The collision that has a date but not a verdict
Andromeda’s closest approach to the Milky Way is pinned to within about a fifth — some time between 3.9 and 4.6 billion years from now. How close it comes is uncertain by a factor of twelve, which is why two teams reading the same sky put the odds of a merger at 50 per cent and at 90 per cent.
You already know the sentence. In roughly 4.5 billion years the Milky Way will collide with the Andromeda galaxy, the two will tear each other into a single featureless ball of stars, and the night sky will never look the same again. It is in textbooks. It is in planetarium shows. It has been the settled answer for most of a century.
In June 2025 a team led by Till Sawala at the University of Helsinki published a paper in Nature Astronomy with a title that reads like a correction notice: No certainty of a Milky Way–Andromeda collision. They had simulated the next ten billion years fifty thousand times, varying every measured quantity across its published uncertainty. In a little over half the runs the galaxies merged. In the rest they did not.
the fate of our Galaxy is still completely open
That is how Sawala and his colleagues ended the paper. Nine months later a second group, led by Hao Wu at Peking University, re-measured the single quantity that mattered most and came back with a merger probability of 90 per cent. Same galaxies. Same physics. Same decade of telescope data. Two very different answers.
What is actually being measured
Andromeda’s motion splits into two parts, and they are not remotely equally known.
The part coming straight at us is easy. Light from an approaching galaxy is shifted toward the blue end of the spectrum, and that shift can be read off a spectrum with great precision. Andromeda is closing at about 110 kilometres every second, and that figure is good to about one kilometre per second.
The part sliding sideways across our sky is the hard one. It cannot be read from a spectrum. It has to be caught as a change in apparent position, measured against more distant objects, over years. For a galaxy 761 kiloparsecs away — about 2.5 million light years — that drift amounts to roughly thirteen millionths of an arcsecond per year.
To put that in something you can hold: the gap between what the two teams measured, about eight millionths of an arcsecond per year, is the angle a United States quarter would cover if you held it up from about 616,000 kilometres away. That is a little over 1.6 times the distance to the Moon.
And the sideways speed is what decides everything. A galaxy falling straight in collides. A galaxy with enough sideways motion sails past.
Run it yourself
The calculator below treats the Milky Way and Andromeda as two point masses with a combined mass of 2.3 trillion suns, and works out where the first close pass lands. Drag the sideways speed and watch two numbers behave completely differently.
A clean miss on the first pass.
Modelled, not measured. Two point masses, no dark-matter drag, no third galaxy. It gives the first pass only and cannot tell you whether the two eventually merge.
The thing neither headline mentioned
Run the slider from one end of the 2026 measurement’s error bar to the other and the closest approach moves from about 15 kiloparsecs to about 157 — and out to 183 at the value the 2025 team used. That is a factor of twelve. At the low end Andromeda ploughs straight through the Milky Way’s disc, which is only about 30 kiloparsecs across. At the high end it never comes closer than six disc widths.
Now watch the clock. Over that same range, the time to closest approach moves from 3.89 to 4.64 billion years. A spread of about nineteen per cent, against eleven hundred per cent for the distance.
The reason is in the arithmetic. The timing is set mostly by how fast Andromeda is falling toward us, and that is the number we know to about one part in a hundred. The outcome is set by the sideways drift, and that is the number whose error bar is larger than half its own value.
Astronomy has always known when. What it never knew was whether. The famous 4.5 billion years was never the shaky part of that textbook sentence. It was attached to the wrong half of it.
What would actually settle it
Both teams name the same fix, and it is a measurement requirement rather than a new idea: Andromeda’s drift needs to be pinned to around two millionths of an arcsecond per year. In velocity that is about 7.2 kilometres per second, roughly three and a third times sharper than today. Do that, and the plausible range of merger probabilities narrows from about thirty-five percentage points to ten.
That measurement is closer than it sounds. Gaia’s fourth data release — the first full release of the mission, covering five and a half years of observations — is scheduled for 2 December 2026. It is the obvious next test.
Which leaves a genuinely open question, and a short window in which to have an opinion about it. If you want to be honest with yourself, write down the number you expect before the data lands.
Classroom packet
Cosmic Collision Course
The 90-minute version of this article, built for Grades 10–12 and first-year undergraduates. Students take both teams’ published figures and compute the encounter themselves.
- They derive the factor-of-twelve result above from scratch, using conservation of energy and angular momentum.
- They solve the inverse problem: what measurement precision would actually settle the question?
- They find out why adding a small galaxy to the simulation lowers the odds of a collision.
- Full answer key with every step worked, plus a 16-point rubric for the closing argument.
- A documented two-period split and a 50-minute short path. Every figure verified in code.
Left: the classroom packet on Teachers Pay Teachers. Right: Andromeda spans about three degrees of sky, six times the width of the full Moon, and this is listed as a wide-field astronomy binocular — the kind that takes in the whole galaxy at once instead of a piece of it — an Amazon affiliate link. As an Amazon Associate this site earns from qualifying purchases, at no additional cost to you.
Sources
The 50 per cent result, the four-galaxy model, the fifty thousand simulations and the quoted sentence come from T. Sawala, J. Delhomelle, A. J. Deason, C. S. Frenk, J. Häkkinen, P. H. Johansson, A. Keitaanranta, A. Rawlings and R. Wright, No certainty of a Milky Way–Andromeda collision, Nature Astronomy, 2 June 2025 (open access). The 90 per cent result, the sideways speed of 46.7 ± 24.0 kilometres per second, the merger time of 6.5 billion years and the two-millionths-of-an-arcsecond target come from H. Wu, Y. Huang, H. Zhang, G. Sun and S. Shao, The Fate of the Milky Way–Andromeda System: To Merge or Not?, arXiv:2603.22863, submitted 24 March 2026 — a preprint, not yet peer reviewed. Andromeda’s distance modulus of 24.407 and the galaxy masses are the values both papers adopt. The Gaia Data Release 4 date of 2 December 2026 is from the European Space Agency’s own announcement of 7 May 2026.
Modelled, not measured: every number the calculator returns. It is a two-body, point-mass approximation with no dynamical friction and no third galaxy, so it gives the first closest approach and nothing beyond it. The published merger probabilities come from full four-body simulations and cannot be reproduced by the calculator above. The 20-kiloparsec merger threshold is a convention both research teams chose, not a measurement.