142 J. Opt. Soc. Am. B / Vol. 23, No.

1 / January 2006 Charles Cerjan

Lateral spatial effects in compressor grating pairs

Charles Cerjan
Lawrence Livermore National Laboratory, 7000 East Avenue, Livermore, California 94550

Received April 20, 2005; accepted August 8, 2005

An effective computational technique to estimate lateral spatial effects in laser beams compressed by diffrac-
tion grating plates is established. In particular, the equivalence of the Kirchoff–Fresnel integral form to the
exponential split-operator Fourier transform representation for a propagating-beam amplitude provides a
more general numerical technique. Thus realistic beam profiles may be accurately propagated through the
compressor pair, including grating-plate misalignment contributions. Additionally, the use of a stationary-
phase approximation to the propagation integral kernel is shown to be a convenient, reliable method to evalu-
ate analytical beam forms. Both the numerical and the approximate approaches generalize previously dis-
cussed ray-tracing approaches to optimize compressor design. © 2006 Optical Society of America
OCIS codes: 320.5520, 320.7160.

1. INTRODUCTION 2. GRATING-PAIR PROPAGATION

The use of diffraction grating plates to compress chirped The general approach relies on some simplifying assump-
laser beams is a well-established short-pulse laser tions about the propagating beam and its interaction with
technique.1,2 Effective recompression of a laser beam is each grating plate. The initial beam, Ainc共x , y , z , t兲, is as-
controlled by the orientation and spacing of the grating sumed to be initially separable in both space and time:
plates,3 but an important design complication arises from
the finite extent of the plates that can introduce spectral Ainc共x,y,z,t兲 = exp共ik · x兲A1共x,y兲B共t兲, 共1兲
clipping. The straightforward approach of simply under-
where the temporal component may be Fourier trans-
filling the diffraction plates is prohibited by the cost and
formed,
complexity of fabricating large diffraction plates. Thus re-

冕
+⬁
cent attempts to construct spatially large-beam petawatt
laser sources must directly confront the size limitations B̂共␯兲 = exp共2␲i␯t兲B共t兲dt, 共2兲
introduced by compressor plates.4–6 −⬁

Previous attempts to include lateral spatial effects and the spatial axes are initially oriented in the x – z
have centered upon ray-tracing methods7–9 or the plane so that k · x = k0x sin ␾ + k0z cos ␾, with k0 = 2␲ / ␭0.
Kirchoff–Fresnel integral-equation approximation.10,11 The reflection of this beam from a diffraction grating is
Ray-tracing methods have the distinct advantage of com- determined by the grating equations for an arbitrary off-
putational speed, leading to rapid parameter searches for plane ray13
optimal operating conditions. On the other hand, this ap-
proach is not readily generalized to more-realistic beam sin ␪⬘ = − sin ␪ 共3兲
profiles and cannot easily estimate beam fluence. The
and
Kirchoff–Fresnel approximation developed by Martinez is
capable of directly capturing many important spatial ef- m␭
fects but was limited to a normal Gaussian profile.10 The sin ␾ + sin ␾⬘ = , 共4兲
dg cos ␪
development presented in this paper generalizes this lat-
ter approach to arbitrary spatial profiles for a compressor where the primed angles are the reflected angles, m is the
grating pair using a well-established pseudospectral (fast order of the grating, dg is the grating period, and ␭ is the
Fourier transform) discretization.12 wavelength. This coordinate system is displayed in Fig. 1.
In the following sections, the underlying simplifications Using the relationship ␭ = 2␲c / ␻, the expression in Eq.
are discussed that establish the basic propgation inte- (4) may be written as
grals in a form that is amenable to numerical treatment.
Section 2 presents the details of the stationary-phase ap- m2␲c
sin ␾ + sin ␾⬘ = + f共⌬␻兲, 共5兲
proximation as applied to these specific propagation ker- ␻0dg cos ␪
nels. It should be noted that this approach yields a first-
order approximation to the previously derived Fresnel- where the higher-order Taylor series expansion of the fre-
Kirchoff Gaussian beam case.10 Finally, representative quency dependence has been collected in the function
numerical results are presented that display the utility f共⌬␻兲:
and limitations of this development, especially for more- f共⌬␻兲 ⬇ ␤1⌬␻ + ␤2共⌬␻兲2 + ␤1共⌬␻兲3 + …, 共6兲
realistic beam shapes, including clipped beams arising
from an overfilling of the grating plates. A brief summary with ⌬␻ = 共␻ − ␻0兲. Assuming that all angular deviations
concludes the discussion. are small upon reflection, the spatial coordinates in the

0740-3224/06/010142-7/$15.00 © 2006 Optical Society of America

Charles Cerjan Vol. 23, No. 1 / January 2006 / J. Opt. Soc. Am. B 143

B̂共␻兲
冕
+⬁
A3共x⬙, ␻,zg兲 = dv1dv2 exp关− 2␲i共vM兲 · x⬙兴
2␲ −⬁

⫻ exp − 冋 2␲2izg
k0
共vM兲 · 共Mtv兲 册
冕
+⬁
⫻ du1du2 exp共2␲iv · u兲A1共u兲
−⬁

⫻exp关ik0⌬␻␤1共M−1u兲1兴, 共12兲

where the substitutions u = Mx and k = Mtv have been

made. The transposition of either a vector or a matrix is
denoted by the superscript t.
In the special case of a Gaussian form and no out-of-
plane contribution, Eq. (12) is
Fig. 1. Coordinate system chosen to orient a diffraction grating.

B̂共␻兲
冕
+⬁
reflected coordinates are related by a linear AG共y, ␻,zg兲 = dv1dv2 exp关− 2␲i共av1y1 + v2y2兲兴
transformation:13 2␲ −⬁

x1⬘ = − x1
cos ␾0
cos ␾0⬘
− x2
m␭0 sin ␪0
dg cos ␪0 cos ␾0⬘
, 共7兲 ⫻ exp − 冋 2␲2izg
k0
共a2v12 + v22兲 册
冕
+⬁
x2⬘ = x2 . 共8兲
⫻ du1du2 exp共2␲iv · u兲
−⬁
Written in matrix form, this transformation becomes

冉 冊 冉 冊冉 冊
x1⬘
x2⬘
=
a b
0 1
x1
x2
, 共9兲
⫻ exp兵− ik0共u12 + u22兲/关2q共z0兲兴其

⫻exp共ik0⌬␻␤1u1/a兲, 共13兲

or x⬘ = Mx, where a and b are the coefficients in Eq. (7). where q共z兲 = z + ik0␴2 / 2 for a beam with spatial width ␴.
The reflected beam, with these first-order approxima- The integrals all have analytic expressions so that
tions, is now expressed as

A2共x⬘, ␻兲 = B̂共␻兲A1共Mx兲exp共ik0x1␤1⌬␻兲. 共10兲 B̂共␻兲 exp关− ik0共y12 + y22兲/共2zg兲兴

AG共y, ␻,zg兲 =
2␲ 关q共z0 + a2zg兲/q共z0兲兴1/2

再
Thus the physical effect of the grating is to transform the
lateral spatial coordinates and to introduce a frequency-
dependent phase.10 To propagate this reflected beam a ⫻ exp ik0 冠 q共z0兲
2zgq共z0 + a2zg兲

冎冡
distance zg to the second grating plate, the pseudospectral
vacuum propagation operator form may be used.12 In
other words, this well-known technique requires multipli- ⫻关共y1 − zg⌬␻␤1兲2 + y22兴 . 共14兲
cation of the Fourier transform of the initial beam by the
exponential form of the momentum space operator, fol-
lowed by an inverse Fourier transform on the combined Qualitatively, the grating displaces the in-plane spatial
function. Mathematically, then, the beam reflected from coordinate by an amount proportional to the propagation
the first grating and propagated to the second grating has distance, frequency shift, and grating parameters. This
the form result agrees with the Kirchoff–Fresnel expression de-
rived previously.10 In effect, the split-step Fourier trans-
B̂共␻兲
冕 form representation provides a numerically efficient
+⬁
A3共x⬙, ␻,zg兲 = dk1dk2exp共− 2␲ik · x⬙兲 evaluation of the Fresnel–Kirchoff integral,14 as may be
2␲ −⬁ immediately seen by interchanging the order of integra-

冉 2␲2izg
冊 tion in Eq. (11) and noticing that

冉 冊
⫻exp − k·k
k0
冕 2␲2izg
+⬁
dk1dk2 exp关2␲ik · 共x − x⬙兲兴exp −
冕
+⬁
k·k
−⬁
k0

冋 册
⫻ dx1dx2 exp共2␲ik · x兲A1共Mx兲
−⬁ ik0
= exp − 共x − x⬙兲2 . 共15兲
⫻exp共ik0x1␤1⌬␻兲. 共11兲 2zg
Now, by switching integration variables in both integrals, Returning to the general development, the lateral spa-
the propagated beam may be expressed as tial coordinates of the beam reflected from the second
144 J. Opt. Soc. Am. B / Vol. 23, No. 1 / January 2006 Charles Cerjan

grating plate are transformed by a matrix M̃, and a The second integral has the explicit value
frequency-dependent phase ˜␤1 appears so that the re-
flected beam has the form 冕−⬁
+⬁
dk1dk2exp关2␲ik · 共M−1u − x⬙兲兴exp − 冉 冊 2␲2izg
k0
k·k

冉 冊 冉 冊
B̂共␻兲
A4共x⬙, ␻,zg兲 = exp关ik0⌬␻˜␤1共M̃x⬙兲1兴 i s12 s22
2␲ = exp i exp i , 共19兲
4␲d1 4d1 4d1

冕
+⬁
⫻ dv1dv2exp关− 2␲i共vM兲 · 共M̃x⬙兲兴 where s = 共M−1u − x⬙兲 and d1 = −zg / 共2k0兲. The remaining
−⬁ double integral thus becomes

⫻ exp − 冋 2␲2izg
k0
共v · MMt · v兲 册 冕−⬁
+⬁
ds1ds2A1关M共s + x⬙兲兴exp关ik0⌬␻␤1共s1 + x1⬙ 兲兴

冕 冉 冊 冉 冊
+⬁
ia s12 s22
⫻ du1du2 exp共2␲iv · u兲A1共u兲 ⫻ exp i exp i . 共20兲
−⬁ 4␲d1 4d1 4d1
⫻exp关ik0⌬␻␤1共M−1u兲1兴. 共16兲 The phase variation is very rapid for propagation dis-
tances on the order of meters and wavelengths in mi-
Ideally, compensating plates are thus required to fulfill
crometers, so the major contribution to the integral is
the condition MM̃ = I, which corresponds to an inversion of given by the regions of stationary phase in the exponen-
the incident and reflected in-plane angles 共␾⬘ ↔ ␾兲 while tial factors, namely ssp = 共−2d1k0⌬␻␤1 , 0兲. Applying this
fixing the out-of-plane angle 共␪⬘ = ␪兲. When this condition result produces the approximation
is fulfilled, an effective calculational grid method arises
for nonanalytic beams. That is, for a given input beam, Asp共x⬙, ␻,zg兲 ⬇ A1关a共x1⬙ − zg⌬␻␤1兲,x2⬙ 兴
possibly defined only on a grid, the algorithm is as fol-
⫻exp关ik0⌬␻␤1共x1⬙ − zg⌬␻␤1兲兴. 共21兲
lows. First, multiply the beam by the appropriate
frequency-dependent phase factor and spatially trans- Thus reflection from a grating adds a frequency-
form the product; second, multiply by the exponential dependent displacement to the in-plane lateral coordi-
propagation kernel; and finally, multiply the inverse Fou- nate.
rier transform of this product by another frequency- This last expression may be used to estimate lateral
dependent phase factor. propagation effects when an analytic form of the incident
Again, for the case of an ideally compensating grating beam is known. For the special case of a Gaussian beam,
pair, the final form of the Gaussian beam in Eq. (14) has this form matches the Kirchoff–Fresnel result10

冉 冊 冋 册
the form
k 0x 2 ik0q共z0兲
B̂共␻兲 exp关ik0zg⌬␻2␤12/2兴 exp − i exp 共x − zg⌬␻␤1兲2 共22兲
AG共x⬙, ␻,zg兲 = 2zg 2zgq共z0 + a2zg兲
2␲ 冑关q共z0 + a zg兲/q共z0兲兴
2 1/2

再冋 册冎
to first order. Specifically, noting that
共x1⬙ − zg⌬␻␤1兲2 + 共x2⬙ 兲2
⫻ exp ik0 . q共z0兲 a 2z g
2q共z0 + a zg兲
2 ⬇1− , 共23兲
q共z0 + a2zg兲 q共z0兲
共17兲
where the wavelength is much smaller than the propaga-
This result is identical to the Kirchoff–Fresnel expression tion distance, the fully propagated Gaussian beam is ap-
derived earlier.10 proximately

3. STATIONARY-PHASE APPROXIMATION
Although Eq. (12) generally requires numerical treat-
exp i 冋 k 0a 2
2q共z0兲
册
共x − zg⌬␻␤1兲2 exp关ik0⌬␻␤1共x − zg⌬␻␤1兲兴.

ment, a useful approximate form may be readily derived 共24兲

from the analytic form of the propagation kernel and the
Application of the stationary phase produces an identical
Fourier transform. Interchanging the order of integration
form. Note also that the approximate form is not re-
in Eq. (12) produces
stricted by the exact compensating grating condition re-
B̂共␻兲
冕
+⬁ quired of the pseudospectral representation MM̃ = I. By
Asp共x⬙, ␻,zg兲 = du1du2A1共u兲exp关ik0⌬␻␤1共M−1u兲1兴 relaxing this restriction, more-general grating-pair mis-
2␲ −⬁ alignment effects may be readily estimated, provided, of

冕 course, that an analytical fit to the beam is possible.

+⬁
⫻ dk1dk2 exp共2␲ik · 关M−1u − x⬙兲兴
−⬁

冉 冊
4. NUMERICAL EXAMPLES
2␲2izg
⫻ exp − k·k . 共18兲 The efficacy of the above development is easily demon-
k0 strated numerically. As a specific example, consider a
Charles Cerjan Vol. 23, No. 1 / January 2006 / J. Opt. Soc. Am. B 145

the second grating. With a spatial extent ␴ (1 / e width) of

5 cm in both lateral dimensions, the pseudospectral and
approximate stationary phase formulations may be di-
rectly compared in several illustrative cases.
Since, as noted above, an analytic result arises for the
simple Gaussian form

冋
exp − ik0
共x2 + y2兲
q0 + z0
册 , 共28兲

where z0 is 450 cm and q0 = i␲␴2 / ␭0, it is possible to gauge

the accuracy of the numerical and stationary phase evalu-
ations. With these parameter choices, the temporal form
of the initial beam is displayed in Fig. 2. The result of ap-
plying the transform technique is displayed in the contour
plot of Fig. 3, where a cross section of the normalized ab-
solute value of the propagated beam immediately after
Fig. 2. Absolute value of the initial Gaussian beam plotted as a the second grating plate is plotted for fixed y against fre-
function of the in-plane coordinate x and time for the fixed out- quency and the in-plane coordinate x. Both the grid rep-
of-plane coordinate y. resentation and the stationary-phase approximation have

Fig. 3. Absolute value of the numerically propagated Gaussian

Fig. 4. Absolute, normalized value of the final temporally com-
beam after the second grating plate as a function of frequency
pressed Gaussian beam as a function of the in-plane coordinate
and the in-plane distance x. The grating plates couple the in- and time for the fixed out-of-plane coordinate.
plane coordinate and the frequency, as noted in the text.

beam with a wavelength of 1.053 ␮m incident at 74°, with

respect to the surface normal, impinging upon a perfectly
compensating grating pair ruled with 1740 lines/ mm
with separation determined by the applied chirp and ␤1.
The parameter ␤1 is derived from the expression
mdg␭02
␤1 = . 共25兲
2␲c cos ␾0
Assuming that the intial temporal response is given by a
chirped Gaussian form
exp关− 共cr − ici兲t2兴, 共26兲
the maximum compression from a grating pair will be at-
tained when
ci k0
= ␤12zg . 共27兲
共c2r + ci2兲 2
Fig. 5. Absolute value of the initial hyperbolic secant beam plot-
Choosing an initial FWHM temporal width of 1.25 ns and ted as a function of the in-plane coordinate x and time for fixed
a bandwidth of 2 Å will compress the beam to 8.2 ps after out-of-plane coordinate y.
146 J. Opt. Soc. Am. B / Vol. 23, No. 1 / January 2006 Charles Cerjan

a maximum relative error of 0.3% with respect to the ana-

lytical result. By applying an inverse Fourier transform
to the frequency, the temporal compression of the beam is
obtained. The contour plot in Fig. 4 demonstrates this ef-
fect, where the normalized absolute value of the beam ap-
pears as a function of time and x.
As a second example, another analytic beam descrip-
tion may be considered with the lateral form

冋
sech − ik0
共x2 + y2兲
q0 + z0
册 , 共29兲

which is plotted in Fig. 5 as a function of x and time be-

fore the first grating plate. The beam parameters and
temporal contribution are identical to those used above
for the Gaussian form. Propagation of this beam through
the grating pair produces a compressed beam, which is
Fig. 6. Absolute value of the numerically propagated hyperbolic plotted in Fig. 6 as a function of x and v. Application of the
secant beam after the second grating plate as a function of fre- stationary-phase approximation reproduces this form
quency and the in-plane distance x. The grating plates couple the with a maximum relative difference of 0.2%. The tempo-
in-plane coordinate and the frequency, as noted in the text.

Fig. 9. Absolute value of the numerically propagated sixth-

Fig. 7. Absolute, normalized value of the final temporally com- order super-Gaussian beam after the second grating plate as a
pressed hyperbolic secant beam as a function of the in-plane co- function of frequency and the in-plane distance x. The grating
ordinate and time for the fixed out-of-plane coordinate. plates couple the in-plane coordinate and the frequency, as noted
in the text.

Fig. 8. Absolute value of the initial sixth-order super-Gaussian

beam plotted as a function of the in-plane coordinate x and time Fig. 10. The grating plates couple the in-plane coordinate and
for the fixed out-of-plane coordinate y. the frequency, as noted in the text.
Charles Cerjan Vol. 23, No. 1 / January 2006 / J. Opt. Soc. Am. B 147

rally compressed expression is obtained by using the in-

verse Fourier transform, and this result is plotted in
Fig. 7.
A related, but more extreme, form that is often used is
a super-Gaussian function, such as

冋
exp −
共x2 + y6兲
␴6
册 . 共30兲

The initial beam, its transform to the frequency domain,

and its compressed temporal form are displayed in Figs.
8–10. Again, the maximum relative difference between
the numerical transform and the stationary-phase ap-
proximation in frequency space was 0.2%. The hyperbolic
secant and super-Gaussian forms are clearly seen to be
qualitatively similar to each other but significantly differ-
ent from the normal Gaussian beam. Thus it is important
to consider the approximate lateral extent of the beam
Fig. 11. Absolute value of the initial truncated Gaussian beam
plotted as a function of in-plane coordinate, x, and time for fixed when designing the compressor pair.
out of plane coordinate, y. Finally, the development above suggests a method to
determine the effects of beam truncation by the finite ex-
tent of the compressor pair. That is, it is possible to esti-
mate the change in the final compressed beam owing to
an overfilling of the grating plates. As an example of this
determination, the first case above (the normal Gaussian
form) was arbitrarily truncated at 5 cm and propagated
through an ideally compensating grating pair. The initial
beam, its frequency transform, and its final temporal
shape are plotted in Figs. 11–13. The truncation intro-
duces the expected rapid variation in frequency space
with a subsequent temporal spread in the final com-
pressed beam. When compared with the full, untrun-
cated, compressed Gaussian beam in Fig. 4, it is clear
that temporal spreading has been induced by the finite
extent of the grating plates with a consequent degrada-
tion in expected beam compression. These effects are im-
portant design considerations in the optimization of spa-
tially large short-pulse laser systems.5,7
Fig. 12. Absolute value of the numerically propagated trun-
cated Gaussian beam after the second grating plate as a function
of frequency and in-plane distance, x. The grating plates couple
the in-plane coordinate and the frequency as noted in the text.
5. CONCLUSION
Grating-plate temporal compression is an essential com-
ponent of many short-pulse, high-power laser systems.
The desired compressed form will be sensitive to plate ori-
entation, plate separation, and diffractive groove spacing,
all of which have been included in this paper. Generally,
the finite lateral extent of the beam propagating between
the grating plates will also induce qualitative changes in
the final temporally compressed beam, as demonstrated
by the several analytic forms considered. The develop-
ment above suggests two possible techniques to more ac-
curately assess changes induced by the lateral extent of
the initial beam. First, the exponential split-step pseu-
dospectral propagation method, commonly used in Fou-
rier optics, may be used to numerically investigate
nonanalytic beam shapes. Second, the stationary-phase
approximation can be applied to the propagation kernel
without performing the numerical transform and inver-
sion, so the temporal compression of an incident beam
Fig. 13. Absolute, normalized value of the final temporally com-
that is reasonably well fit by a standard functional form
pressed truncated Gaussian beam as a function of in-plane coor- may be directly evaluated. For most expected operating
dinate and time for fixed out of plane coordinate. conditions, this latter approximation appears to be ad-
148 J. Opt. Soc. Am. B / Vol. 23, No. 1 / January 2006 Charles Cerjan

equately reliable. Both the purely numerical and approxi- T. Toner, K. W. M. Wigmore, T. B. Winstone, R. W. W.
mate evaluations compare favorably to normal Gaussian Wyatt, and F. Zou, “High contrast multiterawatt pulse
generation using chipred pulse amplification on the
beam Fresnel–Kirchoff propagation and generalize its VULCAN laser facility,” Opt. Commun. 103, 392–397
utility to more-complicated functional forms. Such a gen- (1993).
eralized approach is necessary to enhance the perfor- 5. I. N. Ross, M. Trentelman, and C. N. Danson,
mance of short-pulse, high-power system designs. “Optimization of a chirped-pulse amplification Nd:glass
laser,” Appl. Opt. 36, 9348–9358 (1997).
6. M. D. Perry, D. Pennington, B. C. Stuart, G. Tietbohl, J. A.
ACKNOWLEDGMENTS Britten, C. Brown, S. Herman, B. Golick, M. Kartz, J.
Miller, H. T. Powell, M. Vergino, and V. Yanovsky,
The author would like to thank D. Pennington for sug- “Petawatt laser pulses,” Opt. Lett. 24, 160–162 (1999).
gesting this problem and K. Manes for a careful reading 7. M. Trentelman, I. N. Ross, and C. N. Danson, “Finite size
of the manuscript. This work was performed under the compression gratings in a large aperture chirped pulse
amplification laser system,” Appl. Opt. 36, 8567–8573
auspices of the U. S. Department of Energy by the Uni- (1997).
versity of California, Lawrence Livermore National Labo- 8. P. F. Curley, G. Darpentigny, G. Cheriaux, J. P. Chambaret,
ratory under Contract W-7405-Eng-48. The author can be and A. Antonetti, “High dynamic range autocorrelation
reached na e-mail at [email protected]. studies of a femtosecond Ti:sapphire oscillator and its
relevance to the optimisation of chirped pulse amplification
systems,” Opt. Commun. 120, 71–77 (1995).
REFERENCES 9. K. Osvay and I. N. Ross, “On a pulse compressor with
gratings having arbitrary orientation,” Opt. Commun. 105,
1. M. D. Perry and G. Mourou, “Ultrafast phenomena,” 271–278 (1994).
Science 264, 917–924 (1994). 10. O. E. Martinez, “Grating and prism compressors in the case
2. C. P. J. Barty, C. L. Gordon III, and B. E. Lemoff, of finite beam size,” J. Opt. Soc. Am. B 3, 929–934 (1986).
“Multiterawatt 30-fs Ti:sapphire laser system,” Opt. Lett. 11. O. E. Martinez, “Pulse distortions in tilted pulse schemes
19, 1442–1444 (1994). for ultrashort pulses,” Opt. Commun. 59, 229–232 (1986).
3. C. Fiorini, C. Sauteret, C. Rouyer, N. Blanchot, S. Seznec, 12. G. P. Agrawal, Nonlinear Fiber Optics (Academic, 2001),
and A. Migus, “Temporal aberrations due to misalignments pp. 51–54.
of a stretcher–compressor system and compensation,” 13. J. M. Simon and A. N. Fantino, “Sine relation for optical
IEEE J. Quantum Electron. 30, 1662–1670 (1994). linear systems: matrix formulation,” J. Mod. Opt. 38,
4. C. N. Danson, L. J. Barzanti, Z. Chang, A. E. Damerell, C. 1117–1124 (1991).
B. Edwards, S. Hancock, M. H. R. Hutchinson, M. H. Key, 14. M. Born and E. Wolf, Principles of Optics, 6th Ed.
S. Luan, R. R. Mahadeo, I. P. Mercer, P. Norreys, D. A. (Cambridge U. Press, 1980), Section 8.3.3.
Pepler, D. A. Rodkiss, I. N. Ross, M. A. Smith, P. Taday, W.