PDA

View Full Version : Galaxies Made Simple, or at Least Less Complicated



schlaugh
2008-Oct-28, 02:05 AM
Interesting read on galaxy formation and that maybe there's less going on that we might think:

Galaxies Made Simple, or at Least Less Complicated (http://www.nytimes.com/2008/10/28/science/space/28obgala.html)


By HENRY FOUNTAIN (http://topics.nytimes.com/top/reference/timestopics/people/f/henry_fountain/index.html?inline=nyt-per)
Published: October 27, 2008
Galaxies come in all shapes and sizes, from giant spirals to dwarf irregulars, and this diversity is thought to be a function of how they form. One major theory is that galaxy formation is a hierarchical process involving collisions of chunks of cold dark matter, and that the characteristics of any given galaxy are in part due to these chaotic encounters and to other factors including mass, spin and age.

In other words, itís complicated.

Or maybe it isnít. A study of 200 galaxies by Mike Disney of Cardiff University in Wales and colleagues suggests that galaxies are much simpler ó that their properties and structure are a function of a single parameter, rather than multiple ones.

And...

http://physicsworld.com/cws/article/news/36372

Galaxy survey casts doubt on cold dark matter

The physical properties of most galaxies in the universe can be explained in terms of just a single parameter. Thatís the controversial conclusion of a team of astronomers in the UK and US, who have studied some 200 galaxies using radio and optical telescopes. The team believes that their discovery could mean that cold dark matter ó an invisible substance that some astrophysicists have invoked to explain the formation and motion of galaxies ó does not exist. However, not all astrophysicists are convinced.

Ari Jokimaki
2008-Oct-28, 05:44 AM
This paper is not in astro-ph, but there's a related paper (http://arxiv.org/abs/0809.1434).

PraedSt
2008-Oct-28, 09:04 AM
Interesting. My only quibble: might have been slightly wiser to identify the parameter before going public. Makes for a stronger case if you're up against the mainstream ;)

RussT
2008-Oct-28, 09:59 AM
This paper is not in astro-ph, but there's a related paper (http://arxiv.org/abs/0809.1434).

Wow...........Thanks Ari!!!

It would appear that there are a 'Whole Lot" of "Ghost Galaxies" out there...;)

http://www.astro.washington.edu/HIgalaxies/index.html

http://www.narrabri.atnf.csiro.au/public/images/ngc2915/

parejkoj
2008-Oct-29, 04:14 AM
Hmm... The Nature paper makes no mention of CDM, only the hierarchical merger models that have been used to date (and which are known to be incomplete). Why didn't they put that in the paper?

I also definitely wonder about selection effects in their sample, which are not discussed at all in the Nature paper. They claim that their sample gets rid of optical selection problems (which is true: you can map the HI density to much larger radii than you can the optical flux), but it introduces other selection: if a galaxy has had significant recent interaction, or is undergoing strong star-formation, or probably for other reasons, the HI luminosity will be lower and/or distorted. HI is not a panacea.

For those saying "they should have identified the parameter": PCA (Principal Component Analysis) doesn't give any hint as how things are related, it just suggests what the relations are. And even then, one must be careful.

Oh, and there is a galaxyzoo paper coming out that may be related to all of this. I can't say anything more, as I'm not an author and it isn't public yet.

I'm reading the "survey paper" now, as linked by Ari (Thanks!) and referenced in the Nature paper. This is going to be somewhat stream-of-consciousness and full of typos, since I'll just write comments as I read. One can probably just skip the rest of the post, unless you want to know what goes on in my head as I read a paper... :whistle:

First surprising point is that they are referring to SDSS-DR2, which is several years old at this point. Several groups have redone photometry on large objects in SDSS (this is difficult for several reasons): they cite their own method, but I'd be curious how it compares to the others. Also, since their data is in the equatorial stripe, SDSS has much deeper data available from stacking the multiple epochs (I think it gains ~2 magnitudes, though I can't recall). That should be public next week, so I'd hope they take advantage of it, since they do worry about surface brightness...

Second surprising point (but on reflection, it shouldn't be): yeouch, their beam size is huge! That really makes identifying the true optical counterpart difficult.

Statements such as:


... With such complex noise the selection effects which enter into the discrimination of weak sources cannot be anticipated, and must be recognised retrospectively.

worry me. They didn't do anything at all to try and pre-compute the selection effects? WTF? Oh: looks like they did follow-up of potential sources with longer integrations, but that still makes for a very complicated selection function, so although the survey is "blind" (didn't include optical data in the initial source selection) it is certainly not "uniform" (reaching a given depth). Oh again: they didn't follow-up all sources, just a sub-sample. Still makes one worry about the selection function...

Also, repeated use of "eye-brain." heh... On the other hand, might be a handy use of galaxy zoo people, though probably requires a bit of training...

Ook! Plotting "recessional velocity" instead of z makes my brain hurt. Please just add a second label to the x-axis! Ok, quick math says the sources peak around z~0.0075, which is pretty nearby. This significantly helps their optical correlation, since there aren't many sources at those low redshifts. Ahhh... they do consider random coincidences, which are more important that I would have guessed. But I wonder about what the sources are that they throw out, and I bet they do too! :)

Back to the SDSS equatorial stripe: they definitely want to use the stacked stripe-82 data when it becomes public: the surface brightness limits they need are much better matched by it. There have also been low surface brightness galaxies (LSBs) found just using SDSS. I know some people who are working on them. They are pretty odd objects. I wonder how they compare with this sample? These are the things they call "Inchoates" in their catalog.

I also wonder about their claim that the end-products of "Inchoates" don't appear in their survey: they tossed out ~90 sources because of source confusion, and claimed a ~10% contamination fraction from random coincidence. Could the putative "dark clouds" be hiding among those sources? No good way to identify them, however, since the contaminating matches aren't known specifically, just statistically.

Grah! Their figures are out of numerical order! How did they manage that? LaTeX generally disallows such things, and with good reason...

For those crowing about the "death of CDM", I found this little gem:


... In that case the H -band mass-to-light ratio for both our sample, and the GOLDMine sample, is ∼ 20, leaving plenty of scope for dark matter.

:)

This paper also claims that the 5 observed correlations must be independent, which is contradicted by the Nature papers claims from their PCA work. The Nature paper didn't really address this... If the correlations should be independent, and they make a good case for that here, thenwhat is the PCA showing? "Blind" PCA can be dangerous, sometimes...

Ari Jokimaki
2008-Oct-29, 08:24 AM
First surprising point is that they are referring to SDSS-DR2, which is several years old at this point.
I bet they started working on this project so long ago that they had to use DR2. This quote at least hints to that explanation:


Our Equatorial Survey, based on the same raw data, but covering only 5700 deg2 around the Equator (Garcia-Appadoo, PhD Thesis, 2005) was analysed much earlier (1999) with the intention of providing an early source list in time for SDSS-DR2 release.


Plotting "recessional velocity" instead of z makes my brain hurt.
It might take some time still until cz's are dropped totally. After all, cz is more intuitive to many.

PraedSt
2008-Oct-29, 12:31 PM
I also definitely wonder about selection effects in their sample...
And Ari Jokimaki

Sorry, any chance of an explanation of the selection problem/bias for a layman? I've been doing some reading, and I haven't quite understood it yet. What are you trying to avoid, is it a tough problem, how serious an issue is it; that sort of thing.

Thanks :)

parejkoj
2008-Oct-29, 01:44 PM
Sorry, any chance of an explanation of the selection problem/bias for a layman? I've been doing some reading, and I haven't quite understood it yet. What are you trying to avoid, is it a tough problem, how serious an issue is it; that sort of thing.


In any astronomical survey (or any survey at all, for that matter) there are always certain things you miss (e.g. really dim things), or are less likely to find (e.g. rare things) and certain things that are easy (e.g. bright things). One close to my heart is the quasar selection in SDSS (which I've talked about in some other threads): we know that quasars have certain colors at different redshifts (usually!), but some stars have the same colors as some quasars. Thus, if you want to make do a survey of quasars across a range of redshifts (like SDSS did), you need to understand how many you will miss because of the stellar "contaminants." There are different ways of quantifying this, depending on how your sample is constructed, and how well you understand your instrument, the objects you are observing, the sources of noise and the contaminants.

In the case of this survey, their "selection function" (how likely a given object is to end up in the sample) is biased away from ellipticals and disturbed galaxies, and toward galaxies with lots of cold gas. There are certainly other more subtle effects due to the noise properties of their telescope (radio data is notorious for having complicated noise) and the way the survey was conducted. My comment was expressing surprise that they didn't try to quantify the selection function a-priori; yes, that's hard, but it really should be done if you want to have any hope of understanding your sample.

In this case, they are making strong claims that their sample has these correlations and how that puts strong limits on how galaxies evolve. But if they are only finding those objects that, say, haven't undergone significant mergers, then they are missing a class of objects that we know exists and so their results aren't as general as they might like. I don't know whether this is true or not, I'm just worried that they haven't done enough to quantify it, so I don't believe that their results are as general as they claim.

They aren't the only ones; astronomers in general are often fairly blasť about quantifying selection functions. It's difficult to do and not always very interesting.

PraedSt
2008-Oct-29, 02:16 PM
Hey, perfect answer, thank you. Of course I'm going to be annoying and ask you another one now. :)

This seems to me to have a bit of an identification problem thrown in. Wouldn't it be better to unequivocally identify the object first, and then select a subset of that 'catalogue'? Or alternatively, work with the existing catalogue?
Or am I being naive- maybe there aren't enough objects that have been identified without a doubt?

Thanks

parejkoj
2008-Oct-29, 02:30 PM
This seems to me to have a bit of an identification problem thrown in. Wouldn't it be better to unequivocally identify the object first, and then select a subset of that 'catalogue'? Or alternatively, work with the existing catalogue?
Or am I being naive- maybe there aren't enough objects that have been identified without a doubt?


I'll throw it right back at you: how does one "unequivocally identify [an] object?" How do you know your identification is correct? How do you know you haven't missed things when trying to identify objects? What about objects that you didn't think should be in the sample, but really should be? What about objects that shouldn't be in the sample, but are? What about objects that your telescope can't detect, but that are important in your classification scheme? What about objects that are easy to detect, but end up being rare (think of the brightest stars in the sky: easy to see, but very uncommon when you look at the whole population of stars)?

These are all problems that get folded into the selection function, among others. In the case of this paper, they were trying to get around the known optical selection problems (the night sky is bright, so identifying dim things is hard) by using a different wavelength. Which solves one problem, but introduces its own set of problems.

Does that help?

PraedSt
2008-Oct-29, 02:37 PM
Oh yes, definitely. More complex than I thought. You mention above 'astronomers in general are often fairly blasť about quantifying selection functions. It's difficult to do and not always very interesting.' Would you rather that they weren't? Blasť I mean. Are you extra careful in your work, for example?

parejkoj
2008-Oct-29, 02:47 PM
You mention above 'astronomers in general are often fairly blasť about quantifying selection functions. It's difficult to do and not always very interesting.' Would you rather that they weren't? Blasť I mean. Are you extra careful in your work, for example?

:whistle:

Uh.... Look, that dog has a poofy tail!

*sneak* *sneak*

...

Actually, I try to be, but I fully admit that it is hard and generally not interesting. But not understanding selection biases has bit me hard many times in the past (just this week, in fact!), so that's one reason why I focus on it.

PraedSt
2008-Oct-29, 03:04 PM
:whistle:

Uh.... Look, that dog has a poofy tail!

*sneak* *sneak*

Lol. What does this mean? Astronomer humour? :)
Good luck with your research, and thanks again for your help!

p.s I may bother you again in the future..

parejkoj
2008-Oct-29, 03:47 PM
Lol. What does this mean? Astronomer humour? :)

No, just me humor. I was trying to distract you so I could get away from your embarrassing question...


p.s I may bother you again in the future..

Not a problem.

Ari Jokimaki
2008-Oct-30, 09:59 AM
This seems to me to have a bit of an identification problem thrown in. Wouldn't it be better to unequivocally identify the object first, and then select a subset of that 'catalogue'? Or alternatively, work with the existing catalogue?
In addition to parejkoj's answer, I think it's worth pointing out that existing catalogs also have selection effects in them. In fact, one could argue that you would be better off with creating your own sample instead of using existing catalogs, because in existing catalogs you have to rely the descriptions of catalogs on possible selection effects, and it is more difficult to notice if they have missed anything, but if you do it yourself from scratch, you have a better handle on things, so one would imagine that you are also better aware of all the selection effects involved. Using existing catalogs is less laborous, of course, so it's more tempting to use them. However, they rarely are designed to suite exactly to your project, so you often have to extract a subset out of them anyway.

PraedSt
2008-Oct-31, 10:53 AM
... In fact, one could argue that you would be better off with creating your own sample instead of using existing catalogs...but if you do it yourself from scratch, you have a better handle on things...Using existing catalogs is less laborous, of course, so it's more tempting to use them...

Thanks Ari for clarifying the catalogue aspect. We have a similar, (but much easier), problem in my neck of the woods. How to construct a suitable portfolio of say 50 securities, from the thousands available around the world. And, like you said, although it's better to do the exercise yourself, it's much easier to stick to indices, which are the equivalent of catalogues in our industry. Indices are very popular. ;)