My past academic research focused primarily on methods for the analysis of astrophysical data. I was particularly interested in reliable inference about weak gravitational lensing from large imaging survey experiments such as the Dark Energy Survey, WFIRST, and Euclid. This meant that I used to think a lot about pixels, images, and noisy data.
More recently I have published far less frequently, and read more widely. But I have been working on an idea that has interested me for a decade, wondering whether we might be able to use the entropy of distributions of residuals when fitting models to data. I hope to find the spare time to write as much on this topic as I can!
With Rachel Mandelbaum (Carnegie Mellon) I helped lead the GREAT3 Challenge project (the third GRavitational lEnsing Accuracy Testing Challenge). For information about the Challenge, please see the GREAT3 Challenge web page.
To create the GREAT3 Challenge data we established an open-source software project called GalSim: the modular galaxy image simulation toolkit. This was an ambitious project to provide the community with a free, well-vetted toolkit for simulating realistic galaxy images, based on physical models of telescope optics, the atmosphere, and using real galaxy images from the Hubble Space Telescope. You can read about GalSim in the paper here or here. I am very proud that GalSim is now one of the core image simulation tools adopted by the incredible Large Synoptic Survey Telescope project, among others.
The Dark Energy Survey
At University College London I was part of the Capitalizing on Gravitational Shear (COGS) project (PI: Sarah Bridle), and together with Sarah Bridle, Joe Zuntz, Lisa Voigt, Michael Hirsch & Tomasz Kacprzak we worked towards the accurate measurement of cosmological lensing information in the Dark Energy Survey (DES). This huge telescope project is designed to probe the origin of the accelerating universe and help uncover the nature of dark energy by measuring the 14-billion-year history of cosmic expansion with high precision.
A big part of the work for DES is the open-source im3shape galaxy shape estimation software, described here. This is a maximum-likelihood galaxy shape inference tool, designed from the outset to be as modular and easily extensible as possible. The project was led by Joe Zuntz, and results were very encouraging in terms of both robustness and overall bias.
Optimal Linear Image Combination: IMCOM
In a very enjoyable project funded by the NASA WFIRST mission, I worked on the implementation of an image combination algorithm devised by Christopher Hirata which we called IMCOM. The paper is available here.
The IMCOM algorithm allows for careful treatment of aliasing in undersampled imaging data, and can be used to test the feasibility of multi-exposure observing strategies for space-based survey missions. IMCOM has been used to explore focal plane undersampling for the optical space mission such Euclid, and has been used in laboratory analysis work for infrared focal plane arrays for WFIRST.
The IMCOM software is now freely available on GitHub.
A modern analysis of the current largest area of deep, optical, Canada-France-Hawaii Telescope data using gravitational lensing. A forerunner of DES, the CFHTLenS project has constructed the largest map of dark matter yet made, and placed new constraints on the validity of Einstein's General Relativity on cosmological scales.
Some of the freely-available software I have been involved with helping build in support of the projects described above:
- IMCOM (IMage COMbination) - code for exploring (and combining) undersampled image data (Fortran 95)
- GalSim: The modular galaxy image simulation toolkit - weak gravitational lensing-focused software toolkit for generating realistic galaxy and telescope PSF images (Python wrapping C++)
- great3-public - initially helper and utility scripts for shape measurement method authors preparing submissions to the GREAT3 Challenge , now the entire code used to build GREAT3 (Python)
- im3shape - weak lensing shape measurement code, led by Joe Zuntz at the University of Manchester as part of COGS.