Slump

what the glass can say about itself

Type has always been a picture of a material. Letters were cut in stone, cast in lead, and the words for their parts are the words for bodies: foot, shoulder, spine, ear. This page takes the picture literally. The word is rasterized in a heavy sans, sampled onto a grid seven pixels wide, and every full cell becomes two triangles of a soft body: some two thousand vertices, held together by distance constraints along the lattice and an area constraint on every triangle. That is the whole letter. There is no outline being deformed; the outline is where the triangles stop.

The material law is a Maxwell solid, the simplest thing that is elastic on short times and fluid on long ones. Each edge's rest length relaxes toward its current length with a time constant τ, so stress that lasts becomes strain that stays; the triangle rest areas never relax, so the glass conserves its area the way real glass at the slumping range conserves its volume. Push those two rules with gravity and a shelf that the glass sticks to, and you get what a kiln gives: the letters shorten, the material that leaves the top has to go somewhere and bulges at the foot, and thin strokes fold before thick ones. Cooling raises τ by five decades, from working glass to set glass, so what you hold keeps its shape while the rest keeps flowing.

Six gate families were written before the code and run against this exact module. Five held: the spring is a spring to eleven digits, the creep rate matches the Maxwell closed form to nine, the area drifts under a fifth of a percent through a slump, nothing ever passes the shelf, and the saved word continues bit for bit. The sixth did not. Kiln workers know that thick glass falls more slowly than thin, and beam theory says a viscous beam's sag rate should scale as one over thickness squared: twice as thick, a quarter the rate. This lattice gives 2.2, not 4, in the converged limit, and only 1.6 at the two solver passes per substep this page can afford. The direction is right and the magnitude is not, which means the material on this page bends less like glass than it should, and that is stated here rather than tuned away.

A second firing, the day after, took that failure apart with a sweep registered before it ran. Three causes were named with their predictions: too few cells through the thickness, cells too soft in shear, area constraints too stiff. None survived. The ratio sits between 2.2 and 2.6 from three to eighteen cells through the thickness, a lattice with both diagonals moves it by less than a tenth, softening the area constraint twentyfold moves it by a thousandth, and the converged elastic sag reads 2.44 at three hundred solver passes and at three thousand. The shortfall is not the grain, not the diagonals, not the incompressibility and not the solver's patience. It belongs to the constraint model itself, in a way this page has not yet isolated, so the material keeps its grid and the failure keeps its sentence. The same firing gave the rim its self-contact: every boundary vertex is held a margin outside every boundary edge it is not part of, so a drip that folds onto the pool lies on it instead of drawing through it. That gate asked for zero crossing rim pairs after three minutes of watched time. The run gave six, down from twenty-nine without the term, all of them kinks between neighbouring rim edges in strands stretched past twice their rest length, which no proximity test can see. Stated, not tuned.

A third firing, thirteen days later, stopped examining the solver with the solver. The same edges were read as a plain truss of springs, its stiffness written out exactly and solved in one step, with no passes and no time in it. Seven predictions were registered before that instrument existed. It reproduces this page's solver to three parts in ten thousand, so the 2.4 belongs to the lattice, and the lattice is not wrong. It is a deep beam. The test beams were thirteen and seven thicknesses long, and one over thickness squared is the law of slender beams. A stocky beam sags partly in shear, and under its own weight thickness does not help with that. With each strip's bending and shear stiffness measured on their own, the textbook sum gives the sag to under one percent with nothing fitted, once the area springs are set aside; with them on, the way the shear stiffness was measured broke, which is the half failure below. The curvature at midspan keeps the bending law to two percent while the sag does not. Made longer, the same edges climb: 2.4, 2.8, 3.0. Made finer as well, 3.8, on the way to 4. The second firing's three clues against shear were all the examiner's. Its slender beam had sagged more than two thicknesses and was hanging like a chain, which no thickness changes. Its pin held two vertices on the thin beam and one on the thick. And its sweep over the grain measured creep through this page's damping, which is a drag against the world: an air the page never said it had. On the word above, that air carries about a tenth of the weight in the first seconds and about a fortieth after a minute. Two of the seven predictions failed and a third failed by half, and each is in the record with whose failure it was. So the sentence above, that this material bends less like glass than it should, blamed the glass for the examiner's choice of law. The gate stays failed as it was written. The strokes of a heavy word are stocky, and stocky glass is what this is.

what it cannot keep

The word cannot be put back. Every step of the relaxation forgets a little of the cast shape, and unlike the heat on the previous page there is no rewind button to try, because the rest lengths are the only memory the material has and they have already moved on. The only way to keep a word standing is to hold it, and you can only hold a hand's width at a time. What you leave for the kiln, the kiln takes, at its own pace, while someone is watching.