I mailed comments to Wu on her work on re-identifying Virgo Cluster member galaxies within the SDSS (which covers Virgo almost completely). This is a non-trivial project, because we are trying to cross-identify every member of Binggeli's original Virgo Cluster Catalog.
I tried to understand the role (and values) of the chemical potential in early-universe thermodynamics. I guess it should be obvious, but it has been a long time since I have taken a statistical physics class.
On the plane I worked out (for the first time since about 1993) how one calculates the relic abundance of an annihilating particle species in the expanding universe. Interactions (creation and annihilation processes) try to keep the species in thermal equilibrium (in abundance and in energy). The expansion of the universe dilutes the abundance and redshifts away the energy. The net creation rate (excess creation rate over annihilation rate) is proportional to the deficit in the abundance relative to the thermal equilibrium abundance. Kolb & Turner derive this with a very complicated argument, but it follows nicely from detailed balance (as they note, to their credit). The net creation rate is also proportional to the net mean effective annihilation cross section (sigma), which depends on the creation and destruction processes, the masses of the species and the decay products, and the energy distribution in the species and decay products (since you need to have sufficient energy to create new particles).
I also worked out some things related to RGB to CMYK conversion (in the context of our galaxy atlas). Most of the "algorithms" in use are wrong, because they don't consider the physics of "subtractive" color systems. I am going to write something up about this.
I am off for four weeks of travel; posting may get spotty. I may even violate the rules
at right.
In the morning, Sheldon showed us that the weak lensing signal from massive clusters is consistent with the NFW profile, and that cluster richness is closely related to mass.
In the afternoon, Navarro showed us that CDM profiles really are (plausibly) shallower than r−1.5 at their centers, with a number of arguments. He also showed us that triaxiality of galaxy halos affects rotation curves significantly; plausibly enough to place currently discrepant measurements in agreement with models.
Katarina Kovac (Groningen) came by and showed us a very nice measurement of the HI mass function. She went from raw data (1400 pointings, each producing a 3 million pixel data cube) to mass function; nice work! She has much better sensitivity than HIPASS and finds a "flat faint end slope".
Letters of recommendation are a necessary part of my job, and it is a pleasure to write good ones, but I can't say I got much research done today.
Noam Liebeskind showed us that if black holes suffer large kicks
at merging, you can't get a tight black-hole-mass vs bulge-mass relation. Liebeskind's technique was monte-carlo on the mass growth through mergers, like a technique that Alice Shapley and I discussed (in Chicago) for analysis of the width of the red-galaxy color–magnitude relation (when you allow dry merging
). But Gruzinov noted that you might get even stronger limits on kicks by looking at the centrality of the black holes, since the relaxation time for an off-center black hole might be long!
Bart Pindor dropped by and discussed lensing with me—or at least he listened when I went on about lensing statistics. He was suspicious that arc statistics could be considered evidence for non-gaussianity.
Another great day at New Views, featuring talks (mainly) about galaxies. NYU work got many mentions today, including my polemic (by Dalcanton), Masjedi's LRG–LRG clustering (by Weinberg), and Berlind's halo-occupation stuff (also by Weinberg).
In the course of a long day, I learned things too numerous to mention, but the big shocker for me was Gladders's talk about massive galaxy clusters. He showed, among other things, that the redshift distribution of lensing-arc-selected clusters is totally wrong; the number found per redshift increases with redshift when all predictions (by, eg, Dalal) have it fall with redshift. It ought to fall with redshift because at higher redshifts the clusters are less numerous and less massive and have smaller (angular) Einstein radii (even at fixed mass). The observations are so far from the data, you must either: (1) blame an enormous selection effect (and few can imagine one so strong), (2) postulate a bizarre evolution in the central densities of clusters (opposite to what you would imagine), or (3) seed the early Universe with non-gaussian cluster seeds to derail the evolution in the mass function by gravitational collapse. The last option is clearly insane, but Gladders also noted that there are some huge galaxy clusters observed at substantial redshifts. Nikhil Padmanabhan and I discussed the possibility of a blind arc search in SDSS; it would be shallow but wide.
Before school,
as it were, I started making some pretty
SDSS images of galaxies for a documentary being produced by the American Museum of Natural History.
Today I arrived in Chicago for two days at New Views of the Cosmos, a new Kavli conference in honor of Dave Schramm. I saw talks about dark matter (Frieman), neutrinos (Conrad), the composition of cosmic rays (Hoerandel), neutrino dark matter (Abazajian), accelerator production of dark matter (Baltz), positrons at the Galactic Center (Yuksel) and many others.
Something that interested me greatly was what Jedamzik said about discrepancies (relative to big-bang nucleosynthesis) in the 7Li and 6Li abundances relative to predictions (7Li is too low by a factor of 3 to 4, and 6Li is too high by a lot). This used to be thought to be a problem with stars (maybe 7Li is somehow depleted or destroyed) and cosmic rays (maybe 6Li is produced entirely by CR spallation), but now neither mechanism seems to be working well, especially since both abundances show little system-to-system scatter and a weak dependence on overall metallicity. This might be a window into new physics, such as energy injection at MeV at BBN (decaying SUSY particles?) to spallate 7Li and 4He to 3He (which then makes 6Li). As Jedamzik noted, Dave Schramm would have liked the idea that BBN anomalies could be used to discover new physics. (Later, Dave Tytler also discussed this a bit).
At lunch, Blandford and I discussed finding cosmic strings in HST data by searching for patterns of multiple imaging. Nice, and possible!
Today's talk (it has been a seminar-filled week) was by John Moustakas (Arizona), who gave us unparalleled evidence for evolution in the mass–metallicity relation. Good, hard evidence for evolution is not easy to find. Unfortunately, he can not yet show direct evidence for non-closed-box evolution (as I am always on about), such as merging or accretion or outflows.
John has put a lot of time into taking integrated spectra of a large variety of galaxies in the local Universe; he showed very nice relationships between line strengths, chemical abundances, and dust. His sample of low-redshift galaxies provides a great baseline for high-redshift studies; oddly the low-redshift part is often the hardest to get (and to get right).
Matias Zaldarriaga (Harvard) gave a colloquium about directly observing atomic hydrogen between the recombination (should be combination
) epoch and reionization, in (redshifted) 21-cm radiation. At redshifts 100 to 30, it is visible in absorption (because the spin temperature is locked to the gas temperature, which is colder than the CMB (assuming no stars or AGN have formed). At redshifts around 10, it is visible in emission, because the spin temperature becomes hotter than the CMB (although this doesn't make it very bright, since both the CMB temperature and the spin temperature exceed the characteristic temperature of the transition, about 0.07 K). Anyway, it appears to be a huge store-house of information for fundamental cosmology (better than the CMB because it is 3-d, not 2-d, and it doesn't suffer from Silk damping at small scales), and it will be measured (if all goes well) this decade.
Julio Navarro (UVic) showed us that when small galaxies merge with large, if the large is oblate, the small's orbit is pulled into the "disk plane" of the oblateness by anisotropic dynamical friction. This opens up the possibility that there is structure and accretion history encoded in the disk, especially since the radius at which stuff settles is generally related to the central density of the incoming satellite. Note the contradiction—or at least tension—with the discussion last week.
Eli Waxman (Weizmann) gave a great lunch-time talk on the ultra-high energy neutrinos, the neutrino counterparts of ultra-high energy cosmic rays. He gave a nice argument that the CRs must come from incredibly luminous sources, based on their energies. My summary of this argument is that for any energy/charge, or potential, there is a corresponding power, or luminosity, V2/R. In CGS units, resistance has units of velocity, so the speed of light is a fundamental
resistance (ie, the impedance of free space). Of course, Waxman gives a physical explanation of this, not my mystical one, but the upshot is the same: Very high energy particles means very high potentials, and very high potentials mean very luminous sources or engines. Waxman, pretty convincingly, argues that the engines are probably gamma-ray bursts.
Waxman and the late John Bahcall (IAS) showed that this CR flux requires an ultra-high energy neutrino flux, and places a strong upper limit on the extragalactic neutrino flux over a wide range of energies. Interestingly, these ultra-high energy neutrinos will be detected by the next generation of detectors (Ice Cube and the like). If they come from GRBs, they will point back
to their sources and they will be prompt, ie, simultaneous with the photon burst (both niceties are true because neutrinos don't suffer magnetic deflections). Then Waxman will deserve a big prize.
Thanks to Schlegel and others for writing very useful building blocks, I was able to hack together a piece of code that (relatively) quickly provides all SDSS sources within an angle d of any sky point. This is for fast solving of HST image astrometric WCS, of course.
On the last day of Willman's workshop, Raja Guhathakurta (UCSC) led a very nice discussion of what we know about M31 structure, substructure, and ongoing mergers. The age and metallicity of the infalling stream, apparently, is very similar to that of the bulge. He also showed that the bulge of M31 looks more like a n=2 Sérsic profile than a de Vaucouleurs, confirming my suspicions about bulge/disk decomposition for more distant galaxies. Indeed, he showed that the separation of bulge, disk, and halo is not at all easy for M31.
Steven Barwick (Irvine) gave a nice talk about ANITA, a project to measure radio Cherenkov light from the energetic neutrinos related to ultra-high energy cosmic rays. It occurred to me that it might be possible to make a long-wavelength telescope in Antarctica that measures, simultaneously, radio emission from the sky and radio emission from neutrinos in the ice.
I posted my Marseille review to astro-ph, and Quintero's paper appeared too.
Burles saved my life today by writing an extremely simple and elegant source detection system that will work on the HST and SDSS images we are using to get astrometry.net up and running.
I wrote down on the astrometry.net trac system wiki outlines of the source-detection and WCS-tweak pipelines. This will help organize my work for the next week or two. Unfortunately, I have to get both of these pipelines somewhat functional before I leave for my Christmas travel.
Willman's one-week informal workshop (five participants?) on Local Group substructure and equivalents got off to a nice start today, with discussions about current and future surveys pertaining to the structure of the Milky Way and Local Group. There was pessimism about understanding anything about the Milky Way's accretion history from studies of the very nearby (<100 pc) stellar velocity field (something we have worked on a bit here).
In the afternoon, participant Lucio Mayer gave a nice talk in which he convinced me that a lot of the angular momentum problem
and Tully–Fisher theory–observation discrepancies were highly resolution-dependent. Indeed, he showed that resolution is a much bigger effect than what is assumed about stellar feedback, at least for these problems, and at least for SPH simulations. That's news.
Neal Weiner and I got working on his use of IDL and reading of FITS files. We spent a significant amount of time setting environment variables! He asked "is this what astrophysics research is like?". I answered "absolutely".
Burles took care of the cosmic rays in the HST/ACS data on the COSMOS field. It turns out that STScI flags them for you if you have a CR split or multiple exposures with (close to) the same pointing.
I worked on the problem of least-square fitting with a soft
relationship between data and model; in particular the situation in which you are trying to get the best possible astrometric WCS for an image in the face of severe uncertainty about which source in the image ought to be identified with each astrometric standard star. Right now our WCS tweak code does a hard assignment and then updates it on each iteration, but I think there are better, continuous solutions.
I also worked on the problem of getting ultra-precise relative astrometry for, say, HST/ACS images of the COSMOS field, which is a slightly different problem from getting as-precise-as-possible absolute astrometry for each image using USNO-B1.0. It must be possible to align the HST images in a relative sense to much, much better than 0.05 arcsec, but we can't put things on the USNO-B1.0 system to better than about 1 arcsec (and the USNO-B1.0 system is not even this good, globally).
This is my 200th post (!), but it is just to note that yesterday and today comprise the US Thanksgiving holiday, so nothing was done (if cooking 90 lb of food for 22 people counts as nothing
).
I finished and submitted my Marseille review. In the end, this is what I wrote about morphology–density:
In Figure 5 I show the variation of concentration (a measure of bulge/total ratio and therefore a morphology surrogate [citations]) on environment (clustocentric distance, a high-precision but variable-scale environment measure), in narrow color slices. In Figure 6 I show the variation of color on environment in narrow concentration slices. These Figures look very different: Color depends on environment independently of morphology, morphology does not independently of color. Of course I am over-stating the result by calling this
morphology; take it as you wish, but it is clear that color and concentration are not on an equal footing when we ask what they can tell us about environment.In current thinking about the reasons for the morphology–density relation (things like ram-pressure stripping, mergers, late accretion, tidal perturbations), these results are very difficult to understand. What physical processes can
tellgalaxy star-formation rates about their environments andtellmorphologies to keep track of star-formation rates but not do much to the morphologies independently? I think the conclusion has to be that the processes that set morphology (or, really, concentraion or bulge/total ratio) are somehow internal to the galaxies.
Weiner and I discussed the inelastic dark matter results of Tucker-Smith and Weiner, and the (future) development of a physical model for the abundance that is consistent with cosmology and direct dark-matter detection experiments.
Quintero and I sharpened up his discussion of the morphology–density relation.
I worked on astrometry milestones, when I was supposed to be working on my Marseille review.
I worked on my Marseille review.
Burles assigned me the task of re-solving the astrometric WCS for all of the HST/COSMOS images, because (as he rightly pointed out) this is a perfect overlap of the PRIMUS and astrometry.net projects (see sidebar). I worked out the necessary changes to our system to make this happen. The big deal is that if you want high precision, you can't just fit to USNO-B1.0, you have to propagate faint sources in the HST imaging onto the celestial sphere, and simultaneously fit for (1) the pointing and rotation of each ACS exposure, (2) the pixel scale and camera distortions, and (3) the sky coordinates (RA, Dec) of all the faint sources that appear in any two exposures. Though some people have implemented such systems (it's not rocket science), no-one, to my knowledge, has a general, plug-and-play system. That's one of the long-term goals of astrometry.net, of course.
Rocky Kolb (Fermilab) gave a talk about the possibility that small-scale (tens of Mpc) order-unity density perturbations can mysteriously
produce cosmic acceleration. When various of us (Dvali, Gruzinov, me) asked him how, physically, small-scale variations that average to zero could affect large scales, he pointed to terms that don't obviously sum to small totals in a series expansion of cosmic perturbation theory in GR. In my view, the fact that Kolb can't sum a divergent infinite series is not an argument that acceleration is caused by small-scale inhomogeneities! But in particular, and to be concrete, since the universe he considers is non-relativistic matter (dust
in GR parlance), I asked Kolb for a physical reason that the Milne calculations (ie, the derivation of the FRW Universe in an entirely Newtonian gravity framework) are wrong, as he must assert that they are (since they agree with the conventional GR calculations and show no acceleration). He said simply that it was possible that they are.
Masjedi, Berlind, Blanton, and I discussed hypothesis testing (in the context of Masjedi's project to fit halo occupation models to his small-scale correlation function. For the Nth time, I figured out the answer to a question first asked of me by Scott Tremaine (Princeton): If you have two equally plausible models to explain a set of data, a chi-squared difference of just one or two (in total chi-squared, not per degree-of-freedom) is sufficient to prefer one model over the other. But if you have a model, and you want to show that the model is not allowed by the data, you have to show that the model has a chi-squared that is significantly larger than the number of degrees of freedom. Of course both of these statements are only rigorously justified when you have a linear problem and uniform priors on the parameters and exactly gaussian (and well-known) uncertainties.
Masjedi showed me his result that he cannot fit the LRG–LRG correlation function at small scales with a halo occupation model that uses the NFW profile for the galaxies. This is true even if he allows the concentrations of the halos to vary. This is a nice result and will motivate him to fit for the Masjedi profile
. Look out Moore and NFW!
Quintero made this plot (below), which shows that although there is a relationship between Sérsic index (concentration) and clustocentric distance (environment) when we look at all SDSS galaxies (leftmost panels, top is color histogram, bottom is dependence of quantiles of concentration on environment), if we take narrow color slices, there is no concentration–environment relation within any slice. This is not true if we reverse things; ie, if we take narrow concentration slices, there is a color–environment relation even within each slice. This reflects on the relative fundamental-ness
of the color–environment and concentration–environment relations. By the way, the latter is often called, in recent literature, the morphology–density
relation.
I continued to work on coding the magnitude gaps.
Roman Jackiw gave a talk about symmetry-breaking modifications to E&M and GR.
Blanton and I refined a bit more my ideas about the most conservative possible magnitude gap calculation. I began the implementation of it within the NYU-VAGC.
As I have discussed earlier, there is a hypothesis out there that groups with a large magnitude gap
between brightest and second-brightest galaxies are considered likely candidates to be fossil groups
, in which multiple group members have (in the past) merged into a large, remnant galaxy. I say considered
because this is far from demonstrated. In fact, the distribution of magnitude gaps is very close to what one would expect from a Poisson sampling of an envionment-dependent luminosity function. But we find that there are groups with anomalously large magnitude gaps. The question is, in the face of small redshift incompleteness (and usually spectrograph constraints cause the incompleteness to be higher in compact groups), what is the most conservative estimate of a group's magnitude gap? I think I figured that out today. I will implement it next week.
At group meeting, Blanton discussed his results regarding evolution from redshift unity to one tenth, comparing SDSS and DEEP2. He confirms the data result of Bell et al and Faber et al, but not the punchline, because he can show that the uncertainties in the modeling of stellar evolution (and the photometry of galaxies across redshift) are comparable to the evolutionary effects. So, at the moment, merging is not required to explain the evolution in the luminous part of the red sequence
.
Jim Peebles (Princeton) gave a great talk about the issues, as he sees them, with extending CDM (well tested on large scales) down to very small scales. He put a lot of emphasis on voids (as usual) and merging. The merging predicted by CDM simulations is indeed hard to reconcile with observations; on the voids his argument is "morphological" rather than statistical. But all this was very timely, because it is exactly what I am working on with NSF proposals, review papers, and research papers, but it was also, as always, solid gold Peebles!
Masjedi, Peebles, and I, in separate pairwise interactions, discussed Masjedi's results and the interpretation of the correlation function in terms of steady-state merging (at small scales). We discussed the issue of whether or not the mean, pairwise, galaxy–galaxy infall velocity dr/dt can be reverse engineered
from the correlation function.
I made pretty pictures of Virgo Cluster members (the members that confused the SDSS's photometric pipeline software) for Ronin Wu today.
Masjedi and I discussed the relationships between his limit on the local massive-galaxy merger rate and the usual "pair fraction" that appears in the literature—not a very good statistic, for a variety of reasons, the best being it is never consulted or computed when calculating any good estimate of the merger rate! We also discussed the relationship between his results and those of Eric Bell (MPIA) who, inspired by Masjedi, has started to do a similar project with COMBO-17 data.
Blanton and I also discussed Blanton's apparent conflict with Bell on the evolution in the red sequence since a redshift of unity.
It doesn't count as research
, but Quintero forced me to work out the energy spectrum of positronium, in the context of answering a (difficult) GRE question. We got it, and we got the right answer, but only after a somewhat embarassing amount of calculation and re-calculation!
Here's a question for Steve Thorsett (UCSC), who I am hoping is a lurker here at Hogg's Research:
After a nice talk by Gruzinov here about pulsar emission (Gruzinov has recently discovered a force-free electrodynamics "solution" for a rotating dipole, with rotation and dipole axes aligned), he and I talked a bit about the Eddington limit. No pulsars appear to emit above the Eddington limit (even when you consider total nebular emission and/or implied spin-down luminosity). The Crab and one other are basically at the Eddington limit, and all others are below. Is this a coincidence? Part of me says "yes", because gravity doesn't enter in any of the usual considerations about the emission mechanism. But part of me says "no", because (a) nothing emits above Eddington, and (b) if the pulsar is above Eddington, the emission might significantly distort the plasma in the magnetosphere and "break it" or even distort the outer layers of the NS, and change the moments of inertia.
Interestingly, if people are right about spin-down, the Crab was hugely super-Eddington in the past, and it is a coincidence that we see the Crab only now just as it has passed into the sub-Eddington phase! That sounds like the kind of argument a cosmologist would make.
Of course I am talking about the total nebular emission, not the pulsed emission (which, I understand, is a small fraction of the total—either nebular or spin-down—luminosity).
I spent quite a bit of time this weekend updating, organizing, and creating bug reports on the astrometry.net internal tracking system.
Discussed PRIMUS extensively with Eisenstein (Arizona) and Blanton, and worked a bit on the Steward/Magellan proposal for the pilot project.
Quintero spoke at group meeting about his clustocentric results, and got a good discussion going about the lack of a morphology–density relation; ie, the result that concentration depends on environment only because it depends on color and color depends on environment, and not vice-versa.
Judy Cohen (Caltech) went through the basics of AO, and then showed incredible images of M31 globular clusters taken with the Keck AO system. Awesome! Even more awesome when compared side-by-side with HST images.
I spent a lot of today in discussions with Judy Cohen (Caltech) and Blanton. We discussed the editorial
position that all the data on low and high redshift galaxies (where by high
I mean z=1) is consistent with no evolution in masses, sizes, angular momenta, etc, but only evolution in stellar populations. This is the position Peebles and I both basically hold: that it has been very hard to find incontrovertible evidence that galaxies accrete or merge significantly since redshift unity. There are many conflicting results in the literature, but this just bolsters this position, in my opinion. Until recently, I had been convinced by Bell et al that the galaxies on the red sequence must be growing by mergers, but in fact this study is now in direct conflict with the new (as yet unpublished) work on larger samples by Blanton.
Inspired by a conversation yesterday with Yann LeCun (NYU CS), I wrote an "elevator pitch" (short description, possible to say during an elevator ride) for the possible "data mining" project we have imagined as the full generalization of Willman's search for Milky Way companions: Find all statistical anomalies in the distribution of stars on the sky, using positions and magnitudes in all available bands.
I realized today that any kinematic calculations one might do relating dark matter experimental data to constraints on the dark matter's physical properties are all non-relativistic. Duh! I started calculations relevant to reinterpretation of dark-matter experiments in terms of a dark-matter model in which there are two dark matter species with a small mass difference (the project assigned to me by Weiner). Apparently everything I am doing has been done before, but I am not yet discouraged.
Neal continued his lunch school on "beyond the standard model". I learned something.
I demonstrated (to myself) the (obvious, apparently) fact that when two hard spheres collide elastically in three-space, the cosine of the angle between the incoming and outgoing momenta for each particle has a flat probability distribution function in the center-of-mass frame. Don't ask me why I did this.
Fiddled with some other astrometry issues, and Schiminovich corrected me (quite correctly) that he didn't say we wouldn't be able to solve the GALEX astrometry blind, he said that he would be impressed if we solved the GALEX/FUV astrometry blind, with no help from the NUV channel. And he said that would be useful to boot. He threw down a serious gauntlet today; we expect to pick it up (soon).
I started making the 12 catalogs in the new regime, in which we split the USNO-B1.0 catalog into twelve separate catalogs, each of which does 1/12 of the sky. This allows us to operate with less RAM (but requires us to run simultaneously on more CPUs).
I wrote a few words for Morad's nearly-done paper on LRG clustering at very small scales.
Roweis and I spent all day working on getting SDSS fields to solve in astrometry.net, as well as short-term planning for the project.
Roweis also gave a very nice talk about one of his alogorithms for aligning one-dimensional datasets with variable x and y scalings. We might be able to use it to solve, once and for all, robustly, Burles's perennial problems with automatic arc solutions in spectrograph calibration.
Woah! The WCS tweak code I wrote this summer worked straight out of the box on the GALEX solution Lang found yesterday.
Roweis and I got set up to run all the SDSS fields; we have a bet on whether the 5 arcmin quad index or 10 arcmin quad index will do better.
I wrote material in my Marseille review about the LRG–LRG merger rate. We learned last week in group meeting that it is hard to show just how robust is Masjedi's strict upper limit.
Dustin Lang (hey, this isn't Lang's Research, is it?) successfully found the WCS for a GALEX field blind! This is big news, especially since Schiminovich predicted that we wouldn't succeed. Now on to all GALEX fields. I spent some time fiddling with Keir Mierle's new (and fun) wiki for the astrometry.net project.
Burles took delivery of the prism for the PRIMUS project. It looks awesome!
Yes! The insane idea Phil Marshall and I had yesterday for finding gravitational lenses automatically worked. What we do is (a) subtract smooth models from sources in HST imaging that appear to be massive galaxies at reasonable redshift, producing residuals (including, one hopes, lensed background galaxies), (b) loop over possible parameters for a simple lens model, fixing the center of the lens at the center of the massive galaxy, (c) throw image-plane pixels back to a pixelized "source plane", but (d) taking the minimum image intensity value among the values of the image-plane pixels that hit each source-plane pixel, and (e) measuring the surface brightness inside the high-magnification (principally the multiple-imaging) part of the source plane, as a function of the lens model parameters. This method, because it takes the minimum of the image-plane pixels that hit each source-plane pixels, in effect uses the blank parts of the image plane to "veto" the possible locations on the source plane (and hence models that require that originating location) from which the image-plane intensity might originate.
The amazing thing is, it seems to work, in that it finds a high source-plane intensity (after this min-value filtering) for sources that do look like lenses, and it finds a peak in the intensity at values of lens-model parameters that are consistent with what is found from direct, manual lens modeling. Here's an example plot of the source-plane intensity as a function of the lens (singular isothermal sphere) effective radius for something in Lexi Moustakas's LEGS survey. Note the peak around 20 pixels, which Marshall found previously by modeling this source. Non-lenses show no such peak anywhere.
Neal Weiner gave a nice pedagogical lunch talk setting up the problems with the standard model. He is doing two lunch talks; the first (today) was the standard model, and the second will be "beyond" the standard model.
In other news, Weiner and I are taking on graduate-student-style projects from each other. Weiner is going to find statistically anomalous galaxy groups in Berlind's group catalog, and I am going to work out some cosmological implications of direct dark-matter searches in the context of a few specific dark matter models. More as these projects get underway.
Phil Marshall (SLAC) is in town for 30 hours or so. We discussed, worked out simplifications of, and planned a pair coding session on a version/modification of the Hough transform that might be used to automatically detect gravitational lenses in massive data sets. Tomorrow we code.
Roweis and I discussed the next steps in the astrometry.net project, some of which Mulin has already done, because he has our new machine up and running and on the net with a static IP. I am pretty confident (on the basis of no data) that we will be able to solve Schiminovich's GALEX field, but that is yet to get underway. We discussed some small non-uniformities in the USNO-B1.0 catalog that appear to be related to the photographic plate mapping pattern on the sky. We made some hypotheses about what they are, none of the hypotheses were good!
Willman, Ronin Wu, and I discussed Wu's project to re-determine the properties of Virgo Cluster galaxies. Her plots look great and she is starting to write up the method into a proto-paper. I think there is a lot to do with this data set. Willman mentioned vanGorkom's beautiful HI data on Virgo; I should go up to Columbia and ask for an audience!
Rebecca Bernstein gave a great talk about chemical abundances in globular clusters. She explained how chemical abundances are related to formation history of galaxy stellar populations. More impressively, she explained in detail how chemical abundances can be determined with observations of stars and (harder) with observations of compact populations of stars. All this work was done with a spectrograph she built herself, so it was an impressive show as well as being informative.
I got another page written in my Marseille review.
Rebecca Bernstein (Michigan) is visiting. We discussed galaxy clusters, in particular intracluster light, fossil groups, and environments. We discussed the Dark Energy Survey. Most interestingly, we discussed the process by which she designed the camera for the DES, all the hard and soft requirements, and how to improve an okay design to a good design incrementally.
I wrote code to make full-color images of Berlind's groups, so we can inspect the outliers in the magnitude gap distribution I discussed yesterday. The code isn't quite ready (because Blanton has to give me the Rvir code), but it is close. We find that the gap distribution is slightly larger than expected from the pure independent, Poisson sampling distribution (there are more groups with big gaps); the question is, do these large-gap clusters all have correctly computed magnitude gaps?
Inspired by some results shown by Milos Milosavljevic in his talk yesterday, Blanton and I tried to do a single-afternoon project on the distribution of magnitude gaps between first-ranked and second-ranked and between first-ranked and third-ranked galaxies in Berlind's groups. The distributions are barely different from what you would expect if galaxy groups are simply random, Poisson samplings of the relevant luminosity functions. We are trying to figure out just how different. The big picture is: Are the groups with very large magnitude gaps old "fossil groups" where the original members have all merged into the central galaxy, or are they just the random occurences you get from sampling the luminosity distribution randomly? We didn't finish, but we got everything ready for the quantitative analysis.
Milos Milosavljevic gave a talk elucidating the role and consequences of dissipationless clustering and merging for massive galaxies, under the assumption (well justified observationally) that there is no new star formation in these objects.
I nearly convinced Neal Weiner to do our "binomial outliers" project: looking for groups of N galaxies from Berlind's group catalog that are very unlikely—given binomial probability—of containing m galaxies that have some unusual or remarkable property. An easy project, but certain to produce a few LPU.
I continued to work on my review for Marseille. In working on the end part, where I exhort my colleagues to work on certain crucial problems, I have decided that I am going to say something about constrained realizations. There is nothing—in principle—to stop us from having simulations not just of the cosmological model, but of our actual Universe, constrained by all the multi-wavelength things we know about all lines of sight we have observed (including the actual galaxy properties, their positions, the structures in which they reside, and, in principle, our actual CMB).
There are huge degeneracies, both because the observations are highly incomplete and because the model is extremely uncertain (where it comes to galaxy formation especially). But it is possible—in principle, and soon in practice—to integrate over these degeneracies statistically with Markov chains or Gibbs samplers and the like. The chain
of allowed realizations would make predictions (those things that were common to all models in the chain but not yet observed) and direct decisive new observations (those things that varied enormously among models in the chain). And each link in the chain would give specific initial conditions and formation history for every observed object, and all the not-yet-observed ones too! It is hard to imagine this future, but it is neither impossible nor ridiculous. Indeed, once simulations get a bit faster and more user-friendly, I will start to pursue it myself!
I also adjusted Wiktionary entries on physics subjects (I don't think this counts as research, but hey, it's the weekend).
Todd Thompson (Princeton) gave a great, thought-provoking talk (with lots of lively audience participation) on the possibility that all star-formation (especially in starburst galaxies) is limited by radiation pressure acting on the dust mixed in with the cold gas. Interestingly, the hypothesis is hard to rule out directly, and gets a lot of the order-of-magnitude calculations for galaxies right.
Quintero and I virtually decided that his paper is about the information that galaxies know
about their nearest cluster; we adjusted some of the wording to get this accross.
After a conversation with Sam about the memory limitations on his side of the astrometry project, I re-wrote the USNO subsampler to rank its output such that the user (Sam and Dustin and Keir) can simply truncate the catalog I give them and they will still get a uniform, all-sky catalog. This was not hard, of course. I am re-building the catalogs (yet again).
Brant Robertson (Harvard) gave a nice (informal) talk on the influence of black holes on galaxy formation, in the context of galaxy–galaxy mergers. His disk–disk mergers do seem to produce galaxies that look remarkably like early-type galaxies, and he finds that the feedback from black-hole accretion is extremely important. Unfortunately he has lots of parameters, many of which are going to be hard to explain. But it was very convincing that the properties of ellipticals and lenticulars are not, by themselves, a problem for the dominant paradigm of cosmogony.
Marcelo Magnasco (Rockefeller) gave an interesting talk in which the first half was about how the ear works mechanically and the second half was speculation about how it might work from an information processing point of view.
I worked on my review paper for the Marseille meeting.
Phil Marshall and I discussed the possibility of making a web-based user interface or wiki to get community or student feedback on whether individual extracted HST images of massive galaxies are gravitational lenses. This is similar to things Blanton and I have discussed to get the SDSS galaxies classified morphologically. The idea is crazy, but interesting.
With help from Quintero, I inspected the Juett's XMM targets by eye.
Today, we discussed Morad's results on the correlation function with Milos Milosavljevic (Caltech). Milos said that he has a theoretical result that the inflow of galaxies (subhalos) into clusters (halos) is linear with time, ie, the galaxies flow in at a constant linear velocity. This relatively naturally leads to a 3-space correlation function that goes like r−2.
I gave K+A galaxy targets to Adrienne Juett (Virginia) for an XMM proposal on neutron star binaries (ie, to find binaries with A and later-type stellar companions).
I started (and nearly finished) the job of making the uniform (via healpix) version of the GALEX-optimized astrometric standards catalog (Schiminovich has given us a challenge, if you recall). I gave Morad comments on the second half of his paper.
Sam tells me that his student Dustin Lang has confirmed numerically my (trivial) prediction that the number of valid quads for the astrometry project goes as the sixth power of the angular separations considered. Apparently Dustin has made it possible to quickly assemble the set of all possible (allowed) quads, given a star catalog.
At group meeting we discussed Beth's work on ultra-low-mass companions to the Milky Way: discovery, follow-up, theory, etc. She has enough to give three talks on the subject, at least! Her limits on detections are awesome; it is surprising, frankly, that she finds so few objects, because, eg, she could find things 10 times fainter than Draco, at distances up to a few times further away (ie, in ten or more times the volume).
Romeel Davé (Arizona) gave a characteristically great talk about numerical simulations, where they fail, where they succeed, and prospects for the future. As mentioned previously in this blog, he makes quite a bit of there being two modes of accretion: hot (passing through an accretion shock) and cold (radiating gravitational potential energy as it accretes). There are hints that this may produce the "bimodality" in galaxy properties, or the red sequence. But even sophisticated models do not yet produce the red sequence "naturally". Morad and I got Romeel to agree to make and publish some specific theoretical predictions related to Morad's current and future papers on the galaxy–galaxy merger rate.
I worked more on my review paper.
Blanton, Quintero, and I discussed a project on the vector field described by galaxy dipole moments (eg, the vector offsets between the centroid of the bulge and disk, or old stars and new). It is interesting, because some investigators, including Vogt et al, find that galaxy asymmetries are related to the direction to the nearest rich cluster.
Mikhail Shaposhnikov gave a nice talk on a dark matter model employing very weakly interacting sterile neutrinos. Interestingly, large parts of this parameter space could be ruled out by very high (spectral) resolution x-ray spectroscopy, because generically sterile neutrinos will have decay pathways to the standard neutrinos, emitting a photon.
Today Blanton and I discussed what we would do with 100 nights on a good 2 m telescope. What's the point of doing yet more redshift surveys at redshift of about 2/3? As I see it, you got (a) baryon feature, (b) precision tests of the growth of large-scale structure, or (c) evolution of galaxies and star-formation activity. I can't think of much else, except rare objects, lensing, mergers, etc., most of which fall into category (c). So if you are going to do a new survey, you better be able to show that you can do one of these better than anyone else.
Quintero's paper is nearly ready for prime time.
A combination of meetings, refereeing, and code bugs (yes, bugs in the code to make the damned USNO-B1.0 subset!) kept me from significant research progress today.
Thanks to Fink, I got healpix working and started making Sam his perfectly uniform sky catalog for automated astrometry. I made a healpix grid with Nside=64 (49152 pixels), and am keeping the brightest 1024 USNO stars (subject to R>14 mag) in each pixel. I return the full, all-sky list in brightness order, from brightest to faintest. So Sam will get a very uniform all-sky catalog with a full brightness ranking.
It would be nice, philosophically, if we could operate astrometry.net (both server-side and client-side) with no reference to specific brightnesses beyond ranking. This would relieve our users of anything approximating calibration (or even nonlinearity correction).
I (finally) finished Sam's GALEX-optimized star catalog based on the USNO-B1.0 astrometric standard catalog. It is limited in the B band, and all-sky. Unfortunately, the Galactic Plane is a big issue. I decided that for the next iteration, I need to make a catalog that is uniform in angular density. This means putting a healpix grid on the sphere, and keeping N stars in each healpix pixel. This would be hard work if Finkbeiner hadn't already put healpix into idlutils.
I also went over more of Quintero's paper on clustocentric distance dependencies for SDSS galaxies relative to Berlind's clusters. Quintero's results have a pretty sharp feature (break in dependency slope) at 1 virial radius, which implies that the Berlind mass estimates can't be all that bad.
In my limited research time today, Sam and I tried to figure out what is incompatible about our binary files. Our research was not illuminating.
[Sorry, no posts for a few days because of big—and very, very good—events in my personal life; email for details.]
Today, Morad and I spent a few minutes going over his paper on the LRG–LRG merger rate and small-scale clustering. It is very close now.
Sam and I discussed ways to prioritize quads, both at index time and at test time in the automated astrometry project. I started to make him a list of astrometric standards that is tailored (somewhat) to GALEX data.
Sheldon led a very nice—and lively—group meeting. He talked about the Dark Energy Survey, and, more importantly, dark energy.
Fabio Governato gave an impressive talk about making galaxies in N-body simulations, with approximate treatments of star formation and stellar-driven feedback. Among other things, he showed that the thick disk is likely to have been thickened by a merger, and the age of its stars is likely to be indicative of the cosmic time of that last major merger. I discouraged him from trying to match morphological type fractions in his simulations, since no two people agree on any galaxy's morphological type!
Ari Maller, Blanton, and I discussed the dependencies of observed galaxy properties on axis ratio.
I worked through parts of Quintero's clustocentric distance paper with Quintero. I discussed statistical estimation issues in a paper by Mark Brodwin (JPL) with Blanton and Sheldon. I discussed Willman's paper on Willman 1 stellar populations with Willman. I started to read the latest draft of Masjedi's paper.
Willem Kegel (Utrecht) gave a colloquium in which he showed conditions on colloid particle interactions that lead to the creation of a "cluster" phase. He made a cute analogy with nuclear matter.
I worked out the complete list of free parameters (or equivalent) for the blind astrometry system. There are about 10, which is too many; it would be nice if we could remove some or optimize over them. There are some short-term prospects for getting the list down to of order 5, which would be much more manageable. I also worked on outputting better stellar catalogs for Sam's use.
In related news, we decided to try to solve some GALEX images, which may be non-trivial given the large wavelength difference between GALEX (UV) and USNO (visual). Sam is very optimistic, I am guardedly optimistic, Schiminovich thinks we are very unlikely to succeed!
Morad and I discussed, and Morad went some way towards implementing, a K correction method for his next project that will be super-fast. Because the project (the cross-correlation of spectroscopic SDSS LRGs with essentially all imaging sources) involves doing between 107 and 108 K corrections, it needs to be fast. Fortunately, it does not have to be supremely precise, so our approximations are likely to be allowed.
Anil Seth visited from UW today. He showed us some very nice results on resolved stellar populations and vertical structure in nearby, edge-on, low-mass disk galaxies. Apparently the Milky Way has a thinner disk than most of these neighbors (interesting), but it might be comparing apples to oranges, given that the MW is so much more massive.
Nikhil Padmanabhan and Doug Finkbeiner came to NYU today to talk about astrometry and large data management (for SDSS and Pan-STARRS). We ended up spending most of our time on variable PSF fitting; Fink has a new method that is simultaneously simple, fast, and capable of catching subtle and detailed effects. One of the issues for a future psf.net
is the automatic selection of stars apart from galaxies. That's not trivial.
Sam and I realized that we might be able to speed up astrometry.net by making more use of the flux-ranking of stars. If, at index time, we build quads not out of random stars but out of the brightest set that meet our angular requirements, and if, at test time, we start with the brightest quad in the image and cascade to fainter quads, then we are more likely to make matches sooner. It was a great discussion, but we put many more items onto our to-do list than we took off!
Beth showed beautiful plots today of Willman 1's tidal features. She, Blanton, and I discussed issues relating to making isopleths of the angular density of stars (it requires smoothing). Of course, there appears to be substructure in the outer parts of the system, but is that substructure (a) real, and, if so, (b) physically signficant?
Sheldon, unintentionally, managed to convince me for a whole hour that I had a serious (and publicly humiliating) bug in my SDSS astrometry (or at least routines that use and display it). Fortunately, there was no bug!
Ed Farhi gave a nice colloquium on quantum computing. His group has had some significant success in demonstrating the theoretical power of quantum computing. I still don't fully understand what aspects of the power come from superposition, and what aspects come from the fact that it is analog rather than digital.
I fixed many little bugs in the code that makes Sam's list of x,y lists. I can't believe (actually, I can believe) that such a simple task can take so much time and effort!
Sheldon and I discussed what it would take to make an interactive, google-maps-like interface to the SDSS imaging data.
Among other things, Morad, Blanton, and I discussed how one selects sources for spectroscopy with a spectrograph that has physical constraints on how close any two sources can be and get spectra. The desideratum is that we be able to extract nonetheless the correlation function on all scales, including those smaller than that corresponding to the mechanical constraint (as Morad has done for SDSS). We discussed two schemes, both of which involve random sparse-sampling of the sources in collision groups,
ie, groups of objects that are connected by friends-of-friends with a linking length equal to the mechanical constraint.
I was out sick today, but I did manage to shepherd the creation of a huge number of SDSS images of IRAS-selected sources, after a request (and then helpful information) from Nurur Rahman. Here's one:
I wrote code to overlay arrows on pretty pictures, pointing from one object to another. Here's an obvious realization I had about it: If object positions are given relative to us (eg, in RA,Dec,distance), then their radial distances from us do not come into the directions of the arrows at all. To test my code, I made the arrow pointing from Willman 1 to the UMa dwarf (Willman 2). It is shown below (on a weird coordinate system because I was stress-testing my code).
I started to assemble the x,y positions of every compact source in every SDSS field in order for Sam to run every one of them through the blind astrometry system. The plan is to simultaneously run them all, investigate faliures, and modify and expand the index.
I also worked on the To-do lists for the astrometry.net project and what is now the PRIMUS
project (formerly prism spectroscopy
).
Today was all talk all the time. We had David Schiminovich and Ben Johnson (no, not any of these Ben Johnsons) down from Columbia to discuss Ben's recent results on the inter-comparison of GALEX, SDSS, and Spitzer photometry and spectroscopy of normal
galaxies. Ben finds that he can separate dust and age very clearly (though they do not separate clearly in optical bands), and that there is great promise for understanding galaxy spectral energy distributions from the far UV all the way through to the far IR. At the same time, he confirms a result of ours that there is diversity among galaxies in the mid-IR at low luminsoities. This motivates me to get cracking on my K corrections for Spitzer,
a product that I am (famously) contractually obliged to deliver. We discussed how to proceed on that, especially the desideratum (?) that the spectra we fit not violate energy conservation!
Many discussions today, of Willman I and possible evidence of mass segregation and tidal stripping; of Blanton and Roweis's new k-correct paper (draft); and of Ronin Wu's project of re-identifying the membership of Virgo from SDSS data.
Found a major bug in the WCS tweak code; fixed it, and I now have something that goes from the output of Sam's quad indexer to a passable TAN-SIP WCS header very quickly. Tested it on Sam's first 100 fields and it works on every single one (that makes it through the quad indexer with a match). There is a cryptic QA file here.
I now have functional routines that (a) find the best reverse polynomial distortion terms given forward polynomial distortion terms, and (b) invert any x,y to RA,Dec transformation by Newton's method.
Caught up with Sam (by phone now) on the astrometry.net to-do list. I looked up the IDs of the SDSS fields that failed to match in the current system, and there is nothing obviously strange about them, but Morad is going to check them.
Caught up with Beth on the status of the Willman I observations. We also discussed the coincidence
that Willman I and UMa are so close on the sky.
One of the many unfortunate things about the TAN-SIP polynomial-distorted WCS convention is that it carries around parameters for forward
and reverse
directions, even though these can be derived from one another; indeed, an iterated reverse operation can be correct to machine precision, whereas the polynomial description of the inverse of a polynomial distortion can never be correct to machine precision. But anyway, since we are going to use TAN-SIP (dammit), I wrote code today to make the reverse coefficients with as much precision as possible. It works, but it is not yet ready for prime time.
Officially sanctioned research
(see sidebar) was deprecated today as I readied my office and work for my impending relocation to NYU.
I chose examples of bad sky determination from among our pictures of RC3 galaxies (a particularly egregious example is NGC 4631) and found the SDSS runs from which they are made to create a sky testbed
.
Banged my head against the fitting of polynomial terms in the WCS tweak code. I think I am doing something wrong in the interpret step, where I convert the fitting parameters back into standard WCS parameters.
Made a plan with Blanton to solve the SDSS sky level problem; ie, the determination of the large-scale sky level to very high accuracy (possible because the survey is a drift scan) for finding low surface-brightness galaxies and making pretty pictures. Sky determination is currently the single worst problem with our pretty pictures.
I finished the linearized WCS tweak and it works and it is fast. Now I just have to upgrade idlutils to deal with SIP WCS.
I bit the bullet and figured out how to linearize and fit a reasonable set of WCS parameters, and how to reinterpret the fit output as conventional WCS parameters.
The basic idea is to treat the tangent plane choice as a separate thing from the relationship between x,y in the image and ideal u,v coordinates on the tangent plane. So you fix the tangent point (in the image and on the sky), and then fit for the relationship between x,y and u,v on that fixed tangent plane. Then you re-interpret some parts of the solution as an adjustment to the tangent point on the sky (you don't want to change the tangent point in the image because sometimes that is known a priori). Then you iterate, starting at the updated tangent point.
I started to write the code. The most difficult thing was to figure out how to interpret fit output coefficients in terms of the standard WCS parameters.
My very simple, very slow, very limited WCS tweak code based on MCMC (see two days ago) is tested and works. A code that iteratively linearizes the WCS derivatives around the iteratively updated tangent plane projection will be much faster, but takes more thought to do correctly. The MCMC code can be used, I guess, to check the output of the linearized code once I write the latter.
My friend Darcy Duke helped me find papers (on the "lost in space" problem) in the MIT Engineering Library.
I spent some time working on a new method for finding anomalous groups in the Berlind catalog—anomalous in the properties of the galaxies they contain. Quintero has been assigned the job.
I got lazy about linearizing small changes to WCS (especially since images that are near the celestial poles are nightmares), so I coded up (but have yet to test) a WCS tweak code that takes no derivatives at all—a Markov-Chain Monte Carlo code. I am sure there are a million reasons this is a dumb thing to do, but I can do it, so I did. Debugging and testing will be left for tomorrow.
By my own rules (see sidebar), I can't count what I did for most of the day—refereeing a paper for The Astrophysical Journal—as research,
but I can count giving extremely detailed comments to Quintero on his nice paper on how the distribution of galaxy properties depends on the distance the galaxies are from the centers of rich clusters. He finds some great results, but (like me), no direct evidence that there is anything exciting happening in the infall regions. I think his most interesting result (though perhaps the hardest to get across) is that he confirms, very clearly, the information theory
result from Blanton et al (2005) that morphology-like properties of galaxies are only related to environmental density through their relationships to star-formation-history-like (SFH) properties. The morphology–density relation is actually just a morphology–SFH relation painted onto a more fundamental SFH–environment relation.
I modified the method by which we extract the SDSS x,y lists we input (Sam inputs) into the astrometry engine, to make the cuts not based on specific, calibrated magnitudes, but only on brightness ranking. Sam and I decided that we would go live
on the WWW with the current pointing, rotation, and scale capability, as soon as his new students have tweaked it up and have WWW interfaces working. With only small changes (and little analysis), Sam's success rate is up to about 96 percent.
Blanton and I discussed, among many other things, the scientific output of a pilot program of prism spectroscopy as Burles et al have been contemplating. At worst, it will be better than COMBO-17, on much less telescope time. At best it will be much better, and answer qualitatively new questions about galaxy evolution.
Group meeting. I learned a lot! Spent the rest of the day on things related to moving back to NYU.
I worked on the back end of the automated astrometry system: the part that tweaks up Sam's determination of pointing, rotation, and scale to a high-quality WCS. I keep remembering (and then forgetting) that the general solution to this "tweak" step is not trivial, if you want it to be hands-off but robust to all kinds of stuff you might get.
Today I looked into the "lost in space" problem: the determination of spacecraft orientation based on TV camera images of star fields from cameras fixed on the spacecraft body. This bears a lot of resemblance to the automated astrometry problem, although all solutions to the lost in space problem assume perfect knowledge of the "plate scale" of the images, and usually require very large image fields of view. Unfortunately, I can't link to the literature on this (or read it on the internet), because it is all in proprietary engineering publications. People: please stop publishing public research results in proprietary journals! (Although I love the thought that the journal editors think a theoretical paper on spacecraft orientation from 1995 is such valuable intellectual property that they can't let anyone see it without paying $2.95! I guess I will have to go to the engineering library.)
On my week in the wilderness I started paper I in the automated astrometry series. This was inspired by Sam going "alpha" with the SDSS automatic WCS system. It seems to solve about 90 percent of SDSS fields straight out of the box (ie, give it an x,y list of stars in an SDSS image, it gives you back the pointing, rotation, and scale); it remains to be seen how we can tweak it up so it does them all.
I added a figure to the post-starburst environments paper at the insistence of Morad (someone has to keep me honest), and submitted it.
[post script added next day:] I go on vacation for a full week—with no email or phone even—see
you all again on August 15.
I got the post-starburst environments paper ready to submit! That was painful. I will submit it late tomorrow, God willing.
I got the K+A environments paper closer to being ready for submission.
I discussed with Sam possible projects for his new students working on astrometry projects, including a staged introduction of functionality, from mundane to wondrous, at astrometry.net. In particular, we discussed the value of what I would call a "footprint server" that, for any point on the sky (or every possible image footprint on the sky), returns all the known, public astronomical images whose footprints overlap that point (or image). Although some archive searches come close to this, none of them exactly perform this function, to my knowledge. (I think they fail to do this because the database tables they work off generally know about pointing but generally don't know about imager orientation or geometry.)
Sam has nearly nailed the blind astrometry for SDSS, so it is time to work on the "tweak" step that takes Sam's output and produces a precise, fine-tuned WCS for the image. I started on this today by drafting code to make an approximate WCS header from Sam's output; this and the original x,y list can be passed to the Fink-Hogg code (created as part of the photometricity monitor robot) to get a precise WCS (or report that no such precise WCS can be found).
In other news, the post-starburst environments paper reaches the end of its three-week waiting period (a period when SDSS collaborators can comment on the paper and request authorship or content changes) this week. I worked a bit on getting that ready to submit.
I worked out and wrote up a short note on the statistics of stellar quadrangles. As one might expect, their number on the sky goes as the angular size allowed for the largest star–star separation to the sixth power, and as the total density of stars on the sky to the fourth power.
I did almost no real
research today, but I did investigate some scalings of the current astrometry setup with angular scale of the indexed stellar quadrangles.
After a short vacation in Maine I came back refreshed to work on blind astrometry with Sam and Morad. Sam successfully solved the first-ever SDSS field totally blind, but it was slow because he encountered statistics of false matches that we had not expected. We re-worked the numbers for the millionth time and everything checks out okay. We discussed numerous ways to make everything faster, and we may have settled on a faster scheme, but one that ties the angular scale of the indexed stellar quadrangles to the number density (on the sky) of the magnitude-limited catalog used to construct them. We also tasked Sam and his (future) students with coding work.
I discussed with Burles target selection for the prism-spectroscopy project of Burles, Coil, and Eisenstein. Evidently he is leaning towards the COSMOS survey field, with deep, one-band HST/ACS imaging. We discussed object detection that could take place in the raw frames; ie, without stacking the images. We also discussed slit placement and observing strategy. If fringing can be represented always as additive functions in a restricted range of spatial scales on the detector, it might be possible to estimate it on every source, and then the observing strategy called nod-and-shuffle
(which restricts the slit lengths to be long and increases the noise) might not be necessary.
Quintero, Blanton, and I worked on the dependence of the luminosity distribution on clustocentric distance.
Blanton and I discussed the nascent replacement for STScI drizzle.
With the help of Karen Levay at STScI (because of previously mentioned StarView bugs), Phil and I got a list of every single Northern Galactic Cap ACS and WFPC2 exposure that was longer than 60 s of exposure time and within 3 arcmin of any SDSS LRG target (observed spectroscopically or not). This search returns 5795 public-domain exposures by 1090 LRGs; we expect about 500 of these to actually be in the images. We start the SFTP on Friday. There is lots of science to do in the HST–SDSS overlap.
Phil and I followed this with some arts and crafts: Tweaking color schemes on images of possible HST-discovered lens systems, and bug reporting and bug fixing in the ACS WCS tools.
Phil Marshall and I worked on putting a huge query into the HST Archive, and hit many snags, most of which derive from the fact that HST is not expecting people to query for 42,000 exposures in one shot. I can't imagine why not. No, I am not kidding: aren't we in the world of big data? Anyway, we failed, but the super-helpful people at STScI are on the case and I think we will be able to do a huge query by the end of the week. Interestingly, the STScIers want us to do a huge query to put their system through its paces.
Blanton and I caught up on a lot, as did Eisenstein and I (by phone); Blanton has some extremely good results on the evolution of galaxies as observed and predicted, and nice work reducing the dimensionality of galaxy spectral variations, conditioned on realistic (read: possible
) star-formation histories. Eisenstein (with Burles and Alison Coil) has a great project to take massive numbers of redshifts using a prism rather than a grism or grating.
I helped Morad finish up and post his paper on the very small-scale clustering and merger rate of LRGs to the SDSS collaboration publications archive. It can be submitted after the three-week waiting period (it saves lives, it really does). Other than that, I only did NSF-proposal-related non-research.
Morad and I helped put Sam on the brink of the first real-world test of blind astrometry on the SDSS data. Other than that I worked on a grant proposal, which does not qualify as research.
I made some ugly pictures of Willman 1, Beth's odd (and tiny) Galactic globular cluster, based on our KPNO Mosaic imaging data. I am not posting the pictures here because they aren't yet ready for prime-time. Morad helped me diagnose some problems with our pixel masking and now the pictures are re-making.
Sam and I had yet another conversation about geometry!
Sam and I discussed some of the steps involved in the "resolve" part of automated astrometry, where we have to decide among many computer-generated hypotheses about where each field is pointing and its rotation and scale. We think that a single coincidence among quads will be sufficient, but there are issues for finding even that coincidence (what's a coincidence?).
I got comments on the post-starburst environment paper from Tucker, Knaizev, Goto, Humboldt, and Morad. These comments were great; in particular they pointed out that I have completely ignored an important and relevant literature, much of it emanating from SDSS itself. I should hang my head in shame, but in fact I am happy that everyone is doing my work for me.
Preprints: You can either read them or you can write them
—Penny Sackett
talk, talk, talk!
I talked to Schlegel at length about the conceptual problems with drizzle, and reiterated my concerns that (a) it does not optimize a scalar objective function; ie, the image it produces is not the "best" in any conceivable sense; and (b) philosophically, it treats the telescope as a photon bucket, not a device that measures the intensity. This latter problem is subtle, but if you treat the detector as a flux collector, you have to worry about "flux conservation" and "gaps in the detector" etc, but when you think of it as device that measures the intensity field, you just have to determine each pixel's "beam" and then note that each measurement pixel constrains the beam-convolved intensity map. Of course, drizzle exists, and my ideas are vapor-ware, or maybe bar-ware, since they are just polemics shouted in bars.
I talked to Sam and Morad about the current status of the astrometry project, and we heard that the indexing job we started on Friday finished in an hour (which is good) and we checked some numbers relating to methodologies for determining, at test time, which index matches are correct, and which are flukes.
I discussed projects and drafts with Morad, Quintero, and Blanton.
Sam and I worked through the geometry—and symmetries and ambiguities—of the star quadrangles used for indexing in the current iteration of the blind astrometry project. We checked some code Sam wrote and started an indexing run on the USNO-B subset Morad made us.
Burles found a bug in my ACS WCS code. It was ugly. I fixed it. Then I wept tears of shame.
I discussed Lockman's polemic with Schechter (who is credited with pestering) and Blanton. I think most of Lockman's general, philosophical points are correct, but his conclusions are incorrect: The fact that you need to know the folklore to be a good astronomer is a reason not to maintain the folklore system, but to change it. Of course change requires good people and money. His argument that everyone should get cozy with the raw data is a good one, but it is on a slippery slope: He insists that you must know the raw data (I agree), but he doesn't insist that you build your own correlator. Sounds like an observer (as opposed to an experimentalist) talking.
I briefly discussed the environment- and luminosity-dependences of the merging rate with Morad, and high-density spectroscopy with Burles.
I reminded myself of the details of our future "Gunn Atlas" built from the SDSS data on low-redshift galaxies (it's a great idea), for my next attempt at NSF funding. Unfortunately, NSF funding proposals do not count as "research" for the purposes of this diary, so I am going to have to shut up about this now.
It may sound crazy, but the data are consistent with a picture in which the morphological and star-formation properties of galaxies are in fact set at a relatively early time, before infall, and the relationship with environment comes from the fact that the galaxies that find themselves inside clusters come from a special (ie, not representative) set of environments and initial conditions.
Today I worked on the problem of putting this argument, or a softened version of it, into the post-starburst environments paper.
I cranked out an unintelligible, unreferenced, but complete introduction to the post-starburst environments paper. This brings the current version up to what I call a "zeroth" draft—ie, all there but not yet in a form that is ready for the co-authors to read, let alone the public.
Sam and I spent all afternoon playing with SDSS run 745, camcol 3, field 101, looking at its astrometry, the number of USNO-B1.0 stars in the field, and the probability that the field will contain an astrometric index quadruple in our current plan for automated astrometry. It looks like our index ought to make quads with a maximum separation of between 2 and 7 arcmin to have good coverage and good precision when applied to the SDSS data. We realized that if the USNO-B1.0 and all data we ever see is of high precision, the best plan would be simply to index quads at the smallest possible angular scale; the only reason not to do this is that there is positional jitter in both the USNO catalog and any input image, which makes matching more difficult at the smallest scales. Of course at larger scales, camera distortions enter, so it still might be best always to be looking at the smallest available quads.
I wrote more text, including mentions of the fact that we can rule out tidal triggers for the (majority of the) K+A galaxies, and that we see no excess of K+As in cluster infall regions. This latter is in direct contradiction to many off-hand claims (based on data-free arguments) in the literature. I hope I can trace some of those down.
Wrote text in the post-starburst environment paper.
Helped Burles and Tongyan with color pictures of gravitational lenses.
Worked out (yes, it is obvious) that if the merger rate onto LRGs as a function of secondary magnitude is proportional to the luminosity function, then the mass-weighted merger rate will be dominated by galaxies around L-star. If it is going to diverge to the faint end, then the LRG-galaxy cross correlation will be a very strong function of the secondary galaxy magnitude, with dwarfs having much higher amplitude cross-correlation than giants. Doesn't seem likely; in fact I think we know this not to be true. The assertion in Murali et al 2002 that the mass build-up in galaxies is dominated by smooth accretion rather than merger events (is this the work upon which Romeel's assertion is based?) is not very robust, as it involves a huge extrapolation below the mass resolution of the simulation.
I gave feedback on recent progress on Quintero's clustocentric project, and Sam and Morad's work towards the first milestone on astrometry.net.
I put two more figures into the post-starburst environments paper, but unless I write some text for that paper tonight, I have failed today.
Morad has a great plan for the next project on the merger rate: determination of the mass accretion rate onto LRGs as a function of secondary mass. He will find out what galaxy masses dominate the mass accretion rate. One question: if the accretion onto galaxies is dominated by steady rather than impulsive (merger) events (as Romeel instructs us), does that mean that the mass spectrum of accretion events ought to "diverge" towards the low-mass end?
On day five of The Fabulous, a bit of heat broke out about the merger rate, with the van Dokkums and the Conselices arguing for high rates, and the Pattons and Hoggs for low. Basically, you get low rates if you assume all close pairs merge on a dynamical time (the maximal possible assumption), and you get high rates if you assume all high-asymmetry or faintly-tidal-featured galaxy had a merger in the last dynamical time. Which is more believable? Well, since the close pair rate is an absolute upper limit, I have to assume that either tidal features last for a long time, or they are raised by minor mergers. I don't think there are any other options for resolving this discrepancy.
In related news, it came up at coffee that although the galaxy-galaxy merger rate may or may not be expected to be low, the galaxy group-group merger rate has to be high, because halo mergers are frequent. Now that Berlind has a catalog, lets do it.
On day four of The Fabulous, many people showed believable galaxy luminosity functions for redshift 5, 6, 7, and maybe 8, or 10! Mannuci called redshift 4.5 "low redshift" without any hint of irony.
On day three of The Fabulous, Flores showed amazing IFU spectra that clearly show a cold, rotating disk and a hot bulge in ordinary, moderate redshift galaxies. He could potentially do real bulge-disk decomposition—ie, with no assumption about radial profiles of either component—although the experiment would certainly not be trivial.
Pettini and others discussed chemical abundances. One of Pettini's punchlines (old news, now) is that essentially all gravitationally bound systems with lots of stars have solar abundances, and systems found any other way (eg, by absorption) have abundances all over the hoo. Many were unsurprised; after all it doesn't take long to get to solar. But isn't it surprising that they never get above solar? (BTW, Pettini also showed that some metallicity calibrations are suspect, and many high metallicity results will come down.)
All this is related to the G-dwarf problem. Yesterday, Crampton showed that the evolution he sees in the metallicity-mass relation is consistent with the global star-formation rate density results. If so, then the Milky Way must be a strange outlier (because by Crampton's standards it would show a huge discrepancy between its stellar and metallicity-based star-formation history determinations).
[Disclosure: this post was posted two days late]
On day two of The Fabulous, Ando showed an absolutely incredible mean Subaru spectrum of about 10 redshift 5 (!) galaxies. It shows strong, very well-detected interstellar absorption lines, so presumably the ISM is metal-enriched. Good question: Is the enrichment one infers consistent with the hypothesis that earlier generations of stars in these objects were responsible for re-ionizing the Universe at some earlier redshift?
Romeel Davé gave a very nice overview of the status of simulations/theory in galaxy evolution, including not just the successes, but a thoughtful list of the big, generic conflicts/issues. He also emphasized the importance of accretion, noting that in the simulations, the bulk of galaxy growth is through slow, steady (not major-merger) accretion, and that such accretion is substantial. He distinguished two kinds of accretion, a "hot" mode in which accreted gas hits a shock at the virial radius and is heated to near the virial temperature as it falls in, and a "cold" mode in which no shock occurs. The latter is more common in lower-mass halos and at earlier times, and it can lead to very efficient building of stellar (or black hole) mass. Romeel suggested that the dichotomy of these modes might be related to the "bimodality" in galaxy spectral energy distributions.
In the Fabulous Destiny of Galaxies, day one, two speakers (Devriendt and Ellison) mentioned the issue that detailed chemical abundances are some kind of integral of the star-formation history (a pretty complicated one if you think about it). This brought up two ideas. The first is that the luminosity is also some kind of integral; it would be interesting to ask about what aspects of star-formation history have very different impacts on the two different integrals. There are really many different integrals, because you can look at any element, and you can measure the luminosity in any band.
Two other speakers (Yi and Fontana, I think) mentioned the fact that semi-analytic models (and, presumably, simulations) are unable to make massive galaxies as old and dead, on average, as they are observed; indeed all models produce a significant population of blue, very massive galaxies (which are never observed in the local Universe).
I gave feedback on results from Quintero, and Burles's student Tongyan. For Tongyan I made this picture:
Morad made the figure below, among others, with minor suggestions (and bitmapping services) from me, for Eisenstein's talk at the upcoming SDSS meeting.
The correlation function of the LRGs is incredibly close to a r−2 power law over four orders of magnitude in distance, eight in correlation amplitude!
I outlined the K+A environments paper.
I started to outline my talk for Marseille, but then got stuck on the question of whether or not to go postal on information and statistics. The issues are things like:
Today I worked out what hypotheses we are testing with the K+A environment research.
(1) Post-starburst galaxies lie in the same range of environments as "all" or "ordinary" galaxies.
(2) Post-starburst galaxies lie in the same range of environments as bulge-dominated or "early-type" galaxies.These hypotheses were both ruled out by Quintero et al and by Blake et al; in fact hypothesis (1) was very unlikely from the start, since no spectrally or morphologically selected subsample has the same environment distribution as "all" the galaxies ("all" doesn't really have a meaning here).
Indeed, both Quintero et al and Blake et al concluded that the mean or typical environments of K+As is the same as those of disk-dominated or "spiral" or star-forming galaxies. We already know, from work like Blanton et al, that star formation is the galaxy property most tied to environment. This leads to two new hypotheses
(3-weak) Post-starburst galaxies lie in the same range of environments as disk-dominated or "spiral" galaxies.
(3-strong) Post-starburst galaxies, at every measure of A-star excess (A/K), lie in the same range of environments as the subsample of disk-dominated or "spiral" galaxies that has the same value of A/K.
Our results confirm hypothesis (3-strong): Of all the star-formation-rate indicators, it is A-star excess that predicts the environment distribution, not H-alpha (since K+As have none by definition). This fits in with the idea that A/K evolves on timescales that are long like dynamical times.
The only statistically significant deviations we find from (3-strong) are at very small physical scales, where the dynamical time is likely shorter than the lifetime of an A star.
Worked on re-factoring the K+A environment code and wrote down everything about the K+A environment results that is unique and new. Somewhat concerned about the number of LPUs we have.
Oddly, I am responsible for photometricity monitoring on the photoop team. Today I fixed bugs in ancient 10-micron cloud-camera analysis software (in the ircam subdirectory of the photoop repository at Princeton), and started it running on all the cloud-camera data ever taken at Apache Point Observatory in preparation for the next round of SDSS data processing at Princeton.
I also wasted more time thinking about outlier rejection, and implemented a Markov Chain Monte-Carlo method. D'oh!
I worked on a new idea for totally principled outlier rejection in the determination of weighted means. The idea is to project over all possible rejection permutations and all possible rejection penalties (ie, likelihoods or probabilities that each of the input data is not rejectable). This is in preparation for image combination, but is so far off the critical path, it was a complete waste of time.
Morad and I fixed bugs in the galaxocentric distances calculation (we were being too clever) and discussed the result. It appears that tidal triggering for K+A events is almost certainly ruled out. There is one more test to perform, which is to simply look at the overdensities on a 1 Mpc scale, but I don't expect it to give different results than the nearest-neighbor distances.
I cleaned up some of my ACS thumbnail cutout code (hogg_acs_cutout in astrometry/pro/acs) and shipped it off to Lexi Moustakas for testing and experimentation.
I made thumbnail images of all the galaxies in the clusters in the field of Glennys Farrar's putative high-energy cosmic-ray cluster. To my eye, there is nothing obviously interesting or strange about the field.
I worked on the K+A environment figures and discussed them with Morad. We worked out a new figure to make, analogous to the clustocentric-distance figure, but with distances to nearby, luminous galaxies. We are pretty sure it will show an effect.
Blanton asked for error bars on yesterday's plot, so I calculated the uncertainties for binomial estimation, by taking two derivatives of the log likelihood for the "success" probability p given a sample of N trials in which there are M successes. (I used a flat prior on p.)
As expected (from the first derivative) I get that the maximum likelihood estimate of p is M/N, and (from the second derivative) I get that the (linearized) uncertainty on this is sqrt(pq/N), where q=1−p.
The clustocentric result of yesterday appears marginally significant.
I found a possible excess of post-starburst galaxies in the infall regions of large galaxy clusters. It appears significant. Code is hogg_kplusa_clustocentric in eplusa/pro and the plot (PostScript) is on today's research page.
I hacked together the simplest possible HST/ACS image combination code in IDL, building on the distortions code I wrote earlier in the week. It is not yet checked into the astrometry module. I made this image of one of Schechter's favorite lenses, 1RXS J113155.4-123155, using my new code and some public ACS data
I gave Morad comments on parts of his small-scale LRG-LRG clustering paper.
I modified Burles's ACS image processing script into a set of procedures (prefixed "hogg_acs" in astrometry/pro/hogg) that perform x,y to RA,Dec (and back) transformations for ACS "FLT" (ie, not distortion-corrected) images. This will plug into image combination code downstream, and might be a prototype implementation for future astrometry.net output.
Morad improved his K+A nearby neighbor figures. Oh wait, that only counts for Morad's research blog, not mine. D'oh! We did discuss the problem of making easy-to-read figures.
I discussed Quintero's clustocentric results and strategy with Quintero, Blanton, and Berlind on a phone call.
Morad kicked my butt on computing nearest-neighbor statistics for the K+A galaxies (and, indeed, everything in the VAGC), and he finds that there is evidence for tidal triggering of post-starburst galaxies. So the "environments of post-starburst galaxies" project has a positive result. Time to start writing! We still need to do the clustocentric distances, but the upshot of the Quintero call is that Quintero is making us the FITS file we need.
