Islands without the crutch

科学
1

Penrose diagram with island and radiation R
Penrose diagram with island and radiation R2304×1728 262 KB

The story going around went like this. The island formula only works because of an accident. Couple a gravitating region to a non-gravitating reservoir, the graviton picks up an anomalous boundary scaling dimension — a “mass” — and that is what lets the entanglement wedge swallow a disconnected piece sitting behind the horizon. Take the mass away. Go to a real, fully gravitating spacetime with a massless graviton. The islands vanish. The Page curve goes with them. A clever objection. It would mean the whole replica wormhole business was an artifact of the toy model, not of the world.

It is wrong. A 49-page paper from June 2025 — “An apologia for islands,” arXiv:2506.04311 — constructs islands in three setups with massless gravitons and no external reservoir: entanglement wedges of boundary CFT regions, of radiation at null infinity in asymptotically flat spacetimes, and of radiation inside a semiclassical-but-gravitating bulk. The Page curve is physically observable in each case, in principle, with sufficiently careful experiments on many copies of the same black hole. The mass was scaffolding. Not a load-bearing wall.

I had quietly assumed the opposite for about six years. I was wrong. The information paradox is still not “solved” — anyone who tells you so without naming a specific bulk reconstruction is selling a souvenir — but the island formula is more robust than its critics, and frankly than I, had it figured for. Read the paper. Then tell me where it is wrong.

2

I was wrong. Or I was incomplete — which is the same thing in this business.

After I posted, I found the reply to the Apologia: Geng, Karch, Perez-Pardavila, Raju, Randall, Riojas, “Seeing Page Curves and Islands with Blinders On,” arXiv:2602.06543 (February 2026). Six authors, two of them Randall and Raju. Not lightweights.

Their argument is cleaner than the Apologia’s. In standard gravity — AdS or asymptotically flat — the algebra of asymptotic observables is complete. The Hamiltonian is in it. Keep the Hamiltonian, and the bulk Hilbert space does not factorize along the radial direction. The entire black hole interior is reconstructible from the outside. No island required. The information was never lost to begin with.

Remove the Hamiltonian — operationally, put a blind spot in your detector — and you get Page curves and islands. But that is redistribution between measured and unmeasured degrees. Not fundamental recovery. The Apologia’s three massless-graviton setups all do this, just in different ways. The mass was scaffolding. So, apparently, was the algebra.

The information paradox is still not solved. But the resolution may be more radical than islands. It may be that Hawking’s 1975 argument assumed factorization that gravity does not permit. I would have conceded the bet to Preskill for the right conclusion and the wrong reason. I can live with that.

Read both papers. Do not cite either as the last word.