pop-III stars won't do it

In the morning, I had a long discussion with Whitmore (Swinburne), Finkbeiner (CfA), and Schlafly (MPIA) about Whitmore's spectral calibration issues and the time variation of the fine-structure constant. In that discussion we ended up deciding (a) that the calibration issue is most likely caused by the arc illuminating the instrument differently from any star, and (b) that this outcome is the best possible outcome for Whitmore, because it can be modeled and calibrated out.

In the afternoon, at Hennawi (MPIA) group meeting, many interesting things transpired. One is that Girish Kulkarni (MPIA) can show that when you combine all the competing constraints, it is very unlikely that "population-III" (primordial-abundance) stars can be huge contributors to the cosmic radiation density; they can't provide a significant fraction of the necessary reionizing photons. Another is that we strongly encouraged Beta Lusso (MPIA) to perform her SED modeling on a large number of SDSS quasars, and fast, to smack down some not-so-good recent results!

All the while, Patel kept working on the Sloan Atlas and the statistics of quasar light curves.

does the fine-structure constant vary with cosmic time?

For the last two days, NYU undergrads Ekta Patel and David Mykytyn have been doing great work on a range of projects. Patel has been working on visualizing and vetting the galaxies that make up the Sloan Atlas of Galaxies. This takes serious judgement and patience! Mykytyn has been quantitatively comparing the g and r fluxes of time-variable quasars, on a path to making a statistical model for arbitrarily heterogeneous multi-band, multi-epoch quasar photometry.

This morning, Jonathan Whitmore (Swinburne) gave a nice talk about monitoring the fine-structure constant (alpha) as a function of cosmic time using metal-line absorption in quasar spectra. He finds no signal (contrary to some claims in the literature) and has a plausible explanation for some weak claims: The detailed wavelength calibration of the echelle spectrograph he is using seems to be different for the arc calibrations and for the gas-cell (or solar analog) spectra. This might be because the arc illuminates the spectrograph very slightly differently than the astronomical source. He has to build a working model of all this if he wants to improve the precision of the experiments, which all seem to be limited by these kinds of issues at the present day.

It might all sound crazy, but in fact most current models of inflation or the vacuum energy density do generically predict fundamental-constant variations. They aren't that specific about magnitudes of fluctuations within the Hubble Time, but the predictions are there qualitatively.

defocus for precision?

In a set of summer-student talks, Tanja Schroeder (MPIA) showed absolutely astounding photometry of transiting exoplanets taken with defocused data from Calar Alto and Bologna. Yes, defocused: By defocusing the telescope, she can avoid saturation in longer exposures and sample more of the (noisily known) flat-field, at the expense of more background. It sounds insane but when your noise has larger contributions from systematics (unknown and variable PSF, unknown and variable bias and flat), defocusing can increase precision. She also showed nice exoplanet-transit fits to those data. In the question period, Wolfgang Brandner (MPIA) brought up the idea of learning limb darkening from exoplanet transits, which is the subject of a (sleeping) project by Barclay (Ames), Foreman-Mackey, and me.

Bayes

I spent the day at MPA and MPE in Garching, where I gave a seminar in the "Bayes Forum". I talked about hierarchical inference for quasar target selection, exoplanet characterization, and astronomical image processing (source detection). The audience asked all the right questions. After my talk, Torsten Ensslin showed me some nice results from his "information field theory" approach to inferring density fields and other maps from finite data. His approach is a well-defined, controlled approximation to Bayesian inference that makes use of the equipment of field theory.

lensing cross-correlations, Gaussian Processes

In Galaxy Coffee, Myers (Wyoming) talked about cross-correlations of quasars with the CMB convergence (lensing) map. There are now large-scale convergence maps available, so it is possible to more-or-less automatically generate object-lensing cross-correlations. These are pretty straightforwardly interpretable in terms of halo occupation or mass correlations. This is a great new capability for cosmology, and permits testing of a lot of halo-occupation ideas that have been kicking around for a few years.

In the afternoon, Foreman-Mackey talked about our work with Kepler. He made various important points, but what I learned most clearly is that the biggest advantage we gain over our competitors by using Gaussian Processes is in what we call "search": We can do hypothesis tests or parameter estimation quickly but still marginalize out the stochastic variations of the stars.

undergraduate research

David Mykytyn and Ekta Patel, NYU undergraduates, showed up today for three weeks of sprint in Heidelberg. In addition to whatever Rix and I throw at them, they are working on GALEX calibration and the Sloan Atlas. Mykytyn worked on reformatting some of our GALEX Python code to make it more function based and less script-like. Patel worked on splitting the Sloan Atlas galaxy sample into subsamples by color and intensity.

Fomalhaut b

Hervé Beust (Grenoble) gave a nice talk about inferring orbital parameters for the (directly detected) exoplanet (or exoplanet-like thing) in the Fomalhaut system and then how those parameters might be related to the morphology of the dust debris ring around Fomalhaut. He can't find a consistent dynamical solution, that explains both the exoplanet and the ring, without adding new planets. His inference makes use of priors on exoplanet parameters with which I don't agree (although they are industry standards, so I am not complaining) and it also assumes that the mass of Fomalhaut is very well known. I would be very interested to see if you could explain the system entirely by (a) relaxing the star-mass prior, (b) sharpening the exoplanet orbital-parameter priors, and (c) insisting that the exoplanet orbit be of a size and shape that could explain the ring. Would that fail? Great (but very difficult-to-answer) question!

(In question period, Beust more-or-less admitted that the formation, origin, lifetimes, and morphologies of these dust debris rings are not well understood: They should be very short-lived objects. And yet they are observed in many systems. Interesting, that.)

yet more proposal writing

Spent the weekend and today on boring proposal stuff.

Atlas and proposal

Spent the day (in bed, for uninteresting reasons) working on my Atlas and NYU's Moore–Sloan proposal.

data analysis consulting, mixture models

Julianne Dalcanton (UW) gave a great Galaxy Coffee talk about mapping the dust in M31, by a very clever mixture-model fitting approach, fitting the extinctions towards red-giant stars in the PHAT data. She shows amazing angular resolution and incredible relationships with emission measures from infrared and millimeter. And that after a great Galaxy Coffee talk from Aaron Dutton (MPIA) summarizing the meeting "The Physical Link between Galaxies and their Halos". He did the very clever / sensible thing of choosing the three things about the meeting that most impressed him, and only talking about those.

I got back into my "data analysis guru" mode today, with long conversations with a group (including Smolcic and Groves and many others) that is trying to detect very faint sources in very deep JVLA imaging of blank fields: How do you know that the sources you are seeing are real, and how do you measure their properties? Much of the non-triviality comes from the fact that the raw data are interferometry visibilities and the maps are made with (relatively speaking) black boxes. I was a very strong advocate of jackknife (or full likelihood modeling, which is a great plan but outrageously hard given where everyone is right now).

I also spoke a bit with Watkins (MPIA) and van de Ven (MPIA) about modeling dynamical systems in the presence of a large-amplitude background, such that the model must include not just the dynamical system of interest (a stellar cluster, in this case) but also the foreground or background (the non-cluster stars of the galaxy or Galaxy, in this case). I worked through my understanding of the mixture models that the Loyal Reader (tm) knows so much about. It gets confusing when the mixture amplitudes are conditioned on observables or data that are not explicitly being modeled in the likelihood; for example in the Watkins case, the velocity distribution is a mixture of components, but the mixture amplitudes depend on position. I originally recommended modeling position and velocity simultaneously, but given the crazy selection of stars they face, it is better to model velocity conditioned on position. This makes the mixture less trivial.

In general we would all be better off if we understood mixture models much better. They obviate hard classification and capture a lot of our ideas about how our data are generated.

exoplanet search: hypothesis test or parameter estimation?

As the Loyal Reader (tm) knows, Foreman-Mackey and I have been working on exoplanet detection in Kepler data. Today we decided to switch gears from seeing planet detection as a parameter estimation problem (do we find a planet with sufficiently well-measured depth to be declared Real?) to a hypothesis test problem (does the hypothesis "there is a planet here" beat the null hypothesis "there isn't"?). We are using a Gaussian Process to model the stochastic variations of the star and a rigid but good exoplanet model to model the mean expected behavior of the stellar flux subject to transits. We did a bit of pair-coding, even.

By the end of the day, Foreman-Mackey could show that our detection capability is likely to be an extremely strong function of the hyperparameters of the Gaussian Process, and that this function will very likely vary strongly from star to star. In particular, we expect that sufficiently variable or faint stars will never yield hyperparameter settings sensitive enough to detect Earth-like planets (the planets we need to become internet famous). That will have significant impacts on our decision-making and experimental design. We have yet to insert all this hypothesis testing into a brute-force search loop, in part because our code is so damned slow (but righteous!).

proposals

After four days completely off the grid, I got very little done except the unbloggable email-answering and a few tasks on our Moore-Sloan proposal. The most interesting part of my day was reading a new (and excellent) proposal by Bovy regarding the Milky Way.

wacky ionization states

At Galaxy Coffee, Ben Oppenheimer (the Leiden one, not the AMNH one) gave a nice demonstration (theory) that the ionization states of metals around quasars are very likely to be out of ionization equilibrium with any radiation field, from the quasar or the cosmic mean. The reason is that if the quasar illumination varies, it can take quite a while for the atoms to cascade back to equilibrium. He treated this as a problem—complexifying interpretaton—but it also might (in some circumstances) be an opportunity, as it shows that some kinds of line ratios could be an indicator of illumination history.

Not much else happened today, although both Foreman-Mackey and Weisz spent time discussing code issues in their respective projects.

quasar typology

In Joe Hennawi's (MPIA) group meeting today, lots of discussion broke out about quasars, unification, viewing angles, dust obscuration, and black-hole growth. It was great! Beta Lusso (MPIA) has a very nice argument about dust-obscured quasars: When she and her team fit the spectral energy distributions of non-obscured quasars, they find that more than half of the light from the central engine is being absorbed and re-radiated by an optically thick component (usually called, leadingly, "the torus"). If more than half of the photons are absorbed by an optically thick component, then the simplest, basic prediction is that (for randomly placed observers) more than half of the quasars should be strongly obscured. Great argument! Of course there are ways around it (optically thick material might be in a heterogeneous web of low covering factor; there might be strong temperature gradients in the optically thick material that lead to anisotropies in emission, etc), but it is a solid order-of-magnitude argument.

Before that, and partly starting the obscuration discussion, Nic Ross (LBL) showed some strange BOSS quasars, which were selected in the optical (by Bovy's XDQSO, I presume) but which turn out to be extremely bright in WISE 22-micron data. So bright, indeed, that naïve estimates make them among the most luminous objects in the Universe. (They might even be gravitationally magnified). The odd thing is that they have this huge infrared luminosity but show no sign in the optical of being reddened or obscured. The only clue is that (perhaps) they have large equivalent widths for their narrow lines. Nic: Did I get that right?

flexible rigid models

I worked on Gaussian Processes with Foreman-Mackey today. In our formulation of the "search Kepler data for transiting planet candidates" problem, what is usually called "the model" of the transit sets (in our formulation) the mean of a Gaussian Process, and the photon noise, stellar variability, and spacecraft-induced variability is all subsumed into a covariance function that sets the variance tensor of that Process. The nice thing about this formulation is that the estimation or inference is just what you would do with linear chi-squared inference (optimization or sampling or whatever you like), but just with a much more complicated covariance matrix. In particular, the covariance matrix is now nothing like diagonal.

We are off the usual path of Gaussian Processes, but only because we have a non-trivial function for the mean of the Process. That isn't much of a change, but it is rarely mentioned in the usual texts and descriptions of GPs. I think the fundamental reason for this is that GPs are usually used as incredibly flexible models for barely understood data. In typical astronomical (and physical) problems, there are parts of the problem that are barely understood and require flexible models, but other parts that are very well understood and require pretty rigid, physically motivated models. In our use of GPs, we get both at once: The mean function is set by our physical model of transiting exoplanet; the variance function is set to capture as well as possible all the variability that we don't (officially) care about.

More soon.

nothing

Just more proposal writing.

writing

My only research today was some small writing on our Moore/Sloan proposal.

Milky Way stars

It is a low-research week because of the move to Heidelberg for the summer (consult rules at right). But today I went to Rix's Milky Way group meeting, where Branimir Sesar (Caltech) told us about observations of red giant stars and RR Lyrae stars in the orphan stream. His work suggested a very small velocity dispersion for the stream, and also (apparently) points to a low mass for the Milky Way halo. Like Koposov, Rix, and me, he is fitting the stream with an orbit, not a tidal-stream model, which might lead to some biases.

At the same meeting, Rix showed predictions from Stinson and collaborators for chemical abundances for stars of different ages. They seem to confirm the picture we have been pushing with Bovy that the chemical abundance boxels (in Fe/H and alpha/Fe) are close to age boxels. He also noted that Sesar's work with red giants strongly endorses Rix and Xue's distance estimation work: In the stream, the RRL and K-giant distances seem to agree very well.