Quantum dynamics are being reconstructed without requiring direct measurement access to the entire system. Fast single-qubit control and a connected reference backbone demonstrate that one measurable qubit is sufficient to learn all O(N) independent parameters of a bounded-degree two-body Hamiltonian on N qubits at the Heisenberg limit. Key to this process is the use of strong SWAP gates synthesised by quantum signal processing, which enable coherent transfer of states evolving under distant Hamiltonian parameters to the measurable qubit.

This transfer requires no prior calibration of the Hamiltonian parameters of the intermediate links. A parallel learning architecture accelerates parameter estimation sharply.

Single-qubit measurements efficiently characterise multi-qubit Hamiltonians at the Heisenberg limit

Reconstructing quantum dynamics now requires to be reduced total query time, from eO(N) to widetilde{O}(N), representing a sharp acceleration in understanding N-qubit systems. Employing strong SWAP gates, operations which exchange qubit states, alongside coherent state transfer via a connected reference backbone enabled performance at the Heisenberg limit while matching fundamental lower bounds up to logarithmic factors.

This signifies an extremely precise level of quantum state characterisation as uncertainty decreases proportionally to the square root of measurements taken. The framework also successfully extended beyond simple linear qubit arrangements, demonstrating flexibility in modelling complex systems, and implemented a parallel learning architecture allowing all O(N) independent parameters within the Hamiltonian describing energy and interactions to be learned simultaneously; this sharply boosts efficiency compared with sequential methods. Current limitations regarding error accumulation over longer chains or more intricate network topologies still need addressing before practical applications become fully realised.

Scalable quantum parameter estimation via local qubit readout

Heisenberg-limited precision scaling has been achieved for an N-qubit chain while maintaining total query time eO(N), matching fundamental lower bounds up to logarithmic factors. This framework extends to arbitrary bounded-degree interaction graphs, establishing a scalable route towards learning many-body Hamiltonian parameters through only a local measurement interface.

Complex dynamical systems are rarely observed in their entirety and, across the classical sciences, properties of extended systems are routinely inferred from limited spatially local observations; weather forecasting combines measurements from monitoring stations to reconstruct atmospheric dynamics, seismology infers subsurface structure from sensor signals, and system identification determines internal dynamics via input/output ports.

These examples illustrate that observations obtained through a limited local interface can reveal properties of much larger systems which raises an important question for quantum many-body systems: To what extent can global quantum dynamics be learned to use a local observation window. The work explores this question within Hamiltonian learning, reconstructing the generator of unknown quantum dynamics from experimental data, crucial in calibrating devices, characterising simulators and probing naturally occurring many-body systems. Recent advances have established efficient learning from high-temperature thermal states and real-time evolution with subsequent developments extending thermal-state learning to arbitrary fixed temperatures and improving its temperature dependence.

For dynamic learning coherent control enables Heisenberg-limited precision scaling; further progress encompasses unknown interaction structures, compressed sensing methods and ansatz-free reconstruction while recent protocols achieve similar scaling utilising only static single-qubit control fields. However, these approaches typically assume measurement access distributed throughout the system and retaining such capabilities when measurements are confined to a small subsystem remains challenging.

Restricted readout is natural in several quantum architectures: solid-state spin registers use one optically addressable spin as an ancilla for nearby nuclear spins whose information is coherently transferred to it; reconfigurable neutral atom arrays allow large array operations but have comparatively slow global measurements with developments coupling atoms to optical cavities accelerating measurements within dedicated regions; trapped ion systems utilise physical shuttling of ions or atoms to probing regions providing routing primitives.

These examples motivate treating spatial extent of measurement access as an experimental resource distinct from the size of the quantum system itself. The work studies an asymmetric model featuring fast single-qubit control throughout, evolution under its unknown native Hamiltonian and direct state preparation/measurement confined to a distinguished bright qubit assuming each neighboring pair has a reference interaction bounded away from zero ensuring information propagation along the chain.

Under this model one measurable qubit suffices to reconstruct all local fields and two-body couplings on an N-qubit chain achieving Heisenberg-limited O(1/ε) scaling for target precision ε and total time eO(N). Their approach combines parallel phase encoding with strong information transfer through unknown interactions; first isolating disjoint terms allowing many coefficients to be encoded into separate phases which are then conveyed sequentially avoiding quadratic overhead.

In particular, these transfer operations must themselves be constructed from the unknown Hamiltonian using quantum signal processing to synthesise strong SWAP operations without prior calibration of coupling strengths or signs. Accumulated errors were bounded so that strong phase estimation retains Heisenberg-limited precision scaling extending this to Hamiltonians having bounded-degree interaction graphs with a connected reference backbone. On chains with common axis references the protocol requires only restricted global control removing individual addressing of unmeasured qubits.

The remainder of their work is organised as follows: Section II compares results with previous local probe experiments and protocols; Section III defines the learning task, model and assumptions reviewing reshaping and phase estimation; Section IV presents lower bounds for learning; Section V develops the parallel learning protocol analysing its error/runtime guarantees including implementation with restricted control. Section VI extends the protocol to general interaction graphs while section VII concludes discussing geometry constraints and open questions detailed proofs are in Supplementary Material.

Previous studies have used a local probe to reveal interactions within an extended quantum system such as Abobeih et al. using a nitrogen vacancy centre to characterise 27 nuclear spins or van de Stolpe et al. mapping a 50-spin network through correlated sensing reconstructing interactions in specific solid state spin registers.

A systematic framework exists for reconstructing unknown quantum dynamics via Hamiltonian learning, however existing protocols commonly assume complete measurement access to the entire system. This protocol employs strong SWAP gates synthesised by quantum signal processing, coherently transferring states evolving under distant Hamiltonian parameters to the measured qubit without prior calibration of intermediate links.

A parallel learning architecture achieves total query time eO(N) for an N-qubit chain while retaining Heisenberg-limited precision scaling. These scalings align with fundamental lower bounds for both precision and information propagation up to logarithmic factors; also this framework extends to arbitrary bounded-degree interaction graphs. The paper demonstrates that many-body Hamiltonian parameters can be determined using only local measurement access.

The research demonstrated a method to learn all O(N) independent parameters of a two-body Hamiltonian on N qubits, utilising just one measurable qubit alongside fast single-qubit control and a connected reference backbone. This is important because it establishes a scalable route to characterise complex quantum systems without needing full system measurements. The authors extended this framework to accommodate arbitrary bounded-degree interaction graphs, suggesting broader applicability within many-body physics.

👉 More information
🗞 Learning Many-Body Hamiltonians Using a Local Probe
✍️ Suying Liu, Zitai Xu, Alexey V. Gorshkov and Xiaodi Wu (University of Maryland); Yu-Xin Wang and Zhi-Yuan Wei
🧠 ArXiv: https://arxiv.org/abs/2610.02157

More like this