BRaTS@Home

BRaTS@Home was a volunteer computing project built on the BOINC (Berkeley Open Infrastructure for Network Computing) platform that performed gravitational lensing ray-tracing simulations. The name is a recursive acronym: BRaTS stands for BRaTS Ray Trace Simulator. The project simulated the way dark matter in galaxy clusters and other foreground mass concentrations distorts the images of more distant background objects, work that formed the computational core of developer David Coss's doctoral research at the University of Missouri-St. Louis (UMSL).

BRaTS@Home
Project
Status	Completed
Category	Astrophysics
Compute	CPU
Requires	BOINC client
Development
Developer	David Coss
Author	David Coss
Sponsor	University of Missouri-St. Louis
Maintainer	None (project completed 2010)
Initial release	June 5, 2007  (19 years ago)
Completed	April 13, 2010  (16 years ago) (scheduler shut down; website offline by mid-May 2010)
Repository	https://github.com/kd0kfo/cosmology
Software
Written in	C++
Operating system	Windows, Linux
BOINC statistics
Stats as of	April 12, 2010 (final statistics export)
Performance	~1.1 GFLOPS (average, June 2007 to April 2008)
Active users	0 (at final export)
Total users	115
Active hosts	0 (at final export)
Total hosts	492
Analytics
RAC	0 (at final export)
GPU performance	Not applicable (CPU only)
CPU performance	~1.1 GFLOPS average (June 2007 to April 2008)
Metadata
Website	http://maxwell.dhcp.umsl.edu/brats/ (offline; archived)
License	GPL-3.0 (successor CosApps code)

BRaTS@Home went online on June 5, 2007 and ran for almost three years, attracting volunteers from 26 countries before being retired in April 2010 upon the successful defense of Coss's dissertation. It is remembered as a small, personable project: run single-handedly by a graduate student from a lab computer, invitation-only for most of its life, and famous among its participants for the occasional campus power outage that briefly silenced the server.

History

Origins and launch (2007)

BRaTS@Home was created by David Coss, a physics graduate student at UMSL working under professor Ricardo Flores on the simulation of gravitational lensing. Computing deflection angles by direct integration for realistic cluster mass distributions is computationally expensive, so rather than buy time on a conventional cluster, Coss turned to BOINC, which "allows scientists to build a computer cluster at the price of one server" by harvesting the idle time of volunteers' computers.

The project came online on June 5, 2007, initially offering work for Linux on x86 computers only. The first successful work unit was completed the following day. The project carried an "alpha testing" label from June 13, 2007, and the first small Windows (Windows XP) work units went out on June 18. On July 17, 2007, Coss removed the beta label, writing that "things are running smoothly and I am at a point where science can be done." The statistics service BOINCstats first recorded the project on June 25, 2007.

From early on, enrollment was restricted: new accounts required an invitation code, a state of affairs still in effect as late as 2009. The demand for codes grew so heavy that Coss removed his email address from the front page, directing volunteers to the message boards instead.

Operation (2007 to 2010)

The project was hosted on a computer named maxwell in a UMSL lab, reachable at the address http://maxwell.dhcp.umsl.edu/brats/. Being tied to a campus network had its hazards, and the news archive chronicles a colorful series of interruptions: campus-wide power outages, a router and firewall failure that removed the whole campus from the Internet, an unannounced firewall change that forced a hiatus in early 2008 (used as an opportunity to upgrade to a newer BOINC server), and backbone upgrades by the campus Internet provider.

In June 2008, Coss presented the project at the 212th meeting of the American Astronomical Society in St. Louis, reporting that 182 users in 26 countries had joined, that 795 computers had connected to the server between June 2007 and April 2008, and that the volunteers supplied an average computing power of 1.1 billion floating point operations per second (GFLOPS). At the time, the BOINC platform as a whole averaged more than 1,000 teraFLOPS, putting the scale of the small academic project in perspective.

The project had an active community life for its size, with message boards, participant profiles, a playful "BRaT of the day" profile feature, team competition (45 teams were registered over the project's life), certificates, and an RSS news feed. On January 23, 2010, Coss held an online discussion and question-and-answer session with volunteers about the project's results.

Completion (2010)

BRaTS@Home was, by design, a finite project: it existed to produce the simulations for Coss's dissertation. On March 24, 2010, he announced that he would defend his dissertation on April 9, 2010, and that the project would therefore come to an end. He thanked the volunteers, noting that the dissertation images combined data sets produced on multiple computers, so that "if you have been granted credit... it is likely your computer contributed to part of that."

The scheduler was shut down on April 13, 2010, and volunteers were asked to detach from the project by mid-May 2010, when the university was expected to switch off the server. The final statistics export seen by BOINCstats is dated April 12, 2010, and records 79,464 total credits granted to 115 users and 492 host computers, with 45 teams. Coss received his Ph.D. in physics in 2010, and mirrored the site's content at research.davecoss.com as a keepsake before the original server went dark.

The University of Missouri-St. Louis, which hosted BRaTS@Home in a physics department lab.

Science

Gravitational lensing

Gravitational lensing is the deflection of light by gravity. A point mass $M$ deflects a light ray passing at impact parameter $b$ by the angle

$\hat{\alpha }=\frac{4GM}{c^{2}b},$

where $G$ is the gravitational constant and $c$ the speed of light. In the thin-lens approximation, a mass distribution between a distant source and the observer maps the source position $\mathit{\beta }$ to the image position $\mathit{\theta }$ through the lens equation

$\mathit{\beta }=\mathit{\theta }-\mathit{\alpha }(\mathit{\theta }),\mathit{\alpha }=\frac{D_{ds}}{D_s}\hat{\mathit{\alpha }},$

where $D_{ds}$ and $D_s$ are angular-diameter distances and $\mathit{\alpha }$ is the reduced deflection angle. Because the deflection depends only on the projected mass distribution, lensing offers a way to "see" dark matter, which makes up most of the mass of galaxy clusters but emits no light.

In the weak-lensing regime the distortion is subtle: the shapes of background galaxies are stretched (shear) and magnified (convergence) by small amounts. Both are derived from the two-dimensional lensing potential $\psi$:

$\kappa =\frac{1}{2}({\psi }_{,11}+{\psi }_{,22}),{\gamma }_1=\frac{1}{2}({\psi }_{,11}-{\psi }_{,22}),{\gamma }_2={\psi }_{,12},$

which combine into the complex shear $\gamma ={\gamma }_1+i{\gamma }_2$. Measuring the statistical alignment of many background galaxies (cosmic shear) reveals the projected mass distribution of the foreground cluster.

The BRaTS approach

BRaTS@Home attacked the problem by brute-force ray tracing: each work unit traced light rays through a model mass distribution, integrating the deflection they accumulate along the line of sight, and computing the resulting deflection angles on a grid. The project's main science application was named ray_trace_ellipse, and at its height it was issued for Windows 98 and later on x86 CPUs (version 3.00) and Linux on x86 CPUs (version 5.00), both installed on August 14, 2007; auxiliary applications named test_app and deflector were used for testing. The project's logo was itself a demonstration of the code: a ray-traced, gravitationally lensed image of the Andromeda Galaxy (M31), based on an original astrophoto by Rob Gendler.

For his dissertation, Coss simulated gravitational lensing of realistic galaxy clusters using six high-resolution Adaptive Refinement Tree N-body simulations of clusters (including hydrodynamics), plus a simulation of one group undergoing a merger. Each cluster's three-dimensional particle distribution was projected along three orthogonal lines of sight, yielding 21 projected mass density maps with representative masses and concentrations for the concordance cosmology.

Two simulation methods were compared. The first was the direct-integration ray tracing performed by BRaTS@Home's volunteers; the second computed shear maps with the fast Fourier transform (FFT), a technique also used in observational mass reconstructions. In Fourier space the convergence and shear are related by the Kaiser-Squires relation,

$\hat{\gamma }(\mathit{\ell })=\hat{\kappa }(\mathit{\ell })e^{2i{\varphi }_{\ell }},$

where ${\varphi }_{\ell }$ is the polar angle of the Fourier mode $\mathit{\ell }$; Coss's 2009 consortium paper examined the limitations and artifacts inherent in such FFT-based convolution calculations.

Weak gravitational lensing maps the invisible: in the Bullet Cluster, lensing-derived mass (blue) is offset from the hot gas (red), evidence that most of the cluster's mass is dark matter. BRaTS@Home studied how cluster shape and substructure affect weak-lensing shear signals.

Results

The analysis showed that average shear measured inside and outside half the virial radius of the simulated clusters displays a distinct dispersion, varying by 24 percent from the mean across the 21 maps, while the estimated numerical error is of order 5 percent, meaning the dispersion is a reliable physical consequence of the clusters' shape differences. It correlates most strongly with the ratio of smallest-to-largest principal axis lengths of a cluster's isodensity shell. Image ellipticities, important for mass reconstruction, showed very little variance, but the tangential alignment of the average image distortion was strong enough that the locations of mass density peaks were easily resolvable. In short, the volunteers' computers helped establish how the shapes and substructure of dark matter halos imprint themselves on weak-lensing observables.

Legacy

The project's ray-tracing and shear software was later packaged by Coss as CosApps (Cosmology Applications) and released as open source under the GNU General Public License v3 on GitHub in November 2012. The package provides ray_trace_ellipse, which can compute deflections in parallel across network-connected computers using MPI, and physcalc, which calculates lensing shear from the coupled partial differential equations relating convergence and shear. The code was registered with the Astrophysics Source Code Library in December 2017 as ascl:1712.008. Coss went on to a career in high-performance computing, including building a BOINC-based grid computing system at St. Jude Children's Research Hospital.

The project is also a good early example of a BOINC project that ran to completion and delivered its promised science, wrapping up cleanly with a dissertation, a public results discussion, and a farewell archive for its participants.

Publications

Papers listed on the BOINC publications page

The BOINC project maintains a list of scientific publications arising from BOINC-based computing; the following entries are listed there for BRaTS@Home.

• Coss, David.(2010)."Weak shear study of galaxy clusters by simulated gravitational lensing".(Ph.D. thesis).University of Missouri-St. Louis.Bibcode: 2010PhDT.......285C.Retrieved August 20, 2026.
• Coss, David Raymond.(June 2-5, 2008})."BRaTS@Home and BOINC Distributed Computing for Parallel Computation".In 212th Meeting of the American Astronomical Society, Poster 24.10.St. Louis, Missouri.link.Retrieved August 20, 2026.

Other publications and presentations

Additional papers and presentations by the project author, hosted on the project's research site and documented in the author's curriculum vitae:

• Coss, David.(April 2008})."Simulation of Gravitational Lensing".In NASA Missouri Space Grant Consortium Annual Meeting (conference proceedings).link.Retrieved August 20, 2026.

• Coss, David.(April 2009})."Gravitational Lensing Shear Simulations Using Fast Fourier Transform".In NASA Missouri Space Grant Consortium Annual Meeting (conference proceedings).link.Retrieved August 20, 2026.

• Coss, David.(April 23-24, 2010})."Effects of Dark Matter Halo Shape Variations on Weak Gravitational Lensing".In NASA Missouri Space Grant Consortium Annual Meeting (talk).

(conference talk; results incorporated in the Ph.D. thesis)

• (2008).BRaTS@Home and BOINC Distributed Computing for Parallel Computation (poster abstract, author's copy). research.davecoss.com. Retrieved August 20, 2026.
• (December 2017).CosApps: Simulate gravitational lensing through ray tracing and shear calculation (source code release). Astrophysics Source Code Library. Retrieved August 20, 2026.

See also

• BOINC
• BOINC projects (local project registry)
• Volunteer computing
• Gravitational lensing
• Weak gravitational lensing
• Dark matter
• Cosmology@Home, a contemporary astrophysics BOINC project that also launched in June 2007
• Einstein@Home

External links

• https://web.archive.org/web/20090322190553/http://maxwell.dhcp.umsl.edu/brats/
• https://www.boincstats.com/stats/48/project/detail
• https://ui.adsabs.harvard.edu/abs/2010PhDT.......285C/abstract
• https://github.com/kd0kfo/cosmology
• http://ascl.net/1712.008