arXiv:2610.06709
Out-of-control Hamiltonian Learning
Weiyuan Gong, Muzhou Ma, Sitan Chen, Jordan Cotler, Hsin-Yuan Huang
Abstract
Learning the Hamiltonian of a many-body system from its dynamics is a central task in quantum science, yet the algorithms with the strongest provable guarantees assume some level of quantum control--fast, arbitrary single-qubit gates interleaved with time evolution, and measurements in arbitrary bases--that is beyond the capabilities of near-term analog quantum simulators. Motivated by analog atom- and ion-based platforms, we study Hamiltonian learning under minimal access models. Uniform state preparation and measurements: We first consider the setting where in every experiment, one can rotate each qubit to the same state, perform short-time evolution, and measure every qubit in the same basis. Surprisingly, we show that for generic 2-local Hamiltonians on any interaction graph, all of the parameters can be reconstructed from such experiments. Computational basis state preparation and measurements: We then consider a similarly constrained setting, but where state preparation and measurement are restricted to the computational basis. For nearest-neighbor Hamiltonians with only Pauli $X/Z$ interactions, a class which captures contemporary Rydberg atom platforms, we show that over 1D and 2D rectangular lattices, all of the parameters can be reconstructed from such experiments up to unavoidable gauges. Our protocols introduce new techniques for solving structured polynomial systems over an extensive number of parameters. Taken together, our results suggest that one can learn a great deal from the dynamics of quantum many-body systems even under the most stringent experimental constraints.
AI Summary
The paper asks which many-body Hamiltonians can be learned from dynamics without individually chosen preparation and measurement bases or control during evolution. It studies uniform product-state preparation and common-basis measurement, and computational-basis preparation and readout; both retain site-resolved output bitstrings. With uniform control, it reconstructs all $3|V|+9|E|$ coefficients of a generic two-body Hamiltonian on any known interaction graph, without gauge ambiguity. First-order response leaves isotropic edge couplings and divergence-free skew couplings undetermined; second- and third-order connected cumulants resolve them using marginals on at most four sites and constant-size linear solves. With computational-basis access, it reconstructs generic nearest-neighbor IXZ Hamiltonians on 1D chains and 2D rectangular lattices up to Pauli-$Z$ conjugations and $H\mapsto-H$, giving $2^{n+1}$ sign gauges. This uses second- and fourth-order response and local polynomial systems with algebraic identifiability certificates. Resource bounds are $\widetilde O(\epsilon^{-2}\operatorname{poly}(n,R,\varsigma^{-1},\delta^{-1}))$ for coefficientwise error $\epsilon$; Gaussian smoothing of scale $\varsigma$ controls conditioning. Several translation-invariant families have costs independent of $n$, including an isotropic five-parameter family learnable from all-zero input alone. An explicit 1D IXZ family requires total evolution time $\Omega(\epsilon^{-2})$ without interleaved control, even allowing arbitrary inputs, ancillas, and measurements. Guarantees target the Gaussian-perturbed Hamiltonian, not every instance. Computational-basis results exclude $Y$ terms and generally assume finite-range lattice interactions. Extensions to power-law tails, general SPAM errors, noisy evolution, and spatially varying Hamiltonians with all-zero input remain open.
GPT-6 and Opus 5.5 assisted with literature search and preliminary drafting of the exposition.
Error found: The abstract's computational-basis reconstruction claim needs a genericity assumption: purely diagonal nearest-neighbor IXZ Hamiltonians yield identical computational-basis transition probabilities despite having coefficients unrelated by the stated sign gauges.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 87 pages, 7 figures