1: \documentclass[12pt,preprint]{aastex}
2: %\usepackage{rotating,verbatim,vmargin,graphicx,amssymb,amsfonts,amsmath,calc,epsfig}
3: \usepackage{rotating,amsfonts,epsfig, enumerate}
4: %\usepackage{draftcopy}
5: %% specify the page margins
6: %\setpapersize{USletter}
7: %\setmarginsrb{1.1in}{1.2in}{1.1in}{0.5in}%
8: %{1.0\baselineskip}{1.5\baselineskip}{1.5\baselineskip}{2.5\baselineskip}
9: \begin{document}
10:
11:
12: \bibliographystyle{apj}
13:
14: \title{A New Determination of the High Redshift Type Ia Supernova Rates with the Hubble Space Telescope Advanced Camera for Surveys.\altaffilmark{1}}
15:
16: \author{N.~Kuznetsova\altaffilmark{2,3},
17: K.~Barbary\altaffilmark{2,4},
18: B.~Connolly\altaffilmark{5},
19: A.~G.~Kim\altaffilmark{2},
20: R.~Pain\altaffilmark{6},
21: N.~A.~Roe\altaffilmark{2},
22: G.~Aldering\altaffilmark{2},
23: R.~Amanullah\altaffilmark{7},
24: K.~Dawson\altaffilmark{2},
25: M.~Doi\altaffilmark{8}
26: V.~Fadeyev\altaffilmark{9},
27: A.~S.~Fruchter\altaffilmark{10},
28: R.~Gibbons\altaffilmark{11},
29: G.~Goldhaber\altaffilmark{2,4},
30: A.~Goobar\altaffilmark{12},
31: A.~Gude\altaffilmark{4},
32: R.~A.~Knop\altaffilmark{11},
33: M.~Kowalski\altaffilmark{13},
34: C.~Lidman\altaffilmark{14},
35: T.~Morokuma\altaffilmark{15}
36: J.~Meyers\altaffilmark{2,4},
37: S.~Perlmutter\altaffilmark{2,4},
38: D.~Rubin\altaffilmark{2,4},
39: D.~J.~Schlegel\altaffilmark{2},
40: A.~L.~Spadafora\altaffilmark{2},
41: V.~Stanishev\altaffilmark{12},
42: M.~Strovink\altaffilmark{2,4},
43: N.~Suzuki\altaffilmark{2},
44: L.~Wang\altaffilmark{16},
45: N.~Yasuda\altaffilmark{17} (Supernova Cosmology Project)
46: }
47:
48: \altaffiltext{1}{Based on observations with the NASA/ESA \emph{Hubble Space Telescope}, obtained at the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS 5-26555, under programs GO-9583, GO-9425, GO-9727, and GO-9728.}
49: \altaffiltext{2}{E. O. Lawrence Berkeley National Laboratory, 1 Cyclotron Rd., Berkeley, CA 94720, USA }
50: \altaffiltext{3}{Current address: Physics Department, Hamilton College, Clinton, NY 13323, USA}
51: \altaffiltext{4}{Department of Physics, University of California Berkeley, Berkeley, CA 94720, USA}
52: \altaffiltext{5}{Department of Physics, Columbia University, New York, NY 10027, USA}
53: \altaffiltext{6}{LPNHE, CNRS-IN2P3, University of Paris VI \& VII, Paris, France }
54: \altaffiltext{7}{The Space Sciences Laboratory, University of California Berkeley, Berkeley, CA 94720, USA}
55: \altaffiltext{8}{Institute of Astronomy, School of Science, University of Tokyo, Mitaka, Tokyo, 181-0015, Japan}
56: \altaffiltext{9}{Department of Physics, University of California Santa Cruz, Santa Cruz, CA 95064, USA}
57: \altaffiltext{10}{Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA}
58: \altaffiltext{11}{Department of Physics and Astronomy, Vanderbilt University, Nashville, TN 37240, USA}
59: \altaffiltext{12}{Department of Physics, Stockholm University, Albanova University Center, S-106 91 Stockholm, Sweden}
60: \altaffiltext{13}{Humboldt Universit\"{a}t Institut f\"{u}r Physik, Newtonstrasse 15, Berlin 12489, Germany}
61: \altaffiltext{14}{European Southern Observatory, Alonso de Cordova 3107, Vitacura, Casilla 19001, Santiago 19, Chile}
62: \altaffiltext{15}{Optical and Infrared Astronomy Division, National Astronomical Observatory of Japan, Mitaka, Tokyo, 181-8588, Japan}
63: \altaffiltext{16}{Department of Physics, Texas A\&M University, College Station, TX 77843, USA}
64: \altaffiltext{17}{Institute for Cosmic Ray Research, University of Tokyo, Kashiwa, 277 8582, Japan}
65:
66: \email{nvkuznetsova@lbl.gov}
67:
68: %\offprints{nvkuznetsova@lbl.gov}
69: %\date{Received/Accepted}
70:
71: %\date{\today}
72:
73: \begin{abstract}
74: We present a new measurement of the volumetric rate of Type Ia supernova up
75: to a redshift of 1.7, using the Hubble Space Telescope (HST) GOODS data combined
76: with an additional HST dataset covering the North GOODS field collected in 2004.
77: We employ a novel technique that does not require spectroscopic data
78: for identifying Type Ia supernovae
79: (although spectroscopic measurements of redshifts are used for over
80: half the sample); instead we employ a Bayesian approach using only
81: photometric
82: data to calculate the probability that an object is a Type Ia supernova.
83: This Bayesian technique can easily be modified
84: to incorporate improved priors on supernova properties, and it is well-suited for future
85: high-statistics supernovae searches in which spectroscopic follow up of all candidates
86: will be impractical. Here, the method is validated on both ground- and space-based supernova data having some spectroscopic follow
87: up. We combine our volumetric rate measurements with low redshift supernova data,
88: and fit to a number of possible models for the evolution of the Type Ia supernova rate as
89: a function of redshift. The data do not distinguish between a flat rate at
90: redshift $>$ 0.5 and a previously proposed model, in which the Type Ia rate peaks at redshift
91: $\sim$ 1 due to a significant delay from star-formation to the supernova explosion. Except for the highest redshifts, where the signal to noise ratio is
92: generally too low to apply this technique,
93: this approach yields smaller or comparable uncertainties than previous work.
94: \keywords{supernovae: general}
95: \end{abstract}
96:
97: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
98: \section{Introduction}
99: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
100: The empirical evidence for the existence of dark energy came from observations of
101: Type Ia supernovae~\citep{bib:riess, bib:P99, bib:P03}, which
102: are believed to arise from the thermonuclear explosion of a progenitor white dwarf
103: after it approaches the Chandrasekhar mass limit~\citep{bib:chan}.
104: However, the physics
105: of Type Ia supernova production is not well understood.
106: The two most plausible scenarios for the white dwarf to accrete the
107: necessary mass are the single degenerate case, where the white dwarf is
108: located in a binary system; and the double
109: degenerate case, where two white dwarfs merge. The
110: Type Ia supernova rate is correlated with the star formation history (SFH), and
111: thus a measurement of the rate as a function of redshift helps constrain the possible type
112: Ia progenitor models.
113:
114: In addition to its importance for understanding Type Ia supernovae as astronomical objects,
115: a good grasp of the Type Ia supernova rate to high redshifts is important
116: for the next generation of proposed space-based supernova cosmology experiments,
117: such as SNAP~\citep{bib:snap}.
118: It is therefore of great practical interest to determine the rate of Type Ia supernovae
119: at redshifts $>$ 1.
120:
121: The subject of Type Ia supernova rates has been addressed by many authors
122: in the past.
123: Existing rate measurements have been mostly limited to redshift ranges $<$ 1:
124: the results of~\cite{bib:cappellaro},~\cite{bib:hardin},~\cite{bib:madgwick}, and~\cite{bib:blanc}
125: measure the rates at redshifts $\leq$ $\sim$0.1;~\cite{bib:neill},~\cite{bib:tonry}, and~\cite{bib:pain},
126: at intermediate redshifts of 0.47, 0.50, and 0.55, respectively; and~\cite{bib:barris},
127: up to a redshift of 0.75. The only published measurement of the rates
128: at redshifts $>$ 1 is that of~\cite{bib:dahlen}, who analyzed the GOODS dataset.
129:
130: There are several important
131: differences that distinguish our work from that of~\cite{bib:dahlen}.
132: First, we augment the GOODS sample with the HST data collected during
133: the Spring-Summer 2004 high redshift supernova searches.
134: Second, our methods of calculating the control time (the time during which
135: a supernova search is potentially capable of finding supernova candidates)
136: and the efficiency to identify a supernova are based on a detailed Monte Carlo
137: simulation technique using a library of supernova templates.
138: Third, we adopt a novel approach
139: to typing supernovae, using photometric data and a
140: Bayesian probability method described in~\cite{bib:ourpaper}.
141: The Bayesian technique is able to perform classification
142: using only photometric data, and therefore does not require
143: spectroscopic follow up. Optionally, photometric or
144: spectroscopic redshifts can be used to improve the classification accuracy.
145: Our initial requirements on potential supernova candidates
146: are more stringent in terms of the number of points on the light curve
147: and the signal to noise of those points
148: than those of~\cite{bib:dahlen}; thus some of the candidates they
149: identified will fail our cuts.
150: However, we are able to reliably separate Type Ia supernova
151: from other supernovae types based on their Bayesian probability, with
152: an efficiency that is readily quantifiable, thus allowing
153: us to use larger data samples.
154: Our approach therefore avoids the problems
155: that arise in estimating the efficiency for the decision to schedule
156: spectroscopic follow up based on a potentially low signal-to-noise
157: initial detection.
158:
159:
160: The Bayesian classification technique uses photometric data, and
161: does not require any spectroscopic followup. This is an advantage for
162: future large-area surveys (such as the Dark Energy Survey,
163: Pan-STARRS, and LSST) that will discover thousands of supernova candidates,
164: but are unlikely to be able to obtain spectroscopic data for all of them, to distinguish
165: Type Ia supernovae from core collapse supernovae and other variable objects.
166: The technique described here can be
167: considered a prototype of the kind of analysis that could be performed on
168: these future large data sets to identify Type Ia supernovae for cosmological studies.
169: There is a clear trade-off involved in using photometric measurements alone:
170: if the quality of the photometric data is poor, then
171: the efficiency of this technique to identify Type Ia supernovae
172: is reduced; on the other hand, this technique enables larger samples
173: of Type Ia from imaging surveys to be identified for cosmological studies,
174: without the need for time-consuming spectroscopic follow up.
175:
176:
177: Note that although the method is able to perform the supernova typing with photometric data alone (\emph{i.e.}, it does not require spectroscopic data, either
178: redshifts or types), it is certainly able to use the extra information that is
179: available, and in fact 70\% of the supernova
180: candidates discussed in the present work have redshifts which were
181: obtained spectroscopically.
182: It is also worth noting that while in this paper we only analyze the Type Ia supernova rates,
183: the Bayesian classification technique can be used to classify other types as well,
184: making it possible to measure the rates of non-Type Ia supernovae in a similar fashion.
185: These analyses will be presented in future publications.
186:
187:
188:
189: The paper is organized as follows. In section~\ref{sec:data} we describe
190: the data samples used in the analysis. In section~\ref{sec:candselection}
191: we describe the supernova candidate selection and typing process.
192: In section 4 we calculate the control time, survey area,
193: and search efficiency, and determine the volumetric Type Ia
194: supernova rate from our data sample.
195: A comparison of the rates with those reported in the literature is given
196: in section~\ref{sec:comp}, and fits of the rates to different models
197: relating the Type Ia supernova rates to the SFH
198: are given in section~\ref{sec:sf}.
199: A summary is given in section~\ref{sec:concl}.
200:
201: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
202: \section{Data Sample}
203: \label{sec:data}
204: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
205: For this analysis, we use the Hubble Space Telescope (HST)
206: GOODS dataset collected in 2002-2003~\citep{bib:goods2, bib:goods1, bib:goods0}.
207: In addition to the GOODS data, we use an HST sample collected
208: in the Spring-Summer of 2004, which hereafter we will call the 2004 ACS sample.
209: The GOODS dataset consists of five epochs (data taking periods), separated by approximately
210: 45 observer-frame days. The GOODS data used for this analysis were taken in
211: two HST ACS filter bands: F775W (centered at 775 nm) and F850LP (centered at 850 nm)\footnote[1]{The ACS filter transmission curves
212: are available at http://acs.pha.jhu.edu/instrument/filters/.}.
213: Each F850LP image consists of four exposures; and each F775W image, of two.
214: The GOODS survey includes two fields, GOODS North and GOODS South,
215: and covers approximately 320 square arcminutes. The fields are sub-divided into
216: smaller ``tiles'' that correspond to single ACS pointings
217: (typically 15 or 16), as shown in Fig.~\ref{fig:goods}.
218: %%
219: \begin{figure}[!htb]
220: \begin{center}
221: $\begin{array}{c@{\hspace{0.0in}}c}
222: \epsfxsize=3.in\epsfysize=3.2in\epsffile{f1a.ps} &
223: \epsfxsize=3.in\epsfysize=3.2in\epsffile{f1b.ps}
224: \end{array}$
225: \end{center}
226: \caption[]
227: {\label{fig:goods}
228: The North (left) and South (right) GOODS fields.
229: The fields are subdivided into tiles, which are shown (along with their ID numbers)
230: on the figures. The size of a single tile is $\sim$11.5 sq. arcmin.
231: }
232: \end{figure}
233: %%
234:
235: The 2004 ACS supernova dataset covers only the GOODS North field, with
236: the same tiling as that of the GOODS North dataset. It consists of 4 epochs
237: separated by approximately 45 observer-frame days.
238: The data in this sample were taken in two HST ACS passbands:
239: F775W and F850LP, with one exposure for every F775W image and four
240: for every F850LP image. Two teams (PI Perlmutter and PI Riess) shared this data searching for
241: supernovae in alternate visits;~\cite{bib:riess06} have published the
242: results for the supernovae that were discovered in their team's visits.
243:
244:
245: For convenience, a summary of the datasets used
246: is given in Table~\ref{tab:data}.
247: %%%%
248: \begin{table}[htbp]
249: \begin{center}
250: \begin{minipage}{\textwidth}
251: \begin{tabular}{|c|c|c|c|c|c|c|}
252: \hline\hline
253: Epoch & Filter & Exp.time (s) & Filter & Exp. time (s) & \# Tiles & Taken On \\
254: \hline\hline
255: \multicolumn{7}{|c|}{GOODS South} \\
256: \hline
257: 1 & F775W& 1040 & F850LP & 2120 & 15 & 7/31 - 8/4 (2002)\\
258: 2 & F775W & 1040 & F850LP & 2120 & 16 & 9/19 - 9/22 (2002)\\
259: 3 & F775W & 1040 & F850LP & 2120 & 15 & 10/31 - 11/3 (2002)\\
260: 4 & F775W & 1040 & F850LP & 2120 & 16 & 12/19 - 22 (2002)\\
261: 5 & F775W & 1040 & F850LP & 2120 & 15 & 2/1 - 2/5 (2003)\\
262: \hline\hline
263: \multicolumn{7}{|c|}{GOODS North} \\
264: \hline
265: 1 & F775W & 1120 & F850LP & 2400 & 14 & 11/21 - 11/22 (2002)\\
266: 2 & F775W & 1000\footnote{except for tile 30 (1060)} & F850LP & 2120 & 17 & 1/2 - 1/4 (2003)\\
267: 3 & F775W & 960 & F850LP & 2060 & 16 & 2/20 - 2/23 (2003) \\
268: 4 & F775W & 960 & F850LP & 2000 & 14 & 4/3 - 4/6 (2003)\\
269: 5 & F775W & 960 & F850LP & 2080 & 15 & 5/21 - 5/25 (2003)\\
270: \hline\hline
271: \multicolumn{7}{|c|}{2004 ACS Sample (GOODS North tiles)} \\
272: \hline
273: 1 & F775W & 400 & F850LP & 1600 & 15 & 4/2 - 4/4 (2004) \\
274: 2 & F775W & 400 & F850LP & 1600 & 15 & 5/20 - 5/23 (2004) \\
275: 3 & F775W & 400 & F850LP & 1600 & 15 & 7/9 - 7/10 (2004) \\
276: 4 & F775W & 400 & F850LP & 1600 & 15 & 8/26 - 8/28 (2004)\\
277: \hline\hline
278: \end{tabular}
279: \end{minipage}
280: \end{center}
281: \caption[]
282: {\label{tab:data}
283: A summary of the datasets used in this analysis, listing the data taking epochs,
284: the filters, the exposure times of the combined exposures (in seconds),
285: the number of GOODS field tiles, and the dates when the data were taken.
286: }
287: \end{table}
288: %%%
289:
290:
291: It is worth emphasizing that we are using photometric information from only
292: two filter bands, providing one color measurement.
293: The GOODS dataset has been analyzed before, and
294: 13 out of 42
295: supernovae found were spectroscopically typed~\citep{bib:riess2, bib:strolger}.
296: For the 2004 ACS sample, however, the spectroscopic information is
297: available only for a small fraction of the candidates.
298: We treat both GOODS and 2004 ACS datasets in a consistent fashion, using
299: photometric information only for typing supernovae (note that we still
300: use spectroscopically determined redshifts where available).
301: This allows more data to be searched
302: and more supernovae to be found, but at the expense of neglecting
303: spectroscopic information for the candidates where it is available.
304: In section~\ref{sec:typeia} we discuss
305: in detail the resulting supernova candidate count.
306:
307:
308: We start with the data that have been flat-fielded and gain-corrected
309: by the HST pipeline, and use MultiDrizzle~\citep{bib:drizzle} to
310: perform cosmic ray rejection and to combine dithered observations.
311: The parameters of the drizzling process include a ``square'' kernel, with a pixel fraction of 0.66
312: and a pixel scale of 1.0. The drizzling
313: combines the multiple individual pointings.
314: %We then process the data making use of the STSDAS and
315: %PyRAF (\cite{bib:pyraf}) packagesf\ootnote{STSDAS and PyRAF are products of the Space
316: %Telescope Science Institute, which is operated by AURA for NASA},
317: %as well as the MultiDrizzle package (\cite{bib:drizzle}).
318: %The drizzling combines the multiple individual pointings
319: %to produce an image free of cosmic rays.
320: Drizzling is ineffective for the cosmic ray rejection for
321: the F775W data from the 2004 ACS sample since they contain
322: only a single exposure for each GOODS North tile. We therefore
323: use a morphological cosmic ray rejection package~\citep{bib:lacosmics} to
324: create images with identifiable objects, thus allowing us to generate
325: the geometrical transformations between images; however, the original
326: images are used for extracting photometric information (after verifying
327: that no cosmic rays landed directly at the location of the supernova candidates).
328:
329:
330: Supernovae are identified by subtracting a reference image from each
331: of the HST search epochs.
332: We create four distinct samples summarized in Table~\ref{tab:samples}, which
333: we use for identifying and performing simple aperture photometry on
334: the supernova candidates in each of the five epochs in the GOODS dataset
335: and each of the four epochs in the 2004 ACS dataset.
336: To obtain the multi-epoch photometry
337: for the GOODS North data (sample \#1), we combine all four epochs of the 2004 ACS sample
338: and then subtract these data from each of the five North GOODS epochs in turn.
339: Combining multiple epochs for the reference image allows us to create deeper
340: resulting data, which is important for extracting supernovae with
341: the best possible signal-to-noise ratio (SNR).
342: For sample \#2, we combine the entire North GOODS sample and subtract these data
343: from each of the four 2004 ACS epochs in turn.
344: Because the GOODS and 2004 ACS data were taken with a time separation of approximately
345: a year, these samples should be sensitive to the supernovae that were both on the rise and
346: on the decline during the GOODS and 2004 ACS data taking period for samples \#1 and \#2,
347: respectively.
348: For the GOODS South sample, however, we do not have any additional datasets, and
349: are thus forced to separate the sample into two. This is the reason
350: the three initial data samples (GOODS North and South and the 2004 ACS data set)
351: result in four search samples.
352: We combine South epochs 4 and 5 for sample \#3, and epoch 1 and 2 for sample \#4;
353: we then subtract the two combined samples separately from each of the five
354: South GOODS epochs.
355: %%
356: \begin{table}[htbp]
357: \begin{center}
358: \begin{tabular}{|c|c|c|}
359: \hline\hline
360: Sample \# & Reference Dataset & Supernova Search \\
361: \hline\hline
362: 1 & Combined 2004 ACS data (4 epochs) & Individual North GOODS epochs\\
363: 2 & Combined North GOODS data (5 epochs) & Individual 2004 ACS dataset epochs \\
364: 3 & Epochs 4+5 of the South GOODS data & Individual South GOODS epochs\\
365: 4 & Epochs 1+2 of the South GOODS data & Individual South GOODS epochs\\
366: \hline\hline
367: \end{tabular}
368: \end{center}
369: \caption[]
370: {\label{tab:samples}
371: The samples used in our supernova search. To identify and extract photometry for supernova candidates,
372: we subtract the data listed in column 2 from the data listed in column 3. Note
373: that sample \#2 has the deepest references.}
374: \end{table}
375: %
376: If a supernova candidate has been found in both samples \#3 and \#4, we consider it
377: belonging to the sample in which it had an epoch with the largest
378: SNR. This avoids any possible double-counting of
379: the candidates for the GOODS South data.
380:
381:
382: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
383: \section{The Supernova Candidate Selection and Typing}
384: \label{sec:candselection}
385: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
386: The search for supernova candidates and their subsequent
387: typing as Ia's is a 3-stage process. We will briefly describe them
388: below, and then in detail in sections~\ref{sec:initialsearch}-\ref{sec:typeia}.
389: \begin{enumerate}
390: \item
391: First, potential supernova candidates in individual epochs are identified
392: by the software that is used to subtract the supernova
393: search data from the reference data.
394: The initial candidate selection is done using the F850LP data only
395: because it suffers less from cosmic ray contamination,
396: and because F850LP covers supernovae at redshifts up to $\sim$1.5.
397: The initial supernova selection is primarily directed toward
398: reducing the number of false positives resulting from various image processing artifacts
399: and residual cosmic ray contamination.
400: It is followed by a manual scan to reject any obvious remaining cosmic rays and
401: image processing artifacts.
402: Note that both sources of false detections have specific signatures that
403: real supernovae do not have; this selection therefore is not expected
404: to reduce the number of real supernovae in the sample.
405: This stage is described in detail in section~\ref{sec:initialsearch}.
406: \item
407: For the candidates on individual epochs that pass the first stage of the selection process,
408: we extract the photometric information at the candidate locations
409: in the multi-epoch F850LP and F775W data.
410: We then select candidates with reasonably well-measured light curves
411: by requiring that the candidate's SNR in the
412: subtracted data (in both filters) be greater than 2 for at least 3 search epochs,
413: including at least two with a SNR greater than 3.
414: At the end of this stage, we are left with the majority of candidates that
415: are presumed to be supernovae of some type, as well as some candidates
416: that cannot be modeled as any known supernova type.
417: This stage is described in detail in section~\ref{sec:multibandsearch}.
418: \item
419: The final step applies a Bayesian likelihood technique that assigns each candidate that passed steps 1 and 2
420: a probability to be a Type Ia supernova based on the multi-epoch data in both filters.
421: This stage is described in detail in section~\ref{sec:typeia}.
422: \end{enumerate}
423: For convenience, we summarize the selection process in Table~\ref{tab:selection}.
424: We now describe each of the selection stages in detail.
425: %%
426: \begin{table}[htbp]
427: \begin{center}
428: \begin{tabular}{|c|c|c|}
429: \hline\hline
430: Selection Stage & Data Used & Cuts Applied\\
431: \hline\hline
432: & & SNR$_{\rm exposure}$ $>$ 3 in 4 exposures\\
433: 1 & F850LP & SNR consistency in 3 out of 4 exposures\\
434: & single (discovery) epoch & Percent increase $\geq$ 15\% in combined exposures \\
435: & & shape cuts in combined exposures\\
436: \hline
437: 2 & F850LP, F775W, all epochs & $\geq$ 3 epochs with S/N $>$ 2 \\
438: & & (including $\geq$ 2 epochs with S/N $>$ 3) \\
439: \hline
440: 3 & F850LP, F775W, all epochs & Bayesian Type Ia classification \\
441: \hline\hline
442: \end{tabular}
443: \end{center}
444: \caption[]
445: {\label{tab:selection}
446: A summary of the Type Ia supernova selection and typing process. The meaning of the cuts
447: is explained in the text describing the corresponding stages.}
448: \end{table}
449:
450:
451: %%%%%%
452: \subsection{Stage 1: Single Epoch Supernova Candidate Selection}
453: \label{sec:initialsearch}
454: %%%%%
455: In the first step of the supernova search, we search for supernova candidates
456: in the individual epochs of the F850LP data by looking for signals in the reference-subtracted search
457: images. The reference image is the same for each exposure
458: (recall that each F850LP image consists of four exposures, each
459: with the same exposure time). We use aperture photometry
460: with a radius of 3 pixels, where the pixel scale is 0.03'' (after drizzling).
461: This choice of the aperture optimizes the SNR of supernova candidates.
462: We verified that the photometric extraction procedure is working
463: well by creating ``fake'' supernovae, as described later in this section,
464: and comparing their input and output magnitudes;
465: they agree at the sub-percent level. The procedure for identifying supernovae is as follows.
466: \renewcommand{\labelenumi}{\roman{enumi}.}
467: \begin{enumerate}
468: \item
469: Subtracting the combined (drizzled) exposures of the search data from the (drizzled) reference data,
470: we require that:
471: \begin{itemize}
472: \item
473: The absolute value of the flux within the supernova candidate's aperture
474: in the subtracted data divided by the flux in the reference data (the ``percent
475: increase'' variable) be $\geq$ 15\%.
476: % perinc >= 15 (Percent Increase) -- flux in the subtraction divided by flux in the reference within
477: %the candidate aperture, multiplied by 100. A perinc of 100 indicates that the new signal has a flux \
478: %equal to the flux underlying it in the reference image.
479: \item
480: The candidate's shape in the subtracted data must be consistent with a point source:
481: we require that the candidate's FWHM in both $x$- and $y$- directions
482: be $<$ 4 pixels, and that the absolute value of its normalized $xy$ moment be $<$ 0.5 pixels.
483: % fwx is the candidate's full width in the x-direction.
484: % mxy is the xy moment of the isophotal object
485: %($fw_x$ $<$ 4, $fw_y$ $<$ 4, $|m_{xy}|$ $<$ 0.5).
486: %These are the "profile" variables from which one can judge how close the candidate's shape is
487: %to a perfect point source. Also, normalized moment |mxy| < 0.5.
488: \end{itemize}
489: \item
490: Next, to eliminate false detections resulting from cosmic rays, we do
491: the following:
492: \begin{itemize}
493: \item
494: We consider the four individual exposures of the search
495: images. The SNR measured
496: for a supernova candidate in each of these exposures (SNR$_{\rm exposure}$)
497: should be at least 3. A false positive resulting from cosmic rays will likely
498: not be present in every individual exposure.
499: \item
500: We then subtract each of the individual exposures from the reference
501: image at the location of the supernova candidate
502: and compare the signal to the quadratic sum of the noise. The difference
503: in these SNRs between the exposures must be $<$ 3 for at least 3 out of 4 exposures. We are thus allowing one
504: (and only one) of the four exposures of the search image
505: to be contaminated by a cosmic ray.
506: \end{itemize}
507: \end{enumerate}
508:
509: These cuts eliminate close to 90\% of false detections
510: (\emph{i.e.}, the number of detections decreases from $\sim$100
511: per single tile (see Fig.~\ref{fig:goods}) to $\sim$10).
512: Obvious image processing artifacts or cosmic rays that manage to
513: pass these cuts are rejected by manual screening
514: (typically, there would be a few such candidates per tile,
515: mostly image processing artifacts), with any
516: questionable candidates left in the sample.
517: The efficiency of the manual
518: scan has been checked using a sample of $\sim$100 fake supernovae,
519: generated as described below, and 100\% were correctly identified.
520: The preliminary selection flags any variable objects -- supernovae of various types, as
521: well as active galactic nuclei (AGNs), \emph{etc}. In section~\ref{sec:typeia},
522: we describe our approach to selecting Type Ia supernovae from the sample.
523:
524:
525: In order to measure the efficiency of the selection, we used a Monte Carlo simulation
526: that puts fake supernovae on real F850LP images.
527: Fake supernovae were also used to develop the selection cuts listed above
528: in an unbiased way. The technique follows the
529: approach outlined in~\cite{bib:pain1} and works as follows.
530:
531: First, we run SExtractor~\citep{bib:sex} v2.3 on the search images
532: that have been combined, or drizzled, together from the individual exposures.
533: We do this for a number of both North and South GOODS tiles.
534: Using SExtractor's classification
535: of objects as galaxies and stars, we create a list of the galaxy positions
536: on the image. Because in our analysis we are ignoring candidates near image
537: edges, the galaxies located within 2 galaxy full widths at half maximum (also
538: determined by SExtractor) from the image boundaries are discarded.
539: The fake supernova that is to be put on the image is randomly
540: assigned a magnitude that is drawn from a flat distribution between 23 to 30.
541: The supernova's position is drawn from a Gaussian distribution
542: with half the galaxy's full width at half maximum as the
543: standard deviation and centered on the galaxy's nominal center.
544: We then use STSDAS\footnote[2]{STSDAS and PyRAF are products of the Space
545: Telescope Science Institute, which is operated by AURA for NASA}
546: \emph{tranback} function to convert the fake supernova positions on the
547: drizzled images into coordinates on the raw individual exposures.
548: Fake supernovae themselves are created using the TinyTim software~\citep{bib:tt}, for
549: the ACS WFC1 camera, in filter F850LP. The fake supernova signal,
550: combined with a noise generated using a Poisson distribution with the signal's mean
551: for each pixel,
552: is added onto the input exposures, which are subsequently processed in exactly the same way as real data
553: are.
554:
555:
556: We generated $\sim$13,000 fake supernovae (the 100 supernovae
557: used for the check of the manual scanning efficiency were a subset of this
558: sample).
559: %In the process of creating subtractions, the software performs aperture photometry
560: %at the locations of the fake supernovae, thus extracting
561: %signal-to-noise and other parameters for each of the input fake supernovae.
562: The fake supernovae that pass the stage 1 selection cuts described
563: above are compared with the input list of fakes.
564: This allows us to calculate the efficiency of the selection cuts
565: for the preliminary supernova selection. This efficiency
566: is shown in Fig.~\ref{fig:eff} (upper left) as a function of the candidates' SNR,
567: and in Fig.~\ref{fig:eff} (upper right) as a function of the candidates' magnitude,
568: on the reference-subtracted search images.
569: Note that our reference images are not uniformly deep: they consist of 2, 4, or 5
570: combined epochs, depending on the tile of the GOODS field and the supernova's
571: position on the tile (see Fig.~\ref{fig:goods}).
572: Figure~\ref{fig:eff} (lower left) shows the supernova finding efficiency as a function
573: of the SNR for two representative cases: i) for all locations where
574: two epochs contribute to the reference data; and ii) for all locations
575: where there are four epochs that are available for the reference data.
576: We refer to these cases as ``depth 2'' and ``depth 4'', respectively.
577: It is evident that, within errors, for a given SNR, the efficiency
578: is independent of the depth of the reference image at the location of the
579: fake supernovae, as it should be. We thus use the efficiency curve in Fig.~\ref{fig:eff} (upper left)
580: that combines all of the depths, which we fit to the following four-parameter function:
581: \begin{equation}
582: \epsilon(SNR) = p_1 + \frac{p_2}{1+e^{p_3 \, (SNR - p_4)} }
583: \label{eqn:eff}
584: \end{equation}
585: where we obtain $p_1$ = 0.96, $p_2$ = $-$18.04, $p_3$ = 0.41, and $p_4$ = $-$1.34.
586: The resulting fit is also shown in Fig.~\ref{fig:eff} (upper left).
587: %%
588: \begin{figure}[!htb]
589: \begin{center}
590: $\begin{array}{c@{\hspace{0.0in}}c}
591: \epsfxsize=3.0in\epsffile{f2a.eps} &
592: \epsfxsize=3.0in\epsffile{f2b.eps} \\
593: \epsfxsize=3.0in\epsffile{f2c.eps}&
594: \epsfxsize=3.0in\epsffile{f2d.eps} \\
595: \end{array}$
596: \end{center}
597: \caption{
598: The efficiency of the stage 1 supernova selection
599: in the reference-subtracted search images.
600: The errors are assigned using binomial statistics.
601: The upper left plot shows the combined efficiency for all depths (see text for the definition of depth)
602: of the reference image as a function of the candidate's SNR.
603: Overlaid as a solid line is the fit of the efficiency to the function in Eqn.~\ref{eqn:eff}.
604: The upper right plot shows the same as a function of the candidates' magnitude.
605: The lower left plot shows the efficiencies for two representative depths: 2 (squares)
606: and 4 (down turned triangles). The lower right plot
607: shows the efficiency as a function of the supernova distance from the host galaxy core for
608: all candidates with SNR $>$ 15; the insert shows the efficiency for the candidates
609: with SNR $\leq$ 15.
610: }
611: \label{fig:eff}
612: \end{figure}
613: %%%%%%%
614:
615: One concern in supernova searches is the potential loss of candidates
616: located close to the core of their host galaxies.
617: Figure~\ref{fig:eff} (lower right) shows the efficiency as a
618: function of the supernova's distance from the galaxy core. It is apparent that the efficiency remains
619: essentially flat.
620:
621:
622: %%%%%%
623: \subsection{Stage 2: The Multi-Epoch Selection}
624: \label{sec:multibandsearch}
625: %%%%%
626: The second stage of the supernova candidate selection is where we turn
627: to the multi-epoch photometric data in both filters.
628: Subtracting the stacked image of each epoch of the
629: search data from the reference data,
630: we calculate the candidates' SNRs in the subtracted data
631: and require that there be at least three epochs with a SNR $>$ 2, including at least
632: two epochs with the SNR $>$ 3. These cuts are designed to select
633: candidates with reasonably well-measured light curves.
634: Because the Bayesian technique described in~\ref{sec:typeia} provides
635: a powerful discrimination of Type Ia supernovae, these cuts can be very loose.
636: At the end of stage 2, we have 26 candidates in sample \#1,
637: 17 candidates in sample \#2, 9 candidates in sample \#3, and 5 candidates in sample \#4,
638: for a total of 57 candidates. A list of these candidates is given in
639: Tables~\ref{tab:57cands1},~\ref{tab:57cands2},~\ref{tab:57cands3},
640: and~\ref{tab:57cands4}, for samples \#1, \#2, \#3, and \#4, respectively.
641: The tables specify the supernova names and classifications (gold, silver, or bronze)
642: for the candidates that were also found in~\cite{bib:riess2} and in~\cite{bib:riess06}.
643: The classification refers to
644: the degree of belief in the typing of the candidate, with gold being certain.
645: For sample \#1, we have 8 gold and 1 silver Ia's, and 3 silver core-collapse (CC) supernovae.
646: % missing are 2003eu (Lennon, silver Ia), 2003bb (Raven, silver CC), 2003er (Janis, silver CC)
647: % and 2003et (Jimi, silver CC)
648: For samples \#3 and \#4, we have 5 gold and 2 silver Ia's, and 1 gold and 1 silver CC.
649: % missing are 2002lg (Prometheus, gold Ia), 2002fx (Athena, silver Ia) for South ascending
650: % and 2003ak (Gilgamesh, gold Ia), 2003al (Enki,silver Ia) for South descending%
651: % see bookkeping/strolger_north.dat and strolger_south.txt
652: There were 6 additional gold and silver Ia's found in~\cite{bib:riess2}
653: that failed our stage 2 cuts (SN-2003eu, SN-2002lg, SN-2002fx, SN-2003ak, SN-2003eq, and SN-2003al)
654: because they did not have a sufficient number of epochs with high enough SNR.
655: In other words, these candidates fall below the threshold that is intentionally
656: set high enough that an automated Bayesian classification of candidates
657: (discussed in section~\ref{sec:typeia}) may be possible. Note also that
658: the failure of real SNe Ia to pass stage 2 cuts is taken into account in the
659: control time calculation (Section~\ref{sec:ctandarea}).
660:
661:
662:
663: Spectroscopic redshifts (of the host, the SN or both) were taken from
664: the following sources:~\cite{bib:strolger},~\cite{bib:riess2},
665: ~\cite{bib:cohen},~\cite{bib:cowie},~\cite{bib:keck},~\cite{bib:lefevre},~\cite{bib:vanzella},
666: and~\cite{bib:riess06}.
667: In some cases, the spectroscopic redshift has been determined more than once. We find good
668: agreement in such cases. If a spectroscopic redshift was not available,
669: we used photometric redshifts from~\cite{bib:wolf},~\cite{bib:strolger},
670: and~\cite{bib:private}.
671:
672: The host galaxies of three candidates
673: (candidates \#9 and \#25 in Table~\ref{tab:57cands1} and candidate \#12 in Table~\ref{tab:57cands2}) were observed with the Subaru Faint Object Camera and
674: Spectrograph (FOCAS;~\cite{bib:kash}) on May 17, 2007
675: All three host galaxies
676: were observed with the 300R grism and the SO58 order sorting filter,
677: resulting in spectra covering the 5800-10000\,\AA\ spectral region with
678: a resolving power of $\sim$ 300. Single emission lines were detected in the first
679: two galaxies. If these lines are due to the [OII] doublet at 3727 \AA, then
680: the redshifts of these sources are $z$ = 1.143 $\pm$ 0.001 and $z$ = 0.618 $\pm$ 0.001,
681: respectively. The first measurement confirms the redshift reported in~\cite{bib:strolger}.
682: The second measurement is new. Although the continuum of the third galaxy was
683: detected, no clear spectral features are apparent, so we used the photometric redshift instead.
684:
685:
686:
687: The typical error in the redshift that is measured spectroscopically is
688: 0.001, if the redshift was determined from host galaxy lines, or 0.01, if
689: the redshift was determined from supernova features. For photometric
690: redshifts, the error is larger, ranging from 0.05 to as high as 0.4.
691: The source of the redshift errors is listed in the tables as well.
692: Precision photometric measurements for previously unpublished candidates will be
693: made available in~\cite{bib:suzuki}.
694: %%%
695: \begin{table}[htbp]
696: \begin{center}
697: \tiny
698: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
699: \hline\hline
700: % \multicolumn{6}{|c||}{Sample 1} \\
701: % \hline
702: Cand & RA (2000.) & Dec (2000.) & Redshift & Error & Source & Ref & Comment & $P({\rm Ia}|\{D_i\},z)$\\
703: \hline\hline
704: 1 & 12:37:06.938 & +62:09:15.81 & 0.53 & 0.25 & host phot & M\&D & & 1.0 \\
705: 2 & 12:37:01.537 & +62:11:28.66 & 0.778 & 0.001 & host spec & C04 & & 1.0 \\
706: 3 & 12:36:56.336 & +62:11:55.65 & 0.83 & 0.10 & host phot & M\&D & & 1.0 \\
707: 4 & 12:37:49.350 & +62:14:05.71 & 0.41 & 0.01 & spec & S04 & silver CC (SN-2002kl) & 0.3 \\
708: %2n got de-weighted, and this was its dominant probability
709: 5 & 12:36:21.291 & +62:11:01.24 & 0.633 & 0.001 & host spec & TKRS & & 0.9 \\
710: 6 & 12:37:08.396 & +62:14:23.98 & 0.564 & 0.001 & host spec & TKRS & & 0.9 \\
711: 7 & 12:37:40.658 & +62:20:07.42 & 0.741 & 0.001 & host spec & TKRS & & 1.0 \\
712: 8 & 12:36:16.850 & +62:14:37.30 & 0.71 & 0.05 & host phot & S04 & bronze Ia (SN-2002kh) & 1.0 \\
713: 9 & 12:37:28.421 & +62:20:39.56 & 1.141 & 0.001 & (host + SN) spec & R04 & gold Ia (SN-2002ki)& 1.0 \\
714: 10 & 12:36:38.130 & +62:09:52.88 & 0.513 & 0.001 & host spec & S04, TKRS & silver CC (SN-2003bc) & 0.0 \\
715: 11 & 12:37:25.126 & +62:13:16.98 & 0.67 & 0.01 & SN spec & R04 & gold Ia (SN-2003bd) & 1.0 \\
716: 12 & 12:36:24.506 & +62:08:34.84 & 0.954 & 0.001 & host spec & S04, TKRS & silver CC (SN-2003bb) & 0.8 \\
717: %2n got de-weighted, and this was its dominant probability
718: 13 & 12:36:27.828 & +62:11:24.71 & 0.66 & 0.05 & host phot & S04 & bronze CC (SN-2003ew) & 1.0 \\
719: 14 & 12:37:19.723 & +62:18:37.23 & 1.27 & 0.01 & SN spec & R04 & gold Ia (SN-2003az) & 1.0 \\
720: 15 & 12:37:15.208 & +62:13:33.55 & 0.899 & 0.001 & (host + SN) spec & R04, TKRS & gold Ia (SN-2003eb) & 0.0 \\
721: 16 & 12:36:55.441 & +62:13:11.46 & 0.954 & 0.001 & (host + SN) spec & R04, TKRS & gold Ia (SN-2003es) & 1.0 \\
722: 17 & 12:36:33.179 & +62:13:47.34 & 0.54 & 0.05 & host phot & S04 & bronze Ia (SN-2003en) & 0.9 \\
723: 18 & 12:36:57.900 & +62:17:23.24 & 0.529 & 0.001 & host spec & TKRS & & 1.0 \\
724: 19 & 12:36:39.967 & +62:07:52.12 & 0.48 & 0.05 & host phot & S04 & bronze CC (SN-2003dz) & 0.9 \\
725: 20 & 12:36:31.772 & +62:08:48.25 & 0.46 & 0.05 & host phot & S04 & bronze CC (SN-2003dx) & 0.0 \\
726: 21 & 12:37:28.992 & +62:11:27.36 & 0.935 & 0.001 & host spec & S04, TKRS & silver Ia (SN-2003lv) & N/A \\
727: %residual cosmic ray contamination
728: 22 & 12:37:09.189 & +62:11:28.17 & 1.340 & 0.001 & (host + SN) spec & R04, TKRS & gold Ia (SN-2003dy) & 1.0\\
729: 23 & 12:37:12.066 & +62:12:38.04 & 0.89 & 0.05 & host phot & S04 & bronze CC (SN-2003ea) & 0.4 \\
730: 24 & 12:36:15.925 & +62:12:37.38 & 0.286 & 0.001 & host spec & S04, TKRS & bronze CC (SN-2003ba) & N/A \\
731: 25 & 12:36:26.718 & +62:06:15.16 & 0.618 & 0.001 & host spec & this paper & & N/A \\
732: 26 & 12:36:26.013 & +62:06:55.11 & 0.638 & 0.001 & (host + SN) spec & R04, TKRS & gold Ia (SN-2003be) & 1.0 \\
733: \hline\hline
734: \end{tabular}
735: \end{center}
736: \caption[]
737: {\label{tab:57cands1}
738: The candidates selected at the end of stage 2 for sample \#1.
739: Listed are the candidates' coordinates, redshifts, errors on the redshifts,
740: the sources used for the redshift and redshift error determination, the references
741: for the sources,
742: and $P({\rm Ia}|\{D_i\},z)$ defined in section~\ref{sec:typeia}
743: (the ``N/A'' stands for a special category of candidates designated
744: as ``anomalies'', as described in section~\ref{sec:typeia}).
745: For the candidates found in~\cite{bib:riess2}, the tables also list the
746: supernovae' name and classification (gold, silver, or bronze).
747: C00 is~\cite{bib:cohen},
748: H03 is~\cite{bib:horsh},
749: CO4 is~\cite{bib:cowie},
750: VVDS is~\cite{bib:lefevre},
751: M\&D is~\cite{bib:private},
752: S04 is~\cite{bib:strolger},
753: R04 is~\cite{bib:riess2},
754: TKRS is~\cite{bib:keck},
755: W04 is~\cite{bib:wolf},
756: F2 is~\cite{bib:vanzella}, and
757: R07 is~\cite{bib:riess06}.
758: }
759: \normalsize
760: \end{table}
761: %%%%%%%%%%%%%%%%
762: \begin{table}[htbp]
763: \tiny
764: \begin{center}
765: \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|}
766: \hline\hline
767: % \multicolumn{6}{|c||}{Sample 2} \\
768: % \hline
769: Cand & RA (2000.) & Dec (2000.) & Redshift & Error & Source & Ref & Comment & $P({\rm Ia}|\{D_i\},z)$\\
770: \hline\hline
771: 1 & 12:36:20.889 & +62:10:19.24 & 1.10 & 0.28 & host phot & M\&D & & 1.0\\
772: 2 & 12:36:29.474 & +62:11:41.40 & 1.35 & 0.40 & host phot & M\&D & & 0.0 \\
773: 3 & 12:36:19.901 & +62:13:47.67 & 0.535 & 0.001 & host spec & TKRS & & 0.0 \\
774: 4 & 12:36:27.131 & +62:15:09.27 & 0.794 & 0.001 & host spec & TKRS & & 0.0 \\
775: 5 & 12:36:32.238 & +62:16:58.38 & 0.437 & 0.001 & host spec & TKRS & & 0.4\\
776: 6 & 12:38:03.689 & +62:17:12.23 & 0.280 & 0.001 & host spec & CO4 & & 0.4 \\
777: 7 & 12:37:09.495 & +62:22:15.37 & 1.61 & 0.34 & host phot & M\&D & & 1.0\\
778: 8 & 12:37:06.772 & +62:21:17.46 & 0.406 & 0.001 & host spec & TKRS & & 0.1 \\
779: 9 & 12:36:26.694 & +62:08:29.74 & 0.555 & 0.001 & host spec & TKRS & & 0.8 \\
780: 10 & 12:36:54.125 & +62:08:22.21 & 1.39 & 0.01 & SN spec & R07 & gold Ia (HST04Sas) & 1.0 \\
781: 11 & 12:36:34.363 & +62:12:12.55 & 0.457 & 0.001 & (host + SN) spec & TKRS, R07 & gold Ia (JST04Yow) & 1.0 \\
782: 12 & 12:37:33.918 & +62:19:21.75 & 0.88 & 0.38 & host phot & M\&D & & 1.0 \\
783: 13 & 12:36:34.853 & +62:15:48.86 & 0.855 & 0.001 & (host + SN) spec & R07, TKRS & gold Ia (HST04Man) & 1.0 \\
784: 14 & 12:36:36.009 & +62:17:31.97 & 0.60 & 0.15 & host phot & M\&D & & 0.2 \\
785: 15 & 12:36:55.214 & +62:13:03.75 & 0.952 & 0.004 & (host + SN) spec & C00, R07 & gold Ia (HST04Tha) & 1.0 \\
786: 16 & 12:37:48.435 & +62:13:34.85 & 0.839 & 0.001 & host spec & TKRS & & 1.0 \\
787: 17 & 12:36:01.542 & +62:15:55.16 & 0.086 & 0.001 & host spec & H03 & & N/A \\
788: \hline\hline
789: \end{tabular}
790: \end{center}
791: \caption[]
792: {\label{tab:57cands2}
793: Same as Table~\ref{tab:57cands1} for sample \#2. The redshift of candidate \#17 is uncertain,
794: as the possible host galaxy is 7$\arcsec$ away. Leaving this redshift as unconstrained does
795: not change our results.
796: }
797: \normalsize
798: \end{table}
799: %%%%%%%%%
800: \begin{table}[htbp]
801: \tiny
802: \begin{center}
803: \begin{center}
804: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
805: \hline\hline
806: % \multicolumn{6}{|c||}{Sample 3} \\
807: % \hline
808: Cand & RA (2000.) & Dec (2000.) & Redshift & Error & Source & Ref & Comment& $P({\rm Ia}|\{D_i\},z)$\\
809: \hline\hline
810:
811: 1 & 03:32:18.072 & -27:41:55.83 & 0.88 & 0.05 & host phot & S04 & silver Ia (SN-2002fy) & 0.9 \\
812: 2 & 03:32:13.002 & -27:42:05.75 & 0.421 & 0.001 & host spec & VVDS & & 0.0 \\
813: 3 & 03:32:37.511 & -27:46:46.40 & 1.30 & 0.01 & SN spect & R04 & gold Ia (SN-2002fw) & 1.0\\
814: 4 & 03:32:05.060 & -27:47:02.96 & 0.976 & 0.001 & host spec & VVDS & & N/A \\
815: 5 & 03:32:17.309 & -27:46:23.74 & 0.13 & 0.01 & phot & W04 & & 0.0 \\
816: 6 & 03:32:48.598 & -27:54:17.14 & 0.841 & 0.001 & host spec & S04, VVDS & silver CC (SN-2002fz) & 0.9 \\
817: %2n got de-weighted
818: 7 & 03:32:22.751 & -27:51:09.65 & Unconstrained & Unconstrained & phot & S04 & bronze CC (SN-2002fv) & 0.0 \\
819: 8 & 03:32:42.441 & -27:50:25.08 & 0.58 & 0.01 & spec & S04 & gold CC (SN-2002kb) & N/A \\
820: 9 & 03:32:38.082 & -27:53:48.15 & 0.987 & 0.001 & host spec & S04, VVDS & bronze Ia (SN-2002ga) & 1.0 \\
821: \hline\hline
822: \end{tabular}
823: \end{center}
824: \end{center}
825: \caption[]
826: {\label{tab:57cands3}
827: Same as Table~\ref{tab:57cands1} for sample \#3. The redshift of candidate \#4 is uncertain,
828: as the possible host galaxy is 4$\arcsec$ away. Leaving this redshift as unconstrained does
829: not change our results.
830: }
831: \normalsize
832: \end{table}
833: %%%%%%%%%%%%%%
834: \begin{table}[htbp]
835: \tiny
836: \begin{center}
837: \begin{tabular}{|c|c|c|c|c|c|c|c|c|}
838: \hline\hline
839: % \multicolumn{6}{|c||}{Sample 4} \\
840: % \hline
841: Cand & RA (2000.) & Dec (2000.) & Redshift & Error & Source & Ref & Comment & $P({\rm Ia}|\{D_i\},z)$\\
842: \hline\hline
843: 1 & 03:32:24.782 & -27:46:18.07 & 1.306 & 0.001 & host spec & R04, F2 & gold Ia (SN-2002hp) & 1.0\\
844: 2 & 03:32:22.522 & -27:41:52.26 & 0.526 & 0.001 & (host + SN) spec & R04 & gold Ia (SN-2002hr) & 1.0\\
845: 3 & 03:32:22.318 & -27:44:27.04 & 0.738 & 0.001 & host spec & R04, F2 & gold Ia (SN-2002kd) & 1.0\\
846: 4 & 03:32:05.382 & -27:44:29.76 & 0.91 & 0.05 & host phot & S04 & silver Ia (SN-2003al) & 1.0 \\
847: 5 & 03:32:34.648 & -27:39:58.18 & 0.214 & 0.001 & (host + SN) spec & R04, VVDS & gold Ia (SN-2002kc) & 1.0\\
848: \hline\hline
849: \end{tabular}
850: \end{center}
851: \caption[]
852: {\label{tab:57cands4}
853: Same as Table~\ref{tab:57cands1} for sample \#4.
854: }
855: \normalsize
856: \end{table}
857:
858: Note that the redshifts of candidate \#17 in Table~\ref{tab:57cands2} and
859: candidate \#4 in Table~\ref{tab:57cands3} are uncertain, since the assumed host galaxies
860: of the supernova candidates are 7$\arcsec$ and 4$\arcsec$ away, respectively. However, we have verified
861: that if we leave the redshifts of these candidates as unconstrained, it does not affect
862: our final results.
863:
864:
865:
866: %%%%%%
867: \subsection{Stage 3: The Identification of Type Ia Supernovae}
868: \label{sec:typeia}
869: %%%%%%
870: The candidates that have been selected in stages 1 and 2 are assumed to be
871: real transient objects, most likely supernovae, and must now be classified by type.
872: With only scarce photometric data available, we turn to the Bayesian
873: method of classifying supernovae described in~\cite{bib:ourpaper}.
874:
875: Photometric typing of supernovae has been described
876: in~\cite{pozn02},~\cite{riess2004},~\cite{john} and~\cite{sull1}, among others
877: Most of the existing methods rely on color-color or color-magnitude
878: diagrams for supernova classification.
879:
880: In our method, we consider five possible supernova
881: types (``normal'' Ia~\citep{bib:branch}, Ibc, IIL, IIP, and IIn). We make use of the best
882: currently available supernova multicolor lightcurve templates for each type. When improved
883: supernova templates are available, they can be easily worked into the method.
884: We calculate the probability that a given supernova candidate with photometric data $\{D_i\}$,
885: where $i$ is the index for the number of observational epochs, and redshift $z$ is
886: a Type Ia supernova. By virtue of the Bayes theorem, this probability is given by:
887: \begin{equation}
888: P({\rm Ia}|\{D_i\},\,z)= \frac{ \int_{\vec{\theta}} P(\{D_i\},z|\vec{\theta},{\rm Ia}) P(\vec{\theta},{\rm Ia}) d\vec{\theta}} {\sum_{T} \int_{\vec{\theta}} \, P(\{D_i\},\,z|\vec{\theta},T)P(\vec{\theta},T) d\vec{\theta}}.
889: \label{eqn:bayes}
890: \end{equation}
891: %P({\rm Ia}|\{D_i\}) = \frac{P(\{D_i\} | {\rm Ia}) \, P({\rm Ia})} {\sum_{T} P(\{D_i\} | T) \, P(T) },
892: where $z$ is the measured supernova redshift; $\vec{\theta}$ are the parameters that characterize
893: a given supernova type; $\{D_i\}$ are the data in both F850LP and F775W;
894: $P(\{D_i\}\,z| \vec{\theta}, T)$ is the probability density to obtain data $\{D_i\}$ and redshift $z$
895: for supernova type $T$; $P(\vec{\theta},T)$ contains prior information about type $T$ supernovae;
896: and the denominator contains the normalization (the sum) over all five supernova types $T$ considered.
897: The parameters $\vec{\theta} \equiv (\bar{z}, t_{\rm diff},s,M,R_V,A_V)$
898: are: $\bar{z}$ is the true supernova redshift; $t_{\rm diff}$ is the time difference between
899: the dates of maximum light for the template and the data;
900: $s$ is the stretch parameter~\citep{bib:perl}, which parametrizes the width of the
901: light curve (if $T$ = Ia); $M$ is the absolute magnitude in the restframe $B$-band at maximum light;
902: and $A_V$ and $R_V$ are the Cardelli-Clayton-Mathis
903: interstellar extinction parameters~\citep{bib:ccm}.
904: %($<A(\lambda)/A(V)> = a(x) + b(x)/R_V$
905: We marginalize (integrate over) these parameters as described below.
906:
907:
908: Suppose that we have a photometric template, $\{\bar{D}(\vec{\theta},T)_i\}$, for
909: the expected light curve for a supernova of type $T$ at a given redshift, $\bar{z}$.
910: In this work, we use the templates from P. E. Nugent\footnote[3]{See http://supernova.lbl.gov/$\sim$nugent/nugent\_templates.html},
911: which
912: extend both into the UV (below 3460 \AA\, in the supernova
913: rest frame) and into far red and IR (above 6600 \AA\, in the supernova rest frame) regions.
914: Is it assumed that the measured light curve flux, $\{D_i\}$, can fluctuate from the template $\{\bar{D}(\vec{\theta},T)_i\}$
915: according to Gaussian statistics. It is also assumed that
916: the probability of measuring redshift $z$ fluctuates
917: around a mean $\bar{z}$ according to Gaussian statistics as well. Therefore,
918: \begin{equation}
919: P(\{D_i\},z |\vec{\theta}, T) = \frac{{\rm exp}(-\frac{(z-\bar{z})^2}{2\delta z^2})} {\sqrt{2\pi} \delta z}\prod_{i=1}^{n_{epochs}} \frac{{\rm exp}({-\frac{(\bar{D}(\vec{\theta},T)_i-D_i)^2}{2 \delta D_i^2}}) }{\sqrt{2\pi} \delta D_i },
920: \label{equation:likelihood}
921: \end{equation}
922: where $\delta D_i$ are photometric measurement errors for epoch $i$,
923: and $\delta z$ is the measurement error for the redshift $z$.
924: Note that we assume no errors on the supernova templates themselves; we take them to represent the best currently available
925: knowledge of the supernova behavior. However, it is also worth noting that
926: various parameters that characterize a given template (\emph{e.g.}, the peak
927: restframe $B$-band magnitude, the stretch parameter for Ia's, \emph{etc.}) are varied
928: as described below, thus effectively representing some template variations.
929:
930:
931: The prior $P(\vec{\theta}, T)$ contains all the available information about the
932: behavior of type $T$ supernovae, expressed in terms of parameters ${\vec{\theta}}$.
933: We assume that all constituents of ${\vec{\theta}}$ can be divided as
934: follows where $t_{\rm diff}$, $\bar{z}$ and $T$, $M$, $R_V$ and $A_V$ and $s$ are independent.
935: \begin{equation}
936: P(\vec{\theta} ,T ) = P(t_{\rm diff}|\bar{z},T) \, P(M|\bar{z},T)\, P(s|\bar{z},T)\, P(R_V,A_V|\bar{z},T)\, P(\bar{z},T).
937: \label{equation:prior}
938: \end{equation}
939: The assumed independence of the parameters is certainly an
940: oversimplification. For example, one would expect the stretch and magnitude parameters to be
941: correlated (although the true values of these two parameters should be independent of
942: $t_{\rm diff}$, $R_V$ and $A_V$). Ignoring the correlation
943: might conceivably lead to an overestimation of the probabilities
944: for very bright Type Ia's with a small stretch parameter, or
945: very dim Type Ia's with a large stretch parameter. However,
946: we are exploring every possible combination of stretch and magnitude
947: parameters; the ``correct'' combination should naturally be
948: a better ``fit'' to the data, thus acquiring a larger weight
949: than all the other ones.
950:
951:
952:
953: The prior $P(\bar{z},T)$ includes the relative rates of the various supernova
954: types as a function of redshift. Unfortunately, these rates are not well known, especially at high redshift.
955: We will thus consider three different models for the ratio of the CC supernova rates
956: to the Ia supernova rates. The models are based on~\cite{bib:dahlen2}, and
957: shown in Fig.~\ref{fig:ratio_models}. They correspond to three
958: different values of the characteristic time delay parameter $\tau$:
959: $\tau$ = 1 Gyr, 2 Gyr, and 3 Gyr.
960: Based on~\cite{bib:dahlen2}, we will also assume
961: that the relative (rounded-off) fractions of the CC supernovae are
962: $f_{Ibc}$ = 0.27, $f_{IIL}$ = 0.35, $f_{IIp}$ = 0.35, and $f_{IIn}$ = 0.02,
963: for all three models, regardless of the redshift.
964: %%%%%%%%%%%%%%%%
965: \begin{figure}[!htb]
966: \begin{center}
967: \includegraphics[height=4.0in, width=5.0in]{f3.ps}
968: \end{center}
969: \caption[]
970: {\label{fig:ratio_models}
971: The three models for the ratio of the CC/Ia rates as a function of redshift based on~\cite{bib:dahlen2}.
972: The solid line is for the time delay parameter $\tau$ = 1 Gyr; the dashed line is
973: for $\tau$ = 2 Gyr; and the dashed-dotted line is for $\tau$ = 3 Gyr.
974: }
975: \end{figure}
976:
977:
978:
979: Note that the usage of these models does not bias our answer in any way, as we are not making
980: any assumptions about the \emph{absolute} rates of supernovae, but only about their relative rates.
981: If we assume all three models to be equally likely, then
982: the probability density $P(T)$ of observing a supernova of type $T$ for assumption $n$
983: about the relative rates of the CC to Ia supernovae is given by:
984: \begin{equation}
985: P(\bar{z},T) = \frac{R_n(\bar{z},T)}{\sum_{l=1}^{N_{models}} R_l(\bar{z},T)}
986: \end{equation}
987: where $R_n(\bar{z},T)$ is the rate of type $T$ supernovae
988: for model $n$, and $N_{models}$ = 3.
989:
990:
991: The difference in the dates of maximum light between the template and the data,
992: $t_{\rm diff}$, can also take on any value, making the prior
993: $P(t_{\rm diff}|T)$ flat. In practice, we shift the relative dates of maximum between
994: the measured and the template light curves by increments of one day.
995: The marginalization of this parameter thus amounts to a sum
996: over a finite number (which we take to be 160) of such shifts.
997: \begin{equation}
998: P(t_{\rm diff}|\bar{z},T) = \frac{1}{{t^{max}_{\rm diff}-t^{min}_{\rm diff}}},
999: \end{equation}
1000: where the maximum $t^{max}_{\rm diff}$ and minimum $t^{min}_{\rm diff}$ set the limits on $t_{\rm diff}$.
1001:
1002:
1003: The priors on $P(M|\bar{z}, T)$ and $P(s|\bar{z}, {\rm Ia})$ are taken to be Gaussian:
1004: \begin{equation}
1005: P(M|\bar{z},T) =\frac{e^{-\frac{(M-\bar{M})^2}{2 \delta M^2}}}{\sqrt{2 \pi} \delta M}.
1006: \label{eqn:magrioria}
1007: \end{equation}
1008: \begin{equation}
1009: P(s|\bar{z},{\rm Ia}) =\frac{e^{-\frac{(s-\bar{s})^2}{2 \delta s^2}}}{\sqrt{2 \pi} \delta s}
1010: \label{eqn:sprioria}
1011: \end{equation}
1012: A table of the mean magnitudes $\bar{M}$ and the standard deviations $\delta M$,
1013: as well as the values for the mean stretch $\bar{s}$ and the standard deviation $\delta s$,
1014: are given in~\cite{bib:ourpaper}. For reference, we extract the mean magnitudes
1015: $\bar{M}$ in the restframe $B$-band from
1016: P. E. Nugent\footnote[4]{See http://supernova.lbl.gov/$\sim$nugent/nugent\_templates.html},
1017: and the standard deviations, $\delta M$, from~\cite{bib:rich}.
1018: The stretch parameters are extracted from~\cite{bib:sull}.
1019: Note that for non-Ia's, a complete set of ``virtual'' values for the stretch parameters
1020: are inserted into Eqn.~\ref{eqn:sprioria} and then marginalized with a flat
1021: prior (see Appendix B in~\cite{bib:ourpaper}).
1022:
1023: The effects of interstellar extinction are difficult to parametrize
1024: due to lack of generally accepted models for the behavior of the
1025: Cardelli-Clayton-Mathis parameters $A_V$ and $R_V$. We compromise
1026: by considering a case of no extinction and
1027: two cases of extinction with a moderate value of $A_V$ = 0.4 and two
1028: different values of $R_V$, 2.1 and 3.1. The mathematical framework used
1029: in the analysis easily allows for the implementation of real distributions
1030: for $A_V$ and $R_V$, once they become standardized.
1031: It is known that in simulations $A_V$ is sharply peaked near 0 (\emph{e.g.},~\cite{bib:hatano};~\cite{bib:riel}); therefore, not considering
1032: very large values of $A_V$ is reasonable.
1033: All three cases ($N_V = 3$) are considered
1034: equally possible. In other words, we take:
1035: \begin{eqnarray}
1036: P(R_V,A_V|\bar{z},T) = \frac{1}{N_V}
1037: \end{eqnarray}
1038: It is certainly a simplified extinction model; however, it appears to be sufficient
1039: as demonstrated by the largely successful typing of known Type Ia candidates
1040: in two such diverse samples as the 73 SNLS-identified~\citep{bib:snls}
1041: Type Ia's and the gold and silver Ia's in the HST GOODS data~\citep{bib:ourpaper}. The method correctly
1042: identified 69 out of the 73 SNLS Type Ia's. For the remaining four candidates,
1043: at least one filter band included wavelengths outside of the well-understood
1044: optical range in the supernova rest frame.
1045: It also correctly identified 7 out of 8 gold and silver Ia's,
1046: and 5 out of 5 gold and silver CC's. Another consideration
1047: to note here is that extinction primarily affects the measured magnitudes, and
1048: our model already takes into account wide variations in the magnitudes
1049: (Eqn.~\ref{eqn:magrioria}).
1050:
1051:
1052: Putting everything together, we see that the numerator of Eqn.~\ref{eqn:bayes} is given by:
1053: \begin{small}
1054: \begin{eqnarray}
1055: \label{eqn:final_Ia}
1056: & & \int_{\vec{\theta}} \, P(\{D_i\},z| \vec{\theta}, {\rm Ia}) P(\vec{\theta}, {\rm Ia}) d{\vec{\theta}} = \nonumber \\
1057: & & \sum_{\bar{z}=z_{min}}^{z_{max}} \frac{\Delta \bar{z}}{\sqrt{2 \pi} \delta z} \, e^{-\frac{(z-\bar{z})^2}{2 \delta z^2}} \, \frac{1}{\sum_{l=1}^{N_{models}} R_l(\bar{z},T)}\sum_{n=1}^{N_{models}} R_n(\bar{z},{\rm Ia}) \, \frac{1}{N_v}\sum_{n_v=1}^{N_V}\Delta N_v\sum_{M=M_{min}}^{M_{max}} \frac{\Delta M}{\sqrt{2 \pi} \delta M} \, e^{-\frac{(M-\bar{M})^2}{2 \delta M^2}} \,\frac{\Delta t_{\rm diff}} {{t^{max}_{\rm diff}-t^{min}_{\rm diff}}} \nonumber \\
1058: & & \sum_{s=s_{min}}^{s_{max}} \frac{\Delta s}{\sqrt{2 \pi} \delta s} e^{-\frac{(s-\bar{s})^2}{2 \delta s^2}} \, \prod_{i=1}^{n_{epochs}} \frac{{\rm exp}({-\frac{(\bar{D}_{j}-D_i)^2}{2\delta D_{i}^2}})}{\sqrt{2 \pi} \delta D_i}
1059: \end{eqnarray}
1060: \end{small}
1061: for Ia's, and for types $T'$ that are non-Ia's, it is:
1062: \begin{small}
1063: \begin{eqnarray}
1064: \label{eqn:final_nonIa}
1065: & & \int_{\vec{\theta}} \, P(\{D_i\},z| \vec{\theta}, T') P(\vec{\theta}, T') d{\vec{\theta}} = \nonumber \\
1066: & & \sum_{\bar{z}=z_{min}}^{z_{max}} \frac{\Delta \bar{z}}{\sqrt{2 \pi} \delta z} \, e^{-\frac{(z-\bar{z})^2}{2 \delta z^2}}\,
1067: \frac{1}{\sum_{l=1}^{N_{models}} R_l(\bar{z},T)}\sum_{n=1}^{N_{models}} R_n(\bar{z},T') \, \frac{1}{N_v}\sum_{n_v=1}^{N_v}\Delta N_v\sum_{M=M_{min}}^{M_{max}} \frac{\Delta M}{\sqrt{2 \pi} \delta M} \, e^{-\frac{(M-\bar{M})^2}{2 \delta M^2}} \,\frac{\Delta t_{\rm diff}} {{t^{max}_{\rm diff}-t^{min}_{\rm diff}}} \nonumber \\
1068: & & \prod_{i=1}^{n_{epochs}} \frac{{\rm exp}({-\frac{(\bar{D}_{j}-D_i)^2}{2\delta D_{i}^2}})}{\sqrt{2 \pi} \delta D_i}.
1069: \end{eqnarray}
1070: \end{small}
1071: In Eqns.~\ref{eqn:final_Ia} and~\ref{eqn:final_nonIa}, we marginalize over parameters $\vec{\theta}$,
1072: approximating the integration by summation.
1073: The range of redshifts [$z_{zmin}$, $z_{max}$] is taken to be from 0 to 1.7 in the
1074: denominator of Eqn.~\ref{eqn:bayes}, and over a bin of interest in the numerator
1075: (this point will be explained in more detail later in this section), and we take $\Delta z$ = 0.05.
1076: The mean values of $z$ and the error on the $z$, $\delta z$, are given in
1077: Tables~\ref{tab:57cands1},~\ref{tab:57cands2},~\ref{tab:57cands3}, and~\ref{tab:57cands4}
1078: for the candidates used in the analysis.
1079: $\Delta t_{\rm diff}$ is one day, and $\Delta N_v$ = 1.
1080: We sum $M$ from $M_{min}$ = -3$\delta M$ to $M_{max}$ = +3$\delta M$ with a total of 12 steps,
1081: and we sum $s$ from $s_{min}$ = 0.65 to $s_{max}$ = 1.3 in 14 steps.
1082: For non-Type Ia's, a complete set of ``virtual'' values for the stretch parameters
1083: are inserted into Eqn.~\ref{eqn:sprioria} and then marginalized with a flat
1084: prior (see Appendix B in~\cite{bib:ourpaper}).
1085:
1086:
1087:
1088: The probability that $\alpha^{\rm th}$ candidate is a Type Ia supernova
1089: belonging to the $j^{\rm th}$ redshift bin,
1090: [$\bar{z}_{j\, lower}$, $\bar{z}_{j \, upper}$], is thus:
1091: \begin{equation}
1092: P_j^\alpha = \frac{ \int_{\bar{z}_{j\, lower}}^{\bar{z}_{j\, upper}}d\bar{z} \, \int_{\vec{\theta}} P(\{D_i\}, z| \vec{\theta}, {\rm Ia}) P(\vec{\theta}, {\rm Ia}) \,d\vec{\theta}} {\sum_{T}\int_{0}^{\infty}d\bar{z} \int_{\vec{\theta}} P(\{D_i\},z| \vec{\theta},T) P(\vec{\theta},T)\,d\vec{\theta}}.
1093: \end{equation}.
1094:
1095: %%%%%%% !!!!!!!
1096: Let us now introduce the following variables:
1097: \begin{itemize}
1098: \item
1099: $N_j$ is the total count of the candidates contributing to the $j^{\rm th}$ redshift bin.
1100: \item
1101: $P_j^{\alpha}$ is the Bayesian probability for each candidate $\alpha$ in the $j^{\rm th}$ redshift bin ($\alpha = [1,...,N_j]$).
1102: \item
1103: $\{P^{\alpha}\}_j$ is the full set of probabilities for the candidates in the $j^{th}$ redshift bin.
1104: \item
1105: $d_j$ is the most likely number of Ia candidates in the $j^{\rm th}$ redshift bin.
1106: \end{itemize}
1107: Our goal is to find $d_j$, as well as the error on this number, given $N_j$ and $P_j^{\alpha}$'s.
1108:
1109:
1110: If $N_j$ is large, say of order 100 (which is the case for our Monte Carlo samples), then
1111: $d_j$ can be simply evaluated as:
1112: \begin{equation}
1113: d_j = \sum_{\alpha = 1}^{N_j} P_j^\alpha,
1114: \label{eqn:numcands}
1115: \end{equation}
1116: where the uncertainty on $d_j$ is given by the square root of the binomial and Poisson variances:
1117: \begin{equation}
1118: \Delta d_j = \sqrt{ \sum_{\alpha = 1}^{N_j} P_j^\alpha (1 - P_{j}^{\alpha}) + \sum_{\alpha = 1}^{N_j} P_j^\alpha}.
1119: \label{eqn:numcandserr}
1120: \end{equation}
1121: Note that if all of the probabilities $P_j^\alpha$
1122: were 1 (\emph{i.e.}, the candidates were all known to be Type Ia supernovae), using
1123: Eqn.~\ref{eqn:numcands} would amount to a simple counting of the number of candidates,
1124: and Eqn.~\ref{eqn:numcandserr} would become the usual $\sqrt{N_j}$ error for a large
1125: number of events $N_j$.
1126:
1127:
1128: For a small number of events, $N_j$ $<$ 10, which is typically the case for our data
1129: samples, using Eqn.~\ref{eqn:numcands} and~\ref{eqn:numcandserr} would be incorrect.
1130: A more sophisticated approach is needed.
1131: Let us define a variable $x_\alpha$ such that $x_\alpha = 1$
1132: if the $\alpha^{\rm th}$ candidate is indeed a Type Ia and $x_\alpha = 0$
1133: if it is not, so that there are $k_j$ $\equiv$ $\sum_{\alpha=1}^{N_j} x_\alpha$
1134: Type Ia's in this bin. The probability to obtain $d_j$ is given by:
1135: \begin{eqnarray}
1136: P(d_j|\{P^\alpha \}_j) = \sum_{\{x_\alpha \}} P(d_j|\{x_\alpha \})P(\{x_\alpha \}|\{P^\alpha \}_j) = \\
1137: \sum_{\{x_\alpha \}} \frac{P(\{x_\alpha \}|d_j)P(d_j)}{\int_{d_{j}=0}^{\infty} P(\{x_\alpha \}|d_j)P(d_j) d(d_j) }P(\{x_\alpha \}|\{P^\alpha \}_j),
1138: %P(d_j) = \frac{\sum_{k_j}P(d_j|k_j)P(k_j)}{ \sum_{d_j} \sum_{k_{j}} P(d_j|k_j)P(k_j) },
1139: \label{eqn:bayes2}
1140: \end{eqnarray}
1141: where the sum on $d_j$ can, in principle, extend to arbitrarily large values
1142: (for example, if $N_j$ = 2, there is still a small but non-zero probability that
1143: $d_j$ can be 100). We will assume a flat prior for $P(d_j)$, in which case the
1144: denominator integrates to unity.
1145:
1146:
1147: The first term in Eqn.~\ref{eqn:bayes2} is a normalized Poisson distribution
1148: for the expected $d_j$ number of events while $k_j=\sum_{\alpha = 1}^{N_j}x_\alpha$ events are assumed to be in the $j^{th}$ bin:
1149: \begin{equation}
1150: P(\{x_\alpha \}|d_j) = \frac{d_j^{k_j}e^{-d_j}}{k_j!}~{\rm,\, where}~k_j=\sum_{\alpha =1}^{N_j}x_\alpha .
1151: \end{equation}
1152: The term $P(\{x_\alpha\}|\{P^\alpha \}_j)$ in Eqn.~\ref{eqn:bayes2} is the probability that
1153: certain supernovae do or do not occupy the $j^{th}$ bin. This probability
1154: is simply:
1155: \begin{equation}
1156: P(\{x_\alpha\}|\{P^\alpha \}_j) = \prod_{\alpha=1}^{N_j} [ P_j^\alpha x_\alpha + (1-P_j^\alpha)(1-x_\alpha)].
1157: %\footnote{For clarity, we express Eqn.\,\ref{eqn:bayes2} in terms of $P(d,k|D)$. However, strictly
1158: %speaking, we should re-express Eqn.\,\ref{eqn:bayes2} as $P(d,\{x_i\}|D)$ and thereby avoid marginalizing parameters in
1159: %the priors and thereby maintain the priors' unitarity. }
1160: \end{equation}
1161:
1162:
1163: Because we have no way of knowing \emph{a priori} which candidate belongs in the
1164: $j^{th}$ bin, we must sum over all possible $\{x_{\alpha}\}$'s:
1165: \begin{equation}
1166: P(d_j|\{P^\alpha\}_j) = \sum_{\{x_\alpha\}}\frac{d^{k_j}e^{-d_j}}{k_j!}\prod_{\alpha=1}^{N_j} [ P_j^\alpha x_\alpha + (1-P_j^\alpha)(1-x_\alpha)].
1167: \label{eqn:thed}
1168: \end{equation}
1169: To obtain the best estimate for $d_j$, we must maximize $P(d_j|\{P^\alpha\} _j)$
1170: given in Eqn.~\ref{eqn:thed}.
1171: In practice, this is done numerically for a range of test $d_j$'s from 0 to some maximum $d_{j\, max}$ (we arbitrarily take it
1172: to be 50) to find out which $d_j$ maximizes the probability.
1173:
1174:
1175: Let us consider an example. Suppose that we have two supernovae in a given bin,
1176: with probabilities of being Ia's given by $P^1$ = 0.8 and $P^2$ = 0.9. The possible permutations
1177: of $x_\alpha$'s would be (0,0), meaning that neither candidate is a Type Ia; (0,1) and (1,0), meaning
1178: that only one candidate is a Type Ia; and (1,1), meaning that both candidates are Ia's.
1179: Then we need to maximize
1180: \begin{equation}
1181: \frac{d^0 e^{-d}}{0!} (1 - 0.8) (1 - 0.9) + \frac{d^1 e^{-d}}{1!} 0.8 (1 - 0.9) + \frac{d^1 e^{-d}}{1!} ( 1 - 0.8) 0.9 + \frac{d^2 e^{-d}}{2!} 0.8 \times 0.9
1182: \end{equation}
1183: as a function of $d$. For this particular example, the best estimate for the number of Type Ia's is in fact $1.68^{+2.62}_{-0.58}$, where the errors are estimated as described below.
1184:
1185:
1186: To evaluate the uncertainty on $d_j$, we
1187: find the 68\% confidence regions for $d_j$, [$d_j-\sigma_{j\,low}$, $
1188: d+\sigma_{j\,high}$], by solving:
1189: \begin{equation}
1190: 16\% = \int_0^{d_j-\sigma_{j\,low}} P(d_j|\{P^\alpha\}_j)d(d_j) = \int_{d+\sigma_{j\,high}}^\infty P(d_j|\{P^\alpha\}_j)d(d_j)
1191: \label{eqn:error}
1192: \end{equation}
1193: In the case where $d_j<<1$, we set $\sigma_{j\,low} = 0$ and find
1194: $\sigma_{j\,high}$ by satisfying:
1195: \begin{eqnarray}
1196: 32\%=\int_{\sigma_{j,high}}^\infty P(d_j|\{P^\alpha\}_j)d(d_j).
1197: \end{eqnarray}
1198:
1199: %%%%%%%%%%%%%%%%%%!!!!!!!
1200: We assume that all candidates whose redshift is within $\pm$3 $\delta z$
1201: of the $j^{th}$ bin's boundaries
1202: (where $\delta z$ is the uncertainty on the candidates' redshift, listed in
1203: Tables~\ref{tab:57cands1},~\ref{tab:57cands2},~\ref{tab:57cands3}, and~\ref{tab:57cands4})
1204: will contribute to this bin.
1205: Note that in this formulation, a single candidate with a poorly known redshift may
1206: have a probability distribution that spans several redshift bins.
1207:
1208: We calculate $P({\rm Ia} |\{D_i\},z)$ for all 57 candidates. If a given
1209: candidate's $P(\{D_i\},z | \vec{\theta},T) \, P(\vec{\theta},T)$
1210: is less than 10$^{\rm -15}$ for all types $T$ ,
1211: it is considered to be an ``anomaly'' and is excluded from further consideration.
1212: The 10$^{\rm -15}$ cut was chosen because it is much smaller than
1213: the values calculated for simulated supernovae in the Monte Carlo.
1214: This method thus excludes any need for the often subjective
1215: and time-consuming decision on whether or not a candidate might be a
1216: supernova of a given type;
1217: all dubious candidates are weighted appropriately and
1218: left in the sample for the probability to decide.
1219:
1220: It is a good sanity check to examine the values of
1221: $P({\rm Ia} |\{D_i\},z)$ for the gold and silver Ia candidates
1222: from~\cite{bib:riess2}.
1223: Tables~\ref{tab:57cands1},~\ref{tab:57cands2},~\ref{tab:57cands3}, and~\ref{tab:57cands4}
1224: list $P({\rm Ia} |\{D_i\},z)$ (with [$\bar{z}_{j\, lower}$, $\bar{z}_{j \, upper}$] = [0.0, 1.7])
1225: for all of the candidates. Several candidates have ``N/A'' listed for
1226: $P({\rm Ia} |\{D_i\},z)$: these are the ``anomalous'' candidates, as described above.
1227: It is apparent that the gold and silver Ia candidates are among the largest contributors
1228: to a given redshift bin. All but one of them, SN-2003eb, have probabilities $\geq$ 0.8.
1229: SN-2003eb has only two epochs (epochs 4 and 5 of the GOODS dataset) with
1230: ``appreciable'' SNR ($>$ 10) in both F775W and F850LP bands.
1231: One silver Ia candidate, SN-2003lv, appears to have a rare residual cosmic ray contamination
1232: in the F775W band,
1233: making it appear inconsistent with any of the supernova types considered.
1234: Three silver core-collapse supernovae, SN-2002kl, SN-2003bb, and SN-2002fz
1235: have the probabilities of being Ia's of 0.3, 0.8, and 0.9 respectively.
1236: They are in fact most consistent with being IIn's; however, because the fraction
1237: of IIn's is heavily de-weighted among CC supernovae ($f_{IIn}$ = 0.02), their resulting
1238: $P({\rm Ia} |\{D_i\},z)$ are higher than one would have expected.
1239: How much do our assumptions about the fractions of various supernova
1240: types among the CC supernovae
1241: influence our answer? As we will see in Section~\ref{sec:priors}, if we assume
1242: that all CC types are equally
1243: likely and that the ratio of the CC to Ia rates is redshift
1244: independent, the changes to our final results are within the quoted
1245: uncertainties.
1246:
1247:
1248:
1249: Another sanity check is to make sure that the candidates with
1250: low $P({\rm Ia} |\{D_i\},z)$'s are not all of a particular class (\emph{e.g.}, Ibc's).
1251: We have verified that indeed they are not.
1252:
1253:
1254: It is worth noting that variable objects other than supernovae, such as AGNs,
1255: are selected during the first selection stage. If some
1256: of these objects also pass the second selection stage, they are
1257: unlikely to bias the results significantly, as the
1258: specifically designed cuts in the third stage would likely
1259: reject such candidates. As an extra check, we verified that none of the candidates listed in
1260: Tables~\ref{tab:57cands1},~\ref{tab:57cands2},~\ref{tab:57cands3}, and~\ref{tab:57cands4}
1261: that are close (within 3 pixels) to the core of their host galaxies have a matching
1262: x-ray-bright object in the Chandra Deep Field catalogs~\citep{bib:chandra1, bib:chandra2}.
1263: The only questionable candidate that might have a matching object is candidate \#3
1264: in Table~\ref{tab:57cands2}; however, its $P({\rm Ia} |\{D_i\},z)$ never
1265: exceeds $\sim$10$^{-6}$ for any redshift bin considered.
1266:
1267:
1268: In order to estimate $d_j$'s, one must select some kind of redshift binning.
1269: One must be careful about the selection of the redshift bins in an analysis
1270: whose goal is to estimate the supernova rates, because
1271: the use of binning averages the behavior of the rates over the width of the bin.
1272: However, the uncertainty in the candidates'
1273: redshifts forces us to use finite bins -- or, in other words, it does not make
1274: sense to use infinitely narrow bins when there is significant uncertainty
1275: in the candidate redshifts.
1276: For our analysis, we choose the width of the bins to be $\Delta \bar{z}$ = 0.1.
1277: Table~\ref{tab:obscount} lists the numbers of observed candidates in these bins,
1278: as well as their uncertainties, for the four samples listed in Table~\ref{tab:samples}
1279: ($d_j^m$ refers to a number of candidates in the $j^{\rm th}$ redshift bin
1280: for the $m^{\rm th}$ sample).
1281: All the uncertainties reflect a 68\% confidence region.
1282: In order to calculate the total numbers of supernovae, $d_j$,
1283: we use the procedure described above on the combined candidates from all four samples.
1284: In other words, the total $d_j$ is not a trivial sum of the probability distributions
1285: of the $d_j^m$'s.
1286: %%%%
1287: \begin{table}[htbp]
1288: \small
1289: \begin{center}
1290: \begin{tabular}{|c|c|c|c|c|c|}
1291: \hline\hline
1292: Redshift bin & $d_j^1$ & $d_j^2$ & $d_j^3$ & $d_j^4$ & Total\\
1293: \hline\hline
1294: 0.0 $\leq$ $z$ $<$ 0.1 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$\\
1295: 0.1 $\leq$ $z$ $<$ 0.2 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$\\
1296: 0.2 $\leq$ $z$ $<$ 0.3 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.15}_{-0.00}$\\
1297: 0.3 $\leq$ $z$ $<$ 0.4 & 0.00$^{+1.17}_{-0.00}$ & 0.00$^{+1.41}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.45}_{-0.00}$\\
1298: 0.4 $\leq$ $z$ $<$ 0.5 & 0.00$^{+1.83}_{-0.00}$ & 1.35$^{+2.77}_{-0.41}$ & 0.00$^{+1.14}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 1.84$^{+3.13}_{-0.62}$\\
1299: 0.5 $\leq$ $z$ $<$ 0.6 & 1.74$^{+2.94}_{-0.63}$ & 0.00$^{+1.45}_{-0.00}$ & 0.00$^{+1.16}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 1.98$^{+3.12}_{-0.72}$\\
1300: 0.6 $\leq$ $z$ $<$ 0.7 & 3.31$^{+3.28}_{-1.05}$ & 0.00$^{+1.44}_{-0.00}$ & 0.00$^{+1.19}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 3.58$^{+3.45}_{-1.13}$\\
1301: 0.7 $\leq$ $z$ $<$ 0.8 & 2.17$^{+3.13}_{-0.75}$ & 0.00$^{+2.00}_{-0.00}$ & 0.00$^{+1.29}_{-0.00}$ & 1.00$^{+2.28}_{-0.28}$ & 3.98$^{+3.72}_{-1.29}$\\
1302: 0.8 $\leq$ $z$ $<$ 0.9 & 1.26$^{+3.02}_{-0.46}$ & 1.59$^{+2.95}_{-0.56}$ & 0.85$^{+2.42}_{-0.26}$ & 0.00$^{+1.13}_{-0.00}$ & 4.07$^{+3.78}_{-1.39}$\\
1303: 0.9 $\leq$ $z$ $<$ 1.0 & 2.94$^{+3.21}_{-0.95}$ & 0.72$^{+3.09}_{-0.14}$ & 0.19$^{+1.85}_{-0.19}$ & 0.00$^{+1.28}_{-0.00}$ & 4.89$^{+4.00}_{-1.56}$\\
1304: 1.0 $\leq$ $z$ $<$ 1.1 & 0.00$^{+1.31}_{-0.00}$ & 0.00$^{+2.09}_{-0.00}$ & 0.10$^{+1.85}_{-0.10}$ & 0.00$^{+1.61}_{-0.00}$ & 1.56$^{+3.37}_{-0.59}$\\
1305: 1.1 $\leq$ $z$ $<$ 1.2 & 1.05$^{+2.37}_{-0.30}$ & 0.00$^{+1.92}_{-0.00}$ & 0.00$^{+1.15}_{-0.00}$ & 0.00$^{+1.46}_{-0.00}$ & 1.74$^{+3.09}_{-0.57}$\\
1306: 1.2 $\leq$ $z$ $<$ 1.3 & 1.03$^{+2.34}_{-0.29}$ & 0.00$^{+1.69}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.17}_{-0.00}$ & 1.36$^{+2.78}_{-0.41}$\\
1307: 1.3 $\leq$ $z$ $<$ 1.4 & 1.00$^{+2.28}_{-0.28}$ & 0.00$^{+1.66}_{-0.00}$ & 1.00$^{+2.28}_{-0.28}$ & 0.91$^{+2.29}_{-0.27}$ & 3.27$^{+3.15}_{-1.00}$\\
1308: 1.4 $\leq$ $z$ $<$ 1.5 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.59}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.59}_{-0.00}$\\
1309: 1.5 $\leq$ $z$ $<$ 1.6 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.44}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.44}_{-0.00}$\\
1310: 1.6 $\leq$ $z$ $<$ 1.7 & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.27}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.13}_{-0.00}$ & 0.00$^{+1.27}_{-0.00}$\\
1311: \hline\hline
1312: \end{tabular}
1313: \end{center}
1314: \normalsize
1315: \caption[]
1316: {\label{tab:obscount}
1317: The best estimate (\emph{i.e.}, the most probable) number of Ia's, $d_j^m$, in $\Delta z$ = 0.1 redshift bins ($j$ = [1,..,17]), for the
1318: four samples listed in Table~\ref{tab:samples} ($m$ = [1,..,4]).
1319: The total numbers are the results of applying the counting
1320: procedure described in the text to the combined candidates from all four samples
1321: (in other words, the total probability distribution is not a trivial sum of the probability distributions for the four samples).
1322: All the uncertainties reflect a 68\% confidence region.
1323: }
1324: \end{table}
1325: %%%%
1326:
1327:
1328: %%%%%%%%%%%%%%%
1329: \subsubsection{Sensitivity to Varying Priors}
1330: \label{sec:priors}
1331: %%%%%%%%%%%%%%%%%%
1332: As usual in Bayesian analysis, the errors on the observed number of supernovae $d_j$
1333: calculated as described in section~\ref{sec:typeia} are a combination of
1334: statistical and systematic uncertainties.
1335: However, to gain an appreciation for the effect of the prior assumptions
1336: on the final result, we compute the change in $d_j$'s by varying
1337: the calculation of $P({\rm Ia}|\{D_i\},\,z)$ from Eqn.~\ref{eqn:bayes} in three
1338: different ways:
1339: \begin{itemize}
1340: \item
1341: {\bf Large extinction}: In section~\ref{sec:typeia}, we considered three discrete cases
1342: for extinction: no extinction, ($A_V$, $R_V$) = (0.4, 2.1), and ($A_V$, $R_V$) = (0.4, 3.1).
1343: We now add the case of ($A_V$, $R_V$) = (1.0, 3.1) to the extinction prior,
1344: and consider it to be equally likely as the cases of no extinction and
1345: moderate extinction. It is in fact known that a value of $A_V$ = 1.0 is
1346: much less likely than, say, an $A_V$ = 0; however, it is in cases of strong extinction
1347: that the overlap between the magnitude phase space of Ia's and CC's becomes the largest.
1348: \item
1349: {\bf Overluminous Ibc's}: In~\cite{bib:rich2006}, it is pointed out that there may exist
1350: a sub-class of type Ibc supernovae whose mean restframe $B$-band magnitudes are much closer to those
1351: of normal Ia's, with $\bar{M}$ = -20.08, $\delta M$ = 0.46. We add these supernovae
1352: as one more type to our list of supernova types considered, assuming that $f_{Ibc}$ = 0.18
1353: for normal Ibc's, and 0.09 for the overluminous ones.
1354: \item
1355: {\bf Flat ratio of the CC to Ia rates, all CC types equally likely}:
1356: Instead of using the redshift-dependent models for the ratio of the CC to Ia supernova rates,
1357: we now assume that the ratio is redshift-independent, and taken
1358: to be 2.15, which is roughly the average of the models shown in Fig.~\ref{fig:models}).
1359: We also assume that the relative fractions of the CC supernovae are all 0.25 ($f_{CC}$ = 0.25);
1360: or, in other words, that all classes of the CC supernovae are equally likely.
1361: \end{itemize}
1362:
1363:
1364: The considered alternative
1365: priors are deliberately taken to be such that the effect on $P({\rm Ia}|\{D_i\},\,z)$
1366: should be the most dramatic, without too much regard for whether or not such priors
1367: are realistic.
1368: Table~\ref{tab:systematics} lists the changes in $d_j$ relative to the values
1369: specified in Table~\ref{tab:obscount} as a result of using the alternative priors listed above.
1370: %%%
1371: \begin{table}[htbp]
1372: \small
1373: \begin{center}
1374: \begin{tabular}{|c|c|c|c|c|c|}
1375: \hline\hline
1376: Redshift bin & $A_V = 1$ & Overluminous & Flat CC/Ia rates \\
1377: & & Ibc's & $f_{\rm CC}$ = 0.25 \\
1378: \hline
1379: 0.0 $\leq$ $z$ $<$ 0.1 & 0.00 & 0.00 & 0.00\\
1380: 0.1 $\leq$ $z$ $<$ 0.2 & 0.00 & 0.00 & 0.00\\
1381: 0.2 $\leq$ $z$ $<$ 0.3 & 0.00 & 0.00 & 0.00\\
1382: 0.3 $\leq$ $z$ $<$ 0.4 & 0.00 & 0.00 & 0.00\\
1383: 0.4 $\leq$ $z$ $<$ 0.5 & 0.00 & -0.14 & -0.32\\
1384: 0.5 $\leq$ $z$ $<$ 0.6 & 0.29 & 0.93 & -0.24\\
1385: 0.6 $\leq$ $z$ $<$ 0.7 & -0.09 & -0.86 & -0.21\\
1386: 0.7 $\leq$ $z$ $<$ 0.8 & -0.16 & -0.05 & -0.08\\
1387: 0.8 $\leq$ $z$ $<$ 0.9 & -0.04 & -0.06 & -0.61\\
1388: 0.9 $\leq$ $z$ $<$ 1.0 & 0.06 & -0.12 & -0.56\\
1389: 1.0 $\leq$ $z$ $<$ 1.1 & 0.14 & -0.14 & -0.03\\
1390: 1.1 $\leq$ $z$ $<$ 1.2 & 0.12 & 0.09 & 0.01\\
1391: 1.2 $\leq$ $z$ $<$ 1.3 & -0.01 & -0.02 & 0.00\\
1392: 1.3 $\leq$ $z$ $<$ 1.4 & 0.09 & -0.08 & 0.04\\
1393: 1.4 $\leq$ $z$ $<$ 1.5 & 0.00 & 0.00 & 0.00\\
1394: 1.5 $\leq$ $z$ $<$ 1.6 & 0.00 & 0.00 & 0.00\\
1395: 1.6 $\leq$ $z$ $<$ 1.7 & 0.00 & 0.00 & 0.00\\
1396: \hline\hline
1397: \end{tabular}
1398: \end{center}
1399: \normalsize
1400: \caption[]
1401: {\label{tab:systematics}
1402: A change in the estimates for the numbers of Ia's, $d_j$, as a result
1403: of using alternative priors for the $\vec{\theta}$ parameters, as described
1404: in the text. Listed are the differences between the $d_j$ obtained
1405: for the alternative parameters and for the default priors, in $\Delta z$ = 0.1 redshift bins ($j$ = [1,..,17]),
1406: for the combination of the four samples listed in Table~\ref{tab:samples}.
1407: }
1408: \end{table}
1409: %%%
1410:
1411:
1412: It is clear from Table~\ref{tab:systematics}
1413: that none of the alternative priors considered
1414: leads to a change in the mean that goes beyond the estimated errors
1415: in Table~\ref{tab:obscount}.
1416:
1417:
1418:
1419:
1420: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1421: \section{The Rates Calculation}
1422: \label{sec:analysis}
1423: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1424: Next, we compute the expected number of candidates
1425: in the $j^{\rm th}$ redshift bin whose center is ${\bar z_j}$,
1426: given a volumetric Type Ia supernova rate in the supernova rest frame,
1427: ${\rm r_{V,Ia}({\bar z}) }$,
1428: as a function of redshift $\bar{z}$. The expected number of candidates
1429: is different from the \emph{measured} $d_j$'s: it is \emph{calculated}
1430: entirely based on Monte Carlo simulations of SNe and a given rates model.
1431: \begin{equation}
1432: d_j^{\rm exp} = \Delta \bar{z}_j \frac{r_{V,\rm Ia}({\bar z_j})}{1+{\bar z_j}} \, \frac{\Theta}{4\,\pi}\, \frac{dV}{d\bar{z}}({\bar z_j}) \left [T_{\rm Ia} ({\bar z_j}) \epsilon_{\rm Ia}({\bar z_j}) + \frac{r_{V,\rm CC}({\bar z_j})}{r_{V,\rm Ia}({\bar z_j})} T_{\rm CC}({\bar z_j}) \epsilon_{\rm CC}({\bar z_j}) \right],
1433: \label{eqn:rates}
1434: \end{equation}
1435: %\rm{SNR_{\rm Ia}}(z) = N_{\rm Ia}(z)\frac{4\, \pi}{\Theta} \, \frac{1 + z}{ct (z)} \, \frac{1}{dV/dz},\
1436: where $\Delta \bar{z}_j$ is the width of the redshift bin; $\Theta$ is the survey area covered;
1437: $dV/d\bar{z}$ is the comoving volume
1438: computed assuming a $\Lambda$CDM cosmology with $\Omega_{\Lambda}$ = 0.7,
1439: $\Omega_{M}$ = 0.3, and $H_0$ = 100 $\times$ h$_{70}$ (km/s) Mpc$^{-1}$;
1440: $T_{\rm Ia}(\bar z)$ and $T_{\rm CC}(\bar z)$ are the control times for
1441: Ia and non-Ia candidates, respectively;
1442: $\epsilon_{\rm Ia}$ and $\epsilon_{CC}$ are
1443: the efficiencies of the stage 3 selection for Ia and non-Ia candidates, respectively;
1444: and $r_{V,\rm CC}(\bar z)/{r_{V,\rm Ia}(\bar z)}$ is the ratio of the non-Ia
1445: supernova rate to the Ia supernova rate. Once again, the appearance of this ratio
1446: does not bias our results, since we do not make any assumptions
1447: about the \emph{absolute} Type Ia rate.
1448: The control time in Eqn.~\ref{eqn:rates} enters with a factor of (1 + $\bar z$).
1449: This is a consequence of the fact that it is calculated in the observer frame,
1450: as will be described later.
1451:
1452:
1453: The control time $T$ is defined as the time during which a supernova search is
1454: \emph{potentially} capable of finding supernova candidates.
1455: In order to calculate it, we simulate HST observations of
1456: Type Ia and non-Type Ia supernovae at redshifts up to 1.7,
1457: with the same sampling and exposure times as those of the real data. By
1458: shifting the observing grid along the light curves, we calculate the weighted
1459: sum of the number of days during which a given supernova could be detected.
1460: The weight factors are obtained from the stage 1 efficiency parametrization;
1461: it is also required that the light curves satisfy the stage 2 SNR requirements.
1462: Therefore, stage 1 and 2 supernova selection efficiencies are
1463: naturally built into the control time calculation. However, the stage 3
1464: selection efficiency is \emph{not} part of the control time calculation, and
1465: must therefore be computed separately.
1466: Calculating the area of the survey is straightforward using a Monte Carlo
1467: approach. The calculation of the control time and the survey area
1468: is given in section~\ref{sec:ctandarea}.
1469:
1470:
1471: The stage 3 selection efficiencies $\epsilon_{\rm Ia}$
1472: and $\epsilon_{\rm CC}$
1473: must be calculated for the candidates that passed the control time requirements,
1474: and thus satisfy both stage 1 and 2 cuts. We create a Monte Carlo
1475: sample simulating real supernova candidates of five different types, and apply stage 1 and 2 cuts to them.
1476: We simulate both Ia and non-Ia candidates
1477: and calculate the number of candidates as we would for real data.
1478: This procedure is described in detail in section~\ref{sec:calcnum}.
1479:
1480: The errors on the expected $d_j^{\rm exp}$ are a combination of
1481: statistical and systematic uncertainties.
1482: Apart from the uncertainties inherent in the calculation of
1483: $P({\rm Ia}|\{D_i\},z)$,
1484: the dominant systematic uncertainties come from two sources:
1485: estimating the variation in the control time for Ia's for
1486: values of the lightcurve timescale stretch, $s$, other than 1,
1487: and estimating the effect of varying the
1488: ratio of the rates $r_{V,\rm CC}(\bar z)/{r_{V,\rm Ia}(\bar z)}$.
1489: The former is described in more detail in~\ref{sec:ctandarea}, and
1490: for the latter we use two models described in section~\ref{sec:typeia},
1491: for $\tau$ = 1 Gyr and 3 Gyr.
1492:
1493:
1494: %It is worth noting that instead of putting $\epsilon_{\rm Ia}$
1495: %and $\epsilon_{CC}$ into $d_j^{\rm exp}(z_j)$, as
1496: %in Eqn.~\ref{eqn:rates}, one could instead include them into the prior
1497: %on $P(T)$ in Eqn. 8 of~\cite{bib:ourpaper}. However, because
1498: %we checked that the probabilities for real Ia's are close to 1 and those
1499: %for real non-Ia's are close 0, the introduction of the relative efficiencies
1500: %does not affect the final probability numbers very much.
1501:
1502:
1503: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1504: \subsection{The Control Time and Search Area Calculation}
1505: \label{sec:ctandarea}
1506: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1507: Let us start with describing the calculation of the control time
1508: and search area, $T$ and $\Theta$ from Eqn.~\ref{eqn:rates}.
1509: The control time is the time during which a supernova search is
1510: in principle capable of finding supernova candidates on the area
1511: covered. For the GOODS fields, the orientation of the tiles is such that
1512: a candidate is not necessarily accessible for every search epoch due to edge effects (see Fig.~\ref{fig:goods}).
1513: For example, for sample 1 from Table~\ref{tab:samples},
1514: a given location may only be covered by epochs 1, 3, and 5 (but not by epochs 2 and 4)
1515: of the North GOODS dataset. In both our control time calculation and in the search
1516: area calculation, we thus consider all of the possible epoch permutations
1517: at each location: 31 possible permutations for samples 1, 3, and 4 and 15 possible permutations for sample 2.
1518:
1519: %%% control time
1520: We perform separate control time and search area calculations for the four samples
1521: listed in Table~\ref{tab:samples}; however, the approach is the same.
1522: For the control time calculation, we make use of
1523: the simulation described in some detail in Appendix A of~\cite{bib:ourpaper}.
1524: We use it to create simulated HST observations in both F775W and F850LP bands
1525: for Type Ia supernovae of stretch 1, as well as for non-Type Ia
1526: supernovae, at redshifts up to 1.7 with an increment of 0.1.
1527: Separate sets of observations are generated for each possible permutation
1528: of the available search epochs, for each of the four samples.
1529: For example, for a supernova from Sample 1 that happens to be present in
1530: every one of the North GOODS epochs, there will be five simulated search observations and
1531: a single reference observation.
1532: We use typical epoch separations and exposure times for a given sample.
1533: The observations are realized using an aperture exposure time calculator with a 0.1'' radius.
1534: We initially set the explosion date of the supernova
1535: on the last date of the available search epoch observation set (\emph{e.g.},
1536: for the supernova example mentioned above it would be on the date the last of the North
1537: GOODS data were taken). The observing grid for the search observations is then shifted by one day, and
1538: the procedure is repeated $N_{shifts}$ = 350 times (that is, spanning approximately
1539: a year, which is the longest separation between the search and reference data
1540: for our data samples).
1541: For each such shift, we require that
1542: the simulated data satisfy both the stage 1 and stage 2 requirements listed in Table~\ref{tab:selection}.
1543: The resulting control time thus has stage 1 and 2 efficiencies automatically included. It is given by:
1544: \begin{equation}
1545: T = \sum_{k=1}^{N_{shifts}} \left[ 1 - \prod_{i = 1}^{N_{ep}}(1-\epsilon_i^k(SNR_i)) \right] e_k,
1546: \end{equation}
1547: where the sum is over all the shifts, $N_{ep}$ is the number of available search epochs (in the
1548: example considered above, $N_{ep}$ = 5);
1549: $\epsilon_i^k$ is a function of the $i^{\rm th}$ subtraction's SNR, parametrized
1550: as in Eqn.~\ref{eqn:eff}; and $e_k$ is a binary quantity
1551: \begin{eqnarray}
1552: e_k = \left\{
1553: \begin{array}{ll}
1554: 1 & \mbox{, if $k^{\rm th}$ shift configuration satisfies stage 2 requirements}\\
1555: 0 & \mbox{, if $k^{\rm th}$ shift configuration does not satisfy stage 2 requirements}\\
1556: \end{array}
1557: \right.
1558: \end{eqnarray}
1559: that assesses whether a given configuration has enough epochs with sufficient SNR for
1560: the stage 2 selection.
1561:
1562:
1563: We repeat the control time calculation for Ia's with the lightcurve timescale stretch
1564: values of $s$ = 0.65 and $s$ = 1.30, weight the results by the probability
1565: of obtaining such stretches taken from Eqn.~\ref{eqn:sprioria}, and take
1566: the larger error between the control time computed for these stretch parameters and
1567: that computed for a stretch of 1 as a measure of the systematic error on the control time
1568: for Ia's.
1569: For reference, Table~\ref{tab:ct} lists the control time as a function of redshift for both the nominal
1570: stretch of 1 and for the stretch of 0.65 and 1.30, for the configurations
1571: in which a supernova candidate is assumed present on all of the search epochs.
1572: %% this is get_ct_table.perl in final
1573: \begin{table}[htbp]
1574: \small
1575: \begin{center}
1576: \begin{tabular}{|c|c|c|c||c|c|c||c|c|c||c|c|c|}
1577: \hline\hline
1578: & \multicolumn{12}{c|}{Control time (yrs)}\\
1579: \cline{2-13}
1580: $\bar z$ & \multicolumn{3}{c||}{Sample \#1} & \multicolumn{3}{|c||}{Sample \#2} & \multicolumn{3}{|c||}{Sample \#3} & \multicolumn{3}{|c|}{Sample \#4} \\
1581: \cline{2-13}
1582: & $s$=1 & $s$=0.65 & $s$=1.3 & $s$=1 & $s$=0.65 & $s$=1.3 &$s$=1 & $s$=0.65 & $s$=1.3 &$s$=1 & $s$=0.65 & $s$=1.3 \\
1583: \hline\hline
1584: 0.1 & 0.84 & 0.78 & 0.84 & 0.84 & 0.66 & 0.85 & 0.68 & 0.65 & 0.68 & 0.30 & 0.27 & 0.35\\
1585: 0.2 & 0.84 & 0.77 & 0.84 & 0.84 & 0.65 & 0.85 & 0.68 & 0.63 & 0.67 & 0.32 & 0.29 & 0.42\\
1586: 0.3 & 0.85 & 0.77 & 0.85 & 0.83 & 0.64 & 0.85 & 0.68 & 0.63 & 0.67 & 0.31 & 0.28 & 0.37\\
1587: 0.4 & 0.84 & 0.75 & 0.83 & 0.80 & 0.61 & 0.84 & 0.67 & 0.60 & 0.66 & 0.32 & 0.28 & 0.36\\
1588: 0.5 & 0.84 & 0.73 & 0.84 & 0.77 & 0.59 & 0.84 & 0.67 & 0.58 & 0.66 & 0.32 & 0.28 & 0.36\\
1589: 0.6 & 0.83 & 0.72 & 0.84 & 0.73 & 0.56 & 0.82 & 0.66 & 0.57 & 0.65 & 0.33 & 0.28 & 0.37\\
1590: 0.7 & 0.81 & 0.68 & 0.84 & 0.69 & 0.53 & 0.79 & 0.63 & 0.53 & 0.64 & 0.32 & 0.29 & 0.37\\
1591: 0.8 & 0.78 & 0.64 & 0.83 & 0.64 & 0.51 & 0.75 & 0.59 & 0.49 & 0.63 & 0.33 & 0.29 & 0.37\\
1592: 0.9 & 0.72 & 0.59 & 0.80 & 0.57 & 0.46 & 0.66 & 0.53 & 0.43 & 0.58 & 0.33 & 0.28 & 0.37\\
1593: 1.0 & 0.68 & 0.57 & 0.75 & 0.54 & 0.43 & 0.62 & 0.48 & 0.36 & 0.54 & 0.33 & 0.28 & 0.36\\
1594: 1.1 & 0.65 & 0.55 & 0.72 & 0.51 & 0.40 & 0.59 & 0.46 & 0.32 & 0.49 & 0.32 & 0.27 & 0.37\\
1595: 1.2 & 0.63 & 0.53 & 0.69 & 0.48 & 0.35 & 0.56 & 0.42 & 0.27 & 0.47 & 0.32 & 0.26 & 0.36\\
1596: 1.3 & 0.60 & 0.49 & 0.67 & 0.46 & 0.30 & 0.55 & 0.38 & 0.22 & 0.45 & 0.32 & 0.25 & 0.36\\
1597: 1.4 & 0.57 & 0.42 & 0.64 & 0.44 & 0.23 & 0.53 & 0.37 & 0.17 & 0.43 & 0.31 & 0.20 & 0.36\\
1598: 1.5 & 0.56 & 0.36 & 0.62 & 0.39 & 0.15 & 0.48 & 0.32 & 0.13 & 0.40 & 0.31 & 0.17 & 0.35\\
1599: 1.6 & 0.51 & 0.24 & 0.59 & 0.34 & 0.09 & 0.44 & 0.27 & 0.09 & 0.37 & 0.26 & 0.11 & 0.34\\
1600: 1.7 & 0.47 & 0.17 & 0.56 & 0.21 & 0.04 & 0.38 & 0.25 & 0.06 & 0.33 & 0.25 & 0.07 & 0.31\\
1601: \hline\hline
1602: \end{tabular}
1603: \end{center}
1604: \normalsize
1605: \caption[]
1606: {\label{tab:ct}
1607: The Type Ia control time in years as a function of redshift, for the configurations on which a
1608: supernova candidate is assumed to be present on all of the search epochs.
1609: The control time is given for three different values of the stretch parameter $s$: 1 (nominal), 0.65, and 1.30.
1610: Note that this control time has the stage 1 and 2 efficiencies built into the calculation.
1611: }
1612: \end{table}
1613:
1614:
1615: %%%%%%%%%% area
1616: Calculating the search area is non-trivial because of the complicated orientations of the
1617: GOODS tiles, as well as the overlaps between the tiles (see Fig.~\ref{fig:goods}). In addition, the search area must be
1618: calculated separately for all of the possible epoch configurations, as described above.
1619: We perform this calculation using a Monte Carlo method.
1620: First, we create a 300x300 point
1621: grid between the minimum and maximum right ascensions ($\alpha$) and declinations
1622: ($\delta$) covering the entire North or South GOODS area (\emph{e.g.},
1623: from $\alpha$ = 12:35:34.85 and $\delta$ = 62:4:59.45
1624: to $\alpha$ = 12:38:14.7 and $\delta$ = 62:23:36.78 for epochs 1, 3, and 5
1625: of sample \# 1).
1626: % North epoch 1, 3 and 5:
1627: %12:35:34.85 62:4:59.45 to 12:38:14.71 62:23:36.78
1628: % north epoch 2:
1629: %12:35:40.33 62:6:4.54 to 12:38:9.25 62:22:29.03
1630: % search april & july
1631: %12:35:40.33 62:6:4.54 to 12:38:9.25 62:22:29.03
1632: % search may and august
1633: %12:35:34.85 62:4:59.45 to 12:38:14.71 62:23:36.78
1634: % south epochs 1, 3, and 5:
1635: %3:31:56.50 -27:38:50.35 to 3:33:4.35 -27:57:38.12
1636: % south epochs 2 and 4:
1637: %3:31:54.79 -27:38:17.16 to 3:33:6.30 -27:58:12.00
1638: Then, for a given epoch, and for each
1639: point $i$ on the grid, we check whether this ($\alpha_i$, $\delta_i$) belongs to any of the images
1640: that were used to make subtracted data for this epoch. In other words, we convert
1641: ($\alpha_i$, $\delta_i$) into image coordinates ($x_j$, $y_j$), and check
1642: that: (a) the point falls within the confines of at least one search/reference image pairs;
1643: (b) it does not fall into the gap between the two ACS chips on the search image; and
1644: (c) it does not fall on a known bad pixel or a pixel that has been masked off for any other
1645: reason (\emph{e.g.}, due to a residual cosmic ray contamination) on either image,
1646: although because of the drizzling there are very few affected pixels.
1647: If all of these requirements are satisfied, the point is counted toward the
1648: area calculation. Once counted, a given point can never again
1649: be counted for this particular epoch. This avoids double-counting, an issue particularly
1650: important since most GOODS tiles overlap at least somewhat with their immediate neighbors,
1651: and a point with a given ($\alpha_i$, $\delta_i$) may well be present on several images.
1652: A separate accounting of the number of points is kept for each epoch
1653: permutation.
1654: For example, let us suppose that the number of points that cover all five
1655: of the GOODS North epochs is $a_1$, and that the number of total points tried in the grid
1656: is $A_1$; then the area corresponding to this configuration is $S \, a_1/A_1$, where $S$ is
1657: the area of the entire North GOODS survey.
1658:
1659:
1660: Figure~\ref{fig:ct} shows the resulting product of the control time and surveyed area
1661: ($\Theta \, T(\bar z)$ = \\$\sum_{i = 1}^{n} \Theta_i \, T_i(\bar z)$, where $n$
1662: is the number of all possible permutations) for stretch 1 Type Ia's,
1663: as a function of redshift for the four different samples in Table~\ref{tab:samples}.
1664: %%
1665: \begin{figure}[!htb]
1666: \begin{center}
1667: \includegraphics[height=4.in, width=4.8in]{f4.eps}
1668: \end{center}
1669: \caption[]
1670: {\label{fig:ct}
1671: The product of the control time and surveyed area as a function of redshift for the four
1672: samples listed in Table~\ref{tab:samples}, calculated for a stretch 1 Ia.
1673: The dashed line is for sample \#1,
1674: where the 5 GOODS North epochs were used as the search data, and the combined 2004 ACS sample,
1675: as the reference. The dotted line is for sample \#2, where the 4 epochs of
1676: the 2004 ACS dataset were used as the search data, and the combined GOODS North
1677: data, as the reference. The dashed-dotted line is for sample \#3, where
1678: all 5 epochs of the GOODS South sample were used as the search data; and
1679: the combined epochs 4+5 of the South GOODS dataset, as the reference.
1680: The solid line is for sample \#4, where
1681: all 5 epochs of the GOODS South sample were used as the search data; and
1682: the combined epochs 1+2 of the South GOODS dataset, as the reference.
1683: }
1684: \end{figure}
1685: %%
1686: There are several interesting features in Fig.~\ref{fig:ct}. First,
1687: the product of the control time and area tends to decrease with redshift.
1688: This is a consequence of the fact that it becomes more difficult to satisfy
1689: the stage 2 SNR requirements for higher redshift (dimmer) supernovae.
1690: Second, for a given redshift, the product is smaller
1691: for sample \#2 than for sample \#1, a consequence of the fact that there are only
1692: 4 search epochs in sample \#2 vs. 5 search epochs in sample \#1.
1693: Third, the product is distinctly smaller for the South GOODS
1694: samples (samples \#3 and \#4) than for either of the North samples (\#1 and \#2), a
1695: reflection of the fact that for these samples we are forced to use references made from two of
1696: the GOODS South dataset's own epochs. Finally, the product is smaller
1697: for sample \#4, which uses epochs 1+2 of the GOODS South dataset as its reference data,
1698: than for it is for sample \#5, which uses epochs 4+5. This is simply because
1699: the rise time of a supernova is smaller than its decline time.
1700:
1701:
1702: %%%%%%
1703: \subsection{Calculating $\epsilon_{\rm Ia}$ and $\epsilon_{\rm CC}$}
1704: \label{sec:calcnum}
1705: %%%%%%
1706: In order to determine the efficiency of the stage 3 selection, we generate
1707: four Monte Carlo datasets simulating the data from the four datasets listed in Table~\ref{tab:samples}
1708: (in other words, they have the same sampling, exposure times, \emph{etc.}, as the data).
1709: Each Monte Carlo dataset contains 500 candidates for each of the 5 supernova
1710: types considered (Ia, Ibc, IIL, IIP, and IIn).
1711: The redshifts of these candidates are drawn from a Gaussian distribution that
1712: uses the redshifts and redshift errors of the real data events;
1713: the exposure times and sampling intervals also mimic those
1714: of the real data. The candidates' rest-frame
1715: $B$-band magnitudes, stretch (for Type Ia's), and extinction parameters are drawn from
1716: the appropriate distributions used in Eqns.~\ref{eqn:final_Ia} and~\ref{eqn:final_nonIa}.
1717: The time period between the date of explosion and
1718: the first observation is randomly drawn from a flat distribution.
1719: In addition, because we are simulating a dataset as it would appear by the time it is
1720: ready for the stage 3 selection, we impose the same selection requirements from stages 1 and 2
1721: on these Monte Carlo events as we do on the real data.
1722:
1723: After these Monte Carlo samples are generated,
1724: we calculate the number of candidates in each redshift bin.
1725: Dividing this number by the total number of the generated Ia's
1726: yields the efficiency $\epsilon_{j\, {\rm Ia}}^m$, for
1727: redshift bin $j$ for
1728: Monte Carlo dataset $m$. Similarly, the efficiency for non-Type Ia candidates,
1729: $\epsilon_{j\, {\rm CC}}^m$, is defined as the sum of the probabilities of the
1730: non-Type Ia candidates divided by the total number of all generated non-Type Ia supernovae.
1731: The values of $\epsilon_{j\, {\rm Ia}}^m$'s range from $\sim$10 to 90\%;
1732: and the values of $\epsilon_{j\, {\rm CC}}^m$'s, $\sim$3 to 50\%,
1733: depending on the redshift bin.
1734:
1735:
1736: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1737: \subsection{Comparison of Expected and Observed Numbers of Supernovae}
1738: \label{sec:thenumbers}
1739: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1740: We can now put everything together and compute the expected numbers of supernovae
1741: for a given model of the Type Ia supernova rates using Eqn.~\ref{eqn:rates}.
1742: We calculate the observed numbers of supernovae for redshifts
1743: $\bar{z}$ $\leq$ 1.7, as well as the expected numbers of supernovae for
1744: the two models considered in~\cite{bib:pain}:
1745: a redshift-independent one and one evolving with redshift as a power law.
1746: We perform a least-squares fit of the observed numbers of supernovae
1747: to the predictions for both models.
1748: We also perform a maximum likelihood fit and compare the results.
1749: \begin{itemize}
1750: \item {\bf Redshift-independent rate.}
1751: Assuming the rate is flat as a function of redshift, we obtain the
1752: best-fit value of
1753: ${\rm r_{V,Ia}}$ =
1754: (1.1 $^{+0.2}_{-0.2}$) $\times$ 10$^{-4}$ $N_{\rm Ia}$/(year Mpc$^3$ $h_{70}^{-3}$)
1755: , with a $\chi^2$ = 11.5 for 16 degrees of freedom.
1756: Figure~\ref{fig:numsn} (left) shows the resulting distribution of the predicted
1757: and observed numbers of supernovae.
1758: The errors on the predicted numbers of supernovae are
1759: a quadratic combination of the statistical and systematic errors (the statistical and
1760: systematic errors are comparable).
1761: The maximum likelihood method yields
1762: ${\rm r_{V,Ia}}$ = (0.7$^{+0.2}_{-0.2}$)
1763: $\times$ 10$^{-4}$ $N_{\rm Ia}$/(year Mpc$^3$ $h_{70}^{-3}$),
1764: consistent with the $\chi^2$ method.
1765:
1766:
1767: \item {\bf Rate evolving as a power law with redshift.}
1768: Assuming the rate is varying as a function of redshift as (1 + $\bar z$)$^\alpha$,
1769: using the $\chi^2$ fitter we obtain the best-fitting value for
1770: $\alpha$ = 0.2$^{+0.7}_{-0.7}$ with a $\chi^2$ = 11.4 for 15 degrees of freedom.
1771: This is consistent with~\cite{bib:pain}, who found $\alpha$ = 0.8 $\pm$ 1.6.
1772: Note that the fit results are also consistent with
1773: the $\alpha$ = 0 case that was considered above.
1774: The maximum likelihood method yields $\alpha$ = -0.4$^{1.0}_{1.1}$.
1775: %%
1776: \begin{figure}[!htb]
1777: \begin{center}
1778: $\begin{array}{c@{\hspace{0.0in}}c}
1779: \epsfxsize=3.3in\epsfysize=2.8in\epsffile{f5a.ps} &
1780: \epsfxsize=3.3in\epsfysize=2.8in\epsffile{f5b.ps}
1781: \end{array}$
1782: \end{center}
1783: \caption[]
1784: {\label{fig:numsn}
1785: Left: The total observed candidates for the 4 samples are plotted
1786: as a function of redshift (filled circles).
1787: The errors on the observed candidates are given in Table~\ref{tab:obscount}.
1788: The predicted number of candidates has been computed assuming
1789: a redshift-independent volumetric Type Ia rate of
1790: ${\rm r_{V,Ia}}(\bar z)$ = 1.1 $\times$ 10$^{-4}$ $N_{\rm Ia}$/(year Mpc$^3$ $h_{70}^{-3}$)]),
1791: and is plotted as a dashed histogram.
1792: The shaded region around the predicted numbers indicates the range of combined statistical
1793: and systematic errors.
1794: The contributions from the statistical and systematic errors are comparable.
1795: Right: The calculated rates as a function of redshift (filled circles),
1796: with overplotted fit results to the fits described in the text:
1797: redshift-independent rate (solid line) and power-law redshift dependent rate (dashed line).
1798: Note that the plot does not show the rates in the first redshift bin;
1799: this is because in this bin the rates are effectively unconstrained on the scale shown.
1800: }
1801: \end{figure}
1802: %%
1803: \end{itemize}
1804: Both the redshift-independent model and the power-law model yield acceptable
1805: fit results, judging by the obtained
1806: $\chi^2$'s (note, however,
1807: that the data points in neighboring bins are correlated, leading to lower $\chi^2$ per DOF).
1808: The probability $p(\Delta \chi^2|\Delta DOF)$ = 0.1,
1809: where $\Delta \chi^2$ and $\Delta DOF$ are the difference in the $\chi^2$
1810: and the numbers of degrees of freedom, respectively, for the redshift-independent model
1811: and the power-law model.
1812: In other words, the approximate probability that data would
1813: fluctuate from the redshift-independent model to the power-law model
1814: is 0.1.
1815: This fact indicates that our description of the experiment is
1816: good at both low and high redshifts. One must note, however, that at
1817: redshifts $>$ 1 the samples start becoming sparser, and at redshifts
1818: $>$ ~1.4 the measurement becomes particularly difficult with this dataset.
1819:
1820:
1821:
1822:
1823: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1824: \section{Comparison to Rates in the Literature}
1825: \label{sec:comp}
1826: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1827: To compare our results with those of~\cite{bib:dahlen},
1828: we now compute the Type Ia supernova rates in four large redshift bins,
1829: 0.2 $\leq$ $\bar z$ $<$ 0.6, 0.6 $\leq$ $\bar z$ $<$ 1.0, 1.0 $\leq$ $\bar z$ $<$ 1.4,
1830: and 1.4 $\leq$ $\bar z$ $<$ 1.7.
1831: Table~\ref{tab:candscount} enumerates the estimates for the number of candidates in
1832: these redshift bins for the four samples listed in Table~\ref{tab:samples}.
1833: %%%
1834: \begin{table}[htbp]
1835: \begin{center}
1836: \small
1837: \begin{tabular}{|c|c|c|c|c|c|}
1838: \hline\hline
1839: Redshift bin & $d_j^1$ & $d_j^2$ & $d_j^3$ & $d_j^4$ & Total\\
1840: \hline\hline
1841: 0.2 $\leq$ $z$ $<$ 0.6 & 2.40$^{+3.07}_{-0.86}$ (0) & 1.81$^{+3.05}_{-0.60}$ & 0.00$^{+1.17}_{-0.00}$ (0) & 1.00$^{+2.28}_{-0.28}$ (2) & 5.44$^{+3.90}_{-1.63}$\\
1842: 0.6 $\leq$ $z$ $<$ 1.0 & 13.40$^{+8.28}_{-5.22}$ (6) & 3.85$^{+3.47}_{-1.25}$ & 1.71$^{+2.90}_{-0.63}$ (2) & 1.07$^{+2.42}_{-0.30}$ (2) & 18.33$^{+4.62}_{-4.62}$\\
1843: 1.0 $\leq$ $z$ $<$ 1.4 & 3.23$^{+3.07}_{-0.97}$ (3) & 2.01$^{+3.25}_{-0.75}$ & 1.50$^{+2.74}_{-0.48}$ (1) & 1.70$^{+2.62}_{-0.58}$ (1) & 8.87$^{+3.13}_{-2.36}$\\
1844: 1.4 $\leq$ $z$ $<$ 1.7 & 0.00$^{+1.13}_{-0.00}$ (0) & 0.35$^{+1.72}_{-0.35}$ & 0.00$^{+1.13}_{-0.00}$ (1) & 0.00$^{+1.13}_{-0.00}$ (1) & 0.35$^{+1.72}_{-0.35}$\\
1845: \hline\hline
1846: \end{tabular}
1847: \end{center}
1848: \normalsize
1849: \caption[]
1850: {\label{tab:candscount}
1851: The best estimate (\emph{i.e.}, the most probable) number of Ia's, $d_j^m$,
1852: in $\Delta \bar{z}$ = 0.4 (0.3 for the last bin) redshift bins ($j$ = [1,..,4]), for the
1853: four samples listed in Table~\ref{tab:samples} ($m$ = [1,..,4]).
1854: The numbers in parenthesis are the number of gold and silver Ia's in the sample from~\cite{bib:riess2},
1855: that were used in the rates analysis of~\cite{bib:dahlen}.
1856: The total numbers are the results of applying the counting
1857: procedure described in the text to the combined candidates from all four samples
1858: (in other words, the total probability distribution is not a trivial sum of the probability distributions for the four samples).
1859: All the uncertainties reflect a 68\% confidence region.
1860: }
1861: \end{table}
1862: %%%
1863:
1864: Using all four samples, we can now compute the rates for each bin
1865: using Eqn.~\ref{eqn:rates}.
1866: The values for $\Theta T (\bar z)$, $dV/d\bar{z}$, $r_{V,\rm CC}/r_{V,\rm Ia}$, and $\bar z$ are taken in the middle of the bin.
1867: The errors on the rates are a quadratic combination of the errors on the number of observed
1868: Ia's listed in Table~\ref{tab:candscount}, as well as statistical and
1869: systematic errors on the right-hand-side part of Eqn.~\ref{eqn:rates}.
1870: The resulting rates are summarized in Table~\ref{tab:results}
1871: and plotted in Fig.~\ref{fig:allrates} together with
1872: the rates from~\cite{bib:dahlen} and
1873: results from the literature at lower redshifts.
1874: It is apparent that our results are consistent with those from the literature:
1875: in particular, at higher redshifts our rates are not
1876: inconsistent with those
1877: of~\cite{bib:dahlen}, although obtaining a precise measure of the consistency
1878: would require a careful evaluation of the correlations between the
1879: samples used in both analyses.
1880: %%
1881: \begin{table}[htbp]
1882: \begin{center}
1883: \begin{tabular}{|c|c|}
1884: \hline\hline
1885: Redshift bin & ${\rm r_{V,Ia}(\bar{z}) }$\\
1886: & ([10$^{-4}$ $N_{\rm Ia}$/(year Mpc$^3$ $h_{70}^{-3}$)]) \\
1887: \hline\hline
1888: 0.2 $\leq$ $\bar z$ $<$ 0.6 & $0.53^{+0.39}_{-0.17}$ \\
1889: 0.6 $\leq$ $\bar z$ $<$ 1.0 & $0.93^{+0.25}_{-0.25}$\\
1890: 1.0 $\leq$ $\bar z$ $<$ 1.4 & $0.75^{+0.35}_{-0.30}$\\
1891: 1.4 $\leq$ $\bar z$ $<$ 1.7 & $0.12^{+0.58}_{-0.12}$\\
1892: \hline\hline
1893: \end{tabular}
1894: \end{center}
1895: \caption[]
1896: {\label{tab:results}
1897: The Type Ia supernova rates in the four redshift bins considered.
1898: The errors are a quadratic combination of the errors on
1899: $d_j^m$'s listed in Table~\ref{tab:candscount}, as well
1900: as statistical and systematic errors on the right-hand-side of
1901: Eqn.~\ref{eqn:rates}.
1902: }
1903: \end{table}
1904: %%
1905: \begin{figure}[!htb]
1906: \begin{center}
1907: \includegraphics[height=4.0in, width=5.0in]{f6.ps}
1908: \end{center}
1909: \caption[]
1910: {\label{fig:allrates}
1911: The filled circles are the results of this work.
1912: The first three empty squares at low redshifts (the ``low-$z$ results'') are, from lower to higher redshifts,
1913: the results of~\cite{bib:cappellaro},~\cite{bib:madgwick}, and~\cite{bib:blanc}, respectively. The open upturned triangle at $z$ = 0.47 is from~\cite{bib:neill}.
1914: The open circle at $z$ = 0.50 is from~\cite{bib:tonry}.
1915: The open star at $z$ = 0.55 is from~\cite{bib:pain}.
1916: The open diamonds are the results of~\cite{bib:barris}.
1917: The crosses are from~\cite{bib:dahlen} (including systematic errors).
1918: The horizontal bars are estimated redshift bin sizes.
1919: }
1920: \end{figure}
1921: %%
1922:
1923: %!!!!!!!!!!!
1924: Note that Type Ia supernovae that we have considered encompass a wide range of magnitudes,
1925: stretch parameters, extinction possibilities, etc.. Therefore,
1926: the procedure described in section~\ref{sec:typeia} accounts for not
1927: only the more standard Type Ia's (such as those described in~\cite{bib:branch}), but also
1928: non-standard Type Ia's, such as type 1991bg and 1991T~\cite{bib:fil}.
1929: 1991bg-like supernovae have low values of the stretch parameter
1930: ($s$ = 0.71 $\pm$ 0.05), and are typically $\sim$1.7 magnitudes
1931: fainter in the $V$-band and $\sim$2.6 magnitudes fainter in the $B$-band.
1932: Stretch values of 0.71 are certainly within the range of the stretch parameters
1933: we considered; as for the magnitudes, it is reassuring to note that the
1934: case of strong extinction ($A_V$ = 1) did not significantly alter our results
1935: (see Table~\ref{tab:systematics}).
1936: 1991T-like supernovae are about 0.5-0.9 magnitudes brighter
1937: than normal Ia's, with stretch $s$ = 1.07 $\pm$ 0.06. Both
1938: the stretch and the magnitude values are well within the considered ranges
1939: of these parameters.
1940: Note also that the fact that the Bayesian classification method was able to accurately
1941: type the vast majority of the 73 Type Ia candidates from the SNLS dataset,
1942: as was demonstrated in~\cite{bib:ourpaper}, shows that the method is capable
1943: of identifying Type Ia's in large populations that presumably include
1944: non-standard Ia's.
1945:
1946:
1947:
1948: It is particularly interesting to compare our rate results with that of~\cite{bib:dahlen}.
1949: That study also analyzed the GOODS sample, but there are important differences
1950: in our methods, as pointed out above (Sec. 1). While our results are in statistical agreement,
1951: our measured rate in a given bin can differ from theirs through either the candidate counting or
1952: the calculation of the control time/efficiency.
1953: \begin{itemize}
1954: \item
1955: {\bf Candidate counting: }
1956: In some bins, the final count of the candidates ends up being about the same for
1957: both analyses, but the actual candidates are not the same. This is not unexpected
1958: because the techniques used for the supernova identification in the two analyses are quite
1959: different. Our method provides a probabilistic rather than an absolute identification of each
1960: individual supernova based on its photometric measurements alone; the same probabilistic
1961: approach is used for calculating the efficiency and mis-identification.
1962:
1963: For example, in the highest redshift bin,
1964: we have one candidate but this is from Sample 2, which was taken after the
1965: work of~\cite{bib:dahlen} was published.
1966: However, the two high-redshift Ia candidates from Samples 3 and 4, SN-2002fx and SN-2003ak,
1967: which were used in~\cite{bib:dahlen}, did not pass our stage 2 cuts.
1968:
1969: \item
1970: {\bf Control time/efficiency:}
1971: A rigorous comparison of the control times is difficult due to the
1972: lack of tabulated control time data in~\cite{bib:dahlen}. However, a rough estimation of
1973: the control time times efficiency factor from the data given in~\cite{bib:dahlen} shows that
1974: this factor is approximately half our values for all but the highest redshift bin.
1975: \end{itemize}
1976:
1977:
1978:
1979: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1980: \section{The Star Formation History Connection}
1981: \label{sec:sf}
1982: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1983: A particularly interesting aspect of a Type Ia supernova rates analysis is
1984: the possibility of constraining the delay time between the formation of
1985: a progenitor star and a supernova explosion, which in turn helps constrain
1986: possible models for the Type Ia supernova formation.
1987: There are two leading models that have been considered
1988: in the recent literature: the so-called two-component model and
1989: a Gaussian delay model. We will now consider both of these models.
1990: Unlike section~\ref{sec:thenumbers}, now that we are considering the rates we can add the low-redshift
1991: measurements of~\cite{bib:cappellaro},~\cite{bib:madgwick}, and~\cite{bib:blanc} to
1992: our results and use the combined data in the fits.
1993:
1994:
1995: The two-component model~\citep{bib:scan,bib:mannucci} suggests that that the delay function may be bimodal,
1996: with one component responsible for the ``prompt'' Type Ia supernovae that
1997: explode soon after the formation of their progenitors; and the other,
1998: for the ``tardy'' supernovae that have a much longer delay time.
1999: Following this model, the Type Ia supernova rate can be represented as:
2000: \begin{equation}
2001: {\rm r_{V,Ia}(\bar{z}) } = A\rho_{\ast} (\bar{z}) + B \dot{\rho}_{\ast}(\bar{z}),
2002: \label{eqn:sfh}
2003: \end{equation}
2004: where $\rho_{\ast} (\bar{z})$ is the integrated SFH and
2005: $\dot{\rho}_{\ast}(\bar{z})$ is the instantaneous SFH.
2006: The first term of the equation
2007: accounts for the ``tardy'' population, while the second, for the ``prompt'' one.
2008: We use the parametric form of the SFH as given in~\cite{bib:handb}:
2009: \begin{equation}
2010: \dot{\rho}_{\ast}(\bar{z}) = \frac{ (a + b\bar{z}) \, h_{70} }{1 + (\bar{z}/c)^d},
2011: \end{equation}
2012: where $h_{70} = 0.7$, $a$ = 0.017, $b$ = 0.13, $c$ = 3.3, $d$ = 5.3.
2013:
2014:
2015: The Levenberg-Marquardt least-squares fit of the
2016: combined data to the two-component model is shown in Fig.~\ref{fig:models} (left). We obtain
2017: $A$ = (1.5 $\pm$ 0.7)
2018: $\times 10^{-14}$ yr$^{-1}$M$_\odot^{-1}$ and
2019: $B$ = (5.4 $\pm$ 2.0)
2020: $\times 10^{-4}$ yr$^{-1}$/(M$_{\odot}$ yr$^{-1}$).
2021: These results are entirely consistent with those
2022: obtained by~\cite{bib:neill}:
2023: $A$ = (1.4 $\pm$ 1.0) $\times 10^{-14}$ yr$^{-1}$M$_\odot^{-1}$ and
2024: $B$ = (8.0 $\pm$ 2.6) $\times 10^{-4}$ yr$^{-1}$/(M$_{\odot}$ yr$^{-1}$).
2025: The $\chi^2$ of the fit is 5.4 for 5 degrees of freedom.
2026:
2027: Note that the results of~\cite{bib:barris} at $z$ = 0.55, 0.65, and 0.75
2028: are somewhat inconsistent with our best-fitting two-component model,
2029: with the discrepancy at the level of 4.1, 3.2, and 5.2 $\sigma$, respectively.
2030: This can be seen from Fig.~\ref{fig:allrates}.
2031: It has been argued in~\cite{bib:neill} (who also noted that the results of~\cite{bib:barris}
2032: beyond the redshift of 0.5 appear to be rather high) that contamination by non-Type Ia's
2033: is the most likely source of the problem.
2034:
2035:
2036: It was suggested in~\cite{bib:dahlen} and~\cite{bib:strolger} that the Ia rate
2037: is a convolution of the SFH and a Gaussian time delay distribution function with
2038: a characteristic time delay $\tau$ $\sim$3 Gyr and a $\sigma$ = 0.2 $\tau$.
2039: Using the Hopkins-Beacom SFH, we find that the best-fitting
2040: parameters for such a model are
2041: $\tau$ = (3.2 $\pm$ 0.6) Gyr and $\sigma$ = (0.12 $\pm$ 0.54) $\tau$,
2042: with a fit $\chi^2$ of 2.1 for 4 degrees of freedom.
2043: The fit is shown in Fig.~\ref{fig:models} (right).
2044: For comparison, we also show the rate model obtained using the parameters
2045: from~\cite{bib:strolger} ($\tau$ = 3 Gyr and $\sigma$ = 0.2 $\tau$).
2046: %%%%
2047: \begin{figure}[!htb]
2048: \begin{center}
2049: $\begin{array}{c@{\hspace{-0.1in}}c}
2050: \epsfxsize=3.2in\epsfysize=2.8in\epsffile{f7a.ps} &
2051: \epsfxsize=3.2in\epsfysize=2.8in\epsffile{f7b.ps}\\
2052: \end{array}$
2053: \end{center}
2054: \caption[]
2055: {\label{fig:models}
2056: Left: The least-square fit of the two-component model to the data.
2057: The dashed line represents the prompt component that is proportional to the
2058: instantaneous SFH. The dotted line represents the tardy component that is
2059: proportional to the integrated SFH. The thick solid line is the sum of the two.
2060: Right: The Gaussian time delay model with our best-fitted parameters
2061: (solid line), as well as with the parameters of~\cite{bib:strolger} (dashed line).
2062: In both plots, the first three empty squares at low redshifts are, from lower to higher redshifts,
2063: the results of~\cite{bib:cappellaro},~\cite{bib:madgwick}, and~\cite{bib:blanc}, respectively.
2064: The filled circles are the results of this work.
2065: The horizontal bars are estimated redshift bin sizes.
2066: }
2067: \end{figure}
2068: %%
2069:
2070: One of the main differences between the two-component model and the time delay model
2071: is the predicted behavior at high redshifts: the former predicts an increase in the rates,
2072: while the latter, a decrease. From Fig.~\ref{fig:models} and the results
2073: of the fits of our data to both models, we find that neither scenario
2074: can be ruled out.
2075:
2076:
2077:
2078:
2079: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2080: \section{Summary}
2081: \label{sec:concl}
2082: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2083: We have analyzed the rates of Type Ia supernovae up to a
2084: redshift of 1.7 using two samples
2085: collected with the HST: the GOODS data, and
2086: the 2004 ACS sample
2087: collected in the Spring-Summer 2004 covering the GOODS North field. Using only
2088: the data from two broadband filters, F775W and F850LP, we applied
2089: a novel technique for identifying Type Ia supernovae based on a Bayesian probability
2090: approach.
2091: This method allows us to automatically type supernova candidates
2092: in large samples, properly taking into account all known sources of systematic
2093: error.
2094: We also make use of the best currently available full spectral templates for
2095: five different supernova types for the candidate typing, as well as
2096: for calculating the efficiency of our supernova search,
2097: and the control time.
2098: These templates will undoubtedly be improved over the next several years
2099: as more supernova data becomes available. Current and upcoming supernova
2100: surveys will not only provide a better understanding of individual supernova
2101: types, but may also uncover new types of supernovae,
2102: which can then be added to the Bayesian classification framework.
2103: Likewise, a better understanding of the many parameters that affect supernova
2104: observations will improve the classification scheme, which will
2105: result in better constraints on the measured rates.
2106: The calculations of the supernova finding efficiency, the control time,
2107: and the survey area are all done taking into account the specific observing configurations
2108: pertinent for the surveys, such as exposure times, cadences, and the orientations of the
2109: GOODS tiles.
2110:
2111:
2112: We carried out a comparison of the predicted and observed numbers of
2113: supernovae
2114: in redshift bins of $\Delta \bar{z}$ = 0.1, for two different models
2115: of the Type Ia supernova rates: a redshift-independent rate
2116: and a power-law redshift-dependent rate.
2117: We find that the available data fit both models equally well.
2118:
2119:
2120: For comparison with previous work, particularly that of~\cite{bib:dahlen}, who
2121: also analyzed a large subset of the data used here,
2122: we calculated the volumetric Type Ia
2123: supernova rates in four redshift bins,
2124: 0.2 $\leq$ $\bar z$ $<$ 0.6, 0.6 $\leq$ $\bar z$ $<$ 1.0, 1.0 $\leq$ $\bar z$ $<$ 1.4,
2125: and 1.4 $\leq$ $\bar z$ $<$ 1.7.
2126: We find that our results are generally consistent with
2127: those of~\cite{bib:dahlen}.
2128: Due to the larger of number supernova candidates which this Bayesian
2129: classification technique makes available, we obtain
2130: smaller or equal uncertainties in all the bins up to $z$ = 1.7. In the highest
2131: redshift bin we obtain a larger uncertainty because the signal to
2132: noise ratio is generally too low to apply this technique.
2133:
2134:
2135:
2136: We fitted the resulting rates to two leading models used
2137: in recent literature: the two-component model and a Gaussian time
2138: delay model. The former model implies an increase in the Type Ia supernova
2139: rates at highest redshifts; while the latter, a decrease.
2140: We find that the statistics of the present sample does not
2141: definitively discriminate between the two scenarios --
2142: only one supernova in this work and two supernovae in~\cite{bib:dahlen}
2143: contribute to the important highest-redshift bin.
2144: Significantly larger surveillance time would be required to
2145: arrive at a conclusive statement on the trends for the Type Ia rates
2146: at high redshifts.
2147:
2148:
2149: In the future, several ambitious new surveys are planned that will collect
2150: photometric data for thousands of supernovae in order to improve the
2151: constraints on dark energy. Individual spectroscopic follow up for every
2152: supernova candidate is likely to be impractical in these surveys.
2153: The Bayesian classification method described here has the ability to
2154: classify supernovae using
2155: photometric measurements alone,
2156: and is a promising technique for these future surveys.
2157:
2158:
2159: %%%%%%%%
2160: \section{Acknowledgments}
2161: We would like to thank Tomas Dahlen and Bahram Mobasher for providing us with photometric
2162: redshifts for a number of supernova candidates.
2163: We would also like to thank the anonymous referee for the many
2164: insightful comments and suggestions.
2165: Financial support for this work was provided by NASA through program GO-9727
2166: from the Space Telescope Science Institute, which is operated by
2167: AURA, Inc., under NASA contract NAS 5-26555.
2168: This work was also partially supported by the
2169: Director, Office of Science, Department of Energy, under
2170: grant DE-AC02-05CH11231.
2171:
2172:
2173:
2174: %%%%%%%%%%%%%%%%%%%%%%%%%%%
2175: \begin{thebibliography} {99}
2176: \bibitem[{Aldering {et~al.}(2004)}]{bib:snap} Aldering, G., {\it et al.} 2004, submitted to PASP, preprint astro-ph/0405232
2177: \bibitem[{Alexander {et~al.}(2003)}]{bib:chandra1} Alexander, D. M., {\it et al.} 2003, AJ, 126, 1772
2178: \bibitem[{Astier {et~al.}(2006)}]{bib:snls} Astier, P, \emph{et al.} 2006, A\&A, 447, 31
2179:
2180: \bibitem[{Barris and Tonry(2006)}]{bib:barris} Barris, B. J., \& Tonry, J. L., 2006, ApJ, 637, 427
2181: \bibitem[{Bertin and Arnouts(1996)}]{bib:sex} Berin, E., \& Arnouts, S., 1996, A\&A, 117, 393
2182: \bibitem[{Blanc {et~al.}(2004)}]{bib:blanc} Blanc, G., {\it et al.} 2004, A\&A, 423, 881
2183: \bibitem[{Branch {et~al.}(1993)}]{bib:branch} Branch, D., Fisher, A., Nugent, P. E., 1993, AJ, 106, 2383
2184:
2185: \bibitem[{Cardelli {et~al.}(1998)}]{bib:ccm} Cardelli, J. A., Clayton, G. C., \& Mathis, J. S., 1988, ApJ, 329, L33
2186: \bibitem[{Cappellaro {et~al.}(1999)}]{bib:cappellaro} Cappellaro, E., Evans, R., \& Turatto, M., 1999, A\&A, 351, 459
2187: \bibitem[{Chandrasekhar(1931)}]{bib:chan} Chandrasekhar, S., 1931, ApJ, 74, 81
2188:
2189: \bibitem[{Cohen {et~al.}(2000)}]{bib:cohen} Cohen, J. G., {\it et al.} 2000, ApJ, 538, 29
2190:
2191: \bibitem[{Cowie {et~al.}(2004)}]{bib:cowie} Cowie, L. L., {\it et al.} 2004, AJ, 127, 3137
2192:
2193:
2194: \bibitem[{Dahlen {et~al.}(2004)}]{bib:dahlen} Dahlen, T., {\it et al.} 2004, ApJ, 613, 189
2195: \bibitem[{Dahlen and Fransson(1999)}]{bib:dahlen2} Dahlen, T., \& Fransson, C., 1999, A\&A 350, 349
2196: \bibitem[{Dickinson {et~al.}(2003)}]{bib:goods1} Dickinson, M., \emph{et al.} 2003, in The Mass of Galaxies at Low and High Redshift, ed. R.~Bender and A.~Renzini (Berlin: Springer), 324
2197:
2198:
2199: \bibitem[{Filippenko {et~al.}(1992)}]{bib:fil} Filippenko, A. V., \emph{et al.} 1992, AJ, 104, 1543F
2200: \bibitem[{Fruchter and Hook(2002)}]{bib:drizzle} Fruchter, A., \& Hook, R. N., 2002, PASP, 114, 144
2201:
2202:
2203: %\bibitem[{Garnavich {et~al.}(1998)}]{bib:G98} P.~M.~Garnavich \emph{et al.}, ApJ 493, L53 (1998)
2204: \bibitem[{Giavalisco {et~al.}(2004)}]{bib:goods0} Giavalisco, M., \emph{et al.} 2004, ApJ 600, L93
2205:
2206: \bibitem[{Hardin {et~al.}(2000)}]{bib:hardin} Hardin, D., {\it et al.} 2000, A\&A, 362, 419
2207: \bibitem[{Hatano {et~al.}(1998)}]{bib:hatano} Hatano, K., Branch, D., \& Deaton, J., 1998, ApJ, 502, 177
2208: \bibitem[Hopkins and Beacom(2006)]{bib:handb} Hopkins, A. M., \& Beacom, J. F., 2006, ApJ, 651, 142
2209: \bibitem[{Hornschemeier {et~al.}(2003)}]{bib:horsh} Hornschemeier, A. E., {\it et al.} 2003, AJ, 126, 575
2210:
2211: \bibitem[{Johnson and Crotts(2006)}]{john} Johnson, B. D., and Crotts, A. P. S., 2006, ApJ, 132, 756
2212:
2213:
2214: \bibitem[{Kashikawa {et~al.}(2002)}]{bib:kash} Kashikawa, N., {\it et al.} 2002, PASJ, 54, 819
2215:
2216:
2217: \bibitem[{Krist and Hook(2004)}]{bib:tt} Krist, J., \& Hook, R., 2004, The TinyTim User's Guide, STScI, 339
2218: \bibitem[{Kuznetsova and Connolly(2007)}]{bib:ourpaper} Kuznetsova, N. \& Connolly, B. M., 2007, ApJ, 659, 530
2219:
2220: \bibitem[{Le Fevre {et~al.}(2004)}]{bib:lefevre} Le F\`{e}vre, O., {\it et al.} 2004, A\&A, 428, 1043
2221:
2222:
2223:
2224: \bibitem[Madgwick {et~al.}(2003)]{bib:madgwick} Madgwick, D. S., {\it et al.}, 2003, ApJ 599, L33
2225: \bibitem[Mannucci {et~al.}(2006)]{bib:mannucci} Mannucci, F., Della Valle, M., \& Panagia, N., 2006, MNRAS, 370, 773
2226: \bibitem[Mobasher and Dahlen(2004)]{bib:private} Mobasher, B., \& Dahlen, T., private communications
2227:
2228: \bibitem[{Neill {et~al.}(2006)}]{bib:neill} Neill, J. D, {\it et al.} 2006, AJ, 132, 1126
2229:
2230: \bibitem[{Pain {et~al.}(2002)}]{bib:pain} Pain, R., {\it et al.} 2002, ApJ, 577, 120
2231: \bibitem[{Pain {et~al.}(1996)}]{bib:pain1} Pain, R. {\it et al.} 1996, ApJ, 473, 356
2232:
2233: \bibitem[{Perlmutter {et~al.}(1997)}]{bib:perl} Perlmutter, S., {\it et al.} 1997, ApJ, 483, 565
2234:
2235: %\bibitem[{Perlmutter {et~al.}(1998)}]{bib:P98} S.~Perlmutter {\it et al.}, Nature 391, 51 (1998)
2236: \bibitem[{Perlmutter {et~al.}(1999)}]{bib:P99} Perlmutter, S., {\it et al.} 1999, ApJ, 517, 565
2237:
2238: \bibitem[{for review, see Perlmutter and Schmidt(2003)}]{bib:P03} Perlmutter, S., \& Schmidt, B. P., 2003, in Supernovae and Gamma Ray Bursts, ed K. Weiler (Berlin: Springer), 598
2239:
2240: \bibitem[{Poznanski {et~al.}(2002)}]{pozn02} Poznanski, D. \emph{et al.}, 2002, PASP, 114, 833
2241:
2242:
2243: \bibitem[{Renzini {et~al.}(2002)}]{bib:goods2} Renzini, A., \emph{et al.} 2003, in The Mass of Galaxies at Low and High Redshift, ed. R.~Bender and A.~Renzini (Berlin: Springer), 332
2244:
2245:
2246: \bibitem[{Richardson {et~al.}(2002)}]{bib:rich} Richardson, D.,\emph{et al.} 2002, AJ, 123, 745
2247: \bibitem[{Richardson {et~al.}(2006)}]{bib:rich2006} Richardson, D. \emph{et al.} 2006, AJ, 131, 2233
2248: \bibitem[{for more recent treatment, see also Riello and Patat(2005)}]{bib:riel} Riello, M, \& Patat, F., 2005, Mon. Not. Roy. Astron. Soc., 362, 671
2249:
2250:
2251: \bibitem[{Riess {et~al.}(1998)}]{bib:riess} Riess, A. G., {\it et al.} 1998, AJ, 116, 1009
2252: \bibitem[{Riess {et~al.}(2004a)}]{bib:riess2} Riess, A. G., \emph{et al.} 2004a, ApJ, 607, 665
2253: \bibitem[{Riess {et~al.}(2004b)}]{riess2004} Riess, A. G., \emph{et al.}, 2004b, ApJ, 600, L163
2254: \bibitem[{Riess {et~al.}(2007)}]{bib:riess06} Riess. A. G., {\it et al.} 2007, ApJ, 659, 98
2255: \bibitem[{Rosati {et~al.}(2002)}]{bib:chandra2} Rosati, P., {\it et al.}, 2002, ApJ. Suppl. 139, 369
2256:
2257:
2258: %\bibitem[{Schmidt {et~al.}(1998)}]{bib:Schmidt98} B.~P.~Schmidt \emph{et al.}, ApJ 507, 46 (1998)
2259:
2260: \bibitem[{Scannapieco and Bildstren(2005)}]{bib:scan} Scannapieco, E., and Bildstren, L., 2005, ApJ, 629, L85
2261: \bibitem[{Strolger {et~al.}(2004)}]{bib:strolger} Strolger, L.-G., \emph{et al.} 2004, ApJ, 613, 200
2262: \bibitem[{Sullivan {et~al.}(2005)}]{sull1} Sullivan, M., \emph{et al.}, 2006, AJ, 131, 960
2263: \bibitem[{Sullivan {et~al.}(2006)}]{bib:sull} Sullivan, M., \emph{et al.} 2006, ApJ, 648, 868
2264:
2265:
2266: \bibitem[{Suzuki {et~al.}(2008)}]{bib:suzuki} Suzuki, N., \emph{et al.} 2008, in preparation
2267:
2268:
2269:
2270: \bibitem[{Tonry {et~al.}(2003)}]{bib:tonry} Tonry, J. L., {\it et al.} 2003, ApJ, 594, 1
2271:
2272:
2273:
2274: \bibitem[{van Dokkum(2001)}]{bib:lacosmics} van Dokkum, P. G., 2001, PASP, 113, 1420
2275: \bibitem[{Vanzella {et~al.}(2006)}]{bib:vanzella} Vanzella, E., \emph{et al.} 2006, A\&A, 454, 423
2276:
2277: \bibitem[{Wirth(2004)}]{bib:keck} Wirth, G. D., \emph{et al.} 2004, AJ, 127, 3121
2278: \bibitem[{Wolf {et~al.}(2004)}]{bib:wolf} Wolf, C., \emph{et al.} 2004, A\&A, 421, 913
2279: %%%%%
2280:
2281:
2282: \end{thebibliography}
2283:
2284:
2285:
2286:
2287: \end{document}
2288: