Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Imaging with four spherical mirrors

Joseph M. Howard and Bryan D. Stone

We investigate unobstructed, plane-symmetric imaging systems of four spherical mirrors. Fifteen pa-
rameters are necessary to specify the configuration of such a system. Constraints are determined that
ensure that any resultant system possesses a given set of first-order properties. These constraints
remove four parameters as available degrees of freedom. To illustrate the efficacy of this design
approach, we present two example studies: one for a class of systems with the object at infinity and
another for finite-conjugate projection systems. For each study a global optimizer is used as the primary
search tool. Example systems from these studies are presented. © 2000 Optical Society of America
OCIS codes: 080.2740, 220.2740, 080.3620, 220.3620, 230.4040.

1. Introduction This ray is taken to be approximately at the center of


Design methods are presented here that facilitate the bundle of rays that pass through the system.
studies of unobstructed imaging systems composed of Fifteen construction parameters are associated with
four spherical mirrors. As background, previously the configuration of a plane-symmetric four-
published four-mirror studies have typically assumed spherical-mirror system. By considering a first-
an axis of symmetry in their design methods and order expansion of the imaging properties about the
often used Seidel aberration theory. Examples of base ray, one can constrain four construction param-
this approach can be found in a book by Korsch.1 eters to ensure a given set of first-order properties.
Another published approach is to solve numerically When the object is at infinity, two parameters 共the
for the surface figures of the mirrors such that the object location and the object tilt兲 are predetermined.
resulting system is aplanatic. This paradigm was Thus eleven degrees of freedom remain for the finite-
originally introduced by Schwarszchild2 and was conjugate case; only nine, for the case when the object
later generalized by Wasserman and Wolf.3 Such an is at infinity. When the object and the image tilts
approach generally requires the use of at least two must be chosen for reasons other than image quality,
aspheric surfaces. Yet another technique is to con- the available degrees of freedom are further reduced
catenate two well-corrected systems and then locally by two 共to nine兲 for the finite-conjugate case or by one
optimize the resultant system of four 共or more兲 mir- 共to eight兲 when the object is at infinity. The methods
rors. Examples of systems that use only spherical for developing these imaging constraints are dis-
mirrors can be found in a paper by Shafer.4 In the cussed in Section 2. It turns out that the constraint
study presented here, we assume plane symmetry, equations can be put in a form that can be applied
consider only spherical mirrors, and provide design more generally than just for the four-mirror systems
tools for investigating all feasible configurations of considered here. To avoid impeding the flow of Sec-
such four-spherical-mirror imaging systems.
tion 2, we give these equations, and discuss their
The configuration of the systems in this study is
application to more-general cases, in Appendix A.
specified with respect to a single ray, the base ray.
With this reduction in dimensionality of the con-
figuration space, one can much more effectively
search the remaining degrees of freedom for viable
When this research was conducted, J. M. Howard was with The configurations. In Section 3, systems with the object
Institute of Optics, University of Rochester, Rochester, New York at infinity that operate at a speed of f兾10 are studied.
14627; he is now with the Optics Branch, Code 551, NASA Goddard We use a global optimization routine to search for
Space Flight Center, Greenbelt, Maryland 20771. B. D. Stone is
unobstructed systems over eight degrees of freedom.
with Tropel Corporation, 60 O’Connor Road, Fairport, New York
14450. In Section 4 a similar search is performed for projec-
Received 1 October 1999. tion systems with a magnification of four. Example
0003-6935兾00兾193232-11$15.00兾0 systems are presented in both Sections 3 and 4.
© 2000 Optical Society of America Concluding remarks are offered in Section 5.

3232 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000


simpler with the Hamiltonian approach, but the dif-
ference is slight 共and modern computer algebra pack-
ages mitigate this advantage兲. Therefore, to
emphasize further that the two approaches are equiv-
alent 共and to illustrate their use further兲, we use
matrix methods here to derive imaging constraints
for four spherical mirrors.
In the following discussion, quantities associated
with the object are shown without modification and a
quantity associated with the image is distinguished
by the addition of a prime. To begin, let y collec-
tively represent the y and z coordinates of the point of
intersection of a ray with the object and let yⴕ simi-
larly represent ray coordinates at the image. Let p
denote the y and z optical direction cosines, py and pz,
Fig. 1. Schematic representation of a plane-symmetric system
of a ray in object space and pⴕ be defined analogously
composed of four spherical mirrors. The cross section in the plane
of symmetry is shown. The base ray and tilted object and image
for image space.
planes are also illustrated. The generalized derivative matrix for a system re-
lates the final configuration of a ray to the initial
configuration as follows:
2. Four Spherical Mirrors: Imaging Constraints

A. Fifteen Parameters To Specify a System


冉 冊 冋 册冉 冊
yⴕ
pⴕ

a b y
c d p
⫹ O共2兲. (1)

Before deriving the imaging constraints, we first in- O共2兲 represents all terms of order two and higher in
troduce the construction parameters used here to y and p 共because we have broken symmetry, the
specify a four-mirror system. The cross section in even-order terms do not necessarily vanish兲. The
the plane of symmetry is illustrated in Fig. 1, along elements a, b, c, and d in Eq. 共1兲 are each 2 ⫻ 2
with the base ray. The general notation for the con- diagonal submatrices, and these elements character-
struction parameters, sign conventions, and arrange- ize the first-order properties of the system. That the
ment of local coordinate systems are identical to off-diagonals vanish is a consequence of plane sym-
those presented in an accompanying paper on three metry 共as an aside, the diagonal elements are equal
spherical mirrors5 and are not repeated here. for systems in which the base ray coincides with an
The displacement along the base ray from the basal axis of symmetry兲. The elements a and b are of
object point to the first mirror is denoted d0. The particular significance in the following discussion,
separations between the mirrors are d1, d2, and d3, and the two nonzero elements are labeled with sub-
respectively, and the final displacement along the scripts in the following manner:
base ray to the image is d4. Only positive values of
d1, d2, and d3 are considered, because any system
with a negative separation between mirrors is un-
a⫽ 冋 a11 0
0 a22册, b⫽ 冋 b11 0
0 b22册. (2)

physical. A negative value of d0 or d4 corresponds to One can determine the derivative matrix for any sys-
a virtual object or a virtual image, respectively. The tem by multiplying together a series of transfer and
angles of incidence of the base ray with mirrors 1, 2, reflection matrices. For the four-mirror systems of
3, and 4 are denoted ␪1, ␪2, ␪3, and ␪4, respectively. interest here, this process can be represented as fol-
␪obj and ␪im denote the tilts of the object surface and lows:

冋 册
the image surface. Finally, the curvatures of the
mirrors are written as c1, c2, c3, and c4. The fifteen a b
quantities d0, d1, d2, d3, d4, ␪obj, ␪1, ␪2, ␪3, ␪4, ␪im, c1, ⫽ 共T4R4兲共T3R3兲共T2R2兲共T1R1兲T0, (3)
c d
c2, c3, and c4 completely specify the configuration of a
plane-symmetric system of four spherical mirrors. where the transfer matrices Ti 共i ⫽ 0, 1, 2, 3, 4兲 and
Note that, when the object is at infinity, two of these the reflection matrices Ri 共i ⫽ 1, 2, 3, 4兲 are of the
parameters are determined: d0 ⫽ ⬁ and ␪obj ⫽ 0. form6

冤 冥
B. First-Order Properties
cos ␪i sec ␪i⫹1 0 di sec ␪i sec ␪i⫹1 0
0 1 0 di
In the accompanying paper on three spherical mir- Ti ⫽ , (4a)
rors,5 constraints are derived to ensure that any re- 0 0 cos ␪i⫹1 sec ␪i 0
sultant system possesses a desired set of first-order 0 0 0 1
properties. The methods of Hamiltonian optics were

冤 冥
used in the body of that work, but an alternative 1 0 0 0
approach using first-order matrix methods was dis- 0 1 0 0
Ri ⫽ . (4b)
cussed in an appendix. In comparing the two meth- 2ci cos ␪i 0 1 0
ods, we believe that the intermediate steps are 0 2ci cos ␪i 0 1

1 July 2000 兾 Vol. 39, No. 19 兾 APPLIED OPTICS 3233


Note that, when one is evaluating T0, ␪0 represents where G and H are defined as
␪obj; similarly for T4, ␪5 represents ␪im. For the case
when the object is at infinity, ␪0 is taken to be zero G共t1, t2, t3, t4兲 ⫽ 2兵c2 q2 t2 ⫹ c3共d3 ⫹ d4兲t3 ⫹ c4 d4
and d0 represents the separation from an arbitrarily ⫻ 共2c3 d3 t3 ⫹ 1兲t4 ⫹ c1 t1关2c2 q2 t2 d1
placed reference surface in object space to the first
mirror; it can be taken without loss of generality to be ⫹ d1 ⫹ d2 ⫹ q1 ⫹ 2c3共d1 ⫹ d2兲q1 t3
zero. ⫹ 2c4共d1 ⫹ d2兲t4 d4兴其 ⫹ 1, (10a)
C. Determination of Constraints
H共t1, t2, t3, t4兲 ⫽ d0 ⫹ d1 ⫹ d2 ⫹ d3 ⫹ d4 ⫹ 2兵c2共d0
The first-order imaging properties are completely
characterized by the 4 ⫻ 4 derivative matrix. The ⫹ d1兲q2 t2 ⫹ c3共d0 ⫹ d1 ⫹ d2兲
first-order blur and system magnifications 共or focal ⫻ 共d3 ⫹ d4兲t3 ⫹ c4 d4关d0 ⫹ d1 ⫹ d2 ⫹ d3
lengths when the object is at infinity兲 are of partic-
ular interest here. For a system operating at finite ⫹ 2c3共d0 ⫹ d1 ⫹ d2兲d3 t3兴t4
conjugates, first-order blur is absent only when the ⫹ c1 d0 t1关q2 ⫹ d1共2c2 q2 t2
diagonal elements of b vanish. This fact can be
seen from Eq. 共1兲, where for yⴕ to be solely a function ⫹ 2c3 q1 t3 ⫹ 2c4 d4 t4 ⫹ 1兲兴其 (10b)
of y 共and not of p as well兲 the elements of b must
equal zero. It follows that, when the elements of b and q1 and q2 in Eqs. 共10兲 are given by
vanish, the diagonal elements of a represent the
q1 ⫽ d3 ⫹ d4 ⫹ 2c4 d3 d4 t4, (11a)
principal magnifications 共in and out of the plane of
symmetry兲. These four conditions are represented q2 ⫽ q1 ⫹ d2 ⫹ 2c3 d2 q1 t3 ⫹ 2c4 d2 d4 t4. (11b)
as follows:
To obtain the first-order constraints for a finite-
b11 ⫽ 0, b22 ⫽ 0, (5) conjugate system we substitute the expressions for
the matrix elements in Eqs. 共9兲 into Eqs. 共5兲 and 共6兲
a11 ⫽ m1, a22 ⫽ m2, (6) and solve those equations for individual system pa-
rameters 共as a function of the remaining system pa-
rameters and m1 and m2兲. The following set was
where m1 and m2 in Eqs. 共6兲 represent the principal determined for a four-mirror system operating at fi-
magnifications in the plane of symmetry and trans- nite conjugates:
verse to the plane of symmetry, respectively.
If a similar argument is used when the object is at c3 ⫽ K1共c1, c2, d0, d1, d2, m1, m2, ␪obj, ␪1, ␪2, ␪3, ␪4, ␪im兲,
infinity, there will be no first-order blur when the
diagonal elements of a vanish. When this condition (12a)
is met, the diagonal elements of b represent the ef-
fective focal lengths for the system; this is equivalent d3 ⫽ K2共c1, c2, d0, d1, d2, m1, m2, ␪obj, ␪1, ␪2, ␪3, ␪4, ␪im兲,
to the following equations: (12b)

a11 ⫽ 0, a22 ⫽ 0, (7) c4 ⫽ K3共c1, c2, d0, d1, d2, m1, m2, ␪obj, ␪1, ␪2, ␪3, ␪4, ␪im兲,
(12c)
b11 ⫽ f1, b22 ⫽ f2, (8)
d4 ⫽ K4共c1, c2, d0, d1, d2, m1, m2, ␪obj, ␪1, ␪2, ␪3, ␪4, ␪im兲.
where f1 and f2 in Eqs. 共8兲 are the principal focal
lengths in and out of the plane of symmetry, respec- (12d)
tively. The actual forms of K1, K2, K3, and K4 appear in Eqs.
Expressions for a11, a22, b11, and b22 for a plane- 共A1兲 of Appendix A. Note that, although these ex-
symmetric system of four spherical mirrors follow pressions seem rather complicated, they are readily
from Eq. 共3兲: determined by use of modern computer algebra soft-
ware.
a11 ⫽ cos ␪obj sec ␪imG共sec ␪1, sec ␪2, sec ␪3, sec ␪4兲, When one is solving for the constraints, the choice
of which construction parameters to constrain is
(9a) somewhat arbitrary. Although we found that this
set was relatively straightforward to solve for, other
a22 ⫽ G共cos ␪1, cos ␪2, cos ␪3, cos␪4兲, (9b) sets exist 共e.g., one can also solve for 兵c1, d0, d4, cos
␪1其兲. Note that when a separation between mirrors
b11 ⫽ sec ␪obj sec ␪imH共sec ␪1, sec ␪2, sec ␪3, sec ␪4兲, is constrained 关such as d3 in Eq. 共12b兲兴, it is possible
that values for the independent parameters can be
(9c) found that yield a negative value for this separation.
The resultant system is then unphysical, and new
b22 ⫽ H共cos ␪1, cos ␪2, cos ␪3, cos ␪4兲, (9d) values for 共at least some of 兲 the independent param-

3234 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000


eters must be found if a physically realizable system tive Simulated Annealing 共ASA兲.7 Eight systems
is desired. are illustrated for each study.
When the constraints presented in Eqs. 共12兲 are For this first study we considered f兾10 systems
applied, all the resultant systems contain no first- with focal lengths f1 ⫽ ⫾100.0 and f2 ⫽ ⫾100.0 共there
order blur and possess magnifications of m1 and m2 in are four possible combinations of signs of the two
and out of the plane of symmetry, respectively. Ap- focal lengths兲. A square field that subtended 2.0° on
plying these constraints effectively removes those pa- a side was used. The stop was fixed at the first
rameters as available degrees of freedom to the mirror, and the image tilt was set to zero to allow for
optical designer, which in turn reduces the dimen- conventional rotationally symmetric optics, such as
sionality of the design space and significantly simpli- an eyepiece, to be added to the basic telescope design.
fies the process of determining viable design forms. The figure of merit for each system was the maximum
For the case when the object is at infinity, the root-mean-square 共RMS兲 spot radius at fifteen field
process is repeated but with Eqs. 共7兲 and 共8兲. The points arranged in a rectangular grid of five rows and
following set of constraints was determined: three columns covering half of the object. The first
column bisected the object along the plane of symme-
c3 ⫽ L1共c1, c2, d1, d2, f1, f2, ␪1, ␪2, ␪3, ␪4, ␪im兲, (13a) try 共i.e., along the Y axis兲, the middle was approxi-
mately two thirds of the distance 共65.5%兲 to the outer
d3 ⫽ L2共c1, c2, d1, d2, f1, f2, ␪1, ␪2, ␪3, ␪4, ␪im兲, (13b) edge, and the last was along the edge of the object.
The rows were spaced such that the first and the last
were at opposite edges of the object, the third bisected
c4 ⫽ L3共c1, c2, d1, d2, f1, f2, ␪1, ␪2, ␪3, ␪4, ␪im兲, (13c) the object 共along the Z axis兲, and the second and the
fourth were approximately two thirds of the distance
d4 ⫽ L4共c1, c2, d1, d2, f1, f2, ␪1, ␪2, ␪3, ␪4, ␪im兲. (13d) to the edges. As one would expect, the field point
with the most blur tended to be located at one of the
The forms of Li 共i ⫽ 1, 2, 3, 4兲 appear in Eqs. 共A4兲 of two corners of the image. Distortion was not penal-
Appendix A. When these constraints are applied, ized: The spot radii were calculated from the geo-
all resultant systems are free of first-order blur and metric center of the blur.
possess effective focal lengths of f1 and f2 共in and out The boundaries assigned to the degrees of freedom
of the plane of symmetry兲, respectively. were chosen as c1 ⫽ 关⫺0.1, 0.1兴, c2 ⫽ 关⫺0.1, 0.1兴, d1 ⫽
For the following two example studies, we consider 关1.0, 100.0兴, d2 ⫽ 关1.0, 100.0兴, ␪1 ⫽ 关0.2°, 80.0°兴, ␪2 ⫽
only nonanamorphic systems. Such systems result 关⫺80.0°, 80.0°兴, ␪3 ⫽ 关⫺80.0°, 80.0°兴, and ␪4 ⫽
when the magnitudes 共but not necessarily the signs兲 关⫺80.0°, 80.0°兴. Inasmuch as changing the signs of
of the two magnifications 共or focal lengths when the all the tilt angles does not change the overall system,
object is at infinity兲 are equal, that is, when 兩m1兩 ⫽ one of the tilt angles can be taken, without loss of
兩m2兩 共or 兩 f1兩 ⫽ 兩 f2兩兲. The sign of the magnification or generality, to be positive 共and we chose ␪1兲. ASA
focal length dictates whether there are an even or an was run for 100 iterations for each of the four com-
odd number of internal images. In general, the sys- binations of signs of the focal lengths and took ⬃50 h
tem can have a different number of internal images on a 266-MHz Pentium II machine. The maximum
for tangential rays than it does for sagittal rays. permitted obstruction during the search was 2% of
共Tangential rays are defined as those rays from the rays traced. Eight of the resultant systems are il-
basal object point that lie within the plane of sym- lustrated in Fig. 2, and their parameters are listed in
metry. Sagittal rays are those from the basal object Table 1. Note that these eight are not the only via-
point that initially lie in a plane that contains the ble systems found, but they were chosen to illustrate
base ray segment in object space and is perpendicular the wide range of systems that can be uncovered with
to the plane of symmetry.兲 For tangential rays, a this approach. In Fig. 2 the value for the figure of
four mirror system has an even number 共i.e., zero or merit is also listed next to each system to provide for
two兲 of internal images when m1 ⬍ 0 共or f1 ⬎ 0兲 and a quick comparison among the systems. As a point
an odd number 共i.e., one or three兲 when m1 ⬎ 0 共or of reference, the radius of the first zero in the Airy
f1 ⬍ 0兲. For systems with an even number of mirrors pattern for f兾10 systems with 0.55-␮m light is 6.71 ⫻
the sign convention is the same for sagittal rays: An 10⫺4.
even number of internal images occur when m2 ⬍ 0 As occurs with asymmetric systems, the internal
共or f2 ⬎ 0兲 and an odd number when m2 ⬎ 0 共or f2 ⬍ images for the tangential and the sagittal rays are
0兲. In summary, there are four possible sign combi- not generally coincident. In fact, for four spherical-
nations for m1 and m2 共or f1 and f2兲. In the following mirror systems that are free from first-order blur,
examples, each of these choices is considered. these images can coincide only when they fall be-
tween the second and third mirrors. This is so be-
3. Example 1: Object at Infinity cause a single spherical mirror cannot image a point
In this section the case when the object is at infinity without first-order blur,8 thus excluding the spaces
is considered, and in Section 4 projection systems are between the first and the second mirrors and between
investigated. In both examples, a search over the the third and fourth mirrors because the overall sys-
constrained four-mirror configuration space is per- tems are corrected to first order. Inasmuch as only
formed with the global optimization routine, Adap- tangential rays are illustrated in Fig. 2, we can see

1 July 2000 兾 Vol. 39, No. 19 兾 APPLIED OPTICS 3235


Fig. 2. Systems A–H were found as part of the search of the constrained configuration space. The number next to the images represent
the maximum RMS spot radii of the 15 field points evaluated. A scale is given next to each system. Systems A–C have the same scale,
as do D–H. The sequence of ⫹ and ⫺ after each system label corresponds to the sign of the tilt angle for each of the four mirrors.

the locations of the tangential images by examining system A, the internal image associated with the tan-
that figure. Systems A and B both have negative gential rays is located between the third and fourth
values for f1 and f2, which results in a single internal mirrors, and in system B it is between the second and
image for both tangential rays and sagittal rays. In third. In systems A and B, the internal images as-

3236 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000


Table 1. Parameters for the Systems Illustrated in Fig. 2

Parameter System A System B System C System D System E System F System G System H

␪1 共°兲 2.690 40.512 23.237 26.711 8.166 13.107 26.411 11.880


␪2 共°兲 15.427 ⫺12.151 51.389 9.176 ⫺18.360 ⫺35.143 ⫺24.102 ⫺6.402
␪3 共°兲 ⫺10.808 ⫺21.051 55.559 ⫺35.294 ⫺10.393 ⫺34.221 9.244 ⫺10.501
␪4 共°兲 ⫺4.853 ⫺14.423 31.040 43.244 ⫺5.973 ⫺13.498 8.433 22.955
d1 95.112 36.318 199.251 60.639 60.100 69.308 23.313 53.580
d2 37.889 179.167 39.163 30.437 49.679 27.187 29.541 83.589
d3 199.826 116.825 51.699 17.424 28.644 19.639 26.666 33.393
d4 76.105 50.330 24.562 41.920 30.874 30.457 84.783 33.210
c1 ⫺3.626 ⫻ 10⫺3 9.493 ⫻ 10⫺4 ⫺2.712 ⫻ 10⫺3 ⫺5.062 ⫻ 10⫺4 ⫺2.976 ⫻ 10⫺3 ⫺3.534 ⫻ 10⫺3 ⫺1.026 ⫻ 10⫺3 2.531 ⫻ 10⫺3
c2 1.540 ⫻ 10⫺2 ⫺4.425 ⫻ 10⫺3 ⫺1.790 ⫻ 10⫺2 ⫺6.203 ⫻ 10⫺3 2.021 ⫻ 10⫺3 3.131 ⫻ 10⫺3 2.267 ⫻ 10⫺3 ⫺4.585 ⫻ 10⫺3
c3 ⫺1.010 ⫻ 10⫺3 ⫺1.020 ⫻ 10⫺2 ⫺1.886 ⫻ 10⫺2 4.433 ⫻ 10⫺2 ⫺2.604 ⫻ 10⫺3 ⫺1.617 ⫻ 10⫺3 ⫺4.260 ⫻ 10⫺3 ⫺3.484 ⫻ 10⫺3
c4 ⫺1.038 ⫻ 10⫺4 ⫺1.187 ⫻ 10⫺3 ⫺2.611 ⫻ 10⫺2 ⫺2.639 ⫻ 10⫺3 ⫺6.752 ⫻ 10⫺3 ⫺7.938 ⫻ 10⫺3 2.114 ⫻ 10⫺3 2.421 ⫻ 10⫺3
f1 ⫺100.0 ⫺100.0 100.0 100.0 100.0 100.0 100.0 100.0
f2 ⫺100.0 ⫺100.0 ⫺100.0 100.0 100.0 100.0 100.0 100.0
Rms spot 4.735 ⫻ 10⫺4 1.014 ⫻ 10⫺3 4.535 ⫻ 10⫺4 9.179 ⫻ 10⫺4 4.143 ⫻ 10⫺4 1.058 ⫻ 10⫺3 1.068 ⫻ 10⫺3 8.325 ⫻ 10⫺4
size

sociated with the sagittal rays are located between other in mirror arrangement, as they both have
the same mirrors as the tangential ones but are concave– convex– concave– concave mirrors 共in that
slightly displaced. In system B, an image of the stop order兲 and are arranged with positive–negative–
is located between mirrors three and four, and, as negative–negative tilt angles. System E has the
with the internal images, the tangential and the sag- smallest maximum spot size of those in Fig. 2. Al-
ittal rays from the basal point on the stop are imaged though system F has a figure of merit more than
to points that are slightly displaced from each other. twice as large as system E has, system F has room to
System C also has a positive value for f1, but it has a grow. That is, the system can be made faster, ac-
negative value for f2, giving an even number of inter- commodate a wider field of view, or both without
nal images for tangential rays and an odd number for causing any ray obstruction 共system C also shares
sagittal rays. In this case, two tangential internal this quality兲. System G was is one of the most com-
images can be seen in the figure: one between the pact mirror arrangements found during the search.
first and the second mirrors and the other approxi- Although some rays, such as the rays passing
mately at the third mirror. A single sagittal image through the edges of the second and the third mir-
is located approximately at the second mirror. This rors, are visibly obstructed in this system, recall that
system is unique among those in Fig. 2 because it is a maximum of 2.0% of the rays traced were allowed to
composed strictly of concave mirrors. The existence be blocked. System H has an interesting arrange-
of internal images in systems A, B, and C makes ment in that it allows the first and the fourth mirrors
those systems much longer than the remaining sys- to be placed on either side of a single substrate. If
tems that have no internal images. For our search the back-to-back arrangement of the first and fourth
we found no particularly interesting systems in mirrors is deemed undesirable, d2 共an independent
which f1 was negative and f2 was positive, and we degree of freedom兲 can be slightly increased while the
therefore chose not to illustrate any such systems. constrained parameters are continuously updated.
The remaining systems in Fig. 2 all have negative This process is equivalent to taking a local walk in
values for f1 and f2. We believed that systems with the constrained configuration space and results in a
no internal images were the most interesting and system that has more working space for the two mir-
thus included five from this category. As one would rors with only a slight degradation in image quality.
expect, these systems were the most compact and This is an example of how one can modify a system
typically were among the top performers. Their per- slightly to achieve more-desirable packaging proper-
formance can be compared with those found in the ties.
accompanying paper on three spherical mirrors,5 but
one should note the difference between their respec- 4. Example 2: Finite Conjugates
tive figures of merit. In the three-mirror paper, only In the example presented in this section, we consider
the basal image point was evaluated, whereas for the finite-conjugate projection systems that are nonana-
four-mirror systems here the figure of merit is the morphic with a magnification of 4.0 and have a nu-
maximum spot radius for fifteen field points arranged merical aperture in object space of 0.1. The object
over a 2.0° field of view. chosen for this study is a 2 cm ⫻ 2 cm square, and the
System D has a concave first mirror, as do all the entrance pupil for these systems is set at infinity 共i.e.,
systems except B and H discussed in this section, and the systems are telecentric in the object space兲. Un-
it has its final beam nearly perpendicular to the in- like in the case when the object is at infinity, the
coming beam. Systems E and F are similar to each object and image planes were allowed to tilt to help to

1 July 2000 兾 Vol. 39, No. 19 兾 APPLIED OPTICS 3237


Fig. 3. Systems I–P were found in the constrained configuration space by use of ASA. The numbers next to the images represent the
maximum RMS spot radii of 15 field points evaluated across one half of a 2-cm-square object. Note that systems I, M, N, and P are drawn
with the same scale, as are J, K, and O. The sequence of ⫹ and ⫺ after each system label corresponds to the sign of the tilt angle for each
of the four mirrors.

correct for higher-order aberrations. The magnifi- metric center of the blur 共as opposed to the first-order
cations used were m1 ⫽ ⫾4.0 and m2 ⫽ ⫾4.0, and the image location兲. The boundaries assigned to the 11
figure of merit for each system was the maximum degrees of freedom for the optimization were c1 ⫽
RMS spot radius of 15 field points, arranged similarly 关⫺0.1, 0.1兴, c2 ⫽ 关⫺0.1, 0.1兴, d0 ⫽ 关1.0, 200.0兴, d1 ⫽
as described in Section 3. Distortion was not penal- 关1.0, 200.0兴, d2 ⫽ 关1.0, 200.0兴, ␪1 ⫽ 关0.2°, 80.0°兴, ␪2 ⫽
ized: The spot radii were calculated from the geo- 关⫺80.0°, 80.0°兴 ␪3 ⫽ 关⫺80.0°, 80.0°兴, ␪4 ⫽ 关⫺80.0°,

3238 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000


Table 2. Parameters for the Systems Illustrated in Fig. 3

Parameter System I System J System K System L System M System N System O System P

␪obj 共°兲 26.121 ⫺23.460 ⫺7.469 ⫺1.810 7.417 22.918 15.894 15.762
␪1 共°兲 6.784 1.746 21.767 4.655 13.148 24.066 10.504 3.023
␪2 共°兲 47.675 ⫺19.560 47.590 ⫺12.711 ⫺19.638 ⫺70.117 ⫺10.697 3.754
␪3 共°兲 12.407 6.611 41.385 7.284 5.043 ⫺1.768 2.308 ⫺6.340
␪4 共°兲 ⫺44.877 ⫺10.640 33.058 ⫺10.291 ⫺5.265 11.640 8.129 15.518
␪im 共°兲 ⫺5.607 ⫺4.847 ⫺20.066 0.393 ⫺0.061 ⫺0.107 0.731 ⫺1.226
d0 共cm兲 101.664 95.507 21.023 18.829 48.223 10.010 15.665 90.115
d1 共cm兲 133.271 69.091 38.041 17.314 37.004 12.561 36.327 83.118
d2 共cm兲 152.669 41.901 20.457 19.192 136.741 199.999 102.001 84.010
d3 共cm兲 81.009 20.477 67.876 21.101 118.382 152.754 154.292 84.670
d4 共cm兲 142.733 58.946 118.012 44.787 175.428 90.963 53.178 216.745
c1 共cm⫺1兲 ⫺8.022 ⫻ 10⫺3 ⫺1.183 ⫻ 10⫺2 ⫺3.418 ⫻ 10⫺2 ⫺3.071 ⫻ 10⫺2 ⫺9.399 ⫻ 10⫺3 ⫺1.336 ⫻ 10⫺2 ⫺2.279 ⫻ 10⫺2 ⫺6.659 ⫻ 10⫺3
c2 共cm⫺1兲 ⫺1.624 ⫻ 10⫺2 2.825 ⫻ 10⫺2 ⫺3.359 ⫻ 10⫺2 2.805 ⫻ 10⫺2 6.432 ⫻ 10⫺3 5.959 ⫻ 10⫺5 1.199 ⫻ 10⫺2 6.549 ⫻ 10⫺3
c3 共cm⫺1兲 ⫺8.513 ⫻ 10⫺3 ⫺2.623 ⫻ 10⫺2 ⫺3.927 ⫻ 10⫺2 ⫺2.726 ⫻ 10⫺2 ⫺5.477 ⫻ 10⫺3 ⫺5.054 ⫻ 10⫺3 ⫺6.713 ⫻ 10⫺3 ⫺6.394 ⫻ 10⫺3
c4 共cm⫺1兲 2.033 ⫻ 10⫺2 8.699 ⫻ 10⫺3 ⫺1.199 ⫻ 10⫺2 2.965 ⫻ 10⫺2 7.952 ⫻ 10⫺3 1.615 ⫻ 10⫺3 1.690 ⫻ 10⫺2 7.063 ⫻ 10⫺3
m1 4.0 4.0 ⫺4.0 ⫺4.0 ⫺4.0 ⫺4.0 ⫺4.0 ⫺4.0
m2 4.0 4.0 4.0 ⫺4.0 ⫺4.0 ⫺4.0 ⫺4.0 ⫺4.0
Rms spot 21.0 44.1 45.8 11.7 29.6 10.0 11.2 7.2
size 共␮m兲

80.0°兴, ␪obj ⫽ 关⫺40.0°, 40.0°兴, and ␪im ⫽ 关⫺40.0°, tems shown in Fig. 3, and it happens to be the
40.0°兴. We used ASA to determine one hundred sys- most-compact arrangement. System M has a sim-
tems for each of the four combinations of signs of the ilar configuration but is much larger and has more
magnification, taking ⬃70 h. Eight of the resultant working space for the object than does L. Often-
systems are illustrated in Fig. 3, and their parame- times a system that results from global optimiza-
ters are listed in Table 2. The figure of merit is also tion does not have a desirable layout. Systems L,
listed next to each system. As a point of reference, O, and P, for example, have little working space for
the diffraction-limited spot radius for these systems the object or image because one or the other is
is approximately 13.4 ␮m 共for light with a wave- placed close to a nearby mirror. As we discussed
length of 0.55 ␮m兲. briefly in Section 3, one can alter such systems by
Systems I and J have positive values for both m1 changing the independent degrees of freedom by
and m2, and, for system K, m1 is negative and m2 is small amounts while updating the constrained pa-
positive. The remaining systems all have negative rameters to keep the system corrected to first order.
values for m1 and m2. As in Section 3, we found no Once a more-desirable layout has been achieved,
interesting systems with an odd number of internal the system can be locally optimized to maximize
images for tangential rays and an even number for performance.
sagittal rays 共i.e., with m1 ⫽ 4.0 and m2 ⫽ ⫺4.0兲. In Systems N and O have similar mirror arrange-
system I the internal image for tangential rays is
ments when they are compared with each other, but
located just after the second mirror and the image for
the tilt angles of the third mirror are of opposite
sagittal rays is only slightly displaced from it. Sys-
signs. This implies that these two systems are in
tem J also has a single internal image for both tan-
distinct minima separated by a region of obstructed
gential and sagittal rays, and they can be found
between the second and third mirrors. This system systems. In other words, for the optimizer to
is a good example of a configuration in which the travel from system N to O it must cross through a
centers of curvature of all four mirrors are almost region in the configuration space where the rays to
collinear. This implies that that system is 共nearly兲 and from the third mirror would be obstructed.
an off-axis portion of an axially symmetric system. This simple argument shows that the merit-
As such, system J is illustrated with a dashed line function space is disconnected and, left as such,
representing an approximate axis of symmetry. requires that a true global optimizer be used.
Note that this result was driven solely by the imag- However, a wandering-type global optimizer can be
ery of the system: No attempt was made to look used if these hard boundaries in the configuration
specifically for a system with an underlying axis of space are removed. One strategy for doing this is
symmetry. Unlike system J, system K is highly to penalize the merit function, say, by adding a term
asymmetric, and it is unique in Fig. 3 in that all its proportional to the percentage of rays obstructed.
mirrors are concave. In this way it is comparable This generally results in an upward slope to the
with system C, which has similar constraints on the merit function as a mirror tilt approaches zero and
number of internal images owing to the signs of the then a downward slope as fewer of the rays are
focal lengths. obstructed after the sign of the tilt changes. An-
System L has the best figure of merit of all sys- other source of discontinuities that also must be

1 July 2000 兾 Vol. 39, No. 19 兾 APPLIED OPTICS 3239


considered is the constrained parameter d3. When configuration space would have been able to be re-
this mirror separation turns out to be negative, the duced even further. Even without inclusion of
system is unphysical. To make the merit-function such higher-order constraints, the advantage of ap-
space continuous when d3 is negative one can add a plying the first-order constraints cannot be over-
term proportional to its magnitude to the merit stated. We made a comparison of the results
function. This method of smoothing the topogra- presented here and those that we found by perform-
phy of the configuration space and making it con- ing a search for systems with the object at infinity
tinuous should enable a wandering-type global but with none of the first-order imaging constraints
optimizer also to be used for these searches. imposed. The deviation from the desired focal
System P has an axis of symmetry similar to that of lengths was added to the figure of merit as a pen-
system J but with an added twist. The first and alty, and no attempt was made a priori to correct for
third mirrors are nearly concentric and of the same the first-order blur. The resultant ASA runs con-
curvature, implying that a single mirror could be sumed much more CPU time and failed to yield the
used for the both of them. The overlap of the rays variety of good systems found when the constraints
between these two nearly concentric mirrors in sys- were employed. We believe that the approach that
tem P would be inconsequential if the design were we have developed to the design of unobstructed
adjusted such that a single mirror were used for both mirror systems gives the best chance for finding
reflections. optimal solutions.
5. Concluding Remarks Appendix A. Constraint Equations for Four Spherical
These methods could be extended to encompass mul- Mirrors
tireflection imaging systems similar to the configura- The following equations serve to define Ki 共i ⫽ 1, 2, 3,
tion of system P. In such systems, the base ray 4兲 that appear in Eqs. 共12兲:

cos ␪3共m2d22 ⫺ sin2 ␪4 ⫺ m1d11 cos2 ␪4 cos ␪im兲


c3 ⫽ ⫺ ,
2共m2b22 cos2 ␪3 ⫺ m1b11 cos2 ␪4 cos ␪im兲 (A1a)

共m2b22 ⫺ m1b11 cos ␪im兲共m2b22 cos2 ␪3 ⫺ m1b11 cos2 ␪4 cos ␪im兲


d3 ⫽ , (A1b)
sin ␪4关m2b22 cos2 ␪3共m1d11 cos ␪im ⫺ 1兲 ⫺ m1b11 cos ␪im共m2d22 ⫺ 1兲兴
2

cos ␪4共m1d11 ⫹ 2m1b11c3 sec ␪3 ⫺ sec ␪im兲


c4 ⫽ ⫺ , (A1c)
2m1关d11d3 ⫹ b11共1 ⫹ 2c3 d3 sec ␪3兲兴
d4 ⫽ ⫺m1 cos ␪im关d11d3 ⫹ b11共1 ⫹ 2c3 d3 sec ␪3兲兴. (A1d)

would reflect off an individual mirror twice 共or more兲, It is understood that c3 and d3 that appear in Eqs.
as occurs, for example, in an Offner relay. The mir- 共A1c兲 and 共A1d兲 are given by Eqs. 共A1a兲 and 共A1b兲.
ror with two reflections would be modeled as two Values b11, b22, d11, and d22 appearing in Eqs. 共A1兲
individual mirrors, but geometry constraints would are given by
be necessary to ensure that the curvatures be equal
and that the centers of curvature be coincident. Ei- b11 ⫽ 兵d0 ⫹ d1 ⫹ d2 ⫹ 2c2 d2共d0 ⫹ d1兲sec ␪2
ther a single mirror could be considered as a multire- ⫹ 2c1 d0关2c2 d1 d2 ⫹ 共d1 ⫹ d2兲cos ␪2兴
flector, making, for example, three mirrors into a
four-reflection system, or two mirrors could be mul- ⫻ sec ␪1 sec ␪2其sec ␪obj, (A2a)
tireflectors, forming a two-mirror four-reflection sys- b22 ⫽ d0 ⫹ d1 ⫹ d2 ⫹ 2c2 d2共d0 ⫹ d1兲cos ␪2
tem. 共Of course more mirrors or more reflections
can be added if so desired.兲 ⫹ 2c1 d0共d1 ⫹ d2 ⫹ 2c2 d1 d2 cos ␪2兲cos ␪1, (A2b)
In this paper and in the accompanying paper on
three-spherical-mirror systems5 we originally in- d11 ⫽ 关2c2 d0 ⫹ 共cos ␪2 ⫹ 2c2 d1兲
tended to include constraints based on a second- ⫻ 共1 ⫹ 2c1 d0 sec ␪1兲兴sec ␪2 sec ␪obj, (A2c)
order analysis similar to that performed in a related
paper on two spherical mirrors.9 However, the d22 ⫽ 关2c2 d0 cos ␪2 ⫹ 共1 ⫹ 2c2 d1 cos ␪2兲
constraint equations based on second-order proper-
⫻ 共1 ⫹ 2c1 d0 cos ␪1兲兴. (A2d)
ties turned out to be too complicated to be solved
algebraically, and therefore only first-order proper- Note that Eqs. 共A1兲 have been put in a form that
ties are considered. Had we been able to include can be used to configure the final two spherical mir-
constraints based on second-order properties, the rors in any system to achieve the first-order proper-

3240 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000


ties specified by Eqs. 共5兲 and 共6兲. This can be done, Note that c3 and d3 in Eqs. 共A4c兲 and 共A4d兲 are given
provided the derivative matrix of the subsystem pre- by Eqs. 共A4a兲 and 共A4b兲. In Eqs. 共A4兲, a11, a22, c11,
ceding these mirrors is known, and this derivative and c22 are given by
matrix must be of the form of that of a plane-
symmetric system 共i.e., the off-diagonal elements of
each 2 ⫻ 2 submatrix must vanish兲. We adopted the a11 ⫽ 关共1 ⫹ 2c2 d2 sec ␪2兲 ⫹ 2c1共d1 ⫹ d2
convention in developing Eqs. 共A1兲 that the final ref-
erence surface for this subsystem lie at the penulti- ⫹ 2c2 d1 d2 sec ␪2兲sec ␪1兴, (A5a)
mate mirror 共i.e., the next-to-last mirror兲 in the
system and be normal to the base ray segment before
this mirror. To illustrate this point we consider the a22 ⫽ 关共1 ⫹ 2c2 d2 cos ␪2兲 ⫹ 2c1共d1 ⫹ d2
following matrix: ⫹ 2c2 d1 d2 cos ␪2兲cos ␪1兴, (A5b)

冋 册
a b
c d
⫽ 共T2R2兲共T1R1兲T0, (A3) c11 ⫽ 2关c1 ⫹ c2共2c1 d1 ⫹ cos ␪1兲sec ␪2兴sec ␪1, (A5c)

c22 ⫽ 2关c1 ⫹ c2共2c1 d1 ⫹ sec ␪1兲cos ␪2兴cos ␪1. (A5d)


where Ti 共i ⫽ 0, 1, 2兲 and Ri 共i ⫽ 1, 2兲 are given by
Eqs. 共4兲. To respect the convention described above,
T2 must have its final reference surface normal to the As was done for the finite-conjugate case, Eqs. 共A4兲
base ray segment between mirrors 2 and 3 共mirror 3 have been put in a form in which they can be used
is the penultimate mirror in this case兲. That is, ␪3 in
with any preceding subsystem, provided that a11, a22,
Eq. 共4a兲 with i ⫽ 2 must be taken to be zero. In this
c11, and c22 in those equations are replaced by the
case, b11, b22, d11, and d22 in Eqs. 共A2兲 are precisely
the corresponding elements of b and d that appear in corresponding matrix elements for the preceding sub-
Eq. 共A3兲. If there is a different subsystem preceding system 共with the final reference surface for this sub-
two final spherical mirrors, Eqs. 共A1兲 could still be system located at the penultimate mirror but
used 共with the understanding that c3 is the curvature oriented normally to the base ray segment before this
of the penultimate mirror in the system, c4 is the mirror兲.
curvature of the last mirror, etc.兲. b11, b22, d11, and
The authors thank Tropel Corporation for support-
d22 in Eqs. 共A1兲 simply need to be replaced by the
ing this research.
corresponding elements of b and d for the actual sub-
system preceding these two mirrors.
References and Notes
Of course, Eqs. 共A1兲 fail in cases when the preced-
ing subsystem presents imaging properties that 1. D. Korsch, Reflective Optics 共Academic, Boston, Mass., 1991兲,
make it impossible for the following two mirrors to Chaps. 10 and 12.
2. The original research can be found in K. Schwarzschild, “Un-
achieve the first-order properties specified by Eqs. 共5兲
tersuchungen zur geometrischen Optik. II. Theorie der
and 共6兲. For example, when the preceding sub-
Spiegelteleskope,” Abh. Koenigl. Ges. Wiss. Goettingen Math.-
system forms a sharp image 共to first order兲 on the Phys. Klasse, Folge 9 IV共2兲, 1–28 共1905兲. A modern summary
penultimate mirror in the system, b11 ⫽ b22 ⫽ 0 关and translated into English can be found in R. N. Wilson, “Karl
Eq. 共A1a兲 becomes ill defined兴. In this case the final Schwarzschild Lecture of the German Astronomical Society,”
mirror is essentially required to form a perfect image Rev. Mod. Astron. 7, 1–29 共1993兲.
共to first order兲 of a point object, which cannot be 3. G. D. Wassermann and E. Wolf, “On the theory of aplanatic
achieved unless the tilt angle of the final mirror is aspheric system,” Proc. Phys. Soc. B 62, 2– 8 共1949兲.
zero 共resulting in an obstructed system兲. 4. D. R. Shafer, “Four-mirror unobscured anastigmatic telescopes
For the case when the object is at infinity, the with all-spherical surfaces,” Appl. Opt. 17, 1072–1074 共1977兲.
following equations define Li 共i ⫽ 1, 2, 3, 4兲 that 5. J. M. Howard and B. D. Stone, “Imaging with three spherical
appear in Eqs. 共13兲: mirrors,” Appl. Opt. 39, 3216 –3231 共2000兲.

cos ␪3共 f2c22 ⫹ sin2 ␪4 ⫺ f1c11 cos2 ␪4 cos ␪im兲


c3 ⫽ ⫺ , (A4a)
2共 f2a22 cos2 ␪3 ⫺ f1a11 cos2 ␪4 cos ␪im兲
共 f2a22 ⫺ f1a11 cos ␪im兲共 f2a22 cos2 ␪3 ⫺ f1a11 cos2 ␪4 cos ␪im兲
d3 ⫽ , (A4b)
sin ␪4关 f2a22 cos2 ␪3共 f1c11 cos ␪im ⫹ 1兲 ⫺ f1a11 cos ␪im共 f2c22 ⫹ 1兲兴
2

cos ␪4共 f1c11 ⫹ 2f1a11c3 sec ␪3 ⫺ sec ␪im兲


c4 ⫽ ⫺ , (A4c)
2f1关c11d3 ⫹ a11共1 ⫹ 2c3 d3 sec ␪3兲兴
d4 ⫽ f1 cos ␪im关c11d3 ⫹ a11共1 ⫹ 2c3 d3 sec ␪3兲兴. (A4d)

1 July 2000 兾 Vol. 39, No. 19 兾 APPLIED OPTICS 3241


6. For the mechanics of differential ray tracing through homo- annealing algorithm for global optimization over continuous
geneous media, see, for example, A. Cox, A System of Optical variables,” J. Global Optimization 6, 1–34 共1995兲.
Design 共Focal, London, 1964兲, pp. 112–121; D. P. Feder, “Dif- 8. Details of imaging with a single spherical mirror can be found
ferentiation of ray-tracing equations with respect to construc- in J. M. Howard and B. D. Stone, “Imaging a point to a line
tion parameters of rotationally symmetric optics,” J. Opt. Soc. with a single spherical mirror,” Appl. Opt. 37, 1826 –1834
Am. 58, 1494 –1505 共1968兲. 共1998兲.
7. Details of this global optimization method can be found in 9. J. M. Howard and B. D. Stone, “Imaging a point with two
A. E. W. Jones and G. W. Forbes, “An adaptive simulated spherical mirrors,” J. Opt. Soc. Am. A 15, 3045–3056 共1998兲.

3242 APPLIED OPTICS 兾 Vol. 39, No. 19 兾 1 July 2000

You might also like