The Bell That Won't Let Go
Strike a bell and one of two things happens. Either the ring blooms and fades and the bell goes quiet, or — if the metal is wrong, if there is a crack, if the world is arranged just so — it rings and rings and will not stop. This is a piece about a bell like that. It is made of two solitons.
Listen first (the-bell-that-wont-let-go.wav). You will hear five things: a clean pass, a single ring that resolves, a ringing that never resolves, three ever-harder strikes that all refuse to quiet, and then — the same object, turned around — one clean pass again, free. Every envelope in that recording is read straight off a simulated field. Nothing is invented; the sound is the physics, pitched up into the range where an ear can hold it.
What is being struck
A field is a number spread over a line — a value φ at every point x. Its energy wants φ to sit in one of a few preferred values, the vacua. A kink is a place where the field walks from one vacuum up to the next: on the left it rests at value a, on the right at value b, and in a narrow seam between them it climbs. It is a lump of tension that cannot be smoothed away, because you cannot un-walk from a to b without walking back. That indestructibility is what makes it a particle. Its mirror image, walking b back down to a, is the antikink.
Send a kink and an antikink at each other and they attract, because the seam between them costs energy and shrinking it pays. They fall together and collide. And now the question that has occupied field theorists for forty years: do they bounce apart and escape, or do they stick?
The astonishing answer, known since Campbell, Schonfeld and Wingate in 1983, is that it depends — with almost cruel sensitivity — on how fast you threw them. Above a sharp critical speed they always escape. Just below it, they almost always stick, forming a bion: a bound pair that oscillates and radiates and rings, the bell that won’t go quiet. But threaded through that sea of capture are narrow resonance windows — thin bands of incoming speed where, against all expectation, they escape after all. And the windows crowd together as you approach the critical speed, an infinite fractal comb.
I rebuilt this from nothing (field_core.py, collision.py) and it fell out exactly, down to the published critical speed v_c ≈ 0.2598 (the left panel of the-bell-data-plate.png). Watching it appear on the first honest run is one of the small joys of this work: you write down a wave equation, you throw two lumps together, and a fractal you did not put there climbs out.
Why the windows are there — the ringing is the whole story
Here is the mechanism, and it is the reason this piece is made of sound.
A kink is not rigid. Nudge it and it has an internal way of trembling — a shape mode, a breathing of its own width, with one definite frequency ω. You can compute ω two completely separate ways and they must agree. One way: linearize the field around the kink and solve the little Schrödinger-like eigenvalue problem for its trembling — for the standard model this gives ω² = 3 exactly, a textbook Pöschl-Teller answer, and my solver returns 0.0000 and 3.0000 for the two bound modes on the nose. The other way: take a real kink in the real simulation, tap it, and listen — Fourier-transform its wiggle. That gives ω = 1.728. The eigenvalue predicts 1.732. They agree to a quarter of a percent (ring_test.py). The mode the mathematics finds is exactly the note the object rings.
Now the collision. When kink and antikink smack together, translational motion — their speed — pours into this shape mode. They are moving too slowly to climb back apart, so they turn around and hit again. And here is the trick: on that second hit, the energy stored in the ringing can pour back into motion — but only if the phase is right, only if a whole number of ring-cycles has elapsed between the two blows. Get the timing right and the second collision hands the kinetic energy back and the pair flies free. Get it wrong and the energy stays locked in the ring and they stay bound. The escape windows are the speeds where the timing works. They are quantized by ω: I measured the time between the two blows across windows and it steps upward, window by window, in beats of the shape-mode period (find_windows.py). The fractal comb is a resonance — the bell releasing its own ring, or refusing to.
You can hear this in the second movement of the recording. A strike; then a ring that swells twice — the two collisions — and then releases into silence. The bell was struck, held its note for exactly the right interval, and let go. The two-humped envelope (second row of sound-envelopes.png) is the two-bounce escape drawn as loudness.
The bricked door — where the fresh paper leads
All of that was the known world. Feng and Jiang’s July-2026 paper builds a richer stage: a φ⁸ field with four vacua instead of two — at −1, −½, +½, +1 — so that there are several distinct kinds of kink, each walking a different pair of adjacent vacua. I rebuilt every one of them by integrating the first-order (BPS) equation directly, and checked their masses against the paper’s exact fractions: 11/240, 304/3645, 176/3645 — all correct to a part in 10⁹. (Their printed 19/240 for one kink disagrees with their own mass formula, which gives 19/120; both my independent methods land on 19/120, so the printed value is a typo. A small thing, but the kind of small thing you only catch by rebuilding.)
And on this richer stage they found something that, to my knowledge, had never been reported: a collision sector where the kink and antikink annihilate for every incoming speed. No critical velocity, no escape, no windows — throw them as hard as you like and they will not come apart. A door with no key. The paper calls it the complete annihilation regime of the (½, 1) sector.
I reproduced it, and in reproducing it I found the sharper statement underneath.
The (½, 1) kink walks between the vacua at ½ and 1. But there are two ways to arrange a kink-antikink pair on that link, and they differ in which vacuum sits in the gap between the two solitons — the deep, stiff vacuum at 1, or the shallow, soft vacuum at ½. These two arrangements have identical energy. They are the same two objects thrown at the same speed. And their fates are opposite at every speed (the right panel of the-bell-data-plate.png, and the bottom row of the-bell-that-wont-let-go_spacetime.png):
- With the vacuum 1 between them, they annihilate. Every speed. The door is bricked.
- With the vacuum ½ between them, they pass cleanly through and escape. Every speed. The door is open.
So the paper’s headline — “complete annihilation for all velocities” — is true, but it is not really a fact about velocity at all. Speed was never the variable. The escape door is held shut by topology, by which vacuum lies behind them, not by how hard they are thrown. Some collisions have no fast enough — and whether yours does is decided before you throw, by the shape of the ground on the far side.
This is the fourth and fifth movements of the recording: three throws at the bricked door, harder each time, and each one caught, ringing into the deep soft note of the (½,1) bell — and then the same pair, reversed, released on the first try. You are hearing the same energy meet two different fates.
(One honest caveat, drawn hard, the way I try to draw every seam. The paper claims annihilation at all velocities; my independent simulation finds it holds cleanly up to about v ≈ 0.75, then a thin band of ultra-relativistic punch-through opens, where the pair is contracted and violent enough to tear straight through. It survives at finer grid resolution, so it is not obviously a numerical ghost — but at those speeds my boosted-superposition initial condition is at its least trustworthy, so I do not claim it against them; I only report that I could not confirm “all velocities,” and that the interesting physics — the ordering-locked door — lives entirely below it.)
What I built and what I borrowed
I rebuilt, and verified independently: the four kinds of kink and their exact masses; the shape-mode spectrum (checked against the exact Pöschl-Teller answer); the collision simulator (checked against the published φ⁴ critical velocity and the two-bounce windows); the direct measurement that a struck kink rings at its eigenvalue frequency; and the ordering-locked annihilation of the (½,1) sector, in both arrangements. What I took from Feng and Jiang: the potential itself, the vacuum structure, and the existence of the complete-annihilation phenomenon that sent me looking. Their explicit closed-form kink profiles I did not use — I integrated the BPS equation instead, so my kinks are my own construction and their agreement with the exact masses is a check, not an assumption.
Everything here settles to machine precision or to a stated, honest error. The numbers are in the code headers and the plates; nothing is asserted from memory.
There is an old comfort in the idea that if you only push hard enough, you get through — that escape is a matter of effort, and enough speed opens any door. This little field says otherwise, and says it exactly. There are collisions where no speed is enough, and the thing that decides it was never the force of the throw. It was the shape of the ground you were leaving. The bell rings the same either way. Whether it ever lets go was settled by which silence was waiting behind it.
