Can a small exponential update of an unknown quantum state be sampled from a small number of existing specimens?
This project proves a dimension-dependent obstruction for the normalized Esscher update
Even when the exact spectrum is supplied, producing one updated state can require
For
| Task and interface | Established copy cost |
|---|---|
| One Esscher output, known exact hard-family spectrum, unknown orientation | More than |
| Same task with fixed-probability heralding or uniformly bounded expected budget |
|
| General bounded-spectral-scale prior, collective tomography and classical processing |
|
| Explicit independent-copy Clifford-measurement implementation |
|
| Commuting Esscher update, or ordinary bounded sandwich filtering | At most five copies at the theorem's tilt and error |
The quadratic collective upper bound and linear lower bound leave an open gap. The constructive methods charge tomography, dense classical processing and final state preparation; no QRAM, preparation inverse or modular-Hamiltonian oracle is supplied. Upper bounds distinguish copy complexity from implementation costs.
Ordinary filtering produces
The science guide connects the complete account.
| Question | Authoritative account |
|---|---|
| What is known, supplied, produced and counted? | Model and main theorem |
| How does a bounded tilt reveal hidden orientation? | Exact response and finite-tilt witness |
| Why can arbitrary collective processing do no better? | Copy lower bound and success conventions |
| Why are commuting updates and ordinary filters cheap? | Charged filtering baselines |
| What constructive upper bounds are available? | Stability, tomography and state output |
| What does the theorem mean, and where does it stop? | Interpretation |
| Which definitions, identities and hardness results are inherited? | Related work and bibliography |
The proof combines a finite Esscher-response witness with the established O'Donnell–Wright mixedness bound. It compares fully Haar-oriented and block-commuting priors with identical supplied spectra. Both input ensembles remain difficult to distinguish from a maximally mixed reference at sublinear copy count, while their updated states have different expectation values of one known observable. No typical-input concentration or dimension-dependent accuracy limit is needed.
The selected educational anchor is John Watrous, The Theory of Quantum Information (2018). The learning guide connects the task to the source bridge and the 30-milestone map. The route develops 12 background milestones and 18 narrative milestones, moving from states and copy access to the complete same-spectrum argument and its constructive comparisons. The bridge includes a complete one-qubit example distinguishing exponential updating from ordinary filtering.
The verification guide gives pinned dependencies, commands, expected results and scope. The diagnostics compare independent matrix formulas, exact qubit integration, same-spectrum controls, finite-tilt constants and selected Haar averages. Their figure summarizes the finite experiments. The analytical proofs establish the uniform lower bound.
The source is an unknown generic mixed state with independent-copy access. Its modular Hamiltonian is not promised to be local, known or efficiently simulable. The theorem concerns this particular minimum-relative-entropy exponential update, rather than every quantum inference rule or every noncommuting source. A matching copy upper bound, efficient collective-tomography gate implementation and a practical hardware advantage remain unestablished.
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.