Skip to main content
This page is the demo

A half-full array loads by chance.
Plan it defect-free — provably, at minimum cost.

A neutral-atom processor loads atoms stochastically into a grid of optical traps — each site fills with probability about a half — then mobile tweezers rearrange them into a defect-free target register before the computation. Which atom moves to which site is a combinatorial optimization: the linear sum assignment problem — give every target site a distinct atom, minimizing total move distance — solved exactly by the Hungarian algorithm.

Here it is in pure integer Rust, byte-identical on every machine, turning a low-yield load into a defect-free block and sealing the plan into a signed wai.quantum.rearrange receipt. The success verdict — every target site covered by a valid bijection of atoms actually present — is re-checkable from the bound configurations, and the same check rejects a tampered plan. Honest boundary: a geometric/combinatorial model — the solver and its optimality are exact; tweezer transport, loss, and AOD parallelization are out of scope.

Rearrange to a defect-free register · Hungarian / LSAP

—
—

Yield transformation · the certificate is real

—
—

The combinatorial solver that prepares the register — signed, budgeted, verifiable

The plan panel loads a trap array at random and solves the assignment: every site of a centered defect-free block is matched to a distinct loaded atom minimizing the total Manhattan move distance (the Hungarian algorithm, proven optimal in the tests against brute force), and the arrows show the atoms converging on the target. The receipt binds (initial occupancy, target sites, move plan) → success plus the plan's cost. The yield panel shows the point of it: the target region fills only partway by chance, and completely after the plan — and dropping a single move makes the same deterministic check report not defect-free, so the certificate can say no. A register-preparation plan you can put in an auditor's hands, reproduce from the git repo, and verify with a public key. Reference solver + spec in the open-standards repo, Apache-2.0.