arXiv:2609.35265
Gauge freedom and efficient algorithms for Lindbladian learning
Steven T Flammia, Savar D Sinha, Yu Tong
Abstract
We study the problem of learning a local Lindbladian in the presence of state-preparation-and-measurement (SPAM) noise. Although recent work has developed scalable learning algorithms under idealized access assumptions, SPAM can make distinct Lindbladians experimentally indistinguishable, thus making part of the Lindbladian fundamentally unlearnable. We give a sharp characterization of this obstruction for bounded-degree local Lindbladians. We identify a family of locality-preserving gauge transformations that commutes with trusted single-qubit control, and use it to classify Lindbladian components. For the generically gauge dependent components, we construct Lindbladians to show that they can have $Ω(1)$ uncertainty independent of system size, even after imposing complete positivity. We complement this characterization with SPAM-robust algorithms for learning every universally gauge-invariant component. Our algorithms use only trusted single-qubit operations, require neither ancillas nor entangling control, and use $\widetilde{\mathcal O}\left(ε^{-2}\log(N/δ)\right)$ experiments and total evolution time for constant locality and degree. Our guarantees allow global SPAM to become arbitrarily far from ideal as $N$ grows, requiring only nonvanishing local visibility. We also establish sufficient conditions for boundary-induced gauge invariance of additional Hamiltonian coefficients under physical positivity constraints, though learning these additional parameters in general remains an open problem. More broadly, our gauge-aware framework and SPAM-cancellation techniques offer a practical toolkit for the scalable characterization of open quantum systems.
AI Summary
The paper asks which coefficients of a bounded-degree local Lindbladian remain identifiable under unknown state-preparation-and-measurement (SPAM) noise with trusted single-qubit control, and how efficiently they can be estimated. It assumes a known interaction-support pattern, bounded coefficients, fixed locality and degree, and nonvanishing local SPAM visibility; global SPAM need not approach ideal operations. Local depolarizing similarity transformations commute with the controls, preserve locality, and can be absorbed into SPAM. Analyzing their action in Pauli-transfer and $\chi$ representations identifies exactly three universally invariant classes: diagonal Kossakowski coefficients; $\Re\alpha_{ab}$ when the Pauli strings differ at one site with both factors nonidentity; and $\Im\alpha_{ab}$ when they differ at one site with exactly one identity, together with single-qubit Hamiltonian coefficients. Every other locally allowed component is gauge dependent on an open dense, full-measure subset of physical generators. Positivity-buffered constructions establish system-size-independent $\Omega(1)$ marginal ambiguities despite complete positivity. Protocols learn all universal invariants without ancillas or entangling control. Pauli twirling, local rotations, and SPAM-canceling decay ratios handle the first two classes; partial twirling isolates two-dimensional dynamical blocks for derivative estimation of the third. Experiment counts and total evolution time scale as $\widetilde O(\epsilon^{-2}\log(N/\delta))$; single-qubit gate counts scale as $\widetilde O(N\epsilon^{-3}\log(N/\delta))$, using finite-rate randomized pulses. Positivity also yields sufficient conditions for additional, instance-specific Hamiltonian invariance at boundary generators, but general learning protocols for these remain open. Other stated directions include learning invariant parameter combinations, robustness to imperfect trusted controls, and stability near physical boundaries.
ChatGPT was used to polish and organize these notes into an initial manuscript draft. It was also used to draft the proofs
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.