From apjl@helios.harvard.edu Wed Jun 2 15:37:34 1999 Date: Wed, 2 Jun 1999 11:49:03 -0400 (EDT) From: Ap J Letters To: mrnolta@Princeton.EDU Subject: referee report Referee's report for the paper A MEASUREMENT OF THE ANGULAR POWER SPECTRUM OF THE MICROWAVE BACKGROUND MADE FROM THE HIGH CHILEAN ANDES by Torbet et al. First I address your specific questions: 1. Does the paper contain substantive new results or theories and does it reflect sufficiently high scientific standards to warrant its publication in the Astrophysical Journal? Yes. The paper reports on the results from a major effort to make measurements of the angular power spectrum of the CMB: The results are at least as significant as any previous measurements, and provide important confirmation of the picture which had been evolving from previous measurements. Also, this paper provides an important indication of the quality of their observing site, where other experiments are planned to be performed. These results should undoubtedly be published. 2. Is fast publication as a Letter justified? A Letter should have a significant immediate impact on the research of a number of other investigators or be of special current interest in astrophysics. Permanent, long-range value is less essential. A Letter can be more speculative and less rigorous than an article for Part 1, but should meet the same general standard of quality. Yes. These results are in a fast evolving field. A prolonged delay of could have significant consequences on the excitement of the results. 3. Is it brief and to the point? Papers that are longer than 4 journal pages cannot be accepted as Letters. The paper could be abbreviated in a few places, which I will indicate, without reducing the relevance of the paper. I would claim that there are no new techniques presented here, and so the paper need only address 3 questions: 1) what is the measured spectrum 2) should I believe it? 3) how good is the observing site. Material which does not address these questions need not be kept. (eg, description of well known procedures). On the other hand, there are a couple of places where I will request further enhancement. I leave it to the editors to determine if it meets the page guidelines. 4. Within this space limitation, has sufficient introductory background explanation been included so that the paper can be generally understood by readers who are not specialists in the field? In order to fit into the space required, the authors have made frequent reference to previous longer papers (not letters) where adequate background material is presented. There choice of references seems appropriate. 5. Can you provide comments or criticisms that may be helpful to the author to improve or correct the manuscript? Yes: suggestions will follow ----------------------------------------------------------------------- Changes: Items marked with a (*) I consider required. All others are merely suggestions which I feel will strengthen the paper. section 2: Paragraph 1: i don't like "...is essentially..." Maybe "...is a modified reincarnation of..." or something. paragraph 2: "...three ground based observations..." Don't you mean "...three ground based observing campaigns..." ? "...flat in frequency and linear in phase between 3Hz and twice the sample rate..." (*) What is flat in frequency? I assume the gain. If so, say so. How then can the gain be flat above the sample rate? Don't you have anti-aliasing filters? If so, they are normally set to cut off well before 1/2 the sample rate, not twice the sample rate. And it takes a very special filter indeed to be both flat in frequency and phase up to the cut-off. So either delete this whole sentence (I'm happy to assume that you took the transfer function of your instrument into account properly) or do a better job explaining what you did or didn't have to do. Paragraph 3: The first 2 sentences are fine, and can be added to the previous paragraph in desired. The rest seems either confusing or unnecessary: eg: why did you measure the chopper temperature and current? Of course you calibrated the chopper. Why does the rms deviate from 'a nominal set of values'? What is a 'nominal set of values'? How do you know it has deviated? Why do you then delete the data. If I should care, expand this section. If not, you can delete it. What you could do instead of this 'spotty detail' is simply list the effects which you have though about and dealt with... Section 3: Paragraph 2: Why is it interesting that you observed Jupiter with multiple azimuthal offsets. Presumably because this assists in chopper calibration and verification of the receiver frequency response. You may as well say so, so I don't have to guess. Why do I care what your factors for going from RJ to Thermodynamic are? This is pass band dependent - so its not a generally useful number. I think that you can delete the sentence. It is important to say that the numbers are in thermodynamic units though. Paragraph 3: "...variance...": don't you mean standard deviation? In this paragraph we have once again a fair bit of 'spotty detail'. What I want to know is 1) the calibration uncertainty, and what you included in it. 2) the beam uncertainty 3) that you used Jupiter to calibrate and get the beam. The other stuff is confusing/unnecessary Section 4: paragraph 1: "...record the quadrature..." don't you mean "...generate the..."? paragraph 3: do you compute the 'sample variance' or the just the 'variance'? It sounds like you mean variance. Paragraph 4: * Why did you cut daytime data? It does look terrible in the rms noise plot, but why? Is it atmosphere? Is it side lobes? Is this going to matter for the interferometers heading up there? I think that these are very relevant questions. Also, in the figure, the D data often never hit the noise floor of the receiver. relevant for planned mm-wave mega-projects planned for this site. Figure 1 caption: What NRAO opacity measurements? --------- Paragraphs 5&6: This is a critical section regarding believability and quite frankly, it leaves me a little nervous.... Due to these issues, I am going to require more information in table 2, which may need to be broken up. Paragraph 5: atmospheric cuts "...with a large rms." How large? Do you try different cuts? Does it matter? (*) "...sections with rms > 1sigma..." What?? You cut at 1\sigma!! Am I completely confused here? What does this do to the statistics of the results? The distribution now in no way resembles a Gaussian. Isn't this kind of important? This is significant enough that results at different cut levels need to be reported! "...roughly the same..." Isn't this the same atmosphere? So if it is bad in one channel, shouldn't it be suspect in all of them? Paragraph 6: offsets Any guess as to what causes the offsets? A jump of 3-5 sigma is 60 - 100 uK. How do I know that it doesn't jump in the middle of the night? What is the shape of the offset? Is it the same in each Ka/Q/D channel? Are any of the n-pt analysis clean, or is it in all of them? If so, at what level? <|offset|> would be a good addition to table 2! Last Paragraph of section 5: The first half of this paragraph is meant to alleviate my fears over the results of the previous section. But I have to say it doesn't help much. Do you mean that you really intended on cutting data at 1sigma? Either be specific about what changes didn't matter (preferable!), or don't talk about it. (*)Null tests: I am not happy with the statistic you give for the null tests in table 2: A bunch of 1.9 sigma detections of trouble (which some of these results might actually be given only table 2), while not significant individually, are, as a whole, very significant. And the level of these 2sigma upper limits are not far from the level of the detections in the data channels in many cases. Take as an example, Ka1/2 9 pt result. The signal is ~70uk. The null tests give around 70uK 2sigma upper limits for contamination. So at the two sigma level, can't I say that the true answer could be reported as sqrt((70)^2 +/-(70)^2) or in the range 0uk to 99uk (2 sigma) It is clear that signal on the sky was detected, but it would be great if one could 1) estimate the level of the problems (perhaps a reduced chi-squared) 2) report how significant the trouble is (like perhaps reporting the probability that the null signal would exceed the measured signal if there were no troubles). Given this, you could perhaps find out how much the answer would change if you jacked up the errorbars so that the chi2/nu of the null tests was 1. There are probable better statistics than this, but you get the idea.... In any case, more work could be done in proving that the result is essentially clean. Section 5: paragraph 2: doesn't dI/I from the beam depend on what bin you are looking at? Table 1: The only information here that we really care about in the context of this letter is the beam size. These 4 numbers could be listed in-line in the discussion of the instrument, and the table deleted.