How does the ability to remove entropy during a noisy evolution change the number of oracle calls needed to simulate it?
This project classifies the query cost of a prescribed controlled evolution generated by one unknown Hermitian reflection. The target retains only two qubits, but known controls can repeatedly remove entropy and preserve sensitivity to the unknown noise axis. Their total entropy-decrease capacity determines a sharp crossover between square-root and linear query cost.
Let the unknown reflection satisfy
Entropy uses natural logarithms. Unital controls have zero capacity; replacing a qubit by a state
Given ordinary controlled-
The guarantee covers every input, including one entangled with an untouched reference. The constants are independent of time, capacity, and the number of control boundaries. The model fixes the promise and oracle contract; query complexity proves the matching upper and lower bounds.
| Target control capacity | Query order | Exact two-qubit witness |
|---|---|---|
| Bounded |
Maximally mixed syndrome replacement | |
| Partially polarized syndrome replacement | ||
| Pure syndrome replacement |
Zero capacity permits repeated unital interventions. A single reflection without interventions has an exact sampler requiring at most one query; the interpretation explains why these are different target classes.
For a retained system of dimension
The complete channels obey
The proof combines the entropy that can accumulate during the noisy intervals with a backward-observable variance identity that survives arbitrary common CPTP controls. It includes singular inputs and pure resets, arbitrary untouched references, and bounded piecewise continuous time dependence. It requires no spectral gap or mixing assumption.
A two-qubit mixed-reset construction attains the time-and-capacity dependence at fixed noise rate through an exact finite-distance signal. Local-noise examples establish the dimension dependence and clarify the separate role of retained qubits and pure-reset count.
| Read for | Authoritative page |
|---|---|
| Assumptions, normalization, and principal theorems | Model and results |
| Control capacity, references, and reset accounting | Entropy-decrease capacity |
| Complete entropy-and-variance proof | Channel stability |
| Exact witness in every time and capacity regime | Two-qubit mixed resets |
| Learning upper, Poisson cap, and oracle hybrid lower | Query complexity |
| Dimension, the local query law, and weighted reset sharpness | Memory and local noise |
| Physical meaning and resource boundaries | Interpretation |
| Imported results and closest prior models | Related work and BibTeX |
The science index also provides a connected proof route and navigation by question.
The single educational anchor is John Watrous, The Theory of Quantum Information (2018). Its channel, distance, unitality, and entropy foundations lead into the project's Hermitian-noise and oracle arguments through a focused source bridge.
| Read for | Route |
|---|---|
| The connected study path and proof relationships | Learning overview |
| Relevant Watrous sections, conventions, and the steps into this project | Watrous bridge |
| A complete map of the project's understanding targets | 30 milestones |
The map has 10 background/preliminary milestones and 20 narrative/innovation milestones, progressing from local ingredients to the complete scientific story within each phase. Each target links to its prerequisites, conceptual connections, and authoritative sources.
From the repository root, using Python 3.12:
python3 -m pip install -r requirements.txt
python3 verification/checks.pyThe deterministic suite compares controlled Duhamel, entropy, and variance identities; direct two-qubit circuits against their exact coherence formula; finite capacity witnesses; and local GHZ outputs against the binomial signal and entropy formulas. It prints a summary and writes no file by default. Verification specifies the parameter grids, tolerances, reference environment, and optional JSON output.
The written proofs establish the universal statements. The finite checks supply reproducible evidence for the identities and implementations they exercise; they do not numerically certify the diamond-norm or asymptotic theorems.
This is a fixed-error query theorem for a known sequence of target controls. The simulator may use auxiliary learning probes, memory, measurements, and classical processing; all unknown-oracle calls count. The capacity budget applies to the target controls. Runtime, known-gate count, and arbitrary-dimensional optimal query complexity are separate questions.
The channel theorem compares supplied matched Hermitian jumps and includes every retained quantum or classical record in
Entropy gain, unitary learning, and the informative-noise reset mechanisms have explicit predecessors. The source comparison identifies those ingredients and distinguishes them from the combined capacity-dependent channel bound and controlled-query classification.
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.
MIT.