Neptune's Missing Aurora Was Hiding in the Tropics

For thirty-six years Neptune held out.

In August 1989 Voyager 2 flew past at about 5,000 kilometres above the cloud tops, close enough to feel the magnetic field rather than infer it from far off, and in the few hours it had the spacecraft found a magnetosphere, timed the radio bursts, mapped a field that made very little sense, and picked up faint ultraviolet glows that several people at the time were willing to call auroral [4]. Then it went on out of the Solar System. Nobody has been back.

What was left behind had a specific shape. Every other giant planet in our system shines in the infrared light of H₃⁺, a molecule built from three hydrogen nuclei that share two electrons and bend themselves into a triangle. No simpler polyatomic ion exists. It forms when something knocks an electron off molecular hydrogen high in an atmosphere, and it radiates a set of infrared lines near 3.3 to 4.0 micrometres that are sharp and well catalogued. Jupiter gave it up in 1989 [7]. Saturn four years later [8], with Uranus turning up in the same season [9]. In each case the emission mapped out the magnetic geometry of the planet like ink on a plate, and a whole subfield grew out of that, one that reads a giant planet's field by photographing the glow of its upper atmosphere [10, 11].

Neptune said nothing.

Not for want of asking. A Keck run in 2009 came back with an upper limit [5]. A very long stare with the NASA Infrared Telescope Facility, described by its own authors as a deep burn, came back with a tighter one in 2017 [6]. The models of the day put the ion well above both limits. Nothing appeared at all.

Then JWST looked at Neptune for about an hour on 22 June 2023. There it was [1].

Spend a moment on the molecule. Everything downstream depends on how it behaves.

You have an upper atmosphere made almost entirely of molecular hydrogen. Some energy source, sunlight or a precipitating electron, ionises an H₂ molecule. The resulting H₂⁺ meets another H₂ almost at once and swaps a proton away, and what you have left is H₃⁺ with a spare hydrogen atom beside it. The reaction is fast and it is nearly the only thing that happens, so the quantity of H₃⁺ in a giant planet's thermosphere is a fairly direct record of how much ionising energy is being delivered there.

Now the glowing. H₃⁺ is a floppy triangle, and one of the ways it can wobble is a bending motion in which the triangle squashes and opens. Physicists call that mode ν₂, and it costs about 2,521 wavenumbers of energy to excite, which works out to a photon near 4 micrometres, a long way down the spectrum from anything an eye would call light. The ion cannot stay excited, so it falls back down and emits, and the lines it emits sit in a part of the infrared where hydrogen and helium and methane are all reasonably quiet. Hence the ion's usefulness. A lamp in a dark room.

The catch lies in how the lamp is powered, and this whole article turns on it. Nothing pumps those bending states directly. They are populated thermally, which means the fraction of ions sitting in an emitting state at any moment is governed by a Boltzmann factor, \(e^{-E/k_\mathrm{B}T}\), where \(E\) is the energy of the state and \(T\) is the temperature of the gas, and when \(E/k_\mathrm{B}\) sits near 3,600 kelvin while the gas sits at a few hundred, that exponential does something violent. Halve the temperature and you do not halve the brightness. You divide it by a number with three digits in it.

A thermometer you read by how brightly it shines, on a scale where a modest chill turns the lamp off altogether.

Hold that thought. It comes back in section 7, with numbers.

Here comes the part we think is genuinely lovely. Pure geometry, every step of it, so you never have to trust us.

A planetary magnetic field, to a first approximation, is a dipole, the field of a bar magnet sitting at the centre of the planet, so the field lines leave near one magnetic pole, arch far out into space, and come back down near the other. Charged particles in the surrounding magnetosphere are stuck to those lines, because a charged particle spirals along a magnetic field line and finds it very hard to cross one, and when those particles slide down a line and strike the atmosphere they ionise it, and the ionised patch glows. That patch is the aurora. Its position is set by one thing alone: where the field lines land.

On Earth the bar magnet is roughly aligned with the spin axis, so the field lines land in a ring around the geographic pole and the aurora is a crown. Earth's magnetic axis is off by about 11 degrees, enough to pull a compass needle away from true north and nowhere near enough to lift the crown off the Arctic. Jupiter's by about 10. Saturn's by less than one.

Neptune's is off by about 47.

And not at the centre either. Voyager 2's magnetometer team found a field that could only be fitted by putting the effective dipole roughly 0.55 Neptune radii away from the planet's middle, which is over half way to the surface [2, 3] and one of the strangest arrangements in the Solar System. Its surface strength in one hemisphere runs many times what it is in the other, and the fit needs quadrupole and octupole terms almost as large as the dipole itself.

Take just the 47 degrees, though, and ask the simplest possible question. If the ring of field line footpoints stays where it is relative to the magnet, and you tip the magnet by 47 degrees, where does the ring go?

It goes to the tropics. Not metaphorically. A ring that sat 13 degrees from the pole ends up centred 43 degrees from it, which on Earth would put the northern lights over Marseille and Chicago, and the southern ones over the middle of the Tasman Sea, the whole crown lifted off the axis the planet spins about and hung instead on an axis its weather knows nothing of.

This section is arithmetic. One model, five inputs, one latitude printed at the end.

The field of a dipole with moment direction \(\hat{m}\), evaluated at a point \(\vec{s}\) measured from the magnet, points along

We never need the strength, only the direction, so the magnetic moment cancels out of the whole calculation. To trace a field line you start at a point and step along \(\hat{B}\). Our code uses fourth-order Runge-Kutta with a step that shrinks as the line approaches the planet, and stops when the path first crosses a sphere of one Neptune radius, which is where we have agreed to put the cloud tops. The last step is refined by bisection to a billionth of a radius.

Starting points sit on the dipole's magnetic equator at a distance \(L\), measured in planetary radii, and the letter is standard: an \(L\) shell gathers every field line that crosses the magnetic equator \(L\) radii out. Seed 72 points around a circle at fixed \(L\) and trace each one down. The footpoints draw a closed curve. That curve is the oval.

For a centred dipole the answer is also available in closed form, \(\cos^2\lambda = 1/L\), so we checked the tracer against it before trusting it, and got agreement to four decimal places at every \(L\) we tried. Good.

The paper maps the southern emission onto shells running from under 3 out to 20 planetary radii [1], and we take \(L = 20\) for the headline because it is the outer edge of that range and because it makes the narrowest, most falsifiable prediction of any shell in it. At \(L = 20\) the ring sits 12.92 degrees from the magnetic pole.

Tilt zero: the ring lies at 77.08 degrees south, all the way round.

Tilt 46.9 degrees: the ring runs from 56.02 degrees south to 30.18 degrees south.

Observed: 30 to 60 degrees south [1].

The overlap is 25.8 degrees of the observed 30-degree band, or 86% of it, and the centre of our band is 42.4 degrees south against 45 for theirs. The miss is 2.6 degrees.

Two things about that agreement should be said before anybody gets excited. The comparison runs on latitude alone, and it has to: Neptune's rotation period is known badly enough that nobody can say what longitude was facing JWST during the observation, a point the paper makes about its own analysis [1]. And 86% is a percentage that flatters us, because a model predicting a band twice as wide would score higher still, which is why the 2.6-degree miss is the number to keep.

Something happens on the way from tilt zero to tilt forty-seven that no table of numbers will show you, and Figure 3 exists for it, because what happens is a change of shape on the page rather than a change of value in a column.

The oval never changes shape. Around the magnetic pole it stays a ring of constant magnetic colatitude, and tipping the magnet rotates that ring rigidly, so it remains a circle on the sphere throughout. What changes is the relationship between that circle and the lines of latitude, which are drawn around a different axis entirely, and at zero tilt the two axes coincide and the circle is a line of latitude, flat and featureless, one number for the whole circuit. Tip the magnet five degrees and the ring starts to cut across the parallels. Tip it to forty-seven and the ring, plotted on a map, has become a tall closed loop that climbs from 30 degrees south up to 56 and back down again inside about 35 degrees of longitude, so that a single meridian can cut the aurora twice, once on the way up and once on the way back down.

This matters observationally in a way we did not appreciate until we plotted it. An instrument staring at a narrow strip of latitude will catch a tilted oval only at the longitudes where the loop happens to cross that strip. JWST's integral field unit had no such problem, because it takes a spectrum at every point in a small field of view at once and builds a map rather than a line, and part of why this observation worked and earlier ones did not lies right there, in the shape of the detector rather than the size of the mirror.

Look at the 20-degree curve for a moment. Its lowest point is 57 degrees south and its highest is 83, so at that tilt the oval is already spilling out of the polar region without yet arriving anywhere you would call temperate. Neptune's aurora spent the twentieth century where nobody thought to point.

Session 1 · the wrong ring

First attempt: draw the oval around the magnetic pole, read off the latitude, go home happy, a plan that lasted about four minutes. Somebody asked which magnetic pole. Fair question. A 47-degree tilt gives you two of them, neither anywhere near the geographic poles, and the southern magnetic pole in our frame sits at 43.1 degrees south, which is a latitude and not a pole in any sense a map would recognise.

Session 2 · the offset fight

Longest argument of the term. The offset is 0.55 radii. Fine. In which direction?

Nobody in the room could find a source that gave the direction in a form we could check against a coordinate system we understood, and we were not willing to make one up, which is how an afternoon's question turned into three weeks of reading. So we swept it. Ten directions, from the southern magnetic pole round to the northern one.

The result surprised us. An offset lying along the dipole axis leaves the whole arrangement symmetric about that axis, so it slides the oval but cannot deform it, and it keeps the oval within two degrees of the observed band centre. An offset lying across the axis is a disaster: it drags the footpoints to 74 degrees south, twenty-nine degrees from the emission. So the observation rules out most of the directions we could not choose between.

We did not put that in. It came out. That was a good afternoon.

Session 3 · how lucky are we

Two thousand seeded draws. Tilt normal about 46.9 with a degree of spread, \(L\) uniform between 15 and 25, offset magnitude uniform from zero to 0.55, offset direction uniform over the whole half-circle of possibilities, azimuth around the oval uniform. Then trace every single one.

With the offset off, 94.2% of the traced footpoints land inside the observed band. Median 41.8 degrees south. Turn the offset on, let its direction be anything, and that falls to 70.3%. Median 51.9 south. Both runs are in the output file. The honest reading: the tilt does the predicting, and the offset is the error bar we cannot close from a school computer.

Back to the lamp.

Voyager 2's ultraviolet spectrometer, watching a star pass behind Neptune's atmosphere in 1989, gave an exospheric temperature of 750 ± 150 kelvin [1, 4]. JWST's spectrum, fitted for temperature and column density together, gives 358 ± 8 kelvin [1]. The two instruments sampled somewhat different heights, a caveat the paper itself raises. They also sit a factor of two apart, and a factor of two in temperature is not a subtle thing to lose in a calibration.

So: how much fainter is a cold Neptune?

We built the calculation from the level structure up. The ion's energy levels are those of an oblate symmetric top, which is the standard textbook model for a flattened spinning molecule:

with \(J\) the total rotational quantum number and \(K\) its projection on the symmetry axis. Three identical protons impose nuclear spin statistics, so levels with \(K\) divisible by three get a statistical weight of 4 and the rest get 2, and some levels are forbidden outright. We sum every level to \(J = 22\) across the ground state and two vibrational states above it, build the partition function \(Q(T)\), and work out what fraction of the ions are in a state that can emit in the ν₂ band.

At 750 K the partition function is 265.7. At 358 K it is 84.3.

Watch the ratio in two pieces, because they fight each other. Cooling from 750 to 358 empties the emitting levels by a factor of 624, which is the Boltzmann factor doing what Boltzmann factors do. Cooling also crowds the remaining ions into fewer available states, which raises the fraction in any given one, and that hands back a factor of 3.15. The fight is not close.

Net: the emission per ion at 358 K is 0.51% of what it was at 750 K. A factor of 198.

A second consequence follows, and it explains why this ion matters to people who are not using it as a thermometer. The same ν₂ radiation is one of the main channels by which a giant planet's thermosphere gets rid of energy, so a layer of H₃⁺ is partly its own thermostat [14]. Heat it and it radiates harder and pulls itself back down. Cool it and it radiates far less, which ought to slow any further cooling. That damping is real, and it cuts against the observation, making a drop of several hundred kelvin in thirty-four years harder to arrange rather than easier, since the layer should have been fighting its own cooling the whole way down.

The paper, using a proper line list rather than our rigid-rotor approximation, says 0.8%, which is a factor of 125 [1]. We are out by 1.58, in the direction of predicting too steep a fall. For a model built from four spectroscopic constants and a sum over levels, landing within a factor of two of a professional calculation is roughly what we hoped for and slightly better than we expected.

One more number, because the error bar on 1989 is wide. If Neptune's thermosphere was really at 600 K back then, the fall is only a factor of 59. If it was at 900, the fall is 439. The qualitative story survives the whole range, which is the useful thing about an exponential.

Put the two halves together. The thirty-six years stop looking like bad luck.

Everyone who searched for H₃⁺ at Neptune before 2023 was working from a prediction built on a 750 K atmosphere, because that was the only temperature measurement in existence. On that assumption the ion should have been within reach of a 10-metre telescope. The ion was not there, and the field's explanation drifted toward exotic chemistry: perhaps hydrocarbons falling in from above were destroying it, perhaps metal ions delivered by infalling dust were wiping out the ionosphere before it could build up [17]. Those are real processes and they may well matter.

The simpler answer turns out to be that the atmosphere got cold.

To see how much that costs an observer, we built a sensitivity ladder, and we want to be very clear that it is the crudest thing in this article. It assumes background-limited photon counting and nothing else. Signal to noise goes as brightness times the square root of collecting area times integration time, divided by a penalty for the thermal background that ground-based observing at 3 to 4 micrometres suffers and space-based observing does not. One calibration point fixes the whole scale: JWST detected the emission, so we set its floor at half the brightness the paper measured. Everything else is then predicted rather than fitted.

Three of those four rows tell the story cleanly. Every ground-based attempt had the sensitivity to find the ion on a hot Neptune and none of them had the sensitivity to find it on a cold one. The fourth row is where our ladder falls over. The ladder claims Voyager's infrared interferometer should have caught H₃⁺ during the flyby, and no such detection was ever reported. The reason is a thing the ladder knows nothing about: IRIS had a spectral resolution coarse enough that the individual H₃⁺ lines would have been smeared into the continuum. A figure of merit built from mirror area and exposure time cannot see that. We are leaving the wrong answer in the table rather than quietly tuning it out, because a model that only ever agrees with you is not being asked hard enough questions.

And the brute-force version, for anyone who wants to know what Keck would have needed. Recovering a factor of 198 in brightness at fixed signal to noise costs a factor of 39,170 in area times time. On a 10-metre mirror that is about 9,800 nights of pure integration, which is twenty-seven years of pointing at Neptune and nothing else, on a telescope that has a queue.

Suppose you are the referee who does not want to believe any of this. Your job: find the version of the world where every measurement is correct and the conclusion still fails anyway. Here is the strongest case we can build. The first objection is genuinely strong.

Objection one: the map is built from a very thin signal. The paper is candid about this. Individual spaxels, the small square elements of the integral field unit's image, carry a signal to noise of about 3. To make a map at all, every spaxel is averaged with its eight neighbours [1]. Standard practice, and also a practice that can manufacture smooth structure out of noise, because neighbouring pixels stop being independent the moment you do it. A single one-hour observation, smoothed 3 by 3, showing a 1.7-fold column enhancement in one corner of a disc is not the same class of evidence as a bright, repeatedly imaged Jovian oval. The identification of that patch as auroral rests on it being localised rather than varying smoothly with solar illumination, which is a good argument, and on it coinciding with where the magnetic field model says an oval should be, an argument that borrows its strength from a field model fitted to one flyby in 1989.

Objection two, and this one is sharper than it looks: the longitude agreement may be a coincidence, and the authors say so themselves. Neptune's rotation period carries enough uncertainty that the rotational phase during the observation is formally unknown, which means the predicted oval could be anywhere in longitude. The paper notes that no arbitrary shift was needed to line the prediction up with the observation, and then adds that this may be a coincidence or a sign that the uncertainties are overstated [1]. When a result agrees with a prediction that had a free parameter capable of moving it anywhere, the agreement carries less weight than it appears to.

Objection three: the two temperatures were not measured the same way. Voyager's 750 K came from a stellar occultation in the ultraviolet, sampling the exosphere. JWST's 358 K comes from fitting H₃⁺ line ratios, which samples wherever the ion happens to live. If those are different altitudes in a region with a strong vertical temperature gradient, part of the factor of two is a difference of height rather than a change in time. The paper argues the comparison is fair. The comparison still runs between two instruments separated by thirty-four years with no overlapping calibration.

Objection four, which is ours about our own work. The agreement in section 4 depends on a choice we made out of ignorance. We ran a centred dipole because we could not verify the offset direction, and section 6 shows that a different choice of that direction moves our answer by twenty-nine degrees and destroys the match completely. We did not select the centred case because it agreed with the data. We selected it because it uses fewer unverified numbers, and it happens to agree. Anyone is entitled to think that convenient.

Now the replies, because a steelman nobody answers is a hedge.

On objection one, the paper's strongest defence is not the map at all. The detection of the ion itself carries the weight, a spectroscopic result across the P, Q and R branches rather than a morphological one, and nobody doubts it. The spatial structure is the weaker claim resting on top of a firm one.

On objection three, the authors do the obvious check and it holds: solar Lyman-alpha flux at Neptune was 8.6 mW m⁻² in 1989 and 9.6 in 2023, so the Sun was not quieter, and the change cannot be a solar cycle effect [1]. Nor can it be seasonal, since a Neptunian season runs about forty years and the change happened across thirty-four.

On objection two, the honest position is to treat the latitude agreement as evidence and the longitude agreement as decoration until somebody nails down the rotation period.

What survives is a firm detection with a well-supported temperature, and a spatial identification that is plausible and thin. Normal enough, for a first result.

Four things would move this from plausible to settled. We have put them roughly in order of how decisive they would be.

Observe it again, at a different rotational phase. If the enhancement is the footprint of a fixed magnetic structure, it should reappear at the same magnetic longitude and march across the disc as the planet turns. If it wanders, call it weather or noise.

Pin down the rotation period. Everything about the longitude comparison is hostage to it. A planet with no solid surface and a magnetic field that refuses to behave makes this a hard measurement, and it is the single number that would do the most work.

Get a second temperature by a second method. A stellar occultation in the ultraviolet, the same technique Voyager used, would compare like with like across thirty-six years and settle whether the cooling is real or partly an artefact of comparing different altitudes.

Explain the cooling. Nobody can. Not the Sun, not the season, and the paper says plainly that the cause remains unexplained [1]. Uranus shows a long decline in its own H₃⁺ temperature over three decades [16], which suggests whatever this is may be general to the ice giants rather than peculiar to one of them. The mechanisms people reach for all involve the movement of auroral energy: at Jupiter the polar aurora turns out to warm the entire globe rather than staying where it is generated [15], so a change in how that heat gets spread around is at least the right shape of explanation. Ground-based imaging has also caught Neptune's stratosphere swinging in brightness on timescales far shorter than its forty-year season [18], which at minimum tells you the atmosphere there is capable of changing quickly. A clue, then, and not an answer.

A fifth, quietly: point JWST at Uranus the same way. Uranus tilts 59 degrees, further than Neptune, and on our own model it should carry a band from 44 to 18 degrees from its equator. That work is beginning.

Go back to the picture in Figure 1 and look at the small planets on the right.

Three of them wear something like a crown. Saturn's ring of field lines is drawn as a single line because at its tilt the oval sits inside a degree of the pole. Jupiter's is a little wider. On both of those worlds you could photograph the aurora from directly over the pole and see the whole thing at once, which is exactly what Hubble and Juno have been doing for years.

Neptune's has slid down the side of the planet.

No deep physics lives in that slide. The same dipole, the same trapped particles, the same funnelling down the same kind of field line, and the same glow at the bottom. Somewhere in the ice giant's interior the dynamo that generates the field is running in a thin shell rather than a deep convecting core, and the result is a magnetic axis that points 47 degrees away from the spin axis and sits half a radius off centre, about as far from a tidy bar magnet at the middle of a planet as a dipole can wander and still be called one. Everything else follows by geometry, and the geometry is honest enough that a school club with a laptop can reproduce the observed latitudes to within three degrees.

What we cannot reproduce, and what nobody can yet explain, is the other half of the paper. An atmosphere lost several hundred kelvin in thirty-four years. Not over a season, because the season is longer than that. Not with the solar cycle, because the Sun was as bright at the end as at the start. Something is regulating the energy budget of Neptune's upper atmosphere on a timescale nobody had budgeted for, and the only reason we know is that the loss of heat put the lamp out and it took a telescope above the atmosphere to notice.

Voyager 2 went past in 1989 and took the only close look anyone has had. It saw the magnetic field, sketched the tilt, hinted at ultraviolet glows it could not resolve, and left, carrying instruments designed for Jupiter and already twelve years old when it arrived. Nothing has been back since, and nothing is scheduled. The aurora it half-saw turns out to have been sitting all along where a planet's tropics would be, waiting thirty-six years for an instrument cold enough and quiet enough to find a lamp that had gone dim, while the geometry that put it there sat in Voyager's own magnetometer data the whole time.

References