1
u
_
= n
2
_
1
R

1
:
_
.
and this is a fundamental formula connecting the object distance u, the image
distance :, the radius of curvature R of the surface and the two refractive indices
n
1
and n
2
.
3.4.1 The thin lens
We next apply this result to two surfaces in succession, in other words, to a simple
lens. The image distance after refraction by the rst surface becomes the object
distance for refraction at the second. The distance between the two vertices is taken
as negligible. Then, with sufxes for the two radii and ticks for the distances from
the second surface,
n
1
_
1
R
1

1
u
_
= n
2
_
1
R
1

1
:
_
.
n
2
_
1
R
2

1
u

_
= n
1
_
1
R
2

1
:

_
.
and u

, the object distance for the second surface, equals :, the image distance for
the rst. Then, as before,
n
1
_
1
R
1

1
u
_
= n
2
_
1
R
1

1
:
_
.
14 Geometrical optics
and for the second surface,
n
2
_
1
R
2

1
:
_
= n
1
_
1
R
2

1
:

_
.
Eliminating : between these and adding gives
n
1
R
1

n
1
u

n
2
R
1
+
n
2
R
2

n
1
R
2
+
n
1
:

= 0.
and putting, as usual for lenses in air, n
2
= n and n
1
= 1, the result is
1
:

1
u
= (n 1)
_
1
R
1

1
R
2
_
.
An obvious and probably familiar corollary follows if we put the object distance
equal to , for then
(n 1)
_
1
R
1

1
R
2
_
=
1
f
.
where f is the focal length of the lens. The plane perpendicular to the optic axis
where the image is formed is called the focal plane of the lens. There are obviously
two focal planes, one either side of the lens itself, at distances f and f from the
vertex. Where they cross the optic axis are the two foci or focal points of the lens.
The other familiar equation follows:
1
:

1
u
=
1
f
.
This is the Gaussian lens equation.
NB Distances to the left (object side) of the lens are negative and those to
the right (image side) are positive. This is the so-called sign convention; one of
many which plague geometrical optics, but the one which appears to work best and
which is used throughout here. It is known generally as the Cartesian convention.
Another part of the convention is that surface curvatures are positive if the surface
is convex to the incoming light, that is, if the centre of the sphere lies to the right
of the vertex of the surface.
Notice that in general the optic axis can be dened as the line joining the two
centres of the spheres which formthe lens surfaces. In practice one tries to construct
the instrument so that all the sphere centres lie on one line. This calls for some
subtlety in manufacture and the accuracy with which a given lens element has its
optic axis at the mechanical centre of its (circular) rim is called the centration.
Design note It is much more convenient when doing calculations to use cur-
vatures and powers rather than radii and focal lengths. The Gaussian equation is
3.4 Gaussian optics 15
d
f
Figure 3.2 The geometry of a thick lens. The position of a virtual equivalent thin
lens is one of the two principal planes which denes the position of refraction of a
ray by any lens or system of lenses. The distance from this plane to the focal point
is the focal length of the lens.
then
P = (n 1)(c
1
c
2
) = (n 1)Lc.
where P = 1,f and is measured by opticians in reciprocal metres, or dioptres.
3.4.2 Thick lenses
We assumed above that the vertices of the two surfaces were coincident. In practice
there is a distance d between them, where d is small compared with the object and
image distances.
Repeating the argument, but with u

= : d and some tiresome algebra we nd

1
f
= (n 1)
_
1
R
1

1
R
2
+
(n 1)
2
n
d
R
1
R
2
_
.
This is sometimes called the lensmakers equation.
Confusion often arises when concave and convex mirrors are included. Acertain
way round the difculty is to treat every optical component, lens or mirror, as if it
were a lens, and to lay out the system on the drawing board as if it were a totally
refracting one. Then a mirror concave to the left behaves like a lens with a positive
focal length f and f = R,2 where R is the radius of the mirror. To be pedantic,
the mirror can be treated as a lens of refractive index 1 and a at rst surface.
An important tip is to remember that the angle through which a ray is deviated
on passage through a lens depends only on the height of incidence and the focal
length.
3.4.3 Aplanatic surfaces and lenses
The aplanatic points U and U

of a surface are related by the fact that any ray

through the surface directed towards U will arrive at U

after refraction. Their

16 Geometrical optics
h

= h/f
Figure 3.3 The refraction angle of a ray entering a lens depends in the Gaussian
approximation only on the height of incidence and the focal length.
R
R(1 + n)
R(1 + n)
n
Figure 3.4 The converging aplanatic lens. The object is always virtual. Both object
point and image point are at nite distances from the rst surface vertex and are
related to each other as in the diagram by the refractive index. The second surface
position is not critical but it must be concentric with the paraxial image point.
Perfect imaging holds for any incident ray directed initially towards the object
point. The imaging is virtually aplanatic for a nite eld surrounding the paraxial
image point.
distances from the vertex of the surface are easily shown
2
to be related by
u = R(1 +n) : = R(1 +n),n. (3.1)
An image so formed in the vicinity of U

from a virtual object in the vicinity of U

will thus be free from spherical aberration, coma and astigmatism.
Aconverging lens can be made with these properties provided the second surface
is concentric with the image. The thickness of the lens is not of prime importance
and such a lens is called an aplanat. In spectroscopy its chief use is to increase the
2
A simple exercise for the reader, but failing this, see for example R. Kingslake, Lens Design Fundamentals
(New York: Academic Press, 1978).
3.4 Gaussian optics 17
d
f x
1
Figure 3.5 The geometry of a pair of thin lenses showing the position of a single
equivalent thin lens at the appropriate principal plane.
convergence of a ray bundle so that it can be made to fall on to a smaller detector,
thus reducing the detector noise which is area dependent.
3.4.4 Systems of two lenses
Consider two thin lenses of focal lengths f
1
and f
2
lying on the same axis and
separated by a distance d (Fig. 3.5). It is usual to refer to the whole system as a
lens and to the two components as elements.
A ray coming from parallel to the optic axis at a height h
1
above it suffers
a deviation
1
= h
1
,f
1
at the rst element and
2
= h
2
,f
2
at the second. This
amounts to a total deviation
1
+
2
, so that the focal length of the combination is
F = h
1
,(
1
+
2
).
With h
2
= h
1

1
d the hs and the s can be eliminated to give the focal length
of the whole compound lens as
1
F
=
1
f
1
+
1
f
2

d
f
1
f
2
.
or, more usefully,
F =
_
f
1
f
2
f
1
+ f
2
d
_
.
For practical purposes we need to know more than this. The usual question is
where should one put a single thin lens, equivalent to the combination, which will
produce an image at the same position and with the same focal length? In other
words, where should a single lens of focal length F be placed to secure an image
at V?
18 Geometrical optics
A
f
2
P
2
P
1
f
1
f f
A
Figure 3.6 The principal planes andfocal planes of a systemof lenses. Onlythe rst
and last surfaces are shown and there may be any number of surfaces in between.
Any such system of lenses possesses such a pair of principal planes which serve
to determine the paths of rays traced through the system in either direction. Actual
ray paths are drawn in full, construction paths are shown interrupted.
3.4.5 Principal planes
Suppose that the proper position is at a distance x
1
to the right of the rst element.
Then (Fig. 3.5)
d x
1
=
h
1
h
2

1
+
2
=

1
d

1
+
2
=
Fd
f
1
.
so that
x
1
= d(1 F,f
1
) =
d( f
1
d)
f
1
+ f
2
d
.
and the simple equivalent lens must be placed at a distance x to the right of the rst
element to obtain the same-sized image of an object at at the same place.
If the light were coming from the right:
r
the focal length of the compound lens would be the same;
r
the position of the simple equivalent lens would be
x
2
=
d( f
2
d)
f
1
+ f
2
d
to the left of the second element.
These two positions are calledthe principal planes of the compoundlens. Where
they cross the optic axis are the principal points of the lens.
It can be shown that any centred system of lenses and mirrors, however compli-
cated, has two principal planes and two focal planes. They can be dened as follows
(Fig. 3.6):
A ray, parallel to the optic axis, entering the lens system at P
2
emerges at P
1
at
the same height above the axis and passes through the focus F
1
.
Similarly a ray fromthe right entering the systemat P
1
emerges at P
2
at the same
height and passes through F
2
.
3.4 Gaussian optics 19
f
x
1
(a)
f
x
1 (b)
Figure 3.7 Two examples of lenses with principal planes outside the lens itself.
(a) Represents a telephoto lens and (b) is a combination of positive lenses with a
negative focal length, albeit giving a real erect image of a point at . The single
equivalent lens would be at a distance x
1
to the right of the rst vertex, and would
be a negative lens giving a virtual image a distance f to the left of x
1
.
3.4.6 Nodal planes
Two further points and planes can be dened, although for most purposes they are
coincident with the principal planes. They are the nodal planes. There are several
denitions, but a simple and practical one is as follows:
A ray from the left, crossing the optic axis at the nodal point n
2
at an angle
to the optic axis, emerges from the other nodal point n
1
at the same angle. The
nodal points coincide with the principal points unless the initial and nal refractive
indices are different. This circumstance might happen if a lens were designed to
work with one end in water or oil for example.
For a single-element thick lens the principal planes are inside the glass at a dis-
tance x
i
= f [(n 1),n](d,R
i
) from the surface i (where i = 1 or 2, obviously).
In practice x
i
d,n.
The principal planes are not necessarily between the elements nor even near the
lens itself, as the examples in Fig. 3.7 show. A telescope in proper adjustment, for
example, has no focal length and its principal planes are at .
The principal planes are also called the planes of unit magnication. If some
other optical system to the left produces an image at the P
2
plane, the image will
be transferred with the same size to the P
1
plane and will appear to be there to any
subsequent system.
20 Geometrical optics
3.5 Optical layout
Once the positions of the focal planes and principal planes have been established,
rays can be drawn from an object point, through the system, to the image point.
The rules are:
r
A ray from a point A on the object, parallel to the axis, entering at any height h meets
the principal plane P
2
and emerges from P
1
at this same height and at an angle to make
it pass through the focal point f
1
.
r
If a ray from the same object point, drawn to the nodal (or, in practice, principal) point
of P
2
makes an angle with the optic axis it emerges from the nodal (principal) point of
P
1
at the same angle and continues until it meets the other, previously traced ray, at a
point A

.
Where they meet is the image of the object point A. The two points A and A

are said to be conjugate.

This construction is sufcient to establish the position of the image of any object
and is generally the rst step to be taken when designing a centred optical system.
3.6 Apertures, stops, elds, irises and pupils
3.6.1 The aperture
In a simple lens this is a self-evident concept. It can refer to the clear transmitting
area of the lens or to the diameter if the lens has a circular rim. In a compound
camera lens it is the diameter of the biggest parallel bundle of rays, all parallel to
the optic axis, which is transmitted through to form an image. This may well be
smaller than the aperture of the rst element. This aperture may also be controlled
by an iris diaphragm somewhere on the optic axis (though not necessarily inside
the system), which then forms an aperture stop.
3.6.2 Numerical aperture
This is dened as NA = n sin
o
, where is the angle that the marginal ray makes
with the optic axis, and is a measure of the intensity with which light can be
focused on to an image. A similar and more familiar concept is the F-number or
focal ratio, much used in describing camera lenses. The F-number of a lens is given
by F = 1,2 tan and for small values of is the reciprocal of twice the numerical
aperture.
3.6 Apertures, stops, elds, irises and pupils 21
Figure 3.8 This shows the passage of a beam through a compound lens with air-
spaced elements with and without an iris to control the ray bundles. The stop
ensures a uniformly illuminated eld and, as we shall see later, can be used to
control the aberrations of the lens.
3.6.3 The stop
This is an opaque mask, generally perpendicular to the optic axis, in which there is a
(usually) circular hole through which light can pass. Stops can be found anywhere
in a centred system and may be a useful way of controlling the aberrations and
hence the image quality that is produced. A telescope, for example, may well have
a eld stop placed at the position of an intermediate image just in front of the
eyepiece perhaps where it reduces scattered light, increases the image contrast
and produces a sharp edge to the eld of view.
3.6.4 The eld
In general the eld of an imaging system is considered to be an angle. A telescope
may be said to have a eld of 3 degrees, meaning that parallel bundles of rays may
pass through the telescope at angles up to 1.5 degrees fromthe optic axis. Sometimes
the eld semi-diameter, or the eld semi-angle is quoted. In this book, the eld
angle or obliquity of a parallel ray bundle will be denoted by .
3.6.5 The iris and the pupils of a system
These are dened as follows:
r
The iris There must be one stop which limits and denes the diameter of all the ray
bundles that pass through the system.
r
The entry pupil is the image, seen from the left (the side of the incoming light), of the
iris of the system.
3
Optical elements of the systemmay forma real image of the iris inside
3
Probably virtual, but real if the iris is in front of the rst element, when the iris itself is the pupil.
22 Geometrical optics
aperture
and entry pupil
objective glass
field semi-angle
field stop
field lens
eye-relief
eye lens
exit pupil

Figure 3.9 Arefracting telescope. This is a complete optical instrument illustrating

the concepts in this section. At the prime focus is a eld stop to reduce scattered
light and to dene the edge of the eld. The eld lens follows and its purpose is
to image the objective lens on to the exit pupil where the pupil of the eye will be
located. The third lens is the eye lens and it recollimates the light emerging from
the focal plane so that the image appears to the observers eye to be at .
the system and at this image point there may be another iris, conjugate to the rst. Either
or both of these diaphragms may be regarded as the iris.
r
The exit pupil is similarly dened as the image, real or virtual, viewed from the right of
the iris.
r
The vignette This occurs when there is no single dening stop controlling all the ray
bundles that are to be transmitted. Figure 3.6 shows an example of this, where the two
lens rims each control part of the bundle. The term was originally applied in the Victorian
age to tasteful portrait photographs in which there was no sharp edge, but a general fading
of the scene near the rim. It is highly undesirable in optical instruments and to prevent the
resulting uneven illumination in the image, a stop is placed between the elements so that
all ray bundles have the same size and shape. Nearly all optical instruments have such a
dening stop.
4
A refracting telescope (or a binocular), has the objective lens itself as both entry
pupil and iris and the exit pupil is the image, usually about 2 cm to the right of the
eye lens (the last lens of the system), of the objective. The diameter of this image
is the diameter of the objective divided by the magnication of the telescope. It is
at this point that the pupil of the eye must be placed if the whole eld of view is to
be seen.
This is an example of an important rule:
If two optical systems are to be coupled together the exit pupil of the rst must
be the entry pupil of the second. If both pupils are virtual, a relay lens must be
inserted to image each on to the other.
4
The Michelson Fourier-transform spectrometer is an exception.
3.8 The HelmholtzLagrange invariants 23
3.7 Ray bundles
These are collections of rays which diverge froma point on the object, pass through
different parts of the optical system and converge to form a point on the image.
Their common property is that every ray of the bundle should have the same optical
path length: if not, they interfere with each other. In practice they are not the same
and a good optical image is formed if the differences are less than a couple of
wavelengths or so. If you look with a microscope at an image in monochromatic
light you can sometimes see the interference fringes.
3.7.1 Chief-rays
Every ray bundle has a central ray, the chief-ray, dened as follows:
The chief-ray of any bundle crosses the optic axis at the iris of the system. It
follows that it also crosses the optic axis at the pupils provided they are real images
of the iris, as for example in a telescope eyepiece.
The chief-ray serves to dene the direction in which the bundle is propagating.
If, for example, an instrument is to form a monochromatic image of a source using
an interference lter, all the chief-rays must be parallel to the optic axis where they
pass through the lter, since the wavelength transmitted by an interference lter
depends on the angle of incidence of the light on it. This can be done in a telescope,
for example, by placing at the focus of the objective a second lens, of the same
focal length as the objective and the same diameter as the eld of view desired.
Since the objective lens of the telescope is the entry pupil, the chief-rays of all the
ray bundles pass through the centre of the objective lens and are then collimated
(made parallel) by this second lens placed at (or near, in practice) its focus. The
interference lter follows, and then the eyepiece, camera or whatever is supposed
to follow.
There are two precepts to bear in mind when designing an optical system:
r
There is no such thing as a point source of light.
r
There is no such thing as a parallel beam of light.
Both of these ideas are useful ctions when discussing Gaussian and geometrical
optics, but in the study of photometry we must part company with them.
3.8 The HelmholtzLagrange invariants
It can be demonstrated graphically (Fig. 3.10) that the linear size of an image
multiplied by the tangent of the angle subtended by the preceding and following
24 Geometrical optics
h
1
h
2
h
3

1
Figure 3.10 The HelmholtzLagrange invariant. It is a matter of elementary geo-
metry that the product h
i

i
is conserved throughout the system and there is a
similar product in the plane through the optic axis perpendicular to this diagram.
In practice the two invariants combine to give the law of conservation of etendue.
optical elements is a constant through a given centred system. To be precise,
n
1
h
1
tan
1
= n
2
h
2
tan
2
= n
3
h
3
tan
3
= = L.
where n
1
is the refractive index of the medium in the appropriate section. Usually
n = 1.
If the eld is anamorphic as in a wide-screen cine-projector for example
there are two such invariants, perpendicular to each other, and each including the
optic axis. The concept can be extended as follows:
The area of the eld, multiplied by the solid angle subtended there by the
pupil following it and by the square of the intervening refractive index, is a
constant of the system.
It can alternatively be dened as the area of the pupil multiplied by the solid
angle subtended by the preceding or following eld-stop.
This constant is called the etendue of the system. It is a particularly valuable
quantity to know when designing a photometer, since it immediately allows one to
calculate the ux of light that will be accepted.
In practice it is usually sufcient to dene it as
E =
(area of pupil) (area of eld)
(distance from pupil to eld stop)
2
.
and it is generally quoted as cm
2
sterad.
Etendue is of particular importance when applied to spectrometers. It is also
known as luminosit e and generally denoted by the letter L. In a spectroscope for
example it is easy to see that, when looking at a line emission spectrum, if the slit
width is doubled the amount of light entering the eye is doubled. At the same time
the resolving power the ability to see two lines as separate is halved. We shall
see later that if resolving power is dened as ,d where d is proportional to
the width of the slit, then the product, LR, of etendue and resolving power is a
constant for a particular design and E = LR is dened as the efciency of the
instrument.
3.10 Black body radiation 25
3.9 Surface brightness
The total power radiated by a source depends on its area and its surface brightness.
This is dened in two ways. It is either:
(1) the power emitted per unit area per steradian normal to the surface; or
(2) the photon ux emitted per unit area per steradian normal to the surface.
It can be measured in watts per unit area per steradian, or photons per second per
unit area per steradian. If the source is a Lambert radiator (it usually is not) then
the total power radiated into 2 steradians is I S, where I is the power per unit
area per unit solid angle normal to the surface, and S is the area of the surface.
This follows from Lamberts law.
3.9.1 Lamberts law
This states that the power per unit solid angle emitted by a luminous surface is
proportional to the cosine of the angle which the radiation makes with the surface
normal.
Seen from a distance, a tilted surface of nite area S appears foreshortened to
an area equal to S cos , and by Lamberts law, the radiation is similarly curtailed
in intensity.
Thus the surface brightness of a perfect Lambert radiator is independent of the
direction from which it is viewed. To a distant observer, for example, a radiating
spherical surface will be indistinguishable froma uniformly illuminated plane disc.
The idea is convenient geometrically, but in nature and in the laboratory Lambert
radiators are rare. The Sun, for example, is not a uniformly bright disc: its apparent
surface brightness at the edge is less than half that at the centre, and moreover is
wavelength dependent.
A perfect optical system will convert rays from a point source into a parallel
beam. Since the power radiated by a source is proportional to its area, there will
thus be no power in a parallel beam. The power will depend on the divergence of
the beam, which in turn is measured by the angular diameter of the source at the
entry pupil of the system.
3.10 Black body radiation
The most important practical consequence of Lamberts law is that if a Lambert
surface emits a total of I photons per second per unit area, then the apparent surface
brightness is I , photons per second per unit area per steradian. A black body for
example, emits T
4
watts per unit area into 2 steradians. is Stefans constant
26 Geometrical optics
and T the absolute temperature. The surface brightness normal to the surface is
then T
4
, watts per unit area per sterad.
5
3.10.1 Spectral power density
The total power emitted will depend on the nature of the emitting source. A black
body radiates power as a function of wavelength according to the well-known
Planck formula:
I () =
2hc
2

5
_
1
e
hc,kT
1
_
. (3.2)
and this can be converted to a formula for the photon ux:
N() =
2c

4
_
1
e
hc,kT
1
_
. (3.3)
The units in each case are watts (or photons per second) per unit area per unit
wavelength range radiated into 2 steradians. N or I must be divided by to give
the spectral surface brightness.
Most sources do not radiate as black bodies. A polished reecting surface has
a low emissivity and neither emits nor absorbs radiation efciently. Its emissivity
may be a function of wavelength, e(), which must multiply the Planck formula
above.
The important property of surface brightness is that when multiplied by the
etendue of the photometric instrument observing it, the result is the power or photon
ux entering the detector.
Remember that the units of etendue are (area) (solid angle) and that, multiplied
by the units of surface brightness (ux per unit area per unit solid angle), yields
ux.
These are the twofactors involvedincomputingthe yieldof anoptical photometer
or spectrophotometer with a given source.
Worked example Consider a photometer with a detector area of 1.5 cm
2
at the
focus of a lens 3 cm diameter and focal length 10 cm. Its etendue is (3,2)
2

1.5,100 = 0.106 cm
2
sterad. If it observes a distant extended source of luminosity
10
10
photons cm
2
s
1
sterad
1
it will receive radiationat the rate of 10
10
0.106 =
1.06 10
9
photons s
1
.
Notice that the answer does not depend on the distance of the photometer from
the source. In other words, if you sit facing a blank, uniformly illuminated wall
you cannot tell how far you are from it: its surface brightness does not change as
5
Not T
4
,2, because of the cosine factor.
3.10 Black body radiation 27
you move away. Each cell of the retina is a separate detector with an etendue in the
region of 10
6
cm
2
sterad. At a greater distance it will receive less light from unit
area of the wall, but there will be a correspondingly greater area of the wall imaged
on to each cell. The total light received by each cell is then unchanged.
However, when light arrives at a telescope from an apparently point object such
as a star it is the aperture of the telescope which determines the signal at the
detector, provided all the light from the object arrives at the same detector. The
noise generated by the detector depends on the sensitive area
6
and it is usually
desirable to have as small a detector area as possible. The numerical aperture or
focal ratio of the telescope is then important, as the diameter of the diffraction-
limited image is 1.22F where F is the focal ratio. Even when the size of the
image is determined by seeing to an angular size of about 1

of arc, the focal ratio

still determines the diameter of the image at the detector. This is the point at which
an aplanatic lens becomes important as a way of increasing the numerical aperture
while retaining the spatial resolution.
6
It is generally proportional to the square root of the sensitive area in a photoconductor, a bolometer or a photon
counter such as a photomultiplier tube and is very temperature dependent.
4
Optical aberrations
When parallel rays pass through a centred, image-forming optical system innites-
imally far from the axis, they arrive at the corresponding paraxial image point. Its
position is computed using the Gaussian optical equations.
It is an unfortunate fact that in general parallel rays coming through the system
at the outer parts of the apertures or pupils do not arrive at the same image point as
the paraxial rays. They arrive in the vicinity of the paraxial image point and their
separations from it dene the aberrations of the system. Rays which pass the edges
of the pupils are called the marginal rays and the separation of these rays from the
paraxial image point can be calculated using aberration theory.
In instrument design we arrange the various components of the system so that
the aberrations are reduced to zero where possible and minimised where not. The
whole theory is non-linear; the various causes of aberration interact with each other
and the resulting complication is often deplorable. To ease the analysis we simplify
things by treating each aberration as though the others were all absent.
Optical designers are sometimes eccentric and the routes to their designs are
often highly individual and hard to follow. Conrady, for example, is regarded as the
father of modern optical design but his notation is not to be recommended. In any
case, for plain spectrograph optics we need only a small subset of the theory and
this is laid out below in reasonably familiar language.
There are two principal ways of reducing the aberrations of an instrument.
The rst is by altering the shape of a lens,
1
by thickening and bending it by
changing its radii of curvature, c
1
and c
2
, without altering its power (n 1)Lc.
The second is by placing stops so as to x the position of the pupils
2
of the system,
since the aberration produced by an optical surface depends on the distance of its
1
The shape of a mirror can be changed by guring it, thereby changing its shape from spherical to paraboloidal,
ellipsoidal or to a fourth-order curve.
2
Strictly the iris although the word pupil is used commonly and carelessly in this connection.
28
4.3 First-order aberrations 29
vertex from its pupils. The position of the iris diaphragm in a camera lens, for
example, is carefully chosen as part of the design.
There are various theorems which guide the instrument designer and there are
ray-tracing computer programs which verify (or not!) the calculations which have
been made. The whole subject is an art rather than a science and in the ne detail,
experience tends to be the guiding principle as much as cold analysis.
4.1 The Seidel aberrations
Gaussian optics the rst step in instrument design relies on the approxima-
tions sin ; cos 1. From this the approximate positions of image planes,
magnications and so on can be calculated. This is the zero-order theory.
Seidel theory, sometimes calledrst-order theory or third-order theory, makes
the approximations sin
3
,6, cos 1
2
,2 and tan +
3
,3.
Analysis of ray paths then shows the marginal rays of a bundle passing at cal-
culable distances from the paraxial image point.
NB Seidel theory is not exact, but it brings the design close enough to a starting
point for ray tracing, when experimental renement of the design becomes possible,
formerly by hand calculation using seven-gure logs and trigonometric tables, but
these days by computer-iterative ray tracing.
The aberrations can be distinguished by their severity and by the way in which
they depend for their amplitude on the semi-aperture y and eld of the system.
The derivation of the equations describing them is in many textbooks and they are
distinguished by various coefcients, of which we shall need four: B. F, C and D.
In diminishing order of importance the aberrations are detailed below.
4.2 Zero-order aberration
Defect of focus The rays of a bundle arriving at the Gaussian image point all
pass through the same point, but this point is before or after the image plane. This
aberration is easily corrected by moving the image plane to its proper position.
4.3 First-order aberrations
4.3.1 Spherical aberration
This is caused by employing optical elements with spherical surfaces, and affects
all image points in the same way. If the Gaussian image point is on the optic
axis, the marginal rays cross the axis before or after the paraxial rays. The distance
along the axis fromthe paraxial image point to the marginal image point is called the
30 Optical aberrations
marginal focus
circle of least confusion
paraxial focus
Figure 4.1 Spherical aberration. The longitudinal aberration is the distance be-
tween the points where paraxial rays and marginal rays of a collimated bundle
cross the optic axis. Of more practical importance is the diameter of the circle of
least confusion, which is the quantity which denes the spatial resolution of the
lens.
longitudinal spherical aberration and in the Seidel approximation it is proportional
to the focal length and varies as the square of the aperture. For a lens or mirror of
focal length f the usual formula for longitudinal spherical aberration is
L
sph
= By
2
f. (4.1)
where y is the semi-aperture (the radius of the lens if it is a simple lens) and f the
focal length. B is the spherical aberration coefcient and its value depends on the
bending of the lens. There are formulae in Appendix 1 for various circumstances.
B is well known to be zero for a paraboloidal reector.
It is generally more useful to know the diameter of the circle of least confusion;
that is, the diameter of smallest image patch that can be achieved when the image
plane is appropriately adjusted. The formula for this is
L
colc
= B
y
3
2
. (4.2)
The patch is also known as the blur circle.
Spherical aberration can be corrected in a simple lens by making it an achromatic
doublet. This lens has two components in contact, generally cemented together, of
different types of glass, one component having positive and the other negative
power. They are chosen so that the focal length of the combination is the same for
two different wavelengths, that is for two different refractive indices. By bending
the combination, the spherical aberration of the two components can be made equal
and opposite, and as the components are in contact the two aberrations add simply.
This process will be discussed in detail later.
4.3.2 Coma
This is regarded as the great Satan of optical design, and it is a rule of thumb that,
if coma can be eliminated, the lesser aberrations tend to take care of themselves, or
at least are less troublesome to reduce to acceptable levels. Coma depends on the
4.3 First-order aberrations 31

y
T
S
P
3y
2
Ftan
Figure 4.2 The meridional or tangential coma of a lens. In the absence of other
aberrations, the marginal rays of an oblique bundle in the plane of the optic axis
meet at the tangential image point T, which generally is not on the oblique axis
the line joining the Gaussian image point P to the lens vertex. Tangential coma
is then measured as the distance in the focal plane from point P to the tangential
focus point T. Sometimes it is measured as the angle subtended by this distance
at the lens vertex. The oblique marginal rays perpendicular to this diagram meet
at S, the sagittal focus, and in the Seidel approximation the distance PS is always
1,3 of the distance PT.
eld angle of the chief-ray of a bundle and we therefore consider oblique bundles
of rays where the chief-ray is inclined at an angle to the optic axis.
In an oblique ray bundle, the paraxial rays arrive as before at the paraxial focus.
If there is no spherical aberration, the marginal rays converge to the same focal
surface, but they arrive at different points on that surface. Two pairs of marginal
rays are particularly important and may be said to dene the amount of coma.
Tangential coma
When the two marginal rays in the plane of the optic axis meet at a point which is
not on the paraxial ray, the perpendicular distance of this point from the paraxial
ray is the tangential coma. The height of the coma patch is
L
coma
= 3y
2
Ftan . (4.3)
F, the coma coefcient, and B, the spherical aberration coefcient, are the two
most important design parameters in a spectrograph. Appendix 1 gives formulae
for calculating B and F for single lenses and mirrors.
Sagittal coma
In a simple case, the two marginal rays from the ends of the diameter perpendicular
to the plane of the chief-ray arrive at a point between the paraxial focus and the
point where the meridional marginal rays meet. Its distance from the paraxial focus
is one-third of the distance to the meridional marginal focus and, by symmetry, is
on the line joining them. Other pairs of marginal rays arrive at points on a circle,
of which the meridional and sagittal foci are at the ends of a diameter. Ray pairs
from each annulus of the lens form circles of different diameter and the resulting
32 Optical aberrations
M
S
G
Figure 4.3 The sagittal coma of a lens. This is dened by the point S, where
marginal rays in the plane normal to the plane of Fig. 4.2 meet. Pairs of marginal
rays from other mutually antipodal points on the rim of the lens meet on a small
circle which has the tangential and sagittal foci at the ends of a diameter. Ray pairs
from other, smaller annuli form similar, smaller circles all lying cradled between
two lines of 60

inclination radiating from the Gaussian focus G.

coma patch is triangular with its apex at the paraxial focus and an apex angle of
60

. The illumination is uneven, being fainter nearer the outer edge, and it is from
this comet-like form that the aberration gets its name.
4.3.3 Astigmatism
This aberration is closely linked to eld curvature and the two are usually considered
together. Like the B and F coefcients of spherical aberration and coma, the other
two are C which vanishes for zero astigmatism and D which vanishes if the eld is
at.
Suppose that a distant wheel-like object in the form of concentric black circles
and radial lines on a white eld, centred on the optic axis and perpendicular to it, is
imaged by the system. If astigmatism is present, there are two focal surfaces, both
spherical. One of them will show the rings in focus and the other the radial lines.
The two surfaces have a common tangent at the Gaussian focus.
In neither case does a point on the object correspond to a point in the image.
An annular parallel bundle of rays inclined to the optic axis will rst converge
to form a short line image; a line which, if projected, will cut the optic axis: and
the bundle will subsequently converge to a short line image perpendicular to the
former. This is positive astigmatism. The position of the rst short line is the sagittal
focal surface the spokes of the wheel and the other the meridional or tangential
surface the rimof the wheel. The two do not necessarily happen in the order given
here. Astigmatism, like coma, may be negative.
In classical prism and grating spectroscopes it is the tangential surface which
chiey concerns us. This is where the spectrum lines are focused.
4.3 First-order aberrations 33
M
G
S
Figure 4.4 Astigmatism. In the absence of coma the tangential and sagittal ray
fans meet at different points along the oblique axis forming line images of a point
object. Where the rays of the tangential bundle meet, the tangential line image M
is perpendicular to the optic axis, and at the sagittal focus the sagittal line image S
if produced, would cut the optic axis. There is a circle of least confusion G, lying
between the two foci. In spectroscopy it is the tangential focus which is almost
always required. Consequently a spectrum line is not a precise image of the entry
slit but the aberration, being along the slit direction, does not interfere with the
spectral resolution.
Astigmatism, like coma, is an aberration of the eld and vanishes on the optic
axis. In order for it to exist at all there must simultaneously be eld curvature.
3
One
of the two eld curvatures may be zero. In a grating spectrograph, for example,
the tangential eld may be at and the entry slit will then be imaged, in various
colours, on to a plane. The image of a point on the slit is a short line lying along
the direction of the spectrum line image. Thus we do not notice it unless the slit is
very long or the spectral resolution very high.
4.3.4 Astigmatism and eld curvature
Field curvature on its own means that the image of a scene at innity will not lie
on a plane but on a sphere which has its centre on the optic axis and which cuts the
3
Although the latter may exist even in the absence of astigmatism.
34 Optical aberrations
Focal surfaces
T = tangential or meridional surface
S = sagittal surface
P = (fictitious) Petzval surface.
T S P
Figure 4.5 Field curvature. This is inseparable from astigmatism as the tangential
and sagittal images lie on two different spherical surfaces. If there is no astigmatism
the two surfaces are the same, and if eld curvature is also eliminated the eld is
at. There is a third, ctitious surface called the Petzval surface, the radius of which
is dened in the text, and the order T-S-P of the curvatures is always preserved.
optic axis at the Gaussian focus. The eld curvature is dened by the radius of this
sphere or by its reciprocal.
This sphere may be convex or concave towards the lens, i.e. its radius of curva-
ture may be positive or negative. For many purposes, such as photography, CCD
imaging, etc., it is highly desirable that the radius be innite so that the focal surface
is a plane. There is one notable exception: the Schmidt camera, where all aberra-
tions except eld curvature are perfectly corrected and a plane glass photographic
plate is bent to form a spherical surface (of fairly large radius, although glass is
more elastic than most people realise). The symbol c is used in some books for eld
curvature as well as surface curvature. To avoid confusion lower-case c is reserved
for surface curvatures of lenses and mirrors, and capital C
(sufx)
is used for eld
curvatures. C
P
, C
s
and C
t
refer to Petzval, sagittal and tangential eld curvature
respectively. C refers to the astigmatism coefcient and, like the other aberration
coefcients, is always in bold font.
4.3.5 Distortion
This is characterised by a fth coefcient E, and occurs when the position of the
principal plane depends on the eld angle, so that the focal length is a function of
the eld angle. In the absence of all other aberrations the focal length may either
increase or decrease with eld angle. If it increases with eld angle the result is
4.4 Theorems 35
object
object
sagittal focus tangential focus
barrel
distortion
pincushion
distortion
Figure 4.6 Images of a wheel-object, showing the practical effects of astigmatism
and distortion in the images which they produce. Distortion, usually barrel distor-
tion, is inevitable in lenses corrected to give very wide elds but is small or absent
in so-called symmetrical lenses such as the double Gauss camera objective.
pincushion distortion and if it decreases it produces barrel distortion. In other
words the image of a square object is shaped as the name implies. Barrel distortion
is difcult to avoid with wide-angle lenses.
4
4.4 Theorems
4.4.1 Fraunhofers theorem
This is an important theorem which shows how it is possible to place a stop to
control the coma of a system. If the entry pupil of a simple lens is at the lens itself,
the pupil distance is zero and the coma coefcient is written as F
0
.
Fraunhofers theoremis that if the entry pupil is at a distance Z fromthe principal
point, the coma coefcient becomes
F = (F
0
+ ZB). (4.4)
It follows then that if a stop is placed a distance Z from the vertex, given by
Z = F
0
,B, the lens will be free from coma. The distance is negative and so, by
the Cartesian sign convention, the stop must be to the left of the vertex.
Consider for example a spherical interface between glass and air, with a stop
at the centre of curvature of the glass sphere.
5
Parallel bundles of rays, no matter
4
Take especial care when choosing a lens for a FabryPerot spectrograph, where the dispersion is radial and only
a camera lens with freedom from distortion gives a true square-law wavelength scale.
5
This example is repeated in Appendix 1 with attendant mathematical details.
36 Optical aberrations
Figure 4.7 Controlling coma with a stop. In the upper gure a stop is placed at the
centre of curvature of a glass sphere so that collimated ray bundles coming fromthe
left are always symmetrical and are imaged with spherical aberration but no coma
on a spherical focal surface. In the lower gure the sphere has been truncated to
give a plano-convex lens with a plane rst surface, and the stop position is moved
so that the ray bundles reaching the second surface are unchanged. A stop at this
position in front of a plano-convex lens will thus eliminate coma.
in which direction they are travelling, are symmetrical about their chief-ray since,
because of the spherical symmetry, there is no unique optic axis. There must then
be no coma when the bundle comes to a focus after refraction, although there will
be spherical aberration.
Now place a second, plane, surface to the left of the spherical surface to make
a plano-convex lens. The stop, now in air to the left of the lens, can be moved a
distance R[(n 1),n] towards the lens so that parallel bundles of rays from it will
arrive at the second spherical surface just as if they had come from its centre of
curvature. To put it another way, the stop should be a distance f [(n 1),n] from
the plane surface of the lens. Again there will be no coma and the stop is at the
position required by Fraunhofers theorem.
This is an elementary example showing the underlying physics: the theorem is
of more general use than this.
A word of warning: coma can only be corrected with a stop if there is some
spherical aberration. It happens occasionally that it is worth leaving some spher-
ical aberration in the system deliberately in order that coma may be removed.
For example, the Ross corrector, which was used in large astronomical tele-
scopes with paraboloidal (and therefore spherical-aberration-free) mirrors, did just
this. The pupil distance is so large (from the telescope primary to the vicinity of
the focal plane) that the introduction of a totally insignicant amount of spher-
ical aberration is sufcient to allow coma correction over a large area on the
image.
4.5 Aberration coefcients for mirrors 37
On its own a singlet lens has no coma provided that its surface curvatures are
connected by
c
2
c
1
=
(n
2
n 1)
n
2
.
andtypically, for n = 1.516728, c
2
= 0.94c
1
will give coma belowthe diffraction
limit.
If the refractive index is 1.618 034 the golden section of the pyramid-builders
and Greek architects the coma of a plano-convex lens is zero when the rst surface
is convex. This is no gift of the gods however: all the other aberrations are there in
full measure!
4.4.2 Petzvals theorem
When astigmatism is present the tangential focal surface is a sphere of curvature
C
t
= 4C +2D, and the sagittal surface curvature is C
s
= 2D.
Petzvals theorem relates the two coefcients C and D by the equation
C D =

i
1
n
i
f
i
. (4.5)
where n
i
is the refractive index of the i th component. This called the Petzval sum
and denoted by the symbol

P
z
.
A surface can be dened, the Petzval surface, of which the curvature, C
P
, is
2(C D) and the two astigmatic surfaces have curvatures given by
C
t
= C
P
+
3
2
C.
C
s
= C
P
+
1
2
C.
C D = 0 is the condition for freedom from astigmatism, and C = D = 0 are the
conditions for freedom from both astigmatism and eld curvature.
4.5 Aberration coefcients for mirrors
These are collected here as they can be given in closed form and are most likely to
be used in spectrograph design. Formulae from which the equivalent coefcients
for lenses may be computed are collected in Appendix 1. When using them bear in
mind that these are only rst-order improvements on Gaussian optics and are not
exact. They give reasonable results at apertures slower than F,12 and eld angles
up to 5

and are adequate starting points for iterative ray-tracing programs.

38 Optical aberrations
e is the excentricity of the conic of which the mirror is the surface of revolution.
Z is the vertexpupil distance and is negative when it is to the left of the vertex.
The spherical aberration coefcient is
B = (1 e
2
),8 f
2
. (4.6)
The coma coefcient is
F =
1
4 f
+
Z(1 e
2
)
8 f
2
. (4.7)
The astigmatism coefcients are
C =
1
2 f
+
Z
2 f
2
+
Z
2
(1 e
2
)
8 f
3
. (4.8)
D =
Z
2 f
2
+
Z
2
(1 e
2
)
8 f
3
. (4.9)
The tangential surface curvature is
C
t
= 4C +2D =
2
f
+
3Z
f
2
+
3
4
Z
2
(1 e
2
)
f
3
. (4.10)
The sagittal surface curvature is
C
s
= 2D. (4.11)
Fromthis it is a short step to see that a spherical mirror is free fromastigmatismif
Z = 2 f . The eld curvature is then 1,f . These are the conditions for a Schmidt
camera.
4.5.1 The focal surfaces of spectrographs
These last two expressions are particularly important for

CzernyTurner spectro-
graphs. The tangential eld is at when
2
f
+
3Z
f
2
+
3
4
Z
2
(1 e
2
)
f
3
= 0. (4.12)
When spherical mirrors are used the gratingmirror distance, Z. = 0.843 f .
(The alternative value, Z = 3.25 f , can safely be ignored.) If the mirrors are
paraboloids, then Z = 0.666 f .
4.5.2 Chromatic aberration
A simple single-element lens has a focal length which depends on its refractive
index, n, and refractive index generally increases with decreasing wavelength. In
4.6 The achromatic doublet 39
crown glass this variation is comparatively low and in int glass it is higher. There
is no discontinuous change in this rate of variation with glass type and a quantity
called the partial dispersion or Abb e v-value is used to measure it. The partial
dispersion of a glass is dened by
: =
n
D
1
n
F
n
C
.
where n
C
. n
D
and n
F
are the refractive indices at 656.3 nm, 589.3 nm and 486.1 nm
respectively.
6
Typical values are
Glass Index v-value
Crown (BK7) 1.516 728 64.2
Flint (SF12) 1.648 143 33.8
and there is an arbitrary division between crown and int glasses at : = 50. Notice
that the ints, with their higher dispersions, have the lower v-values.
The consequences of this change of focal length with colour are:
r
The image of a white luminous point object is a coloured point surrounded by a disc of the
complementary colour. This is known as longitudinal colour. The actual colours depend
on where the focusing screen is placed.
r
The red image of a nite object is larger than the blue image (lateral colour).
Correction of chromatic aberration then requires both that the focal length be
the same for different wavelengths and that the principal planes shall coincide for
different wavelengths. Only then will the images of an object be superimposed on
the focusing screen.
4.6 The achromatic doublet
Two elements, of two glasses with different refractive indices and dispersions, can
be combined to make a doublet lens which has the same focal length for two dif-
ferent wavelengths. The parameters available are the two bendings (that is the
actual shapes of the two elements), the two refractive indices and the two disper-
sions; although these latter, being physical properties, depend on the glass types
available.
The algebra is elementary. From the Gaussian formula, P = (n 1)C, we nd
P = P
1
+ P
2
= (n
1
1)C
1
+(n
2
1)C
2
= (n

1
1)C
1
+(n

2
1)C
2
.
6
These wavelengths correspond to the similarly lettered Fraunhofer lines in the solar spectrum.
40 Optical aberrations
where n

1
and n

2
are the refractive indices at the second wavelength. Then
Ln
1
C
1
= Ln
2
C
2
.
P = (n
1
1)C
1
(Ln
1
,Ln
2
)(n
2
1)C
1
= Ln
1
C
1
_
n
1
1
Ln
1

n
2
1
Ln
2
_
C
1
Ln
1
(:
1
:
2
)
and the sign indicates that an intermediate wavelength has been used for the
factors (n
i
1).
After a few more lines of algebra you will nd
P
1
=
_
:
1
:
1
:
2
_
P
and
P
2
=
_
:
2
:
1
:
2
_
P.
From these two equations the curvatures of the two elements can be computed.
However, nothing has been said about the individual curvatures of the four sur-
faces. These can be adjusted to make the spherical aberrations equal and opposite.
7
A minor caution, however: the chromatic aberration is not perfectly corrected. Al-
though the formulae above provide the same focal length for the two colours, the
principal planes and hence the focal planes are not exactly coincident and small
longitudinal colour remains. At F,10, though, this is comparable with the diffrac-
tion disc and is not signicant. Yet the ray tracer will correct for this also and will
remove one or other of these two aberrations. Finding tool radii to match is another
problem entirely and best left to the lensmakers.
The search for a respectable at eld of 10

20

has been going on since the

invention of photography and has resulted in the elaborate photographic objectives
discussed by Kingslake,
8
to be found especially on 35 mm cameras of quality.
7
In the old days this would have been done by hand using aberration formulae. Today the iterative ray-tracing
program does it instantly.
8
R. Kingslake, A History of the Photographic Lens (San Diego: Academic Press, 1989).
5
Fourier transforms: a brief revision
5.1 Fourier transforms
Fourier transforms and Fourier methods generally are such a powerful tool in de-
scribing physical optics and spectrography that a familiarity with them is a sine
qua non of the optical designer. The spectrograph is the physical manifestation of
a Fourier transformer. It takes an incoming signal, a function of time, and displays
the frequencies present in it. It performs in fact the power transform of the signal,
a concept which will be described later.
This chapter is chiey to introduce the reader to the notation and to summarise
the more important properties and theorems of Fourier theory. The notation in
particular differs from that of mathematicians because we prefer to deal with real,
measured quantities rather than the angular variables which they usually employ.
The conjugate variables x and p traditionally used in Fourier theory may by replaced
by t and , that is time in seconds and frequency (not angular frequency) in hertz,
and functions of these two variables are connected by the Fourier inversion theorem:
() =
_

f (t )e
2it
dt.
f (t ) =
_

()e
2it
d.
which we usually write symbolically as
f (t ) ().
Note that this particular notation frees the denitions fromthe normalising constants
which otherwise appear in front of the integral sign.
In geometrical optics we also have to deal with a spatial dimension and its
inverse, spatial frequency. On a grating for example, we may have a ruled width
x measured in centimetres and the neness of the grating measured in lines per
41
42 Fourier transforms: a brief revision
centimetre p. Similarly in photography, the resolution of a lens or a photographic
emulsion may be dened in line-pairs per centimetre, and by convention this uses
the symbol s.
5.2 Theorems
Various elementary theorems follow here and for brevity are given without proof.
Proofs are generally simple and to be found in any treatise on Fourier theory.
1
5.2.1 The addition theorem
f
1
(t ) + f
2
(t )
1
() +
2
(). (5.1)
5.2.2 The multiplication theorem
_

1
()

2
() d =
_

f
1
(t ) f

2
(t ) dt. (5.2)
where the asterisk denotes a complex conjugate. An important special case of this is.
5.2.3 Parsevals theorem
_

| () |
2
d =
_

| f
1
(t ) |
2
dt.
5.2.4 The shift theorem
f (t +t
0
) ()e
2it
0
. (5.3)
where t
0
is a constant. Its corollary is important:
f (t +t
0
) + f (t t
0
) 2() cos 2t
0
.
5.3 Convolutions
These are conveniently described by a spectroscopic example. No spectrometer is
perfect. It it were and if monochromatic light really existed then the spectrum
1
The exact and rigorous proof of the fundamental inversion theorem is anything but simple and to be found for
example in Titchmarsh, a bible to be consulted only by the most didactic reader.
5.3 Convolutions 43
I ()
(a)
(
0
)

I ()
(b)
C(
0
)

I ()
(c)
S(
1
)
1
S()

I ()
(d)
C(
2

1
) S(
1
)
1

1

2
Figure 5.1 Convolution of an instrumental prole with a continuous spectrum.
Each element of the spectrumis treated as a monochromatic emission line which is
converted by the shortcomings of the spectrograph to the shape of the instrumental
prole. The sum of all these conversions is added to give the observed spectrum.
of the light would be a Dirac delta function A(
0
) at the appropriate place in the
spectrum, where A is the intensity. In the real world the spectrometer produces a
curve of intensity against wavelength which we denote by AI (
0
) and which
either peaks at
0
or has its centre of gravity at
0
. The function I () is called the
instrumental prole of the spectrometer and
_

I () d = 1. The instrumental
prole in general does not hold its shape over the whole spectrum but the shape
changes slowly enough with wavelength that it serves to demonstrate the idea of a
convolution.
Consider now a real spectrum, S(), and in particular an innitesimal part, of
intensity S(
1
) d
1
at wavelength
1
. This approximates to a -function and will be
converted by the defects of the spectrometer to a function S(
1
) d
1
I (
1
) and
this in turn will contribute an innitesimal amount of power S(
1
)I (
2

1
) d
1
apparently at a wavelength
2
. The apparent spectrum, which is the total power
appearing at
2
from all parts of the true spectrum, is then the integral
P(
2
) =
_

0
S(
1
)I (
2

1
) d
1
.
and this is a practical example of the mathematical operation of convolution:
c(x

) =
_

f
1
(x) f
2
(x

x) dx. (5.4)
44 Fourier transforms: a brief revision
usually written symbolically as
c(x

) = f
1
(x) f
2
(x).
The operation can be visualised by drawing the function f
1
(x) on a sheet of paper,
drawing f
2
(x) on transparent paper, turning the drawing over and laying it on top
of the sheet with f
1
(x) and sliding it along, integrating the product of the two as
you go. Each integral gives one point on the convolution.
5.3.1 The convolution theorem
This theorem is a most important and powerful one, particularly in Fraunhofer
diffraction theory, and it greatly simplies the mathematics of diffraction.
It states that if c(x) is the convolution of f
1
(x) and f
2
(x) as above, then their
respective Fourier transforms ( p),
1
( p) and
2
( p) are related by
( p) =
1
( p)
2
( p). (5.5)
In words: the Fourier transform of a convolution is the product of the Fourier
transforms of the components.
5.3.2 Convolution algebra
There is an algebra of convolutions which can be used to manipulate them. It is not
difcult. The rules are as follows:
The commutative law
f
1
(x) f
2
(x) = f
2
(x) f
1
(x).
The associative law
f
1
(x) [ f
2
(x) f
3
(x)] = [ f
1
(x) f
2
(x)] f
3
(x).
The distributive law
f
1
(x) [ f
2
(x) + f
3
(x)] = f
1
(x) f
2
(x) + f
1
(x) f
3
(x).
but note that convolutions and multiplications must be kept separate. The distribu-
tive law does not hold in a mixture of them:
f
1
(x) [ f
2
(x) f
3
(x)] = [ f
1
(x) f
2
(x)] f
3
(x).
Nevertheless the algebra can be extended. For example,
[ f
1
(x) f
2
(x)] [ f
3
(x) f
4
(x)] [
1
( p)
2
( p)][
3
( p)
4
( p)]
where Greek and Roman letters denote Fourier pairs as before.
5.5 Useful functions 45
5.4 The WienerKhinchine theorem
We rst of all dene spectral power density (SPD), which is what a spectrometer
measures.
The autocorrelation function of a function f (t ) is similar to the convolution of a
function with itself, and is dened as
A() = lim
T
1
2T
_
T
T
f (t ) f (t +) dt.
where the factor before the integral implies a time average.
If the Fourier pair of f (t ) is (), the shift theorem gives the Fourier transform
of f (t +) as ()e
2i
and the multiplication theorem then gives
_
T
T
f (t ) f (t +) dt =
_

()()e
2i
d =
_

|()|
2
e
2i
d.
On the (reasonable) assumption that the power is zero before the beginning and
after the end of the signal receipt, we dene the SPD to be G() given by
G() =
|()|
2
2T
.
The WienerKhinchine theorem states, as in the equation above, that the SPD is
the Fourier transform of the autocorrelation function of the signal f (t ):
A() G().
or, more practically,
1
2T
_
T
T
f (t ) f (t +) dt
|()|
2
2T
. (5.6)
5.5 Useful functions
There are some mathematical expressions which recur frequently in physical sci-
ence in conjunction with their Fourier transforms or Fourier pairs. Some of them
are described here.
5.5.1 The top-hat function
This is a function which is zero from to a,2 and from a,2 to . Between
a,2 and a,2 it has the value 1. It is generally denoted by H
a
(x).
46 Fourier transforms: a brief revision
Its Fourier transform is
_

H(x)e
2i px
dx =
_
a,2
a,2
e
2i px
dx =
e
i pa
e
i pa
2i p
= a
sin pa
pa
= a sinc(pa).
and the so-called sinc function is much encountered in physics and engineering,
not least because data strings have a nite length and the length in turn limits the
frequency resolution. This is because the observed spectrum the Fourier transform
of the data is the convolution of the true spectrum with the sinc function.
5.5.2 The Gaussian function
The Fourier transform of the Gaussian function e
x
2
,a
2
is obtained directly by
completing the square of the exponent in the integrand:
_

e
x
2
,a
2
e
2i px
dx = e

2
p
2
a
2
_

e
(x,a+i pa)
2
dx
= a

2
p
2
a
2
.
The full width at half maximum(FWHM) of a Gaussian is 2a

ln 2, and numeri-
cally is 1.386a where a is the width parameter. The convolution of two Gaussians
of width parameters a and b is another Gaussian of width parameter

a
2
+b
2
, i.e.
the Pythagorean sum of the two width parameters.
5.5.3 The Lorentz function
When a damped electrical oscillator decays the amplitude it radiates varies with time
according to f (t ) = A
0
e
2i
0
t
e
t ,
, where
0
is its natural oscillation frequency and
the decay constant.
The spectrum of the oscillation is the Fourier transform of this:
A() = A
0
_

0
e
2i
0
t
e
t ,
e
2it
dt.
and the integral lower limit is 0 because the oscillation is deemed to begin at this
time. The integral is then
A() =
A
0
2i(
0
) 1,
.
5.5 Useful functions 47
This gives the spectral amplitude of the radiation. The spectral power density, the
square modulus of this, is A()

A(), which is
I () =
| A
0
|
2
4
2
(
0
)
2
+1,
2
.
and this is the prole of a spectrum line emitted in an atomic dipole transition.
The same function can be derived quantum mechanically. It is known as a Lorentz
prole.
5.5.4 The Voigt prole
This is a common shape for spectrum lines which have been broadened by temper-
ature in the emitting gas or plasma.
Temperature alone will broaden a monochromatic line to a Gaussian:
I () = I
0
e
(
0
)
2
,a
2
.
The width parameter, a, comes from the Maxwellian distribution of velocities, and
a
2
= 2
2
kT,mc
2
, where T is the absolute temperature, k is Boltzmanns constant
and m the mass of the emitting species.
Substituting values we nd that a monochromatic spectrum line becomes a
Gaussian with a FWHM L given by L, = 7.16 10
7

T,M where M is
the atomic or molecular weight of the emitter.
The observed spectrum line shape is the convolution of this with the Lorentz
prole resulting from the nite decay time of the oscillator, the so-called natural
width of the line.
5.5.5 Separation of the components of a Voigt prole
Given a spectrum line with a Voigt prole the width parameters of the two com-
ponents can be separated. Suppose that the width parameter of the Gaussian com-
ponent is a and that of the Lorentz component is b. Then the Voigt prole is the
convolution
V() = G() V() = e

2
,a
2
1,(
2
+b
2
).
Let n be the Fourier pair of . Then g(n) G() and l(n) L().
The Lorentz prole L() is, by the WienerKhinchine theorem, the Fourier trans-
form of the autocorrelation of the truncated exponential function which represents
the decaying amplitude of a damped oscillator. The truncation is because the decay
48 Fourier transforms: a brief revision
is deemed to begin at n = 0. Then,
l(n

) =
_

n

e
bn
e
(nn

),b
dn
= b,2e
n

,b
for positive values of n

. Then the Fourier transform of V() is, for positive values

of n,
:(n) = b,2 e

2
n
2
a
2
e
n

,b
.
A graph of log :(n) is a parabola and the values of a and b can be extracted by
simple measurement.
5.5.6 The Dirac delta function
This has the property that (x) is zero everywhere except at the origin x = 0, and
at x = 0 is innite. The important property is
_

(x) dx = 1.
It may be regarded as the limiting case of a top-hat function as it becomes
innitesimally wide and innitely high with its area always unity, or as the limiting
case of a Gaussian:
lim
a0
1
a

e
x
2
,a
2
.
In either case it is an improper function which gives angst to mathematicians but
is useful to physicists.
2
5.5.7 The Dirac comb
This is an innite set of equally spaced -functions each separated from its neigh-
bours by a distance a. It is generally denoted by the Cyrillic letter III (Shah) and
written III
a
(x):
III
a
(x) =

n=
(x na).
Its Fourier transform is then

n=
_

(x na)e
2i px
dx =

n=
e
2inpa
. (5.7)
2
For example, it has no upper bound, which violates one of the Dirichlet conditions that dene a function to be
Fourier-transformable.
5.6 More theorems 49
f (x)
f
p
(x)
a
0
a
1
a
2
a
3
( p)
( p).III( p)
Figure 5.2 The sampling theorem. A continuous truncated function together
with its mirror image produces a real Fourier pair. The same function repeated
indenitely can be synthesised from the fundamental a
0
and its overtones
a
1
. a
2
. a
3
. a
4
. . . . . and these have the same values as the Fourier pair at the appro-
priate values of p. In practice it is the set of regular samples which provides the
experimental database and the function which produced them can be computed
via a FFT computer program.
and it can be shown
3
that this is another Dirac comb, (1,a)III
1,a
( p), so that we can
write
III
a
(x) (1,a)III
1,a
(x).
It is useful for example when describing the aperture of a diffraction grating as
the convolution of a top-hat function of width a with a Dirac comb of spacing b,
the convoluted product then being multiplied by a wide top-hat function of width
Nb to limit the width to N rulings.
5.6 More theorems
5.6.1 The sampling theorem
This is especially important in Fouriermultiplex spectrometry but it applies also
to the theory of diffraction grating spectrographs. It comes from Whittakers inter-
polary function theory
4
where it is called the cardinal theorem, and states, in effect,
that the amplitudes of all the frequencies that a signal contains can be measured
provided it is measured (sampled) at intervals of 1,2
f
, that is, to the reciprocal
of twice the highest frequency present.
5
It serves also to link Fourier series with
the Fourier transform.
3
R. N. Bracewell, The Fourier Transform and Its Applications (New York: McGraw-Hill, 1965).
4
J. M. Whittaker, Interpolary Function Theory (Cambridge University Press, 1935).
5
Sometimes called the Nyquist or folding frequency.
50 Fourier transforms: a brief revision
The signal which enters a spectrograph or spectrometer has an alternating electric
eld which can be described by F(t ) and this has a spectrum, S(), which may
extend from = 0 to
0
.
Consider a periodic function S
p
() which is identical with this spectrum from 0
to
0
, is the mirror image from
0
to 2
0
and is thereafter periodic with period 2
0
.
In other words, it is the convolution of S() and its mirror image in the origin with
a Dirac comb of interval 2
0
.
This periodic function can be written as the sum of an innite series:
S
p
() =
a
0
2
+

n=1
a
n
cos
2n
2
0
=
a
0
2
+

n=1
a
n
cos
n

0
. (5.8)
and then the amplitudes a
n
are given by
a
n
=
1

0
_

0
0
S() cos
n

0
d.
and the integral here need go only from 0 to
0
because the integrand is zero from
there on.
Now suppose we have a signal, F(t ), which contains frequencies up to
0
and
is empty from then on. Its spectrum, S(), is continuous from = 0 to =
0
.
Corresponding to this spectrum we may construct a symmetrical function S
s
()
such that S
s
() = 1,2(S() + S()), which is symmetric and stretches between

0
and
0
. The convolution of this function with III
2
0
is symmetric and periodic
with a period 2
0
and can therefore be synthesised from a discrete innite set of
cosines with coefcients which are a set of numbers a
n
. In Fourier language:
S
s
() =
1
2
[S() + S()] III
2
0
and the corresponding F
s
(t ) is
F
s
(t ) =
1
2
[F(t ) + F

(t )]III 1
2
0
(t ).
Since F(t ) is real, F(t ) = F

(t ) and F
p
(t ) = F(t )III 1
2
0
(t ), which is the set, F(nt
0
),
of samples of F(t ) at intervals t = 1,2
0
. From this set of samples we recover
S
s
() by the Fourier sum:
S
s
() =

F(nt
0
)e
2in,
0
=

F(nt
0
)e
int
0
.
5.7 Aliasing 51
Thus, in principle, a measurement of the incoming signal at intervals 1,2
0
is
sufcient to give the spectrum of the signal. The spectral resolution depends on the
number of samples taken of the signal, when the signs in the sum are replaced
by 0 and N, the (nite) number of samples taken.
The application of this theorem is particularly apparent in Fourier spectroscopy,
but the analogy in grating spectroscopy is apparent also when considering the sum
of amplitudes reected from a grating. The angle of diffraction and the implied
phase delay in the signal are analogous to the exponential in the Fourier sum and
different values of correspond to different angles.
The detector receives the sum of the complex amplitudes from the rulings, in
other words the sumof the samples, each delayed in time fromits neighbours by the
extra path difference it has travelled. This sum multiplied by its complex conjugate
is the power detected at each point of the spectrum.
5.7 Aliasing
The highest frequency,
0
, in the last section is also called the folding frequency
of the spectrum: the frequency beyond which there should be no power. If the sam-
pling interval is too large, the corresponding folding frequency may be too low for
the spectrum and the power beyond it at frequency appears in the spectrum at
frequency
0
. Power at frequency 2
0
appears at frequency 0. For example,
consider a cosine of amplitude a. It must be sampled at 0. . 2. 3 etc., and alter-
nate samples will have values a. a. a. a . . . From this the amplitude is correctly
recovered by the summing process. Had the samples been taken at 0. 2. 4. 6 . . .
they would all have had the same value, a, and the cosine would appear to be at
zero frequency.
Again, had the samples been taken at 0. 3. 6. . . the frequency would have
appeared in its proper place.
Aliasing can thus be put to good use. If the spectrum is conned to a small part
of the frequency domain it can be under-sampled deliberately. It may in fact occupy
the third, fth etc. period of the periodic function, S
s
(), but the Fourier-summing
process will put it apparently in the rst period. This is perfectly satisfactory pro-
vided we knowbeforehand what the sampling interval and the range of the spectrum
are, and it avoids the tediumof sampling for the empty part of the spectrumat lower
frequencies.
Again the analogy with grating spectra will be clear: we may look at the second
or third orders of diffraction, knowing beforehand that the rst-order spectrum at
longer wavelengths is absent, either by ltering or by detector impotence.
6
Physical optics and diffraction
6.1 Fraunhofer diffraction
Elementary textbooks dealing with a transparent aperture in physical optics reduce
the question of Fraunhofer or far-eld diffraction to a simple two-dimensional
form in which a plane wavefront passes through an opaque surface. There is no
variation perpendicular to the plane of the diagram and in this third dimension the
aperture is assumed to have unit length. A narrow strip of width dx along the x-
direction is a source of secondary wavelets and transmits, according to Huygens
principle, a wave with an amplitude proportional to the area A dx, where A is the
amplitude per unit area on the incident wavefront. In the far-eld approximation
the inverse-square law is ignored and the amplitude, including the phase,
1
at the
point P is dU, where
dU = A dx e
2i

(Rx sin )
.
where R is large compared with and x. If the amplitude A varies from point to
point along the aperture, the total amplitude arriving at point P is
U() = e
2i

R
_
A(x) e

2i

x sin
dx.
and the integral is over the whole of the transparent part of the aperture.
If A varies in the x-direction and if we replace sin , by a new variable p and
take the innite integral we nd
U( p) = C
_

A(x)e
2i px
dx (6.1)
so that
The amplitude U( p) diffracted in a direction is proportional to the Fourier
transform of the incident amplitude A(x) across the aperture
1
Remember that phase change =
2

path change.
52
6.1 Fraunhofer diffraction 53

x
s
in

x
dx
R
P
Figure 6.1 Fraunhofer diffraction by a slit aperture. Each element of the aperture is
treated as a separate source coherent with all the others and the several amplitudes
are added at the image point to produce a resultant which depends on the various
phases as well as the magnitudes of the elements.
and as usual in optics, A(x), the variation of incident amplitude with x, when
multiplied by its complex conjugate, is the transparency of the aperture.
2
In practice it is the intensity of the diffracted light which we measure and this
in turn is the (complex) diffracted amplitude multiplied by its complex conjugate,
and we write, remembering that C = e
2i

R
so that CC

= 1,
I ( p) = U( p)U

( p).
This can now be applied to various apertures and the usual diffraction patterns are
described.
6.1.1 Single-slit diffraction
A single slit of width n will diffract an amplitude a( p) where
a( p) = A
_
n,2
n,2
e
2i px
dx
= A
1
2i p
(e
i pn
e
i pn
)
= An
1
pn
sin(pn)
= Ansinc(pn).
where the sinc function (sin x,x) is as described in the previous chapter.
2
A(x) itself may be complex, since there may be variations in the optical thickness in what is otherwise a
transparent part of the aperture, and this will cause the phase to vary with x. This happens for instance in the
blazed transmission grating.
54 Physical optics and diffraction
The diffracted intensity then varies according to
I ( p) = I (0)sinc
2
(pn).
which is the familiar expression for diffraction by a single slit.
6.1.2 Two-slit diffraction
Two slits, as in the Youngs slits experiment, are represented by an aperture function
which is the convolution H
n
(x) [(x a,2) +(x +a,2)], where nis the width
of each slit and a their separation.
The Fourier transform of this is 2 sinc(pn) cos(2pa), and this, when multi-
plied by its complex conjugate, describes the familiar two-slit diffraction pattern.
Bear in mind that we can rejoin physical reality at any time by replacing
the abstract variable p by sin ,.
6.1.3 N-slit diffraction: the diffraction grating aperture
The original crude diffraction gratings were transparent slits ruled on an opaque
screen of smoked glass. Such an aperture can be represented by narrow top-hat
functions representing a single ruling, a Dirac comb representing the grid of aper-
tures and a broad top-hat function dening the limited width of the grating. The
result is
G(x) = H
Na
(x)[H
n
(x) III
a
(x)]. (6.2)
The width of a ruling is n, the spacing between adjacent rulings (the grating con-
stant) is a and there are a total of N rulings.
The Fourier transform of this is
I( p) = Na sinc(Npa)
_
nsinc(pn)
1
a
III
1,a
( p)
_
. (6.3)
and with p = , sin as before, this is the amplitude diffracted in the direction
. Notice in particular that because of the convolution with a Dirac comb, I( p)
will have a maximum whenever p = n,a, that is when sin = n,a. n is any
integer, positive or negative and ,a is the ratio of wavelength to slit spacing. The
application of this to diffraction grating theory will be elaborated in Chapter 8.
6.1.4 Diffraction with oblique incidence
If the wavefronts are incident on the diffracting aperture at an angle to the aperture
normal, we must take account of the phase at the aperture. The complex amplitude
6.2 Two-dimensional apertures and oblique incidence 55

x
s
i
n

x
s
in

R
P
R
Figure 6.2 Fraunhofer diffraction with oblique incidence. The different times of
arrival of the incoming wavefront at the different elements of the aperture imply
an extra phase term in the eventual sum of the elementary amplitudes at the image
point.
at z = 0 is then
A(x)e
2

x sin
.
and the diffracted amplitude is
U( p) =
_

A(x)e
2

x sin
e
2

x sin
dx.
For N slits, where A(x) = H
Na
(x)[H
n
(x) III
a
(x)], the diffracted amplitude has
a maximum whenever sin +sin = n,a.
6.2 Two-dimensional apertures and oblique incidence
Assume that the diffracting aperture in three-dimensional Cartesian coordinates
occupies the plane z = 0. A plane wavefront, incident obliquely, is conveniently
described by its direction cosines, l. m. and n. If the phase is zero at the origin,
the wavefront occupies the plane l x +my +nz = 0 and the phase at the point
P(x. y. 0) is (2,) times the perpendicular distance (l x +my) from the point to
the diffracting plane. The two-dimensional complex amplitude at the diffracting
aperture is then
A(x. y)e
2i

(l x+my)
.
and the diffracted amplitude is
U( p. q) =
_

A(x. y) e
2i

(l x+my)
e
2i

( px+qy)
dx dy.
where p and q are, as before, the direction cosines of the diffracted wavefront.
56 Physical optics and diffraction
A(x. y) for a diffraction grating is separable into A
1
(x) A
2
(y) and A
2
(y) is a top-
hat function with a width equal to the length of the grating rulings. The condition
nowfor a maximumof the diffracted amplitude is obtained as before, when l + p =
n,a and m +q = 0, but l and p are now direction cosines. l = sin i cos and
p = sinr cos , and the equation for a maximum, in Cartesian coordinates, becomes
(sin i +sinr) cos = n,a. (6.4)
where is the angle of inclination of the incoming chief-ray to the xz plane.
As with the diffraction grating, the amplitude at the diffracting aperture is often
separable into functions of x and y, and the integrals are separated and elementary.
However, if there is circular symmetry, the whole integral equation must be referred
to polar coordinates. The problem is not usually to be found in spectrograph design
and will not be pursued further here.
7
The prism spectrograph
7.1 Introduction
This is the traditional form of spectrograph or spectroscope, rst perfected by
Bunsen and Kirchhoff. In its transmission formthe prismis generally an equilateral
triangle in section. There is no absolute virtue in this as it is the prism base length
which determines the resolving power, and the equilateral is a compromise between
too great an apex angle, which results in more light reection at the surfaces, and too
small an apex angle, which results in an extra weight of glass at no extra resolution.
The angle of 60

is also close to the Brewster angle

1
for one of the planes of
polarisation, which gives in principle perfect transmission for that plane and a
guaranteed 50% throughput.
7.2 The traditional prism spectrograph
This is still useful for rapid determination of refractive index.
The resolving power is determined by the wavelength range corresponding to the
FWHM of the diffraction pattern produced at the image of a slit in monochromatic
light. In Fig. 7.1 we have
h = s cos
(A + D)
2
. b = 2s sin
A
2
. n =
sin
_
(A+D)
2
_
sin
_
A
2
_ .
whence
n

=
cos
_
(A+D)
2
_
sin
_
A
2
_
1
2
D

.
1
It would be exact for a refractive index of

3 = 1.732 . . .
57
58 The prism spectrograph
A
h
D
Figure 7.1 The spectrograph derived fromthe classic prismspectroscope of Kirch-
hoff and Bunsen. The resolution depends on the width, h, of the emerging beam.
The ratio ,h determines the optical resolution via diffraction theory and the
dispersion requires an angle greater than this between two resolved wavelengths.
and on making the necessary substitutions:
b n, = h D,.
The Rayleigh criterion for resolution is D = ,h in this instance so that

d
= R = b
n

.
where b is the base length of the prism. Visual inspection with the telescope on a
goniometer table divided into minutes of arc will allow the refractive index to be
measured with a precision of 1 part in 10
4
, and interpolation between two known
lines on a photograph will achieve a wavelength measurement with an accuracy
of a few parts in 10
6
, bearing in mind that the centre of a diffraction-limited line
image can be estimated to 1%.
n, tends to vary rapidly with wavelength in highly dispersive glasses and
dispersioncurves for UV-qualityfusedquartz anda typical short int glass suitable
for prisms are shown in Fig. 7.2.
One of the chief reasons why prism spectrographs are still worth considering is
their freedom from grating ghosts. The line prole, provided that the prism is the
limiting aperture, is a sinc
2
function, and may have its secondary maxima further
reduced by a mask. This makes it sensitive to faint satellite lines on either side of a
parent line and this, together with the relative freedom from scattering when best-
quality glass is used, makes it suitable for Raman spectrography, where spectral
purityandnot highresolutionis the major criterion. Ina Littrowmounting, where the
base of the 30

60

90

reecting prism is 75 mm, the resolving power of a Schott

SF11 glass prism at 5000

A is 40 000, so that a Raman satellite 0.12

A (0.5 cm
1
or 6 10
5
eV) from the parent line can in principle be resolved provided it is
bright compared with the rst satellite of a sinc
4
line shape.
7.4 The Littrow mounting 59
7000
6000
5000
4000
3000
SiO
2
SF11
2000
1000
2000 3000 4000 5000 6000
Wavelength, Angstroms
7000 8000 9000 10,000
cm
1 n

Figure 7.2 The wavelength dispersion curves for two materials suitable for spec-
trograph prisms. The short int SF11 is satisfactory at wavelengths above
3200

A and fused quartz is unsurpassable between 3000

A and 2000

A. The
curves show resolving power of a prism with a 1 cm base length.
7.3 The focal curve theorem
This derives from the elementary consideration of chromatic aberration in a lens. If
most of the lens is removed, the images of a point object will appear in focus along
the optic axis, the red image most distant from the lens vertex. By extrapolation, a
prismwith lens of the same glass type to collimate and focus the light will showthe
spectrum of a point source lying along a line joining the red image to the source.
This is not exact of course, but it explains why the detector in such a spectrograph
lies at a steep angle to the emerging chief-rays. The focal surface is generally not
at but is a gentle curve which must be determined by calculation or ray tracing.
Achromatic or apochromatic lenses may be computed for collimator and camera
which will give a at focal surface, and camera lenses can be pressed into service
on both sides to give a low focal ratio, but it must be borne in mind that such lenses
are computed strictly for the visible region and lose resolution very rapidly outside
their intended spectral extent, especially at the UV end of the spectrum.
7.4 The Littrow mounting
This uses a 30

60

90

prism with a mirror to reect the light, giving in effect a

60

prism with half the weight of glass. With a high focal ratio and a singlet lens
60 The prism spectrograph
Figure 7.3 The traditional quartz Littrow mounting. With a resolution of about
0.1

A throughout the visible and near UV, this was state-of-the-art spectrography
in the mid twentieth century and much used for the analysis of molecular spectra.
acting as collimator and camera diffraction-limited resolution is possible, although
long exposure times are needed, and the photographic plate is at a steep angle to
the output chief-rays.
The Hilger 1
1
2
-metre quartz Littrow spectrograph (Fig. 7.3), which gave yeoman
service in the rst half of the twentieth century, was a typical example of a high-
performance Littrow prism spectrograph, working at a focal ratio of F,30 and
giving resolving power in the region of 40 000 in the near UV. The recording plates
were 250 mm long by 100 mm wide, were curved gently by pressing them against
a mandrel in the plateholder and gave full resolution along the whole length of
the spectrum. Ten or more spectra could be recorded on one plate by moving the
plateholder vertically with a rack-and-pinion motion.
7.5 The PellinBroca prism
This was an ingenious device (Fig. 7.4) for producing a monochromator with con-
stant deviation in this case 90

by burying the two halves of the usual prism into

one block of glass with a 90

45

45

prism acting as a reector. The wavelength

presented at the output slit or eyepiece could be varied by rotating the prism slowly
with a turntable and worm-screw.
It is in effect a 60

prism, split into two and with a 45

prism using total internal

reection in place of a mirror. It was a feature of the constant deviation spectroscope,
where wavelength scanning was by simple rotation of the prism table rather than
the whole of the telescope. Many remain far back in the store rooms of university
physics departments.
2
The resolving power is determined by n, 2 base
length of the smaller of the two buried 30

60

90

prisms (Fig. 7.4).

A serviceable high-speed, low-resolution spectrograph can be made from this
prism, using photographic lenses for collimation and photography and Fig. 7.5
shows an example using a 200 mm F,3.5 Tessar as collimator and a 90 mm F,1.8
double-Gauss 35 mmcamera lens to focus the spectrumon to a CCD. The dispersion
gives a spectrum from 4000

A to 7500

A of about 11 mm length, depending on the
glass type in the prism. With 5 m pixels in the CCD the resolution is about 1.8

A.
2
The authors own example has a face length of 105 mm on the larger 30

60

90

prism, a thickness of 45 mm,

a mass of 886 g and acts as a paperweight on his desk.
7.6 Focal isolation 61
90
Figure 7.4 The PellinBroca prism, which at the minimum deviation position
gave an output beam perpendicular to the input beam. This in turn gave rise to the
constant deviation spectroscope, which was scanned in wavelength by rotating
the prism on a turntable while the collimator and telescope were xed.
Figure 7.5 The PellinBroca prism in a serviceable spectrographic mounting.
Standard camera lenses can be used for collimator and camera and in the vicinity
of 5000

A for example will give a resolution in the region of 0.5

A over a limited
range.
7.6 Focal isolation
This is a technique invented by H. Rubens and R. W. Wood initially for investigating
heat waves.
3
Essentially it is a prism technique, relying on chromatic aberration
for its effect. In its original form it comprised a lens, a source and a screen with a
small hole to transmit light through to the detector. With the chromatic aberration
3
R. W. Wood, Physical Optics (New York: McMillan, 1911), 414.
62 The prism spectrograph
detector
Fabry lens
grid
aplanatic lens
oscillating
plane mirror
Figure 7.6 The focal isolation principle. A not yet fully explored possibility
for infra-red spectrometry when a very high etendue is required and new high-
dispersion infra-red transmitting materials may be employed. It has the merit of
simplicity and portability. Modulation of the focused wavelength can be by rock-
ing the reecting mirror or by vibrating the grid in its own plane perpendicular to
its rulings.
of the lens, only one small wavelength band is in focus and fully transmitted. They
achieved some success in crude spectrophotometry using a Welsbach lamp and
observed radiation out to 0.1 mm wavelength.
It was developed considerably, at rst into an ultra-violet spectrometer
4
but
more lately into a high-resolution infra-red spectrometer
5
by manually guring a
lens with a high infra-red dispersion so as to correct it for spherical aberration. One
or more aplanatic elements were added of the same material to improve the relative
aperture, and it imaged a ne grid on to itself at strictly 1 magnication. The grid
was made to oscillate, mechanically or optically, perpendicular to the grid elements
through one half-space and this modulated the wavelength for which the focus was
good. It was scanned in wavelength by moving the lens along the optic axis. It had
the advantage of a very large etendue, unlimited by the usual constraint of a xed
L-R product.
Its demerits were (1) the line prole, which as a result of false resolution
6
is a
sinc function with its huge attendant side-lobes, and (2) it suffered the multiplex
disadvantage, whereby radiation is received fromall spectral elements but only one
is modulated. Consequently, like the Michelson Fourier spectrometer, it is effective
only when detector noise is the limiting noise-source.
The device is not yet well developed and more experiment, in modulation meth-
ods and especially in lens design to improve the dispersion and line prole while
retaining the resolution, is probably needed in order to realise its full potential.
4
C. S. Forbes, L. J. Heidt & L. W. Spencer, Phil. Mag., 5 (1934), 253.
5
J. F. James & R. S. Sternberg, J. de Phys., 20 (1967), C2-326; L. McNaughton, Ph.D. thesis, University of
Manchester (1969).
6
H. H. Hopkins, Proc. Roy. Soc., A, 231 (1955), 98.
8
The plane grating spectrograph
8.1 The shape of a monochromatic line spectrum
In Chapter 6 the expression was derived for the diffracted amplitude from a diffrac-
tion grating:
I( p) = Na sinc(Npa)
_
nsinc(pn)
1
a
III
1,a
( p)
_
. (8.1)
where a is the interval between rulings and n is the ruling width.
The intensity of light diffracted by a grating is then
I ( p) = N
2
n
2
sinc
2
(Npa).
and this is modulated by the intensity represented in the square bracket, that is, by
sinc
2
(pn) repeated at p = 1,a. 2,a. 3,a. . . . . n,a.
The rst factor gives the shape of each spectrumline, a very narrowsinc
2
function
determined by the total ruled width.
In the second factor (Fig. 8.2b), the rst component of the convolution de-
scribes the broad sinc function that controls the intensities of the various orders of
diffraction, and the second component gives the values of sin , at which diffrac-
tion maxima occur.
Figure 8.2c shows the complete graph describing the diffracted intensity.
From the shape of the line and using Rayleighs criterion
1
as a guide, we can
deduce the resolving power of the grating. Maxima of the pattern, spectral lines
that is, occur when p = n,a. Using the sinc
2
function of Fig. 8.2a, the shape of the
line is
I () = sinc
2
(Na sin ,).
1
Lord Rayleigh, Phil. Mag., 8 (1879), 261.
63
64 The plane grating spectrograph
w a
Na
Figure 8.1 The primitive geometry of the grating. The transparent width of each
slit is n and the distance between adjacent rulings is a. The overall width of the
ruled area is Na and these three parameters are used in the derivation of the grating
equation.
Rayleighs criterion for resolution required a 20% fall in intensity between two
maxima. This was doubtless due to the fact that when two identical sinc
2
functions
are added, each displaced so that its central maximum lies on the rst minimum
of the other (Fig. 8.3a), the resulting curve (Fig. 8.3b) has a 20% dip from the two
maxima. This follows from the simple fact that the value of sinc
2
(x) at x = ,2 is
0.405.
Apart from anything else, this is mathematically convenient. At a maximum:
pNa = n sin = n,Na.
where n is any integer the order of diffraction. The adjacent zero is when
p = n +,N sin = (n +1,N),a.
If this is the same as the maximum of wavelength + then
sin = (n +1,N),a = n( +),Na.
whence
, = Nn = R = the resolving power.
This is the theoretical maximum resolving power available. Two comments are
required.
(1) The imperfections of manufacture make this practically unrealisable. In practice one
may achieve 8590% of this. Again in practice it is rare for a grating to be worked
anywhere near its maximum theoretical performance. For example, a 200 mm ruled
width at 1200 rulings/mm implies 240 000 rulings (with a total length of 36 km) and it
is rare indeed for a spectrometer to be required to give this resolving power over a wide
spectral range.
(2) Aspectral feature can be shown to comprise two separate lines even though they are not
resolved: the line shape changes, becoming fatter for example, or asymmetrical when
adjacent singlet spectrum lines are not.
8.1 The shape of a monochromatic line spectrum 65
(a)
(b)
(c)
I ( p)
3

a
3
a
P 2

a
1

a
2
a
1
a
0
Figure 8.2 (a) The shape of the spectrumline when strictly monochromatic light is
diffracted. This is the rst factor in the grating equation and the width (FWHM) is
determined by the whole ruled width Na of the grating. (b) The second component
of the convolution in Eq. (8.1). The amplitudes of teeth of the Dirac comb are con-
trolled by the overall sinc
2
function derived from the width, n, of one ruling. The
abscissae here are the sines of the angles of diffraction divided by the wavelength
of the diffracted light. (c) When each tooth of the comb is replaced by the sinc
2
function illustrated in (a), the diffraction pattern of the grating is displayed.
66 The plane grating spectrograph
(a)
(b)
Figure 8.3 (a) Two sinc
2
functions representing two adjacent spectrum lines lie
with one principal maximum lying on the rst zero of the other. (b) When the
two are added the combined intensity prole has a 20% dip between the principal
maxima.
A more elaborate criterion may be taken as a measure of resolution, similar to
the optical transfer function used for photographic lenses, in which an image is
formed of an object that is a grid of alternate black and white lines. The visibility
2
of the fringes in the image, as a function of spatial frequency, is called the contrast
transfer function and is regarded as a measure of the spatial lter response of the
lens.
In the spectrograph a set of EdserButler fringes
3
may be observed, using a
Michelson interferometer rather than the usual FabryPerot etalon as the fringe
generator. With a small diameter source of white light, well collimated as it passes
through the interferometer, the intensity at the entry slit of the spectrograph will
vary sinusoidally with wavenumber or approximately with wavelength over a
short range of the spectrum. The visibility of the EdserButler fringes as a function
of path difference in the interferometer is then a measure of the spectral transfer
function, which in the absence of aberrations should fall to zero when the Michelson
path difference is equal to the ruled width, W, of the grating.
8.1.1 The grating equation
The elementary analysis assumed that light was incident normally on to the grating.
If the incidence is oblique, at an angle to the grating normal, there is an additional
phase term
4
covering the whole aperture and then a = ,(sin +sin ) or, more
2
Dened as : = (I
max
I
min
),(I
max
+ I
min
).
3
E. Edser & C. P. Butler, Phil. Mag., 46 (1898), 207.
4
As in Fig. 6.2.
8.2 Blazing of gratings 67
b
a
reflecting face
scarp face
Figure 8.4 A section through the reecting surface of a blazed grating. The height
of the step between rulings, the scarp face, is one half-wavelength of the blaze
or rst-order blaze wavelength. At this wavelength incident light at an angle b to
the grating normal is all reected at this same angle and in rst order of diffraction.
This is the angle and wavelength of maximum efciency.
familiarly, the grating equation:
sin +sin = n,a. (8.2)
Remember that and are on the same side of the grating normal.
8.2 Blazing of gratings
In early gratings where the width of each ruling was about 1,2 the ruling interval,
i.e. when n was a,2, only odd orders were present, the even orders having been
suppressed by the zeros of the broad sinc function lying on top of the diffraction
maxima.
In practice the grating rulings have the same width a as the spacing and are
inclined at the blaze angle b of the grating, so that a cross-section of the surface
has the appearance of Fig. 8.4. The effect of this blazing on a reection grating
can be described by changing in Eq. (6.2) the amplitude factor of one ruling from
H
a
(x) to H
a
(x)e
(2,)2x sin b
. The factor of 2 allows for the reection. Replacing
2 sin b, by q changes the corresponding Fourier transformfactor in Eq. (6.3) from
nsinc(pn) to nsinc(pa) ( p q) which, by the shift theorem (Eq. (5.3)) is
the same as nsinc(( p q)a). The maximumof the diffraction pattern then comes
when p = q, or =2b.
Thus for one wavelength, the blaze wavelength, all the incident light is diffracted
into rst order and the zero order is suppressed as are all other orders, by zeros of
the sinc function.
For neighbouring wavelengths the enveloping broad sinc function changes its
width at the same time as the diffraction peaks move in the opposite direction so
68 The plane grating spectrograph
that the maxima are not at the peak transmission. The efciency is less, and some
light is lost into other orders and scattered inside the spectrograph, but nevertheless,
as a rule-of-thumb, the efciency is greater than 50% over a range of wavelengths:
_
2
2n +1
_

b
- -
_
2
2n 1
_

b
.
where n is the order of diffraction and
b
the blaze wavelength.
5
8.3 Apodising
It is the rst factor in the diffraction equation which describes the shape of each
spectrum line. A purely monochromatic line is thus presented as a sinc
2
function
of FWHM ,Nn and with secondary maxima on either side of approximately
4% of the height of the parent. It is possible that these secondary maxima may
disguise real satellite lines or impersonate ctitious ones, and some effort has
been devoted to their suppression. The sinc
2
function occurs when the grating is
uniformly illuminated, giving a H
a
(x)-factor in the aperture function. This factor
can be modied by masking the grating with an aperture of a shape such that its
Fourier transform alters the shape of the line prole. A diamond-shaped mask for
instance will change the rst term of the aperture function from H
Na
(x) to A
Na
(x)
where the A-function, or triangle function, is the convolution of two H-functions:
A
a
(x) = H
a,2
(x) H
a,2
(x).
Here, two top-hat functions each of width Na,2 are combined to form A
Na
(x).
By the convolution theorem the Fourier transform of A
Na
(x) is
(N
2
a
2
,4)sinc
2
(pNa).
and the intensity distribution in a spectrum line is
(N
2
a
2
,4)
2
sinc
4
(pNa).
This has the merit of suppressing the secondary maxima from 4.5% to 0.2%
of the principal line, a possibly useful gain if the ne structure of a line is being
examined. The concomitant demerits are
(1) that by covering half the grating area, half the intensity is lost and
(2) that by restricting the effective width, the resolving power is reduced by 40%.
There are various shaped masks, known as windows, illustrated in Fig. 8.5,
which will redistribute the power in the secondary maxima in various ways and the
5
For ne details of the often complicated efciency curves of diffraction gratings see: C. Palmer, Diffraction
Grating Handbook, 6th edn. (Rochester, NY: Newport Corporation, 2005).
8.4 Order overlap and free spectral range 69
Figure 8.5 Apodising masks for reection gratings. The Hanning mask with
the sinusoidal variation of amplitude across the face of the grating has many
advantages, but the area is still open to experiment. The light which fails to reach
the principal maximum can be redistributed almost at will among the secondary
maxima.
problem of side-lobes is one that is common to all spectroscopy. The Hanning
and Hamming windows of communication theory
6
are typical but in the normal
spectrograph the combination of the rectangular grating aperture and the usually
circular pupils which are superimposed on it compromise all these windows.
The masking process is generally known as apodising
7
and there are legions
of apodising masks known and advocated, chiey by their inventors.
8.4 Order overlap and free spectral range
Despite blazingandother improvements it is not possible tosuppress bythese means
unwanted radiation from other orders of diffraction from appearing in a spectrum.
A wavelength diffracted in rst order is accompanied by ,2 in second order, by
,3 in third order and so on, to the confusion of the investigator. It may well be that
the detector is insensitive to this unwanted radiation but in general precautions must
be taken to exclude it. Usually simple colour ltering is adequate and in particular in
the visible region in rst-order working, a glass which absorbs UVlight is sufcient
6
See R. B. Blackman & J. E. Tukey, The Measurement of Power Spectra (New York: Dover, 1958).
7
From the Greek without feet. The same apodising process is used in Fourier-multiplex spectrometry where
it is much more effective and more necessary, since the raw instrumental prole there is a sinc function. In
that case apodising is part of the computation process and involves interferogram samples being appropriately
weighted to form the apodising mask.
70 The plane grating spectrograph
to allow an octave of the spectrum from 8000 to 4000 to be photographed
unambiguously. However, it may be necessary to use a higher order with an ap-
propriately higher blaze angle to obtain the necessary resolution, and in this case
narrow-band lters are required to suppress both higher and lower orders.
8.4.1 Free spectral range
If a wavelength is to be examined in nth order, then sin i +sinr = n,a. At
these same angles i and r, n will appear in rst order, n,2 in second order, and
generally, n,m in mth order. To avoid confusion all these other wavelengths must
be excluded. The extent of the spectrogram, the range of wavelengths that is, which
can be examined without confusion in nth order, stretches from to L at
which point appears again in (n +1)th order. L comes from
(n +1) = n( L) whence L = ,n.
8.5 Grating ghosts and periodic errors
8.5.1 Rowland ghosts
It sometimes happens that the rulings of a grating are not separated by precisely
equal spaces. Particularly in older ruling engines which relied on mechanical pre-
cision, a periodic error in the master-screw would cause a cyclic variation in the
position of successive rulings. When this happens a monochromatic spectrum line
may be accompanied on each side by faint satellite lines known as ghosts, some-
times by a whole series of themof steadily decreasing intensity. There is an analogy
with the side-bands of a frequency-modulated radio transmission, where the side-
bands contain the information signal.
The problemmay be described using Fourier theory. The Dirac comb which rep-
resented the positions of the rulings is replaced by a set of -functions in which the
nth is displaced from its proper place by a cyclically varying amount sin(2sna).
is the amplitude of the periodic error and 1,s, the pitch of the lead-screw of the
ruling engine, is its period. The aperture function for the grating then replaces the
Dirac comb III
a
by
A(x) =

n=
[na + sin(2sna)]. (8.3)
and its Fourier transform is
I( p) =

n=
e
2i p[na+ sin(2sna)]
. (8.4)
8.5 Grating ghosts and periodic errors 71
with p = sin , as usual. From Bessel function theory we have the Jacobi
expansion:
e
ix sin y
=

m=
J
m
(x)e
imy
.
where J
m
(x) is the mth-order Bessel function. Using this with x = 2pn and
y = 2sna, the sum becomes
I( p) =

n=
e
2i pna

m=
J
m
(2p)e
2imsna
. (8.5)
which, when rearranged is
I( p) =

n=

m=
J
m
(2p)e
2ina( p+ms)
. (8.6)
We suppose now that the amplitude, , of the periodic error is small compared
with the grating constant a. For small values of the argument the Bessel functions
approximate to
J
0
(x) = 1. J
1
(x) =
x
2
. J
2
(x) =
x
2
4
. etc.
and
I( p) =

n=
e
2inap
+

n=
p e
2ian( p+s)
+

n=
p e
2ian( ps)
+

n=

2
p
2

2
e
2ian( p+2s)
+

n=

2
p
2

2
e
2ian( p2s)
+ (8.7)
This represents an innite series of Dirac combs. The rst is the one which
appears in Eq. (8.1), representing the perfect grating. The second pair represent
Dirac combs with amplitudes attenuated by a factor p and with periods 1,( p +s)
and 1,( p s). The third pair are attenuated by
2
p
2

2
, with periods 1,( p +2s)
and 1,( p 2s). These describe the ghost lines, and in rst order the strongest
are separated from the parent line by L =
2
,(lead-screw pitch). The effect is
shown in Fig. 8.6.
8.5.2 Lyman ghosts
These are faint images of a line which appear apparently at sub-integer orders and
are chiey found in XUV spectroscopy when older, mechanically ruled concave
gratings are used in rst order. They may be ascribed to defects on the ruling tip
72 The plane grating spectrograph
x
2
x
2
4
(x = 2 p)
Figure 8.6 Rowland ghosts surrounding a spectrum line. These are separated by
amounts which depend on the pitch of the ruling engine lead-screw and by ampli-
tudes which depend on the amplitude of the pitch error. In a modern ruling engine,
where the position of the ruling tip is optically controlled, there should be no such
periodic error and hence no Rowland ghosts.
which make an irregular but constantly repeated ruling prole so that, perhaps as a
result of aliasing, the III(x)-function which describes the grating appears to contain
another such function in its interstices.
8
This imitates a grating with two, three or
more times the number of rulings and consequently two, three or more times the
dispersion.
Lyman ghosts are a possible but very unlikely source of confusion unless very
large dynamic range and sensitivity are factors in the investigation.
8.6 The complete grating equation
A complete description of the diffraction grating must take account of the length
of the rulings as well as the ruled width. The Fraunhofer diffraction pattern is then
two dimensional and the two-dimensional Fourier transform must be used with
allowance made for incidence which is oblique in both axes.
8
Crudely, if the H-function which naively describes one ruling were replaced by two narrower H-functions,
the grating would have twice the number of rulings, not necessarily uniformly spaced. If, for example, the
H-functions were separated by 1,3 of the grating constant, there would be a contribution to the spectrum at
order 1,3.
8.6 The complete grating equation 73
R
2
R
1
B
P(x, y, 0)
A
R
1
R
2
y
x
z

Figure 8.7 The geometry of incident and diffracted rays. R

1
and R
2
are incoming
rays from a collimated beam and R

1
and R

2
are the same rays after reection.
Perpendiculars from the reection point of R
2
on to R
1
and R

1
lie in the incoming
and outgoing wavefronts and cut off lengths which add to give the path difference
between the two. When the reection point is one ruling away from the origin the
path difference must be one wavelength.
However, the alternative derivation of the full grating equation is both simple
and instructive. Suppose that the grating occupies the plane z = 0 and that the
rulings are parallel to the y-axis. Consider two parallel rays from the collimated
incident beam incident on the grating, ray R
1
at the origin and ray R
2
at the point
P(x. y. 0) (Fig. 8.7). Suppose that their direction cosines are l
1
. m
1
. n
1
and that the
directioncosines of the correspondingdiffractedrays R

1
and R

2
arel
2
. m
2
. n
2
. Byan
elementarytheoreminCartesiangeometry, perpendiculars fromthe point P onto R
1
and R

1
intercept lengths PA = l
1
x +m
1
y +0z and PB = l
2
x +m
2
y +0z. This
path difference between the two rays must be an integer n number of wavelengths.
Thus,
AO + BO = (l
1
+l
2
)x +(m
1
+m
2
)y = n.
where n is the order of diffraction.
This must hold for any value of y, so that (m
1
+m
2
) = 0 and (l
1
+l
2
)x = n.
If point P is on a ruling adjacent to the y-axis, then the grating equation appears:
(l
1
+l
2
)x = n,a. or (sin i +sinr) cos = n,a.
8.6.1 Distortion of line shapes
Some important consequences come from this equation. Consider rst of all the
chief-rays of wavelength incident on to the grating at its vertex from points on
74 The plane grating spectrograph
a
(a,i)
(a, r)
i
r
0
Figure 8.8 The diffraction of rays fromdifferent points on a straight entry slit. The
complete grating equation requires a conical fan of diffracted rays, which in turn
results in a curved image of a straight entry slit.
a straight entry slit parallel to the y-axis. All the x-direction cosines, l
1
, are zero.
The diffracted chief-rays will therefore all have the same value of l
2
, so that they
generate a circular cone with its axis on the x-axis of coordinates. If the focal plane
of the spectrograph is parallel to the grating surface, the images of the entry slit will
occupy a hyperboloid, where the focal plane intersects the chief-rays. Alternatively,
if the focal plane is perpendicular to the surface and parallel to the rulings, the
images will be arcs of circles. This curvature of line images is potentially a major
cause of loss of resolution. The image curvature varies with wavelength and in the
calibration of the spectrograph due account must be taken of it.
The proper descriptive tool here is spherical trigonometry, in which the grating
normal lies on the equator and the rulings are parallel to the polar axis. In Fig. 8.8,
angles like i and r are longitudes and chief-rays approach the centre of the unit
sphere from various latitudes and longitudes. A chief-ray coming from latitude
and longitude i is diffracted to latitude and longitude r, connected by
sinr = (n,a) sec sin i.
and for small values of the approximation sec = 1 +
2
,2 is adequate. If
the equatorial chief-ray is diffracted to r
0
, the inclined one arrives at longitude r
where
sinr sinr
0
= (n,2a)
2
.
r = (n,2a)
2
sec r
0
. (8.8)
For small the curve of against r is a parabola and describes the shape of
the Gaussian image of a straight input slit. In practice this may sometimes be
8.6 The complete grating equation 75
i
,
Figure 8.9 An entry slit which is the arc of a circle will produce an image at one
wavelength which is a continuation of the same circle. In a scanning monochro-
mator, images of one half-circle will appear on the other half-circle at whatever
wavelength is returned from the tilted grating. The wavelength scanning is then
done simply by rotating the grating about an axis parallel to the rulings.
compromised by the aberrations of the system but is nevertheless a useful guide to
the correction process. The constant of the parabola changes with wavelength and
at high resolution each wavelength requires separate treatment to straighten its line
image.
8.6.2 The circular slit theorem
Another theorem, chiey of interest to monochromator constructors, is worthy of
mention in this context.
In Fig. 8.9, consider a point at latitude 0, longitude 0 and suppose that this makes
an angle i with the optic axis of the spectrometer. Draw a small circle of angular
radius with its centre at this point. The coordinates (. ) of a point on this circle
are connected by another well-known theorem in spherical trigonometry:
cos cos = cos .
Two diametrically opposite points on the circumference of the small circle then
have coordinates (. ) and (. ), and if they are also points where incoming
and outgoing chief-rays cut the sphere, then
cos [sin(i +) +sin(i )] = 2 cos sin i cos = m,a.
or
2 cos sin i = m,a.
76 The plane grating spectrograph
Since i is constant, every diametrically opposite pair of points on the small circle
are conjugate points for the same wavelength. An input slit which is the arc of a
circle will result in an image at wavelength on another arc of the same circle.
Different wavelengths will provide images at the same circular output slit as the
grating is turned about its axis.
8.7 Differential dispersion
In normal practice the spectrograph slit widths are not determined by the ultimate
resolving power theoretically available. The entry slit width in such an instance
would be such that the rst zeros of its diffraction pattern the sinc
2
function
would fall at the edges of the grating. However, in a spectrograph the resolution
may be set by the width of the CCD pixels or the average size of the silver halide
crystals of a photographic emulsion. The slit width is then chosen to match this
width and there is no resolution to gain and much light to lose in making it narrower.
From the grating equation
(sin i +sinr) = n,a
we get
cos i di = n d,a.
whichrelates the resolutiondtothe angular slit widthdi andfor a xedwavelength,
cos i di = cos r dr. (8.9)
which relates the width of the input slit to the width of its image. This is the lateral
magnication, which is a function of wavelength, and must be taken into account
during the design.
8.8 Mounting congurations
Since the time of Bunsen and Kirchhoff and until the end of the nineteenth century
the optical arrangement for a spectroscope had comprised a slit, a collimator, a
disperser, and a telescope. For photography the eyepiece could be replaced by a
photographic emulsion. It was Ebert, following Rowlands production of research-
quality diffraction gratings, who proposed an achromatic spectrograph using a
single spherical mirror both as collimator and telescope and eliminating the un-
wanted longitudinal dispersion inherent in a simple lens system.
9
9
H. Ebert, Annalen der Physik, 38 (1889), 489.
8.8 Mounting congurations 77
entry slit
grating
Ebert mirror
photographic
plate
Figure 8.10 The EbertFastie mounting. The rst achromatic method of mounting
a high-resolution spectrograph, with almost miraculous correction of the gross
aberrations. However, it proved to be endishly difcult to achieve good optical
alignment and it consequently lost ground to the

CzernyTurner mounting with
similar achromatic properties, where the separate mirrors gave more degrees of
freedom and greater exibility in practical alignment.
Strangely it seems to have been ignored until it was rediscovered by Fastie,
10
and meanwhile

Czerny and Turner had replaced the two traditional lenses by
two spherical mirrors.
11
Littrow similarly had evolved an arrangement using
a single spherical or paraboloidal mirror which would yield diffraction-limited
resolution.
8.8.1 The EbertFastie mounting
One of the features of this mounting (Fig. 8.10), which it shares with the

Czerny
Turner mounting, is the partial correction of tangential coma, leaving spherical
aberration and astigmatism as the two main aberrations. A proper choice of pos-
ition for the grating ensures a at meridional eld. Astigmatism is tolerated be-
cause only the meridional focus is required and distortion is ignored since the
spectral dispersion is in any case non-linear. Although it is capable of yielding
a high resolution at a fairly high numerical aperture, the restriction imposed by
the single mirror makes the optical alignment of the Ebert mounting unnecessarily
difcult and it has largely been discarded in favour of the

CzernyTurner mount-
ing in most commercial designs. Only two portions of the mirror are used there
can be no overlap between incoming and outgoing beams and a mask must be
inserted in the centre of the aperture to prevent undispersed light falling on to
the detector. This reduces the device essentially to the same conguration as the

CzernyTurner mounting and the EbertFastie arrangement will not be considered

further.
10
W. G. Fastie, J. Opt. Soc. Am., 42 (1952), 641 and 647.
11
M.

Czerny & A. F. Turner, Z. Phys., 61 (1930), 792.
78 The plane grating spectrograph
off-axis paraboloid
mirror axis
exit slit (below the plane of the diagram)
entry slit
(above the plane of
the diagram)
grating
Figure 8.11 The Littrow mounting. Like its prism counterpart it returns the spec-
trum through the same optical path as the incident radiation and focuses the light
close to the entry slit. This version may be constructed with the grating rulings
parallel to the diagram, or perpendicular in which case the grating is rotated
instead of tilted as shown here. There is no appreciable difference in the aberration
amplitudes or in the spectral resolution available.
8.8.2 The Littrow mounting
This is illustrated in Fig. 8.11. The chief application of this mounting is as a
monochromator to give an aberration-free image of a short straight entry slit, and
for this reason an off-axis paraboloidal mirror is employed. To keep the eld angle
as small as possible an over-and-under slit arrangement may be used, with an
exit slit for a monochromator and a photographic plate for a spectrograph. As the
full coma of a paraboloid is present, the mounting offers no advantage over the

CzernyTurner mounting as a spectrograph. As a monochromator, high resolution

is available provided the entry slit is short
12
and there is the curvature of a line
image which follows from the full grating equation.
As a monochromator, the performance is the same when the grating rulings
are parallel to the plane of Fig. 8.11 and the grating is turned instead of tilted.
The dispersion is then perpendicular to the plane of the gure and the spectrum is
formed next to the entry slit. There may be some structural advantage in such an
arrangement but in neither instance does the mounting have any particular merit as
a spectrograph when compared with the

CzernyTurner mounting. Amajor demerit
is the need for an off-axis paraboloid, which in turn implies an inordinately large
and heavy mirror or one which has been cut to shape from such a mirror. Then
there exists the possibility of distortion of the surface gure because of this cutting
process, with a consequent loss of resolution. This is a point for discussion with
the mirror manufacturer.
An unusual application for the Littrow mounting is in the monochromatic imag-
ing of an extended object, something which is generally done with interference
lters. Because the system is virtually stigmatic close to the paraboloid axis,
12
Two per cent of the focal length is usual.
8.8 Mounting congurations 79
entry slit
grating
exit slit
concave mirror
Figure 8.12 The Pfundt mounting. It enjoyed some success as a monochroma-
tor mounting but an extended eld requires a large prism and consequently an
inconveniently large obstruction to the collimated light.
two-dimensional resolution is possible and a rotating mirror may be used to sweep
an image of a luminous object such as the Sun past the entry slit while a
similar mirror or another part of the same mirror reects the output slit syn-
chronously across the eld of a camera or telescope. Various auxiliary optical
arrangements using mirrors or rotating glass cube prisms are possible and permit
the two-dimensional resolution of an object at any wavelength of choice. A ray
trace indicates that a spectral resolution of 0.05 is achievable with a 2 m focal
length, a 75 mm ruling width of grating and 15 m slit widths.
A similar adaptation will convert a Littrow or a

CzernyTurner monochromator
into a spectrograph with a spectral range of a few angstroms. A rotating square-
section prism as long as the exit slit and parallel to it can be placed just before the
exit slit (i.e. inside the spectrometer) with an electric motor to spin it rapidly about
its axis. This sweeps an image of the entry slit across the exit slit and reveals the
local structure of a monochromatic emission line multiplet. The detector output
swept synchronously across a CRT monitor will show real-time changes to the line
structure when, for example, the source temperature is changing or when a magnetic
eld is applied to it. Self-absorption of the mercury green line as the source warms
up is easily demonstrated with this technique, for example.
8.8.3 The Pfundt mounting
This is basically a variant of the EbertFastie mounting where the off-axis aberra-
tions are reduced by using Newtonian mirrors for inserting and extracting the light.
It is illustrated in Fig. 8.12. It is of little practical use as a spectrograph because of
the narrow eld and consequently low spectral range available, but it is a possible
alternative to the Littrow mounting when low astigmatism is important in line im-
ages. However, the astigmatism does not vanish completely and its chief virtue is
its small mirror diameter and consequent low weight, not to mention cost.
80 The plane grating spectrograph
entry slit
grating
focal
surface
M1
M2
Figure 8.13 The basic

CzernyTurner mounting. It is often constructed with a
symmetrical layout, the grating vertex being at the midpoint between the incoming
and outgoing chief-rays, but proper coma correction, according to the Rosendahl
criterion, requires sometimes a markedly asymmetric arrangement.
Clearly it can be used with

CzernyTurner mirrors and then has zero astigma-
tism. It should probably be called the Newtonian mounting and, like the Littrow,
is stigmatic. With paraboloidal mirrors it also has zero spherical aberration. The
practical limit, as always, is the curved image of a straight slit which it gives. A
possible minor demerit is the necessary obstruction of part of the grating by the
Newtonian mirrors with the consequent loss of etendue.
8.8.4 The

CzernyTurner mounting
This mounting, illustrated in Fig. 8.13, is by far the most popular arrangement
for plane reection grating spectrographs. The two mirrors are used with off-axis
chief-rays and there is consequently some coma in the incoming beam. If there were
a plane mirror in the grating position reecting the light to the second mirror, the
coma would be exactly corrected, leaving only astigmatismand spherical aberration
in the nal converging beam. However, after diffraction the outgoing collimated
beam is either wider or (more correctly) narrower than the incoming beam, so that
if the systemis made symmetrical as many early spectrographs were the coma is
only partially corrected. The result nevertheless at focal ratios of F,12F,16 gave
satisfactory resolution for most purposes.
In the

CzernyTurner spectrograph the grating tilt should normally be towards
the input side. This is an elementary precaution because the output beams are
then narrower and allow the greatest length of spectrum to be focused from a
given output (M2) mirror diameter.
13
Various alternative arrangements, such as
the over-and-under mounting, where the slit, ruling and dispersion directions are
perpendicular to the plane of the optics, have been considered from time to time
both in EbertFastie and

CzernyTurner congurations, but they generally confer
13
Or conversely, allow the smallest possible mirror to collect light without vignetting from a given spectral
bandwidth.
8.8 Mounting congurations 81
no obvious advantage, and ray tracing will quickly reveal the huge extra aberration
and varying line-tilt incurred.
With the standard layout there are more free design parameters available than are
necessary to ensure best performance, and a general rule in design is to keep the off-
axis angles as small as possible consistent with the length of spectrum that is to be
recorded, and to make the input and output chief-rays parallel. This latter condition
greatly simplies the practical detail of design and construction of the instrument.
A note of caution is needed at this point. There must be enough margin between
the various ray bundles to allow bafes to be included. Attempts to save space by
folding the optical paths with plane mirrors may compromise this feature and it
is too important to be neglected or dismissed. For this same reason the so-called
crossed

CzernyTurner mounting is to be regarded with suspicion.
8.8.5 The Rosendahl condition
There was a major improvement in optical design when Rosendahl analysed the
aberrations of the

CzernyTurner conguration and showed that, provided a certain
condition, the Rosendahl cos-cubed condition were fullled, the tangential coma
of the system would be completely corrected at one wavelength.
14
The condition
is that
cos
3
i
cos
3

sin =
cos
3
r
cos
3

sin . (8.10)
where i and r are as usual the incident and diffraction angles, and and are the
eld angles of the incoming and outgoing beams at the M1 and M2 mirrors. The
condition can be derived from elementary Seidel theory as follows.
In Fig. 8.14, parallel rays starting from the grating and reected from the M1
mirror will converge to a comatic image at the entry slit, the marginal rays showing
an angular coma
coma
. Conversely, marginal rays starting precisely from the entry
slit will arrive at the grating edge at this same angle
coma
to the paraxial ray. After
diffraction this angle will be multiplied by the factor (cos i,cos r) of Eq. (8.7)
and this must be added to the coma produced by the output mirror M2. The total
tangential coma is then
C
m
= 3F
1
y
2
1
tan (cos i , cos r) +3F
2
y
2
2
tan . (8.11)
where subscripts refer to the respective mirrors.
The coma coefcient F (Eq. 4.8) is given by
F = f
2
,4 Z(1 e
2
),8 f
2
.
14
G. R. Rosendahl, J. Opt. Soc. Am., 52 (1962), 412.
82 The plane grating spectrograph
r
i

Figure 8.14 The angles involved in the calculation of the coma of the

Czerny
Turner mounting. The rays here are of course the chief-rays of each bundle.
For paraboloidal mirrors, which as we shall see later are preferable to spherical
mirrors, this reduces to f
2
,4.
The geometry of the

CzernyTurner arrangement provides that y
1
=
(W,2)(cos i,cos ) and y
2
= (W,2)(cos r,cos ) and if as usual (though not neces-
sarily) both mirrors are identical, then F
1
= F
2
= F and the total tangential coma is
C
m
= (3,4)F[(cos i,cos )
2
(cos i,cos r) tan (cos r,cos )
2
tan ]. (8.12)
and when C
m
= 0 this is equivalent to the Rosendahl condition.
The condition for vanishing of tangential coma derived here is more general
than the Rosendahl condition as it allows for M1 and M2 mirrors of different
focal lengths and it also allows exact computation of the coma when there is some
offence against the Rosendahl condition. The condition holds strictly only for one
wavelength and it is only the tangential coma which is corrected. Sagittal coma
remains, albeit somewhat reduced. There is no foreshortening of the aperture on
diffraction and the sagittal coma, one-third of the tangential coma, is F(L,2)
2
tan .
This is converted to F(L,2)
2
tan (cos i,cos r) on diffraction, and supplemented by
F(L,2)
2
tan from the M2 mirror. When the Rosendahl condition is met this gives
C
s
= F(L,2)
2
tan
_
1 (sin ,sin )
2,3
_
. (8.13)
which is substantially less than the coma of a paraboloid at the same eld angle
away from the zero-coma wavelength.
8.8.6 The generalised Rosendahl condition
To satisfy the Rosendahl condition we abandon the cherished concept of the sym-
metrical

CzernyTurner conguration and allow the grating to be nearer (usually)
8.8 Mounting congurations 83
to the input-side optic axis. If we also make allowance for the M1 and M2 mirrors
to be of different focal lengths, the condition can be rearranged in a form more
convenient for computation:
_
cos i
cos r
_
3
=
_
f
1
f
2
_
2
_
cos
cos
_
3
_
sin
sin
_
. (8.14)
Although it is not necessary for the input and output chief-rays to be parallel, it is
convenient and it gives us another constraint:
r i = 2( +) = Q. (8.15)
With r replaced by Q +i , the grating equation becomes
2 cos(i + Q),2 sin(i Q),2 = ,a. (8.16)
where is the desired zero tangential coma wavelength. FromEq. (8.14), i is found
by iteration. From Eq. (8.13), r is found. Then by substituting values in Eq. (8.12),
is obtained to any degree of accuracy by another iteration. Finally the grating tilt,
t , is given by t = ,2 2 i .
8.8.7 The at-eld condition
While photographic plates and lms can be bent gently to t the eld curvature of
an optical instrument, the crystalline nature of a CCD or similar detector requires
an accurately at focal surface. The expression for tangential eld curvature of a
convex mirror is (Eq. 4.12)
C
t
=
2
f
+
3Z
f
2
+
3
4
Z
2
(1 e
2
)
f
3
.
and for a spherical mirror C
t
is zero when Z = 2 f (1 1,

3), which in practice

means that Z = 0.843 f
2
, where f
2
is the focal length of the output M2 mirror.
Similarly, for a paraboloid the eld is at when Z = 2 f ,3.
In each case, the pupil of the mirror is the grating aperture, and these expressions
give the proper dispositions of the M2 mirror following the grating.
The computation does not end here. We must consider the possibility that the
optical train will continue beyond the rst focus, in which case it is desirable that
the output chief-ray at the centre of the eld be perpendicular to the focal plane. The

CzernyTurner conguration is a decentred system where the pupil (the grating)

is displaced laterally from the optic axis of the M2 mirror. The problem of the
Seidel aberrations of a decentred system is resolved by considering two centred
pupils of different diameters and tracing the two marginal rays from one side at
different eld angles. Their intersection points mark the focal curve for a decentred
84 The plane grating spectrograph
grating
mirror
camera
CCD
collimator
entry slit
Figure 8.15 The rst of two simple mountings which use an oversized grating and
make the camera pupil the controlling aperture of the system. Its chief merit is the
compact design and the use of standard optical components, thus requiring little
design effort or exceptional expense.
pupil. It can be shown that the effect of a decentred stop is to tilt the focal surface
about the paraxial focus, and that for a paraboloid the tilt is exactly the tilt of the
central chief-ray.
15
As remarked then, this is a fortunate result and means that with
a paraboloidal M2 mirror the focal plane can be re-imaged and demagnied with a
concomitant useful increase in numerical aperture without loss of resolution. The
details of this are in Chapter 13.
8.8.8 Alternative spectrographic mountings
In general the grating is the iris of the system and it is the grating constant
which chiey determines the systems performance. There are simple alternative
congurations which the author has found suitable for the examination of very
feeble sources at moderate (R = 10005000) resolution and high numerical aper-
ture, and which are radical departures from the usual rules for spectrograph design.
In this case an oversize grating is used and it is the camera iris which determines the
performance of the system. The optical diagrams are shown in outline in Figs. 8.15
and 8.16. The important design parameters are the grating constant, its position and
its tilt. In each case the chief-ray of the centre of the spectrumleaves the grating close
to the normal to the surface and the grating blaze must be towards the input mirror.
The wavelength falling on a pixel is proportional to sinr: i.e. sinr = ,a sin i
and i is xed. Thus sinr = (
c
),a where
c
is the wavelength at the centre of
the spectrum. Alternatively, =
c
+a sinr.
The spectral range available is set generally by the focal length of the camera
lens, the width of the detector and its number of pixels. The position of the camera
pupil is determined by the need for the main mirror to image this pupil on to the
distant light source the night sky for example so that the camera pupil should
be at the rear focus of the mirror. The coma and astigmatism of the mirror are
15
J. F. James, J. Mod. Opt., 41 (1994), 2033.
8.8 Mounting congurations 85
entry slit
grating
CCD
F/2 double-Gauss 35 mm camera lens
paraboloidal
mirror
Figure 8.16 This is an alternative simple mounting using an astronomers 120 mm
paraboloid as the collimator. Because of the exceptionally high input-side focal
ratio, the coma resulting from the off-axis use is quite negligible. A mottled ex-
tended source such as the night sky must rst be imaged on to the entry slit to
ensure that the grating is uniformly illuminated across its width: otherwise the
CCD receives light at different wavelengths from different parts of the source,
making spectrophotometry unreliable.
negligibly small, given the lownumerical aperture on the input side, and the quality
of a 35 mm camera lens is generally sufcient to image the entry slit on to the
detector pixels.
8.8.9 Transmission gratings
These are plane gratings with dimensions and grating constants similar to those
of reection gratings. They are transparent mouldings from a master, mounted on
a silica blank and have a refractive index, n, in the region of 1.6. They are fully
transparent, with inclined grooved rulings which act as a blaze. However, there
is refraction as well as diffraction, with attendant total internal reection at some
wavelengths, and this limits their use in spectrography. They are at their best when
the wavelength range to be studied is known and xed and the whole instrument is
constructed without moving parts.
They are generally used as in Fig. 8.17 with light incident normally on the
plane face and with the camera inclined at the blaze angle. If the grooved faces
are inclined at an angle b to the surface, then with normal incidence there will be
refraction (neglecting diffraction for the moment) so that the light emerges at an
angle r to the normal where
cos r = n

a tan b
.
where a is the grating constant.
16
16
This refers to the peak of the broad sinc function coming from diffraction by a single ruling, as shifted by
refraction.
86 The plane grating spectrograph
b
r
Figure 8.17 The geometry of the transmission grating, illustrating the various
angles involved in the text.
This places a limit on the blaze angle allowed. Beyond the limit the refraction
causes total internal reection at the inclined surface of the ruling, and just below it
the angle of emergence is so close to the surface that little light is collected by the
camera: the camera in fact is looking at the scarp faces of the rulings, from which
no diffracted light emerges.
The theory is set by two equations:
n sin b = sin(b +r) refraction.
sinr = ,a diffraction.
where b is the inclination of the ruling surface, r the blaze angle of diffraction and
a the grating constant. After some elementary algebra we nd the grating geometry
set by
,a = (n,2) sin 2b sin b
_
1 n
2
sin
2
b. (8.17)
At the same time, the grating ruled width, W
grat
, is foreshortened and the width of
ruling available to the camera is W
cam
, given by
W
cam
= W
grat
cos b cos(b +r). (8.18)
The camera also sees scattered light which has been totally internally reected from
the scarp faces of the rulings.
The limitations of performance are shown in Table 8.1. The rst column shows
the inclination angle of the rulings to the grating surface. The second is the refraction
angle of light fromone ruling, independent of wavelength and assuming a refractive
index, n = 1.6. The third is the ratio of the blaze wavelength to the grating constant,
a. The fourth is the ratio of the transparent width seen by the camera to the actual
ruled width of the grating and the fth is the corresponding blaze wavelength on
the assumption that the grating has 6000 rulings/cm and that the light is normally
incident on the plane face.
8.8 Mounting congurations 87
Table 8.1
Ruling angle Refraction angle ,a W
cam
,W
grat

blaze
38.5 46.389 0.724 0.070 1.117 m
35 31.595 0.524 0.325 8131
30 23.130 0.393 0.520 6546
25 17.547 0.301 0.668 5024
20 13.117 0.228 0.787 3799
15 9.463 0.164 0.879 2740
entry slit
collimator
grating
CCD
Figure 8.18 A spectrograph with a transmission grating, showing the passage of
an oblique ray bundle. Of necessity the camera iris is the dening stop of the
system, just as in the spectrographs of Figs. 8.15 and 8.16, and the grating must
be proportionately oversized.
The moral of the tale is that no great dispersion is possible because for a long
blaze wavelength the grating constant must be large. The broad sinc
2
function
caused by the individual rulings ensures that all the diffracted light is either in rst
order or zero order. The transmission grating is therefore coarsely ruled, typically
30006000 rulings/cm, and suitable for fairly low resolution spectrographs. In-
evitably, longer wavelengths, being diffracted at larger angles, will suffer a re-
duction in intensity because of the ruling foreshortening, and shorter wavelengths
likewise are attenuated, because the broad maximum mentioned above shifts to
larger angles as the refractive index increases while the diffraction angle decreases.
8.8.10 Holographic gratings
These are made by photographing localised interference fringes falling on a photo-
resist material, followed by chemical or physical removal of those parts unaffected
by light or alternatively of those parts which are affected by light. The use of a photo-
resist is essential in order to get the ne detail required, the reason being that the
88 The plane grating spectrograph
photosensitivity is at the molecular-bond level rather than the comparatively crude
crystal-size level of silver halide emulsions.
The result is a sinusoidal grating in which the grooves follow the lines of the
interference fringes. If for example the fringes come from interfering monochro-
matic point sources the resulting grating, being a hologram, will be stigmatic for
that wavelength.
17
The groove shape however cannot be controlled to give a blaze
as in a ruled grating and the effect of the non-parallel, non-rectilinear rulings on the
normal aberrations of a grating spectrometer as well as on its efciency are open to
question. Before deciding to incorporate a holographic grating in a spectrometer,
one of the books dealing specically with the subject must be consulted.
17
And for adjacent wavelengths too, presumably.
9
The concave grating spectrograph
9.1 The Rowland grating
This is a grating ruled on a concave spherical mirror, in consequence of which it
has a self-focusing property and needs no other optical component to produce a
spectrum from a point or short line source. It is the invention of H. A. Rowland
of Johns Hopkins University who gave an account of it in 1883.
1
The rulings are
the intersections of the sphere with a set of parallel, equi-spaced planes, so that the
interval between rulings is not constant on the surface of the sphere, and the device,
although it is a spherical mirror, has a denite optic axis.
The focusing property is as follows:
The grating is ruled on a spherical mirror of radius R. There is a circle, the
Rowland circle, touching the sphere at its vertex perpendicular to the rulings and of
radius R,2. The elementary theory of the grating states that monochromatic light
from any point on this circle is diffracted according to the grating equation, and is
subsequently focused at another point on the Rowland circle.
The theorem is not exact. There is no point focus but an approximation to the
proof can be given by elementary geometry. In Fig. 9.1 the grating is displaced by
an innitesimal amount along the optic axis and intersects the Rowland circle at
points x. x

. Suppose the grating constant at x is a. A ray coming from the centre of

the Rowland circle at angle to the axis meets the grating edge normally at x and
is diffracted according to sinr = ,a. A similar ray directed along the grating axis
meets it normally at o and is diffracted according to sinr

= ,a

. Both diffracted
rays intersect the Rowland circle at y. Then by geometry,
sinr =
H
2R
. sinr

=
H
r
g
.
R
g
2R
= cos .
1
H. A. Rowland, Phil. Mag., 16 (1883), 197.
89
90 The concave grating spectrograph
r
r
H
o
y
x
r
g
r
g
r
g
sec
r
g
sec

Figure 9.1 The basic geometry of the concave grating. The grating has been moved
slightly (about 1 mm) to ease the geometry. Elementary diffraction theory predicts
a focus at the point y provided the grating constant changes appropriately from
the vertex to the edge.
By grating theory,
sinr =

a
. sinr

=

a

.
and by elimination, a = a

sec .
The focusing property therefore follows from the requirement for the spacing a
to be constant along a chord. This argument, like the theorem, is not exact but the
device works well enough within the limits of practical grating construction.
In practice it is the tangential focal surface which is supposed to lie on the
Rowland circle, and a monochromatic point source is imaged as a line segment,
long or short according to the dimensions of the grating.
2
A theory of the focusing property of the Rowland grating is to be found, with
some approximations, in R. W. Woods Physical Optics.
3
However, the strict, three-
dimensional theory is derived by Beutler
4
and a restricted form of this is presented
here as an illustration of the different ways in which the grating may be used.
Beutler used the so-called eikonal method in which the length of a ray is traced
from an object point to an image point and measured in terms of the positions of
2
To be more precise the tangential focal surface is a cylinder coaxial with the Rowland circle and containing it,
and the sagittal focal surface is a similar cylinder turned through a right angle about the grating normal. These
are in accordance with Eqs. (4.11) and (4.12).
3
R. W. Wood, Physical Optics (New York: MacMillan, 1905).
4
H. G. J. Beutler, J. Opt. Soc. Am., 35 (1945), 311.
9.1 The Rowland grating 91
x, w, h
r
cos r, sin r
cos i, sin i
i
Figure 9.2 The geometry of the eikonal method used by Beutler for a strict descrip-
tion of the geometry of the concave grating. The method measures the path length
from an arbitrary point on the Rowland circle to another arbitrary point where
an image is formed, and the length is given as a function of wavelength, grating
constant and the coordinates of the point of reection. In the present analysis a
simplied eikonal, ignoring the third dimension, is used.
the object and image and the coordinates of the intersection with the grating.
5
Polar
and Cartesian coordinates are used as in Fig. 9.2.
6
The object and image positions
are (. i. 0) and (

. r. 0), the grating surface is x

2
2Rx + y
2
+ z
2
= 0 and the
intersection with the grating is (x. n. h). The condition for an image to be formed
is that the optical path E shall change by one wavelength when the y-coordinate,
n, of the intersection point changes by one ruling-width.
In the simplied eikonal used here, the z-coordinate h is neglected and the
problem treated as two dimensional. The distance from object to image is then
E = +

n(sin i +sinr) +
mn
a
+
n
2
2

0
__
sin i

_
n
_
cos
2
i

cos i
R
_
+
_
sinr

_
n
_
cos
2
r

cos r
R
__
.
(9.1)
The vanishing of various powers of n gives ever more stringent conditions for
an image to be formed at the image point.
(1) The vanishingof the rst power of ngives the standardgratingequation(sin i +sinr)
m
a
= 0.
(2) The vanishing of the second power gives
_
cos
2
i

cos i
R
_
+
_
cos
2
r

cos r
R
_
= 0. (9.2)
5
See e.g. M. Born & E. Wolf, Principles of Optics (Cambridge University Press, 1999), Chapter 3. Not always
to be recommended because although it is precise, it involves miserably complicated and tedious elementary
algebra and in practice yields little that is not available through Seidel theory.
6
Beutler has the x-coordinate as the optic axis.
92 The concave grating spectrograph
(3) The vanishing of the third power gives
sin i

_
cos
2
i

cos i
R
_
+
sinr

_
cos
2
r

cos r
R
_
= 0. (9.3)
and conditions (2) and (3) can be satised simultaneously by
_
cos
2
i

cos i
R
_ _
sin i

sinr

_
= 0. (9.4)
The rst factor gives the Rowland condition. The second factor yields some
alternative congurations which do not use the Rowland circle. This factor can be
combined with Eq. (9.2) to yield
=
R[cos
2
i sinr +cos
2
r sin i ]
sinr[cos i +cos r]
(9.5)
and

=
R[cos
2
i sinr +cos
2
r sin i ]
sin i [cos i +cos r]
. (9.6)
Two results follow:
(1) If i = r, then =

= R cos i . This describes the Eagle conguration, where the image

is formed above or below the entry slit.
(2) If r = 0, then = and

= R,(1 +cos i ). This is the Wadsworth stigmatic cong-

uration.
This latter is important because of its optical qualities. It is almost diffraction-
limited and suffers only from spherical aberration and eld curvature. A resolving
power of 25 000 is possible with a grating of ruled area 60 mm80 mm. As with
other arrangements of the Rowland grating it requires a curved focal surface, and
this restricts it, for the moment at least, to silver halide emulsion photography.
7
9.2 The concave grating as a spectrograph
9.2.1 The Rowland mounting
The original conguration was Rowlands and is shown in Fig. 9.3. He used the
Euclidean theorem that the angle in a semicircle is a right angle, and mounted the
grating cell on the end of a bar which could slide along a xed rail. The other end of
the bar was xed to another rail at right angles to the rst where the photographic
plateholder was xed. The entry slit was at the junction point of the two rails and
7
There must in any case be some doubt whether a semiconductor detector could withstand the power of XUV
radiation without irreversible changes to the crystalline structure of the detector.
9.2 The concave grating as a spectrograph 93
grating
entry slit
photographic plate
Figure 9.3 The Rowland mounting. The grating is xed to one end of a bar of
length equal to the grating radius. The photographic plate is xed to the other end
and the plate and the grating are face-to-face. The plate is then moved along a slide
at right angles to another slide which constrains the grating to move towards the
entry slit, thus preserving the illumination and etendue.
an optical bench for external optics ensured that the optic axis pointed directly at
the grating vertex.
The distance between the grating vertex and the plateholder was equal to the
diameter of the Rowland circle and the two pivots by which the bar was attached to
the slippers were separated by the same distance. All this ensured that the grating
was lled with light and that the photographic plate which was bent on a mandrel
in the plateholder to t the Rowland circle stayed in focus while the wavelength
range was altered. This mounting still has its merits.
9.2.2 The Abney mounting
This is shown in Fig. 9.4. It was a variant on the Rowland mounting intended
for use with very large spectrographs (and some concave grating spectrographs
were on a huge scale, possibly lling a whole room) and very large plates. The
plateholder and the grating were kept xed and the input optics and source were
moved from one entry slit to another when different regions of the spectrum were
to be photographed. The inconvenience of this was so obvious that it was never
particularly popular and is long consigned to history.
9.2.3 The PaschenRunge mounting
The same is true of this mounting, also intended for large concave gratings: but
now the input optics and grating are xed on the Rowland circle and instead the
94 The concave grating spectrograph
grating
entry slit 1
entry slit 2
photographic plate
Figure 9.4 The Abney mounting. A relic of the time when spectrographs were
sometimes constructed on by modern standards an enormous scale with focal
lengths of 5 or 6 metres or more.
plateholder is moved on a railway track which runs around the circumference of
the circle. This was another of the heroic mountings mentioned in Chapter 1.
Asmaller version of it with a 1-metre radius Rowland grating used a long curved
plateholder with a strip of 35 mm lm taped to a former of the same radius as the
Rowland circle in a specially made lm cassette. Some of these are still to be found
in use.
8
9.2.4 The Eagle mounting
In common with the Rowland mounting this uses the angle-in-a-semicircle theorem.
In this case however, the plateholder is placed rigidly above (or below) the entry
slit as in Fig. 9.5, and the bar serves to change the region to be photographed. The
result is the most compact and useful conguration for the concave grating. It also
serves as a monochromator mounting as described in the next section. The extra
aberration resulting from out-of-plane working is negligible.
9.2.5 The Wadsworth stigmatic mounting
The Rowland focusing property is that the tangential focus lies on the Rowland
circle. The sagittal focus at the same time is a plane tangent to the Rowland circle
so that there is increasing astigmatism as the images are formed further and further
from the grating normal.
8
One at least, nearing 100 years old, was still in use at the end of the last century in the teaching laboratory in
Manchester University.
9.2 The concave grating as a spectrograph 95
grating
Figure 9.5 The Eagle mounting. This is a three-dimensional version of the Row-
land mounting, and like it, uses the angle-in-a-semicircle theorem to maintain
focus. The entry slit is above the plane of the diagram and the photographic plate,
bent to t the Rowland circle, is beneath it.
entry slit
collimating mirror
photographic plate
Rowland grating
Figure 9.6 The Wadsworth stigmatic mounting. The Rowland focusing property
is abandoned and a separate mirror is used as a collimator. The focal surface is a
parabola andstigmatic images of the entryslit are formedonit at appropriate points.
The main advantage, that the Rowland grating requires no other optical focusing
element, is lost and this makes the mounting unsuitable for XUV spectrography.
The Wadsworth mounting
9
dispenses with the Rowland focusing property and
illuminates the grating with collimated light (Fig. 9.6). If the grating is pivoted
together with the plate or lm holder attached to a bar on the grating normal,
the line images are formed stigmatically on a paraboloidal focal surface to which
the plate or lm must be bent. The distance from the grating vertex to the focal
9
F. L. O. Wadsworth, Astrophys. J., 3 (1896), 54.
96 The concave grating spectrograph
surface varies with the tilt angle away from the input chief-ray so that the position
of the plateholder on the bar must be adjustable.
The mounting is awkward to adjust and focus and its advantage, that it is stig-
matic, is that the images produced are more intense and that, as in old-fashioned
spectrophotography, a Hartmann diaphragm can be used.
10
There also remains the problem of the collimated input beam which in turn
requires an extra optical element for collimation. This tends to remove the chief
advantage of the Rowland grating: its freedom from other attenuators of XUV
radiation.
9.3 The concave grating as a monochromator
This was the normal way of using the Rowland grating after the invention of the
photomultiplier tube. XUV radiation, after diffraction, could be focused on to a
uorescent screen usually a nely ground translucent layer of sodium salicylate
andthe photocurrent fromdetectionof the resultingvisible radiationmeasuredinthe
usual way, by a double-beam method if necessary.
11
Chiey the Eagle arrangement
was used but various Japanese authors published alternative linkages which relied
on a single rotating vacuum joint instead of the cumbersome and sometimes
troublesome linkage of the Eagle mounting.
First among these was the SeyaNamioka mounting
12
(Fig. 9.7), which held the
slits xed at two points on the Rowland circle separated by 141

. Input and output

chief-rays are thus separated by 70

30

. Tangential focus is then maintained as the

grating is rotated about its own vertex. The Rowland focusing condition was clearly
abandoned but there was the merit that good tangential focus could be maintained
over a large range of wavelengths with no adjustment other than grating rotation.
This property considerably eased the construction of vacuum spectrographs.
More practical was the Johnson mounting
13
modied by Onaka
14
in which the
grating is rotated about an offset axis and the input and output chief-rays are sepa-
rated by an angle 2. They describe an instrument with a grating of 40 cm radius,
the angle 2 as 30

and the offset of the rotation axis from the grating vertex as
l = R sin [1 tan (tan
0
tan
0
),2].
10
This is a device for wavelength calibration. It is a plate covering the input slit with three rectangular holes,
staggered, in it. It can be slid across the slit so that the lower third, the middle third or the upper third of the
slit can be photographed in succession, with a reference spectrum in the rst and last cases. Line curvature
sometimes makes it of dubious value.
11
J. F. James, J. Sci. Instrum., 36 (1959), 186.
12
M. Seya, Sci. Light, 2 (1951), 8. Seyas mounting follows fromthe solution of a transcendental equation derived
from the Beutler eikonal.
13
P. D. Johnson, Rev. Sci. Instrum., 28 (1957), 833.
14
R. Onaka & I. Fujita, Sci. Light, 9 (1960), 31.
9.4 The aberrations of the Rowland grating 97
grating rotates
about its vertex
70 30
0.8179 r
g
both slits are fixed
Figure 9.7 The SeyaNamioka mounting. This tooabandons the Rowlandfocusing
property, but, as discovered by Seya, two points are conjugate provided they are
on the Rowland circle and separated by an angle of 70

30

. Wavelength scanning
is then by rotation of the grating about its vertex. The astigmatism however is a
serious drawback with this type of mounting.
However, a great demerit appears when these designs are subjected to ray tracing.
Giventhe usual lengthof gratingrulings, there is a verylarge amount of astigmatism,
up to half the length of the grating rulings, with its accompanying defocus. It is
sufcient to dilute line images to an unacceptable degree and only a few milli-
metres of each ruling on either side of the plane of symmetry make any effective
contribution to the measurable line images.
9.4 The aberrations of the Rowland grating
Astigmatism is the chief source of imperfection in any mounting. The full Beutler
eikonal is required to discuss it, and if this is done an expression in closed form for
astigmatism emerges. A point source on the Rowland circle with the usual angles
of incidence and diffraction i and r, will produce an image at the tangential focus,
also on the Rowland circle, of length L where
L = l
_
sin
2
r +sin
2
i
cos r
cos i
_
= Al. (9.7)
where l is the length of the rulings. The coefcient A is required later.
9.4.1 Astigmatic curvature
The astigmatic image of a point source is not straight and neither is the image of a
long input slit, for the reason outlined in Subsection 8.6.1. The two curvatures are
98 The concave grating spectrograph
in opposite directions and the overlapping images from different parts of the input
slit yield a coma-like aberration which limits the resolution.
15
There are two cases
to consider.
(1) Grating rulings of nite length and a point source on the Rowland circle.
(2) A long straight slit and innitesimal grating rulings.
To simplify the analysis we dene by
1

2
= [i sin
2
i +sinr tan
2
r +sin i sin
2
r tan
2
r sin 2i sinr tanr].
Then dr, the width of the line image, is in each instance:
(1) Long rulings, short input slit:
dr =
l
2
2R
2
whence l =

2R dr.
and l is the length of the longest rulings that can be used.
(2) Short rulings, long input slit:
dr =
s
2
2Rl
2

2
whence s = A

2R dr.
where A is dened in Eq. (9.7) and s is the length of the longest slit that can be used.
It must be emphasised that these computations are based solely on geometry,
and the effective resolution, d, in a concave grating spectrograph is obtained in
each case from the usual formula d = (a,m) cos r dr.
9.5 Practical details of design
There are some peculiarities of vacuum grating spectrography which are worthy of
particular mention, beyond what is discussed in Chapter 14.
There are various ways of constructing vacuum spectrographs and some provide
more practical difculties than others. It is a sine qua non in vacuum technology
that you have as few vacuum seals as possible, particularly if they are to be broken
frequently.
So far as possible, remote operation of any interior moving parts by electric
motor or magnetic coupling is to be preferred to a rotating vacuum seal. Low-
voltage operation is essential, especially if low-pressure gases are to be tolerated
or even required perhaps, in absorption spectrography.
15
Coma itself depends on the coefcient of nl
2
in the Beutler eikonal, and as l is ignored, it does not appear here.
9.5 Practical details of design 99
source, detector etc.
to cold trap and diffusion pump
Figure 9.8 Apossible arrangement for an Eagle mounting, with a separate vacuum
chamber which can be removed for adjustments to be made to the optical system.
All controls are led to the optical table through the end plate via vacuum seals
which in general are not rotating and therefore leak-proof.
It is imperative that the instrument be designed so that it can be adjusted and
focused in visible light. The awful alternative is to reduce the pressure to a vacuum
each time a focus adjustment is made.
The author offers this advice from his own experience.
Build a monochromator rather than a spectrograph. Make it as small as possible,
compatible with the research end in view. Choose the Eagle mounting, with a
1-metre radius grating with 6000 rulings/cm. Design for 10 mm-long slits centred
15 mmabove andbelowthe plane of symmetry. The defocus andcoma are negligible
out to 5000

A. The resolving power is of the order of 5 10
4
. Build the instrument
as shown schematically in Fig. 9.8, on a horizontal table cantilevered out from a
vertical face-plate which itself is carried on a suitable structure. Allow for a grating
adjustment and traverse for a maximum wavelength of at least 5461

A to permit
visual adjustment of the optical alignment and focus. The table should have large
holes for faster pumping. The vacuum chamber should be a simple cylinder able
to run its whole length on horizontal rollers, wheels or linear bearings to make a
single large-diameter vacuum joint at the face-plate. The seal may then be like that
of a simple evaporator plant. This may be expensive, but saves much time in the
long run.
Pass all controls through the face-plate in seals which need not normally be
demountable. The face-plate also holds the slits, fore-optics and detectors. The
lower half of the face-plate below the optical table holds the vacuum pump
port, and the pump should have the largest pumping speed that the budget will bear.
100 The concave grating spectrograph
It is unlikely that a vacuum lower than 10
6
mm Hg will be required, so that a
turbo-molecular pump is probably unnecessary. Nevertheless for a 1-metre concave
grating spectrograph in a 500-litre stainless steel chamber, a diffusion pump with at
least 1520 cm throat diameter should be included in the design. A cold-trap (LN
2
)
must be included.
10
The interference spectrograph
Interferometers were originally conceived as devices for measuring the wavelength
of light or the precise measurement of small distances, and with spatial resolution
and localised fringes,
1
contour maps of reecting surfaces could be made using
Fizeau fringes or Tolansky fringes. They also served as very high resolution spec-
trographs with which the hyperne structure of spectral lines could be studied. It
was only later that their very high efciency was noted and that they could be used
with advantage at moderate resolution for the spectral examination of very feeble
sources of radiation such as chemical phosphorescence, the night airglow and the
zodiacal light.
10.1 The phase angle
There is an abstract angle, the phase angle, dened by
= (2,)2nd cos . (10.1)
This is the most important general concept in interferometry, and describes all the
varieties of interferometer. The phase angle comprises four variables: (1) the path
difference, 2d; (2) the angle of incidence, ; (3) the incident wavelength, ; and
(4) the refractive index in the gap, n. The product 2nd represents the optical path
difference between interfering wavefronts and (2,)2nd is the consequent phase
difference.
Holding two of these four quantities constant and using a third as independent
variable, the behaviour of the fourth describes one of a large number of possible
fringe patterns. These include Fizeau fringes, constant inclination fringes, Mach
Zehnder fringes, EdserButler fringes, FabryPerot fringes, Tolansky fringes and
1
There is a myth that in localised fringes the interference takes place in the gap between two reecting surfaces
rather than at innity as in fringes of constant inclination. The interference of course takes place everywhere
in space and in particular at the detector, in the retina of the eye or on the plate of the camera.
101
102 The interference spectrograph
so on. Six types of interferometer result. So far as spectroscopy is concerned, is
the dependent variable and can be made to depend on the variation of . d or n.
Interferometers for spectroscopy can be divided into two categories: spectrom-
eters and spectrographs. In the former category there is a further subdivision into
Michelson Fourier-transform spectrometers and etalon scanning monochromators,
and in the latter category the FabryPerot spectrograph is the sole survivor of a
century of experiment. Other devices such as Michelsons echelon and Lummers
plate, although they yielded very high resolution, were difcult to make, inconve-
nient to use and had far lower etendue than the etalon spectrograph. They are all
now museum pieces.
10.2 The FabryPerot etalon spectrograph
The FabryPerot etalon was rst described by Charles Fabry and Alfred Perot
in 1899.
2
It comprises two at transparent plates worked optically so that the
gap between them can be held constant over the whole surface to an accuracy
of better than ,120. The constant gap is the important aspect: the surfaces of
the individual plates need not be at to better than ,4.
3
When the device is
dismountable, the angular orientation of the two plates may be important and is
then indicated by a pair of arrows inscribed on the sides to ensure correct alignment.
Etalons are oftenmanufacturedandassembledas sealedunitarydevices withthe gap
permanently xed so that no adjustment is required or possible, and these, dedicated
to a particular investigation, are to be preferred when constructing a FabryPerot
spectrograph.
The obvious advantage of a dismountable etalon is that the gap can be changed
for different purposes and when this is needed it is desirable that there be a servo-
stabilising mechanismfor holding the gap parallel and constant during observations.
10.3 FabryPerot theory
The theory of the FabryPerot etalon is to be found in every textbook on physical
optics. The transmitted intensity is given by the Airy formula:
I () =
I (0)
1 +
4r
(1r)
2
sin
2
_

2
_. (10.2)
where r is the reection coefcient for intensity at the reecting surfaces and, as
before, = (2,)2nd cos . For values of close to zero, sin(,2) (,2) and
2
C. Fabry & A. Perot, Ann. Chim. Phys., 16 (1899), 115.
3
Although the modern tendency is to make the plates individually so that they can be assembled without regard
to orientation.
10.3 FabryPerot theory 103

0 2

Figure 10.1 The Airy function of Eq. (10.2), illustrating the free spectral range
L and the resolution, .
the Airy function can be approximated by
A() = I (0)
U
2
U
2
+
2
.
where U
2
= (1 r)
2
,r.
This is the equation of a Lorentz curve, and the Airy function (Fig. 10.1), which is
periodic, may be regarded as the convolution of a Lorentz curve with a Dirac comb
of period 2. As we shall see in a moment, this Lorentz curve is the instrumental
prole of the FabryPerot spectrometer and the number of such proles that can be
tted into one order of interference is the ratio of the FWHM to the order spacing.
The FWHM, expressed as an angle, is 2(1 r),

r and the ratio of this to the

period, 2, is
F = 2

r
2(1 r)
=

r
1 r

3
1 r
.
This quantity F is called the nesse of the interferometer and is a measure of
the number of resolved spectral elements in one order, that is, in one free spectral
range (FSR). The nesse is affected by the quality of the reecting coatings, by
the wavelength to be examined and by the atness of the gap. In addition there is
a small effect caused by the polarisation of the light, as the effective gap width for
oblique incidence at the edge of the fringe pattern that is depends on the state of
polarisation, that is, whether the electric vector is radial or tangential. The effect,
which vanishes on the optic axis, in the end controls the outer ring diameters and
limits the available nesse.
For purely practical reasons, larger etalons of 50 mm diameter and above are
likely to have a lower nesse than smaller ones of less than 25 mmaperture. Typical
values in practice for a 50 mm etalon are F = 3545.
104 The interference spectrograph
The free spectral range L is the range of wavelengths that can be tted between
two orders of the same wavelength. The gap width, d = n,2, is also equal to
(n 1)( +L),2, so that L = ,n =
2
,2d.
The resolution, the smallest range of wavelengths, , that can be observed as
two distinct lines, is equal to the FWHM of the Airy function maximum, that is to
L,F, and can be written in terms of wavelength as
=

nF
=

2
2dF
=

2
(1 r)
6d
and r, the reection coefcient of the plates, is usually in the region of 0.93.
4
10.4 The FabryPerot monochromator
10.4.1 Scanning in wavelength
This is done by changing the optical path between the two plates.
If the etalon is followed by a focusing lens the FabryPerot fringes from
monochromatic light appear at the focus as ne concentric circles. An annular
hole might be made in the screen with its inner and outer edges at the half-power
points of a ring, and its diameters come from the Airy formula (Eq. 10.1) with the
usual approximation cos = 1
2
,2. At the maximum of intensity,

2
2
= 1
n
2d
.
and at the half-power points,

2
1
2
= 1
n( ,2)
2d
.

2
2
2
= 1
n( +,2)
2d
.
whence

2
1

2
2
2
=
n
2d
=
_
n
2d
__

_
=
cos
R
.
It is usual but not universal to let the inner diameter of the annulus go to zero
so that the wavelength-dening stop is a circle. Then
2
= 0 and cos
1
1 and the
angular diameter, , of the circular stop is given by

2
8
=
1
R
. (10.3)
4
With multilayer reecting coatings this high reection coefcient holds good only over a hundred angstroms or
so, and must be specied when ordering the plates.
10.4 The FabryPerot monochromator 105
FabryPerot talon
chief-ray collimator
Jacquinot stop
and detector
interference filter
as order-sorter
Figure 10.2 The classical mounting for a FabryPerot spectrometer. The incoming
light, which must be from an extended source, is focused on to the Jacquinot stop
but rst passes through a short section where the chief-rays are collimated before
passing through the interference lter which is the order-sorter in this type of
arrangement. Wavelength scanning is by one of various methods of changing the
optical path between the plates. Fore-optics, usually in the formof an astronomical
telescope, could be added if the source was of small extent or if ne detail in the
source was to be examined.
These relationships between the various quantities . d. r and can be used to
dene the important quantities associated with the FabryPerot spectrometer and
spectrograph.
(a) The free spectral range (FSR) = L. = ,n. = 2d,n
2
. =
2
,2d.
(b) The nesse, F = L,. =

r,(1 r). 3,(1 r).

(c) The resolving power R = ,. = nF. = 2Fd,.
A typical arrangement for a scanning FabryPerot spectrometer is shown in
Fig. 10.2. There is a short section near the Jacquinot stop where the chief-rays
are all made parallel to the optic axis and this is where the order-sorting interfer-
ence lter is placed. The lter, which is a FabryPerot etalon working at a very
high nesse and a very low order, requires a divergence less than F,8 or higher (the
higher the better), and this focal ratio should be built-in as part of the initial design.
10.4.2 Mechanical scanning
The mode of mechanical displacement of one plate is generally by piezo-electric
transducers, mountedtypicallyas inFig. 10.3. Anindividual piezo-electric
5
ceramic
disc of barium titanate, for example, will change its thickness by 3

A V
1
applied
across its faces and a stack of ten such ceramic discs of alternating polarity will
multiply this by 10, to 30

A V
1
. A scanning range of one order is thus achieved
with a ramp of 100 V, a ramp function which can be repeated slowly over 1500 s,
or rapidly at 50 Hz or more. The efciency and the signal/noise ratio obtained is the
5
Not electrostrictive, but a disc which has been electrically polarised during manufacture.
106 The interference spectrograph
electrodes
connected
to earth
piezo-electric
discs of alternating
polarity
brass discs
epoxy-resin sheath
electrodes
connected
to ramp voltage
Figure 10.3 The mounting of piezo-electric feet for the plate when mechanical
scanning is used to change the gap. Control of the gap width was either by capaci-
tance measurement or by transmission of EdserButler fringes through opposite
edges of the gap.
beam splitter
polaroid analyser
detector
/4 plate
/4 plate
Figure 10.4 An optical screw for controlling and changing the gap. This device is
also applied to the gap in Michelson Fourier spectrometers and allows changes of
the thickness of ,500 to be made under computer control.
same in both cases and both scan rates may have advantages. The voltage applied
can be controlled by an optical screw which changes the gap by one order (i.e. by
,2) for every half-turn of the screw.
6
The mechanism is simple: a stabilised laser (Fig. 10.4) sends a plane-polarised
beam into a Michelson interferometer with the plane of polarisation parallel to the
beam-splitter surface. One beamis reected and returned to the output port while the
other passes through a quarter-wave plate and is converted to circularly polarised
light. After reection it passes again through the quarter-wave plate and is then
converted back to plane-polarised light but now with the plane of polarisation per-
pendicular to the original plane. It too passes through the output port, but although
6
Y. P. Elsworth & J. F. James, J. Phys. E: Sci. Instrum., 6 (1973), 1134.
10.4 The FabryPerot monochromator 107
it is coherent with the other output beam there is no interference because of the
mutually perpendicular planes of polarisation. The two beams now pass together
through another quarter-wave plate, and one is converted to left circularly and the
other to right circularly polarised light. The result is plane-polarised light, but the
plane of polarisation depends on the path difference in the interferometer. A piezo-
electric transducer vibrating on one of the Michelson reectors causes a minuscule
sinusoidal variation in the path difference, enough to cause the plane of polarisa-
tion to oscillate in direction and a phase-sensitive detector measures the amplitude
and causes a servo-mechanism to turn an analyser to the extinction direction. The
angle of the analyser indicates the path difference. Alternatively the analyser can
be turned and the path difference changed to match by a servo-mechanism in the
interferometer. The interferometer is very small large enough to allow a laser
beam to traverse and one of its reectors is a small patch on the etalon plate.
During scanning the atness of the gap can be controlled by two types of servo-
mechanism. In one case
7
a narrowcollimated beamof white light is directed through
the edge of the etalon and EdserButler fringes would be observed if the light were
analysed. Instead it is passed a second time through the opposite edge and there
will be full transmission of the EdserButler fringes only if the two gaps are exactly
equal. If one plate is made to rock through a small amplitude a few angstroms
is sufcient the transmitted intensity varies rapidly and again a phase-sensitive
detector tuned to the rocking frequency will sense the direction and a voltage
amplier will apply a restoring force to the transducer. Two such servo-loops,
mounted normally to each other, will stabilise the gap in two dimensions.
Alternatively there is a mechanism described by Hicks et al.
8
Two small ca-
pacitors can be made by aluminising patches at opposite points on the etalon rim,
and the two capacitances (c
0
A,d), will be the same only if the gap d is constant.
An a,c bridge then measures any change in the two capacitances and applies an
appropriate restoring movement.
10.4.3 Pressure scanning
This, like mechanical scanning, describes a scanning monochromator, but now the
etalon is held in a pressure chamber. The gap width is held constant and the same
diameter of circular stop denes the wavelength range transmitted. The refractive
index, n, of the gas in the chamber is changed slowly by a pressure valve according
to n = (1 +kp). For air k typically is 0.000 11 atmosphere
1
but may be higher
for various halogen-substituted hydrocarbon vapours (CFCs). The method is thor-
oughly redundant and obsolete and not to be considered further.
7
J. V. Ramsay, Appl. Opt., 1 (1963), 411; M. J. Smeethe & J. F. James, J. Phys. E: Sci. Instrum., 4 (1971), 429.
8
T. R. Hicks, N. K. Reay & R. J. Scaddan, J. Phys., E, 7 (1974), 27.
108 The interference spectrograph
10.4.4 Multislit scanning
The efciency of the scanning monochromator can be improved by an order of
magnitude by using a focal plane mask with concentric annular apertures.
9
The
instrument must be designed and built before the mask can be made, since the
angular diameters of the apertures must coincide with the ring system. The mask is
conveniently made by setting the instrument to look at a monochromatic source with
a wavelength close to the wavelength to be observed, then to photograph the Fabry
Perot fringes with a high-contrast, ne-grain emulsion on a glass plate. The mask
can then be made by printing the negative or by reversal development, in which
the original reduced silver image is bleached away and the remaining emulsion
exposed and developed. Up to ten rings can be included in the mask before the
polarisation effect mentioned above interferes seriously with the nesse.
10.5 The FabryPerot CCD spectrograph
The original FabryPerot spectrograph was abandoned for serious research when
the photomultiplier arrived in the 1950s. Since the CCD reversed the situation the
FabryPerot spectrograph now leads the eld in this kind of interference spec-
troscopy. The advantages are manifold. The etalon is xed, and such a sealed,
unitary device is ideal for its purpose where the resolving power required is known
beforehand. There are then no moving parts other than electrons to the spec-
trograph. Although photography may still be by silver halide emulsion, the CCD
is the detector of choice whenever fast transient phenomena or very feeble light
sources are to be examined.
The efciency of the device is determined by the area of the CCD multiplied
by the solid angle subtended at the etalon aperture, or equivalently, by the square
of the numerical aperture of the camera lens. The statement is subject to various
constraints. These are:
(1) There must be at least one complete ring diameter photographed, and in practice 2 or
2
1
2
rings are desirable. The rings need not be complete, because this depends on the
aspect ratio of the CCD chip, but a means of determining the x- and y-coordinates of
the pixel at the centre of the ring system must be available.
The ring diameters, D
n
, are given by the well-known formula
D
2
n
= 8 f
2
(1 n,2d). (10.4)
where n refers to the nth ring fromthe centre of the eld, d is the gap width in the etalon
and f is the camera lens focal length. At the centre of the ring system, n is given by
9
Amal Al-Hillou, Ph.D. Thesis, University of Manchester (1976).
10.5 The FabryPerot CCD spectrograph 109
n = 2d, and is not usually integer. Remember that the order, n, decreases as the ring
diameters increase.
(2) The outer ring thickness for strictly monochromatic light must be several (at least 3 or
4) pixels, otherwise the spectral resolution suffers.
(3) The camera lens must have an aperture equal to the etalon clear diameter and the highest
numerical aperture available.
(4) There must be provision for an order-isolating lter. This is usually an interference lter
with a FWHMbandwidth equal to half the free spectral range of the etalon. Such a lter
requires that all chief-rays passing through it must be parallel to the optic axis. This
is to ensure that the same spectral range is photographed at all points of the detector,
remembering that the lter itself is a FabryPerot etalon with an effective gap 1,40
that of the principal etalon.
The precursors of the design are:
(1) The resolution required. This determines the gap thickness, which in turn determines
the angular diameter of the rst two rings.
(2) The dimensions of the CCD chip. This determines the focal length of the camera lens,
which must be such as to allow at least two and not more than three complete rings to
fall on the CCD.
(3) The numerical aperture of the necessary lens. This determines the diameter of etalon
required after the optical paths have been laid out. The diaphragm of the camera lens
and not the etalon is the pupil of the system and enough clearance between the front
surface of the lens and the rear surface of the etalon must be given.
10.5.1 The L-R product
Like the scanning FabryPerot spectrometer, the spectrograph has an L-R product
and it is greater than that of the scanning spectrometer by a factor 16F, where F is
the nesse of the etalon.
This can be derived from the need for two rings on the CCD. If the order of
interference is n
0
at the centre of the ring system it must be (n
0
2) at the edge of
the CCD chip, and the diameter of this ring is connected to the resolving power by
D
2
2
= 8 f
2
_
1
(n
0
2)
n
0
_
=
16 f
2
n
0
=
16 f
2
F
R
.
D is the width of the CCD chip and its area is D
2
if it is square (or proportional
to D
2
if it is not), so that with an etalon of diameter D
e
, the etendue of the whole
spectrograph is
L =
D
2
e
D
2
f
2
=
D
2
e
f
2
16 f
2
F
R
.
and the efciency, E = LR = 16D
2
e
F.
110 The interference spectrograph
etalon
interference filter
chief-ray
collimator
objective
F/2 CCD camera collimator
Figure 10.5 The FabryPerot CCD spectrograph. A telescope appropriate to the
extent of the object under examination produces an image of the object at its focus
and, as before, a short section is included where the chief-rays are collimated before
passing the order-sorting interference lter. The object is then re-collimated before
passing through the etalon. The auxiliary function of the optical system is to form
an image of the objective on the camera pupil and an image of the object on the
CCD. The evolution of this type of spectrograph is discussed in more detail in
Appendix 3.
10.6 Fore-optics
Figure 10.5 shows a FabryPerot spectrograph designed for airglowspectrography.
The telescope objective is followed by a chief-ray collimator near its focus. This
is a lens of the same focal length as the objective and it serves to pass all ray
bundles through the interference lter in the same orientation. The light is then
re-collimated preparatory to its passage through the etalon and simultaneously the
objective is imaged on to the CCD camera pupil. Two lenses are employed to avoid
using an inconveniently large collimator. The evolution of this design is described
as a worked example in Appendix 3.
10.7 Reference fringes
It is generally a good idea to record a reference wavelength at the same time as the
spectrum under investigation. One method is to use a keyhole projector, imaging
the refernece source through a spatial lter on to a small Newtonian mirror on
the optic axis of the spectrograph (Fig. 10.6). The reference wavelength, from a
198
Hg lamp for instance, need not be within the pass-band of the order-sorting
interference lter but it must be within the wavelength range for which the etalon
coatings were made. To avoid confusing the reference fringes with details in the
observed spectrum, the light from the reference source may be passed through a
spatial lter to interrupt the fringes it produces. The light coming through the lter
is collimated and an image appears on the CCD. Since no sharp image of the mask
is needed, the projector optics may be simple plano-convex lenses.
10.8 Extraction of the spectrum 111
Reference source
condenser
Spatial filter mask
etalon
mask collimator
main collimator
Newtonian flat
CCD
Figure 10.6 The reference fringe injector uses keyhole optics. The condenser
images the reference source on to the Newtonian mirror. The mask is at the back
focus of its collimator, so that at the camera it appears to be at and is imaged
with interrupted reference fringes at the detector. No elaborate optical system
is required and only a few per cent of the etalon area need be obscured by the
Newtonian mirror.
10.8 Extraction of the spectrum
The CCD output will be in the form of a two-dimensional array of numbers repre-
senting the content of each pixel at read-out time. At the time of writing the FITS
format is popular and adequate because the contents of each pixel can be read and,
if necessary, altered individually.
10
Finding the pixel coordinates of the centre of the ring system can be done by
software, but in practice it is a simple task to do so adequately by inspection.
The following software procedure will extract the spectrum:
(1) Three linear arrays are required in the computer program.
(2) The two-dimensional read-out of the eld is inspected visually and the pixel at the
centre of the ring system is identied. A computer program can be written to do this,
but in practice a simple visual inspection will nd the appropriate pixel.
10
Flexible Image Transport System: visit an internet site for details of its format.
112 The interference spectrograph
(3) Each pixel is taken in turn and the square of its distance in pixels from the central pixel
is computed and used as the address in all three arrays. In general this number will not
be an integer and the array should be long enough (i.e. should have enough addresses)
that the integer part of the calculated number or its multiple can be used. (For example,
if d
2
comes to 102.776, this can be made to correspond to address number 1028.)
(4) The contents of the pixel are stored at the corresponding address in array [a].
(5) The number in the corresponding address in array [b] is increased by 1 to indicate the
number of times the intensity at that distance from the centre has been added to array
[a]. This is to make allowance for incomplete rings that may be on the CCD image.
(6) Array [c] is constructed containing array [a] divided by array [b] and this contains the
average intensity at (the square of) each distance from the centre. The addresses are
numbers proportional to the wavelength at each ring via Eq. (10.3) above. A graphical
plot of array [c] is then a linear plot of the required spectrum, repeated one order later.
The geometry of the spectrograph allows the wavelength corresponding to each address
to be computed.
The method is particularly designed to cope with incomplete rings, but it happens
from time to time that there is need for spatial resolution in the interferogram. The
corona of the Sun or a luminous interstellar cloud may be in focus on the CCD
together with the rings. In this case the same reduction technique may be used, but
with the area restricted to correspond to a feature or a particular small area in the
source.
10.9 Choice of the resolution and gap
This depends on the nature of the spectrum to be investigated. For hyperne struc-
ture, thenclearlythe resolutionmust be sufcient toresolve the various components.
For small changes such as Doppler shifts, the intention must be to maximise the
change that appears and this in turn depends on the spatial frequencies present in the
displayed spectrum. Higher resolution implies less luminosity and hence a longer
observing time, and this, with faint sources, could be the limiting factor. There is
no point in using more resolution than is needed. The matched lter theorem
11
guides the gap selection here. The Fourier transform of the instrumental prole of
a FabryPerot spectrograph is a Dirac comb multiplying a decaying exponential:
III( p)e
p,b
III(x)
1
4
2
x
2
b
b
+1
.
the right hand side being the instrumental prole of the etalon. If the observed
spectrum line is regarded as a Gaussian of FWHM n then its shape is represented
11
The signal/noise ratio in a receiver is greatest when the noise-lter transmission prole has the same shape as
the power spectrum of the incoming signal. Although this was derived originally for radio transmission, it is
quite general and applies throughout information and communication theory, including spatial ltering.
10.10 The crossed FabryPerot spectrograph 113
by
e
1.92x
2
,n
2
e

2
p
2
n
2
,1.92
.
These two curves can not be overlaid and matched exactly but a reasonable approx-
imation is that they cross at the 1,e-point. Then,
n,1.386 = 1,b
and the FWHM of the FabryPerot instrumental prole should be that of the ob-
served line FWHM divided by 1.386.
This is still approximate, and in practice much the same efciency is obtained
provided that the etalon line width is about the same as the observed line or some-
what less (rather than more). It is impossible to be more precise as the line shapes
being observed are not always conveniently Gaussian and the spectrummay contain
lines of different half-widths.
10.10 The crossed FabryPerot spectrograph
A technique for improving the resolution of emission line spectra in a spectrograph
by an order of magnitude was developed in the early days by imaging the rings
of a FabryPerot etalon on to the entry slit of a Littrow spectrograph, the centre
of the ring system lying on the line of the slit. The slit was then opened wide,
subject to not allowing line images at different wavelengths to overlap and the
spectrogram revealed such things as pressure broadening, Zeeman splitting and
hyperne structure simultaneously at a large number of wavelengths for example
in a molecular band spectrum. The technique depends essentially on having small or
zero astigmatismin the grating or prismspectrograph and is therefore most suitable
for the Littrow conguration. The image of the ring system should contain both
parts of the innermost ring in order that the centre of the system be located.
12
The
same criteria apply as for a normal FabryPerot spectrograph: the number of rings
should be at least two, and with a CCD detector the instrumental prole should be
at least 5 pixels wide at FWHM if the line prole is required.
The technique is not yet moribund and, if applied, the same rules apply for
imaging the source with the proviso that if the source has structure of its own an
oxy-acetylene ame for example the magnication and defocus should be such
that the local variations do not affect the interferogram.
12
The alternative is a tedious amount of programming to linearise the square-law wavelength dispersion along
the slit direction.
11
The multiplex spectrometer
The theory and practice of Michelson Fourier-transform spectrometry is strictly
speaking beyond the purview of this book, but it is in such universal use that a
brief description is warranted in order that a reasoned choice can be made when
contemplating infra-red spectroscopy.
11.1 The principles of Fourier spectrometry
When a monochromatic beam of light passes through an ideal Michelson inter-
ferometer with no absorption and exactly 50% reection at the beam-splitter, the
intensity of the beam which emerges depends upon the path difference between the
two recombined components.
Suppose that the incident beam has wavenumber and complex amplitude A.
Suppose, for simplicity, that the phase is zero at the beam-splitter. After passing
the beam-splitter the two emerging amplitudes will be
A

2
e
i
1
and
A

2
e
i
2
.
where
1
and
2
are the phase changes on reection and transmission.
After passing through the two arms of the interferometer the two beams recom-
bine at the beam-splitter. Before recombination the two amplitudes are
1
A

2
e
i
1
e
2id
1
and
A

2
e
i
2
e
2id
2
.
and after recombination, when each beam has undergone one transmission and one
reection, the resultant amplitude transmitted is
A
2
e
i(
1
+
2
)
_
e
2id
1
+e
2id
2
_
.
1
Bearing in mind the precept that phase change = 2 path change.
114
11.1 The principles of Fourier spectrometry 115
This quantity, multiplied by its complex conjugate gives an expression for the
transmitted intensity:
I (L) =
AA

2
[1 +cos(2L)].
where L = 2(d
1
d
2
) is the path difference between the two beams.
The transmitted intensity can be seen to vary sinusoidally with L. The intensity
and the frequency of this wave depend on the rate at which the path difference
changes and on the intensity and wavenumber of the incident radiation.
In practice the input has a continuous spectrum S() and if the intensity in the
range + is S() d, it will contribute an innitesimal power dI (L) =
(S() d,2)[1 +cos(2L)] to the output beam.
The total transmitted intensity will then be
I (L) =
1
2
_

=0
S() d[1 +cos 2L]
=
1
2
_

=0
S() d +
1
2
_

=0
S() cos 2Ld
and the second term is the Fourier cosine transform of the spectral power density,
S(), as a function of L.
An alternative way of looking at Fourier spectrometry is to regard the signal
strength at the detector as the time average of the product of the signal and its
delayed counterpart:
I (t

) = [ f (t ) + f (t +t

)][ f

(t ) + f

(t +t

)]
= f (t ) f

(t ) + f (t +t

) f

(t +t

) +2| f (t ) f (t +t

)|.
where t

= L,c and L is the path difference. The third term is the autocorrelation
function of f (t ) and by the WienerKhinchine theorem its Fourier transform is the
SPD, or the spectrum of the signal.
The basis of Fourier-transform spectroscopy is that the output signal from the
interferometer is recorded as a function of L (the interferogram) and the spectrum
is obtained from this by a Fourier transform, usually in a digital computer.
The sampling theoremis used, and samples of the transmitted intensity are taken
at equal intervals of path difference. The number of samples may vary from 1024
to 10
6
, depending on the investigation being followed. This set of numbers takes
the place of the interferogram and is sent as a one-dimensional array or vector to
the computer.
The art of multiplex spectrometry lies in three directions.
Firstly, the optical alignment must be maintained to interferometric accuracy
during the whole displacement phase of several centimetres or more while the
interferogram is being recorded.
116 The multiplex spectrometer
(a)
beam-splitter
R1
R2
output port
Detector
(c)
beam-splitter
R1
R2
output port
input port
Detector
Detector
(b)
(d)
beam-splitter
R1
R1
R2
R2
output port
Detector
input port input port
input port
output port
beam-splitter
Figure 11.1 Variations on a theme. Four different types of Michelson Fourier
spectrometer which have found commercial employment in different wavelength
regions. (a) The classical Michelson interferometer; (b) the same, with cube-corner
reectors for automatic alignment maintenance; (c) the cats-eye autocollima-
tor reector which returns any incidence ray anti-parallel to its input direction;
(d) the tilting Michelson, in which the path difference is changed without
physically moving the reectors.
Secondly, the samples must be collected at exactly equal intervals of path dif-
ference. The process is analogous to ruling a new diffraction grating every time
a spectrum is observed, and the precision required is the same as that of a ruling
engine,
2
although the time devoted to each spectrum is considerably less.
Thirdly, an efcient way of doing the Fourier transformmechanically is required.
Much research was devoted to these three problems in the 1960s and 1970s and
various solutions were found. In far IR spectroscopy, simple mechanical displace-
ment of a reector is adequate. At shorter wavelengths, cats-eye reectors and
cube-corner reectors allow linear translation of the Michelson mirrors without
misalignment, and the tilting Michelson changes the path difference by rotating
a jig containing the beam-splitter and a rigorously parallel mirror about a hinge.
Examples of successful applications of these ideas in manufactured Fourier spec-
trometers are shown in Fig. 11.1.
If samples of the interferogram must be taken when the reectors are stationary,
piezo-electric transducers can be made to oppose a smooth steady change in path
2
The late Dr H. J. J. Braddick once remarked that designing a grating-ruling engine was the same problem as
designing a screw-cutting lathe to be made of rubber.
11.2 The multiplex advantage 117
difference with servo-stabilisation of both the alignment and path difference for
each sample. Wavelength-stabilised lasers offer a standard of step length either
by ensuring that the step length is a constant number of half-wavelength fringes
or, by an optical screw
3
or ratchet device, a constant fraction (often vulgar) of a
fringe.
The Fourier transformation of the interferogram is by digital computer, usually
a dedicated laptop, using the fast Fourier transform or FFT, one of a number of
routines derived from the CooleyTukey algorithm of 1964.
4
The reasonfor doingspectroscopyinthis wayis the enormous gaininsignal/noise
ratio over grating spectroscopy which is obtained when the dominant noise source
is the detector noise. This follows from P. J. Fellgetts discovery in the 1950s of
the multiplex advantage.
5
This merits some discussion as it is typical of the way in
which detection methods are assessed.
11.2 The multiplex advantage
Consider a grating spectrometer and a multiplex spectrometer of the same etendue
E and observing the same spectral source S(), in each case to be explored over a
bandwidth
0
. The spectrum is everywhere dense and the total power is S.
Suppose each instrument dwells at each sample point for a time T. Suppose there
are to be N resolved elements in the spectrum.
In the monochromator the ux to the detector is
F
g
=
ES
0
T
N
.
which is the total signal energy in each sample. The detector noise is proportional
to

T and the signal,noise ratio achieved is

S,N =
ES
0
T
Nk

T
=
ES
0

T
Nk
.
In the multiplex spectrometer the total signal in each sample is much greater:
F
m
=
ES
0
T
2
.
and again the noise energy at each sample is k

T.
3
Y. P. Elsworth & J. F. James, J. Phys. E: Sci. Instrum., 6 (1973), 1134.
4
J. W. Cooley and J. W. Tukey, Mathematics of Computation, 19 (1965), 297. See also: E. Oran Brigham, The
Fast Fourier Transform (Prentice-Hall, 1974).
5
P. Fellgett, J. de Phys., 19 (1958), 187, 237.
118 The multiplex spectrometer
In this case the samples are used to compute the energy in each element of the
spectrum by
S(m) =
N1

n=0
_
ES
0
T
2
_
cos
2mn
N
.
and since cosines are as often negative as positive the total energy in the computed
sample will be
F
m
=
ES
0
T
2
.
The noise samples, also negative as often as positive will average to 0 but the
RMS value will be

_
N1

n=0
k
2
T cos
2
2mn
N
=
_
Nk
2
T
2
since cos
2
=
1
2
.
and the signal/noise ratio is then
S,N =
ES
0
2k
_
T
N
.
which is greater than that of the monochromator by a factor

N,2.
Asimilar calculation for the same spectrometers with photoelectron shot-noise as
the principal source of noise shows that there is no such multiplex advantage and that
the S,N ratio is apparently the same for each type of instrument. The calculation
however does not show that there is in fact a multiplex disadvantage since the
shot-noise on any one strong emission line is spread throughout the interferogram
and hence throughout the spectrum, and it effectively drowns any faint emission
lines which may be present. Similarly, the shot-noise from a continuous emission
spectrum lls in or obscures any narrow shallow absorption lines which may be
present.
The above calculations are not exact since source strengths may vary from point
to point in the spectrum as well as from moment to moment during the observation,
and the multiplex advantage is usually present only in order of magnitude. When
the number of samples (or the resolving power) is high, 100 000 for example,
the multiplex advantage is 300, the time taken to obtain the spectrum is reduced
from 5 minutes to 1 second, and in the infra-red beyond 1 m, where sensitive area
detectors still await invention, the Michelson Fourier spectrometer is the instrument
of choice.
Fourier spectroscopy is most suited to absorption spectroscopy where (usually)
vapour samples are observed with a constant, well-regulated infra-red source
11.2 The multiplex advantage 119
providing the radiation power, in which case it has a marked advantage over grat-
ing spectroscopy both in accuracy and time. It is only to be undertaken on variable
sources whenthe frequencyspectrum
6
of the variationdoes not include the sampling
frequency or, more prudently, when a separate detector monitors the moment-to-
moment intensity uctuations of the source. But beware: this latter precaution does
not work if the spectral content varies frommoment to moment. It is the sort of thing
which happens in atmospheric research for example, by the selective absorption of
sunlight in the water vapour of occasional passing clouds.
To summarize: none of the techniques involved in Fourier-multiplex spectrom-
etry is to be undertaken lightly and for this reason the description of Fourier spec-
trometers is curtailed here.
6
The frequency spectrum is not to be confused with the infra-red spectrum: it refers to the frequencies (measured
in Hz) present in the moment-to-moment variation of the source strength.
12
Detectors
In spectrography, where spatial resolution is required, there is a clear subdivision
in detection technology between photochemical and photoelectric methods. So far
as chemical methods are concerned they possess fewadvantages over photoelectric
methods except possibly for observations at remote sites where a reliable supply
of electricity is uncertain or unavailable. They comprise the reduction of silver
halide salts to colloidal silver in a solid emulsion coated on to glass or on to a
transparent exible substrate.
1
The detective quantum efciency (DQE) of silver
halide crystals is in the region of 10
4
10
3
, the dynamic range is in the region
of 100:1 and as photodetectors they are severely non-linear in response, especially
where long recording times greater than 100 s are involved. ISOratings, usually
quoted in the range of 12 for high-resolution ne-grain emulsions, to 2000 at the
higher end of the range for severely coarse high- high-sensitivity emulsions,
2
fall rapidly at exposures longer than 1030 s. The effect, known as reciprocity
failure, is the notion that the blackening produced depends solely on the product
of intensity and time and is a necessary assumption if spectrophotometry is to be
done.
This sensitivity range may be compared with a liquid-nitrogen cooled CCD
detector with a DQE in the region of 0.5 and strict linearity over a dynamic range
of 25 000:1, a sensitivity broadly equivalent to an ISO of 3 10
6
.
3
Liquid-nitrogen
cooled CCD detectors require far more careful screening from light than silver
halide emulsions. An ordinary photograph with a cooled CCD camera and an F,2
lens in a typical photographers darkroom will give a clear, detailed picture of the
interior in a few seconds exposure.
1
Gold salts can also be made to follow the same photoselective reduction process.
2
Sometimes known as soot-and-whitewash emulsions.
3
No strict equivalence is possible because of the non-linearities and different dynamic ranges involved, and this
estimate is made by comparing the exposures at a shutter speed of 0.02 s needed for photographs of similar
quality.
120
12.1 Silver halide photography 121
12.1 Silver halide photography
Emulsions on glass plates have become comparatively rare in spectrography and it is
now the custom to rely on sheet lm. Spectrographic plates have traditionally been
large, 75 mm 250 mm being typical, and the emulsions coating them have been
specialised to cope with the peculiar requirements of spectrography. Chief among
these is the need for uniform sensitivity across the whole spectral range covered by
the instrument and generally instruments which deal chiey with infra-red radiation
are not required to cope simultaneously with the visible and UV regions.
12.1.1 Infra-red emulsions
Emulsions are available from photographic material manufacturers which will
record radiation to 1.2 m in the infra-red. These have a comparatively coarse
grain so that a spectrograph with high dispersion is needed for high-resolution
spectrography. The roll lm, sheet lm or plates must be kept in a refrigerator until
shortly before they are to be exposed, to avoid themfogging fromthermal radiation.
Condensation is then a problem especially in tropical regions, and the installation
of a drying agent such as silica gel in the spectrograph is advisable well before the
plates are introduced. Appropriate development conditions are required afterwards
and the manufacturers instructions must be followed closely for satisfactory re-
sults. If any sort of photometric precision is intended, the technique is an exacting
and unforgiving one.
12.1.2 Visible and near UV emulsions
For the visible region and near UV, so-called astrographic emulsions are available
which were developed initially for astronomical research and which possess rea-
sonably ne grain and uniform sensitivity throughout the visible region. The point
which must be emphasised here is that if spectrophotometry is to be attempted,
it is important that all the plates or sheet lms come from the same batch, that
they be developed and xed under rigidly identical conditions of time, temperature
and agitation, and preferably with the same batch and conditions of mixing of the
developer. De-ionised water is recommended but is not of overriding importance.
4
This is provided that the water is free from hostile cations, i.e. that it has a pH
higher than 7.
The bugbear of reciprocity failure at long exposure times has been the subject of
a great deal of experiment, chiey with the use of reducing agents to remove any
4
The author was advised by a manufacturer that, on an eclipse expedition to a tropical island (Manuae, Cook
Islands, 30 May 1965), their plates could if necessary be washed in sea-water, the nal clearance washing being
deferred until a later date.
122 Detectors
trace of free oxygen. Hydrogen bathing and soaking in a solution of ammonia have
been claimed to be effective, but following the development of the CCD, they are
no longer worth considering seriously.
With experience, sufcient uniformity can be reached with the silver halides that
spectrophotometry to 1% is possible. This seems to be a statistical limit, such that
there is little improvement on it even with large amounts of repetitive data.
Emulsions can be sensitised to the far UV by suitable phosphorescent materials
such as uorescein, and sensitised emulsions are available commercially. For the
extreme UV (XUV) below 2000

A Ilford Q plates are to be preferred. These
are manufactured, chiey for the nuclear industry, without the clear top-coating
of emulsion, so that the silver halide crystals actually reach the surface. Such
emulsions, on thin glass plates, can be used in vacuum spectrographs and can be
bent on a mandrel or in a plateholder to t a 500 mm radius Rowland circle. The
emulsion surface is sensitive to touch and so the plates require careful handling
in the darkroom when they are loaded. As usual with silver halide emulsions,
different grades are available with different crystal sizes and corresponding different
sensitivities.
Included in the paraphernalia of traditional silver halide spectrophotometry must
be a plate measuring machine or travelling microscope, and a densitometer is re-
quired if accurate photometry is to be done. Calibration exposures must also be
made, although the method of the common calibration curve is a useful alterna-
tive.
5
12.2 Elementary electronic detectors
12.2.1 Photocells and photomultipliers
In the age of the monochromator the photomultiplier led the way for 50 years. It
is a vacuum tube and comprises typically a photocathode deposited on an end-
window a thin lm of silver-silver oxide overcoated with caesium, antimony
or other substances with a low work-function, to act as a photoelectron emitting
surface, and a succession of dynodes: electrodes with a slatted structure similar
to a Venetian blind and held at successively increasing positive voltages to attract
electrons sufciently strongly to provoke secondary emission and thus to act as
current ampliers. A single photoelectron expelled from the photocathode will
provoke a cascade through anything up to 13 successive dynodes until a pulse of
10
7
electrons arrives at the anode. The pulse may have a duration of substantially
less than a nanosecond
6
allowing high pulse rates to be collected. The pulses can
5
See Appendix 4.
6
Bearing in mind that the speed of light is 1 foot per nanosecond, this is not so surprising.
12.3 Detectors with spatial resolution 123
be collected as a photocurrent or, especially if the radiation rate is very low, can
be amplied and counted as individual pulses in a digital recording system. Either
way, the DQE of the device is in the region of 0.150.35. Photomultipliers must be
cooled, usually to 80

C by dry-ice or a Peltier cooler to reduce thermal electron

emission to a low enough level to allow faint light photometry.
The sensitivity of the photosensitive surface is wavelength dependent and there is
little or no sensitivity at longer than 1.2 m. Ultra-violet sensitivity can be improved
by using quartz windows or by covering the glass with a thin layer of phosphorescent
material wavelength converters typically uorescein, vacuum-grease or sodium
salicylate. Sensitivity to the phosphorescence from these materials continues down
through the XUV to the soft X-ray region.
The chief sources of noise (dark-counts) are thermal emission from the photo-
cathode, cosmic ray

Cerenkov emission in the glass and radioactivity in some of
the materials, usually
40
K, in the glass.
There is an hour or two of high dark-current after the PMT has been exposed to
light and this is due to phosphorescence in the glass.
12.3 Detectors with spatial resolution
12.3.1 The charge-coupled device
The detailed electronics need not concern us. It is sufcient here that a CCD is a
rectangular plane surface covered with a two-dimensional array of pixels, each
one a capacitor capable of holding up to half a million electrons; that the capacitor
becomes charged with photoelectrons when light falls on it; and that the DQE, which
is strongly wavelength dependent, is of the order of 0.5. The pixels can be read-out
(destructively) in sequence and the charge contained in each can be measured with
an accuracy of 5e. The size of each rectangular pixel may be as large as 25 m
square (they are mostly but not necessarily square) and, at the time of writing, as
small as 5 m5 m. The number on a chip, again at the time of writing, is as low
as 80 80 for some applications, and with 9 m pixels, as large as 4096 4096.
Small pixel size is not necessarily an advantage in spectroscopy, bearing in mind
that the efciency of the spectrograph depends on the total area of the chip, not on
the number of pixels it holds. The resolution afforded by a given size is described
in the next section.
12.3.2 The dispersion relation
There is a fundamental relation, irrespective of the type of mounting or grating in
use, between the spectral bandwidth falling on to one pixel of a CCD, the grating
124 Detectors
constant, a, the order of diffraction, n, the pixel width, p, the width, W, of the ruled
area of the grating and the nal camera focal ratio, F.
(1) From the grating equation we have cos r dr = n d,a.
(2) The pixel width receiving the bandwidth d is f
2
dr, where f
2
is the focal length of the
output mirror (or lens).
(3) The projected width of the grating on to the output mirror is W cos r.
The focal ratio of the output mirror is then f
2
,W cos r and W cos r dr,f
2
is a
HelmholtzLagrange invariant of the system which holds good even when there is
re-imaging of the spectrum by a new output lens.
From these equations we can eliminate cos r and dr to give a universal relation-
ship
= pa,nWF. (12.1)
This may well be taken as the starting point for the design of a CCD spectro-
graph.
One caveat: this relation holds only when the grating is the iris of the system. In
the mounting described in Subsection 8.8.6 at the end of Chapter 8, for instance, it is
the camera pupil diameter rather than the grating which determines W in Eq. (12.1).
12.4 Exposure limitations
The chief cause of problems with the cooled CCDdetector comes frombackground,
chiey cosmic, radiation. At sea level, particles leave tracks or more often clumps
of 510 lled pixels. The diurnal variation is large but on average is 50 CREs
cm
2
hr
1
. These clumps, called CREs (cosmic ray events) can usually be recog-
nised because they contain a much greater density of charge than normally exposed
pixels. There is some dependence on the orientation of the CCD. It is no great
matter to write a short program to detect such clusters in the read-out array and,
if appropriate, change the affected pixel contents to zero. Due allowance can then
be made in the reduction process summing pixels along a spectrum line for in-
stance to disregard such empty pixels. The rate at sea level is such as to interfere
substantially with spectrometric exposures much longer than 30 minutes, and a
sequence of several exposures, median-ltered after read-out,
7
may be preferable
to one long exposure.
Another, similar cause of blackened pixels is occasionally to be found in the lens
of the CCD camera. Modern lenses with focal ratios of F,2 or less often contain
7
The median is taken of the corresponding pixels in the sequence of exposures and a new photograph is compiled
from all these chosen medians. This process is adequate to eliminate nearly all the CREs on the frame. Three
frames is a minimum, four or ve is better.
12.5 CCD software 125
elements made froma glass whichincludes a thoriumsalt inits recipe. Sucha glass is
more than usually radioactive, not seriously enough to be a health hazard but enough
to cause a rash of lled pixels on the CCD. A Geiger counter is a useful accessory
when buying a CCD camera lens. Other possible sources of read-out noise are the
materials of construction of the laboratory building housing the camera. A CCD
photograph of a completely dark sub-basement room with unpainted brick walls
will show the outline of the concrete between the bricks after about ve minutes
exposure. This is fromthe radioactivity (chiey
40
K) in the cement. Simultaneously
it will be discovered that most materials in everyday use, including emulsion paints,
are phosphorescent with decay times of up to several hours. This is a factor to
consider seriously when choosing the materials of construction of the spectrograph.
Test exposures with a newly constructed spectrograph with the entry slit completely
closed may cause some dismay to the designer at rst.
12.5 CCD software
Close inspection of the software attending a LN
2
- or Peltier-cooled CCD camera
is advisable. It is most likely that suites of programs have been written by the
manufacturer to take care of spectroscopic applications, but there are certain items
which are important if the best use is to be made of the camera. Most importantly,
the whole of the raw data, that is, the two-dimensional array of pixel contents, must
be available, pixel-by-pixel. All data processing must be accessible to the worker,
and for this the FITS format is desirable.
8
Useful operations on the data include:
(1) All individual pixel contents must be available for inspection and alteration, preferably
in a frame of about 8 8 pixels. Occasionally it is necessary to smooth the data by
replacing pixel contents with numbers in adjacent pixels. This is particularly so when
CREs are obtrusive.
(2) There should be a graphic display of raw data either as a photograph or line-by-line in
graphical form. Similar displays of processed data are also highly desirable.
(3) All arithmetic operations must be available, such as adding, subtracting, multiplying,
dividing and logic operations on corresponding pixels in two or more frames.
(4) Median ltering of contiguous sets of frames is required.
(5) Binning of frames in the x- and y-directions separately or together by user-prescribed
amounts should be possible, in order to alter the aspect ratio, with or without averaging
the binned pixels. This includes binning down to a one-dimensional output to make
spectrograms.
(6) Output should be available in ASCII format for further processing by external, probably
spreadsheet programs.
8
See Section 10.8.
126 Detectors
(7) Rapid scanning of a chosen line comprising a few (2030) pixels is useful for focusing
the camera. Similarly, rapid scanning of a small frame of perhaps 50 50 pixels is
useful for focusing and camera adjustment generally.
12.6 CCD calibration
The CCD sensitivity varies from one pixel to another, and from one exposure to
another, andone maynot relyonreproducibilitytobetter than1%. EachCCDmust
be tested by itself for consistency and the results are sometimes disappointing. CCD
photographs are generally divided by a frame which has (hopefully) been exposed
to uniform illumination in order to eliminate errors due to pixel-to-pixel variation.
This is sometimes called pre-whitening or eld attening, although the latter is
really an optical term.
This must be done through the camera with the same lens and at the same
focal ratio as the observational photographs.
The pixel contents are then normalised to unity so that the division does not
affect the overall exposure.
The chief problem lies with the phrase uniform illumination, which is more
difcult to achieve than is generally realised. Plain white (i.e. high albedo) surfaces
are rarely perfect Lambert radiators. Thick, translucent sheets of polythene are a
reasonable approximationanda 75mmradius hemisphere cut fromPTFE, andmade
to cover the lens, is suitable for 35 mm camera lenses, provided it is illuminated
by diffuse light. A straight-sided sh tank, lled with fresh water and half a litre of
fresh milk, illuminated from a 10-metre distance by a small diameter source, will
serve for larger apertures, but reections are a problem and there must be complete
darkness behind the camera lens. One may test for uniformity by making successive
exposures with the optic axis at different angles to the surface normal of the glass.
The local variations of sensitivity will depend on the wavelength of the light and
the resultant reference exposures are at best no more than an average.
12.7 Spectrograph calibration
There are so many variables in the optical path of a spectrograph that calibration is
totally empirical. A black body at a known temperature is the only feasible method
of absolute calibration, and even then one must be sure that there has been a total
suppression of other orders and of scattered light in the spectrograph. Reliable
absolute calibration is such an onerous task that one must question the need for
it in most circumstances. For practical purposes a comparison with a standard
source is sufcient. A reasonable approximation no more can be obtained by
exposure to a card coated with freshly deposited magnesium oxide from burning
12.7 Spectrograph calibration 127
magnesium ribbon and exposed to the midday sun, which in the visible at least, is
an approximation to a black body at 6000 K. A white card will not do, neither will
a painted white matt surface. Both are likely to be uorescent under the solar UV
radiation.
The resultingspectrum, after correctionfor the CCDarea variations of sensitivity,
may not look like a black body spectrum and for good reason. Apart from the basic
variations with wavelength of the CCD overall sensitivity, there are the reection
coefcients of the various mirrors and, above all, the variations with wavelength
and polarisation of the grating rulings, and the variation of these with angle of
diffraction. Spectro-polarimetry with a reection grating is probably a lost cause
for this reason.
13
Auxiliary optics
13.1 Fore-optics
The correct use of a spectrograph requires that its aperture and its eld be lled with
light. More than this is not possible. Less than this is inefcient and may record the
spectrum incompletely or incorrectly.
For most efcient use, the light source should be imaged on to the entry slit with
a condensing lens, at a magnication which allows the whole slit length and width
to be illuminated. The aperture of the condensing lens must be enough to allow the
diverging light beam, after passing the slit, to ll the grating with light.
1
When this is done the condensing lens must itself be imaged on to the grating (not
the collimator) by a relay lens
2
somewhere near the entry slit, at a magnication
sufcient toll the gratingwiththe image. If the condenser has beenchosenproperly
in accordance with the rst paragraph this will happen automatically. The aperture
of the collimator must itself be large enough to avoid vignetting the grating, and
the length of the entry slit and the distance of the collimator from the grating may
well determine whether this is the case. The sort of fore-optical arrangement which
meets these criteria is shown in Fig. 13.1.
However, there are perils in imaging a structured source on to the entry slit unless
perhaps the source structure itself is being examined. In particular, in double-beam
spectrometry, it is particularly important that the two beams see the same light
distribution, and in this case the source should be imaged on to the grating
3
and not
on to the entry slit(s).
1
The use of lens and grating here is for economy. The words mirror and prism are implicit, as the principles
are the same for all optical instruments.
2
Also called a eld-lens, or sometimes a Fabry lens after its inventor.
3
Or on to its image formed by a Fabry lens before the entry slit.
128
13.1 Fore-optics 129
source
condenser lens
Fabry lens
spectrograph collimator
entry slit grating or prism
Figure 13.1 The standard method of illuminating a spectrograph. The essence of
the arrangement is that both the eld (the entry slit) and the aperture (the grating)
are lled with light from the source. More is not possible; less is inefcient.
13.1.1 Telecentric operation
It sometimes is important when examining a remote extended source such as the
night sky to be certain that all the line images receive their light from the same area
of the source. To do this the source should be imaged on to the entry pupil of the
instrument.
The simplest way to ensure this is to place the input (M1) mirror of a

Czerny
Turner spectrograph at its focal length fromthe grating. The chief-rays frombundles
from all parts of the entry slit are then necessarily parallel to the optic axis when
they reach the grating, so that all ray bundles have come from the same part of the
distant source. The source is itself focused on to the grating which itself acts as a
telecentric stop.
The input mirror may thus be at a different distance from the grating from the
output M2 mirror which, if it is a paraboloid, should be at two-thirds of its focal
length from the grating to ensure a at eld.
The alternative way of ensuring telecentric input is to put a Fabry lens at or near
the entry slit, its focal length chosen so that together with the M1 mirror it images
the distant source on to the grating. If the M1 mirror is also at the two-thirds focal
length position, this may need a very long possibly impractically long focal
length in the Fabry lens.
13.1.2 The telecentric stop
Telecentric is a technical term taken from the discipline of prole projection and
cathetometry. It implies that chief-rays of ray bundles fromthe object are all parallel
to the optic axis, so that a true silhouette of an object can be photographed.
In spectrography it may happen that an active beam of particles a beam of
excited molecules for instance is to be examined and that a spectrum is desired
with the beam imaged on to the entry slit, perhaps to measure the decay rates of
various transitions. It is important then that there should be telecentric illumination
130 Auxiliary optics
molecular beam
molecular beam
condenser lens
entry slit
grating
Fabry lens
telecentric
stop
beam collimator
spectrograph collimator
focusing
lens
spectrograph M1 mirror
grating
spectrograph
entry slit
telecentric stop
(a)
(b)
beam collimator
Figure 13.2 Two methods of achieving telecentric input to a spectrograph. Pre-
cautions like this are necessary for avoiding Doppler effects for example, when
examining phosphorescent decay times in molecular beams. All chief-rays are
perpendicular to the beam direction.
(in the object space only) in order that errors of wavelength due to the Doppler
effect are avoided.
For this a lens or mirror is necessary with a diameter greater than the length
of beam to be examined.
This lens acts as a collimator and at its rear focus is placed a stop, the telecentric
stop, which ensures that all ray bundles have their chief-rays parallel to the optic
axis. At or near this stop is placed the condenser lens which focuses an image of
the beam on to the entry slit. A Fabry lens in front of the slit images the condenser
on to the grating as shown in Fig. 13.2. Alternative arrangements are possible but
this one allows the smallest diameter lenses to be used.
If accurate imaging of the beam is required, astigmatism must be avoided and it
may be necessary to use Littrow optics to achieve this.
13.2 The astronomical telescope as fore-optics
If the spectrograph is to be attached to an astronomical telescope, then in the ideal
case the star would be focused on to the entry slit and the telescope objective
simultaneously imaged on to the grating. This however is a doctrine of perfection.
The chief problem is that even if the imaging is perfect the diffraction-limited star
image is far too large for the spectrograph entry slit. If the spectrograph input
13.2 The astronomical telescope as fore-optics 131
collimator is made long enough for the stellar image to t the slit width, the image
of the telescope objective is far too large for the grating. Neglecting atmospheric
seeing turbulence, the spectrograph used at its full resolving power should have
an entry slit width such that the principal lobe of its diffraction pattern just lls the
grating. For perfect efciency therefore, the grating would have to be the same size
as the telescope objective when the diffraction-limited image of a star falls on to
the slit. Therein lies the essence of the problem.
Many ingenious solutions have been tried with varying degrees of success, none
of themoverwhelming. The image slicer family takes sections of the stellar seeing
disc on either side of the slit aperture and redirects the light through a more or less
complicated set of mirrors to a part of the slit above or below the seeing disc.
Specially formed bre optic bundles offer a more reasonable solution to the
problem.
4
For example, at the Cassegrain focus of a 4-metre telescope, working
at F,12.5, a 0.5 arcsecond seeing disc the smallest one may reasonably expect
at or near the bottom of the atmosphere has a diameter of 125 m. An F,12.5
spectrograph with a 200 mm wide grating at a resolution of 10
5
requires an entry
slit width of 12.5 m, and a focal reducer to convert the F,3 output beam from
the bre to the F,12.5 of the spectrograph must start with an object diameter
the end of the bre of 3 m. Therefore 1800 such bres are required to collect
the light from the whole of the seeing disc and will convert it to a line 3 m wide
and 5.4 mm long. This becomes, on magnication, 12.5 m wide and 22.5 mm
long, about as long a slit as the spectrograph can tolerate without unacceptable
aberration.
13.2.1 Slitless spectrography
The stellar astronomer has the advantage that the light sources to be investigated
are already collimated and are (optically) at , and the function of the telescope
is to collect as much light as its aperture will allow while retaining the collimation.
The problem has already been pointed out, that the stellar image, even when not
broadened by atmospheric turbulence, is too large to allowefcient high-resolution
spectrography. The chief advantage of slitless spectrography therefore is in its use
with low magnication, or no magnication at all, to examine simultaneously a
large number of sources, each of which will record its own spectrum on a different
part of the eld. The application is therefore in taxonomy the measurement of
the relative abundance of different stellar spectral types for example, or the relative
4
Fibre bundles with individual bres a few micrometres in diameter may have a circular form at one end to
receive the starlight, and be shaped to a single column of bres at the other end to match the input slit.
132 Auxiliary optics
abundance of galactic spectral types Seyferts for example in a cluster: this latter
at high magnication of course.
The spectrograph input optics then are similar to those of the eye: the eld of stars
at the telescope focus is re-collimated by the eye lens, and the eld lens
5
images
the telescope objective on to the grating. The output optics of the spectrograph
focuses the star eld on to the photographic plate or CCDdetector. In the latter case
the focal reducer, described in the next section, may be necessary.
High resolution is not to be expected in slitless spectrography, spectrograph
aberrations being what they are, but enough information can be gathered by this
technique to make it a valuable asset in astrophysics.
13.3 Focal reducers
The focal reducer was invented originally by Meinel
6
and separately by Court` es
7
as a means of increasing the relative aperture of an astronomical telescope. In effect
it converts a 5-metre telescope into a 35 mm camera with a 10-metre, F,2 lens.
The principle is that of a giant eyepiece. Following the Cassegrain (or any other)
focus, the paraxial beam is allowed to expand until it has the same diameter as the
nal camera pupil. It is then collimated with a simple achromat working at the same
focal ratio, typically F,14, as the Cassegrain telescope. Simultaneously a eld lens
just beyond the Cassegrain focus images the telescope objective on to the camera
entry pupil.
The device can be further simplied.
8
Using a modern camera lens with a macro
facility, one may dispense with the collimator, and a simple eld lens of appropriate
focal length is the only addition required. The depth of focus is such that it need
not be achromatised and a simple plano-convex lens is adequate. The camera pupil
is placed at the image of the telescope objective.
The device may usefully be applied to spectrography, especially for the study of
faint sources. Aproperly designed focal reducer will convert a

CzernyTurner CCD
spectrograph to a focal ratio of F,2 or thereabouts without loss of resolution. The
principle is the same: a collimator captures the beam expanding after the primary
focus, at a diameter equal to that of the nal camera entry pupil. Simultaneously a
eld lens just beyond the

CzernyTurner focus images the grating on to the camera
entry pupil. The focal reducer then designs itself, being constrained by these two
requirements. The design should be that of a simple Ramsden eyepiece and both
eld lens and eye lens are plano-convex with the convex surfaces facing each other.
5
These words being used by analogy with a telescope eyepiece.
6
A. B. Meinel, Astrophys. J., 124 (1956), 652.
7
G. Court` es, Ann. dAstrophys., 23 (1960), 115.
8
J. F. James, Q. J. R. Astron. Soc., 22 (1981), 244.
13.3 Focal reducers 133
entry slit
M1
M2
grating
CCD camera
eye lens
field lens
primary focal plane
focal reducer
Figure 13.3 A

CzernyTurner spectrograph with a focal reducer. The device
achieves a low focal ratio without loss of resolution, but cannot be added to an
existing design. The

CzernyTurner spectrograph itself must be designed to have
a at primary focal surface perpendicular to the zero-coma chief-ray, and this is a
matter for experiment with a ray-tracing program. Asimple Ramsden-type reducer
then adapts the spectrum to the CCD surface while the grating is simultaneously
imaged on to the camera pupil.
As Conrady pointed out, attempts to improve the resolution by bending either lens
are doomed to failure or, in his own elegant phrase, do not open any very alluring
prospects.
9
The eld lens focal length is chosen to image the grating on to the
camera entry pupil at a distance of [(n 1),n] f
eye
beyond the eye lens, at the
position that is, of the coma-correcting stop for the plano-convex eye lens. An
example of the arrangement is shown in Fig. 13.3.
The Ramsden type of eyepiece is to be preferred because the

CzernyTurner
focus is accessible so that, if necessary, stops can be placed there to intercept
line images which are overwhelmingly bright and scatter unwanted light into the
camera. A uorescent screen can also be placed there for observing UV spectra
beyond the scope of the detector. (Such a screen may be of plane glass coated with
a thin transparent lm of a uorescent material.)
A disadvantage is that this simple focal reducer will require refocusing if a
different part of the spectrum is to be examined, and the eye lens may need to be
an achromatic doublet to avoid this. The proper position for the camera entry pupil
may be calculated or, more likely, will be found by experiment with the ray-tracing
program. There is no sensible contribution to the chromatic aberration by the eld
lens which may remain a simple plano-convex element.
A small amount of chromatic aberration can be deliberately introduced to the
spectrograph itself, by moving the entry slit inside or outside its true focus. This
causes the grating to be illuminated by slightly convergent or divergent light and
the convergence after diffraction is diffraction-angle, i.e. wavelength, dependent via
the cos i , cos r factor in Eq. (8.10). The nal focus then depends on the wavelength,
9
A. E. Conrady, Applied Optics and Optical Design (London: Oxford University Press, 1929), Vol. 1, 482.
134 Auxiliary optics
and the slit position shift a few mm is small enough that there is no appreciable
alteration in the other aberrations.
A1-metre

CzernyTurner spectrograph with a grating of 100 mmruled width and
12 000 rulings/cm will spread an 800

A band of the visible region across 110 mm
at F,10 at the primary focus, and the focal reducer will convert this to 26 mm at
F,2, covering a 25.6 mm CCD with 1152 pixels at a resolution of 0.7

A.
13.4 Schmidt-camera spectrography
The Schmidt camera is a well-known optical device for achieving a very high
numerical aperture at a large eld angle. It has been applied to spectrography for
the observation of very faint extended sources such as the zodiacal light and the
night airglow. The input optics is conventional, with a long-focus collimating mirror
or lens and a plane grating, but with the output optics replaced by a Schmidt camera.
The property to be used is the high numerical aperture, and no great eld angle is
required. With a grating constant of 6000 rulings/cmthe visible spectrumis covered
in rst order by a eld of 3.5

.
Briey, the Schmidt camera uses the condition for zero coma in a spherical
mirror given by Eq. (4.8): the pupil must be at z = 2 f , that is, at the centre
of curvature of the mirror. At this pupil a zero-power corrector plate gured to a
fourth-order surface curvature is placed to remove the spherical aberration, which
at this high numerical aperture would be severe. Symmetry precludes astigmatism
(Eq. 4.9) but eld curvature must be accepted as the price to pay for removal
of the other aberrations at such a high numerical aperture, and herein lies the
problem with spectrography. In a large Schmidt camera a photographic plate can
be bent over a mandrel to accommodate the eld curvature and in a small one the
emulsion can be coated on to a spherical glass substrate; or a eld-attening optical
system can be placed in the appropriate position. But a CCD detector requires such
a bulk of electronics and refrigeration that a totally unacceptable obstruction to
the beam would be incurred and it is this which makes the Schmidt conguration
impracticable for CCDspectrography. Ahigh numerical aperture depends therefore
on a focal reducer followed by a good quality camera lens of low focal ratio.
10
13.5 Scattered light and bafing
The purpose of including bafing in the design is the reduction of scattered light.
10
For most spectrographic purposes a standard double-Gauss 35 mmF,2 camera lens, of which there are probably
a hundred designs on the market, is adequate.
13.5 Scattered light and bafing 135
13.5.1 Causes of scattering
Scattering comes chiey fromuneven edges to the input slit, fromdust and scratches
on any of the optical surfaces, from imperfections in the grating rulings and from
light coming through the entry slit which reaches any of the walls of the interior of
the spectrograph. Bafes are simply opaque materials in the form of walls or tubes
which intercept and absorb unwanted light. They may be inside the spectrograph
or outside or both.
Decient optics are of course a matter of quality control by the manufacturer,
and large, thick lenses of glass or quartz are more likely to have striae, stones, sleaks
and scratches than small ones. Mirrors are superior in this respect but still must be
manufactured, coated and chosen with care. Dust particles on optical surfaces are a
matter of cleanliness and accessibility. Gratings, by the nature of their construction,
scatter more than most optical elements and for this reason prisms are still to be
considered when scattered light is more of an obstacle to research than lack of high
resolution.
13.5.2 Scattering from edges
The edges of the entry slit were mentioned above. Any edge which can be seen
from the grating is suspect, because of the possibility of Fresnel diffraction effects.
The sharp edge of a lens-hood is a particular culprit even in a camera, and may
scatter enough light on to the lens surface to be re-scattered into the interior. For
true bafing, several such edges must shield the lens, each one shielding from the
last: alternatively, a trumpet-shaped bafe will show a smooth horizon rather than
a sharp edge to the input optics. The problem is particularly acute in space-borne
spectrographs where it is desired to acquire stellar spectra from anywhere within
90

of the Sun.
13.5.3 External bafes
The principle to followwhen judging where to incorporate bafes is that the grating
must see nothing except light from the external condenser, and that light entering
through the slit can go nowhere except to the grating.
An external bafe in addition to the internal bafe may help to achieve this.
There is an image of the grating formed by the collimating mirror and the input
Fabry lens. It is situated before the entry-slit/Fabry-lens assembly. There may well
be a condensing lens there, but whether there is or no, this is the place to put a real
stop, its aperture cut to the shape of the grating, inserted in the optical train and
136 Auxiliary optics
tilted to match the image of the tilted grating. The tilt is not large, bearing in mind
that the longitudinal demagnication is the square of the lateral demagnication of
the grating image which it covers. It must follow then that light which enters the
spectrograph through this stop, unless scattered by the input slit, must fall on to the
grating. Similarly, if there is a re-imaging of the spectrum, for example through a
focal reducer, there may be another opportunity to insert a made-to-measure stop
at the image of the grating on the output side, although this will coincide with the
camera entry pupil.
13.6 Absorption cells
Absorption spectra of liquids and solids require only the simplest of absorption
cells, typically with paths from a few micrometres to a few centimetres, and such
cells are normally placed just before the entry slit, sometimes covering only one
half of it so that comparison spectra come through the other half. If this method is
adopted, then the designer must be careful to see that the whole slit length receives
identical illumination and it is the light source and not the condenser which must
be imaged on to the grating by the Fabry lens. Strictly speaking, the light should
be collimated, but in practice a low numerical aperture, equivalent to F,12, is
adequate to give the accuracy normally required. The marginal rays in the cell
differ in length by less than one part in a thousand from the paraxial ray.
13.6.1 White cells
Serious problems only begin to arise when the absorption is very low so that long
paths are required.
In the White cell,
11
long path absorption is achieved through the periscopic
principle, in which a beam of light is sent down a tube with relays of lenses which
achieve a useful eldof viewwhile conningthe light tothe interior. Inanabsorption
cell the lenses are replaced by mirrors, and absorption path-lengths of tens of metres
can be achieved in a 1-metre long tube.
Three concave spherical mirrors are employed, all of the same radius of curvature
(Fig. 13.4). Light is admitted through an entry slit and is reected off mirror 1 to a
focus at a point on the surface of mirror 2. From mirror 2 it is returned to the centre
of mirror 3 and thence to a focus at another point on mirror 2. The process continues
until a nal focus is reached outside mirror 2, whence it goes to the spectrograph.
Figure 13.4 shows the chief-ray as it passes through the cell with seven reections.
11
J. V. White, J. Opt. Soc. Am., 32 (1942), 285.
13.7 Fibre optical input 137
1
6
5
2
7
4
3
8
entry slit
exit slit
Figure 13.4 The White cell. This is a device for measuring feeble absorption coef-
cients where long path-lengths are required. The device uses multiple reections
between three mirrors and depending on the reectivity of the mirrors, the number
of reections may be as high as 30 in the visible and several hundred in the FIR
where reection coefcients are higher.
The actual number will depend on the tilt of the two smaller mirrors, which may
be adjustable.
The number of reections available depends on the reection coefcients of
the mirror coatings. An untarnished silver coating has a reection coefcient
in the region of 0.92 and transmits 10% of the incident light after 27 reec-
tions. Higher reection coefcients and longer paths are possible over restricted
wavelength ranges with multilayer coatings. Gold coatings in the infra-red, with
reection coefcients in the region of 0.980.99, will allow several hundred
reections.
13.7 Fibre optical input
The salient facts which concern us in spectrography are that multiple bre bundles
are used rather than single bres, and that a circular luminous object, such as a star
image, can be converted to a line image suitable for the input slit of a spectrograph.
A single bre in a bundle may have a diameter anywhere from 1 m to 25 m,
10 m being typical, and the output beam from the bundle emerges at a focal ratio
in the region of F,3. The illumination is Gaussian rather than uniform, but should
be converted by a lens to F,12 so that all the emerging light will be accepted
by the collimator. The apparent slit width will then be four times greater than the
diameter of the bres at the output end and the apparent slit length will also be four
138 Auxiliary optics
times greater than the length of the output end of the bre. Alow-power microscope
objective may be a suitable converter, focusing the bre-bundle end on to the entry
slit. A Fabry lens at the slit will be needed to image the converter pupil on to the
grating. Alternatively a virtual magnied image of the bre end may act as the entry
slit, in which case aplanatic lenses are indicated.
14
Optical design
Le mieux est lennemi du bien.
(The best is the enemy of the good.)
Voltaire
14.1 First steps
Once the task for the spectrograph has been dened, a suitable type may be chosen
and a catalogue search made for a possible manufactured model. Many factors
affect the decision. The rst of course is tness for purpose. Time and cost of
manufacture may be factors and the facilities available for local or in-house design
and construction are primary considerations. A decision on whether to buy or to
design and build a dedicated instrument must rest on such an appreciation.
It is a mistake when doing fundamental or academic research to construct a
more elaborate, high-performance or expensive instrument than the immediate task
demands, possibly in the hope or expectation that it may prove useful for some other
investigation at some later date. In the authors experience this is almost never the
case and the chief result is delay and unnecessary expense. In the long term the
instrument is a sad relic, cannibalised of its optical components and left to decay
in the attic with its ingenious mechanisms and precision micrometers.
14.2 Initial layout
Once the spectrograph type has been chosen and its main parameters such as wave-
length range, resolution, type of detector, number of resolved elements, etendue etc.
have been decided, a sketch can be made, and the traditional back-of-an-envelope is
as good a place as any for this. Manufactured mirrors and lenses must be considered
rst before taking a decision to have optical parts made, and it may well be dis-
covered that the design can be adapted to use bought-in elements. Almost certainly
139
140 Optical design
the optical components will come from a makers catalogue. Rarely a bespoke item
must be ordered from a specialist manufacturer and if this is so, optical manufac-
turers are usually happy to let the designer have lists of their tool radii, of suitable
glasses at their disposal, with their optical properties and the costs of manufacture.
Catalogues of optical, opto-mechanical and opto-electronic components must be
available to the designer, quite probably from the Internet.
14.3 The drawing board
The next stage is to compute the Gaussian optics and to make the outline optical
drawing, and this will determine the sizes and positions of the optical components.
When making the rst sketch it is both clearer and more instructive to lay
out all optical components in position as if they were transmitting light rather
than reecting it. Consider all the mirrors as if they were lenses and the grating
as if it were a transmission grating.
The x- and y-scales on the diagram need not be the same. The ray paths are not
affected and the positions of stops and pupils will be unaltered. A CAD (computer
aided design) program may lend some precision to the drawing, but the old ruler,
pencil and compasses is probably adequate and may well be more convenient at
this stage.
If the instrument is to be of the

CzernyTurner or EbertFastie type, this is
the time to choose input- and output-mirror off-axis angles for the chief-rays. It
is a simple matter to write a short computer program which, given the zero-coma
wavelength and the sumof the two off-axis angles, will compute, on the assumption
of parallel input and output chief-rays, the and angles, the grating tilt and the
meridional coma for rays on either side of the ZMC direction. If the instrument is
to have a Littrow conguration, the off-axis angle should be chosen so as to allow
adequate spacing for the various components with room for the mirror and grating
cells, auxiliary mirror mounts, slits and CCD/plateholders. A computer ray trace
will determine whether the aberrations are under adequate control. Assume at this
stage that all mirrors will have a thickness equal to one-sixth of the diameter, and
that all positive lenses will have a rim thickness of at least 2 mm. Grating blank
sizes are to be found in the manufacturers catalogue.
14.4 Computer ray tracing
This is the point to emphasise once again that one cannot design an optical
instrument on a computer with a ray-tracing program.
The design must be done rst and set up on the ray tracer for renements to be
made. Remember that the Seidel aberration calculations are not exact and that small
14.5 Renement of the optical design 141
changes in the dispositions of the optical surfaces can be made by the program to
minimise the aberrations, reducing them sometimes to zero.
A typical ray-tracing program invites the operator to insert the x-, y- and z-
coordinates of the vertex of each surface. The z-axis is normally the optic axis. It
then requires the pitch and tilt angles of the surface, its curvature and excentricity,
whether it is reecting or transmitting, whether it is a grating and if so the number
of rulings per millimetre. Refractive indices will generally follow each vertex,
referring to the material following.
Next, the x-, y- and z-coordinates of the launch points of each ray are required
together with the sines of the angles which each one makes with the z-axis in the
xz and yz planes, and the wavelength associated with the ray.
Bear in mind that the outer 3 mm at the rim of a mirror cannot be relied on, as
there is more often than not a turn-down as the result of the polishing action. If
the manufacturer is supplying a 150 mm diameter mirror, it is prudent to assume
that the aperture is 140 mmwhen doing the optical design. In any case the retaining
plate which holds the mirror in its cell may well have such an overlap at the edge.
The program will trace the ray through the system with an accuracy far beyond
anything that can be measured in practice. It can be asked to adjust any (or all)
of the input parameters until the ray arrives at a denite given goal, usually either
the x- and y-coordinates of a point on the last surface or, if the output is to be a
collimated beam ready for a bought-in camera lens, the sines of the angles at which
the ray leaves the last surface. A multiplicity of rays, 20 or 30 perhaps, can be
treated simultaneously.
14.5 Renement of the optical design
Thus, when the outline design appears satisfactory sufcient room left for baf-
es, detector optics and fore-optics unlikely to interfere with each other etc. the
optical train can be set up in the ray-tracing program as outlined above. This is the
design stage which formerly was done by hand using seven-gure logarithm and
trigonometrical tables, when an experienced ray tracer could trace a ray through
eight or nine surfaces in 20 minutes and an experienced lens designer could make
an intelligent guess as to which parameter needed changing next before repeating
the process. A modern ray-tracing program will trace 30 rays through 20 surfaces
in as long as it takes to press the start key on the computer keyboard.
There is a sufcient variety of these ray-tracing programs on the market that it is
pointless to particularise too much. No great elaboration is required here, as focal
ratios are unlikely to be belowF,8, the limit at which simple achromatic doublets
provide adequate eld and resolution for a spectrograph. One aims generally at a
spatial resolution in the neighbourhood of 35 mdepending on the type of detector
142 Optical design
and the resolving power required. A simple geometrical ray-tracing program such
as BEAM 3 from Stellar Software
1
is adequate although it is strictly geometrical
and does not include renements such as diffraction effects at the image points.
In the event that a manufactured camera lens is to be used for the camera,
2
then
the spectrograph designers aim may be for parallel ray bundles at each computed
wavelength at the nal output pupil of the spectrograph.
14.5.1 Initial alignment
Once the optical elements are put in their places, a single ray, the chief-ray, at
the central wavelength is launched from the centre of the entry slit to the vertex
of the rst (collimating) mirror or lens. The mirror is given the tilt required by the
outline design and the grating (or prism rst surface) position may be adjusted by
the program to receive the ray at its centre.
The grating tilt will already be known from the outline design and this will have
already been entered in the ray-tracing program. Alternatively, if the zero-coma
wavelength has been decided, the grating can be tilted by the program to reect the
chosen wavelength to the vertex of the next surface.
The position of the second mirror is then adjusted to receive the diffracted chief-
ray at its vertex and its tilt is then adjusted by the program to send the ray either
towards the detector centre or possibly to be made parallel to the input ray. In the
latter case the detector vertex may be adjusted by the program to receive the
chief-ray.
Exact positions are not set rmly at this stage. The distance from the entry slit
to the M1 mirror of a

CzernyTurner mounting is found by tracing parallel rays
backwards, starting from the corners of the grating and nding the place where the
marginal rays would meet to forma comatic image of a virtual point at innity. This
is where the entry slit must go, and the procedure takes care of the eld curvature
of the M1 mirror and avoids any focus defect. Rays launched from a point source
at the slit centre towards the M1 mirror then arrive at the grating with the proper
amount of coma.
Procedures like this set out the initial dispositions of the major components, and
ray bundles of six or eight marginal rays may be launched from the same point on
the entry slit and adjusted (again by the program, which can adjust the sines of the
launched rays) to arrive initially at eight equally spaced points around the margin
of the iris of the system usually the grating. These are then traced on through the
system so that the nal image quality can be assessed.
1
Stellar Software, PO Box 10183, Berkeley CA 94709, USA.
2
Recall the advice in Section 12.4 regarding radioactive elements in low focal-ratio camera lenses for CCD
cameras.
14.5 Renement of the optical design 143
This is the stage where such items as coma, astigmatism and eld curvature can
actually be measured (as opposed to being calculated) and necessary adjustments
made to the positions and tilts of the various components to ensure compliance with
the original design goals.
The process is a lengthy one and experiments on the computer are generally
needed to nd the optimum positions. The departures from the positions calculated
by Seidel theory will not be large a small percentage probably but they will be
signicant, and the nal aberrations may well be substantially less than the original
estimates.
14.5.2 Elaboration of the design
Once optimum positions have been found for imaging a point source on the entry
slit, the length of the slit can be added as a newcondition and the effects of spectrum
line curvature can be measured. It is at this stage that compromises will be needed
to secure the best image and to discover the spectral range over which the whole
spectrograph will operate satisfactorily.
When the process is complete and a satisfactory optical design has been reached,
small changes should be made in the ray-tracing program to the positions on the
optical layout, to determine the tolerances that may be allowed in positioning the
various components. This in turn determines the accuracy which must be required
in the workshop when the instrument is nally constructed. More importantly, it
may well reveal that the manufacturing tolerances are too tight and that the design
cannot be realised in practice. This sort of defect occurs, for example, when a ray
is incident on a refracting surface at a large angle, and where a small displacement
of the surface vertex along the optic axis would entail a large change in the height
of incidence and consequently the ruin of all that follows. Considerations like this
play a signicant part in the design of camera lenses for example.
14.5.3 Local minima
A warning is appropriate at this point a general warning in fact to everyone using
a ray-tracing program. The program varies the positions, tilts, curvatures etc. as
it searches for a minimum departure from the goals set by the programmer a
focus perhaps or parallelism of rays in a collimated beam. The initial xed points
of the program may be the launch points of the rays and the refractive indices
of the transmitting elements. The variables are the bendings, excentricities, vertex
positions, tilt and pitch of some or all of the optical components (some, taken froma
manufacturers catalogue, may be taken as xed, except in position). Vertices can be
ganged together so that a lens or a group of lenses can be moved together as one unit.
144 Optical design
It is inevitable, with such a multiplicity of variables, even when only two or three
are allowed to vary at one time, that many local minima will be found such that any
small change in any one of the variables will produce a worse result. In practice a
great deal of experiment will be needed to nd the ultimate best arrangement of
the components, and the nal result, which may sometimes be surprising, must be
tested with a large variety of possible ray-input parameters.
Metaphorically, the climber is climbing a mountain in a fog, and must be certain
that she has actually climbed the mountain and is not standing on the summit of
one of the foothills.
The design found by computing the Seidel aberrations will almost certainly
need minor corrections. The at-eld condition of Eq. (4.12) will not give an
exact position for the grating and a small movement of the grating or the M2
mirror, of several per cent of the M2 focal length, may be required to produce a
eld at enough for the CCD detector. Computer experiment is the answer, and
it is instructive to plot, for example, a rough graph of the way the eld curvature
actually changes with M2 position.
14.6 Requirements of a ray-tracing program
The obvious function of a ray-tracing program is to allow you to make small
variations in such variables as you choose and to re-compute the rays to see whether
the variation moves them towards or away from a goal which you have chosen.
Several variables may be chosen at one time and this is the chief cause of the
programs nding local minima.
There are other useful functions of the programbeyond this. It should be possible
to compute spot diagrams at the nal focus or at any other intermediate point in
order to check the contributions of the various aberrations. In a CCD spectrograph
a square or rectangular iris at the position of a pixel or a column of pixels should
measure the proportion of randomly launched rays which enter the target area, a
potent measure of the efciency of the optical design.
There are other, more subtle variables which can usefully be displayed. Apart
from drawing the ray diagram, an xy plotting program should be available to
make graphs of one variable against another. For instance the so-called Htan U

plot,
3
in which an oblique parallel bundle of rays at some angle with the optic axis
is traced and a graph made of the intersection height, H, above the z-axis at the
image plane, against some other variable such as tanU

, the tangent of the angle at

which each ray emerges from the exit pupil, or H, the intersection height of each
ray at the entry pupil on the y-axis. A perfect lens would show the plot as a straight
3
R. Kingslake, Lens Design Fundamentals (New York: Academic Press, 1978), p. 144.
14.6 Requirements of a ray-tracing program 145
horizontal line. The ray-tracing program will show, in the shape of the actual curve
that is drawn, the effect of various aberrations of the system. A straight line with
a slope indicates eld curvature. An S-shape indicates spherical aberration. U- and
V-shaped lines indicate positive and negative coma respectively, and if the curve is
asymmetrical it indicates spherical aberration as well. Details of this technique are
to be found in Kingslakes book.
There is a world of difference between lens design and instrument design. Nearly
all spectrographs work at focal ratios greater than F,12 and eld angles less than
3

, the points at which serious lens design begins. Seidel theory is adequate to
give the iterative ray tracer a place to begin, and when low focal ratios are needed,
photographic lenses can be made to look at the collimated beams which emerge
from F,12 optical systems. The spectrograph designer will rarely have to deal with
anything more complex than the design of an air-spaced achromatic doublet. It
is the proper positioning of simple elements along the optic axis which chiey
gives renement, resulting eventually in a high resolution and etendue over a wide
spectral range.
14.6.1 Worked example
This is a 1-metre focal length

CzernyTurner CCD spectrograph intended for night
airglow spectroscopy, with angles chosen to remove tangential coma at 5000

A in
second order. The spectrograph was intended to photograph the spectrum between
4000

A and 6000

A in tranches of 500

A at a resolution of 0.5

A, turning the grat-
ing for each new spectrum and moving the focal reducer on a sliding carriage if
necessary to refocus.
A Ramsden-type focal reducer was added to put the spectrum on to a CCD with
1152 325 pixels of 25 m width covering 28.8 mm. The plano-convex lenses are
of BK7 glass and have focal lengths of 497 mm (eld lens) and 429 mm (eye lens).
The focal reducer requires a longitudinal movement of a few millimetres to bring
all sections of the visible region into focus. As it is a tranche of 500

A can be
photographed at any one time.
It was designed using the BEAM 3 program mentioned earlier.
Table 14.1 shows the layout of the ray-tracing program for the

CzernyTurner
spectrograph with a focal reducer made to order, taking surface radii from the
manufacturers catalogue. The specication was that the grating ruled width was
120 mm, the (paraboloidal) mirror focal lengths were 1000 mm and the nal CCD
camera lens was to be an F,1.8 50 mm focal length 35 mm camera lens. The design
requirement then was for parallel ray bundles to emerge fromthe focal reducer with
a diameter of 25 mm.
T
a
b
l
e
1
4
.
1
14.6 Requirements of a ray-tracing program 147
0
0
200
200
400
200
Figure 14.1 The optical layout of the

CzernyTurner spectrograph described in
the text. The focal reducer is a simple Ramsden-eyepiece design, that is to say, with
plano-convex lenses having the plane surfaces outwards. This serves to collimate
the emerging spectrum at a diameter of 23.4 mm and simultaneously images the
grating on to the nal image-plane where the pupil of the CCD camera lens would
be placed. The camera lens (not shown) may typically be a 35 mm camera lens
with a focal ratio of F,2 or less, and its optical quality is likely to determine the
ultimate resolving power of the instrument.
The initial layout had the angles appropriate to zero coma at 5000

A, with the
mirrors at z = 1000 and the grating at 2 f ,3 from the M2 mirror vertex, as required
by the Seidel at-eld condition. Rays were launched from z = 2.65 for minimum
coma at the M1 mirror, and the nal positions for M2 and the following optics were
achieved by the requirement for a at eld at the CCD, that is, for the emerging ray
bundles at 5000

A (paraxial) and 5250

A (marginal) to be parallel, and in particular
to have the x-direction cosines all as closely as possible equal to each other. The
y-direction cosines which describe the astigmatism were not subjected to the
iteration process. The question marks in the delimiter columns following the lens
vertices were instructions to the iterator to vary the vertex positions of the focal
reducer to minimise the aberrations. The letters d and f , in the column below each,
were instructions to move the second lens surface with the rst to keep the lens
thickness constant.
Figure 14.1 shows the layout of the resulting spectrograph. In the nal ver-
sion the ray bundles have diameters of 23.4 mm. This compromise came from
the best choice of available tool radii when the lenses for the focal reducer were
ordered.
Figure 14.2a shows the spot diagrams of the nal collimated output beams at
5000.00

A and 5000.15

A, showing the eventual resolution at the zero-coma wave-
length. Figure 14.2b shows a similar pair at = 5250

A. The 1000 plotted points
are of random rays from the same launch point at z = 2.56; x = 0 and are of U
f
148 Optical design
0.002
(a)
0.000 V
f
0.002
0.0000 0.0002
U
f
0.002
(b)
0.000
0.2362 0.2364
U
f
0.2366 0.2368
V
f
0.002
Figure 14.2 Spot diagrams of images in this spectrograph at (a) 5000

A and
5000.15

A at the centre of the eld and (b) 5250

A and 5250.15

A at the long-
wavelength margin. The astigmatism increases at the margin but the spectral res-
olution is not signicantly compromised. Notice that the units are radians, the
inclination to the optic axis of the outgoing rays, and that the vertical and horizon-
tal scales are different. These gures must be multiplied by the focal length of the
CCD camera lens to give the actual sizes of the spots on the CCD surface. With a
50 mm camera lens for example, the widths would be 7 m and the length at the
eld centre would be 150 m.
and V
f
, the angles
4
(in radians) which they make with the optic axis. These must
be multiplied by 50 to show the spot positions on the CCD. An angle of 0.0002
radian corresponds to 10 m at the CCD surface.
4
The sines of the angles in fact, but clearly the difference is negligible.
14.6 Requirements of a ray-tracing program 149
0.04
0.02
0.0000
U
f
0.0005
V
f
0.00
Figure 14.3 This is the spot diagram of rays from the top half of a straight entry
slit 20 mm long. The horizontal scale is magnied to 100 times the vertical. Line
curvature is now dominant and it may be necessary, before co-adding rows of
pixels, to shift (in the computer) the upper and lower rows sideways in order to
straighten the image and give the best spectral resolution. Care is needed here
because the line curvature varies with wavelength. There is room in this data-
reduction process for considerable sophisticated software writing, since each row
of pixels is, in effect, a separate spectrum with a slightly different dispersion.
Figure 14.3 shows the spot diagramfor = 5000

Awhen a nite (10 mm) entry-
slit length is included. Nine rays have been launched in two annular bundles from
x = 0 mm and from x = 10 mm, both directed to arrive at the same places on the
grating. A thousand rays are then launched randomly from all points with x = 0,
z = 2.56 and y between 0 and 10, to arrive at the same points on the grating.
There is little change in the magnitude of the aberrations but the line curvature
described in Eq. (8.8) is apparent. As in the other spot diagrams, notice the different
vertical and horizontal scales, which here show that with a 50 mm focal length
camera lens the line images are 3 mm or 120 pixels long but that there is still less
than 1 pixel width of curvature.
The nal layout of the instrument was the result of many hours of investigation,
and the position in particular of the M2 mirror is the position which determines the
at eld at the CCD. It is some distance from the position demanded by Eq. (4.12)
but takes account of the eld curvature added by the focal reducer. In the end the
resolving power of 30 000 achieved was better than the CCD resolution, and a
CCD with a 10 m pixel width would have been useable.
With a sufciently bright source, the CCD camera could have been replaced by
a traditional 35 mm camera with ne grain lm.
15
Mechanical design and construction
Mechanical motions, automatic adjustments and electro-mechanical monitoring
and position sensors are so diverse, and the range of transducers and electric motors
for remote focusing, wavelength shifting and measuring so vast that there is no
point here in listing them or criticising their relative merits. So much depends on
the specication of the instrument and the capability of the available workshops that
it must be left to the builder to decide how the instrument shall be controlled. The
simplest possible mechanical construction is shown here handles to turn to adjust
grating turntables, capstan-bar mirror adjustments and so on, and the improvement
of these by other means is clearly up to the designer and the size of his purse. This
chapter therefore is conned to those matters which are peculiar to spectrograph
construction.
15.1 The optical layout
The positions of the optical components will have been decided by the nal version
of the design coming from the ray-tracing program. The position tolerance for the
mirrors is 0.5 mm on both x- and z-axes, the small errors being taken up when
focusing.
It remains to dress them in appropriate mountings and assemble the mountings
either on to an optical table or into a tube or a space-frame.
The art of constructing a scientic instrument is different from that of engineer-
ing mechanisms. The loadings are generally lighter and the precision as high as
that of a machine tool or an internal combustion engine, and different criteria apply.
Kinematic mounting, for example, may involve high point-loading with the conse-
quent risk of fretting corrosion but yields a precision of fractions of a micrometre
where necessary. It is suitable for optical instruments where there is no sliding
motion between parts.
150
15.1 The optical layout 151
15.1.1 Kinematic mounting
Briey, the principle is that to x the relative positions of two solid objects, six
restraints are required, three of translation and three of rotation. The common
example is that of a surveyors cube-corner reector which is to stand on a xed
table. The cube-corner mounting has three steel balls press-tted into holes in its
base to act as feet and the table has three radial grooves, each of which xes a ball
by making contact with it in two places. The device can be removed and replaced
with a precision limited only by any particles of dust which may have landed in
a groove at a point of contact; in other words, to a fraction of a wavelength of
light. Occasionally a hole-slot-and-plane arrangement is used instead to give the
six constraints, although the trihedral hole required for this is more difcult to make.
The penalty for this precision is the high point-loading at the contacts between
the surfaces, and the method fails if there is a weight or contact force sufcient to
distort the material signicantly. It also fails when heavily loaded sliding contacts
are involved. Dust in the air consists largely of micrometre-sized crystals of SiO
2
and these embed themselves in the softer of two surfaces and abrade the harder.
This is a classical example of the apparent anomaly that if two surfaces of unequal
hardness rub together it is the harder one which wears away. If the two surfaces are
of the same material, the frictional heat generated at the point of contact may be
enough to weld them together. It that case more robust methods of support are used
and a ball-race, for example, will provide adequate precision with rolling friction
for a grating spectrograph.
There is little in the literature on kinematic mounting and Braddicks classic
work was published over 50 years ago.
1
The principles however are simple and still
leave much room for inventive design.
For any optical instrument the order of design is important. In a

CzernyTurner
spectrograph for example the largest item is likely to be the M2 mirror, and the
mounting for this will determine the height of the optic axis above the baseplate,
or the internal diameter of the tube, or the datum surface if a space-frame is to be
used. From this the designs of the other mirror mounts, the slit and detector holders
and the grating turntable will follow.
15.1.2 The main frame
There are broadlyfour ways of constructinga spectrographas illustratedinFig. 15.1.
(1) A light alloy casting, forming an optical bench or table with raised pads at appropriate
places to be milled at to hold the optical components.
1
H. J. J. Braddick, The Physics of Experimental Method (London: Chapman & Hall, 1954).
152 Mechanical design and construction
(a)
(b)
(c)
Figure 15.1 Various methods of constructionfor spectrographs. For manypurposes
the plain optical table (a) is adequate but may prove too heavy for instruments much
greater than 1 m focal length. The cover, which must of course be removable, may
be of sheet metal, wood or GRP, depending on the skills available. The tube (b),
which is its own shroud, is appropriate to large instruments of 3 m focal length
and is also very convenient for small, portable spectrographs of 30 cm focal
length. The Serrurier truss (c) allows maintenance of the optical alignment when
the direction of g is changing as on a large astronomical telescope.
(2) An optical table of sheet aluminiumwith holes bored and threaded at appropriate places
to hold the optical components. It can be a box girder and, if very large and heavy, may
be self-standing, resting its Airy points on two pillars with the optical table vertical for
greater stiffness. The optical components are then cantilevered out sideways on their
holders. (This, incidentally, takes up much less space in a crowded laboratory.)
The covering shroud may be of metal, wood or GRPdepending on the skills available.
(3) An open-frame lattice to which the optical component-holders are bolted, with a
wooden,
2
GRP or sheet-aluminium shroud. If the instrument is to be attached to an
astronomical telescope the Serrurier truss type of space-frame (Fig. 15.1c) is worth
considering.
(4) A tube, drawn if of small (-20 cm) diameter, but probably rolled and welded if larger,
with shrouded tables, plates or boxes at the ends to hold the optical components.
The choice depends on the size, the application and the quantity to be constructed.
Access is always important for focus adjustment and cleaning, and large, light-tight,
hinged doors at appropriate places should be part of the design, as are access holes
for screwdrivers, spanners and capstan bars for ne adjustment while the instrument
is operating.
2
The author once used this type of construction with the optical shroud doubling as the packing case into which
everything was packed for transport to a remote observing site.
15.1 The optical layout 153
The optical table is probably the simplest method of construction for spectro-
graphs up to 1 m focal length. The mounting positions on the table are set by the
optical design and from this the screw-hole positions follow. Threaded holes are to
be preferred to through-bolting for ease of access and assembly.
Space-frame construction may be considered for large instruments, with 23 m
focal length mirrors. This is because optical tables with adequate stiffness become
unacceptably heavy and unwieldy and one must consider the feasibility of moving
them from room to room and from building to building. Space-frames are more
transportable although equally bulky, and the dimensions of corridors and goods-
lifts
3
which may be needed to move them about must be measured. Spectrographs
with focal lengths much greater than 1 m are not generally transportable. If they
are to have carrying handles, then the limit is about 0.5 m.
When designing a space-frame, the criterion is, again, stiffness. Because of
interference with the optical path, internal cross-bracing is scarcely possible but
Warren-girder construction of welded 25 mm square-section steel tubing with solid
12.5 mm light alloy end-plates will answer the purpose. Mirror mounts similarly
will act as cross-braces if designed into the frame. There is much to be said for a
vertical optical table if a

CzernyTurner mounting is specied. The space-frame is
comparatively stiffer and both sides of the optical table are accessible in the event
that electronic equipment is situated there. If this arrangement is chosen, the input
side should be below the detector, so that the grating tilt, facing the input side, is
downwards, reducing the amount of dust falling on the grating surface.
The principle of the Serrurier truss is that the frame is allowed to sag but in such a
waythat the ends remainparallel toeachother andthe optic axis is not compromised.
Such an arrangement is worth considering if there is a weight constraint and the
spectrograph is to be mounted on an astronomical telescope where the direction of
gravity is varying with respect to the instrument during the observation. Elaborate
civil-engineering design is not called for and the rule is: if in doubt make it stiffer.
A large laboratory- or Coud e-focus instrument, by contrast, is not expected to
move and once it is set in position the optical alignment can be completed in the
knowledge that it will remain aligned until the desired measurements are made.
15.1.3 Mirror and lens holders
The construction of these is governed by the need to avoid any sort of pressure or
loadingwhichmight cause distortionof the optical surface, especiallyas the result of
a change of temperature. Small mirrors, of diameter less than 50 mmcan be glued
to a backing plate with an epoxy resin, provided that a single small area of contact
3
Freight-elevators if you are an American reader!
154 Mechanical design and construction
Figure 15.2 The mirror holder. For mirrors of intermediate size, in the region of
150400 mm diameter, a simple semi-kinematic mounting is adequate. The rim
rests on two of three cylindrical bearings covered by thin polythene sheaths, and is
held by retainers of berylliumcopper screwed to the ends of the bearings, exerting
minimum pressure on the mirror.
Figure 15.3 The mirror cell. This is suitable for smaller mirrors of a fewcentimetres
diameter. The mirror rests on a polythene pad and is retained by a counter-cell held
in place by screws (not threaded as lens counter-cells sometimes are).
(5 mmdiameter) is sufcient to carry the weight. Multiple attachment points must
be avoided because of the bending which may result from differential expansion.
Mirrors and lenses larger than this require either a cell, or quasi-kinematic support.
Figure 15.2 illustrates a suitable mounting for mirrors of 150400 mm diameter.
Mirrors smaller than this may be contained in a mirror cell such as that in Fig. 15.3,
15.1 The optical layout 155
screw for
altitude alignment
transverse hole for
capstan-bar to make
azimuthal alignment
Figure 15.4 The mirror mount. The principle is that the mirror can be rotated in
azimuth and altitude almost exactly about its vertex and, once adjusted satisfactor-
ily, is xed in place by a single bolt. The same mounting is suitable for gratings
when no rotation is needed to change the wavelength range.
where they are backed by a thin polythene lm and held in place by a counter-cell
xed by screws which are not over-tightened.
The thin polythene lm is used to avoid high local pressure, which may damage
brittle materials like glass. The accuracy with which a mirror is positioned is not
high by engineering standards, and an error in centration
4
of 0.1 mm is generally
acceptable when focal lengths are of the order of 1 m.
15.1.4 Mirror adjustments
The mirrors require adjustment and an altazimuth set of axes is convenient and
practicable. In Fig. 15.4, the mirror holder may be attached to an L-bracket by
means of a horizontal berylliumcopper strip-spring clamped in place with a keeper,
and the L-bracket is rotated about a vertical axis which passes through the mirror
vertex. To ensure this is a simple matter of design and the rotation is secured by a
journal bearing a simple drilled hole in the baseplate with a brass dowel passing
through a similar hole in the L-bracket. This arrangement works well for mirrors
4
Centration here means the accuracy with which the optic axis of a lens or the vertex of a paraboloidal mirror
coincides with the centre of the circular rim.
156 Mechanical design and construction
up to 400 mm diameter. The L-bracket is clamped in place with a bolt tting a
threaded hole in the baseplate through an arc in the L-bracket base. A washer at the
bolt head is vital here to retain the accurate positioning of the azimuth as the bolt
is tightened.
The optical alignments are then carried out as follows: the tilt about the horizontal
axis is governed by a brass screw with a 1 mm pitch bearing against the rear of the
mirror cell and held in contact either by the force of gravity or if necessary by a
retaining spring. The adjustment about the vertical axis is by a silver-steel capstan
bar inserted into a hole at the elbow of the L-bracket. The capstan bar should be
long enough to allow the positioning of the mirror to 3

of arc a comfortable
tolerance in practice. Multiple attachment of the L-bracket to the baseplate is not
necessary, provided of course that the single bolt is big enough.
If the instrument is subject to shaking or vibration a more solid xing is needed.
In this case three pairs of push-pull screws will hold plates against plates essentially
as a solid, and the holding of the mirror in its cell becomes a more difcult task when
allowance is made for differential thermal expansion and the brittleness of glass and
fused quartz. An oversized cell may be used with a gap around the mirror edge
approximately three times the thickness of the berylliumcopper leaf-springs which
are inserted to keep the mirror centration exact. A front-plate, also of beryllium
copper, with a 3 mmoverlap at the edge, will do no more than obscure the turn-down
at the edge of the mirror. These extra precautions are only needed if a very large
temperature range between midday and midnight in the desert for example is
expected.
For small mirrors, of less than 50 mm diameter, the vertical axis is sometimes
carried on a berylliumcopper strip-spring, but this foregoes the opportunity to have
the axis of rotation through the vertex, and requires a holding spring to locate the
azimuth adjustment.
15.1.5 Grating holders
Much depends on the portability of the nished spectrograph. In the case of a large,
immobile instrument with a horizontal optical table, the grating may be held free as
in Fig. 15.5, resting on a lower pad of stiff plastic such as polythene, and constrained
by a light alloy bar which rests on an upper polythene pad and in turn is held in
place by capstan-head screws. The rear face of the grating may be insulated from
its mount by a thin sheet of polythene.
If a cell is required it may be fabricated, or milled from a single block of light
alloy, with relieving holes drilled at the corners as in Fig. 15.6. Again space should
be left for insulating polythene sheets and if necessary PTFE clamping screws
will hold it immobile. Gratings are generally formed on quartz blanks and the low
15.1 The optical layout 157
Figure 15.5 The grating cell. The grating can stand on a shelf at the appropri-
ate height above the table, and is retained in place by a bar exerting minimum
downward pressure through two capstan screws.
cover plate
Figure 15.6 An alternative grating cell, in which the grating is fully enclosed
and retained by a face-plate of light alloy with an appropriate mask-shape as in
Fig. 8.5.
158 Mechanical design and construction
thermal expansion of these must be borne in mind. The grating may be held in
its cell by a face-plate which restrains the corners, exible enough that when the
screws are fastened it places only the necessary pressure on the grating, sufcient
to maintain it against temperature changes and to bear its weight when it is possibly
face-down at the Cassegrain focus of a telescope.
Optical adjustments are probably needed only in the initial alignment process,
in which case no elaborate rotation about the grating normal is called for. Rotating
the grating in its own plane to align the rulings can be done with copper or bronze
shims under the lower edge.
5
A metal shim, a sliver 25 m thick, under one end of
a 150 mm wide grating will tilt the rulings by 0.5

of arc.
15.1.6 The grating turntable
In a high-resolution spectrograph it would be unusual if the whole of the spectral
range were available to the detector without rotating the grating. If the adjustment
is infrequent the grating cell can be mounted on a circular brass pad which rotates
about a simple journal bearing in the form of a spindle lying in a hole in the optical
table. It is then held in position by a clamping screw, much like the provision of the
mirror holders.
If adjacent regions of the spectrum are to be photographed in rapid succession
a more elaborate arrangement is needed and a proper turntable must be used. The
standard method in this instance is to use two large-diameter, light-duty angular
contact bearings inopposition, heldinpermanent contact bymeans of a compression
spring as in Fig. 15.2. A simple worm-gear can be used for rotation, but there is
an advantage in using a sine-bar, as in Fig. 15.7. The valuable property of a sine-
bar motion is that the wavelength presented at a particular pixel varies linearly
with the angle turned by the screw. This in turn leads to convenient calibration
and re-setting. In a scanning monochromator, such a drive, usually motorised, is
regarded as virtually essential. Of various possibilities in practice, the double-
hinged version is to be preferred. The moving disc bearing against a ball engenders
wear and fretting corrosion with consequent loss of calibration and smoothness of
scan.
In the asymmetrical

CzernyTurner mounting the grating face must make an
angle withthe sine-bar towhichit is xedsince the gratingequation, Eq. (8.2),
which governs the relation between captive-nut position and grating tilt angle, t ,
can be rewritten as
n,a = 2 cos( +) sin(t + ). (15.1)
5
With good modern workshop practice this will almost certainly not be necessary.
15.1 The optical layout 159
Figure 15.7 The grating turntable. In most cases it is not possible to record the
entire spectrum in one exposure, and different regions are brought into view by
rotating the grating. It is important to be able to do so accurately and for this a grat-
ing turntable is required. A standard form is shown here with manual adjustment
which can obviously be replaced by a stepper-motor under computer control. The
sine-bar link ensures that the screw moves the central wavelength of the spectrum
linearly with rotation angle.
In the original Ebert mounting and were the same and + was called the
ebert angle.
Whichever method is used, there is backlash in the drive screw to be considered.
Although there exist sophisticated anti-backlash split-nuts and similar devices, a
light pressure applied to the turntable by a spring is adequate for most purposes.
15.1.7 Entry and exit slits
Many constructors have spent time, ingenuity and energy on manufacturing double-
openingslits, where bothjaws move apart symmetrically. This is probablya mistake.
Traditionally only one jaw moved and the other, xed jaw was the ducial mark
for measuring deviation angle in a prism spectroscope. An emission line spectrum
could then be observed with slits as wide as convenient to allow plenty of light to
enter the eye, and this in turn made for easy location of the xed edge with a pair
of cross-hairs at 45

to the line of the slit edge.

There is still no particular virtue in the double-opening slit. If wavelengths are to
be measured, the xed edge is still the ducial mark even on a photographic plate,
160 Mechanical design and construction
land thickness
incident light
Figure 15.8 Asection through the slit jaws, illustrating the lands of the jaws. Light
should always be incident on the plane faces of the jaws.
and unless there are close doublets to be resolved, there is no virtue in a narrow slit
width when observing emission spectra.
Slits should always be mounted with the light incident on the plane, unchamfered
face as in Fig. 15.8. The purpose of the chamfer is to permit a slit width as small as
10 m with the thickness still less than the width. In other words, slit edges must
be knife-sharp.
Much ingenuity has been expended on slit-adjusting mechanisms. The simplest
appears to follow the practice of R. V. Joness parallel motion, a device much
used in instrumental practice. It is illustrated in Fig. 15.9. A ball-ended micrometer
pushes its ball against an anvil such that the point of contact is midway between the
moving and xed bars. The width of the slit is read directly off the micrometer which
can have a large diameter drum for easy reading in dim light. The arrangement is
such that the slits cannot be damaged by over-tightening of the micrometer and a
preventer screw holds the moving jaw at the point of closure to avoid damage to
the slit lands.
Slit jaws themselves should be replaceable. They are normally made of hardened
and ground tool-steel, or gauge-plate as it is sometimes called, and can be made
straight enough with a grinder that has not previously been misused in the workshop.
If ultimate precision is needed, then three similar edges can be ground and lapped
alternately in the same fashion employed in making optical ats.
It is probably convenient to shot-blast or grit-blast the plane faces of the input-slit
jaws before the nal grinding of the edges, so that they act as a focusing screen for
the fore-optics.
The slit assembly will normally be attached to a mounting, which itself may
stand on the optical table, but which may more conveniently be tted to the edge
15.1 The optical layout 161
(a)
(b) (c)
Figure 15.9 The parallel motion for opening slit jaws. One jaw is attached to the
upper, xed bar and the other to the lower L-shaped bar which moves when pushed
by the ball-ended micrometer. The anvil of the micrometer must always be midway
between the two bars.
of the table. Some attention must then be given to any fore-optics which may be
attached to this mounting and sufcient strength must be given to enable it to bear
the weight and moment. The classical approach was to mount an optical bench
along the input optical axis, but this is sometimes inconvenient and such things as
external bafes, colour lters and order-sorters might as well be made with threaded
ends to be attached directly to the entry-slit mounting, much as they are on camera
lenses.
15.1.8 The CCD camera attachment
The LN
2
-cooled CCD camera may be a common-user component which must be
removed occasionally from one instrument to another, or from a single instrument
for focus adjustment and cleaning. In any case a quick-release mechanism is de-
sirable. One possibility is the hook-on-shelf attachment shown in Fig. 15.10 which
allows adequate precision, which avoids the need for extensive screw-ttings and
which interfaces easily to different spectrographs with a simple common interface.
If the attachment is to an astronomical telescope, one substantial additional bolt will
secure the camera against variations in the direction of the gravitational g-vector.
15.1.9 Access
It is likely from time to time that the alignment and focus will be checked and it is
convenient if this can be done while the instrument is in a well-lit laboratory. Access
ports allowing a hand or a screwdriver shaft to reach in and turn an adjusting knob
162 Mechanical design and construction
Figure 15.10 A quick-release attachment for a CCD camera. This was designed
to attach a Wright Instruments LN
2
-cooled camera to two different spectrographs.
The essence, which is easily adaptable for other types of camera, is a pair of
interlocking hooked shelves which bear the weight and allow the camera to be slid
horizontally against a stop to achieve correct alignment. If necessary the binding
can be made rmby a screwat the bottom. The camera can be released and removed
to another spectrograph with a similar tting in a matter of seconds.
should be incorporated at suitable places, and in particular a means of adjusting the
grating table from behind the grating will probably be required. Access holes must
be planned in advance before the shroud is made.
The shroud itself as well as the base should be thermally insulated to give the
greatest possible temperature relaxation time. This is not so important in the normal
thermostatted laboratory, but the interior of the dome of a 2-metre telescope may
fall in temperature by 1015

C during a nights observation, with a corresponding

change in length of a 1-metre spectrograph optical table of a millimetre or so unless
the insulation is adequate.
15.2 Optical materials
Mirrors and grating blanks are generally made of fused silica, ground and polished
to an accuracy of ,8 of green light. This is to ensure that a reected plane wave
from a grating for example remains at to ,4. Their advertised focal lengths
are not generally accurate to better than 0.3% and should be measured with a
Foucault knife-edge test. The measured focal lengths may then be used to revise
computed values in the ray-tracing program.
Mirrors are made more cheaply from low-expansion borosilicate glasses and
more expensively fromvitreous ceramics with a zero thermal expansion coefcient
15.4 Reectors 163
over a limited temperature range. Such materials are also useful for interferometer
construction.
Transparent materials include various crown and int glasses for the visible
and near ultra-violet, fused silica for the ultra-violet, calcium uoride and lithium
uoride for the extreme ultra-violet as far as 1200

A. Belowthis wavelength nothing
is transparent until the soft X-ray region, and gas retention for example must rely
on thin lms of collodion or some similar plastic material which can be formed
into lms 1 m thick. Such lms tend to be porous and a double-layer lm is to
be recommended. They will withstand a pressure difference of 5 mm Hg over an
aperture of 3 mm diameter. Since gases are as opaque as other matter, this is an
adequate pressure for absorption spectrometry in the VUV.
15.3 Transparent optical materials
Glasses are available with different optical and physical properties and can be
chosen by refractive index, which is between 1.44 and 1.95, by dispersion or Abb e
v-value, by weathering properties, which makes them sometimes unsuitable for
the outer component of a lens, and not least by cost, with extreme examples of
high-index lanthanum crowns being beyond the purse of most scientists. Glasses
are not generally transparent below 3000

A.
For most general purposes Schott BK7, a borosilicate crown appears to be the
glass of choice among makers of singlet lenses. The Schott catalogue of optical
glass is the bible on the subject.
In the infra-red a whole gamut of transparent materials exists with refractive
indices upto2.5or higher. These include polycrystalline pressings andsophisticated
plastics materials. Among these are sulphones which have Abb e v-values less than
20, implying enormous dispersion and possible use in prism spectrographs. The
Eastman Kodak companys products will ll most transparency requirements in the
infra-red. For reection, a gold coating has the highest reection coefcient.
In general a study of the properties of transparent materials is not necessary
for the construction of a spectrograph as nearly always a commercial product
especially a photographic lens will be found for most purposes. The question
chiey of interest to the spectrograph designer is the refractive index, dispersion,
transparency and cost.
15.4 Reectors
The choice between a spherical mirror and a paraboloid depends to some extent
on the optical requirements, but at focal ratios above F,10 and focal lengths below
1000 mm the difference is not signicant. The surface error of a sphere, that is its
164 Mechanical design and construction
departure from a paraboloid, is Lx = y
4
,64 f
3
, where y is the semi-aperture and
f the focal length. Alternatively, the maximum tolerable aperture, D (millimetres),
where parabolising makes no difference, is given by D = 0.064F
3
where F is the
focal ratio.
Mirrors in optical instruments generally have a diameter/thickness ratio of about
6:1 to give the necessary elastic stability. Diameters up to 600 mm are solid but
when weight becomes a problem, then a honeycomb structure may be indicated.
Interferometer plates, especially for the visible or near UV may be proportionately
thicker, with a diameter/thickness ratio of 4:1. They will invariably be made of
fused silica.
15.4.1 Cleaning optical materials
Mirrors will probably arrive with the reection coating in place and this will have
been overcoated with silicon monoxide or some similar protective layer. The surface
is fairly durable and can be cleaned, but prevention being better than cure, they
should be protected when not in use with covers to exclude dust, insects and oily
vapours which may condense.
The surface of an aluminised mirror will be hard enough that, in the words of
one eminent astronomical telescope mirror-maker, it can be washed with anything
you would use to wash your face. Nevertheless, most people will be more cautious
and will clean their optical sufaces with de-ionised water,
6
analytical reagent grade
iso-propyl alcohol and a soft cotton-wool or tissue swab which has been removed
from a just-opened packet. The point is not trivial: aerosol particles of silica a few
wavelengths in diameter, which are universal at sea level, falling on to a clean
polished surface, are wonderfully abrasive. Cleaning cloths should always be kept
folded, should be washed frequently and dried in a closed desiccator, and only the
inside of the folded cloth should be used. The blower brushes commonly used with
camera lenses may be of inadequate size for 200 mmdiameter mirrors. Compressed
air may not be free of oil droplets.
Uncoated lens surfaces may be cleaned with the aforementioned recently un-
folded cloth, and if there is a natural bloom to be removed, jewellers rouge can be
applied before rubbing gently. Surfaces with anti-reection coatings should not be
treated thus and cleaning is conned to the usual solvents softly applied.
15.4.2 Diffraction gratings
It goes without saying that the grating surface must not be touched. However, grating
surfaces do become contaminated eventually and they can be cleaned. Although
6
This includes exhaled breath, a perfectly adequate cleansing uid.
15.5 Metals for construction 165
no solid may make contact with the surface, suitable liquids will remove dust
particles and lipid condensation. Analytical quality chemical solvents such as iso-
propyl alcohol and diethyl ether, used in well-ventilated, warm, dry surroundings,
poured gently over a grating held at 45

to the horizontal in a photographers

developer tray will make a remarkable restoration to a contaminated grating: but
solvent evaporation, particularly of ether, will cool the surface and lower the local
dewpoint so that water vapour may condense on to the grating and leave a faint
stain.
7
Therefore nish the cleaning with iso-propyl alcohol.
15.5 Metals for construction
The days when scientic instruments were made of brass are unhappily long past.
Light alloys of one sort or another are the materials of choice both for laboratory
and eld instruments, with various sophisticated lightweight honeycomb materials
for space-borne instrumentation, with metal inserts at strategic places to support
the items which require rm anchorage.
Thanks to computer-controlled machine tools, fabricated light alloy construction
is not the time-consuming bugbear it once was and parts once tted and bolted
together are now machined integrally. Nevertheless, light alloy castings for the
main optical table are worth considering, especially for instance if several examples
are to be made for a teaching laboratory; and there is then a considerable saving of
workshop time because of the comparatively small number of machining operations
to be done.
8
Set against this is the need to mature the casting, which may be liable
to gradual creep and distortion. Traditionally, iron castings were laid on a at
roof to weather for a year, but this is probably excessive with modern light alloys
and casting techniques. The largest casting that can be accepted by the available
workshop machine-tools is also a factor, especially for spectrographs of more than
a metre focal length. Turning is an unlikely process for large castings but the milling
of several separate surfaces to the same plane requires an adequately large milling
machine.
Bolted sheet aluminiumalloy is a convenient method of construction for small to
medium-sized instruments. Welding is not recommended where strength is essential
and epoxy-gluing may be too permanent when possible later modications are
required. The weight advantage of magnesium and its alloys is completely offset
by its instability and brittleness and magnesiumis generally worth considering only
if the weight-saving requirements are extreme. Brass and gunmetal are expensive
but useful for journal bearings.
7
The ideal environment for cleaning a grating is a dust-free room in the middle of the Sahara Desert.
8
In the vicinity of rust-belt cities, search the scrapyards for superannuated lathe-beds and milling-machine tables,
which make superbly stable optical tables: but be sure your workshop can handle the necessary modications.
166 Mechanical design and construction
The bending tolerance for the optical table is 0.5 arcminute, rather than the
0.01 arcsecond required in an uncompensated interferometer. As a guide, adequate
stiffness for a 1-metre spectrograph is achieved by using two parallel light alloy
plates 12.5 mm thick, 1.2 m0.4 m in area, connected by two parallel U-section
girders 100 mm60 mm of 10 mm thickness. The weight can be relieved by re-
moving those sections of the lower plate which do not affect the torsional stiffness.
The whole structure, in accordance with kinematic principles, should be sup-
ported on three feet, placed at the Airy points which, for an unencumbered beam of
length L are at a distance L(

3 1),2

3 = 0.211L from the ends. This balances

out sag and hump and ensures that the ends are horizontal. Bear in mind that the
weight of the optical table is considerably greater than the combined weight of the
optical components which will stand on it.
The qualities of alloys, especially aluminium alloys, vary according to compo-
sition, and some are more amenable to turning and milling than others. Materials
to avoid are (1) pure aluminium, (2) stainless steel except in nished form such as
nuts and bolts, (3) mild steel, which corrodes easily and (4) pure copper, which is
ductile and difcult to shape with machine tools.
Berylliumcopper
9
in thin sheet form is superlative for making leaf-springs, but
after forming it must be tempered (precipitation hardened in fact) by boiling it
in benzyl benzoate under reux for 90 minutes. Thicknesses between 0.5 mm and
1.0 mm are suitable for most laboratory instrument spring-hinges. All the forming,
cutting and drilling must be complete before the tempering.
Plastics such as polythene and PTFE (poly-tetra-uoro-ethylene) have appropri-
ate properties for insulation, padding etc. PTFE is machinable, durable and soft,
but has a large thermal expansion coefcient. It is also available as screwed-thread,
nuts, bolts and washers. Washers should always be used on both sides with nuts and
bolts and with bolts tted into threaded holes. Bolts in threaded holes are always to
be preferred to nut-and-bolt fastening unless there is a risk of overtightening and
consequent stripping of the thread.
15.5.1 Journal bearings
Bearings for light loads, whether journal or sliding, may be simple contact sliding
faces. They should always be of different metals: light alloy and brass for example
or silver-steel and gunmetal. Sliding faces of light alloy, in particular, will weld
together sooner rather than later and must be avoided. Any sort of grease or oil is
anathema in a spectrograph and all lubrication should be dry, either by graphite or
9
Copper alloyed with 2% beryllium, much to be preferred to phosphor-bronze which can only be hardened by
cold working.
15.6 Other materials 167
by a colloidal suspension of graphite such as that sold under the name of Aquadag.
For heavy loads or for precision rotation, ball-races should be used
10
and circulating
ball-race bearings will support translational motion. They are not lubricated and
must be protected from dust.
15.6 Other materials
An American friend and colleague, a virtuoso at interferometer construction, now
sadly deceased, occasionally made his prototype Michelson Fourier spectrometers
out of wood. This is not recommended for ordinary mortals. He also believed in
rapid scanning to give an interferogram every few milliseconds, co-added until the
signal/noise ratio was satisfactory. This, by contrast, has its merits.
Wood nevertheless is an acceptable material for the shroud the light-tight
cover of a spectrograph. It is light in weight for easy removal, is easily shaped, with
cut-outs for access and for sliding over necessary metal protrusions. For eld instru-
ments, marine ply has excellent strength/weight ratio and can be made weatherproof
and durable. The alternatives, for those skilled in the art, are either GRP (glass
reinforced plastic), the material from which boats are constructed, or frameworks
clad with sheet-metal.
15.6.1 Protective coatings and paint
Zinc chromate is the accepted protective coating for light alloys but is probably
unnecessary unless the nished instrument is to stand in the open air. Even then it
must be overcoated with a suitable weatherproof paint. For laboratory work light
alloys can be anodised and dyed, and black-anodising of a previously shot-blasted
or abraded surface is a convenient and durable way of treating the interior. Other-
wise matt black paints such as optical black can be applied with an appropriate
undercoating according to the manufacturers instructions.
11
Beware of the peeling
of paint fromunprepared metal surfaces. Exterior ttings and controls are tradition-
ally painted white for easy visibility in low-light conditions and it is worth making
different hand-controls of different shapes so that they can be recognised in the
dark by feel. It sometimes happens that it is necessary to work in complete dark-
ness and pilot-lights, LEDs and such should have covers which can be closed. The
extreme sensitivity of CCD spectrographs to faint light must not be disregarded.
10
Except in interference spectrometers where kinematic hinges should be used.
11
There is no such thing as black paint. There are various shades of dark grey, or sometimes dark blue or dark
green, and a low albedo is all one can hope for. Some black paints are anything but black in the infra-red and
the shape of a bafe is more important than its colour.
16
Calibration
16.1 Sensitivity calibration
There are so many variables in the optical path of a spectrograph that sensitivity
calibration is necessarily totally empirical. A black body at a known temperature is
the only feasible method of absolute calibration, and even then one must be sure that
there has been a total suppression in the spectrograph of other orders and of scattered
light. Reliable absolute calibration is such an onerous task that one must question
the need for it in most circumstances. For almost all practical purposes a comparison
with a standard source is sufcient. In the visible, a reasonable approximation no
more can be obtained by making a spectrogram of a surface coated with freshly
deposited magnesiumoxide fromburning magnesiumribbon which will scatter the
light of the midday sun on a clear day. In the tropics and at moderate temperate
latitudes, the midday sun is an approximation to a black body at 6000 K. Ordinary
white card will not do as a scatterer, neither will a painted white matt surface
because both are likely to be uorescent under the UV solar radiation.
The resulting spectrum, after correction for the CCD pixel-to-pixel variations of
sensitivity, may not look much like a textbook black body spectrum and for good
reason. Firstly there are the basic variations with wavelength of the CCD overall
sensitivity. Added to these are the reection coefcients of the various mirrors and
above all the variations with wavelength and polarisation of the grating rulings and
the change of these with angle of diffraction. (Spectro-polarimetry with a reection
grating is a lost cause for this reason.) Nevertheless, since the radiation intensity
is calculable from Eq. (3.1), the spectral sensitivity of the spectrograph can be
estimated.
168
16.2 Wavelength calibration 169
16.2 Wavelength calibration
The most obvious and most useful way of making a wavelength calibration curve
or table is to select ten or twelve known wavelengths covering the spectral range
required, as evenly spaced as circumstances permit, and to tabulate them against
pixel number, then to draw or compute a cubic spline
1
to t them. In general, with a
grating spectrograph, the calibration curve of pixel number vs. wavelength will be
close to a straight line and a table of interpolated wavelengths vs. pixel number will
not deviate by more than 0.1

A from the true values. Single sharp emission lines
will usually occupy two adjacent pixels and if the line is symmetrical, its wavelength
can be interpolated using the ratio of the two pixel counts. Causes of line asymmetry
can be physical (e.g. pressure broadening) as well as optical shortcomings, chiey
the result of coma, in the spectrograph.
There are manycubic-spline-computingprograms invarious computer languages
to be found as public domain software, ready for loading, compiling and using.
The output from a cubic spline program should be available as a table of wave-
length against pixel number as well as a graph.
2
The spectrogram, binned down
to a one-dimensional table of intensity vs. pixel number by the camera software,
together with the spline program output, can then be loaded into a spreadsheet
program and thence plotted as a perfectly linear xy graph of intensity against
wavelength. Various arithmetic functions can be applied, such as subtraction or div-
ision of successive spectrograms, to measure such things as time variation, optical
lter proles and absorption coefcients.
16.2.1 Infra-red calibration
There is a relative poverty of suitable sharp emission lines for wavelength cali-
bration in the infra-red and it may be necessary to resort to the method of Edser
Butler fringes.
3
These are most suitably generated in a FabryPerot etalon with
metal coatings and quartz FabryPerot plates. These will give an adequate nesse
(F 1012) to allow wavenumber measurements to be made. Accuracy depends
on knowing precisely the gap width between the FabryPerot plates, and this is
measured if necessary by the (tedious) method of exact fractions with known cali-
bration wavelengths in the visible.
4
Bear in mind that the EdserButler fringes are
1
Cubic splines are what you get when you plank the ribs of a ship with single long strakes of wood. Drawing-pins
xed on a drawing board, with a length of ne piano wire threaded so as to touch each pin, give the same effect.
2
The graph is a useful weapon against misplaced entries in the data, which show up instantly in loops and whorls
in the plotted graph.
3
These are the fringes obtained when , in Eq. (4.1), is the independent variable in and I is plotted as a function
of in Eq. (4.2).
4
See for example, R. W. Ditchburn, Light (New York: Dover, 1991), Chapter 9.
170 Calibration
at equal intervals of wavenumber, so that it is important that the gap be an ade-
quate multiple of the longest wavelength to be recorded by the spectrograph under
calibration.
16.3 Small spectral shifts and radial-velocity measurement
A well-focused CCD grating spectrograph has a remarkable ability to detect small
shifts of wavelength, provided one is condent of its thermal stability. Thermal
insulation is important: one may alter the calibration just by standing too close,
allowing body-heat to warm one side of the optical bench, expanding it and so
distorting the optical path. One 25 m pixel subtends 9

of arc in a 1-metre
spectrograph. The ultimate resolution at 5000

A typically is 0.015

A, subtending
1

of arc, and a shift of illumination of 5%fromone pixel to its neighbour implies

a wavelength shift of 0.5

of arc, i.e. 0.008

A or 0.5 km s
1
in astronomical terms.
This is just enough to reveal the morning-to-evening Doppler shifts of solar
Fraunhofer lines caused by the diurnal rotation of the Earth. A laboratory lamp to
furnish a calibration wavelength is essential of course.
16.4 Absorption measurements
Absorption spectroscopy is traditionally the most difcult spectroscopic technique
5
and the precautions to be taken are severe. The shape of an absorption line is
modied by the instrumental prole of the spectrograph and an accurate line prole
may not be available. Astronomers in particular are content often with the equivalent
width, as this measures the amount of power removed from the continuum by the
absorption, which is the usual purpose of the investigation. Elaborate precautions
must still be taken to exclude scattered light, but provided this is done, the equivalent
width is not affected by the instrumental prole. Equivalent width is determined as
follows.
Assume the spectrum is normalised to the continuum level. If the spectrum in
Fig. 16.1 is 1 S() and the instrumental prole is I (),
_
b
a
S() d = S
and S is the total power removed from the continuum by the absorption feature.
By denition:
_

I () d = 1
5
Especially in XUV region where it is next to impossible.
16.4 Absorption measurements 171
S() I () C()
=
Figure 16.1 The principle of the equivalent width measurement. The true absorp-
tion line S(), when convoluted with the instrumental prole I (), produces a
shallower, broader absorption line in the recorded spectrum. Provided there is no
contribution to the recorded continuumby scattered light, the area of the absorption
line is the same as that in the original, and measures the total amount of continuum
radiation that has been absorbed. The equivalent width then is the width of an
idealised rectangular absorption feature of full depth with the same area.
and if the limits a and b are far removed from the absorption region we can replace
them as usual by .
The output from the spectrograph, which is the convolution of the two, is
_

S()I (

) d = C(

)
then
_

C(

) d

= S
because with

= y. d

= dy so that
_

S()I (

) d =
_

S()I (y) d dy.

which separates into
_

S() d
_

I (y) dy = S.
The equivalent width is thus dened independently of the spectrometer instru-
mental prole as L, where I ()L = S. In other words the area still measures
the amount of continuum power removed from the spectrum by an absorption line.
For those occasions where there is a need for the absorption line prole for
example where there are narrow deep cores to comparatively broad shallow lines,
showing perhaps that a cool, optically thick outer layer of stellar atmosphere or
an interstellar cloud has been traversed by the radiation there is no option but to
use a very high resolving power so that the slit width is narrow compared with the
half-width of the absorption line (which is not always well dened).
17
The alignment of a spectrograph
The method described here is suitable for the assembly and adjustment of a

Czerny
Turner type of spectrograph. Essentially similar methods are indicated for other
types and mountings.
Alignment is best carried out systematically after the instrument has been in-
stalled in its place of work or, if it is newly made, close to the workshop which
made it.
17.1 The optical alignment
The optical table has a number of holes drilled at the vertices of the mirrors and the
grating. Into these holes will normally t the vertical axle about which the angular
adjustments of the optical components are made. If a precision milling machine is
available, similar holes can be drilled along the optic path so that alignments can
be checked at any time.
An alignment tool (Fig. 17.1) should be made, preferably but not necessarily of
metal, which will stand on the optical table with a base extension which will t these
holes. It should carry a vertical disc with a 1 mmhole at the height of the optic axis.
1
Stage 1. A laser beam, adjusted to the height of the optic axis, is made to shine
through the centre of the entry slit so that it is parallel to the optical table surface.
Parallel here means that the light from the laser would be able to pass through the
hole in the alignment tool no matter where the latter is placed on the table.
Stage 2. The laser beamdirectionis nowadjusteduntil it shines towards the vertex
of the collimator (M1) mirror. The alignment tool, placed in the hole corresponding
to the vertical axle of the mirror mount, has the central hole at the mirror vertex
position. The tool can then be removed and the M1 mirror installed. The procedure
ensures that the beam is shining along the optic axis.
1
I am indebted to colleagues at Perkin-Elmer Ltd. for this method of optical alignment.
172
17.2 The focus 173
optic axis
base plate
Figure 17.1 The optical alignment tool. Its use depends on having holes for the
dowel drilled in the spectrograph baseplate along the optic axis, with x- and y-
coordinates accurate to 25 m. When a laser beam shines through the central
hole with maximum intensity, the alignment in a 1.5-metre spectrograph is good
to better than 1

of arc. The dowel holes should be close to (-25 mm) the mirror
and lens vertices and to the grating surface.
Stage 3. The alignment tool is next transferred to the grating table vertical axle
hole and the M1 collimator mirror is adjusted in both axes until the reected laser
beam passes through the hole in the alignment tool. The tool is then removed and
the grating with its mount is installed in its proper place.
Stage 4. The alignment tool is transferred to the vertical axle of the focusing (M2)
mirror and the grating adjusted in zero order so that the light passes through the
alignment tool. When this adjustment is satisfactory the grating should be rotated
until the rst-order reected beam passes through the hole in the alignment tool.
It is at this point that the grating may need shimming in case the rulings are not
parallel to the rotation axis. The adjustment is then checked by rotating the grating
so as to pass other orders through the alignment hole.
The M2 mirror and any auxiliary optics are installed, and the M2 mirror is
adjusted until the laser beam arrives satisfactorily at the pupil of the camera.
17.2 The focus
Stage 1. The grating is rotated on its table until zero order is reected back to the
input slit. A small (5 mm) 90

45

45

prism is useful here to shine light from a

spectral lamp such as a low-pressure mercury laboratory lamp through the lower
174 The alignment of a spectrograph
half of the input slit. The slit can be opened wide for this adjustment and the prism
attached by double-sided sticky-tape. When the grating rotation is correct an image
of the slit should appear in the upper half of the slit. A ne adjustment of the
grating tilt will allow a Foucault knife-edge test to be made using the slit edge and
its image. The eye applied to the slit and looking through it at the M1 mirror will see
the illumination appear uniformly on the mirror surface when the focus is correct:
otherwise it will appear to wipe across the mirror as the grating rotates. Adjustment
is by sliding the slit mechanism in-and-out along the optic axis.
Bear in mind that the focusing cannot be done by turning the M1 mirror to face
the entry slit, since the eld curvature of the tilted M1 mirror requires the entry slit
to be slightly closer to it than its axial focal length (which will have been previously
measured) would imply. For this focus adjustment it is essential that the light
be reected from the grating.
(This test is very precise and if the operator is unfamiliar with knife-edge testing
it can be rst practised directly on the M1 mirror without the encumbrance of the
input slit mechanism. If the M1 mirror is spherical rather than paraboloid it may
show radial zones at the proper focus condition. These correspond to the focus
being paraxial or marginal.)
Stage 2. When the M1 focus is satisfactory the grating should be turned to allow
a bright rst-order line, such as Hg 546.1 nm, coming through the entry slit to fall
on to the M2 mirror. A white card is needed here to ensure that the mirror area is
fully illuminated. Again the entry slit can be wide open, as only an edge is going
to be used. A sharp edge, such as a razor blade, may now be placed at the focus
position and the same Foucault test used to adjust the M2 focus position.
If the parts have been made by a good quality workshop all these focus adjust-
ments will be very ne if they are needed at all. The depth-of-focus of a spec-
trographic instrument working at F,14 is 0.25 mm and the tolerances on
focus are not difcult to achieve. A few millimetres defocus at the entry slit of
a 1 m M1 mirror is compensated without signicant aberration deterioration by a
complementary adjustment of the focal plane of the M2 mirror.
It is the initial alignment of the optic axis which is the sharp and tedious but
essential process.
Appendix 1
Optical aberrations
Aberration coefcients
These somewhat complicated expressions are worth including here since they can be set
up easily in a computer program and allow the spherical aberration and coma of any lens
to be evaluated in a few moments.
Aberration coefcients for lenses
The important quantities are
(a) the spherical aberration coefcient, B and
(b) the coma coefcient, F.
For a single thin lens with radii of curvature R
1
and R
2
these coefcients are computed
from the G-sums:
1
The G-sums are constructed from eight coefcients which depend only on the
refractive index of the glass:
G
1
= n
2
(n 1),2
G
2
= (2n +1)(n 1),2
G
3
= (3n +1)(n 1),2
G
4
= (n +2)(n 1),2n
G
5
= 2(n
2
1),n
G
6
= (3n +2)(n 1),2n
G
7
= (2n +1)(n 1),2n
G
8
= n(n 1),2
and various sums can be computed from these:
The spherical aberration G-sum is
G
sph
= G
1
c
3
G
2
c
2
c
1
+ G
3
c
2
s + G
4
cc
2
1
G
5
cc
1
s + G
6
cs
2
.
1
A. E. Conrady, Applied Optics and Optical Design (London: Oxford University Press, 1929), Vol. 1, p. 95.
175
176 Appendix 1
The coma G-sum is
G
coma
= G
5
cc
1
,4 G
7
cs G
8
c
2
.
Longitudinal spherical aberration = G
sph
:
2
y
2
Tangential coma
2
= 3G
coma
y
2
: tan
where
: tan = h, the height of the image above the axis
: is the image distance (= f , if s = 0)
s, = 1,u, is the reciprocal of the object distance
c
1
is the rst surface curvature (= 1,R
1
)
c. =
1
(n1)
1
f
= c
1
c
2
, is the total curvature of the lens
3
y is the height of incidence of the marginal ray on the lens.
A special case of particular importance is when the object distance = ; that is
when rays come from . Then s = 0; : = f , and equations simplify to
G
0.sph
= G
1
c
3
G
2
c
2
c
1
+ G
4
cc
2
1
.
G
0.coma
= G
5
cc
1
,4 G
8
c
2
.
Alternatively and conveniently when there is an innite conjugate on the object side we
can dene the aberration coefcients, B, F
0
and F directly from the refractive indices and
radii of the lens or mirror by
B =
_
n
n 1
_
2
1
f
2

_
2n +1
n 1
_
1
R
1
f
+
_
n +2
n
_
1
R
2
1
.
F
0
=
_
n
n 1
_
1
f

_
n +1
n
_
1
R
1
.
For the lens coefcients there are alternative expressions using the curvatures c
1
and c
2
instead of the radii. They are sometimes useful because they do not involve the focal
length:
B = c
2
1
[n
3
2n
2
+2],n c
1
c
2
[2n
2
2n 1] +c
2
2
n
2
.
F
0
= c
1
[n
2
n 1],n nc
2
.
These expressions simplify greatly if the lens is plano-convex.
Aberration formulae
Longitudinal spherical aberration
L
sph
=
1
2
y
2
B f.
The diameter of the circle of least confusion
L
colc
= L
blur
= y
3
B,4.
2
Sagittal coma is 1,3 of this.
3
Also referred to as Lc elsewhere in the book.
Optical aberrations 177
Meridional or tangential coma
The height of the coma patch is given by
L
coma
= 3y
2
(F
0
+ZB) tan .
Sagittal coma is 1,3 of this.
Fraunhofers theorem
The coma coefcient for a thin lens or mirror with its entry pupil at a distance Z from the
vertex is
F = (F
0
+ZB).
Worked example
4
For a plano-convex lens with a at rst surface (c
1
= 0) the second surface curvature is
negative (the centre of curvature is to the left of the surface) and the aberration coefcients
are
B = c
2
2
n
2
. F
0
= nc
2
.
and the lens is free from coma if the pupil is at Z = F
0
,B = 1,nc
2
, and since c
2
is
negative, so is Z and the pupil is to the left of the lens. In terms of focal length it is at
Z = [(n 1),n] f .
It is easy to verify that if the rst surface of the lens is curved (c
2
= 0), there is in
general no real position for a coma-correcting pupil.
5
Astigmatism and eld curvature
The difference in the curvatures of the two focal surfaces C
s
and C
t
is given by
LC = C
t
C
s
= (2,f )(Z
2
B +2ZF
0
+1).
Petzvals theorem
This gives the relation between the Petzval sum P
z
= Y
i
_
1
n
i
f
i
_
, the difference of
curvature LC and the two curvatures of the sagittal and tangential elds:
C
t
= YP
z
+(3,2)LC.
C
s
= YP
z
+(1,2)LC.
For a mirror, the Petzval sum is found by treating it as a lens with a refractive index 1.
Then
YP
z
= 1,f.
4
From Chapter 4, page 35.
5
But for a fortuitous exception see Chapter 4, page 37.
178 Appendix 1
Flat-eld conditions
Generally a at eld free from astigmatism is obtained only when the coefcients C and D
are both zero. The condition for a at tangential eld in a spectrograph is
4C +2D = 0.
Appendix 2
Wavelengths of spectral lines for calibration
Laboratory discharge-lamp spectra
Laboratory gas-discharge lamps are commonly available with a wide variety of elemental
contents and Table A2.1 contains the wavelengths usually encountered in the popular
ones. Avoid laboratory discharge lamps such as zinccadmiummercury which are
operated at high pressure. It is most important that low-pressure lamps be used to
ensure narrow spectrum lines which suffer no pressure- or temperature-broadening.
Table A2.1 lists the wavelengths in angstroms of the bright emission lines in the
common low-pressure laboratory discharge lamps. Wavelengths, are air wavelengths.
Wavenumbers, for which the unit is cm
1
, are always vacuum measurements and must be
increased by a factor (n 1) to convert them to air wavenumbers. The refractive index, n,
of air is given to sufcient accuracy for most practical purposes by
1
(n 1) 10
6
= 8340.78 +[2 405 640,(130
2
)] +[15 994,(38.9
2
)].
where is the vacuum wavenumber (= 1,
vac
).
Solar Fraunhofer lines
These are chiey for rough calibration of lowmedium resolution instruments and the
lines are sometimes broad and not well dened. Their advantage is that they are always
available, although visual identication is not always easy and an annotated photographic
copy of the solar spectrum is useful for eld-work. It is worth bearing in mind that
although lines are sometimes over 1

A wide at half-depth the centroids can be located to
0.1

A or less, which is adequate for calibration of a spectrographic resolving power of
10 000.
Lines C, D and F are usually used to specify the refractive indices and Abb e v-values
of optical glasses and other transparent materials. Some caution is needed. Older
astrographic telescope objectives, used with orthochromatic emulsions, were computed to
be achromatic in the violet and near UV and used different lines as the standard correction
wavelengths.
1
J. C. Owens, Appl. Opt., 6 (1967), 51.
179
180 Appendix 2
Table A2.1 Wavelengths (

A) of discharge lamps
Hydrogen Mercury Sodium Helium Cadmium Zinc
6562.79 11287 11404.2 10830 6438.4696 6362.35
4861.33 10140 11382.4 7065.19 5085.82 5181.99
4340.47 5790.65 8194.81 5875.62 4799.91 4810.53
4101.74 5769.59 8183.27 4685.75 4687.15 4722.16
3970.07 5460.75 5895.92 4471.68 3610.51 4680.14
3889.06 4358.35 5889.95 4026.19 3466.20 3345.02
3835.40 4046.56 5688.22 3888.65 3403.65 3282.33
3797.91 3663.28 5682.66 3261.06
3770.64 3650.15 3303.00 2748.58
3750.16 2536.52 3302.32
Notes
(1) The cadmiumline at 6438.4696

Ais an old international wavelength standard. Being the
result of a singlet transition it is a particularly narrowline and eminently suitable for demon-
strating the normal Zeeman effect. It was subsequently replaced by the 6056.125

A line of
krypton-86 but cadmium lamps are still in common use and readily available.
(2) The zinc line at 5181.99

A is a convenient standard wavelength for measuring small
Doppler shifts in a moving source of the Mg triplet, a group of closely spaced Fraunhofer
lines in a solar or stellar spectrum.
(3) The HeNe laser line is at 6328.17

A.
Arc spectra
The spectrum of an iron arc is less convenient than a set of laboratory lamps and the
necessary apparatus is mostly to be found in older, well-equipped spectroscopic
laboratories. The iron arc in fact is the traditional source for the visible and near UV
region, with many hundreds of well-measured wavelengths. Sundry authors have
recommended a vertical arc about 12 mm long taking 5 amps from a 230-volt source and
with only the central 2 mm of the arc focused on to a horizontal slit. Welders dark glasses
are needed when adjusting it and it is a source of skin-damaging UV.
Alternatively, a hollow-cathode lamp will yield a similar spectrum but at much lower
intensity.
Passive wavelength standards
There are sundry methods of calibrating spectrographs which are independent of
electricity or sunlight and therefore suitable for eld-work at remote sites. Chief among
these is absorption spectroscopy using a tungsten lamp, possibly battery-powered, as a
light source.
One of the most useful calibrators for the visible region is an absorption cell preferably
of glass or quartz, with a diameter of a few centimetres and an optical absorption path
length of about 75 cm. It should be evacuated or possibly lled with an inert gas such as
helium, and should contain some crystals of iodine. At the equilibrium vapour pressure of
Wavelengths of spectral lines for calibration 181
Table A2.2 Solar Fraunhofer lines
Letter Wavelength,

A Element Source
A 7621 O Telluric absorption band
7594 O Telluric absorption band
B 6869.95 O Telluric absorption band
C 6562.79 hydrogen Balmer H
D 5895.92 Na Sodium resonance line
D 5889.95 Na Sodium resonance line
E 5269.54 Fe
b
1
5183.62 Mg Magnesium triplet
b
2
5172.70 Mg Magnesium triplet
b
3
5167.51 Fe
b
4
5167.33 Mg Magnesium triplet
F 4861.33 H Balmer H
G

4340.46 H Balmer H
G 4307.91 Fe
4307.74 Ca
g 4226.73 Ca
h 4101.74 H Balmer H
H 3968.47 Ca II Ionised calcium resonance
K 3933.67 Ca II Ionised calcium resonance
L 3820.43 Fe
M 3727.67 Fe
N 3581.21 Fe
O 3440.99 Fe
P 3361.21 Ti II
Q 3286.76 Fe
R 3181.28 Ca II
3179.33 Ca II
s
1
3100.67 Fe
3100.30 Fe
s
2
3099.94 Fe
T 3021.07 Fe
iodine at room temperature the electronic transition X
1
Y
+
g
B O
+
u
of I
2
gives a
well-known series of absorption bands.
2
These are sufciently deep and have a rotational
structure open enough that the narrow individual absorption lines give a large number of
accurate calibration points throughout the yellow and green. The amount of absorption
and therefore the depth of the lines depends markedly on the ambient temperature but the
wavelength calibration is very precise.
More compact, with some absorption lines in the visible but a large number in the UV,
is europium.
3
An absorption cell with an absorption path length a few millimetres long
2
I. J. McNaught, J. Chem. Educ., 57 (1980), 101.
3
G. Smith & F. S. Tomkins, Proc. Roy. Soc., A342 (1975), 149; Phil. Trans. Roy. Soc., A283 (1976), 345.
182 Appendix 2
lled with an aqueous solution of europium chloride is highly portable. The lines result
from atomic transitions betweeen energy levels of unlled inner electron shells in the
europium atom. The lines are not generally as narrow as those in the vapour phase but as
calibration marks are comparable with solar Fraunhofer lines.
Appendix 3
The evolution of a FabryPerot
interference spectrograph
This is to show the stages of evolution of a design for a FabryPerot spectrograph for
night airglow or astronomical applications.
(1) Astronomical considerations required that the resolution should be 1.2

A at
5183.6

A, the wavelength of the Mg triplet of solar Fraunhofer lines. The free spectral
range was to be about 20

A, set by the FWHM of the interference lter which was to act as
order-sorter. The available nesse was 40 and this constrained the etalon gap thickness to
be 28 m. The order at the centre of the ring system is then 56 m,0.5183.6 m = 108.
(2) The available CCD camera used a 50 mm F,1.4 Takumar lens and a CCD with
1152 325 pixels each 25 m square. The eld semi-angle was then 16

. A primary
requirement was for spatial resolution, so that both the sky and the FabryPerot fringes
were to be in focus on the CCD. It was also required that the angle subtended by the CCD
on the sky should not be greater than 5

.
(3) The etalon gap thickness determined the angular radii of the FabryPerot rings
(assuming m
0
= 108 at the centre) which are given by cos
n
= (m
0
n),m
0
with
n = 1. 2. 3 . . . so that cos
1
= 7.8

. The next rings are at 11.04

and 13.53

. Their
diameters on the CCD are at 50 mmtan
n
which are 0, 13.7 mm, 19.5 mm and 24.07 mm,
so that the CCD records approximately three orders of interference. This would be so even
if the order at the centre is not integer.
(4) Conservation of etendue then required a telescope with an aperture
1
32,5 times that
of the camera lens, i.e. 160 mm. The design then compromised with an available
plano-convex lens of 6

(152 mm) aperture and 1000 mm focal length. This would give
adequate resolution on the sky and the small spectral range meant that chromatic
aberration was no problem. It remained to collimate the image so that the sky could be
received at the CCD with the FabryPerot fringes superimposed on it. At the same time
this collimator had to image the objective on to the pupil of the camera. The result was the
conguration shown in Fig. A3.1a in which the image collimator needed is clearly
inconveniently large. This diagram, like the others, is not to scale, the vertical dimension
having been greatly exaggerated for clarity.
(5) The eld of the objective required by the camera is in the ratio (again conserving
etendue) of their pupil diameters, so that the eld semi-angle for the objective is
25,150 16

= 2.66

. The actual eld radius is then 46.5 mm, requiring the order-sorter
to have an effective diameter of 93 mm. In the actual instrument a standard 100 mm
diameter interference lter was used.
1
The ratio of the eld angles.
183
184 Appendix 3
etalon
CCD camera
image collimator objective
objective Chief-ray collimator Order-sorter image collimator CCD camera
objective
(a)
(c)
(b)
Figure A3.1 The evolution of a FabryPerot CCD spectrograph following the
precepts of Chapter 10. The vertical scale has been exaggerated for clarity. In
practice a 150 mm diameter objective would result in an instrument 2 m long.
(a) The primitive design. Objective collimator camera.
(b) The addition of a chief-ray collimator immediately to the left of the order-
sorting interference lter moves the exit pupil (and hence the CCD camera) in to
the F
1
focal plane of the image collimator.
(c) The splitting of the image collimator into two smaller lenses, with the same
focal length and P
1
principal plane as the original singlet lens, reduces the overall
diameter of the spectrograph to that of the objective. Ideally the rst member of the
pair should have a focal length which makes the marginal ray parallel to the optic
axis, but in practice one may compromise with the contents of a manufacturers
catalogue.
(6) It was a sine qua non that the order-sorter should have all chief-rays parallel to the
optic axis as they passed through it. This was achieved by inserting, just in front of the
focus of the 1000 mm objective lens, a chief-ray collimator lens of sufcient diameter
(100 mm) to cover the necessary 5

eld and of 1000 mm focal length. This came from

the same catalogue as the objective.
The order-sorter could then be placed close to, but not exactly at the focus of the
objective. Any other position for the order-sorter would have entailed steeper marginal
rays than those from the F,6.66 of the objective and would have reduced the efciency of
the component.
(7) This chief-ray collimator made no difference to the image collimator but meant that
its diameter could be reduced somewhat. The other effect was to move the exit pupil of the
system to the back-focus of the image collimator, shortening the system slightly.
These various criteria resulted in the conguration shown in Fig. A3.1b. A scale
drawing of the optical layout then determined the diameter needed for the etalon plates.
The evolution of a FabryPerot interference spectrograph 185
(8) The image collimator was still inconveniently large in diameter and so, with the
object of reducing the diameter of the whole instrument, it was replaced by a pair of lenses
with the same focal length and the same positions for the principal planes. There was
some latitude in design here and items from the catalogue were found which would give
the result required.
This resulted in the nal form of the instrument shown in Fig. A3.1c.
Appendix 4
The common calibration curve in silver
halide spectrophotometry
The problem of photometry with silver halide emulsions is that of nding the relation
between density and exposure. Density is dened as minus the logarithm of the
transmission ratio, i.e. of the fraction of incident light transmitted through the emulsion to
the detector (on the assumption that the response of the detector is linear). Exposure is
dened as the product of incident intensity during the photographic exposure and the
duration of the exposure, and it is assumed, justiably, that the density depends only on
the product.
The method of the common calibration curve
1
is to take several photographs of the
same spectrum under identical circumstances, but with different neutral-density lters, the
lter densities ascending for example in factors of 2. The densities of a particular
spectrum line, measured from the different exposures, will give a set of points on the
calibration curve.
2
A different spectrum line can then be interpolated on this initial line
and the different exposures of this second line provide further points which extrapolate the
curve at one end or the other. By continuing in this fashion with other spectrum lines, the
curve can be continued, always by interpolating new initial material, gradually extending
the curve in both directions as far as necessary and well beyond the linear portion.
The result is not of course absolute, but will allow comparison of intensities (provided
the wavelength range is not too great) with an accuracy of up to 1% provided that proper
care has been taken with all the processing details.
1
I have been unable to nd any reference to the original inventor of this technique. I myself learned it and used it
while doing spectrography in R. W. Ditchburns XUV laboratory in Reading University during the last century.
2
Two will do at a pinch. Three or four is better.
186
Bibliography
Blackman, R. B. & Tukey, J. W., The Measurement of Power Spectra (New York: Dover,
1959).
Braddick, H. J. J., The Physics of Experimental Method (London: Chapman & Hall, 1963).
Conrady, A. E., Applied Optics and Optical Design (London: Oxford University Press,
1929).
Harrison, G. R., Lord, R. C. & Loofbourow, J. R., Practical Spectroscopy (New York:
Prentice-Hall, 1948).
James, J. F. & Sternberg, R. S., The Design of Optical Spectrometers (London: Chapman
& Hall, 1969).
Kingslake, R., Lens Design Fundamentals (New York: Academic Press, 1978).
Kingslake, R., A History of the Photographic Lens (San Diego, CA: Academic Press,
1989).
Loewen, E. G. & Popov, E., Diffraction Gratings and Applications (London: Marcel
Dekker Ltd., 1977).
Moore, J., Davis, C., Coplan, M. & Greer, S., Building Scientic Apparatus (Boulder, CO:
Westview Press Inc., 2005).
Palmer, C., Diffraction Grating Handbook, 6th edn. (Rochester, NY: Newport
Corporation, 2005).
Rieke, G., Detection of Light (Cambridge University Press, 2002).
Stroke, G. W., Diffraction Gratings, in Handbook of Physics, Vol. 29 (Berlin: Springer,
1967).
187
Index
aberrations, 28
aberration coefcients
chromatic, 39
for lenses, 175
for mirrors, 37
formul, 176
of Rowland grating, 97
aberration theory, Seidel, 29
Abney mounting for concace grating, 93
absorption
cells, 136
lines, 182
measurements, 170
addition theorem, 42
Airy function, 102
aliasing, 72
alignment, optical, 142, 156, 172
alignment tool, 173
aperture, 20
numerical, 20
aplanatic lens, 16
aplanatic surface, 15
apodising, 68
masks, 69
arc spectra, 181
astigmatism, 32, 33, 177
autocorrelation function, 45, 115
auxiliary optics, 128
bafes, 81, 141
external, 135
bearings, journal, 166
angular contact, 158
Bessel functions, 71
Beutler eikonal theory, 97
black body radiation, 25, 26
blazing, of gratings, 67
Bunsen, R., 1
Calibration
CCD sensitivity, 126
lines, laboratory, 180
lines, solar, 181
passive wavelength standards, 180
spectrograph, 126
wavelengths, 169, 180
cameras, CCD, 124
camera lenses, 40, 60, 85, 132
thorium in, 125
Cassegrain focus, 132
centred systems, 28
charge-coupled device (CCD), 123 et seq.
circle of least confusion, 30
circular slit theorem, 75
cleaning of optical materials, 164
diffraction gratings, 164
coma
control with a stop, 35, 36
equations for, 31, 35, 176,177
sagittal, 31, 82, 176
Seidel coefcients, 38
tangential, 81, 145, 177
coma patch, height of, 31, 177
common calibration curve, 185
congurations, spectrograph
Abney, 93, 94
alternative, 84, 85

CzernyTurner, 80
Eagle, 94, 95, 99
EbertFastie, 77
Littrow prism, 60
grating, 78
PaschenRunge, 93
PellinBroca, 60, 61
Pfundt, 79
Rowland, 93
SeyaNamioka, 96, 97
JohnsonOnaka, 96
Wadsworth stigmatic, 94, 95
Connes, P., 4
Conrady, A. E., 28, 111, 133, 175
convolutions, 42
convolution algebra, 44
convolution theorem, 44
cosmic ray events (CREs), 124
Coud e focus, 41, 53
188
Index 189
Court` es, G., 132
Crookes, Sir William, 2
cubic splines, 169
curvature, astigmatic, 97
curvature, eld, 33, 177
curvature, surface, see surface curvature

CzernyTurner mounting, 77, 80 et seq.

aberrations of, 81 et seq.
alignment of, 172
asymmetrical, 158
crossed, 81
vertical table, 153
worked example of, 145
decentred stop, 84
dispersion, 76
Diffraction, Fraunhofer, 52
single slit, 53
two-slit, 54
N-slit, 54
with oblique incidence, 54
Dirac comb, 48
Dirac delta function, 48
dispersion relation, in CCD spectrographs, 123
dispersion
and resolution, 4
curve for quartz, 59
differential, 76
partial, 6, 39
distortion, 34
doublet, achromatic, 39
DQE (detector quantum efciency), 3
Eagle mounting for concave grating, 94
Ebert, H., 2
EbertFastie mounting, 77
echelle, 3
EdserButler fringes, 66
emulsions, silver halide, 5, 76, 121
equivalent width, 170
etalon, see FabryPerot
etendue, 5, 24, 62
F-number, 20
Fabry, C., 3
Fabry lens, 129, 130
FabryPerot
etalon, 102
monochromator, 104
spectrometer, 105
mechanical scanning, 105
CCD spectrograph, 108, 183
spectrum extraction, 111
theory, 102
Fellgett, P., 5, 117
bre optical input, 137
eld, 21
angle, 31, 34, 81
stop, 21
at-eld condition, 83
focal isolation, 61
focal reducer, 132
in

CzernyTurner spectrograph, 133
focus, as zero-order aberration
paraxial, 31
sagittal, 32
tangential, 32
fore-optics, 128
Fouriers inversion theorem, 41
Fourier spectrometry, principles, 114
Fourier transforms in optics, 52
Fraunhofer, J. von, 1
Fraunhofer, diffraction, see diffraction, Fraunhofer
Fraunhofer, lines, 180, 181
free spectral range, 69, 103
FWHM (full width at half maximum), 46
Gauss, C. F., 12
Gaussian function, 46, 112, 137
Gaussian optics, 12, 28
ghosts
Lyman, 71
Rowland, 70
glass, 39, 58, 163
wavelength dispersion, 59
grating, diffraction, 63 et seq.
cleaning, 164
concave, 89 et seq.
holders, 156
transmission, 85
turntable, 158
HelmholtzLagrange invariant, 23
Herschel, Sir William, 1
Huygens, J. C., 52
incidence, oblique, 54
infra-red spectrometry, 62, 114
emulsions for, 121
iris diaphragm, 20 et seq., 84
Jacquinot, P., 4, 105
JohnsonOnaka mounting, 96
Kinematic mounting, 151
principles, 166
Kingslake, R., 40, 145
Kirchhoff, G., 1
Lamberts law, 25
radiator, 126
lenses
coma of, 36, 175
double-Gauss, 60, 134
focal planes and points, 14, 18
holders, 153
nodal planes and points, 19
plano-convex, 36, 132, 145, 176
principal planes and points, 18
Tessar, 60
thick, 15
lensmakers equation, 15
190 Index
Littrow mounting
grating, 78
prism, 59
LN
2
, 120, 161
Lorentz function, 46
luminosity, 24, 26, 112
luminosityresolution product, 4
Lummer, O., 3
mechanical design, 150
Meinel, A. B., 132
meridional coma, see tangential coma
Mertz, L., 2, 4
Michelson, A. A., 2
Michelson Fourier spectrometers, 62, 114 et seq.
cats eye, 116
cube-corner, 116
tilting, 116
mirrors
adjustments, 155
holders, 153
paraboloidal, 30, 36, 38, 77, 80, 163
spherical, 30, 38, 76, 163
multiplex spectrometer, 114
advantage, 117
multiplication theorem, 42
multislit scanning, 108
Newtonian mirrors, 79, 111
numerical aperture, 20
oblique incidence, 54
Onaka, R., 96
optical design, 139
layout, 139, 150
transparent materials, 163
renement, 29, 140 et seq.
optical screw, 106
order overlap, 68
paint
optical black, 167
protective, 167
Parsevals theorem, 42
PaschenRunge mounting for concave grating, 93
path
geometrical, 11
optical, 11
reduced, 12
PellinBroca prism, 60
Peltier cooling, 123
Periodic errors, 70
Perot, A., 3
Petzval sum, 37
Petzvals theorem, 37
Pfundt mounting, 79
phase angle, 101
photography, silver halide, 3, 76, 92, 120, 185
photomultipliers, 122
piezo-electric transducers, 105, 106
pixels, 84, 108, 123, 145, 169, 183
Planck formula, 26
pressure scanning, 107
prism spectrographs, 57
PellinBroca, 61
Littrow, 60
Prole, instrumental, 43, 47, 69, 103
FabryPerot etalon, 111
pupil, control of aberrations by, 21
pupils, laterally displaced, 84
Q plates, for XUV spectrography, 122
quartz, wavelength dispersion of fused, 59
Ramsden, J., 132, 145
ray-tracing, computer, 140
methods, 142
programs, 144
Rayleigh criterion, 58
reciprocity failure, 3, 5, 120
refractive index, 11, 58, 85
determination of, 57
of mirror, 15
resolution, 58
resolving power
dened, 24
of diffraction grating, 64
of FabryPerot etalon, 105
of prism, 58
Rosendahl condition, 81
Rowland, H. A., 2
Rowland circle, 89
Rowland grating theory, 89 et seq.
Rowland mounting for concave grating, 93
Rubens, H., 61
sagittal coma, 31
sagittal focus, 33
sampling theorem, 49
scattering
causes of, 135
from edges, 135
Schmidt camera spectrography, 134
Schuster, Sir Arthur, 2
screw, optical, 106
seidel aberrations of decentred stop, 84
SeyaNamioka mounting, 96
shift theorem, 42
shroud, spectrograph cover, 167
sinc function, 46
slitless spectrography, 131
slits, construction of, 159, 161
spectral power density (SPD), 45, 115
spectro, slitless, 131
spectrograph mountings, see congurations
spectrum lines, 179, 180, 181
shapes of, 47
spherical aberration
Seidel coefcient, 30
equation for, 176
indicator, 145
spot diagrams, 148
Index 191
Stefans constant, 25
stops, 20, 28
eld
lateral displacement of, 129
telecentric, 84
sulphones, 163
surface brightness, 25
surfaces curvature
focal, curvature of eld, 33 et seq., 177
focal, at eld condition, 83
Petzval, 37
sagittal, 37
tangential, 37
tangential coma, 85, 145, 177
tangential focus, 94, 96
telecentric operation, 129
stops, 129
telescope
astronomical, 130
refracting, 22
temperatures
addition, 42
circular slit, 75
convolution, 44
focal curve, 59
Fraunhofers, 177
Parsevals, 42
Petzvals, 177
sampling, 49
shift, 42, 45, 67
WienerKhinchine, 45
top-hat function, 45
ultra-violet, 7, 122, 163, 181
UVOIR, 7
vacuum ultra-violet, 8
vertex, of optical surface, 12
vignette, 22
Voigt prole, 47
Wadsworth stigmatic mounting, 92, 94
wavelength calibration, 126, 169, 179
standards, 168
passive, 180
tables, 180 181
White cells, 136
WienerKhinchine theorem, 45
Wollaston, W. H., 1
Wood, R. W., 1
XUV, 7, 71, 92, 96
photographic plates for, 122
zero meridional coma (ZMC), 140