A Model-Independent Photometric Redshift Estimator for Type Ia Supernovae

Draft version February 5, 2008

Preprint typeset using LAT^^ style cmulateapj v. 11/12/01

O
O

>
O

>

cn

a^
o
\o
o

Oh:

6

A MODEL-INDEPENDENT PHOTOMETRIC REDSHIFT ESTIMATOR FOR TYPE lA

SUPERNOVAE

YuN Wang^

Draft version February 5, 2008

ABSTRACT

The use of type la supernovae (SNe la) as cosmological standard candles is fundamental in modern
observational cosmology. In this letter, we derive a simple empirical photometric redshift estimator for
SNe la using a training set of SNe la with multiband {griz) light-curves and spectroscopic redshifts
obtained by the Supernova Legacy Survey (SNLS). This estimator is analytical and model- independent;
it does not use spectral templates. We use all the available SNe la from SNLS with near maximum
photometry in griz (a total of 40 SNe la) to train and test our photometric redshift estimator. The
difference between the estimated redshifts Zphot and the spectroscopic redshifts Zspec, [zphot — •Zspec)/(1 +
Zspec), has rms dispersions of 0.031 for 20 SNe la used in the training set, and 0.050 for 20 SNe la not
used in the training set. The dispersion is of the same order of magnitude as the flux uncertainties at
peak brightness for the SNe la. There are no outlyers.

This photometric redshift estimator should significantly enhance the ability of observers to accurately
target high redshift SNe la for spectroscopy in ongoing surveys. It will also dramatically boost the
cosmological impact of very large future supernova surveys, such as those planned for Advanced Liquid-
mirror Probe for Astrophysics, Cosmology and Asteroids (ALPACA), and the Large Synoptic Survey
Telescope (LSST).

Subject headings: distance scale - methods: data analysis - supernovae: general

1. INTRODUCTION

The use of type la supernovae (SNe la) as cosmo logical
standard candles (lPhillipslll993t [Riess. Press, fc KirshneH
[19951 IWa^ et al.ll2003fl is fundamental in modern obser-
vational cosmology. Obtaining the spectroscopic redshifts
of SNe la is the most costly aspect of supernova surveys.

The use of broadband photometry in multiple filters
to esti mate redshifts of galax ies has become well estab-
lished (jWevmann et al.l ll999D . There are two different
approaches in estimating photometric redshif ts of galax-
ies. In the empirical fitting metho d (Co nnollv et al.l
119951: IWang. Bahcall. fc Turnal ll998D . a training set
of galaxies with measured spectroscopic redshifts are
used to derive analytical formulae relating the red-
shift to colors and magnitude s. In the template fit-
ting t echni que (see for example, Puschell. Owen, fc Laind
(I1982D: iLanzetta. Yahil. fc Fernandez-Sotd (I1996D :

iMobasher et all (|1996D : ISawicki. Lin, fc Yed (|1997[ )). the
observed colors are compared with the predictions of a set
of galaxy SED templates.

The multiband photometry of SNe has been used
to select SNe la candidates (iDahln fc Goobai 20021:
iRiess et al.1l2004l : I Johnson fc Crottsll2006f ) . IStrolger et al.l
(|2004[ ) used the template fitting method to estimate the
p hotometric redshift of SN host galaxies.

ICohen etaP (I200GD carried out a blind t est of the pre-
dictions of lWang. Bahcall. fc Turned ([1991) for galaxies in
the Hubble Deep Field North (HDF), and demonstrated
that this technique is capable of reaching a precision of
(^[{zphot - Zspec)/ii + Zspec)] = 0.05 for the majority of
galaxies with z < 1.3. In this paper, we modify and furthe r
develop the technique of lWang. Bahcall. fc Turne"rl ( 19981 ).
to derive a simple and model-independent empirical pho-
tometric redshift estimator for SNe la. This work is made

possible by the Su pernova Legacy S urvey (SNLS) First
Year Data Release (jAstier et aLll2006[ ).

We present our method in Sec. 2, and the results in Sec. 3.
Sec. 4 contains a simple guide to the use of our photometric
redshift estimator. We discuss and summarize in Sec. 5.

2. THE METHOD

We derive the empirical photometric redshift estimator
for SNe la by using observables that reflect the properties
of SNe la as calibrated standard candles.

If SNe la were perfect standard candles, the most im-
portant observable in estimating their redshifts is the peak
brightness. Since the SNLS has the best sampled light-
curves in the i band, we use the i band maximum flux.
We use the fluxes in grz at the epoch of i maximum flux
to make an effective K-correction to the i flux. Our first
estimate of redshift is given by

4hot = ci -f C2gf + csrj + aif + c^zj + cgij (1)
where gj = 2.51og(/g), rj = 2.51og(/r), if = 2.51og(/j),
and Zf = 2.51og(/^), with fg, fr, fi, fz are fluxes in ADU
counts in griz at the epoch of i maximum flux.

Next, we calibrate each SN la in its estimated restframe.
We define

Azi5 = 2.51og(/i5V/0, (2)

where //^'' is the i band flux at 15 days after the i flux
maximum in the estimated restframe, corresponding to the

"phot

) days after the epoch of i

epoch of At^^'^ = 15 * (1
flux maximum.

We now arrive at the final photometric redshift estima-
tor

Zphot ci + C2gf + c^rf + aif + c^Zf + caij + cyAiis (3)

The coefficients Ci (i=l,2,...,7) are found by using a
training set of SNe la with griz light-curves and measured

Homer L. Dodge Department of Physics & Astronomy, Univ. of Oklahoma, 440 W Brooks St., Norman, OK 73019; email: wang@nhn.ou.edu

1

2

Wang

spectroscopic redshifts. We use the jackknife technique
(jLuptonl I1993D to estimate the bias-corrected mean and
the covariance matrix of Ci .

3. RESULTS

The S upernov a Legacy Survey (SNLS) First Year Data
Release (jAstier et al.,.2006i ) consists of the photometry and
redshifts of 71 SNe la. Of these, only 40 have griz light-
curves with gz photometry covering the epoch of the max-
imum flux in the i band. For each of these SN la, we fit
the fluxes in the i band light-cur ve to a n asym metrically
stretched Gaussian introduced bv I Wan j (jl999[ ):

/ = /o exp < -

peak

w{t - tstartY

(4)

All the SNe la we used are well fitted by this form in the
regions of interest (not too close to the tails) . This yields a
smooth light curve without spurious features. The i band
maximum flux and its corresponding epoch are given by
/o (/i = /o) and tp^ak respectively. The fluxes in grz at
the same epoch {tpeak), fg, fr, fz are obtained from the
grz light-curves using linear interpolation. We assume a
floor of fg — 200 (about the size of the flux errors) for
SNe la with fg<0. Eq.([T|) is then used to obtain a first
estimate of the SN redshifts, which allows an estimate of
At^sd. Eq.Q then gives //^'^ needed for Eq.®. Eq.®
gives the final result for the estimated redshifts of the SNe
la.

Fig.l shows the photometric redshifts estimated using
the estimator derived in this paper, compared to the mea-
sured spectroscopic redshifts. The upper panel shows the
results for using all 40 SNe la in deriving the coefficients in
Eq.©. The rms dispersion in {zphot - Zspec)/ (1 + Zgpec) is
0.036. Note that there are no outlyers. This demonstrates
the tight correlation between the griz fluxes and Aiis with
redshifts. The lower panel shows the results for using only
the set containing the most recently discovered 20 SNe la
in deriving the coefficients in Eq. ^ . These coefficients are
then used to predict the redshifts of the other 20 SNe la.
The rms dispersion in {zphot — •Zspec)/(1 + Zspec) is 0.031
for the 20 SNe la used in the training set, and 0.050 for
20 SNe la not used in the training set.

To avoid biases that arise from hand picking the SNe
la used in the training set, we have kept the 40 SNe la
from SNLS in the order of their discovery, then split them
evenly in the middle into two subsets, and used these as
the training set and testing set respectively for our photo-
metric redshift estimator.

Note that the lowest redshift SN la in the SNLS griz
sample is at 2; = 0.263. Thus this photometric redshift
estimator is not calibrated for SNe la at z < 0.263. It
will be straightforward to modify and extend this estima-
tor to lower redshifts, as larger uniform samples of SNe la
covering a greater range of redshifts become available.

Peculiar or highly extincted SNe la can sometimes be
mistaken as high redshift SNe la. Table 1 lists 5 peculiar
and 2 highly extincted SNe la (all are nearby). It demon-
strates that the photometric redshift estimator presented
here generally yields accurate or negative estimated red-
shifts for peculiar and highly extincted SNe la. Thus it
will not lead to contamination of the high z sample by
nearby peculiar and highly extincted SNe la.

4. A RECIPE FOR USING THE PHOTOMETRIC REDSHIFT
ESTIMATOR

We now give a practical guide to the use of our photo-
metric redshift estimator, Eq.Q. The coefficients Ci have
been derived using 20 SNe la, and tested using another
20 SNe la [see Fig. 1(b)]. The bias-corrected mean and
standard deviations of Ci, computed using the jackknife
technique, are given by

ci = 6.122 ±2.006

C2 = -0.06545 ± 0.04548

C3 = -0.03268 ±0.08201

C4 = -0.8225 ±0.4513

C5 = -0.06292 ± 0.08601

C6 0.03979 ± 0.02104

C7 = 0.04552 ± 0.04335 (5)
We give the covariance matrix of Ci in Table 2.

Here are the steps one should follow in using our pho-
tometric redshift estimator: (1) Estimate the flux and
epoch of the i band maximum, fi and tpeak- (2) Esti-
mate the fluxes in grz at the same epoch, fg, fr, and
f^. (3) Use Eq.© with c, (i=l,2,...,6) given in Eq.®
and C7 = to obtain a first estimate of z^f^^^. (4) Use

phot

to estimate At^^'^ = 15 * (1 + z^/^oJ. (5) Esti

mate the i band flux at At^^"^ days after tpeak, fl^'^, and
compute Aii5 = 2.5 \og{ f l^"^ / fi). (6) Use Eq.Q with c,
(i=l,2,...,7) given in Eq.® to obtain Zphot- (7) Use the
covariance matrix of Ci given in Table 1 to compute the
standard deviation of Zphot-

Note that the SNLS zeropoints should be used to
conver t gf, rf, if, z f in Eqs.([T|) and ([3]) into magni-
tudes (jAstier et al.ll2006l) , and the appropriate conversions
should be made if the available pho tometry is in BVRI
instead of griz (jFukugita et al.|[l996f ).

5. DISCUSSION AND SUMMARY

We have derived a model-independent photometric red-
shift estimator, Eqs.® & (O, for SNe la that uses only
multiband photometry near maximum light, and a training
set of SNe la with multiband photometry and measured
spectroscopic redshifts. This estimator is simple and ana-
lytical, thus is very easy to implement (see Sec. 4).

The test of our photometric redshift estimator using SNe
la not used in the training set demonstrates that this es-
timator is robust, and with an accuracy that is of the
same order of magnitude as the photometric errors (see
Fig. 1(b)). The large uncertainties in the coefficients Cj
used in our photometric redshift estimator (see Eq. (O are
indicative of the relatively small sample (20 SNe la) used
for the training set, as well as photometric errors. Eq.®
also shows that the most key constraints on redshift comes
from the g and i band photometry.

As the quality of data improves, we expect that our
photometric redshift estimator can be further improved to
provide more accurate redshift estimates. This photomet-
ric redshift estimator can be easily modified and trained
to apply to SNe la photometry in any choices of multiple
bands.

The measurement of spectroscopic redshifts of SNe is
the most costly and constraining aspect of a supernova
survey. Our results will allow observers to estimate the

Photometric Redshift Estimator for SNe la
Table 1

Behavior of the photometric redshift estimator for peculiar and highly extincted SNe Ia.

SN name

^photo

characteristic

reference

SN2005hk -0.14

SN2002CX -0.12

SN1999ac -0.1

SN1999by 0.00

SN1997br -0.01

SN2006X -0.01

SN1997cy -0.12

pecuhar
pecuhar
pecuhar
pecuhar
pecuhar
highly extincted
highly extincted

Phillips at al. (2006b)
Li et al. (2003)
Philhps et al. (2006a)
Garnavich et_aL (200^
Li et al. (1999)
Krisciunas (2006)
Germany et al. (2000)

Table 2
The covariance matrix of d

0.4022E-h01
0.3364E-02
0.6515E-01
-0.8817E+00
0.2031E-01
0.3988E-01
0.3912E-01

0.3364E-02
0.2068E-02
0.3714E-03
0.2112E-02
-0.2602E-02
-0.2421E-03
0.9723E-03

0.6515E-01
0.3714E-03
0.6726E-02
-0.1750E-01
0.1978E-03
0.4210E-03
0.1924E-03

-0.8817E-I-00
0.2112E-02
-0.1750E-01
0.2037E+00
-0.1203E-01
-0.9101E-02
-0.5894E-02

0.2031E-01
-0.2602E-02

0.1978E-03
-0.1203E-01

0.7398E-02

0.5995E-03
-0.1549E-02

0.3988E-01
-0.2421E-03
0.4210E-03
-0.9101E-02
0.5995E-03
0.4429E-03
0.2314E-03

0.3912E-01
0.9723E-03
0.1924E-03
-0.5894E-02
-0.1549E-02
0.2314E-03
0.1879E-02

redshifts rather accurately based on near maximum light
multiband photometry only (after obtaining spectroscopic
redshifts for a modest training set), thus greatly increase
the efficiency of supernova spectroscopy of high redshift
candidates.

In order to model the systematic uncertainties of SNe
la as standard candles, it is critical to obtain a very large
number of SNe. Future supernova surveys can easily ob-
tain the multiband photometry of a hu ge number of su-
pernovae (|Wandl2()O0t IWang et al.ll2004l ). for example, us-
ing the Advanced Liquid- mirror Probe for Astrophysics,
Cosmology and Asteroids (ALPACA) 0, and the Large
Synoptic Survey Telescope (LSST) 0. It will not be prac-

tical to obtain spectroscopic redshifts for all the SNe la
found by such surveys. With the dense sampling and ac-
curate multiband photometry expected for future super-
nova surveys, it will be possible to refine our photometric
redshift estima tor to the accuracies suitable for cosmology
(jHuterer et al.i 2004). This will dramatically boost the cos-
mological impact of very large future supernova surveys.

Acknowledgements I am grateful to the SNLS team
for releasing their data, including multi-band light-curves
in flux units, and to Kevin Krisciunas for providing me
with unpublished photometry on SN2006X. This work is
supported in part by NSF CAREER grant AST-0094335.

REFERENCES

Astier, P., et al. 2006, A & A 447, 31-48 (2006)
Cohen, J. G., et al. 2000, ApJ, 538, 29
Connolly, A. J., et al., 1995, AJ, 110, 2655
Dahln, T.; Goobar, A. 2002, PASP, 114, 284
Fukugita, M., et al. 1996, AJ, 111, 1748
Garnavich, P.M., et al. 2004, ApJ, 613, 1120
Germany, L.M., et al. 2000, ApJ, 533, 320
Huterer, D., et al. 2004, ApJ, 615, 595
Johnson, B. D.; Crotts, A. P. S. 2006, AJ, 132, 756
Lanzetta, K. M.; Yahil, A.; Fernandez-Soto, A. 1996, Nature, 381,
759

Li, W., et al. 1999, AJ, 117, 2709
Li, W., et al. 2003, PASP, 115, 453

Lupton, R. , 1993, "Statistics in Theory and Practice", Princeton

tjniversity Press

Krisciunas, K., http:/ /www.nd.edu~k krisciu/sn2006X.html|
Mobasher, B., et ai. 1996, IvINRAS, 282, LV

Phillips, M.M. 1993, ApJ, 413, L105

Phillips, M.M., et al. 2006a, astro-ph/0"601684i AJ, in press

Phillips, M.M., et al. 2006b, astro-ph/06112y5

Puschell, J. J.; Owen, F. N.; Laing, R. A. i9»^, ApJ, 257, L57

Riess, A.G., Press, W.H., and Kirshner, R.P. 1995, ApJ, 438, L17

Riess, A. G., et al. 2004, ApJ, 607, 665

Sawicki, M. J.; Lin, H.; Yee, H. K. C, 1997, AJ, 113, 1

Strolger, L., et al. 2004, ApJ, 613, 200

Wang, L. et al., 2003, ApJ, 590, 944

Wang, Y. 1999, ApJ, 525, 651

Wang, Y. 2000, ApJ 531, 676

Wang, Y.; Bahcall, N.; Turner, E. L., 1998, AJ, 116, 2081

Wang, Y., et al. 2004 (the JEDI Collaboration), BAAS, v36, n5,

1560. See also http://jedi.nhn.ou.edu/
Weymann, R.J.; Storrie-Lombardi, L.J.; Sawicki, M.; Brunner,

R.J. 1999., Editors, "Photometric Redshifts and High Redshift

Galaxies", ASP Conference Series, Vol 191

^ http:/ /www. astro. ubc.ca/LMT/alpaca/
^ http://www.lsst.org/

4

Wang

N

+

N

I

N

+

N

I

0.3

0.2

0.1

-0.1

-0.2

-0.3
0.3

— I 1 r-

— I 1 1 r-

using 40 SNe la for fitting Zp^^^
rms of (Z|„.,-z,J/<' + ^.t..> = ""^^

SNe used in fitting

_l I L_

_l I I L_

using 20 SNe la for fitting Zp^^^
0.2 P rms of (z„,„,-z,„J/(l+z,„J = 0.031

0.1

-0.1

-0.2

-0.3

— I [—

— I 1 1 r-

(a)

(b)

SNe not used in fitting (rms of ^ph°t '^■pec = nsnj-

1+Z.p.c

_l L_

_l I I L_

0.5

spec

Fig. 1. — The photometric redshifts estimated using the estimator in Eq.l(3ll, compared to the measured spectroscopic redshifts. (a) The
results for using all 40 SNe la in deriving the coefficients in Eq.l(3ll. (b) The results for using only the set containing the most recently
discovered 20 SNe la in deriving the coefficients in Eq. These coefHcients are then used to predict the redshifts of the other 20 SNe la.