Merger driven quenching
-
by
mlpeck
Is anyone still here?
I haven't done much directly related to this project lately, but I did notice something possibly interesting. I turned the "merging" category into a dichotomous variable: if the GZ classifiers called an object "merging", "tidal tail", "both", or "disturbed" I call it "disturbed", otherwise not. By my count there are 665/3000 (22%) disturbed in the quench sample vs. 362/3001 (12%) in the control. Now, here's a crosstab of disturbed against classification by BPT diagnostic. In the table cut & pasted below 0 is unclassified, 1=starforming, 2=transitional or composite, 3=AGN/Seyfert, 4=LINER:
> temp bpt.quench disturbed.quench 0 1 2 3 4 FALSE 629 873 448 157 228 TRUE 213 142 151 68 91 > temp[2,]/(temp[1,]+temp[2,]) 0 1 2 3 4 0.2529691 0.1399015 0.2520868 0.3022222 0.2852665So, only 14% of the starforming galaxies are "disturbed" and conversely about 30% of the AGN/LINERs are "disturbed." This suggests that mergers are driving AGN formation and quenching star formation.
For comparison here's the same table for the control group, which shows no systematic trends:
bpt.control disturbed.control 0 1 2 3 4 FALSE 1321 904 274 66 74 TRUE 167 130 42 10 13 > temp1[2,]/(temp1[1,]+temp1[2,]) 0 1 2 3 4 0.1122312 0.1257253 0.1329114 0.1315789 0.1494253I'll work on testing statistical significance later.
Posted
-
by
mlpeck
Apparently not.
A simple χ2 test is good enough for now:
> table(disturbed.quench,bpt.quench) bpt.quench disturbed.quench 0 1 2 3 4 FALSE 629 873 448 157 228 TRUE 213 142 151 68 91 > chisq.test(table(disturbed.quench,bpt.quench)) Pearson's Chi-squared test data: table(disturbed.quench, bpt.quench) X-squared = 63.2674, df = 4, p-value = 5.961e-13 > chisq.test(table(disturbed.control,bpt.control)) Pearson's Chi-squared test data: table(disturbed.control, bpt.control) X-squared = 2.4579, df = 4, p-value = 0.6522This tells us that we can reject the hypothesis that the cell frequencies are the product of the marginal frequencies in the quench sample, but not in the control.
Posted
-
by
jules
moderator
I'm still here! Just working on some more BPT diagrams using subsets of mergers - so I'll post asap. Interesting you have posted this as my plots will have something in common re showing the effect of mergers on AGN formation and quenching star formation. My "merging" category comprises galaxies classified as "merging", "tidal debris" and "both.
Posted
-
by
ChrisMolloy
in response to mlpeck's comment.
I'm still here too. Have looked at asymmetrical and merging or not signatures (including disturbed) and are about to start looking at BPT examples. Am also looking at log mass and red shift. Still working on all of this.
Posted
-
by
trouille
scientist, moderator, admin
Hello,
Yes, this is a really interesting series of posts, particularly your follow up about whether the difference between Control and Quench and between AGN and star formers is statistically significant. I'd be interested in hearing more about your reasoning to use the Chi-square test and how it relates to the following:
AGN w/ merger signatures:
Quench: 30% +/- 2%
Control: 15% +/- 3%
SF w/ merger signatures:
Quench: 14% +/- 1%
Control: 12% +/- 1%
Uncertainties shown are the 1-sigma uncertainties based on the number of sources in each subsample.
Based on this, I would think that the % and their associated uncertainties show that this is a statistically significant difference between the Control and the Quench sample.
Is a chi-square test in this context is appropriate? I'm not sure and definitely want to hear more. In the past I've used KW-test results to say whether a distribution of results is from the same 'parent population' as another distribution -- for example, student test scores from one semester to another. And this tells me whether I can aggregate the results or not. If you could explain more your reasoning and how it works and how it relates to the % uncertainties I've posted here, that would be extremely useful.
Additional notes:
My merging category is everything that's not 'Neither'. I.e., << mergSigs = where(category ne 'Neither') >>. Please let me know if you've found that by doing it this way, rather than Jules' <<merging + tidal debris + both>>, I might be bringing in non-merging sources. I haven't double checked this.
Also, I only include sources whose emission line flux is greater than 3 times the error on that emission line's flux (i.e., a signal-to-noise greater than 3). And my AGN category are sources to the right of the Kauffman03 line and my SF category are sources to the left of the Kewley01 line.
Here is the link to my code. It's IDL code, which unfortunately is not a free program. I am in the process of learning Python, with the hopes of being able to only code in Python soon enough, but I'm not quite there yet. I annotated the code to help it be more clear, but I'd happily explain any parts that are unclear.
Posted
-
by
trouille
scientist, moderator, admin
As you mention in your first post, there are a lot of Quench sample sources (and control) that can't be put on the BPT because not all 4 emission lines are present or not all 4 emission lines have enough signal-to-noise (I use a cut at S/N > 3). When I check the Quench sample, I get 2209 sources with S/N>3 for all 4 lines. This leaves 793 sources off the BPT.
I was hopeful that we could use the MEx diagnostic (Juneau et al. 2011, https://sites.google.com/site/agndiagnostics/), which uses Mass versus [OIII]/Hb to separate AGN from star formers and be able to say something about these 793 sources. But when you take away the restriction to have strong NII and Halpha, you only add 7 more sources for which you can say something (i.e., there are 2216 sources with S/N>3 for NII and Halpha, regardless of their OIII and Hb). So it's not a fruitful route to follow.
Has someone tried doing this analysis but taking away any S/N restriction and using upper limits for [OII]/Hb and [NII/Ha. In this scenario, all we need is for [OIII] > 0 and [NII] > 0. In this case you can work with 2983 of the sources (i.e., you only miss out on 19 sources). Would someone like to give that a try and see what they get?
Posted
-
by
trouille
scientist, moderator, admin
By S/N > 3, I mean that I put in statements like the following into my code:
SN = 3.
onBPT = where(O3 ge SNO3err and Hb ge SNHberr and N2 ge SNN2err and Ha ge SNHaerr)
Note: * is the symbol for 'times' in IDL.
Posted
-
by
mlpeck
in response to trouille's comment.
Is a chi-square test in this context is appropriate? I'm not sure and definitely want to hear more. In the past I've used KW-test results to say whether a distribution of results is from the same 'parent population' as another distribution
A χ2 test is a fairly standard (if flawed) test of independence in a two-way contingency table. What it's testing basically is whether the proportions in the table in the first post equal the product of the marginal distributions. In this case there was no comparison of the quench sample with the control sample. What caught my eye was that while 22% of the quench sample was classified as disturbed or merging only 14% of the SF subset was disturbed, and at the same time the AGN fraction went up. The χ2 test says that some proportions changed (or at least it rejects the hypothesis that the cell counts came from the marginal distributions multiplied by the total sample size), but it doesn't quite answer the question I'd really like to ask.
I was looking at Bayesian approaches to contingency table analysis, and if I find the time I'll try something Bayesian.
Posted
-
by
mlpeck
in response to trouille's comment.
But when you take away the restriction to have strong NII and Halpha
The majority of the quench sample unclassifieds did have detectable emission in both Hα and NII. Based on the ratio of those two I convinced myself that most of them are weak LINERs, but I didn't look hard for a literature reference to justify that conclusion.
Posted
-
by
trouille
scientist, moderator, admin
And what do you think of the result (in terms of the statistical significance of the result) when viewed in terms of?:
AGN w/ merger signatures:
Quench: 30% +/- 2%
Control: 15% +/- 3%
SF w/ merger signatures:
Quench: 14% +/- 1%
Control: 12% +/- 1%
Posted
-
by
mlpeck
And what do you think of the result (in terms of the statistical
significance of the result) when viewed in terms of?:AGN w/ merger signatures:
Quench: 30% +/- 2%
Control: 15% +/- 3%
Is there a meaningful comparison to be made here? I was looking separately at the quench and control samples: in the former having a merger signature had a statistically significant effect on the proportions in the BPT diagram. In the latter, not. The pattern of changes in the quench sample seems to suggest a narrative. I'm not sure the null result for the control sample really needs explaining.
Anyway, there were many more objects with AGN signatures in the quench sample than the control and many more with merger signatures and AGN activity: 159 ± 13 (1σ) in QS vs. 23 ± 5 in QC by my count and assuming Poisson statistics.
Posted
-
by
mlpeck
in response to trouille's comment.
I was hopeful that we could use the MEx diagnostic (Juneau et al. 2011, https://sites.google.com/site/agndiagnostics/), which uses Mass versus [OIII]/Hb to separate AGN from star formers and be able to say something about these 793 sources.
Following the link to that page and going to the "WHAN" page suggests more promising diagnostics based on [NII]/Hα ratios and equivalent widths based on work by Cid Fernandes et al. (2011). Most of the quench sample unclassified objects have detections in both Hα and NII, and equivalent widths are tabulated in the MPA pipeline.
Posted