reentry

A graphite heat shield, simulated.

We heat a graphite slab from room temperature to 10,000 K and track every atom that leaves. Orb-v3, a machine-learned potential, computes the forces.

Video demo

The bottom layer stays frozen and stands in for the cold shield underneath. The rest heats to 10,000 K in one picosecond. The top layers buckle, unzip into carbon chains and drift off.

time 0.00 psslab 286 Kfree 0escaped 0

intact graphite chain dangling end sp³ free atom

Rendered from the saved trajectory · one frame per 10 fs · colour counts carbon neighbours within 1.9 Å

Four picoseconds

10,000 K graphite sublimes near 4,000 K 0 ps 1.1 3.6 4 ps

Six flat graphene layers, slate grey, at 286 K — 0.00 ps · 286 K
Six graphene layers, all intact.

The top layers buckle and amber carbon chains form, at 7,714 K — 1.10 ps · 7,714 K
The lattice buckles and chains form. The first fragment broke free at 1.06 ps.

Chains and fragments fill the space above a thin remaining layer, at 10,120 K — 3.60 ps · 10,120 K
86 atoms float free of the shield. 32 have left the box.

By 4 ps, 41 atoms have left and 40 more float free, 22.5% of the atoms that can move. They leave as chains and clusters. In real carbon vapour, C₂ and C₃ dominate.

Real graphite sublimes near 4,000 K. Here the first fragment leaves at 7,450 K, since 4 ps is too short to see it happen at 4,000 K. Running hotter makes it about a thousand times faster.

The run

// results/run.json
{
  "model": "orb_v3_direct_20_omat",
  "n_atoms": 432,
  "cell": [14.76, 14.76, 52.75],
  "dt_fs": 0.5,
  "stride": 10,
  "bond_cutoff": 1.9,
  "fragment_max": 8,
  "wall_s": 5965.3
}

Bottom layer of 72 atoms frozen · 300 K, then a 1 ps ramp to 10,000 K, held to 4 ps · 99 minutes on an M2 CPU

Limitations

Real reentry ablation runs on oxygen. The hot gas burns the surface into CO. So we fired oxygen atoms at a 3,300 K slab at 5 and 10 eV.

Red oxygen atoms linked in chains cover a small graphite slab; the panel shows 200 oxygen atoms fired and 150 bound

5 eV probe at 1 ps · 200 oxygen atoms fired, 150 bound, no CO

The oxygen stuck, made no CO and formed O₂ to O₇ chains that glued the layers together. This checkpoint learned from inorganic crystals and doesn't know this chemistry. Don't use it for oxidation. The orbmol checkpoints are next.

Two more caveats. This checkpoint predicts forces without energies, so energy drifts and the thermostat makes up for it. And the slab is an endless sheet with no gas or pressure above it.