ArticlePDF Available

Tailored freeform optical surfaces


Abstract and Figures

Freeform optical surfaces embedded in three-dimensional space, without any symmetry, are tailored so as to solve the archetypal problem of illumination design: redistribute the radiation of a given small light source onto a given reference surface, thus achieving a desired irradiance distribution on that surface. The shape of the optical surface is found by solving a set of partial nonlinear differential equations. For most cases, a few topologically distinct solutions exist, given suitable boundary conditions.
Content may be subject to copyright.
Tailored freeform optical surfaces
Harald Ries
Optics & Energy Concepts, Paul-Gerhardt-Allee 42, 81245 Munich, Germany,
and Philipps-University, Renthof 5, 35032 Marburg, Germany
Julius Muschaweck
Optics & Energy Concepts, Paul-Gerhardt-Allee 42, 81245 Munich, Germany
Received January 24, 2001; accepted August 31, 2001; revised manuscript received September 4, 2001
Freeform optical surfaces embedded in three-dimensional space, without any symmetry, are tailored so as to
solve the archetypal problem of illumination design: redistribute the radiation of a given small light source
onto a given reference surface, thus achieving a desired irradiance distribution on that surface. The shape of
the optical surface is found by solving a set of partial nonlinear differential equations. For most cases, a few
topologically distinct solutions exist, given suitable boundary conditions. © 2002 Optical Society of America
OCIS codes: 080.2740; 350.4600; 080.2720.
The general problem of optical design for illumination can
be stated as follows: Given a specified light source, de-
sign an optical system that redirects the light from the
source onto a target surface of given shape such as to pro-
duce a given desired irradiance distribution on the target.
In this formulation, for simplicity, multiple optical sur-
faces and gradient index distributions are not considered.
A large amount of work has been devoted to this classical
design problem. Often a simple system consisting of a
single reflective or refractive surface is desired. In this
work we also refer to single optical surfaces.
Heuristic approaches are based on the known optical
properties of conics. Imaging optical systems tradition-
ally rely almost exclusively on spherical surfaces. Even
nonimaging systems traditionally involve combinations of
conics. Recently we introduced the idea of tailoring an
optical surface, which consists of translating the desired
optical properties into a suitable differential equation,
which is then solved numerically to determine the shape
of the optical surface.15In this way smooth surfaces are
found, which in general are not conic sections. The pre-
vious work on tailoring was applicable to two-dimensional
systems, i.e., systems that possess translational or rota-
tional symmetry. In this paper we follow the same line of
thought for designing freeform three-dimensional sur-
faces. Although the extension from two to three dimen-
sions may seem a minor step, this is by no means true.
Two-dimensional tailoring leads to ordinary differential
equations. Such equations can easily be solved even by
heuristic approaches. Three-dimensional tailoring, in
contrast, leads to nonlinear partial differential equations
that are much more demanding.
Optimization approaches first attempt a suitable pa-
rameterization of the surface with a finite and in general
low number of parameters. The value of these param-
eters are then to be determined by optimization based on
ray tracing and some objective function to measure how
well a surface characterized by certain values of the pa-
rameters performs. Optimization approaches are very
successful for many applications. However, if freeform
optical surfaces are to be designed for illumination tasks,
procedures based on optimization suffer from the fact that
adequate representation of optical surfaces tends to re-
quire a large number of parameters. Even slight varia-
tions of the surface shape can result in large variations in
the produced irradiance distribution. Ray tracing in-
volves a statistical element and therefore cannot assess
the optical performance precisely. At best it can provide
an estimate that becomes better as more rays are used.
For freeform surfaces whose purpose is to produce irradi-
ance distributions that are rich in detail, each evaluation
of the objective function with reasonable accuracy re-
quires a huge computational cost. For these reasons we
do not elaborate further on optimization approaches here.
In this paper we outline a constructive method of de-
sign that translates the desired optical performance into
a differential equation that is solved by numerical meth-
ods to yield the desired shape of the optical surface.
Consider a ray from a small source located at saimed
at a point on the optical surface located at pas shown
in Fig. 1. The laws of reflection (refraction) govern
the direction into which this ray is redirected. In effect
these laws establish a relation between the surface-
normal unit vector N, the direction of the incoming ray
In Unit(ps) , and the direction of the outgoing ray
Out Unit(tp).
1n22nOut In1/2NOut nIn. (1)
Here Unit denotes the unit vector: Unit(v)v/
vv. Equation (1) applies for both reflection (n
1) and refraction (n1), where ndenotes the index
of refraction of the medium surrounding the source. To
analyze the power density carried by the radiation, we
need to consider an ensemble of adjacent rays. This
leads to the concept of a wave front; see Fig. 2.
590 J. Opt. Soc. Am. A /Vol. 19, No. 3 /March 2002 H. Ries and J. Muschaweck
0740-3232/2002/030590-06$15.00 © 2002 Optical Society of America
A wave front is a family of surfaces, each perpendicular
to the local direction of the radiation. For example, the
undistorted radiation emitted by a point source is de-
scribed by concentric spheres centered at the location of
the source. Wave fronts may be distorted by optical sur-
faces or nonconstant refractive index and thereby have
shapes different from spherical. The power density
transported by the radiation along the direction of propa-
gation is proportional to the absolute value of the Gauss-
ian curvature of the wave front. The curvature tensor K
is the derivative of the unit normal of a surface with re-
spect to the tangent directions t1and t2:
i,j1, 2. (2)
Since Nis a unit vector, its derivative is perpendicular to
N, i.e., restricted to the tangent plane. The curvature is
thus a two-dimensional tensor. The Gaussian curvature
is the product of the two eigenvalues c1and c2of the cur-
vature tensor defined in Eq. (2). For free propagation,
the curvature of the wave front changes along the ray
such that the two curvature centers (also referred to as
foci) remain fixed. For the radiation of a point source,
the two foci f1and f2coincide with the location of the
source. For a general wave front, the position of the foci
is calculated from the eigenvalues c1and c2of the curva-
ture tensor:
N. (3)
Here we adopt the convention that positive curvature im-
plies a convex surface, where the focus is behind. The ir-
radiance of a freely propagating wave front changes in the
direction of propagation according to Ref. 6:
. (4)
In general, the two foci for each point of the wave front
form two (curved) surfaces in space where the optical sur-
face has imaging properties. Equation (5) diverges if p
coincides with one of the foci. In reality, sources are not
true points and their brightness is not infinite; therefore
the irradiance approaches the brightness of the image of
the source. This phenomenon is illustrated by the bright
lines (caustics) formed on a sunny day at the bottom of a
swimming pool. The two caustic surfaces are the gener-
alization of the line caustic in two dimensions that is use-
ful for illumination design.7The tailoring algorithm as-
sumes a small source, which implies that the source
brightness is much larger than the desired irradiance
anywhere on the target. Therefore the caustic surfaces
must never intersect the target surface. This fact further
illustrates that our approach is rooted in nonimaging op-
To quantify the irradiance on the target surface, we
first need to derive how the curvature of a wave front is
changed by an optical surface. To this end we compute
the derivative of Eq. (2) with respect to displacements of
the point pin the tangent plane of the optical surface.
We are interested only in changes of the normal vectors
confined to the tangent plane of the optical surface:
1n22nOut InKo
POut KOut JOut nPIn KIn JIn . (5)
Here Kodenotes the curvature tensor of the optical sur-
face, KIn the curvature tensor of the incoming optical
beam at the intersection with the optical surface, and
KOut the curvature tensor of the outgoing optical beam
immediately after reflection (refraction) at the optical sur-
face. The tensors PIn and POut are projection operators
from the tangential plane of the incoming or the outgoing
wave front onto the tangential plane of the optical sur-
face. The tensors JIn and JOut are the Jacobians, which
project the differentials from the tangent plane of the
wave fronts onto the tangent plane of the optical surface:
JIn i
tjO,JOut i
tjO. (6)
If we choose a coordinate system on each tangent plane
such that the first coordinate is sagittal, i.e., co-planar
with the incoming and the outgoing ray, and the second
coordinate is tangential, i.e., parallel on all surfaces, then
the projection operators as well as the Jacobians are di-
agonal and simple:
Fig. 2. Curvature of the optical surface, illustrated here as a re-
flector, and the curvature of the incoming wave front combine to
determine the curvature of the outgoing wave front.
Fig. 1. Sketch of source location s, a point pon the optical sur-
face, and points ton the target surface. Ndenotes the unit nor-
mal to the optical surface.
H. Ries and J. Muschaweck Vol. 19, No. 3/ March 2002 /J. Opt. Soc. Am. A 591
NIn 0
POut JOut
NOut 0
. (7)
In particular, for a reflector surface this leads to n1,
Out In ⫽⫺cos(2
) and NOut ⫽⫺NIn
), where
is the incidence angle, and Eq. (5) be-
0 cos
KIn KOut
. (8)
Equation (8) can be interpreted as follows: The curva-
ture of the optical surface is the difference between the
projected curvatures of incoming and the outgoing wave
fronts, which is a direct extension of Eq. (1). Projection
essentially increases curvature in the sagittal direction
and decreases it in the tangential direction. This phe-
nomenon was outlined in an earlier paper.8
Given the curvature of the outgoing wave front, we can
calculate the foci of the outgoing beam, g1and g2, and the
irradiance produced at the target surface according to Eq.
NtOut. (9)
Here I(In) is the intensity of the source in the direction
In, i.e., the irradiance produced by the source on the unit
sphere in that direction. The first fraction in Eq. (9) de-
scribes the irradiance produced by the source at the posi-
tion of the optical surface, measured in the tangent plane
of the incoming radiation. The second fraction describes
propagation of the outgoing beam characterized by the
foci g1and g2between the optical surface at pand the
target point t. Finally, the last scalar product accounts
for the incidence angle with respect to the target surface
normal Nt. Note that the positions of the foci g1and g2
refer to the outgoing wave front. Their relation to the
curvature of the incoming radiation and that of the opti-
cal surface is quantified by Eq. (5).
If one chooses orthonormal coordinate systems aligned
with a sagittal and a tangential plane on all three sur-
faces, the projection operators are diagonal, which consid-
erably simplifies Eq. (5). But in this coordinate system,
in general the curvature tensor of the optical surface is
not diagonal. One might be tempted to restrict the cur-
vature of the optical surface by requiring that the eigen-
vectors of the curvature tensor also be aligned with the
sagittal and tangential directions, defined by the inci-
dence direction. After all, the distance to these foci is all
that matters for the irradiance. However, for reasons
outlined below, this choice violates the integrability con-
dition and therefore in general rules out a solution of the
differential equations as a smooth, continuous optical sur-
Equation (5) relates the curvature of the optical surface
to its slope. The great achievement of Gauss was to
prove that slope and curvature are inherent properties of
a surface, which do not depend on a particular parameter-
ization. We made use of this fact by choosing an abstract
formulation in the previous derivations in order to better
illustrate the underlying principles.
If, however, one proceeds to actually construct an opti-
cal surface, one cannot avoid resorting to a particular rep-
resentation based on some coordinate system. As we
have seen, it is not possible to diagonalize the projection
and the curvature tensors simultaneously. Equations
(1), (4), and (5) translate to a set of two elliptical nonlin-
ear partial differential equations. Equivalent equations
were derived by Monge and Ampe
`re more than 200 years
ago.9The general problem for a point source was pre-
sented in the form of an elliptical equation of the Monge
`re type by Komissarov and Boldyrev.10 The prob-
lem that we have treated has also been analyzed in detail
by Schruben.11 This author used an equivalent but nev-
ertheless different notation and also showed that the
problem is governed by an elliptic MongeAmpe
`re differ-
ential equation. However, the earlier authors give no
hint about actually generating a solution for a particular
desired distribution. The reason for us to use a different
formulation is that our primary goal is to actually find an
algorithm that generates a numerical solution in the gen-
eral case. We did not attempt a closed-form solution be-
cause we did not wish to restrict the design input to func-
tions represented as closed-form expressions. Indeed,
although the numerical problem is quite demanding and
challenging, we were able to devise a numerical algorithm
for finding a numerical solution following standard nu-
merical procedures outlined in Ref. 12. On the basis of
the examples we have tackled so far, we are confident that
we have developed a powerful tool applicable to an ex-
tremely large class of problems.
Suppose we have integrated the second-order equation
and determined the normal to the surface. Is it possible
to determine the reflector surface? It has been shown13
that a necessary and sufficient condition for the existence
of a surface everywhere perpendicular to a vector field N,
i.e., for the existence of a quasipotential, is that this vec-
tor field satisfy
NcurlN0. (10)
This constitutes an additional condition to be imposed on
the on the second derivative of the surface parameteriza-
tion. An equivalent condition was found by Parkyn.14
One is free to choose boundary conditions for a particular
solution; however, one cannot freely choose a complete
boundary curve for the optical surface. In general, such
a choice would fix at least one component of the normal
vector and therefore violate either Eq. (1) or, most prob-
ably, the integrability condition. Therefore we chose
to require the boundary of the optical surface to lie on an
a priori defined two-dimensional manifold such as a cyl-
592 J. Opt. Soc. Am. A /Vol. 19, No. 3 /March 2002 H. Ries and J. Muschaweck
inder or a cone. We found that with this boundary con-
dition the solution is unique up to a scaling factor.
In order to determine the shape of the optical surface, we
have to solve the set of differential equations. Let us re-
view them in turn:
Equation (1), which specifies which point on the tar-
get surface is illuminated by a given point on the optical
Equation (9), which connects the desired irradiance
on the target to the focal points of the outgoing wave
Equation (5), which connects the curvature of the
outgoing wave front to the curvature of the incoming
wave front and the curvature of the optical surface.
Equation (10), the integrability condition.
Equation (10) has a key role. It may seem that the de-
sired irradiance distribution on the target may be
achieved with an entire family of curvature tensors, and
one may simply choose a convenient representative. In-
deed, solutions based on such an arbitrary choice have
been shown to work in particular, usually highly symmet-
ric, configurations.15 However, we found that an arbi-
trary choice in the general case violates the integrability
condition. Therefore we have added the integrability
condition explicitly to the set of determining equations.
The set of equations is nonlinear and elliptical. As out-
lined in Ref. 12, equations of this type need special care.
We have been able to devise an algorithm by discretiza-
tion of the optical surface to a grid of points and subse-
quently identifying derivatives with finite differences.
In addition to the differential equations, one can
specify boundary conditions. The boundary conditions
also need special consideration as indicated by
Schruben.11 If we were simply to specify the outer
boundary of the optical surface as a curve in space, this
would also specify one tangent at the rim and therefore in
general violate Eq. (1). Therefore we can require only
that the boundary of the optical surface lie on some pre-
defined surface. This allows us, for example, to require
that the boundary of the optics be on a cylinder but not
that it be on a circle.
In addition, we found that because in the limit of a
small source the optical laws scale, we can require a cer-
tain average size of the optical surface (or average dis-
tance from the source). However, this condition may vio-
late energy conservation, because the entire optical
surface must intercept a total power from the source that
precisely matches the integral of the desired irradiance
on the target.
An additional restriction applies to refractive optics.
The maximum angle by which a ray can be redirected by
a lens is limited by total internal reflection. When tailor-
ing refractive optics, we did encounter occasionally un-
physical solutions, which attempt to change the direction
of the rays by too large an angle, meaning that in reality,
total internal reflection would occur for part of or the en-
tire optical surface.
The two caustic surfaces are the reason for the exis-
tence of three topologically distinct solutions. We have
explained that the caustic surfaces must never intersect
the target surface. Therefore in general there are dis-
tinct solutions: one in which both caustic surfaces lie be-
tween source and target, one in which both caustic sur-
faces lie outside the region between source and target,
and a third in which one caustic surface lies between
source and target and one does not. Apart from these
distinctions and the restrictions outlined above, we found
that if a suitable set of boundary conditions is specified, a
solution can be found by tailoring, and this solution ap-
pears to be unique.
Fig. 3. Sketch of the setting: The lens is made to collect light
from an isotropic point source and redistribute it onto a target
perpendicular to the axis. Length units are arbitrary; the draw-
ing is not to scale.
Fig. 4. Shape of the freeform lens that casts the letters OEC on
a square background, as it would appear under oblique illumina-
tion. A slight bulging is visible, which remotely resembles the
letter shape. Note that the rim lies on a cone as required by the
boundary condition.
H. Ries and J. Muschaweck Vol. 19, No. 3/ March 2002 /J. Opt. Soc. Am. A 593
As a demanding example and proof of concept, we speci-
fied a desired irradiance distribution that is constant over
a square target except for the contours of the letters
‘‘OEC,’’ for which we specified an intensity three times
that of the background. All rays incident on the optical
surface are to be redistributed on the target with no rays
outside. For this case we designed a refractive (lens) sur-
face to perform the task in a setting sketched in Fig. 3.
Of course, it would be a simple task to create any de-
sired irradiance distribution in this setting with a simple
slide projector. However, such an approach wastes radia-
tion by absorption in the slide. In contrast, the method
that we outline here creates the desired irradiance distri-
bution purely by redirecting the radiation. Thereby all
intercepted radiation is redirected to the target.
The shape of the lens is depicted in Figs. 4 and 5. By
careful inspection, one has the impression of recognizing
regions on the lens surface singled out by slightly differ-
ent curvatures, which resemble the letters on the target
although the shape is distorted. In retrospect, this is
easy to understand since the regions of the lens contrib-
uting to the letters have to be three times as large in area
as a result of energy conservation. The rim of the lens
was made to lie on a cone, oriented with its axis toward
the center of the desired distribution and a tip half-angle
of roughly 33 deg. The energy balance determines the
exact value of the tip angle.
We verified the optical performance of the tailored free-
form lens by importing the shape into a commercial ray-
tracing program16 and performed an independent evalua-
tion by tracing 9 106rays from a point source through
the lens and computing their intersection with the target
plane. The result is shown in Fig. 6. Figure 7 plots the
intensity along the horizontal cross section through the
center in order to quantitatively verify the intensity ratio
of zero (outside) to 1 to 3.
We have presented a method for designing freeform opti-
cal surfaces tailored to specific illumination tasks. Given
a small source with known intensity distribution, given
also a desired irradiance distribution on a target surface,
we have devised a constructive procedure to design the
shape of an optical surface, either reflective or refractive,
that redistributes the radiation from the source onto the
target, precisely matching the desired distribution. Tai-
loring draws on principles of nonimaging optics, applied,
as far as we know for the first time, to problems without
Fig. 5. Shape of the freeform lens in an isosag representation.
Lines indicate equal sag. Spacing is 0.01 unit. The ragged rim
is an artifact of the representation. The rim of the lens was
made to lie on a cone oriented with its axis toward the center of
the desired distribution.
Fig. 6. Irradiance distribution produced by the freeform tailored
lens based on Monte Carlo ray tracing of 9 106rays. Inter-
section of the rays with a plane at the position of the nominal tar-
get was binned by 129 129 pixels. Intensity variations are
entirely compatible with shot noise.
Fig. 7. Horizontal cross section through the irradiance distribu-
tion in Fig. 5 at vertical position 0. Irradiance is in arbitrary
units. Note the contrast of 0/1/3. Noise is completely within
statistical bounds.
594 J. Opt. Soc. Am. A /Vol. 19, No. 3 /March 2002 H. Ries and J. Muschaweck
any symmetry. Because of its generality this approach is
very promising for a large variety of applications.
We have presented idealized results in which we delib-
erately excluded many real-world effects such as finite
source size, Fresnel reflections and, above all, manufac-
turing errors such as slope errors of the surface. We did
this in order to show the strength and limitations of the
mathematical procedure in the sense of a proof of concept.
We found that the numerical procedure is extremely fast
and efficient. For the example presented, in which the
surface was rendered as a grid of 129 129 points, the
computation required roughly one hour on a standard
contemporary personal computer. The numerical preci-
sion was far beyond manufacturing capabilities.
In fact, we found it more difficult to assess the perfor-
mance of the tailored optical surface with ray tracing
than to design it. For an illustration, imagine an ambi-
tious claim that within a target area of, say, 1 m2there is
not a single squared millimeter on which the irradiance is
more than 1% off the nominal value. In fact, we believe
that apart from manufacturing tolerances, the pure com-
putational procedure outlined here is quite capable of
such outstanding performance because we have verified
that the tolerance in selected outgoing rays is of the order
of 1
m. To substantiate such a claim by a Monte Carlo
ray-tracing technique, one would need more than 104rays
per bin in order to reach a statistical error below 1%. On
the other hand, to match the spatial resolution, one would
need at least 106pixels. These requirements combine to
a requirement of 1010 rays. This is beyond the capability
of today’s computers, which typically require of the order
of 103s per ray.
We are well aware that our example is of no practical
value. The reason we chose it was, again, to demonstrate
a capability that other known design methods can hardly
match. The desired irradiance distribution is what we
would call rich in detail. It is hard to imagine that a pa-
rameterization of a surface based on few parameters for
such a task would be feasible.
The example chosen deliberately involves a discontinu-
ous desired irradiance distribution that features steps in
irradiance by a factor of 3. Because irradiance on the
target essentially is linked to the curvature of the tailored
surface, the optical surface designed is continuous and
smooth. Only the curvature exhibits steps. The tailored
surface is geometrically smooth (G1) , even for disconti-
nous desired intensity distribution. Since intensity is re-
lated to the curvature of the optical surface, in general, if
the desired irradiance is cncontinuous, the tailored opti-
cal surface will feature Gn2geometric continuity. This
facilitates manufacturing.
We thank Wolfgang Spirkl and Andreas Timinger for
valuable suggestions and discussions. We are grateful to
the reviewer for pointing out Refs. 10, 11, and 13.
Harald Ries can be reached by e-mail at harald.ries
1. J. M. Gordon and H. Ries, ‘‘Tailored edge-ray concentrators
as ideal second stages for Fresnel reflectors,’’ Appl. Opt. 32,
22432251 (1993).
2. H. Ries and R. Winston, ‘‘Tailored edge-ray reflectors for il-
lumination,’’ J. Opt. Soc. Am. A 11 , 12601264 (1994).
3. H. Ries and J. Gordon, ‘‘Double-tailored imaging concentra-
tors,’’ in Nonimaging Optics: Maximum Efficiency Light
Transfer V, R. Winston, ed., Proc. SPIE 3781, 129 134
4. J. M. Gordon and H. Ries, ‘‘Tailord edge-ray concentrators
(TERC’s): Approaching the thermodynamic limit with new
ideal second stages for Fresnel reflectors,’’ presented at The
Biennial Congress of the International Solar Energy Soci-
ety, Budapest, August 1923, 1993.
5. R. P. Friedman, J. M. Gordon, and H. Ries, ‘‘Compact high-
flux two-stage solar collectors based on tailored edge-ray
concentrators,’’ Sol. Energy 56,607615 (1996).
6. L. D. Landau and E. M. Lifschitz, Klassische Feldtheorie.
Vol. 2 of Lehrbuch der Theoretischen Physik, G. Heber, Ed.
(Akademie Verlag, Berlin, 1904).
7. H. Ries and W. Spirkl, ‘‘Caustic and its use in designing op-
timal absorber shapes for 2D concentrators,’’ in Nonimag-
ing Optics: Maximum Efficiency Light Transfer III,R.
Winston, ed., Proc. SPIE 2538,29 (1995).
8. R. Zaibel, E. Dagan, J. Karni, and H. Ries, ‘‘An astigmatic
corrected target-aligned heliostat for high concentration,’’
Sol. Energy Mater. Sol. Cells 37, 191202 (1995).
9. G. Monge, Application de l’analyse a
`la ge
´trie, 5th ed.
Corrected and annotated by M. Liouville (Bachelier, Paris,
10. V. D. Komissarov and N. G. Boldyrev, ‘‘The foundations of
calculating specular prismatic fittings,’’ Trudy VEI 43,661
11. J. S. Schruben, ‘‘Formulation of a reflector design problem
for a lighting fixture,’’ J. Opt. Soc. Am. 62, 14981501
12. W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vet-
terling, Numerical Recipes, 2nd ed. (Cambridge U. Press,
New York, 1992).
13. P. Moon, D. E. Spencer, The Photic Field (MIT Press, Cam-
bridge, Mass., 1981).
14. W. A. Parkyn, ‘‘The design of illumination lenses via extrin-
sic differential geometry,’’ in International Optical Design
Conference, K. P. Thompson and L. R. Gardner, eds., Proc.
SPIE 3482,191193 (1998).
15. W. A. Parkyn, ‘‘Illuminating Lens designed via extrinsic dif-
ferential geometry,’’ U.S. patent 5,924,788, July 20, 1999.
16. LightTools 3.1, 2001, Optical Research Associates, www.op-, 3280 East Foothill Boulevard, Pasadena, Cali-
fornia 91107.
H. Ries and J. Muschaweck Vol. 19, No. 3/ March 2002 /J. Opt. Soc. Am. A 595
... In dO, the merit function can also be at the irradiance level, when optimizing in the image space rather in the intersection space, e.g. in Refs. [48], [13]. This requires integrating a sensor image I(p) from intersections p, given a pre-defined pixel size. ...
Full-text available
Computational imaging systems algorithmically post-process acquisition images either to reveal physical quantities of interest or to increase image quality, e.g., deblurring. Designing a computational imaging system requires co-design of optics and algorithms, and recently Deep Lens systems have been proposed in which both components are end-to-end designed using data-driven end-to-end training. However, progress on this exciting concept has so far been hampered by the lack of differentiable forward simulations for complex optical design spaces. Here, we introduce $\mathbf{dO}$ (DiffOptics) to provide derivative insights into the design pipeline to chain variable parameters and their gradients to an error metric through differential ray tracing. However, straightforward back-propagation of many millions of rays requires unaffordable device memory, and is not resolved by prior works. $\mathbf{dO}$ alleviates this issue using two customized memory-efficient techniques: differentiable ray-surface intersection and adjoint back-propagation. Broad application examples demonstrate the versatility and flexibility of $\mathbf{dO}$ , including classical lens designs in asphere, double-Gauss, and freeform, reverse engineering for metrology, and joint designs of optics-network in computational imaging applications. We believe $\mathbf{dO}$ enables a radically new approach to computational imaging system designs and relevant research domains.
... Non-imaging free-form lenses are formulated as an inverse problem with the aim to redistribute a light source with a known intensity profile into a desired output intensity [21]. A variety of different strategies have been implemented to tackle this inverse problem [21,[25][26][27][28][29], but a particularly elegant and general way to do it is through an optimal transport formulation [30,31]. This method makes no a priori symmetry assumptions on the lens shape and has been demonstrated to work for a variety of incident light distributions and output illumination patterns [32]. ...
Full-text available
Metasurfaces provide a versatile platform for realizing ultrathin flat optics for use in a wide variety of optical applications. The design process involves defining or calculating the phase profile of the metasurface that will yield the desired optical output. Here, we present an inverse design method for determining the phase profile for shaping the intensity profile of a collimated incident beam. The model is based on the concept of optimal transport from non-imaging optics and enables a collimated beam with an arbitrary intensity profile to be redistributed to a desired output intensity profile. We derive the model from the generalized law of refraction and numerically solve the resulting differential equation using a finite-difference scheme. Through a variety of examples, we show that our approach accommodates a range of different input and output intensity profiles, and discuss its feasibility as a design platform for non-imaging optics.
... Tailoring design method is also used to develop illumination system design, and to build 3D freeform surfaces by solving nonlinear PDEs. The relationship between light vectors such as incident, normal, and exit, as well as constraints among the source of light with the irradiance distribution at the reference surface, are taken into account by the formed freeform optical surface [79]. ...
Full-text available
Freeform optics has become the most prominent element of the optics industry. Advanced freeform optical designs supplementary to ultra-precision manufacturing and metrology techniques have upgraded the lifestyle, thinking, and observing power of existing humans. Imaginations related to space explorations, portability, accessibility have also witnessed sensible in today's time with freeform optics. Present-day design methods and fabrications techniques applicable in the development of freeform optics and the market requirements are focussed and explained with the help of traditional and non-traditional optical applications. Over the years, significant research is performed in the emerging field of freeform optics, but no standards are established yet in terms of tolerances and definitions. We critically review the optical design methods for freeform optics considering the image forming and non-image forming applications. Numerous subtractive manufacturing technologies including figure correction methods and metrology have been developed to fabricate extreme modern freeform optics to satisfy the demands of various applications such as space, astronomy, earth science, defence, biomedical, material processing, surveillance, and many more. We described a variety of advanced technologies in manufacturing and metrology for novel freeform optics. Next, we also covered the manufacturing-oriented design scheme for advanced optics. We conclude this review with an outlook on the future of freeform optics design, manufacturing and metrology.
... The design process often involves cumbersome calculations in order to determine the surface needed to project the desired intensity profile. Indeed, these geometries are typically defined as sum of spherical or aspherical lenses [151], through Q-polynomials description [152] or nonlinear partial differential equations [153]. Their fabrication is also a demanding task, in fact, manufacturing processes with high degrees of freedom are required [154], such as grinding, polishing [155], and ultra-precision turning [156]. ...
Full-text available
In this thesis I will explain the properties of intersubband polaritons by a theoretical and experimental point of view. After a brief introduction to explain some general aspects of the intersubband polariton generation I will describe an experimental campaign with the aim to verify the possibility to create a polariton laser. The emission of light from a polaritonic state lacks a clear experimental verification, therefore this experiment may be the first observation of a polariton laser. The possibility to use graphene to design a better optical resonance will be largely discussed. In the end I will illustrate a concept to manipulate the light with 3D printed windows. The final objective being the design of windows able to remove aberration from laser source (in particular TeraHertz sources).
Wide-field-of-view (FoV) Offner imaging spectrometers with freeform surfaces have been studied extensively in recent years. However, a design result with a large numerical aperture (NA) cannot be simultaneously obtained with this layout. We present the concept of a limited system in the tangential direction. Based on this insight, we present a new design method, to the best of our knowledge, based on the decenter anamorphic stop, which can achieve large NA in compact wide-FoV Offner imaging spectrometers with freeform surfaces. Compared to conventional imaging spectrometers with the same parameters, the light-gathering capacity of the decenter anamorphic stop-based imaging spectrometer is increased by more than 40%. In addition, based on the presented method, we design a compact imaging spectrometer with a wide FoV and large NA. The designed imaging spectrometer with a freeform surface has excellent performance. Finally, we fabricate and measure the freeform mirror. The surface irregularity of the freeform mirror is better than 1 / 30 λ ( λ = 632.8 n m ). The result shows that the Offner imaging spectrometer with a freeform surface can be fabricated and will play a significant role in the fields of aeronautical and astronautical remote sensing.
Freeform tailoring optics are core components in advanced illumination and laser technologies. In this work, we propose an efficient and streamlined optics design method to face the challenges in its implementation. We integrate ray-mapping least squares calculations with freeform surface optimization tools embedded in commercial optical design software. As a result, the complexity of freeform illumination and laser beam shaping design is significantly reduced by only adapting a built-in merit function. We demonstrate the feasibility of our approach through three design examples that provide tailored irradiance distributions. Furthermore, the results highlight the efficiency, simplicity and versatility of our approach that can be readily applied to multiple configurations.
A key challenge in tailoring compact and high-performance illumination lenses for extended non-Lambertian sources is to take both the étendue and the radiance distribution of an extended non-Lambertian source into account when redirecting the light rays from the source. We develop a direct method to tailor high-performance illumination lenses with prescribed irradiance properties for extended non-Lambertian sources. A relationship between the irradiance distribution on a given observation plane and the radiance distribution of the non-Lambertian source is established. Both edge rays and internal rays emanating from the extended light source are considered in the numerical calculation of lens profiles. Three examples are given to illustrate the effectiveness and characteristics of the proposed method. The results show that the proposed method can yield compact and high-performance illumination systems in both the near field and far field.
A freeform reflector for illumination in a space-based remote sensor calibration system is designed to produce the desired patterns and improve efficiency. Owing to the excellent cosine properties and light source stability of the integrating sphere, it is widely used as the calibration source in optical remote sensing instruments. However, the light source cannot obtain uniform illumination on a large scale. Therefore, we propose a freeform reflector design for an integral sphere source, which generates an independent irradiance distribution and achieves large-scale illumination by arraying. The irradiance uniformity and efficiency obtained were greater than 98% and 90%, respectively.
Full-text available
Adoption of LED lighting products have many advantages such as energy conservation and environmental protection. But the luminescent characteristics of LEDs cannot provide satisfactory lighting solutions to meet the requirements of today’s sophisticated applications. Freeform optics constitutes a new technique for regulating the energy output of LEDs into a customized form. However, many freeform optical elements are designed to be bulky since the point-like sources assumption. The designs that consider the spatial extents of LEDs usually have poorly controlled light patterns or low efficiencies. Here, we propose a very robust design method based on spatially variant convolution for 4D light field tailoring, which enables high-performing freeform illumination optics with small sizes. The proposed method is holistic that is suitable for various kinds of design requirements with no limitations of far-field approximation and Lambertian luminescent property. Several design examples including fabricated lenses and experimental tests are presented. The results realize compact, low-cost, and highly efficient freeform optics with customized lighting solutions. We believe that the proposed work can promote the research on enhancing the emission quality of LEDs and generating artificial light for application-specific lighting systems.
Nonimaging optical design aiming at energy control has a wide range of applications in optoelectronics. A nonimaging optical system is composed of a light source, optical components, and a target screen, and can be described by an equation named light taming equation(LTE). Given the light source and prescribed target spot, the required freeform surfaces of the optical component can be obtained by solving the LTE. If the light source profile does not change, the optical surface has to make some suitable morphs when the target spot translates on the screen, and the group theory can describe these morph operators well. The basic LTE was established for a normal nonimaging optical system, which is to design an optical element for redirecting the lights from the source so that a prescribed light distribution is generated on a given target. A translation light taming equation(T-LTE) was derived for the case of only spot translating on the target screen, and an optical translation group(OTG) was introduced for describing all of the morph operators of the optical surface caused by light spot translation. There are multiple solutions for the same T-LTE, but the uniqueness of the T-LTE solution is necessary for OTG. Fortunately, the eikonal-energy(KE) mapping method can guarantee the uniqueness of the T-LTE solution, where K is the optical path length. The supporting quadric method(SQM) is one of the KE mapping methods when the nonimaging optical system has only one optical surface to be resolved. The LTE with SQM was deduced, and the OTG can be discussed in K-space. A deep neural network(DNN) was introduced to fit the KE mapping and spot translating operators to obtain the required optical surface. Taking the uniform square spot as an example, the SQM generated the sample data of spot translation to train DNN. The optical simulation results show that the error between the light distribution generated by the DNN and the standard uniform square spot is small, all within the order of 10⁻³, which indicates that the DNN and KE mapping method successfully realized the function of the OTG. There is a guiding significance for intelligent nonimaging optical design.
Full-text available
Conventional heliostats suffer from astigmatism for non-normal incidence. For tangential rays the focal length is shortened while for sagittal rays it is longer than the nominal focal length. Due to this astigmatism it is impossible to produce a sharp image of the sun, and the rays will be spread over a larger area. In order to correct this the heliostat should have different curvature radii along the sagittal and tangential direction in the heliostat plane just like a non axial part of a paraboloid. In conventional heliostats, where the first axis, fixed with respect to the ground, is vertical while the second, fixed with respect to the reflector surface, is horizontal such an astigmatism correction is not practical because the sagittal and tangential directions rotate with respect to the reflector. We suggest an alternative mount where the first axis is oriented towards the target. The second axis, perpendicular to the first and tangent to the reflector, coincides with the tangential direction. With this mounting sagittal and tangential direction are fixed with respect to the reflector during operation. Therefore a partial astigmatism compensation is possible. We calculate the optimum correction and show the performance of the heliostat. We also show predicted yearly average concentrations.
Full-text available
The edge-ray principle can be used to tailor a reflector. However, one set of edge rays already fully determines the reflector profile. We present a design method for tailoring compact compound elliptical concentrator (CEC)-type reflectors to a given source and a desired angular power distribution. Two reflected images of the source, one on each side of the source, contribute together with the direct radiation from the source to produce the desired power distribution. We determine the reflector profile by numerically solving a differential equation. No optimization is required. Beyond the angular region in which the power distribution can be strictly controlled, the power drops to zero in a finite decay range. This decay range becomes narrower as the reflector increases in size. We show a reflector for producing a strictly constant irradiance from −43 to 43 deg from a cylindrical source of constant brightness. The reflector extends to a maximum distance of 8 source diameters. No power is radiated beyond ± 50 deg.
The caustic of a set of edge rays is defined as the set of intersection points of adjacent edge rays. For a body having a smooth differentiable contour, the caustic of its edge rays coincides with the contour of the body. Therefore one would assume that by calculating the caustic of the edge rays as they are produced by a 2D concentrator such as a trough, the optimal shape for the absorber, e.g. the minimal surface absorber capable of intercepting all rays, should also coincide with the shape of the caustic. We show that this conjecture is not valid in general, but only if the caustic indeed forms a closed smooth curve. For parabolic trough systems, the caustic intersects and forms closed domains for half rim angles of around 60 degrees and 120 degrees. In both cases the contour is not smooth. Therefore the optimal shape is not given by the domain enclosed by the caustic. We present a general recipe of how to construct minimum surface absorbers for given caustics in 2D and apply this to the case of trough parabolic concentrators. We show practical absorber shapes for parabolic troughs with various rim angles. The optimal contour depends discontinuously on the rim angle. The area of the optimum shape for a rim angle of 90 degrees is 0.72 of the area of the smallest cylindric absorber capable of intersecting all rays.
The differential geometry of surfaces describes both intrinsic and extrinsic properties: the intrinsic ones, such as the metric, are inherent, whereas extrinsic ones describe shape relative to 3D space, such as the distribution of surface normal vectors. This paper describes how the surface normals of a non-rotationally-symmetric illumination lens are generated, via Snell's law in vector form, from (1) the distribution of light from a sources, and (2) a desired directional distribution of light exiting the lens. These two distributions are expressed as grids with cells of variable size but constant photometric flux, on the Gaussian sphere of directions. The grids must be sufficiently fine so as to generate enough surface normal vectors for accurate numerical generation of the requisite surface. The grids must have the same number of cells, and the same topology (i.e., rectangular vs. polar). The source-intensity grid must be adjusted to account for Fresnel reflection losses. For an unfaceted (smooth-surfaced) lens, the array of surface normal vectors must be adjusted for equality of the crossed partial derivatives. This class of lenses has only recently become producible due to the advent of electric- discharge machining for the shaping of non-rotationally symmetric injection molds for plastic lenses.
A novel method is described of designing lenses for general illumination tasks that are not circularly symmetric. The crux of the method is the specification of the illumination task by a grid on the unit sphere of directions. This grid, or tessellation, has cells which vary in solid angle such that each encompasses the same luminous flux: high intensity corresponds to small cells, and vice versa. Another grid, having the same topology and number of cells, is formed according to the intensity distribution of the source. An illumination lens must then transform the source distribution into the task distribution, via one or more refractions. Thus the direction vector of a cell in the source grid must be redirected into that of the corresponding cell in the task grid. Snell's law in vector form enables the derivation of a corresponding surface normal vector, or sequence of normal vectors, that will accomplish this redirection. Extrinsic differential geometry is then used to generate a lens surface having, as closely as possible, this distribution of surface normals, which must be irrotational to generate a smooth surface. This class of lenses has only recently become producible due to the advent of electric-discharge machining for the shaping of non-rotationally symmetric injection molds for plastic lenses.
We present a new approach in optical design whereby two- stage axisymmetric reflectors are tailored with a completely imaging strategy, and can closely approach the thermodynamic limit to radiation concentration at near-maximum collection efficiency. Practical virtues include: (1) an inherent large gap between the receiver and the second-stage mirror; (2) an upward-facing receiver; (3) the possibility of compact units (large rim angles), i.e., low ratios of total depth to total width; and (4) no chromatic aberration. We describe how one can tailor both the primary and secondary mirrors so as to insure that spherical aberration is eliminated in all orders, and circular coma is canceled up to first order in the angle subtended by the radiation source. An illustrative solution that attains about 93% of the thermodynamic limit to concentration is presented for a far-field source, as is common in solar energy and infrared detection applications. Double-tailored imaging concentrators are similar in principle to complementary Cassegrain concentrators that comprise a paraboloidal primary mirror and the inner concave surface of a hyperboloid secondary reflector, but have monotonic contours that are substantially different with far superior flux concentration.
The problem of designing a reflector to distribute the illumination of a nonisotropic point source on a plane aperture according to a pre-assigned pattern is analyzed. An integral equation and equivalent partial differential equation are derived. The form of the latter reveals this reflector-design problem to be a singular elliptic Monge—Ampère boundary-value problem.