Distant Supernova Photometry
This is a brief introduction to how distant supernova photometries are done
and calibrated.
For high red-shift supernovae we are interested in, separation from the
host galaxies is very small, typically far less then .5". The usual ground-base
images have seeing (=FWHM of point spread function) of about 1" or worse. This
means that aperture photometries of these supernovae will inevitably include
significant amount of host galaxy light. Supernova Cosmology Project has been
using host galaxy subtraction technique where host galaxy contamination is
subtracted out from a supernova plus host galaxy image (NEW), by a reference
image (REF) of the same host galaxy without the supernova.
For each NEW image and REF image, we calculate their seeing by fitting a
gaussian to the stellar objects in the image. We
then convolve the better seeing image to match the worse seeing image. For the
supernova and its nearby "fiducial" objects, one FWHM radius aperture
photometries are performed on the two images separately to avoid unnecessary
resampling of pixels to superimpose the two images. Fiducial objects are any
objects (mostly galaxies in our case) close enough to the SN so that they are
under the similar seeing when the seeing varies spatially over the image. In
this process, we fix the position of the supernova, i.e., position of the
supernova is predetermined using all the available images and then translated
back to each image. (This positioning usually requires some iterations).
Multicolor (R & I) imaging allows us to do rough color cut on fiducial objects
to select object of similar color to the supernova host galaxy. From this subset
of fiducial objects we can find "fidratio", the ratio between (convolved) NEW
and REF. (The idea of subset of fiducial objects are not yet implemented and
currently we are using all the fiducial objects for the fidratio.) Then one
FWHM radius aperture of SN only at each new image is given by
SN(new) = conv_NEW - fidratio * conv_REF
Subtraction above takes care of underlying host galaxy but the problem of
calibrating each data point still remains, especially when many of the SN images
are taken under non-photometric condition. Calibration of SNe magnitude is also
more complicated by the rapid evolution of SNe spectra. Color correction and
instrumental correction require color of the object, which depends on the date
of maximum light. This means that iterations are inevitable in SN photometry.
As a starting point, we can ignore some of the corrections needed for the final
photometry and define an crude magnitude system to save time and effort.
Thus we choose one particular image called "Primary Reference" for of each
SN for a practical purpose, requiring it to have good seeing, good signal-to-
noise ratio, and SN preferably in the center of the image, not to close to the
edges of the image. (The last requirement is to be able to match it with the
most other images of the same SN.) Then we transform each subtraction data point
to our primary reference system by multiplying by the stellar object ratio of
primary reference and each new image in 1 FWHM radius aperture.
SN(prime) = < stars(prime) / stars(new) > * SN(new)
We do take some calibration fields, mostly Landolt stars, while we take SN
fields. Using 4 FWHM radius aperture for these standard stars, we fit the
zeropoints, A, B, C, and D of the following formula on each night.
R(calib-star) = -2.5*log(Rcounts/sec) + ZP_r + A*(R-I) + B*Airmass
I(calib-star) = -2.5*log(Icounts/sec) + ZP_i + C*(R-I) + D*Airmass
These numbers enables us to iteratively determine R & I magnitudes of secondary
standard stars in our supernova image, provided that we took the SN field in
both R and I on the same night as the standards were takeN. If we took data in
one band only, say in R, we can still get the R magnitudes of our secondary
standard stars upto A*(R-I), where we use (R-I)=0.5 +/- 0.4. This process is
repeated for each night we took both calibration fields and SN fields.
Then we can average the magnitudes of our secondary standard stars in SN fields.
Once magnitudes of our secondary standard stars are determined, we may define R
zeropoint for 4 FWHM radius aperture on our primary reference image as follows;
zeropoint = < 2.5*log(COUNTS) + R >
or
zeropoint = 2.5*log(COUNTS) + R + E*(R-I)
where the first is averaged over, and in the second, E and zeropoint are to be
fitted over secondary standard stars in the primary reference image. One more
simple conversion gives zeropoint for 1 FWHM radius aperture.
ZP_1FWHM = 2.5*log(< stars_1FWHM / stars_4FWHM >) + ZP_4FWHM
This allows us to convert SN(prime) above to R magnitude.
(We can get theoretical (R-I) of SN by integrating their expected spectra)
There are a few corrections we have to apply to our magnitude. To do these
corrections correctly, we need possibly extincted, redshifted and
extincted-again spectra of SN at the right epoch having the right stretch. But
even the date of max and stretch may depend on amount of corrections we put in,
especially extinction. Either we have to do some iterations of all our correction
processes or have to find some clever alternatives.
Once we determine the R & I magnitude of secondary standard stars in the
same image as our SN, we can use them to do the "color correction" using them on
each SN image separately.
R_sn(new) = -2.5*log(SN_counts) + ZP_new + A*(R-I)_sn
If we did convert it to primary reference system as our first iteration, then
R_sn(prime) = -2.5*log(SN_counts * )
+ ZP_prime + A'(R-I)_sn
= < exp((ZP_prime - R_stars + A'*(R-I)_stars)/2.5)
/ exp((ZP_new - R_stars + A*(R-I)_stars)/2.5) >
= < exp((ZP_prime - ZP_new + (A'-A)(R-I)_stars) / 2.5) >
R_sn(prime) = -2.5*log(SN_counts) + ZP_new
+ A'*((R-I)_sn - _stars) + A*_stars
The correction for using primary reference system is then
R_sn(new) - R_sn(prime) = (A - A') * ((R-I)_sn - _stars)
Let's define Instrumental Magnitude :
r_inst(object) = -2.5*log(| F_object(v)*S_inst(v)dv
/ | F_vega(v)*S_inst(v)dv)
where F_object(v) is flux of an object and S_inst(v) is filter function of
an instrument. R_inst(object) is then the magnitude of an object in the
instrument system used, defining magnitude of Vega to be 0.
The magnitude in section 3 is in Bessel system for stars. In other
words, if there were a "star" which happened to have the same counts and the
same R-I color as our SN in our image, Bessel magnitude for that star is
given by the procedures described in section 3. Note that having the same
counts means having the same instrumental magnitude. Therefore,
R_inst(SN) = R_inst("star")
SN correction needed to convert the magnitude of section 3 to what we are
after, namely our SN magnitude in Bessel system, is then :
SN Correction = R_bessel(SN) - R_bessel("star")
= [R_bessel(SN) - R_inst(SN)] - [R_bessel("star") - R_inst("star")]
In the second form, each term in square brackets is an "Instrumental
Correction", where normalization cancels out nicely. Here we can relax the
condition for the "star" to have the same counts as the SN.
1) Introduction
Luminosity distance in astronomy is definded as
(Luminosity Distance)^2 = Luminosity / (4*pi*Flux)
The idea behind this definition is very simple. Observed brightness is
inverse-proportional to the surface area of detector 2-sphere surrounding the
source in the center. This definition of luminosity distance is bolometric,
meaning all wavelength of energy integrated.
In reality, measurements of flux is done only in a small window of wavelength,
called filter pass-band. The luminosity distance will remain the same for these
filtered measurements of flux as long as luminosity L is also "filtered
correspondingly". Therefore, in the absence of red-shift and reddening, the
differences between the apparent magnitudes and the absolute magnitudes of any
arbitrary object will be the same across any filter band of non-zero flux.
(2) Traditional K-correction
Red-shift makes the idea of being "filtered correspondingly" a little
complicated. Let's say we are to compare apparent brightness of intrinsically
identical objects, one nearby and one far away. To measure luminosity distance
ratio, we have to make sure that we are integrating the same part of the
spectra for their apparent magnitude. Otherwise, we are making a mistake similar
to comparing R magnitude and I magnitude at the same distance.
Let F(y) be the spectral distribution of flux of the nearby object. Then
that of the redshifted object can be written as f(y) = A*(1/(1+z))*F(y/(1+z)),
where A is a scalar representing all the dimming due to luminosity distance
effect and (1/(1+z)) shows the fact that the energy in a unit y bin is move to
a (1+z)-strethed y bin. Flux measurements are then integration of these with
the filter function S(y).
Nearby : |F(y)S(y)dy
Distant : |f(y)S(y)dy = (A/(1+z)) * |F(y/(1+z))S(y)dy
Because of the red-shift, the two measurements are done on intrinsically
different parts of the spectra. What we needed for distant object measurement
was red-shifted filter pass-band, S'(y) = S(y/(1+z)).
Distant (ideal filter) : |f(y)S'(y)dy
= (A/(1+z)) * |F(y/1+z)S(y/(1+z))dy
= A * |F(y)S(y)dy = A * (Nearby)
Comparing the ideal filter measurement value to the real one, K-correction for
the distant magnitude is given by :
K-correction = 2.5*log(1+z) + 2.5*log(|F(y)S(y)dy / |F(y/(1+z))S(y)dy)
Then bolometric apparent magnitude is obtained by
m_bol = m_inst - K_corr(inst) + dm_bol(inst)
where dm_bol(inst) is the "bolometric correction" that accounts for difference
in zeropoint defintions of the magnitudes involved and not much of physical
interest for us.
(3) Generalized K-correction
When we observe an SN with a small z, spectral region probed by one filter
function S(y) will have significant overlap with that of a nearby SN in the same
filter. But by insisting the same filter function for both nearby and far away
observations, the overlap will decrease as z increases. Eventually we are
comparing two disjoint parts of spectra. Even though the traditional
K-correction is still mathematically correct in these cases, in practice it
becomes more susceptible to systemetic errors because of the uncertainties in SN
spectra, and so forth. As we saw in the derivation of traditional K-correction,
what we are trying to do is convert our "apparent" distant measurements to be
on "equal footing" with nearby ones so that we can get relative luminosity
distances.
For that purpose alone, the ideal filter for high z observation is the filter
used in nearby observation red-shifted by the same z. But continuously
red-shiftable filters are not realistic and also very difficult calibrate.
Still, we are better off using different filters that will match our target
redshift range than insisting one identical filter for both nearby and distant
observation. For instance, when z is around 0.5, R filter matches very well to
redshifted B filter and it is better to compare R filter measurement of SNe at
z~5 with not R but B magnitude of nearby ones.
Let G(y) be the filter function used for the distant observation. Then
the flux measurement is :
Distant : |f(y)G(y)dy = (A/(1+z)) * |F(y/(1+z))G(y)dy
Comparing it to the ideal value again, the generalized K-correction needed is give by :
K-correction = 2.5*log(1+z) + 2.5*log(|F(y)S(y)dy / |F(y/(1+z))G(y)dy)
-2.5*log(|V(y)S(y)dy / |V(y)G(y)dy)
where the last term is added to take care of the difference between the
zero-points of the two filters. V(y) is the idealized stellar spectra at z=0 whose
magnitude is define to be zero in all filter bands. Although this zeropoint term
may not be interesting physically, this should be included because of the way
our measurements are calibrated by standard stars using their magnitude system.
SN light may suffer from extinction by "dust" clouds. The extinction may happen
in our galaxy, in the SN host galaxy, and/or in between. For now, we will
ignore the possibility of extinction in between and assume all the extinction
happened in our Milky Way and/or in the SN host galaxy. We also assume the same
kind of dust as prescribed by Cardelli, Clayton and Mathis. They give
A(x) / A(V) = a(x) + b(x)/Rv
where x is inverse wavelength and Rv = A(V) / E(B-V) is a parameter
characterizing dust property, ranging from 2.5 to 5.5 (???). We are using Rv=3.1
for our calculations.
First we undo the extinction in our galaxy by using E(B-V) value of Burstein and
Heiles. A(V) = 3.1 * E(B-V) is the normalization specifying the amount of dust.
Integration of A(x) * (red-shifted SN spectra) * (filter function) gives the
extinction due to dust in our Milky Way for that filter.
For the extinction in the host galaxy, we have to come up with color
excess at the host galaxy. We use our light-curve fit in R and I to determine
stretch and (observed) R & I magnitudes. Since we do generalized K-correction
for B to R and V to I, R-I we got here is B-V at the host galaxy. Unextincted
B-V is given by integration of theoretical SN spectra. Once we get E(B-V) =
(B-V)_observed - (B-V)_theory at the host galaxy, we do the similar procedure
as above, using un-redshifted spectra of SN there.
To avoid non-trivial amount of iterations in lightcurve fitting and corrections,
we use the empirical fact that the K-correction is "stretchable". This allows us
to build up a look-up table for given z, all possible stretch,extinction, and
SN color. (More details to be added)