1: \pdfoutput=1
2:
3: \documentclass[12pt,preprint]{aastex}
4:
5: \def\textfrac#1#2{{\textstyle\frac{#1}{#2}}}
6: \def\ainf{\alpha}
7: \def\acrit{\alpha_{\rm crit}}
8: \def\Dl{D_L} \def\Ds{D_S} \def\Dls{D_{LS}}
9: \def\hover#1{\setbox0\hbox to 0pt{\hss$\scriptscriptstyle
10: \rightharpoonup$}%
11: #1\kern0.4ex\raise 1.5ex\box0\kern-0.01ex}
12: \def\btheta{{\hover\theta}}
13: \def\kpc{\;{\rm kpc}}
14:
15:
16: \shorttitle{}
17:
18: \shortauthors{}
19:
20: \begin{document}
21:
22: \title{The cluster lens ACO~1703: \\ redshift contrast and the inner profile}
23:
24: \author{Prasenjit Saha}
25: \and
26: \author{Justin I. Read}
27: \affil{Institute for Theoretical Physics, University of Z\"urich, \\
28: Winterthurerstrasse 190, 8057 Z\"urich, Switzerland}
29:
30: \begin{abstract}
31:
32: ACO~1703 is a cluster recently found to have a variety of strongly
33: lensed objects: there is a quintuply-imaged system
34: at $z=0.888$ and several other
35: lensed objects from $z=2.2$ to 3.0 (the cluster itself is at
36: $z=0.28$). It is not difficult to model the lens, as previous work
37: has already done. However, lens models are generically non-unique.
38: We generate ensembles of models to explore the non-uniqueness. When
39: the full range of source redshifts is included, all models are close
40: to $\rho\propto r^{-1}$ out to $200\kpc$. But if the quint is
41: omitted, both shallower and steeper models (e.g., $\rho\propto
42: r^{-2}$) are possible. The reason is that the redshift contrast
43: between the quint and the other sources gives a good measurement of
44: the enclosed mass at two different radii, thus providing a good
45: estimate of the mass profile in between. This result supports
46: universal profiles and explains why single-model approaches can give
47: conflicting results. The mass map itself is elongated in the NW-SE
48: direction, like the galaxy distribution. An overdensity in both
49: mass and light is also apparent to the SE, which suggests
50: meso-structure.
51: \end{abstract}
52:
53: \keywords{gravitational lensing --- galaxies: clusters: general}
54:
55: \section{Introduction}
56:
57: An important prediction of hierarchical clustering of cold dark matter
58: is a universal profile for halos. Universal profiles were originally
59: indicated by cosmological simulations
60: \citep{1997ApJ...490..493N,1998ApJ...499L...5M} but more recently
61: phenomenological models have also produced them
62: \citep[e.g.,][]{2005ApJ...634..756A,2006MNRAS.368.1931L}, as have more
63: specialized simulations \citep{2006ApJ...653...43M}. The precise form
64: of a universal profile and the expected variation from halo to halo
65: are still being debated; \cite{2006AJ....132.2685M} discuss some
66: competing parametrizations. A further complication arises in that
67: infalling baryons would steepen the dark-matter profile somewhat
68: (adiabatic contraction), but in large clusters this effect is expected
69: to be small exterior to $\sim 20\kpc$ \citep{2004ApJ...616...16G}.
70: There is general agreement that in the inner regions of clusters the
71: dark matter profile would have $\rho\sim r^{\rm -1\ to\ -1.4}$
72: \citep{2004MNRAS.353..624D}.
73:
74: Lensing is an obvious way to measure the profiles of clusters, and in
75: this paper we will address specifically multiple-image (i.e.,
76: ``strong'') lensing. Weak lensing by outer regions of clusters is
77: also of interest \citep[for a recent example,
78: see][]{2008MNRAS.385.1431H}. Other possibilities are to use
79: generalized virial theorems and the kinematics of cluster galaxies
80: \citep{2006MNRAS.367.1463L}, or hydrostatic equilibrium of x-ray gas
81: \citep{2008MNRAS.386.1092L}. Combinations of these approaches are
82: also possible.
83:
84: Extracting a mass profile from lensing is not trivial, however. The
85: reason is that lensing depends on {\em integrated\/} mass. For
86: example, the light deflection near the Sun
87: $4GM_\odot/(c^2R_\odot)=1.75''$ says nothing about the mass profile of
88: the Sun. Similarly, a single Einstein ring measures only the mass
89: enclosed within a cylinder. Having two or more images at different
90: radii provides some more information, but is still not sufficient to
91: specify the mass profile, because it is possible to redistribute the
92: mass in a way that leaves the image configuration invariant (steepness
93: degeneracy). Much more information becomes available if the
94: background sources are at different distances behind the lens. Light
95: from further sources experiences a larger
96: effective lensing deflection and hence a larger Einstein radius.
97: Multiple Einstein radii, resulting from multiple source redshifts or
98: ``redshift contrast'', measure the mass within multiple cylinders,
99: thus providing a robust constraint on the mass profile. ACO~1703
100: turns out to be a striking example of this type of constraint.
101:
102: The non-uniqueness of mass models that fit a given set of observables
103: has been much studied in lensing theory. The steepness degeneracy was
104: discovered by \cite{1985ApJ...289L...1F}, and then rediscovered
105: multiple times; \cite{2000AJ....120.1654S} gives a short history of
106: these (re)discoveries, and a unified picture of degeneracies then
107: known. Further degeneracies continue to be uncovered from theory
108: \citep{2008MNRAS.386..307L}.
109:
110: But in practice, lensing degeneracies are most likely to be noticed in
111: the modeling process, when different kinds of models are tried out.
112: \cite{1991ApJ...373..354K} was the first to try a range of models and
113: distinguish robust features from model-dependent features. Subsequent
114: work found parameter degeneracies appearing in models in many
115: different ways
116: \citep{1994AJ....108.1156W,1999AJ....118...14B,2000ApJ...535..671T}.
117: Recently \cite{2008ApJ...674..711S} report non-uniqueness in models of
118: clusters.
119:
120: Faced with the non-uniqueness of models, three strategies are possible.
121: \begin{enumerate}
122: \item One could justify the model-type being fit from astrophysical
123: arguments, and then assume it is correct. This approach goes back
124: to the first models of the first two known lenses
125: \citep{1981ApJ...244..723Y,1981ApJ...244..736Y}. With the benefit of
126: hindsight, we find that these pioneering models inferred correctly
127: that the ``triple-quasar'' was really a quad, but predicted time
128: delays for the double quasar several times too long.
129: \item One could try and introduce new information such as stellar
130: kinematics or x-rays, as indeed \cite{1981ApJ...244..736Y}
131: advocated, to reduce model-dependence in the results.
132: \cite{2008MNRAS.384..987C} is a recent example.
133: \item One can explore model degeneracies and quantify uncertainties by
134: generating large ensembles of models, thereby also identifying the
135: best-constrained systems.
136: \end{enumerate}
137:
138: With this background in mind, the recent discovery by
139: \cite{2008arXiv0802.4292L} of 13 different multiple-image systems in
140: ACO~1703 is very exciting. Though not a well-known cluster, ACO~1703
141: is a strong x-ray source \citep{1992MNRAS.259...67A} and hence a
142: natural place to search for lensing. For modeling, the authors adopted
143: strategy~1 above, and their model is a generalized NFW profile, having
144: \begin{equation}
145: \rho = \frac{\rho_0}{{\left(\frac{r}{r_s}\right)^\alpha}\left(1+\frac{r}{r_s}\right)^{3-\alpha}}
146: \label{eq-nfwlike}
147: \end{equation}
148: where $r$ is an elliptical radius. The normalization $\rho_0$ is
149: defined indirectly in terms of the concentration $c$, such that the
150: average density within $r=cr_s$ is 200 times the
151: critical cosmological density. The models give $\alpha\simeq1.1$,
152: $r_s\simeq170''$ or $\simeq 700\kpc$, and $c\simeq3$. These results
153: appear to verify the prediction of a universal profile, but they also
154: raise some questions: (i)~Is the model an acceptable fit to the data,
155: according to a $\chi^2$ or other goodness-of-fit test? (ii)~Can
156: other, very different, models provide equally good fits? (iii)~What
157: do the fitted values of $r_s$ and $c$ really mean, when the data go
158: out to only $\sim50''$ ($\sim200\kpc$)?
159:
160: In this paper we analyze the same multiple-image data following
161: strategy~3. The technique is basically the same as in our previous
162: work on the inner profiles of J1004+411 and ACO~1689
163: \citep{2006ApJ...652L...5S}. But in this work we study more carefully
164: just where the constraints come from.
165:
166: \section{Modeling the cluster}
167:
168: \subsection{The image configurations}
169:
170: Of the 13 multiply-imaged background objects, making a total of 42
171: lensed images identified by \cite{2008arXiv0802.4292L}, some sources
172: are probably groups of galaxies, leading to similar image systems. Of
173: completely independent sources, there are at least six. Their image
174: configurations are typical of cluster lenses, and can be summarized as
175: follows.
176:
177: \begin{itemize}
178:
179: \item {\em The quint.} A five-image system \citep[1 in the numbering
180: system of][]{2008arXiv0802.4292L} with a spectroscopic redshift $z=0.888$.
181: The morphology is like the well-studied galaxy lens J0911+055,
182: except that the fifth image (which in galaxy lenses is almost never
183: observable) is here distinctly present.
184:
185: \item {\em The central quads.} Two four-image systems (15, 16),
186: resulting from a two-component source. This configuration is common
187: in galaxy lenses.
188:
189: \item {\em The incipient quint.} One two-image system (2), probably
190: two-nearly merging images of a quint. The well-known galaxy lens
191: PG1115+080 would look like this, if we knew only the two brightest
192: images in that lens.
193:
194: \item {\em The northern lemniscates.} Three three-image systems (7, 8,
195: 9), again from a multi-component source, with a typical naked-cusp
196: or lemniscate configuration. This type of image configuration
197: requires a strong quadrupole in the lens potential and is not shown
198: by galaxy lenses, but is common in cluster lenses.
199:
200: \item {\em The southern lemniscates.} Two three-image systems (10, 11)
201: with morphology similar to the northern lemniscates.
202:
203: \item {\em The northern long arcs.} Two three-image system (4, 5)
204: which could also be lemniscates, but where modelling suggests
205: further incipient images.
206:
207: \end{itemize}
208:
209: For two systems (nos.~3 and 6) in \cite{2008arXiv0802.4292L}, the
210: image configurations (that is, image parities, time ordering, etc.)
211: are less clear, though both are somewhat similar
212: to the northern lemniscates. For this reason, we do not consider these
213: two image systems in this paper. However, model ensembles that do
214: include these systems, with plausible image parities and time
215: ordering, agree with the results presented here within quoted
216: uncertainties. Many more candidate lensed objects are evident in the
217: cluster image, and it seems likely that several more lensed systems in
218: this cluster will be identified in the future.
219:
220: Fig.~\ref{fig-simp} conveniently summarizes the image configurations,
221: with the help of a very simple model: a cored isothermal sphere with
222: external shear. Such a model can roughly reproduce the image
223: positions, except that the relative sizes of the image systems are
224: incorrect. In particular, Fig.~\ref{fig-simp} gives roughly the same
225: size for the quint and the quad, whereas the real quint is much
226: smaller than the quad. The reason is that the source redshift is much
227: smaller for the quint than the quad, and this fact becomes significant
228: below.
229:
230: \subsection{The modeling method}
231:
232: We model the cluster using the {\em PixeLens\/} method. The technique
233: is described in detail in \cite{2008ApJ...679...17C}, and involves two
234: ideas. (Neither of these is unique to {\em PixeLens,} though the
235: combination is.)
236:
237: \begin{enumerate}
238: \item Instead of being cast as a parameter-fitting problem, lens
239: reconstruction is formulated as an inversion problem. The image
240: data are treated as constraints on the mass distribution. For
241: example, the five images of the quint are required to map to a
242: common source position. Astrometric errors are assumed negligible.
243: (Hence the goodness-of-fit problem does not appear, as the mass map
244: is required to fit the image data precisely.) In addition, a mass
245: map is required to satisfy some prior constraints. Specifically, in
246: this work we require that (i)~the mass distribution must be
247: non-negative everywhere, (ii)~the local density gradient must
248: point within $60^\circ$ of the direction to the brightest cluster
249: galaxy, and (iii)~no pixel other than the central pixel can be
250: higher than twice the sum of its neighbors.
251: \item Rather than a single model or a few models, an ensemble of 100
252: models is generated, which automatically explores degeneracies and
253: provides uncertainties. If a single model is desired for
254: illustration, the ensemble-average model can be used; this model is
255: in Bayesian terms the expectation over the posterior.
256: \end{enumerate}
257:
258: Formulation of mass reconstruction in strong lensing as an inversion
259: problem was
260: introduced by \cite{1997MNRAS.292..148S} and extended to combined
261: strong and weak lensing
262: \citep{1998AJ....116.1541A,2001ASPC..237..279S}. These works used a
263: form of regularization to find a single model consistent with the data
264: and optimal according to some criterion (for example, minimal
265: variation of $M/L$). Varying the regularization gave some idea of the
266: uncertainties. The basic scheme has been adapted in different ways,
267: \citep[e.g.,][]{2005A&A...437...39B,2008arXiv0802.0004D,2008arXiv0803.1199C}.
268: An interesting new idea comes from \cite{2007MNRAS.380.1729L} who
269: incorporate a constraint that additional bright images are not
270: produced.
271:
272: The technique of generating ensembles of models to explore
273: degeneracies and estimate uncertainties (no regularization is
274: involved) was introduced in \cite{2000AJ....119..439W} and later
275: packaged in {\em PixeLens\/} \citep{2004AJ....127.2604S}, although a
276: theoretical justification for the precise model-sampling strategy was
277: not available until \cite{2008ApJ...679...17C}. Other kinds of model
278: ensembles, involving parametric models, have also been used
279: \citep{2003ApJ...590...39K,2004ApJ...605...78O}.
280:
281: \subsection{Model ensembles for ACO~1703}
282:
283: We now model ACO~1703 from the lensing data, considering two cases in
284: detail: first with all the image systems discussed above included, and
285: then with the quint excluded. The possible models in the `with-quint'
286: case are naturally a subset of the possibilities in the `no-quint'
287: case. Figs.~\ref{fig-mass} to \ref{fig-crit} show some results from
288: the ensemble-average models.
289:
290: Fig.~\ref{fig-mass} shows the ensemble-average mass maps for the two
291: cases. Fig.~\ref{fig-arriv} shows the morphologies in the {\em
292: PixeLens\/} models; the with-quint case is shown, but the no-quint
293: case is similar (apart from the quint itself). Then Fig.~\ref{fig-crit}
294: compares the critical curves for the quint and one of the
295: quads---these are the first and second objects shown in
296: Fig.~\ref{fig-arriv}.
297:
298: Qualitatively, we see that the cluster is similar to the simple model
299: illustrated in Fig.~\ref{fig-simp}, provided we take 11~o'clock in
300: Fig.~\ref{fig-simp} as `north'. The arrival-time contours for the
301: {\em PixeLens\/} models are like those of the simple model, as are the
302: image configurations. The long axis of the lens is oriented
303: perpendicular to the long direction of the quint in both cases. The
304: critical curves are also similar, although the {\em PixeLens\/} curves
305: are jagged because of pixelation. (The arrival-time contours are not
306: jagged because they depend on integrals over the mass distribution and
307: hence are smoother.) Clearly, these qualitative features depend only
308: on the image identification and would appear in all models.
309:
310: But of course, examining the {\em PixeLens\/} models in detail reveals
311: many more features. First, we see in Fig.~\ref{fig-mass} that the
312: with-quint and no-quint models look similar but the latter look
313: slightly shallower. This is only true of the ensemble average; as we
314: will see below, the no-quint ensemble contains a much larger variation
315: of models, including steeper and shallower models than the with-quint
316: ensemble. A second feature is that in Fig.~\ref{fig-mass} the mass
317: contours seem to trace out the distribution of galaxies. (Recall that no
318: information about the cluster galaxies, except the location of the
319: brightest galaxy, was used in the modeling.) This suggests
320: meso-structure, or extended dark-matter structure correlated with
321: galaxies, similar to J1004+411 and ACO~1689
322: \citep{2007ApJ...663...29S}. We will not attempt to analyze this
323: possible meso-structure in the present paper, but we will see some
324: further indications below. A third feature is that the quint is
325: smaller than other image systems. In Fig.~\ref{fig-arriv} the quint
326: is shown zoomed to make it comparable to the others in size, while in
327: Fig.~\ref{fig-crit} we see that the quint has smaller critical curves.
328: This is all simply because the quint has a smaller source redshift,
329: and hence a smaller $\Dls/\Ds$, resulting in a smaller Einstein
330: radius. Thus, as a consequence of what we may call the redshift
331: contrast between the quint and other sources, we have distinct
332: Einstein radii within which the enclosed mass is well-measured. This
333: enables a good estimate of the steepness of the mass profile, as we
334: see below.
335:
336: \section{Source-redshift contrast and the inner slope}
337:
338: We now specialize to the radial profiles, but unlike the previous
339: section we take the ensemble-derived uncertainties into account.
340:
341: \subsection{Deprojecting the mass map}
342:
343: From the mass map we now take the circular average to obtain
344: $\Sigma(R)$, and then deproject to derive $\rho(r)$ by numerically
345: solving the usual Abel integral equation. Appendix~\ref{ap-tests}
346: gives details and tests of the deprojection method.
347:
348: Fig.~\ref{fig-surfden} shows $\Sigma(R)$ and $\rho(r)$ for ACO~1703
349: with 68\% and 99\% confidence intervals, when the cluster is
350: reconstructed with and without the quint. Notice what a difference
351: the quint makes: without the quint, the mass profile between the
352: innermost and outermost images could be like $r^{-1}$, but it could
353: also be like $r^{-2}$ within the uncertainties; when the quint
354: is included, $\rho(r)$ is inferred as close to $\propto r^{-1}$.
355:
356: Note that to arrive at the conclusion that the redshift constrast
357: reduces the available `model-space', the model-ensemble strategy is
358: essential. Within the paradigm of a single best-fit model, one cannot
359: formulate such a conclusion. However, a single-model strategy may
360: still provide a hint; \cite{2008arXiv0802.4292L} found that the slope
361: of their best-fit model was sensitive to the redshift of the quint,
362: and this seems to be such a hint. We emphasise that redshift contrast
363: does not act simply by increasing the radius range. Disregarding the
364: three innermost images of the quint gives a radius range almost
365: identical to the no-quint case, but still adds a redshift contrast. In
366: this case the uncertainties within the image region still shrink
367: dramatically, as with the full quint data.
368:
369: The spherically averaged density profile $\rho(r)$ in
370: Fig.~\ref{fig-surfden} also shows a `shelf' (a shallowing of the
371: log-slope) for $r \sim 100\kpc$.
372: This radius corresponds to the possible meso-structure, which we have
373: suggested above is indicated by Fig.~\ref{fig-mass}. Tests
374: (cf.~Appendix~\ref{ap-tests}) show that deprojection of substructure
375: will lead to a shelf or even a bump in $\rho(r)$.
376:
377: It is possible to fit the spherically averaged density using the
378: NFW-like form (\ref{eq-nfwlike}), and values like $\alpha=1$,
379: $r_s=500\kpc$, $c=4$ \cite[such as][derive]{2008arXiv0802.4292L}
380: are plausible within the uncertainties.
381: Formal least-squares parameter fits can be carried out, but are not
382: always meaningful, because they involve the tacit assumptions that the
383: radial bins are uncorrelated, and standard deviation over the model
384: ensemble represent Gaussian dispersions, both of which are poor
385: approximations. With this caveat in mind, the
386: formal best fit to the region between the innermost and the outermost
387: images gives $\alpha=0.95\pm0.17$, but the scale radius $r_s$
388: and concentration $c$ are unrealistic.
389: (The models ensemble without the quint gives
390: $\alpha=0.5\pm0.5$, with the formal uncertainty increasing to
391: $\pm1.5$ at 99\% confidence.)
392: We find systematically different parameters depending on whether the
393: radius range with meso-structure is included in the fit: excluding the
394: meso-structure gives $\alpha=1.22\pm0.48$. We find smaller errors if
395: we put priors on the scale length $r_s$ \citep[as
396: in][]{2008arXiv0802.4292L}, but then the best fit simply chooses the
397: largest possible scale length.
398:
399: It is easy to see why the derived $r_s$ is so unstable. To
400: infer $r_s$ and $c$ from data interior to $r_s$, one would have to
401: accurately measure the gradual steepening of $\rho(r)$. But in
402: Fig.~\ref{fig-surfden}, we see that in the well-constrained region,
403: $\rho(r)$ is probably getting shallower rather than steeper. In other
404: words, in these data meso-structure is a stronger effect than scale
405: radius and concentration.
406:
407: It is interesting to ask whether the above meso structure can be
408: explained simply by the dark matter associated with visible galaxies
409: in the cluster. There are five galaxies at a projected distance of
410: $\sim 100$\,kpc that could contribute. Four lie to the south east of
411: the cD galaxy; one to the north. We obtain a rather crude estimate of
412: the total mass in the meso-structure by subtracting the cumulative
413: mass of the best fit equation \ref{eq-nfwlike} for $r<100$\,kpc from
414: the cumulative mass of the ensemble average, and measuring the
415: residual over the range 100\,kpc$<r<$150\,kpc. This gives
416: $M_\mathrm{meso} = 7\times 10^{12}\,M_\odot$, which implies an average
417: mass per galaxy of $\sim 1.4\times 10^{12}\,M_\odot$. This indicates
418: that all the mass in the meso-structure could be associated with
419: visible galaxies \citep[as the model of]
420: [assumes]{2008arXiv0802.4292L} but does not require it. Whether
421: associated with the visible galaxies or not, however, the
422: meso-structure must be a transient phenomenon. There are four cases of
423: interest. The meso-structure could be several galaxies or one large
424: group, or it could be viewed in projection, or really lie at $\sim
425: 100$\,kpc. If projected along the line of sight, then it is transient
426: because it must rapidly move to larger projected radius. If it is really at
427: $\sim 100$\,kpc then, at the inferred densities of a few times
428: $10^6M_\odot\kpc^{-3}$, the crossing time is $\sim 0.2$\,Gyrs. If the
429: meso structure is comprised of individual galaxies, these will rapidly
430: phase mix; if instead the meso-structure is a large group then it will
431: be rapidly tidally stripped, since its density is significantly lower
432: than that of the main cluster \citep[see e.g.,][]{2006MNRAS.366..429R}.
433: In all cases, the meso-structure will be transient.
434:
435: \subsection{Towards direct comparison of lenses and simulated clusters}
436:
437: While an inferred inner profile of $\sim r^{-1}$ provides some
438: evidence in favor of universal profiles, it is desirable to be able to
439: compare lens reconstructions with theory and simulations more
440: directly. Ideally, the comparison would involve quantities clearly
441: related to the cluster-formation process. Failing that, one would
442: like to compare quantities that are well-constrained by the
443: observations. Current parameterizations do not achieve either of
444: these.
445:
446: With this in mind, let us consider the lensing deflection for a source
447: at infinity, as a function of projected radius
448: \begin{equation}
449: \ainf(R) = \frac{4GM(R)}{c^2R} .
450: \label{def-ainf}
451: \end{equation}
452: Here $M(R)$ is the mass within a circle of projected radius $R$. The
453: deflection angle can also be expressed in velocity units: if we write
454: $\sigma^2(R)=c^2\ainf(R)/(4\pi)$ then for an isothermal sphere $\sigma$
455: will be the velocity dispersion.
456:
457: While $\ainf(R)$ has no obvious connection to the formation process,
458: comparing `bending-angle curves' for galaxy and cluster simulations
459: (see Fig.~\ref{fig-sim}) indicates a clear qualitative difference: for
460: a galaxy $\ainf(R)$ tends to rise steeply at first and then stay flat
461: over an extended range, whereas for a cluster $\ainf(R)$ tends to rise
462: and then gradually turn over. This is just another way of stating
463: that galaxies tend to have $\sim r^{-2}$ profiles and flat rotation
464: curves, whereas clusters tend to have shallower profiles and rising
465: rotation curves. Put in still another way, galaxies have higher
466: concentration than clusters.
467:
468: In lensing, $\ainf$ is the quantity best constrained. Fig.~\ref{fig-ainf}
469: shows
470: $\ainf(R)$ for ACO~1703, reconstructed with and without the quint.
471: Also shown is what we may call the critical bending angle
472: \begin{equation}
473: \acrit(R) = \frac{\Ds}{\Dls} \frac{R}{\Dl} .
474: \label{def-acrit}
475: \end{equation}
476: An intersection
477: \begin{equation}
478: \ainf(R_E) = \acrit(R_E)
479: \label{def-RE}
480: \end{equation}
481: corresponds to an Einstein ring. Strictly speaking, the Einstein
482: radius is only defined for a perfectly circular lens, but it is useful
483: to take (\ref{def-RE}) as a working definition of $R_E$.
484:
485: From Fig.~\ref{fig-ainf} it is clear that all the sources except the
486: quint have a very similar $R_E$, whereas for the quint $R_E$ is
487: significantly less. This tightly constrains $M(R)$ at two different
488: radii, and is the reason for the constraint on the profile. Without
489: the redshift contrast provided by the quint, the profile becomes much
490: more uncertain, and even isothermal-type profiles are marginally
491: permitted.
492:
493: It is interesting to note that the shallower the profile, the more
494: $R_E$ depends on the source redshift. This can be inferred from
495: Fig.~\ref{fig-ainf}: for shallower profiles, $M(R)$ and hence
496: $\ainf(R)$ would rise more steeply, and hence the intersections with
497: the $\acrit$ lines would get more widely spaced.
498:
499: In particular, we see $\ainf(R)$ curve and the $\acrit$ lines
500: intersecting at $R\simeq125\kpc$ for the high-redshift sources and at
501: $R\simeq75\kpc$ for the quint. Had the $\ainf(R)$ curve been constant
502: at the $\ainf(R=125\kpc)$ value, the $\acrit$ line for the quint would
503: have intersected it at $R\simeq100\kpc$. We may say that the
504: characteristic size of the quint is 60\% that of the other systems,
505: whereas for an isothermal lens we would expect it to be 80\% the size.
506: Thus the simple observation that the quad is much larger than the
507: quint is already an indicator that the profile is shallow.
508:
509: More formally,
510: let us denote the ratio $\Dls/\Ds$ by $f$, and suppose that $M(R)$
511: mass varies in some range of $R$ as $R^\beta$. Substituting in
512: Eqs.~(\ref{def-ainf}--\ref{def-RE}) it follows that
513: \begin{equation}
514: \frac{d\ln R_E}{d\ln f} = \frac1{2-\beta} .
515: \end{equation}
516: The point-mass case ($\beta=0$) has the weakest dependence on the
517: redshift contrast, while a constant-density sheet ($\beta=2$) becomes
518: infinitely sensitive. An isothermal lens has $\beta=1$, and inner
519: clusters are expected to locally have $\beta>1$, and hence a small
520: change in $\Dls/Ds$ implies a larger change in $R_E$.
521:
522: \section{Discussion}
523:
524: This paper has been an analysis of the multiple-image lensing systems
525: reported by \cite{2008arXiv0802.4292L} in the cluster ACO~1703, and
526: comparison of the inner profile with what hierarchical structure
527: formation predicts. The main conclusions are as follows.
528: \begin{enumerate}
529: \item It is easy to model the lens, and in simple qualitative features
530: all models will agree. But models are highly non-unique in
531: important details. Readers of the online version can explore models
532: interactively: see Appendix \ref{online}.
533: \item Robust constraints can, however, be derived from ensembles of
534: models. We find the inner profile is well constrained and supports
535: the prediction of universal profiles. Similar work earlier on the
536: clusters SDSS~J1004+411 and ACO~1689 led to the same conclusion
537: \citep{2006ApJ...652L...5S}.
538: \item It is possible to identify where the inner-slope constraint
539: comes from. Behind the cluster at $z=0.28$ there is on the one hand
540: a source at $z=0.888$ lensed into a quint, and on the other hand
541: several sources at $z$ from 2.2 to 3.0 that are variously
542: multiply-imaged. The redshift contrast between the quint and the
543: other sources is responsible for the inner-slope constraint. Without
544: the quint, any constraint on the inner slope would have been much
545: weaker, and even an isothermal-type profile would fit.
546: \item The projected-density contours (see Fig.~\ref{fig-mass})
547: correlate with the cluster galaxies. Since the mass involved is much
548: more than the stellar mass of the galaxies, this suggests
549: meso-structure, i.e., extended dark-matter substructure correlated
550: with galaxies.
551: \end{enumerate}
552:
553: These results raise a further question: How can one most effectively
554: compare a reconstructed lensing cluster with simulations and/or
555: phenomenological models for structure? A non-parametric test would be
556: very welcome. We suggest that the deflection angle (and specifically
557: its redshift dependence) may form the basis for such a test. In
558: particular, we note that the shallower the lensing profile, the more
559: sensitive the lensed images are to source-redshift contrast.
560:
561: \acknowledgments We thank Andrea Macci\`o for providing the $N$-body
562: and hydro models used for comparisons and tests, John Stott for
563: providing a composite image of the cluster, Liliya Williams for
564: comments on the manuscript, and the referee, Dan Coe, for many useful
565: recommendations.
566:
567: \appendix
568: \section{Deprojection of the density profile}\label{ap-tests}
569:
570: To derive the projected density $\rho(r)$ shown in Fig.~\ref{fig-mass}
571: we first computed the circularly-averaged $\Sigma(R)$ from the lensing
572: mass map and then applied an Abel deprojection
573: \begin{equation}
574: \rho(r) = -\frac{1}{\pi}\int_r^{\infty}
575: \frac{d\Sigma(R)}{dR}\frac{dR}{\sqrt{R^2-r^2}} .
576: \end{equation}
577: To evaluate the numerical derivative and then the integral, we
578: linearly interpolate $\Sigma(R)$ up to the $R$ of the outermost image.
579: For the contribution beyond the outermost image, we assume $\Sigma
580: \propto R^{-2}$. This makes the total mass formally divergent, but
581: that is harmless for the region of interest. In fact, as noted in our
582: earlier work \citep{2006ApJ...652L...5S,2007ApJ...667..645R} and also
583: by \cite{2008arXiv0801.1875B}, the inferred $\rho(r)$ is very
584: insensitive to $\Sigma(R)$ for $R\gg r$, as long as $\Sigma$ is
585: asymptotically steeper than $R^{-1}$.
586:
587: Since we have an ensemble of models, we automatically derive
588: uncertainties. Previously, we deprojected the maximum and minimum of
589: an uncertainty band in $\Sigma(R)$ and showed the result as an
590: uncertainty band in $\rho(r)$. This caused a problem in that the
591: upper and lower range of $\rho(r)$ could switch, giving the illusion
592: of zero uncertainty at certain $r$ \citep[see the right panels of
593: Fig.~1 in][]{2006ApJ...652L...5S}. Here we deproject each $\Sigma(R)$
594: profile from the model ensemble separately, and then derive an
595: uncertainty band in $\rho(r)$. At any $r$, the band represents the
596: uncertainty at that $r$, marginalized over all the other radial bins.
597:
598: The above procedure assumes spherical symmetry and a possible concern
599: is that this may introduce a large systematic error if the cluster is
600: significantly non-spherical, as ACO~1703 evidently is. To test for
601: this, we projected and then deprojected (using the above method) 14
602: triaxial cluster halos taken from a cosmological $N$-body simulation.
603: In Fig.~\ref{fig:test} we show the worst example --- where the cluster
604: is really two merging clusters. We see that if the substructure is
605: evident in the projection, the spherically-averaged $\rho(r)$ is
606: mostly recovered to within the claimed uncertainties in the real
607: cluster. The secondary cluster appears as a shelf $\rho(r)$ when it
608: is actually a bump. The worst-case scenario is when the system is
609: projected `along the barrel' such that it appears spherical. In this
610: case, the inferred $\rho(r)$ is overestimated by a about a factor of
611: two.
612:
613: To summarize, the errors in the deprojection are generally smaller than
614: the uncertainty from lensing, except in the case where a major
615: substructure is hidden along the line of sight.
616:
617: \section{Online modeling}\label{online}
618:
619: The online version of this paper includes the {\em PixeLens\/}
620: modeling program as a Java applet. Readers can model the lens
621: interactively within a web browser. The example input uses lower
622: resolution than the paper, and a subset of the image systems (in fact,
623: the six systems illustrated in Figure~\ref{fig-arriv}), but still
624: enables one to verify the significance of redshift contrast. The
625: input can be edited online, and additional data typed or pasted in.
626:
627: The input syntax is as described in \cite{2004AJ....127.2604S} but
628: with one important new feature, namely multiple source redshifts. An
629: $M$-image lens system with source redshift $z_S$ is input as
630: $$ \matrix{ \hbox{\tt multi} & M & z_S \cr
631: x_1 & y_1 & p_1 \cr
632: & \ldots \cr
633: x_M & y_M & p_M }
634: $$
635: where $(x_i,y_i)$ are the image positions and $p_i$ encode the image
636: types (1 for a mininum, 2 for a saddle point, 3 for a maximum).
637: \cite{2008ApJ...679...17C} describes recent developments of the
638: modeling technique, including parallelism.
639:
640: \bibliographystyle{apj}
641: \bibliography{ms.bbl}
642:
643: \clearpage
644: \begin{figure}
645: \begin{center}
646: \epsscale{0.2}
647: \plotone{f1a.png}\plotone{f1b.png}\plotone{f1c.png}
648: \goodbreak
649: \plotone{f1d.png}\plotone{f1e.png}\plotone{f1f.png}
650: \goodbreak
651: \plotone{f1g.png}\plotone{f1h.png}\plotone{f1i.png}
652: \goodbreak
653: \plotone{f1j.png}\plotone{f1k.png}\plotone{f1l.png}
654: \caption{Qualitative summary of the image systems, using a very simple
655: lens model.
656: \endgraf
657: The model is a cored isothermal sphere with external
658: shear, the lens potential being
659: $$ \psi(\btheta) = \left(\theta_x^2 + \theta_y^2 + \theta_c^2\right)
660: ^{\frac12} +
661: \textfrac12\gamma(\theta_x^2-\theta_y^2)
662: $$ with $\theta_c=0.1$ and $\gamma=0.3$.
663: \endgraf
664: The top row of panels corresponds to the quint, followed by the quad,
665: the incipient quint, and the northern lemniscate. (The southern
666: lemniscate and the northern long arc are qualitatively similar to the
667: northern lemniscate.) In each row, the left panel shows the source
668: position and the caustics, the middle panel shows the image positions
669: and the critical curves, while the right panel shows the image
670: positions and arrival-time contours. The scales are not all the same,
671: but in all panels the distance between ticks is 0.1 in model units.}
672: \label{fig-simp}
673: \end{center}
674: \end{figure}
675:
676: \begin{figure}
677: \begin{center}
678: \includegraphics[width=0.49\textwidth]{f2a.png}
679: \includegraphics[width=0.49\textwidth]{f2b.png}
680:
681: \caption{Mass reconstruction of ACO~1703, superimposed on an optical
682: image of the cluster \citep{Stott:2007}. North is up and East to the left.
683: The two panels correspond
684: to all image systems included (left panel) and quint excluded (right
685: panel). Gray filled circles show image positions. Contours refer
686: to the ensemble-average PixeLens mass model. These are labelled in
687: units of the critical density for sources at infinity, or
688: $3.4\times10^{10}M_\odot/\rm arcsec^2$. The distance scale is
689: $4.2\kpc/\rm arcsec$. Note the overdensities in both mass and
690: light, especially to the SE.}
691:
692: \label{fig-mass}
693: \end{center}
694: \end{figure}
695:
696: \begin{figure}
697: \begin{center}
698: \epsscale{0.3}
699: \plotone{f3a.png}\plotone{f3b.png}\plotone{f3c.png}
700: \goodbreak
701: \plotone{f3d.png}\plotone{f3e.png}\plotone{f3f.png}
702: \caption{Arrival-time contours for six of the multiple-image systems
703: in the ensemble-average model (left panel of Fig.~\ref{fig-mass}).
704: The panels, in reading order, refer to the quint \citep[1 in the
705: numbering system of][]{2008arXiv0802.4292L}, a quad (16), the
706: incipient quint (2), a northern lemniscate (7), a southern
707: lemniscate (10), and a northern long arc (4). In all panels, ticks
708: are $10''$ apart; thus the quint is shown zoomed in.
709: Each panel uses its own contour step, chosen to best illustrate the
710: image configuration, but the step is always a few months of light
711: travel time.}
712: \label{fig-arriv}
713: \end{center}
714: \end{figure}
715:
716:
717: \begin{figure}
718: \begin{center}
719: \epsscale{0.5}
720: \plotone{f4a.png}\plotone{f4b.png}
721: \caption{Critical curves of the quint and a quad, in the ensemble
722: average model. Qualitatively, these two panels resemble the middle
723: panels of Fig.~\ref{fig-simp}. But note the size difference between
724: the quint and the quad. Ticks are $10''$ apart.}
725: \label{fig-crit}
726: \end{center}
727: \end{figure}
728:
729:
730: \begin{figure}
731: \begin{center}
732: \epsscale{0.5}
733: \includegraphics[width=0.49\textwidth]{f5a.png}
734: \includegraphics[width=0.49\textwidth]{f5b.png}\\
735: \includegraphics[width=0.49\textwidth]{f5c.png}
736: \includegraphics[width=0.49\textwidth]{f5d.png}\\
737: \caption{Surface density $\Sigma(R)$ (top panels) and deprojected
738: density $\rho(r)$ (bottom panels) for ACO~1703. The left column is
739: for models with the quint, the right column for models without the
740: quint. The dark gray and light gray bands give 68\% and 99\%
741: confidence intervals respectively. The vertical dotted lines are the
742: projected radii of the inner and outermost image. The oblique
743: line in the bottom panels shows $r^{-1}$. With the quint
744: included, the inner profile is constrained to be quite close to
745: $\rho\propto r^{-1}$. Without the quint, even $\rho\propto r^{-2}$
746: is admissible. Notice also the `shelf' in $\rho(r)$ for $r\sim
747: 100\kpc$, which may indicate meso-structure (see text).}
748: \label{fig-surfden}
749: \end{center}
750: \end{figure}
751:
752: \begin{figure}
753: \begin{center}
754: \epsscale{0.3}
755: \plotone{f6a.png}
756: \plotone{f6b.png}
757: \plotone{f6c.png}
758: \caption{The bending angle $\ainf(R)$ for two simulated clusters (left
759: and middle panels) and a galaxy (right panel), out to $\sim25\%$ of
760: the total mass. The three curves denote three orthogonal projections
761: of a simulation to produce a lens.}
762: \label{fig-sim}
763: \end{center}
764: \end{figure}
765:
766: \begin{figure}
767: \begin{center}
768: \epsscale{0.6}
769: \plotone{f7a.png}
770: \plotone{f7b.png}
771: \caption{Deflection angle $\ainf(R)$ of the circularly averaged
772: profile (for sources at infinity) against projected radius $R$.
773: The upper panel come from a
774: models with the quint included, and the lower panel from models with
775: the quint excluded. The points show $\ainf(R)$, with 90\%
776: uncertainties derived from the ensemble, against the projected radius
777: $R$. The barcode-like patterns show the $R$ values of individual
778: images. The oblique lines show the critical deflection $\acrit(R)$
779: for each source redshift; note how the quint stands out.}
780: \label{fig-ainf}
781: \end{center}
782: \end{figure}
783:
784: \begin{figure}
785: \begin{center}
786:
787: \includegraphics[width=0.3\textwidth]{f8a.png}
788: \includegraphics[width=0.3\textwidth]{f8b.png}
789: \includegraphics[width=0.3\textwidth]{f8c.png}\\
790: \includegraphics[width=0.3\textwidth]{f8d.png}
791: \includegraphics[width=0.3\textwidth]{f8e.png}
792: \includegraphics[width=0.3\textwidth]{f8f.png}\\
793: \includegraphics[width=0.3\textwidth]{f8g.png}
794: \includegraphics[width=0.3\textwidth]{f8h.png}
795: \includegraphics[width=0.3\textwidth]{f8i.png}
796:
797: \caption{Test of the Abel deprojection method for an extremely
798: non-spherical system. The upper panels show the density of an
799: $N$-body system of two merging clusters projected along three
800: orthogonal axes. The middle panels show the density $\rho(r)$
801: spherically averaged with respect to the centroid of the more massive
802: cluster (dotted curves), and the value recovered by taking the
803: circularly averaged projected density $\Sigma(R)$ and then
804: deprojecting (solid curves). The lower panels are similar, except
805: that $\Sigma(R)$ has been truncated at $200\kpc$ to mimic the strong
806: lensing regime. The worst case is in the left column, when the two
807: clusters are aligned along the line of sight.}
808:
809: \label{fig:test}
810: \end{center}
811: \end{figure}
812:
813:
814:
815: \end{document}
816:
817: