Você está na página 1de 29

ASTRONOMICAL DATA

The astronomical data I have in my disposal are the measurements of the distance modulus
and the redshift of 681 supernova-Ia. These 681 supernova have been chosen of a list of 715
supernova from which have been excluded those with redshift less than 0.01. I did this to avoid
errors due to their extra velocity, other than Hubble velocity. You can see these data bellow.

10

11

12

13

14

15

On column 4 there is the distance modulus . The distance modulus is the difference between the
apparent magnitude m and the absolute magnitude of a supernova (=m-).
On column 5 the dimensionless observational luminosity distance d obv is related to the distance
modulus by
H
d obv = 0 100.2 5
c
H 0 is measured in km/sec/Mpc and c in km/sec.
On column 7 the Gaussian error in d obv is computed by

= d obv 0, 2ln (10 )


On column 8 the simulation program computes the dimensionless theoretical luminosity distance
using the formula

eu
du
H (u )

d the = e
0

This formula comes out as follows

z du
z du
e u
H
H
= (1 + z )
= e
du
dth = 0 dth = 0 c (1 + z )
0 H (u )
0 H (u )
0 H (u )
c
c

a
where we have changed the variable from z to through the equation = ln = ln (1 + z ) .
a0
On column 9 there are the values of residuals Rn , which are given by
Rn =

d nobv d nthe

16

Residuals are very useful because by comparing them with a Gaussian distribution we can test the
goodness of fit of a model to the observational data. I describe below the method of residuals I have
used as more reliable than the very popular chi-squared [1].
THE METHOD OF RESIDUALS [2]
In order to assess the goodness of fit of some model to the astronomical data we will inspect
the residuals. For the true model the distribution of residuals is by definition Gaussian with mean
=0 and variance 2 =1. Consequently, all we need to do is to plot the distribution of residuals in a
diagram and compare it to a Gaussian with =0 and 2 =1. If the diagram exhibits a statistically
significant deviation from the Gaussian, then it is not the true. If there is no significant difference
between diagram and the Gaussian, then this means we probably found the truth. The comparison
of the residuals to this Gaussian will be objectively quantified, by using a modification of the
Kolmogorov-Smirnov test of goodness of fit, generally referred to as the Lilliefors test of
normality. [3]
The table bellow illustrates the test.

On column 3 there are the values of the empirical cumulative distribution function (cdf) of
the sample. The empirical cdf is a step function that jumps up by 1/681 at each of the 681 data
points. On column 4 there are the values of the normal cumulative distribution function. The test
quantifies a distance L (column 8) between the empirical cumulative distribution function of the
sample and the cumulative distribution function of the normal distribution. The lower the value of
L, the better the goodness of fit. A result at the simulation program is acceptable if L is smaller
than a critical value of L ( L Lcritical ) . If we want a result to be 99% true, then Lcritical = 1.035 / f N ,
where f N = ( 0.83 + N ) / N 0.01 and N is the size of the sample. In our case N = 631 so that
Lcritical = 0.0396 .

However a small value of L is not enough. We also need the distribution of the residuals
along the axis to be symmetrical. Lets make it clear by the following example. For two CDM

17

models, the first with 0 = 0.68 and H 0 = 70 , the second with 0 = 0.88 and H 0 = 72 , the
value L for both is L 0.0557 . That means the chances for both to be the true model are equal.
However, comparing the distributions of the residuals we see that the seconds is not symmetrical
along the axis and hence is not acceptable (see figure bellow).

The distribution is symmetrical along the axis when the least square line of the residuals,
R = + b , coincides with the axis. That is, when = 0 and b = 0 .
The fit parameters and b (assuming a straight line fit) are [4]
N

R R
n n

a=

N
N n
n
n
N

2
n

n2 Rn n Rn n
b=

N
N n2 n
n
n
However, there is an uncertainty in and b . It turns out that the uncertainty in is [5]
N
a =
2
N
N
2
N n n
n =1
n =1
and the uncertainty in b is
N

b =

2
n

n =1
2

N
N n2 n
n =1
n =1
The values n are given from the redshift data zn through equation n = ln (1 + zn )
With N=681 we obtain a = 0.1681 and b = 0.0577
Consequently, a model will be acceptable if
L Lcritical = 0.0396
and
a 0.1681
N

b 0.0577

18

NUMERICAL SIMULATIONS

We apply equation
H2 =

0e3 f

f
f
at the simulation program and we give as input data the values 0 = 0.200, 0.201, ...., 0.350 and
H 0 = 69.0, 69.1, ... , 71.0 Km/sec/Mpc.
1+

H 0 is the present Hubbles parameter


0 is the present density in matter
As output data we have the following parameters:
L is the maximum distance between empirical cdf and normal cdf (Lilliefors test of normality).
q0 is the present deceleration parameter
t is the transition time. That is, the when the universe made a transition from deceleration to
acceleration.
zt is the transition redshift. Is computed through equation t = ln (1 + zt ) .

is given by equation m / m and computed via equation = (1 0 ) H 0


All the acceptable results we get are in the table bellow.

19

20

21

22

23

It is obvious from the table above that the decreasing mass model fits to the astronomical data if
H 0 = 70 0.3 km / sec/ Mpc

0 = 0.26 0.02
q0 = 0.62 0.03

t = 0, 44 0.03
zt = 0.56 0.04

= (1.69 0.05) 1018 sec1


Now lets do the same with CDM model, which is the best model we have today to
explain the accelerating expansion of the universe, and compare the two models.
The CDM model obeys the equation
H 2 = 0 e 3 + (1 0 )
We apply this equation at the simulation program and the acceptable results we get are in the table
bellow.

24

25

26

27

28

From the table above we conclude, that CDM model fits to the astronomical data if
H 0 = 69.8 0.3 km / sec/ Mpc

0 = 0.30 0.02
q0 = 0.55 0.03

t = 0,52 0.04
zt = 0.68 0.06

References
[1] http://arxiv.org/pdf/1012.3754v1.pdf Dos and donts of reduced chi-squared. Rene Andrae,
Tim Schulze-Hartung &Peter Melchior. Page 1.
[2] http://arxiv.org/pdf/1012.3754v1.pdf Dos and donts of reduced chi-squared. Rene Andrae,
Tim Schulze-Hartung &Peter Melchior. Page 8: Residuals.
[3] http://www.utd.edu/~herve/Abdi-Lillie2007-pretty.pdf Lilliefors/ Van Soests test of normality.
Herve Abdi & Paul Molin
[4] http://www.physics.buffalo.edu/phy410-505-2010/topic1/app1/app1.pdf Hubbles Law and
Model Fitting. Page 5. Algorithm for fitting to a straight line. Compare equations (10) with
equation (6).
[5] http://young.physics.ucsc.edu/250/lsfit.pdf Physics 250, Least squares fitting, Peter Young.
Page 5. FITTING TO DATA WITH ERROR BARS, see equation (23) on page 6.

29

Você também pode gostar