I'm spending some time over the break thinking about possible long-form writing projects. I have an Atlas of Galaxies to finish, and I have ideas about possible books on introductory mechanics, and ideas about something on the practice and deep beliefs of scientists. And statistics and data analysis, of course! I kicked those around and wrote a little in a possible mechanics preface.
I spent a piece of the morning exhaustively going through short-term priorities with Bedell (Flatiron). We discussed strategy given her stage. She has enough projects to last a decade! I guess we all do, but it is still amazing when we list them. We decided to focus on things that make direct use of the technologies we have built and not particularly build new technology for a bit. We also decided to submit the wobble paper right after the break.
After this, we segued into a conversation about the (badly named) Rossiter-McLaughlin effect with Luger (Flatiron) and Beale (Flatiron). The effect is the effective change in a star's radial velocity as a planet transits its surface, since it is rotating and has a spatial gradient in surface RV. We discussed what is involved in modeling this more accurately than is currently done. There were some philosophical issues coming up around flux conservation, limb darkening, and continuum normalization. All hard issues!
At the end of the day I got in a short quality conversation (over wine) with Alex Barnett (Flatiron) so I could pre-flash him the correlation-function and power-spectrum problems that Storey-Fisher (NYU) and I will bring him in January. He agreed that we are going to effectively unify fourier-space and real-space approaches when we make them all more efficient and more accurate. So excited about a winter of clustering!
I was out sick today. The only thing I did was think a bit about the project code-named myspace in which Adrian Price-Whelan (Princeton) and I warp phase-space coordinate systems to sharpen up velocity-space structure in Gaia data in the local disk. This looks like it works and maybe will provide insights about where the structure is coming from.
I spent a bit of research time on the weekend on writing projects. In one, I am writing about the algorithmic observing strategies that involve sensible objectives, adaptation to what's known previously, look-ahead to the future, and a discount rate. The idea is that observing decisions should be made algorithmically but also just-in-time. And perhaps simply and interpretably, which is even harder.
In another writing project—which is perhaps not reslly research by my strict rules—I am trying to set down my thoughts about the moderation of the submissions to arXiv. Why? Because this blew up this week and I didn't agree with a lot of the things that people were saying, on all sides. Of course if I really write something good, will the arXiv accept it? I think it will get rejected by moderation!
Kate Storey-Fisher (NYU) and I continue to discuss tools for cosmology. She is working on a new estimator for the correlation function, which is fun! But we were tipped off a month or two ago by Tom Abel (KIPAC) about the point that a non-adaptive Poisson random catalog is pretty much known to be the worst possible way to integrate a function. That is, the random catalogs used for correlation-function estimators are almost certainly the dead-wrong methods. And this also connects to the comments we got by Roman Scoccimarro (NYU) this week about the point that the random–random term in the correlation-function estimators being used in eBOSS take weeks to execute! We discussed this in more detail today, and I made a mental note to check in with our math colleagues in January.
My research highlight for the day was a visit by Eric Vanden-Eijnden (NYU) in Math. He showed me (among other things) some new MCMC methods that can make use of dynamics to sample difficult probability distributions or compute fully marginalized likelihoods (evidences). They involve dissipative dynamics, not fully Hamiltonian dynamics like Hamiltonian methods. We resolved to try them out on the binary-star problem that we solved with The Joker, because these problems are multi-modal but in ways we understand.
Discussions continued at Flatiron about Galactic dynamics and actions. We laid out uses for actions and then discussed more results from Beane (Flatiron) on the inconsistency of actions when you have wrong coordinate systems or potential.
Stars meeting featured various interesting discussions. But during a discussion led by Kreidberg (Harvard) about temperature-mapping hot rocky planets, I had an idea: We could use the strong absorption lines in stellar spectra to increase the planet-to-star brightness ratio. If we have full-phase coverage with high-resolution spectroscopy, we can look for the hot planet to “fill in” some of the absorption lines at full phases, and the amount it fills in for lines at different wavelengths would tell you the temperature (or low-resolution spectrum) of the planet! I want to do this with our HARPS data and our wobble pipeline!
In the afternoon, Boris Leistedt (NYU) and I made a plan with David Blei (Columbia) and Andrew Miller (Columbia) to build our 3-d dust model out of dust measurements. There are many problems to solve! But we are starting by assuming that Leistedt's data-driven dust measurements are correct and have Gaussian noise, the stellar positions are well known, and the dust field can be represented by a Gaussian process. In terms of challenges, we are starting by working on the scaling problem: How to make things run on millions or hundreds of millions of stars at a time? One dispute we had is about what line-of-sight integral of the dust corresponds to the extinction?
Today Kate Alexander (Harvard) gave the Astro Seminar. She talked about the observational properties of jets across wavelength but especially in the radio. And unresolved jets, understood through their spectral energy distributions. One point which came up is that there does still seem to be a beaming puzzle: The models of the observations imply high beaming factors, but off-axis examples are very hard to find. So is the model ruled out? MacFadyen (NYU) implied yes, even though he is one of the principal authors of the theories! I think this is a super-important area for multi-messenger and time-domain astrophysics.
Before lunch, Kate Storey-Fisher (NYU) and I had an absolutely great discussion with Roman Scoccimarro (NYU) about our correlation function estimator. He started off very skeptical and ended up a huge fan, which was fun to see, because I am pretty stoked about it! But then he said something off-topic but super-interesting: He has a standard experience on huge projects of the following form: While the correlation-function team is waiting for the data center to compute the correlation-function estimator (which involves an enormous pair-count operation in data and (much more importantly) random catalogs), he computes the power spectrum for the same data sample on his laptop! And yet the correlation function and the power spectrum are (in principle) the same information! What gives?
The answer—which I have to say I haven't fully figured out yet—is in part that the standard power-spectrum estimation doesn't consider explicitly the off-diagonal (k not equal to k-prime) mode cross-correlations, and in part that the standard power-spectrum estimation assumes that the window function is simple enough that a random catalog is not necessary. Those are huge approximations! However, if they are good enough for the power spectrum on baryon-acoustic scales, then they must be good enough for the correlation function on those same scales and maybe we can build a far, far faster estimator?
It is a low-research time of year! But a highlight today was the PhD candidacy exam of Shengqi Yang (NYU), who is working with Anthony Pullen (NYU) on new data analysis techniques for cosmology. She is doing a number of things, but the part I am most interested in is manipulation of combinations of cross- and auto-correlation functions to determine other cross- and auto-correlation functions. These combinations are very simple and valuable! And you can combine observed and theoretical functions as you see fit. I got an argument started in the room about the conditions under which these relationships are exact, or true in the limit, or approximations. I would like to understand that better!
In the morning I met with Gus Beane (Flatiron) to discuss his use and understanding of actions in empirical work on the Milky Way, following up on the blow-up of last week. We discussed the point that small issues with the Galactocentric coordinate system could totally mess any action calculation, even if the actions make sense, and the point that there are many possible galaxies we might live in for which the actions don't even make sense. We vowed to move the conversation / argument going on at Flatiron towards the question of what we are trying to achieve with these calculations. Are they just orbit labels? Or are they quasi-invariants? Or are we using them to match up stars that are far apart?
Stars Meeting was a whirlwind of interesting things! Kim Bott (UW) told us about polarimetry for exoplanet discovery and characterization. Evan Bauer (UCSB) told us about accretion signatures on white dwarfs and how they have probably been mis-interpreted (but in a way that makes accretion more important!). And Suroor Gandhi (NYU) told us about relationships between ages, abundances, and actions in the Milky Way that suggest that there are interesting relationships at all ages, and at all abundances. Hard to summarize, but there are lots of things to think about in there.
Today was the first day of a meeting hosted by the NASA IPAC (home of IRSA and Spitzer among other important projects) to start the discussion of the response of the astronomical data archives to the US Astrophysics Decadal Survey. Not everyone agreed on the point of the meeting, but I think it is to create talking points that connect to archives, but which could be incorporated into community science white papers. These white papers are due in February.
There were several highlights for me at the meeting today. One was Hillenbrand (Caltech) summarizing the white paper process from last decade, and giving advice for white-paper submitters. She emphasized that the white paper text should be cut-and-paste ready for inclusion in the final report. That is, it isn't like a proposal to be approved; it is like a community contribution to the writing of the report. And she emphasized that it doesn't make sense to make points in white papers that will be obvious to the committee!
One of the technical concepts that was discussed today by archives was that of science platforms, in which archives might provide compute resources or other scientific facilities to their users: The idea is to bring the code and the analysis to the data, since the alternatives might be too expensive. But (as I brought up in discussion) then that gets into the space of archives making decisions about what science they do and don't support, which might conflict with peer review, or put scientific projects under various kinds of double jeopardy. And it might mean that projects like LSST, which are doing related things, might end up interfering in unintended ways with the astronomical community and its scientific priorities. These are interesting issues to keep track of.
Today was the NASA Spitzer Science Center Oversight Committee meeting. As usual, the day was filled with insights about how you run a large integrated engineering and science operation. The project team particularly called out Lockheed Martin and the Deep Space Network (separately) for praise in how they have both supported the mission. One of the interesting stories of the day is how the mission depends on legacy hardware, including ground-based computers that are way out of date! The project has no programmers, so it has to keep its software running on late-1990s computer hardware!