REVIEW
Gauge freedom and efficient algorithms for Lindbladian learning
Quantum Physics · Submitted 28 Sep 2026 · 41 references
Pith's verdict: A genuinely new classification of which Lindbladian coefficients survive unknown SPAM under single-qubit control, mostly well-proved, with one gap worth closing before it goes out.
T0 · AI review
Why Pith says this
Highlight any text to comment on it.
Referee report
The paper
The paper claims
Unknown SPAM noise makes most Lindbladian components unlearnable, and the paper pins down exactly which ones survive: the diagonal dissipative rates, one class of real off-diagonal couplings, and the imaginary couplings together with…
The reading
What carries the argument
The engine is the $N$-parameter family of gauge transformations $S_\theta = e^{\Gamma_\theta}$, where $\Gamma_\theta$ rescales each Pauli operator $P$ by $e^{-\theta\cdot s(P)}$; these commute with every trusted single-qubit unitary, preserve locality, and can be absorbed into the unknown SPAM channels so that all outcome probabilities are unchanged (Equation 19). Theorem 13's phase-parity argument, in the $\chi$-matrix representation, shows the coefficients of $L$ rescale as $\chi_{U,V}\to e^{\theta\cdot(s(V)-s(U))}\chi_{U,V}$ plus mixing terms, and a parity bookkeeping over the differing-qubit set $D(U,V)$ decides exactly which real or imaginary component admits a gauge-dependent summand. For learning, the protocols use Pauli and partial twirls to isolate decay-like signals whose SPAM factors cancel in ratios, a back-substitution over support inclusion to recover individual coefficients from restricted ones, and a Chebyshev-endpoint derivative estimator (Lemma 28) for the Type II imaginary parts, which cannot be converted into decay rates.
Extended reading notes
Core claim
Under unknown SPAM noise with only trusted single-qubit control, two physically distinct local Lindbladians can produce data that is exactly identical in every experiment, so part of the generator is unlearnable in principle. The paper pinpoints the boundary: a component survives exactly when it is invariant under an $N$-parameter family of local depolarizing gauge transformations $S_\theta$ that commute with all single-qubit unitaries and can be absorbed into the SPAM operations. That yields an exact list — the diagonal dissipative coefficients $\alpha_{aa}$, the real parts $\mathrm{Re}\,\alpha_{ab}$ for pairs differing on a single qubit with both nonidentity there (Type I), and the imaginary parts $\mathrm{Im}\,\alpha_{ab}$ for pairs differing on a single qubit with exactly one identity there, together with single-qubit Hamiltonian coefficients $h_c$ (Type II). Every other component is generically gauge dependent among physical complete-positive Lindbladians with any prescribed bounded-degree support pattern, and there exist generators for which each such component shifts by an $\Omega(1)$ amount — independent of system size — under an admissible gauge transformation, so no estimator can beat a constant error floor. For the survivors, the paper gives SPAM-robust protocols using only single-qubit gates with $\widetilde{\mathcal O}(\epsilon^{-2}\log(N/\delta))$ experiments and total evolution time for constant locality and degree.
Load-bearing premise
The load-bearing assumption is local SPAM visibility (Definition 8): the noisy prepared state and noisy measurement must retain overlap at least $r_p$ and $r_m$, bounded away from zero independently of $N$, with the intended local operators; if that local visibility decays with system size, then the identification of the invariant components, the bounded signals the protocols rely on, and every sample-complexity guarantee collapse.
Editorial extensions
If this is right
- With unknown SPAM and trusted single-qubit control, no experiment can determine the full Lindbladian: the generically gauge-dependent components carry an irreducible uncertainty bounded below by a constant independent of system size, even under complete positivity (Theorem 18).
- All universally gauge-invariant components — diagonal dissipative rates, Type I real couplings, and Type II imaginary couplings plus single-qubit Hamiltonian terms — are efficiently learnable with $\widetilde{\mathcal O}(\epsilon^{-2}\log(N/\delta))$ experiments and total evolution time using only single-qubit gates, within polylogarithmic factors of near-optimal lower bounds.
- The guarantees hold even when global SPAM is arbitrarily far from ideal, as long as local preparation and measurement visibility are bounded away from zero independently of $N$ (Definition 8).
- At boundary instances where the Kossakowski matrix has zero diagonal entries, physical positivity can freeze additional Hamiltonian coefficients under admissible gauge transformations, though learning these instance-specific invariants remains open (Theorems 31 and 32).
Reading between the lines
- The commuting-gauge characterization suggests a quantitative transition: adding trusted entangling gates should shrink the invariant family and unlock more components, by contrast with the multiqubit-control setting; this transition could be probed numerically on small systems.
- The $\Omega(1)$ error floor implies that, without extra calibration assumptions, only the invariant projection of the Lindbladian is operationally meaningful in a continuous-time gate-set-tomography sense — a statement worth testing on device data where SPAM is far from ideal.
- The boundary-induced invariance of Hamiltonian coefficients not in the span of dissipative operator products echoes the Hamiltonian-not-in-Lindbladian-span condition for Heisenberg-limited metrology; a concrete question is whether those boundary-invariant Hamiltonian terms admit shot-noise-beating estimation, which the paper leaves open.
Original 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 $\Omega(1)$ 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 $\widetilde{\mathcal O}\left(\epsilon^{-2}\log(N/\delta)\right)$ 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.
The compiled record
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
The abstract, sentence by sentence
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 $\Omega(1)$ 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 $\widetilde{\mathcal O}\left(\epsilon^{-2}\log(N/\delta)\right)$ 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.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | We study the problem of learning a local Lindbladian in the presence of state-preparation-and-measurement (SPAM) noise. | background 0.55 | 0.79 | 0.04 | asserted 0.80 |
| s1 | 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. | background 0.62 | 0.08 | 0.14 | asserted 0.80 |
| s2 | We give a sharp characterization of this obstruction for bounded-degree local Lindbladians. | contribution 0.82 | 0.95 | 0.15 | asserted 1.00 |
| s3 | We identify a family of locality-preserving gauge transformations that commutes with trusted single-qubit control, and use it to classify Lindbladian components. | method 0.67 | 0.95 | 0.08 | asserted 1.00 |
| s4 | For the generically gauge dependent components, we construct Lindbladians to show that they can have $\Omega(1)$ uncertainty independent of system size, even after imposing complete positivity. | result 0.93 | 0.94 | 0.14 | asserted 0.97 |
| s5 | We complement this characterization with SPAM-robust algorithms for learning every universally gauge-invariant component. | contribution 0.92 | 0.96 | 0.27 | asserted 1.00 |
| s6 | Our algorithms use only trusted single-qubit operations, require neither ancillas nor entangling control, and use $\widetilde{\mathcal O}\left(\epsilon^{-2}\log(N/\delta)\right)$ experiments and total evolution time for constant locality and degree. | method 0.69 | 0.95 | 0.17 | asserted 1.00 |
| s7 | Our guarantees allow global SPAM to become arbitrarily far from ideal as $N$ grows, requiring only nonvanishing local visibility. | result 0.80 | 0.90 | 0.12 | asserted 0.99 |
| s8 | 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. | limitation 0.71 | 0.92 | 0.06 | asserted 0.98 |
| s9 | More broadly, our gauge-aware framework and SPAM-cancellation techniques offer a practical toolkit for the scalable characterization of open quantum systems. | interpretation 0.91 | 0.77 | 0.27 | asserted 0.83 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
References
Discussion
Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.