Referee report from ApJ: two weak lensing fixes needed

[drphilmarshall]

Over in the base repo of the LSST overview paper Zeljko @ivezic has been working through the ApJ referee report. He needs DESC's help with two weak lensing issues. We'll prepare an improved paper in the apj-v1 branch - and then submit it in a PR against the base repo apj-v1 branch including proposed response text in the PR comment.

---

• [x] https://github.com/lsst-pst/overview_paper/issues/45 Weak lensing background comparison

  Referee: How does the effective number density of weak lensing background galaxies compare with Euclid and WFIRST? Euclid is observing a similarly large area with a better PSF, no seeing, no weather, no Moon, no daytime, and a darker sky.

This refers to a sentence on line 79 in science.tex, which says "The sample to $i=25.3$ will include several billion galaxies. At a slightly brighter cut, there will be around 30 galaxies arcmin$^{-2}$ with shapes measured well enough for weak lensing measurements \citep{2013MNRAS.434.2121C,2015MNRAS.447.1746C}, with the number realized in practice being dependent on the performance of the deblending and shape measurement algorithms." What we need is an additional sentence that gives the corresponding number for the Euclid Gold Sample. Can you please provide such a sentence, @timeifler @chihway @rmandelb?

---

• [x] https://github.com/lsst-pst/overview_paper/issues/46 Update Figure 22 on weak lensing

  Referee: The weak lensing results will be competing with Euclid which has a similar start date. There are several surveys that have been "skimming the cream on weak lensing science", including CFHTLenS, PanSTARRS and DES. So far the cream has not produced spectacular results. Figure 22 seems frozen in the DETF days. Figure 22 should acknowledge the surveys in the can, and show an "All" that include Euclid WL and then "All minus LSST WL". So you should include a fair assessment of how much the LSST will advance the field over where it will be given current and under construction surveys.

I believe the Figure 22 that the referee is referring to is now Fig 23 in the current version: We have new figures from the DESC SRD that we could include. I am inclined to argue that the LSST overview paper should not include Euclid contours - that's for the Euclid team to present. But, we can include the appropriate Y1 and Y10 contours from the v1 DESC SRD, no problem. I'll insert this and see what it looks like.

[rmandelb]

Hi @drphilmarshall - I'm commenting here so @timeifler @chihway and I can discuss. However, once we've converged, I can help to put this into the branch for a PR.

1. Euclid and WFIRST number density: The referee is correct that Euclid has a number of advantages that she/he has listed. However, it's also a lot shallower, so the number density should be similar to LSST (depth gains you a lot). The most defensible number to quote is 30/arcmin^2 in the 2011 Euclid definition study report, Laureijs et al., unless some Euclid WL paper has a more recent estimate. (I mean: we should be quoting something by the Euclid team, I think.) @chihway @timeifler do you have some other suggestion for what we could say here, e.g., from a more up-to-date Euclid paper? We can also quote a WFIRST number from the 2015 Spergel+ WFIRST SDT report. That one is substantially larger than LSST/Euclid because of the depth, around 48/arcmin^2.

2. Constraining power: I am inclined to agree that we should update this with the DESC SRD figure and a suitably descriptive caption. I actually would argue quite strongly, on principle, that we shouldn't be doing "All [including Euclid]" and "All minus LSST", because it gives the impression that total additions to the FoM are everything. That is not the impression we want to make. Part of the era of precision cosmology will be cross-checks, independent measurements with different limiting systematics and methodology, etc. So we should not show a figure that encourages people to just count the beans (FoM) and use that as a means to evaluate the value of these surveys happening at the same time. My suggestion is that this argument should be made in the referee response (while updating the figure to the DESC SRD one, so we are being responsive to a good chunk of that comment while disagreeing with just part of it.).

[rmandelb]

Oh, and if we want to acknowledge surveys that are "in the can" we could show a DES 3x2pt+SN Y1 contour, if @timeifler is willing to add that? (Seems safer than including a DES and/or HSC full survey forecast, since those aren't really "in the can".)

[drphilmarshall]

Thanks very much, @rmandelb - while you were thinking, I pushed a couple of minimal starter commits to the apj-v1 branch, take a look at the diff. I think it's nice to show the figs from the DESC SRD if we can - can we argue that our Stage III is a fair approximation of what Stage III will actually give us? I didn't cite Laureijs et al for the 30 per sq arcmin number but we can add that no problem.

[ivezic]

Re: Rachel's proposal in "2. Constraining power": I agree! I will use any words you suggest in the reply to the editor/referee.

[drphilmarshall]

Good: I hope we are close. Here's a screenshot of the current DESC fork apj-v1 branch PDF on the one page we need to fix up:

[rmandelb]

@drphilmarshall - no, our Stage III is not an approximation of what Stage III will actually give us. It's only Planck + BOSS BAO + JLA + H0 (the Planck + ext chains, basically). We deliberately didn't try to include DES WL for example, because there are covariances with LSST that we'd have to account for, and it would get ugly. We didn't include eBOSS because I think we didn't have forecasts for it. It was basically "Stage III that was already in the can and that we could argue is independent of the LSST constraints so the combination is easy".

[rmandelb]

@ivezic - glad to hear you agree. Once I hear from Tim and Chihway about the other points, I can suggest further text edits and also some text for the referee response to push back on that second point.

[timeifler]

Hello, I am not sure where to start... :)

• Yes, Euclid has 30 ngal/arcmin^2 (Laureijs et al 2011 is a good reference).
• Showing the DESC-SRD figure is certainly a good idea for LSST forecasts. Please also cite Krause & Eifler 2017 (http://adsabs.harvard.edu/abs/2017MNRAS.470.2100K) for the DESC-SRD forecasting methodology and for an even more detailed forecast of the LSST static probes including systematics.
• Systematics control (astrophysics and observational) will be the name of the game for all these surveys and comparing statistical constraining power of individual endeavors does not give a good picture of reality. We will have to cross-calibrate and correlate data from all these missions and hence these should be considered synergistic and not in competition. I would phrase the answer to the referee in that light.
• The Euclid ngal number relies on the VIS band only, which is 1 band 550-900nm. The claim is that this will go down to i-lim mag 24.5, which is much shallower than LSST (shallower redshift distribution=less constraining power). Other than that Euclid has even more shallow Y,J,H data which is not useful for WL (it is somewhat useful for photo-z).
• Euclid relies on additional imaging from the ground to get photo-z for their 30 galaxies/arcmin^2. Without this info the Euclid FoM is very close to zero (perhaps better not to mention that...) In the south this basically means that Euclid requires LSST information or it will not be able to make use of that part of its survey area.
• Euclid survey strategy is "step and stare", which means it will build up area over time (2500 deg^2 in the first year and so on). LSST will probably have covered ~18000 square degree by 2023, with sufficient ngal to not being noise dominated. This means that LSST will reach Euclid constraining power ~ 5years earlier. Probably also better not to mention since Euclid does not like this statement.
• Spergel SDT report is a good ref for WFIRST; ngal in that report is 45 galaxies/arcmin^2, but newer unpublished numbers are closer to 50 indeed. WFIRST baseline survey is just 2000 deg^2, so higher ngal than LSST but less area.
• For a stage 3 survey I'd recommend using the plot I sent to Dan and Michelle for the white paper which includes DES 3x2pt Y6. This has been computed with the same methodology as the LSST forecasts and is the closest to an apples-apples plot I can think of at this point. DESC_SRD_DES_LSST_Y10.pdf

[drphilmarshall]

Thanks @timeifler ! I included that plot, and modified the caption to read, in addition to what I had before, the following:

"In the lower panel we show, for reference, the expected constraints from the Y6 Dark Energy Survey 3x2pt analysis (brown): the corresponding LSST analysis is already expected to be higher precision than this after just 1 year.

I also added citations to Laureijs et al and Krause and Eifler, and included you as a co-author on the overview paper. Hope that's OK! Do please give the WL section a read in case it is missing anything or needs bringing up to date.

[rmandelb]

@drphilmarshall - the latest version of the diff looks very good to me, just two minor points.

1. it would be good to add WFIRST to address the referee's first suggestion, since she/he explicitly asks about it. Is the 45-50/arcmin^2 and reference to the 2015 WFIRST SDT report enough for you to go on or do you need more info / suggested text?

2. In addition to the Jain+ white paper, it would be good to cite Rhodes+17, which builds on the earlier Jain paper and is more quantitative in a few respects when it talks about survey synergies and tradeoffs in strategy and so on.

@timeifler - I agree with you that we shouldn't get into these Euclid vs. LSST tradeoffs (going deep vs. going wide and what it does to constraining power over time; Euclid's need for optical data) in either the paper or in the referee response.

[rmandelb]

@drphilmarshall @ivezic - given @timeifler 's agreement with my suggestion, I would suggest the following sentences to be inserted in some form into the referee response (I assume somebody else will write the sentences about how we did update the plot to have a better understanding of what LSST and DES will do, and this is just the bit that pushes back on the comparison with Euclid):

We have not added cosmological constraint contours that correspond to "All" including Euclid WL and then "All minus LSST WL", as you suggested. The reason for this is that a key part of the cosmological community's Stage IV dark energy program is that there are multiple surveys that are dominated by different systematic uncertainties and use (at least mostly) independent analysis algorithms. These enable multiple cross-checks that provide confidence in our control of astrophysical and observational systematic uncertainties when placing tight constraints on dark energy (or modified gravity, neutrino masses, and other measurements that will also be made using those surveys). The Weinberg+13 review article on cosmology, Jain+15 white paper, and Rhodes+17 ApJ paper more thoroughly explore the value of these tradeoffs in survey strategy, analysis methodology, and so on that will enable us to validate that we have adequate control of astrophysical and observational systematic uncertainties. If we provide a FoM plot that appears to emphasize the purely statistical constraining power of these surveys, we would be missing this key element of the community's Stage IV program. Finally, as a purely practical matter, we do not have Euclid forecasts that are in the same 7-dimensional cosmological parameter constraints as the LSST ones, or with reasonably consistent assumptions, nor do we think that such constraints provided by the LSST community would necessarily be seen as credible (as compared to whatever the Euclid consortium provides based on their best current understanding of the instrument and survey).

[ivezic]

Thanks! I can use these words in the reply to the referee (as soon as I have all responses, I will check it in). As for edits in the paper itself (including the figure update), I am assuming that at some point, I will get a pull request. Please advise if not.

[timeifler]

Thanks @drphilmarshall ! Sure, happy to be a co-author. I had a quick look at the relevant sections and of course one can write more details, but it is definitely a good overview and I would not recommend changes beyond the ones already mentioned above in response to the referee.