672 EDWARD C. DOERNER Vol.

52

in Fig. 2 or Fig. 3, is scanned with an optical system
whose transfer function is lower than that of the
diffraction-limited microscope system used in the
microphotometer, the Wiener spectrum of the sample
is modified by the transfer function of the system. The
curve in Fig. 4 represents the modulation transfer
function of a certain lens obtained in the routine
manner. This lens was used in the microphotometer to
scan a uniformly exposed and developed sample of
Kodak Tri-X Film, and the resulting Wiener spectrum
divided by the Wiener spectrum of this "white-noise
generator" gave the values of responses indicated by
the open circles. These circles fall practically on the
curve and therefore agree with the modulation transfer
factors of the lens as measured directly.
IV. APPLICATION TO X-RAY FILMS
One of the most useful applications of this technique
has been to x-ray films. When these films are exposed
between two fluorescent intensifying screens, a mottle
is produced which has been traced to the quantum
fluctuations in the exposing x rays.9 The spectrum-
analysis problem here is complicated by the fact that
the x-ray emulsion is coated on both sides of the film
base ("duplitized") and is exposed from screens on both
sides. Diffusion of the light within each screen means
that there must be a transfer function associated with
the screen as well as a Wiener spectrum associated with
the quantum mottle. In fact, the mottle is confined to
very low spatial frequencies-the region where it is most
easily visible. Using the technique of Wiener-spectrum
I K. Rossmann, J. Opt. Soc. Am. (to be published).
analysis, Rossmann obtained values for the response
function of x-ray film-screen combinations in a manner
similar to that used to obtain the lens function in Fig. 4.
Figure 5 gives the Wiener spectra for the cases of
quantum mottle produced by three different screens,
showing characteristic differences between them.
Figure 6 illustrates a problem encountered in scanning
x-ray film samples that will help to clarify the experi-
mental approach to the analysis of the Wiener spectrum.
The question arises when scanning a duplitized x-ray
material whether the two surfaces are contributing
equally to the measured granularity spectrum. An
experimental sample of an x-ray material coated on
only one side was exposed with the aid of intensifying
screens, processed as usual, and scanned in the analyzer.
Its Wiener spectrum is shown in Fig. 6 as the "focused"
curve. The sample was then reversed in the micro-
photometer and scanned again without being refocused,
the resulting Wiener spectrum being the "reversed"
curve. In this position, the spectrum as measured
should be similar to that of the rear surface alone during
the scan of an ordinary duplitized material. It is obvious
that the spectrum for the reversed position shows a
considerable loss of high-frequency content. In addition,
the spectrum representing the sum of the two individual
spectra (and also representing the scan of the duplitized
film), which should have the same shape as the spectrum
of a single focused layer on a logarithmic plot, is
distorted.
These examples show what a powerful tool the Wiener
spectrum is proving to be for investigating complex
problems n granularity, mottle, and the transfer of
images in general.

JOURNALOF THE OPTICALSOCIETYOF AMERICA VOLUME 52, NJMBER 6 JUNE 1962

General Ray-Tracing Proceduret

G. H. SPENCER* AND M. V. R. K. MURTY
Institute of Optics, University of Rochester, Rochester, New York
(Received October 4, 1961)

Computing formulas are presented for tracing skew rays through optical systems containing cylindrical
and toric surfaces of arbitrary orientation and position. Particular attention is given to the treatment of
diffraction gratings and a vector form of the diffraction law is suggested.

INTRODUCTION require the use of unusual surface shapes. Spectro-

ADVANCES
in the technology of optical-system
fabrication have made practical the design of
systems containing nonspherical surfaces. The demands
made of modern optical systems, in fact, frequently
t This work was partly supported by the National Aeronautic..
and Space Administration under a contract.
* The manuscript in its present form was completed while the
first author was at International Business Machines Corporation,
Thomas J. Watson Research Center, Yorktown Heights, New
York, during the summer of 1961.
graphic and related systems may employ one or more
nonspherical gratings, together with refracting or
reflecting surfaces in an uncentered configuration.
In order to facilitate the design and analysis of such
systems, various techniques of ray tracing have been
proposed. Feder' has described a computer-oriented
procedure for skew-ray tracing through rotationally
symmetric systems containing aspheric surfaces. Allen
I D. Feder, J. Opt. Soc. Am. 41, 630 (1951).
Jtlne 1962 GENERAL RAY-TRACING PROCEDURE 673

and Snyder2 have extended Feder's procedure to allow

for uncentered configurations of rotationally symmetric 6V Iz so
surfaces. This procedure requires that the designer
specify the direction cosines of the line joining the
vertices of adjacent surfaces. Murray3 has presented a
x
toric skew-ray trace which is suitable for surfaces whose
radii are not too large, and Herzberger4 has given a SI
detailed treatment of quadratic surfaces (conics of
revolution, parabolic cylinders, and elliptic paraboloids). FIG. 1. Reference coordinate system (X,F,Z) and local system
More recently, Rosendahl5 has described a method for (X, Y,Z), showing incident ray and computed distances, so and sf .
ray tracing through systems containing parallel ruled
gratings. Yoshinaga, Okazaki, and Tatsuoka 6 also used referred to a coordinate system (X, Y,Z) having its
ray-tracing procedures for evaluating prism and grating origin at a point 0. The orientation of the (X,Y,Z)
spectrometers. system relative to the (XYZ) system and the coordin-
It is the purpose of the present paper to describe a ates of 0 with respect to 0 are given. The ray-tracing
unified ray-tracing procedure applicable to systems of problem may now be separated as follows:
quite general type. The procedure has been developed (I) Transform the ray-point coordinates (X0,Y,Zo) 0
in accordance with the following requirements: and direction cosines (k,i,rn) into their values in the
(1) Cylindrical, toric, and conic surfaces must be (X,Y,Z) system.
accommodated with provision for specifying departures (II) Find the point of intersection of the ray with
from these forms. the surface S.
(2) Provision must be made for the arbitrary orienta- (III) Find the change in direction of the ray refrac-
tion and positioning of all surfaces with relative ease of tion, reflection (or diffraction in the case of a grating)
specification. at the surface S.
(3) Diffraction gratings generated by linear or con- (IV) Transform the new ray-point coordinates and
centric circular ruling motions on any of the allowed direction cosines back to the (XYZ) system (optional).
surface types must be accommodated with provision for (V) Repeat I through IV for succeeding surfaces in
specifying a variable ruling separation. sequence.
(4) The procedure should be capable of extension to Step IV is obviously omitted if the coordinate system
cover new surface types or new modes of ruling without associated with the next surface is referred to the system
major modification. I associated with S instead of to the reference system.
In addition, the computing formulas have been We now proceed to treat each of the foregoing steps
tailored for digital-computer use. For this reason, as in detail.
well as for conceptual simplicity, an algebraic, rather
I. Transformation to the Local
than a trigonometric, approach has been adopted.
Coordinate System of S
DIVISION OF THE PROBLEM The orientation of the local system (X,Y,Z) may be
specified with respect to the reference system in a
A given ray is specified by the coordinates (X0 ,Yo,Zo) variety of ways. We
have chosen a specification in
of a point P through which the ray passes and by its
terms of Euler angles. Figure 2 shows the generation of
direction cosines (k,l,Th)in a reference coordinate system the (X, Y,Z)
system from the (X,Y,Z) system by three
(X,Y,Z) having its origin at a point 0 (see Fig. 1). A successive
rotations a, fi, and T. The complete rotation
surface S is specified by an equation matrix is the product of matrices representing the
F(X,Y,Z)=O, (1) individual rotations and is

cos'y -siny 0 1 0 o iFCosa 0 -sina

R= siny cosy 0 0 cost -sin:O 1 0
0 0 1, 0 sin: cosJ sina 0 cosa
(cosa cosoy+sina sink siny) -cosg sin-y (-sina cosy+ cosa sin3 sin-y) (2)
= (cosa siny-sina sin: cos-y) cosf cos'y (-sina siny-cosa sin3 cosy) .
sina coso sin: cosa cosJ
All angles shown in Fig. 2 are positive.
2 W. Allen and J. Snyder, J. Opt. Soc. Am. 42, 243 (1952).
3 A. Murray, J. Opt. Soc. Am. 44, 672 (1954).
4M. Herzberger, Modern Geometrical Optics (Interscience Publishers, Inc., New York, 1958), Chap. V.
I G. Rosendahl, J. Opt. Soc. Am. 51, 1 (1961).
6 H. Yoshinaga, B. Okazaki, and S. Tatsuoka, J. Opt. Soc. Am. 50, 437 (1960).
674 G. H. SPENCFR AND M. V. R. K. MURTY Vol. 52

V.Y. -
X
Y' ZW
zz and where
Z,
F'(Xj, Yj,Zj)= dF/ds(s_,i)
i = (F)jk+ (F)jl+ (F)jm. (11)
We have used the notation (F )j to represent aF/OX
FIG. 2. Generation of (X,Y,Z) system from (XYZ) system evaluated at (XjYjZj), and similarly for the deriva-
after translation of origin to 0. (X,Y,Z) - (X',Y',Z') by rota- tives with respect to Y and Z. The process may be
tion a; (X',Y',Z') - (X",Y",Z") by rotation fl; (X",Y",Z") -
(X, Y,Z) by rotation y. started from a first approximation
s1= 0 (12)
The coordinates of the ray point Po in the (XY Z)
system will be denoted by (X0 ,YoZ o), and the ray and is terminated with the value sf for which
direction cosines in this system will be (k,l,m). If the
coordinates of 0 are (xtyo,), as measured in the Isf-sf-1 < , (13)
reference system, then the transformation equations for
the ray data are where e is a small preassigned tolerance whose value
depends on the degree of accuracy required.
Yo =R |Yo-yoJ (3) The choice of a starting point at the Z= 0 plane will
ordinarily ensure convergence of the process represented
and by Eq. (9) for the forms of the surface equation which
we shall define. There are, however, circumstances
[ilJl=R(Z] 0. (4) under which the process will break down or converge to
an incorrect value. If F'-- 0 at any point in the process,
Eq. (9) will either become indeterminate or will result
II. Determination of the Ray Intersection in s - oo. In particular, Eq. (9) will become indeter-
with S minate if the ray meets the surface at grazing incidence.
The parametric equations of the given ray may be It is also possible for F'-- 0 under other circumstances
written if the surface is nonspherical, but the likelihood of such
X=Xo+ks, Y=Ya+s, Z=ZO+ms, (5) a situation occurring in practice is fortunately slight.
When the ray, in fact, fails to intersect the surface at
where s is the parameter of distance along the ray all, the process will naturally diverge and will usually
measured from (X0 ,YoZo). result in an oscillating sequence of values for s with the
It will be convenient to determine first the inter- possibility of s ->oo at some point.
section of the ray with the Z=O plane. Denoting the It is further possible for a ray to intersect a surface
required value of s by so, and the resulting values of X at several points, and the first approximation (12) may,
and Y by X 1 and YI, we have from Eqs. (5) in such circumstances, result in convergence to an
incorrect point. Again, however, the situation is not
s0= -Zo/in, (6) likely to be encountered in practice.
XI=Xo+kso, Y 1 = Yo+lso. (7) The total path along the ray from (XoYoZo) to the
surface is
With the understanding that s now be measured from P= So+Sf. (14)
(X1 ,Y1 ,0), we may replace Eqs. (5) by
While this quantity is not directly needed for the trace,
X=Xi+ks, Y=Y1+ls, Z=ms. (8) it provides a quick check on the nature of the ray path.
For a true ray path, p must always be positive. A
Our problem now is to find a value of s such that the negative value for p indicates a virtual path, i.e., an
resulting values (X,Y,Z) from Eqs. (8) satisfy the extension along the line of the ray in the direction
surface Eq. (1). We might substitute Eqs. (8) directly opposite to the direction of propagation.
in Eq. (1) and attempt to solve the resulting equation Having completed the iteration, we have the inter-
for s. This can be accomplished directly, however, only section coordinates (Xf,Yf,Zf). In addition, we have
in special cases. Instead, we shall apply the Newton- available a set of direction numbers for the surface
Raphson iteration technique. Using the subscript j to normal at the intersection point. These will be needed
denote the iteration number, we write, in step III, and are given by
sj+1= sj-F(Xj, Yj,Zj)/F' (Xj,Yj,Zj) (9) K=(F,.)f, L=(r,,,)f, M=(F.-)f. (15)
where from Eqs. (8),
We now turn our attention to the form of the surface
Xj= Xl+ksi, Yj= Yi+lsj, Zj=ms3, (10) equation and its derivatives for various surface types.
 June 1962 GENERAL RAY-TRACING PROCEDURE 675

TABLE I. Surface obtained for various values of K. AXIS OF REVOLUTION

Range or value of Type or surface

K<O hyperboloid
K=0 paraboloid FIG. 4. Generation
X /CIRCULAR PROFILEIN
0 <K< 1 - u PLANEY* CONST
hemelipsoid of revolution of toric surface.
about major axis III
K =1 hemisphere
II
K>1 hemelipsoid of revolution
about minor axis Z-fIy)

Rotationally Symmetric Surfaces giving c a large value and setting

All of the rotationally symmetric surfaces encountered K= -tan2O. (22)
in optical design may be represented by
The error in this representation satisfies the following
N conditions:
F(X,Y,Z) = Z- 2 2
p /[1+ (1 -Kcp )i]-E aip 2
i= 0, (16)
i=1 error in F < IKC -l, (23)
2 2
error in dF/dp < 21C
where p2 = X 2+ Y 2 . In most cases, N need not exceed 5. p -, (24)
If the terms of the summation are omitted, Eq. (16) Finally, it is worthwhile noting that when the terms
represents revolved conic sections. The vertex curvature of the summation in Eq. (16) are omitted, we may
of these surfaces is c, and the type of surface is deter- avoid taking a square root by writing the surface
mined by the value of K. Table I indicates the type of equation in the simpler form
surface obtained for various values of K:
Figure 3 clarifies the relationship between c, K, and F=Z-2'c(p2+KZ2)=o. (25)
the type of surface represented. Note that when K>O This is essentially the form used by Herzberger in his 5
the surface Eq. (16) is limited to a range treatment of conics of revolution. For this case, we have
p2 < 1/ (KC2). (17) F,=-cX, F=-cY, F =1-KCZ. (26)
The derivatives associated with Eq. (16) are needed,
and these are Torics and Cylinders

F. =-XE, F =-YE, F = 1, (18) In the last section, we considered surfaces of revolu-

where tion about the Z axis. Here, we consider surfaces
N
generated by revolving a curve contained in the YZ
2 p 2 )+2
E=c/(-KC E jajp2(bl) (19) plane about an axis parallel to the Y axis, which inter-
i=1 sects the Z axis at a distance R from the origin (see
Fig. 4). We shall let the equation of the curve in the
Cones of revolution, or axicons, may be represented YZ plane be
by Z=f(Y). (27)
F=Z-p/tan6=O, (20)
The profile of the surface in a plane parallel to the XZ
where 0 is the half-angle of the cone. The associated plane cutting the surface at a height Y will be a circle
derivatives are, then, of radius R-f(Y) with center at Z=R, having the
Fz=-X/ptan0, F=-Y/p tan, F=1. (21) equation
X2 +(Z-R) 2 = [R-f(Y)]2, (28)
It should be mentioned that Eq. (20) may be closely or
approximated by the hyperboloid form of Eq. (16) by 2 _f2(y)].
Z= f(Y)+ (1/2R)[X2+Z (29)

p
Defining the curvature of revolution by
K-co
cR= 1/R, (30)
we may thus write
Ia
h-b-
F(X,Y,Z)=Z-f(Y) -2CR[X2+Z2- f2(y)I= 0, (31)
. b)
K aZ/bZ
with derivatives
C (b)' I Fz= -cRX,
FIG. 3. Relationship of K and c to geometrical parameters of the F8= EcR(Y)-Z. (dldY), (32)
conics represented by Eq. (25). The dashed lines in the figure for
g<O represent asymptotes. Fz=1-CRZ.
676 G. H. SPENCER AND M. V. R. K. MURTY Vol. 52

We may take a form for f(Y) suggested by Eq. (16): It is necessary to choose a first approximation r which
N
will ensure convergence to the physically correct root of
f (Y) = cY 2 /E1+ (1 -C 2
Y2) ]+ aY2i. (33) Eq. (38). A suitable value is
j=1
ri=-b/2a. (44)
The derivative needed in the second of the Eqs. (32)
This approximation is seen to break down as a -- 0.
is thus However, this corresponds to the case of grazing inci-
N
df/d Y=cY/ (1-Kc2Y2)i+2E jajY (- ). (34) dence and, hence, would have resulted in a breakdown
2 1
i=1 in II. The condition under which Eq. (38) fails to have
a real root is b>a2 , corresponding to total internal
An aspherized cylinder with axis parallel to the X reflection.
axs is obtained by setting CR =O, while a circular
cylinder with axis parallel to the Y axis is obtained by Reflection
setting f(Y)=0.
At a reflecting surface we may use. the foregoing
III. Change in Direction of the Ray after equations with u= 1 and the other root of Eq. (38). In
Refraction, Reflection, or Diffraction this case, b = 0, and we may immediately write
Refraction r=-2a. (45)
Snell's law of refraction is most conveniently ex- Equations (37) thus become
pressed in the vector form
k'=k-2aK, '=l-2aL, m'=m-2aM. (46)
N'S'x r=NS x r. (35)
Here S is a unit vector in the incident-ray direction with Diffraction
components (k,l,m), and S' is a unit vector in the In view of the fact that we have to trace rays in all
refracted-ray direction with components (k',l',m'); r is
possible orientations, the usual diffraction equation for
a vector normal to the refracting surface at the ray the principal section of the grating is inadequate. The
intersection point and has components (K,L,M); N is general diffraction equations for a plane grating were
the refractive index of the medium in which the ray is used in scalar form by, among others, Minkowski in
incident and N' is the index of the medium into which connection with the curvature of spectral lines of
the ray is refracted. grating spectrographs, by Toraldo di Francia 9 in con-
It follows from Eq. (35) that nection with the theory of the Ronchi test, and by
0
S'=,,S+rr, (36) Guild in connection with the formation of moir6 fringes
by two crossed-diffraction gratings. The scalar diffrac-
where .=N'N' and r is an undetermined multiplier. tion equations may be conveniently put in a vectorial
The component equations of (36) are form as indicated below.
k'=Puk+rK, '= ul+rL, m'=Am+rM. . (37) Consider a plane grating having parallel linear rulings
with separation d. Let us associate with the grating a
Squaring and adding Eqs. (37) we obtain a quadratic right triple of unit vectors, p, q, and r. Let r be normal
in r, to the grating and q parallel to the rulings; p will then
r2+2ar+b=O, (38) be normal to the rulings. Let a beam of light be incident
in which on the grating in a direction defined by the unit vector
a=,u(kK+IL+mM)/(K+L +M ), 2 2 (39) S. It may be shown that the diffracted beam has a
maximum of intensity in the direction of the unit vector
and
b(=(-1)/(K 2 +L 2 +M 2 ). (40)
S', satisfying the relation

As suggested by Herzberger,7 Eq. (38) may be so lved S' x r=uS x r+ (nX/N'd)q, (47)
by the Newton-Raphson iteration technique. We vrite where n is the integral order number, X the vacuum
v(r)=r 2+2ar+b, (41) wavelength, and and N' are as previously defined.
and use Note that, for the zero order, Eq. (47) becomes Snell's
r (+P=
rn)- (rn)/v (rn)- (42) Law.
The following geometrical interpretation of the gener-
Now,
r (r.,)=2(r..+a), alized diffraction relation may be of interest. This is the

so Eq. (42) becomes 8 R. Minkowski, Astrophys. J. 96, 305 (1942).

9 T.di Francia, Contributed article on the Ronchi Test, "Optical
r_,,,+=(rn2 -b)/2(rF+a). (43) image evaluation," NBS Circ. 526, 165 (1954).
10J. Guild, The Interference Systems of Crossed Diffraction
7 M. Herzberger, J. Opt. Soc. Am. 41, 805 (1951). Gratings (Oxford University Press, New York, 1956), p. 23.
June 1962 GENERAL RAY-TRACING PROCEDURE 677

fact that all orders of diffraction lie on a cone with ruling Y

direction as the axis and the semi-apex angle equal to

the angle made by the zero-order ray with the direction
of the rulings.
Now the geometrical-optics approximation allows the
laws governing the propagation of a plane wave across
an infinite plane boundary to be applied to the propaga-
tion of a vanishingly thin pencil of light-the ray-
-Z
across a local region of a nonplanar boundary. Thus we
shall use Eq. (47) as the law determining the direction
of the diffracted ray at a point on a grating surface

JJ
whose shape corresponds to any of the previously given Case 1.
forms of Eq. (1). It will be necessary to determine the
local value of d and the local direction of the rulings.
First, let us derive a set of component equations for ~~~~~~~~
S' analogous to Eqs. (37). For convenience, we define
A= nXIN'd. (48)
Then, substituting q= - p x r in Eq. (47), we obtain
Case H.
(S'-yS+Ap) x r=0, FIG. 5. Mode of grating ruling generation. In Case I, the rulings
from which are defined by the surface intersections with parallel planes. In
S'=yS-Ap+Ir. Case II, the rulings are defined by the intersections with concen-
tric cylinders.
Now, r need not have unit magnitude in Eq. (49). The
value obtained for the multiplier r will effectively com- root is of the form given in the discussion of refraction:
pensate for a nonunit magnitude of r. Thus, we may b'>a2 .
take r to have the components (K,L,M) as determined It is now necessary to determine the local grating
in II. Denoting the components of p by (u,v,w)we may space d (from which A is computed), and the com-
write the component equations of (49): ponents of p. These quantities depend on the way in
which the grating rulings are generated, and we shall
k'=,uk-Au+rK, consider two cases of interest (see Fig. 5).
l'=iil-Av+rL, (50) Case I. Surface intersectionswith parallel planes. Here,
m=m-Aw+rfM. the grating rulings are defined by the intersections with
the surface of a family of parallel planes. The planes are
Squaring and adding Eqs. (50) and making use of the taken to be parallel to the YZ plane and the separation
fact that p-r=0, we obtain a quadratic equation in r: of adjacent planes, d, is assumed given by a relation of
r 2 +2ar+b'=o, (51) the form
d,= g(X). (56)
which differs from Eq. (38) only in the value of the last
term, given by In most cases, d will be constant, but in special situa-
2 2 2 2 tions it may be desirable to use a variable spacing. In
b'= [j2- 1+A - 2,.A(kA+lv+mw)]/(K +L +M ). particular, the effects of a periodic ruling error might be
(52) studied by including sinusoidal terms in Eq. (56).
The iteration formula (43) may thus be used to deter- Now, q has no component in the X direction, since it
mine r1: is directed along a grating line which lies in a plane
r,,+,= (rn2- b')12(r,+ a). (53) parallel to the YZ plane. Hence,
For a transmission grating, the root of Eq. (51) having q=- (px r).=Lw-Mv=O. (57)
the smaller magnitude is required. In this case a suitable Also,
first approximation is p r=Ku+Lv+Mw=0, (58)
and
ri=-b'/2a. (54) p- p=u 2 +v+w 2
= 1. (59)
For the case of a reflection grating, the larger root of
Eq. (51) must be taken, and convergence to this root is From Eqs. (57)-(59), we obtain
assured by the first approximation u= 1/[1+Ks/(L2+M2)]1,
r = b'/2a- 2a. (55) v= -KLu/(L+M 2
), (60)
The condition under which Eq. (51) fails to have a real w= -KMu/ (L2 +M2 ).
678 G. H. SPENCER AND M. V. R. K. MURTY Vol. 52

The local grating space is not given directly by Eq. (56) The local grating space is determined in the same
since the generating planes intersect the surface fashion as before:
obliquely. Instead, it is given by d= d,/i- p, where i is d=d/9- p,
the unit normal to the generating planes. Thus,
where is the unit normal to the generating cylinders
d=d,/u=g(X)/u. (61) at the point of interest. Thus, we obtain for the grating
space
Case 1I. Surface intersectionswith concentriccylinders. d= pdr/(Xu+ Yv) =pg(p)/(Xu+ Yv). (66)
In this case, the grating rulings are defined by the
intersections with the surface of a family of concentric IV. Transformation Back to the Reference
circular cylinders centered on the Z axis. The radial System
separation of adjacent cylinders may be taken to be
The new ray-point coordinates (X, Y,Z) and direction

Z+
dr = g(p), (62) cosines (k',l',m') may be transformed to the reference
where system using the inverse forms of Eqs. (3) and (4).
p= (X2+Y2). Using barred symbols to represent coordinates and
cosines in the reference system, we have
Here, we make use of the fact that p is constant along
a ruling. This leads to Xq,+ Yq,,= 0, from which rX 'X+ton
Y =R-' Y+y-oJ (67)
X(Lw-Mv)+ Y(Mu-Kw) =O. (63)
= R
Equations (58), (59), and (63) yield and

u= [M2 X+L(LX-KY)]/G, I' = R- ' . (68)

v=[M 2Y-K(LX-KY)]/G, (64)
w= -M(KX+LY)/G, Now, R is a product of unitary matrices and is, there-
where fore, itself unitary. Since a unitary matrix has the
property that its inverse equals its transpose, we may
G= {(K 2+L2 +M2 )EM2 p2 + (LX-KY)']} i. (65) write
-

F (cosa cosoy+sina sin3 siny) (cosa siny - sina sink cosy) sine costly
R-' = - cost siny cos3 cosy sink l (69)
(-sina cosy+cosa sino siny) (- sina siny - cosa sink3cosy) cosa cosigi

In some situations, particularly when large element
separations are involved, it may be desirable from the
standpoint of accuracy to retranslate only the X and Y
coordinates, setting 2o=0 in Eq. (67). This establishes
a new reference system whose axes are mutually parallel
to those of the old system, whose Z axis is collinear with
that of the old system, and whose origin is at the point
(0,0,2o) as measured in the old system.
In other situations, it may be more convenient to
specify the orientation and position of each surface with
respect to the local coordinate system of the preceding
surface. In these cases, Eqs. (67) and (68) are omitted
from the procedure.
ACKNOWLEDGMENTS
The authors are indebted to Professor R. E. Hopkins
for his continued interest and encouragement.