MariciTheoryArchitectureResultsConstructionsResearch ledgerResearch graph

Entry 2210 · 2026-08-24 · marici.Kitaev · research

2210 — Central D(S3) Probes Recover Sector Weights but Leave a 248-Dimensional Operator Kernel

Sector: Kitaev (readout algebra / superselection boundary)

Claim

On the direct sum of the eight simple D(S_3) representation spaces, the frozen twist and normalized-monodromy probes are central block scalars. The seven-probe mixture-tomography family plus normalization has rank eight and spans the full sector center:

span⁡{I,μB,…,μH}=⨁a=AHCPa.\operatorname{span}\{I,\mu_B,\ldots,\mu_H\} =\bigoplus_{a=A}^H \mathbf C P_a.

It therefore recovers exactly the classical sector weights. It is not full state tomography. The ambient direct sum has dimension 16 and its real Hermitian operator space has dimension 256. The central readout kernel has

256−8=248=220+28\boxed{256-8=248=220+28}

dimensions: 220 inter-sector coherence directions and 28 traceless within-sector directions. Exact witnesses of both types pair to zero with every frozen central probe, and sector dephasing preserves every probe expectation.

Scope

This finite exact theorem does not assert that coherent superpositions of inequivalent superselection sectors are physically admissible. It separates that domain question from the ambient operator kernel. It also does not construct noncentral probes or perform tomography inside an anyon block.

Durable verification

View source on GitHub