Hi Jon and Roger, Thanks for yesterday's update on BICEP2 FTS status: On 12/20/12 5:03 AM, Obrient, Roger (Guest) wrote: I have been following the plots as they are added to Jon's posting. The news looks mostly encouraging, but I am left with many questions--some just about what is actually being plotted since there is limited text and some ambiguous figure labels, but some questions are more significant. I'm glad Roger and perhaps others at Pole will help review and clarify your posting while you push to complete final analysis in the next 24 hours. We need to wrap this up so BICEP2 can warm. The most important comments below are probably about repeatability under pointing changes--make sure you have close enough spacing in multiple pointings (9? 16?) at at least 2 or 3 DK angles to judge systematic variation, and to be able to pull out close-to-on-axis data for each detector if needed to give our best estimates of the A, B, and differential spectra. That is my guess, based on the plots I've seen so far--but your analysis should tell you how well you need to do in this. --------------------------------------------------------------------------------------- Here are comments posting in the order that it stands right now: Setup: can you include a photo? "It's clear that we can get spectra for one tile at a time." In terms of S/N maybe, but of course you should not assume that they are all good--off axis pixels may have increased systematics (e.g. freq dep coupling) so until you've proven that's not a significant concern you should assume more pointings may be necessary. (There is more on this below...) Spectra: Fig 4: shows repeatable structure at 200-300 GHz--if real, we'd want to convince ourselves that is OK (I actually expect some level of out of band response from island coupling if nothing else, but that's probably well-matched A/B). So: how repeatable are those features under dk rotation, scan speed change, signal strength change, etc? Fig 5: left vs right slope is almost surely an analysis artifact, I'm not worried about it per se but it needs to be understood to interpret other kinds of systematic repeatability. Fig 6: UL, What does this pattern look like for alt pointings? Is the taper seen here a tile center-to-edge effect, or a systematic of off-axis coupling into the FTS? LL, that's a surprisingly large range of measured BW's, from 30-40 GHz, by your definition. Fig 7: L vs R repeatability is probing whatever analysis artifact is producing the slope of Fig 5. From the LL scatter plot, looks like it varies for each tile plotted here--is it random with each run? Probably low priority... Fig 8: UL, What is being plotted?? Is it percent? If so, totally saturated color scale so not a very useful plot, nor is the LL histogram (which is misleading, because I guess all the outliers are off range.) You probably want to fix your definition of BW to interpolate to continuous rather than discretized values. Low priority right now, but fix it later. Fig 11, BW repeatability under DK rot: looks very poor, judging from UR and LL figures I'd say we can't trust your BW's to better than 10-20%. Makes me wonder about the robustness of your algorithm. You are not just picking out 50% crossing points, are you? An integral will be more robust. You don't have a link or a description of your algorithm in this posting. Positive vs Negative from WLF scans: "Here we definitely see a systematic shift of order ~1%. Here, were definitely running into some systematic that is slanting the spectra. This may be a beam illumination effect as our beam is somewhere between a point source and beam-filling. " Agreed. The pattern in Fig 13 UL seems telling. Can I assume that this, and all your other plots that don't say otherwise, is constructed from data taken with a single pointing per tile? The systematic divergence is greatest around the tile edges--this looks like direct evidence for systematic skewing of spectra from pixels that are off-axis through the FTS. Scan speed: Good to see generally nice agreement, at least for the aspects of the spectra you are plotting. The different (small) offsets in frequency axis for the different tiles (Fig 15 UL, LL) is telling you something about a non-repeatable systematic error in your frequency axis for each run--you should understand this, but fortunately it appears to be small. Are you using the encoder data in your analysis (if not, will you add it?) How do the out-of-band features respond to the different scan speeds? I'm talking about the inset of Fig 4. Multiple Paths: You are labeling plots here "misaligned"...are they? Your description says that you moved the stationary mirror by 0.5", not that you misaligned the FTS. Please clarify. A 2x0.5"= 2.5cm change in total path might be expected to alter any systematic fringing on ~10 GHz periodic scales. I'm reassured to see that the dominant fringing in the spectra don't appear to change significantly, lending evidence it is real and in the detectors. There are significant systematic shifts in the spectra, perhaps consistent with other alignment/pointing systematics being explored by the DK rot, +/-, and pointing consistency checks. Nine Pointings This is probably the analysis you should have started with. The question you need to ask here is how repeatable are the spectra and quantities derived from them for a given pixel that enters the FTS at different angles and/or different orientations. As we discussed yesterday, you can make focal plane coordinate plots of the mismatch in e.g. band center or (more interestingly) A/B relgain mismatch for pointings that closely overlap vs. those that are offset by a larger amount. See if you can draw conclusions like: "spectra taken for multiple pointings where the FTS central ray is within a 3 pixel radius vary by < 1.5 GHz, but pointings further off-axis give larger discrepencies." Make a similar statement for A/B relgain mismatch. This informs how many pointings you need to get full coverage with well-controlled systematic uncertainty. Spectra Gain Mismatch Please fold these results into the appropriate sections above--it is more confusing to have them separated out at the end like this. Fig 23, scan direction: Are these the exact same data as was processed in Fig 7? The pattern of inconsistency does not appear similar to me, so it would appear that this consistency check is probing a different kind of divergence in the left vs right spectra than the simple slope of Fig 5...assuming that is what is behind Fig 7. Odd. Worth looking at the A vs B spectra to understand this. Fig 24, DK rotation: A LOT of seemingly random variation. As an aside, please fix your histograms. As usual the scatter plots are very informative. There is some correlation here but a very disappointing level of spread. You make a statement: "The spread looks pretty large but if you look at the dxi_df (which is integrated to get spectra gain mismatch) for each detector, they are clearly sub-percent." Sub-percent compared to what? You sound like you are trying to reassure us this is a small effect, but it does not appear to be. There is a hint in the Fig 24 LL scatter plot that actual spectral gain mismatches of order 1% are being measured in these spectra with S/N of no better than 1. I think that is worse than BICEP1 results, but I'm not sure--can you make a direct comparison? Obviously other consistency checks need to be added as well. It would be helpful to see plots that illustrate the A and B spectra time the atm model separately, and then their difference (which is what I think you are plotting here...but you need to define dxi_df or link to a definition). How does B2 average (not differential) spectral response in the region of the 120 GHz and 180 GHz atm lines compare with B1 150 GHz spectra? Are our B2 detectors hitting those lines harder than B1 did? -- ___________________________________________________________________ John Kovac jmkovac(a)cfa.harvard.edu Assistant Professor, Astronomy and Physics, Harvard University 160 Concord Ave rm 310, Cambridge MA 02138, 617-496-0611