BitBOINC

🌠 Nostos: tracing interstellar objects back to the stars they came from 🌠

Nostos is a new research project here on bitboinc. Since it's a very different kind of problem from Kampylos or the puzzle hunt, here's the long version of what it actually computes, and why each step is there.

☄️ Visitors from another star

In October 2017 the Pan-STARRS telescope in Hawaii picked up a small object moving through the inner Solar System far too fast to be bound to the Sun. Its orbit wasn't an ellipse at all but a hyperbola: it came in from interstellar space, swung once around the Sun, and left for good. That was 1I/'Oumuamua, the first interstellar object ever seen. In 2019 came the second, the comet 2I/Borisov, found by an amateur astronomer.

Objects like these are thought to be planetesimals and comets thrown out of young planetary systems around other stars, by close passes with giant planets for example. Our own Solar System is believed to have ejected huge numbers of them in its early history. So every interstellar object is a piece of some other star's planetary system, and the natural question is: which star? That's what Nostos ("homecoming" in Greek) tries to answer.

🧭 The basic idea: run the clock backwards

Everything in the Milky Way moves. The interstellar object moves, and so does every star around us. If we know where an object is and how fast it moves today, we can compute where it was in the past by running its motion backwards under gravity. Do the same for the stars around us, and you can ask, for every star: how close did this object come to it, and when?

A star the object passed very close to in the past, and slowly, is a candidate for its home system. "Slowly" matters as much as "close": an object ejected from a planetary system leaves its star at a fairly low speed (typically a few km/s), so a star that the object merely flew past at 30 km/s can't be where it was born, however close the pass was.

🌌 The ingredients

[*]The interstellar object's path. Astronomers measured the orbits of 'Oumuamua and Borisov very precisely while they were here. From the orbit we compute the "inbound asymptote": the straight line and speed the object had before the Sun's gravity bent its path, i.e. how it was moving through interstellar space on its way in (about 26 km/s for 'Oumuamua, 32 km/s for Borisov, relative to the Sun).
[*]The stars. The European Space Agency's Gaia satellite has measured positions, distances (parallaxes) and motions across the sky for over a billion stars. For about 33 million of them, Gaia DR3 also has a radial velocity, the motion towards or away from us. Only with all six numbers (3D position + 3D velocity) can a star's past path be computed, so these 33 million stars are what Nostos searches. Gaia parallaxes have a small known systematic offset; it's corrected with the official recipe (Lindegren et al. 2021) before anything else happens.
[*]The Milky Way's gravity. Over millions of years, stars and interstellar objects don't move in straight lines: they follow orbits around the centre of the Galaxy. Nostos uses a standard published model of the Galaxy's gravitational field (Allen & Santillán 1991: a central bulge, a flattened disk and a dark-matter halo) and integrates every path step by step, one thousand years per step.

🔁 What one task does

Each task is one interstellar object against 1500 Gaia stars. For every star it:

[*]converts the star's Gaia measurements into a position and velocity in the Galaxy;
[*]integrates the star and the interstellar object backwards together for 100 million years, tracking the distance between them at every step;
[*]records the closest approach: how close, and how long ago;
[*]does all of that 100 times, each time with a slightly different version of the interstellar object (see below), and reports the closest approach over the 100 copies and its average;
[*]keeps the star only if it came within 50 parsecs (about 160 light years); everything else is thrown away so results stay tiny.

That's 100 copies × 100,000 steps × 1500 stars per task. The whole search is 33 million stars per interstellar object, about 2.4 CPU-years of computing for each one.

🎲 Why 100 copies of the object?

No measurement is perfect. The interstellar object's incoming velocity is known to a few hundredths of a km/s, which sounds tiny, but over a million years a 0.02 km/s difference shifts its position by about 0.02 parsecs, and over tens of millions of years by much more. A single "best guess" path would give a single closest-approach number with no idea how reliable it is. So Nostos uses the published uncertainty of each object's velocity (including how its three components are correlated) to draw 100 plausible versions of it, and runs all of them. If all 100 pass close to a star, that encounter is robust; if only a few do, it isn't. Each chunk of stars gets its own fixed random seed, so any task can be re-run and gives exactly the same answer.

🔍 Checked twice

What volunteers compute is the first stage: the exhaustive scan. Every star that comes out close is then checked again in a second stage on the server, this time also varying the star's own measurement errors (its distance, proper motion and radial velocity, with their correlations from the Gaia catalogue), thousands of times. Only an encounter that stays close under both the object's and the star's uncertainties counts as a candidate.

✅ Proving it works before trusting it

Before a single task went out, the pipeline had to reproduce independent published results: the closest known encounters of 'Oumuamua with the stars HIP 3757 and HD 292249 (Bailer-Jones et al. 2018), and of Borisov with Ross 573 (Bailer-Jones et al. 2020). All three now come out inside the published uncertainty ranges; HIP 3757, for example, passed about 0.6 parsecs from 'Oumuamua roughly a million years ago, exactly as published.

Getting there wasn't automatic, and that's the point of the check. Two real bugs turned up along the way, both in how the interstellar object's starting state was set. First, the code used the object's outgoing direction instead of its incoming one; the Sun bends the path by over 100° for 'Oumuamua, so the backwards trace pointed in the wrong direction entirely. Second, the starting speed was 2-4% too high, because it was taken from a point that was still too close to the Sun. Both are fixed, and the published encounters now reproduce.

📏 What the numbers mean, and their limits

[*]Distances are in parsecs (1 pc = 3.26 light years). For comparison, the nearest star to the Sun today, Proxima Centauri, is 1.3 pc away. An encounter well under 1 pc means the object passed through the outskirts of that star's own system.
[*]"How long ago" is in millions of years. The further back, the less certain: small errors in the measurements and in the Galaxy model grow over time, so encounters older than about 50 million years are treated as lower confidence.
[*]Gaia only covers stars that exist and are bright enough today. If the object was ejected billions of years ago, its home star may be far across the Galaxy, too faint, or no longer a star at all. So finding no convincing birthplace is a real and possible answer, not a failure.

🎯 Why do this exhaustively?

The best earlier searches (Bailer-Jones and colleagues, 2018 and 2020) used the roughly 7 million stars with radial velocities in Gaia's previous data release. Gaia DR3 has about 33 million, measured more precisely. Nostos runs every one of them, with the uncertainty included for every star. What can come out of it:

[*]a complete catalogue of close stellar encounters for 'Oumuamua and Borisov over the last 100 million years, which doesn't exist yet and will be published openly;
[*]a fresh, independent check of the origin candidates proposed so far, with newer data;
[*]and maybe a star the object passed slowly and very close to: a real candidate for its home.

The Vera C. Rubin Observatory, which recently started its survey of the whole southern sky, is expected to discover many more interstellar objects in the coming years. Each one can go into the same search, so Nostos is meant to keep running as the list grows.

'Oumuamua goes first, then Borisov. Progress and the closest stars found so far are on the Nostos page. Questions are welcome below.

---

this project looks really cool!

---

I crunch for Ukraine

I am up for this one, I like astronomy and this sounds very interesting.

Conan

The application generally runs smoothly. On a modern processor, the process is expected to take approximately 20 minutes. Please let us know if you encounter any problems. A GPU version will be available soon; this version is expected to take 50 to 70 minutes. The GPU calculates 27 times the area of ​​the CPU application.

---

Cool again a new scientific Subproject! I love!

I would think about run a second Mainproject with scientific subprojects...perhaps that one becomes a place in Statspages after that bad start. But i dont know how much they want to communicate now about that possibility.

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

Cool Project, I think my 6500XT will be crunching it this evening.

By the way...
Could You provide HIP-Apps für AMD GPU?
Should be easily translatet from the CUDA-Apps.
I would think there would be a performance benefit compared to OpenCL.

Hi!
First:
Thank You Alperen for starting all this nice and interresting Projects, To be honest I've never seen such a well organized and such helpful
community and all this just after just a week 👍🤓🙂😍👏👏

Second:

Are there some problems with the statistics and badges for this project?
I can't see any Nostos wus crunched neither on Per_App credits nor on Computing Stats for my Account, I also think I've got at least Wanderer badge 🤔

With regards and very happy to help,

Hans S.
Oslo, Norway

I think badges service runs not very often - I also noticed i crunched enough WUs but did not get a badge yet ...

---

I crunch for Ukraine

Is this Project Quorum 2?
My WU are waiting for validation. :)

In reply to Krümel's message of 9 Oct 2026:

Is this Project Quorum 2?
My WU are waiting for validation. :)

Yes, it is.

In reply to Krümel's message of 9 Oct 2026:

Cool Project, I think my 6500XT will be crunching it this evening.

By the way...
Could You provide HIP-Apps für AMD GPU?
Should be easily translatet from the CUDA-Apps.
I would think there would be a performance benefit compared to OpenCL.

We tested the HIP but decided against it because it ran slower than native OpenCL.

---

Thank You for the information.