1: %% Beginning of file 'snid.tex'

2: %%

3: %% This is a sample manuscript marked up using the

4: %% AASTeX v5.x LaTeX 2e macros.

5: 

6: %\documentclass[12pt,preprint]{aastex}

7: \documentclass{emulateapj}

8: 

9: %% manuscript produces a one-column, double-spaced document:

10: %% \documentclass[manuscript]{aastex}

11: 

12: %% preprint2 produces a double-column, single-spaced document:

13: %% \documentclass[preprint2]{aastex}

14: 

15: %% load packages

16: 

17: %% NOT SUPPORTED BY ApJ

18: %\usepackage{multirow}

19: 

20: %% new commands

21: \def\kms{\hbox{$~$km$~$s$^{-1}$}}

22: \def\l{\ifmmode\lambda\else$\lambda$\fi}

23: \def\snid{\ifmmode{\rm SNID}\else{SNID}\fi}

24: 

25: % ions

26: \def\ctwo{C~{\sc ii}}

27: \def\cfour{C~{\sc iv}}

28: \def\catwo{Ca~{\sc ii}}

29: \def\cotwo{Co~{\sc ii}}

30: \def\fetwo{Fe~{\sc ii}}

31: \def\fethree{Fe~{\sc iii}}

32: \def\ha{H$\alpha$}

33: \def\heone{He~{\sc i}}

34: \def\mgtwo{Mg~{\sc ii}}

35: \def\ntwo{N~{\sc ii}}

36: \def\naone{Na~{\sc i}}

37: \def\naoned{Na~{\sc i}D}

38: \def\nitwo{Ni~{\sc ii}}

39: \def\oone{O~{\sc i}}

40: \def\otwo{O~{\sc ii}}

41: \def\othree{O~{\sc iii}}

42: \def\stwo{S~{\sc ii}}

43: \def\sitwo{Si~{\sc ii}}

44: \def\sithree{Si~{\sc iii}}

45: \def\sifour{Si~{\sc iv}}

46: \def\titwo{Ti~{\sc ii}}

47: 

48: %% You can insert a short comment on the title page using the command below.

49: \slugcomment{The Astrophysical Journal, 666:1024--1047, 2007 September 10}

50: \shorttitle{Type, Redshift, \& Age of SN spectra}

51: \shortauthors{Blondin \& Tonry}

52: 

53: 

54: 

55: \begin{document}

56: 

57: \title{Determining the Type, Redshift, and Age of a Supernova Spectrum}

58: 

59: \author{St{\'e}phane Blondin}

60: \affil{Harvard-Smithsonian Center for Astrophysics, Cambridge, MA 02138}

61: \email{sblondin@cfa.harvard.edu}

62: 

63: \and

64: 

65: \author{John L. Tonry}

66: \affil{Institute for Astronomy, University of Hawaii, Honolulu, HI 96822}

67: \email{jt@ifa.hawaii.edu}

68: 

69: 

70: 

71: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

72: %%

73: %%   Abstract

74: %%

75: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

76: 

77: \begin{abstract}

78: We present an algorithm to identify the type of an SN spectrum

79: and to determine its redshift and age. This algorithm, based on the 

80: correlation techniques of Tonry \& Davis, is implemented in the

81: Supernova Identification (\snid) code. It is used by members of 

82: ongoing high-redshift SN searches to distinguish between

83: type Ia and type Ib/c SNe, and to identify ``peculiar'' SNe~Ia. We

84: develop a diagnostic to quantify the quality of a correlation between

85: the input and template spectra, which enables a formal evaluation of

86: the associated redshift error. Furthermore, by comparing the

87: correlation redshifts obtained using \snid\ with those determined from

88: narrow lines in the SN host galaxy spectrum, we show that

89: accurate redshifts (with a typical error $\sigma_z \lesssim 0.01$) can

90: be determined for SNe~Ia without a spectrum of the host galaxy. 

91: Last, the age of an input spectrum is determined with a

92: typical accuracy $\sigma_t \lesssim 3$ days, shown here by using

93: high-redshift SNe~Ia with well-sampled light curves. The success of

94: the correlation technique 

95: % provides an additional argument in favor of

96: % the similarity of SNe~Ia at low and high redshifts. 

97: confirms the similarity of some SNe Ia at low and high redshifts.

98: The \snid\ code,

99: which will be made available to the community, can also be used for 

100: comparative studies of SN spectra, as well as comparisons

101: between data and models.

102: \end{abstract}

103: 

104: \keywords{methods: data analysis ---  methods: statistical ---

105:   supernovae: general}

106: 

107: 

108: 

109: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

110: %%

111: %%   Introduction

112: %%

113: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

114: 

115: \section{Introduction}\label{Sect:intro}

116: 

117: Supernovae (SNe) play a major role in the recent revival of observational

118: cosmology. It is through comparison of high-redshift Type Ia

119: supernova (SN~Ia) magnitudes with those at low-redshift

120: \citep{Hamuy/etal:1996,Riess/etal:1999,Jha/etal:2006a} that two teams

121: independently found the present rate of the universal expansion to be

122: accelerating \citep{Riess/etal:1998a,Perlmutter/etal:1999}. This

123: astonishing result has been confirmed in subsequent years out to

124: redshift $z \lesssim 1$

125: \citep{Tonry/etal:2003,Knop/etal:2003,Barris/etal:2004}, but also at

126: higher redshifts where the universal expansion is in a decelerating

127: phase \citep{Riess/etal:2004}. Currently, two ongoing projects have

128: the more ambitious goal to measure the equation-of-state parameter,

129: $w$, of the ``dark energy'' that drives the expansion: the ESSENCE

130: (Equation of State: SupErNovae trace Cosmic Expansion;

131: \citealt{Miknaitis/etal:2007,Wood-Vasey/etal:2007}) and SNLS

132: (SuperNova Legacy Survey; \citealt{Astier/etal:2006}) projects. Both

133: teams have published their initial results, which indicate that $w$,

134: if constant, is consistent with a cosmological constant ($w=-1$).

135: 

136: 

137: The success of these cosmological experiments depends, among other

138: things, on the assurance that the supernovae in the sample are of the

139: correct type, namely, SNe~Ia. The classification of supernovae is

140: based on their optical spectra around maximum light (for a review see

141: \citealt{Filippenko:1997}). At high redshifts, obtaining

142: sufficiently high signal-to-noise ratio (S/N) spectra of such objects

143: requires 1--2\,hr integrations at 6.5--10\,m-class telescopes (see,

144: e.g., \citealt{Matheson/etal:2005}), and constitutes the limiting

145: factor for these experiments. Recently, alternative classification

146: methods based on photometry alone have been suggested

147: \citep{Poznanski/Maoz/Gal-Yam:2006,Kuznetsova/Connolly:2007,Kunz/etal:2006},

148: in anticipation of the next generation of wide-field all-sky surveys

149: optimized for the detection of transient events

150: (Dark Energy Survey, \citealt{Frieman:2004}; Pan-STARRS,

151: \citealt{Kaiser/etal:2002}; SKYMAPPER, \citealt{Schmidt/etal:2005};

152: ALPACA, \citealt{Crotts/etal:2006}; LSST,

153: \citealt{Tyson/Angel:2001}). Inclusion of supernovae that are 

154: of a different type leads to biased cosmological parameters

155: \citep{Homeier:2005}. Exclusion of genuine SNe~Ia from the sample

156: leads to increased statistical errors on these same parameters.

157: 

158: 

159: The secure classification of supernovae is a challenge at all

160: redshifts, however. Even with high S/N spectra, the distinction

161: between supernovae of different types (or between subtypes within a

162: given type) can pose problems. This points to the inadequacy of the

163: present purely empirical SN classification scheme in establishing

164: distinct classes of supernovae, whose observational properties can be

165: directly traced back to an explosion mechanism and a progenitor

166: system. The two major types of supernovae are defined based on the

167: presence (Type II) or absence (Type I) of hydrogen in their

168: spectra, a distinction that does not reflect the differences in

169: their explosion mechanisms and progenitors: through the

170: thermonuclear disruption of a carbon-oxygen white dwarf star (Type

171: Ia), or through the collapse of the degenerate core of a massive

172: star (Types Ib, Ic, and II). For the latter case, it is now thought

173: that a there exists a continuity of events between the types

174: II$\rightarrow$Ib$\rightarrow$Ic, corresponding to increasing mass

175: loss of the outer envelope of the progenitor star prior to explosion

176: \citep{Chevalier:2006}. SNe~IIb are an

177: intermediate case between Type II and Type Ib, and illustrate the

178: tendency of some supernovae to ``evolve'' from one type to

179: another. SNe~Ic supernovae are only defined by the {\it absence} of

180: elements 

181: (hydrogen and helium; although see \citealt{Elmhamdi/etal:2006} for

182: the presence of hydrogen in SNe~Ib/c) in their atmospheres and thus

183: form a heterogeneous class--- which includes the supernovae associated

184: with gamma-ray bursts

185: \citep{Kulkarni/etal:1998,Matheson/etal:2003}. The classification 

186: scheme is further complicated by ``peculiar'' sub-classes of events

187: associated with the four types (Ia, Ib, Ic, and

188: II). Nonetheless, this classification scheme provides a means to keep

189: track of general spectroscopic properties associated with the many

190: supernovae discovered each year 

191: % (more than per year since 2000 according

192: (more than 550 in 2006 according

193: to the International Astronomical Union\footnote{{\tt

194:     http://www.cfa.harvard.edu/iau/lists/Supernovae.html}}) and is

195: useful for comparative studies of supernovae with similar

196: characteristics.

197: 

198: 

199: The spectrum of a supernova also contains information on its redshift

200: and age (

201: %hereafter ``phase'', 

202: defined as the number of days from

203: maximum light in a given filter). Knowledge of the SN redshift is

204: necessary for the use of SNe~Ia as distance indicators (although see

205: \citealt{Barris/Tonry:2004} for redshift-independent distances), and

206: is usually determined using narrow lines in the spectrum of the host

207: galaxy. When such a spectrum is unavailable, one has to rely on

208: comparison with SN template spectra for determining the redshift,

209: although we note that \citet{Wang:2007} has recently presented a 

210: purely photometric redshift estimator for SNe~Ia, albeit with 3--5

211: times larger errors.

212: The SN age is usually determined (to within 1 day) using a

213: well-sampled light curve, but a single spectrum can also provide a

214: good estimate (to within 2--3 days for SNe~Ia;

215: \citealt{Riess/etal:1997}), since the relative strengths and

216: wavelength location of spectral features evolve 

217: significantly on the timescale of days. Knowledge of

218: the age of the supernova and its apparent magnitude and color at a

219: single epoch can yield a distance measurement accurate to $\sim 10\%$  

220: \citep{Riess/etal:1998b}. Moreover, comparison of spectral and

221: light-curve ages of high-redshift supernovae can be used to test the

222: expected time-dilation factor of $(1+z)$, where $z$ is the redshift,

223: in an expanding universe \citep{Riess/etal:1997,Foley/etal:2005}.

224: 

225: 

226: We have developed a tool (Supernova Identification [\snid]) to determine

227: the type, redshift, and age of a supernova, using a single

228: spectrum. The algorithm is based on the correlation techniques of 

229: \citet{Tonry/Davis:1979}, and relies on the comparison of an input

230: spectrum with a database of high-S/N template spectra. Fundamental to

231: the success of the correlation technique is its application to objects

232: that have counterparts in the template database. We briefly

233: describe the cross-correlation technique in the next section, before

234: presenting the algorithm for determining the redshift

235: (\S~\ref{Sect:z}). We then briefly comment on the composition of our

236: spectral database (\S~\ref{Sect:sniddb}), before testing the accuracy

237: of correlation redshifts and ages using \snid\ in

238: \S~\ref{Sect:zt}. Last, in \S~\ref{Sect:type}, we tackle the issue of

239: supernova classification by focusing on specific examples, some of

240: which are particularly relevant to SN searches at high redshifts.

241: 

242: 

243: 

244: 

245: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

246: %%

247: %%   Section 2 - Cross-correlation formalism

248: %%

249: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

250: 

251: \section{Cross-correlation Formalism}\label{Sect:xcorform}

252: 

253: The cross-correlation method presented in this section is extensively

254: discussed by \citet{Tonry/Davis:1979}, where it is exclusively applied

255: to galaxy spectra. We reproduce part of this discussion here to

256: highlight the specificity of determining supernova (as opposed to

257: galaxy) redshifts.

258: 

259: Sections \ref{Sect:defs} and \ref{Sect:xcorz} are rather

260: technical, while \S~\ref{Sect:preproc} presents the more practical

261: aspects of spectrum pre-processing necessary for the

262: cross-correlation method.

263: 

264: \subsection{A Few Definitions\label{Sect:defs}}

265: 

266: The correlation technique is straightforward: a

267: supernova spectrum $s(n)$ whose redshift $z_s$ is to be found is

268: cross-correlated with a template spectrum (of known type and age)

269: $t(n)$ at zero redshift. We want to determine the $(1+z_s^\prime)$

270: wavelength scaling that maximizes the cross-correlation $c(n) = s(n)

271: \star t(n)$, where $\star$ denotes the cross-correlation product. In

272: practice, it is convenient to bin the spectra on a

273: logarithmic wavelength axis. Multiplying

274: the wavelength axis of $t(n)$ by a factor $(1+z)$ is equivalent

275: to adding a $\ln (1+z)$ shift to the logarithmic wavelength axis of

276: $t(n)$, i.e. a (velocity) redshift corresponds to a uniform linear

277: shift. Supposing we bin $s(n)$ and $t(n)$ into $N$ bins in the range

278: $[\l_0,\l_1]$, each wavelength coordinate $\l_{{\rm ln},n}$ is given

279: by

280: 

281: \begin{equation}

282: \label{Eqn:dllog}

283: \l_{{\rm ln},n} = \l_0\ e^{n \times d\l_{\rm ln}},

284: \end{equation}

285: 

286: \noindent

287: where $d\l_{\rm ln} = \ln{(\l_1/\l_0)}/N$ is the size of a

288: logarithmic wavelength bin, and assuming $n$ runs from 0 to $N$. We

289: then have:

290: 

291: \begin{equation}

292: \label{Eqn:nlnl}

293: n = A \ln{\l_{{\rm ln},n}} + B,

294: \end{equation}

295: 

296: \noindent

297: where $A = N / \ln{(\l_1/\l_0)}$ and $B = -N\ln{\l_0} /

298: \ln{(\l_1/\l_0)}$. In what follows we assume that $s(n)$ and $t(n)$

299: have been normalized such that their mean is zero (see

300: \S~\ref{Sect:preproc}).

301: 

302: For computational ease and for pre-processing purposes, the

303: cross-correlation is computed in Fourier space. Let $S(k)$ and $T(k)$

304: be the discrete Fourier transforms of the supernova and template

305: spectra, respectively ($k$ is the wavenumber):

306: 

307: \begin{equation}

308: S,T(k) = \sum_{n=0}^{N-1} s,t(n)\ e^{-2\pi ink/N}.

309: \end{equation}

310: 

311: \noindent

312: Let $\sigma_s$ and $\sigma_t$ be the rms of the

313: supernova and template spectrum, respectively:

314: 

315: \begin{equation}

316: \label{Eqn:sigma2}

317: \sigma_{s,t}^2 = \frac{1}{N} \sum_{n=0}^{N-1} s,t(n)^2.

318: \end{equation}

319: 

320: \noindent

321: The {\it normalized} correlation function $c(n)$ is defined as 

322: 

323: \begin{equation}

324: \label{Eqn:cnorm}

325: c(n) = s(n) \star t(n) = \frac{1}{N\sigma_s\sigma_t}

326: \sum_{m=0}^{N-1} s(m)t(m-n), 

327: \end{equation}

328: 

329: \noindent

330: such that if the supernova spectrum is the same as the template

331: spectrum, but shifted by $\delta$ logarithmic wavelength bins--- i.e. $s(n) =

332: t(n-\delta)$, then $c(\delta) = 1$. The Fourier transform of the

333: correlation function is

334: 

335: \begin{equation}

336: \label{Eqn:C(k)}

337: C(k) = \frac{1}{N\sigma_s\sigma_t} S(k) \overline{T(k)},

338: \end{equation}

339: 

340: \noindent

341: where $\overline{T(k)}$ denotes the complex conjugate of $T(k)$.

342:  

343: 

344: 

345: \subsection{Cross-correlation Redshifts\label{Sect:xcorz}}

346: 

347: Following \citet{Tonry/Davis:1979} we assume that $s(n)$ is some

348: multiple $\alpha$ of $t(n)$, but shifted by $\delta$ logarithmic

349: wavelength bins:

350: 

351: \begin{equation}

352: s(n) = \alpha t(n-\delta).

353: \end{equation}

354: 

355: \noindent

356: Unlike \citet{Tonry/Davis:1979}, however, we do not need to assume

357: that $t(n-\delta)$ is further convolved with a broadening symmetric

358: function [$b(n)$ in \citealt{Tonry/Davis:1979}, their eq.~6] that

359: accounts for galaxy stellar velocity dispersions and spectrograph

360: resolutions. While there exists a velocity dispersion residual in

361: $c(n)$ primarily due to differences in the dynamics of the expanding

362: envelope for different supernovae, this residual carries important

363: information on the age of the supernova, which we also want to

364: determine (see \S~\ref{Sect:zt}), and on the specificity of $s(n)$,

365: which is important for more general comparative supernova

366: studies. Second, the nominal Doppler width of a supernova spectral

367: ``feature'' is $\sim$1--2 orders of magnitude greater than the

368: resolution of a typical low-resolution spectrograph.

369: 

370: 

371: To estimate $\alpha$ and $\delta$, we need to minimize the following

372: expression:

373: 

374: \begin{eqnarray}

375: \label{Eqn:chi2}

376: \chi^2(\alpha,\delta) &=& \sum_{n=0}^{N-1} [\alpha t(n-\delta)-s(n)]^2 \\

377: \label{Eqn:chi2exp}

378: \Rightarrow \chi^2(\alpha,\delta) &=& \alpha^2 N \sigma_t^2 -

379: 2\alpha N \sigma_s \sigma_t c(\delta) + N \sigma_s^2,

380: \end{eqnarray}

381: 

382: \noindent

383: using eqs.~\ref{Eqn:sigma2} and \ref{Eqn:cnorm}. We then obtain the

384: condition for minimizing $\chi^2$ with respect to $\alpha$:

385: 

386: \begin{equation}

387: \frac{\partial \chi^2}{\partial \alpha} = 

388: 2N[\alpha \sigma_t^2 - \sigma_s \sigma_t c(\delta)] = 0,

389: \end{equation}

390: 

391: \noindent

392: from which we derive $\alpha_{\rm min}$ satisfying the above:

393: 

394: \begin{equation}

395: \alpha_{\rm min} = \frac{\sigma_s}{\sigma_t} c(\delta).

396: \end{equation}

397: 

398: \noindent

399: Substituting this value for $\alpha$ back into eq.~\ref{Eqn:chi2exp},

400: we obtain a new expression for $\chi^2$:

401: 

402: \begin{equation}

403: \chi^2(\alpha_{\rm min},\delta) = N \sigma_s^2 [1 - c(\delta)^2].

404: \end{equation}

405: 

406: \noindent

407: As expected, minimizing $\chi^2$ is equivalent to maximizing the

408: normalized correlation function $c(\delta)$.

409: 

410: 

411: Thus, the input supernova spectrum $s(n)$ is cross-correlated with a

412: template spectrum $t(n)$, and a smooth function (here a fourth-order

413: polynomial, as in \citealt{Tonry/Davis:1979}) is fitted to the highest

414: peak in $c(n)$, whose height and center determine $\alpha$ and

415: $\delta$, respectively. The cross-correlation redshift is then

416: trivially computed as

417: 

418: \begin{equation}

419: z_s^\prime = e^{\delta \times d\l_{\rm ln}} - 1,

420: \end{equation}

421: 

422: \noindent

423: where $d\l_{\rm ln}$ is the logarithmic wavelength bin defined in

424: eq.~\ref{Eqn:dllog}. The width of the peak is a measure of the error

425: in $z_s^\prime$ and is of the order of the typical width of a

426: supernova spectral feature, modulated by the signal-to-noise ratio of

427: the input spectrum (see \S~\ref{Sect:zerr}).

428: 

429: 

430: It is important to note that we assume the noise per pixel to be

431: constant in the input spectrum. This is clearly not the case for

432: ground-based optical spectra, where sharp emission features from the

433: sky background leads to increased noise at specific

434: wavelengths. Recently, \citet{Saunders/Cannon/Sutherland:2004} found

435: that scaling the input spectrum by the inverse-variance yielded a

436: dramatic improvement in the derived cross-correlation redshifts;

437: specifically, \citet{Saunders/Cannon/Sutherland:2004} rewrite

438: eq.~\ref{Eqn:chi2} as

439: 

440: \begin{equation}

441: \chi^2(\alpha,\delta) = \sum_{n=0}^{N-1} \left[ \frac{\alpha

442:     t(n-\delta)-s(n)}{\sigma(n)} \right]^2,

443: \end{equation}

444: 

445: \noindent

446: where $\sigma(n)$ is the noise per pixel of $s(n)$, and find that this

447: is equivalent to simply scaling $s(n)$ by $1/\sigma(n)^2$.

448: 

449: 

450: This modification is well suited for determining galaxy redshifts,

451: since sharp features in the variance spectrum (due to sky noise) have

452: widths comparable to galaxy lines and hence will affect the Fourier

453: transform of the correlation function, $C(k)$, at similar

454: wavenumbers. However, we have found no such improvement for

455: determining supernova redshifts. This is expected since supernova

456: spectra consist of overlapping Doppler-broadened lines whose widths

457: ($\sim 10000$\kms) are 1--2 orders of magnitude greater than sky

458: emission features. However, noise from an underlying galaxy continuum

459: can yield power at similar wavenumbers as $C(k)$ and can

460: significantly degrade the redshift accuracy when the fraction of

461: galaxy light in the supernova spectrum is high (see

462: \S~\ref{Sect:zt}).

463: 

464: 

465: \subsection{Pre-processing the Supernova Spectrum\label{Sect:preproc}}

466: 

467: As already mentioned, the input and template spectra are binned on a

468: common logarithmic wavelength scale, characterized by $(\l_0,\l_1,N)$

469: (eq.~\ref{Eqn:dllog}). We show the result of mapping an input

470: supernova spectrum onto a logarithmic wavelength axis with $(\l_0,

471: \l_1, N) = (2500$\,\AA$, 10000$\,\AA$, 1024)$ in

472: Fig.~\ref{Fig:lnlambda}{\it a,b}. The size of a logarithmic

473: wavelength bin in this case is $d\l_{\rm ln} = \ln{(10000/2500)} /

474: 1024 \approx 0.0014$, from eq.~\ref{Eqn:dllog}. So a shift by one bin

475: in $\ln \l$ space corresponds to a velocity shift of $d\l_{\rm ln}c

476: \approx 400$\kms. This is 1 order of magnitude less than the typical

477: width of a supernova spectral feature, Doppler-broadened by the $\sim

478: 10000$\,\kms expansion velocity of the SN ejecta.

479: 

480: %%% Figure: pre-processing the spectrum

481: \begin{figure*}

482: \plotone{f1.eps}

483: \caption{Pre-processing the spectrum for \snid. ({\it a}) Spectrum of the

484:   SN~Ia SN~2003lj at $z=0.417$

485:   \citep{Matheson/etal:2005}. ({\it b}) The result of mapping the 

486:   spectrum to $\ln \l$ coordinates, with $(\l_0, \l_1, N) =

487:   (2500$\,\AA$, 10000$\,\AA$, 1024)$. ({\it c})

488:   A 13-point spline has been divided out and the result normalized to

489:   zero mean flux. ({\it d}) A bandpass filter with

490:   $(k_1,k_2,k_3,k_4)=(1,4,25,102)$ has been applied to the

491:   spectrum. 

492: \label{Fig:lnlambda}}

493: \end{figure*}

494: 

495: The next step in preparing the spectra for correlation analysis 

496: is continuum removal \citep{Tonry/Davis:1979}. 

497: For galaxy spectra, the continuum is well defined and is easily

498: removed using a least-squares polynomial fit. In

499: supernova spectra, however, the apparent continuum is ill-defined

500: due to the domination of bound-bound transitions in the total opacity

501: \citep{Pinto/Eastman:2000}. Dividing out a 13-point cubic 

502: spline fit to the spectrum (over the entire 2500--10000\,\AA\

503: wavelength interval) is akin to removing a {\it pseudo} continuum

504: from the supernova spectrum. We then subtract 1 from the resulting

505: spectrum and apply a normalization constant for the mean flux to equal

506: zero (Fig.~\ref{Fig:lnlambda}{\it c}). This effectively discards any

507: spectral color information (including reddening uncertainties and flux

508: mis-calibrations), and the correlation only relies on the {\it

509:   relative} shape and strength of spectral features in the input and

510: template spectra. We note that continuum division is also used by

511:   \citet{Jeffery/etal:2006} for measuring the goodness-of-fit between

512:   supernova spectra. We see below (\S~\ref{Sect:zt})

513: that the loss of color information has surprisingly little impact on

514: the redshift and age determination. Continuum removal also minimizes

515: discontinuities at each end of the spectrum, which would cause

516: artificial peaks in the correlation function. Further discontinuities

517: are removed by apodizing the spectra with a cosine bell ($\sim 5\%$ at

518: either end).

519: 

520: 

521: The final step is the application of a bandpass filter. While it is

522: actually applied at a later stage, directly to the correlation

523: function, we show its effect on the input spectrum in

524: Fig.~\ref{Fig:lnlambda}{\it d}. The goal is to remove low-frequency

525: residuals left over from the pseudo-continuum removal and

526: high-frequency noise components. Formally, the Fourier

527: transform of the normalized correlation function, $C(k)$

528: (eq.~\ref{Eqn:C(k)}), is multiplied by a real bandpass function

529: (so that no phase shifts are introduced) $B(k)$, such that

530: 

531: \begin{equation}

532: B(k) = \left\{ \begin{array}{ll}

533: 0 & \textrm{for $k   \le k_1$ or $k \ge k_4$} \\

534: \onehalf \left[ 1 - \cos \left( \frac{k-k_1}{k_2-k_1}\right) \right] & \textrm{for $k_1 < k < k_2$} \\

535: 1 & \textrm{for $k_2 \le k \le k_3$} \\

536: \onehalf \left[ 1 - \cos \left( \frac{k-k_3}{k_4-k_3}\right) \right]   & \textrm{for $k_3 < k < k_4$} \\

537: %0 & \textrm{for $k   \ge k_4$.} \\

538: \end{array} \right.

539: \end{equation}

540: 

541: The exact choices for the wavenumbers $(k_1,k_2,k_3,k_4)$ depend on

542: the size of each 

543: $k$ bin and on the spectral energy distribution of a supernova

544: spectrum. Supernova spectral lines have typical widths of

545: $\sim$100--150\,\AA, due to the large expansion velocities of the

546: ejecta ($\sim 10000$\,\kms). The mean 

547: size of a logarithmic wavelength bin with $(\l_0, \l_1, N) =

548: (2500$\,\AA$, 10000$\,\AA$, 1024)$ is $\sim 7.2$\,\AA, so a typical SN

549: line will have a width $w_{\rm line} \sim$ 14--21 logarithmic wavelength

550: bins. In Fourier space, most of the information will be at wavenumbers

551: $k = N / (2\pi \times w_{\rm line}) \approx$ 8--12. Since SN 

552: spectra consist of overlapping spectral lines \citep{Baron/etal:1996},

553: a typical SN feature may have a lower width ($\la 50$\,\AA). This

554: translates to $k \sim 25$, so most information is at wavenumbers less

555: than 25 and almost everything above wavenumber $k \sim 50$ is

556: noise. Also, low wavenumbers ($k \la 5$) contain information about

557: the low-frequency residuals from continuum removal. In

558: Fig.~\ref{Fig:cfnamp} we show the amplitude of the Fourier transform

559: of typical unfiltered correlation functions as a function of

560: wavenumber. As expected, most of the correlation power is at wave

561: numbers $k \sim$ 10--20, and virtually no information is contained in

562: wavenumbers $k > 50$.

563: 

564: %%% Figure: amplitude of cfn and bandpass filter

565: \begin{figure}	

566: \plotone{f2.eps}

567: \caption{Normalized amplitude of the Fourier transform of typical 

568:   unfiltered cross-correlation functions vs

569:   wavenumber, $k$. Note how most of the power is

570:   concentrated at low wavenumbers ($k \lesssim 50$), justifying our

571:   choice for the bandpass filter ({\it dashed line}). In this example

572:   the maximum correlation amplitude is at $k=13$, corresponding to a

573:   wavelength scale of $\sim 90$\,\AA.

574: \label{Fig:cfnamp}}

575: \end{figure}

576: 

577: 

578: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

579: %%

580: %%   Section 3 - Redshift Estimate

581: %%

582: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

583: 

584: \section{Redshift Estimate}\label{Sect:z}

585: 

586: In this section we introduce the correlation height-noise ratio

587: $r$ (\S~\ref{Sect:r}) and the spectrum overlap (${\rm lap}$) parameter (\S~\ref{Sect:lap}),

588: the product of which (the $r{\rm lap}$ quality parameter) conveys quantitative information about

589: the reliability of a cross-correlation redshift output by \snid. We

590: then briefly describe the redshift estimation

591: (\S~\ref{Sect:initrevz}) and associated error (\S~\ref{Sect:zerr}).

592: 

593: \subsection{The $r$-value\label{Sect:r}}

594: 

595: \citet{Tonry/Davis:1979} introduce the correlation height-noise ratio, $r$, to quantify the

596: significance of a peak in the normalized correlation function,

597: $c(n)$. It is defined as the ratio of the height, $h$, of the peak to

598: the rms of the antisymmetric component of $c(n)$,

599: $\sigma_a$, about the correlation redshift (Fig.~\ref{Fig:rdef}):

600: 

601: \begin{equation}

602: \label{Eqn:rdef}

603: r = \frac{h}{\sqrt{2} \sigma_a}.

604: \end{equation}

605: 

606: %%% Figure: definition of the r-value

607: \begin{figure}	

608: \plotone{f3_col.eps}

609: \caption{Correlation height-noise ratio, $r$, is defined as the

610:   ratio of the height, $h$, of the highest peak in the normalized

611:   correlation function ({\it solid line}) to the rms of

612:   its antisymmetric component, $a(n)$ ({\it dashed line}), about the

613:   redshift corresponding to that peak ($z_{\rm peak}$). The width of

614:   the peak, $w$, is used to compute the redshift error ({\it

615:   see text for details}). {\it [See the electronic version of the

616:   Journal for a color version of this figure.]}

617: \label{Fig:rdef}}

618: \end{figure}

619: 

620: In order to compute $\sigma_a$, \citet{Tonry/Davis:1979} assume that

621: $c(n)$ is the sum of an auto-correlation of a template spectrum $t(n)$

622: with a shifted template spectrum $t(n-\delta)$ and of a random function

623: $a(n)$ that can distort the correlation peak:

624: 

625: \begin{equation}

626: \label{Eqn:tshift}

627: c(n) \equiv t(n) \star t(n-\delta) + a(n).

628: \end{equation}

629: 

630: \noindent

631: The first term on the right-hand side of eq.~\ref{Eqn:tshift} is

632: supposed to give a correlation peak of height $h=1$ at the exact

633: redshift (corresponding to a shift $\delta$ in logarithmic wavelength units),

634: while the second part can distort the peak. Since $t \star

635: t(n-\delta)$ is symmetric about $n=\delta$, the antisymmetric part of

636: $c(n)$ about $n=\delta$ equals the antisymmetric part of $a(n)$ about

637: $n=\delta$. We further assume that the symmetric part of $a(n)$ has

638: roughly the same amplitude as its antisymmetric part and that the

639: symmetric and antisymmetric parts of $a(n)$ are uncorrelated. In that

640: case, the rms of $a(n)$ is $\sqrt{2}$ times the rms of its

641: antisymmetric component.

642: 

643: 

644: A perfect correlation will have a peak with $h=1$ at the exact

645: redshift, and $c(n)$ will be symmetric about $n=\delta$,

646: thus  $\sigma_a=0$ and so $r \longrightarrow \infty$

647: (Fig.~\ref{Fig:cfnegs}, {\it left panel}). Conversely, $r$

648: will be small ($r \lesssim 5$) for a spurious correlation peak

649: (Fig.~\ref{Fig:cfnegs}, {\it right panel}), and large ($r \gtrsim 10$) for a

650: significant peak, since $h$ will be close to 1 and $\sigma_a$ will be

651: small (Fig.~\ref{Fig:cfnegs}, {\it middle panel}).

652: 

653: %%% Figure: examples of cfns

654: \begin{figure}	

655: \epsscale{1.2}

656: \plotone{f4_col.eps}

657: \caption{Examples of perfect ({\it left}), good ({\it middle}), and

658:   poor ({\it right}) normalized correlation functions ({\it solid

659:   line}). The antisymmetric component of the correlation

660:   function about the \snid\ redshift ($z_{\rm \snid}$) is also shown

661:   ({\it dashed line}). {\it [See the electronic version of the

662:   Journal for a color version of this figure.]}

663: \label{Fig:cfnegs}}

664: \end{figure}

665: 

666: 

667: \subsection{Spectrum Overlap\label{Sect:lap}}

668: 

669: In \snid, the correlation height-noise ratio $r$ alone does not

670: provide the estimator by which 

671: a correlation peak is deemed reliable. It is further weighted by the

672: overlap in $\ln \l$ space between the input spectrum and each of

673: the template spectra used in the correlation. In practice, the

674: template spectra are trimmed to match the wavelength range of the

675: input spectrum at the redshift corresponding to the correlation

676: peak. For an input spectrum with rest-frame wavelength range

677: $[\l_0,\l_1]$, the overlap in $\ln \l$ space, ${\rm lap}$, with each

678: template spectrum is in the range

679: 

680: \begin{equation}

681: \label{Eqn:laprange}

682: 0 \le {\rm lap} \le \ln{\left( \frac{\l_1}{\l_0} \right)}.

683: \end{equation}

684: 

685: \noindent

686: Thus for an input and template spectra both overlapping the

687: rest-frame wavelength interval 3500--6000\,\AA,

688: ${\rm lap}=\ln(6000/3500)\approx 0.54$.

689: % For a spectrum in the range 4000--9000\,\AA, the maximum value for

690: % ${\rm lap}$ is $\sim 0.8$.

691: 

692: 

693: The spectrum overlap parameter conveys important absolute information

694: about the quality of the correlation, complementary to the correlation

695: height-noise ratio $r$. Supposing a typical SN~Ia spectral feature has an FWHM of

696: $\Delta \l \approx 200$\,\AA\ at $\l \approx 5000$\,\AA, any correlation

697: with ${\rm lap}\lesssim \ln{(5400/5000)} \approx 0.08$ will be meaningless: {\it

698: any} feature will match any other at practically any redshift. Only

699: when a correlation has an associated ${\rm lap}$ that is several times

700: $\ln{(\Delta\l/\l)}$ can one rely on the redshift output by \snid. 

701: 

702: In what follows, we usually discard correlation redshifts that have an 

703: associated ${\rm lap} < {\rm lap}_{\rm min} = 0.4$ and a quality parameter $r{\rm lap} = r

704: \times {\rm lap} < r{\rm lap}_{\rm min} = 5$. In Fig.~\ref{Fig:rlapfigs} we

705:   show contour plots for both the ${\rm lap}$ and $r{\rm lap}$ parameters, central

706: to the use of \snid.

707: 

708: %%% contour plots for lap and rlap parameters

709: \begin{figure}	

710: \centering

711: \includegraphics[width=3.25in]{f5a.eps} \\

712: \vspace{.35in}

713: \includegraphics[width=3.25in]{f5b.eps}

714: \caption{{\it Top:} Contours of equal

715:   ${\rm lap}=\ln(\l_1/\l_0)$ for different rest-frame wavelength

716:   ranges $[\l_0,\l_1]$ of overlap between input and template

717:   spectra. We usually discard correlations with ${\rm lap} < 0.4$.

718:   {\it Bottom:} Contours of equal $r{\rm lap} = r \times {\rm lap}$ for a

719:   broad range of values for the correlation height-noise ratio

720:   ($r$) and spectrum overlap parameter (${\rm lap}$). We usually

721:   discard correlations with $r{\rm lap} < 5$ (and ${\rm lap} < 0.4$).

722: \label{Fig:rlapfigs}}

723: \end{figure}

724: 

725: 

726: \subsection{Initial and Revised Redshift Estimates\label{Sect:initrevz}}

727: 

728: For each template spectrum $t_i(n)$, we compute the correlation

729: function $c_i(n)=s(n)\star t_i(n)$. In general, $c_i(n)$ has many

730: peaks in redshift space (Fig.~\ref{Fig:rdef},\ref{Fig:cfnegs}). The

731: true redshift is most likely the one corresponding to the highest peak

732: in $c_i(n)$, although in poor signal-to-noise ratio cases some peaks

733: can distort or surpass the true redshift peak (Fig.~\ref{Fig:cfnegs},

734: {\it right panel}). In practice, \snid\ selects the 10 highest peaks

735: (labeled with index $j$) in $c_i(n)$ one by one and performs a fit

736: with a smooth function to determine the peak height and position,

737: $h_{ij}$ and $\delta_{ij}$, respectively. The corresponding redshift

738: is $z_{ij}=\exp(\delta_{ij}d\l_{\rm ln})-1$. The wavelength

739: regions of $s(n)$ and $t_i(n)$ that do not overlap at $z_{ij}$ are

740: trimmed, and, if the resulting spectrum overlap ${\rm lap} \ge {\rm lap}_{\rm min}$, a new

741: ``trimmed'' correlation function, $c_{ij}(n)$, is computed, and the

742: corresponding correlation height-noise ratio ($r_{ij}$), spectrum

743:   overlap (${\rm lap}_{ij}$), and redshift ($z_{ij}$) are stored.

744: 

745: 

746: Once all templates have been cross-correlated with the

747: input spectrum, \snid\ computes an initial redshift, $z_{\rm

748:   init}$, based on an $r{\rm lap}$-weighted median of all $z_{ij}$. Each

749: redshift $z_{ij}$ is replicated $W_{ij}$ times according to the following

750: weighting scheme: 

751: 

752: \begin{equation}

753: W_{ij} = \left\{ \begin{array}{ll}

754: 5 & \quad \textrm{for $\quad r{\rm lap}_{ij} >  6$} \\

755: 3 & \quad \textrm{for $\quad r{\rm lap}_{ij} >  5$} \\

756: 1 & \quad \textrm{for $\quad r{\rm lap}_{ij} >  4$} \\

757: 0 & \quad \textrm{for $\quad r{\rm lap}_{ij}\le 4$,} \\

758: \end{array} \right.

759: \end{equation}

760: 

761: \noindent

762: If all $r{\rm lap}_{ij} \le 4$, $z_{\rm init}$ is set to 0.

763: 

764: 

765: \snid\ then computes a revised redshift based on the initial estimate, $z_{\rm init}$. The

766: input and template spectra (again labeled $i$) are  

767: trimmed such that their wavelength coverage coincides at $z_{\rm

768:   init}$. If the resulting spectrum overlap ${\rm lap}_{i} \ge {\rm lap}_{\rm min}$, a

769: second trimmed correlation function is computed and the

770: correlation height-noise ratio ($r_{i}$), spectrum overlap

771:   (${\rm lap}_{i}$), and redshift ($z_{i}$) corresponding

772: to the {\em highest} correlation peak are stored. The width

773: $w_{i}$ of the correlation peak is also saved and is used to

774: compute the redshift error (see next section). The revised redshift,

775: $z_{\snid}$, is computed as the non-$r{\rm lap}$-weighted median of all

776: redshifts $z_{i}$ that satisfy $r{\rm lap}_{i} \ge r{\rm lap}_{\rm min}$ with

777: ${\rm lap}_{i} \ge {\rm lap}_{\rm min}$, with the additional requirement that

778: the individual redshifts $z_i$ do not differ significantly from the

779: initial redshift estimate: $|z_{i} - z_{\rm init}|

780:   < z_{\rm filt}$, where $z_{\rm filt} = 0.02$, typically.

781: 

782: 

783: \subsection{Redshift Error\label{Sect:zerr}}

784: 

785: One of the advantages of using the cross-correlation technique

786: for redshift determination is the ability to estimate the redshift

787: error, $\epsilon_z$. \citet{Tonry/Davis:1979} derive a formal

788: expression for $\epsilon_z$ based on the idea that spurious peaks

789: (positive and negative) in the antisymmetric component, $a(n)$,  of

790: the correlation function, $c(n)$, can distort the true correlation

791: peak. Obviously, $\epsilon_z$ is proportional to the number of peaks

792: in $a(n)$, and hence to the mean distance between peaks. Assuming

793: $c(n)$ and $a(n)$ have similar power spectra, the mean distance

794: between a peak in $c(n)$ and the nearest peak in $a(n)$ can be

795: estimated as $N/8B$ (\citealt{Tonry/Davis:1979}, their eq.~22), where

796: $N$ is the total number of bins and $B$ is the highest wavenumber at

797: the half-maximum point of the Fourier transform of $c(n)$ ($B \approx

798: 25$ here; see Fig.~\ref{Fig:cfnamp}). One can then show that

799: (\citealt{Tonry/Davis:1979}, their eq.~24)

800: 

801: \begin{equation}

802: \epsilon_z = k_z \times \frac{1}{1+r},

803: \end{equation}

804: 

805: \noindent

806: where $k_z=N/8B$ ($\approx 5$ here) and $r$ is the correlation

807: height-noise ratio defined in

808: eq.~\ref{Eqn:rdef}. With the additional assumption of sinusoidal

809: noise in $c(n)$, \citet{Kurtz/Mink:1998} find $k_z=3w/8$, where $w$ is

810: the width of the correlation peak.

811: 

812: 

813: In practice, $k_z$ is calibrated using additional redshift measurements,

814: either through a different technique (e.g., 21\,cm measurements for

815: galaxy redshifts in \citealt{Tonry/Davis:1979}) or using the same

816: cross-correlation technique on two spectra of the same object

817: \citep{Kurtz/Mink:1998}. For supernova spectra, additional redshift

818: information potentially comes from narrow emission and absorption

819: lines in the host galaxy, while duplicate spectra of the same

820: supernova (at the same age) are not common

821: (Table~\ref{Table:sndb}). We find that including the spectrum overlap 

822: parameter (${\rm lap}$) yields a more robust estimator of the redshift error

823: (see also \S~\ref{Sect:zreserr}):

824: 

825: \begin{equation}

826: \label{Eqn:epsilonz}

827: \epsilon_z = k_z \times \frac{1}{1+r{\rm lap}},

828: \end{equation}

829: 

830: \noindent

831: with $k_z \approx$ 2--4$w$.

832: 

833: 

834: % \subsection{Confidence Level}

835: % 

836: % Following \citet{Tonry/Davis:1979}, we can compute the formal

837: % confidence, $C$, that the correct peak in $c(n)$ is chosen. $C$ is

838: % related to the probability of choosing a wrong peak in $c(n)$, which

839: % in turn scales as the number of positive peaks, $B$ ($\approx 25$

840: % here; see \S~\ref{Sect:zerr}), in $c(n)$. If we restrict the

841: % correlation to a specific redshift range, for instance $-0.01 \le z

842: % \le 1.2$, only a fraction $f$ of these peaks needs to be

843: % considered. For $(\l_0,\l_1) = (2500$\,\AA$,10000$\,\AA), 

844: % the maximum redshift range one can probe is $10000/2500 - 1 =3$, and

845: % thus $f=1.21/3\approx 0.40$. Thus, the confidence $C$ can be expressed

846: % as (see \citealt{Tonry/Davis:1979}, their eqs.~30 \& 33):

847: % 

848: % \begin{equation}

849: % \label{Eqn:C}

850: % C = \left[ 1 - \mathrm{erfc}\left( \frac{r{\rm lap}}{\sqrt{2}} \right)

851: %   \right]^{fB}, 

852: % \end{equation}

853: % 

854: % \noindent

855: % where erfc $=(2/\sqrt{2 \pi}) \int_{r{\rm lap}}^\infty \exp{(r{\rm lap}^{\prime

856: % 2}/2)} dr{\rm lap}^{\prime}$ and $fB\approx 10$. Since $fB$ is essentially

857: % fixed for a given bandpass filter and type of spectrum, $C$ has a

858: % one-to-one correspondence with $r{\rm lap}$. From eq.~\ref{Eqn:C} above,

859: % confidence intervals of (1,2,3,4,5)$\sigma$ correspond to $r{\rm lap}

860: % \approx (2.1,2.8,3.6,4.5,5.4)$, respectively.

861:  

862: 

863: 

864: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

865: %%

866: %%   Section 4 - The SNID database

867: %%

868: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

869: 

870: \section{The \snid\ Database}\label{Sect:sniddb}

871: 

872: \subsection{Nomenclature and Age Distribution}

873: 

874: The current \snid\ spectral database comprises 

875: % 796 spectra of 64

876: % SN~Ia, 301 spectra of 18 SN~Ib/c, and 353 spectra of 10 SN~II 

877: 879 spectra of 65 SNe~Ia, 322 spectra of 19 SNe~Ib/c, and 353 spectra of

878: 10 SNe~II (Table~\ref{Table:sndb}). The spectra are drawn from public archives 

879: (SUSPECT\footnote{{\tt

880:     http://bruford.nhn.ou.edu/$\sim$suspect/index1.html}} and the CfA

881: Supernova Archive\footnote{{\tt

882: http://www.cfa.harvard.edu/supernova/SNarchive.html}}) 

883: and from the CfA Supernova Program \citep{Matheson/etal:2007}. The spectra are

884: chosen to have high signal-to-noise ratio (typically $\gtrsim 10$ per

885: \AA) and to span a sufficiently large optical rest frame wavelength

886: range ($\l_{\rm min} \le 4000$\,\AA; $\l_{\rm max} \ge 6500$\,\AA) to

887: include all the identifying features of SN spectra. We

888: remove telluric features in all the spectra, either using the

889: well-exposed continua of spectrophotometric standard stars for the

890: CfA data \citep{Wade/Horne:1988,Matheson/etal:2000}, or using a

891: simple linear interpolation over the strong A- and B-bands. We show

892: the full suite of spectra for the local SN~Ia spectral template

893: SN~1992A \citep{Kirshner/etal:1993}, which also includes some UV data

894: from the {\it Hubble Space Telescope} at some epochs, shifted to zero

895: redshift in Fig.~\ref{Fig:doit}.

896: 

897: 

898: While we have included all the supernovae available to us for which

899: there are a large number of epochs of spectroscopy, there are still

900: many more ($> 1000$) supernovae for which there are only one to two

901: epochs of spectroscopy that we have yet to include in the

902: database. We also include spectra of galaxies, active galactic nuclei, stars (including

903: variable stars, such as luminous blue variables), and novae. This can

904: be particularly useful when trying to weed out contaminants

905: from large surveys of high-redshift supernovae

906: (cf. \citealt{Matheson/etal:2005}).

907: 

908: %%% Figure: example of Snid template (1992A)

909: \begin{figure}	

910: \plotone{f6.eps}

911: \caption{Example of an \snid\ template, here the SN~Ia SN~1992A

912:   \citep{Kirshner/etal:1993}. We also indicate the age of each

913:   spectrum (in days from $B$-band maximum). Discontinuities in

914:   flux around $\sim3500$\,\AA\ are due to a calibration mismatch

915:   between the UV and optical components of the spectrum.

916: \label{Fig:doit}}

917: \end{figure}

918: 

919: We show the age distribution of \snid\ supernova templates for the

920: main supernova types (Ia, Ib, Ic, II) in Fig.~\ref{Fig:thist}. 

921: For each type, we show the age distribution of ``normal''

922: representatives of that type, as well as spectra that show deviations

923: from the latter (in the ``other'' category). We note that this

924: division is somewhat qualitative and relies on the identification by

925: eye of certain characteristic spectroscopic features in the spectra

926: \citep{Filippenko:1997}. We are currently working on a statistical

927: scheme to separate our template spectra in these various categories

928: (see also Fig.~\ref{Fig:meanspec}). The nomenclature for the

929: different supernova types and their associated subtypes is given in

930: Table~\ref{Table:stype}. From Fig.~\ref{Fig:thist}, it is clear that

931: the age distribution of the \snid\ templates is not uniform, and

932: even bi-modal for SNe~Ia. This potentially introduces age

933: ``attractors'' that could in principle bias the age and

934: redshift determination (although see \S~\ref{Sect:zt}). The 

935: fact that there are more SN~Ia templates than SNe~Ib, Ic, and II {\it

936:   combined} also leads to a type ``attractor,'' with the risk for

937: low-S/N spectra to be preferentially classified as SN~Ia, regardless

938: of their type (see \S~\ref{Sect:type}). 

939: 

940: %%% Figure: phase distribution of Snid templates

941: \begin{figure}

942: \epsscale{1.1}

943: \plotone{f7_col.eps}

944: \caption{Age distributions of \snid\ templates for supernovae of

945:   different types. The number of supernovae corresponding to a given

946:   type is indicated in square brackets. Note the larger

947:   ordinate range for SNe~Ia. For SNe~II, the

948:   age is given in days from the estimated date of explosion as

949:   opposed to days from maximum in a specific band. The transitional

950:   SNe~IIb are included in both Type Ib and Type II

951:   histograms. {\it [See the electronic version of the Journal for a

952:   color version of this figure.]}

953: \label{Fig:thist}}

954: \end{figure}

955: 

956: %%% Table: type/subtype nomenclature

957: \begin{deluxetable}{ l | l l l l }

958: \tabletypesize{\scriptsize}

959: \tablenum{2}

960: \tablewidth{0pt}

961: \tablecaption{Supernova types and subtypes\label{Table:stype}}

962: \tablehead{

963: \colhead{Type} &

964: \colhead{Ia} & 

965: \colhead{Ib} & 

966: \colhead{Ic} &

967: \colhead{II}

968: }

969: \startdata

970: ``normal'' & Ia-norm & Ib-norm & Ic-norm  & II-norm (IIP) \\ \hline

971:            & Ia-pec  & Ib-pec  & Ic-pec   & II-pec        \\

972: ``other''  & Ia-91T  & IIb     & Ic-broad & IIL           \\

973:            & Ia-91bg &         &          & IIn           \\

974:            &         &         &          & IIb           

975: \enddata

976: \tablecomments{

977: ``norm'' and ``pec'' refer to ``normal'' and ``peculiar'' subtypes

978:   of the corresponding type; see Table~\ref{Table:sndb} for specific

979:   examples. ``Ic-broad'' is used to identify broad-lined SNe~Ic

980: (``hypernovae''), some of which are associated with Gamma-Ray

981: Bursts. The transitional Type IIb supernovae are included in both

982: Type Ib and Type II categories.

983: }

984: \end{deluxetable}

985: 

986: 

987: The execution time of \snid\ scales linearly with the number of

988: templates\footnote{Execution time $t_{\rm exec} \approx 6s ({\rm cpu}/{\rm 2.86

989: GHz})(N_{\rm temp}/1000)$, where $N_{\rm temp}$ is the number of 

990: templates.} and is remarkably low compared with $\chi^2$-based

991: methods (see \S~\ref{Sect:comparison})--- although see

992:   \citet{Rybicki/Press:1995} for fast statistical methods that can

993:   compete with the cross-correlation technique. It is trivial to

994: include large spectroscopic data sets--- such as those from

995: the CfA Supernova Group (for example, 431 spectra of 32 SNe~Ia,

996: included in the present \snid\ database, will soon be published by

997: \citealt{Matheson/etal:2007}).

998: 

999: 

1000: 

1001: \subsection{Intrinsic Spectral Variance}

1002: 

1003: In Fig.~\ref{Fig:meanspec} we show the standard and maximum deviation

1004: from the mean spectrum of all ``Ia-norm'' templates at $-10$, +0, +10,

1005: and +20 days from maximum light. One clearly sees the rapid variation

1006: of SN spectra around maximum light, but also the change in intrinsic

1007: scatter with age. For instance, the intrinsic spread in the strength

1008: and position of the defining \sitwo\ \l6355 feature (which causes the

1009: deep blueshifted absorption around $\sim 6100$\,\AA) decreases between

1010: $-10$ to $+10$ days from maximum light. At +20 days, the scatter is

1011: large in that wavelength region due to increasing contribution from

1012: other ions (mainly \fetwo).

1013: 

1014: %%% Figure: mean/variance spectra and Nugent templates

1015: \begin{figure*}

1016: \plottwo{f8a_col.eps}{f8b_col.eps}

1017: \caption{

1018: {\it Left:} Standard ({\it light gray}) and maximum ({\it dark gray})

1019: deviation from the mean spectrum for all Ia-norm templates, at

1020: four different ages. We overplot the corresponding Nugent template

1021: at each age ({\it dashed line}). All spectra have been pre-processed

1022: in the same way as any \snid\ template (\S~\ref{Sect:preproc}).

1023: {\it Right:} Fractional difference from the mean Ia-norm

1024: spectrum. We also show the ratio of the mean spectrum to the

1025: corresponding Nugent template ({\it dashed line}). {\it [See the

1026:     electronic version of the Journal for a color version of this

1027:     figure.]}

1028: \label{Fig:meanspec}}

1029: \end{figure*}

1030: 

1031: The residual variation about the mean spectrum

1032: (Fig.~\ref{Fig:meanspec}, {\it right panel}) shows that normal SNe~Ia

1033: are typically within 10\%--20\%\ from the mean spectrum, although

1034: deviations greater than 40\%\ are seen at certain wavelength intervals

1035: (again depending on the age). The fact that all Ia-norm

1036: template spectra are within two standard deviations from a mean

1037: spectrum suggests a possible statistical classification scheme to

1038: differentiate normal SNe~Ia from the other Ia subtypes. With more

1039: data, it is in principle possible to do this more reliably for SNe~Ia,

1040: as well as other supernova types. 

1041: 

1042: 

1043: The intrinsic variation of the Ia-norm templates points to the

1044: inadequacy of describing a given SN subtype with a single

1045: representative template, unless the latter includes this

1046: variance explicitly. Past attempts to create grids of such template

1047: spectra, such as those presented by

1048: \citet{Nugent/Kim/Perlmutter:2002}, do not account for the variability

1049: within a given SN type at a given age. We show the corresponding

1050: Nugent template (ver. 1.2) at each age in Fig.~\ref{Fig:meanspec} ({\it

1051:   dashed line}). While most of the Nugent template is included within

1052: the standard deviation from the mean spectrum in our database, there

1053: are also significant deviations. We do note, however, that the

1054: comparison is somewhat misleading since

1055: \citet{Nugent/Kim/Perlmutter:2002} had less data available to them for

1056: the elaboration of these templates. Nevertheless, we have tested their

1057: use in the \snid\ spectral database, but have found them to lead to

1058: systematic errors in both the redshift and age determination.

1059: 

1060: 

1061: 

1062: 

1063: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1064: %%

1065: %%   Section 5 - Accuracy of redshift and age determination

1066: %%

1067: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1068: 

1069: \section{Accuracy of Redshift and Age Determination}\label{Sect:zt}

1070: 

1071: We use a simple simulation to test the accuracy of \snid\ in

1072: determining the redshift and age of a supernova spectrum. Here we

1073: focus on normal SNe~Ia since they are the most

1074: represented in our spectral database, although the conclusions of

1075: this section are qualitatively valid for all other supernova

1076: types. Even though normal SNe~Ia form a homogeneous class, the

1077: spectra reveal intrinsic variations at any given age that affect

1078: directly the redshift and age determination. The redshift precision

1079: depends primarily on the typical width of a spectral feature

1080: (decreasing from broad-lined SNe~Ic to SNe~IIn), which

1081: affects the width of the correlation peak (see

1082: Fig.~\ref{Fig:rdef}). The redshift accuracy depends primarily on the

1083: intrinsic variation of line positions at a given age. The age

1084: determination strongly correlates with the redshift determination

1085: (\S~\ref{Sect:ztcovar}), and depends on how quickly the SN spectra

1086: evolve at a given age.

1087: 

1088: 

1089: \subsection{Presentation of the Simulation\label{Sect:simulation}}

1090: 

1091: In this simulation, each Ia-norm spectrum in the \snid\ database

1092: (cf. Table~\ref{Table:sndb}) is correlated with all other Ia-norm

1093: spectra, except for those corresponding to the input supernova (to

1094: ensure unbiased results). We require all spectra used in the

1095: simulation to include the rest-frame wavelength interval

1096: 3700--6500\,\AA and to have an age (in days from $B$-band

1097: maximum, hereafter $t_B$) $-10 \le t_B \le +20$.

1098: 

1099: 

1100: We show the simulation parameters in Table~\ref{Table:simparams}. The

1101: input spectrum is first redshifted to $z$ by simply multiplying

1102: the wavelength axis by $(1+z)$. We then ``contaminate'' the input

1103: supernova spectrum with galaxy light (up to 50\% of the total flux),

1104: using the elliptical and Sc galaxy templates of

1105: \citet{Kinney/etal:1996}, and add noise (both random Poisson noise and

1106: sky background) to reproduce the range of typical signal-to-noise

1107: ratio of SN spectra at the simulation redshifts, when observed

1108: with 6.5--10\,m-class telescopes (e.g., VLT, Keck, Gemini, Magellan)

1109: used in cosmological SN~Ia surveys. Note that we do not scale

1110: the input spectral flux to match a given simulation redshift, as

1111: \snid\ normalizes the input and template spectra in a

1112: similar fashion (Fig.~\ref{Fig:lnlambda}).

1113: 

1114: %%% Table: Simulation parameters

1115: \begin{deluxetable*}{ll}

1116: \tabletypesize{\scriptsize}

1117: \tablenum{3}

1118: \tablewidth{0pt}

1119: \tablecaption{Simulation parameters\label{Table:simparams}}

1120: \tablehead{

1121: \colhead{Parameter} & 

1122: \colhead{Range}

1123: }

1124: \startdata

1125: Redshift, $z$ & $0.1 \le z \le 0.7$ \\

1126: Galaxy contamination fraction, $f_{\rm gal}$ & $0 \le f_{\rm gal} \le 0.50$ \\

1127: Signal-to-noise ratio, S/N (per 2\,\AA) & $1 \le {\rm S/N} \le 15$ \\

1128: Age (days from $B$-band maximum), $t_B$ & $-10 \le t_B \le +20$ \\

1129: Minimum rest frame wavelength coverage, $\l_{\rm rest}$ (\AA) & $3700 \le \l_{\rm test} \le 6500$ \\

1130: Observed wavelength range, $\l_{\rm obs}$ (\AA) & $4000 \le \l_{\rm obs} \le 9000$ \\

1131: \enddata

1132: \end{deluxetable*} 

1133: 

1134: 

1135: We restrict the {\it observed} wavelength range over which \snid\

1136: computes the correlation to $4000 \le \l_{\rm obs} [$\AA$]\le 9000$,

1137: to mimic the coverage of the FORS1 optical spectrograph mounted on the

1138: VLT. We have not studied the impact of a change in this wavelength

1139: range on the redshift or age determination. Furthermore, we force

1140: \snid\ to only consider correlation redshifts in the interval

1141: $[0,1]$. For each correlation, we record the template name, type,

1142: subtype, and age; the correlation redshift and its associated 

1143:   correlation height-noise ratio ($r$) and spectrum overlap parameter

1144:   (${\rm lap}$); and the width $w$ of the correlation peak (to estimate the 

1145: redshift error).

1146: 

1147: 

1148: To study the effects of constraints on redshift and age, we run

1149: \snid\ three times on the input spectrum: once with no constraints; a

1150: second time with a flat constraint on redshift ($\pm 0.01$), and a third

1151: time with a flat constraint on age ($\pm 3$ days). We note that the

1152: distribution of redshift residuals is remarkably Gaussian

1153: (\S~\ref{Sect:zreserr}), and we are currently implementing Gaussian

1154: priors in \snid. A total of 4 billion correlations were computed

1155: with \snid\ for this simulation, in just under 70 CPU hr.

1156: 

1157: 

1158: \subsection{Redshift Residuals and Redshift Error\label{Sect:zreserr}}

1159: 

1160: We show the distribution of redshift residuals, $\Delta z$, versus

1161: the $r{\rm lap}$ quality parameter in the top-right panel of

1162: Fig.~\ref{Fig:zsigrlap}, for input parameters  $0.3 \le z \le

1163: 0.5$, $-5 \le t_B \le +15$, and $2 \le$ S/N (per 2\,\AA)$\le

1164: 10$. The residuals are shown as a two-dimensional (2D) histogram, with 

1165: a linear gray-scale scheme reflecting the number of points in a given

1166: $(\Delta z,r{\rm lap})$ bin. We only show correlations for which the overlap

1167: between input and template spectra ${\rm lap} \ge 0.4$. For good

1168: correlations ($r{\rm lap} \gtrsim 5$), the distribution of redshift

1169: residuals is a Gaussian centered at $\Delta z = 0$. In the bottom right

1170: panel, we show the standard deviation of redshift residuals,

1171: $\sigma_z$, in $r{\rm lap}$ bins of size unity. For $r{\rm lap} \gtrsim 5$, we

1172: have a typical error in redshift of order $\sigma_z \lesssim 0.01$.

1173: 

1174: %%% Figure: z residuals vs. r,rlap (2d)

1175: \begin{figure}

1176: \epsscale{1.2}

1177: \plotone{f9.eps}

1178: \caption{{\it Top:} 2D histograms of redshift

1179:   residuals vs. the correlation height-noise ratio $r$

1180:   ({\it left panel}) and the $r{\rm lap}$ quality parameter (with ${\rm lap} \ge

1181:   0.4$; {\it right panel}), with the following parameters: $0.3 \le z \le

1182:   0.5$, $-5 \le t_B \le +15$, $2 \le$ S/N (per

1183:   2\,\AA)$\le 10$. The linear gray scale reflects the number of points

1184:   in a given 2D bin (the more points the darker). 

1185:   {\it Bottom:} Standard deviation, $\sigma_z$, of

1186:   redshift residuals in $r,r{\rm lap}$ bins of size unity. For the

1187:   $\sigma_z(r)$ curve ({\it filled circles, bottom left}), we show the

1188:   effect of additionally requiring that ${\rm lap} \ge 0.4$ ({\it open

1189:   circles}).

1190: \label{Fig:zsigrlap}}

1191: \end{figure}

1192: 

1193: For poor correlations ($r{\rm lap} \lesssim 3$) there is a

1194: concentration of points around $\Delta z \approx -0.01$. 

1195: This is an artifact of the {\it pseudo}-continuum removal, which

1196: enhances the contrast between emission peaks and

1197: absorption troughs in the input and template spectra and biases

1198: poor correlations to later ages. In this simulation, many input

1199: spectra at maximum are attracted to $\sim +10$ days,

1200: % This is due

1201: % to the bimodality in the SN~Ia template age distribution 

1202: % (Fig.~\ref{Fig:thist}, {\it top-left}): many input spectra at

1203: % post-maximum ages ($t \lesssim +10$ days) correlate with template

1204: % spectra at higher

1205: % ages ($\gtrsim +10$ days), 

1206: where the position of SN spectral

1207: features has shifted redward in wavelength due to the expansion of the

1208: supernova envelope (Fig.~\ref{Fig:vabst}). The template needs to be

1209: shifted less in $\ln \l$ space to match the redshift of the input

1210: spectrum, which leads to an under-estimation of the redshift by $\sim

1211: 0.01$. This corresponds to a combination of the typical velocity shift

1212: in SN~Ia absorption features from maximum to $\sim 10$ days past

1213: maximum, and the spread of these velocities at a given age ($\sim

1214: 3000$\kms; see \citealt{Benetti/etal:2005,Blondin/etal:2006a}).

1215: We note that this artifact has no impact on correlations with $r{\rm lap}

1216: > 3$.

1217: 

1218: %%% Figure: Evolution of SiII

1219: \begin{figure}

1220: \epsscale{1.1}

1221: \plotone{f10_col.eps}

1222: \caption{Evolution of the blueshifted \sitwo\ \l6355 absorption

1223:   profile in the SN~Ia SN~1994D

1224:   \citep{Hoeflich:1995,Patat/etal:1996} between $-11$ and +7 days

1225:   from $B$-band maximum. The dotted line shows the velocity location

1226:   of the locus of maximum absorption, $v_{\rm abs}$. We highlight the

1227:   \sitwo\ profile at maximum light ({\it dashed line}). Over the course

1228:   of 18 days, the locus of maximum absorption shifts redward in

1229:   wavelength by $\sim 80$\,\AA, corresponding to $\sim 4000$\,\kms\ in

1230:   velocity. Note the more rapid evolution of $v_{\rm abs}$ before

1231:   maximum light. {\it [See the electronic version of the Journal for a

1232:   color version of this figure.]}

1233: \label{Fig:vabst}}

1234: \end{figure}

1235: 

1236: In the left panels of Fig.~\ref{Fig:zsigrlap}, we show the same

1237: distribution of $\Delta z$, this time only as a function of the

1238:   correlation height-noise ratio, $r$ (note

1239: the change in the abscissa range). To first order, the 

1240: 2D histogram of redshift residuals looks remarkably

1241: similar to that as a function of $r{\rm lap}$ (Fig.~\ref{Fig:zsigrlap}, {\it

1242:   right panels}), with again a concentration of

1243: points around $\Delta z \approx -0.01$ for low values of $r$. However,

1244: the variation of $\sigma_z$ with $r$ ({\it bottom left panel, filled

1245:   circles}) gives a different picture: the lack of constraint on ${\rm lap}$

1246: causes in some cases a mis-estimate of the redshift, at all $r$,

1247: thereby greatly biasing $\sigma_z$ to higher values ($\sigma_z > 0.03$, for

1248: all $r$). Requiring that ${\rm lap} \ge 0.4$ leads to a significant improvement

1249: ({\it open circles}), with $\sigma_z \lesssim 0.01$

1250: for $r \gtrsim 10$. It is therefore imperative to consider the overlap

1251: between the input and template spectra to yield accurate supernova

1252: redshifts with the cross-correlation technique.

1253: 

1254: 

1255: The formal redshift error, $\epsilon_z$, is proportional to

1256: $w/(1+r{\rm lap})$ (eq.~\ref{Eqn:epsilonz}), $w$ being the width of the

1257: correlation peak (Fig.~\ref{Fig:rdef}). We illustrate the

1258: determination of the constant of 

1259: proportionality, $k_z$, in Fig.~\ref{Fig:zsigrlapw}, where we show the

1260: same 2D histograms of redshift residuals $\Delta z$, this

1261: time as a function of $(1+r)/w$ ({\it left panels}) and $(1+r{\rm lap})/w$ ({\it

1262:   right panels}). A best fit to the $\sigma_z$ curves in the bottom panels

1263: yields a value for $k_z$: 5.3 for $(1+r)/w$, and 3.1 for

1264: $(1+r{\rm lap})/w$. Only correlations with ${\rm lap} \ge 0.4$ are shown. As in

1265: Fig.~\ref{Fig:zsigrlap}, the product of the $r$-value and the

1266: overlap yields a more robust error estimator than the

1267: $r$-value alone. In what follows we study variations of redshift and

1268: age determinations using \snid\ only as a function of the $r{\rm lap}$

1269: quality parameter, with ${\rm lap} \ge 0.4$.

1270: 

1271: %%% Figure: z residuals vs. 1+r,rlap/w (2d)

1272: \begin{figure}

1273: \epsscale{1.2}

1274: \plotone{f11.eps}

1275: \caption{Same as Fig.~\ref{Fig:zsigrlap}, except the abscissae now

1276:   correspond to $(1+r,r{\rm lap})/w$, where $w$ is the width of the

1277:   correlation peak. A fit to the binned $\sigma_z$ distributions ({\it

1278:   bottom}) yields the value for $k$ used in estimating the error

1279:   (eq.~\ref{Eqn:epsilonz}). In both panels, only

1280:   correlations with ${\rm lap} \ge 0.4$ are shown.

1281: \label{Fig:zsigrlapw}}

1282: \end{figure}

1283: 

1284: In principle, $k_z$ needs to be evaluated for every template spectrum in

1285: the database, through either internal or external comparisons (as done

1286: for galaxy spectral templates in

1287: \citealt{Tonry/Davis:1979,Kurtz/Mink:1998}). While this is impractical

1288: for supernova spectra--- there are few duplicate spectra of the same

1289: supernova at a given age (Table~\ref{Table:sndb}), we have

1290: computed $k_z$ using subsets of templates used in our simulation, as

1291: well as for other supernova types, and have found that $k_z$ is

1292: typically in the range $2 \lesssim k_z \lesssim 4$, with $k_z\approx 3$

1293: being the median value.

1294: 

1295: 

1296: The above holds for a {\it single} spectral template; to use \snid\ to

1297: its full capacity, we need to combine redshifts for {\it all} templates

1298: for which the $r{\rm lap}$ quality parameter is greater than a certain cutoff (generally, $r{\rm lap}

1299: \ge r{\rm lap}_{\rm min} = 5$). In \S~\ref{Sect:initrevz}, we favored the

1300: non-$r{\rm lap}$-weighted median of all correlation redshifts with $r{\rm lap} \ge

1301: r{\rm lap}_{\rm min}$ as being ``the'' \snid\ redshift, but did not justify

1302: this. In Fig.~\ref{Fig:deltazhist} ({\it top panels}), we show

1303: distributions of \snid\ redshift residuals, when the \snid\ redshift

1304: is taken to be the redshift corresponding to the highest $r{\rm lap} \ge 5$

1305: value (``best;'' {\it left panel}); the median ({\it middle panel}) or

1306: $r{\rm lap}$-weighted mean ({\it right panel}) of all redshifts with $r{\rm lap} \ge

1307: 5$. Both the median and mean distributions are consistent with a

1308: Gaussian distribution, with the median redshift providing a slightly

1309: better match. This is expected since the use of the median redshift

1310: guards us from systematic errors produced by spurious or ill-defined

1311: correlation peaks for some templates. The distribution of ``best''

1312: redshift residuals is broader and non-uniform. We therefore consider

1313: the median redshift to provide the best estimate.

1314: 

1315: %%% Figure: z{best,median,mean} residuals distribution and error estimators

1316: \begin{figure}

1317: \epsscale{1.2}

1318: \plotone{f12.eps}

1319: \caption{{\it Top:} Normalized distributions of redshift

1320:   residuals, when the \snid\ redshift is assumed to be the redshift

1321:   of the best-match template ({\it left}), the median redshift ({\it

1322:   middle}), and the $r{\rm lap}$-weighted mean redshift ({\it right}). The

1323:   distribution of median redshifts is the most consistent with a

1324:   normal distribution. {\it Bottom:} Normalized distributions of

1325:   the ratio of the absolute redshift residual (corresponding to the

1326:   different redshift estimators in the top panel) to the redshift

1327:   error, $\epsilon_z$, estimated in different ways (see text for

1328:   details). 

1329: \label{Fig:deltazhist}}

1330: \end{figure}

1331: 

1332: In the bottom panels of Fig.~\ref{Fig:deltazhist} we show the normalized

1333: distributions of the ratio of the absolute redshift residual

1334: (corresponding to the different redshift estimators in the top

1335: panel) to the redshift error, $\epsilon_z$, estimated in different

1336: ways. A ratio equal to or above unity indicates that the actual redshift is

1337: consistent with the \snid\ redshift within the estimated error, while a ratio

1338: below unity indicates that the error is under-estimated. For a good error

1339: estimator, we expect those distributions to peak at a ratio near

1340: unity, with a long tail to higher ratios and a sharp drop below

1341:  unity. Such is the case for the formal redshift error

1342: (eq.~\ref{Eqn:epsilonz}) associated with the ``best'' redshift ({\it

1343:   left panel}). It is not obvious which error to associate with the median

1344: redshift. We found that the standard deviation of all correlation

1345: redshifts with $r{\rm lap} \ge 5$ provided a satisfactory estimate of the

1346: error ({\it middle panel}). This same estimator was used by

1347: \citet{Matheson/etal:2005} for high-$z$ SN~Ia spectra from the ESSENCE

1348: survey. The error in the $r{\rm lap}$-weighted mean ({\it right panel}), on the

1349: other hand, systematically underestimates the true redshift error by

1350: a factor of $\sim 3$.

1351: 

1352: 

1353: \subsection{Age Residuals}

1354: 

1355: Unlike redshift, the supernova age is not (and cannot be)

1356: a free parameter in \snid, as it is a discrete variable tied in with a

1357: specific spectral template. Nevertheless, since the 

1358: cross-correlation technique relies solely on the relative strengths

1359: and position of broad spectroscopic features, which themselves are a

1360: strong function of the supernova age

1361: (Figs.~\ref{Fig:doit},\ref{Fig:meanspec}, \&~\ref{Fig:vabst}), we

1362: expect a strong correlation between the $r{\rm lap}$ quality parameter and the age

1363: residual, $\Delta t$, between input and template spectra.

1364: 

1365: 

1366: We show the distribution of age residuals versus $r{\rm lap}$ in

1367: Fig.~\ref{Fig:tsignf} ({\it top panel}), where the gray scale has the

1368: same meaning as in the previous 2D histograms. For $r{\rm lap}

1369: \gtrsim 6$, the distribution of age residuals is a Gaussian centered

1370: at $\Delta t = 0$. In the bottom panel, we show the standard deviation

1371: of age residuals, $\sigma_t$, in $r{\rm lap}$ bins of size unity. For

1372: $r{\rm lap} \gtrsim 6$, we have a typical error in age of order

1373: $\sigma_t \lesssim 5$ days.

1374: 

1375: %%% Figure: t residuals vs. rlap (2d)

1376: \begin{figure}

1377: \epsscale{1.1}

1378: \plotone{f13.eps}

1379: \caption{{\it Top:} 2D histogram of age

1380:   residuals vs. the $r{\rm lap}$ quality parameter (with ${\rm lap} \ge 0.4$), with the same

1381:   parameters as in Fig.~\ref{Fig:zsigrlap}. The gray scale reflects the

1382:   number of points in a given 2D bin (the more points the darker). 

1383:   {\it Bottom:} Standard deviation, $\sigma_t$, of age

1384:   residuals in $r{\rm lap}$ bins of size unity. For $r{\rm lap} \gtrsim 6$,

1385:   $\sigma_t \lesssim 5$ days. 

1386: \label{Fig:tsignf}}

1387: \end{figure}

1388: 

1389: 

1390: The most striking feature in the gray scale of Fig.~\ref{Fig:tsignf} is

1391: the near absence of points around $\Delta t=0$ for low values of

1392: $r{\rm lap}$. For poor correlations, the age is systematically

1393: mis-estimated, with a tendency to overestimate the age by $\sim10$

1394: days. Again, this is an artifact of the {\it pseudo-}continuum

1395: removal, which causes many maximum-light spectra to correlate with

1396: $\sim +10$-day templates (see \S~\ref{Sect:zreserr}).

1397: % a direct consequence of the bimodality in the

1398: % SN~Ia template age distribution (Fig.~\ref{Fig:thist}), where the

1399: % two peaks are separated by roughly ten days. 

1400: It is also due to the

1401: nature of the supernova evolution, as the spectra evolve more rapidly

1402: around maximum light than they do around 10 days past maximum

1403: (Fig.~\ref{Fig:meanspec}), increasing the likelihood of correlations

1404: with templates at these ages.

1405: 

1406: 

1407: There is no formal estimator for the age error. We have examined the

1408: distribution of age residuals above a certain $r{\rm lap}$ cutoff 

1409: (cf. Fig.~\ref{Fig:deltazhist} for redshift) and find the median age of

1410: all templates with $r{\rm lap} \ge 5$ to be a good estimate of the spectral

1411: age. However, the standard deviation of all template ages with $r{\rm lap}

1412: \ge 5$ tends to systematically overestimate the age error by $\sim

1413: 20\%$.

1414: 

1415: \subsection{Covariance Between Redshift and Age\label{Sect:ztcovar}}

1416: 

1417: The determination of redshift and age is intrinsically connected,

1418: and in principle one should marginalize over one 

1419: parameter to infer the other. Marginalizing out the redshift (a

1420: continuous variable) to infer the age is straightforward, but the

1421: reverse is more complex, as it involves marginalization over sparsely

1422: sampled variables. The techniques to do this abound in the Bayesian

1423: literature, but we have yet to implement them in \snid. Nevertheless,

1424: we illustrate the covariance between redshift and age using the

1425: 2D histogram of age versus redshift residuals, for

1426: correlations satisfying $r{\rm lap} \ge 5$ (Fig.~\ref{Fig:ztres_covar}). As

1427: expected (see \S~\ref{Sect:zreserr}), over(under)-estimating the

1428: age leads to under(over)-estimating the redshift, since the loci of

1429: maximum absorption shift to the red with age (Fig~\ref{Fig:vabst}).

1430: 

1431: %%% Figure: covariance between redshift/phase residuals (2d)

1432: \begin{figure}

1433: \epsscale{1.1}

1434: \plotone{f14.eps}

1435: \caption{Age residuals vs. redshift residuals, illustrating the

1436:   covariance between the two quantities. We show the $1\sigma$ ({\it

1437:   solid line}) and $2\sigma$ ({\it dashed line}) contours. The

1438:   parameters are the same as those used in Fig.~\ref{Fig:zsigrlap},

1439:   with the requirement that $r{\rm lap} \ge 5$. 

1440: \label{Fig:ztres_covar}}

1441: \end{figure}

1442: 

1443: The anti-correlation between redshift and age residuals shown in

1444: Fig.~\ref{Fig:ztres_covar} suggests that constraints on one parameter

1445: should improve the accuracy of the

1446: other. Fig,~\ref{Fig:deltazthist_prior} shows the effect on the

1447: distributions of redshift ({\it left panel}) and age ({\it right panel})

1448: residuals (for $r{\rm lap} \ge 5$; {\it open histograms}) of adding a flat

1449: $\pm 3$-day constraint on age and a flat $\pm 0.01$ constraint on redshift,

1450: respectively ({\it hatched histograms}). A constraint on the

1451: age leads to a $\sim 30\%$ narrower distribution of redshift

1452: residuals (from $\sigma_z = 0.006$ to $\sigma_z = 0.004$) and a

1453: constraint on redshift improves the age determination by $\sim 15\%$

1454: ($\sigma_t = 3.4$ days to $\sigma_t = 2.9$ days). In practice, the

1455: constraint on redshift generally comes from a spectrum of the SN host

1456: galaxy and one can impose a constraint on age using a well-sampled

1457: light curve of the supernova (one for which the date of maximum light

1458: is easily determined).

1459: 

1460: %%% Figure: z,t residuals distribution with t,z priors

1461: \begin{figure}

1462: \epsscale{1.2}

1463: \plotone{f15_col.eps}

1464: \caption{Effect of age and redshift constraints on redshift ({\it

1465:   left}) and age ({\it right panel}) residuals, respectively, with the same

1466:   parameters as in Fig.~\ref{Fig:zsigrlap}. Here $z_{\rm \snid}$

1467:   ($t_{\rm \snid}$) corresponds to the median of all redshifts

1468:   (ages) with $r{\rm lap} \ge 5$. The open and hatched histograms

1469:   correspond to residuals with no constraint and a constraint on $(z,t)$,

1470:   respectively. {\it [See the electronic version of the Journal for a

1471:   color version of this figure.]}

1472: \label{Fig:deltazthist_prior}}

1473: \end{figure}

1474: 

1475: The age distribution of SN spectral templates in the database

1476: affects the accuracy of both cross-correlation age and redshifts.

1477: % It is instructive to see which age distribution

1478: % would be sufficient for accurate redshift and age determinations. 

1479: In Fig.~\ref{Fig:opthist} we show the result of a Monte Carlo simulation

1480: where we compute the number of SN~Ia spectra in bins of 3 days

1481: that would be sufficient for accurate redshift ({\it left panel}) and

1482: age ({\it right panel}) determinations with \snid.

1483: % in

1484: % age bins of 3 days in the interval $-10 \le t_B \le

1485: % +20$, for most accurate redshift ({\it left panel}) and age ({\it right panel})

1486: % determinations with \snid. 

1487: The solid histogram is the actual age distribution of normal SN~Ia

1488: templates in the interval $-10 \le t_B \le +20$ and the dotted

1489: histogram is the Monte Carlo distribution.

1490: %computed through Monte Carlo. 

1491: We compute 1000 Monte Carlo

1492: realizations for each unity increment in the number of spectra in a

1493: given 3-day age bin. The number of spectra was chosen

1494: such that adding more spectra would not change the mean redshift and

1495: age residuals (for $r{\rm lap} \ge 5$) by more than 0.0001 and 0.1 days,

1496: respectively. For this Monte Carlo distribution, at least eight

1497: correlations with $r{\rm lap} \ge 5$ are needed for the median redshift and

1498: associated error (Fig.~\ref{Fig:deltazhist}) to provide an accurate

1499: estimate.

1500: 

1501: %%% Figure: optimal phase distributions for z,t determination for normal SNe Ia

1502: \begin{figure}

1503: \epsscale{1.2}

1504: \plotone{f16_col.eps}

1505: \caption{Actual ({\it solid line}) and Monte Carlo ({\it dotted

1506:   line}) age distributions of normal SN~Ia templates for redshift

1507:   ({\it left panel}) and age ({\it right panel}) determination. The Monte

1508:   Carlo distribution was computed such that adding more spectra would

1509:   not change the mean redshift and age residuals (for $r{\rm lap} \ge 5$)

1510:   by more than 0.0001 and 0.1 days, respectively.

1511:   {\it [See the electronic version of the Journal for a color version

1512:   of this figure.]}

1513: \label{Fig:opthist}}

1514: \end{figure}

1515: 

1516: The Monte Carlo age distribution for redshift determination

1517: (Fig.~\ref{Fig:opthist}, {\it left panel}) has an initial peak around $-10$

1518: days and a bell-shaped envelope roughly centered around maximum ($0$

1519: days)--- akin in fact to a supernova light curve. This is due to the

1520: faster evolution of supernova spectra around maximum light than around

1521: 1--2 weeks past maximum (Figs.~\ref{Fig:meanspec}

1522: \&~\ref{Fig:vabst}). In other words, \snid\ can accurately determine

1523: the redshift of an input spectrum at $+10$ days using a template at

1524: $+15$ days, since the wavelength (velocity) positions and relative

1525: strengths of spectral features change little over this age

1526: interval, but will be less accurate when an input spectrum at maximum

1527: light is correlated with template spectra at +5 days, since the

1528: evolution of the spectra is more significant then. The initial peak

1529: around $-10$ days is due to the rapid decrease in spectral line

1530: blueshifts from $\lesssim -10$ days to $\sim -5$ days

1531: (Fig.~\ref{Fig:vabst};

1532: \citealt{Benetti/etal:2005,Blondin/etal:2006a}), rather than to a 

1533: change in the relative strengths of spectral features. 

1534: % This minimum histogram shows we 

1535: We are currently lacking normal SN~Ia spectral templates around +5

1536: days past maximum light.

1537: %  and are thus still

1538: % subject to systematic redshift errors due to the bimodality of the

1539: % actual age distribution. 

1540: This gap will be filled shortly with a new

1541: set of spectra from the CfA Supernova Program (almost 50 SNe~Ia with

1542: more than 10 epochs of spectroscopy since 2000).

1543: 

1544: 

1545: The Monte Carlo age distribution for age determination

1546: (Fig.~\ref{Fig:opthist}, {\it right panel}) is altogether different, but the

1547: same reasons apply: due to the rapid evolution of SN spectra around

1548: maximum light, it is easier to accurately determine the age then

1549: than at 1--2 weeks past maximum, where the spectra evolve on

1550: longer timescales. Hence, more spectra are needed at later ages than

1551: around maximum light. 

1552: % The minimum histogram shows that the 

1553: The current number of normal SN~Ia templates in our database is

1554: sufficient for accurate age determinations out to $t \lesssim +15$

1555: days, but we need twice the number of templates in the last age

1556: bin. Again, this is within reach with the new set of CfA spectra.

1557: 

1558: 

1559: To minimize the impact of our currently non-optimal age distribution

1560: of SN~Ia templates, we have studied the redshift and age residual

1561: distribution when imposing a $1/N_{\rm temp}(t)^a$ weighting scheme,

1562: where $N_{\rm temp}(t)$ is the template age distribution (in 3-day

1563: bins) and $0.0 \le a \le 2.0$ ($a=0.5$ corresponds to a Poisson-like

1564: weighting scheme). This way, the artificial attractors in the actual

1565: age distribution around maximum light and +10 days are down-weighted

1566: with respect to templates at other ages. The weighting scheme does

1567: not lead to any improvement in either the redshift or age

1568: determinations, namely the distribution of residuals for $r{\rm lap} \ge 5$

1569: does not get any narrower. Clearly, a more elaborate method

1570: is necessary to break the redshift-age degeneracy and the

1571: covariance between the two quantities will be included explicitly in a

1572: future version of \snid.

1573: 

1574: 

1575: Fig.~\ref{Fig:opthist} only shows the age distribution of

1576: normal SN~Ia templates, for which we have sufficient spectra in

1577: the database to construct a viable Monte Carlo simulation. The faster

1578: evolution of supernova spectra around maximum light is common to all

1579: supernova types, but the homogeneity at a given age may vary

1580: significantly. We are not in a position to test this thoroughly, due

1581: to the limited number of Type Ib/c and Type II templates in the

1582: current \snid\ database. Again, we are confident that new data from

1583: the CfA Supernova Program will better constrain the variance of SN

1584: spectra at a given age (almost 20 SNe~Ib/Ic/II with more than 10

1585: epochs of spectroscopy since 2000). Therefore, while the overall shape

1586: of the Monte Carlo distributions should remain the same, the

1587: absolute scale should be different for the various supernova types.

1588: 

1589: 

1590: \subsection{Variation of Redshift and Age Accuracy with Redshift, Age,

1591:   S/N and Galaxy Contamination\label{Sect:zts2ngf}}

1592: 

1593: The previous studies are valid for the following

1594: parameter space: $0.3 \le z \le 0.5$; $-5 \le t_B \le

1595: +15$; $2 \le$ S/N (per 2\,\AA)$\le 10$. However, we expect the

1596: accuracy of cross-correlation redshifts and ages to change with

1597: redshift, age and signal-to-noise ratio of the input spectrum. 

1598: 

1599: 

1600: In top panels of the left group of panels of  Figure~\ref{Fig:ztvar_zts2n} we show the

1601: variation of the standard deviation of redshift residuals, $\sigma_z$,

1602: with the $r{\rm lap}$ quality parameter for varying redshift ({\it left}), age ({\it middle}),

1603: and S/N ({\it right}). We expect the degrading accuracy with redshift,

1604: since the rest-frame overlap (${\rm lap}$) between the input and 

1605: template spectra in our database decreases with redshift. Even

1606: requiring that ${\rm lap} \ge 0.4$ can lead to degenerate redshifts at the

1607: higher end of the redshift range ($z \gtrsim 0.5$). Increasing the

1608: number of spectra extending blueward to $\l \gtrsim 2000$\,\AA\ would

1609: partially alleviate this problem, although the flux is strongly

1610: depleted at these wavelengths (due to line-blanketing from iron-group

1611: elements) and the most prominent features in supernova spectra are at

1612: optical wavelengths. UV spectra of (nearby) supernovae are rare, but

1613: the database could be expanded at these wavelengths (for SNe~Ia, at

1614: least) by including the higher-S/N publicly-available 

1615: spectra of ongoing high-$z$ SN searches, such as the spectra from the

1616: ESSENCE project \citep{Matheson/etal:2005}.

1617: 

1618: %%% Figure: sigz vs. rlap for varying z, t and S/N

1619: \begin{figure*}

1620: \plottwo{f17a_col.eps}{f17b_col.eps}

1621: \caption{{\it Left:} Variation of $\sigma_z$ with redshift ({\it top

1622:   left}), age (in days; {\it top middle}) and S/N (per 2\,\AA; {\it

1623:   top right}). We show the effect of applying 

1624:   a $\pm 3$ day constraint on age in the bottom panels. {\it Right:} Same

1625:   as the left group of panels, but for the variation of $\sigma_t$ with redshift ({\it

1626:   top left}), age ({\it top middle}) and S/N

1627:   ({\it top right}). We show the effect of applying a $\pm 0.01$

1628:   constraint on redshift in the bottom panels. {\it [See the electronic

1629:   version of the Journal for a color version of this figure.]}

1630: \label{Fig:ztvar_zts2n}}

1631: \end{figure*}

1632: 

1633: The variation of $\sigma_z(r{\rm lap})$ with S/N of the

1634: input spectrum (Fig.~\ref{Fig:ztvar_zts2n}, {\it left group of panels;

1635:   top right panel}) is also expected, with a significant degradation below S/N

1636: $\lesssim 3$ per 2\,\AA. The degradation with increasing age of the

1637: input spectrum is again due to the slower evolution of SN spectra at

1638: later ages. The input spectrum will correlate well with template

1639: spectra over a larger range of ages, where  the scatter in the

1640: velocity location of spectral features will translate directly into an

1641: error in redshift.

1642: 

1643: 

1644: In the bottom panels of the left group of panels of

1645: Figure~\ref{Fig:ztvar_zts2n} we show 

1646: the effect of applying a flat $\pm 3$ day age constraint on the

1647: $\sigma_z(r{\rm lap})$ curves. The improvement is significant in all cases

1648: (although less so for $z=0.7$). 

1649: 

1650: 

1651: In the top panels of the right group of panels

1652: Figure~\ref{Fig:ztvar_zts2n} we show the 

1653: variation of the standard deviation of age residuals, $\sigma_t$,

1654: with the $r{\rm lap}$ quality parameter for varying redshift ({\it left}), age ({\it middle}),

1655: and S/N ({\it right}). Again, the degradation at the highest redshift

1656: ($z=0.7$) is expected, although it is surprising that the

1657: $\sigma_t(r{\rm lap})$ curves for $z=0.3$ and $z=0.5$ lie atop the one

1658: corresponding to $z=0.1$. It appears that the optimal rest frame

1659: wavelength range of the input spectrum is different for redshift and

1660: age determination. This has already been mentioned by

1661: \citet{Foley/etal:2005} concerning the age determination and

1662: points towards the need for an age- and wavelength-dependent

1663: ${\rm lap}(t,\l)$ parameter to weight the correlation height-noise ratio, $r$, instead of

1664: the constant ${\rm lap}$ currently implemented in \snid. The difficulty of

1665: determining the age of an input spectrum at later times is due to

1666: the less rapid evolution of the spectra at these ages. At high

1667: values of $r{\rm lap}$ ($\gtrsim 7$), however, $\sigma_t$ {\it decreases}

1668: with age. This behavior is unexpected, given the discussion in

1669: \S~\ref{Sect:ztcovar} and could again be due to the

1670: wavelength-independent nature of our ${\rm lap}$ parameter.

1671: 

1672: 

1673: Even more surprising is the apparent independence of $\sigma_t(r{\rm lap})$

1674: on S/N: for fixed redshift and age (here $z=0.5$

1675: and $-5 \le t_B \le +5$), the $r{\rm lap}$ quality parameter gives an {\it absolute}

1676: measure of the age accuracy, regardless of the S/N of the input

1677: spectrum. Of course, the probability of having correlations with high

1678: $r{\rm lap}$ values drops with S/N, but we have checked that our simulation

1679: yielded a sufficient number of correlations at $r{\rm lap} \gtrsim 7$ for

1680: this result to be statistically significant.

1681: 

1682: 

1683: We next study the impact of contamination from the underlying spectrum

1684: of the host galaxy affecting the input supernova spectrum. The

1685: contamination fraction will depend both on the projected position of

1686: the supernova within its host (higher contamination closer to the

1687: nucleus) and on the relative flux difference between the supernova and

1688: the portion of the galaxy located in the same aperture

1689: (i.e. immediately underlying the SN trace, when extracting the

1690: spectrum). Several techniques are commonly used to separate the

1691: supernova light from that of the host galaxy, either through galaxy

1692: template subtraction (e.g., in the algorithm presented by

1693: \citealt{Howell/etal:2005}), or using more elaborate techniques such

1694: as two-channel deconvolution directly applied to the 2D spectrum

1695: \citep{Blondin/etal:2005}. However, neither of these techniques works

1696: well in cases where the SN lies on top of the nucleus of a bright

1697: galaxy and in all other cases there still remains some fraction of

1698: galaxy light in the SN spectrum.

1699: 

1700: 

1701: We contaminate each input spectrum in our simulation with galaxy

1702: light, using the elliptical and Sc galaxy templates of

1703: \citet{Kinney/etal:1996}. The top panels of Figure~\ref{Fig:zvar_gf} 

1704:  show the effect on redshift residuals,

1705: $\sigma_z(r{\rm lap})$, of increasing the galaxy contamination fraction

1706: (expressed in fractions of the total flux) from 0.00 to 0.50. The

1707: impact of the elliptical galaxy ({\it top left panel}) is most

1708: severe, since the spectra of early-type hosts contain broad continuum

1709: structures that yield strong power at similar wavenumbers as

1710: supernova features in Fourier space. Late-type galaxies have smoother

1711: continua and their narrow emission lines are filtered out using the

1712: bandpass filter. In the bottom panels of Fig.~\ref{Fig:zvar_gf} we

1713: apply a flat age constraint of $\pm 3$ days. The improvement, if

1714: any, is negligible for both the elliptical and Sc galaxy types.

1715: 

1716: %%% Figure: z residuals distribution with different galaxy

1717: %%%         contamination fractions

1718: \begin{figure}

1719: \epsscale{1.1}

1720: \plotone{f18_col.eps}

1721: \caption{{\it Top:} Impact of galaxy contamination fraction on

1722:   $\sigma_z(r{\rm lap})$, for both elliptical (E; {\it left panel}) and spiral

1723:   (Sc; {\it right panel}) galaxies. The different lines correspond to

1724:   different galaxy contamination fractions, expressed in fractions of

1725:   the total flux. {\it Bottom:} Effect of applying a $\pm 3$-day

1726:   constraint on the age. {\it [See the electronic version of the

1727:   Journal for a color version of this figure.]} 

1728: \label{Fig:zvar_gf}}

1729: \end{figure}

1730: 

1731: We have run several simulations to test whether the $r{\rm lap}$ quality parameter

1732: could be used to evaluate the amount of galaxy contamination for

1733: various galaxy types, but the results were inconclusive. This

1734: constitutes the real limit of \snid:

1735: % , although the galaxy contamination

1736: % fraction will in general be less than 50\% in real-life

1737: % situations. 

1738: some extra pre-processing of the supernova spectrum is

1739: necessary to ensure that the input to \snid\ is as ``clean'' as

1740: possible. Other algorithms (see \S~\ref{Sect:comparison}) perform a

1741: simultaneous fit of the galaxy fraction when comparing the input SN

1742: spectrum to the set of templates in the database, which enables

1743: the classification of supernovae when the galaxy contamination is

1744: $\lesssim 75\%$ \citep{Howell/etal:2005}.

1745: 

1746: 

1747: \subsection{Comparison with External Measurements}

1748: 

1749: In this section we test the accuracy of correlation redshifts using

1750: \snid\ by comparing them with those of the host galaxy. Galaxy redshifts

1751: ($z_{\rm GAL}$) are routinely determined using nebular emission lines

1752: in their spectra or by cross-correlation with absorption-line galaxy

1753: spectral templates \citep{Kurtz/Mink:1998}. They are typically

1754: accurate to $< 0.001$. However, the redshift of the supernova can

1755: differ from $z_{\rm GAL}$, since the supernova event may have

1756: occurred in a region where its velocity (in the galaxy rest frame) is

1757: different from the mean value, due to the velocity dispersion of the

1758: galaxy's light-emitting component ($\sim 100$\kms\ and

1759: $\sim 200$\kms\ in early- and late-type galaxies, respectively;

1760: \citealt{McElroy:1995}). Nevertheless, $z_{\rm GAL}$ gives

1761: a more accurate determination of the SN redshift than \snid\ (which

1762: has typical redshift errors of $\lesssim 0.01$ for $r{\rm lap} \ge 5$), so a

1763: comparison of the two gives a valuable indication on the accuracy of

1764: \snid\ redshifts, determined from real data.

1765: 

1766: 

1767: We have selected high-redshift SN~Ia spectra taken by the

1768: ESSENCE team (\citealt{Matheson/etal:2005}; Foley et al., {\it in

1769:   prep}; Blondin  et al., {\it in prep}), for which a redshift of the

1770: host galaxy was obtained. This amounts to 57 SN~Ia spectra in

1771: the redshift range $0.164 \le z \le 0.782$. The result of this

1772: comparison is shown in the left panels of Figure~\ref{Fig:zgaltlcsnid}. The

1773: dispersion about the one-to-one correspondence of the redshifts is

1774: excellent, with $\sigma_z \approx 0.005$ over the whole redshift

1775: range. This is in good agreement with the expected redshift residual

1776: found from simulations (with no constraint on the age;

1777: Fig.~\ref{Fig:deltazthist_prior}). The bottom left panel shows a plot of the

1778: redshift residuals as a function of the galaxy redshift. The mean

1779: residual is $\sim 4 \times 10^{-4} \ll \sigma_z$, which shows that

1780: there are no systematic effects in using \snid\ to determine the SN

1781: redshift.

1782: 

1783: %%% Figure: z_SNID vs. z_GAL & t_SNID vs. t_LC

1784: \begin{figure*}

1785: \begin{center}	

1786: \plottwo{f19a.eps}{f19b.eps}

1787: \caption{{\it Left:} Comparison of redshifts determined from

1788:   cross-correlations with SN~Ia spectral templates using \snid\

1789:   ($z_{\rm \snid}$) and from narrow lines in the

1790:   host galaxy spectrum ($z_{\rm GAL}$; {\it top}). We show the

1791:   residuals vs. $z_{\rm GAL}$ in the bottom panel. {\it Right:}

1792:   Comparison of supernova spectral ages determined using \snid\

1793:   ($t_{\rm \snid}$) and rest-frame light-curve ages ($t_{\rm LC}$)

1794:   of high-$z$ SNe~Ia ($0.164 \le z \le 0.587$; {\it top}). We

1795:   show the residuals vs. $t_{\rm LC}$ in the bottom panel. The

1796:   data are from the ESSENCE project

1797:   (\citealt{Matheson/etal:2005,Miknaitis/etal:2007}; Foley et al.,

1798:   {\it in prep}; Blondin et al., {\it in prep}). 

1799: \label{Fig:zgaltlcsnid}}

1800: \end{center}

1801: \end{figure*}

1802: 

1803: To compare the supernova age determined through cross-correlation

1804: with external measurements, we select ESSENCE high-redshift SN~Ia

1805: spectra for which a well-sampled light curve is available around maximum

1806: light \citep{Miknaitis/etal:2007}. This way we can determine the time

1807: difference (in the {\it observer} frame) between maximum light,

1808: ($t_{\rm max}$) and the time the spectrum was obtained ($t_{\rm spec}$)

1809: and compare this time interval with the {\it rest-frame} age ($t_{\rm

1810: \snid}$) determined through cross-correlation with local SN~Ia

1811: templates. For this comparison to make sense we must correct the

1812: light-curve age for the $(1+z)$ time-dilation factor expected in an

1813: expanding universe

1814: \citep{Wilson:1939,Rust:1974,Leibundgut/etal:1996,Goldhaber/etal:2001}.

1815: We expect a one-to-one correspondence between

1816: 

1817: 

1818: \begin{equation}

1819: t_{\rm LC} = \frac{t_{\rm spec}-t_{\rm max}}{1+z}

1820: \end{equation}

1821: 

1822: 

1823: \noindent

1824: and $t_{\rm \snid}$. The result is shown in the right panels of

1825: Figure~\ref{Fig:zgaltlcsnid}. We used a total of 54 spectra in the

1826: redshift range 

1827: $0.205 \le z \le 0.687$, 27 of which had an associated galaxy

1828: redshift--- which we used as a constraint when determining

1829: the age. The dispersion about the $t_{\rm \snid} = t_{\rm LC}$

1830: line is $\sigma_t \approx 2.9$ days over a age interval $-10 \lesssim

1831: t_{\rm LC} \lesssim +20$, again in good agreement with the expected

1832: residuals (Fig.~\ref{Fig:deltazthist_prior}). We show the

1833: residuals versus $t_{\rm LC}$ in the bottom right panel. The mean  residual

1834: is approximately $-0.7$ days. 

1835: % Two points at $t_{\rm LC} < -10$ days are

1836: % $\sim$2--3$\sigma$ off the null residual line. This is due to the low

1837: % number of SN~Ia templates at these early ages

1838: % (Fig.~\ref{Fig:thist}). 

1839: The excellent correspondence between $t_{\rm

1840:   LC}$ and $t_{\rm \snid}$ shows that \snid\ can be used in studies of

1841: time dilation effects in high-redshift multiepoch SN~Ia spectra 

1842: \citep{Riess/etal:1997,Foley/etal:2005}.

1843: 

1844: 

1845: The correlation technique could not have yielded such good results had

1846: the high-$z$ SNe~Ia in the sample been significantly different from

1847: the SN~Ia template spectra in the \snid\ database. The fact that the

1848: correlation redshifts and ages agree so well with the galaxy

1849: redshifts and light-curve ages, respectively, is a strong argument

1850: in favor of the similarity of 

1851: % SNe~Ia at low and high redshifts.

1852: these SNe~Ia with local counterparts.

1853: 

1854: 

1855: 

1856: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1857: %%

1858: %%   Section 6 - Type Determination

1859: %%

1860: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1861: 

1862: \section{Type Determination}\label{Sect:type}

1863: 

1864: The results of \S~\ref{Sect:zt} are only valid if we assume that we know

1865: the type of the input supernova spectrum--- in this case a normal

1866: SN~Ia. Although \snid\ is tuned to determining SN redshifts, we

1867: investigate its potential in determining the SN type in an impartial

1868: way. We base our investigation on a simple frequentist

1869: approach as opposed to a more elaborate Bayesian one, but we discuss the

1870: future implementation of the latter in \snid\ in

1871: \S~\ref{Sect:comparison}.

1872: 

1873: 

1874: In what follows we focus on five distinct examples, the first three

1875: being particularly relevant to ongoing high-redshift SN~Ia searches:

1876: the distinction between 1991T-like SNe~Ia and other SNe~Ia

1877: (\S~\ref{Sect:1anorm1a91t}); the distinction  

1878: between SNe~Ib/c and SNe~Ia at high redshifts

1879: (\S~\ref{Sect:1cnorm1a}); the identification of peculiar SNe~Ia

1880: (\S~\ref{Sect:1apec}); finally, the distinction between SNe~Ib and

1881: SNe~Ic and between SNe~IIb and both SNe~II 

1882: and Ib (\S~\ref{Sect:otheregs}), more relevant to ongoing nearby ($z

1883: \lesssim 0.1$) supernova searches. We used the same simulation setup

1884: as in \S~\ref{Sect:simulation}, except we consider correlations with

1885: all supernova types in the database.

1886: 

1887: 

1888: The reader must keep in mind that, while the age distribution of

1889: SN~Ia templates is close to optimal in the current \snid\ database

1890: (see \S~\ref{Sect:ztcovar}), those for supernovae of other types are

1891: most certainly not. While the results presented in this section are

1892: encouraging, they are no doubt biased by the relatively low number of

1893: SN~Ib/Ic/II templates with respect to SNe~Ia.

1894: 

1895: 

1896: \subsection{Normal versus 1991T-like SNe Ia\label{Sect:1anorm1a91t}}

1897: 

1898: It can be challenging to distinguish the subtypes of SNe~Ia from one

1899: another. 1991T-like SNe~Ia have a peak luminosity at the bright end of

1900: the SN~Ia distribution and although their light curves still obey the

1901: width-peak luminosity, or ``Phillips,'' relation

1902: \citep{Phillips:1993}, it is useful to have an independent  

1903: confirmation of their high intrinsic luminosity from their

1904: spectra. Spectra of 1991T-like SNe~Ia are characterized by the

1905: near-absence of \catwo\ and \sitwo\ lines in the early-time spectra,

1906: and prominent high-excitation features of \fethree--- not found in

1907: normal SNe~Ia. The \sitwo, \stwo and \catwo\ features develop

1908: during the post-maximum ages and by $\sim 2$ weeks past maximum the

1909: spectra of ``1991T-like'' objects are similar to those of normal

1910: SNe~Ia

1911: \citep{Filippenko/etal:1992,Ruiz-Lapuente/etal:1992,Phillips/etal:1992,Jeffery/etal:1992}.

1912: 

1913: 

1914: In the top panels of Figure~\ref{Fig:Ia91t} we illustrate the ability

1915: of \snid\ to identify 1991T-like SNe~Ia at $z=0.5$. The input

1916: spectrum is a ``Ia-91T'' template in the \snid\ database 

1917: (Table~\ref{Table:sndb}) in the age interval $-5 \le t_B

1918: \le +5$, i.e. when the spectroscopic differences with normal

1919: SNe~Ia are most apparent. We show the fraction of templates in the

1920: \snid\ database that correlate with the input spectrum, as a function

1921: of the $r{\rm lap}$ quality parameter: 1991T-like SNe~Ia ({\it solid line}),

1922: other SNe~Ia ({\it dashed line}), and supernovae of other types

1923: ({\it dotted line}). From left to right, we show the effect of

1924: having no constraint on either age or redshift, a flat $\pm 0.01$ constraint

1925: on redshift, a flat $\pm 3$ day constraint on the age, and a combined constraint

1926: on both redshift and age.

1927: 

1928: %%% Figure: Ia-91T vs. other Ia's

1929: \begin{figure*}

1930: \begin{center}	

1931: \includegraphics[width=7in]{f20_col.eps}

1932: \caption{Attempt to identify a 1991T-like SN~Ia at $z=0.5$ in the

1933:   age interval $-5 \le t_B \le +5$. {\it Top:}

1934:   Fraction of templates in the \snid\ database corresponding to a

1935:   certain supernova type (1991T-like SN~Ia: {\it solid line}; SN~Ia of

1936:   other subtypes: {\it dashed line}; supernova of other types: {\it

1937:   dotted line}), in $r{\rm lap}$ bins of size unity. {\it Left to

1938:   right:} With no constraints on the redshift or age, with a $\pm

1939:   0.01$ constraint on the redshift, with a $\pm 3$ days constraint on

1940:   the age, and with a combined constraint on redshift and age.

1941:   {\it Bottom:} Same lines as above, but for post-maximum

1942:   spectra in the age interval $+5 \le t_B \le +15$.

1943:   The right panel shows representative spectra of 1991T-like and

1944:   normal SNe~Ia, around maximum light ({\it top}) and

1945:   $\sim$1--2 weeks past maximum ({\it bottom}), as observed with a

1946:   typical optical spectrograph at $z=0.5$ (the cross-hatched area represents

1947:   the rest frame portion of the spectrum that is lost due to the

1948:   redshift). Note that the relative differences in {\it

1949:   pseudo}-continuum shapes have no impact on the \snid\ results. {\it

1950:   [See the electronic version of the Journal for a color version of

1951:   this figure.]} 

1952: \label{Fig:Ia91t}}

1953: \end{center}

1954: \end{figure*}

1955: 

1956: When there is no constraint on either redshift or age, the fraction of

1957: Ia-91T templates ({\it solid line}) for $r{\rm lap} \gtrsim 10$ is

1958: greater than that for other SN~Ia templates ({\it dashed line}). Note

1959: that for $r{\rm lap} \gtrsim 5$ the confusion 

1960: with supernovae of other types is practically non-existent ($\lesssim

1961: 2\%$). Adding a constraint on the redshift does not lead to a

1962:   significant improvement, but a constraint on age reduces the cross-over

1963:  $r{\rm lap}$ value between other SN~Ia and 1991T-like templates from

1964:   $r{\rm lap} \gtrsim 15$ to $r{\rm lap} \approx 6$.

1965: % or age reduces the cross-over

1966: % $r{\rm lap}$ value between a domination of other SNe~Ia to 1991T-like

1967: % templates from $r{\rm lap} \approx 10$ to $r{\rm lap} \approx 8$ and $r{\rm lap} \approx

1968: % 6$, respectively. 

1969: The negative noise spikes around $r{\rm lap} \approx 15$ are

1970: statistical noise due to the small number of templates with $r{\rm lap}$

1971: values in excess of that cutoff.

1972: 

1973: 

1974: The bottom panels of Figure~\ref{Fig:Ia91t} show the same lines for

1975: input spectra in the age interval $+5 \le t_B \le +15$. At

1976: these post-maximum ages, the differences between 1991T-like and

1977: normal SNe~Ia are less apparent (see the rightmost panel) and the

1978: impact on the ability of \snid\ to distinguish between the different Ia

1979: subtypes is readily apparent. In the absence of constraints on redshift or

1980: age, the fraction of Ia-91T templates reaches a peak of 

1981:   $\sim 70$\%, while it increases to 100\% with constraints

1982: on redshift and age. The variation for $r{\rm lap} \ge 15$

1983: is again statistical noise due to the limited number of templates with

1984: such good correlations.

1985: 

1986: 

1987: The difficulty of distinguishing between normal and 1991T-like

1988: SNe~Ia at high redshift could partly explain the apparent lack of

1989: 1991T-like SNe~Ia at high redshifts ($2/52 \approx 4$\% in the

1990: SN~Ia sample published by \citealt{Matheson/etal:2005}) with respect

1991: to the fraction expected locally (up to $\sim20$\% according to

1992: \citealt{Li/etal:2001b}).

1993: 

1994: 

1995: \subsection{SN Ia versus SN Ib/c\label{Sect:1cnorm1a}}

1996: 

1997: The misidentification of supernovae of other types as SNe~Ia is a

1998: major concern for ongoing high-redshift SN~Ia searches. Including only

1999: a small fraction of non-Ia supernovae in a sample would lead to a

2000: mis-calibration of the absolute magnitudes of these objects and to

2001: biases in the derived cosmological parameters \citep{Homeier:2005}. A

2002: particular concern is the contamination of high-z SN~Ia samples with

2003: SNe~Ib/c. At redshifts $z \gtrsim 0.4$, the defining \sitwo\

2004: $\l 6355$ absorption feature of SNe~Ia (also present, although somewhat

2005: weaker, in SNe~Ic) is redshifted out of the range of most optical

2006: spectrographs and one has to rely on spectral features blueward of

2007: this to determine the supernova type. Some of these features, such as

2008: the \catwo\ H and K $\l \l 3934, 3968$ doublet, are common to both SNe~Ia

2009: and SNe~Ib/c. Other features characteristic of SN~Ia spectra around

2010: maximum light (e.g. \stwo\ $\l \l 5454, 5640$) are generally weak and

2011: can be difficult to detect in low-S/N spectra. One has to invoke

2012: external constraints, such as the SN color evolution, light-curve

2013: shape, host galaxy morphology (only SNe~Ia occur in early-type hosts;

2014: \citealt{Cappellaro/etal:1997}), or the expected apparent peak

2015: magnitude: SNe~Ib/c at maximum light are often $\gtrsim

2016: 1$~mag fainter than SNe~Ia (\citealt{Richardson/etal:2006};

2017: although \citealt{Clocchiatti/etal:2000} have reported on one SN~Ic

2018: with an absolute magnitude similar to normal SNe~Ia) and hence are

2019: only expected to ``pollute'' magnitude-limited samples of SNe~Ia at

2020: the lower-redshift end. If the redshift is not known, SNe~Ib/c can be a

2021: serious contaminant for high-redshift SN~Ia searches.

2022: % and hence are not expected to ``pollute''

2023: %magnitude-limited samples of SNe~Ia at high redshift (although

2024: %\citealt{Clocchiatti/etal:2000} have reported on one Type Ic SN with an

2025: %absolute magnitude similar to normal SNe~Ia).

2026: 

2027: 

2028: In the top panels of Figure~\ref{Fig:Icnorm} we illustrate the ability

2029: for \snid\ to identify SNe~Ib/c at $z=0.3$. The input spectrum is a

2030: ``Ib-norm'' or ``Ic-norm'' template in the \snid\ database 

2031: (Table~\ref{Table:sndb}) in the age interval $-5 \le t_B

2032: \le +5$. We show the fraction of templates in the \snid\ database that

2033: correlate with the input spectrum, as a function of the $r{\rm lap}$

2034: quality parameter: all SNe~Ib/c (including SNe~IIb; {\it solid line}), SNe~Ia ({\it dashed

2035:   line}), and SNe~II (excluding SNe~IIb; {\it dotted line}).

2036: 

2037: %%% Figure: Ic-norm vs. other Ia

2038: \begin{figure*}

2039: \begin{center}	

2040: \includegraphics[width=7in]{f21_col.eps}

2041: \caption{Same as Fig.~\ref{Fig:Ia91t}, but for normal SNe~Ic

2042:   in the age interval $-5 \le t_B \le +15$, at

2043:   $z=0.3$ ({\it top}) and $z=0.5$ ({\it bottom}). Here

2044:   the lines correspond to fractions of SNe~Ib/c ({\it solid

2045:   line}), SNe~Ia of all subtypes ({\it dashed line}), and SNe~II

2046:   (excluding SNe~IIb, {\it dashed line}). {\it [See the electronic

2047:   version of the Journal for a color version of this figure.]}

2048: \label{Fig:Icnorm}}

2049: \end{center}

2050: \end{figure*}

2051: 

2052: In the absence of a constraint on age, correlations with $r{\rm lap}

2053: \gtrsim 4$ are sufficient to recover a dominant fraction of SNe~Ib/c over

2054: SNe~Ia. The confusion with SNe~II is practically nonexistant ($<5\%$

2055: for $r{\rm lap} \lesssim 3$; 0\%\ for $r{\rm lap} > 3$). A constraint on age only

2056: reduced the cross-over $r{\rm lap}$ between SNe~Ia and SNe~Ib/c from

2057: $r{\rm lap} \approx 4$ to $r{\rm lap} \approx 3$ but leads to less correlations

2058: with $r{\rm lap} \gtrsim 5$ (hence the noise spikes in the recovered

2059: template fractions). A combined constraint on redshift and age yields

2060: a 100\%\ SN~Ib/c fraction for $r{\rm lap} \ge 4$, but no correlations with

2061: $r{\rm lap} > 8$.

2062: 

2063: %In the absence of constraints on redshift or age, correlations with $rlap

2064: %\gtrsim 10$ are needed to recover a dominant fraction of SNe~Ic over

2065: %SNe~Ia. The confusion with supernovae of 

2066: %other types ({\bf essentially only} SNe~Ib) is low {\bf ($\lesssim 25\%$)}, but persists

2067: %out to $rlap \approx 10$. A constraint on the redshift {\bf leads to a

2068: %marginal improvement, while a constraint on age leads to a dominant

2069: %fraction of normal SN~Ic templates for $rlap \gtrsim 8$.}

2070: %% or age leads to a

2071: %% significant improvement, with the fraction of ``Ic-norm'' templates

2072: %% dominating for $rlap \gtrsim 8$ and $rlap \gtrsim 6$, respectively. 

2073: %A constraint on the age prevents a confusion that can arise between spectra

2074: %of SNe~Ic around maximum light and SNe~Ia at $\sim$1--2 weeks past

2075: %maximum.

2076: 

2077: At $z=0.5$, the results are essentially unchanged from $z=0.3$,

2078: although the absolute number of good correlations ($r{\rm lap} \gtrsim 5$)

2079: is generally less. We also note that these results are biased by the

2080: confusion between SNe~Ib and SNe~Ic at this redshift (up to

2081: $\sim40\%$). The constraint on age leads to a more significant 

2082: improvement than at $z=0.3$, suggesting that the similarity between

2083: spectra of SNe~Ic around maximum light and SNe~Ia at  $\sim$1--2

2084: weeks past maximum is less problematic over this restricted

2085: wavelength range.

2086: 

2087: %At $z=0.5$, however, the fraction of SN~Ic templates is the lowest in

2088: %the absence of constraints on redshift or age 

2089: %(Fig.~\ref{Fig:Icnorm}, {\it bottom panel}) {\bf  and is still

2090: %  superseded by SN~Ia templates with a constraint on redshift. A

2091: %  constraint on age leads to similar fractions of SNe~Ia, Ib and Ic

2092: %at $rlap \approx 9$. For $rlap \gtrsim 10$ the fraction of SN~Ib

2093: %templates reaches 100\%, but this result is based on a single

2094: %template.}

2095: %% . A constraint on the redshift

2096: %% leads to a more significant improvement than a constraint on

2097: %% age, but this is in part due to the small number of

2098: %% correlations with $rlap \gtrsim 5$ in this particular simulation. 

2099: %{\bf Only a} combined constraint on redshift and age leads to a

2100: %domination of Type Ic templates for {\bf $rlap \gtrsim 6$}.

2101: 

2102: 

2103: Note that the mis-classification of SNe~Ib/c as SNe~Ia (or the reverse)

2104: can sometimes pose problems with the high-S/N spectra of nearby

2105: objects, especially at later ages or if no age information is

2106: available. A striking example is the nearby SN~Ic SN~2004aw

2107: \citep{Taubenberger/etal:2006}, originally classified as an SN~Ia

2108:  by \citet{Benetti/etal:2004b}. More recently, two nearby

2109: supernovae originally announced as Type Ic events around maximum light

2110: (SN~2006bb and SN~2006bk;

2111: \citealt{Kinugasa/Yamaoka:2006a,Kinugasa/Yamaoka:2006b}) were

2112: re-classified as SNe~Ia at 2--3 weeks past maximum light

2113: based on cross-correlations with SN spectra of all types using \snid\

2114: \citep{Blondin/etal:2006b,Blondin/etal:2006c}.

2115: 

2116: 

2117: \subsection{Identifying SN~Ia ``Oddballs''\label{Sect:1apec}}

2118: 

2119: Some SNe~Ia, which we refer to as peculiar, do not

2120: belong to any of the normal, 1991T-like, or 1991bg-like

2121: categories. Such is the case of SNe~2000cx \citep{Li/etal:2001a}, 2002cx

2122: \citep{Li/etal:2003} and 2005hk \citep{Phillips/etal:2007}. The first

2123: of these has pre-maximum spectra similar to those of SN~1991T,

2124: although the \sitwo\ lines that appear around maximum light remain

2125: strong several weeks past maximum. SN~2002cx, 

2126: ``the most peculiar known SN~Ia''

2127: \citep{Li/etal:2003,Branch/etal:2004,Jha/etal:2006b} and SN~2005hk are

2128: even more difficult to accommodate in the current classification

2129: scheme: their early-time spectra show signatures of high-ionization

2130: lines of iron, as in the overluminous SN~1991T, but their luminosity

2131: is similar to that of the subluminous SN~1991bg. Moreover, their

2132: $I$-band light curves are devoid of the secondary maximum present in

2133: all other Ia subtypes. Both objects possibly originate from a pure

2134: deflagration explosion (see \citealt{Phillips/etal:2007}) and could

2135: form an altogether separate class of SNe~Ia.

2136: 

2137: 

2138: These peculiar SNe~Ia do not obey the Phillips relation and thus

2139: cannot be used as calibrated distance indicators. They must be weeded

2140: out of high-redshift SN~Ia samples in order to avoid significant

2141: biases in the derived cosmological parameters. Until recently, there

2142: was no evidence for such peculiar events at high redshifts. This has

2143: changed with the recent discovery of the overluminous SNLS-03D3bb

2144: (SN~2003fg) at $z=0.244$ \citep{Howell/etal:2006}.

2145: 

2146: 

2147: In the top panels of Figure~\ref{Fig:Iapec} we illustrate the ability

2148: of \snid\ to identify peculiar SNe~Ia at $z=0.3$. The input

2149: spectrum is a ``Ia-pec'' template in the \snid\ database 

2150: (Table~\ref{Table:sndb}) in the age interval $-5 \le t_B

2151: \le +5$. We show the fraction of templates in the \snid\ database

2152: that correlate with the input spectrum, as a function of the $r{\rm lap}$

2153: quality parameter: peculiar SNe~Ia ({\it solid line}), other SNe~Ia

2154: ({\it dashed line}), and supernovae of other types ({\it dotted line}). In

2155: the absence of a constraint on redshift, the maximum recovered fraction of

2156: Ia-pec templates is $\sim 5$\%, with no correlations at $r{\rm lap} >

2157: 6$. With a constraint on redshift, the Ia-pec template fraction peaks

2158: at $\sim 70$\%, but the best correlations are always for another Ia

2159: subtype (most frequently 1991T-like). At $z=0.5$, the recovered

2160: fraction of Ia-pec templates is complete for $r{\rm lap} \ge 5$ for both

2161: a constraint on redshift and a combined constraint on redshift and

2162: age. This apparent improvement is counterbalanced by the absence of

2163: correlations with $r{\rm lap} > 9$. Note that in the absence of any

2164: constraint, the recovered fraction of Ia-pec templates is dominant

2165: for $r{\rm lap} \gtrsim 6$.

2166: % input spectrum has a higher probability of correlating

2167: % with a supernova of another type.

2168: 

2169: %%% Figure: Ia-pec vs. other Ia's

2170: \begin{figure*}

2171: \begin{center}	

2172: \includegraphics[width=7in]{f22_col.eps}

2173: \caption{Same as Fig.~\ref{Fig:Ia91t}, but for a peculiar SN~Ia

2174:   in the age interval $-5 \le t_B \le +15$, at

2175:   $z=0.3$ ({\it top}) and $z=0.5$ ({\it bottom}). Here

2176:   the lines correspond to fractions of peculiar SNe~Ia ({\it solid

2177:   line}), other SNe~Ia ({\it dashed line}), and all other

2178:   supernova types ({\it dashed line}). {\it [See the electronic

2179:   version of the Journal for a color version of this figure.]} 

2180: \label{Fig:Iapec}}

2181: \end{center}

2182: \end{figure*}

2183: 

2184: Despite the limited number of peculiar SNe~Ia in our database

2185: (SN~2000cx, SN~2002cx and SN~2005hk, for a total of 15 spectra in the

2186: age interval $-5 \le t_B \le +5$; see Table~\ref{Table:sndb}), the

2187: change in template fraction from $z=0.3$ to $z=0.5$ gives indications

2188: as to which rest frame wavelength range is most valuable for

2189: determining the SN type. This again calls for a wavelength (and age)

2190: weighting of the spectrum overlap parameter, ${\rm lap}(\l,t)$ (see \S~\ref{Sect:zts2ngf}), which we

2191: are working to implement in a future version of \snid.

2192: 

2193: 

2194: We note that \snid\ unambiguously confirms the similarity between

2195: SN~2005hk and SN~2002cx (see \citealt{Phillips/etal:2007}): for all the

2196: spectra of SN~2005hk in the \snid\ database (Table~\ref{Table:sndb}), the

2197: best-match template spectrum is SN~2002cx, whether a constraint on redshift

2198: and age is applied or not. In the absence of a constraint on redshift,

2199: however, the fraction of 1991T-like templates that correlate with the

2200: input SN~2005hk spectrum increases dramatically, leading to an

2201: overestimation of the redshift by $\sim 0.02$, roughly corresponding

2202: to the difference in absorption velocities between SN~2005hk and

2203: SN~1991T at a given age ($\sim 6000$\kms;

2204: \citealt{Phillips/etal:2007}).

2205: 

2206: 

2207: SNLS-03D3bb (SN~2003fg) at $z=0.244$ \citep{Howell/etal:2006}, on

2208: the other hand, illustrates the limitations of the cross-correlation

2209: approach in determining the SN type, when applied to objects that are

2210: not part of the library of spectral templates. Its spectrum (at +2

2211: days; A.~Howell, private communication) is unique among supernova

2212: spectra and we have no similar examples in our database. 

2213: In the absence of constraints, the best-match template is the

2214: 1991T-like SN~1999dq \citep{Matheson/etal:2007} at $t=+6.2$ days

2215: ($z=0.254 \pm 0.005$). With constraints on age or redshift, the

2216: best-match template is in all cases the normal SN~Ia SN~1989B at

2217: $t=+3.5$ days ($z=0.251 \pm 0.005$). The spectrum of SNLS-03D3bb is

2218: now part of the \snid\ database and will prove useful to identify

2219: such peculiar objects at all redshifts.

2220: 

2221: 

2222: \subsection{Further Specific Examples\label{Sect:otheregs}}

2223: 

2224: We next focus on two further specific examples, relevant to the

2225: spectroscopic classification of supernovae in nearby ($z \lesssim

2226: 0.1$) SN searches: the distinction between SN~Ib and Ic,

2227: and that between SN~IIb and II/Ib.

2228: 

2229: \subsubsection{SN~Ib versus SN~Ic}

2230: 

2231: SN~Ib and Ic are often difficult to tell apart and are

2232: sometimes referred to as ``SNe~Ib/c'' in the literature

2233: (e.g., SN~1999ex; \citealt{Hamuy/etal:2002})

2234: and IAU circulars. This difficulty is inherent to the SN

2235: classification scheme, rather than from a physical mis-conception of

2236: these events, both of which are believed to originate in the core

2237: collapse of a massive star, stripped of its outer layers

2238: through either stellar winds or interaction with a binary companion 

2239: \citep{Woosley/Langer/Weaver:1993,Woosley/Langer/Weaver:1995}. SNe~Ib

2240: are defined by the presence of conspicuous lines of \heone\ in their 

2241: optical spectra, whereas SNe~Ic are defined by their

2242: quasi-absence \citep{Matheson/etal:2001}. The \sitwo\

2243: \l6355 feature is weak in SNe~Ib/c, which enables one to

2244: differentiate them from SNe~Ia, at least in principle (see

2245: \S~\ref{Sect:1cnorm1a}). Both are of Type I and are thus also defined

2246: by the absence of hydrogen lines in their spectra, although the case

2247: has recently been made for some hydrogen being present in SNe~Ib/c 

2248: \citep{Branch/etal:2006,Elmhamdi/etal:2006}.

2249: 

2250: 

2251: In Fig.~\ref{Fig:Ib}, we show the result of cross-correlating SN~Ib

2252:  spectra at low redshift ($z=0.1$) within 10 days from

2253: $V$-band maximum with SNe of all types in the \snid\ database:

2254: Type Ib (excluding Type IIb; {\it solid line}), Type Ic ({\it dashed

2255:   line}), and other SN types ({\it dotted line}). The lines correspond to

2256: the fraction of template supernovae of a given type, as a 

2257: function of the $r{\rm lap}$ quality parameter. We deliberately exclude the IIb

2258: subtype from this analysis, as this type is a hybrid between the

2259:  II and Ib subtypes (see below). The results of Fig.~\ref{Fig:Ib}

2260: are encouraging: for $r{\rm lap} > 6$, the recovered fraction of

2261: SN~Ib dominates over SNe~Ic, with less than 25\%\

2262: confusion with other supernova types.

2263: % more than $80\%$ of templates

2264: % correspond to the Ib type (with less than 20\%\ confusion with Type Ic

2265: % supernovae). 

2266: For $r{\rm lap} \ge 13$, only Type Ib templates correlate with

2267: the input spectrum.

2268: 

2269: %%% Figure: Ib vs. Ic

2270: \begin{figure}

2271: \epsscale{1.15}

2272: \plotone{f23_col.eps}

2273: \caption{Attempt to classify an SN~Ib at $z=0.1$ in the 

2274:   age interval $-10 \le t_V \le +10$. The lines

2275:   correspond to fractions of matching templates of Type Ib ({\it solid

2276:   line}), Type Ic ({\it dashed line}) and all other types ({\it

2277:   dotted line}), in $r{\rm lap}$ bins of size unity. The Type IIb subclass,

2278:   included in both the SN~Ib and SN~II categories, is excluded for

2279:   this purpose. The right panel shows representative spectra of

2280:   normal SN~Ib and Ic around maximum light. Note that

2281:   the relative differences in {\it pseudo}-continuum shapes have no

2282:   impact on the \snid\ results. {\it [See the electronic version of

2283:   the Journal for a color version of this figure.]} 

2284: \label{Fig:Ib}}

2285: \end{figure}

2286: 

2287: 

2288: \subsubsection{SN~IIb versus SN~II/Ib}

2289: 

2290: Some supernovae ``evolve'' from one type to another as they age. 

2291: Such is the case for SNe~IIb 

2292: \citep{Woosley/etal:1987}, whose early-time spectrum is characterized

2293: by prominent P Cygni lines of the hydrogen Balmer series (as in

2294: SNe~II), but which later develop conspicuous lines of

2295: \heone, as in SNe~Ib. A proto-typical example of such a

2296: supernova is SN~1993J \citep{Nomoto/etal:1993,Matheson/etal:2000}. The

2297: \snid\ database currently contains spectra for three SNe~IIb:

2298: SN~1993J, SN~2000H and SN~2006T (Table~\ref{Table:sndb}).

2299: 

2300: 

2301: In Fig.~\ref{Fig:IIb} we study the fraction of templates corresponding

2302: to Type IIb ({\it solid line}), Type II ({\it dashed line}), and Type Ib

2303: ({\it dotted line}), when the input spectrum is a Type IIb at low

2304: redshift ($z=0.1$), within 15 days past explosion. At these ages,

2305: the \heone\ lines are somewhat weaker than after maximum and the

2306: confusion with SNe~II is greater: only for $r{\rm lap} \gtrsim

2307: 7$ is the fraction of recovered IIb templates greater than ordinary

2308: SNe~II. The confusion with Type Ib templates is small ($< 20\%$)

2309: up to $r{\rm lap} \approx 9$ and null for larger values of

2310: $r{\rm lap}$. Templates corresponding to SNe~Ia and Ic do not correlate

2311: well with input Type IIb spectra and the confusion is

2312: practically nonexistent ($< 5\%$, not shown here). Again, more SNe~IIb

2313: are needed to truly test the ability for \snid\ to

2314: correctly identify them.

2315: 

2316: %%% Figure: IIb vs. Ib/II

2317: \begin{figure}

2318: \epsscale{1.15}

2319: \plotone{f24_col.eps}

2320: \caption{Same as Fig.~\ref{Fig:Ib}, but for an SN~IIb at

2321:   $z=0.1$ in the age interval $t_{\rm exp} \le +15$

2322:   ($t_{\rm exp}$ is the number of days past explosion). The lines

2323:   correspond to fractions of matching templates of Type IIb ({\it solid

2324:   line}), Type II ({\it dashed line}) and Type Ib ({\it dotted

2325:   line}), in $r{\rm lap}$ bins of size unity. {\it [See the electronic

2326:   version of the Journal for a color version of this figure.]}

2327: \label{Fig:IIb}}

2328: \end{figure}

2329: 

2330: 

2331: \subsection{Using \snid\ for SN Identification}

2332: 

2333: The previous examples illustrate the ability of \snid\ to recover a

2334: significant fraction of supernovae in the database corresponding to

2335: the input SN type. While Figs.~\ref{Fig:Ia91t}--\ref{Fig:IIb} are

2336: informative, they do not provide a unique answer to the following: how

2337: does one use \snid\ to determine the type of an SN spectrum and

2338: can one relate the $r{\rm lap}$ quality parameter to a formal confidence that an

2339: input spectrum is of a certain type (and, sometimes more importantly,

2340: that it is not of another type)? The answer is far from settled,

2341: and its resolution will probably involve a more sophisticated Bayesian

2342: approach (see next section). Nevertheless, we have tested the following

2343: classification schemes:

2344: 

2345: \begin{itemize}

2346: 

2347: \item[1.]{The SN type (subtype) is simply determined as the type

2348:   (subtype) of the best-match template(s) for $r{\rm lap}$ greater than some

2349:   cutoff value $r{\rm lap}_{\rm min}$.}

2350: 

2351: 

2352: \item[2.]{The SN type (subtype) is the one corresponding to the

2353:   highest fraction of templates corresponding to a given type

2354:   (subtype) for $r{\rm lap} \ge r{\rm lap}_{\rm min}$, with the possible

2355:   additional requirement that this fraction exceeds some cutoff.}

2356: 

2357: \item[3.]{The SN type (subtype) is the one whose fraction increases

2358:   most with increasing $r{\rm lap}$ (i.e. the lines shown in

2359:   Figs.~\ref{Fig:Ia91t}--\ref{Fig:IIb} that have the highest positive

2360:   gradient for $r{\rm lap} \ge r{\rm lap}_{\rm min}$), with the possible

2361:   additional requirement that this gradient exceeds some cutoff.}

2362: 

2363: \end{itemize}

2364: 

2365: The first of these is the one commonly used for the identification of

2366: supernovae, at both low and high redshift. In IAU circulars, a

2367: supernova is announced to be of a given type when it is ``(most)

2368: similar to supernova $X$ at $N$ days from maximum light.'' In high-$z$

2369: SN~Ia searches, a secure type is determined when a given spectrum is

2370: sufficiently similar to a nearby SN~Ia (e.g.,

2371: \citealt{Matheson/etal:2005}), but it is never clear how {\it 

2372: different} it is from supernovae of other types. However, the

2373: best-match template is not always the best indicator of the SN type

2374: (see the distinction between post-maximum spectra of 1991T-like SNe~Ia

2375: and other SNe~Ia in Fig.~\ref{Fig:Ia91t}) and the large spectral

2376: database used in \snid\ offers the possibility to use the statistical

2377: power of correlations exceeding a certain $r{\rm lap}$ cutoff. This second

2378: classification scheme has already been used for the classification of

2379: high-$z$ SN~Ia spectra using \snid\ \citep{Miknaitis/etal:2007}. A

2380: drawback of such an approach is that the determination of the SN type

2381: depends on the completeness of the SN database--- which comprises few 

2382: template spectra of core-collapse SNe (Ib,Ic, and II; see

2383: \S~\ref{Sect:sniddb}). Thus, for instance, it is not possible to

2384: identify peculiar SNe~Ia this way (Fig.~\ref{Fig:Iapec}). Last, the

2385: ``gradient method'' for classification is generally more robust, but

2386: also requires some other constraint on either the template fraction or

2387: the type of the best-match template(s).

2388: 

2389: 

2390: We are extensively testing these different combinations, to

2391: optimize the type determination for all SN types, at varying $z$, $t$,

2392: and S/N, although we suspect that a more elaborate Bayesian treatment will

2393: be required to properly account for the probability of an input

2394: spectrum to be of a certain type (as well as {\it not} being of some

2395: other type).

2396: 

2397: 

2398: 

2399: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2400: %%

2401: %%   Section 7 - Comparison with other methods

2402: %%

2403: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2404: 

2405: \section{Comparison with Other Methods and Further Improvements}\label{Sect:comparison}

2406: 

2407: Several other methods are used to determine the type,

2408: redshift, or age of a supernova spectrum. We briefly describe them

2409: here, with a distinction between cross-correlation and

2410: ``$\chi^2$-based'' methods. Last, we comment on the alternative

2411: Bayesian approach to supernova classification--- only applied so far

2412: to photometric measurements and its possible implementation in \snid.

2413: 

2414: 

2415: Other spectral classification methods involve principle component

2416: analysis (PCA), possibly in combination with artificial neural

2417: networks. These are beyond the scope of this paper and we do not

2418: discuss them here, although the PCA method has already been applied to SN~Ia

2419: classification by \citet{James/etal:2006}.

2420: 

2421: \subsection{Cross-correlation Methods}

2422: 

2423: We are aware of two other algorithms based on the cross-correlation

2424: techniques of \citet{Tonry/Davis:1979}. Both are aimed at determining

2425: redshifts of galaxies (or stars) in large surveys, but could easily

2426: be tuned (e.g., by modifying the shape of the bandpass

2427: filter and including age information) to supernovae as in \snid.

2428: 

2429: 

2430: XCSAO \citep{Kurtz/Mink:1998} is a program part of the

2431: IRAF\footnote{IRAF is distributed by the National Optical Astronomy

2432: Observatories, operated by the Association of Universities for

2433: Research in Astronomy, Inc., under contract to the National Science

2434: Foundation of the United States.} RVSAO package, aimed at obtaining 

2435: redshifts and radial velocities from digital spectra. It has been used

2436: extensively in the past for redshift surveys (for references see

2437: \citealt{Kurtz/Mink:1998}) and is currently used in

2438: the Smithsonian Hectospec Lensing Survey (SHELS;

2439: \citealt{Geller/etal:2005}). The basic algorithm is the same as that

2440: described in \S~\ref{Sect:xcorform}, although some important

2441: differences exist. First, the overlap in wavelength between the input and

2442: template spectra at the correlation redshift (the ${\rm lap}$ parameter

2443: discussed in \S~\ref{Sect:lap}) is not taken into account

2444: explicitely; rather it is maximized by including spectral templates

2445: at different redshifts. Second,

2446: XCSAO directly selects the best peak in the correlation

2447: function, rather than looking at the 10 best peaks individually. Last,

2448: the reported correlation redshift is that associated with the

2449: best-match template (i.e., the one with the highest correlation

2450:   height-noise ratio, $r$), rather

2451: than being the median of all redshifts above a certain cutoff

2452: $r_{\rm min}$. It may be that not including the ${\rm lap}$ parameter is less 

2453: important for galaxy redshift determination, due to the narrower

2454: widths of spectral lines in galaxy spectra and the iterative scheme

2455: implemented in XCSAO ensures that an optimal correlation redshift is

2456: computed (D.~Mink 2007, private communication). For supernova spectra, however, the

2457: inclusion of the ${\rm lap}$ parameter is fundamental to obtain reliable

2458: redshifts (see \S~\ref{Sect:zreserr}). The other two differences are

2459: of a less fundamental nature.

2460: 

2461: 

2462: Another algorithm based on the correlation techniques of

2463: \citet{Tonry/Davis:1979} is RUNZ

2464: \citep{Saunders/Cannon/Sutherland:2004}, a program used by members of

2465: the 2dF Galaxy Redshift Survey \citep{Colless/etal:2001} and the 6dF

2466: Galaxy Survey \citep{Jones/etal:2004}. An essential difference with

2467: \snid\ is the scaling of the input spectrum by its inverse-variance

2468: (see \S~\ref{Sect:xcorz}), which leads to improved

2469: cross-correlation redshifts for galaxy spectra. As mentioned

2470: earlier in \S~\ref{Sect:xcorz}, no such improvement is found for

2471: supernova redshifts, since the power spectrum of a typical variance

2472: spectrum (for ground-based observations) peaks at higher wavenumbers

2473: than for SN spectra. One advantage of RUNZ over \snid\ is the

2474: implementation of Gaussian constraints on redshift, a feature that will

2475: be part of a future version of \snid.

2476: 

2477: \subsection{$\chi^2$ Methods}

2478: 

2479: An alternative to cross-correlation techniques involves the

2480: minimization of a $\chi^2$-like quantity at discrete redshift

2481: intervals, to find the best match between an input and template

2482: spectrum. Such techniques do not enable a formal evaluation of the

2483: redshift error as in \snid\ (see \S~\ref{Sect:zerr}). An elaborate

2484: implementation of this approach is ``superfit''

2485: \citep{Howell/etal:2005}, in which an input spectrum is compared to a

2486: combination of a (possibly reddened) template supernova spectrum and a

2487: template galaxy spectrum at a given redshift. A similar program to

2488: superfit is $\mathcal{SN}$-fit (\citealt{Sainton:2004}; see also

2489: discussion in \citealt{Blondin/etal:2005}). The number of free

2490: parameters means that the execution time is 2--3 orders of magnitude greater

2491: than for \snid. For this reason we could not test superfit in the

2492: same way we conducted the simulations presented in the previous

2493: sections.

2494: 

2495: 

2496: The advantage of this technique lies in the evaluation of the

2497: fraction of galaxy light in the input spectrum and its

2498: subsequent subtraction to obtain a ``pure'' supernova

2499: spectrum. 

2500: % Obviously, this fraction is covariant with extinction (and

2501: % especially so for red, early-type hosts) and relies on accurate flux

2502: % calibration of the template spectra (the spectra are normalized, but

2503: % not ``flattened'', as in \snid--- see Fig.~\ref{Fig:lnlambda}). 

2504: Consequently, supernovae can be classified even when the galaxy

2505: contamination fraction is high ($\lesssim 75\%$;

2506: \citealt{Howell/etal:2005}).

2507: Given the impact of galaxy contamination on 

2508: correlation redshifts (Fig.~\ref{Fig:zvar_gf}), we expect more

2509: accurate redshifts when this contamination is removed 

2510: from the input spectrum. However, the typical errors on redshift are

2511: similar to those in \snid\ 

2512: % (and even slightly larger; 

2513: \citep{Hook/etal:2005}. This suggests that the 

2514: redshift accuracy is limited by physical properties of supernovae

2515: (namely, the velocity location of their prominent spectral features)

2516: rather than by differences in algorithm. The accuracy on the

2517: age determination is similar to that in \snid, with a $\sim 3$-day

2518: dispersion about the one-to-one correspondence with the light-curve

2519: age \citep{Hook/etal:2005,Howell/etal:2005}. We note that the

2520: current version of superfit has not been extensively tested for

2521: redshift and age determination and has not been tested at all for

2522: type determination (A.~Howell 2006, private communication).

2523: 

2524: 

2525: A variant of the $\chi^2$ approach implemented in superfit is the

2526: spectral feature aging (SFA) algorithm of \citet{Riess/etal:1997},

2527: used to determine the age of a normal SN~Ia of known

2528: redshift. In SFA, the input spectrum is divided up into several

2529: wavelength intervals (typically eight) and each of these spectral

2530: regions is compared with corresponding ones in a database of template

2531: spectra. The age accuracy is similar to that in \snid but is

2532: largely sensitive to the wavelength intervals of the spectral regions

2533: used to divide up the input spectrum \citep{Foley/etal:2005}. Another

2534: tool, based on the SFA algorithm, plans to extend the age

2535: determination to spectra of all types \citep{Harutyunyan/etal:2005}.

2536: 

2537: 

2538: \subsection{Bayesian Approach to SN Classification}

2539: 

2540: Recently, several authors have presented Bayesian methods to

2541: determine the type of a supernova based on a single-epoch photometric

2542: measurement (potentially in multiple bands;

2543: \citealt{Poznanski/Maoz/Gal-Yam:2006}), or on multiband light curves

2544: \citep{Kuznetsova/Connolly:2007,Kunz/etal:2006}. The motivation behind

2545: these purely photometry-based approaches is the planned

2546: next-generation of wide-field all-sky surveys (such as Pan-STARRS and

2547: LSST), for which many SN~Ia candidates will be too faint for

2548: spectroscopy. However, these techniques are in principle   

2549: applicable to ongoing high-redshift SN~Ia surveys, which are limited

2550: by the amount of available spectroscopy time

2551: \citep{Matheson/etal:2005,Howell/etal:2005}. In fact,

2552: \citet{Tonry/etal:2003} already adopt a Bayesian approach in

2553: fitting high-$z$ SN~Ia light curves with BATM (Bayesian Adapted

2554: Template Match; \citealt{Tonry/etal:2003}).

2555: 

2556: 

2557: These Bayesian-based approaches assign a probability for a supernova

2558: to be of a certain type, based on a set of measurements (e.g., light

2559: curve points in a given photometric band) and given a model--- or

2560: template, that depends on a set of parameters. As pointed out by

2561: \citet{Kuznetsova/Connolly:2007}, when computing this probability it

2562: is generally assumed that the input is indeed a supernova of a known

2563: type, although in principle one could extend the formalism to

2564: incorporate all known astrophysical objects. Moreover, such methods

2565: invoke ``marginalization over type,'' which poses some conceptual

2566: problems, since it assumes the SN classification scheme to be both

2567: complete (i.e. to include all possible SN types) and 

2568: hermetic. Yet, there appears to be a continuum of properties relating

2569: different SN types (e.g., Type Ib and Ic), some supernovae evolve from

2570: one type to another (e.g., Type IIb) and others still seem to defy

2571: classification (such as the ``peculiar SNe~Ia'' SN~2002cx or SN~2005hk).

2572: 

2573: 

2574: The current version of \snid\ does not incorporate such a methodology in

2575: its supernova classification. Nevertheless, the change in the relative

2576: fraction of templates of a given type as a function of the $r{\rm lap}$

2577: quality parameter could be folded in as an extra constraint on the SN

2578: type in a Bayesian framework. 

2579: 

2580: 

2581: 

2582: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2583: %%

2584: %%   Section 7 - Conclusion

2585: %%

2586: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2587: 

2588: \section{Conclusion}\label{Sect:ccl}

2589: 

2590: We have presented an algorithm, based on the correlation techniques of

2591: \citet{Tonry/Davis:1979}, that can be used to

2592: determine the redshift and age of a supernova spectrum and place

2593: constraints on its type. We develop a diagnostic, the $r{\rm lap}$

2594: quality parameter, to quantify the reliability of a given correlation between the

2595: input and a template spectrum. This parameter is simply the

2596: product of the \citet{Tonry/Davis:1979} correlation height-noise ratio ($r$) and the overlap in

2597: rest-frame $\ln \l$ space between the input and template spectrum at the 

2598: correlation redshift (${\rm lap}$). This $r{\rm lap}$ diagnostic has the advantage of

2599: enabling the formal computation of the redshift error, proportional to

2600: $1/(1+r{\rm lap})$. We show, based on simulations, that for $r{\rm lap} 

2601: \gtrsim 5$, the typical error on redshift and age is $\sigma_z \lesssim

2602: 0.01$ and $\sigma_t \lesssim 3$ days, respectively. The former

2603: accuracy on redshift is confirmed through a comparison of correlation

2604: redshifts with host-galaxy redshifts (determined from narrow lines in

2605: the spectrum) out to redshifts $z \lesssim 0.8$. The latter accuracy

2606: on age is confirmed through a comparison of (rest frame) spectral

2607: ages using \snid\ and (observer frame) light curve ages corrected

2608: for the $(1+z)$ time-dilation factor expected in an expanding

2609: universe. The fact that both age estimates agree so well is itself a

2610: verification of the cosmological nature of redshifts, previously

2611: tested with multiepoch SN~Ia spectra

2612: \citep{Riess/etal:1997,Foley/etal:2005}. Furthermore, the success of

2613: \snid\ in determining the redshift and age of the high-redshift SN~Ia

2614: spectra in our sample shows that these are similar to

2615:   low-redshift counterparts.

2616: % they are similar to their low-redshift

2617: % counterparts.

2618: 

2619: 

2620: We present first results of an impartial and effective spectroscopic

2621: classification of supernovae, based on the fraction of correlations

2622: exceeding a certain $r{\rm lap}$ cutoff. We illustrate this through various

2623: examples, three of which are relevant to ongoing SN~Ia searches at

2624: high redshift: we are able to distinguish 1991T-like SNe~Ia from other

2625: SNe~Ia at $z=0.5$;

2626: % , if the input spectrum is within five days from maximum light

2627: we identify SNe~Ib/c as such at both $z=0.3$ and

2628: $z=0.5$. The identification of peculiar SNe~Ia, on the other

2629: hand, proves easier at $z=0.5$ than $z=0.3$, although this

2630: result remains to be verified with more peculiar SN~Ia spectral

2631: templates. In all cases we assume the galaxy contamination fraction,

2632: to which \snid\ is {\it not} sensitive, to be negligible.

2633: %difficult both at $z=0.3$ and $z=0.5$ without a constraint on

2634: % redshift.

2635:  

2636: These examples both illustrate the success and 

2637: limitations of such an automated classification scheme and highlight

2638: the complementarity between spectroscopic and photometric observations

2639: in determining the supernova type. We are currently testing various

2640: combinations of other classification schemes to improve the

2641: classification of all SN types using \snid.

2642: 

2643: % A rather surprising fact is that the loss of all spectral color

2644: % information in the input and template spectra through a {\it

2645: %   pseudo}-continuum removal (\S~\ref{Sect:preproc}) does not affect

2646: % \snid\ in determining the redshift and age of a supernova

2647: % spectrum. Nevertheless, a combination of \snid\ with multi-color light

2648: % curve information would be a powerful tool in better constraining the

2649: % SN type.

2650: 

2651: Supernova discoveries will continue to increase dramatically with the

2652: advent of wide-field imaging telescopes optimized for the detection of

2653: transient events, such as Pan-STARRS and LSST. These experiments

2654: are expected to find tens to hundreds of thousands of new supernovae

2655: each year, few of which will have spectroscopic

2656: confirmation. Thus, most identifications will have to rely solely on

2657: photometric properties, a difficult task in view of the present

2658: difficulty of distinguishing between SN types (and subtypes) with

2659: spectra. It is likely that those experiments will have to rely on a

2660: simpler classification scheme, focusing on the main SN types (Type Ia,

2661: Ib/c, \& II) with little or no distinction between the associated

2662: subtypes.

2663: 

2664: 

2665: Future planned space-based high-redshift SN~Ia surveys within the

2666: NASA/DOE Joint Dark Energy Mission (ADEPT, DESTINY, SNAP) will

2667: incorporate a spectrograph and could benefit from a tool such as

2668: \snid. A secure identification of SNe~Ia in such experiments will

2669: require sufficient rest frame wavelength coverage beyond $\sim 5500$\,\AA,

2670: as the distinction between SNe~Ia and Ic (and between SN~Ia subtypes,

2671: including peculiar events) is otherwise problematic

2672: (Figs.~\ref{Fig:Ia91t}--\ref{Fig:Iapec}).

2673: 

2674: 

2675: The current version of \snid\ will be made available to the

2676: community and we will set up a Web-based interface for

2677: instantaneous supernova typing (including redshift and age

2678: determination). Future versions of \snid\ will include a

2679: wavelength- and age-weighted spectrum overlap parameter, ${\rm lap}(\l,t)$, an explicit 

2680: treatment of the covariance between redshift and age and a Bayesian

2681: approach to type determination, as currently used for photometric

2682: classification of supernovae

2683: \citep{Poznanski/Maoz/Gal-Yam:2006,Kuznetsova/Connolly:2007}.

2684: Moreover, more spectral templates are continuously being included in

2685: the \snid\ database through the CfA Supernova Program (more than 3000

2686: spectra of over 700 supernovae since 1997), which directly

2687: impact the ability of \snid\ to securely identify input spectra. This

2688: further enables comparative studies of SN spectra and quantitative

2689: evaluation of synthetic spectral fits to observations.

2690: 

2691: 

2692: 

2693: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2694: %%

2695: %%   Acknowledgments

2696: %%

2697: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2698: 

2699: \acknowledgments

2700: The authors wish to thank Tom Matheson for use of CfA SN~Ia

2701: spectra ahead of publication, as well as the ESSENCE team for providing

2702: spectra and light curves of the high-redshift SNe~Ia used in

2703: Fig.~\ref{Fig:zgaltlcsnid}. We also thank Weidong Li and Rubina Kotak

2704: for providing spectra of SN~2002cx and SN~2002er, respectively. The

2705: present paper benefited from discussions with the CfA Supernova Group,

2706: and we thank Malcolm Hicken for his work on light-curve data that led

2707: to the inclusion of several extra supernovae in the \snid\

2708: database. We thank the referee andy Howell, for his comments

2709: that led to an improvement of the paper and for making available to

2710: us the spectrum of SNLS-03D3bb \citep{Howell/etal:2006}.

2711: % We appreciate the use of Andy Howell's code ``superfit'', as

2712: % well as fruitful email exchanges with its author and thank him for 

2713: % making available to us the spectrum of SNLS-03D3bb. 

2714: We also thank Robert Kirshner, Bruno Leibundgut, Brian Schmidt and

2715: Kevin Krisciunas for useful comments on the manuscript.

2716: 

2717: S.~B. thanks the Kavli Institute for Theoretical Physics at

2718: the University of California, Santa Barbara for its hospitality during

2719: the program: ``Accretion and Explosion: the Astrophysics of

2720: Degenerate Stars.''

2721: 

2722: This work has been funded in part by the US National Science

2723: Foundation through grants AST 04-43378, AST 05-7475, AST

2724: 06-06772 and PHY 05-51164. This research has made use of the CfA

2725: Supernova Archive, which is funded in part by the National Science

2726: Foundation through grant AST 06-06772.

2727: 

2728: 

2729: 

2730: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2731: %%

2732: %%   Bibliography

2733: %%

2734: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2735: 

2736: \begin{thebibliography}{}

2737: 

2738: \bibitem[Anupama et al.(2005)]{Anupama/Sahu/Jose:2005} Anupama, G.~C., Sahu, 

2739: D.~K., \& Jose, J.\ 2005, \aap, 429, 667 

2740: \bibitem[Astier et al.(2006)]{Astier/etal:2006} Astier, P., et al.\ 

2741: 2006, \aap, 447, 31 

2742: \bibitem[Barbon et al.(1995)]{Barbon/etal:1995} Barbon, R., Benetti, S., 

2743: Cappellaro, E., Patat, F., Turatto, M., \& Iijima, T.\ 1995, \aaps, 110, 

2744: 513

2745: \bibitem[Baron et al.(1996)]{Baron/etal:1996} Baron, E.,

2746:   Hauschildt, P.~H., Nugent, P., \& Branch, D.\ 1996, \mnras, 283, 297 

2747: \bibitem[Baron et al.(2000)]{Baron/etal:2000} Baron, E., et al.\ 2000, 

2748: \apj, 545, 444 

2749: \bibitem[Barris et al.(2004)]{Barris/etal:2004} Barris, B.~J., et al.\

2750: 2004, \apj, 602, 571 

2751: \bibitem[Barris \& Tonry(2004)]{Barris/Tonry:2004} Barris, B.~J., \& 

2752: Tonry, J.~L.\ 2004, \apjl, 613, L21 

2753: \bibitem[Benetti et al.(2004)]{Benetti/etal:2004a} Benetti, S., et al.\ 

2754: 2004, \mnras, 348, 261 

2755: \bibitem[Benetti et al.(2004)]{Benetti/etal:2004b} Benetti, S.,

2756:   Elias-Rosa, N., Blanc, G., Navasardyan, H., Turatto, M., Zampieri,

2757:   L., Cappellaro, E., \& Pedani, M.\ 2004, \iaucirc, 8312, 3  

2758: \bibitem[Benetti et al.(2005)]{Benetti/etal:2005} Benetti, S., et al.\ 

2759: 2005, \apj, 623, 1011 

2760: \bibitem[Blondin et al.(2005)]{Blondin/etal:2005} Blondin, S., Walsh, 

2761: J.~R., Leibundgut, B., \& Sainton, G.\ 2005, \aap, 431, 757 

2762: \bibitem[Blondin et al.(2006a)]{Blondin/etal:2006a} Blondin, S., et al.\ 

2763: 2006, \aj, 131, 1648 

2764: \bibitem[Blondin et al.(2006b)]{Blondin/etal:2006b} Blondin, S., Modjaz, 

2765: M., Kirshner, R., Challis, P., \& Peters, W.\ 2006, Central Bureau 

2766: Electronic Telegrams, 488, 1 (2006).~ Edited by Green, D.~W.~E., 488,

2767: 1

2768: \bibitem[Blondin et al.(2006c)]{Blondin/etal:2006c} Blondin, S., Modjaz, 

2769: M., Kirshner, R., Challis, P., Matheson, T., \& Antoniou, V.\ 2006, Central 

2770: Bureau Electronic Telegrams, 461, 1 (2006).~ Edited by Green, D.~W.~E., 

2771: 461, 1 

2772: \bibitem[Branch et al.(2002)]{Branch/etal:2002} Branch, D., et al.\ 

2773: 2002, \apj, 566, 1005 

2774: \bibitem[Branch et al.(2003)]{Branch/etal:2003} Branch, D., et al.\ 

2775: 2003, \aj, 126, 1489 

2776: \bibitem[Branch et al.(2004)]{Branch/etal:2004} Branch, D., Baron, E., 

2777: Thomas, R.~C., Kasen, D., Li, W., \& Filippenko, A.~V.\ 2004, \pasp, 116, 

2778: 903 

2779: \bibitem[Branch et al.(2006)]{Branch/etal:2006} Branch, D., Jeffery, 

2780: D.~J., Young, T.~R., \& Baron, E.\ 2006, \pasp, 118, 791 

2781: \bibitem[Brown et al.(2007)]{Brown/etal:2007} Brown, P.~J., et al.\ 

2782: 2007, \apj, 659, 1488 

2783: \bibitem[Cappellaro et al.(1997)]{Cappellaro/etal:1997} Cappellaro, E., 

2784: Turatto, M., Tsvetkov, D.~Y., Bartunov, O.~S., Pollas, C., Evans, R., \& 

2785: Hamuy, M.\ 1997, \aap, 322, 431 

2786: \bibitem[Chevalier(2006)]{Chevalier:2006} Chevalier, R.~A.\ 2006,

2787:   ArXiv Astrophysics e-prints, arXiv:astro-ph/0612734  

2788: \bibitem[Clocchiatti et al.(1996)]{Clocchiatti/etal:1996} Clocchiatti, A., et 

2789: al.\ 1996, \aj, 111, 1286 

2790: \bibitem[Clocchiatti et al.(2000)]{Clocchiatti/etal:2000} Clocchiatti, A., et 

2791: al.\ 2000, \apj, 529, 661 

2792: \bibitem[Clocchiatti et al.(2001)]{Clocchiatti/etal:2001} Clocchiatti, A., et 

2793: al.\ 2001, \apj, 553, 886 

2794: \bibitem[Colless et al.(2001)]{Colless/etal:2001} Colless, M., et al.\ 

2795: 2001, \mnras, 328, 1039 

2796: \bibitem[Crotts \& Consortium(2006)]{Crotts/etal:2006} Crotts,

2797:   A.~P., \& Consortium, A.\ 2006, American Astronomical Society

2798:   Meeting Abstracts, 209, \#99.05  

2799: \bibitem[Dessart et al.(2007)]{Dessart/etal:2007} Dessart, L., et al.\ 

2800: 2007, \apj, submitted

2801: \bibitem[Elmhamdi et al.(2004)]{Elmhamdi/etal:2004} Elmhamdi, A., 

2802: Danziger, I.~J., Cappellaro, E., Della Valle, M., Gouiffes, C., Phillips, 

2803: M.~M., \& Turatto, M.\ 2004, \aap, 426, 963 

2804: \bibitem[Elmhamdi et al.(2006)]{Elmhamdi/etal:2006} Elmhamdi, A., 

2805: Danziger, I.~J., Branch, D., Leibundgut, B., Baron, E., \& Kirshner, R.~P.\ 

2806: 2006, \aap, 450, 305

2807: \bibitem[Fassia et al.(2001)]{Fassia/etal:2001} Fassia, A., et al.\ 

2808: 2001, \mnras, 325, 907 

2809: \bibitem[Filippenko et al.(1992)]{Filippenko/etal:1992} Filippenko,

2810:   A.~V., et al.\ 1992, \apjl, 384, L15 

2811: \bibitem[Filippenko(1997)]{Filippenko:1997} Filippenko, A.~V.\ 1997,  

2812: \araa, 35, 309 

2813: \bibitem[Foley et al.(2005)]{Foley/etal:2005} Foley, R.~J., Filippenko, 

2814: A.~V., Leonard, D.~C., Riess, A.~G., Nugent, P., \& Perlmutter, S.\ 2005, 

2815: \apjl, 626, L11 

2816: \bibitem[Fransson et al.(2005)]{Fransson/etal:2005} Fransson, C., et al.\ 

2817: 2005, \apj, 622, 991

2818: \bibitem[Frieman(2004)]{Frieman:2004}

2819:   Frieman, J., \& Dark Energy Survey 2004, Bulletin of the American

2820:   Astronomical Society, 36, 1462  

2821: \bibitem[Gal-Yam et al.(2002)]{Gal-Yam/Ofek/Shemmer:2002} Gal-Yam, A., Ofek, 

2822: E.~O., \& Shemmer, O.\ 2002, \mnras, 332, L73

2823: \bibitem[Garnavich et al.(2004)]{Garnavich/etal:2004} Garnavich, P.~M., et 

2824: al.\ 2004, \apj, 613, 1120 

2825: \bibitem[Geller et al.(2005)]{Geller/etal:2005} Geller, M.~J., 

2826: Dell'Antonio, I.~P., Kurtz, M.~J., Ramella, M., Fabricant, D.~G., Caldwell, 

2827: N., Tyson, J.~A., \& Wittman, D.\ 2005, \apjl, 635, L125 

2828: \bibitem[Goldhaber et al.(2001)]{Goldhaber/etal:2001} Goldhaber, G., et 

2829: al.\ 2001, \apj, 558, 359 

2830: \bibitem[G{\'o}mez et al.(1996)]{Gomez/Lopez/Sanchez:1996} G{\'o}mez, G., L{\'o}pez, R., \& 

2831: S{\'a}nchez, F.\ 1996, \aj, 112, 2094 

2832: \bibitem[G{\'o}mez \& L{\'o}pez(2000)]{Gomez/Lopez:2000} G{\'o}mez, G., 

2833: \& L{\'o}pez, R.\ 2000, \aj, 120, 367 

2834: \bibitem[Hamuy et al.(1996)]{Hamuy/etal:1996} Hamuy, M., et al.\

2835:   1996, \aj, 112, 2408  

2836: \bibitem[Hamuy et al.(2001)]{Hamuy/etal:2001} Hamuy, M., et al.\ 2001, 

2837: \apj, 558, 615 

2838: \bibitem[Hamuy et al.(2002)]{Hamuy/etal:2002} Hamuy, M., et al.\ 2002, 

2839: \aj, 124, 417 

2840: \bibitem[Harutyunyan et al.(2005)]{Harutyunyan/etal:2005} Harutyunyan, A., 

2841: Benetti, S., Cappellaro, E., \& Turatto, M.\ 2005, ASP Conf.~Ser.~342: 

2842: 1604-2004: Supernovae as Cosmological Lighthouses, 342, 258 

2843: \bibitem[H\"oflich(1995)]{Hoeflich:1995} H\"oflich, P.\ 1995, \apj,

2844:   443, 89 

2845: \bibitem[Homeier(2005)]{Homeier:2005} Homeier, N.~L.\ 2005, \apj, 

2846: 620, 12 

2847: \bibitem[Hook et al.(2005)]{Hook/etal:2005} Hook, I.~M., et al.\ 2005, 

2848: \aj, 130, 2788 

2849: \bibitem[Howell et al.(2005)]{Howell/etal:2005} Howell, D.~A., et al.\ 

2850: 2005, \apj, 634, 1190 

2851: \bibitem[Howell et al.(2006)]{Howell/etal:2006} Howell, D.~A., et al.\ 

2852: 2006, \nat, 443, 308 

2853: \bibitem[Iwamoto et al.(2000)]{Iwamoto/etal:2000} Iwamoto, K., et al.\ 

2854: 2000, \apj, 534, 660

2855: \bibitem[James et al.(2006)]{James/etal:2006} James, J.~B., Davis, 

2856: T.~M., Schmidt, B.~P., \& Kim, A.~G.\ 2006, \mnras, 370, 933 

2857: \bibitem[Jeffery et al.(1992)]{Jeffery/etal:1992} Jeffery, D.~J., 

2858: Leibundgut, B., Kirshner, R.~P., Benetti, S., Branch, D., \& Sonneborn, G.\ 

2859: 1992, \apj, 397, 304 

2860: \bibitem[Jeffery et al.(1994)]{Jeffery/etal:1994} Jeffery, D.~J., et al.\ 

2861: 1994, \apjl, 421, L27 

2862: \bibitem[Jeffery et al.(2006)]{Jeffery/etal:2006} Jeffery, D.~J., 

2863: Ketchum, W., Branch, D., Baron, E., Elmhamdi, A., \& Danziger, I.~J.\ 2006, 

2864: ArXiv Astrophysics e-prints, arXiv:astro-ph/0607084 

2865: \bibitem[Jha et al.(1999)]{Jha/etal:1999} Jha, S., et al.\ 1999, 

2866: \apjs, 125, 73 

2867: \bibitem[Jha et al.(2006a)]{Jha/etal:2006a} Jha, S., et al.\ 2006,

2868:   \aj, 131, 527

2869: \bibitem[Jha et al.(2006b)]{Jha/etal:2006b} Jha, S., Branch, D., 

2870: Chornock, R., Foley, R.~J., Li, W., Swift, B.~J., Casebeer, D., \& 

2871: Filippenko, A.~V.\ 2006, \aj, 132, 189 

2872: \bibitem[Jones et al.(2004)]{Jones/etal:2004} Jones, D.~H., et al.\ 

2873: 2004, \mnras, 355, 747 

2874: \bibitem[Kaiser et al.(2002)]{Kaiser/etal:2002} Kaiser, N., et al.\

2875:   2002, \procspie, 4836, 154  

2876: \bibitem[Kinney et al.(1996)]{Kinney/etal:1996} Kinney, A.~L., Calzetti, 

2877: D., Bohlin, R.~C., McQuade, K., Storchi-Bergmann, T., \& Schmitt, H.~R.\ 

2878: 1996, \apj, 467, 38 

2879: \bibitem[Kinugasa \& Yamaoka(2006a)]{Kinugasa/Yamaoka:2006a} Kinugasa, K., \& 

2880: Yamaoka, H.\ 2006, Central Bureau Electronic Telegrams, 454, 1 (2006).~ 

2881: Edited by Green, D.~W.~E., 454, 1 

2882: \bibitem[Kinugasa \& Yamaoka(2006b)]{Kinugasa/Yamaoka:2006b} Kinugasa, K., \& 

2883: Yamaoka, H.\ 2006, Central Bureau Electronic Telegrams, 475, 1 (2006).~ 

2884: Edited by Green, D.~W.~E., 475, 1 

2885: \bibitem[Kirshner et al.(1993)]{Kirshner/etal:1993} Kirshner, R.~P., et 

2886: al.\ 1993, \apj, 415, 589 

2887: \bibitem[Knop et al.(2003)]{Knop/etal:2003} Knop, R.~A., et al.\ 2003, 

2888: \apj, 598, 102 

2889: \bibitem[Kotak et al.(2005)]{Kotak/etal:2005} Kotak, R., et al.\ 2005, 

2890: \aap, 436, 1021

2891: \bibitem[Kulkarni et al.(1998)]{Kulkarni/etal:1998} Kulkarni, S.~R., et 

2892: al.\ 1998, \nat, 395, 663 

2893: \bibitem[Kunz et al.(2006)]{Kunz/etal:2006} Kunz, M., Bassett, B.~A.,

2894: \& Hlozek, R.\ 2006, ArXiv Astrophysics e-prints, arXiv:astro-ph/0611004 

2895: \bibitem[Kurtz \& Mink(1998)]{Kurtz/Mink:1998} Kurtz, M.~J., \& Mink, 

2896: D.~J.\ 1998, \pasp, 110, 934 

2897: \bibitem[Kuznetsova \& Connolly(2007)]{Kuznetsova/Connolly:2007} 

2898: Kuznetsova, N.~V., \& Connolly, B.~M.\ 2007, \apj, 659, 530 

2899: \bibitem[Leibundgut et al.(1991)]{Leibundgut/etal:1991} Leibundgut, B., 

2900: Kirshner, R.~P., Filippenko, A.~V., Shields, J.~C., Foltz, C.~B., Phillips, 

2901: M.~M., \& Sonneborn, G.\ 1991, \apjl, 371, L23 

2902: \bibitem[Leibundgut et al.(1993)]{Leibundgut/etal:1993} Leibundgut, B., et 

2903: al.\ 1993, \aj, 105, 301 

2904: \bibitem[Leibundgut et al.(1996)]{Leibundgut/etal:1996} Leibundgut, B.,

2905:   et al.\ 1996, \apjl, 466, L21  

2906: \bibitem[Leonard et al.(2002a)]{Leonard/etal:2002a} Leonard, D.~C., et al.\ 

2907: 2002, \pasp, 114, 35 

2908: \bibitem[Leonard et al.(2002b)]{Leonard/etal:2002b} Leonard, D.~C., et al.\ 

2909: 2002, \aj, 124, 2490 

2910: \bibitem[Lentz et al.(2001)]{Lentz/etal:2001} Lentz, E.~J., et al.\ 

2911: 2001, \apj, 547, 406

2912: \bibitem[Li et al.(1999)]{Li/etal:1999} Li, W.~D., et al.\ 1999, 

2913: \aj, 117, 2709

2914: \bibitem[Li et al.(2001a)]{Li/etal:2001a} Li, W., et al.\ 2001, \pasp, 

2915: 113, 1178 

2916: \bibitem[Li et al.(2001b)]{Li/etal:2001b} Li, W., Filippenko,

2917:   A.~V., Treffers, R.~R., Riess, A.~G., Hu, J., \& Qiu, Y.\ 2001,

2918:   \apj, 546, 734 

2919: \bibitem[Li et al.(2003)]{Li/etal:2003} Li, W., et al.\ 2003, \pasp, 

2920: 115, 453

2921: \bibitem[Matheson et al.(2000)]{Matheson/etal:2000} Matheson, T., et al.\ 

2922: 2000, \aj, 120, 1487 

2923: \bibitem[Matheson et al.(2001)]{Matheson/etal:2001} Matheson, T., 

2924: Filippenko, A.~V., Li, W., Leonard, D.~C., \& Shields, J.~C.\ 2001, \aj, 

2925: 121, 1648 

2926: \bibitem[Matheson et al.(2003)]{Matheson/etal:2003} Matheson, T., et

2927:   al.\ 2003, \apj, 599, 394 

2928: \bibitem[Matheson et al.(2005)]{Matheson/etal:2005} Matheson, T., et al.\ 

2929: 2005, \aj, 129, 2352 

2930: \bibitem[Matheson et al.(2007)]{Matheson/etal:2007} Matheson, T., et al.\ 

2931: 2007, submitted

2932: \bibitem[Mazzali et al.(1993)]{Mazzali/etal:1993} Mazzali, P.~A., Lucy, 

2933: L.~B., Danziger, I.~J., Gouiffes, C., Cappellaro, E., \& Turatto, M.\ 1993, 

2934: \aap, 269, 423 

2935: \bibitem[Mazzali et al.(1995)]{Mazzali/etal:1995} Mazzali, P.~A., 

2936: Danziger, I.~J., \& Turatto, M.\ 1995, \aap, 297, 509 

2937: \bibitem[McElroy(1995)]{McElroy:1995} McElroy, D.~B.\ 1995, \apjs, 

2938: 100, 105 

2939: \bibitem[Miknaitis et al.(2007)]{Miknaitis/etal:2007} Miknaitis, G., et al.\ 

2940:  2007, \apj, submitted

2941: \bibitem[Millard et al.(1999)]{Millard/etal:1999} Millard, J., et al.\ 

2942: 1999, \apj, 527, 746

2943: \bibitem[Modjaz et al.(2006)]{Modjaz/etal:2006} Modjaz, M., et al.\ 

2944: 2006, \apjl, 645, L21 

2945: \bibitem[Nomoto et al.(1993)]{Nomoto/etal:1993} Nomoto, K., Suzuki, T., 

2946: Shigeyama, T., Kumagai, S., Yamaoka, H., \& Saio, H.\ 1993, \nat, 364, 507 

2947: \bibitem[Nugent et al.(2002)]{Nugent/Kim/Perlmutter:2002} Nugent, P., Kim,

2948:   A., \& Perlmutter, S.\ 2002, \pasp, 114, 803

2949: \bibitem[Parrent et al.(2007)]{Parrent/etal:2007} Parrent, J., et

2950:   al.\ 2007, \pasp, 119, 135 

2951: \bibitem[Patat et al.(1996)]{Patat/etal:1996} Patat, F., Benetti, S., 

2952: Cappellaro, E., Danziger, I.~J., della Valle, M., Mazzali, P.~A., \& 

2953: Turatto, M.\ 1996, \mnras, 278, 111 

2954: \bibitem[Patat et al.(2001)]{Patat/etal:2001} Patat, F., et al.\ 2001, 

2955: \apj, 555, 900 

2956: \bibitem[Perlmutter et al.(1999)]{Perlmutter/etal:1999} Perlmutter,

2957: S., et al.\ 1999, \apj, 517, 565 

2958: \bibitem[Phillips et al.(1988)]{Phillips/etal:1988} Phillips, M.~M., 

2959: Heathcote, S.~R., Hamuy, M., \& Navarrete, M.\ 1988, \aj, 95, 1087

2960: \bibitem[Phillips et al.(1990)]{Phillips/etal:1990} Phillips, M.~M., 

2961: Hamuy, M., Heathcote, S.~R., Suntzeff, N.~B., \& Kirhakos, S.\ 1990, \aj, 

2962: 99, 1133 

2963: \bibitem[Phillips et al.(1992)]{Phillips/etal:1992} Phillips, M.~M.,

2964:   Wells, L.~A., Suntzeff, N.~B., Hamuy, M., Leibundgut, B.,

2965:   Kirshner, R.~P., \& Foltz, C.~B.\ 1992, \aj, 103, 1632  

2966: \bibitem[Phillips(1993)]{Phillips:1993} Phillips, M.~M.\ 1993, \apjl, 

2967: 413, L105 

2968: \bibitem[Phillips et al.(2007)]{Phillips/etal:2007} Phillips, M.~M., et 

2969: al.\ 2007, \pasp, 119, 360 

2970: \bibitem[Pinto \& Eastman(2000)]{Pinto/Eastman:2000} Pinto, P.~A., \&

2971:   Eastman, R.~G.\ 2000, \apj, 530, 757  

2972: \bibitem[Poznanski, Maoz, \& Gal-Yam(2006)]{Poznanski/Maoz/Gal-Yam:2006} 

2973: Poznanski, D., Maoz, D., \& Gal-Yam, A.\ 2006, ArXiv Astrophysics

2974: e-prints, arXiv:astro-ph/0610129  

2975: \bibitem[Richardson et al.(2006)]{Richardson/etal:2006} Richardson, D., 

2976: Branch, D., \& Baron, E.\ 2006, \aj, 131, 2233 

2977: \bibitem[Riess et al.(1997)]{Riess/etal:1997} Riess, A.~G., et al.\ 

2978: 1997, \aj, 114, 722 

2979: \bibitem[Riess et al.(1998a)]{Riess/etal:1998a} Riess, A.~G., et al.\

2980: 1998, \aj, 116, 1009  

2981: \bibitem[Riess et al.(1998b)]{Riess/etal:1998b} Riess, A.~G.,

2982:   Nugent, P., Filippenko, A.~V., Kirshner, R.~P., \& Perlmutter, S.\

2983:   1998, \apj, 504, 935  

2984: \bibitem[Riess et al.(1999)]{Riess/etal:1999} Riess, A.~G., et

2985:   al.\ 1999, \aj, 117, 707 

2986: \bibitem[Riess et al.(2004)]{Riess/etal:2004} Riess, A.~G., et al.\ 

2987: 2004, \apj, 607, 665 

2988: \bibitem[Ruiz-Lapuente et al.(1992)]{Ruiz-Lapuente/etal:1992}

2989:   Ruiz-Lapuente, P., Cappellaro, E., Turatto, M., Gouiffes, C.,

2990:   Danziger, I.~J., della Valle, M., \& Lucy, L.~B.\ 1992, \apjl,

2991:   387, L33 

2992: \bibitem[Rust(1974)]{Rust:1974} Rust, B.~W.\ 1974, Ph.D.~Thesis, 

2993: Oak Ridge National Lab., TN.

2994: \bibitem[Rybicki \& Press(1995)]{Rybicki/Press:1995} Rybicki, G.~B., \& 

2995: Press, W.~H.\ 1995, Physical Review Letters, 74, 1060 

2996: \bibitem[Sainton(2004)]{Sainton:2004} Sainton, G.\ 2004, Ph.D. thesis, 

2997: Universit{\'e} Lyon I, UCBL 131-2004 

2998: \bibitem[Salvo et al.(2001)]{Salvo/etal:2001} Salvo, M.~E., Cappellaro, 

2999: E., Mazzali, P.~A., Benetti, S., Danziger, I.~J., Patat, F., \& Turatto, 

3000: M.\ 2001, \mnras, 321, 254 

3001: \bibitem[Saunders et al.(2004)]{Saunders/Cannon/Sutherland:2004}

3002:   Saunders, W., Cannon, R., \& Sutherland, W.~J.\ 2004, AAO

3003:   Newsletter, December 2004, pp.16--17

3004: \bibitem[Schmidt et al.(1994)]{Schmidt/etal:1994} Schmidt, B.~P., 

3005: Kirshner, R.~P., Leibundgut, B., Wells, L.~A., Porter, A.~C., 

3006: Ruiz-Lapuente, P., Challis, P., \& Filippenko, A.~V.\ 1994, \apjl, 434, L19 

3007: \bibitem[Schmidt et al.(2005)]{Schmidt/etal:2005} Schmidt, B.~P.,

3008:   Keller, S.~C., Francis, P.~J., \& Bessell, M.~S.\ 2005, Bulletin

3009:   of the American Astronomical Society, 37, 457  

3010: \bibitem[Stanishev et al.(2007)]{Stanishev/etal:2007} Stanishev, V., et 

3011: al.\ 2007, \aap, 469, 645 

3012: \bibitem[Taubenberger et al.(2006)]{Taubenberger/etal:2006} Taubenberger, S., 

3013: et al.\ 2006, \mnras, 371, 1459 

3014: \bibitem[Tominaga et al.(2005)]{Tominaga/etal:2005} Tominaga, N., et al.\ 

3015: 2005, \apjl, 633, L97

3016: \bibitem[Tonry \& Davis(1979)]{Tonry/Davis:1979} Tonry, J., \& Davis,

3017:   M.\ 1979, \aj, 84, 1511

3018: \bibitem[Tonry et al.(2003)]{Tonry/etal:2003} Tonry, J.~L., et al.\ 

3019: 2003, \apj, 594, 1 

3020: \bibitem[Turatto et al.(1996)]{Turatto/etal:1996} Turatto, M., Benetti, 

3021: S., Cappellaro, E., Danziger, I.~J., della Valle, M., Gouiffes, C., 

3022: Mazzali, P.~A., \& Patat, F.\ 1996, \mnras, 283, 1 

3023: \bibitem[Turatto et al.(1998)]{Turatto/etal:1998} Turatto, M., Piemonte, 

3024: A., Benetti, S., Cappellaro, E., Mazzali, P.~A., Danziger, I.~J., \& Patat, 

3025: F.\ 1998, \aj, 116, 2431

3026: \bibitem[Tyson \& Angel(2001)]{Tyson/Angel:2001} Tyson, A., \&

3027:   Angel, R.\ 2001, ASP Conf.~Ser.~232: The New Era of Wide Field

3028:   Astronomy, 232, 347  

3029: \bibitem[Valentini et al.(2003)]{Valentini/etal:2003} Valentini, G., et 

3030: al.\ 2003, \apj, 595, 779 

3031: \bibitem[Wade \& Horne(1988)]{Wade/Horne:1988} Wade, R.~A., \& Horne, 

3032: K.\ 1988, \apj, 324, 411 

3033: \bibitem[Wang(2007)]{Wang:2007} Wang, Y.\ 2007, \apjl, 654, L123 

3034: \bibitem[Wells et al.(1994)]{Wells/etal:1994} Wells, L.~A., et al.\ 

3035: 1994, \aj, 108, 2233 

3036: \bibitem[Wilson(1939)]{Wilson:1939} Wilson, O.~C.\ 1939, \apj,

3037:   90, 634  

3038: \bibitem[Wood-Vasey et al.(2007)]{Wood-Vasey/etal:2007} Wood-Vasey, W.~M., et al.\ 

3039:  2007, \apj, submitted

3040: \bibitem[Woosley et al.(1987)]{Woosley/etal:1987} Woosley, S.~E., Pinto, 

3041: P.~A., Martin, P.~G., \& Weaver, T.~A.\ 1987, \apj, 318, 664 

3042: \bibitem[Woosley et al.(1993)]{Woosley/Langer/Weaver:1993} Woosley, S.~E., Langer, 

3043: N., \& Weaver, T.~A.\ 1993, \apj, 411, 823 

3044: \bibitem[Woosley et al.(1995)]{Woosley/Langer/Weaver:1995} Woosley, S.~E., Langer, 

3045: N., \& Weaver, T.~A.\ 1995, \apj, 448, 315 

3046: 

3047: \end{thebibliography}

3048: 

3049: %%% Table: phase distribution of Snid templates

3050: \clearpage

3051: \LongTables

3052: \begin{deluxetable*}{llll}

3053: \tabletypesize{\scriptsize}

3054: \tablenum{1}

3055: \tablewidth{0pt}

3056: \tablecaption{\snid\ Supernova Database\label{Table:sndb}}

3057: \tablehead{

3058: \colhead{IAU Name} & 

3059: \colhead{Subtype} & 

3060: \colhead{Ages}&

3061: \colhead{Ref.} \\

3062: \colhead{(1)} & 

3063: \colhead{(2)} & 

3064: \colhead{(3)} & 

3065: \colhead{(4)}

3066: }

3067: \startdata

3068: 1981B  & Ia-norm  & 0,17,20,24                                                                                                        & 1              \\

3069: 1986G  & Ia-91bg  & $-$3,$-$2,0--2,30+(4)                                                                                             & 1              \\ %46,50,55,57

3070: 1987A  & II-pec   & 2,4--9,11--27,31--39,40+(69)                                                                                      & 2,3            \\ %40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,56,58,60,62,66,68,69,71,74,76,78,80,82,86,86,88,93,94,96,98,102,103,105,108,110,112,114,116,118,122,124,126,129,130,131,198,289,340,340,345,362,376,400,408,423,471,498,514,526,589,713,719,786,804

3071: 1989B  & Ia-norm  & $-$6,$-$1,4,6,8,10,12(2),13,14,16--25,30+(3)                                                                      & 1              \\ %31,37,49

3072: 1990B  & Ic-norm  & 5,6(2),7,9,10,15,28(2),30+(6)                                                                                     & 4,5,CfA        \\ %31,41,47,58,88,116

3073: 1990I  & Ib-norm  & 11,19,30+(6)                                                                                                      & 6              \\ %40,41,51,65,88,235

3074: 1990N  & Ia-norm  & $-$13(2),$-$6,3,5,15,18,30+(5)                                                                                    & 7,8            \\ %39,213,245,246,310

3075: 1990O  & Ia-norm  & $-$[7--5],0,19,20                                                                                                 & CfA            \\

3076: 1991M  & Ia-norm  & 25,26,30+(1)                                                                                                      & CfA            \\ %112

3077: 1991T  & Ia-91T   & $-$12,$-$10,$-$9,$-$[7--5],0,19,30+(3)                                                                            & 9--11          \\ %47,75,260

3078: 1991bg & Ia-91bg  & 2,3(3),15,16,19(2),20,26,30+(16)                                                                                  & 12,14          \\ %30,33,34,34,46,47,51,53,54,55,88,113,117,117,143,203

3079: 1992A  & Ia-norm  & $-$5(2),$-$1,0,2,3,6(2),7,9(2),12(2),16,17,24,28                                                                  & 15             \\

3080: 1992H{\tablenotemark{a}} & IIP & 16,29,40+(8)                                                                                         & 16,17          \\ %49,108,118,135,141,146,270,315

3081: 1992ar & Ic-norm  & 3                                                                                                                 & CfA            \\

3082: 1993J  & IIb      & 3,4(2),5,11,16,17,18,19(3),22,24(2),25(2),26--28,32--34,38,40+(51)                                                & 18--21         \\ %41,45,50,51,53,56,56,57,57,58,59,60,62,83,85,89,91,93,109,116,118,123,139,142,167,171,172,182,207,209,226,233,235,238,257,266,286,298,300,315,347,348,363,369,387,413,433,442,473,523,553

3083: 1993ac & Ia-norm  & 7                                                                                                                 & CfA            \\

3084: 1994D  & Ia-norm  & $-$11(2),$-$10(2),$-$9,$-$8,$-$6,$-$5(2),$-$4(2),$-$3,$-$2,0,2,3,10--12,13(2),14,15(3),16,17(2),19,24,26,30+(11)  & 22,23,CfA      \\ %40,43,46,48,50,50,54,55,57,59,83

3085: 1994I  & Ic-norm  & $-$6(2),$-$4,$-$3,1,2(2),3,21--24,26,30+(5)                                                                       & 24             \\ %30,36,38,40,64

3086: 1994M  & Ia-norm  & 3--5,8,13,14                                                                                                      & CfA            \\

3087: 1994Q  & Ia-norm  & 19                                                                                                                & CfA            \\

3088: 1994S  & Ia-norm  & $-$3(2),1                                                                                                         & CfA            \\

3089: 1994T  & Ia-norm  & 1                                                                                                                 & CfA            \\

3090: 1994ae & Ia-norm  & 1,2,3(2),4,6,9(2)10,11,30+(7)                                                                                     & CfA            \\ %30,36,40,70,88,124,153

3091: 1995D  & Ia-norm  & 4,6,8,10,11,14,16,30+(3)                                                                                          & CfA            \\ %38,43,70

3092: 1995E  & Ia-norm  & $-$2,0,2,7,10,30+(1)                                                                                              & CfA            \\ %33

3093: 1995ac & Ia-norm  & 24                                                                                                                & CfA            \\

3094: 1995al & Ia-norm  & 17,25                                                                                                             & CfA            \\

3095: 1995bd & Ia-norm  & 11,21,30+(2)                                                                                                      & CfA            \\ %37,72

3096: 1996C  & Ia-norm  & 8                                                                                                                 & CfA            \\

3097: 1996X  & Ia-norm  & $-$3,$-$2,$-$1(2),0,1(2),2(2),3,5--7,8(2),9,13,21,23,25,30+(1)                                                    & 25,CfA         \\ %32

3098: 1997br & Ia-91T   & $-$9,$-$8,$-$7(2),$-$6(2),$-$4,8,9,12,17,18,21,24,30+(6)                                                          & 26,CfA         \\ %38,42,49,51,66,72

3099: 1997cn & Ia-91bg  & 4,29,30+(1)                                                                                                       & 27,CfA         \\ %43

3100: 1997do & Ia-norm  & $-$11,$-$10,$-$7,$-$6,9,11,12,13,15,16,20,21                                                                      & 28             \\

3101: 1997dt & Ia-norm  & $-$[10--7],$-$4,1,3                                                                                               & 28             \\

3102: 1997ef & Ic-hyper & $-$14,$-$12,$-$[11--9],$-$6,$-$5(2),$-$4,7,13,14,16,17,19,22,24,27,30+(4)                                         & 29             \\ %40,44,75,80

3103: 1998S  & IIn      & 5,6,17,19,20(2),21,28,30--32,34,40+(44)                                                                           & 30--32         \\ %49,50,58,59,62,64,65,75,77,79,89,91,93,94,100,108,112,114,117,120,124,137,139,146,147,259,260,266,275,287,294,296,319,324,329,330,343,357,374,386,402,409,432,441

3104: 1998V  & Ia-norm  & 1--3,13,14,15,30+(3)                                                                                              & 28             \\ %31,41,44

3105: 1998ab & Ia-91T   & $-$7,7,8,18,19,20,21,22,23,30+(3)                                                                                 & 28             \\ %34,36,46

3106: 1998aq & Ia-norm  & $-$9,$-$8,0--7,19,21,30+(15)                                                                                      & 28,33          \\ %31,32,34,36,51,55,58,60,63,66,79,82,91,230,240

3107: 1998bp & Ia-91bg  & $-$2,$-$1,0--2,13,15,25,26,28,30+(1)                                                                              & 28             \\ %30

3108: 1998bu & Ia-norm  & $-$[3--1],1,9--14,28,29,30+(21)                                                                                   & 28,34          \\ %30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,57,179,190,208,217,243

3109: 1998bw & Ic-hyper & 8,9,12--14,16,18,19,21,24,26--28,30+(8)                                                                           & 35             \\ %34,37,43,59,66,87,139,213

3110: 1998de & Ia-91bg  & $-$[7--5],$-$3,$-$2,0,3                                                                                           & 28             \\

3111: 1998dh & Ia-norm  & $-$[9--7],$-$5,$-$3,0,30+(4)                                                                                      & 28             \\ %37,41,44,46

3112: 1998dk & Ia-norm  & 10,11,13,16,18,21,23,30+(2)                                                                                       & 28             \\ %30,45

3113: 1998dm & Ia-norm  & 5,6,8,11,13,16,18,25,30+(2)                                                                                       & 28             \\ %40,48

3114: 1998dt & Ib-norm  & 0,1,4,7,11,12,17                                                                                                  & CfA            \\

3115: 1998ec & Ia-norm  & $-$2,$-$1,13,21,27,30+(1)                                                                                         & 28             \\ %40

3116: 1998eg & Ia-norm  & 0,5,18,20,23                                                                                                      & 28             \\

3117: 1998es & Ia-91T   & $-$[10--1],1--3,16,18,19,20,24,26,30+(7)                                                                          & 28             \\ %30,45,52,57,74,81,90

3118: 1999X  & Ia-norm  & 12,13,15,16,21,29                                                                                                 & 28             \\

3119: 1999aa & Ia-91T   & $-$[9--1],1,15--18,27--29,30+(11)                                                                                 & 28             \\ %30,43,45,47,49,51,53,55,57,73,82

3120: 1999ac & Ia-91T   & $-$4,$-$3,$-$1,9--12,25,27,29,30+(9)                                                                              & 28             \\ %31,33,35,37,39,41,55,63,86

3121: 1999by & Ia-91bg  & $-$[5--2],2--8,11,25,29,30+(3)                                                                                    & 28,36          \\ %31,33,42

3122: 1999cc & Ia-norm  & $-$3,$-$1,0,2,19,24,25                                                                                            & 28             \\

3123: 1999cl & Ia-norm  & $-$8,$-$7,$-$[5--1],1,8,30+(1)                                                                                    & 28             \\ %38

3124: 1999dq & Ia-91T   & $-$[10--2],1--4,6,18,19,27,30+(6)                                                                                 & 28             \\ %30,34,47,48,58,89

3125: 1999ee & Ia-norm  & $-$9,$-$7,0,2,7,11,16,20,22,27,30+(2)                                                                             & 28             \\ %32,41

3126: 1999ej & Ia-norm  & $-$1,2,4,9,12                                                                                                     & 28             \\

3127: 1999em & IIP      & 6(2),7--9,10(2),11,12,15,16(3),17,19,21,26,37,40+(27)                                                             & 37--39         \\ %42,44,45,47,49,54,68,70,72,74,78,80,81,82,97,100,111,125,129,130,137,143,164,168,318,338,521

3128: 1999ex{\tablenotemark{b}} & Ib-norm  & $-$5,0,9                                                                                       & 40             \\

3129: 1999gd & Ia-norm  & 3,9,27,30+(2)                                                                                                     & 28             \\ %31,36

3130: 1999gh & Ia-91bg  & 5--9,11,12,28,30+(7)                                                                                              & 28             \\ %34,36,40,42,66,71,73

3131: 1999gi & IIP      & 5,7,8,31,36,39,40+(5)                                                                                             & 41             \\ %89,93,141,156,172

3132: 1999gp & Ia-91T   & $-$5,$-$2,0,3,5,7,22,30+(3)                                                                                       & 28             \\ %35,50,55

3133: 2000B  & Ia-norm  & 9,14,22,30+(2)                                                                                                    & 28             \\ %38,44

3134: 2000E  & Ia-norm  & $-$6,$-$3,$-$1,8,30+(1)                                                                                           & 42             \\ %121

3135: 2000H  & IIb      & 28,29(2),31--34,40+(5)                                                                                            & 43,CfA         \\ %46,50,61,77,86

3136: 2000cf & Ia-norm  & 3,4,15,17,25,26                                                                                                   & 28             \\

3137: 2000cn & Ia-91bg  & $-$[9--7],9,11,13,22,26,27,30+(1)                                                                                 & 28             \\ %51

3138: 2000cx & Ia-pec   & $-$[3--1],0--2,6--8,10,12,15,19,22,24,26,28,30+(9)                                                                & 44             \\ %30,31,41,61,71,93,97,125,146

3139: 2000dk & Ia-91bg  & $-$5,$-$4,1,4,10,30+(3)                                                                                           & 28             \\ %34,59,89

3140: 2000fa & Ia-norm  & $-$10,$-$9,2,3,5,9,11,14,16,18,21,30+(3)                                                                          & 28             \\ %32,38,43

3141: 2001V  & Ia-norm  & $-$[13--9],$-$[7--5],$-$3,10,11(2),13,14,18,19,20(2),21(2),22--24,27,28,30+(13)                                   & 28             \\ %40,47,48,49,50,52,52,53,57,73,78,79,105

3142: 2002ap & Ic-hyper & $-$6,$-$5,$-$2,$-$1,0--2,4--6,7(2),10,12,26,30+(5)                                                                & 45,CfA         \\ %30,31,149,155,267

3143: 2002bo & Ia-norm  & $-$13(2),$-$12(2),$-$[9--6],$-$5(2),$-$4(2),$-$3(3),$-$2(2),$-$1(3),0,6,11--22,24,28,29(2),30+(12)                & 46,CfA         \\ %30,39,41,44,44,46,50,52,55,73,79,83

3144: 2002cx & Ia-pec   & $-$5,$-$2,10,14,18,23,24,30+(1)                                                                                   & 47             \\ %53

3145: 2002er & Ia-norm  & $-$11,$-$[9--5],$-$4(2),$-$[3--1],2--4,6,9,11,12,16(2),20,24,30+(2)                                               & 48             \\ %33,213

3146: 2003du & Ia-norm  & $-$12,$-$10(3),$-$[9--5],$-$3,$-$1(2),0--2,3(2),4,5,8--11,14,17--19,20(2),22(2),24,26,27,29,30+(24)               & 49,50          \\ %32--35,37,40,46,50,52(2),57,59,61,62,63,72,84,87,109,138,139,141,221,375

3147: 2004ao & Ib-norm  & 7--13,16,17,20--23,30+(16)                                                                                        & CfA            \\ %38,41,44,49,52,66,68,74,80,103,106,135,185,188,190,198

3148: %2004aw & Ic-norm  & $-$2,2,3,5,12,18,19,23,25,26,30+(3)                                                                               & 51             \\ %32,229,254

3149: 2004aw & Ic-norm  & $-$5,$-$3,$-$2(2)$-$1,0,2(2),3--5,12,18(2),19,21,23,25,26,30+(6)                                                  & 51,CfA         \\ %30,32,46,48229,254

3150: 2004et & IIP      & 13--15,17,18,20,22,26,28,40+(15)                                                                                  & CfA            \\ %40,41,43,49,50,51,58,79,80,227,266,284,346,385,398

3151: %2004gt & Ic-norm  & 9,11,15,17,30+(8)                                                                                                 & CfA            \\ %36,41,62,92,124,134,160,179

3152: 2004gt & Ic-norm  & 15,17,21,23,30+(8)                                                                                                & CfA            \\ %42,47,69,98,130,140,166,185

3153: 2005bf{\tablenotemark{c}} & Ib-pec   & $-$4,$-$2,$-$1,0,2,16,18--23,25--27,29,30+(6)                                                  & 52             \\ %30,32,47,50,54,57

3154: 2005cs & IIP      & 7--14,16--19,35,36,40+(1)                                                                                         & 53,54          \\ %185

3155: %2005hg & Ib-norm  & $-$13,$-$12,$-$10(2),$-$[9--1],12,16,26                                                                           & 55,CfA         \\

3156: 2005hg & Ib-norm  & $-$[13--1],0,13,17,27                                                                                             & CfA            \\

3157: 2005hk & Ia-pec   & $-$9$-$8(2),$-$7(2),$-$6(3),$-$5(2),$-$4(3),$-$3(2),$-$2,$-$1,4,12,14(2),20,23(2),26,27,30+(6)                    & 55,CfA         \\ %37,39,42,43,54,66

3158: 2005mf & Ic-norm  & $-$4,$-$3,3,6                                                                                                     & CfA            \\

3159: 2005mz & Ia-91bg  & $-$7,11,13,19                                                                                                     & CfA            \\

3160: 2006T  & IIb      & 8,10,28,36                                                                                                        & CfA            \\

3161: 2006aj & Ic-broad & $-$5,$-$3,$-$2,$-$1,0,2,3                                                                                         & 56             \\

3162: \enddata

3163: \tablenotetext{\ }{

3164: {\it Column headings:}

3165: (1) IAU designation.

3166: (2) Supernova subtype, as defined in Table~\ref{Table:stype}.

3167: (3) Rest-frame SN age, rounded to closest whole day, in days from

3168: $B$-band maximum (for SN~Ia), from $V$-band maximum (for SN~Ib/c), or from the 

3169: estimated date of explosion (for SN~II). Adjacent ages are listed in

3170: between square brackets; a ``(n)'' indicates that $n$ spectra

3171: correspond to a same rounded age. Spectra of SN~Ia/b/c whose age

3172: exceeds +30 days are grouped together, e.g. 30+(5) indicated there are

3173: 5 spectra with ages $\ge +30$ days (past maximum); same for spectra of SN~II whose

3174: age exceeds 40 days (past explosion).

3175: (4) Reference of refereed articles presenting optical spectroscopic

3176: data (see ``References'' below); ``CfA'' refers to unpublished spectra

3177: obtained by members of the CfA SN Group, some of which are available

3178: {\it via} the CfA Supernova Archive ({\tt

3179: http://www.cfa.harvard.edu/supernova/SNarchive.html}).

3180: }

3181: \tablenotetext{a}{The light curve of SN~1992H exhibited a truncated

3182: plateau \citep{Clocchiatti/etal:1996}, but its spectra are otherwise

3183: indistinguishable from other Type IIP supernovae.}

3184: \tablenotetext{b}{\citet{Hamuy/etal:2002} classify SN~1999ex as an

3185:   intermediate Ib/c event, while \citet{Parrent/etal:2007} support the

3186:   Ib classification, highlighting the similarity with the peculiar

3187:   SN~Ib 2005bf \citep{Tominaga/etal:2005}. We classify SN~1999ex as a

3188:   normal Type Ib supernova and note that the essential spectroscopic

3189:   peculiarity of SN~2005bf (namely the increasing absorption velocity

3190:   of the \heone\,\l5876 line; \citealt{Tominaga/etal:2005}) is not

3191:   present in SN~1999ex.}

3192: \tablenotetext{c}{The $V$-band light curve of SN~2005bf had two

3193: distinct maxima. The age is expressed in days from the {\it first}

3194: maximum.}

3195: \tablerefs{

3196: (1)  \citealt{Wells/etal:1994}; 

3197: (2)  \citealt{Phillips/etal:1988};

3198: (3)  \citealt{Phillips/etal:1990}; 

3199: (4)  \citealt{Matheson/etal:2001};

3200: (5)  \citealt{Clocchiatti/etal:2001};

3201: (6)  \citealt{Elmhamdi/etal:2004};

3202: (7)  \citealt{Leibundgut/etal:1991};

3203: (8)  \citealt{Mazzali/etal:1993};

3204: (9)  \citealt{Jeffery/etal:1992};

3205: (10) \citealt{Schmidt/etal:1994};

3206: (11) \citealt{Mazzali/etal:1995};

3207: (12) \citealt{Leibundgut/etal:1993};

3208: (13) \citealt{Turatto/etal:1996};

3209: (14) \citealt{Gomez/Lopez/Sanchez:1996};

3210: (15) \citealt{Kirshner/etal:1993};

3211: (16) \citealt{Clocchiatti/etal:1996};

3212: (17) \citealt{Gomez/Lopez:2000};

3213: (18) \citealt{Jeffery/etal:1994};

3214: (19) \citealt{Barbon/etal:1995};

3215: (20) \citealt{Matheson/etal:2000};

3216: (21) \citealt{Fransson/etal:2005};

3217: (22) \citealt{Hoeflich:1995};

3218: (23) \citealt{Patat/etal:1996};

3219: (24) \citealt{Millard/etal:1999};

3220: (25) \citealt{Salvo/etal:2001};

3221: (26) \citealt{Li/etal:1999};

3222: (27) \citealt{Turatto/etal:1998};

3223: (28) \citealt{Matheson/etal:2007}; 

3224: (29) \citealt{Iwamoto/etal:2000};

3225: (30) \citealt{Lentz/etal:2001};

3226: (31) \citealt{Fassia/etal:2001};

3227: (32) \citealt{Fransson/etal:2005};

3228: (33) \citealt{Branch/etal:2003};

3229: (34) \citealt{Jha/etal:1999};

3230: (35) \citealt{Patat/etal:2001};

3231: (36) \citealt{Garnavich/etal:2004};

3232: (37) \citealt{Baron/etal:2000};

3233: (38) \citealt{Hamuy/etal:2001};

3234: (39) \citealt{Leonard/etal:2002a};

3235: (40) \citealt{Hamuy/etal:2002};

3236: (41) \citealt{Leonard/etal:2002b};

3237: (42) \citealt{Valentini/etal:2003};

3238: (43) \citealt{Branch/etal:2002};

3239: (44) \citealt{Li/etal:2001a};

3240: (45) \citealt{Gal-Yam/Ofek/Shemmer:2002};

3241: (46) \citealt{Benetti/etal:2004a};

3242: (47) \citealt{Li/etal:2003};

3243: (48) \citealt{Kotak/etal:2005};

3244: (49) \citealt{Anupama/Sahu/Jose:2005};

3245: (50) \citealt{Stanishev/etal:2007};

3246: (51) \citealt{Taubenberger/etal:2006};

3247: (52) \citealt{Tominaga/etal:2005};

3248: (53) \citealt{Brown/etal:2007};

3249: (54) \citealt{Dessart/etal:2007};

3250: (55) \citealt{Phillips/etal:2007};

3251: (56) \citealt{Modjaz/etal:2006}

3252: }

3253: \end{deluxetable*}

3254: \clearpage

3255: 

3256: \end{document}

3257: