Quilt friction
On the fourth evening the quilt still looked like cloth until you shoved it. Then it swam. The batting had heft. The table did not. Four commits later, sitting on the glass is contact, a fold has nap, and a wrinkle on a flat iPad stops.
The watery shove
Chapter 13 had already given the quilt heft: stiffer in a fold, heavier in the air, a little thickness so a fold’s edge shades the layer under it. Shoved, a loose quilt was supposed to slide a few millimetres and settle. On the iPad it whispered. It kept going. It looked like water.
sometimes it seems almost watery and that might just be lack of friction underneath
Oct 4, evening
It was. Three things can slow a particle. Air drag only kills motion through the sheet, along the normal. Damping is a light global hush. The only force meant to keep the quilt where it lands is the glass. That force was a keep-fraction: when a particle went through the plane, this substep’s in-plane move kept 35 percent and threw away 65. If that law had run twelve times a frame, the quilt would have glued itself. The watery look meant the law was not running.
Sitting on the plane is ice
Friction ran only when a particle’s height was strictly below the floor, the glass plus six millimetres of half-thickness. After a hit, the particle was placed on the floor. Next substep, stretch, pins, or self-collision often held it there. The test failed. It skated. Gravity had to punch it through again before the glass grabbed. A sheet that looked as if it were on the table was often numerically at the table, which the code treated as ice.
The test that claimed the glass had friction tilted gravity a little and asked that a point move less than five centimetres in two seconds. That punch goes through every substep, so the keep-fraction fired and the test passed. A wrinkle skating on a flat iPad does not punch. The test could not see the watery case.
And a fold sliding on itself had no nap at all: self-collision is a ball push along the line between two particles, with no shear. Cotton on cotton was ice on ice. That is why a hitch looked as if it were swimming around the twist instead of catching.
Not a knot
A severe wring of fabric looks like a knot. It is not one. A quilt is a disk in three-space, and a disk cannot knot. Knot theory is about closed curves. What you have is a crumple: self-contact packed so tight the sheet locally inverts. Geometry people call the sharp bits ridges and d-cones. Cloth papers call the failure interpenetration, or an impact zone. Wring, or hitch, is closer than knot.
when a severe knot of fabric (is it called a knot in curved 2-space?) happens, i get significant artifacting around the twist
Oct 4, evening
Around that hitch the solver was doing the wrong thing twice. The balls were Jacobi, so twenty non-neighbors inside one cell all shoved from the same positions and the shoves added. And at ninety and a hundred and twenty particles across, those balls were still eighteen millimetres, the 60×33 default: bigger than the in-plane spacing, so the same sheet shoved itself apart. The sparkle around the twist was inverted triangles and flipped normals, which is what a smashed grid does when it cannot draw a d-cone.
A cone, not a keep-fraction
Coulomb friction asks: is the sideways force under µ times the load? If yes, stick. If no, slide against the motion. Load matters. A heavy fold grips more than a lifted corner. The keep-fraction never saw the load, and never stuck.
The glass and the table a lip below it are one implicit plane. Contact when a particle is on or through it. Load is how far we just projected, and if we were already on it, gravity into the surface times the substep squared. Then a cone in the tangent plane: stick if this substep’s in-plane move is inside µ static times the load, otherwise slide with µ kinetic. Cotton on glass starts at 0.5 and 0.35. A millimetre-of-slide gate would have been the wrong static test: gravity only drives a particle about two hundredths of a millimetre through the glass each substep, and that gate would have frozen the hero drop’s few millimetres.
The same cone on a fold, with a slightly smaller µ, so cloth on cloth catches but does not weld. A hitch cannot shove a particle more than one layer-gap in a pass. The layer gap itself now tracks the particle spacing, 1.05 times, so 60×33 stays about twenty-five millimetres and 90 and 120 shrink with the mesh. The eighteen millimetre floor is gone.
go but keep your commits granular so we can analyze step by step
Oct 4, evening
Four commits
They went on a branch of their own, engine/friction, each a step you can check out, the CPU and the GPU in lockstep, the fabric numbers left alone.
- the glass is a Coulomb cone. Contact on or through the plane. Gravity-backed load. µs 0.5, µk 0.35.
- tests: the old tilt still holds; a wrinkle on a flat glass stops; with µ at zero the tilt still slides.
- the layer gap tracks particle spacing. No eighteen millimetre floor.
- cloth on cloth learns shear. A hitch cannot move a particle more than one gap in a pass.
Twenty-one cloth tests still green, two of them new, the watery case among them. The keep-fraction is still on the surface struct, unused. The screen-edge walls still keep half. The seven millimetre step off the glass onto the table is still a stair, not a fillet. Those wait. The batting was not retuned to fake grip. Grip is a contact law.
What the hand is for
The iPad has the stack. Drag it, drop it, crank the mesh, wring a hitch. Sitting on the plane is contact. A fold has nap. A wrinkle on a flat table stops. That is the whole of this evening’s physics, told so the next person who shoves the quilt knows why it used to swim.