This bell is not a recording and not a sample. Its voice is the eigenmode spectrum of a real bell shape cast in ice — Young's modulus 9.3 GPa, density 917 kg/m³ — computed by finite-element modal analysis and re-solved at twenty-two stages of melting. Every tap excites those modes the way a soft mallet would: the partial amplitudes are the mode shapes at the strike point, the decay times come from measured internal friction of ice near 0 °C.
As the wall thins the stiffness dies faster than the mass, so the bell goes flat — this one falls nearly six semitones before its last strike, and the full melt sweep spans more than a fifth. And each strike is a stress test, literally: the page computes the peak tensile stress of your blow against the strength of ice — which itself falls from ≈2.0 toward ≈0.8 MPa as the ice warms (Petrovic 2003). Rising stress, falling strength: the two numbers in the readout are walking toward each other. The bell can only break if you ring it.
The full piece rings four of these bells in change-ringing rounds (Plain Bob Minimus) while they melt: the treble shatters first, silences open in the grid where dead bells should speak, and the tenor tolls alone before its own last strike. Path-traced in Blender, scored entirely by the physics above.
Modeled: free-free elastic modes (a bell hung at the crown), uniform surface recession on both faces of the wall, constant-Q damping sliding from cold-cast ice (Q≈300) to temperate ice (Q≈35, Clee 1969; Meyer et al. 2019), mallet contact rolloff, and a root-sum-square modal stress estimate against a tensile strength that weakens with warming (2.0 → 0.8 MPa). Not modeled: gravity sag, sunlight, water films re-freezing, the exact crack path — a real ice bell would be messier and lovelier. The doublet warble is added explicitly (±0.2%): a perfect axisymmetric bell does not beat, and mine is mathematically perfect until you break it.