Pith. sign in

REVIEW 4 major objections 4 minor 3 references

Quantitative Phase Retrieval and Characterization of Magnetic Nanostructures via Lorentz (Scanning) Transmission Electron Microscopy

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Electron ptychography retrieves the total phase of magnetic nanostructures more accurately and at higher resolution than TIE, RMAD, or off-axis holography in this comparison.

desk verdict A useful head-to-head benchmark of phase retrieval for magnetic nanostructures, with a plausible ePIE ranking but a fitted and unquantified saturation magnetization that needs work. read the letter →

arxiv 2412.21005 v1 pith:5BS77KX5 submitted 2024-12-30 cond-mat.mtrl-sci

Kayna L. Mendoza , Haoyang Ni , Georgios Varnavides , Miaofang Chi , Colin Ophus , Amanda Petford-Long , Charudatta Phatak This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords LorentzTEMSTEMphaseretrievalmagneticmaterialstransport-of-intensityequationautomaticdifferentiationoff-axisholographyelectronptychography
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper compares four phase-retrieval routes for magnetic nanostructures—transport-of-intensity (TIE), reverse-mode automatic differentiation (RMAD), off-axis holography (OAH), and the extended ptychographic iterative engine (ePIE)—and asks which one gives the most faithful reconstruction of the electron phase shift. On simulated Permalloy island arrays and on an experimental 36 nm NiFe nanowire, ePIE retrieves the total phase with the highest spatial resolution and phase accuracy, especially in the magnetic contribution, while TIE blurs at large defocus and RMAD produces nonphysical stray fields for the irregular nanowire. OAH remains accurate but is noisier and requires a reference vacuum region. The recovered phase-gradient maps expose stray-field and inter-island proximity effects, and yield a nanowire saturation magnetization of $478.8\times10^3$ A/m, about 57% of bulk Permalloy.

What carries the argument

The load-bearing quantity is the total electron phase shift $\phi_t(\mathbf{r}_\perp)=\phi_e+\phi_m$, related by the standard Aharonov-Bohm relation to the projected electrostatic potential and the magnetic vector potential; the gradient $\nabla\phi_t$ maps the projected magnetic induction, so the stray fields and inter-island interactions appear as faint phase signals outside the sample edges. The four algorithms constitute the machinery: TIE solves the transport-of-intensity equation from through-focal intensities; RMAD fits the same focal series by reverse-mode automatic differentiation, iteratively updating a phase model to match measured intensities; OAH reconstructs the phase from a biprism interference pattern; and ePIE alternates updates of the complex object and illuminating probe using overlapping diffraction patterns recorded in Lorentz-mode 4D-STEM. ePIE with batch size one is described as a limiting case of stochastic gradient descent, and it is the reconstructed phase gradient, not the raw phase, that serves as the quantitative magnetic observable.

What would settle it

Reconstruct the same 36 nm NiFe nanowire from the published 4D-STEM dataset with a multislice ptychographic model that accounts for the wire's thickness and compare the magnetic phase shift outside the wire with the single-slice ePIE result; a change larger than about 0.1 radians would show that the single-slice weak-phase assumption, not the algorithm, controls the reported agreement.

Watch

Extended reading notes

Core claim

The central claim is that electron ptychography, specifically the extended ptychographic iterative engine (ePIE) implemented as stochastic gradient descent with a batch size of one, is the most robust of the compared methods for quantitative magnetic phase retrieval. In simulations of Permalloy nanoislands, ePIE's reconstructions track the ground truth closely, with structural similarity indices of 93.36% for the total phase and 98.24% for the phase gradient, while TIE loses low-spatial-frequency signal at $\pm100\,\mu$m defocus. In the experimental NiFe nanowire, the ePIE phase gradient matches the micromagnetic simulation in orientation and intensity distribution, with the magnetic phase shift outside the wire differing from simulation by 0.106 radians; OAH differs by 0.278 radians and RMAD's stray field is nonphysical. The paper further reports a measured saturation magnetization of $478.8\times10^3$ A/m for the 36 nm nanowire and shows that phase-gradient maps can visualize proximity effects between neighboring magnetic islands.

Load-bearing premise

The numbers assume the nanowire is thin enough that the electron wave changes in one flat slice and that the microscope's probe and focus are known exactly; a strongly scattering, irregular 36 nm wire can break that assumption.

Editorial extensions

If this is right

  • A single 4D-STEM dataset should be sufficient for quantitative magnetic phase maps, removing the reference vacuum region that off-axis holography needs and opening extended or embedded samples to measurement.
  • RMAD phase retrieval should not be relied on for inhomogeneous, strongly scattering nanostructures unless the forward model accounts for multiple scattering or geometric irregularity.
  • TIE remains a fast screening tool for magnetic induction, but its resolution penalty at high defocus makes it unsuitable for few-nanometer quantitative work.
  • The reported $478.8\times10^3$ A/m saturation magnetization implies that Lorentz phase measurements can detect property changes—oxidation, size effects, or contamination—in individual nanowires.
  • Phase-gradient maps can serve as a practical way to visualize stray-field coupling in arrays of coupled magnetic islands, relevant to artificial spin ice and related systems.

Reading between the lines

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

  • An implicit extension of the ePIE result is that the same single-dataset workflow could map magnetic textures in continuous films and devices where OAH cannot be used, and in principle be combined with tilt series for three-dimensional magnetic induction retrieval.
  • A testable consequence the paper does not draw: if RMAD's failure on the nanowire comes from the single-slice approximation, a multislice forward model applied to the same published through-focal dataset should remove the nonphysical upward stray-field trajectory.
  • The low saturation magnetization invites a dedicated check: spatially resolved core-loss spectroscopy across the same nanowire would test whether the reduction is a surface oxide shell or an intrinsic size effect.
  • The reversed magnetization direction between ePIE and OAH, attributed to specimen handling, suggests a controlled before-and-after transfer experiment could quantify how much mechanical handling perturbs the magnetic state of such nanowires.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The manuscript compares three phase-retrieval methods—TIE, RMAD, and ePIE—for Lorentz TEM and Lorentz 4D-STEM imaging of magnetic nanostructures. It first benchmarks the methods on simulated images of Permalloy nano-islands using SSIM and line profiles, then applies OAH, RMAD, and ePIE to experimental data from a 36 nm diameter NiFe nanowire, comparing the reconstructed total phase and phase gradient against a micromagnetic simulation. From this comparison, the authors report a saturation magnetization of 478.8 × 10^3 A/m for the nanowire and conclude that ePIE is the most robust method in terms of spatial resolution and phase accuracy, especially for magnetic contributions.

Significance. If the conclusions are supported, the paper would provide a useful practical comparison of widely used phase-retrieval tools for nanoscale magnetic imaging, with potential impact on how researchers choose between LTEM and Ltz-4D-STEM methods. The manuscript benefits from using open-source software (PyLorentz, ADLTEM, py4DSTEM), publicly archived raw data, and a clear statement of experimental parameters. However, the significance is currently limited by unresolved quantitative issues: the key experimental ePIE reconstruction uses an estimated defocus rather than a measured one, the saturation magnetization is a fitted simulation parameter without an uncertainty estimate, and the central claim that ePIE is 'most robust' is not fully supported by the paper's own simulation metrics.

major comments (4)
  1. [§2.2.2, Methods] The ePIE reconstruction of the NiFe nanowire relies on a defocus value that is only estimated ('Defocus of the dataset is estimated to be 11 µm'). In single-slice ptychography, an incorrect probe defocus can be partially absorbed into the reconstructed object as low-order phase errors, which then propagate directly into the total phase gradient and the stray-field analysis. Since the paper's headline claim about ePIE's magnetic phase accuracy depends on this unmeasured parameter, the authors should provide an independent probe calibration, a defocus sensitivity analysis, or a quantitative demonstration that the extracted phase differences (2.096 rad versus 1.990 rad from simulation) are stable over a plausible range of defocus values.
  2. [§3, Discussion] The reported saturation magnetization of 478.8 × 10^3 A/m is presented as an experimental finding, but according to the text it is obtained by running micromagnetic simulations and matching their phase to the experimental phase. This is a fitted input parameter, not an independent prediction, and the fitting procedure, the parameter ranges explored, and any uncertainty estimate are not reported. As a fitted value, it cannot serve as corroboration of the ePIE reconstruction. The authors should either clearly label this as a model-dependent fit with error bars derived from the reconstruction uncertainties, or remove it from the list of validated findings.
  3. [§2.1, Fig. 1 and §3, Discussion] The claim that ePIE 'proves to be the most robust' in the simulated comparison is not supported by the manuscript's own quantitative metric: RMAD yields a higher SSIM than ePIE for both the total phase (94.87% vs 93.36%) and the phase gradient (98.995% vs 98.24%). The text justifies the preference for ePIE using the line profiles in Fig. 2 and the weak-phase variations, but that argument is not reconciled with the global SSIM scores. The authors should either present a consistent quantitative ranking or weaken the claim to say that ePIE and RMAD are comparably accurate in simulations, with ePIE showing specific advantages in certain line-profile features.
  4. [§2.2.3, Fig. 5] Using the OAH reconstruction as the 'experimental ground truth basis' is problematic for a comparison whose goal is to assess phase accuracy, because OAH itself contains noise and distortions that the authors acknowledge. In addition, the ePIE reconstruction shows a reversed magnetization direction relative to both OAH and the simulation, which is attributed to sample handling but is not quantitatively reconciled. If the ePIE phase is sign-flipped for comparison, the reported phase differences need to be recomputed with the sign convention stated; if not, the quoted agreement between ePIE and the simulation (difference of 0.106 rad) is not a direct comparison. Please clarify the sign handling and report the sensitivity of the quantitative comparisons to the OAH noise level.
minor comments (4)
  1. [Throughout] There are multiple typographical errors and duplicated words, e.g., 'Morevover' in the Introduction, 'we explored we explored' in §2.2, 'betwen' in §2.2.3, and inconsistent use of 'SFIG' versus 'Fig.' for supplementary figures.
  2. [Fig. 1 caption] The caption references subfigures (d), (e), (f) as total phase reconstructions and (g), (h), (i) as phase gradients, but the main text refers to 'Fig. 1(d)' as a line plot and 'Fig. 1(e)' as the TIE reconstruction, which is confusing. Please renumber the panels or correct the cross-references.
  3. [Methods, §5] The statement that 'we note that these will vary based on individual datasets' after listing ePIE reconstruction parameters is vague; it would be more useful to state explicitly which parameters are robust and which are dataset-dependent.
  4. [References] Several references are incomplete or malformed, for example Ref. 18 ends with 'zhou (2021)' instead of the full citation, and Ref. 21 lacks author and venue information.

Circularity Check

2 steps flagged · score 6.0 of 10

The simulated ePIE benchmark is self-referential because the synthetic 4D-STEM dataset was generated with ePIE, and the reported saturation magnetization is a simulation-matched fit; the experimental OAH comparison supplies the main independent support.

  1. self definitional [Methods, Section 2.1; Discussion, Section 3]
    "Ltz-4D-STEM dataset was generated using ePIE with microscope parameters equivalent to LTEM simulation parameters presented above. ... It is apparent that the extended ptychographic iterative engine (ePIE), i.e. stochastic gradient descent with a small batch size, proves to be the most robust in retrieving the object’s phase with the highest spatial resolution and phase accuracy, especially as it pertains to magnetic contributions."

    The synthetic 4D-STEM data used to benchmark ePIE are produced by the same single-slice forward model that the ePIE reconstruction inverts. Evaluating an algorithm on data generated by its own model is a self-consistency check, not an independent test against TIE and RMAD, which were tested on a separate PyLorentz through-focal simulation. The simulated phase retrieval therefore partly returns the generative model's own assumptions, so the 'most robust' conclusion drawn from the simulations is forced by construction rather than by an independent physical benchmark.

  2. fitted input called prediction [Abstract; Discussion, Section 3]
    "determine the magnetization saturation through corroborations with micromagnetic simulations ... the saturation magnetization of the 36 nm diameter NiFe nanowire was found to be 478.8 ×10^3 Amperes/meter, approximately 57% that of Permalloy (840×10^3 Amperes/meter) which we used to simulate our conditions initially."

    The only stated route to the experimental Ms is 'corroborations with micromagnetic simulations,' and the Methods specify that those simulations were initialized with Ms = 840×10^3 A/m. If Ms is varied until the simulated phase matches the experimental phase, the 'found' value is the fitting parameter itself, not an independent measurement; reporting it as a discovered property renames the input of the matching loop as an output. No inversion equation, uncertainty, or independent magnetometry is provided, so the derived value is not shown to be independent of the simulation input.

full rationale

The paper's strongest independent content is the experimental comparison: OAH is an interferometric measurement with a reference hologram, and ePIE's phase difference (2.096 rad) is compared against OAH (2.268 rad) and a micromagnetic simulation (1.990 rad), so the experimental portion is not circular. However, the simulation-based argument for ePIE's robustness is partially self-referential because the synthetic Ltz-4D-STEM dataset was generated with ePIE itself, making the simulated ePIE reconstruction a self-consistency test. The reported saturation magnetization is asserted without a shown derivation and, per the abstract, comes from simulation corroboration, which makes it a simulation-matched parameter rather than an independently measured constant. The reliance on an estimated 11 µm defocus and single-slice weak-phase assumptions is a correctness risk, not a circularity. Self-citations (Zhou et al. [25]; Trujillo et al. [41]) document prior demonstrations and are not load-bearing in the present derivation chain. These factors justify a partial circularity score of 6 rather than 8, because the experimental OAH benchmark remains external and falsifiable.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the Aharonov-Bohm phase model, micromagnetic simulations as ground truth, an assumed mean inner potential of 26 V, and the single-slice approximation in ePIE. The most consequential free parameter is the saturation magnetization of the nanowire, which is adjusted to match the experimental phase and then reported as a measured value. No new physical entities are postulated.

free parameters (3)
  • Saturation magnetization of NiFe nanowire (Ms) = 478.8 x 10^3 A/m
    Obtained by matching micromagnetic simulation phase to experimental phase (Section 2.2.3 and Discussion). The procedure is not described, no uncertainty is given, and the simulation initially used 840 x 10^3 A/m for Permalloy, so the reported value is an adjusted input, not an independent measurement.
  • ePIE probe defocus = 11 um (estimated)
    Used for the 4D-STEM ptychographic reconstruction (Methods 2.2.2); not directly measured, and the absolute phase values depend on it.
  • RMAD defocus subset selection = moderate and minimum defocus images
    Chosen post hoc from 21 defocus values to improve the RMAD reconstruction (Section 2.2.1). This is an algorithm hyperparameter, not a physical parameter.
assumptions (5)
  • standard math Aharonov-Bohm relation (Eq. 1) relates total phase to electrostatic scalar potential and magnetic vector potential.
    Foundational equation for all phase reconstructions in the paper; accepted.
  • domain assumption Multiplicative object / single-slice approximation for ePIE phase retrieval is valid for the NiFe nanowire.
    Invoked in Methods 2.2.2 for Ltz-4D-STEM single-slice ptychography; strongly scattering irregular samples can violate it.
  • domain assumption Mean inner potential of Permalloy V0 = 26 V (from reference 25) applies to the NiFe nanowire.
    Used in PyLorentz simulations to generate ground truth phase (Methods 2.1); variations due to oxidation or carbon would shift the electrostatic phase.
  • domain assumption Micromagnetic simulation parameters (exchange stiffness 13e-12 J/m, damping 0.02, geometry) represent the real nanowire.
    Micromagnetic simulations with MuMax are used as ground truth and for Ms extraction; no independent experimental verification of the magnetic configuration beyond phase comparison.
  • domain assumption TIE assumes linearity of the microscope transfer function and small defocus; the paper uses large defocus (±100 um) for simulations, knowingly violating it.
    Acknowledged in Section 1 and demonstrated by TIE's lower SSIM; the central comparison uses this violation to show TIE's limitation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantitative Phase Retrieval and Characterization of Magnetic Nanostructures via Lorentz (Scanning) Transmission Electron Microscopy." pith.science (2026). https://pith.science/paper/5BS77KX5

@misc{pith2026241221005,
  author       = {Pith},
  title        = {Pith review of: Quantitative Phase Retrieval and Characterization of Magnetic Nanostructures via Lorentz (Scanning) Transmission Electron Microscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5BS77KX5}},
  note         = {Machine review of arXiv:2412.21005}
}
read the original abstract

Magnetic materials phase reconstruction from Lorentz transmission electron microscopy (LTEM) measurements has traditionally been achieved using longstanding methods such as off-axis holography (OAH) and the transport-of-intensity equation (TIE). Amidst the increase in access to processing power and the development of advanced algorithms, phase retrieval of nanoscale magnetic materials with higher fidelity and resolution, potentially down to the few nanometer limit, becomes possible. Specifically, reverse-mode automatic differentiation (RMAD) and the extended electron ptychography iterative engine (ePIE) are two methods that have been utilized for high confidence phase reconstructions using LTEM through-focal series imaging and Lorentz scanning TEM (Ltz-4D-STEM), respectively. This work evaluates phase retrieval using TIE, RMAD, and ePIE in simulations consisting of an array of Permalloy (Ni80Fe20) nanoscale islands. Extending beyond simulations, we demonstrate total phase reconstructions of a NiFe nanowire using OAH and RMAD in LTEM and ePIE in Ltz-4D-STEM experiments and determine the magnetization saturation through corroborations with micromagnetic simulations. Finally, we show how the total phase shift gradient can be utilized to observe and characterize the proximity effects emanating from neighboring magnetic island interactions and an isolated NiFe nanowire.

Figures

Figures reproduced from arXiv: 2412.21005 by the authors.

Figure 1
Figure 1. Comparative performance of phase retrieval techniques on simulated data. (a) Simulated magnetization distribution in Permalloy nanoscale islands used for phase reconstructions. The total phase shift (b) and the phase gradient (c) are generated using the Mansuripur algorithm. Simulated phase reconstructions depict the total phase shift using TIE (d), RMAD (e), and ePIE (f) alongside the respective phase gradients usi… view at source ↗
Figure 2
Figure 2. Simulated total phase reconstruction of the vertical line (1) from top to bottom and the horizontal line (2) from left to right. Scalebar within inset is 200 nm. line plots, as shown in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. LTEM bright field micrographs. Through-focal-series in-focus image (a) and under-focus (b). The scalebar in (a) extends to (b). Off-axis hologram (c) with scalebar of 50 nm [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Ltz-4D-STEM single-slice ptychographic reconstruction. High-angle annular dark field (HAADF) image (a) reconstructed probe (b), and reconstructed diffractogram (c). 2.2.3 Comparison of Phase Retrieval Methods Finally, we compare the successful reconstructions of the to…
Figure 5
Figure 5. Figure 5: Comparative performance of the reconstructed total phase shift. Phase reconstructions for (a) OAH, (b) RMAD, (c) ePIE, and (d) simulation. Gradient of the total phase for (e) OAH, (f) RMAD, (g) ePIE, and (h) simulation. Line profiles are depicted from the position of t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

3 extracted references · 2 canonical work pages

  1. [2016]

    Electron Ptychography for Investigating Magnetic Textures in Micro- and Nano- Scale Magnets via Lorentz Transmission Electron Microscopy

    41K. L. M. Trujillo, R. N. S. Divan, Y. Li, Y. Jiang, A. R. C. McCray, A. K. Petford- Long, and C. Phatak, “Electron Ptychography for Investigating Magnetic Textures in Micro- and Nano- Scale Magnets via Lorentz Transmission Electron Microscopy”, en, Microscopy and Microanalysis 29, 1847–1848 (2023). 42M. R. R. Azad, A. Kobs, B. Beyersdorff, P. Staeck, G....

    work page 2023

  2. [2023]

    MuMax: a new high-performance micromagnetic simulation tool

    37A. Vansteenkiste and B. Van de Wiele, “MuMax: a new high-performance micromagnetic simulation tool”, Journal of Magnetism and Magnetic Materials 323, arXiv:1102.3069 [cond-mat, physics:physics], 2585–2591 (2011). 38Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image Quality Assessment: From Error Visibility to Structural Similarity”, en, IEEE Transa...

    work page Pith review arXiv 2011

  3. [2024]

    Simultaneous mapping of magnetic and atomic structure for direct visualization of nanoscale magnetoelastic coupling

    27S. Kang, M. Tollner, D. Wang, C. Minnert, K. Durst, A. Caron, R. E. Dunin-Borkowski, J. McCord, C. Kubel, and X. Mu, “Simultaneous mapping of magnetic and atomic structure for direct visualization of nanoscale magnetoelastic coupling”, in (Mar. 2024). 28A. M. Maiden and J. M. Rodenburg, “An im- proved ptychographical phase retrieval algorithm for diffra...

    arXiv 2009

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.