How much precision is lost when quantum demonstrations are measured too early?
An unknown unitary acts on fresh random states. Each example supplies two registers: an input specimen and its transformed partner. The preparation record is private. The pair can be separable across its registers, yet retaining coherence within a pair and across different pairs gives two distinct precision gains.
The source is the full-rank separable family
where
| Measurement access | Exact coefficient |
|---|---|
| Collective measurement across examples | |
| Arbitrary joint measurement of each pair; classical records between pairs | |
| Separable measurement effects within each pair; classical records between pairs |
Joint pair readout improves the last coefficient by exactly
These are fixed-dimension high-precision limits: first fix
The sister project Quantum Learning Correspondence removes the input–output matching across a whole batch of the same demonstrations. Its exact kernel gives an information ceiling. The joint result goes further: as one anonymous batch grows at fixed dimension and visibility, separate collective measurements of the two lists preserve all surviving statistical information asymptotically. The full experiment becomes one classical Gaussian covariance observation.
Read the self-contained application here and the authoritative general proof in Correspondence. Each side may retain its full list coherently, so this is different from the per-pair separable class in the table. The result proves uniform recovery and bounded-decision equivalence; it supplies no efficient receiver, convergence rate or automatic new precision coefficient for many growing batches.
Start with the scientific route. The account joins sharp measurement-information budgets to a global converse that permits biased, adaptive learners, and to actual attaining designs.
| Question | Scientific account |
|---|---|
| What is supplied, measured and estimated? | Model and exact theorem |
| Why does the converse include biased adaptive estimators? | Global risk bound |
| What is the optimum with separate-register measurements? | Separable readout |
| What does one coherent pair buy? | Joint-pair readout |
| Why is the collective coefficient attainable without knowing the unitary? | Collective attainment and localization |
| What survives a growing anonymous batch? | Separate-list collective readout |
| What do the gains mean? | Interpretation |
| What must preparation, storage, control and output cost? | Resources |
| Which results are inherited? | Related work and bibliography |
For example, at
The common educational source with Correspondence is Dénes Petz, Quantum Information Theory and Quantum Statistics (2008). The learning route, 30-milestone map, Petz bridge and worked tutorial connect quantum statistical estimation, conditional expectations and sufficiency to the paired precision hierarchy and its anonymous-batch application.
python3 -m pip install -r verification/requirements.txt
python3 verification/run_all.pyThe verification guide explains the direct matrix, derivative, covariant-seed and group identities checked, together with their limits. Finite computations support the constructions; the written proofs establish the universal and asymptotic conclusions.
Faithful separable process probes, Werner-state quantum Fisher information, and coherent-readout advantages are established foundations. The contribution here is the sharp full-unitary three-tier learning comparison with matching acquisition schemes, together with its application of Correspondence’s general anonymous-batch sufficiency theorem. Shared theorem ownership and inherited tools are stated in the related-work account. This is a supplied-data experiment: the learner receives neither the hidden preparation record nor active access to the unknown unitary.
This repository serves as a record of the work and a guide for the author’s self-directed learning. For discussion or potential collaboration, please contact Ruge Lin at gogoko699@gmail.com.