gravitational lensing · graviitational lensing stephen mullens and simon notley, 2005 history 4...
TRANSCRIPT
Gravitational Lensing
Stephen Mullens
Simon Notley
�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Abstract
Presented here is a detailed consideration of the modelling of gravitational lensing
using a ray-tracing computer simulation. The model discussed deals exclusively
with forward-engineering techniques – constructing the visible image from the
source and the lens. The mechanics of constructing the model are discussed at
length, including an analysis of the effect on image quality of different computational
parameters. Numerous gravitational lensing phenomena are reproduced using
computer simulation, including the Einstein Ring, arcs and lesser distortions due to
weak lensing. Astronomical images of real galaxies are also processed. The results
of these simulation are shown to be similar to observed gravitational lenses. A brief
discussion of the Einstein Cross is undertaken, although the model discussed here
fails to reproduced observed images in this case.
�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
History 4
Theory of Gravitational Lensing 4
Strong Lensing 5
Weak Lensing 5
Microlensing 5
Computational Methods: Simple Model 6
Initial Assumptions 6
The Basic Model 7
Monte Carlo Methods 8
Building the Lensed Image 10
The Einstein Ring 1�
Off-axis and Elliptical Sources 15
Computational Methods: Advanced Model 17
Non point-like gravitational lenses 17
Calculating the mass felt by each ray 18
Lens Pixelation Problems �0
Interpolation Schemes �1
Central defects in lensed images ��
Lensed Images 23
Elliptical Lenses �5
Simulation of Real Life Lensing Observations �6
Conclusions 29
References, Bibliography & Sources 30
�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
History
The first written occurrence of the idea of lights path being bent by massive objects came in 1804 from
a German geodesist, mathematician and astronomer, Johann Soldner. A rough translations of the title of
his article was “On the Deflection of a Light Ray from its Straight Motion due to the Attraction of a World
Body which it Passes Closely’”. This publication pre-empts modern ideas of curved space time by over
a hundred years (see Wambsganss 1998)
In 1916 Einstein published his General Theory of Relativity which outlined the quantitative affect of massive
objects in the fabric of space-time. The first confirmation of his theory came from explaining Mercury’s
perihelion shift, the second came during the solar eclipse of 1919. Arthur Eddington and his team
measured the apparent shift in the position of a star viewed close to the sun’s corona and the shift
matched the result predicted by Einstein to approximately �0%, this has since been re-measured and
is correct to within 0.0�%. The solar eclipse was essential for this observation because it blocked out a
significant amount of light, hence making faint stars visible.
In recent years the gravitational lensing phenomena has been used to infer the existence of black holes
and other massive dark matter objects such as MACHOs (Massive compact halo objects). This is
achieved by observing a stellar object over time and watching to see if its intensity changes, if a good
sized lensing object moves in between the source and observer then the star will wink in and out.
Theory of Gravitational Lensing
A gravitational lens can be described as a natural telescope because they are capable of magnifying and
distorting light from distant objects. A gravitational lens occurs because massive objects bend and distort
space-time around them, light rays passing near follow the straightest possible worldline (path of least
action) which causes them to change direction to an external observer. Gravitational lensing is analogous
to normal refraction of light by a glass lens, as the light crosses the lens surface, the refractive index
changes and its direction of travel is changed. As shown by Carroll and Ostlie (1996), the deviation in the
angle in radians of a ray of light as it passes a massive object is
Where r is the closest distance the ray passes from the centre of mass, M is the enclosed mass and
all other symbols have their normal definitions. From this formula it is possible to analytically solve simple
lensing situations with point mass lenses.
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
Eqn. 1
5
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Strong Lensing
This type of lensing occurs when there is a particularly massive lens with a source close to it. There are
three significant phenomena resulting from strong lensing (Cohn 2000)
1) Multiple images - light from a source is distorted and travels over more than one path resulting in
multiple images of the source forming. Well known multiple images include Einstein Crosses where a �
or 5 multiple images are visible in a cross shape, the lens geometry needed to create this is complex and
not well understood.
2) Einstein Ring - If all objects are in alignment it is possible for an Einstein ring to form whereby a single
object appears as a ring of light on the sky. Einstein rings are often distorted and not perfect rings
because lenses rarely have perfect spherical symmetry and perfect alignment
�) Arcs - this is where a partial Einstein ring forms, this may be due to off axis situations, strange shaped
lenses or other objects interfering. Arcs are the most common phenomena as they can be produced in
many situations.
Weak Lensing
Weak lensing occurs when the lensing is not strong enough to produce multiple images or arcs, but is
capable of distorting the apparent shape of the source, ie stretched and magnified. A possible example
of this effect could be a spherical source appearing to be elliptical on the sky; in reality it is not possible to
use this information to infer the shape of the lens because you never really know the shape of the source.
A method to get a good idea of the distribution of mass in the lens is to find a lensing situation where the
distribution of light coming from the lens can only really infer one type of source which nearly always has
the same shape. If you then know the source and what it looks like after lensing, computer reconstruction
can be used to estimate the mass distribution in the lens.
Microlensing
In this case the lensing is not strong enough to produce multiple images, instead the additional light bent
towards the observer increases the brightness of the apparent source. This can be helpful in viewing
sources which are too far away (faint) to be directly visible.
6
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Computational Method
Initial Assumptions
In the first instance, only point-like lensing objects were considered, although care was taken to adopt a
method that could be simply extended to cover more complex cases. The method used is based on the
ray-tracing model of light propagation and encompasses the following characteristics.
Firstly, rays of light are considered to travel in straight lines, only being deflected by the lens at a single
discontinuity in the plane of the lens and parallel to the observer. The magnitude of that deflection is a
function of the distance from the lens at the point of deflection.
Secondly, the reciprocity of ray paths means that the path of a ray is independent of the direction of
propagation along it. For this reason, we consider paths from the observer to the source. The gain in
efficiency from this can be seen by consideration of a source object emitting light isotropically. Each ray
has a vanishingly small probability of reaching the observer. By considering rays ‘fired’ from the observer,
efficiency can be increased as time is not wasted following rays that can never be observed.
Clearly, it makes no sense to fire beams in all directions from the observer. Instead we limit them to a small
angular range in the direction of the source, thus maximising efficiency. This also has physical significance
because the observer, typically a telescope, can only take in light from a very limited range of angles. The
range of angles used for this is calculated by consideration of the limiting case when a ray fired from the
telescope will only just strike the edge of the source galaxy.
The final in-built assumption is that, when an angle is being used, small angle approximations can be used
to simplify trigonometric functions. The angles of interest in gravitational lensing are of the order of a few
arc seconds or 10-5 radians.
7
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
The Basic Model
The structure of the model devised for this investigation is illustrated in Figure 1. The source is treated as
a two-dimensional image in the source plane. From a computational point of view, this can be either an
imported image or an artificially generated one. The gravitational lensing effect is encompassed entirely in
the deflection angle phi as described by Equation 1.
The dimension xyscale is derived from the relative positions of lens, source and observer as well as the
mass of the lens by the relation
This is the parameter that controls the distribution of rays fired from the observer. Rays are chosen that
pass through a square of side length xyscale positioned in the lens plane. The defining property of this
square is that a ray xyscale/� from the central axis will fall imscale/� from the axis at the source plane.
This assumes an axially symmetrical system; observer, lens and source all aligned. The reason for this
The layout of the model to be used. Rays travel in straight lines between planes and all deflection
takes place in the lens plane. The incidence of a ray or projected ray on the lens, source and im-
age plane is described by the co-ordinates within those planes (xL,yL), (xS,yS) and (xI,yI) respectively.
‘Imscale’ is the physical size of the source image, xyscale is defined below.
Lens Plane Image / Source PlaneObserver
Dol
Dls
imsc
ale
xysc
ale
xI
xL x
S
xI, xS and xL have corresponding y co-ordinates within the relevant plane. When using a point source as the lensing object, it is placed in the Lens Plane at (xg,yg).
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
Fig. 1
Eqn. 2
8
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
choice of parameter is that it ensures most of the rays fired will fall upon the image. Note that there is still
a possibility of rays passing very close to the lens, such that r is very small and the deflection angle very
large. These rays will mostly fall outside the source image.
Monte Carlo Methods
A Monte Carlo process is one which involves using random numbers to quickly find an approximate
answer to a problem. In this instance, rays will be fired at random angles within the established parameters,
an image of what the observer sees will eventually be built up. The level of detail in the image will be
dependent on the number of these random rays used. This may seem an unnecessary sacrifice of
quality for reduced run-time at this stage, but it was realised that these methods would be vital for later
parts of the investigation and as such they were used from the start.
9
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Start
Input source image and imscale,
the mass of the lense,Dol, Dls, xg and yg.
Calculate maximum value of xL that results in a ray from O falling within the source image.
Define xyscale as twice this value
Randomly generate two vectors of length S, scale the values to have
range ±xyscale/2. Call these xL and yL
Choose number of samples
Calculate corresponding (xS,yS) and (xI,yI) and store in vectors
Take the nth vector pair created above, record source intensity
at (xS,yS)
Create new image array of size (1+Dls/Dol)*xyscale. This willbecome the lensed image
n=1
Set element of lensed image array at (xI,yI) equal to recorded
intensity
n<S?
n=n+1
YES
NO
Output lensed image
Add recorded intensityto element of lensed image
array at (xI,yI)
Mapping Addition
End
The operation of the computational model used to map the incidence of the rays on
the three planes and build an image by one of two methods – addition or mapping
Fig. 2
10
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Building the lensed image
Figure 2 illustrates the general structure of the program used to generate the image seen by the observer.
In compiling this image, we considered two possible methods of evaluating the relative light intensity at
each point. We named these methods intensity mapping and intensity addition. The two techniques
utilise the same method to trace the path of the beam through the lens, but differ when it comes to
interpreting this to create the final image.
As shown by the flowchart in Figure 2, the program operates on simple premise. It fires a large number of
random beams from the observer, towards the lens, constrained by the parameter xyscale as explained
earlier. Simple geometry is then applied to calculate the point at which the deflected ray strikes the source
plane and the point in the source plane from where it appears to have come. Note that the latter set
of points are said to comprise the image plane. With these sets of co-ordinates stored in arrays, the
lensed image can be built-up one of two ways. Intensity mapping simply involves mapping the intensity
in the source plane to its corresponding image plane co-ordinate. This is computationally efficient and
quickly creates a clear, noiseless image. However, it does not have the physical realism of the additive
approach.
Consider a lens such that the light from a large but dull star was concentrated into a very small area by
the deflection. One would expect this small area to appear brighter to an observer than the original as
the same amount of light is now coming from a smaller area. The mapping approach does not allow for
this sort of intensity magnification or indeed for any similar effect where more than one pixel of the source
contributes to the brightness of a single pixel in the lensed image. The additive method avoids this by
adding the intensity of the corresponding source area to the lensed image each time a ray is fired. This
makes intensity changes possible at the expense of taking much longer to create a noise-free image.
This reduction in efficiency is a product of the Monte Carlo technique. In the mapping model, since
beams falling on the same pixel over write, the criteria for a perfect image is that each pixel is struck by
at least one beam. In the additive model, two beams striking one pixel will add together. Clearly some
pixels will end up brighter or darker simply due to random coincidence rather than the nature of the lens.
In order to minimise this, we need considerably more beams to be fired. The image will tend towards
being ‘perfect’ as the number of beams fired tends to infinity.
For a 512x512 pixel image, we can estimate the number of samples required for both methods statistically.
As stated above, the mapping method requires each pixel to be hit once. Imagine firing one ray at a
time, each one hitting a random pixel (we neglect the small number of rays not hitting any pixel here) and
somehow marking that pixel. The probability of the first ray hitting an unmarked pixel is clearly 1. The
probability of the second doing the same is �6�1��/�6�1�� and so on. Using these probabilities to
estimate a the number of rays needed to hit an unmarked pixel each time (from 1 for the first pixel through
11
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
to 262144 for the final one) we sum all these numbers to get an estimate of the total number of rays
required. For the 512x512 image, this is ~3x106.
Figure 3 shows the results of using increasing numbers of rays. Note that the image is smooth in the
places that have been imaged, but that it has ‘dead’ white pixels where insufficient rays have been fired.
The additive approach is somewhat more complicated. Here, we need enough rays that the whole of
the image is sampled equally. Clearly this will mean many more rays than before when each pixel had to
be hit only once. The method used to estimate the number of rays required here involves looking at the
distribution of rays over the 51�x51� image. A simple computer program was used to generate random
numbers in the range 1-�6�1��. The frequency of each of these random numbers was recorded. The
aim for an even distribution is to have equal frequencies to within a certain tolerance. The tolerance in this
case is one grey-level in the final image. Using a 64-level image, this means that the frequencies should
be equal to within 1/6� of their value. The statistical condition applied to represent this is that �s=(1/6�),
where s is the standard deviation of the frequency values after scaling such that the mean frequency is
one.
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Images of a galaxy constructed using the mapping method with 5x104, 5x105, 1x106 and
5x106 rays. As predicted, the image is totally smooth in the final image as all pixels are filled.
Fig. 3
1�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Figure 4 shows the reduction on standard deviation as the number of samples is increased. It is clear
that a large number of samples will be required for smooth images. To meet our criteria for totally smooth
images, the number of rays needed would be in excess of 1010. This would take far too long to run,
simply calculating the standard deviation for this case would take around 1� hours on a �Ghz machine.
Some compromise will have to be reached between performance and run time.
Level of Random Error as a Function of Number of Rays for 512x512 Image
0
5
10
15
20
25
30
35
1.00E+06 1.00E+07 1.00E+08 1.00E+09 1.00E+10
Number of Rays
Sta
nd
ard
Devia
tio
n in
Gre
y L
evels
(fo
r 6
-bit g
reyscale
)
The standard deviation of the grey-levels clearly falls as the number of rays is increased.
However the decrease is extremely slow as the error drops to around 1 grey-level.
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
The additive method applied using 5x106, 1x107, 5x107 and 1x108 rays. The final picture took
some time to create and appears very smooth although it doesn’t fit the criteria specified earlier.
Fig. 4
Fig. 5
1�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
The effect of increasing ray numbers in the additive method is shown in Figure 5. Notice how the images
become grainy rather than simply having dead pixels as before. It was decided that the additive methods
was fast enough to be useable in the investigation. However, creating images with more than 1010 rays
was simply not time-efficient, so a compromise was reached and most of the images used in this report
were created using the additive method and 5x107 rays.
The Einstein Ring
For the case where the observer, lens and source are perfectly in line, light rays emitted from the source
are bent in a manner which produces a ring shape from the observers perspective.
Figure 6 shows the path of a ray which travels from the source and is bent just enough by the lens to
strike the observer. If the path of the ray is traced back from the perspective of the observer then it will
appear to come from the top point on the circle. The system illustrated is clearly cylindrically symmetric
about the central axis. This means that the point produced by the single ray can be rotated around the
axis to produce a ring. The ring in this case will have an angular radius of q radians on the sky. This effect
is known as an Einstein Ring, The rays contributing to the ring satisfy the condition
where r is the distance from the origin in the lens plane. This can then be rearranged to find q and the
formula for the angular deflection substituted in.
This formula shows that as the mass of the lens rises, it is possible to produce much larger rings, however
even with massive objects such as black holes, the radius is only of the order of 10-5 radians.
Dls Dol
ObserverSource
φ
θLens Plane
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
The path of a ray from source to observer and the projection of these rays onto the image plane
Fig. 6
Eqn. 3
Eqn. 4
1�
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
The first test of the computer model was whether it could reproduce this phenomenon if given a small,
spherical source galaxy and a point like lensing object, both aligned with the observer. Figure 7 shows the
results. The ring calculated above is shown in red and the ring produced by ray addition perfectly aligned
with it, the soft edges being a consequence of the slightly diffuse source.
It is no great surprise that the program and the calculation on the previous page agree so well, as they
are derived from the same ‘ray-tracing’ model. A better test would be whether the program can produce
similar effects to those observed by astronomers. Nearly all observed rings exhibit two bright spots at
opposite sides of the ring, connected by some kind of arc, as seen in Figure 8. Could the program
reproduce these features?
An Einstein Ring (PKS 1830-211) observed using the Australia Telescope. Notice
the dimension agree well with the model, but the shape is more complex.
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
An Einstein Ring formed with Dol=Dls=1x1025m, M=5x1042kg and a simple spherical
source of ~ 0.5 arcsecs diameter. The analytically calculated ring is shown in red
Fig. 7
Fig. 8
15
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Off-axis objects and elliptical sources
In order to introduce some asymmetry to the lensed image, it makes sense to break the cylindrical
symmetry of the system. The program used for the point-like lens allows the source to be moved to a
position (xg,y
g) within the lens plane. The distance to the lens at each point is modified accordingly and
thus the image produced is altered. It is not immediately apparent what effect this will have and it is not
easy to solve this case analytically due to the lack of symmetry.
Figure 9 shows promising results with clear evidence of the two bright spots observed in real-life images.
In fact the clear presence of these two intense areas allows the formulation and solution of a second
geometrical problem, similar to that of the Einstein ring. It is apparent from the results illustrated in Figure
9 that the axis with the image plane along which the bright spots develop is defined by the direction of
movement of the lens. For example, moving the lens along the horizontal axis results in the bright areas
forming on this axis, one either side of the origin. Given this additional insight it is possible to calculate
analytically the locations of the bright spots. Working within the plane containing the spots – defined as
the plane containing lens, source and observer where these are all point-like – the problem reduces to
that solved earlier for the Einstein ring with a slight modification. The condition which needs to be satisfied
is
where
This reduces to a quadratic equation in q, the solutions representing the positions of the two bright
areas:
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Three lensed images formed with a point-like lens positioned at (1,0), (1,1) and (1,3) in the lens
plane - units are 1020m. Note the bright areas on opposing sides of the origin in each image.
Fig. 9
Eqn. 5
Eqn. 6
Eqn. 7a & 7b
16
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
To plot these, the components xg and yg are used to split r into x and y components. It is interesting to
note that moving the lens galaxy is largely equivalent to moving the source as the system is defined by the
relative position of these element. In fact, the same derivation can be used if the substitution
is made, where (i,j) are co-ordinates of the centre of the source. This is a more convenient formulation for
computational purposes, so the remaining images in this section were generated by moving the source
whilst leaving the lens at the origin.
Figure 10 Shows a series of images depicting a source passing behind a gravitational lens. Note that
the program produces results in very good agreement with the result is derived above. An extra element
of asymmetry has been introduced in this series of pictures by using an elliptical galaxy as the image.
The effect of this is largely as would be expected, it adds an element of ellipticity to observed shapes.
Following the sequence of Figure 10, when the source is far from the lens, it is relatively unaffected, save
for some minor distortions. As it gets closer, it forms arcs and then a ring. Some of the images at this
point are reminiscent of the real case shown in Figure 8.
Although these images are broadly encouraging there are some drawbacks of the point-like lens model.
In an infinitely small lens, the deflection angle will tend to infinity as rays pass close to the lens. This means
r D D
GMrcDD
GMc
ol ls
ls
ol
= = −( )=
=⎛
⎝⎜⎞
⎠⎟⎛
θ φ θ
φ
θ
4
4
2
2⎝⎝⎜⎞⎠⎟ +( )
− = − = −( )
=±
1D D
r r D r D
r r
ls ol
g ol g ls
g
θ φ θ
θ gg
ol
g g g
ol ls
ls ol
C
D
r x y
CD D GM
D D c
2
2 2
2
2
16
−
= +
=+( )
(xx yDD
i j
xyscaleimscale
g gls
ol
, ) ( )( , )
( / )
= − +
=
1
2 ±± + +⎛
⎝⎜⎞
⎠⎟⎛⎝⎜
⎞⎠
( / )imscale DDD
GMcls
ls
ol
2 4 1 422 ⎟⎟
+⎛
⎝⎜⎞
⎠⎟1
DD
ls
ol
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Source passing a point−like stationary lensing object
x − arcseconds
y −
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
An elliptical source passing behind a point-like lens mass positioned centrally. The theoretical
Einstein ring and the calculated locations of the bright spots are shown in red. The source used
originally had its major axis running horizontally, it has been twisted due to weak lensing effects.
Eqn. 8
Fig. 10
17
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
that, however far from the lens the source gets on the sky, there will still be at least one path that a ray
could follow from source to lens to observer. This is what causes the persistent small image close to the
origin in Figure 10. This is an intrinsic weakness of the point-lens model. To advance beyond this stage,
a more sophisticated model allowing for lens objects with physical size must be developed.
Computational Methods: Advanced Model
Non point-like gravitational lenses
So far all the lenses which have been looked at are perfect point like lenses which exist as a single point
on a 2D plane. For each ray the distance between the point lens and is computed and deflections
calculated.
For a lens which is distributed over a 2D plane perpendicular (to the vector source-observer), each ray
feels mass from all over the distribution with the resultant mass in some unknown direction. In real life a
lens would be a three dimensional object with a continuous mass distribution, the lens would also have
absorption and emission characteristics. For the sake of this investigation the lens will be simplified with
a few assumptions and computational simplifications and further complexities added where physically
justifiable and computationally cheap.
Lens mass will be distributed into a pixel structure matrix, that is to say that the lens will consist of a matrix
of pixels, with each pixel having a weight which corresponds to a specific mass. The total mass of the
lens will be normalised to a fix number for each experiment to guarantee consistency.
The lens will not emit any radiation and will be totally transparent to all incoming radiation.
Whether the ray is inside or outside the lensing mass, the position of the centre of mass will always be
the same. The implication of this is that at the start of the program, the centre of mass of the lens can be
worked out and the light rays will always be bent in this direction, but because each ray feels a different
mass, they are bent to a different degree.
18
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Calculating the mass felt by each ray
For simplicity lens distributions will be built in MatLab to ensure maximum control over all the factors such
as position. The lens will be a square n by n array of pixels which will be normalised once created.
Now that there is a mass distribution of pixels, the goal is to work out the net mass felt by each pixel in
that array. For example in the middle of a spherically symmetric distribution of mass, a ray would feel no
force as all the mass vectors cancel, but outside of the distribution the ray would feel a point mass at the
centre of mass of the distribution.
The first step should be to find the centre of mass of the lens distribution by vector summing over all
of the elements. This can be done by giving each pixel an additional weight based upon its x and y
displacement from the (1,1) point. Considering x and y separately, multiply each pixel mass by the vertex
displacement then add them all together, this can be algebraically:
Now that the centre of mass of the lens is known (Cx and Cy) this can be set as xL and yL, the position of
a point lens. The next step is to work out how much net mass is felt by a ray at each point in the mass
array. The expectation is that all rays are lensed towards the same point but can feel drastically different
masses.
For each point on the mass matrix with coordinates (x,y), it is necessary to work out the mass felt. With
d being the distance between the coordinate of interested and a point on the array, the vector mass in
both the x and y dimensions can be calculated using:
Cf i j i
f i j
Cf i j j
f i j
xji
ji
yji
ji
=( )
( )
=( )
( )
∑∑∑∑∑∑∑∑
,
,
,
,
Total Mass== f i jji
,( )∑∑
d i x j y
i xd
Mj yd
ji
xji
= −( ) + −( )( ) −( )
=( ) −( )
=
∑∑
∑∑
2 2
M =f i,j
f i,j
M
y
tot MM Mx y2 2+
Eqn. 9a-c
Eqn. 10a-d
19
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
The vector masses in terms of x and y can then be combined together as the sum of two squares to
find the total mass felt at the point (x,y). When programming this code it may be necessary to restrict the
summations to prevent cases where d is zero, hence causing an undefined result, this occurs when i=x
and j=y.
It is then necessary to carry out this operation on every point on the lens mass distribution array. At this
point it is clear that this is a very computationally demanding process and further experiments will be
required to find a suitably high resolution to give detail, while keeping the program running quickly. To
check this produces accurate results, quick examination of elements in the mass felt matrix for physical
results is useful. At a large distance from the distribution, the mass felt should be the normalised total
mass, whereas in the middle of the distribution the mass felt should tend to zero.
Figure 11 shows some examples of possible lensing masses and their associated mass felt distributions.
The black in the Lens mass distributions indicates the presence of mass, the darker the pixel, the more
mass in the area. The black areas in the Mass felt images indicates the ray feels the total mass of the
lens, as the black becomes more grey it feels less mass until it is white where it feels no mass.
From inspection of figure 11 it is clear that the non point lens method is successful, or at least producing
the expected trend.
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Mass Felt
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass Distribution
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lens mass distributions converted to mass felt distributions using equations 9 and 10
Fig. 11
�0
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Lens Pixelation Problems
A significant problem with using a matrix of mass elements instead of a continuous mass distribution for the
lens is that there will be only a few specific quantised mass states in the matrix. Rays which intersect the
lens plane at similar positions will likely feel the same mass until the difference is significant enough to push
the ray into a different pixel and then there will be a sudden change of mass. These discontinuities will
produce visual defects in the lensed image, to minimise these defects the obvious solution is to increase
the resolution of the lens mass matrix so that there are more discrete quantised states, unfortunately as
the matrix increases, the computational requirements exponentially scale up.
Figure 12 shows the image defect which is produced when using a 16 pixel by 16 pixel lens, where the
limits on the on the x and y axis represent approximately they maximum range of rays which the telescope
can detect. In this manner no rays will exist further out than this image shows. Conversely figure 13
shows that the effect can be diminished by increasing the resolution of the lens to 1�8 by 1�8 pixels over
the same angular size
Some structure is still clearly visible in the 128x128 lensed image but it much less significant. Further
increases in lens resolution are possible and this reduces the size of the square edge enhancement
patterns.
Lensed Image
arcseconds
arcs
econ
ds
0.1 0.2 0.3 0.4
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Galaxy
arcseconds
arcs
econ
ds
−0.15 −0.1 −0.05 0 0.05 0.1 0.15
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
0 0.2 0.4 0.6
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
Galaxy
arcseconds
arcs
econ
ds
−0.15 −0.1 −0.05 0 0.05 0.1 0.15
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
Lens Mass
arcsecondsar
csec
onds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
A 16x16 resolution lens which produces significant patterns in the lensed image due to rounding
numbers - zoomed image
A 128x128 resolution lens showing significantly reduced pixel edge enhancements - zoomed image
Fig. 12
Fig. 13
�1
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Interpolation Schemes
Another method to reduce the pixelation rounding effect is to introduce an interpolation scheme which
estimates the mass felt more precisely by considering where the ray falls into a pixel. Unfortunately this
process can be very costly as it would require approximately at least another 10 operations for each ray,
where the number of rays used in this experiment are of the order of 50 million.
For a linear first order interpolation, the mass felt by the ray is calculated by examining the four nearest
array vertices in the mass felt array and weighing them based on the proximity of the ray to each specific
vertex.
Where the function f(x,y) denotes the mass felt by a particular ray according to first order interpolation, and
f(xi,yj) denotes one of the surrounding pixels. The mass felt can by found by:
This operation requires 10 additions/subtractions and 4 multiplications to find an approximate value for
the mass felt for each ray. Higher order interpolations schemes such as bicubic (order 3) can be used to
further improve the estimate of the mass felt at each point but these require exponentially more operations
to be carried out. Using these interpolation algorithms the lensed images can be improved with a similar
effect that can be achieved by raising the resolution of the lensed image.
f x y f x y
f x y f x y x
, ,
, ,
( ) = ( )+ ( ) − ( )( )
0 0
1 0 0 0 Δ
+ ( ) − ( )( )+ ( ) + ( ) −
f x y f x y y
f x y f x y f0 1 0 0
1 1 0 0
, ,
, ,
Δ
xx y f x y x y
x x xy y y
0 1 1 0
0
0
, ,( ) − ( )( )
= −= −
Δ Δ
ΔΔ
(0,0) (1,0)
(0,1) (1,1)
(x,y)
Diagram to show how each surrounding vertex influences the mass felt by the ray at position (x,y).
Fig. 14
Eqn. 11a-c
��
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Central defects in lensed images
In the model being used it is assumed that the lens is totally transparent. At the centre of a circularly
symmetric lens, the mass felt by a ray is approximately zero and this will produce a slightly distorted image
of the source at the centre of any lensed image. As before the issue of pixelation of the lens will show its
hand here.
Figures 15 and 16, show the diminished angular size of the central region anomaly, as the resolution rises
the central region fixes itself. Physically these rays of light would have definitely been absorbed by the
lensing object and not reach the observer, so it is good that they can be diminished computationally.
A further improvement to the program could be applied whereby rays are actually absorbed by the lens,
and the probability of absorption is related to the mass felt at that point. If a ray felt ~100% of the lensing
mass then it is safe to say it wont be absorbed but alternatively if it only feels �% of the mass it will be in
the central part of the lens and definitely absorbed. Modification of the code to allow this further increase
the computational work needed and does help remove the central defects.
Lensed Image
arcseconds
arcs
econ
ds
0.005 0.01 0.015 0.02 0.025 0.03
0
0.005
0.01
0.015
0.02
Galaxy
arcseconds
arcs
econ
ds
−0.15 −0.1 −0.05 0 0.05 0.1 0.15
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
0 0.02 0.04 0.06 0.08 0.1 0.12
0
0.02
0.04
0.06
0.08
0.1
0.12
Galaxy
arcseconds
arcs
econ
ds
−0.15 −0.1 −0.05 0 0.05 0.1 0.15
−0.15
−0.1
−0.05
0
0.05
0.1
0.15
Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
A 32x32 resolution lens with zoomed in section on the lensed image showing the central low mass
defect.
A 128x128 resolution lens with zoomed in section on the lensed image, note the axis labels which
show how small the defect is in this case.
Fig. 15
Fig. 16
��
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Lensed Images
The first type of non point like lens to be considered is a spherically symmetric lens with a mass distribution
which falls away from a constant value at the centre by 1/radius�.
By changing the size of the lens it is possible to alter the Einstein ring phenomena when compared to total
alignment of a point lens. As the lens expands from 1 pixel (approximate point lens) to a large lens with
the specified profile in figure 17, the Einstein ring collapses inwards. The constant narrow outer line is the
theoretical Einstein ring for a point lens.
Radius
Len
s D
ensi
ty
r2
1
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lens density profile for possible lens type
Lensed images produced by a perfectly aligned spherical source and spherical lens
Fig. 17
Fig. 18
��
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Figure 18 shows how the Einstein rings collapse in on itself as the mass distribution of the lens changes,
note the total mass of the lens is constant over each image. The images in question were created by
simulating fifty million rays of light passing over the area of the sky in a solid angle of approximately two
arcseconds squared, and using a 1�8x1�8 lens matrix.
The collapse of the Einstein ring in this case is due to the rays feeling a smaller mass than for a point lens
situation. In the point situation the rays which are detected are those at ~0.5 arc seconds radius, for
the non point cases rays which pass through the lens at this point will possibly be absorbed, but more
importantly will feel a smaller mass, this in turn means that the radius of the ring must shrink to compensate
for this loss of mass. As the lens continues to expand in size the rays will continue to feel less and less
mass, up to the point where they will enter the constant density region where they will mainly be absorbed
and never seen by the observer.
�5
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Elliptical Lenses
Figure 19 shows what happens as an elliptical lens is moved off centre to produce arcs. In the first image
the lens is exactly central and it is slowly moved off centre in steps of equal x and y simultaneously. This
is not the only possible situation which can produce arcs.
In the first stage of figure 19 there is diminished intensity of radiation at either side of the ring, this is due
to the Lens mass absorbing some of the light which would have formed the ring. The ring is also slightly
squashed in on the sides do to a smaller mass being felt in these areas.
The next stage is to produce lensing images which match observations as closely as possible, the areas
to focus on will be the three major phenomena, arcs, rings and crosses. By replicating observed images
it should be possible to infer the possible lens mass distribution which produced them in the first place.
Typically the source objects being used will be spherical but other shapes will be tested as well because
a lot of objects in space are elliptical.
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcsecondsar
csec
onds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Production of basic arcs and rings using an elliptical lens
Fig. 19
�6
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Simulation of Real Life Lensing Observations
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Figure 20 - Giant arcs
A = Cl2244-02 - A giant arcs viewed via the Hubble
Space Telescope
B = Simulated arc using a circular source galaxy and a
slightly off centre elliptical lens.
A
B
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Figure 21 - Distorted Einstein Ring
A = MG 1654+1346. Point like quasar firing two radio
jets in opposite directions, one is lensed from Earths frame
of reference, the other is not.
B = Simulated distorted ring shape. Note it is very difficult
to establish which lens mass distribution would produce
the required result.
A
B
�7
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Figure 22 - Einstein Cross
A = G2237 + 0305. Einstein Cross
B = Einstein Cross produced via different computer
modelling approach
A
B
Figures 20 and 21 show how basic phenomena can be reproduced using the code created in this
project. In the case of the Einstein ring, the mass distribution was fudged by a trial and error approach
to try to replicate the observed image; many other mass distributions could also have worked. Minor
changes in the lens mass distribution can have a profound effect on the lensing patterns produced and
this made it very difficult to produce an Einstein cross. The Einstein cross shown in Figure 22B was
taken from Newbury (1997) and shows that it is possible to produce an Einstein cross using the basic
assumptions in this report. The key difficultly is finding the correct lens distribution, Newbury (1997) states
it to be an elliptical structure, but does not mention how the mass is actually distributed in this structure.
�8
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Lensing real images of Galaxies
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Figure 23 - Lensing of an elliptical galaxy
A= Source galaxy
B= Galaxy seen after being lensed by a point source with total
alignment of the system
A
B
Lensed Image
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Galaxy
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1−1
−0.5
0
0.5
1Lens Mass
arcseconds
arcs
econ
ds
−1 −0.5 0 0.5 1
−1
−0.5
0
0.5
1
Figure 24 - Lensing of a spiral galaxy
A= Source galaxy
B= Galaxy seen after being lensed by a point source with
total alignment of the system
A
B
Figures 23 and 24 show what happens if pictures of galaxies are lensed using a point lens at the origin.
The fuzziness in figure 23A has the effect of making the Einstein Ring in 23B fuzzy in specific areas. Figure
24A shows a spiral galaxy with a selection of bright objects around it, the spirals in figure 24A produce
small arcs around the Einstein ring in the lensed image of figure 24B, the bright objects occur as dots
around the ring in this image.
�9
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
Conclusion
This investigation has looked into how gravitational lensing can be simulated using Monte Carlo ray
tracing techniques. The computational model was able to reproduce some basic gravitational lensing
phenomena, but producing more complex situations such as the Einstein Cross proved difficult because
they require a very complicated lensing mass. Even the example shown in the previous section of a
program that successfully predicts a cross, doesn’t give the four circular images seen in the G���7 +
0�05 formation.
Some of the subtleties of real life lensing behaviour might be better modelled with a three dimensional lensing
mass rather than a two dimensional projection. A three dimensional lensing mass would allow a source
galaxy to be repeatedly lensed and hence re-allow previously forbidden rays to strike the observer.
The backwards mapping ray tracing method which was used is possibly an overly simplistic approach
to the problem of gravitational lensing, this may have further complicated producing the Einstein cross
phenomena. This method assumes that the probability of a ray striking any location on the telescope’s
CCD is constant over the whole array, this is unlikely to be exactly the case because the lens architecture
may make some locations preferable. It would make more sense to assume isotropic emission from the
source, rather than isotropic rays reaching the observer. However, this would be very difficult to model to
implement with any reasonable degree of computational efficiency.
Despite these limitations and the failure to produce a convincing Einstein cross, the investigation has
shown that many simple lensing effects can be easily modelled with a fairly straightforward ray-tracing
model. It performed particularly well in dealing with point-like and simple diffuse lenses, producing results
similar to those observed by astronomers, and for this reason it can be considered a broad success.
�0
Graviitational Lensing Stephen Mullens and Simon Notley, 2005
References
Carroll & Ostilie,1996, An Introduction to Modern Astrophysics, Page 1�05-1�11, Addison Wesley
Longman
Cohn J, �000. Gravitational Lensing, http://astron.berkeley.edu/~jcohn/lens.html
Newbury P, 1997, http://www.iam.ubc.ca/~newbury/lenses/lenses.html
Wambsganss J, 1998, Gravitational Lensing in Astronomy, http://www.livingreviews.org/lrr-1998-1�,
Max-Planck-Gesellschaft.
Bibliography
Chae, Turnshek, Khersonsky, 1998,The American Astronomical Society. A Realistic Grid Of Models
For The Gravitationally Lensed Einstein Cross (Q2237]0305) And Its Relation To Observational Constraints,
The Astrophysical Journal, �95 : 609 E 616
Ratnatunga et al,1995, New ‘‘Einstein Cross’’ Gravitational Lens Candidates In Hubble Space Telescope
Wfpc2 Survey Images, The Astrophysical Journal, �5� : L 5–L 8.
Sources
All figures used in this report were created by the authors, with the exception of those listed below
Fig. 8 http://www.atnf.csiro.au/people/jlovell/
Fig. 20A http://serweb.oamp.fr/kneib/ast_now96/
Fig. 21A www.mira.org/fts0/s_system/161/text/txt005z.htm
Fig. 22A http://www.laughtergenealogy.com/bin/space/einstein-cross.jpg
Fig. 22B http://www.iam.ubc.ca/~newbury/lenses/mfk/src/mfk128.gif
Fig. 23A http://en.wikipedia.org/wiki/NGC_��1�
Fig. 24A http://www.hero.ac.uk/resources/R_2_Galactic_ghost_300.jpg