A full list of my publications can be found on ADS and arxiv. A few selected publications are listed below along with summaries of their key results.
Speagle & Eisenstein (2017a,b): Deriving Photometric Redshifts using Fuzzy Archetypes and Self-Organizing Maps
Building on our previous work on approaching photometric redshifts from an unorthodox viewpoint (see below), Daniel Eisenstein and I looked into how numerous spectral ''archetypes'' combined with local, physical perturbations could be used to derive photometric redshifts in a more flexible, data-driven way. Using Self-Organizing Maps, we explored how our these ''fuzzy archetypes'' could be re-organized in an observationally-oriented way to better deal with degeneracies where very intrinsically different galaxies actually look quite similar in our particular dataset. We then explored several sampling algorithms to exploit these methods, including Markov Chain Monte Carlo and adaptive importance sampling. Tests on mock data showed that most of our methods were reasonably competitive with other template-based approaches both in speed and accuracy. We also used our results to provide some rough forecasts for the upcoming Euclid survey.
Speagle et al. (2016): Exploring Photometric Redshifts as an Optimization Problem: An Ensemble MCMC and Simulated Annealing-Driven Template-Fitting Approach
Traditionally, photometric redshifts have been derived from a series of simple models based on brute-force sampling from a combination of local galaxy templates, dust attenuation curves, extinction, and redshift. While this has been adequate for smaller/medium sized surveys, such a low dimensional, brute-force approach will not be fast nor accurate enough for future large-scale surveys such as, e.g., Euclid. Using a large grid of models adapted from the COSMOS survey, we proceeded to map out what this surface actually looked like so we could brainstorm possible algorithmic solutions. As it turns out, the surface is extremely ugly, with numerous local minima and large degeneracies that confound most greedy gradient descent-oriented algorithms. Using this knowledge, however, we were able to come up with a solution that combined simulated annealing with ensemble Markov Chain Monte Carlo sampling, which we found could explore these surfaces ~40x more efficiently than the corresponding brute-force approaches while retaining similar levels of accuracy.
Steinhardt & Speagle (2014): A Uniform History for Galaxy Evolution
Recent observations indicate a remarkable similarity in the properties of evolving galaxies at fixed mass and redshift, most notably represented by the star-forming ''main sequence'' (SFMS). This prompted us to consider the possibility that most galaxies may actually evolve with a common history, passing through star-forming, quasar, and quiescent phases in a deterministic manner in a similar way that humans pass through childhood, adulthood, and old age. To quantify how ''similar'' galaxies are, we translated the spread in star formation rate (SFR) at fixed stellar mass and time to a spread in time at fixed mass and SFR. This ''synchronization timescale'' then characterizes the extent to which galaxies at a common mass are evolving as a uniform cohort at various cosmic epochs. We then measured this synchronization timescale using nine different star-forming galaxy observations from the literature (taken from Speagle et al. 2014) and Sloan Digital Sky Survey quasar observations spanning z=0-6. Surprisingly, we found this synchronization timescale was a constant ~1.5 Gyr for all combinations of mass and time across both datasets. Based on this and some additional arguments/results presented in the paper, we proposed a simple model in which the SFMS, analogous quasar behavior, and other observations form a galactic evolution ''main sequence'' where star formation occurs first, supermassive black hole accretion comes afterwards, and feedback between the two are dominated by deterministic (rather than stochastic) processes.
Steinhardt, Speagle et al. (2014): Star Formation at z~4-6 From the Spitzer Large Area Survey with Hyper-Suprime-Cam (SPLASH)
One of the more surprising results in galaxy evolution over the past two decades has been the strong observational evidence for ''downsizing'', whereby more massive galaxies evolve more rapidly than there lower mass peers. This is a marked contrast to what one might be expect from simple scenarios for hierarchical stucture formation (i.e. large galaxies form through the merging of smaller galaxies). Observations of the star-forming ''main sequence'' (SFMS) have continued to support this picture, with the most massive star-forming galaxies at fixed redshift becoming increasingly more massive and more active (i.e. have higher star formation rates; SFRs) at higher redshifts. However, at some point this behavior has to stop: not only will the universe eventually be too young for galaxies to have enough time to assemble that much mass, but the implied SFRs will also eventually exceed the expected ''Eddington limit'' for star formation. Using early data collected from the Spitzer Large Area Survey with Hyper-Suprime-Cam (SPLASH) program in the Cosmological Origins Survey (COSMOS) field, we decided to check whether a stellar mass-limited SFMS measured at z=4-6 would begin to show evidence for this turnover. To our surprise, we found exactly what you would anticipate from extrapolating lower-redshift trends to higher-redshift observations: increasingly massive objects with increasingly high SFRs. It is difficult to explain this continued correlation, especially for the most massive systems, unless the most massive galaxies are forming stars near their Eddington-limited rate from the moment of their first collapse. This mirrors a corresponding tension observed with massive, high-redshift quasars.
Speagle et al. (2014): A Highly Consistent Framework for the Evolution of the Star-Forming ''Main Sequence'' from z~0-6
In the last decade or so, a growing number of extragalactic studies have shown that there exists a strong correlation between star formation rates (SFRs) and stellar masses (M) among star-forming galaxies (SFGs) at a wide range of redshifts. However, there remains substantial debate as to how exactly this star-forming ''main sequence'' (SFMS) evolves as a function of redshift, both in terms of the relationship between SFR and M (i.e. the star formation efficiency) and the absolute normalization (i.e. the average star formation rate). Part of the reason for this confusion is that most studies adopt a variety of different assumptions when calibrating their data, making it difficult to compare published results. In an effort to remedy this, we compiled a collection of SFMS measurements from 25 different studies in the literature and went through the arduous task of re-calibrating their results to a common set of assumptions. After going through this process, we found that the data agreed remarkably well with each other across a broad range of observational techniques and most selection criteria. We then used this re-calibrated data to derive a functional form for the SFMS over time, which we used to study some of its possible time-dependent properties. Finally, we explored some of the implications of the mean relation and scatter about our consensus model assuming that galaxies actually evolve ''on'' the SFMS.
UPDATE: A newer version of this figure with data from seven additional studies can be found in this paper.