Pith. Sign in

REVIEW

Out-of-control Hamiltonian Learning

PaperPDF from arXiv

data files

ReviewThe objections and verdict, as JSON Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Weiyuan Gong, Muzhou Ma, Sitan Chen, Jordan Cotler, Hsin-Yuan Huang

Quantum Physics · Submitted 5 Oct 2026 · 53 references

Pith's verdict: Genuinely new uniform-control learning result that deserves a serious referee; the real risk is appendix certificate arithmetic, not the architecture of the argument.

T0 · AI review

How this review was made: An AI referee (deepseek/deepseek-v4-flash) read the paper on October 6, 2026 and answered a fixed list of questions about it. T0 is the starting level every paper gets. The levels describe how deeply Pith has checked the paper, never the authors.

  1. T0 AI review An AI read the paper and answered a fixed list of questions.
  2. T1 Audit Automatic checks that the references and files the paper relies on exist and match.
  3. T2 Compute The paper's code or calculations re-run with the same results.
  4. T3 Formal statement The claims written out in a formal math language a computer can read.
  5. T4 Proof Those claims proved, and the proof checked by computer.

How Pith checks papers →

· 2 major objections · 5 minor · Ask about this paper · The compiled record · Read the full review →
Challenge this review Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2610.06709 (version 1)
Pith IDA3YPOSUL, Pith's permanent record of this paper and its review
CitationHsin-Yuan Huang, Jordan Cotler, Muzhou Ma, Sitan Chen, Weiyuan Gong. "Out-of-control Hamiltonian Learning." arXiv:2610.06709 (2026). https://pith.science/paper/A3YPOSUL
Cite this reviewBibTeX for Pith's review of this paper
Claim paper

Why Pith says this

Highlight any text to comment on it.

Referee report

How Pith checks papers → · Full audit of this paper →

The paper

The paper claims

With only uniform rotations or Z-basis shots, generic 2-local Hamiltonians on any known graph are fully learnable, up to unavoidable sign gauges under Z-basis access.

The reading

This paper asks what can be learned about a many-body Hamiltonian when the experimenter has no per-qubit control: either every qubit is rotated and measured along a common direction, or preparation and measurement are restricted to the computational basis. The authors claim that surprisingly much is learnable in both settings, at the standard quantum limit. Under uniform control, all 3|V|+9|E| parameters of a generic 2-local Hamiltonian on any known interaction graph are recovered with no gauge ambiguity from short-time second- and third-order response. Under Z-basis access, nearest-neighbor Hamiltonians built only from X and Z Paulis on chains and rectangular lattices are recovered up to exactly the Pauli-Z conjugation and time-reversal gauges. A matching lower bound shows the 1/$epsilon^{2}$ total evolution time is unavoidable without interleaved control, so these results attain the best possible scaling for the access models they target.

What carries the argument

The load-bearing object is a staged decomposition of the Hamiltonian into components that become detectable at different Taylor orders of short-time observables. For uniform control, J_uv = T_uv + [d_uv]_x + lambda_uv I separates a symmetric traceless part (visible at first order), a skew-symmetric curl part (second order, via connected three-site cumulants whose leading monomials require two distinct edges spanning three sites), and an isotropic trace living in the swap direction (third order, via connected four-site cumulants); the first-order kernel is exactly the curl gauge of dimension 3(|E|-|V|+c) plus one trace per edge. For Z-basis access, H = D + K makes the off-diagonal part visible at second order as squared direct amplitudes and the diagonal part visible at fourth order through Eq. (30), which isolates a squared energy difference and an energy-weighted sum over two-step paths. Around both, the paper builds an algebraic certification toolkit showing that the solving polynomial systems are generically well conditioned, with explicit determinant values exhibited at chosen parameter points.

Extended reading notes

Core claim

The core claim is that the standard assumption of individually addressable gates is not needed for generic Hamiltonian learning. For the uniform-control model, the paper decomposes each 2-local coupling matrix into a symmetric traceless part, a skew-symmetric part, and a trace part, and shows these become visible at successive orders: first-order data fix the symmetric part, the fields, and the vertex divergences, leaving only curl gauges and edge traces; second-order connected three-site cumulants remove the curl ambiguity; third-order connected four-site cumulants recover the traces. For the Z-basis model, the Hamiltonian is split into a diagonal part D and an off-diagonal part K; second-order data give the squared amplitudes of direct transitions and thereby the off-diagonal coefficients up to sign, while fourth-order interference terms resolve relative signs and determine the diagonal coefficients, with the only remaining freedom being the 2^(n+1) choices from Pauli-Z conjugation and time reversal. Both reconstructions reduce to constant-size systems of polynomial equations whose unique solvability and conditioning are certified by explicit nonzero-polynomial determinants under smoothed analysis.

Load-bearing premise

The entire stability of the protocol rests on the smoothed-genericity event that every determinant, denominator, and certificate polynomial of the unknown instance is bounded away from zero; if the true Hamiltonian sits at or near a zero of any of these polynomials, the error amplification is uncontrolled, a constraint the authors explicitly acknowledge by restricting to generic smoothed instances.

Editorial extensions

If this is right

  • Analog Rydberg and trapped-ion simulators, which natively support only global rotations or computational-basis readout, can in principle be characterized without adding individually addressed gates, at the standard quantum limit in total evolution time.
  • The uniform-control protocol applies to any known interaction graph regardless of degree or geometric locality, so it covers long-range and nonlocal couplings as long as the graph is known.
  • For translation-invariant or isotropic families, the experimental cost becomes independent of the number of qubits, making constant-size experiments sufficient even for very large arrays.
  • The 1/epsilon^2 lower bound shows that no control-free protocol, even with arbitrary entangled inputs and ancillas, can reach Heisenberg scaling, so any future improvement must interleave at least some control during evolution.
  • Both reconstructions rely only on up-to-four-site marginal probabilities and constant-size polynomial solves, keeping the classical post-processing polynomial and the measured regions small.
  • The unobservable gauges under Z-basis access are exactly characterized: 2^(n+1) sign flips generated by site-wise Pauli-Z conjugation and time reversal, with no additional generic ambiguity in the IXZ family.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The explicit determinant values and certificate polynomials supplied in the appendices mean each local identifiability claim is numerically checkable offline: one can evaluate the certificates at random smoothed instances and verify the claimed conditioning without running any experiments.
  • The gauge analysis suggests a natural next question: for Hamiltonians containing Y terms, conjugation by diagonal unitaries and time reversal may not be the only generic ambiguities under Z-basis access, and the paper leaves open whether families that admit an IXZ representative under a locality-preserving diagonal conjugation are learnable by the same method.
  • The polynomial-certification approach, which trades generic-instance assumptions for explicit nonzero witnesses, is likely portable to other restricted-access inverse problems, such as learning a local Lindbladian generator or SPAM parameters from a single global drive.
  • The two-qubit family behind the lower bound is a concrete probe: any claim of Heisenberg scaling in a control-free model would have to fail on that specific family, which provides a quick experimental or analytical falsification test.
Original 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.

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.

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-10-06: 1 request, 6,413 input tokens.

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.59 · human participants 0.01 · controlled comparison 0.04 · randomized assignment 0.01 · states quantitative result 0.12 · compares with prior results 0.60 · mentions code or data availability 0.02.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

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.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 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. background 0.77 0.05 0.04 asserted 1.00
s1 Motivated by analog atom- and ion-based platforms, we study Hamiltonian learning under minimal access models. contribution 0.95 0.61 0.04 asserted 0.64
s2 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. method 0.89 0.25 0.03 asserted 0.80
s3 Surprisingly, we show that for generic 2-local Hamiltonians on any interaction graph, all of the parameters can be reconstructed from such experiments. result 1.00 0.94 0.25 asserted 1.00
s4 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. method 0.79 0.50 0.03 unclear 0.69
s5 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. result 0.99 0.94 0.13 asserted 1.00
s6 Our protocols introduce new techniques for solving structured polynomial systems over an extensive number of parameters. method 0.55 0.91 0.06 asserted 1.00
s7 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. interpretation 1.00 0.59 0.80 hedged possibility 1.00

What kind of paper the classifier reports this is: theoretical result 0.74 · evidence from proof or derivation 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

Figures
Figure 1. (a) Uniform control (𝑎ˆ-to-𝑏ˆ model): all qubits are prepared along one common direction 𝑎ˆ, |0 𝑛 𝑎ˆ ⟩ = (𝑈|0⟩)⊗𝑛 , and measured along one common direction 𝑏ˆ. (b) 𝑍-basis access (𝑍-to-𝑍 ′ model; 0-to-𝑍 ′ for |0 𝑛 ⟩ input): the qubits start in a computational basis state |𝑧⟩ and are measured in the computational basis. (c) From the data of (a), all parameters can be learned, for any interaction graph. (d) From the data of (b), a generic NN IXZ Hamiltonian on a 1D chain or square lattice can be learned up to unavoidable Pauli-𝑍 conjugation and time-reversal gauges. 2.1 Uniform co
Figure 2. Three-round reconstruction under uniform control. (a) Local data: Taylor coefficients 𝐿𝑆, 𝑄𝑆 and W𝑆 (orders one to three) of the Fourier coefficients 𝑚 𝑎,ˆ 𝑏ˆ 𝑆 (𝑡) on sets 𝑆 of at most four sites, with 𝐽𝑒 = 𝑇𝑒 + 𝐴𝑒 + 𝜆𝑒 𝐼 for 𝐴𝑒 = [d𝑒]×. (b)–(d): checks denote what is learned in that round, crimson and question marks denote the remaining ambiguities. (b) Round 1: first-order data identifies 𝑇𝑒, h𝑣, as well as 𝐴𝑒 up to a global curl gauge. (c) Round 2: Second-order terms 𝑅𝑢𝑣𝑤 of three-site cumulants determine the skew-symmetric components 𝐴𝑒 for edges on induced paths. For the…
Figure 3. Local reconstruction of a generic NN IXZ Hamiltonian from 𝑍-to-𝑍 ′ data. (a) For transitions that flip the marked site (ring), vary the neighboring input bits (double arrows) and extract the second- and fourth-order coefficients of the local transition probabilities 𝑝𝑧→𝑧 ′ (𝑡). (b) Stage 1: Second-order data recover each transverse field and its incident mixed couplings up to a common site sign (±). Fourth-order tests on nonadjacent pairs resolve the chain-endpoint ambiguities and determine relative 𝐽𝑋𝑋 signs along the shaded paths, leaving one common sign. (c) Stage 2: Subtracting…
Figure 4. Experimental access to the unknown Hamiltonian 𝐻. (a) Uniform control (𝑎ˆ-to-𝑏ˆ): Prepare every qubit along a common direction 𝑎ˆ, evolve once under 𝐻, and measure along a common direction 𝑏ˆ. Uniform rotations implement the preparation and readout, and the two directions may vary between experiments. (b) Computational-basis access (𝑍-to-𝑍 ′ ): Prepare |𝑧⟩ in the computational basis, evolve once under 𝐻, and measure in the computational basis, obtaining 𝑧 ′ with probability 𝑝𝑧→𝑧 ′ (𝑡). Arbitrary inputs require site-resolved bit flips, and the all-zero-input model fixes 𝑧 = 0 𝑛 . Our…
Figure 5. First-order reconstruction under uniform control. (a) Single-site and two-site responses determine the local fields h𝑣 and the symmetric traceless parts 𝑇𝑒 of the couplings. The skew-symmetric parts 𝐴𝑒 enter only through their divergence at each vertex. (b) The trace terms 𝜆𝑒 𝐼 are invisible at first order, leaving one undetermined parameter per edge. (c) Adding a skew-symmetric matrix around a cycle, with signs following the edge orientations, preserves these divergences: the incoming and outgoing contributions cancel at every vertex. The illustrated graph has two independent…
Figure 6. Local reconstruction of the skew-symmetric edge vectors. (a) For an induced path 𝑢-𝑣-𝑤, six fixed settings (basis pairs) give a 6 × 6 system in (d𝑢𝑣, d𝑣𝑤). Its determinant is a nonzero polynomial of degree six, so independent Gaussian smoothing gives the simultaneous lower bound in Eq. (21). Each selected path is solved independently. In particular, every edge of a triangle-free component with at least three vertices has a path. (b) For general graphs, first solve every edge lying in an induced path, then recover each remaining edge in a non-complete connected component from a…
Figure 7. Recovering the edge traces from local responses. (a) The trace term (crimson) exchanges two spins between pair flips generated by the known couplings (amber), making it visible in the four-site third-order response. (b) A vertex and three of its neighbors give a linear system for at most six edge traces. Subtracting the known traceless contribution from the connected third-order response isolates these traces and exterior neighbors do not contribute. (c) At vertices with one or two neighbors, single-site second-order responses recover the incident traces after the known contribution…

Figures reproduced from arXiv:2610.06709 by the authors.

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.