Escaping Gas Is Pushing 3I/ATLAS. Its Orbit Cannot Say Which Ice.

For anyone trying to measure how hard a comet's own gas pushes it, the best observations are the ones taken close to the Sun, where the ice is most active and the recoil is strongest. Małgorzata Królikowska, Federico Spada and Luke Dones threw all of those away on purpose.

What the three of them kept is 3I/ATLAS at a distance. The manuscript went up on arXiv on Oct. 28, 2026, and says it has been accepted for publication in Astronomy & Astrophysics; the journal has not published it yet. The analysis uses 4,403 position measurements, every one taken farther from the Sun than 3.49 astronomical units before the comet's closest approach and 2.67 astronomical units after it. An astronomical unit is the Earth's distance from the Sun. The approach itself, at 1.356 of them, is gone from the data set, and with it every measurement from the part of the orbit where the gas pushes hardest. Królikowska works at the Space Research Centre of the Polish Academy of Sciences in Warsaw. Spada is at the Università di Catania and Dones at the Southwest Research Institute in Boulder, Colorado. Their question was not whether the comet is being pushed. It was whether distant measurements can say what is doing the pushing.

The push itself is not news. NASA's Jet Propulsion Laboratory carries a nongravitational term in its cataloged orbit for 3I/ATLAS, and Eubanks and colleagues fitted one in 2025 from their own selection of the measurements. A third determination, published in 2026, shares an author with this work and is not an outside check on it. Nor is anything exotic on the table. All three of the nongravitational models tested here are activity laws for ice turning straight to gas in sunlight, the ordinary engine of every comet. They differ only in how steeply that production of gas falls away as the comet retreats.

Taken separately, neither distant arc says much. The stretch before closest approach is described well by gravity alone: its residuals, the gaps between where the comet was seen and where the orbit says it should have been, scatter the way pure measurement error does. The stretch after it does not, and the authors are careful about what that does and does not mean. The excess could be dynamics their model leaves out, or systematic differences between the observatories that supplied the two halves of the data, or both, and the position measurements cannot separate them.

Together the two arcs are a different matter. Fitted with gravity only, the combined data scatter by 0.328 arcseconds, worse than either arc managed alone, and the leftover errors drift steadily with time in a way a correct orbit does not produce. Add a nongravitational term and the scatter falls to about 0.25 arcseconds and the drift disappears. That is the first result: the push is visible in the comet's distant motion, measured over the part of the orbit where it was least active.

The second result is the one that matters for how this comet gets written about. The three activity laws take very different shapes with distance from the Sun, yet over the range the observations cover they produce fits that differ by one or two thousandths of an arcsecond. "The astrometric data do not uniquely distinguish between the tested activity laws," the authors write. The two flatter laws would separate from each other only beyond roughly 20 astronomical units, far outside anything in the record.

Nor are those laws species identifications. The paper is explicit that they stand for generic classes of volatile ice rather than for carbon dioxide or carbon monoxide in particular, even though those two gases are the reason to try them. Webb and SPHEREx found a carbon-dioxide-rich coma, the cloud of gas around the nucleus, far from the Sun, and the comet's output shifted toward water as it came in. The authors call a preference for one of those laws "a physically motivated interpretation supported by independent observational evidence, rather than a requirement imposed by the astrometric data themselves."

Even with a push in the model the fit is not clean. In declination, the north-south coordinate on the sky, the leftover errors still fail a normality test under every law tested. The test statistic sits around six where 0.785 would be a pass, so none of the tested models fully explains the observations. One reading, which the authors raise and decline to press, is that the recoil changes along the orbit: a wandering high-latitude jet reported in 2026, and the lopsided sunward outgassing Webb recorded, would both do that. Distant position measurements cannot test it.

A smaller result arrives with its own warning. All three asymmetric fits put the comet's activity peak a few days before closest approach, while published solutions disagree even about which side of it the peak falls on. Królikowska and her colleagues read their own value as a description of the stretch of orbit they sampled rather than a property of the comet, and when they re-cut the data more symmetrically the shift came out no larger than its own uncertainty.

What the work adds up to is one sentence in its conclusions: "The existence of nongravitational acceleration is therefore considerably better constrained by astrometry than its physical origin or heliocentric dependence." Positions on the sky can show that something other than gravity is moving a comet. Naming the ice takes spectroscopy, which has already shown what comes off this one, or position measurements from much farther out than anyone has yet collected.