Galaxy Zoo Starburst Talk

Quench sample environment

  • mlpeck by mlpeck

    I'm going to continue the analysis I started in the topic "Can we say anything about the quench sample environments?", but using the galaxy and group catalog recently published by Tempel et al. (2014). Having definite redshift and magnitude cuts will make this easier: I just applied the same cuts to the Tempel data set to get a subset of 252,480 galaxies in the same redshift range and magnitude upper limit as subset 2 of the quench sample1.

    The graph below doesn't have much to do with the analysis that will follow: it just shows a color-magnitude diagram (specifically g-r vs. Mr)for the Tempel data with the quench sample overlaid as color points coded by position in the BPT diagram. The graph is similar to one that was used in Tojeiro et al. (2013) and earlier in Masters et al. (2010).

    The red line was the one used by Masters+ and Tojeiro+ to separate red from blue galaxies. This seems a bit too blue to me: I'm going to use the blue line as my divider between red sequence and blue cloud galaxies -- this is shifted redward by 0.05 magnitude from the Masters+ relation. The blue line falls right in the green valley region of this color-magnitude plot.

    enter image description here

    1A couple details about the Tempel catalog: They adopted a Hubble constant H0 = 100 km/sec, so 5 log10(0.7) must be added to their tabulated absolute magnitudes. Also, they used K+E corrected Petrosian magnitudes and I think we have model mags or perhaps Cmodel mags. This shouldn't really matter because all I'm using this for is to make a rough cut of the full sample into red and blue sequence galaxies.

    Posted

  • JeanTate by JeanTate in response to mlpeck's comment.

    and I think we have model mags or perhaps Cmodel mags.

    We use modelmags. However, the details of how Laura (?) did the k-corrections and produced estimates of absolute mags was not posted (that I can remember; a quick search here in Talk turns up nothing, but that also means very little).

    Posted

  • mlpeck by mlpeck

    I don't think details of the k-corrections were given. I thought it was KWillett who did the calculations, but my memory may be faulty.

    I compared our K-corrected magnitudes to the ones in the Tempel catalog and there was no systematic offset after correcting for the assumed Hubble constant. There were scattered outliers as large as ± 1-2 magnitudes. Typical photometric precision.

    Posted

  • mlpeck by mlpeck

    I'm going to post some results, but first I'll get a couple caveats out of the way.

    • Some quench sample objects are absent from the Tempel data set. A join on DR8+ specObjID's produced 931 (81%) matches. A position based match added another 42, for just under 85% completeness. I don't know why some objects are missing, but I checked some attributes of the missing ones and didn't see any obvious differences from the subset that's in the Tempel database.

    • I had noticed some systematics in the DR8 based Tempel density estimates that I didn't really understand (well to be honest I don't really understand what they're estimating and certainly don't understand their methodology). The systematics haven't gone away as shown below: for z >≈ 0.05 the minimum density rises systematically even though they made a correction for Malmquist bias. However by making the same redshift and magnitude cuts for the Tempel sample as for the Quench sample any bias this produces should apply equally to both sets of data, and statistical comparisons of density distributions should still be valid. If this proves to be a concern we can use a density estimate for a larger characteristic distance, at the cost of lower contrast between blue and red galaxies

    • As I noted in the first post they tabulated Petrosian magnitudes while we're using model mags. K corrections may have been implemented differently, but I saw no systematic offset in magnitudes. I'm just using the photometry to make a rough separation into blue and red sequence galaxies, so none of these differences should matter.

    enter image description here

    With that out of the way, the blue line in the CMD of the first post has equation (g-r) = 0.68 -0.02 (Mr + 20). That's what I used as the dividing line between blue and red galaxies. That line is offset by 0.05 magnitudes to the red of the one used by Masters+ to select red spirals. This does make a quantitative difference in results. Blue cloud galaxies live in lower density environments using the Masters+ cut than they do in the one I'm using.

    In the remainder of these posts I'm going to be comparing cumulative distribution functions of Tempel's density estimates. I'm bootstrapping quantiles to give an idea of the uncertainty of the CDFs of the quench sample. Those will be the cloud of gray lines surrounding the CDFs. Blue and red sequence galaxies from the Tempel+ data set are in blue and red, respectively. Here's the result for the full quench sample (at least the 973 survivors from subset 2):

    enter image description here

    This is completely consistent with the conventional wisdom: post starburst galaxies in the local universe are thought to favor the same environments as blue galaxies, and our quench sample density distribution is indistinguishable from the blue cloud.

    Later I'll break down the full sample into various subsets.

    Posted

  • mlpeck by mlpeck

    One more for the full sample. This compares the density functions of log(group size). Color coding is the same as in the previous post.

    enter image description here

    I kind of like using bootstraps of quantiles as a visual alternative to Kolmogorov-Smirnov tests. A KS test says the blue sequence and quench sample distributions are not significantly different (p=0.72), while the red sequence and quench emphatically are significantly different (p ≈ 0).

    Posted

  • mlpeck by mlpeck

    Today I'm going to examine the density distribution for various cuts on BPT diagnostic classification. Sample sizes for each cut are in parentheses.

    First, starforming only (N=541)

    enter image description here

    Everything else (N=608)

    enter image description here

    Starforming + Brinchmann's "Low S/N starforming" (N=688)

    enter image description here

    AGN + LINERs only (N=204)

    enter image description here

    Composite or transitional (N=247)

    enter image description here

    Brinchmann's 'Low S/N SF' + unclassifiable (N=157, of which 147 are Low S/N SF)

    enter image description here

    Conclusions: Starforming galaxies in the quench sample live in the same environments as blue cloud galaxies in the general population. Everything else favors intermediate densities, or at least disfavors the lowest density environments. This is most apparent for the AGN/LINER group and 'Low S/N starformers', although small numbers make statistical comparisons less than compelling.

    A KS test comparing the blue cloud density distribution to that of the AGN/LINERs says they are significantly different at the p=0.001 level. The blue cloud density distribution is significantly different from the 'Low S/N SF' + unclassifiable group at the p=0.002 level, while the red sequence distribution is significantly different only at p=0.03.

    Composite or transitional region galaxies favor the same environments as starforming galaxies.

    Posted

  • mlpeck by mlpeck

    Today's post: Quench sample 'blue cloud' and 'red sequence' galaxies, defined by the same cuts as the Tempel sample in the first post.

    Blue cloud (N=938)

    enter image description here

    Red sequence (N=211)

    enter image description here

    Quench blue cloud galaxies actually favor slightly denser environments than blue cloud galaxies in the general population (KS test is significant p=0.02). Quench red galaxies are indistinguishable in their environments from blue cloud galaxies.

    This result is perhaps not too surprising since photometric properties played no role at all in quench sample selection. It turns out that broad band colors are fairly strongly correlated with internal extinction as measured by the Balmer decrement, so much of the variation in color within the quench sample is due to variations in internal reddening.

    Posted

  • jtmendel by jtmendel scientist, moderator

    Thanks mlpeck, this is a really nice comparison! I wonder if it's possible to add an additional constraint to the comparison of cumulative density, which is to require that the tracer populations have the same mass and redshift distribution as the quench sample (unless I've missed it and you've done that already!).

    My thinking is that we know massive galaxies are preferentially found in high-density environments, and so we'd want to control for this is a possible systematic effect in the comparison of densities; if you like, it's changing the question slightly to "What are the environments of quenched galaxies relative (star-forming or passive) galaxies of the same stellar mass?"

    Posted

  • mlpeck by mlpeck in response to jtmendel's comment.

    I wonder if it's possible to add an additional constraint to the
    comparison of cumulative density, which is to require that the tracer
    populations have the same mass and redshift distribution as the quench
    sample (unless I've missed it and you've done that already!).

    Hi jtmendel, thanks for joining us.

    We made cuts on redshift and magnitude to get from the original Quench sample of 3002 to the subset that the remaining "citizen scientists" are working with now. Specifically we chose objects with 0.02 < z < 0.10 and Mz < -20, which left us with a sample of 1149.

    I made the same cuts to the Tempel dataset using their tabulated redshifts and absolute magnitudes. The comparison sample has 252480 galaxies out of the full dataset of 588193, of which 99120 (39%) are in the "blue cloud" as I defined it.

    Is this what you had in mind more or less, only substituting luminosity for mass? Stellar masses aren't tabulated in the Tempel database, but they can certainly be retrieved easily enough from SDSS. Just from looking at the CMD in the first post I would guess that we have some giant ellipticals in the Tempel sample that are more massive than anything in the quench sample.

    We also have a control sample that is matched one for one in redshift and stellar mass to the quench sample. I plan to do some comparisons later.

    Posted

  • mlpeck by mlpeck

    I'm going to look a little bit at some of the classifications: so far I've just taken a brief look at smooth vs. features or disk and the merging question. Merging seems more interesting, so here goes. I defined a "merger signature" vote fraction as the sum of the vote fractions for "Merging", "Tidal debris", or "Both". I tried different thresholds for the vote fraction, but for now I'm just going to show the results for "merger signature" vote ≥ 0.5 vs. "merger signature" vote < 0.5. Any threshold above ≈ 0.4 produces similar results.

    enter image description here

    enter image description here

    So, the left tail of the "shows signs of a merger" distribution lies at higher densities than even red sequence galaxies in the general distribution. This makes sense I suppose since a merging galaxy needs something to merge with and therefore can't be completely isolated unless it has swallowed up everything nearby.

    The median "shows signs of a merger" galaxy lives in intermediate density environments while the right tail follows the blue cloud distribution.

    Posted

  • mlpeck by mlpeck

    For now I'm going to look at just one more classification category: "Smooth" vs. "Features or disk". The plots are for a vote threshold ≥ 0.5 for the feature. As with the merger category these look similar at least over the range 0.4 ≤ vote fraction ≤ 0.6.

    Smooth (N=807)

    enter image description here

    Features or disk (N=332)

    enter image description here

    So, quench galaxies that the GZ classifiers saw as smooth live in the same distribution of environments as blue cloud galaxies, while those we saw had features favor intermediate densities.

    My recollection is that seeing apparently disturbed morphology was one of the features that would lead me to click "Features or disk." I wonder if others followed the same protocol. If so this category could contain mergers as well as disk galaxies.

    At this point it would be useful to do some comparisons with the control sample, but nothing can be done with the control classifications since our data remain incomplete. I may look at a few things that don't depend on them.

    Posted

  • trouille by trouille scientist, moderator, admin

    Ramin Skibba is a Galaxy Zoo astronomer (http://cass.ucsd.edu/~rskibba/) who has worked on galaxy environment studies. He wrote the following in a recent email to me about advice on measuring environment around our Quench galaxies:

    I work on environments and large-scale structure, so if you want to include some things about that, I could give some input. The Baldry density (using 4th & 5th nearest neighbors) is a popular and pretty good one. If you want to see comparisons between environment measures, you might want to look at this paper I was involved in: http://arxiv.org/abs/1109.6328

    The Baldry densities are correlated with halo mass at M GT 1e13 Msun, but there is a lot of scatter below that. In our other "Measures of Galaxy Environment" papers, we showed that the Baldry densities are primarily sensitive to small-scale environments (a few hundred kpc/h scales) and that they can be affected by projection effects and redshift errors, but those effects are probably small (unless the redshift errors are large).

    The Tempel et al. group catalog seems like a robust one. If you're interested, the Yang et al. and Tinker et al. group catalogs seem to be a little more popular, and more people will be familiar with them. But I think all three should yield approximately similar results.

    Posted

  • mlpeck by mlpeck in response to trouille's comment.

    I've taken a brief second look at the Baldry dataset. They imposed redshift limits of 0.01 ≤ z ≤ 0.085, more restrictive on the high end than our subset 2. I found 787 matches (on position) from the quench sample in Baldry. The density distribution of the matched quench sample galaxies is the same as "blue cloud" galaxies in the full dataset, so that agrees with Tempel.

    The Yang+ catalog is available "on request." I haven't investigated Tinker et al.

    Thanks for the link to the paper. I'd be interested in any comments Skibba has on Tempel et al.'s methodology.

    In the meantime I will look at some properties of the control sample.

    Posted

  • mlpeck by mlpeck

    Here's a quick look at some properties of the control sample environment. This is version 2 of subset 2 of the control sample. First a CMD similar to the one that led off this topic:

    enter image description here

    I did the same blue cloud/red sequence cut on the control sample as the others. It turns out there are 679 blue cloud and 470 red sequence galaxies in the sample, which is quite different from the quench sample (938 and 211, respectively). This is also rather different from the Tempel data set, which has about 39% in the blue cloud. Note also in the CMD above the control red sequence galaxies appear to be rather bluer than the densest part of the red sequence in the general population.

    Does this have any significance? I have no idea.

    Here are comparisons of density distributions, first for the blue cloud controls, next red sequence.

    enter image description here

    enter image description here

    So blue cloud controls have the same density distribution as blue cloud galaxies in the general population. The red sequence density distribution doesn't quite look like the general population, but it's closer to the red sequence than blue.

    Posted

  • mlpeck by mlpeck

    Today I'll present density distributions for the controls broken down by BPT class in various ways. By the way my sample sizes in the previous posts were wrong since not all galaxies in either the quench or control samples are found in the Tempel data set. I'll put the actual sample plotted sample sizes in brackets in this post.

    Control starforming (N=484 [410]):

    enter image description here

    Everything else (N=665 [571])

    enter image description here

    Composite or transitional only (N=133 [118])

    enter image description here

    AGN+LINER+"Low S/N SF" (N=462 [397])

    enter image description here

    No detectable emission (N=100 [76])1

    enter image description here

    The control sample galaxies with no emission are truly "red and dead" passively evolving systems, and they favor dense environments.

    Overall we see that the red sequence/blue cloud division is meaningful for the control sample, and it divides fairly cleanly starforming galaxies from the rest. Galaxies with composite or transitional emission line ratios seem to favor intermediate density environments.

    1 Note added 5 March: Brinchmann's "Low S/N starforming/AGN" classes don't really seem to add much information. For one thing there's only one Low S/N AGN in the entire quench and control samples, and for another it's not clear that some emission in NII or Hα indicates star formation.

    I've taken the liberty of modifying the Brinchmann "Low S/N" category as follows: if one or more of the lines that go into the BPT diagram is undetected or missing but at least one of [O II] 3727, [O III] 5007, Hα, [N II] 6584 is detected in emission at the 3σ level I'm calling it "Weak emission". This splits the spectra that are unclassifiable by the BPT diagnostic into some emission detected and truly passive.

    Posted

  • mlpeck by mlpeck

    In their galaxy zoo red spiral paper Masters+ looked at the effect of environment on some proxies for star formation, using density measures from Baldry et al. I'm not quite sure why they used proxies when star formation rate estimates are available from the same source as their proxy data (i.e. MPA-JHU), and I didn't really understand why they made the plots they did. Today I'm going to look at star formation rates and specific star formation rates as a function of Tempel's density measure den1. I may get to proxies later.

    The plotted points in these graphs are for the quench sample color coded by position in the BPT diagram, overlaid on contours representing the density of points in the control sample. First star formation rates:

    enter image description here

    The line here is a weighted1 least squares fit of star formation rate on density for spectroscopically starforming galaxies only2. The slope of the line is 0.28 ± 0.05, so it's significantly different from 0 at nearly the 6σ level. Why is the line offset so much from the center of the cloud of starformers? Basically it's because the estimated uncertainties are strongly correlated with the estimated star formation rates, with the highest SFR's getting the highest weights.

    There is no significant difference between quench and control sample starformers in this relationship.

    Here are specific star formation rates - log(SSFR) = log(SFR) - log(M*):

    enter image description here

    Specific star formation rates for starformers show no significant dependence on environment. The only apparent difference between the quench and control samples is that the control distribution is bimodal with a second mode at low SSFR and high density. This is where the "red and dead" presumably mostly elliptical galaxies live, a population that is completely absent from the quench sample.

    If star formation rates increase with density while specific star formation rates do not, that must mean that galaxy masses increase with density. Yup (line is weighted least squares fit for the entire quench sample):

    enter image description here

    1 Weighted least squares? The MPA pipeline provides uncertainty estimates for star formation rates and stellar masses in the form of several quantiles of the marginal posterior distributions of the estimates. If the distributions were gaussian (probably not, but no matter) the difference between the 84th and 16th percentile would be approximately 2σ. I've retrieved those quantiles from the SDSS database (DR10 version) and used them to estimate uncertainties in SFR, M*, and SSFR. The uncertainty estimates are highly variable especially for SFR, so weighting the sum of squares by the inverse variance should give a better estimate of the slope than ordinary least squares.

    Significance levels should not be taken too seriously. For one thing we have no uncertainty estimates for the densities and if those are significant other methods should be used.

    2 Why did I fit to starformers only? As I mentioned late in this topic on star formation rates there is a very significant difference in SFR estimates between DR7 and DR8+ for objects that occupy regions of the BPT diagram other than starforming.

    Until some credentialed scientist weighs in on why the difference is there and which version should be used I don't think anything can be said with any confidence about star formation in galaxies that don't exhibit starforming spectra.

    Posted

  • mlpeck by mlpeck

    I'll get the two most interesting proxies that Masters+ looked at out of the way, Lick HδA and D4000n.

    enter image description here

    There's no significant dependence on environment in Hδ strength for the quench sample. Controls have starforming galaxies with relatively strong Hδ at low densities and passive galaxies with low Hδ at high densities, so there is a correlation although with huge scatter.

    D4000n:

    enter image description here

    The line is a weighted least squares fit for the full quench sample. The slope is significantly different from 0 at a little under 4σ: 0.019 ± 0.005 (1σ). To the extent that D4000 is an age proxy this would indicate a slight dependence of starburst age on environment.

    Posted