Gaia@home

Gaia@home
Project
Status	Active
Category	Astronomy
Compute	CPU
Development
Developer	Gaia@home project team
Author	Astronomical Observatory Institute, Faculty of Physics, Adam Mickiewicz University
Initial release	August 21, 2019  (7 years ago)
Completed	No
Software
Written in	C, C++
Operating system	Linux (64-bit), Windows (64-bit)
BOINC statistics
Stats as of	November 6, 2026  (0 years ago)
Performance	4,264.96 GigaFLOPS
Active users	320
Total users	771
Active hosts	833
Total hosts	4,237
Metadata
Website	https://gaiaathome.eu/gaiaathome/
License	Free software

Gaia@home is a volunteer distributed computing project that runs on the BOINC middleware platform. It is operated by the Astronomical Observatory Institute of the Faculty of Physics at Adam Mickiewicz University in Poznań, Poland.

The project was created to give researchers working with data from the Gaia space observatory of the European Space Agency (ESA) access to a large pool of volunteer computing power. Between 2019 and 2025 the volunteer machines integrated orbits of long-period comets and computed stellar close approaches to the Solar System. On 2 November 2026 the project was relaunched around an entirely new scientific goal, the reconstruction of three-dimensional asteroid shapes from their lightcurves, using an application called SAGE (Shaping Asteroids with Genetic Evolution).

Artist impression of the Gaia spacecraft of the European Space Agency. The project is named after the mission and began as a service for Gaia data users.

History

Origins and the Gaia catalogue era (2019 to 2025)

Gaia@home was implemented by Adam Mickiewicz University under ESA contract No. 4000120180, within the framework of the Permanently Open Call for Proposals of the Polish Industry Incentive Scheme (AO8433), and formed part of GAVIP-GC, the Grid Computing extension of the GAVIP service for Gaia data. The BOINC side of the project opened in 2019, and the project's BOINC pages are archived from 28 September 2019 onward.

The design was unusual among BOINC projects. Instead of one fixed science application, Gaia@home was presented as a general purpose service for astronomers: a scientist supplied an executable, optional input files and a prescription for splitting the Gaia archive into independent chunks, and the service turned this into thousands of small BOINC jobs, each receiving one portion of catalogue data. From 2023 the service could also send queries to the Gaia DR3 catalogue, so that the query results themselves became input files for volunteer computations.

Two scientific programmes were advertised on the project site and were shipped as the first numbered applications, which ran only on 64-bit Linux:

• Long-period comets, the numerical integration of cometary orbits under simultaneous Galactic and stellar perturbations, posed as an N-body problem with on the order of 400 perturbing bodies.
• Stellar close approaches, the recomputation of the encounter parameters of nearby stars from Gaia DR2 and Gaia DR3 astrometry, with ten thousand clones drawn from the covariance matrix to estimate the most probable position and velocity of a star at its minimum distance from the Sun.

The applications on the project server were numbered rather than named, and instances identified as 1_Gaia@home through 6_Gaia@home were created between May 2021 and February 2025. The science pages of the project described both programmes with reference to the published studies on cometary dynamics and stellar encounters.

Relaunch in 2026

In a news item dated 2 November 2026, 07:12:24 UTC, the team announced that the earlier research "has reached the limit of what it could deliver" and that the project had been rebuilt around a new task: reconstructing asteroid shapes from lightcurves, the way an asteroid's brightness changes as it rotates. The changes announced at the same time were:

• Existing accounts, credit, computers, teams and forum posts were carried over unchanged, so attached BOINC clients reconnected without user action.
• Old tasks were retired. Volunteers who saw errors for old Gaia@home tasks were advised to use "Reset project" in the BOINC Manager.
• The new application runs on 64-bit Linux and Windows. The earlier Linux only restriction therefore ended.
• The website moved to a new server, and finished asteroid models appear in a public model gallery.

The relaunched server identifies the workunit as "SAGE: asteroid shape models" and runs dedicated daemons named sage_work_generator, sage_validator and sage_assimilator. The site footer changed from "GAVIP-GC" to "Gaia@home / SAGE".

Scientific background

A lightcurve of asteroid (3671) Dionysus obtained at ESO in 1997. Brightness is plotted against time; the modulation is produced by the rotating, irregular body, and the small dip near the middle revealed a satellite.

Almost all asteroids are unresolved points of light in even the largest telescopes. What can be measured from the ground is the disk-integrated brightness as a function of time: the lightcurve. Because the illumination and the viewing geometry change as the asteroid orbits the Sun and as the Earth moves, a set of lightcurves taken over several apparitions contains information about the shape of the body, the direction of its spin axis and its rotation period.

Recovering that information is an inverse problem. The classical method introduced by Kaasalainen and Torppa assumes that the body is convex, which keeps the inversion tractable, and underlies a very large fraction of the published asteroid shape models, including those collected in the DAMIT database at Charles University. Spacecraft images, however, showed that small bodies are often concave, lobed or bilobed, so a convex hull is only an approximation of the real object.

SAGE, the method now distributed by Gaia@home, drops the convexity assumption and searches for a non-convex shape, a spin axis and a rotation period that together reproduce the observed lightcurves. Because the parameter space is large and strongly non-linear, the search is performed with a genetic algorithm rather than with gradient descent, and it is computationally expensive.

A convex shape model of asteroid (9) Metis derived by classical lightcurve inversion and published in the DAMIT database. SAGE models the same body without the convexity assumption, and (9) Metis is one of the two asteroids in the current Gaia@home model gallery.

The SAGE application

Shape representation

A model is a closed triangular mesh. In the original implementation a shape is described by 242 free parameters, one distance along each of 242 fixed rays evenly distributed on a sphere, so that a vertex can only move radially and the topology of the mesh never changes. This coarse mesh is refined with Catmull-Clark surface subdivision into the mesh actually used for rendering, which has 3,842 vertices and 7,680 triangular faces. The Gaia@home description quotes the same 7,680 triangles.

Because the shapes are non-convex, self-shadowing has to be modelled: for every observation the body is rendered exactly as it was seen from the Earth at that moment, and the parts of the surface hidden from the Sun or from the observer are removed. The reflected light is computed with a linear combination of the Lambert and Lommel-Seeliger scattering laws,

$S=(1-c)\frac{\mu {\mu }_0}{\mu +{\mu }_0}+c\mu {\mu }_0,$

where $\mu$ and ${\mu }_0$ are the cosines of the angles between the surface normal and the directions to the observer and to the Sun respectively, and the mixing factor is $c=0.1$.

Fitting and evolution

Each task is described by the project as "an island of evolution": starting from a parent model it applies small random changes, bumps and dents on the surface, stretching, a shift of the spin axis, keeps the variants that fit the observations better, and after about an hour of computing reports the best models it found. The server then collects the results of thousands of islands and promotes the best models to parents of the next generation, so that the population converges slowly towards shapes that explain the data.

The quality of a model is measured by the deviation between the computed and the observed lightcurves. In the published method the fit is quantified by the root mean square deviation of all lightcurves,

$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{D}=\sqrt{\frac{1}{N}\sum _{i=1}^N(m_i-M_i{)}^2},$

where $m_i$ are the observed and $M_i$ the computed magnitudes, the synthetic lightcurve is allowed to shift vertically to absorb the unknown zero point of relative photometry, and the rotation period is searched on a fine grid, with a step of $10^{-6}$ hours in the original code.

Rendering millions of synthetic lightcurves is the dominant cost. The method was written for graphics processors, and the authors reported a speed-up of about 100 times for rasterisation on a GPU compared with equivalent CPU code, and the original runs of the method were made on a cluster of GPU equipped nodes. The distributed version keeps the per-task cost to about one hour by splitting the population into many independent islands.

Non-uniqueness and uncertainty

Brightness alone often permits several solutions, for example two spin axes that are mirror images in ecliptic longitude. Gaia@home keeps a separate niche for each distinct solution so that one of them does not crowd out the others too early, and it also stores every model that fits the data within the measurement uncertainty. The published result of a run is therefore not only the best shape but also a map of which parts of the surface are well determined and which are not. Pole solutions are conventionally quoted as an ecliptic longitude $\lambda$ and latitude $\beta$, together with the sidereal rotation period $P$.

The same machinery is used for binary asteroids. Many small asteroids are pairs of bodies, and SAGE models both synchronous pairs, which rotate together as one rigid system, and asynchronous systems in which a small moon orbits a fast-spinning primary, rendering the mutual eclipses and occultations of the two components directly.

Observations

The lightcurves come from published archives such as DAMIT and the Asteroid Lightcurve Data Exchange Format (ALCDEF) and from observing campaigns of the research team. The positions of the Sun, the Earth and the asteroid are taken from the JPL Horizons ephemeris service. Unlike the 2019 to 2025 programme, the new application does not query the Gaia archive: its input is asteroid photometry, and the Gaia connection is now historical and institutional rather than operational.

The model gallery

Finished and preliminary models are published on the project website. Each entry is the best solution of one niche, that is, of one distinct spin axis that fits the observations, and every model that lies within the measurement uncertainty of a niche is counted and kept. At the time of the relaunch the gallery contained two runs, both marked as preliminary and both computed by volunteers:

Asteroid	Solution	Spin axis $\lambda$, $\beta$	Rotation period	Models within uncertainty	Tasks computed
(433) Eros	best solution	17°, 11°	5.270256 h	137,560	19,395
(9) Metis	solution 1	184°, 21°	5.079175 h	732,288	64,837
(9) Metis	solution 2	2°, 7°	5.079175 h	176,716

The two Metis rows illustrate the non-uniqueness of the inversion: both solutions reproduce the lightcurves with the same rotation period but with spin axes in nearly opposite directions, and each is maintained as a separate niche with its own population of models. Independent ground truth exists for both targets, which is why they were used to validate SAGE in the original paper: the shape of Eros was measured in situ by the NEAR Shoemaker spacecraft and that of Metis by adaptive optics imaging and stellar occultations.

The gallery credits the volunteers whose computers produced a winning model as co-authors of that model, and lists all participants of a run together with the number of tasks their machines returned, updated hourly; the Eros model alone is credited to 233 volunteer co-authors. The project home page reported 195,287 tasks computed and 6 runs in progress on 6 November 2026.

Asteroid (433) Eros seen by the NEAR Shoemaker spacecraft. The mission measured the shape of Eros directly, which makes the asteroid a benchmark for lightcurve inversion methods such as SAGE.

The same machinery has been applied outside the project. For the Hayabusa2 extended mission target 1998 KY26, SAGE produced a non-convex model with a hint of a large concavity close to the southern hemisphere, and the preferred pole solution was $\lambda =36^{\circ }$, $\beta =-44^{\circ }$, in agreement with the convex solution at $\lambda =29^{\circ }$, $\beta =-41^{\circ }$.

The kind of result SAGE produces. For asteroid 1998 KY26 the non-convex SAGE model (left of each pair) is compared with a convex lightcurve inversion solution (right). The non-convex model shows a hint of a large concavity, which a convex model cannot represent.

Statistics

The project server reported the following figures on 6 November 2026 at 11:51:59 UTC, four days after the relaunch:

• 771 users with credit, of which 320 had recent credit, and 39 newly registered users in the preceding 24 hours
• 4,237 computers with credit, of which 833 had recent credit, and 38 newly registered computers in the preceding 24 hours
• 4,264.96 GigaFLOPS of current computing performance
• 679 tasks ready to send and 5,937 tasks in progress, with 229 users active in the preceding 24 hours
• An average runtime of recent tasks of 58.80 minutes, with a minimum of 36 seconds and a maximum of 1.22 days

The server software version was 1.1.0 on that date.

Software and infrastructure

Gaia@home uses the BOINC middleware developed at the University of California, Berkeley, which lets volunteers donate idle processor time to research projects and handles task distribution, replication and credit accounting. Volunteers attach to the project with the project URL and receive work automatically; the SAGE tasks require a 64-bit operating system and run on Linux and Windows.

The scientific code is written in C and C++, and the original implementation used CUDA for rasterisation on graphics cards; the volunteer version distributes the same calculations as ordinary CPU tasks, which is why a task is sized to take about an hour. On the server side the project runs a work generator, a validator and an assimilator specialised for SAGE, alongside the standard BOINC scheduler, feeder, transitioner and file deleter programs.

Scientific publications

Publications that explicitly credit Gaia@home volunteer computing or describe the Gaia@home project itself are listed here.

Scientific results using Gaia@home computing

• Oszkiewicz, D., Bartczak, P., Colazo, M. and Penttilä, A. (2026). Asteroid phase curve modeling with empirical correction for shape and viewing geometry. Icarus, 446, 116886. The method normalises sparse photometry to a pole-on geometry using precomputed spin and shape models and was applied to more than 25,000 asteroids observed by the ATLAS survey. The paper states that all calculations were performed using the gaiaathome.eu service and thanks the volunteers of the BOINC Gaia@home project for providing the computing power used in the research. A preprint is available on arXiv.

Project descriptions (conference abstract and e-poster)

These publications describe the Gaia@home service and example applications; they are not scientific results produced by volunteer computing.

• Bartczak, P., Santana-Ros, T., Hypki, A., Langner, K. and Dudziński, G. (2022). Gaia@home: Combining Gaia catalogue with the power of volunteer computing. Europlanet Science Congress 2022, Granada, Spain, 18 to 23 September 2022, EPSC2022-659. Describes the original Gaia@home service for running Gaia-archive-based computations on BOINC volunteer computers.
• Bartczak, P. and Santana-Ros, T. (2023). Examples of scientific research using gaia@home. Asteroids, Comets, Meteors Conference 2023, e-poster 2185. Describes example scientific applications and the addition of Gaia DR3 querying to the service.

Project descriptions

• Bartczak, P., Santana-Ros, T., Hypki, A., Langner, K. and Dudziński, G. (2022). Gaia@home: Combining Gaia catalogue with the power of volunteer computing. Europlanet Science Congress 2022, Granada, Spain, 18 to 23 September 2022, EPSC2022-659.
• Bartczak, P. and Santana-Ros, T. (2023). Examples of scientific research using gaia@home. Asteroids, Comets, Meteors Conference 2023, e-poster 2185.

Project team and funding

The project is operated by the Astronomical Observatory Institute, Faculty of Physics, Adam Mickiewicz University in Poznań, at Słoneczna 36, 60-286 Poznań, Poland. The helpdesk address published on the site is [email protected].

The original implementation was funded by ESA under contract No. 4000120180, in the framework of the Permanently Open Call for Proposals under the Polish Industry Incentive Scheme (AO8433), and was part of GAVIP-GC. Scientific work connected with the project has been supported by the National Science Centre of Poland (grant 2022/45/B/ST9/00267) and by the Spanish María Zambrano retraining plan. Authors of the relaunched application include Przemysław Bartczak and Grzegorz Dudziński, who developed SAGE, and the wider team has included Toni Santana-Ros, Arkadiusz Hypki and Krzysztof Langner.

See also

• Asteroids@home
• MilkyWay@home
• Einstein@Home
• SPACIOUS@home
• Universe@Home
• Cosmology@Home
• BOINC projects
• BOINC
• Gaia spacecraft
• Asteroid
• Light curve
• Volunteer computing

External links

• Official Gaia@home website
• How SAGE works
• Gaia@home model gallery
• Gaia@home server status
• Shaping Asteroids with Genetic Evolution (SAGE) on arXiv
• DAMIT, Database of Asteroid Models from Inversion Techniques
• ALCDEF, Asteroid Lightcurve Data Exchange Format
• ESA Gaia mission
• BOINC official website