Star formation rates and specific star formation rates -- comparing the quench and control samples

by JeanTate

This is very impressive! 😃

To my eye, the major difference¹ between the QS objects and the controls is in the last plot: QS LINER, unclass (and AGN?) objects are to be found in a region that is completely empty in the QC plot!

Dumb question, if I may: if the controls (objects in the QC catalog) had been selected to carefully exclude 'red sequence' galaxies, do you think the results would be dramatically different? Expressed in another way: how do the objects in the QS catalog differ from matched (stellar mass, redshift) 'blue cloud + green valley' galaxies?

¹ Other than that which very likely is due to the almost complete absence of red sequence QS objects

Posted September 23, 2013 9:20 PM
by JeanTate

Suppose we did a color and magnitude cut to separate red sequence galaxies from the rest. Would a redefined control sample still look different from the quench sample? Is that what you're asking?

~half of it, yes. One great thing about the QS/QC is that they are almost exactly matched; if you remove the red sequence objects from QC, you'll have ~half the number of QS ones. Comparisons will still be possible, of course, but more work will be needed to adjust for fact that one group has only ~50% of the other (by number). On top of that, it's possible the red sequence QC objects have a different (stellar mass, redshift) distribution than the QS ones, making comparisons even more challenging.

Posted September 24, 2013 12:34 AM
by JeanTate

The last two posts suggest that it'd likely be a major undertaking to come up with a coherent story ... and that may not be possible at all (different methodologies, different selection effects, different 'known problems/limits', ...). Especially as Quench galaxies are, by their very selection, unusual. Perhaps that's the take-away, that given the present-day all-over-the-map status, no real conclusions can be drawn?

Posted January 20, 2014 3:46 PM
by mlpeck

I'm not quite so pessimistic. I see a narrative starting to cohere, especially if the DR8+ star formation rate estimates from the MPA group are more believable than the alternatives.

It might help to inquire how and why the SFR estimates changed between DR7 and DR8, either through a literature search or just asking someone at MPA (presumably Jarle Brinchmann).

Posted January 21, 2014 6:04 PM
by mlpeck

Kauffmann 2014: "Quantitative constraints on starburst cycles in galaxies with stellar masses in the range 108-1010 Msol". Note figure 1, which shows model locations in the SSFR-D4000, SSFR-Hδ, and Hδ-D4000 planes.

Posted February 5, 2014 2:39 AM
by trouille scientist, moderator, admin

Note that there's a core group consisting mostly of starforming galaxies (per BPT diagnostic) that remained essentially unchanged between DR7 and DR9. However the MPA people evidently did a major recalibration of SFR estimates for everything else.

Well I looked at the SFR and mass estimates from the other two groups that provide them in DR10 and they just further muddied the waters. Stellar mass estimates show reasonably tight correlations, but star formation rate estimates are all over the place.

Good work in noticing these differences. Your posts and findings prompted me to search through the SDSS databases for DR7 through DR10 for possible explanations.

1st: a presentation of the three stellar mass estimate approaches is provided here: https://www.sdss3.org/dr10/spectro/galaxy.php.

A comparison between the Portsmouth SED-fit and the Wisconsin spectral PCA stellar masses is discussed in Appendix A of Maraston et al. (2012).

The Wisconsin group includes Christy Tremonti and Yanmei Chen who are members of the Quench science team and led the post-quenched galaxy sample selection effort, which made possible our GZ Quench project.

Question: Am I missing another Star Formation Rate estimate, other than from MPA-JHU (https://www.sdss3.org/dr10/spectro/galaxy_mpajhu.php) and Portsmouth (https://www.sdss3.org/dr10/spectro/galaxy_portsmouth.php #kinematics)?

I believe the Wisconsin group only provides stellar mass estimates; no star formation rate estimates. The same appears to be true for the Granada group.

I've emailed colleagues to find out what they'd recommend.

Posted March 27, 2014 6:44 PM
by trouille scientist, moderator, admin in response to mlpeck's comment.

Well I looked at the SFR and mass estimates from the other two groups that provide them in DR10 and they just further muddied the waters. Stellar mass estimates show reasonably tight correlations, but star formation rate estimates are all over the place. I may post more elsewhere some other time.

I emailed colleagues and Ramin Skibba had a quick response:

I think either the MPA-JHU or the Portsmouth-Maraston SFRs should be fine. Are you finding major differences in the SFR distributions? I don't know of any reason not to use the Portsmouth-Maraston ones.

Note: I'd use the Kroupa IMF based Portsmouth results, rather than the Salpeter IMF based results (if you're given a choice in the SQL query you might do).

For completeness -- could you take a moment and post your plots showing the difference you found between the MPA-JHU and Portsmouth-Maraston SFRs?

Thanks!
Laura

Posted March 27, 2014 8:27 PM
by JeanTate in response to trouille's comment.

The Wisconsin group includes Christy Tremonti

The same person who is the lead author on the key Tremonti+ 2004 MPA-JHU paper? Wow! 😮 😄

Perhaps we can ask her directly about the questions I posted, in the How to decide which objects' MPA-JHU derived parameters are unreliable because ... thread?

Posted March 27, 2014 10:34 PM
by mlpeck in response to trouille's comment.

Note: I'd use the Kroupa IMF based Portsmouth results, rather than the
Salpeter IMF based results (if you're given a choice in the SQL query
you might do).

For completeness -- could you take a moment and post your plots
showing the difference you found between the MPA-JHU and
Portsmouth-Maraston SFRs?

I'll try to post some graphs later. The "Granada group" also calculates specific star formation rates, which combined with their stellar mass estimates allows calculating total SFRs. They have 2 different assumptions about star formation histories, models with and without dust, and both Kroupa and Salpeter IMFs although it appears only one of those is in the DR10 database.

Most of the Portsmouth groups SFR estimates are identically zero -- there's a rough correlation with the MPA estimates for non-zero estimated SFRs but with an apparent systematic offset.

Posted March 28, 2014 2:16 PM
by JeanTate in response to trouille's comment.

Note: I'd use the Kroupa IMF based Portsmouth results, rather than the Salpeter IMF based results

Why?

Because, for whatever reason, the Portsmouth results are more robust?

Or because there are good reasons to prefer a Kroupa IMF (or, more accurately, the Kroupa IMF which Portsmouth uses) over a (the) Salpeter IMF?

Posted March 28, 2014 3:32 PM
by KWillett scientist

I'd agree with the recommendation to use the Portsmouth SFR results; those are "officially" endorsed by the SDSS as of Phase III of the project. For context, the Salpeter IMF is more than 50 years old and as a consequence was based on worse data. In particular, since surveys weren't as sensitive back then, the Salpeter IMF isn't sensitive to low-mass stars (less than 0.1 solar masses). Updated IMFs in the last 20 years, including the Kroupa model, have revised this. So I think that's one reason to use the Portsmouth model.

Posted March 28, 2014 4:27 PM
by trouille scientist, moderator, admin

Finally, just as a reminder of what concerned me in the first place,
here is the difference between the DR8+ version of the MPA SFR
estimates and the DR7 version (I think this is also the full sample of
3000). To my knowledge the changes from DR7 to DR8 are undocumented.

Again, great that you noticed this and brought it up. The following is a summary of what I learned from http://www.mpa-garching.mpg.de/SDSS/DR7/sfrs.html and https://www.sdss3.org/dr10/spectro/galaxy_mpajhu.php and Brinchmann et al. (2004) (B04).

For star-forming galaxies: MPA-DR8+ computes star formation rates (SFRs) using the nebular emission lines as described in B04.

This is the same procedure/approach as was used for creating the MPA-DR7 results for the star forming galaxies.

This use of the same procedure/approach is likely why you get similar values from both DR8+ and DR7 in your plot for the star forming galaxies.

For non-star-forming galaxies: MPA-DR8+ computes SFRs using galaxy photometry (SED fitting) following Salim et al. (2007).
MPA-DR7 computes SFRs for these galaxies using only D4000 (as done in B04) plus a better dust attenuation model.

This use of a different procedure/approach is likely why you get quite different values between DR8+ and DR7 for many non-star-forming galaxies in your plot.

The galaxy stellar mass values remained the same from DR7 to DR8+, as did the redshifts (for our subsample of 1149 QUench sources with z=0.02-0.1 and absZ LT -20).

I find this to be reassuring (i.e., understanding why there's the difference between the DR7 and DR8 SFRs for non-star-forming galaxies).

As you noticed, in https://www.sdss3.org/dr10/spectro/galaxy_mpajhu.php they write that the MPA-DR8+ results are deprecated in favor of the other available model/approaches (i.e., Wisconsin, Portsmouth, & Granada).

Posted April 2, 2014 1:41 AM
by trouille scientist, moderator, admin in response to mlpeck's comment.

Here's my version of mlpeck's plot, just showing the results for our subsample 2 (1149 quench sources with z=0.02-0.1 and absZ LT -20). I see the same trends as his (i.e., SF galaxies showing similar SFRs between MPA-DR7 and MPA-DR8+ while non-star-forming galaxies are significantly different).

Comparison between MPA-DR7 and MPA-DR8+ star formation rate estimates

Posted April 2, 2014 2:08 AM
by trouille scientist, moderator, admin in response to KWillett's comment.

I'd agree with the recommendation to use the Portsmouth SFR results; those are "officially" endorsed by the SDSS as of Phase III of the project. For context, the Salpeter IMF is more than 50 years old and as a consequence was based on worse data. In particular, since surveys weren't as sensitive back then, the Salpeter IMF isn't sensitive to low-mass stars (less than 0.1 solar masses). Updated IMFs in the last 20 years, including the Kroupa model, have revised this. So I think that's one reason to use the Portsmouth model.

Looking more into the SDSS documentation (including https://www.sdss3.org/dr10/spectro/galaxy_portsmouth.php #fitting) and speaking with colleagues here at Northwestern University who work with SDSS data as well, the recommendation continues to be the Portsmouth model with the Kroupa IMF because of how they do their modeling.

https://www.sdss3.org/dr10/spectro/galaxy_portsmouth.php describes the models used as well as the fact that the Portsmouth group has released results for both DR8 and DR10. The DR10 results are focused on the BOSS survey, which has less sky coverage to date. This sparser sky coverage may be less of an issue when you retrieve data through CAS JOBS (I haven't checked yet if CAS Jobs compiles both DR8 and DR10 results from the Portsmouth group).

If, like me, you're using the files provided on the link above, you need to stick with the DR8 files, i.e.,:
portsmouth_stellarmass_starforming_krou-26.fits.gz [SDSS DR8].

The sky coverage for the data contained in this file means that you get information for all our Quench sources.

If instead you use the file portsmouth_stellarmass_starforming_krou-v5_5_12.fits.gz [BOSS DR10], your sky coverage is as shown in the image below, and you recover very few of our Quench sources in the data set (it's not as clear in the image, but the spatial coverage within the darkened area is also sparser than for the DR8 data).

If you're using CAS Jobs, please do make sure you are recovering Quench sources.

Posted April 2, 2014 2:53 AM
by trouille scientist, moderator, admin in response to KWillett's comment.

Also, like mlpeck, I also find significant difference between the Portsmouth DR8 and MPA-DR8 SFR results. Again, restricting to subsample 2 (1149 quench sources with z=0.02-0.1 and absZ LT -20).

I've restricted the SFR range to LT 10 Msun/year, just as a way to zoom in and show how the Portsmouth DR8 SFRs are mostly LT 2 Msun/year. The MPA-DR8 SFRs are above 30Msun/year for some of our Quench sources.

Given our selection of sources that have had their star formation recently quenched (according to Yanmei's PCA analysis), it makes more sense for the SFR to be relatively low or not significant. For example, the Portsmouth-DR8 SFRs are LE 0 for 768 of our Quench subsample 2 sources. This is in contrast to only 13 sources with SFR LE 0 when using the MPA-DR8 results.

Comparing Star Formation Rates for MPA-DR8 and Portsmouth-DR8 for Quench Subsample 2

Posted April 2, 2014 3:03 AM
by JeanTate

This seems to be as good a place to ask this as any ...

The most recent - and hopefully the penultimate, all-but-definitive - version of the Quench project catalogs (in this post, by Kyle) contains 'the original' SFR values. Per an earlier post in this thread, mlpeck independently verified that these are the same as "the DR7 MPA pipeline estimates".

The post I wrote which contains links to files with the clean version of the "021020"* IDs and key data is the OP of the Clean "021020" galaxies: 11 April catalogs, comparisons, and discussion thread:

There are 1084 QS "clean 021020" galaxies, and 1131 QC ones. I have uploaded two CSV files with OBJID, uid, redshift, and Z_ABSMAG values for these, from the 11 April catalogs, to Google spreadsheets: 0210QS_excl_ppo, and 0210QC_excl_ppo (please let me know if you are unable to access/download them, or if you have any problems opening them).

For whatever analyses we plan to do, on 'the 021020', will we be replacing the DR7 MPA SFR estimates with the DR8 ones? What about the 27 QS objects, and one QC object, which have SFR values of -99?

*redshift between 0.02 and 0.10 AND z-band absolute magnitude brighter than -20.0 AND excluding ppos; the one QS ppo I found, and wrote about in that thread, has yet to be removed

Posted April 20, 2014 2:05 PM