{"id":"433479b5-9d49-4334-952a-32021ca50e40","arxiv_id":"2608.13106","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Scan-coil settling during microsecond dwells anisotropically smears fast 4D-STEM data along the scan direction, and a phase-correlation sub-frame alignment recovers a substantial fraction of the lost signal.","lead":"Fast 4D-STEM imaging at microsecond dwell times suffers from a hidden artifact: the microscope's scan coils take tens of microseconds to settle, so each recorded diffraction pattern is smeared along the scan direction. The authors measure this effect directly and show that a software-only alignment of sub-frames recovers much of the lost signal, which matters for low-dose imaging of beam-sensitive samples.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Within-subframe probe motion is not a pure translation, so the reported recovery factor may overstate the signal actually restored; a ground-truth simulation is needed to validate the correction.","rationale":"The paper's qualitative claims—coil settling exists, is directional, and is reproducible across instruments—are well supported by direct probe images, CoM drift curves, and multi-instrument data. The load-bearing gap is quantitative: the software correction is the paper's main contribution, but its benefit is measured only relative to the raw sum, not against ground truth. The reader's weakest assumption concerns spatial uniformity of the scalar shift; we partially agree, but the more serious failure mode is temporal: even a perfectly uniform shift cannot describe a sub-frame that is itself a time integral of a moving probe. This is acknowledged in the Discussion but not quantified. A simulation with known shifts and known within-window motion would settle both issues at once. We therefore keep the reader's CONDITIONAL verdict: the paper is plausible and useful, but the claimed recovery factors should be treated as upper bounds until validated on synthetic data with ground truth.","tokens_in":14499,"tokens_out":14392,"duration_ms":147798,"concrete_test":"Simulate a known ground-truth object and generate 4D-STEM sub-frames by integrating the true signal over the measured intra-dwell probe trajectory (e.g., per-sub-frame boxcar or trapezoid kernels matching the CoM curves in Figure 2b), add Poisson noise at the same probe current and dwell time, then run the paper's phase-correlation alignment pipeline. Compare (1) the estimated per-sub-frame shifts to the known kernel centroids and (2) the aligned power spectrum to the ground-truth power spectrum as a function of kx. If the aligned spectrum remains substantially below ground truth at high kx while R(k) > 1, the recovery factor overstates the signal actually restored; if the aligned spectrum approaches the ground-truth spectrum, the correction is validated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2.4 models each early sub-frame as a scalar translation of the final sub-frame along the unrolled scan path. But each sub-frame is a 10 microsecond integration over a continuously moving probe (Section 2.1, Discussion), so an early sub-frame is a directionally blurred version of the sample, not a shifted copy. Aligning centroids removes the between-sub-frame displacement but cannot undo the convolution with the probe trajectory within each window. The recovery factor R(k) = P_aligned/P_raw (Section 2.5) can therefore exceed unity merely because centering differently shifted blur kernels reduces their mutual phase cancellation, even when the aligned image remains far from the true settled object. Residual shift maps (Supplementary Figures 3-4) only demonstrate that a global shift removes the displacement between sub-frames; they do not demonstrate that the aligned image matches a settled acquisition. The Discussion concedes that \"a residual smearing therefore persists within every sub-frame,\" but the paper never quantifies this residual or compares the corrected result to a ground truth. Without such a comparison, the reported recovery factors (up to about 3x at Au(200)) cannot be interpreted as the fraction of signal actually recovered, which is the paper's central practical claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports that at microsecond dwell times the scan deflection coils in a 4D-STEM instrument do not settle instantaneously; the probe continues to move during the detector integration window, smearing the recorded diffraction signal anisotropically along the fast-scan direction. Using direct real-space probe imaging and sub-frame diffraction analysis on a gold cross-grating, the authors document a settling timescale of several tens of microseconds across multiple scan strategies, step sizes, and two independent microscopes. They propose a phase-correlation-based alignment that registers each early sub-frame to the final settled sub-frame along the unrolled scan trajectory, and they quantify the benefit with a recovery factor R(k) = P_aligned/P_raw. The correction is reported to restore signal across a broad range of spatial frequencies, with the largest gains at large step sizes relevant for low-dose biological imaging.","tokens_in":14674,"tokens_out":6108,"duration_ms":51401,"significance":"The direct observation of intra-dwell scan-coil settling is a timely and important characterization for the fast 4D-STEM community, and the proposed software-only correction is practically appealing because it requires no hardware modification. The claim is supported by direct probe imaging (Figure 1), diffraction CoM drift measurements that are independent of scan strategy (Figure 2), residual-shift collapse after correction (Supplementary Figures 3-4), and reproduction on a second microscope (Supplementary Figures 1 and 8). The manuscript also makes data and code available. However, the quantitative recovery factors are not yet convincingly tied to a ground truth: because each sub-frame is an integration over continuous probe motion, the alignment removes only the between-sub-frame displacement, not the intra-sub-frame blur, and the recovery factor is computed on the same data used to estimate the shifts. A ground-truth simulation or slow-scan comparison is needed before the reported factors can be read as the actual fraction of signal recovered.","major_comments":[{"comment":"The scalar-translation model in Section 2.4 treats each early sub-frame as a shifted copy of the final settled sub-frame. However, Section 2.1 states that each sub-frame is a 10 μs integration window, and the probe motion during that window is continuous (Figure 1). An early sub-frame is therefore a directionally blurred version of the object, not a pure translation. Phase-correlation alignment can remove the centroid displacement between sub-frames, but it cannot undo the convolution with the intra-window probe trajectory. The Discussion correctly concedes that 'a residual smearing therefore persists within every sub-frame,' yet the paper never quantifies this residual or compares the corrected result against a settled-acquisition ground truth. Without such a comparison, the recovery factors in Figures 4-6 (up to about 3× at Au(200)) cannot be interpreted as the fraction of signal actually recovered. I recommend adding a simulation with a known ground-truth object and known probe trajectory, or a comparison to a slow-scan (long-dwell) acquisition of the same field of view.","section":"Section 2.4 (Sub-frame alignment) and Section 4 (Discussion)"},{"comment":"The recovery factor R(k) = P_aligned(k)/P_raw(k) is computed on the same data used to estimate the per-sub-frame shifts in Section 2.4. This creates an internal consistency loop: R(k) measures how well the alignment procedure removes the between-sub-frame displacement, not how well the aligned image matches the true settled object. The directional control (R≈1 along the slow-scan direction) demonstrates that the correction does not boost power uniformly, but it does not establish that the fast-scan aligned image is closer to ground truth. The manuscript would be strengthened by a calibration experiment or simulation that reports R(k) against a known ground-truth signal loss.","section":"Section 2.5 (Recovery factor analysis)"},{"comment":"For the 50 μs total dwell time datasets used in Figures 4-6, the data are divided into five 10 μs sub-frames, and the alignment uses the final sub-frame as the settled reference. The settling timescale measured in Figure 2 is approximately 40 μs, so the reference sub-frame (40-50 μs) may still contain a small residual drift. The manuscript should specify which sub-frame serves as the reference for each dataset and quantify the residual drift in that reference, since a biased reference would affect both the measured shifts and the recovery factors.","section":"Section 3 (Figures 4-6) and Section 2.2"}],"minor_comments":[{"comment":"The phrase 'a.k.a., the probe' is colloquial; consider replacing it with 'the electron probe.'","section":"Section 1"},{"comment":"The sentence 'For measurements at LMU, microscope was operated at 200 kV' is missing an article; it should read 'the microscope was operated at 200 kV.'","section":"Section 2.2"},{"comment":"The text contains the typo 'powersepctrum'; it should be 'power spectrum.'","section":"Section 3 (Figure 6)"},{"comment":"The statement that intra-dwell drift 'manifests as a single scalar translation of one 1D signal relative to another' is presented as exact, but it is an approximation given the within-sub-frame integration; a brief acknowledgment of this approximation at this point would help the reader.","section":"Section 3 (Sub-frame alignment description)"},{"comment":"In the sentence beginning 'The software correction discussed here, applies directly,' the comma after 'here' is incorrect and should be removed.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of physics.ins-det and the direct characterization of scan-coil settling is solid. The main risk is the over-interpretation of the recovery factor as the actual recovered signal fraction without ground-truth validation. I recommend requesting a simulation or a slow-scan comparison before acceptance. The authors have declared the Dectris consultancy, which is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should read this paper. The authors have documented a genuine and underappreciated effect: at microsecond dwell times, scan coils keep moving for tens of microseconds after each commanded jump, so each detector sub-frame integrates over a moving probe rather than a single position. The direct evidence is strong — real-space probe imaging, diffraction center-of-mass drift across linear/serpentine/spiral scans and multiple step sizes, and reproduction on a second microscope. That part is solid and will be useful to anyone doing fast 4D-STEM. The software fix is simple and sensible: split each dwell into 10 microsecond sub-frames, reconstruct each sub-frame separately, unroll along the scan path, and use phase correlation to register early sub-frames to the final settled one. The residual-shift collapse after correction is convincing evidence that the scalar-shift model works for these datasets, and they’ve posted data and code on Zenodo.\n\nThe soft spot is the recovery factor. Because each sub-frame is an integral over continuous probe motion, aligning centroids removes the between-sub-frame displacement but cannot undo the intra-frame convolution. The paper acknowledges this residual smearing in the Discussion, but the reported recovery factors (up to ~3x at Au(200)) are ratios of aligned to raw power spectra, so they conflate true signal recovery with the simple removal of phase cancellation between misregistered sub-frames. A single simulation with known shifts and known intra-frame blur would have distinguished these and given an honest upper bound on what the correction can achieve. The absence of such a ground truth, plus no error bars on the recovery factors, makes the quantitative headline weaker than the qualitative conclusion. There is a second minor concern: the uniform scalar-shift model is validated on their gold cross-grating fields, but larger fields or different coil drivers could show spatially varying settling; the residual maps suggest this is not a problem here, but it limits generalization.\n\nThis is a solid experimental characterization paper with a practical correction. It deserves peer review. I’d recommend the authors add a simulation-based validation and report uncertainties on the recovery factors; neither request undermines the core finding. I’d cite it if I were working on fast 4D-STEM, and I’d bring it to a reading group to discuss the motion-integration issue.","headline":"A real, well-documented scan-coil settling effect in fast 4D-STEM, with an honest but slightly overclaimed software correction; deserves peer review after adding ground-truth validation.","tokens_in":15253,"tokens_out":1774,"would_cite":true,"duration_ms":20559,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["68.37.Ma"],"model":"deepseek-v4-flash","headline":"At microsecond dwell times, scan-coil settling smears 4D-STEM signal along the fast-scan direction, and a phase-correlation sub-frame alignment recovers most of the lost signal without hardware changes.","keywords":["4D-STEM","scan-coil delay","intra-dwell smearing","sub-frame alignment","phase-correlation registration","low-dose electron microscopy","pixelated detectors","anisotropic resolution loss"],"falsifier":"A decisive test would be to record a fast 4D-STEM dataset on a field of view several times larger than the few-Ångström steps tested here, apply the scalar per-sub-frame shift, and plot the residual shift vectors against position; if the residuals grow systematically toward the edges of the field or develop a rotational component, the uniform-translation assumption fails. A complementary check is to re-bin event-driven TimePix4 data into 1 µs (or finer) slices and ask whether the earliest slices still show within-slice motion that a single shift cannot remove.","tokens_in":14290,"feed_emoji":"🔬","tokens_out":7653,"duration_ms":64534,"temperature":0.7,"pith_summary":"The paper sets out to show that a finite scan-coil response is a measurable and systematic problem in fast 4D-STEM, not a rare edge case. At dwell times around ten microseconds, the probe is still moving while the detector integrates, so each diffraction pattern mixes signal from several neighbouring beam positions; the smearing lies almost entirely along the fast-scan direction. The paper argues this can be corrected after the fact by splitting each dwell into sub-frames, registering each early sub-frame to the final settled sub-frame with a phase-correlation shift along the unrolled scan trajectory, and re-summing. A sympathetic reader should care because the correction works on existing data and existing microscopes, and the largest gains appear exactly where low-dose biological imaging needs them: large step sizes and high spatial frequencies.","feed_headline":"Scan-coil settling smears fast 4D-STEM images; realignment restores them","feed_subtitle":"Microsecond dwells leave the probe moving while the detector integrates; per-sub-frame alignment restores lost signal.","key_machinery":"The load-bearing mechanism is the time-to-space mapping of the data and its inversion: with the detector running at 100 kHz while the scan steps at 10 kHz, each commanded probe position is recorded as ten consecutive 10 µs sub-frames, and the intra-dwell drift appears as a translation of early sub-frame images relative to the settled final sub-frame. The paper unrolls each per-sub-frame reconstructed image along the actual 1D scan path, measures the translation by upsampled phase cross-correlation at 1/20-pixel precision, applies the shift with Fourier phase shifting, and re-rasterises the result. Direct probe imaging and centre-of-mass analysis carry the argument that this translation comes from the deflection coils rather than stage drift, because the shift direction follows the scan trajectory and is independent of scan strategy.","core_discovery":"On the paper's own terms, the discovery is that intra-dwell scan-coil settling is real, generic, directional, and correctable in software. Time-resolved probe images show the probe centroid converging to its commanded position only after roughly 40 µs on one microscope and about 20 µs on another, and the same settling appears in diffraction-space centre-of-mass shifts across linear, serpentine, and spiral scans. The paper's correction measures one scalar translation per sub-frame relative to the final sub-frame, applies it via Fourier phase shifting along the unrolled scan path, and yields recovery factors above unity along the scan direction at every step size tested, reaching nearly 3× at the Au(200) reflection and producing broadband recovery in defocused parallax imaging at 5.025 Å steps, while the perpendicular direction remains near unity.","pith_inferences":["The near independence of settling time from scan strategy suggests the coil response can be modeled with a simple linear system; if so, the shift curve could be predicted for arbitrary dwell times and step sizes without acquiring a calibration dataset each time.","Because the smear acts as a directional low-pass filter whose cutoff moves lower as step size grows, a deconvolution-based alternative to registration might recover even the signal that remains spread within individual sub-frames.","The residual within-sub-frame motion sets a floor on what any post-acquisition registration can restore; event-driven detectors that time-stamp individual electrons could probe this floor and tell whether hardware pre-emphasis is ultimately needed.","The scalar-shift assumption should be stress-tested at much larger fields of view than reported; if coil settling is position-dependent, a single uniform shift will fail at the edges, and a spatially varying warp would be required."],"forward_implications":["Any 4D-STEM dataset recorded with sub-frame or event-binned structure can be corrected with the same software-only procedure, without touching the microscope or the dose.","Recovery grows with both scan step size and spatial frequency, so the method returns the most signal in exactly the low-dose biological regime where large steps are mandatory.","The per-sub-frame shift curves can serve as a quantitative instrument diagnostic for scan-coil settling performance, reproducible across scan patterns and step sizes.","The alignment carries over to focused and defocused reconstruction modes, so it can be combined with ptychography by treating early sub-frames as intermediate probe positions."],"supporting_citations":[{"why":"Supplies the Dectris ARINA hybrid-pixel detector whose 100 kHz sub-frame rate makes the intra-dwell measurements possible.","marker":"[20]"},{"why":"Establishes the event-driven TimePix3/4 acquisition context and motivates the microsecond-dwell regime where coil delay matters.","marker":"[21]"},{"why":"Prior characterization of the dynamic response of the beam deflection system that this work extends to intra-dwell smearing.","marker":"[32]"},{"why":"Provides py4DSTEM, used to compute integrated centre-of-mass and defocused parallax reconstructions.","marker":"[33]"},{"why":"Provides the phase_cross_correlation function used for upsampled sub-pixel shift measurement in the alignment.","marker":"[34]"},{"why":"Demonstrates atomic-resolution low-dose ptychography of a radiation-sensitive MOF, the regime where recovery factors matter.","marker":"[14]"},{"why":"Demonstrates sub-nanometer low-dose cryo-electron ptychography of proteins, the biological context requiring large scan steps.","marker":"[15]"},{"why":"Provides the parallax imaging contrast-transfer analysis used to explain the oscillating power spectra and the broadband recovery.","marker":"[38]"}],"fun_headline_variants":["Software fix recovers signal lost to scan-coil lag in fast 4D-STEM","Anisotropic blur from scan-coil settling fixed by sub-frame alignment","Correcting scan-coil lag in 4D-STEM: software-only boost for low-dose imaging","Fast 4D-STEM's hidden foe: scan-coil settling, fixed with sub-frame shifts","Scan-coil lag steals signal in 4D-STEM; realignment gives it back"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The correction rests on the assumption that every sub-frame's intra-dwell drift is exactly one scalar translation along the scan path, the same everywhere in the field of view; if coil settling varies with position or includes distortion, parts of the image would be misregistered after alignment.","fun_headline_variants_meta":{"raw":{"variants":["Software fix recovers signal lost to scan-coil lag in fast 4D-STEM","Anisotropic blur from scan-coil settling fixed by sub-frame alignment","Correcting scan-coil lag in 4D-STEM: software-only boost for low-dose imaging","Fast 4D-STEM's hidden foe: scan-coil settling, fixed with sub-frame shifts","Scan-coil lag steals signal in 4D-STEM; realignment gives it back"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000782,"raw_usage":{"total_tokens":3422,"prompt_tokens":885,"completion_tokens":2537,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":501,"completion_tokens_details":{"reasoning_tokens":2421}},"tokens_in":501,"tokens_out":2537,"duration_ms":15137,"temperature":1.0,"reasoning_tokens":2421,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:38:51.446876+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive test would be to record a fast 4D-STEM dataset on a field of view several times larger than the few-Ångström steps tested here, apply the scalar per-sub-frame shift, and plot the residual shift vectors against position; if the residuals grow systematically toward the edges of the field or develop a rotational component, the uniform-translation assumption fails. A complementary check is to re-bin event-driven TimePix4 data into 1 µs (or finer) slices and ask whether the earliest slices still show within-slice motion that a single shift cannot remove.","supporting_citations":[{"cited_title":"Stroppa, M","cited_arxiv_id":null,"evidence_quote":"Supplies the Dectris ARINA hybrid-pixel detector whose 100 kHz sub-frame rate makes the intra-dwell measurements possible."},{"cited_title":"Jannis, C","cited_arxiv_id":null,"evidence_quote":"Establishes the event-driven TimePix3/4 acquisition context and motivates the microsecond-dwell regime where coil delay matters."},{"cited_title":"Kumar, H","cited_arxiv_id":null,"evidence_quote":"Prior characterization of the dynamic response of the beam deflection system that this work extends to intra-dwell smearing."},{"cited_title":"Savitzky, S.E","cited_arxiv_id":null,"evidence_quote":"Provides py4DSTEM, used to compute integrated centre-of-mass and defocused parallax reconstructions."},{"cited_title":"Van der Walt, J.L","cited_arxiv_id":null,"evidence_quote":"Provides the phase_cross_correlation function used for upsampled sub-pixel shift measurement in the alignment."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Demonstrates atomic-resolution low-dose ptychography of a radiation-sensitive MOF, the regime where recovery factors matter."},{"cited_title":"Varnavides, J.M","cited_arxiv_id":null,"evidence_quote":"Provides the parallax imaging contrast-transfer analysis used to explain the oscillating power spectra and the broadband recovery."}],"review_version":1}