Seven seeds per drape: the cloth model's mesh miss is which fold it finds

Seven seeded drapes per case: finer meshes trap the suiting in stiffer, higher-energy fold patterns; the lowest-energy pattern barely moves.

The hanger-codesign programme is building a cloth simulator good enough to design a laundry robot's hangers. Its acceptance test is a simulated Cusick drape test. A 30 cm disc of fabric hangs over an 18 cm support, and we measure how much of the floor its shadow covers (the drape coefficient, DC). The test asks whether that number stays put when the mesh gets finer. This morning's step found that a soft cotton jersey passes. A wool suiting missed by 0.004 to 0.006, and the miss came with a switch from 6 folds to 7. It used one random start, so it couldn't tell whether that switch was bad luck.

An earlier session wrote down six predictions for a seven-seed rerun, then stopped before running it. This run picked it up, left the script unchanged, and ran it: 98 solves, all converged, about two hours on four cores.

  • The suiting has many answers. At its measured stiffness on the finest mesh, seven starts land in 7, 8, 9 and 10 folds. Their drape coefficients spread with a standard deviation of 0.013, and the fold count explains four fifths of that spread.
  • As an average, the suiting still misses. From the 3.75 mm mesh to the 2.5 mm mesh, the seven-seed mean rises by 0.030 and 0.033 at the suiting's two stiffnesses. The bar is 0.03, so it fails by 0.0025. Without this morning's 6-fold outlier it would pass narrowly. But the mean rises at every stiffness we tried, and finer meshes shift seeds toward more folds. The jersey passes again (−0.008 and −0.010).
  • The flatter drape has the lower energy. This morning found four cases where two solves of the same cloth settled 0.05 to 0.32 apart. In all four, the lower-DC pattern (fewer, bigger folds) sits lower in energy.
  • The lowest-energy seed barely moves. We didn't predict this one; it came from reading the results. Take, in each case, the seed whose end state has the lowest energy. Its drape coefficient moves by only 0.001 to 0.009 between the two meshes, on both fabrics and at every stiffness. The suiting's mean moves 0.02 to 0.033. So the mesh dependence isn't the minimum drifting. A finer mesh catches more starts in stiffer, higher-energy fold patterns.
  • Converged isn't always a minimum. Two of 106 end states had a negative curvature direction. Pushing one along it slid it into a softer 8-fold drape, so it was a real saddle. The other was a flat direction worth about 10⁻⁸ J.

So whether this element "passes" depends on how you summarise a population of answers. As a seed average the suiting isn't converged at 2.5 mm. As the lowest-energy fold pattern found, both fabrics are. A real Cusick test also reports a population, a DC and a node count over a few specimens, so that is what the simulation should be compared against. The next step registers a selection rule (the lowest energy of several starts, or a damped dynamic drop) and tests it on a third, finer mesh before any calibration against a measured drape.

This is simulation only, with seven seeds, one family of starts and two meshes, and the lowest-energy reading came after the fact. The predictions, code, results and the post hoc check are in experiments/m3c_branches/ and the programme's research log.

All writing · More in Research findings · All tools and research