PDA

View Full Version : Questions about Yun+ (2001)



Jean Tate
2015-Sep-24, 10:53 PM
I'm going through "Radio Properties of Infrared Selected Galaxies in the IRAS 2 Jy Sample", by Yun, Reddy, and Condon; link to the ADS abstract (http://adsabs.harvard.edu/abs/2001ApJ...554..803Y): http://adsabs.harvard.edu/abs/2001ApJ...554..803Y

I have some questions about this, and am hoping someone can help.

Look at Fig 6; the caption reads "Distribution of q values plotted as a function of IRAS 60 μm luminosity. The solid line marks the average value of q = 2.34 while the dotted lines delineate the “radio-excess” (below) and “infrared-excess” (above) objects, delineated for having 5 times larger radio and infrared flux density than the expected values from the linear radio-FIR relation, respectively."

The caption mentions "the linear radio-FIR relation", but I cannot find it anywhere in the paper (in quantitative form)! :eek:

The closest to such a thing that I can find is equation (4), "A formal fit to the observed radio-FIR luminosity correlation", which relates 1.4 GHz and 60 μm luminosities, as defined in equations (1) and (2).

You might think that "the linear radio-FIR relation" and "the observed radio-FIR luminosity correlation" are the same; however, that cannot be ... because, by definition, q includes "S100μm", which is the IRAS 100 μm band flux density in Jy.

So, quantitatively, what is "the linear radio-FIR relation"?

If you can't work out what "having 5 times larger radio and infrared flux density than the expected values from the linear radio-FIR relation, respectively" is, quantitatively, from what's in the text of the paper, you can work it out from the y-intercepts in Fig 6, right?

Now, look at Fig. 6; in particular the axes ... notice something, um, peculiar? No? OK, try this: reading from the y-axis, what is the value of q, for the two dotted blue lines, and the red line? See it now?

So, what are the values of the y-intercepts (of the two blue lines*) in Fig 6?

Last (rhetorical) question: this paper has been cited over 500 times, per ADS; what did all those papers use, quantitatively, for "the linear radio-FIR relation"?

* the intercept for the red line is 2.34 ... it says so in the caption

Reality Check
2015-Sep-25, 12:43 AM
Hi Jean Tate, have a look at section 3 in the paper:

3. Radio-FIR Correlation
The correlation between the radio and FIR luminosity in galaxies is one of the tightest and best-studied in astrophysics. It holds over five orders of magnitude in luminosity (Price & Duric 1992), and its nearly linear nature is interpreted as a direct relationship between star formation and cosmic-ray production (Harwit & Pacini 1975; Rickard & Harvey 1984; de Jong et al. 1985; Helou et al. 1985; Wunderlich & Klein 1988).
The quantitative form will be in the references.

Jean Tate
2015-Sep-25, 02:05 PM
Thanks Reality Check.

Actually, the quantitative form of the linear radio-FIR relation is not in such references.

For example, the key paper cited - Price & Duric 1992 (ADS link here (http://adsabs.harvard.edu/abs/1992ApJ...401...81P)) - reports several different linear radio-FIR relations, as this part of the abstract makes clear:


We present new results that have a direct bearing on the radio-far-infrared (FIR) relation for galaxies. By separating the thermal bremsstrahlung and synchrotron components of radio emission we have decomposed the radio-FIR relation into thermal bremsstrahlung-FIR and synchrotron-FIR relations for [...] The new relations manifest strong relations of their own. [...] These results shed new light on the radio-FIR relation and explain a number of previously unresolved issues.
1. We find that both radio emission components are tightly correlated with the FIR emission [...] It follows that any mixture of of the radio components produces a tight universal radio-FIR relation.

Further, this 'tight universal radio-FIR relation' is (radio) frequency dependent, and not linear* at all frequencies. From the last section (Summary):


We find that the slope of the radio-FIR correlation steepens with radio wavelength [...] it increases from unity at short wavelengths (where the thermal bremsstrahlung dominates the radio emission) to 1.15 at 75 cm where the synchrotron emission is dominant.

Clearly, the authors' intended audience is other professional astronomers who are intimately familiar with the subject, not amateurs (citizen scientists). But what about professional astronomers whose experience does not include the FIR, observationally?

A clue to what's going on is this (from Yun+2001): "This 60 μm luminosity (L60μm) is the luminosity contribution from the IRAS 60 μm band to the FIR luminosity LFIR6, i.e. [equation (3)] (see Helou et al. 1988)". Here (http://adsabs.harvard.edu/abs/1988ApJS...68..151H) is a link to the ADS abstract of that paper. An interesting (apparent) implication from this, and a central result reported in Yun+(2001), the observed radio-FIR luminosity correlation (which contains L60μm but not L100μm), is that there must be a pretty good correlation between the observed radio luminosity (L1.4GHz in this paper) and L100μm, right? If that's so, then why bother with LFIR?

Lastly, did you look at Fig 6? What did you notice? I urge all readers to also look at Fig 5 (a), and read the text which describes this (not just the caption).

Question for all senior high school science teachers, and those who teach some branch of science at undergrad level in university: if one of your students turned in work with the elementary mistakes evident in these two figures, what would do? Specifically, would you consider such work to be worthy of publication in ApJ?

* In context, "linear" mean "with a slope of 1.00"

Ken G
2015-Sep-30, 02:20 PM
You might think that "the linear radio-FIR relation" and "the observed radio-FIR luminosity correlation" are the same; however, that cannot be ... because, by definition, q includes "S100μm", which is the IRAS 100 μm band flux density in Jy.I think those two are the same, since the correlation has a slope of 0.99, which is indistinguishable from 1. I'm not sure the problem you are seeing in q having both 60 and 100 micron contributions, it is just some kind of convenient mix of the two to characterize the "FIR" luminosity. The overarching issue is whether there is some way to characterize the FIR luminosity such that it is proportional to the radio emission. Physically, this is thought to express a connection between the AGN accretion disk thought to relate to cosmic rays, and the star formation rate. I'm not sure about what feedback is being envisioned there.


So, quantitatively, what is "the linear radio-FIR relation"?Just the physical ramifications of that proportionality, exactly how it is expressed is kind of a detail that is not of great significance. The significance is that there is a way to get it to be proportional, and the physical ramifications of that surprising result.


If you can't work out what "having 5 times larger radio and infrared flux density than the expected values from the linear radio-FIR relation, respectively" is, quantitatively, from what's in the text of the paper, you can work it out from the y-intercepts in Fig 6, right?
Yes, but remember q is a log of a ratio, not the ratio itself.

Now, look at Fig. 6; in particular the axes ... notice something, um, peculiar? No? OK, try this: reading from the y-axis, what is the value of q, for the two dotted blue lines, and the red line? See it now?I'm not sure I see the problem. A factor of 5 corresponds to an offset in a log plot of a little less than 0.8, which is where those lines are. Am I missing something, or is it that you are thinking q is a ratio rather than a log of a ratio?

Jean Tate
2015-Oct-01, 07:41 PM
Thanks Ken G.

I'll respond at length later, but for now: look at Fig 5(a), and read the axes. Take the x-axis, pick a value, say "9". Per the axis title, that's a 60 μm luminosity of 9 'sols', the equivalent of nine Suns, in the IRAS 60 μm band.

Really?!?

No, it's 10^9 ... the axis title is missing "log". Ditto the y-axis.

In Fig 6, there are eight 'tick marks' between the major (unit) marks; making them 0.125 apart. While this is not necessarily a booboo - nothing to say minor tick marks have to be decimal - do the values of the datapoints correspond to these unusual markings? Or to decimal marks; i.e. is the plot accurate per the marks, or not?

Reality Check
2015-Oct-01, 09:06 PM
Thanks Reality Check.

Actually, the quantitative form of the linear radio-FIR relation is not in such references.
That makes "the linear radio-FIR relation" in the caption a reference to the observation that there is a relation and that it is linear, not that there is one and only one ("the") quantitative form of the linear radio-FIR relation. As you have found there are several quantitative forms of the linear radio-FIR relation.

ngc3314
2015-Oct-02, 03:21 PM
The best I can say about this is it's extremely sloppy - full of shorthand informally used by people working in that exact subfield and looking hurriedly prepared. I'm a bit surprised (but not shocked, from experience) that these lapses made it through both refereeing and journal editing. OTOH, some papers have issues that a referee will be so incensed by that other things sort of go under the radar.

Jean Tate
2015-Oct-03, 06:38 PM
Thanks again, Ken G; thanks too to Reality Check and ngc3314.

I'll be posting several (1+) responses; this is just the first.


You might think that "the linear radio-FIR relation" and "the observed radio-FIR luminosity correlation" are the same; however, that cannot be ... because, by definition, q includes "S100μm", which is the IRAS 100 μm band flux density in Jy.I think those two are the same, since the correlation has a slope of 0.99, which is indistinguishable from 1.
They cannot be the same, as "LFIR" is defined as including S100μm - see equation (3).

It may be the case that a correlation between LFIR and L1.4GHz also has a slope of ~1; however, no evidence for that is presented in the paper (as far as I can see).

Further, it may be, for the ~1+k objects they report, that S100μm and S60μm are highly correlated (or, saying this another way, these objects have ~the same (100μm-60μm) colors). However, again, they did not report anything like this (and, as far as I have been able to tell by reading the relevant literature, such FIR colors are not uniform for the range of galaxies reported in this paper).


I'm not sure the problem you are seeing in q having both 60 and 100 micron contributions, it is just some kind of convenient mix of the two to characterize the "FIR" luminosity. The overarching issue is whether there is some way to characterize the FIR luminosity such that it is proportional to the radio emission. Physically, this is thought to express a connection between the AGN accretion disk thought to relate to cosmic rays, and the star formation rate. I'm not sure about what feedback is being envisioned there.
At the level of detail reported, for q, it's not important (except, perhaps, how "the rms scatter of the data is less than 0.26 in dex" is factored into the data displayed in Fig 6).

However, if one were trying to replicate the paper's reported results, using different (FIR, radio) data, I think it's pretty important.



So, quantitatively, what is "the linear radio-FIR relation"?Just the physical ramifications of that proportionality, exactly how it is expressed is kind of a detail that is not of great significance. The significance is that there is a way to get it to be proportional, and the physical ramifications of that surprising result.
This goes to the heart of my frustration in trying to understand this paper.

By not writing down, in a clear, quantitative form, exactly what "the linear radio-FIR relation" is, how can anyone use it, in their own research? Of course, it's easy enough to take independent radio and FIR data (even from a different mission, AKARI say) and produce plots like Fig 5 (a) and (b); however, if those plots are different from what's in Yun+(2001), how can you tell what the cause of those differences is?



If you can't work out what "having 5 times larger radio and infrared flux density than the expected values from the linear radio-FIR relation, respectively" is, quantitatively, from what's in the text of the paper, you can work it out from the y-intercepts in Fig 6, right?Yes, but remember q is a log of a ratio, not the ratio itself.
Now, look at Fig. 6; in particular the axes ... notice something, um, peculiar? No? OK, try this: reading from the y-axis, what is the value of q, for the two dotted blue lines, and the red line? See it now?I'm not sure I see the problem. A factor of 5 corresponds to an offset in a log plot of a little less than 0.8, which is where those lines are. Am I missing something, or is it that you are thinking q is a ratio rather than a log of a ratio?
At one level, I was over-thinking this; "having 5 times larger radio and infrared flux density than the expected values from the linear radio-FIR relation, respectively" is simply ±log(5) :doh:

At another level, I was referring to the fact that the minor tick marks are not decimal, thus raising the question of whether the y-axis (and x-axis) is correctly marked/scaled. Yes, you can digitize the plots, and do an independent analysis ... but why should you have to?

Jean Tate
2015-Oct-03, 06:46 PM
Actually, the quantitative form of the linear radio-FIR relation is not in such references.That makes "the linear radio-FIR relation" in the caption a reference to the observation that there is a relation and that it is linear, not that there is one and only one ("the") quantitative form of the linear radio-FIR relation. As you have found there are several quantitative forms of the linear radio-FIR relation.
Yeah.

And the authors do not state - in a concise, quantitative form - what "linear radio-FIR relation" they used.

So let me ask you, Reality Check: if you were a referee, would you have let this pass? If something like this was given to you by a (PhD) student, would you have said nothing?

Personally, I do not have sufficient experience to judge, but my naive impression is that any PhD thesis adviser would be quite incensed at the evident sloppiness.

ngc3314
2015-Oct-03, 11:28 PM
This thread has just handed me a new homework exercise for a class in observational and data-analysis techniques...

Ken G
2015-Oct-04, 01:31 AM
Thanks Ken G.

I'll respond at length later, but for now: look at Fig 5(a), and read the axes. Take the x-axis, pick a value, say "9". Per the axis title, that's a 60 μm luminosity of 9 'sols', the equivalent of nine Suns, in the IRAS 60 μm band.

Really?!?

No, it's 10^9 ... the axis title is missing "log". Ditto the y-axis.

In Fig 6, there are eight 'tick marks' between the major (unit) marks; making them 0.125 apart. While this is not necessarily a booboo - nothing to say minor tick marks have to be decimal - do the values of the datapoints correspond to these unusual markings? Or to decimal marks; i.e. is the plot accurate per the marks, or not?Those are sloppy mistakes, but sometimes you have to look at what people meant to say, when it's clear enough. It should have been caught by the referee, but it's not guaranteed to be. It's not so important what "the" relationship is, it's that it can be viewed as pretty close to linear-- regardless of the details of how it is defined (such as what kind of mixture of the different FIR bands). The reason to include a mixture is that some galaxies might contribute more of one or more of the other, but combining them creates a kind of net to capture that type of emission. The main idea is that the radio comes from one source, and the FIR another, so a linear correlation between them suggests a proportional connection between those sources. Such a connection will not be a law, so it will be rather informal, but that it exists at all is surprising and significant-- though I don't know how.

Jerry
2015-Oct-04, 04:53 AM
This thread has just handed me a new homework exercise for a class in observational and data-analysis techniques...

Love it! Would love to be in the class. There is a lot of seat-of-the-pants flying in astrophysics, and what is an obvious 'tight' correlation to some is a severe margin of error to others.

Jean Tate
2015-Oct-04, 07:53 PM
The best I can say about this is it's extremely sloppy - full of shorthand informally used by people working in that exact subfield and looking hurriedly prepared.
Wow, thanks ngc3314! :)

When I first read the paper, I thought it was just my lack of knowledge and understanding of the field. However, the more times I (re-)read it, the more it seemed, um, odd that it has garnered >500 cites. I've certainly read plenty of papers that are sloppy (and plenty which are pure gold, in terms of their clarity and precision), but this one seems to stand out, if only because of how often it's been cited.


I'm a bit surprised (but not shocked, from experience) that these lapses made it through both refereeing and journal editing. OTOH, some papers have issues that a referee will be so incensed by that other things sort of go under the radar.
Quite eye-opening to me, a mere citizen scientist! I wonder if it managed to get through the review process because a giant of the field - J. J. Condon - is a co-author? I mean, if the authors had been unknown citizen scientists who'd never published anything before, the reviewers would have been very harsh, wouldn't they?

Jean Tate
2015-Oct-04, 08:26 PM
Those are sloppy mistakes, but sometimes you have to look at what people meant to say, when it's clear enough. It should have been caught by the referee, but it's not guaranteed to be.
Hmm ... look at this from a different perspective: Consider a self-taught citizen scientist (CS), who wishes to write a paper (on astronomy) and get it published in, say, MNRAS. A CS who wants to be lead author and who has some other CSs as co-authors. To guide them, they read heavily cited papers in the general field, such as Yun+(2001). What lessons - if any - should they take away from papers such as Yun+(2001)?

Sure, such CSs are total outsiders, so cannot possibly expect to clearly grasp what the authors meant to say (but didn't); is their only practical course to go get PhDs in astronomy?


It's not so important what "the" relationship is, it's that it can be viewed as pretty close to linear-- regardless of the details of how it is defined (such as what kind of mixture of the different FIR bands).
Actually, I have to disagree with you on this. If only because (100μm-60μm) colors were not - at the time - known to be pretty uniform. To me, the lack of anything in the paper (or any relevant reference) on S100μm seems, um, reckless. It even crossed my mind that the only reason they used FIR (instead of 60μm) was so they could introduce stuff on q, which is a parameter pioneered by one of the co-authors (Condon).


The reason to include a mixture is that some galaxies might contribute more of one or more of the other, but combining them creates a kind of net to capture that type of emission.
Sure.

Except why not actually do the L1.4GHz vs L100μm analysis, along with the L1.4GHz vs L60μm one?

As they make clear in the paper, they had the S100μm data to hand, and all the sources they used for the S60μm analysis also have (had) quality IRAS S100μm data ("The IRAS 100 μm flux is not available for a small subset of the IRAS redshift catalog sources, but all such sources are excluded from our complete sample by other selection criteria").

May I ask you, Ken G, if one of your students had submitted work like this part of the Yun+(2001) paper (on "the linear radio-FIR relation"), would you have allowed the lack of any L1.4GHz vs L100μm analysis to go unnoted, much less given them high marks?


The main idea is that the radio comes from one source, and the FIR another, so a linear correlation between them suggests a proportional connection between those sources. Such a connection will not be a law, so it will be rather informal, but that it exists at all is surprising and significant-- though I don't know how.
Oh it's an important relation, no doubt about that! :) After all, the 500+ cites alone establishes that.

Jean Tate
2015-Oct-04, 08:37 PM
Love it! Would love to be in the class.
As would I! :)


There is a lot of seat-of-the-pants flying in astrophysics,
From my own reading, I can agree with you only to some extent.

Radio luminosity vs x-ray/UV/optical/FIR/... luminosity correlations are certainly not 'seat-of-the-pants flying', and the Yun+(2001) results have withstood very detailed, subsequent research* (that's one reason why the paper has 500+ cites).

From my own reading, 'green valley' is perhaps an example of some rather sloppy research and analyses (have you read up on that?)


and what is an obvious 'tight' correlation to some is a severe margin of error to others.
I'm not sure if you're referring specifically to Yun+(2001) or not; if you are, then my own, independent analyses show that it's not too bad.

Personally, my take-away from this (well, one take-away) is to not trust results presented in such a sloppy way, especially if your own research is going to depend on it to a significant extent. That means a lot of work to find data sources, understand them, and do your own analyses. And if you find something at odds with a key result in a paper as often cited at Yun+(2001), you'd better not write it up anything remotely resembling the sloppy way they did!

* in a general sense; the details of where they have not are very interesting.

Ken G
2015-Oct-10, 02:10 AM
Hmm ... look at this from a different perspective: Consider a self-taught citizen scientist (CS), who wishes to write a paper (on astronomy) and get it published in, say, MNRAS. A CS who wants to be lead author and who has some other CSs as co-authors. To guide them, they read heavily cited papers in the general field, such as Yun+(2001). What lessons - if any - should they take away from papers such as Yun+(2001)?Several lessons I can think of:
1) there is a surprising tendency for brightness measures in the radio to be proportional to brightness measures in the infrared. The specifics of those measures will alter the correlation somewhat, but only in detailed ways-- that there is any such correlation at all is the surprising thing, the thing that needs explaining.
2) sometimes an idea can be a bit vaguely defined quantitatively, yet still have value. There is not always a "canned" formula for expressing important discoveries, and sometimes, looking for too much of a detailed formula can give a kind of cook-booky feel to what is supposed to be a more general conceptual discovery. On the other hand, more precise language does have the advantage of being less ambiguous, so there is a kind of tradeoff there to be navigated.


Sure, such CSs are total outsiders, so cannot possibly expect to clearly grasp what the authors meant to say (but didn't); is their only practical course to go get PhDs in astronomy?
This is the part I don't understand in your criticism. It seems to me what the authors are saying is relatively clear-- there is a linear correlation between radio and IR emissions, expressed in some appropriate way to allow a wider net to be cast to include more systems. The details of how that net is cast are not so important, and might be a bit frustrating to piece together I grant you, but presumably that's because the details of how you cast that net doesn't really matter so much. What matters is that it can be done at all. Also, they are saying that a few systems escape the net, and there must be something special about them. So the key discovery is we now have two things to explain: why there is a linear correlation at all, regardless of the details of how it can be defined (which is actually somewhat arbitrary), and why some systems don't fit it.



Except why not actually do the L1.4GHz vs L100μm analysis, along with the L1.4GHz vs L60μm one? Because some systems emit more in the 100 micron range, and others more in the 60 micron one, but the point is to cast a net that catches all the IR emission, or at least a proxy for it. The IR emission is from one type of source, the radio from another, that's the key point-- not the particulars of the wavelengths. It would be like, what if you discovered that some measure of the gravitational wave emission from most galaxies is proportional to some measure of the starlight the galaxy emits? You wouldn't pick just one wavelength of starlight, because maybe you don't get the correlation if you do that-- maybe starburst galaxies have more blue light, and ancient ellipticals have more red light, and you have to add them together to get the linear correlation. In that case, it wouldn't matter the exact details of how you need to combine the blue and red light to get the linear correlation, only that you could do it at all.


May I ask you, Ken G, if one of your students had submitted work like this part of the Yun+(2001) paper (on "the linear radio-FIR relation"), would you have allowed the lack of any L1.4GHz vs L100μm analysis to go unnoted, much less given them high marks? If they discovered some fundamentally mysterious linear correlation, or a small set of objects that don't obey it, they'd get high marks. I might work with them to clean up some of the informalities, but the discovery would still be the focus.