1: %\documentclass[12pt,preprint]{aastex}
2: \documentclass[apj]{emulateapj}
3:
4: \newcommand{\etal}{et~al.}
5: \newcommand{\deltah}{\ensuremath{\delta H / H}}
6:
7: %Accepted, proofed version. Lots of typos, added some
8: % sources.
9:
10: \begin{document}
11: \title{Is There Evidence for a Hubble Bubble? The Nature
12: of Type~Ia Supernova Colors and Dust in External Galaxies}
13: \shorttitle{The Hubble Bubble And SN~Ia Colors}
14: \shortauthors{Conley \etal}
15: \author{
16: A.~Conley\altaffilmark{1},
17: R.~G.~Carlberg\altaffilmark{1},
18: J.~Guy\altaffilmark{2},
19: D.~A.~Howell\altaffilmark{1},
20: S.~Jha\altaffilmark{3},
21: A.~G.~Riess\altaffilmark{4,5},
22: M.~Sullivan\altaffilmark{1}
23: }
24: \altaffiltext{1}{Department of Astronomy and Astrophysics, University
25: of Toronto, Toronto, ON M5S 3H4, Canada}
26: \altaffiltext{2}{LPNHE, CNRS-IN2P3 and University of Paris
27: VI and VII, 75005 Paris, France}
28: \altaffiltext{3}{Kavli Institute for Particle Astrophysics and Cosmology,
29: SLAC, Menlo Park, CA 94025, USA}
30: \altaffiltext{4}{Space Telescope Science Institute, MD 21218, USA}
31: \altaffiltext{5}{Johns Hopkins University, MD 21218, USA}
32: \email{ conley@astro.utoronto.ca }
33:
34: \begin{abstract}
35: We examine recent evidence from the luminosity-redshift relation of
36: Type~Ia Supernovae (SNe~Ia) for the $\sim 3\, \sigma$ detection of a ``Hubble
37: bubble'' -- a departure of the local value of the Hubble constant from
38: its globally averaged value \citep{Jha:07}. By comparing the MLCS2k2
39: fits used in that study to the results from other light-curve fitters
40: applied to the same data, we demonstrate that this is related to the
41: interpretation of SN color excesses (after correction for a
42: light-curve shape-color relation) and the presence of a color gradient
43: across the local sample. If the slope of the linear relation
44: ($\beta$) between SN color excess and luminosity is fit empirically,
45: then the bubble disappears. If, on the other hand, the color excess
46: arises purely from Milky-Way like dust, then SN data clearly favors a
47: Hubble bubble. We find that SN data give $\beta \simeq 2$,
48: instead of the $\beta \simeq 4$ one would expect from purely
49: Milky Way-like dust. This suggests that either SN intrinsic colors
50: are more complicated than can be described with a single light-curve
51: shape parameter, or that dust around SN is unusual. Disentangling
52: these possibilities is both a challenge and an opportunity for large-survey
53: SN~Ia cosmology.
54:
55: \end{abstract}
56:
57: \keywords{cosmology: observations --- supernovae: general}
58:
59: \section{INTRODUCTION}
60: \nobreak
61: In an analysis of the luminosity distances of Type~Ia supernovae
62: (SNe~Ia), \citet{Jha:07} (hereafter J07) presented evidence for an
63: offset between the Hubble constant measured from SNe~Ia with $2500 <
64: cz < 7400$ km s$^{-1}$ and from those with $7400 < cz < 45000$ km
65: s$^{-1}$. Specifically, the more distant SNe are slightly fainter
66: than one would expect, implying a high local value of $H_0$. One
67: natural explanation is a ``Hubble bubble'' -- a local monopole in the
68: peculiar velocity field, perhaps caused by a local void in the mass
69: density. The analysis of J07 was carried out using the most recent
70: manifestation of the MLCS light-curve fitter, MLCS2k2, and found
71: $\deltah = 6.5\% \pm 1.8\%$. The $\Delta m_{15}$ method
72: \citep{Prieto:06} gives similar results. In contrast, galaxy-cluster
73: distances give $\deltah \simeq 1.5\% \pm 2\%$ \citep{Giovanelli:99,
74: Hudson:04}, which is marginally inconsistent with the SN result ($1.9
75: \sigma$). Earlier studies using SNe~Ia can be found in \citet{Kim:97}
76: and \citet{Zehavi:98}. A Hubble bubble could have serious
77: implications for precision SN~Ia cosmology programs.
78:
79: In this Letter we test the evidence for the Hubble bubble using three
80: other light-curve analysis packages: SALT \citep{Guy:05}, SALT2
81: \citep{Guy:07}, and an unpublished package developed for the 3rd year
82: Supernova Legacy Survey \citep[SNLS,][]{Astier:06} data (SiFTO). An
83: analysis of the same data set with these tools does not support a
84: bubble if SN data are used to derive the relationship between SN color
85: excess and peak luminosity, but does if this relationship is required to be
86: that of Milky Way-like dust. Therefore the question of the Hubble
87: bubble is one of the nature of SN colors. A similar result was found
88: independently by \citet{Wang:07}.
89:
90: \section{LIGHT-CURVE FITTERS AND THE MEANING OF SN COLORS}
91: \label{sec:fitters}
92: \nobreak
93: A considerable amount of effort has been devoted to the question of
94: how to best fit SN~Ia light curves. The goal is to measure the
95: relative luminosity distance of different SNe, usually after
96: correcting for the width-luminosity and color-luminosity
97: relationships. These approaches differ considerably in implementation
98: and assumptions, but generally produce very similar results when used
99: to estimate the cosmological density parameters ($\Omega_m,
100: \Omega_{\Lambda}$) \citep{WoodVasey:07}.
101:
102: In this analysis we consider four fitting packages: MLCS2k2, as well
103: as three packages developed for use with SNLS: SiFTO, SALT, and SALT2.
104: The latter three are more similar to each other than to MLCS2k2. The
105: details of SiFTO will be presented elsewhere (Sullivan \etal , in
106: preparation); it is broadly similar to SALT except that the color excess
107: relation is not imposed during the light-curve fit, but rather when
108: the results of the fit are converted into a distance estimate.
109:
110: The critical issue for the question of the Hubble bubble relates to
111: how the different fitters handle SN color excesses. All of the models
112: have two parameters: one that describes the light-curve shape, and
113: one the color, both of which affect the peak luminosity. For MLCS2k2
114: these are $\Delta$ and $A_V$, for SALT they are $s$ and $c$. These
115: have different technical meanings, but essentially describe the same
116: effects. MLCS2k2 breaks the color relation into two pieces: a
117: time-independent piece\footnote{Technically, the effective value of $R_V$
118: depends on the SED, so the effects of this term are weakly
119: time-dependent.}, parameterized by $A_V$, and a time-dependent piece
120: determined by the shape of the light curve, $\Delta$. The time
121: independent piece is assumed to follow the CCM \citep{CCM:89} dust
122: extinction law. In effect, $A_V$ parameterizes the color of the SN
123: after correction for a shape-color relation. SALT follows a similar
124: prescription, with some critical differences. The wavelength behavior
125: of the time-independent piece (parameterized by $c$) is derived as
126: part of the SALT training process; the results are shown in
127: \citet{Guy:07}. This relation is reasonably similar to the CCM law,
128: and is arbitrarily normalized so that $c$ is the $B-V$ color at peak
129: luminosity (plus a small constant) for any $s$; the same is not true
130: of $U-B$ and $V-R$, which are functions of $s$. In the following we
131: refer to the time-independent piece as the color excess relation.
132:
133: The real difference is how the color excess relation is used to
134: correct the peak luminosities. MLCS2k2 makes the simplest assumption --
135: that it is due to dust. The wavelength behavior of the CCM law is
136: parameterized by the ratio of selective to total extinction ($R_V$),
137: and this is also used to convert the color excess into the magnitude
138: correction. The other fitters considered here take a different
139: approach. One of the steps in the process of turning their results
140: into a distance estimate is to empirically measure the slope of the
141: relationship between the color $c$ and the luminosity correction
142: ($\beta$), as well as between the light-curve shape and luminosity
143: ($\alpha$) by minimizing the residuals with respect to the Hubble
144: line. Here the model for the predicted peak $B$ magnitude $m_B$ is
145: $m_B = M_B + 5 \log_{10} d_L - \alpha \left(s-1\right) + \beta c$,
146: where $M_B$ is the absolute peak magnitude, $d_L$ is the luminosity
147: distance, and $\alpha$ and $\beta$ are determined from SN data; the
148: slopes can be cleanly separated from the intercept without any
149: difficulties. In this formulation, the shape-color relation is
150: absorbed into the $\alpha$ coefficient, and therefore the $\beta$
151: correction only applies to the color excess. In this analysis we
152: generally work with $B$ magnitudes, and therefore if the color excess
153: relation is purely due to dust then we expect $\beta = R_B = R_V + 1
154: \simeq 4$. Interestingly, the current constraint on $\beta$ from
155: combined SNLS and low-z data is $\beta \sim 2 \pm 0.2$, which differs
156: from 4 by $> 10 \sigma$. A similar value was found by \citet{Tripp:98}.
157:
158: What does $\beta \neq 4$ mean? If $\beta$ is interpreted as arising
159: from dust, then this would require $R_B = 2$, which is extreme. On
160: the other hand, we know that dust {\it must} be present at some level.
161: One possible explanation is that the various color parameters are
162: actually measuring some combination of two effects: dust, as well as some
163: additional intrinsic color variations that are not related to $\Delta$
164: or $s$. The $\beta$ for this additional piece is not known, but if it
165: is less than the value for dust, then the effective combined value will be
166: driven down. This additional color parameter could be interpreted as
167: additional scatter in the measured colors of SN~Ia; most of the
168: fitters studied here (except SALT) do include terms for such scatter,
169: but assume that it has no effect on the overall luminosity
170: ($\beta=0$). A worry, then, is that the relative balance of dust and
171: the additional intrinsic color might vary between SNe in different
172: environments. Alternatively, it is possible that the dust along the
173: line of sight to some SNe~Ia is different than standard dust, perhaps
174: due to scattering effects local to the SN \citep{Wang:05}.
175:
176: \section{TESTING THE HUBBLE BUBBLE WITH OTHER FITTERS}
177: \label{sec:fits}
178: \nobreak
179: The data sample used in the analysis of J07 incorporates a number
180: of SN with sparsely sampled light curves. We have placed the
181: following additional requirements on our sample: First we require at
182: least one rest frame $B$ observation within $\pm 7$ days of the
183: estimated $B$ peak, in order to ensure that $m_B$ is
184: well measured, and at least one $U$ or $V$ observation within -7 to
185: +10 days for the color. We exclude all SN with $s < 0.7$ or $s >
186: 1.3$, since these are outside the model bounds for most of the
187: fitters. Finally, we require that $E\left(B-V\right)_{MW} < 0.4$ mag
188: as calculated from the dust maps of \citet{Schlegel:98} and $c < 0.6$
189: mag, both to avoid regions of anomalous dust. This reduces the sample
190: of 95 SN used by J07 to 61 objects. Since only MLCS2k2 is trained to
191: fit $I$ data, we remove this from all fits. In order to check that
192: this reduction in the sample does not affect the results, we then
193: refit all of these SN using the most recent version of MLCS2k2. We
194: find good agreement with J07, with $\deltah = 6.0\% \pm 1.9\%$ for
195: $cz_{\mbox{void}} = 7400$ km s$^{-1}$ and assuming a flat $\Omega_m=0.3$
196: $\Lambda$CDM universe.
197:
198: We then fit the same photometry using the other light-curve
199: packages\footnote{The raw light-curve parameters are available from
200: \url{http://qold.astro.utoronto.ca/conley/bubble}}.
201: The different packages show impressive agreement in their basic
202: derived light-curve parameters, as demonstrated in
203: figure~\ref{fig:salt2_mlcs_color}. In order to measure \deltah with
204: SALT/SALT2/SiFTO, we must determine $\beta$. This is calculated
205: from the SN data, but incorporating high-redshift SNLS data as well.
206: SALT, SALT2, and SiFTO give $\beta =$ 1.82, 1.75 and 2.31,
207: respectively, with errors of about $\pm 0.16$. The different values
208: are not particularly important for the current analysis; all three can
209: be fixed to the mean value without qualitatively changing the results.
210: After applying the color and light-curve shape corrections, we obtain
211: $\deltah = 0.9\% \pm 2.0\%$, $-1.2\% \pm 2.1\%$, and $0.4\% \pm 2.1\%$
212: -- in other words, they do not support a Hubble bubble if $\beta$ is
213: fit to the data. SALT and MLCS2k2 are compared in
214: figure~\ref{fig:deltaH} for various values of $cz_{\mbox{void}}$. The
215: exact values are somewhat sensitive to the cuts applied to the sample.
216: However, it requires considerable hand tuning to find regions of
217: parameter space where any of the fitters besides MLCS2k2 detect a
218: Hubble bubble at greater than $1.4\, \sigma$, and in all cases MLCS2k2
219: gives much greater significance.
220:
221: \begin{figure}
222: \plotone{f1.eps}
223: \caption{SALT2 $c$ parameter compared with the equivalent MLCS2k2
224: value, $A_V/R_V$ for our 61 SNe. The $A_V$ prior has been removed
225: from the MLCS2k2 fits to allow easier comparison of the underlying
226: fits. Good agreement is also found in the times of maximum, peak
227: magnitudes, and light-curve shape parameters (although for the latter
228: the relations are non-linear).
229: \label{fig:salt2_mlcs_color} }
230: \end{figure}
231:
232: \begin{figure}
233: \plotone{f2.eps}
234: \caption{Hubble bubble as a function of the velocity of the
235: step in the Hubble constant, $cz_{\mbox{void}}$. Left:
236: Results for SALT (with $\beta=1.82$). Right: Results for MLCS2k2.
237: The grey band is the error in \deltah .
238: The constant value for $16000 < cz < 19000$ km s$^{-1}$ simply reflects the
239: lack of any SN in this range.\label{fig:deltaH}
240: }
241: \end{figure}
242:
243: One might be tempted to interpret this as an error in one or more of
244: the packages. However, the actual cause is more interesting: if one
245: sets $\beta = 4.1$ for SALT/SALT2/SiFTO, they {\it do} find evidence
246: for a Hubble bubble at $> 2.5 \sigma$, as shown in
247: figure~\ref{fig:deltaHbeta}. The color model for the most recent
248: version of $\Delta m_{15}$, which also favors the bubble, is similar
249: to that of MLCS2k2. The question of the Hubble bubble therefore boils
250: down to the appropriate value of $\beta$ -- is it the $\sim 4$ of
251: Milky Way-like dust, or is something more complicated going on?
252:
253: \begin{figure}
254: \plotone{f3.eps}
255: \caption{\deltah\ vs.\ the value of $\beta$ used to convert the measured
256: color parameter into a luminosity correction for SiFTO. The
257: relation is almost, but not quite, linear. Fits to low-z and SNLS
258: data give $\beta = 2.35 \pm 0.16$. \label{fig:deltaHbeta} }
259: \end{figure}
260:
261: The value of $\beta$ is relevant because the nearby portion of the
262: low-z SN sample is redder than the distant portion
263: (figure~\ref{fig:malmquist}). This is probably due to Malmquist bias
264: \citep{Malmquist:36} or other selection effects, since redder
265: SNe are fainter and harder to detect. The effects of the color
266: correction are to make the blue SNe dimmer and the red SNe brighter
267: (relatively). Malmquist bias has little effect as long as the
268: appropriate value of $\beta$ is applied across the whole sample;
269: however, if the right value is $\beta = 2$ and instead 4 is used, then
270: the distant portion will be made too faint, and the nearby portion too
271: bright, which is exactly the effect observed in J07. MLCS2k2 usually
272: includes a prior on $A_V$, and the default version does not take into
273: account the redshift-dependent effects of Malmquist bias. Adjusting
274: the prior to reflect this increases the value of \deltah .
275:
276: \begin{figure}
277: \plotone{f4.eps}
278: \caption{Colors of the low-z supernova sample using SALT.
279: The distant portion of the sample is bluer than the nearby
280: portion, which is probably caused by selection effects. \label{fig:malmquist}}
281: \end{figure}
282:
283: \section{THE VALUE OF $\beta$}
284: \nobreak
285: \label{sec:beta}
286: If, in fact, $\beta = R_B$ (i.e., it is purely dust), then we should
287: note that a large range of $R_B$ values have been observed in different
288: environments. However, a sample mean value of $R_B = 2$ would be quite
289: surprising \citep{Draine:03}, even with selection effects. Most
290: previous studies of large samples of SNe~Ia have found $R_B \simeq 3.5 \pm
291: 0.3$ \citep{dellaValle:92, Riess:96, Phillips:99, Altavilla:04}. This
292: value weakens the evidence for the Hubble bubble slightly to
293: $\sim 2\, \sigma$.
294:
295: Our empirical values for $\beta$ were calculated in a model that
296: assumed a smooth local Hubble flow. It is possible that this could be
297: artificially suppressing \deltah , so we checked this by refitting
298: $\beta$ using only high-z SNLS SN, only the low-z SN below
299: $cz_{\mbox{void}}$, and simultaneously with our fits to \deltah.
300: These give essentially the same result ($\beta \sim 2$), albeit with
301: larger errors than the full sample. Therefore, $\beta$ is robust
302: against the presence of a Hubble bubble. Monte Carlo studies indicate
303: that the bias in $\beta$ ($b_{\beta}$) due to errors in the measurement
304: uncertainties and covariances is $\left| b_{\beta} \right| < 0.02$ for
305: fairly extreme cases.
306:
307: We can also analyze the results of the MLCS2k2 fits using the $\beta$
308: framework. This differs from what is meant in J07 by fitting $R_V$,
309: which enforces a certain relation between the wavelength dependence
310: and scaling of the color excess relation. We first remove the
311: extinction correction from the MLCS2k2 distance estimates, then
312: convert $A_V$ into $E\left(B-V\right)$ using $R_V$, and finally use
313: this to fit for the value of $\beta$ by minimizing the residuals with
314: respect for the Hubble line via a $\beta E\left(B-V\right)$ term. We
315: work in $B$, and find $\beta = 2.7 \pm 0.3$, which again should be
316: compared with the expected value of 4.1. Restricting the fit to only
317: SNe below $cz_{\mbox{void}}$ does not change the results. The fits are
318: shown in figure~\ref{fig:betafits}.
319:
320: \begin{figure}
321: \plotone{f5.eps}
322: \caption{Fits to $\beta$ for SiFTO (top left), SALT (top right),
323: and MLCS2k2 (bottom left). The residuals are compared to the best
324: fitting Hubble line but without correction for the various color
325: parameters, which are shown along the abscissas. Bottom right:
326: Results for SiFTO using the late time colors as
327: described in the text. The solid lines are the best fit to $\beta$
328: as given in the text. The dashed lines show $\beta = 4.1$, which is
329: the expected value if the color excess relation is caused by
330: Milky Way-like dust. The black circles are for SNe with $cz < 7400$
331: km s$^{-1}$, and the blue squares for $cz > 7400$ km s$^{-1}$.
332: The grouping around $A_V/R_V = 0$ for MLCS2k2 is the result of the $A_V$
333: prior. \label{fig:betafits} }
334: \end{figure}
335:
336: Another technique for estimating the amount of extinction is to use
337: late time ($\sim 45$ days after peak) color measurements. The idea is
338: that at late times all SNe~Ia have a simple relationship between intrinsic
339: color and epoch, and this can be used to measure extinction \citep[][hereafter
340: P99]{Phillips:99}. The evidence for this is based on a handful of SN
341: for which there is independent evidence for low extinction, and the
342: distribution of the measured late-time colors (J07, figure~6). A
343: version of this is used in the training process of MLCS2k2.
344:
345: This suggests one more test of $\beta$. We take the subsample of our
346: 61 SNe that have late-time color measurements (from J07) and use these
347: values as our color estimate to fit for $\beta$. Carrying out this
348: analysis out for all four fitters, we find $\beta \simeq 2.3 \pm 0.3$
349: for 37 SNe. In order to eliminate any ``cross-talk'' between the color
350: parameters and the peak magnitudes, we also tested this procedure
351: using only $B$ band data to fit the peak magnitude and light-curve
352: shape and obtained the same results. Using the late-time colors
353: provided by P99 gives $\beta \simeq 1.5$ for 28 SNe. These results
354: again differ considerably from the expectation of 4.1.
355:
356: \section{CONCLUSIONS AND DISCUSSION}
357: \nobreak
358: We have demonstrated that the SN~Ia evidence for a Hubble bubble is
359: related to how SN colors are modeled. All of the approaches
360: considered agree that the wavelength dependence of the color excess
361: relation is similar to that of dust, but disagree on whether or not
362: the relation between the measured color excess and the peak luminosity
363: is also dust-like. Our fits give a value of $\beta \sim 2$, which if
364: interpreted via the CCM dust law, requires the extreme value $R_V \sim
365: 1$. Therefore, either a more complicated model of intrinsic SN
366: colors is required, which goes beyond a single light-curve shape-color
367: relation, or dust in the host galaxies of SNe~Ia is quite atypical of
368: Milky Way dust. If the former, then the late-time colors of SNe~Ia
369: vary from SN to SN, and therefore do not provide a simple measure of
370: extinction. If one does favor the single-parameter model with
371: Galactic dust, then the evidence for the Hubble bubble from SNe~Ia is
372: fairly strong. These results depend on how accurately the light-curve
373: shape-intrinsic color relationship has been modeled, so it is
374: reassuring that the four fitters are in approximate agreement on the
375: value of $\beta$.
376:
377: Requiring an additional intrinsic color relation beyond the
378: shape-color relation raises some challenges for SN research. Unless
379: this relation can be disentangled from dust, we must consider the
380: possibility that the balance of the two effects will change with
381: environment and redshift, perhaps even within the low-z sample, and
382: affect precision SN cosmology. MLCS2k2 might require a more
383: sophisticated prior that takes into account the relation between
384: extinction, light-curve shape, intrinsic color, and would also have to
385: allow for any effects of this intrinsic color on SN luminosity.
386: SALT/SALT2/SiFTO might require different values of $\beta$ in different
387: environments or for different color thresholds. Using an
388: inappropriate value of $\beta$ will mostly affect the most distant SN
389: in any survey, where Malmquist bias is important. The potential
390: systematic for a given SN survey can be evaluated by multiplying the
391: uncertainty in $\beta$ by the change in color excess across the
392: sample. If the two-component color model is correct, then constraining
393: it will not be a trivial task, but it does hold out the possibility of
394: making SNe~Ia even better standard candles.
395:
396: How can we resolve this issue? A deeper nearby supernova sample that
397: has a similar color distribution at all distances out to $cz \sim
398: 25000$ km s$^{-1}$ would provide a good test of the Hubble bubble, since an
399: incorrect value of $\beta$ would no longer introduce such an effect.
400: Note that SN cosmology analyses that restrict themselves to $z_{\rm
401: min} \gtrsim 0.015$ such as \citet{Astier:06, Riess:07, WoodVasey:07}
402: are not strongly affected by the existence of the bubble. Determining
403: if a more complicated color model is necessary requires a different
404: approach. A larger sample of SNe in low-extinction environments (like
405: elliptical galaxies), or at least in a narrow color range, could be
406: used to search for a non-dust-like color relation. A wider baseline
407: of color measurements could also help this problem; it seems unlikely
408: that the wavelength dependence of the color excess relationship will
409: continue to look like dust at all wavelengths unless it really is
410: dust. The SALT/SALT2 color excess relationship
411: \citep[figure~3]{Guy:07} displays tantalizing hints of departures from
412: the CCM law. If these can be conclusively demonstrated, it would at
413: least prove that there is more going on than Milky Way-like dust, even
414: if it might not elucidate the underlying mechanism.
415:
416: \acknowledgements
417: We thank Reynald Pain and Kathy Perrett for useful discussions.
418:
419: \begin{thebibliography}{}
420: \bibitem[Altavilla \etal (2004)]{Altavilla:04} Altavilla, G.\ \etal\
421: 2004, \mnras\ 349, 1344
422: \bibitem[Astier \etal (2006)]{Astier:06} Astier, P.\ \etal\ 2006,
423: \aap\ 447, 31
424: \bibitem[Cardelli \etal (1989)]{CCM:89} Cardelli, J.~A.\ \etal\
425: 1989, \apj\ 345, 245
426: \bibitem[della Valle \& Panagia(1992)]{dellaValle:92} della Valle, M.\
427: and Panagia, N.\ 1992, \aj\ 104, 696
428: \bibitem[Draine(2003)]{Draine:03} Draine, B.~T. 2003, \araa\ 41, 241
429: \bibitem[Giovanelli \etal (1999)]{Giovanelli:99} Giovanelli, R.\ \etal\
430: 1999, \apj\ 525, 25
431: \bibitem[Guy \etal (2005)]{Guy:05} Guy, J.\ \etal\ 2005,
432: \aap\ 443, 781
433: \bibitem[Guy \etal (2007)]{Guy:07} Guy, J.\ \etal\ 2007,
434: \aap\ 446, 11
435: \bibitem[Hudson \etal (2004)]{Hudson:04} Hudson, M.\ \etal\ 2004,
436: \mnras\ 352, 61
437: \bibitem[Jha \etal (2007)]{Jha:07} Jha, S.\ \etal\ 2007,
438: \apj\ 659, 122 (J07)
439: \bibitem[Kim \etal (1997)]{Kim:97} Kim, A.~G.\ \etal\ 1997,
440: \apjl\ 476, 63
441: \bibitem[Malmquist \etal (1936)]{Malmquist:36} Malmquist, K.~G.\ 1936,
442: Stockholm Observatory Medd., no.\ 26
443: \bibitem[Phillips \etal (1999)]{Phillips:99} Phillips, M.~M.\ \etal\ 1999,
444: \aj\ 118, 1766
445: \bibitem[Prieto \etal (2006)]{Prieto:06} Prieto, J.~L.\ \etal\ 2006,
446: \apj\ 647, 501
447: \bibitem[Riess \etal (1996)]{Riess:96} Riess, A.~G.\ \etal\ 1996,
448: \apj\ 473, 588
449: \bibitem[Riess \etal (2007)]{Riess:07} Riess, A.~G.\ \etal\ 2007,
450: \apj\ 659, 98
451: \bibitem[Schlegel \etal (1998)]{Schlegel:98} Schlegel, D.~J.\ \etal\ 1998,
452: \apj\ 500, 525
453: \bibitem[Tripp(1998)]{Tripp:98} Tripp, R.\ 1998, \aap\ 331, 815
454: \bibitem[Wang(2005)]{Wang:05} Wang, L.\ 2005, \apj\ 635, 33
455: \bibitem[Wang(2007)]{Wang:07} Wang, L.\ 2007, \apj\ submitted
456: (arXiv:0705.0368)
457: \bibitem[Wood-Vasey \etal (2007)]{WoodVasey:07} Wood-Vasey, W.~M.\ \etal\
458: 2005, \apj\ in press (astro-ph/0701041)
459: \bibitem[Zehavi \etal (1998)]{Zehavi:98} Zehavi, I.\ \etal\ 1998,
460: \apj\ 503, 483
461: \end{thebibliography}
462:
463: \end{document}
464:
465: