Hopf Quantum Backpropagation

How can one quantum measurement record contribute to the complete gradient of a variational state? Hopf-QBP turns the Hopf ansatz's coordinate tangents into an addressed frame. Inverting that frame makes every global magnitude outcome informative about every magnitude coordinate; a separate leaf record supplies complex phase derivatives.

Ruge Lin and Ranyiliu Chen — Compass in the Mirror: Quantum Backpropagation with the Hopf Ansatz (2026).

The task is to return the complete classical vector of raw Hopf-coordinate derivatives with simultaneous absolute coordinate error at most $\varepsilon$ and confidence $1-\delta$. For $n$ qubits, $N=2^n$, and $M=N-1$ real or $M=2N-1$ complex coordinates:

Result	Resource or guarantee
Global record	$O((1+\log(n/\delta))/\varepsilon^2)$ independent executions
Complete checkpoint schedule	$O(n(1+\log(n/\delta))/\varepsilon^2)$ independent executions
Direct-angle global Hopf work	$O(M\log M)$ assigned Hopf work per magnitude execution
Classical decoding	$O(M\log M)$ Walsh processing
Returned output	$M$ classical entries

The controlled objective must be a known, phase-calibrated reflection. Each execution makes one controlled-reflection call; its implementation cost is accounted for separately from the assigned Hopf ledger. The fixed-accuracy $O(\log\log M)$ global execution count describes record reuse. The full gradient remains an exponentially large output in $n$. Exact multiplexed recompilation has its own circuit contract and resource account.

Read the science

The complete scientific account contains the paper's main text and every appendix, with original equation, statement, protocol, figure, table, and reference identifiers. Follow the topic route below, or use the reviewer crosswalk to locate an item from the paper.

To understand	Read
The question and relation to earlier work	Motivation
Objective access, output accuracy, and circuit contracts	The reverse task
Hopf differentials and forward interfaces	Balanced differential
Addressed frames and shared magnitude records	Global real method
Phase-dressed magnitudes and direct phase records	Global complex method
Layer-local reverse transport	Checkpoint method
Concentration, decoding, storage, and matched costs	Accuracy and resources
A complete concrete construction	Four-qubit example
Full analytical arguments and compiler derivations	Proofs and compilation
Implementation conventions and API	Engineering guide

The supporting guides explain statistical output norms, matched method comparisons, exact recompilation, and reflection sums and readout. The bibliography preserves the paper's attribution.

Learn the project

The single educational anchor is Michael A. Nielsen and Isaac L. Chuang, Quantum Computation and Quantum Information, 10th Anniversary Edition. The educational bridge selects relevant sections, translates conventions, and develops the Hopf differential, interference, and statistical ideas needed for this work.

The 30-milestone map contains 12 background/preliminaries milestones and 18 narrative/innovation milestones. Both phases progress from local objects and mechanisms to the complete scientific story. Every milestone has an understanding target, dependencies, a conceptual connection, and source links.

Verify and reproduce

Start with the reviewer route: paper item → full statement and proof → implementation → executable check. The reproduction guide gives environment setup, commands, figure reproduction, expected results, and tolerances.

Use Python 3.10–3.13. From the repository root:

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt -r requirements-optional.txt
python validate_qbp.py

Use python validate_qbp.py --analytic for Qibo-free checks or --smoke for representative circuit contracts. The deterministic suite compares circuits, output distributions, decoded gradients, and assigned ledgers against independent references. General concentration and complexity guarantees are established by their written proofs.

These supplementary implementation figures summarize finite-dimensional checks. The original paper figures appear with their captions in the scientific account.

Companion paper

Hopf-ansatz develops the coordinates, inverse map, geometry, and indexed gradient access used here. Its paper is A Compass on the Quantum State Sphere: The Hopf Ansatz for Arbitrary Pure-State Optimization. The repositories use the same educational anchor and reading structure and are independently runnable.

Citation

Ruge Lin and Ranyiliu Chen, “Compass in the Mirror: Quantum Backpropagation with the Hopf Ansatz,” arXiv:2610.14177 (2026).

@misc{lin2026hopfqbp,
  title = {Compass in the Mirror: Quantum Backpropagation with the Hopf Ansatz},
  author = {Lin, Ruge and Chen, Ranyiliu},
  year = {2026},
  eprint = {2610.14177},
  archivePrefix = {arXiv},
  primaryClass = {quant-ph},
  url = {https://arxiv.org/abs/2610.14177}
}

Cite the companion Hopf-ansatz paper when using its state-preparation or geometric constructions. Machine-readable citation metadata is available in CITATION.cff.

For questions about the article, feel free to contact Ruge Lin at [email protected].

License

The software is released under the MIT License. The scientific account identifies the authors and the arXiv source of the paper.