Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

STEPS TOWARDS A PERIOD-AGE

RELATION FOR CEPHID VARIABLE


STARS
Calibrating the Period-Age Relation for Cephid Variables
from the Age of Their Associated Cluster

Author: Giorgia Line


Supervisor: Dr David Bersier
Abstract
Cephid variables are blue giant variable stars that are leaving the main sequence. These stars
undergo predictable pulsations with periods that should relate directly to their age, due to higher
mass stars leaving the main sequence earlier, and also having lower densities which cause them to
have longer pulsation periods. A Period-Age relation for these bright stars could allow for age
determination of lone cephids, and age determination for parent clusters without sufficient turn-off
data for isochrone comparisons. Photometry was performed on variable objects located in NGC 1831
in order to locate potential cephid variables and determine their periods. These periods, along with
isochrone comparisons for NGC 1831, were then used to calibrate the Period-Age relation for these
variables. The results were found to agree well with the currently accepted relation of log 𝑇 =
8.50 − 0.65 log 𝑃. Further calibration could be performed with cephid period data from multiple
clusters of varying ages.

Introduction
Cepheids
Cephids are blue giant intrinsic variable stars, which undergo predictable pulsations with a pulsation
period that is directly related to their luminosity. As their pulsations are so predictable, and as the
stars themselves are extremely luminous, cephid variables are a key primary distance indicator for
measuring cosmological distances, as their absolute magnitude can be calculated from their
pulsation period, which is independent of the star’s distance.

Cephid variables are thought to originate from main sequence B type stars, but are now leaving the
main sequence and entering a phase of instability. Higher mass stars leave the main sequence
earlier, and also have lower densities which results in them having longer pulsation periods.
Therefore, there should be a relation between a cephid’s pulsation period and its age.
(Yu.N.Efremov, 2003) The spread of ages between stars in the same cluster is small compared to
their lifetimes, meaning it can be reasonably approximated that all stars in a cluster formed
simultaneously. From this, with an independently measured age for the cluster (eg; from isochrone
comparisons) and a measurement of the pulsation periods of cephids within that cluster, it is
possible to determine the relationship between the pulsation period and cephid age. This age-period
relationship can then be used to determine the ages of clusters with cephids for which turn-off
photometry data is poor, or to age lone cephids as the pulsation period of a cephid is an observable
factor that is only marginally affected by systematic errors. (M. Marconi, 2005).

The most probable empirically determined period-age relation for cephid stars is

log 𝑇 = 8.50 − 0.65 log 𝑃 equation 1

There are disadvantages to using the empirically determined relation; it is calibrated using cluster
cephids, and is therefore susceptible to errors caused by reddening uncertainties, distance
uncertainties, and the range of cephid periods covered by the cluster. The relation also ignores the
effects that chemical composition and temperature can have upon the star’s pulsation period.

The most probable theoretical period-age relation as derived from evolutionary and pulsation
models is between

log 𝑇 = (8.41 𝑡𝑜 8.08) − (1.07 𝑡𝑜 0.39) log 𝑃 equation 2

for First Overtone Pulsations, and


log 𝑇 = (8.49 𝑡𝑜 8.31) − (0.79 𝑡𝑜 0.67) log 𝑃 equation 3

for Fundamental Overtone Pulsations. This range in values is caused by accounting for the changes
due to the chemical composition of the modelled cephids. (G. Bono, 2005) The empirical relation
from Efremov is in good agreement with the theoretical relations for fundamental overtone
pulsations from Bono et.al.

1.2 NGC 1831


NGC 1831 is a large and highly populated blue globular cluster in the Large Magellanic Cloud, with a
binary population of 30% and a metallicity of Z=0.004. The age of the NGC 1831 cluster can be
determined independently by comparison of the present cluster to theoretical stellar isochrones.

Stellar isochrones show the HR diagram of a collection of stars which are all at the same age, as is
approximately true of stars in the same cluster. As different classes of stars leave the main sequence
at different ages, the type of stars currently leaving the main sequence, called the main-sequence
turn-off, indicate the age of the cluster. From isochrone comparisons it is thought to be an
intermediate age cluster of between 0.8 x109 to 1.4x109 years, as the main sequence band
terminates at around M=-0.3 magnitudes, with very little magnitude difference between the red
cluster and the main sequence band, and with red stars scattered towards bluer colours. (R.
Capuzzo-Dolcetta, 2012) Though outside the scope of this experiment, it has been suggested that
NGC1831 may have undergone prolonged star formation in the form of several bursts of activity
throughout its lifetime, so that not all the stars in the cluster can necessarily be assumed to be the
same age. (Vangioni-Flam, 1990)

Image Processing
Determining the period of a variable star requires taking photometry of many images taken on
different nights over a long period of time. These images will therefore be taken under different
observational conditions, may sometimes be taken with different instruments, and so will not be
perfectly aligned. Before image subtraction and photometry can be successful, the images must be
re-aligned both astrometrically and photometrically.

Templates
In order to align the images, templates were first made for both the I and V band images. The images
chosen were the 7 in the V band and 6 in the I band with the smallest PSF and sky noise. It is
important to take the images with the least noise as the goal of the image stacking process is to
increase the overall signal to noise ratio, and so adding in unnecessary noise is counterproductive.
The images chosen for the templates first had to be aligned with each other. This was done by taking
10 of the brightest objects present across all the template images and measuring their centre
coordinates using the IMEXAMINE task in IRAF. The bright objects chosen to be anchor points are
marked in figure 1. It was important that the objects chosen be spread out spatially across the image
rather than clustered in one area in order for the calculated transformation to be reliable across the
whole image. The task XYXYMATCH was then used to produce a file that matched up the different
coordinates for the same objects in each image. This file was then input into the GEOMAP task,
along with the coordinate file from the first image in each band, allowing the task to calculate the
transformation that would need to be performed on each subsequent image in order to bring it into
astrometric alignment with the first. The GEOMAP output file was then input along with the chosen
images into the GEOTRAN task, which performed the transformation and output a series of aligned
images.
To make the coadded templates, a list of these aligned images was then entered into the
IMCOMBINE task, which output the template images for the I and V bands.

Astrometric Alignment
The next stage was aligning all images in each band to the template image. Again, the centres of 10
of the brightest objects present in all good images were measured using IMEXAMINE. At this stage,
several images were discarded as the image quality was so poor due to high levels of noise, which
made it impossible to locate any anchor points within the images. These were the images from
2015/10/29, 2015/11/02, 2015/12/09, 2016/01/11 (0086) and (0087), 2016/01/20, 2016/02/20,
2015/10/29 (0124). An example of the discarded images can be found in Figure 2, below. Once again
the coordinates were entered into the task XYXYMATCH, this time with a coordinate list from the
centres of these objects in the template images being used as reference. The XYXYMATCH output
and the coordinates from the template image were then input into the GEOMAP task to calculate
the transformation needed to bring each of the images into alignment with the template, and this
transformation was then performed for each of the non-discarded images using the GEOTRAN task.
The blink function of DS9 was then used to confirm the images had been correctly aligned, with only
the brightness of the stars changing as the were images cycled through and not their physical
locations.

V I

2015/10/06 0142 2015/10/06 0141

2015/10/06 0153 2015/10/08 0152

2016/01/09 0132 2016/01/09 0131

2016/01/11 0159 2016/01/11 0158

2016/01/30 0124 2016/01/30 0125

2016/02/09 0099

2016/02/09 0100 2016/02/09 0101

TABLE 1
FIGURE 1 FIGURE 2

Photometric Alignment
In order for useful photometry to be performed on the images, they must be brought into the same
photometric system, where any changes in brightness across the images are due to intrinsic
variation in the star and not caused by variation in observation conditions on different nights.

HOTPANTS, aka High Order Transform of PSF ANd Template Subtraction, was then used to perform
PSF matching and image subtraction on the aligned images (Becker, 2013). PSF matching is
performed by calculating a convolution kernel K for a pair of input images T, the template image,
and I, the image to be measured. K is a gaussian which can be convolved with the gaussian PSF of the
objects in I in order to match the gaussian PSF of the objects in T. K varies spatially throughout the
images, and so the image is split into multiple regions or ‘stamps’ by the program and K is calculated
for each stamp in order to minimize error caused by this spatial variation in the gaussian fit applied
to each object. There is generally one stamp per 100 pixels. When PSF matching is performed on
multiple images to bring them into photometric alignment with the same template, it also brings
them into photometric alignment with each other, overcoming the problem of them being taken
under different conditions.

Once the images are brought into photometric alignment, HOTPANTS then performs image
subtraction on the input images in relation to the template. A shell code was written to run each of
the aligned images through HOTPANTS along with their respective template in each band. These
shell codes can be found in Appendix 1. This process leaves behind only the objects that vary in
brightness in comparison to the template, with objects that remain the same throughout the
observation period removed from the field. However, image subtraction also introduces noise to the
image, and can lead to some spuriously low values at some points in the image, which can lead to
objects being missed in later photometry.

Photometry
Variable Detection
Before photometry could be performed on the variables, it was first necessary to locate them in the
subtracted images. To improve the detectability of the variable stars, the task IMEXPR was used on a
few good-quality subtracted images to make every pixel positive via the absolute value function. At
this stage the subtracted images for 2015/12/09(0102), 2016/01/08(0103), 2016/01/30(0124), and
2016/02/10(0103) in the V band were removed from consideration, as an undetermined error
during psf matching and image subtraction lead to any stack containing these images returning
absurd and inconsistent results when attempting to determine parameters for the DAOFIND task.
The absolute value images were then stacked, and the task DAOFIND was run on the stacked image
for each band.

Preliminary measurements were taken via IMEXAMINE on the stacked image, and those entered into
DAOFIND are shown in Table 2. The detection threshold was decided via trial and error until most
detections were real stars. It is possible that some of the detected variables are not in fact
associated with the cluster, but are simply within the same field of view. Dealing with this however is
outside the scope of this investigation.

The TVMARK task was then used to mark each of the detected objects on the original non-
subtracted template for each band. Any ‘detection’ that was not a real star was then manually
removed from the coordinate list through this task. The remaining coordinate lists could then be
used to perform photometry on the detected variables.

V I

FWHM 3 3

Standard Deviation of Sky 0.3 0.3

Detection Threshold 5 5

TABLE 2

Photometry
The PHOT task was used to perform photometry at the coordinates found in DAOFIND. There were
40 variable objects located in the V band but only 14 located in the I band. As the PHOT task will
return strange results if the flux in an image is negative, it was necessary to create an inverted copy
of each subtracted image and perform photometry at the relevant coordinates on both images. This
was done by using the IMARITH task to calculate a new fits image with pixel values equal to 0-
{image}. The measured fluxes were then extracted from the output files and sorted to produce files
containing the flux measurements for each detected variable across all the subtracted images.

The average standard deviation on the flux measurements from PHOT was 1.5 counts, varying 0.5
counts around this value with no apparent pattern across images and objects.

Data Analysis
Period Determination
In order to calibrate the period-age relation using the variables located in NGC 1831, it was
necessary to fit a pulsation periods to each of the detected variables. Periods for the variables were
determined using Phase-Dispersal Minimisation via the PDM task in IRAF.

The PDM task was used to calculate the probable period of each detected variable. PDM, or Phase
Dispersion Minimisation, is a least-squares fitting method that instead of fitting to a given curve such
as a Fourier component, fits to the mean curve defined by the means of each bin. This
simultaneously fits for the optimum light curve and optimum period. (Stellingwerf, 1978)

The PDM task is known to produce some strange-looking curves for fits with a low number of data
points, where a low number is <20. As only the I band had more than 20 data points remaining, and
even then only had 21, it was expected that the task might not perform as well as desired. However,
the help page for the task states that the periods fitted are still generally okay. Ideally the process
would be carried out again with more data points, and preferably more regular sampling.

The periods determined for each detected variable in days are shown in table 3 for the V band and
table 4 for the I band. However, the significance values given by PDM for these periods are all ~1,
meaning these are likely not the true periods. The task was performed with the full continuous data
set (top line), and then again using the auto-ranges option to account for gaps in the data by finding
the period on ranges decided by the program instead (bottom line). Changing this did not alter the
periods significantly for most of the variables in the V band, though did result in smaller periods for
most of the I band, and did nothing to improve the significance of the results.

V1 2.35864 V11 14.04043 V21 45.30548 V31 2.623696

2.76524 13.97175 45.47931 6.361251

V2 31.20824 V12 12.96258 V22 33.80322 V32 3.424074

31.40580 3.327633 33.66360 4.292282

V3 62.36003 V13 9.932573 V23 51.75658 V33 6.150754

61.44225 15.32682 52.25350 3.650366

V4 9.932573 V14 6.841956 V24 8.466111 V34 8.466111

41.45843 3.485267 14.29886 3.104451

V5 41.82749 V15 14.80833 V25 2.35864 V35 51.75658

41.45843 14.63361 14.2989 3.033438

V6 14.04043 V16 3.516464 V26 9.932573 V36 2.35864

14.29886 18.02204 9.874095 4.70857

V7 20.3827 V17 9.932573 V27 3.611346 V37 124.6073

10.3418 9.874096 13.97176 81.10921

V8 8.466111 V18 22.67324 V28 106.21 V38 60.72162

7.654988 22.71467 112.1434 61.44226

V9 138.6102 V19 14.41927 V26 45.30548 V39 4.468614

81.1092115, 13.97175 45.47931 4.495612

V10 4.970781 V20 45.30548 V30 9.932573 V40 6.487156

4.931622 46.54405 6.662589 62.88071

TABLE 3
I1 3.516464 I8 16.47245

3.327634 3.327634

I2 2.918538 I9 11.65305

2.896236 8.205316

I3 4.970781 I10 142.3502

2.520766 67.40128

I4 4.713014 I11 98.05657

2.896236 2.520766

I5 2.296671 I12 12.96258

2.964041 2.765244

I6 7.610831 I13 4.713014

4.600860 4.495612

I7 3.611346 I14 60.72162

3.735827 56.01008

TABLE 4

Cephid Identification
No variables were found with periods less than a day, which would have marked them as not being
Cepheid variables. The upper limit on Cepheid variable periods is less clear from the literature, with
different sources citing anything from 40 to 100 days. Still, it seems safe to conclude that the
variables with periods greater than 100 are also not cepheids and should be discounted from the age
determination process.

The way to determine which objects are likely cepheids is to study the shape of their light curve, as
Cepheid light curves have a distinctive shape shown in figure 3:
FIGURE 3, (L. INNO, 2017)
However, the light curves produced are eclectic and it is hard to determine whether any have the
desired shape.

The objects whose light curves appeared to best fit the expected shape of cephid variables were are
given in table 5 along with their periods.

V2 31.40580 V26 9.874095 V34 3.104451 I1 3.327634

V6 14.29886 V29 45.47931 V35 3.033438 I3 2.520766

V12 3.327633 V30 6.662589 V36 4.70857 I13 4.495612

V17 9.874096 V32 4.292282 V38 61.44226

V23 52.25350 V33 3.650366 V39 4.495612

TABLE 5
Those with periods greater than 10 days appear to be outliers and are possibly cephids not truly
associated with the cluster but simply in the same field.
The light curves and attempted fits for V36 and I3 are shown in figures 4 and 5 below.

FIGURE 4

FIGURE 5
500
400
300
200
100
0
Flux

2457280 2457300 2457320 2457340 2457360 2457380 2457400 2457420 2457440


-100
-200
-300
-400
-500
JD

FIGURE 6
Figure 6 shows the flux data from V36 fitted with the period and amplitude calculated by the PDM
task.

Calibration
According to the isochrone comparisons from Capuzzo-Dolcetta et. al., NGC 1831 has a dominant
age of the order of 109 years. Taking the age of the cluster as 109 years and assigning this age to the
cephids located in the cluster, according to the current relation it would be expected to find cephids
with periods of approximately 5.878 days. The average period of probable cephids in the cluster is
4.618 days, which is a fair agreement with the current relation.

With data for only one cluster, and therefore cephids of only one age, it is not possible to plot a
graph of age against period in order to determine a new empirical relation. Ideally, this process
would be repeated on many clusters with independent isochrone data in order to calibrate the
equation further. With what is available, it is possible only to say that either the intercept (8.50)
should perhaps be higher and/or the gradient (-0.65) should be higher, and due to the lack of
reliable significance for the calibrated periods it is not truly possible even to say that.

Conclusion
Cephid variables already play an important role in determining the cosmic distance scale due to their
high luminosities, and to their absolute magnitude being directly linked to their pulsation periods,
which are measurable independent of distance. These same variables may also play an important
role in determining the cosmic age scale, as their age is also related to their pulsation period,
allowing them to be used to determine the ages of parent clusters without sufficient main sequence
turn off photometry data for isochrone comparisons, and also for determining the age of lone
cephids that do not have an associated parent cluster to estimate their age from.

The likely cephids located in this experiment had an average age of 4.6 days, which is not far from
the expected value of 5.9 days obtained for a cluster aged 109 years using the current relation. With
data for only one cluster, and therefore only one age measurement, it is not possible to truly
calibrate the relation.
Repeating this process with other clusters of different ages, for which turn off photometry is
available, would allow the relation to be calibrated further.

References
R. Capuzzo-Dolcetta, C. C. (2012). Physical Processes in Fragmentation and Star Formation. Physical
Processes in Fragmentation and Star Formation (p. 431). Rome: Springer Science & Business
Media.

Vangioni-Flam, E. (1990). Astrophysical Ages and Dating Methods. Astrophysical Ages and Dating
Methods (p. 221). Paris: Editions Frontires.

Becker, A. (2013). High Order Transform of Psf ANd Template Subtraction code (hotpants). Retrieved
from https://github.com/acbecker/hotpants

G. Bono, M. M. (2005). Classical Cephid Pulsation Models. X. The Period-Age Realation. The
Astrophysical Journal(621), 966-977.

L. Inno, e. a. (2017, 09 15). Studying Cephied Light Curves. Retrieved from astrobites:
https://astrobites.org/2014/12/15/studying-cepheid-light-curves/

M. Marconi, G. B. (2005). Classical Cephids as Age Indicators. Memorie della Società Astronomica
Italiana, 75, 282-285.

Stellingwerf, R. F. (1978). Periodd Determination using Phase Dispertion Minimisation. The


Astrophysical Journal, 224, 953-560.

Yu.N.Efremov. (2003). Cephids in the LMC Clusters and the Period-Age Relation. Steinberg
Astronomical Institute Astronomy Reports, 47(12), 1000-1012.

Appendix 1
Hotpants Shell:
while read VList;

do

./hotpants -inim "${VList}-A.fits" -tmplim TemplateV.fits -outim "${VList}-S.fits" -n t

done < VList

You might also like