The actual adaptive computational readout #
Reading a seed and retaining its CL prefix and selected next coordinates is exactly the prefix projector tensored with the next local Z measurement. The outcome type remembers the prefix, so the selected register can vary.
Stated in a bipartite model (Phase 4 of planning/mipco-track.md): the readouts are register
operators over ℂ, and their relabelling identities (registerOp_readout_fst,
registerOp_readout_pair) are unchanged. They enter the first player's algebra Matrix I I 𝒜 as
smulKron 1, so the factorizations through a register split (registerReadout_wZ) and through the
prefix reinsertion prefixResidualOp (prefixResidualOp_readout, adaptiveZ_readout_factor) are
identities of block matrices over any algebra 𝒜. The one step they share: X ⊗ (P ⊗ N), read
along e : I ≃ J × R, is regSplitHom e applied to the block matrix (X ⊗ N) ⊗ P.
The Z readout of a linear map on the selected coordinates of a register is the readout of the map on those coordinates, as a register operator.
A register readout reinserted behind its prefix projector is the joint readout of the prefix and of the residual label.
The actual seed coarse-graining keeps the CL prefix and its selected next register.
Equations
- MIPRE.Introspection.adaptiveZOutcome P k x = ⟨(MIPRE.CL.CLFun.truncate k P).eval x, (MIPRE.Introspection.coordinateRestrict (P.factorOfPrefix k ((MIPRE.CL.CLFun.truncate k P).eval x))) x⟩
Instances For
Exact factorization of the actual adaptive Z readout, including zero-weight prefixes.