Cleave
A sheet under tension, notched at one edge. The light is the elastic energy the material is holding. The crack is the dark path that releases it, and how fast it runs decides whether it stays one crack or comes apart.
Nothing here is drawn. The sheet is a triangular lattice of point masses joined by springs; a spring carries force until its strain passes a threshold and then it is gone. That is the entire fracture model. The crack accelerating, choosing its path, and splitting are all consequences of it.
Turn the tension down far enough and the crack will not move at all. There is a real threshold: below it the energy released by opening the next bit of crack is less than the energy needed to break the bonds, so nothing happens. Above it, the extra energy goes into speed.
The speed has a ceiling, and it is the Rayleigh wave speed — the speed of the surface waves running along the crack’s own new faces. A crack cannot outrun the signal that tells the material behind it to relax. For this lattice that ceiling works out at 0.5705 in lattice units, computed from the spring constants before the simulation was ever run, and the crack never passes it.
What happens instead is the interesting part. Push the tension high and the crack stops being a line. It throws off short side branches, most of which die immediately, and the energy that would have gone into forward speed goes into making extra surface. Held at low tension the crack stays within about half a row of the centreline; at high tension it spreads across eight or nine. That was measured against a control at matched geometry, five runs each, and the two populations do not overlap.
The honest limits. Real brittle solids do this too, and the measured onset is around 0.36 of the Rayleigh speed in PMMA and 0.40 to 0.44 in soda-lime glass (Sharon and Fineberg, 1996 and 1998). This lattice breaks up in a comparable range, and that resemblance should not be read as agreement, because the mechanism is different: a discrete lattice breaks up largely through lattice trapping, an artefact of being made of discrete bonds, whereas the current continuum account of PMMA attributes it to a genuine elastic nonlinearity very close to the tip. Matching numbers would not mean matching causes. A triangular lattice is also only mostly isotropic — six-fold symmetry reduces the tendency of a crack to follow the mesh but does not remove it, and the randomised bond strengths here are doing part of that work. There is no plasticity, no crazing, no rate dependence and no third dimension. It is a real dynamical solve of an idealised solid, and it is not a prediction about any particular material.