Professional Documents
Culture Documents
Aop 10 2 484
Aop 10 2 484
Received February 21, 2018; revised April 27, 2018; accepted April 30, 2018; published May 24, 2018 (Doc.
ID 306663)
Nonimaging optics is the theory of thermodynamically efficient optics and as such, de-
pends more on thermodynamics than on optics. Historically, nonimaging optics that work
as ideal concentrators have been discovered through such heuristic ideas as “edge
ray involutes,” “string method,” “simultaneous multiple surface,” and “tailored edge ray
concentrator,” without a consistent theoretical definition of what “ideal” means. In this
tutorial, we provide a thermodynamic perspective of nonimaging optical designs to shine
light on the commonality of all these designing ideas, or what “ideal” nonimaging
design means. Hence, in this paper, a condition for the “best” design is proposed based
purely on thermodynamic arguments, which we believe have profound consequences.
Thermodynamics may also be the most intuitive way for a reader who is new to this subject
to understand or study it within a certain framework, instead of learning from sporadic
designing methodologies. This way of looking at the problem of efficient concentration
and illumination depends on probabilities, the ingredients of entropy, and information
theory, while “optics” in the conventional sense recedes into the background. We attempt
to link the key concept of nonimaging optics, étendue, with the radiative heat transfer
concept of view factor, which may be more familiar to some readers. However, we do
not want to limit the readers to a single thermodynamic understanding of this subject.
Therefore, two alternative perspectives of nonimaging optics will also be introduced
and used throughout the tutorial: the definition of a nonimaging optics design according
to the Hilbert integral, and the phase space analysis of the ideal design. The tutorial will be
organized as follows: Section 1 highlights the difference between nonimaging and
imaging optics, Section 2 describes the thermodynamic understanding of nonimaging op-
tics, Section 3 presents the alternative phase space representation of nonimaging optics,
Section 4 describes the most basic nonimaging designs using Hottel’s strings, Section 5
discusses the geometric flow line designing method, and Section 6 summarizes the various
concepts of nonimaging optics. © 2018 Optical Society of America
OCIS codes: (080.4295) Nonimaging optical systems; (080.4298) Nonimaging
optics; (220.4298) Nonimaging optics; (080.2175) Etendue; (000.6850)
Thermodynamics; (010.5620) Radiative transfer
https://doi.org/10.1364/AOP.10.000484
1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486
1.1. Sphere Ellipse Paradox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
2. Thermodynamic Origin of Nonimaging Optics . . . . . . . . . . . . . . . . . . . 487
2.1. Maximize Radiation Flux at the Absorber . . . . . . . . . . . . . . . . . . . 488
Tutorial Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics 485
1. INTRODUCTION
Nomenclature
px optical momentum according to axis x
L directional cosine x
M directional cosine y
N directional cosine z
E étendue
Pij probability of ray from i arrives at j
C concentration ratio
J~ geometric flux vector.
The difference between imaging optics and nonimaging optics depends on whether
ray sources are treated as points or extended objects. The latter have temperature and
entropy and therefore we can apply thermodynamics, while point sources do not
because they do not possess any internal degrees of freedom.
Conventional imaging optics is most concerned with the idea of point to point
mapping of an optical system. A point from the object space is ideally mapped
via an optical device to a corresponding point in the image space. For example, a
Gaussian optical system starts with mapping the center point of the object to its image
on the optical axis, and then the aberrations of the off-axis points are minimized by
optimization. Figure 1 shows a comparison between imaging (left) and nonimaging
(right) optics. A “perfect” concentrating imaging system free from aberration cannot
be achieved without infinite degrees of design freedom ([1], Appendix B).
Nonimaging optics, however, starts with the idea of energy transfer. For this purpose,
only the boundary of the object must be considered for the designing process. The
mapping of points inside such a boundary is not required. In a sense, nonimaging
optics design only guarantees that the edge rays, or the boundary consisting of ex-
treme positions or directions, is carried over. The contents within the edge rays, or the
rays of the positions and directions in between, can be scrambled.
Figure 1
Figure 2
which will have impacts on the thermodynamics of the system, are not our concern.
For the readers who are interested in a more general analysis for effects such as
luminescent concentrators, please refer to [3]. Compared to imaging optics, a non-
imaging concentrator is concerned with transferring energy from the source to the
absorber (sink) within the theoretical limit of thermodynamics. Instead of designing
an optical system based on fine tuning and engineering light rays, we first need to
look at the maximum concentration ratio allowed by the fundamental laws of physics.
To do so, we must enter the regime of geometrical optics and probability.
The general setup of a concentrating system requires three components, as shown in
Fig. 3: the source of radiation, the aperture of the concentrator, and the absorber (sink)
of the radiation. We define the geometric concentration ratio as C Area Area of the aperture
of the absorber.
To explore the thermodynamic limitations of a concentrator design using geometrical
optics, we assume that the source and absorber are perfect Lambertian surfaces
(blackbody).
The probability of radiation leaving one surface and arriving at another surface is
defined as
radiation from i to j
Pij : (1)
total radiation emitted from i
Starting from this simple concept, we will describe nonimaging optics with two
thermodynamic arguments.
Figure 3
Equation (3) represents a fundamental concept: given two blackbodies at the same
temperature, the radiative power from one to the other is equivalent.
From here, we derive the radiative heat flux at the surface of the absorber:
Q13 A1 P13 4
q13 σT 1 P31 σT 41 : (4)
A3 A3
Because P31 ≤ 1, q13 reaches a maximum radiative flux equal to that of the source,
which is consistent with the second law of thermodynamics:
Equation (6) provides the limit of concentration given the additional requirement that
the optical efficiency of the device is also ideal. Such a requirement is not always
enforced in the real-world application; theoretically, however, providing a device with
the ideal design is at least an option to be considered.
2AD − AC DE 1 1
P21 sinθ, C max : (10)
2CD CD P21 sinθ
Equation (10) gives the maximum ratio of concentration within the framework of
thermodynamics, for an infinitely far away radiation source in 2D. One can easily
generalize this limit to 3D:
1 1
C max ,3D : (11)
P21 sin2 θ
Here dA1 , dA2 are two infinitely small areas on the surfaces of two objects undergoing
radiative heat transfer (Fig. 5). Variables β1 and β2 are the angles formed between the
direction connecting them and the norms of the specific surface, and r is the distance
between them. Because we are in a geometric optical setup, some of the rays from the
source of radiation to the sink can be reflected or refracted. Probability P instead of
view factor F is used to include these rays. In other words, compared to the definition
of view factor F, which only describes geometric configurations without considering
optical devices, probability P is a more general description of radiative heat transfer
which includes ray paths that connect the source and the absorber through optical
devices such as mirrors and lenses.
Figure 4
This result connects the étendue with view factor analysis, which is an important con-
cept in radiative heat transfer studies, and its tabulated analytical results are widely
available in the literature [5]. Notice that for nonimaging optical system designed in
2D, E 12 2A1 P12 due to the integral in Eqs. (12) and (13) being limited in 2D.
The second law of thermodynamics forbids a colder body to transfer heat to a hotter
body. Therefore [also from Eq. (12)], this can be described from the étendue perspec-
tive as
We can implement Eq. (16) with Eqs. (6) and (7) and conclude that the shorthand
expression of the two important thermodynamic arguments for nonimaging optics
is [10]
Equation (17) shows that the ideal nonimaging optical designs are geometrically
étendue matching devices. If we treat étendue as the ensemble of all the rays connect-
ing two objects (such as radiation source, aperture, or absorber), then we can observe
that for the nonimaging geometric optical design that achieved maximum concentra-
tion ratio with theoretical 100% optical efficiency, Eq. (17) is required from a thermo-
dynamic point of view.
Figure 5
The conservation of étendue can be derived from Hamiltonian optics; other forms of
proofs have also been offered ([8], Appendix A.2, A.3). In Appendix A we also offer
Figure 6
a proof due to Fermi (Appendix A, or [13], page 34). For a system described in Fig. 7,
it can be visualized as follows. First, we choose an arbitrary position on the z axis.
The rays of an optical system that intersect a screen at this position [Fig. 7(a)] can be
represented, one to one, as points in a phase space with x, px coordinates [Fig. 7(b)].
As we move the screen along the z axis, the boundary of such a phase space repre-
sentation of all the rays in the optical system will also shift its shape. However, the
area, or étendue, within this boundary will remain constant. In an analogy, if we treat z,
or the position of the screen as time, then as we move away from the screen (changing
the time z), the étendue of the system is like the volume of a noncompressible fluid.
As the fluid takes on different shapes, its volume will remain the same.
Figure 7
Figure 8
lines. Then we can always find a new boundary with vertical line x x0 (blue dashed
line) that allows part of the phase space area outside the new boundary to be folded
inside, resulting in a smaller absorber area, where the new physical size of the absorber
starts with x x0 . In other words, having a nonvertical boundary for phase space
volume means that certain small areas on the absorber are under-utilized. Or, some
of the angles for such small areas are not being used to receive incoming radiation.
In such a situation, one ends up “wasting” the phase space volume that the absorber
has for accepting incoming light.
A quick calculation shows that the size of the new absorber is 2a sinθ; therefore, the
ideal concentration ratio is 1∕ sinθ, which agrees with Eq. (10):
1
C max : (19)
sinθ
Here the phase space “volume” is calculated based on the four-dimensional coordi-
nates x, y, px , py . For any point x, y, the set of rays coming from the source looks
the same:
Figure 9
When the aperture is illuminated with identical pencil rays, in other words, px , py are
independent according to x, y [e.g., Fig. 10(a)],
Z Z Z Z Z
étendue dpx dpy dxdy dpx dpy dxdy A dpx dpy n2 A dLdM:
(23)
Here A is the area of the aperture. L and M are directional cosine x and directional
cosine y, respectively.
As we mentioned before, although it is hard to visualize such a four-dimensional
volume, for certain configurations, such a calculation can be greatly simplified.
For example, given an aperture receiving light from an infinitely far away source,
here we can pick any point (x, y) on the receiver, and the set of rays going through
a small area dxdy will not be dependent on the (x, y) position [Fig. 10(a)]. We can
visualize the 4D phase space of such an aperture by projecting it on the 2D LM
(or, px py ) space. The phase space projection into LM space in such an example will
be of a circular area of radius sinθ, as shown in Fig. 10(b). The phase space volume
is therefore πA sin2 θ.
R
We can also calculate the dLdM using the unit sphere method [5]. It is crucially
important to understand that the directional cosine space is different from the 2D
phase space. Although they are both two-dimensional, one represents the directional
x and directional y (L and M , or Px Py ), which is a projection of the full four-
dimensional phase space. The other is the full representation of x and Px under a
2D configuration.
Figure 10
(a) 3D configuration of light coming into an aperture A subtending angle θ. (b) The
L∕M space representation of the incoming rays.
496 Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics Tutorial
concentrator, it does provide a theoretical limit and some intuitions for the designing
process. It also serves as a guideline on how nonimaging optical systems can be
designed, as shown below.
Figure 11
Figure 12
Mapping these two groups of parallel rays into two points will result in two parabolic
curves (Fig. 13). We determine the aperture AA0 when the tangent of parabola becomes
vertical. Because under this condition the aperture of the concentrator will be at a
maximum. This, however, is only one way of designing optics that satisfy both
thermodynamic requirements of Eq. (7).
CA AB0 A0 B BB0
and
A0 B AB0 :
Therefore,
CA BB0 :
AC sin θ · A0 A,
A0 A 1
0 :
BB sin θ
Figure 13
the incident angle, we plot the direction according to the sine of the incident angle, or
px n sinincident angle, where n is the index of refraction. We also call px the
optical momentum or, when the refractive index is 1, directional cosine. The incident
angle of ray 1 is θ, while as ray 2 is at −θ. Therefore, they are plotted as points 1 and 2
in the phase space. The same can be done for rays 3 and 4, which are represented as
points 3 and 4 in the phase space plot. All the other rays incident at point A0 have the
same x coordinates, with different values for px between sinθ and − sinθ. As a
result, they will form a line between points 1 and 2. As we plot these phase space
points one by one based on their position of x and directional cosine px , we find that
this “cloud” of points occupies an area shaped as a rectangle. The same plot can be
created for the absorber. In this case the extreme rays will intersect with the absorber
at angles of 90 deg; therefore, their px coordinates correspond to 1 and −1, corre-
sponding to the limits we saw for L in Fig. 8(b).
Equation (24) shows that, from the phase space conservation perspective, the CPC is
achieving the ideal concentration limit.
Figure 14
absorber. We call such rays the “edge rays.” When plotted in phase space, they are
represented by the boundaries of the phase space volume (area for 2D).
We use the phase space representations of an optical device to prove this concept,
which is similar to the Liouville theorem in classical mechanics [16]. As shown
in Fig. 16, it is impossible for a ray to retire from its position as a boundary point
in its phase space representation. For example, if ray 1 at the boundary (red square)
wants to become a ray that occupies a position in phase space located internally, the
following must happen: ray 1 would need to select one of its neighboring points
(for example, the blue square) to exchange positions. However, the Liouville theorem
requires that the phase space volume occupied by all the rays of the system to have
no source or sink. In other words, there is no generation, vanishing, or merging of
the little squares representing each ray.
In order for blue and red squares to interchange their positions in phase space at one
point, they must be compressed, as in Fig. 17, and eventually merged (the green square
in Fig. 18). However, merging the area of the phase space representations of any two
rays will result in these rays then processing at the same direction and position (green
square), which renders them indistinguishable from then on. This makes it clear that
two separate rays in a system cannot during their propagation within the optical sys-
tem acquire both the same position and same direction, or occupy the same phase
space area. This proves that the edge rays cannot interchange their positions with
internal rays, they are stuck as edge rays. Any edge rays from the absorber must
always be mapped to the edge rays of the radiation source, and vice versa.
Figure 15
Figure 16
Ray 1 cannot (red square) interchange its position with a ray (blue square) that is not
originally inside the boundary.
500 Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics Tutorial
Figure 17
In order for the ray 1 to interchange its phase space position, it will have to merge
with a small phase space that was originally not on the boundary. This violates the
Liouville theorem.
Figure 18
Merging of the edge ray with an internal ray in their phase space representations.
Similarly, it violates the Liouville theorem.
Tutorial Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics 501
Figure 19
bb0 A2 2A2 1
C 0 C max : (28)
cc A3 A4 A5 − A6 − A7 P21
This method is not limited to cc0 being a straight line; with cc0 being any convex shape,
a similar process will also produce ideal nonimaging optics [19]. An example can be
found here [20].
Therefore, we can arrive at the relationship between étendue and flow line vector as [9]
prescribes (eq 5.24),
Z Z Z Z
E dpx dpy dxdy dpx dpz dxdz dpz dpy dzdy J~ · d~s, (30)
which is also a Poincare invariant [19,26]. Here d~s represents the differential surface
area with its direction pointing at the surface normal, and E is étendue. We establish
here that the étendue between two geometric surfaces is the flux of the geometric flow
line going through such an area.
For a two-dimensional configuration as shown in Fig. 21, AB is a blackbody radiation
source. The J~ is a vector with a magnitude jJ j sinθ1 −θ
2 , and a direction of
2 θ1 θ2
2 ,
or the direction bisecting the two extreme rays.
Figure 20
Figure 21
Figure 22
in Fig. 24. The flat radiation absorber is extended with two “shadow lines” from the
source, which extend to infinity at the acceptance angle (e.g., line PR). Such a shape
for the flow line generator can be shown to work as the following:
(a) At the source edge, the flow line at P will bisect the direction of PA and PR.
(b) As P moves away from the source, such a bisecting feature is maintained, effec-
tively tracing out a tilted parabola with its focus at A.
(c) The process is continued until it crosses the dotted line.
As we observe such a process, we can see that the flow line design is capable
of transforming the view of any point within the concentrator to be seeing the
whole “mountain top” even if we only keep the flow line source AP. The phase
space that AP occupies for any point is effectively the same as the phase space
the mountain top shape occupies. This setup does not, however, work for all 3D
configurations [8].
Figure 23
Figure 24
“Mountain top” shape as the source for flow line that generates CPC.
Tutorial Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics 505
A2 1
C : (33)
A3 sin 2 1 sin θ1 θ2
θ −θ
2 2
Figure 25
Figure 26
where n is the local refractive index, k~ is a unit vector along the ray direction at the
current point, and d~s is an element along the path P1 P2 . Thus, I1, 2 is simply the
Figure 27
optical path length along any ray between the wavefronts that pass through P1 and P2 ,
so it is independent of the form of the path of integration. We can now use this to find
the étendue of the beams in Fig. 27. The Hilbert integral from A to B for the pencil is
seen from (1) to be
Z B
I α AB n sin ϕds, (35)
A
where ϕ is the angle of incidence of a ray on the line element ds. Thus,
where h i denotes the average and LAB is the length of the curve from A to B. It follows
that provided the aperture AB is filled with rays from the line of point sources just
indicated, the étendue is simply
But
where the square brackets denote optical path lengths along the rays so that we obtain
for the étendue
Now let there be some kind of system constructed that achieves the desired transfor-
mation of incident extreme pencils with emergent extreme pencils, as in Fig. 27. The
system takes Σα into Σ0α and Σβ into Σ0β , and we want it to do so without loss of
étendue. We write down the optical path length from Pα to P0α and equate it to that
from Pα to P0α . Similarly, for the other pencil
Figure 28
where BA0 β denotes the optical path from B to A0 along the β ray and similarly for the
other symbols. From these we find
The left-hand side of this equation can be treated as the difference between the
étendues at the entry and exit apertures. Since we require this difference to vanish,
we have to make the right-hand side of Eq. (41) vanish. A simple way to do this would
be to ensure that the optical system of the concentrator is such that the α and β ray
paths from A to B0 coincide, and the same for those from B to A0 . We can do this by
starting segments of mirror surfaces at A and A0 in such directions as to bisect the
angles between the incoming α and β rays. In this way we effectively made the
Pα A and Pβ A the same through the optical system between AB and A0 B0 . In other
words, the two extreme rays became one. We then continue the mirror surfaces in
such a way as to make all β rays join up with the corresponding emerging rays;
in other words, we image the β0 pencil exactly into the β0 pencil, and similarly for
the other mirror surface connecting B and B0 . We have E12 E34 thus completed
the construction and used up all degrees of freedom in doing so.
In Fig. 28 we construct a simplified configuration of Fig. 27 with only mirrors AA0
and BB0 . The result of satisfying Eq. (41) is E 12 E34 , and if 3 and 4 overlaps,
E 34 πA3 . This brings us back to Eq. (17). Notice that Eq. (41) is not the only
way to enable E12 E34 . However, A0 B0 overlapping with Σ0α Σ0β , being either the
same or part of it, is essential to meet the condition E34 πA3 .
6. CONCLUSION
Other designing methods include the simultaneous multiple surface, and tailored edge
ray concentrators have been built based on the generalized method presented in
Subsection 5.3. These methods either directly utilize the edge ray principle as de-
scribed previously, or use a variation of the edge ray concept to form the optical
device. The breakthrough of nonimaging optics compared to the traditional imaging
optical design, however, lies within its capability of connecting the thermodynamic
principles with the designing process through edge rays. This tutorial aims to show
that the intuition of the multitude of nonimaging optics designing methods originates
from a thermodynamic understanding of the optical system. This concept continues
to fuel the development of this field. Therefore, to understand both how a nonimaging
optical design works, and how to develop new ways of such optical designs, one could
rely on the guidance of the thermodynamic understanding of the optical designing
process.
Tutorial Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics 509
This results in ∂H _ ∂H _ ∂H _ ∂H
∂x −px , ∂y −py , ∂px x, ∂py y, which are familiar to
_
Hamiltonian equations.
Notice that ∂_
px
∂px − ∂ ∂H
∂px ∂x − ∂ ∂H ∂_x
∂x ∂px ∂x, etc, which are the Hamilton’s equa-
tions of motion.
~ _x, y_ , p_ x , p_ y , notice that ∇ · W
Now construct a vector W ~ ∂W ∂W ∂W
∂_x ∂_y ∂_px
∂_py ~
∂W
∂_py ∂_
px ∂_x ∂_y
∂px ∂x ∂py ∂y 0. This means the field of four-dimensional vector W
has the important property of divergence being zero. In other words, the four-
dimensional hyperspace of x, y, px , py has the property of conservation of volume
as all the light rays evolve in an optical system.
Here we will also offer an intuitive analogy using an incompressible fluid (Fig. 29).
Let V be the volume surrounded by a closed Surface, where !v is the velocity field of
the small elements within. As the fluid starts to flow according to the change of t,
and the Surfacet starts to evolve into shape Surfacet Δt, the enclosed volume
of Vt Δt will also change as
Vt Δt − Vt ∯ Surfacet ~vΔt · d~s, (A2)
where d~s is the surface vector pointing outward along the normal direction.
Using Gauss’s theorem,
ZZZ
∯ Surfacet ~vΔt · d~s ∇ · ~vdτ, (A3)
FUNDING
University of California (UC) Advanced Solar Technologies Institute.
ACKNOWLEDGMENT
The authors want to thank Aaron Brinkerhoff for the flow line orange cone photo. We are
grateful to our colleagues Bennett Widyolar, Jonathan Ferry, and Jordyn Brinkley, who
provided expertise assisting the research. We also want to thank the anonymous reviewers
who greatly helped to improve the quality of the paper. Last but not the least, we want
to thank Sarah Boyd, who moderated this paper. The research was supported by
University of California Advanced Solar Technologies Institute (ucsolar.org).
REFERENCES
1. W. Welford, R. Winston, and D. Sinclair, “The optics of nonimaging concentra-
tors: light and solar energy,” Phys. Today 33(6), 56–57 (1980).
2. W. T. Welford and R. Winston, “The ellipsoid paradox in thermodynamics,”
J. Stat. Phys. 28, 603–606 (1982).
3. H. Ries, “Thermodynamic limitations of the concentration of electromagnetic
radiation,” J. Opt. Soc. Am. 72, 380–385 (1982).
4. B. Widyolar, L. Jiang, and R. Winston, “Thermodynamics and the segmented
compound parabolic concentrator,” J. Photonics Energy 7, 28002 (2017).
5. M. F. Modest, Radiative Heat Transfer (Academic, 2013).
6. R. Siegel and J. R. Howell, Thermal Radiation Heat Transfer (McGraw-Hill,
1972), pp. 204–207.
7. G. Derrick, “A three-dimensional analogue of the Hottel string construction for
radiation transfer,” J. Mod. Opt. 32, 39–60 (1985).
8. R. Winston, J. C. Miñano, and P. Benítez, Nonimaging Optics (Academic, 2005).
9. J. Chaves, Introduction to Nonimaging Optics (CRC Press, 2015).
10. L. Jiang and R. Winston, “Flow line asymmetric nonimaging concentrating
optics,” Proc. SPIE 9955, 99550I (2016).
11. L. Jiang, “2D Phase space representation example,” https://github.com/wormite/
PhaseSpace/.
12. T. Sekiguchi and K. B. Wolf, “The Hamiltonian formulation of optics,” Am. J.
Phys. 55, 830–835 (1987).
13. E. Fermi, Notes on Thermodynamics and Statistics (Univsersity of Chicago,
1966).
14. H. Ries and A. Rabl, “Edge-ray principle of nonimaging optics,” J. Opt. Soc. Am.
A 11, 2627–2632 (1994).
15. R. Winston, “Light collection within the framework of geometrical optics,” J. Opt.
Soc. Am. 60, 245–247 (1970).
16. H. Goldstein, Classical Mechanics (Addison-Wesley, 1950).
17. L. Jiang, “CPC program for any convex absorber shape,” https://github.com/
wormite/AnyCPC2.
Tutorial Vol. 10, No. 2 / June 2018 / Advances in Optics and Photonics 511