Trigonometry leads to
qf-
=
-- -
--
()
()
()()
()()
()
r
r
rt
nnt
nt
tan
111
11
.2.2
22
22
The intensity at R in the image plane is the reciprocal Jaco-
bian determinant
=
¶
¶
=
-
⎜⎟
⎛
⎝
⎜
⎜
⎛
⎝
⎞
⎠
⎞
⎠
⎟
⎟
()
()
()
()
R
Rr
r
I det , 2.3
rrR
1
in which r(R) is the inverse function of R(r) in (2.1).
In the regime we are considering, r(R) is a singlevalued
function. Physically, this means that h(r) is sufficiently gentle,
and (n − 1)z sufficiently small, that no caustics are generated
between the object and image planes. Alternatively stated, z
does not exceed the principal radii of curvatures of the
wavefront (n − 1)h(r) produced by the surface relief (where
there are caustics, (2.3) still applies, in the form of a sum over
the different solutions r(R), with det(K) replaced by its
modulus |det(K)|).
The first step in deriving the Laplacian image is to
approximate the ray position (2.1) for gentle surfaces:
=+ - () ( ) () ( )Rr r rzn h1. 2.4
Lap
The second step is to calculate the Jacobian in (2.3), also to
leading order, i.e. neglecting terms quadratic in the curvatures
∂
xx
h, etc. Finally, we note that to leading order we can replace
r by R in the Jacobian, obtaining
=- - () ( ) () ( )RRInzh11 . 2.5
Lap
2
This is the Laplacian image, generated by what we are
calling the Laplacian magic window. To confirm that the
minus sign is correct, simply note that
<h 0
2
corresponds
to local convexity, which for refractive indices n>1
increases the intensity by concentrating light. Equation (2.5)
also applies to oriental magic mirrors, with index n=−1,
corresponding to reflection and light is concentrated by local
concavity.
Finding the surface h(R) that will generate a desired
image I
Lap
(R) − 1 involves solving (2.5), interpreted as
Poisson’s equation. This does not determine h(R) uniquely,
because
+++-() [( ) ( )] ( )RhfXYgXYRe i i , 2.6
in which f and g are arbitrary functions, is also a solution and
so would also generate the desired image. One way to resolve
this ambiguity is to exploit the fact that the desired images
are always finite in extent, and specify boundary conditions
on h(R). This can be achieved with standard numerical
Poisson solvers.
3. Illustrations
To explore the accuracy of the simple approximation (2.4) for
the refracted rays, we employ the one-dimensional surface
profile
=++
+-
() ( () ( )
()) ()
hx x x
x
1
100
cos cos 2 3
sin 3 1 , 3.1
and index n=1.5. Figure 2 shows a comparison between the
family of refracted rays calculated exactly from (2.1) and
(2.2), and the rays calculated approximately from (2.4). The
slight differences ( near z=10) are almost impossible to
discern by eye, even though the comparison extends almost as
far as the first caustic.
A more discriminating comparison is between the exact
intensity (2.3) and its approximate counterpart ( 2.5). This is
shown in figures 3(a)–(c) for image planes at three distances
z. As could have been anticipated, the accuracy gets worse as
z increases and the prefocal brightening intensifies. In
figure 3(c), the incipient caustics are clear in the exact
intensity, and smoothed in the approximation. Nevertheless,
the approximation always qualitatively follows the increase
and decrease of the exact intensity.
Also shown, in figures 3(d)–(f), is the intermediate
approximation
=
+-
()
()()
()R
R
I
nzh
1
11
,3.2
Lap0
2
avoiding the expansion of the denominator in the approxi-
mated (2.3) that led to the final Laplacian image (2.5). This
captures the incipent caustics more accurately, with the only
discrepancy being a slight broadening. I have checked that the
minor complication of incorporating this improvement into
the generation of magic windows makes little difference to
the appearance of the images.
With the units in figures 2 and 3 chosen as centimetres,
the window in (3.1), with the prefactor 1 / 100, would be
10 cm wide, and its surface relief would have maximum
height about 0.2 mm and lateral variations on scales of
approximately 1 cm. On such a window the relief would so
gentle as to be barely discernable. These numerical values
were chosen simply for graphical convenience; the magic
windows that I envisage would have relief heights con-
siderably smaller, and the Laplacian approximation would
survive over image distances correspondingly longer than the
0<z<10 cm in figures 2 and 3.
Figure 1. Geometry and notation for refraction by a magic window.
2
J. Opt. 19 (2017) 06LT01