511 keV: thunderstorms make antimatter and fire it into space

Here's a number from a gamma-ray detector: 511.

It's an energy, in keV, a unit made for single particles. And it's what you'd get by turning the mass of one electron completely into energy.

Every electron has an antimatter twin, the positron: the same mass, the opposite electric charge. When the two meet, both disappear, and what's left is two flashes of light, each carrying that same energy: gamma rays, light far too energetic for eyes to see.

So a pile of gamma rays at 511 keV is a receipt: proof that antimatter met matter.

This story follows that receipt above, inside and below thunderstorms: how a cloud becomes a particle accelerator, where the line has been caught, how sure anyone is, and what a flash would do to a plane.

Guess first

Where do storms get their antimatter?

See the answerWhy

Storms make positrons the ordinary way, from gamma rays hitting atoms. Pulling antimatter out of empty space would take a field about 5 trillion times the avalanche threshold (chapter 4).

Chapter 1 Pre-flight briefing

The energy in an electron

An electronvolt is the energy one electron picks up crossing one volt. A keV is a thousand of those; an MeV is a million, or 1,000 keV.

Turn all of an electron's mass into energy and you get 511 keV. A positron has the same mass, so it carries the same.

When a positron meets an electron that's barely moving, both vanish. A single flash of light can't carry energy away without also carrying motion, and the pair had almost none. So the energy leaves as two gamma rays of 511 keV, flying apart in opposite directions. (Sometimes the two cling together for a moment first, as positronium; one kind of it ends in three weaker gamma rays instead. In air, two is the usual outcome.)

It also runs backwards, and that's how storms make positrons. A gamma ray with more than twice 511 keV, about 1 MeV, passing close to an atom's nucleus, can turn into an electron and a positron.

Both ways

Meet. Both vanish into two gamma rays of 511 keV, flying apart.

Backwards. A gamma ray above about 1 MeV, passing close to a nucleus, can turn into an electron and a positron.

A detector sorts light by energy, and light that all has one energy piles up as a sharp spike: a line. A spike at 511 keV says positrons were there, though near the ground it needs care (chapter 9). It doesn't say how many.

Chapter 2 Sea level

A guess from 1925

In 1925 C. T. R. Wilson, who had invented the cloud chamber (a box that shows the tracks of fast particles), guessed that the strong electric fields in thunderclouds could speed fast electrons up to very high energies. (His 1927 Nobel Prize was for the cloud chamber; this idea waited decades for evidence.)

Air drags on a moving electron, which loses energy knocking electrons off molecules. But the faster it goes, the less time it spends near each one, so the drag falls as it speeds up, to a low point near 1 MeV. After that it barely grows.

So if a field pushes harder than the drag at that low point, an electron that's already fast gains more than it loses, and keeps gaining. It runs away.

A cyclist with a tailwind is different: their drag keeps rising with speed, so they always hit a top speed. Nothing like that holds the runaway back.

At sea level that takes about 218 kilovolts per metre (kV/m). A spark in dry air needs about 3,000 kV/m, more than ten times as much. A storm needn't make a spark, only beat the drag on electrons already fast, and cosmic rays (fast particles from space that crash into the air) fill the air with those.

Thinner air drags less. At 12 km, near airliner height, the air is about a quarter as dense, so a field there counts for about four times as much. That's what the next figure means by "scaled to sea-level air".

Chapter 3 Sea level to 12 km

One electron becomes thousands

In 1992 Alexander Gurevich, Gennady Milikh and Robert Roussel-Dupré showed that runaways multiply. Now and then one knocks an electron out of a molecule hard enough that it runs away too. One becomes two, two become four: an avalanche, started by fast electrons that cosmic rays leave in the air, or by ones that lightning itself makes.

Real electrons get knocked off course, which raises the push needed by about 30%: an avalanche needs about 284 kV/m at sea level, in Joseph Dwyer's 2003 calculation (others differ by several per cent). That's about ten times less than a spark. Thinner air needs less: about 206 kV/m at a mountain station 3,200 m up in Armenia.

At 350 kV/m, 1.23 times the threshold, the number of fast electrons doubles about every 68 m, so in 1 km one starting electron, a seed, becomes about 25,000. Just above the threshold it's far gentler: at 1.1 times, about 150 in a kilometre. At the threshold itself, growth stops. (This page's arithmetic, from a formula by Dwyer's group.)

Thin air needs a weaker push but more room: at 12 km the same 25,000-fold growth takes 3.9 km. Those numbers are rough; a 2026 computer study of mountain stations found slower growth.

The runaways average about 7 MeV, so they move at about 99.8% of light speed (this page's arithmetic). The average stays there because fresh slow ones keep joining.

It's like a snowball rolling downhill, except that snow never climbs back uphill to start new snowballs. Positrons do (next chapter).

Are storm fields that strong? Just. The strongest measured, scaled to sea-level air, rarely pass about 400 kV/m, and a balloon 5.77 km up in 2005 measured a field about 118% of the threshold there (by this page's sums; its team, with a different model of the air, says about 130%).

Fields at sea level, on a log scale

about 218 kV/mAn electron that's already fast runs away
about 284 kV/mAn avalanche (Dwyer's 2003 calculation)
about 400 kV/mThe strongest measured in storms, rarely more
about 3,000 kV/mA spark in dry air

Drag the push past the threshold and release a seed. The count stops near 100,000, and the amber dots running back up are the reason 👇

See how violently the growth depends on the push: from 1.1 to 1.23 times the threshold, a kilometre's count jumps from about 150 to about 25,000.

Chapter 4 Inside the cloud

Antimatter inside the machine

An avalanche shines.

When a runaway swerves past an atom's nucleus it throws off a gamma ray, "braking radiation", up to tens of MeV.
Those above about 1 MeV can make electron and positron pairs.
The new electron joins the avalanche. The positron, with the opposite charge, is pushed back up the field.
Back at the end where the avalanche began, it can knock loose fresh runaways, which start a whole new avalanche.
and round again

Joseph Dwyer called this relativistic feedback. In 2003 he argued it sets "a fundamental limit on electric fields in air": past a point, each avalanche seeds another, the discharge keeps itself going, and on his calculation the field stops growing. That's a calculation, not a measurement.

Feedback also caps one avalanche below about 100,000 electrons from one seed, for reasonable storm fields. That's an upper limit, not a switch. In 2015, aircraft data showed one strong glow (gamma rays a cloud gives off for seconds or minutes; chapter 8) already close to where feedback matters, with a growth of 4,500. So antimatter isn't just something the storm makes. It's part of the machine.

Could a storm pull antimatter out of empty space instead, as a strong enough field can (Nº 5)? Only with a field about 5 trillion times the avalanche threshold. Storms make positrons the ordinary way, from gamma rays hitting atoms.

Chapter 5 Orbit, looking down

Flashes from below

The first evidence from space that storms speed up electrons came from an instrument pointed the wrong way. NASA's Compton Gamma Ray Observatory carried a detector built to catch bursts of gamma rays from deep space. In 1994 Gerald Fishman and colleagues reported it was also catching brief flashes of gamma rays coming up from thunderstorms.

These terrestrial gamma-ray flashes last well under a thousandth of a second. Their gamma rays reach 20 MeV (NASA's RHESSI satellite, 2005) and about 100 MeV (as Italy's AGILE team reported in 2011). Their spread of energies is what an avalanche makes, and they leave in a broad upward cone.

The flashes satellites see usually start about 12 km up, near airliner height, and need about 100 million billion fast electrons, or more.

Where do all those electrons come from? Far too many for avalanches started by cosmic rays' fast electrons: with each one held below about 100,000, it would take a storm region at least 800 km across. The leading ideas are runaways made at the tips of lightning leaders (the channels of hot air a lightning flash grows along), or feedback. As Dwyer wrote in 2025, "it has been challenging to tell which mechanism is correct when lightning is present".

Chapter 6 Orbit, over Egypt

A beam from over the horizon

NASA's Fermi telescope sees these flashes too, mostly from storms not far below it. A few are different: they last longer, and can come from far away. They aren't gamma rays by then. A flash's gamma rays knock electrons out of the thin air above the storm and make electron and positron pairs, and the charged particles that escape are steered along Earth's magnetic field, sometimes for thousands of kilometres.

On 14 December 2009 Fermi was over Egypt when one hit it. The storm was over Zambia, about 4,500 km away and below the horizon. The beam travelled on until the magnetic field turned it round, then hit Fermi again 23 thousandths of a second later.

14 December 2009, not to scale

The storm, over Zambia.
Its beam follows Earth's magnetic field, for thousands of kilometres.
Fermi, over Egypt: about 4,500 km away, below the horizon.
Turned round by the field, the beam hit Fermi again 23 thousandths of a second later.

In 2011 Michael Briggs and colleagues reported a strong 511 keV line in the three brightest beams like this. So they held many positrons, which met electrons in the spacecraft itself. The team thinks most likely every gamma-ray flash from a storm does this; that's their view.

Fermi didn't see antimatter hovering over a storm. It saw a beam the storm had thrown into space.

Chapter 7 Into the storm

Inside the cloud

Aircraft can get closer, sometimes closer than planned.

21 Aug 200914.1 kmResearch jet511 keV × 2 bursts

On 21 August 2009 a research jet flying at 14.1 km near Brunswick, on the coast of Georgia in the US, strayed into the top of a storm. Its instrument saw two bursts of 511 keV gamma rays, each about 12 times the usual level and a fifth of a second long. The team could match them with positron clouds more than a kilometre across, but how those would form is unclear, and the plane itself may have played a part.

Jan 201612 kmAirbus test plane511 keV × 10

In January 2016 an Airbus test plane flew through a storm at 12 km over northern Australia and saw the line ten times. Its team traced the positrons' meetings to the plane's own surroundings: right beside it, or in its body.

Neither shows antimatter clouds waiting for airliners. A 511 keV gamma ray tells you where a positron met an electron, not where it was made.

Chapter 8 20 km, above the storms

A boiling pot

A flash is over in a blink, but thunderclouds can also glow: give off gamma rays for seconds to minutes, sometimes cut off by lightning.

Between 3 and 29 July 2023 NASA's high-flying ER-2 made ten flights from Tampa, 20 km up, over the Gulf of Mexico, Central America and the Caribbean: the ALOFT campaign. The results came out in Nature in October 2024.

Tropical thunderclouds over the sea and coasts, it turned out, commonly glow, for hours, over up to a few thousand square kilometres, bubbling up and dying away within seconds like a "boiling pot" (the team's image of the pattern, not of heat). The plane recorded more than 500 separate glows (608 in the paper's flight table), on nine of its ten flights. One flight, over the Bay of Campeche on 24 July, found 392 of them: nearly two in three (this page's arithmetic). One flight found none.

Flickering flashes

The team also found a new kind of flash. A flickering gamma-ray flash starts as a glow, then rises fast and flickers, for 20 to 250 thousandths of a second. There were 24, on five flights, and the team suggests they're the missing link between glows and the brief flashes. In 17 of the 24, a sharp burst of radio waves followed within tens of thousandths of a second, and then lightning: a sequence in time (chapter 11).

Of 96 brief flashes the flights recorded, only 3 or 4 were bright enough to be seen from space. "At least in the Caribbean and Central America during summer", the team concludes, storms make almost a hundred times more gamma-ray flashes than satellites can detect. Two computer models can grow glows into flashes, but neither yet shows that it happens in nature.

96 brief flashes the flights recorded

bright enough to be seen from spaceperhaps (3 or 4)too faint to be seen from space

Press Fly the campaign 👇

Look how lumpy it is: one flight, nearly two-thirds of the glows; another, none. And it's about tropical oceans and coasts in one summer, not every storm on Earth.

Chapter 9 The ground

A gamma-ray rainbow

A detector never sees positrons, only light, sorted by energy like a rainbow. From a storm, most of it is a smooth slope: lots of gamma rays at low energies, fewer and fewer out to tens of MeV. About 41% of those above 100 keV could make a positron (this page's arithmetic), though most bounce off electrons in the air first and lose energy.

The positrons that are made meet electrons and add a sharp spike at 511 keV.

It's a fingerprint, not a count.

Any band of energies around 511 keV also catches light from the smooth slope, and near the ground it catches natural radioactivity too, including a line of thallium-208 at 510.8 keV, too close to 511 for any detector to tell apart.

A blurry detector smears the spike out; a sharp one keeps it narrow.

13 Jan 2012Sea of Japan coastSpike at 508 keV

The standard example from the ground is from Japan. On 13 January 2012, detectors on the Sea of Japan coast caught about a minute of gamma rays from a winter thundercloud, with a spike at 508 keV that the team identified as the line from positrons meeting electrons: about 10 in every 100 of the gamma rays they caught. The team suggests a beam of positrons heading down made it; that's a suggestion, not an established result. As far as we can find, it's the earliest such report from the ground in a journal that had other scientists check it.

Feb 2017Japan511 keV for about a minute

Lightning can make positrons another way. In February 2017, in Japan, lightning's gamma rays knocked neutrons out of nitrogen in the air, leaving radioactive nitrogen-13, which gives off positrons, and the 511 keV line kept going for about a minute. Its team calls lightning the second known natural way, after cosmic rays (Nº 12), to make atoms such as carbon-14. How much it makes, this page doesn't say.

Jul 2024Mount Etna, 1,741 m511 keV count × 5

On Mount Etna in July 2024, during one burst in a storm, the 511 keV count 1,741 m up jumped fivefold.

Press Let positrons appear, then try a sharper detector 👇

Slide the spike down to nothing: the 511 keV window (the narrow band of energies around 511 keV) still catches about 3 gamma rays in every 100, all from the smooth slope. A window count is not a positron count.

Chapter 10 3,200 m up A preprint

Mount Aragats, May 2026

The newest result from the ground comes from a station 3,200 m up Mount Aragats in Armenia. In 2024 its group reported, in a journal, a rise in the 511 keV line during a storm, and proposed that charge low in the cloud pushes positrons down.

The new work is a preprint: a paper posted online before any journal has checked it. Chilingarian and Sargsyan submitted it on 27 June 2026, and it was posted in July. As of 25 October 2026 it isn't in a journal.

It reports three events on 16 and 17 May 2026. In each, the count in a narrow energy window around 511 keV rose: by about 15%, 35% and 24%.

The catch: during storms, gamma rays from radon's decay products also rise near the ground (radon is a radioactive gas from the ground), and some bounce down into the 511 keV window. So the window rises with the radon even without positrons. The team's radon window rose too, by about 19%, 30% and 11%.

Two windows, three storm events (a preprint)

How much each window rose in the three events of 16 and 17 May 2026
Event	511 keV window	Radon window
16 May	about +15%	about +19%
17 May	about +35%	about +30%
17 May, later	about +24%	about +11%

Each window starts from its own count, so the two columns can't be subtracted (this page's reading).

What's new is timing. On 17 May an ordinary burst of radiation peaked at 14:14 UT (Universal Time, the world's standard clock), and the 511 keV window peaked 12 minutes later. The team argues that a small pair of opposite charges, between a positive pocket at the cloud's base and the ground, pushed positrons down. That's their idea, and as far as we can find nobody else has tested it.

It has no separated spike, no number for how easily chance alone could have made the rises, no simulation of the positrons' path, and no word on thallium-208's line.

It's tempting to subtract, 35 minus 30, and call the rest positrons. You can't: the two windows start from different counts, so their percentage rises don't subtract (this page's reading). What it shows is a rise of up to about a third in the 511 keV count. How much is antimatter, and where it came from, it doesn't settle. Rises like these at the ground, here and on Etna, are local changes in one narrow band of energies, and harmless.

Two teams have proposed that storms push positrons down; neither idea is established. The rainbow's Aragats setting shows the three events beside the radon.

Chapter 11 Cloud to ground

Who lights the spark?

All this happens in fields too weak for a spark, which raises the oldest puzzle: how does lightning start? The strongest fields measured, rarely above about 400 kV/m scaled to sea-level air, are about 7.5 times below what a spark needs in dry air, and only about a third of what it needs even when raindrops and hail make sparking easier. Three main ideas compete, each with evidence against it.

Cosmic rays as the trigger. Alexander Gurevich and Kirill Zybin proposed that the shower of particles from one cosmic ray, with the avalanches, starts the lightning flash. Computer models found the avalanches wouldn't make the air carry enough current to start lightning. It hasn't been ruled out, but nobody has shown how it would work.
Ordinary breakdown. In 2016 William Rison and colleagues, listening to storms' radio waves, found that many lightning flashes begin with a fast burst of ordinary breakdown (the way an everyday spark starts), which seems to need no runaways. They suggested that all flashes might start this way. One later study found that most of the 31 flashes it imaged began without it, and another found that a quarter of these bursts started no flash at all.
Feedback. Victor Pasko and colleagues argue that light from avalanches freeing fresh electrons explains both the brief gamma-ray flashes and how lightning starts. And 17 of the 24 flickering flashes in 2023 were followed by lightning: a sequence, not proof of cause.

The ideas overlap: cosmic rays supply the first electrons in all of them, and the argument is about what turns those into lightning. As of October 2026, how lightning starts is still an open question.

Chapter 12 Cruising height

A plane in the wrong cloud

The flashes satellites see start about 12 km up, close to where airliners cruise, about 10 to 12.5 km. Every dose here is a calculation, not a measurement.

Inside the electron beam, the flash's runaway electrons would give the biggest doses, in under a thousandth of a second. They depend mostly on how wide the beam is, which nobody knows.

One model, from 2010, gives about 0.7 sieverts for a beam 50 m wide, a tenth of a sievert at 140 m, and under a millisievert for beams wider than about 1.4 km. Another, from 2021, gives more: from about 30 millisieverts for a beam 2 km wide to several sieverts for the narrowest, and more in a stronger field.

At the same width the two differ by about 60 to 70 times, and a tenth of a sievert is about 3,700 transatlantic flights (this page's arithmetic).

Outside, in the cone of gamma rays, the dose is far weaker. A published calculation for a bright flash gives at most about 90 microsieverts, 1 km above it: below every yearly limit, and too small, its team says, to count in flight-dose calculations.

This page's own rough model gives about 65 microsieverts 1 km above a flash of the brightness satellites usually see, and about 2.2 microsieverts at 3 km. For the published flash it gives about 2.4 times more than the published calculation, which stands where the two differ.

The dose ladder, on a log scale

A London to New York flight≈ 27 microsieverts
In the cone, 1 km above a bright flash (published, at most)≈ 90 microsieverts
The public's recommended limit, a year1 millisievert
Aircrew, a year, on average≈ 3 millisieverts
Inside an electron beam 140 m wide (one model, 2010)0.1 sievert
Inside an electron beam 50 m wide (the same model)≈ 0.7 sieverts

Each fifth of the width is ten times more. The flash doses are calculations, not measurements.

A plane 1 km above a 12 km flash is at 13 km, above nearly every airliner. To be in the cone, an airliner needs a flash below its cruising height: with the flash at 10.3 km and the plane 1 km above it, the model gives about 52 microsieverts from a typical flash. The figure's very bright flash, squeezed into its narrowest cone, with the plane 1 km above it and the flash at 12 km, reads about 1.4 millisieverts, 140% of a year's public limit, in this page's model, which runs high.

Drag the plane up the cone, then try the electron beam 👇

How likely is any of this? A 2023 study of real flight routes found fewer than one plane every two years would fly into one of these electron beams, counting all the world's flights. That's an upper limit, and the team thinks it likely no aircraft ever has.

Chapter 13 Landed: the debrief

The receipt

Guess first

A 511 keV flash reaches your detector. Where was its positron made?

See the answerWhy

A 511 keV gamma ray tells you where a positron met an electron, not where it was made.

A 511 keV line is a receipt. It proves a positron existed and met an electron. Like a receipt, it records the deal, not where the money came from: a gamma ray in the cloud, a beam carried along Earth's magnetic field from a storm 4,500 km away, perhaps a push towards the ground, or nitrogen that lightning made radioactive.

1925C. T. R. Wilson guessed in 1925 that thunderclouds could speed up electrons.
1992Gurevich and colleagues showed in 1992 how they multiply into avalanches.
2003Joseph Dwyer worked out in 2003 how positrons could feed the avalanche back on itself.
2011Fermi caught the positron beams in space (reported in 2011).
2024And the 2023 flights, published in 2024, showed how common and restless the glows are.

Storms don't make antimatter from nothing; they make positrons from gamma rays hitting atoms. Still open, as of 25 October 2026: how the brief flashes get their electrons, how lightning starts, and whether storms push positrons to the ground.

Let's add it all up.

🎫 A thunderstorm, itemised

One electron's mass, as energyCODATA511 keV

Avalanche threshold, sea levelDwyer 2003284 kV/m

An ordinary spark needsdry airabout 10× more

Runaway electrons' speedthis page's arithmetic99.8% of light

Fermi's storm, 2009over the horizon4,500 km away

Glows, July 20239 of 10 flights500+

Flickering flashes, then lightning2023 flights17 of 24

Flashes bright enough for space2023 flights, tropics3 or 4 of 96

Aragats 511 window, May 2026a preprintup to +35%

Gamma rays, 1 km above a bright flashpublished model≈ 90 microsieverts

What the line proves511 keVa positron met an electron

storms accelerate ✦ gamma rays make positrons ✦ 511 keV is the receipt

A storm top over Zambia.

Electrons racing along its field, more and more of them.

A gamma ray, passing an atom, becoming a pair.

A positron riding Earth's magnetic field to a spacecraft 4,500 km away,

meeting an electron there,

and two gamma rays flying apart.

511 keV is the energy in one electron's mass, and thunderstorms keep writing it across the sky.

Part IThe line and the machine

Chapter 1511 keV

On 14 December 2009 NASA's Fermi Gamma-ray Space Telescope, passing over Egypt, recorded a burst whose spectrum carried a sharp line at 511 keV. A detector sorts the photons it catches (the particles of light) by energy into a spectrum, and photons that all carry the same energy pile up in it as a narrow spike: a line. The storm that made this burst was not beneath the spacecraft: it was over Zambia, about 4,500 km (NASA's 2,800 miles; this page's conversion) to the south and below the spacecraft's horizon, and what reached Fermi was a beam of charged particles that had travelled along Earth's magnetic field.^1,2 The line said that the beam contained positrons: a thunderstorm had made antimatter and launched it into space. Chapter 6 tells that story; this one is about the number.

An electron's rest energy, the energy locked up in its mass, is the same for its antimatter twin, the positron, which has the same mass and the opposite charge:³

m_ec² = 510.998 950 69(16) keV ≈ 511 keVCODATA 2022; the (16) is the uncertainty in the last two digits. The CODATA electron mass times c² is 8.187 × 10⁻¹⁴ J, or 510.999 keV (this page's arithmetic); 511 keV is 0.511 MeV.

When a positron and an electron meet nearly at rest, both vanish. They have almost no momentum between them, and a single photon cannot carry energy without carrying momentum, so the energy leaves as two photons flying apart back to back:

e⁺ + e⁻ → γ + γ, E_γ = m_ec² = 511 keV eachFor a pair annihilating at rest, directly or through the short-lived form of positronium (below). The total, 1.022 MeV, is the whole of the pair's mass.

Each pair turns all of its mass into light. Nº 0's efficiency scale counts that energy against the antimatter's mass alone, which is why antimatter sits at 200% on it; and the same annihilation, in the first minutes after the Big Bang, heated the photons of the early universe (Nº 4).

A slow positron can first bind to an electron as positronium. In para-positronium the spins are opposite, and it decays into the same two 511 keV photons; in ortho-positronium they are parallel, and it decays into three photons sharing 1.022 MeV, spread over a continuum of energies below 511 keV. In air, not every positron forms positronium first, and collisions with molecules turn some of the ortho form into two-photon annihilation, so two photons is the usual outcome in a storm; we have found no measurement of the three-photon share that remains in storm air.

The reaction also runs backwards, which is how storms make their positrons. A gamma ray passing close to an atomic nucleus can turn into an electron and a positron if it carries at least the energy of the two masses:

E_γ > 2m_ec² = 1.022 MeVPair production needs a nucleus nearby to take up the recoil; near an electron the threshold is 4m_ec² = 2.044 MeV.

Thunderstorm gamma rays reach tens of MeV (Chapter 5), so many of them clear that bar. A sharp line at 511 keV in a gamma-ray detector is therefore a fingerprint of positrons: not a count of them, as Chapter 9 explains, but a sign that they were there. Near the ground the sign needs care, because natural radioactivity puts a line at almost the same energy (Chapter 10).

The line is not unique to storms. Hospital PET (positron emission tomography) scanners build their images from it, positron-emitting nuclei produce it, and astronomers have watched it coming from the direction of the Galactic Centre since balloon flights in 1970 and 1971;^4,5 in 1977 a germanium detector placed it at 510.7 ± 0.5 keV.⁶ In the Galaxy's thin gas essentially every positron forms positronium first, and positronium forms in its two-photon (para) state only one time in four, so most annihilations there give three photons and the 511 keV line comes from that quarter.^7,8 Where the Galaxy's positrons come from was still debated in a 2011 review,⁷ and a 2023 review was still calling it a puzzle.⁹

Chapter 21925

In 1925 C. T. R. Wilson, the physicist who had invented the cloud chamber, proposed that the strong electric fields of thunderclouds could accelerate fast electrons ("β-particles") to very high energies.¹⁰ His 1927 Nobel Prize was for the cloud chamber; this idea waited decades for its evidence.

The argument rests on how air resists a fast electron. The electron loses energy ionising the molecules it passes, and the faster it moves the less time it spends near each, so the friction falls as its energy rises: down to a minimum near 1 MeV, then roughly flat, rising only slowly at higher energies, as the 2012 review of the field by Joseph Dwyer, David Smith and Steven Cummer sets out.¹¹ A field E gives the electron an energy eE for every metre it travels along it, where e is the electron's charge. If that beats the minimum friction, an electron that is already fast gains more than it loses and keeps on gaining: it runs away. For an electron moving straight along the field with no scattering, gain and loss balance at

E_b = 2.18 × 10⁵ V/m × nThe break-even field (Dwyer, Smith and Cummer 2012), 218 kV/m at sea-level density; n is the air density relative to sea level. Elastic scattering, which knocks the electron off the field line, raises the field needed by about 30%, to the avalanche threshold of Chapter 3.

An ordinary spark in dry air needs about 3 × 10⁶ V/m (30 kV/cm) at sea level,¹² more than ten times as much. That gap is why Wilson's idea matters: a storm need only beat the friction on electrons that are already fast, and cosmic rays fill the air with those all the time.

Picture a cyclist with a tailwind: the wind pushes, the air drags, and the cyclist speeds up while the push is the stronger. The picture breaks at the decisive point. Air drag on a cyclist keeps rising with speed, so however strong the tailwind the cyclist settles at a top speed; the drag on an electron falls as it speeds up, to about 1 MeV, and then barely grows, so once the field beats the drag at that minimum nothing holds the electron back.

Friction comes from molecules, so every field in this story scales with the air density. Where this page calls a field "sea-level equivalent", it means the local field divided by n, with n from the US Standard Atmosphere 1976:¹³ about 0.73 at the 3,200 m Aragats station in Armenia, 0.54 at 6 km and 0.25 at 12 km, near airliner cruising height (this page's arithmetic). At 12 km a field of 72 kV/m is, for a fast electron, as strong as 284 kV/m at sea level, the avalanche threshold of Chapter 3.

Chapter 3Avalanche

Wilson's runaway was a single electron. In 1992 Alexander Gurevich, Gennady Milikh and Robert Roussel-Dupré showed that runaways multiply.¹⁴ Now and then a runaway knocks an electron out of a molecule with enough energy to run away too, so the number grows exponentially along the field: a relativistic runaway electron avalanche, or RREA, seeded by the secondary electrons of cosmic rays (or by runaways from lightning leaders, the hot, conducting channels of ionised air along which a lightning flash grows). It works at fields about ten times below those of ordinary breakdown, and Gurevich and colleagues proposed it as a "preconditioning" of the air for lightning (Chapter 11).

Scattering puts the threshold for an avalanche a little above the break-even field. Dwyer's 2003 calculation gives¹⁵

E_th = 2.84 × 10⁵ V/m × n = 284 kV/m × n2.84 kV/cm at sea-level density; about 206 kV/m at the Aragats station (3,200 m) and 72 kV/m at 12 km (this page's arithmetic, US Standard Atmosphere 1976).

Dwyer's value is the one most used; other calculations differ from it by several per cent,¹¹ and simulations of mountain stations with CORSIKA, a standard air-shower code, find that fields must exceed the thresholds they test (2.67 and 2.80 kV/cm × n) by about 10–20% before avalanches develop, about 7–15% above Dwyer's value (this page's arithmetic).¹⁶ Above the threshold, an approximate expression from Dwyer's group describes the growth:^17,11

λ ≈ 7.3 MeV / (eE − F_d) = 7,300 kV / (E − 276 kV/m × n), N = e^L/λλ is the e-folding length, over which the number of runaways multiplies by e ≈ 2.718 (the number, not the electron's charge e of eE); N is what one seed becomes over a field region of length L; F_d = 0.276 MeV/m × n is an effective drag. From Coleman and Dwyer (2006), as given in the 2012 review; later papers quote it for fields above about 300 kV/m × n. The offset 276 and the threshold 284 are different constants.

At 350 kV/m, 1.23 times the sea-level threshold, λ is 99 m and the count doubles every 68 m, so over 1 km one seed electron becomes about 25,000 (this page's arithmetic). Nearer the threshold the growth is far gentler: at 1.1 times it, 312.4 kV/m, λ is about 200 m and one seed becomes about 150 in a kilometre. Below 300 kV/m the expression is used outside the range it was derived for, and at the threshold itself the growth stops by definition, so this page quotes no multiplication there.

Thin air lowers the threshold in proportion to n and stretches every length by 1/n: at 12 km, 89 kV/m is the same 1.23 times the local threshold of 72 kV/m, but λ is about 390 m, and the 25,000-fold growth needs 3.9 km. The growth depends only on the voltage the field supplies across the region beyond what the drag takes back (L/λ is that surplus divided by 7,300 kV), and the surplus is the same at 12 km over 3.9 km as at sea level over 1 km. Near threshold and at altitude the expression is rough. A 2026 simulation study of four mountain stations finds longer avalanches: between 1.1 and 1.3 times the threshold its fitted coefficients make them about 1.4 to 1.8 times this expression's lengths, so over a kilometre at Aragats one seed becomes 4 to 25 times fewer than Figure 1 shows, and 11 to 14 times fewer at 350 kV/m (this page's arithmetic). The study also finds an extra dependence on density, and warns that its results "should not be generalized to arbitrary thunderstorm electric-field structures without further validation".¹⁸ Figure 1's altitude readouts are rough guides only.

The runaways average about 7 MeV, with an energy spectrum close to e^{−ε/7.3 MeV}, where ε is a runaway's energy (each 7.3 MeV step up holds e times fewer), nearly independent of field and density.¹¹ That is a property of the population: each runaway keeps gaining energy until it leaves the field region, and the average stays near 7 MeV because fresh, slow runaways keep joining. Their speed follows from the Lorentz factor γ, the total energy divided by the rest energy (this γ is not a gamma ray):

γ = 1 + 7.3/0.511 = 15.3, v/c = √(1 − 1/γ²) = 0.9979A 1 MeV electron moves at 94% of light speed and a 20 MeV one at 99.97% (this page's arithmetic).

Typical runaways move at 99.8% of light speed (this page's arithmetic); the avalanche as a whole advances at about 89% of light speed.¹⁷ The usual picture is a snowball rolling down a slope steep enough to keep it going, growing as it gathers snow. It breaks twice: snow does not climb back uphill to start new snowballs, but positrons do (Chapter 4); and a snowball keeps speeding up, while the avalanche's average energy stays near 7 MeV.

The fields come from the cloud's charge: usually a positive layer near the top, a negative layer in the middle and often a smaller positive pocket near the base, with the ground, a conductor, acting as if it held an opposite "mirror" charge beneath each.¹² The strong fields lie between these layers; electrons run one way along them, positrons the other.

Are storm fields that strong? Just. From 23 balloon soundings through thunderstorms, Marshall, McCarthy and Rust found that fields tend to be limited to about the runaway break-even field, which, they suggested, "may explain why electric field magnitudes greater than 150 kV m⁻¹ are rarely found inside thunderstorms"; 150 kV/m is about the break-even field at 3.7 km and the avalanche threshold at 6 km (this page's arithmetic: 218 × 0.69, 284 × 0.54).¹⁹ A 2005 sounding measured 186 kV/m at 5.77 km: about 130% of the avalanche threshold by its authors' own density model, and about 118% with the US Standard Atmosphere this page uses (n = 0.553, a local threshold of 157 kV/m; this page's arithmetic).²⁰ The strongest fields measured in storms, scaled to sea level, rarely exceed about 400 kV/m:¹¹ above the threshold, but not far above it.

Figure 1 puts these numbers under your fingers, at sea level and higher up. Its count stops near 100,000, and the amber dots running back up the column are the reason, which Chapter 4 explains.

Figure 1📈 Try it

One seed, one avalanche

Drag the field past the threshold

Field strength (sea-level equivalent)How hard the cloud pushes (scaled to sea-level air)

218284400750

Length of the strong-field region

Where is the cloud?

Threshold here, E_thThreshold here

E ÷ E_thPush ÷ threshold

e-folding length λDoubles every

One seed becomes

A relativistic runaway electron avalanche in a uniform field region, from the approximate expression of Coleman and Dwyer (2006) as given by Dwyer, Smith and Cummer (2012), with n, the air density relative to sea level, from the US Standard Atmosphere 1976. The slider sets the sea-level-equivalent field E/n, where E is the local field; the scale under the axis gives the actual field at the chosen altitude, where every length stretches by 1/n. Avalanches need E above E_th = 284n kV/m (Dwyer 2003); the grey band up to 1.1 E_th (1.15 E_th on the mountain settings, where simulations need fields up to about 15% above it; Chilingarian et al. 2026) marks where the formula is least reliable. The mark at 400 kV/m is the level that the strongest fields measured in storms, scaled to sea level, rarely exceed (Dwyer, Smith and Cummer 2012); the dot is a 2005 balloon reading, 186 kV/m at 5.77 km (Marshall et al. 2005): 336 kV/m sea-level equivalent with this page's standard atmosphere, 370 kV/m with the authors' density model. The count stops at about 100,000, the upper bound that relativistic feedback sets (Chapter 4), not a switch-on point; the positrons drawn running back up the field show that feedback. At altitude the readouts are rough: simulations for mountain stations give avalanche lengths about 1.5 times these at 350 kV/m, and on the Aragats setting a count over a kilometre 11 to 14 times smaller (Hovhannisyan 2026; this page's arithmetic). The figure shows how an avalanche starts and grows, as in a gamma-ray glow; a cosmic-ray seed and one avalanche do not by themselves make a terrestrial gamma-ray flash, whose source is still argued over. This page's arithmetic.Slide how hard the cloud pushes on its electrons. Below the threshold line, drag and the knocks off course win, and nothing grows. Above it, each one knocks loose more: well above it, at 350 kV per metre, their number doubles about every 68 m; just above the line it grows far more slowly. Press Release a seed to watch one cosmic-ray electron start an avalanche. It can't grow for ever: feedback from positrons running back up the field limits it to somewhere below about 100,000. Avalanches like this make a cloud glow with gamma rays for seconds or minutes (chapter 8); how the brief, far brighter flashes get their electrons is still argued over. Up a mountain or high in a storm the air is thinner: a weaker push is enough, but the avalanche needs proportionally more room, so over the same distance it grows less. Those numbers are rough.

Chapter 4Feedback

An avalanche of 7 MeV electrons shines. When a runaway is deflected by a nucleus it emits bremsstrahlung, "braking radiation", and the avalanche's gamma rays reach tens of MeV. Those above 1.022 MeV can make electron–positron pairs (Chapter 1). The new electron joins the avalanche. The positron, with the opposite charge, runs back up the field, and if it reaches the upstream end of the high-field region it can knock loose fresh runaway electrons that start a new avalanche. Gamma rays scattered backwards can do the same.

Joseph Dwyer called this relativistic feedback, and in 2003 he argued that it sets "a fundamental limit on electric fields in air". Once the product of the field's strength and the region's size passes a feedback threshold, each avalanche seeds at least one more on average: the discharge becomes self-sustaining and, on his calculation, stops the field growing further.¹⁵ Antimatter, in this picture, is part of the machine and not only one of its products.

Feedback also limits how far a single avalanche can grow. The 2012 review puts it plainly: "relativistic feedback limits the RREA multiplication factor to below about 10⁵ for reasonable thundercloud electric fields".¹¹ That is an upper bound, not a switch that flips at 100,000. Where the limit falls depends on the size and shape of the field region, and in a 2015 analysis of aircraft data the strongest gamma-ray glow (a storm's gamma-ray emission lasting seconds or minutes; Chapter 8) was already "close to the value at which relativistic feedback processes become important, with an avalanche multiplication factor of 4,500".²¹ The 25,000-fold example of Chapter 3 sits near the upper limit that feedback allows.

Positron feedback usually dominates for average fields below about 750 kV/m × n, and feedback by backscattered photons above that (Dwyer 2007, as summarised in the review).¹¹ Dwyer's case that feedback caps the fields of real storms rests on calculation; it is not a measured universal limit.

Quantum electrodynamics, the quantum theory of light and charged particles, also allows a strong enough field to pull electron–positron pairs out of empty space by tunnelling, the process Nº 5 met when the string between two quarks breaks. The field needed is Schwinger's:²²

E_S = m_e²c³ / (eħ) = 1.32 × 10¹⁸ V/mSchwinger 1951; this page's arithmetic from CODATA constants: 4.7 × 10¹² times the 284 kV/m avalanche threshold.

Storm fields never come near it: thunderstorms make their positrons the ordinary way, from gamma rays striking atoms, and never tear them out of the vacuum. (Nº 11 met yet another route to pairs, predicted near nuclei with a charge of about 170.)

Part IICatching the line

Chapter 5TGFs

The first evidence from space that storms accelerate electrons came from an instrument pointed the other way. BATSE, on NASA's Compton Gamma Ray Observatory, was built to catch gamma-ray bursts from deep space. In 1994 Gerald Fishman and colleagues reported that it was also catching brief flashes of gamma rays coming up from the atmosphere over thunderstorms (they placed the sources above 30 km; later work put them inside the clouds).²³ These terrestrial gamma-ray flashes, TGFs, last tens to hundreds of microseconds, well under a millisecond, and their photons are energetic: NASA's RHESSI satellite recorded them up to 20 MeV in 2005,²⁴ and the team of the Italian AGILE satellite reported photons up to about 100 MeV in 2011.²⁵

The spectrum at the source is what a runaway avalanche should make:²⁶

dN/dE ∝ E⁻¹ e^{−E/7.3 MeV}Photons per unit energy at the source (E here is the photon's energy, not the field): the bremsstrahlung of a runaway avalanche, which fits TGF observations (Dwyer and Smith 2005). The 7.3 MeV is the runaways' mean energy (Chapter 3). Figures 3 and 4 use this spectrum.

In words: many photons at low energies, fewer at each step up, and a steep fall beyond about 7 MeV (∝ means "in proportion to").

Two counts describe a flash: the runaway electrons that make it, and the gamma rays above 1 MeV that leave the source. The flashes seen from orbit are typically made about 12 km up, near airliner cruising height. Radio measurements place some sources at 11.8–11.9 km and imply about 10¹⁸ runaway electrons;²⁷ to be seen from space at all, a TGF needs about 10¹⁷ or more of them.¹¹ In gamma rays, a bright Mediterranean flash analysed by Gjesteland and colleagues in 2015 needed about 3 × 10¹⁸ above 1 MeV, about ten times more than earlier estimates for an average flash seen by RHESSI.²⁸ This page takes a tenth of that, about 3 × 10¹⁷, as a typical value (this page's reading). Pallu and colleagues, modelling doses in 2021, took 10¹⁸, which they call both bright and typical of the flashes observed from space.²⁹ A typical space-detected TGF, then, sends out a few times 10¹⁷ to 10¹⁸ gamma rays above 1 MeV. All of these numbers describe the flashes satellites see; Chapter 8 shows that most TGFs are far fainter.

The gamma rays leave in a broad upward cone. RHESSI data are "most consistent with wide gamma-ray beams with half angles > 30 degrees" (measured from the cone's axis to its edge), and deeper sources fit better with about 45°;^11,30,31 individual flashes seen by Fermi differ, some favouring narrow beams and others wide ones.³² Since 2018 ASIM, a European instrument on the International Space Station, has tied TGFs to lightning leaders and to ultraviolet emissions in the ionosphere, the electrically charged upper atmosphere.³³

Where does a TGF get its electrons? About 10¹⁷ of them in well under a millisecond is far more than an avalanche working on the ordinary cosmic-ray background can supply. With feedback holding each avalanche below about 10⁵ and a seed flux of about 10⁴ per square metre per second, "the RREA mechanism would require a lateral diameter of the TGF source region to be at least 800 km", and cosmic-ray air showers, the particle cascades one energetic cosmic ray sets off, do not supply enough runaways either.¹¹ Dwyer argued in 2008, as the 2012 review summarises,¹¹ for two viable mechanisms: relativistic feedback (Chapter 4), or runaway electrons made in the strong fields at the tips of lightning leaders and of streamers, the thin, cold sparks that branch out ahead of a leader. In 2025 he wrote that "it has been challenging to tell which mechanism is correct when lightning is present".³⁴

Chapter 6Fermi 2009

Fermi's Gamma-ray Burst Monitor sees TGFs too, and most arrive in a fraction of a millisecond from storms not far from the point below the spacecraft. A few arrive differently: they last longer than a millisecond, and they can come from storms far from the spacecraft. These "electron TGFs" are not gamma rays when they reach orbit: a TGF's gamma rays knock electrons out of air molecules and make electron–positron pairs in the thin air above the source, and the charged particles that escape are guided along Earth's magnetic field lines, which can carry them thousands of kilometres.

In January 2011 Michael Briggs and colleagues reported that the spectra of the three brightest electron TGFs show strong 511 keV annihilation lines. The beams therefore contain many positrons, which annihilate with electrons in the spacecraft itself. The authors concluded that most likely all TGFs inject beams of electrons and positrons into near-Earth space.¹ The detection is established; "all TGFs" is the authors' interpretation.

The event NASA used to show it was the one this page began with: Fermi over Egypt and the storm over Zambia, on 14 December 2009. The beam hit Fermi, travelled on to a magnetic mirror point, where the converging field turned it round, and hit Fermi again 23 ms later.²

The positrons carried the storm's signature thousands of kilometres before they met their electrons. Fermi did not see antimatter hovering over a storm; it saw a beam the storm had thrown into space.

Chapter 7Inside

Aircraft can get closer than satellites, sometimes closer than planned. On 21 August 2009 a Gulfstream V jet carrying ADELE, a gamma-ray instrument, was flying at 14.1 km over the coastal estuaries of the US state of Georgia, near Brunswick, when it inadvertently entered the upper part of an active thunderstorm cell. ADELE recorded two enhancements of 511 keV gamma rays, each about 12 times the background, about 0.2 s long and 35 s apart, with electrical activity in the storm. Joseph Dwyer and colleagues fitted them with clouds of positrons more than a kilometre across, and wrote that the cause is unclear and that the aircraft itself may have contributed.³⁵

In January 2016 an Airbus A340 test aircraft carrying ILDAS, an instrument system built to record lightning strikes on aircraft, flew through a thunderstorm at 12 km over northern Australia and recorded a sequence of ten gamma-ray enhancements, each lasting about a second, whose spectrum consisted mainly of 511 keV photons and their lower-energy tail from Compton scattering, in which a photon loses part of its energy to an electron (Chapter 9). They came with fast changes in the electric field, and the electrical activity close by was a series of static discharges from the aircraft itself. A simulation of the whole aircraft with Geant4, a standard particle-transport code, "indicated that the positrons annihilated in direct vicinity or in the aircraft body".³⁶

Neither result shows a cloud of antimatter waiting in storms for airliners to fly into. The ADELE positron clouds are the team's fit to two bursts, and how the clouds formed is unclear; the ILDAS line was traced by its team to the aircraft's own surroundings. What the two flights show is that instruments at and just above airliner heights see the line inside storms, and that reading it needs a model of the aircraft as well as of the cloud: a 511 keV photon records where a positron met an electron, not where it was made.

Chapter 8Glows

A TGF is over in a fraction of a millisecond. Thunderclouds can also glow: gamma-ray emission lasting seconds to minutes, sometimes ended abruptly by a lightning flash. Aircraft caught the first signs. In 1981 Parks and colleagues reported X-rays (photons of lower energy than gamma rays) of 3 to more than 12 keV, measured aboard a NASA F-106 jet flown into thunderstorms,³⁷ and a follow-up flight saw X-ray rises lasting seconds, at energies from 5 to more than 110 keV, some ending within 0.1 s of lightning.³⁸ Glows have since been seen from aircraft, from balloons and from the ground.³⁹ ADELE's aircraft data, analysed in 2015, showed glows to be relatively common near storm tops.²¹ In 2017 NASA's high-altitude ER-2 flew 45 hours at 20 km over storms in the United States and reported two glows, both over Colorado on 8 May.⁴⁰

In July 2023 the ER-2 went to the tropics. The ALOFT campaign, led by the University of Bergen with NASA, made ten flights between 3 and 29 July 2023 from Tampa, Florida, at 20 km above the Gulf of Mexico, Central America and the Caribbean. The campaign's two main papers appeared in Nature in October 2024.^39,41

Martino Marisaldi and colleagues report that tropical thunderclouds over ocean and coastal regions commonly emit gamma rays, for hours, over areas up to a few thousand square kilometres. The emission is tied to the storms' deep convective cores, their tallest rising towers, and varies on timescales of seconds and less, a pattern the authors liken to a "boiling pot": glows bubbling up and dying away, a picture of the pattern and not of heat. The aircraft recorded more than 500 individual glows of 1 to 10 seconds, on nine of the ten flights.³⁹ The flight table lists 608: 392 of them, 64% (this page's arithmetic), on one flight over the Bay of Campeche on 24 July, and none on 12–13 July off Florida.

Nikolai Østgaard and colleagues report a new kind of flash, the flickering gamma-ray flash (FGF): 24 of them, on five of the ten flights, lasting 20–250 ms, with more and longer pulses than ordinary TGFs and no radio or optical signal of their own. An FGF starts as a glow, then rises exponentially and flickers; the authors propose FGFs as the missing link between glows and TGFs. In 17 of the 24 the flash was followed, within tens of milliseconds, by a narrow bipolar event, a short, intense radio burst, and then by lightning.⁴¹ That is an observed sequence; Chapter 11 asks what it might mean.

Figure 2 maps the campaign and, in its world view, every other detection on this page.

Figure 2🗺️ Try it

Ten flights, 608 glows

Fly the campaign

Flight

GlowsSeparate glows

Share of the campaignOut of all 608

10 flights, 3–29 July 2023 · glows on 9 · 24 flickering flashes on 5 flights · 96 TGFs, only 3 or 4 bright enough for space10 flights, 3–29 July 2023 · glows on 9 · 24 flickering flashes on 5 flights · 96 brief flashes, only 3 or 4 bright enough for space

The ALOFT campaign: ten flights of NASA's ER-2 at 20 km between 3 and 29 July 2023 from MacDill Air Force Base, Tampa (Marisaldi et al. 2024, Extended Data Table 1; published October 2024). Each flight is drawn as the region the flight table names, not as a track; flight 8 appears as a labelled region, "Inland Mexico, as the paper's flight table names it", because the main text places all glows over ocean or coastal regions. Bars: glows per flight, 608 in the table. Diamonds: regions where flickering gamma-ray flashes were seen (Østgaard et al. 2024). The small circle is 3,000 km² for scale; the abstract says glowing areas reach up to a few thousand km². The world view draws the page's other detections as small shaded regions around the places the sources name; the Fermi arc joins the storm and the spacecraft, whose beam followed Earth's magnetic field. Coastlines: simplified outlines drawn for this page. As of October 2026.Where NASA's high-flying ER-2 went looking in July 2023. Each bubble is a region one of the flights went to (some were visited more than once), and the bars under the map count the separate gamma-ray glows on each flight. Press Fly the campaign to step through the ten flights. One flight, over the Bay of Campeche on 24 July, found 392 glows on its own. One bubble says inland Mexico because that is what the paper's flight table calls it; the paper itself places all the glows over sea or coasts. Switch to Around the world to see everywhere else in this story that storms' gamma rays and positrons have been caught. Places and results as of October 2026.

The flights also recorded 96 TGFs, of which only 3 or 4 were bright enough to have been seen from space. "At least in the Caribbean and Central America during summer", the authors conclude, thunderclouds produce almost a hundred times (two orders of magnitude) more gamma-ray flashes than space instruments can detect.⁴¹ Analyses in 2024–2026 sharpened the point. Of six TGFs that ALOFT recorded on 24 July 2023 while the space station passed overhead, ASIM detected none: Bjørge-Engeland and colleagues put five at about 10¹²–10¹⁵ photons at the source (counted from 100 keV to 40 MeV in their model) and the brightest at about 10¹⁶, the only one ASIM might have caught.⁴² Fuglestad and colleagues, estimating by simulation the source brightness (photons above 100 keV) of the 43 flashes they located and fitting 37 of them, conclude that at least 97% of the ALOFT TGFs cannot be detected from space.⁴³

A ground instrument agrees. The Telescope Array, a 700 km² cosmic-ray observatory in south-western Utah, recorded gamma-ray bursts at the ground coinciding with downward lightning leaders in 2014–2016, from sources about 3–5 km above the ground and, for the 16° half-angle beams its simulations assume, of about 10¹²–10¹⁴ photons (no energy cut given); from that one site, its authors conclude that satellites see only the most powerful flashes.⁴⁴

Simulations try to join glows, flickering flashes and TGFs. Dwyer's 2025 calculations make strong glows from rapid storm charging alone that grow into pulse trains closely resembling the multi-pulsed TGFs seen by BATSE and by ALOFT,³⁴ and a 2026 model from the ALOFT team, by Færder and colleagues, makes all three from one relativistic-feedback process as the field rises at different rates.⁴⁵ Both are simulations: neither yet demonstrates that a glow boils over into a flash in nature.

Chapter 9Spectrum

A detector does not see positrons, only photons, and from a storm most of their spectrum is the bremsstrahlung continuum of Chapter 5, a smooth curve falling out to tens of MeV. On the source spectrum, 41% of the photons above 100 keV (counting up to 100 MeV) have more than the 1.022 MeV needed to make a pair (this page's arithmetic). That is the share able to make pairs, not the share that do: most of them Compton-scatter in the air first, losing energy to electrons.

The positrons that are made slow down and annihilate, adding a line, a narrow spike at 511 keV, to the continuum. The line is a fingerprint, not a count. Any energy band around 511 keV also holds continuum photons and, near the ground, photons from natural radioactivity: some Compton-scattered down from the higher-energy lines of radon's decay products, and a line of thallium-208 at 510.8 keV, too close to 511 keV for any detector to resolve; its size has to be estimated from thallium-208's other lines, such as its stronger one at 583 keV (Chapter 10). Positronium adds a smear, the three-photon spectrum of ortho-positronium worked out by Ore and Powell in 1949, rising to a sharp edge at 511 keV.⁴⁶

How sharp the spike looks depends on the detector's energy resolution, the width it gives a single energy (the full width at half maximum, FWHM). A typical sodium iodide (NaI) scintillator, a crystal that flashes when a gamma ray strikes it, smears 511 keV over about 40 keV; a germanium detector of the kind used to measure annihilation lines in gases, over 1.16 keV. The line has a width of its own, because the electrons the positrons meet are moving: for positrons slowed to room-temperature energies, the widths measured in molecular gases ran from 1.7 keV for hydrogen to about 3.1 keV for fluorocarbons.^47,48 That is negligible next to a scintillator, but not next to germanium. A measure that does not depend on the resolution is the equivalent width:

EW = N_line / (dN/dE)_{E = 511 keV}The number of photons in the line divided by the continuum's photons per unit energy at the line: the width of continuum that holds as many photons as the line does.

On the ground, the reference case is a winter thunderstorm on the Sea of Japan coast. At 00:27 UT on 13 January 2012 the GROWTH experiment, detectors set up to watch winter thunderclouds, recorded a flash of under 300 ms with an intracloud discharge, lightning inside the cloud, then about 60 s of emission reaching about 10 MeV, containing "a distinct line emission at 508 ± 3(stat.) ± 5(sys.) keV" (statistical and systematic uncertainties), identified as the annihilation line. The line was no wider than the instrument's resolution of about 80 keV and held 520 ± 50 of the 5,340 ± 190 photons detected between 0.1 and 10 MeV, about 10%; its equivalent width was 280 ± 40 keV. Umemoto and colleagues suggest that a downward positron beam produced both the continuum and the line photons.⁴⁹ As far as we can find, it is the earliest refereed report of the line from a thundercloud at ground level; the beam is their suggestion, not an established result. On Figure 3's source continuum the same share gives an equivalent width of about 210 keV, not 280 (this page's arithmetic), because the figure draws the spectrum as it leaves the source, before the air reshapes it.

Lightning can also make positrons through the nucleus. On 6 February 2017, near Kashiwazaki in Niigata, Japan, Teruaki Enoto and colleagues saw lightning 0.5–1.7 km away produce a TGF-like flash, then gamma rays from the capture of neutrons the flash had freed, fading with a decay time of 40–60 ms, then a 511 keV line lasting about a minute.⁵⁰ The flash's gamma rays had knocked neutrons out of nitrogen:

¹⁴N + γ → ¹³N + n, ¹³N → ¹³C + e⁺ + νA gamma ray knocks a neutron (n) out of a nitrogen-14 nucleus; the nitrogen-13 left behind is radioactive and decays to carbon-13 by emitting a positron and a neutrino (ν), so positrons kept appearing after the flash had ended.

Enoto and colleagues call lightning the second known natural channel on Earth, after cosmic rays (Nº 12), for isotopes such as ¹³C, ¹⁴C and ¹⁵N. This page claims nothing about the size of lightning's share.

On Mount Etna on 22 July 2024 the ESTHER detector at the Citelli Refuge, 1,741 m up, recorded a 6.3-minute event; in its final burst more than 12% of the counts fell within 511 ± 25 keV, and the count rate in that band reached five times its usual background, with the nearest lightning under 500 m away. The team interprets it as possible annihilation inside the parent cloud.⁵¹

Figure 3 draws the source spectrum with its line; change the detector to see the same photons as a tall, narrow spike or a low, wide one. Even with no line, the window around 511 keV is never empty. The Aragats setting belongs to Chapter 10.

Figure 3🔬 Try it

The line in the rainbow

Let the positrons appear

Detector

Line strength (illustrative, not a measured positron fraction)How strong the spike is (illustrative)

Line sharePhotons in the spike

Equivalent width

511 window: share that is line photonsSpike's share of the 511 window

Continuum and line photons able to make pairs (> 1.022 MeV)Strong enough to make a positron

Gamma rays at the source of a runaway avalanche: bremsstrahlung close to dN/dE ∝ E⁻¹e^{−E/7.3 MeV} (Dwyer and Smith 2005), drawn from 0.1 to 10 MeV and normalised over that range (photons per keV; logarithmic axes; the strip below zooms on 300–700 keV with a linear axis). Photons above 1.022 MeV (shaded) are able to make electron–positron pairs: 41% of the source's photons above 100 keV (counting up to 100 MeV), 39% of those drawn here, though most Compton-scatter first. The 511 keV line is a Gaussian whose width combines the detector's resolution (scaled as √E from typical values at 662 keV for NaI(Tl) and LaBr₃; 1.16 keV at 511 keV for germanium; 80 keV for the 2012 Japanese detector, as its authors give it) with the line's own width, about 2.4 keV here, inside the 1.7–3.1 keV measured in molecular gases (Iwata, Greaves and Surko 1997, as reviewed by Gribakin, Young and Surko 2010). Line strength is the share of photons between 0.1 and 10 MeV that are line photons: illustrative except in the Japan 2012 setting, which uses Umemoto et al.'s 520 ± 50 of 5,340 ± 190 photons. The positronium smear (Ore and Powell 1949) has an illustrative share, since none has been measured in storm air; its photons count in the window share, not in the line share. The Aragats setting marks the energy windows that team used and prints the numbers its preprint reports; it draws no line, only the model continuum for the window positions, and does not reproduce that detector's spectrum. Preprint status as of October 2026. This page's model.A gamma-ray rainbow from a storm: low energies on the left, high on the right. The smooth slope is the light of fast electrons braking in air, and only photons (particles of light) to the right of the 1 MeV mark can make a positron. That's about 4 in every 10 of the slope's photons. The tile reads a little less, because the figure draws only up to 10 MeV and counts the spike's photons too. Each line up the side means 10 times more photons. Press Let positrons appear and watch a sharp spike grow at 511 keV: the two gamma rays from each positron meeting an electron. The strip underneath zooms in on the spike; a blurrier detector smears it (try Blurry against Sharpest; the Japan 2012 detector is the blurriest). Positronium smear is something else: it adds the three weaker gamma rays of chapter 1, spread out below 511 keV. Nobody has measured how many there are in storm air, so its size here is only an illustration. The Aragats setting (a preprint, not checked by a journal as of October 2026) shows that the radon window rose in the same storms. That is why a rise in the 511 keV window isn't a positron count: radon's decay products, which rise in storms, send photons into that window too.

Chapter 10Aragats

The newest ground result is a preprint from Armenia. The Aragats station of the A. I. Alikhanyan National Laboratory (the Yerevan Physics Institute) stands at 3,200 m beside Lake Kari; its detectors record thunderstorm ground enhancements (TGEs), rises in the electrons and gamma rays reaching the ground during storms.

The group had reported positrons before. In 2024 Chilingarian and colleagues described a thunderstorm-related rise in the 511 keV line and linked it to a lower positive charge region near the cloud base, the pocket of Chapter 3, and the field it sets up near the ground.⁵² In a second 2024 paper, on an event they described as an "enormous increase" in positron flux, they proposed a localised dipole, a pair of opposite charges, between that charge region and the ground, which accelerates positrons downwards (as their 2026 preprint summarises it).⁵³

The new work, "Positron acceleration in thunderstorms" by Chilingarian and Sargsyan, was submitted to arXiv on 27 June 2026 and posted in July; as of 25 October 2026 it has no journal reference and no later version.⁵⁴ It reports three events on 16 and 17 May 2026 as rises in the counts of a spectrometer called RT56 in fixed energy windows: 478.5–543.5 keV around the annihilation line, and two that track radon's decay products, around lead-214's 352 keV line (325.2–378.8 keV) and bismuth-214's 609 keV line (572.9–645.1 keV).

Why radon? During storms the gamma-ray count from the short-lived decay products of radon, a radioactive gas from the ground, also rises near the detectors: precipitation washes those decay products out of the air (shown for rain;⁵⁵ two of the three May 2026 events came with graupel, soft hail), and the Aragats group adds that storm fields lift them. Bismuth-214's gamma rays at 609 keV and above can Compton-scatter down into the 511 keV window, so the window rises with the radon even without positrons. (Lead-214's 352 keV line lies below the window, and scattering only lowers a photon's energy, so it cannot spill upwards into it.) The window also holds a line from natural radioactivity: thallium-208, a decay product of thoron (radon-220), emits gamma rays at 510.8 keV, inside the window, and at 583 keV, inside the window the preprint labels bismuth-214's;⁵⁶ the preprint does not discuss it. The 511 keV window therefore has to be read beside the radon windows, as the preprint does.

The preprint's Table 1, side by side:

Event (UT)	511 keV window	²¹⁴Bi window	The authors' reading
16 May 2026, 09:34	+14.8%	+19.0%	"not cleanly separated from the radon response"
17 May 2026, 14:26	+35.3%	+29.9%	"significantly larger than the average radon progeny response"
17 May 2026, 17:03, during an ordinary burst of electrons and gamma rays	+24.1%	+11.0%	pairs made by the burst's gamma rays, plus Compton-scattered photons from enhanced bismuth-214

Rises as printed in the preprint, to one decimal place; this page does no arithmetic on them.

What the preprint adds to the 2024 work is timing. On 17 May an ordinary TGE, a burst of electrons and gamma rays, peaked at 14:14 UT, and the 511 keV window peaked about 12 minutes later, at 14:26 UT: the table's second row. The burst in its third row, at 17:03 UT, was a later one. Calling these the first time-resolved observations of such events, the authors argue that the delay shows positrons accelerated towards the ground by a small dipole between a positive charge pocket at the cloud base and its mirror charge in the ground, a dipole that also slows electrons. It is one half of what they call a "dual-dipole" model; the other half, the main negative charge layer and its mirror charge, drives the ordinary bursts. They propose "positron TGEs" as a new subclass. The cloud base was estimated, not measured, at about 35–50 m above the station, from the gap between the air temperature and the dew point. The model is their interpretation and, as far as we can find, has not been tested by others.

What the preprint reports

Data: three rises in a 65 keV window around 511 keV, each beside a bismuth-214 rise (Table 1)
Timing: on 17 May, the 511 keV peak at 14:26 UT, 12 minutes after an ordinary burst at 14:14 UT
Claim: positrons pushed to the ground by a small dipole under the cloud

What it does not contain

A fitted line: no net 511 keV line separated from radon-related, thallium-208 and Compton counts
Significance: no number; 16 May is "statistically significant in the selected window"
Transport: no simulation of positrons moving through the proposed dipole
Detector: RT56's crystal and energy resolution are not given

It is tempting to subtract the windows, 35.3% minus 29.9%, and call the rest positrons, but the two windows have different baselines and different radon shares, so percentage rises in them cannot be subtracted (this page's reading of the preprint). The same subtraction would make the ordinary 17:03 UT burst, 24.1% against 11.0%, the most positron-rich of the three, which the authors explain differently. What the preprint shows is a rise of up to about a third in the 511 keV count; how much of it is annihilation, and where the positrons came from, it does not settle.

Ground-level positron signals have now been reported from Japan (2012 and 2017), Armenia and Italy, by different methods. Two teams have proposed that storms drive positrons downwards, the GROWTH team for the 2012 event and the Aragats group, and neither proposal is established. As of 25 October 2026 whether storms do so is an open question.

Part IIIWhat it means

Chapter 11Lightning

Everything so far has happened in fields too weak to make a spark, which raises the subject's oldest puzzle: how does lightning start? Dry air breaks down at about 3 × 10⁶ V/m at sea level, and with raindrops and hail present at 1.0–1.4 × 10⁶ V/m. Yet decades of measurements inside storms "have failed to find electric field strengths anywhere near the conventional breakdown threshold, even when the effects of precipitation are included"; the strongest, scaled to sea level, rarely exceed about 4 × 10⁵ V/m (the 400 kV/m of Chapter 3), "one third the required value with precipitation".¹¹ That is about 7.5 times below dry breakdown and 2.5–3.5 times below breakdown with precipitation (this page's arithmetic): several times too weak, not ten times. The factor of ten is the gap between a spark and the avalanche threshold, and the strongest fields do exceed that threshold. An avalanche at a tenth of a spark's field is not itself a spark. Three views give runaways very different roles, and each has evidence against it.

Cosmic rays as the trigger. Alexander Gurevich and Kirill Zybin proposed that a cosmic-ray air shower together with runaway breakdown (their name for the avalanche of Chapter 3) provides the trigger.^57,58 The 2012 review finds "little observational evidence and only limited theoretical work to support this idea": in Babich and colleagues' models air showers and avalanches together cannot start lightning, and in Dwyer's Monte Carlo simulations, which follow particles collision by collision, the avalanches' sideways spreading and the feedback limit keep the air's conductivity too low to do so. The idea has not been ruled out, but how it would work has not been shown.¹¹

Ordinary (dielectric) breakdown. In 2016 William Rison and colleagues located discharges from their radio waves (radio interferometry) and found that "fast positive breakdown", the process behind narrow bipolar events, starts many lightning discharges; it appears to be purely dielectric, positive streamers in a locally intense field with no runaway electrons needed, and they suggested that many or possibly all flashes begin this way.⁵⁹ Later radio studies limit how general that is: most of the 31 intracloud flashes that Lyu and colleagues imaged began without any identifiable fast positive breakdown;⁶⁰ Tilles and colleagues showed that breakdown can also begin with the opposite polarity;⁶¹ and in Kolmašová and colleagues' sample a quarter of 35 positive narrow bipolar events over south-eastern France started no flash at all.⁶²

Feedback and high-energy processes. Victor Pasko and colleagues argue that photoelectric feedback, in which photons from the avalanches free new seed electrons from air molecules, explains TGFs from compact runaway regions and offers "a straightforward solution" to what they call the long-standing problem of lightning initiation;⁶³ a 2026 follow-up, published in October, extends the idea to weaker and slower initiating discharges and matches their current moments (current times the length of the channel), pulse durations and speeds.⁶⁴ Dwyer's 2025 simulations show glows amplifying fields locally, potentially towards lightning.³⁴ And on the 2023 flights 17 of the 24 flickering flashes were followed within tens of milliseconds by a narrow bipolar event and then lightning; the ALOFT authors raise the possibility that flickering flashes play a role at least in some lightning initiation.⁴¹ That is a sequence in time, not a demonstrated cause, and none of these models is yet the consensus.

The views overlap more than their labels suggest: cosmic rays provide seed electrons in all of them, and the dispute is about what turns seeds into a leader. In 2026 a group of lightning modellers called for an integrated, end-to-end model of lightning, set out as a five-year plan.⁶⁵ As of October 2026, how lightning starts is still an open question.

Chapter 12Aircraft

The flashes seen from orbit are born about 12 km up, near the cruising heights of airliners, about 10–12.5 km (flight level 370, FL370, is 37,000 feet or 11.3 km, and FL410 is 41,000 feet or 12.5 km). Airliners are struck by lightning once or twice a year each on average, as Dwyer and colleagues noted in 2010,⁶⁶ but a strike does not put a plane inside a TGF. Every dose from a flash in this chapter is a calculation, not a measurement.

For scale: a London–New York flight (6.8 hours at FL370) gives about 27.3 µSv from cosmic rays; aircrew receive about 3 mSv a year on average.⁶⁷ The International Commission on Radiological Protection recommends limits of 1 mSv a year for the public and 20 mSv a year, averaged over five years, for workers.⁶⁸ The solar storms of Nº 12, which raise aircrew doses through bursts of high-energy protons, are a different matter.

The largest doses would come from the runaway electrons themselves, to an aircraft inside the source region. They depend mostly on the beam's width, which nobody knows, so both groups that calculated them give curves against width. For a flash with 2 × 10¹⁷ runaway electrons, Dwyer and colleagues (2010) give about 0.7 Sv for a 50 m beam, about 0.1 Sv for one about 140 m wide (their abstract says the dose "could potentially approach 0.1 Sv"), about 7 mSv at 500 m, and under 1 mSv for beams wider than about 1.4 km (values read from their figure, as reproduced in the 2012 review).^66,11 Pallu and colleagues (2021) give more once a field region is added. Their first model, a lightning leader alone, peaks at about 0.3 Sv 50 m above the source; in their second, a leader plus a uniform field over 1 km, the dose at the end of the avalanche runs from about 30 mSv for a 2 km-wide beam to 7 Sv for the narrowest, 130 m, with a field of 400 kV/m scaled to ground level, and from 54 mSv to 18 Sv (at 107 m) with 500 kV/m. Their summary covers both: "at least approaching the very high dose of 1 Sv in both models".²⁹ At the same width, Dwyer's curve and Pallu's 400 kV/m case disagree by about 60 to 70 times: at 2 km, about 0.45 mSv on Dwyer's curve against Pallu's 30 mSv (this page's arithmetic). Two groups, then, and their spread is part of the answer. The 0.1 Sv on Dwyer's curve for a 140 m beam is about 3,700 London–New York flights, or 33 years of the average aircrew dose (this page's arithmetic).

Outside the electron source only the gamma rays arrive, and they are far weaker. Pallu and colleagues modelled a flash of 10¹⁸ gamma rays above 1 MeV from a point at 12 km into a 45° cone and found at most about 90 µSv (as ambient dose equivalent, H*(10), which they treat as comparable to effective dose here), for a plane 1 km above the source. Within 1 km they calculated no photon dose, because there the runaway electrons, which travel at least 700 m, dominate.²⁹ That maximum is below every yearly limit, which is why a 2023 paper from Pallu's group treats photon doses as too low to count in flight-dose calculations.⁶⁹

This page checks the scale with a simple model of its own, in which the gamma rays spread out through a cone from a point and the air takes some of them out of the beam on the way up:

D(r) = N / (Ωr²) × ∫ f(E) e^{−(μ/ρ)(E) X(r)} k(E) dE, Ω = 2π(1 − cos θ)N gamma rays above 1 MeV leave a point source into an upward cone of half-angle θ; the aircraft is a distance r straight up the axis. f(E) is the source spectrum E⁻¹e^{−E/7.3 MeV} normalised over 1–30 MeV; μ/ρ is the mass attenuation coefficient of air (NIST); X(r) is the column of air between source and aircraft, thinning with height (US Standard Atmosphere 1976); k(E) converts photon fluence to effective dose (ICRP Publication 116, isotropic geometry). Scattered photons count as lost, so values far from the source are lower bounds. This page's estimate.^70,71,13

For a typical space-detected flash, 3 × 10¹⁷ gamma rays above 1 MeV into a 45° cone, the model gives about 65 µSv at 1 km straight up the beam and 2.2 µSv at 3 km (this page's estimate), a fall of about 30 times from the spreading cone and the air in the way. For Pallu and colleagues' case it gives about 220 µSv of effective dose, 2.4 times their 90 µSv. They average the dose over rings around the beam's axis, which they note blurs the value on the axis some kilometres above the source; this page cannot say why the two differ at 1 km.

Geometry matters too. A plane 1 km above a flash at 12 km is at 13 km, above nearly every airliner, so an airliner in the upward cone needs a flash below its cruising height: with the flash at 10.3 km, a plane 1 km above it at FL370 would receive about 52 µSv from a typical flash in this model. The model's brightest cases pass the yearly public limit: the bright Mediterranean flash of Chapter 5 (3 × 10¹⁸, three times Pallu and colleagues' case) into a narrow 30° cone, with the plane 1 km up, reads about 1.4 mSv with the flash at 12 km, 140% of the 1 mSv limit. That is this page's simple model, which reads high; the published verdict rests on Pallu and colleagues' 90 µSv.

Figure 4🔦 Try it

A plane in the beam

Move the plane up the beamMove the plane up the cone

How bright a flash

Height above the flash, straight up the beamHeight above the flash, straight up its cone of gamma rays

Beam half-angleHow wide the cone is

Flash altitude (this page's assumption)

Gamma-ray dose (this page's estimate)Dose (this page's estimate)

Aircraft altitudePlane's height

London–New York flightsTransatlantic flights' worth

Share of the 1 mSv yearly public limitOf a year's public limit

Gamma rays (tab A): the model of the box above, a flash as a point source at 12 km (adjustable, an assumption) sending N photons above 1 MeV into an upward cone of half-angle θ (30°–60° here; RHESSI data favour more than 30°, Gjesteland et al. 2011), with the dose taken on the axis a distance r straight up; scattered photons count as lost, so far values are lower bounds. Brightness presets are labelled with their sources (photons above 1 MeV). For Pallu et al.'s 2021 case the model gives about 220 µSv at 1 km against their Monte Carlo maximum of 90 µSv (H*(10)); where the two differ, the published value stands. Within 1 km of the source the runaway electrons dominate and no photon value is shown. Electrons (tab B): the published dose curves for a plane inside the electron source region, against beam width: Dwyer et al. 2010, values read from their figure as reproduced in Dwyer, Smith and Cummer (2012), and the end points of Pallu et al.'s 2021 leader-plus-field model at 400 and 500 kV/m; Dwyer's curve and Pallu et al.'s 400 kV/m case differ by about 60 to 70 times at the same width, and the real width is unknown. All values are calculations, not measurements. Comparisons: FAA (London–New York, 27.3 µSv; the aircrew average, UNSCEAR's figure as the FAA gives it), ICRP 103 (limits). This page's estimate where marked.How much radiation would reach a plane above a flash? In the first tab, drag the plane up the flash's cone of gamma rays. Gamma rays thin out fast with distance: 1 km above a flash of the brightness satellites usually see, this page's rough model gives about as much as two or three transatlantic flights, and far less farther away. For a brighter flash the published calculation gives about 90 microsieverts, less than this page's model gives for the same flash. In the second tab you're inside the electron beam itself, where two published models give anything from under a millisievert to over ten sieverts, depending on how wide the beam is. Nobody knows that width, and probably no airliner has ever been inside one.

How likely is any of this? In their 2023 paper, Pallu and colleagues combined the first Fermi TGF catalogue with 2.5 years of real Air France routes and put the chance of an aircraft entering a TGF's electron beam at "lower than one aircraft hit every 2 years for the total air traffic". That upper bound ignores flight altitudes and the fact that pilots steer round storms, and they think it likely that no aircraft has ever flown through a TGF's source.⁶⁹

Chapter 13Receipt

A 511 keV line is a receipt. It proves that a positron existed and met an electron. Like a receipt, it records the transaction and not where the money came from: whether the positron was made by a gamma ray in the cloud, carried along Earth's magnetic field from a storm 4,500 km away, pushed towards the ground, or emitted by a nitrogen-13 nucleus that lightning had made.

Read that way, the story holds together. C. T. R. Wilson guessed in 1925 that thunderclouds could accelerate electrons; Gurevich, Milikh and Roussel-Dupré showed in 1992 how they multiply into avalanches; Joseph Dwyer worked out in 2003 how positrons could feed the avalanche back on itself; Fermi (Briggs and colleagues, 2011) caught the positron beams in space; and the ALOFT flights (Østgaard, Marisaldi and colleagues, 2024) showed how common and restless the glows are.

Six things the line does not mean:

Antimatter from nothing: Storms make positrons from gamma rays striking nuclei, or from the decay of nuclei that gamma rays create; pairs from empty space would need a field about 5 × 10¹² times the avalanche threshold (Chapter 4).
A storms-only signal: 511 keV is the universal annihilation line: from PET scanners, radioactive decays and the centre of the Galaxy (Chapter 1).
Antimatter falling on us: Ground detections are local rises in a narrow energy window, from tens of per cent at Aragats to a fivefold rate in one burst on Etna, partly tangled with radon, and harmless (Chapters 9 and 10).
A proven downward beam: Two teams have proposed that storms drive positrons to the ground; neither proposal is established (Chapter 10).
Lightning lit by cosmic rays: One hypothesis among several, and modelling has found it falls short (Chapter 11).
A dangerous flight: The large doses are model estimates for an aircraft inside a TGF's electron source, where probably no airliner has ever been (Chapter 12).

Where things stand, as of 25 October 2026: how TGFs get their electrons, from lightning leaders, feedback or both, is argued over; how lightning starts is an open question, as is whether storms push positrons to the ground; and the 2026 Aragats preprint has no journal reference yet.

And the title's other meaning stands. The same line shines from the centre of the Milky Way, where its positrons' origin is still debated, and from a thundercloud a few kilometres up, where the rest of the evidence follows it back to the gamma rays of a runaway avalanche. Both receipts carry the same number: 511 keV, the energy in one electron's mass.

Glossary

Annihilation: An electron and a positron vanishing together; nearly at rest they become two 511 keV gamma rays flying apart.
Line (spectral line): Photons of one energy, which a detector's spectrum shows as a narrow spike; the annihilation line sits at 511 keV.
Photon (gamma ray, X-ray): A particle of light. Gamma rays are the most energetic photons and X-rays the next band down, with a loose boundary; a storm's gamma rays reach tens of MeV.
Positron, pair production: The electron's antiparticle: the same mass, the opposite charge. Storms make positrons by pair production: a gamma ray above 1.022 MeV turning into an electron and a positron near a nucleus.
Positronium (para-, ortho-): A positron and an electron briefly bound. Para-positronium decays into two 511 keV photons, ortho-positronium into three of lower energy.
Electronvolt (keV, MeV): The energy an electron gains crossing one volt. A keV is a thousand, an MeV a million; one electron's mass is 511 keV.
Runaway electron, RREA: An electron gaining more from a field than it loses to the air. A relativistic runaway electron avalanche is many of them multiplying.
Avalanche threshold, sea-level-equivalent field: The field above which runaways multiply, about 284 kV/m × n, where n is the air density relative to sea level; the break-even field, 218 kV/m × n, ignores scattering. A sea-level-equivalent field is the local field divided by n.
e-folding length: The distance over which a growing count multiplies by e ≈ 2.718; it doubles in 0.69 of it.
Relativistic feedback: Positrons and backscattered photons from an avalanche seeding new avalanches upstream; it limits the multiplication to below about 100,000.
Bremsstrahlung: "Braking radiation": gamma rays a fast electron emits when a nucleus deflects it.
Compton scattering: A photon bouncing off an electron and losing part of its energy.
TGF: Terrestrial gamma-ray flash: a burst of gamma rays from a thunderstorm lasting well under a millisecond.
Gamma-ray glow: Gamma rays from a thundercloud lasting seconds to minutes, sometimes ended by lightning.
FGF: Flickering gamma-ray flash: a pulsing flash of 20–250 ms, found on the 2023 ALOFT flights.
Energy resolution (FWHM): The width a detector gives a single energy, measured at half the peak's height.
Equivalent width: A line's photon count divided by the continuum's photons per unit energy at the line.
Radon progeny (²¹⁴Pb, ²¹⁴Bi): Short-lived decay products of radon gas, with gamma-ray lines at 352 and 609 keV; their signal at the ground rises during storms.
Sievert (µSv, mSv): The unit of radiation dose to people. Effective dose weights the organs; ambient dose equivalent, H*(10), is its measurable stand-in.
Preprint: A paper made public, for example on arXiv, before or without review by a journal.

Sources

M. S. Briggs et al., "Electron-positron beams from terrestrial lightning observed with Fermi GBM," Geophys. Res. Lett. 38, L02808 (2011), doi:10.1029/2010GL046259.
NASA, "NASA's Fermi Catches Thunderstorms Hurling Antimatter into Space" (press release, 10 January 2011), nasa.gov.
NIST, CODATA 2022 recommended values of the fundamental physical constants: electron mass energy equivalent in MeV (web page), physics.nist.gov/cuu.
W. N. Johnson, F. R. Harnden and R. C. Haymes, "The spectrum of low-energy gamma radiation from the Galactic-Center region," Astrophys. J. 172, L1 (1972), doi:10.1086/180878.
W. N. Johnson and R. C. Haymes, "Detection of a gamma-ray spectral line from the Galactic-Center region," Astrophys. J. 184, 103 (1973), doi:10.1086/152309, ADS 1973ApJ...184..103J.
M. Leventhal, C. J. MacCallum and P. D. Stang, "Detection of 511 keV positron annihilation radiation from the galactic center direction," Astrophys. J. 225, L11 (1978), doi:10.1086/182782.
N. Prantzos et al., "The 511 keV emission from positron annihilation in the Galaxy," Rev. Mod. Phys. 83, 1001 (2011), doi:10.1103/RevModPhys.83.1001, arXiv:1009.4620.
T. Siegert et al., "Gamma-ray spectroscopy of positron annihilation in the Milky Way," Astron. Astrophys. 586, A84 (2016), doi:10.1051/0004-6361/201527510, arXiv:1512.00325.
T. Siegert, "The positron puzzle," Astrophys. Space Sci. 368, 27 (2023), doi:10.1007/s10509-023-04184-4, arXiv:2303.15582.
C. T. R. Wilson, "The acceleration of β-particles in strong electric fields such as those of thunderclouds," Math. Proc. Camb. Phil. Soc. 22, 534 (1925), doi:10.1017/S0305004100003236.
J. R. Dwyer, D. M. Smith and S. A. Cummer, "High-energy atmospheric physics: terrestrial gamma-ray flashes and related phenomena," Space Sci. Rev. 173, 133 (2012), doi:10.1007/s11214-012-9894-0.
J. R. Dwyer and M. A. Uman, "The physics of lightning," Phys. Rep. 534, 147 (2014), doi:10.1016/j.physrep.2013.09.004.
NOAA, NASA and US Air Force, U.S. Standard Atmosphere, 1976 (NOAA-S/T 76-1562, US Government Printing Office, Washington, D.C., October 1976), NASA NTRS 19770009539.
A. V. Gurevich, G. M. Milikh and R. Roussel-Dupré, "Runaway electron mechanism of air breakdown and preconditioning during a thunderstorm," Phys. Lett. A 165, 463 (1992), doi:10.1016/0375-9601(92)90348-P.
J. R. Dwyer, "A fundamental limit on electric fields in air," Geophys. Res. Lett. 30, 2055 (2003), doi:10.1029/2003GL017781.
A. Chilingarian, L. Hovhannisyan and M. Zazyan, "Threshold atmospheric electric fields for initiating relativistic runaway electron avalanches: theoretical estimates and CORSIKA simulations," Geosci. Model Dev. 19, 621 (2026), doi:10.5194/gmd-19-621-2026.
L. M. Coleman and J. R. Dwyer, "Propagation speed of runaway electron avalanches," Geophys. Res. Lett. 33, L11810 (2006), doi:10.1029/2006GL025863.
L. Hovhannisyan, "Relativistic runaway electron avalanches: unified density-dependent scaling and transport," Geosci. Model Dev. 19, 9289 (2026), doi:10.5194/gmd-19-9289-2026.
T. C. Marshall, M. P. McCarthy and W. D. Rust, "Electric field magnitudes and lightning initiation in thunderstorms," J. Geophys. Res. 100, 7097 (1995), doi:10.1029/95JD00020.
T. C. Marshall et al., "Observed electric fields associated with lightning initiation," Geophys. Res. Lett. 32, L03813 (2005), doi:10.1029/2004GL021802.
N. A. Kelley et al., "Relativistic electron avalanches as a thunderstorm discharge competing with lightning," Nat. Commun. 6, 7845 (2015), doi:10.1038/ncomms8845.
J. Schwinger, "On gauge invariance and vacuum polarization," Phys. Rev. 82, 664 (1951), doi:10.1103/PhysRev.82.664.
G. J. Fishman et al., "Discovery of intense gamma-ray flashes of atmospheric origin," Science 264, 1313 (1994), doi:10.1126/science.264.5163.1313.
D. M. Smith et al., "Terrestrial gamma-ray flashes observed up to 20 MeV," Science 307, 1085 (2005), doi:10.1126/science.1107466.
M. Tavani et al., "Terrestrial gamma-ray flashes as powerful particle accelerators," Phys. Rev. Lett. 106, 018501 (2011), doi:10.1103/PhysRevLett.106.018501.
J. R. Dwyer and D. M. Smith, "A comparison between Monte Carlo simulations of runaway breakdown and terrestrial gamma-ray flash observations," Geophys. Res. Lett. 32, L22804 (2005), doi:10.1029/2005GL023848.
S. A. Cummer et al., "The source altitude, electric current, and intrinsic brightness of terrestrial gamma ray flashes," Geophys. Res. Lett. 41, 8586 (2014), doi:10.1002/2014GL062196.
T. Gjesteland et al., "Observation of intrinsically bright terrestrial gamma ray flashes from the Mediterranean basin," J. Geophys. Res. Atmos. 120 (2015), doi:10.1002/2015JD023704, arXiv:1605.05529.
M. Pallu et al., "Estimation of radiation doses delivered by terrestrial gamma ray flashes within leader-based production models," J. Geophys. Res. Atmos. 126, e2020JD033907 (2021), doi:10.1029/2020JD033907 (accepted manuscript at HAL, insu-03177542).
T. Gjesteland et al., "Confining the angular distribution of terrestrial gamma ray flash emission," J. Geophys. Res. Space Phys. 116 (2011), doi:10.1029/2011JA016716.
B. J. Hazelton et al., "Spectral dependence of terrestrial gamma-ray flashes on source distance," Geophys. Res. Lett. 36 (2009), doi:10.1029/2008GL035906.
B. G. Mailyan et al., "The spectroscopy of individual terrestrial gamma-ray flashes: constraining the source properties," J. Geophys. Res. Space Phys. 121 (2016), doi:10.1002/2016JA022702.
T. Neubert et al., "A terrestrial gamma-ray flash and ionospheric ultraviolet emissions powered by lightning," Science 367, 183 (2020), doi:10.1126/science.aax3872.
J. R. Dwyer, "Energetic particles produced by thunderstorm electric fields," J. Geophys. Res. Atmos. 130, e2024JD042193 (2025), doi:10.1029/2024JD042193.
J. R. Dwyer et al., "Positron clouds within thunderstorms," J. Plasma Phys. 81, 475810405 (2015), doi:10.1017/S0022377815000549, arXiv:1505.03782.
P. Kochkin et al., "In-flight observation of positron annihilation by ILDAS," J. Geophys. Res. Atmos. 123, 8074 (2018), doi:10.1029/2018JD028337.
G. K. Parks, B. H. Mauk, R. Spiger and J. Chin, "X-ray enhancements detected during thunderstorm and lightning activities," Geophys. Res. Lett. 8, 1176 (1981), doi:10.1029/GL008i011p01176.
M. McCarthy and G. K. Parks, "Further observations of X-rays inside thunderstorms," Geophys. Res. Lett. 12, 393 (1985), doi:10.1029/GL012i006p00393.
M. Marisaldi et al., "Highly dynamic gamma-ray emissions are common in tropical thunderclouds," Nature 634, 57 (2024), doi:10.1038/s41586-024-07936-6; accepted manuscript at NASA NTRS 20240010779.
N. Østgaard et al., "Gamma ray glow observations at 20-km altitude," J. Geophys. Res. Atmos. 124, 7236 (2019), doi:10.1029/2019JD030312.
N. Østgaard et al., "Flickering gamma-ray flashes, the missing link between gamma glows and TGFs," Nature 634, 53 (2024), doi:10.1038/s41586-024-07893-0; accepted manuscript at NASA NTRS 20240010723.
I. Bjørge-Engeland et al., "Evidence of a new population of weak terrestrial gamma-ray flashes observed from aircraft altitude," Geophys. Res. Lett. 51, e2024GL110395 (2024), doi:10.1029/2024GL110395.
A. Fuglestad et al., "The source brightness distribution of terrestrial gamma-ray flashes from the ALOFT flight campaign," J. Geophys. Res. Atmos. 131, e2025JD045235 (2026), doi:10.1029/2025JD045235.
R. U. Abbasi et al. (Telescope Array), "Gamma ray showers observed at ground level in coincidence with downward lightning leaders," J. Geophys. Res. Atmos. 123, 6864 (2018), doi:10.1029/2017JD027931.
Ø. Færder et al., "Numerical parameter-space studies of various types of thundercloud gamma-ray emissions," J. Geophys. Res. Atmos. 131, e2025JD045214 (2026), doi:10.1029/2025JD045214.
A. Ore and J. L. Powell, "Three-photon annihilation of an electron-positron pair," Phys. Rev. 75, 1696 (1949), doi:10.1103/PhysRev.75.1696.
K. Iwata, R. G. Greaves and C. M. Surko, "γ-ray spectra from positron annihilation on atoms and molecules," Phys. Rev. A 55, 3586 (1997), doi:10.1103/PhysRevA.55.3586 (line widths as reviewed in ref. 48).
G. F. Gribakin, J. A. Young and C. M. Surko, "Positron-molecule interactions: resonant attachment, annihilation, and bound states," Rev. Mod. Phys. 82, 2557 (2010), doi:10.1103/RevModPhys.82.2557, arXiv:1009.4069.
D. Umemoto et al., "On-ground detection of an electron-positron annihilation line from thunderclouds," Phys. Rev. E 93, 021201(R) (2016), doi:10.1103/PhysRevE.93.021201.
T. Enoto et al., "Photonuclear reactions triggered by lightning discharge," Nature 551, 481 (2017), doi:10.1038/nature24630, arXiv:1711.08044.
A. Ursi et al., "Results of the first ESTHER summer campaign: detection of an intense positron burst during a summer thunderstorm on Mount Etna," Atmosphere 17(1), 20 (2026; online 24 December 2025), doi:10.3390/atmos17010020.
A. Chilingarian and B. Sargsyan, "Atmospheric positron flux modulation during thunderstorms," Phys. Rev. D 109, 062003 (2024), doi:10.1103/PhysRevD.109.062003.
A. Chilingarian, B. Sargsyan and M. Zazyan, "An enormous increase in atmospheric positron flux during a summer thunderstorm on Mount Aragats," Radiat. Phys. Chem. 222, 111819 (2024), doi:10.1016/j.radphyschem.2024.111819.
A. Chilingarian and B. Sargsyan, "Positron acceleration in thunderstorms" (preprint, submitted 27 June 2026, posted July 2026; no journal reference as of 25 October 2026), arXiv:2607.11902.
R. J. Livesay, C. S. Blessinger, T. F. Guzzardo and P. A. Hausladen, "Rain-induced increase in background radiation detected by Radiation Portal Monitors," J. Environ. Radioact. 137, 137 (2014), doi:10.1016/j.jenvrad.2014.07.010.
M. J. Martin, "Nuclear data sheets for A = 208," Nucl. Data Sheets 108, 1583 (2007), doi:10.1016/j.nds.2007.07.001 (²⁰⁸Tl gamma-ray lines, via the IAEA LiveChart of Nuclides, www-nds.iaea.org).
A. V. Gurevich and K. P. Zybin, "Runaway breakdown and electric discharges in thunderstorms," Phys.-Usp. 44, 1119 (2001), doi:10.1070/PU2001v044n11ABEH000939.
A. V. Gurevich and K. P. Zybin, "Runaway breakdown and the mysteries of lightning," Physics Today 58(5), 37 (2005), doi:10.1063/1.1995746.
W. Rison et al., "Observations of narrow bipolar events reveal how lightning is initiated in thunderstorms," Nat. Commun. 7, 10721 (2016), doi:10.1038/ncomms10721.
F. Lyu, S. A. Cummer, Z. Qin and M. Chen, "Lightning initiation processes imaged with very high frequency broadband interferometry," J. Geophys. Res. Atmos. 124, 2994 (2019), doi:10.1029/2018JD029817.
J. N. Tilles et al., "Fast negative breakdown in thunderstorms," Nat. Commun. 10, 1648 (2019), doi:10.1038/s41467-019-09621-z.
I. Kolmašová et al., "Properties of positive narrow bipolar events observed in South-Eastern France," J. Geophys. Res. Atmos. 131, e2025JD045415 (2026), doi:10.1029/2025JD045415.
V. P. Pasko et al., "Photoelectric effect in air explains lightning initiation and terrestrial gamma ray flashes," J. Geophys. Res. Atmos. 130, e2025JD043897 (2025), doi:10.1029/2025JD043897.
V. P. Pasko, S. Celestin and A. Bourdon, "Weak positive discharges leading to lightning initiation," Geophys. Res. Lett. 53, e2026GL125703 (2026), doi:10.1029/2026GL125703.
S. A. Behnke et al., "Introducing a grand challenge for lightning modeling," Bull. Am. Meteorol. Soc. 107, E869 (2026), doi:10.1175/BAMS-D-25-0021.1.
J. R. Dwyer et al., "Estimation of the fluence of high-energy electron bursts produced by thunderclouds and the resulting radiation doses received in aircraft," J. Geophys. Res. 115, D09206 (2010), doi:10.1029/2009JD012039 (dose curve as reproduced in ref. 11, Fig. 29).
US Federal Aviation Administration, "In-flight radiation exposure," Advisory Circular 120-61B (21 November 2014), Tables 1 and 2, faa.gov.
ICRP, "The 2007 Recommendations of the International Commission on Radiological Protection," ICRP Publication 103, Ann. ICRP 37(2–4) (2007), icrp.org.
M. Pallu et al., "Radiation risk assessment associated with terrestrial gamma ray flashes for commercial flights," J. Geophys. Res. Atmos. 128, e2022JD037569 (2023), doi:10.1029/2022JD037569.
NIST, X-ray mass attenuation coefficients, dry air (web table), physics.nist.gov.
ICRP, "Conversion coefficients for radiological protection quantities for external radiation exposures," ICRP Publication 116, Ann. ICRP 40(2–5) (2010), icrp.org.