Universal Fault-Tolerant Computation on Three-Dimensional Layer Codes at Optimal Distance
Year: 2026
Type: preprint
Abstract: Layer codes are three-dimensional quantum error-correcting codes that can reach the largestdistance possible for their size when every check acts on nearby qubits. Blocks of Θ(N3) qubits builtfrom balanced good input codes of length N have distance d = Θ(N2). We prove a threshold theoremfor universal fault-tolerant quantum computation on these codes that uses only nearest-neighboroperations and keeps the full distance d in every logical operation. The difficulty is measuring logicaloperators. Existing gauging and surgery methods guarantee the distance through expansion of thegraph of helper qubits. A graph expands when every region of it has many edges leaving it, comparedwith the size of the region. No large graph of nearest-neighbor connections in three dimensionsexpands. A measurement layer of bounded thickness provably lowers the distance of joint layer-codemeasurements to O(N). We use the shape of the logical operators instead. We give an explicitrealization of the Williamson–Baspin layer code for any CSS input. In it, every pure-X logical classhas a representative on a single flat layer at every height. Every pure-Z class has one in a singleflat transverse plane at every position. We call the qubits of such a layer or plane a slab port. Anearest-neighbor grid that fills the block expands relative to these ports, that is, compared withthe number of port qubits in each region. It therefore measures them at the full distance with aconstant number of helper qubits per edge. We add encoded resource transfer, distillation, and alocal schedule of check measurements. A circuit of depth T on W logical sites in a box of boundedaspect ratio then uses O(W f + f3/2) physical qubits and O(T f 3/2Ψ) depth under local-stochasticnoise. Here Ψ = O(flog2 15) and f = Θ(log(W T /ε)) for target error ε, with a decoder that finds thefewest faults consistent with the data. A hierarchical cluster decoder that runs in polynomial timegives the same bounds with f = O(log6.65(W T /ε)). An exact form of the distance condition showsthat much smaller (compressed) grids work at finite size. For two blocks built from the [[4, 2, 2]]code, a grid of 107 qubits instead of 369 keeps circuit fault distance 5. Circuit-level simulations ofexplicit layer-code blocks and runs of small blocks on IQM and IBM superconducting processorstest these mechanisms. They show that a thin connection lowers the distance, that a spanning gridrestores it, and that compressed grids keep it.
Author: Katoch, Aryaman
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225990466
