Accepted for publication in Ap. J.

Preprint typeset using LATEX style emulateapj v. 03/07/07

TIME DILATION IN TYPE Ia SUPERNOVA SPECTRA AT HIGH REDSHIFT1

S. Blondin, T. M. Davis,3,4 K. Krisciunas,5 B. P. Schmidt,6 J. Sollerman,3 W. M. Wood-Vasey,2 A. C. Becker,7
2

P. Challis,2 A. Clocchiatti,8 G. Damke,9 A. V. Filippenko,10 R. J. Foley,10 P. M. Garnavich,11 S. W. Jha,12

R. P. Kirshner,2 B. Leibundgut,13 W. Li,10 T. Matheson,14 G. Miknaitis,15 G. Narayan,16 G. Pignata,17
A. Rest,9,16 A. G. Riess,18,19 J. M. Silverman,10 R. C. Smith,9 J. Spyromilio,13 M. Stritzinger,3,20
C. W. Stubbs,2,16 N. B. Suntzeff,5 J. L. Tonry,21 B. E. Tucker,6 and A. Zenteno22
Accepted for publication in Ap. J.

ABSTRACT
We present multiepoch spectra of 13 high-redshift Type Ia supernovae (SNe Ia) drawn from the
literature, the ESSENCE and SNLS projects, and our own separate dedicated program on the ESO
Very Large Telescope. We use the Supernova Identification (SNID) code of Blondin & Tonry to
determine the spectral ages in the supernova rest frame. Comparison with the observed elapsed time
yields an apparent aging rate consistent with the 1/(1 + z) factor (where z is the redshift) expected
in a homogeneous, isotropic, expanding universe. These measurements thus confirm the expansion
hypothesis, while unambiguously excluding models that predict no time dilation, such as Zwicky’s
“tired light” hypothesis. We also test for power-law dependencies of the aging rate on redshift. The
best-fit exponent for these models is consistent with the expected 1/(1 + z) factor.
Subject headings: cosmology: miscellaneous — supernovae: general

1. INTRODUCTION
1 Based on observations made with ESO Telescopes at the
Paranal Observatory under programs 67.A-0361, 267.A-5688, The redshift, z, is a fundamental observational quan-
078.D-0383, and 080.D-0477; at the Gemini Observatory and tity in Friedman-Lemaı̂tre-Robertson-Walker (FLRW)
NOAO, which are operated by the Association of Universities for models of the universe. It relates the frequency of light
Research in Astronomy, Inc., under cooperative agreements with emitted from a distant source to that detected by a local
the NSF; with the Magellan Telescopes at Las Campanas Observa-
tory; and at the W. M. Keck Observatory, which was made possible observer by a factor 1/(1 + z). One important conse-
by the generous financial support of the W. M. Keck Foundation. quence is that the observed rate of any time variation
2 Harvard-Smithsonian Center for Astrophysics, 60 Garden
in the intensity of emitted radiation will also be propor-
Street, Cambridge, MA 02138; [email protected] tional to 1/(1 + z) (see Weinberg 1972 and Appendix A).
3 Dark Cosmology Centre, Niels Bohr Institute, University of
Copenhagen, Juliane Maries Vej 30, DK–2100 Copenhagen Ø, Den- Due to their large luminosities (several billion times
mark. that of the sun) and variability on short timescales (∼
4 Department of Physics, University of Queensland, QLD, 4072,
20 d from explosion to peak luminosity; Riess et al. 1999;
Australia.
5 Department of Physics, Texas A&M University, College Sta- Conley et al. 2006), Type Ia supernovae (SNe Ia) are
tion, TX 77843-4242. ideally suited to probe these time-dilation effects across a
6 The Research School of Astronomy and Astrophysics, The Aus-
large fraction of the observable universe. The suggestion
tralian National University, Mount Stromlo and Siding Spring Ob- to use supernovae as “cosmic clocks” was proposed by
servatories, via Cotter Road, Weston Creek, PO 2611, Australia.
7 Department of Astronomy, University of Washington, Box Wilson more than six decades ago (Wilson 1939) and
351580, Seattle, WA 98195-1580. tested on light curves of low-redshift SNe Ia in the mid-
8 Pontificia Universidad Católica de Chile, Departamento de As-
1970s (Rust 1974), but only since the mid-1990s has this
tronomı́a y Astrofı́sica, Casilla 306, Santiago 22, Chile. effect been unambiguously detected in the light curves of
9 Cerro Tololo Inter-American Observatory, National Optical As-
tronomy Observatory, Casilla 603, La Serena, Chile. high-redshift objects (Leibundgut et al. 1996; Goldhaber
10 Department of Astronomy, University of California, Berkeley, et al. 2001).
CA 94720-3411.
11 Department of Physics, University of Notre Dame, 225 Nieuw-
These latter studies show that the light curves of dis-
tant SNe Ia are consistent with those of nearby SNe Ia
land Science Hall, Notre Dame, IN 46556-5670.
12 Department of Physics and Astronomy, Rutgers University, whose time axis is dilated by a factor (1 + z). However,
136 Frelinghuysen Road, Piscataway, New Jersey 08854. there exists an intrinsic variation in the width of SN Ia
13 European Southern Observatory, Karl-Schwarzschild-Strasse
light curves which is related to their peak luminosities
2, D-85748 Garching, Germany. (Phillips 1993), such that more luminous SNe Ia have
14 National Optical Astronomy Observatory, 950 North Cherry
Avenue, Tucson, AZ 85719-4933. broader light curves (Fig. 1). This “width-luminosity”
15 Fermilab, P.O. Box 500, Batavia, IL 60510-0500. relation is derived using low-redshift SNe Ia where the
16 Department of Physics, Harvard University, 17 Oxford Street,
time-dilation effect, if any, is negligible (Phillips 1993;
Cambridge, MA 02138.
17 Departamento de Astronomı́a, Universidad de Chile, Casilla Hamuy et al. 1995; Riess et al. 1995; Phillips et al. 1999;
36-D, Santiago, Chile. Jha et al. 2007).
18 Johns Hopkins University, 3400 North Charles Street, Balti-
It is problematic to disentangle this intrinsic variation
more, MD 21218. of light-curve width with luminosity and the effect of time
19 Space Telescope Science Institute, 3700 San Martin Drive,
Baltimore, MD 21218.
20 Las Campanas Observatory, Carnegie Observatories, Casilla lawn Drive, Honolulu, HI 96822.
22 Department of Astronomy, University of Illinois at Urbana-
601, La Serena, Chile.
21 Institute for Astronomy, University of Hawaii, 2680 Wood- Champaign, 1002 West Green St, Urbana, IL 61801-3080.
2 Blondin et al.

As of today there are two published examples of ag-

ing rate measurements using spectra of a single SN Ia
(SN 1996bj at z = 0.574, Riess et al. 1997; SN 1997ex
at z = 0.362, Foley et al. 2005). In both cases, the null
hypothesis of no time dilation is excluded with high sig-
nificance (> 95%).
In this paper we present data on 13 high-redshift
(0.28 ≤ z ≤ 0.62) SNe Ia for which we have multiepoch
spectra. We use the Supernova Identification (SNID)
code of Blondin & Tonry (2007) to infer the age of each
spectrum in the supernova rest frame, from which we de-
termine the apparent aging rate of each SN Ia. These ag-
ing rate measurements are then used to test the 1/(1+z)
time-dilation hypothesis expected in an expanding uni-
verse. The data enable us for the first time to directly
test the time-dilation hypothesis over a large redshift
range.
This paper is organized as follows. In Section 2 we
explain how one can determine the age of a SN Ia based
on a single spectrum and present the SNID algorithm
used for this purpose. The aging rate measurements are
presented in Section 3, and the time-dilation hypothesis
Fig. 1.— Bolometric light curves of 5 low-redshift SNe Ia taken (amongst others) is tested against the data in Section 4,
from Stritzinger (2005) (from top to bottom: SNe 1991T, 1999ee, with the help of model selection statistics (information
1994D, 1992A, 1993H, and 1991bg). More luminous SNe Ia have criteria). Conclusions follow in Section 5.
broader light curves. SN 1991bg is an example of intrinsically
subluminous SNe Ia (maximum Lbol < 109 Lsun ), which are less 2. DETERMINING THE AGE OF A SN Ia SPECTRUM
likely to be found at high redshifts.
2.1. SN Ia Spectral Evolution
dilation. To directly test the time-dilation hypothesis one The spectra of SNe Ia consist of blended spectral lines,
needs to accurately know the distribution of light-curve with a profile shape characteristic of stellar outflows.
widths at z ≈ 0 and its potential evolution with redshift, This line profile (also known as a “P Cygni” profile) con-
whether due to a selection effect (not taken into account sists of an emission component symmetric about the line
by Goldhaber et al. 2001) or an evolution of the mean center, and an absorption component that is blueshifted
properties of the SN Ia sample with redshift— as possibly by the ∼ 10, 000 km s−1 expansion velocity of the SN
observed by Howell et al. (2007). Moreover, one needs ejecta (Pinto & Eastman 2000). The expansion also
to probe sufficiently high redshifts (z & 0.4, as done by causes a Doppler broadening of both components, such
Leibundgut et al. 1996; Goldhaber et al. 2001) such that that a typical spectroscopic feature in SN Ia spectra has
the observed widths of the SN Ia light curves are broader a width of ∼ 100 Å. As the ejecta expand, the photo-
than the intrinsic width of any nearby counterpart. sphere recedes in the comoving frame of the supernova,
Furthermore, one might argue that at high redshift we such that the spectra probe deeper layers of the ejecta
are preferentially finding the brighter events (akin to a with time. Given the homologous nature of the expan-
Malmquist bias). Such a selection effect would produce sion (velocity proportional to radius), and the chemical
a spurious relation in which there would be broader light stratification in the SN ejecta (Nomoto et al. 1984; Stehle
curves at higher redshifts, without any time dilation. et al. 2005; Mazzali et al. 2008), deeper layers correspond
The spectra of SNe Ia provide an alternative and a to lower expansion velocities and an increased abundance
more reliable way to measure the apparent aging rate of iron-peak elements. The impact on the spectra is
of distant objects. Indeed, the spectra of SNe Ia are twofold. First, the blueshift of SN Ia spectral lines de-
remarkably homogeneous at a given age, such that the creases with time (by as much as ∼ 1000 km s−1 per day;
age of a SN Ia can be determined from a single spectrum Benetti et al. 2005; Blondin et al. 2006). Second, due to
with an accuracy of 1–3 d— with no reference to its corre- the varying chemical composition at the photosphere, the
sponding light curve (Riess et al. 1997; Howell et al. 2005; relative shapes and strengths of spectral features change
Blondin & Tonry 2007). More importantly, the spectra on a timescale of days.
of SNe Ia spanning a range of luminosities (and hence This complex spectral evolution is nonetheless pre-
different intrinsic light-curve widths) evolve uniformly dictable to a large extent. At a given age, the spectra
over time (Matheson et al. 2008). The use of spectra are remarkably homogeneous among different “normal”
thus avoids the degeneracy between intrinsic light-curve SNe Ia. According to Li et al. (2001b), these constitute
width and time-dilation effects. While there are some ∼ 65% of the local SN Ia sample, the rest consisting
notable examples of deviations from homogeneity in sev- of intrinsically subluminous (∼ 15%) or overluminous
eral SNe Ia (e.g., SN 2000cx, Li et al. 2001a; SN 2002cx, (∼ 20%) events, whose spectra show deviations from
Li et al. 2003; SN 2002ic, Hamuy et al. 2003; SN 2003fg, those of normal SNe Ia. Subluminous SNe Ia are less
Howell et al. 2006; SN 2006gz, Hicken et al. 2007), these likely to be found at high redshifts; in fact, no such object
outliers are readily identifiable spectroscopically through has been spectroscopically confirmed in any high-redshift
comparison with a large database of supernova spectra supernova search to this day (e.g., Matheson et al. 2005;
(see Section 2 and Blondin & Tonry 2007). Howell et al. 2005). In what follows we only consider
 Time Dilation in SN Ia Spectra 3

Tonry 2007). SNID automatically determines the type,

redshift, and age of a supernova spectrum. We refer the
reader to that paper for a more detailed discussion.
The redshift of the input spectrum is a free parameter
in SNID, although it can be fixed to a specific value.
Comparison of the SNID redshifts with those determined
from narrow emission and absorption lines in the host-
galaxy spectra (typically accurate to . 100 km s−1 ; see
Falco et al. 1999) yields a dispersion about the one-to-
one correspondence of only σz ≈ 0.005 out to a redshift
z ≈ 0.8 (Blondin & Tonry 2007).
Similarly, comparison of the SNID ages with those de-
termined using the corresponding light curves yields a
typical accuracy < 3 d, comparable to other algorithms
(Riess et al. 1997; Howell et al. 2005). However, the age
error is systematically overestimated. In this paper, we
estimate the error as follows: each spectrum in the SNID
database is trimmed to match the rest-frame wavelength
range of the input spectrum, and is correlated with all
other spectra in the database (except those correspond-
ing to the same supernova). The age error is then given
by the mean variance of all template spectra whose SNID
age is within one day of the initial estimate.
The success of SNID and similar algorithms lies pri-
marily in the completeness of the spectral database. In
Fig. 3 we show the age distribution of the SN Ia templates
used in SNID for this paper (these do not include sub-
Fig. 2.— Standard (light gray) and maximum (dark gray) devi-
luminous SNe Ia). This database comprises 959 spectra
ation from the mean spectrum (black line) for the 22 low-redshift of 79 low-redshift (z . 0.05) SNe Ia with ages between
SNe Ia for which we report an aging rate measurement (see Sec- −15 and +50 d from maximum light. The spectra are
tion 4), at four different ages — given in days from B-band max- taken from the literature, from public databases (such as
imum light. A low-order curve has been divided out from each
spectrum to reveal the relative shapes and strengths of the various
SUSPECT23 or the CfA Supernova Archive24 ), or from
spectroscopic features. a set of unpublished spectra from the CfA Supernova
Program. A full reference to all spectra in the SNID
normal and overluminous SNe Ia. In a separate paper database is given by Blondin & Tonry (2007). It is im-
we will show that none of the SNe Ia in the high-redshift portant to note that each template spectrum is shifted
sample presented here (see Section 4) has a spectrum or to zero redshift and that each template age is corrected
light curve consistent with the subluminous variety of for the expected (1 + z) time-dilation factor. Because all
SNe Ia. the template SNe Ia are at low redshift (z . 0.05), this
The spectroscopic homogeneity of SNe Ia holds even is a very small correction, and we will see in Section 4
when we consider both normal and overluminous objects that this has no impact on the aging rate measurements.
in a representative sample of nearby events. In Fig. 2 we Thus, SNID determines the age a supernova would have
show the mean spectrum for the 22 low-redshift SNe Ia at z = 0 — that is, in the supernova rest frame.
for which we report an aging rate measurement (see Sec- The number of SNe Ia shown in Fig. 3 is large enough
tion 4), at four different ages. While there is an in- that we can select a subsample (shown as an open his-
trinsic spectral variance among these different SNe Ia togram) on which to conduct age determinations and ag-
— some spectroscopic features correlate with luminosity ing rate measurements on low-redshift SNe Ia. The size
(e.g., Nugent et al. 1995; Matheson et al. 2008), the av- of this subsample is set by the requirement that remov-
erage deviations from the mean spectrum are small, and ing it from the SNID database would leave a sufficient
all spectra evolve in a similar manner over the course of number of templates in a given age bin for a reliable
several days, independent of light-curve width. age determination (see Blondin & Tonry 2007). It was
Both the homogeneity and rapid evolution of SN Ia also chosen to include a sufficient number of intrinsically
spectra enables an accurate determination of the age of overluminous SNe Ia: indeed, there are five such SNe Ia
a single spectrum. We explain how this is achieved in (SNe 1997br, 1998ab, 1999dq, 1999gp, and 2001eh) in
practice in the following section. this subsample, accounting for ∼ 20% by number of ob-
jects and spectra. For these specific tests, the templates
2.2. The SNID Algorithm corresponding to the 22 SNe Ia in this subsample are
Given a large database of finely time-sampled SN Ia temporarily removed from the SNID database to avoid
spectral templates, we can determine the age of a given biasing the age determination.
input spectrum by finding the best-match template(s) We deliberately restrict this subsample to ages between
in the database. There are several standard techniques −10 and +30 d from maximum light. Before −10 d, the
to do this (see Blondin & Tonry 2007 for a review). In
23 http://bruford.nhn.ou.edu/∼suspect/index1.html
this paper, we use an implementation of the correlation 24 http://www.cfa.harvard.edu/supernova/SNarchive.html
techniques of Tonry & Davis (1979): SNID (Blondin &
4 Blondin et al.

TABLE 1
Comparison of galaxy and supernova
redshifts

SN zGAL zSN zSNID

(1) (2) (3) (4)
1996bj 0.574 0.581 (0.005) 0.580 (0.008)
0.582 (0.008)
1997ex 0.361 0.362 (0.002) 0.362 (0.005)
0.361 (0.004)
0.362 (0.004)
2001go 0.552 0.552 (0.005) 0.552 (0.008)
0.556 (0.008)
0.550 (0.009)
2002iz 0.427 0.425 (0.004) 0.422 (0.006)
0.428 (0.006)
b027 ··· 0.315 (0.003) 0.315 (0.006)
0.315 (0.004)
2003js 0.363 0.361 (0.003) 0.359 (0.004)
0.363 (0.006)
Fig. 3.— Age distribution of the 79 SN Ia templates used in 04D2an 0.621 0.614 (0.006) 0.608 (0.007)
SNID (hatched histogram). There are a total of 959 spectra with 0.625 (0.011)
ages between −15 and +50 d from B-band maximum light. The 2006mk 0.475 0.477 (0.003) 0.479 (0.005)
open histogram shows the subsample of 145 spectra from 22 SNe Ia 0.478 (0.007)
(restricted to ages between −10 and 30 d from maximum) for which 0.474 (0.008)
we report an aging rate measurement (see Section 4). 0.476 (0.006)
2006sc 0.357 0.356 (0.004) 0.355 (0.007)
number of spectral templates in the SNID database drops 0.357 (0.007)
rapidly, and the age determination is inaccurate. Past 0.356 (0.006)
+30 d, the spectra of SNe Ia do not evolve as rapidly as 2006tk ··· 0.312 (0.003) 0.312 (0.006)
around maximum light, and the age determination is less 0.310 (0.003)
0.315 (0.006)
precise (Blondin & Tonry 2007). 2007tg ··· 0.502 (0.004) 0.503 (0.009)
0.503 (0.008)
3. AGING RATE AT HIGH REDSHIFTS 0.502 (0.007)
3.1. Spectroscopic Data 2007tt 0.374 0.376 (0.004) 0.367 (0.008)
0.379 (0.007)
Our aging rate measurements at high redshifts are 0.377 (0.005)
based on a sample of 35 spectra of 13 SNe Ia in 2007un 0.283 0.285 (0.004) 0.287 (0.006)
0.285 (0.007)
the redshift range 0.28 ≤ z ≤ 0.62. These include 0.285 (0.005)
previously published data by the High-Z Supernova Column headings: (1) SN name. (2) Galaxy red-
Search Team (SN 1996bj, Riess et al. 1997), the Super- shift (the typical error is < 0.001). (3) SN redshift,
nova Cosmology Project (SN 1997ex, Foley et al. 2005; quoted as the error-weighted mean of the individ-
SN 2001go, Lidman et al. 2005), and the ESSENCE ual redshifts for each epoch. (4) SNID redshift for
each epoch, in order of increasing age.
project (SNe 2002iz, b027, and 2003js, Matheson et al.
2005). For SN 2001go we present our own reductions shifts (zSNID ) for each of its spectra (Table 1). For ten
of the three epochs of spectroscopic data obtained from SNe Ia, we also have a redshift determination from the
the ESO Science Archive Facility25 , as only the first spec- host galaxy (zGAL ). Comparison with the supernova red-
trum was published by Lidman et al. (2005). The spectra shift shows an excellent agreement between the two mea-
of SN 04D2an (the highest-redshift SN Ia in this sample) surements (better than 1%). For the three remaining
were obtained by members of the Supernova Legacy Sur- SNe Ia (SNe b027, 2006tk, and 2007tg), only the SNID
vey (SNLS), and will be published as part of a larger redshift is available, but the different redshift measure-
sample of SNLS data by Stéphane Basa and cowork- ments for individual spectra all agree to within 1σ, and
ers. SN 2006tk will be published alongside the complete we are confident about their accuracy. In what follows,
ESSENCE supernova dataset in the near future. we will use the galaxy redshift when available for the age
The other five SNe Ia (SNe 2006mk, 2006sc, 2007tg, and aging rate measurements. Given the excellent agree-
2007tt, 2007un) are ESSENCE targets which were ob- ment between zGAL and zSN , this choice has negligible
served spectroscopically through two dedicated “Target- impact on our results.
of-Opportunity” programs at the ESO Very Large Tele-
scope26 . 3.2. Accuracy of Relative Age Determination
Details on the instrumental setup and data reduction
can be found in the aforementioned references. The rest An accurate determination of the rate of aging involves
of the data will be presented more thoroughly in a sepa- accurate knowledge of age differences. In what follows
rate publication. All of the data are shown in Fig. 4. we test how well SNID determines differential ages using
We use SNID to determine the redshift of each spec- the subsample of low-redshift SNe Ia presented in Sec-
trum. The redshift of a given supernova (zSN ) is then tion 2.2. While the determination of absolute ages has
computed as the error-weighted mean of the SNID red- no impact on the main result of this paper, we discuss
their accuracy in Appendix B.
25 http://www.eso.org/sci/archive/ . We determine the rest-frame age (tspec ) of each of the
26 Programs 078.D-0383 and 080.D-0477; PI: Jesper Sollerman. 145 spectra in the low-redshift subsample of 22 SNe Ia.
 Time Dilation in SN Ia Spectra 5

Fig. 4.— Multiepoch spectra of the 13 high-redshift SNe Ia used in this study, binned to 10 Å (gray). The vertical offset between each
spectrum is for clarity only, and does not reflect differences in flux density (F λ ; erg s−1 cm−2 Å−1 ) between them. In each plot, the age
of the supernova increases downwards, and the observed time (in days) from the first spectrum is indicated. Overplotted in black is a
smoothed version using the inverse-variance-weighted Gaussian algorithm of Blondin et al. (2006).

We then compute the absolute age difference (∆tspec ) spondence. For ∆tobs & 30 d, however, SNID systemati-
between each unique pair of spectra corresponding to a cally underestimates the age difference by ∼ 1.5 d. This
given supernova. This amounts to 631 pairs. This age is more apparent in the plot of residuals in the middle
difference is then compared with the absolute observer- panel. it is mainly due to a systematic overestimate of
frame age difference (∆tobs ) for each spectrum pair. rest-frame ages tspec . −7 d from maximum light, due
Since z ≈ 0 for this subsample, ∆tspec can be directly to the lack of spectral templates in the SNID database
compared to ∆tobs with no correction for time dila- with similar ages (see Fig. 3 and Blondin & Tonry 2007).
tion. Given the restriction to ages between −10 and The lower panel of Fig. 5 shows the absolute fractional
+30 d from maximum light in the low-redshift subsam- age difference vs. ∆tobs . The quantity |∆tobs /∆tspec |
ple, ∆tspec (and hence ∆tobs ) is at most 40 d. is a direct measure of the accuracy we can achieve for
The results are displayed in the upper panel of Fig. 5. the aging rate determination. As expected, the frac-
There is good agreement between ∆tspec and ∆tobs , with tional age difference decreases with increasing age differ-
a dispersion of only 2.0 d about the one-to-one corre- ence. For ∆tobs > 6 d, this difference drops below 20%.
6 Blondin et al.

TABLE 2
Observer-frame and rest-frame age differences

SN tobs tspec ∆tobs ∆tspec

(1) (2) (3) (4) (5)
1996bj 367.99 −2.2 (3.0) 0.00 0.0 (3.0)
378.04 3.1 (2.2) 10.05 5.3 (2.2)
1997ex 815.08 −1.6 (1.6) 0.00 0.0 (1.6)
839.96 17.4 (2.1) 24.88 19.0 (2.1)
846.03 21.2 (2.0) 30.95 22.8 (2.0)
2001go 2021.70 7.9 (2.3) 0.00 0.0 (2.3)
2027.69 9.8 (1.7) 5.99 1.9 (1.7)
2059.17 31.2 (1.6) 37.47 23.3 (1.6)
2002iz 2586.95 −0.5 (2.2) 0.00 0.0 (2.2)
2614.57 17.6 (1.2) 27.62 18.1 (1.2)
b027 2589.95 −3.5 (1.8) 0.00 0.0 (1.8)
2616.57 18.4 (1.6) 26.62 21.9 (1.6)
2003js 2942.46 −4.9 (1.6) 0.00 0.0 (1.6)
2966.71 12.5 (1.2) 24.25 17.4 (1.2)
04D2an 3026.20 −2.5 (1.6) 0.00 0.0 (1.6)
3032.20 0.9 (1.3) 6.00 3.4 (1.3)
2006mk 4031.71 −6.2 (1.0) 0.00 0.0 (1.0)
4040.72 −0.6 (2.2) 9.01 5.6 (2.2)
4051.77 7.3 (1.9) 20.06 13.5 (1.9)
4063.78 18.5 (1.8) 32.07 24.7 (1.8)
2006sc 4063.58 0.9 (1.6) 0.00 0.0 (1.6)
4076.65 9.8 (1.4) 13.07 8.9 (1.4)
4084.68 13.4 (2.2) 21.10 12.5 (2.2)
2006tk 4089.57 −8.8 (2.4) 0.00 0.0 (2.4)
4100.57 0.3 (2.0) 11.00 9.1 (2.0)
Fig. 5.— Upper panel: Rest-frame age difference (∆tspec ) vs. 4103.59 2.9 (0.9) 14.02 11.7 (0.9)
observer-frame age difference (∆tobs ) for each spectrum pair for a 2007tg 4381.75 −6.1 (2.0) 0.00 0.0 (2.0)
given supernova. There are 631 such pairs, with a dispersion of 4391.65 −0.5 (1.8) 9.90 5.6 (1.8)
2.0 d about the one-to-one correspondence. Middle panel: Residu- 4405.61 10.0 (1.5) 23.86 16.1 (1.5)
als in the upper panel vs. ∆tobs . Lower panel: Ratio of ∆tspec to 2007tt 4415.81 −5.0 (2.1) 0.00 0.0 (2.1)
∆tobs , again vs. ∆tobs . For ∆tobs > 6 d, the fractional difference 4430.65 6.1 (1.4) 14.84 11.1 (1.4)
is less than 20%. 4443.58 14.9 (2.2) 27.77 19.9 (2.2)
2007un 4441.61 3.2 (2.2) 0.00 0.0 (2.2)
The high-redshift data presented in the previous section 4451.60 11.1 (1.3) 9.99 7.9 (1.3)
span a sufficient range of observer-frame age difference 4460.58 17.7 (1.4) 18.97 14.5 (1.4)
that the aging rate determination is accurate. Note that Column headings: (1) SN name. (2) Julian date (JD)
the systematic underestimate of the age difference for minus 2,450,000 at midpoint of observation. (3) SN
∆tobs & 30 d results in a negligible fractional difference. rest-frame age in days from maximum light, derived
from the cross-correlation with spectral templates us-
ing SNID. (4) Observer-frame days from first spectrum.
3.3. Aging Rate Determination (5) Rest-frame days from first spectrum.
The rest-frame age of each high-redshift SN Ia spec-
trum (tspec ) is determined as outlined in Section 2.2. In is illustrated in Fig. 7 using a Monte Carlo simulation of
each case, we fix the redshift to that determined in the the age measurements for SN 1997ex presented by Fo-
previous section. The results are displayed in Table 2, ley et al. (2005). Using the same errors on the individual
along with the corresponding observed date of each spec- age measurements, the distribution of the slope measure-
trum (tobs ). However, since the aging rate determination ments is highly non-Gaussian in (1 + z) space, while it is
depends on age differences (see previous Section), we also normally distributed in 1/(1 + z) space.
report the observer-frame and rest-frame age from the The individual aging rate measurements presented here
first spectrum, respectively denoted ∆tobs and ∆tspec in alone reject models that predict no time dilation at a high
Table 2. significance (up to ∼ 6σ), and all (except for SN 2006mk)
We can then trivially compute the aging rate for each are within 1σ of the expected 1/(1+z) factor. In the next
supernova. This is simply done through a least-squares section we combine all aging rate measurements (includ-
fit of a line to ∆tspec versus ∆tobs . The slope of the line is ing those for the low-redshift sample) to test each hy-
a measure of the aging rate, which should equal 1/(1 + z) pothesis more thoroughly.
in an expanding universe (see Appendix A). Were there
no time dilation, the aging rate would equal one. The 4. TESTING THE 1/(1 + z) TIME-DILATION HYPOTHESIS
results are displayed in Fig. 6. We have determined the aging rate for the subsample
We note that comparing the inverse of the slope in of 22 low-redshift SNe Ia. We combine these aging rates
Fig. 6 (denoted “age factor” by Foley et al. 2005) and with those determined for the 13 high-redshift SNe Ia of
(1 + z) leads to asymmetric errors. The errors on the our sample (see Table 3) to test the 1/(1+z) time-dilation
“age factor” [≡ (1 + z)] become highly non-Gaussian hypothesis. As noted in Section 3.2 and Appendix B,
when the uncertainties of the individual age measure- these measurements rely on a database of SN Ia spectra
ments are large (& 1 d, as is the case in this paper, and whose ages have already been corrected for the expected
in Foley et al. 2005 for SN 1997ex), whereas the errors on 1/(1 + z) time-dilation factor. However, since all SNe Ia
the aging rate [≡ 1/(1 + z)] are always Gaussian. This in the SNID database are at redshifts z ≤ 0.05, the cor-
 Time Dilation in SN Ia Spectra 7

Fig. 6.— Comparison of rest-frame (∆tspec ) and observer-frame (∆tobs ) time from the first spectrum, for each of the 13 high-redshift
SNe Ia in our sample. The abscissa and ordinate ranges are both set to [−3,+40] d in all cases. The slope of the best-fit line (solid line)
gives a measurement of the apparent aging rate of the supernova, which is compared to the expected 1/(1 + z) value. The dotted line in
each plot corresponds to ∆tspec = ∆tobs .

All aging rate measurements are shown in Fig. 8. The

solid line shows the expected 1/(1 + z) time-dilation fac-
tor, while the dashed line represents the “tired light”
hypothesis of Zwicky (1929). According to this hypothe-
sis, photons lose energy as they interact with matter and
other photons in a static universe. The energy loss is pro-
portional to the distance from the source, and causes a
redshift in spectra as in an expanding universe. However,
this hypothesis does not predict a time-dilation effect,
and so the aging rate should equal one for all redshifts.
As expected, the measurement of a time-dilation effect
is more obvious at larger redshift, and the precision im-
Fig. 7.— Monte Carlo results illustrating the advantage of work-
ing in 1/(1 + z) space for time-dilation measurements. Solid lines: proves as the number and time span of spectra for each
recovered slope (in standard deviations from the mean, µ) in (1+z) supernova increases. This latter effect explains why the
(left) and 1/(1 + z) space (right), using the SN age errors reported aging rate measurements for SN 1996bj (z = 0.574) and
by Foley et al. (2005). The distribution is highly non-Gaussian in SN 04D2an (z = 0.621) have a large associated error
the former case.
— despite being the two highest-redshift SNe Ia in our
rection is small (. 1 d) and has a negligible impact on sample, since only two spectra separated by ∼ 10 (for
the aging rate measurements. SN 1996bj) and ∼ 6 (for SN 04D2an) observer-frame
8 Blondin et al.

Fig. 8.— Apparent aging rate versus 1/(1 + z) for the 13 high-redshift (0.28 ≤ z ≤ 0.62) and 22 low-redshift (z < 0.04) SNe Ia in our
sample. Overplotted are the expected 1/(1 + z) time dilation (solid line) and the best-fit 1/(1 + z) b model (with b = 0.97 ± 0.10; dotted
line and gray area). The dashed line corresponds to no time dilation, as expected in the tired-light model — clearly inconsistent with the
data. The inset shows a close-up view of the low-redshift sample. These data are summarized in Table 3.

days are available. ses, since the impact of time dilation is small at such low
A simple χ2 analysis is sufficient to confirm what the redshifts.
eye sees: the hypothesis of no time dilation is not a good In Fig. 9 a different view of Fig. 8 shows the distribu-
fit to the data (χ2 = 150.3 for 35 degrees of freedom; tions of the ratio between the aging rate and 1/(1 + z)
see Table 4), with a goodness-of-fit of ∼ 0% (defined as for both the low-redshift (open histogram) and high-
GoF= Γ(ν/2, χ2 /2)/Γ(ν/2), where Γ(ν/2, χ2 /2) is the redshift (hatched histogram) samples. Both distributions
incomplete gamma function and ν is the number of de- are within ∼ 20% of a unit ratio, again validating the
grees of freedom) — namely, a null probability of ob- hypothesis of time dilation over a large redshift range.
taining data that are a worse fit to the model, assuming The apparent bias to lower values of the ratio for the
that the model is indeed correct. The expected 1/(1 + z) low-redshift sample is not statistically significant, as the
time-dilation factor, on the other hand, yields a good mean error on the aging rate is of order one bin size
fit to the data (χ2 = 27.0 for 35 degrees of freedom), (. 0.1; see Table 3).
with GoF= 83.2%, and is largely favored over the null In what follows we test whether the data favor a non-
hypothesis of no time dilation (∆χ2 ≈ 123). linear dependence of the aging rate on redshift, namely
This result holds (and in fact improves) when we con-
sider only the high-redshift sample (see Table 4). This 1
aging rate = , (1)
works because the z = 0 end of the aging rate versus red- (1 + z)b
shift relation (Fig. 8) is fixed to unity by theory, so the
low-redshift sample is not needed to anchor the theoret- where b is a free parameter. While Eq. [1] satisfies the
ical curve at z ≈ 0 (although it is still used to calibrate same zero point as the two previous hypotheses (aging
the tspec measurement). The low-redshift data alone do rate equal to 1 at z = 0), no model actually predicts such
not enable us to distinguish between the two hypothe- a dependence of the aging rate on redshift. Nonetheless,
small deviations from the expected 1/(1+z) factor would
 Time Dilation in SN Ia Spectra 9

b = 0.95 ± 0.10 for the high-redshift sample. As ex-

TABLE 3 pected, the low-redshift sample alone is insufficient to
Aging rate measurements
constrain the free parameter (b = 3.18 ± 1.28). Nonethe-
SN z 1/(1 + z) Aging rate less, the samples that include the high-redshift objects
have a best-fit value for b that is consistent with b = 1,
Low redshift (z < 0.04) and thus with the expected 1/(1+z) time-dilation factor.
1981B 0.006 0.994 1.099 (0.071)
1989B 0.002 0.998 1.036 (0.079) Since this model has an additional free parameter, it
1992A 0.006 0.994 1.040 (0.077) is instructive to use information criteria to compare it to
1994D 0.002 0.998 1.011 (0.152) the simple 1/(1+z) prediction. These model comparison
1996X 0.007 0.993 0.886 (0.116) statistics favor models that yield a good fit to the data
1997br 0.007 0.993 1.120 (0.149)
1998V 0.018 0.983 1.047 (0.166) with fewer parameters. As in Davis et al. (2007), we use
1998ab 0.027 0.974 0.987 (0.151) the Akaike information criterion (AIC; Akaike 1974). For
1998dm 0.007 0.993 0.778 (0.119) Gaussian errors (which is the case here, see Section 3.3),
1998eg 0.025 0.976 0.883 (0.085) this criterion can be expressed as
1999cl 0.008 0.992 0.870 (0.122)
1999dq 0.014 0.986 0.928 (0.055)
1999ej 0.014 0.986 1.147 (0.159) AIC = χ2 + 2k, (2)
1999gp 0.027 0.974 0.839 (0.060)
2000fa 0.021 0.979 0.884 (0.076) where k is the number of free parameters (Davis et al.
2001V 0.015 0.985 0.973 (0.072) 2007). Comparison of models simply involves comput-
2001eh 0.037 0.964 1.148 (0.107) ing the difference in AIC (∆AIC) with respect to the
2001ep 0.013 0.987 0.925 (0.086) model with the lowest value for this criterion. A differ-
2002ha 0.014 0.986 0.973 (0.134)
2003cg 0.004 0.996 1.006 (0.131) ence in AIC of 2 is considered positive evidence against
2003du 0.006 0.994 0.909 (0.091) the model with the higher AIC, whereas a difference of
2006lf 0.013 0.987 0.941 (0.110) 6 is considered strong evidence (Liddle 2004; Davis et al.
High redshift (z > 0.2) 2007). In the models considered here (see Table 4), the
1996bj 0.574 0.635 0.527 (0.369)
1997ex 0.361 0.735 0.745 (0.076)
expected 1/(1 + z) time-dilation model has the lowest
2001go 0.552 0.644 0.652 (0.062) AIC (although this is not true for the low-redshift sam-
2002iz 0.427 0.701 0.655 (0.089) ple), and we compute AIC differences with respect to
b027 0.315 0.760 0.823 (0.092) that model.
2003js 0.363 0.734 0.718 (0.082)
04D2an 0.621 0.617 0.567 (0.341) With ∆AIC = 1, we conclude that the information
2006mk 0.475 0.678 0.753 (0.060) criteria do not provide positive evidence against a 1/(1 +
2006sc 0.357 0.737 0.619 (0.121) z)b dependence of the aging rate. The χ2 per degree of
2006tk 0.312 0.762 0.835 (0.181)
2007tg 0.502 0.666 0.687 (0.102)
freedom is also satisfactory for the samples that include
2007tt 0.374 0.728 0.718 (0.108) the high-redshift objects.
2007un 0.283 0.779 0.759 (0.135) The two other models considered previously have no
free parameters (k = 0 in Eq. [2]), hence ∆AIC = ∆χ2 ,
and the information criterion is reduced to a simple χ2
test.
5. CONCLUSION
We have presented 35 spectra of 13 high-redshift
(0.28 ≤ z ≤ 0.62) SNe Ia, which include previously un-
published data from the ESSENCE and SNLS projects
and from our own dedicated program at the ESO Very
Large Telescope. Given the rapid and predictable evolu-
tion of SN Ia spectral features with age, as well as the
relative homogeneity of SN Ia spectra at a given age, one
is able to determine the (rest-frame) age of a single spec-
trum with a typical accuracy of 1–3 d (Riess et al. 1997;
Foley et al. 2005; Hook et al. 2005; Howell et al. 2005;
Blondin & Tonry 2007).
Using the Supernova Identification (SNID) code of
Blondin & Tonry (2007), we determine the ages of each
spectrum in the supernova rest frame. Comparison with
Fig. 9.— Ratio of the aging rate to 1/(1 + z) for all SNe Ia in
the observed time difference between the spectra yields
Fig. 8. Both the low-redshift (open histogram) and high-redshift an apparent aging rate consistent with 1/(1 + z), as ex-
(hatched histogram) samples are shown. pected in a homogeneous and isotropic expanding uni-
verse. Moreover, the data unambiguously rule out the
have profound implications for our assumption of FLRW “tired light” hypothesis (Zwicky 1929) in which photons
cosmology. lose energy as they interact with matter and other pho-
Again, we performed a least-squares fit to the entire tons in a static universe.
sample, and also to the individual high- and low-redshift The fact that the age determination is so accurate over
samples (see Table 4). The data constrain the b expo- a large redshift range also shows that the deviations be-
nent to 10% (1σ), and yield b = 0.97 ± 0.10 for the en- tween spectra of low- and high-redshift SNe Ia in our
tire sample (Fig. 8; dotted line and gray region) and sample are small.
10 Blondin et al.

TABLE 4
Time-dilation model comparison

All SNe High-redshift SNe only Low-redshift SNe only

Modela χ2 /dof GoF ∆AIC χ2 /dof GoF ∆AIC χ2 /dof GoF ∆AIC
(%) (%) (%)
1/(1 + z) 27.0/35 83.2 0 3.6/13 99.5 0 23.4/22 38.2 0
1/(1 + z)b 26.9/34 80.2 1 3.4/12 99.2 1 20.3/21 50.0 −1
tired light 150.3/35 0.0 123 123.4/13 0.0 119 26.9/22 21.4 3
a The best-fit values for the b exponent in the second model are as follows. All SNe: b =
0.97 ± 0.10; high-redshift SNe only: b = 0.95 ± 0.10; low-redshift SNe only: b = 3.18 ± 1.28

We also test for alternate dependencies of the aging The authors wish to thank the Supernova Legacy Sur-
rate on redshift, namely 1/(1 + z)b , although these are vey (SNLS) collaboration, and in particular Stéphane
not predicted by any model. Whether we consider the Basa and Tianmeng Zhang, for providing spectra of su-
entire sample or only the high-redshift sample, the best- pernova 04D2an prior to publication. The ESSENCE
fit value for the b exponent is consistent with b = 1, and survey is supported by the US National Science Foun-
thus with the expected 1/(1 + z) factor. dation (NSF) through grants AST 04-43378 and AST
That these data provide a confirmation of the time- 05-07475. Support for supernova research at Harvard
dilation factor expected in an expanding universe should University, including the CfA Supernova Archive, is pro-
be of no surprise. Nonetheless, previous use of SN Ia light vided in part by NSF grant AST 06-06772. The Dark
curves to test this hypothesis (Leibundgut et al. 1996; Cosmology Centre is funded by the Danish National Re-
Goldhaber et al. 2001) are prone to the spread in intrinsic search Foundation. A. C. acknowledges the support of
light-curve widths and its possible variation with redshift CONICYT, Chile, under grants FONDECYT 1051061
(which includes selection effects; see Section 1). and FONDAP 15010003. T. M. D. appreciates the sup-
The data presented here are unique in that they en- port of the Villum Kann Rasmussen Fonden. A. V. F.
able the most direct test of the 1/(1 + z) time-dilation is grateful for the support of NSF grant AST 06-07894.
hypothesis over a larger redshift range than has yet been G. P. acknowledges the support of the Proyecto FONDE-
performed. This hypothesis is favored beyond doubt over CYT 3070034.
models that predict no time dilation. With more data, Facilities: VLT:Kueyen (FORS1), VLT:Antu
the focus will shift to testing more thoroughly the alter- (FORS2), Gemini:South (GMOS), Gemini:Gillete
native 1/(1+z)b dependence of the aging rate on redshift. (GMOS), Keck:I (LRIS), Keck:II (DEIMOS, ESI),
Any significant deviation from b = 1 would have a pro- Magellan:Baade (IMACS), Magellan:Clay (LDSS3).
found impact on our assumption of a FLRW cosmology
to describe the universal expansion.
APPENDIX
A. TIME DILATION IN AN EXPANDING UNIVERSE
In a homogeneous, isotropic expanding universe, the interval dτ between two space-time events is given by the
Robertson-Walker (RW) metric (Robertson 1935, 1936a,b; Walker 1936),
dr2
 
2
dτ 2 = c2 dt2 − a2 (t) + r 2
(dθ 2
+ sin θdφ 2
) , (A1)
1 − kr 2
where c is the speed of light in vacuum, t is the cosmic time, (r, θ, φ) are the comoving spatial coordinates, k is the
curvature parameter, and a(t) is the dimensionless scale factor. In what follows we assume the present-day value of
the dimensionless scale factor a0 = 1.
Photons travel along null geodesics (dτ 2 = 0). In what follows we consider radial null rays only (dθ = dφ = 0). For
a photon emitted at time t1 from an object located at (r1 , θ1 , φ1 ) and observed at time t0 , Eq. [A1] implies
Z t0 Z r1
cdt dr
= √ ≡ f (r1 ). (A2)
t1 a(t) 0 1 − kr 2
Here we assume that the object from which the photon was emitted has constant coordinates (r 1 , θ1 , φ1 ) such that
f (r1 ), also known as the comoving distance, is time independent. Thus, for a photon emitted at time t 1 + δt1 and
observed at time t0 + δt0 , Eq. [A1] also implies
Z t0 +δt0
cdt
= f (r1 ). (A3)
t1 +δt1 a(t)
For small δt1 (and hence small δt0 ), the rate of change of the scale factor remains roughly constant and Eqs. [A2] and
[A3] imply
δt0 δt1
= . (A4)
a0 a(t1 )
 Time Dilation in SN Ia Spectra 11

Hence, a light signal emitted with frequency ν1 will reach us with frequency ν0 such that
ν0 δt1 a(t1 )
= = . (A5)
ν1 δt0 a0
Using the standard definition of redshift, z = (λ0 − λ1 )/λ1 = ν1 /ν0 − 1, we obtain a relationship between observed
and rest-frame time intervals in a RW metric as a function of redshift z:
δt0
= 1 + z. (A6)
δt1
A supernova at redshift z will thus appear to age (1 + z) times more slowly with respect to a local event at z ≈ 0.
The prediction of time dilation proportional to (1+z) is generic to expanding universe models, whether the underlying
theory be general relativity (e.g., the Friedmann-Lemaı̂tre-Robertson-Walker universe), special relativity (e.g., the
Milne Universe), or Newtonian expansion. A point of confusion can occur in the special relativistic case for which the
well-known time-dilation factor is given by
  v 2 −1/2
γSR = 1 − (A7)
c
 
1 1
= 1+z+ , (A8)
2 1+z
which evidently differs from (1+z). Thus it might be assumed that a special relativistic expansion can be distinguished
from the FLRW universe using a time-dilation test27 .
This is not the case. Special relativistic expansion of the universe assumes there is an inertial frame that extends
to infinity (impossible in the non-empty general relativistic picture) and that the expansion involves objects moving
through this inertial frame. The time-dilation factor from Eq. [A8] relates the proper time in the moving emitter’s
inertial frame (δt1 ) to the proper time in the observer’s inertial frame (δt0 ). To measure this time dilation the observer
has to set up a set of synchronized clocks (each at rest in the observer’s inertial frame) and take readings of the
emitter’s proper time as the emitter moves past each synchronized clock. The readings show that the emitter’s clock
is time dilated such that δt0 = γSR δt1 .
We do not have this set of synchronized clocks at our disposal when we measure time dilation of supernovae in an
expanding universe and therefore Eq. [A8] is not the time dilation we observe. We must also take into account an
extra time-dilation factor that occurs because the distance to the emitter (and thus the distance light has to propagate
to reach us) is increasing. In the time δt0 the emitter moves a distance vδt0 away from us. The total proper time we
observe, δt0,tot , is δt0 plus an extra factor describing how long light takes to traverse this extra distance (vδt 0 /c),

δt0,tot = δt0 (1 + v/c). (A9)

The relationship between proper time at the emitter and proper time at the observer is thus

δt0,tot = γSR δt1 (1 + v/c) (A10)

s
1 + v/c
= δt1 (A11)
1 − v/c
= δt1 (1 + z), (A12)
which is identical to the GR time-dilation equation.
Non-cosmological redshifts (i.e., not due to universal expansion) also cause a time-dilation effect described by
Eq. [A6]. However, these additional effects from peculiar velocities and gravitational redshifts contribute random
error only, and do not bias the measurements presented here.
B. COMPARISON OF SPECTRAL AND LIGHT-CURVE AGES
To test the accuracy of the age determination using SNID, we select the SNe Ia for which a well-sampled light curve
is available around maximum light. Only the ESSENCE and SNLS SNe Ia in our sample have associated light curves
for which we could determine the date of maximum brightness (tmax ). To do so we used the MLCS2k2 light-curve
fitting code of Jha et al. (2007), as done by Wood-Vasey et al. (2007). This way we can determine the time difference
(in the observer frame) between maximum light (tmax ) and the time the spectrum was obtained (tobs ). We compare
this time interval with the rest-frame age determined through cross-correlation with local SN Ia spectral templates
using SNID (tspec ). We expect a one-to-one correspondence between
tobs − tmax
tLC = (B1)
1+z
27 In fact, such an erroneous assumption was made by one of the current authors of Davis & Lineweaver (2004).
12 Blondin et al.

Fig. 10.— Upper panel: comparison of supernova rest-frame ages (in days from maximum light) obtained from cross-correlation with
spectral templates (tspec ) and from fits to the light curve (tLC ). 145 age measurements for the subsample of 22 low-redshift SNe Ia are shown
in gray. The dashed line represents the one-to-one correspondence between t LC and tspec . Middle panel: Age residuals, ∆t = tspec − tLC .
We also indicate the standard deviation (σ) and mean residual (µ). Lower panel: Same as above, where each point has been corrected for
the mean offset between tspec and tLC for a given supernova.

and tspec .
The result is shown as black points in Fig. 10. While the agreement is good, there is a mean systematic offset of −1.6
d between tspec and tLC , as shown in the middle panel. If this offset were to affect only a subset of age measurements
for a given supernova, the impact on the aging rate determination would be severe. To check this, we correct the
spectral ages of a given supernova for the mean difference between tLC and tspec . This “corrected” age residual, ∆tcorr
is plotted in the lower panel of Fig. 10. The mean residual drops to −0.1 d and the scatter decreases slightly.
Since there are 2 to 4 tspec measurements for a given supernova, and only one measurement of tmax , the source of the
discrepancy between the spectral and light-curve ages is most likely due to the determination of the date of maximum
using the light-curve fitter. Indeed, using a different light-curve fitter (SALT2; Guy et al. 2007) yields values for t max
that differ from the MLCS2k2 measurements by more than one day in 9 out of 10 cases, and by more than two days
for three objects (SNe 2003js, 2007tg, and 2007un). These discrepancies are due to a combination of differences in
light-curve fitter algorithms and data quality (light-curve sampling around maximum light and signal-to-noise ratio of
each light-curve measurement; see Miknaitis et al. 2007).
Therefore, while there is a systematic offset between part of these different age determinations, this offset affects all
measurements in a similar fashion and has no impact on the determination of the rate of aging. In fact, the main result
of this paper (see Section 4) is completely independent of tLC , and hence of tmax . Nonetheless, the comparison between
spectral and light-curve ages confirms the accuracy of age determination using spectra alone (Blondin & Tonry 2007).
The age measurements for all the high-redshift SNe Ia in our sample are reported in Table 5.
In making the comparison we have implicitly assumed what we are trying to show, namely a time-dilation factor of
(1+z). Accordingly, we also make the same comparison for our subsample of 22 low-redshift SNe Ia (0.002 ≤ z ≤ 0.04).
At such low redshifts, the (1 + z) correction present in tLC is negligible (the mean correction is ∼ 0.06 d). The result is
shown as gray points in Fig. 10. The mean residual between tLC and tspec for this low-redshift sample is close to zero
with a small scatter (σ ≈ 1.5 d), and unlike the high-redshift sample there is no significant systematic offset between
the two age measurements.
The age measurements presented in Table 2 also enable us to infer the date of maximum light for each supernova
using spectra alone (corresponding to tspec = 0). This way we are able to determine the time of maximum for the SNe Ia
in our sample for which a well-sampled light curve was unavailable (SNe 1996bj, 1997ex, and 2001go; see Table 6). We
 Time Dilation in SN Ia Spectra 13

TABLE 5
Comparison of rest-frame light-curve and spectral
ages

SN tLC tspec ∆t ∆tcorr

(1) (2) (3) (4) (5)
2002iz 0.1 (1.1) −0.5 (2.2) −0.6 (2.4) 0.8 (1.5)
19.4 (1.1) 17.6 (1.2) −1.8 (1.6) −0.4 (2.3)
b027 −2.4 (0.5) −3.5 (1.8) −1.1 (1.9) −0.9 (1.7)
17.9 (0.5) 18.4 (1.6) 0.5 (1.7) 0.7 (1.4)
2003js −3.2 (0.3) −4.9 (1.6) −1.7 (1.6) 0.3 (2.0)
14.6 (0.3) 12.5 (1.2) −2.1 (1.2) −0.1 (2.3)
04D2an −3.2 (0.9) −2.5 (1.6) 0.7 (1.8) 0.2 (1.4)
0.5 (0.9) 0.9 (1.3) 0.4 (1.6) −0.1 (1.3)
2006mk −3.6 (0.7) −6.2 (1.0) −2.6 (1.3) −0.5 (2.7)
2.6 (0.7) −0.6 (2.2) −3.2 (2.3) −1.1 (3.3)
10.0 (0.7) 7.3 (1.9) −2.7 (2.0) −0.6 (2.8)
18.2 (0.7) 18.5 (1.8) 0.3 (1.9) 2.4 (0.9)
2006sc 0.9 (0.5) 0.9 (1.6) 0.0 (1.7) 0.9 (1.0)
10.5 (0.5) 9.8 (1.4) −0.7 (1.5) 0.2 (1.2)
16.4 (0.5) 13.4 (2.2) −3.0 (2.3) −2.1 (3.2)
2006tk −6.1 (0.5) −8.8 (2.4) −2.7 (2.5) −0.8 (2.8)
2.3 (0.5) 0.3 (2.0) −2.0 (2.1) −0.1 (2.2)
4.6 (0.5) 2.9 (0.9) −1.7 (1.0) 0.2 (1.9)
2007tg −6.6 (0.9) −6.1 (2.0) 0.5 (2.2) 0.2 (1.2)
0.0 (0.9) −0.5 (1.8) −0.5 (2.0) −0.8 (1.2)
9.3 (0.9) 10.0 (1.5) 0.7 (1.8) 0.4 (1.3)
2007tt −2.5 (0.6) −5.0 (2.1) −2.5 (2.2) −0.1 (2.7)
8.3 (0.6) 6.1 (1.4) −2.2 (1.6) 0.2 (2.5)
17.7 (0.6) 14.9 (2.2) −2.8 (2.2) −0.4 (3.0)
2007un 4.3 (0.4) 3.2 (2.2) −1.1 (2.3) 0.1 (1.4)
12.1 (0.4) 11.1 (1.3) −1.0 (1.3) 0.2 (1.4)
19.1 (0.4) 17.7 (1.4) −1.4 (1.5) −0.2 (1.7)
Column headings: (1) SN name. (2) SN rest-frame age in
days from maximum light, derived from the light curve. (3)
SN rest-frame age in days from maximum light, derived from
the cross-correlation with spectral templates using SNID. (4)
∆t = tspec − tLC . (5) ∆t corrected for the mean offset between
tspec and tLC .

TABLE 6
Comparison of dates of maximum light

SN tLC
max tspec
max ∆tmax
(1) (2) (3) (4)
1996bj ··· 372.16 (3.73) ···
1997ex ··· 817.16 (1.98) ···
2001go ··· 2011.47 (3.12) ···
2002iz 2586.83 (1.51) 2587.71 (3.21) 0.88 (3.54)
b027 2593.09 (0.65) 2594.20 (1.90) 1.11 (2.01)
2003js 2946.80 (0.47) 2949.29 (1.67) 2.49 (1.73)
04D2an 3031.36 (1.50) 3030.61 (1.82) −0.75 (2.36)
2006mk 4036.95 (0.96) 4040.32 (1.02) 3.37 (1.40)
2006sc 4062.39 (0.64) 4061.71 (2.77) −0.68 (2.84)
2006tk 4097.56 (0.66) 4100.13 (0.98) 2.57 (1.18)
2007tg 4391.62 (1.34) 4391.34 (1.58) −0.28 (2.08)
2007tt 4419.29 (0.83) 4422.47 (1.83) 3.18 (2.01)
2007un 4436.12 (0.47) 4437.15 (3.12) 1.03 (3.16)
Column headings: (1) SN name. (2) JD − 2,450,000 of
maximum light, derived from the light curve. (3) JD −
2,450,000 of maximum light, derived from the spectra. (4)
∆tmax = tspec LC
max − tmax .

can also compare the dates of maximum as inferred from a fit to the light curve (t LC
max ) with those determined from
the spectra alone (tspec
max ). The results are also shown in Table 6. For four objects (SNe 2003js, 2006mk, 2006tk, and
2007tt) the disagreement is larger than 1σ, and explains the systematic negative offset between t spec and tLC seen in
Fig. 10.

REFERENCES
Akaike, H. 1974, IEEE Transactions on Automatic Control, 19, 716 Blondin, S. & Tonry, J. L. 2007, ApJ, 666, 1024
Benetti, S., et al. 2005, ApJ, 623, 1011 Conley, A., et al. 2006, AJ, 132, 1707
Blondin, S., et al. 2006, AJ, 131, 1648
14 Blondin et al.

Davis, T. M. & Lineweaver, C. H. 2004, Publications of the Nomoto, K., Thielemann, F.-K., & Yokoi, K. 1984, ApJ, 286, 644
Astronomical Society of Australia, 21, 97 Nugent, P., Phillips, M., Baron, E., Branch, D., & Hauschildt, P.
Davis, T. M., et al. 2007, ApJ, 666, 716 1995, ApJ, 455, L147+
Falco, E. E., et al. 1999, PASP, 111, 438 Phillips, M. M. 1993, ApJ, 413, L105
Foley, R. J., Filippenko, A. V., Leonard, D. C., Riess, A. G., Phillips, M. M., Lira, P., Suntzeff, N. B., Schommer, R. A., Hamuy,
Nugent, P., & Perlmutter, S. 2005, ApJ, 626, L11 M., & Maza, J. 1999, AJ, 118, 1766
Goldhaber, G., et al. 2001, ApJ, 558, 359 Pinto, P. A. & Eastman, R. G. 2000, ApJ, 530, 757
Guy, J., et al. 2007, A&A, 466, 11 Riess, A. G., et al. 1997, AJ, 114, 722
Hamuy, M., Phillips, M. M., Maza, J., Suntzeff, N. B., Schommer, Riess, A. G., et al. 1999, AJ, 118, 2675
R. A., & Aviles, R. 1995, AJ, 109, 1 Riess, A. G., Press, W. H., & Kirshner, R. P. 1995, ApJ, 438, L17
Hamuy, M., et al. 2003, Nature, 424, 651 Robertson, H. P. 1935, ApJ, 82, 284
Hicken, M., Garnavich, P. M., Prieto, J. L., Blondin, S., DePoy, —. 1936a, ApJ, 83, 187
D. L., Kirshner, R. P., & Parrent, J. 2007, ApJ, 669, L17 —. 1936b, ApJ, 83, 257
Hook, I. M., et al. 2005, AJ, 130, 2788 Rust, B. W. 1974, PhD thesis, Oak Ridge National Lab., TN.
Howell, D. A., Sullivan, M., Conley, A., & Carlberg, R. 2007, ApJ, Stehle, M., Mazzali, P. A., Benetti, S., & Hillebrandt, W. 2005,
667, L37 MNRAS, 360, 1231
Howell, D. A., et al. 2006, Nature, 443, 308 Stritzinger, M. 2005, PhD thesis, Technische Universität München,
Howell, D. A., et al. 2005, ApJ, 634, 1190 Tonry, J. & Davis, M. 1979, AJ, 84, 1511
Jha, S., Riess, A. G., & Kirshner, R. P. 2007, ApJ, 659, 122 Walker, A. G. 1936, in Proceedings of the London Mathematical
Leibundgut, B., et al. 1996, ApJ, 466, L21 Society, 90–127
Li, W., et al. 2003, PASP, 115, 453 Weinberg, S. 1972, Gravitation and Cosmology: Principles and
Li, W., et al. 2001a, PASP, 113, 1178 Applications of the General Theory of Relativity (Gravitation
Li, W., Filippenko, A. V., Treffers, R. R., Riess, A. G., Hu, J., & and Cosmology: Principles and Applications of the General
Qiu, Y. 2001b, ApJ, 546, 734 Theory of Relativity, by Steven Weinberg, pp. 688. ISBN 0-471-
Liddle, A. R. 2004, MNRAS, 351, L49 92567-5. Wiley-VCH , July 1972.)
Lidman, C., et al. 2005, A&A, 430, 843 Wilson, O. C. 1939, ApJ, 90, 634
Matheson, T., et al. 2005, AJ, 129, 2352 Wood-Vasey, W. M., et al. 2007, ApJ, 666, 694
Matheson, T., et al. 2008, accepted for publication in AJ Zwicky, F. 1929, Proceedings of the National Academy of Science,
Mazzali, P. A., Sauer, D. N., Pastorello, A., Benetti, S., & 15, 773
Hillebrandt, W. 2008, ArXiv e-prints, 803
Miknaitis, G., et al. 2007, ApJ, 666, 674