{"id":"c69a6c09-1942-4903-94ee-7d6cfd59f222","arxiv_id":"2411.15323","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"The authors demonstrate quantitative 3D reconstruction of the magnetisation vector field in a ferromagnetic cobalt nanowire from electron holographic phase measurements, with sub-50 nm spatial resolution.","lead":"Using electron holography and model-based iterative reconstruction, the authors reconstructed the 3D magnetisation vector field inside a cobalt nanowire junction, revealing a curved vortex domain wall. The work demonstrates a route to quantitative 3D magnetic imaging at the nanoscale, relevant for future 3D magnetic memory and spintronic devices.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Reconstruction accuracy rests on an unvalidated 3D geometry model; mask sensitivity is not demonstrated, so the 50-nm accuracy claim remains conditional.","rationale":"The paper is a careful proof-of-concept with several internally consistent diagnostics: Fourier shell correlation indicates feature SNR above the half-bit threshold, optimal estimation gives a 43 nm point-spread function and 0.01 T precision, the reconstructed Ms values match EELS-based composition estimates, and the solution is reported stable over three orders of magnitude of regulariser weights. These are genuine supporting checks. However, none of them validates the absolute accuracy of the 3D M vector field because all are computed under the same forward model and geometry. The reader's weakest assumption identifies this precisely: the mask threshold was adjusted after seeing a first reconstruction, which creates a risk of circularity, and the quoted volume-mismatch simulation only covers a global 10% error, not the spatially structured mask errors that would arise from the carbon shell or missing-wedge elongation. A mask sensitivity test with eroded and dilated masks would settle whether the reconstruction is robust to plausible geometry errors. Since such a test is not in the paper, the CONDITIONAL verdict remains appropriate; no change is needed based on this stress-test pass.","tokens_in":16079,"tokens_out":6405,"duration_ms":64087,"concrete_test":"Using the released dataset and documented software, rerun the MBIR pipeline with at least three masks: (i) the published mask, (ii) the mask dilated by one 10.2 nm voxel, and (iii) the mask eroded by one voxel; if an EELS-based Co mask at the stated 45% to 60% cobalt boundaries can be generated from Fig. 3e, include that as a fourth mask. Then compare the reconstructed M fields, quantifying the shift in vortex core position and the change in mean Ms in the two nanowire arms. If the vortex core moves by more than approximately 20 nm or Ms changes by more than 0.1 T, the 50-nm accuracy claim is not robust to geometry uncertainty. This test requires no new experiment and directly probes the weakest assumption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the reconstructed magnetisation is accurate for domains larger than 50 nm depends on the forward model in Eq. 5 being a faithful mapping from the true sample to the measured phase. That forward model uses a 3D geometrical mask (Fig. 5d) generated by back-projecting thresholded electrostatic phase masks and cropping to SEM dimensions. The threshold was refined after the first reconstruction to balance surface artefacts, and the paper's own error estimate in Supplementary Material [S1] assumes a 10% volume mismatch. This is not a validation of the mask: a 10% global volume error can accompany a spatially structured misassignment, such as including the non-magnetic carbon shell detected by EELS or excluding magnetic protrusions, which would bias M locally even if the global volume error is within 10%. The observed phase residual RMS of 0.38 rad, much larger than the 0.016 rad noise floor, means the data do not independently confirm the model. Because the mask defines where M may be non-zero, any mask error directly corrupts the reconstruction; the reported stability of the solution when varying regulariser weights does not address this. In the absence of an independent 3D measurement of M on the same sample, the accuracy claim is conditional on the geometric model being correct.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The authors present an experimental application of model-based iterative reconstruction (MBIR) to off-axis electron holographic vector field tomography (EH-VFT) data from an L-shaped FEBID cobalt nanowire. They record two tilt arcs of holograms, separate magnetic and electrostatic phase contributions, align the images, and generate a 3D geometrical mask from back-projected electrostatic phase images cropped to SEM dimensions. An inverse regularised forward model (Eq. 5) is minimised against the measured magnetic phase images to obtain a 3D magnetisation vector field, revealing a U-shaped vortex domain wall at the nanowire intersection. The paper reports diagnostics including a Fourier shell correlation, a point-spread/averaging-kernel analysis, residual analysis, and comparison of reconstructed µ0Ms with EELS cobalt composition, and claims that the reconstructed magnetisation is accurate for magnetic domains larger than about 50 nm.","tokens_in":16313,"tokens_out":7968,"duration_ms":75553,"significance":"If the central claim holds, this is a notable step beyond earlier EH-VFT reconstructions of the B field: it is a quantitative 3D reconstruction of M from TEM phase data, with publicly available data and a realistic discussion of null-space limitations. The paper's strengths include a careful phase-separation protocol, explicit error diagnostics (FSC, averaging kernel, residual analysis), robustness of the reconstruction to regulariser weights over three orders of magnitude, and consistency between reconstructed Ms and EELS composition. The main weakness is that the quantitative accuracy claim rests on assumptions about the 3D geometric mask and on simulations and comparisons with similar, rather than identical, samples, rather than on a direct validation of the same sample.","major_comments":[{"comment":"The 3D mask generated by back-projecting thresholded electrostatic phase masks and cropping to SEM dimensions defines the support of the reconstructed magnetisation, so mask errors enter the reconstruction directly. EELS shows a nonmagnetic carbon shell, yet the mask threshold was refined after the first reconstruction 'so that similar amounts of over- and underestimating surface artefacts were present' (Results, Uniform region reconstruction). The only quantitative geometry-error estimate cited is the 10% volume-mismatch assumption in Supplementary Material S1; a 10% global volume error does not exclude a spatially structured misassignment, such as including part of the carbon shell or excluding a magnetic protrusion. Because the phase residual RMS is 0.38 rad, the phase data do not independently validate the mask. I request a mask-sensitivity analysis (for example, dilate and erode the mask by several nanometres, or re-run the reconstruction with independent threshold choices) and, if such an analysis is not possible, an explicit statement that the 50-nm accuracy claim is conditional on the geometrical model.","section":"Alignment of phase images / Three-dimensional reconstruction of M (Fig. 5d, Eq. 5)"},{"comment":"The stated precision of 0.01 T per pixel is obtained by propagating the 0.016 rad phase-noise floor through the error gain matrix G. The actual reconstruction residual has an RMS of 0.38 rad, about 24 times larger, and is not shown to be random. If the residual arises from mask error, residual misalignment, or missing-wedge artefacts, the linear error propagation underestimates the systematic uncertainty in µ0Ms and in the 50-nm accuracy claim. The authors should either demonstrate that the residual becomes compatible with the noise floor once model errors are accounted for, or propagate the residual structure explicitly (for example, by adding the residual field to the measured phases and re-running the reconstruction) to estimate a model-error contribution.","section":"Diagnostics of the reconstruction (Eq. 6, Fig. 8)"},{"comment":"The external validation is made against X-ray laminography and micromagnetic simulations of similar, but not identical, FEBID cobalt nanostructures. The reconstructed vortex and the Ms values are consistent with that body of work, but this consistency does not by itself quantify the accuracy of the particular reconstruction presented here, especially for the 50-nm-domain claim. I recommend that the paper separate internal diagnostics (FSC, averaging kernel, residual analysis, EELS correlation) from external comparison, and state explicitly which components of the accuracy claim each type of evidence supports. Without this separation, the abstract's claim that the reconstruction 'is shown to be accurate' is stronger than the evidence in the paper.","section":"Discussion / Conclusions"}],"minor_comments":[{"comment":"The Fourier shell correlation is computed by randomly halving the 3D reconstruction and interpolating missing values, which is not the standard split-dataset FSC and may overestimate the correlation because the interpolated values are not independent. The 14.8 nm figure should be treated with caution, or the analysis should be replaced by an FSC calculated from two independent reconstructions (for example, from odd and even tilt angles).","section":"Results / Diagnostics (Fig. 7)"},{"comment":"The tilt ranges are stated inconsistently: the acquisition ranges are given as -60° to 30° and -60° to 0° in the Materials and methods, while the abstract and Outlook refer to ±60° tilt arcs. Please reconcile these statements.","section":"Materials and methods / Abstract / Outlook"},{"comment":"Please correct typographical errors, including 'cotains' (Results), 'inflluece' (Diagnostics), 'show in Fig. 3b' (Materials and methods), and 'I addition' (Outlook).","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a solid proof-of-concept with a novel experimental pipeline, but the central accuracy claim is more conditional than the abstract suggests. The main gap is the absence of a mask-sensitivity study, which is load-bearing because the geometric mask defines the support of the reconstructed magnetisation. If the authors can add such a study, or rephrase the accuracy claim as conditional on the model, the paper would be acceptable after revision. I do not see grounds for rejection on novelty or circularity."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is the first experimental reconstruction of the 3D magnetisation vector field, not just the induction B, from electron holographic vector field tomography, and it looks like a genuine step forward. Second, the load-bearing assumption is the 3D geometric model that says where magnetic material is, and the paper does not do the sensitivity analysis that would make the 50 nm accuracy claim solid.\n\nThe authors get a lot right. They acquire two tilt arcs up to ±60°, use a forward model that directly relates M to the measured phase, and vary the regulariser weights over three orders of magnitude with the same features surviving. The diagnostics are serious: Fourier shell correlation, an averaging-kernel analysis giving a 43 nm FWHM point spread, and error propagation from the 0.016 rad phase noise to 0.01 T per voxel. The reconstructed Ms tracks the EELS cobalt content, and the vortex domain wall geometry is plausible. The writing is candid about surface artefacts, null spaces, and the fact that the mask threshold was adjusted after the first pass.\n\nThe soft spot the reader flagged is real. The forward model uses a geometric mask built by back-projecting thresholded electrostatic phase images and cropping to SEM dimensions. The threshold was refined to balance artefacts, and the supplement's error estimate assumes a 10% volume mismatch. A 10% global volume error doesn't rule out a spatially structured misassignment—for example, treating the carbon shell as magnetic, or missing a magnetic protrusion. That would bias M locally even if the phase residuals are small. The residual RMS is 0.38 rad against a 0.016 rad noise floor, so the data are not tightly constraining the model. The averaging-kernel analysis is done for one voxel; the authors say it's similar elsewhere but don't show it. These are conditional-support issues, not fatal ones.\n\nThis paper is for TEM method developers and anyone working on 3D magnetic nanostructures. It deserves a serious referee. The method is novel, the data are public, and the limitations are stated honestly. I'd ask for a mask sensitivity study in revision—perturb the threshold, erode/dilate the mask, re-reconstruct, and show how the vortex core and Ms histograms move. I'd also ask for code. My verdict would be conditional acceptance, not reject.","headline":"First experimental quantitative 3D M reconstruction from EH-VFT – real novelty, solid diagnostics, but the 50 nm accuracy claim depends on an unvalidated geometric mask.","tokens_in":16898,"tokens_out":3091,"would_cite":true,"duration_ms":27219,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Model-based iterative reconstruction recovers the 3D magnetisation vector field inside a cobalt nanowire from electron holography, resolving magnetic domains down to about 50 nanometres.","keywords":["off-axis electron holography","electron tomography","model-based iterative reconstruction","magnetisation vector field","3D magnetic imaging","FEBID cobalt nanowire","vortex domain wall","Lorentz microscopy"],"falsifier":"Image a uniformly magnetised FEBID cobalt nanowire with known saturation magnetisation and compare the MBIR reconstruction to that known value: if the reconstructed $\\mu_0 M_s$ varies spatially or deviates from the known value, or if eroding or dilating the geometrical mask by 10% materially changes the reconstructed domain structure, the method's central claim would be called into question.","tokens_in":15882,"feed_emoji":"🧲","tokens_out":8356,"duration_ms":64448,"temperature":0.7,"pith_summary":"This paper establishes that model-based iterative reconstruction (MBIR) can recover the three-dimensional magnetisation vector field $\\vec{M}$ inside a real nanostructure from electron holographic phase measurements, rather than only the magnetic induction $\\vec{B}$ as in previous electron holographic vector field tomography. The method is applied to an L-shaped cobalt nanowire grown by focused electron beam induced deposition, and the reconstruction reveals multiple magnetic domains and a curved vortex domain wall at the wire intersection. The reconstruction is quantitative, with a per-voxel precision of 0.01 T, and is stated to be accurate for magnetic domains larger than approximately 50 nm. If correct, this provides a direct TEM-based route to 3D magnetisation mapping in nanoscale magnetic devices, where the local magnetisation governs switching and domain-wall behaviour.","feed_headline":"Nanowire magnetisation mapped in 3D by electron holography","feed_subtitle":"MBIR reveals magnetic domains and a vortex wall in a FEBID cobalt nanowire with sub-50 nm accuracy.","key_machinery":"The central machinery is a model-based iterative reconstruction (MBIR) loop: a forward model simulates the magnetic electron phase shift that a trial 3D vector field $\\vec{M}$ would produce; a cost function $C = \\sum_i (\\phi_{i,\\mathrm{meas}} - \\phi_{i,\\mathrm{sim}}(\\vec{M}))^2 + \\lambda_1 \\sum_j (\\vec{\\nabla} M_j \\cdot \\vec{\\nabla} M_j) + \\lambda_2 \\mathrm{var}(|\\vec{M}|)$ is minimised by conjugate gradient, with regularisers that favour ferromagnetic order and uniform magnetisation magnitude. The magnetic phase is separated from the electrostatic phase by flipping the specimen 180°, and a 3D geometrical mask, built by back-projecting thresholded electrostatic phase images and cropping to SEM dimensions, defines where magnetic material is assumed to sit. Alignment of the tilt series uses a common-lines symmetry method and affine distortion corrections. The paper also stresses the role of null spaces — magnetisation configurations that produce no phase shift in any projection — which limit reconstructibility but are absent in this sample.","core_discovery":"The paper reports an experimental reconstruction of a 3D magnetisation distribution $\\vec{M}_{\\mathrm{rec}}$ from an electron holographic vector field tomography dataset, using MBIR to find the $\\vec{M}$ that best fits 16 magnetic phase images recorded over two tilt arcs (up to $\\pm 60^\\circ$ and $-60^\\circ$ to $0^\\circ$). The reconstruction of the L-shaped FEBID cobalt nanowire shows single-domain regions in the two arms with $\\mu_0 M_s$ values consistent with the measured cobalt content, and a U-shaped vortex domain wall occupying the full intersection volume. The point spread function of the reconstruction has a FWHM of 43 nm, and a Fourier shell correlation indicates features above 14.8 nm have sufficient signal-to-noise; the authors therefore state that the reconstruction is accurate for domains larger than about 50 nm.","pith_inferences":["The dependence of the reconstruction on the geometrical mask suggests that combining MBIR with an independent structural measurement (e.g., atomic-scale STEM tomography or EELS-based volume segmentation) could eliminate the main source of systematic bias identified in the paper.","The null-space analysis implies that the method will struggle with Néel-type walls or certain spin textures that produce no phase signal; a testable extension would be to reconstruct a sample known to contain such a texture and compare the result with micromagnetic simulations.","Since the paper shows the regulariser weights can vary over three orders of magnitude without changing the solution, the reconstruction is measurement-dominated; this suggests that phase-resolved tomographic data alone may be enough to constrain $\\vec{M}$ when the geometry is known, which is useful for automated pipelines.","The 50 nm accuracy bound is tied to this dataset's alignment and voxel size; the paper's own reasoning implies that with 1 nm voxels and improved distortion correction, a 3 nm resolution is in reach, making MBIR competitive with X-ray laminography."],"forward_implications":["If the reconstruction is correct, TEM phase measurements can now deliver the full 3D magnetisation vector, not just the projected induction, giving a compact local descriptor of magnetic order inside a nanostructure.","The method should apply to other Lorentz microscopy techniques (differential phase contrast, ptychography) because they all sense the same magnetic phase shift.","For samples with no null spaces, two complete tilt arcs are sufficient; the paper's simulation-based error analysis states that a 10% volume mismatch between model and sample leads on average to less than 10% error per voxel in $\\vec{M}$.","The demonstrated accuracy for domains above roughly 50 nm sets a benchmark; sub-50 nm textures would require smaller voxels, better alignment, and more advanced algorithms, which the paper argues could ultimately approach the atomic scale.","The ability to measure $\\mu_0 M_s$ locally allows direct correlation between reconstructed magnetisation and chemical composition maps (e.g., from energy-loss spectroscopy)."],"supporting_citations":[{"why":"Supplies the theoretical development and forward model for MBIR of magnetisation from electron optical phase images.","marker":"[Caron, 2018a]"},{"why":"Provides simulation-based tests of MBIR on uniformly-magnetised nanowires and vortex configurations, justifying tilt increments.","marker":"[Caron, 2018b]"},{"why":"Introduces vector field electron tomography for magnetic materials, the foundational theory for EH-VFT.","marker":"[Phatak et al., 2008]"},{"why":"Demonstrates holographic vector field electron tomography of 3D nanomagnets, the approach this paper extends from B to M.","marker":"[Wolf et al., 2019]"},{"why":"An earlier MBIR implementation for 3D magnetisation reconstruction from vector field electron tomography, which this work builds upon.","marker":"[Aditya Mohan et al., 2018]"},{"why":"Common-lines method used to refine projection orientations in the tilt series alignment.","marker":"[Penczek et al., 1996]"},{"why":"Specimen holder enabling high tilt angles and accurate tilt measurement needed for two-arc acquisition.","marker":"[Diehle et al., 2021]"},{"why":"Provides the half-bit Fourier shell correlation threshold used to judge reconstruction resolution.","marker":"[van Heel and Schatz, 2005]"},{"why":"Supplies optimal estimation diagnostics methodology for error analysis.","marker":"[Ungermann et al., 2010]"}],"fun_headline_variants":["3D magnetisation of a nanowire reconstructed via MBIR","Electron holography yields 3D spin map of a cobalt nanowire","MBIR and electron holography unveil magnetic domains in a nanowire","Nanowire's 3D magnetisation resolved by model-based iteration","Holographic tomography and MBIR map nanoscale magnetism"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reconstruction is only as good as the 3D geometrical model that defines where magnetic material sits; that model is built from thresholded electrostatic phase images and SEM cropping, and the paper's error analysis assumes a 10% volume mismatch between model and sample, so if the mask is wrong the forward model and reconstructed magnetisation are biased.","fun_headline_variants_meta":{"raw":{"variants":["3D magnetisation of a nanowire reconstructed via MBIR","Electron holography yields 3D spin map of a cobalt nanowire","MBIR and electron holography unveil magnetic domains in a nanowire","Nanowire's 3D magnetisation resolved by model-based iteration","Holographic tomography and MBIR map nanoscale magnetism"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000702,"raw_usage":{"total_tokens":3181,"prompt_tokens":970,"completion_tokens":2211,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":586,"completion_tokens_details":{"reasoning_tokens":2117}},"tokens_in":586,"tokens_out":2211,"duration_ms":13493,"temperature":1.0,"reasoning_tokens":2117,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:26:31.235422+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Image a uniformly magnetised FEBID cobalt nanowire with known saturation magnetisation and compare the MBIR reconstruction to that known value: if the reconstructed $\\mu_0 M_s$ varies spatially or deviates from the known value, or if eroding or dilating the geometrical mask by 10% materially changes the reconstructed domain structure, the method's central claim would be called into question.","supporting_citations":[{"cited_title":"Vector field electron tomography of magnetic materials: Theoretical development","cited_arxiv_id":null,"evidence_quote":"Introduces vector field electron tomography for magnetic materials, the foundational theory for EH-VFT."},{"cited_title":"Holographic vector field electron tomography of three-dimensional nanomagnets","cited_arxiv_id":null,"evidence_quote":"Demonstrates holographic vector field electron tomography of 3D nanomagnets, the approach this paper extends from B to M."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"An earlier MBIR implementation for 3D magnetisation reconstruction from vector field electron tomography, which this work builds upon."},{"cited_title":"Dunin-Borkowski","cited_arxiv_id":null,"evidence_quote":"Specimen holder enabling high tilt angles and accurate tilt measurement needed for two-arc acquisition."},{"cited_title":"Fourier shell correlation threshold criteria","cited_arxiv_id":null,"evidence_quote":"Provides the half-bit Fourier shell correlation threshold used to judge reconstruction resolution."},{"cited_title":"Towards a 3- D tomographic retrieval for the air-borne limb-imager GLORIA","cited_arxiv_id":null,"evidence_quote":"Supplies optimal estimation diagnostics methodology for error analysis."}],"review_version":1}