Sequential Spatiotemporal Magnetic-Field Reconstruction via Quantum Hamiltonian Learning with NV-Center Spin-1 Hamiltonians

Hiroshi Yamauchi; Samuel Tovey; Sophie Colleen Stearn

arxiv: 2605.23455 · v2 · pith:4UIKUR2Jnew · submitted 2026-05-22 · 🪐 quant-ph

Sequential Spatiotemporal Magnetic-Field Reconstruction via Quantum Hamiltonian Learning with NV-Center Spin-1 Hamiltonians

Hiroshi Yamauchi , Sophie Colleen Stearn , Samuel Tovey This is my paper

Pith reviewed 2026-05-25 04:42 UTC · model grok-4.3

classification 🪐 quant-ph

keywords quantum hamiltonian learningnv-center spin-1magnetic field reconstructionsequential bayesian inferencespatiotemporal reconstructiondipolar couplingsynthetic field sequences

0 comments

The pith

Sequential Bayesian updates over overlapping scan windows reconstruct evolving 2D magnetic fields from local NV-center spin-1 measurements.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a reconstruction pipeline in which local likelihoods are generated from the NV spin-1 Hamiltonian that depends on the unknown local field value and a shared dipolar coupling J. These likelihoods are fed into sequential Bayesian updates performed on sliding windows, with the posterior from one frame propagated forward as the prior for the next. The approach is tested on synthetic time-varying fields that contain maze-like spatial structure. A sympathetic reader would care because the latent field cannot be observed directly, so any workable method must convert the indirect spin dynamics into accurate field maps without assuming the field is static. The reported final-frame root-mean-square error of 7.037×10^{-7} T supplies the concrete performance benchmark.

Core claim

The central claim is that sequential Bayesian updates over overlapping scan windows, combined with temporal posterior propagation and a likelihood model derived from the NV-center spin-1 Hamiltonian, recover the dominant spatial structure of an evolving two-dimensional magnetic field. On synthetic maze-like sequences the method achieves a root-mean-square error of 7.037×10^{-7} T at the final frame. Adaptive diagnostics indicate decreasing expected information gain and stable local convergence, while combined horizontal and vertical scans outperform single-direction acquisition. The shared coupling parameter J reaches a posterior standard deviation of 87.0 Hz yet remains biased by 326.9 Hz.

What carries the argument

Sequential Bayesian updates over overlapping scan windows with temporal posterior propagation, using the NV spin-1 Hamiltonian as the local likelihood model that depends on the latent field and shared coupling J.

If this is right

The method recovers the dominant spatial structure of the evolving field on the tested synthetic sequences.
Combined horizontal and vertical scans improve reconstruction accuracy relative to single-direction acquisition.
The posterior for the shared coupling J narrows to 87.0 Hz standard deviation while remaining biased.
Fisher-information and leakage diagnostics reveal a sensitivity-leakage tradeoff under long-interrogation controls.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same windowed-update structure could be tested on fields whose time scale is comparable to the scan duration to check whether propagation remains stable.
Partial identifiability of J appears compatible with accurate field recovery, suggesting that downstream applications may tolerate the observed bias.
The leakage-aware control scoring could be examined for its effect on information gain when the interrogation time is varied systematically.

Load-bearing premise

Local measurements are generated accurately through NV spin dynamics governed by local magnetic-field values and a shared dipolar coupling parameter J whose posterior can be propagated reliably across frames.

What would settle it

An independent experimental run in which the true time-dependent field is known from a separate calibrated sensor and the reported reconstruction error at the final frame exceeds 10^{-6} T would falsify the performance claim.

Figures

Figures reproduced from arXiv: 2605.23455 by Hiroshi Yamauchi, Samuel Tovey, Sophie Colleen Stearn.

**Figure 2.** Figure 2: FIG. 2. Adaptive measurement behavior. (a) Distribution [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

**Figure 5.** Figure 5: FIG. 5. Reconstruction error and scan strategy comparison. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗

**Figure 4.** Figure 4: FIG. 4. Global coupling estimation. (a) Posterior mean and [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗

**Figure 6.** Figure 6: FIG. 6. Qualitative comparison of scan strategies. (a) H [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗

read the original abstract

We propose a quantum-Hamiltonian-learning-based sequential reconstruction framework for dynamic two-dimensional magnetic-field maps using a local likelihood model derived from a nitrogen-vacancy-center spin-1 Hamiltonian. Local measurements are generated through nitrogen-vacancy spin dynamics governed by local magnetic-field values and a shared dipolar coupling parameter, rather than by direct observation of the latent field. Sequential Bayesian updates over overlapping scan windows are combined with temporal posterior propagation to reconstruct the evolving field. Numerical proof-of-concept experiments on controlled synthetic maze-like magnetic-field sequences show that the proposed method reconstructs the dominant spatial structure of the tested field class, achieving a final-frame RMSE of \(7.037\times10^{-7}\,\mathrm{T}\). Adaptive diagnostics show decreasing expected information gain and stable local convergence, while Fisher-information and leakage diagnostics reveal a sensitivity--leakage tradeoff under long-interrogation controls. Combined horizontal and vertical scans yield better reconstruction than single-direction acquisition in the tested setting. In contrast, the shared coupling parameter \(J\) is only partially identifiable: its posterior becomes narrow but remains frame-dependent and biased. At the final checkpoint, \(J_{\rm std}=87.0\,\mathrm{Hz}\), close to a finite-time product-state reference benchmark of \(73.3\,\mathrm{Hz}\), while remaining \(3.35\times\) above a gain-extrapolated ideal-state benchmark. The posterior mean remains biased by \(326.9\,\mathrm{Hz}\), indicating that posterior concentration alone does not imply unbiased coupling recovery. These results demonstrate feasibility for the tested structured field class and identify coupling estimation as the main identifiability bottleneck.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The sequential Bayesian reconstruction idea has some engineering merit but the frame-dependent bias in J directly weakens the field estimates.

read the letter

The paper's main contribution is a sequential Bayesian framework for reconstructing time-varying 2D magnetic fields from NV center measurements, but the persistent bias in the shared dipolar coupling J weakens the support for reliable field reconstruction. What is new here is the use of overlapping scan windows with temporal posterior propagation to handle dynamic fields, along with leakage-aware adaptive controls. The work does a solid job of demonstrating the approach on synthetic data, recovering the main spatial features of maze-like field sequences and reporting a low RMSE of about 7 times 10 to the minus 7 Tesla at the final frame. The inclusion of Fisher information diagnostics and combined horizontal-vertical scans shows some thoughtful engineering of the acquisition strategy. The soft spots are more concerning. The abstract makes clear that J's posterior narrows but remains biased by 326.9 Hz and varies across frames, even though the likelihood model ties B and J together. This bias is not just a side note; it directly impacts the local field inferences, and the paper does not quantify how much it affects the reported RMSE. All validation is synthetic, with no real NV measurements to test against. The partial identifiability of J is acknowledged but seems to be a load-bearing assumption for the temporal propagation to work cleanly. This paper is aimed at researchers in quantum sensing and metrology who are thinking about dynamic field mapping. Someone looking for new algorithmic ideas in Hamiltonian learning might find the sequential update scheme worth a look, though they would need to address the bias issue themselves. I think it deserves peer review. Referees could examine whether the bias can be reduced or if the method still holds up despite it.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a sequential Bayesian reconstruction framework for dynamic 2D magnetic fields using NV-center spin-1 Hamiltonians. Local measurements are generated via spin dynamics depending on both the latent local field B and a shared dipolar coupling J. The method combines Bayesian updates over overlapping scan windows with temporal posterior propagation of J; on synthetic maze-like sequences it recovers dominant spatial structure and reports a final-frame RMSE of 7.037×10^{-7} T, together with diagnostics on information gain, leakage, and adaptive control scoring. The abstract explicitly notes that J is only partially identifiable, with its posterior narrowing (std = 87.0 Hz) yet remaining frame-dependent and biased by 326.9 Hz.

Significance. If the effect of the reported J bias on the inferred B field can be quantified and the framework extended beyond synthetic data, the approach would offer a concrete route to Hamiltonian-learning-based sensing of time-varying fields, with built-in adaptive diagnostics and a demonstrated sensitivity-leakage tradeoff. The explicit comparison of the J posterior to product-state and ideal-state benchmarks is a strength, as is the open reporting of partial identifiability; however, the absence of real-measurement validation and unquantified bias propagation limit the immediate impact.

major comments (3)

[Abstract] Abstract: the reported RMSE of 7.037×10^{-7} T is obtained by propagating the J posterior into the local likelihood p(data | B_local, J); the documented mean bias of 326.9 Hz (frame-dependent and 3.35× above the gain-extrapolated benchmark) therefore directly distorts the B inference, yet no propagation of this systematic error into the RMSE or spatial-structure recovery is provided.
[Abstract] Abstract / Numerical experiments: the reconstruction rests on learning the shared J from the identical measurement model used for field inference; the observed frame-dependence and deviation from both the 73.3 Hz product-state and ideal-state benchmarks indicate that the validation is not fully independent, weakening support for the central claim that the sequential updates reliably reconstruct the evolving field.
[Abstract] Abstract: the soundness assessment notes the lack of full derivation details, error propagation, or validation against real measurements; because the likelihood is jointly sensitive to B and J, these omissions are load-bearing for assessing whether the reported RMSE and adaptive-policy results are robust.

minor comments (2)

[Abstract] The abstract could state the J-identifiability limitations more prominently in the opening summary rather than only in the final paragraph.
Notation for the NV spin-1 Hamiltonian and the precise form of the local likelihood could be introduced earlier to aid readability for readers outside quantum sensing.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below. We agree that further quantification of bias propagation is needed and will revise the manuscript to include this analysis. The synthetic nature of the validation is explicitly stated, and we note the scope limitations accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: the reported RMSE of 7.037×10^{-7} T is obtained by propagating the J posterior into the local likelihood p(data | B_local, J); the documented mean bias of 326.9 Hz (frame-dependent and 3.35× above the gain-extrapolated benchmark) therefore directly distorts the B inference, yet no propagation of this systematic error into the RMSE or spatial-structure recovery is provided.

Authors: We agree that explicit propagation of the J bias into the B inference metrics is warranted. In the revised manuscript we will add a sensitivity analysis comparing B reconstructions obtained with the learned (biased) J posterior versus the true generative J, quantifying the resulting change in RMSE and spatial structure fidelity. revision: yes
Referee: [Abstract] Abstract / Numerical experiments: the reconstruction rests on learning the shared J from the identical measurement model used for field inference; the observed frame-dependence and deviation from both the 73.3 Hz product-state and ideal-state benchmarks indicate that the validation is not fully independent, weakening support for the central claim that the sequential updates reliably reconstruct the evolving field.

Authors: The synthetic experiments deliberately employ the known generative model to enable quantitative ground-truth evaluation of both field recovery and J identifiability; this is standard practice and is accompanied by explicit reporting of partial identifiability, frame dependence, and benchmark comparisons. The low RMSE despite the documented J bias already illustrates robustness of the sequential updates. We will add clarifying language in the abstract and discussion to emphasize the controlled, model-consistent nature of the validation. revision: partial
Referee: [Abstract] Abstract: the soundness assessment notes the lack of full derivation details, error propagation, or validation against real measurements; because the likelihood is jointly sensitive to B and J, these omissions are load-bearing for assessing whether the reported RMSE and adaptive-policy results are robust.

Authors: We will expand the methods section with additional derivation steps and a dedicated error-propagation subsection in the revision. Validation on real NV-center measurements lies outside the present scope, which focuses on the algorithmic framework and controlled synthetic demonstrations. revision: partial

standing simulated objections not resolved

Validation against real experimental NV-center measurements

Circularity Check

0 steps flagged

No significant circularity detected in the derivation chain

full rationale

The paper applies standard sequential Bayesian updating to a likelihood derived from the NV spin-1 Hamiltonian, jointly inferring local B fields and the shared coupling J from synthetic data generated under the same model. The reported RMSE is an empirical validation metric on held-out synthetic sequences rather than a quantity forced by construction from fitted inputs. No self-citations appear as load-bearing premises, no uniqueness theorems are imported from prior author work, and no ansatz or renaming reduces the central reconstruction claim to its inputs. The acknowledged bias and frame-dependence in the J posterior constitute a correctness or identifiability concern but do not create a circular reduction in the method itself.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on standard Bayesian inference assumptions plus the NV spin-1 Hamiltonian model; the only explicit free parameter is the shared coupling J learned from data.

free parameters (1)

shared dipolar coupling J
Learned via posterior updates; reported as frame-dependent with bias of 326.9 Hz relative to benchmarks.

axioms (2)

domain assumption Bayesian sequential updates and temporal posterior propagation correctly capture the inference dynamics
Invoked throughout the sequential reconstruction description.
domain assumption NV-center spin-1 Hamiltonian accurately generates the local likelihood from field and J
Core modeling assumption stated in the abstract.

pith-pipeline@v0.9.0 · 5824 in / 1246 out tokens · 48778 ms · 2026-05-25T04:42:36.014393+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

local likelihood model derived from a nitrogen-vacancy center spin-1 Hamiltonian... Sequential Bayesian updates over overlapping scan windows... shared dipolar coupling parameter J
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

root-mean-square error of 7.037×10^{-7} T... J_std=87.0 Hz... biased by 326.9 Hz

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.