An encoder receives one unknown quantum register. It must store the register before learning which local X or Z measurement will be requested. Classical records are free; only the quantum memory is limited.
The central question is when retaining original qubits is optimal, and when a collective encoding helps. The supplied results give exact common-accuracy costs through four inputs, a complete allocation rule for one memory qubit at every input size, and a collective advantage for a specified unequal-accuracy task.
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.
The teaching anchor is Gühne, Haapasalo, Kraft, Pellonpää and Uola, Incompatible measurements in quantum information science, Reviews of Modern Physics 95, 011003 (2023). The learning path selects the relevant sections and connects them to this project. It is the only external tutorial required by the path; the additional arguments are explained here.
| Pass | Route | Purpose |
|---|---|---|
| The question | This page, then the learning path and first tutorial | Understand the memory task and work the two-input example |
| The proof | Memory to proofs, then the core argument | Connect an arbitrary encoder to the exact memory bounds |
| The evidence | Scientific scope, prior comparison and evidence freeze | Check assumptions, attribution and proof evidence |
With only classical memory, a single qubit can support later noisy X and
Z measurements with contrasts
For two inputs, keep one qubit chosen by a fair coin and measure the other with the compatible X/Z measurement. Record the choice and the measurement outcome. Each site is retained half the time, so all four possible queries have contrast
Here
Reading the figure. A is the compatible classical boundary; B is perfect retention. Their midpoint C is the contrast at each site when exactly one of two input qubits is retained, chosen with equal probability.
The worked tutorial derives the disk, constructs this protocol and then treats unequal accuracies.
| When | What happens |
|---|---|
| Before the query | An arbitrary collective encoder acts on one unknown |
| What survives | A quantum register of dimension at most |
| After the query | A site |
The guarantee holds for every input and each allowed query. The quantum cap is branchwise, not an average. All branches are accepted; there are no extra copies, source reaccess, preshared entanglement or free quantum bypass. The decoder may depend on the query. Its task is to reproduce the requested measurement statistics, not to reconstruct the whole input state.
Retaining a random subset of
The converse covers arbitrary collective encoders, spectra and binary
decoders. These are all 14 integer-qubit budgets through four inputs.
The same formula holds for
For arbitrary local contrasts
where
Here
A complete encoder for 31 inputs using five memory qubits has
It exceeds the defined original-site-retention class, including joint measurements of discarded sites. This is a class separation, not a claim of unrestricted optimality or improved common accuracy.
The balanced-spectrum theorem and stability
are optional structural companions. The general common-accuracy optimum,
sharp entropy bound and asymptotic rate remain open. The smallest remaining
finite common-accuracy case is
The core argument connects the selected proofs. The focused source map distinguishes established frameworks and ingredients from the supplied evaluations. The manuscript background guide connects the required concepts to those sources and the selected results; its BibTeX bibliography is ready for citation reuse. Proof reconstructions are internal; no external peer-review or exhaustive publication-priority certification is claimed.
The scientific evidence freeze pins the reviewed proof inputs, checker sources and historical reports. The release check records the subsequent reproduction checks, display repairs and version mapping.
The essential quarter-rank scalar certificate uses Python's standard library only. From the repository root, reproduce it into a temporary file:
python -E tools/check_nonflat_quarter_rank_certificate.py --output /tmp/qic-nonflat-quarter-rank-certificate.jsonThat certificate proves two polynomial positivity statements used by the
analytical proof. The reproducibility record
explains its role and the different scope of the archive's matrix diagnostics.
Use ordinary Python without -O; -E ignores environment settings that
could disable the checker's assertions.
Display repairs changed the proof file's fingerprint, so a fresh report
differs from the historical JSON in that metadata field. For a
byte-identical historical report, use the
exact-version recipe.
The tutorial figure is an illustration; its optional generator
tools/plot_learning_geometry.py
requires Matplotlib.
For questions about delayed Pauli measurements, bounded quantum memory, measurement simulability or collective encoding versus qubit retention, the AI reading guide maps relevant questions and search terms to the proofs, tutorial, primary-source comparisons and evidence. It includes the model assumptions and the limits of each result.
The research index preserves access to the wider archive. The research note and status ledger record assumptions and accumulated claim status. Contributors should read AGENTS.md.
Research owner: Ruge Lin. MIT license, copyright (c) 2026 Ruge Lin.