{"id":"565905ae-0630-400f-a918-dcad27151a4e","arxiv_id":"2607.29471","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Randomized probe imaging, needing only one far-field diffraction pattern per projection, produced a 62 nm-resolution 3D phase map of gold nanocubes that matches ptychography on large features.","lead":"A single randomized-probe diffraction pattern per angle was enough to reconstruct a 3D phase image of gold nanoparticles at 62 nm half-pitch resolution. This could make nanoscale X-ray phase tomography roughly 200 times faster, enabling in-situ studies of materials dynamics.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 62 nm 3D-resolution claim is not established: both FSC comparisons share the same missing wedge and the ptychography reference, so the stated 'sub-100 nm 3D resolution' may be an artifact.","rationale":"The reader's weakest assumption focused on probe stability and the absence of an actual fast RPI scan. That is a valid limitation for the speed extrapolation, but the existing four-hour dataset already provides partial evidence that the fixed probe model worked: RPI projections matched ptychography, so drift did not catastrophically break the reconstruction. The most load-bearing part of the central claim is the sub-100 nm 3D resolution, because that is the quantitative headline and it is supported only by FSC analyses that are vulnerable to shared systematic artifacts. The missing wedge is severe (70°), and both FSC references inherit the same missing wedge. The paper even acknowledges degraded depth resolution in the supplementary. Without directional FSC curves, the 62 nm number cannot be interpreted as an isotropic 3D resolution. The proposed directional FSC test can be run on the existing data and would settle whether the claim is over-stated. The verdict should remain conditional, with the additional condition that the resolution be reported by direction and that the FSC be shown not to be inflated by common missing-wedge artifacts.","tokens_in":5947,"tokens_out":15053,"duration_ms":169446,"concrete_test":"Compute directional Fourier shell correlations for (i) RPI vs ptychography and (ii) the two RPI half-volumes, binning frequencies into shells whose normal is along the rotation-axis/depth direction and perpendicular to it. If the half-bit intersection along the depth direction exceeds 100 nm half-pitch, or if the split-RPI FSC is high only in the missing-wedge-affected directions, the isotropic 'sub-100 nm' claim fails. For robustness, mask the reconstructed volumes to the sample support and recompute; also report the Fourier coverage map to show which orientations are untested.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is the 62 nm half-pitch 3D resolution. This rests on two FSC analyses. The RPI-vs-ptychography FSC uses the ptychography volume as reference; that volume has its own systematic errors and shares the same 70° missing wedge, so common missing-wedge streaks can push the FSC upward. The split-RPI FSC compares two half-datasets reconstructed with the same RPI probe model and the same missing wedge; common artifacts are just as correlated as signal. The half-bit threshold does not remove this common-artifact bias. The paper's own observation that the 3D FSC resolution (~62 nm) is no worse than the 2D FRC resolution (~76 nm) despite the missing wedge is a strong claim and is explained more simply by FSC inflation than by tomography 'overcoming' the Nyquist limit. No directional FSC is shown; with a 70° missing wedge, depth resolution should be degraded, and the supplementary already notes limited depth resolution. Thus 'sub-100 nm 3D resolution' is not currently established. I do not think the probe-stability/speed extrapolation is the weakest point: the four-hour dataset actually provides evidence that the fixed probe model remained usable, and a fast scan would have less drift.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper demonstrates tomographic phase imaging of a cluster of cubic gold nanoparticles using randomized probe imaging (RPI). A random zone plate creates a speckled illumination; for each tomographic projection, a single far-field diffraction pattern is used to retrieve a complex-valued real-space image via the RPI algorithm with a ptychographically calibrated probe. The resulting 2D projections are fed into a standard SART/FBP tomographic workflow. The data are extracted from a four-hour ptychography-tomography scan, using one diffraction pattern per projection. The authors report a 3D half-pitch resolution of 62 nm by Fourier shell correlation, compare the RPI volume with a ptychography-tomography volume of the same sample, and estimate that a dedicated RPI scan could be performed in 73 s, a factor of 200 faster than the ptychography scan.","tokens_in":1561,"tokens_out":2059,"duration_ms":74010,"significance":"If the resolution and speed claims are robust, this is a valuable contribution to fast nanoscale X-ray phase tomography. The experimental demonstration is nontrivial: it uses a real heterogeneous sample, a random zone plate probe calibrated by ptychography on a Siemens star, and cross-validates the RPI volume against an independent ptychography-tomography reconstruction. The use of established software (Pty-Chi, CDTools, pyxalign) and the comparison of RPI with ptychography on the same dataset are strengths. The main weakness is that the central resolution claim rests on FSC analyses that may be inflated by common missing-wedge artifacts, and the speed improvement is extrapolated from a ptychography dataset rather than demonstrated in a dedicated fast RPI acquisition. The paper is the first application of RPI to tomography, but the method itself is prior work [12], so the novelty lies in the demonstration and in identifying the practical potential and limitations.","major_comments":[{"comment":"The 62 nm half-pitch 3D resolution is not convincingly established because both FSC curves share confounds. The RPI-vs-ptychography FSC uses a reference volume reconstructed from the same 110 degree angular range with the same 70 degree missing wedge; common missing-wedge streaks and shared low-frequency artifacts can inflate the correlation. The split-RPI FSC compares two reconstructions made with the same probe model, the same reconstruction schedule, and the same missing wedge; common artifacts are as reproducible as signal. The half-bit criterion does not correct for this common-mode bias. The paper's observation that the 3D FSC resolution is no worse than the 2D FRC resolution despite the missing wedge is a red flag and is more simply explained by FSC inflation than by tomography overcoming the Nyquist limit. Please report directional FSC, wedge-masked FSC, or a phantom simulation w","section":"Results, Fig. 1(f); Methods, tomography scan paragraph"},{"comment":"The factor-200 speed improvement and the 73 s / 7.3 s scan times are extrapolated from a ptychography-tomography dataset acquired at 2.5 Hz with a 750 nm step scan. No dedicated RPI fast scan was performed, and the manuscript does not demonstrate that the fixed calibrated probe remains valid under a fast angular step or fly scan, where vibration, drift, and partial coherence may differ from the slow ptychography scan. The speed claim is important to the paper's central motivation, so it should either be demonstrated in a proof-of-principle fast acquisition or clearly labeled as a projected capability. If the 3 Hz step-scan rate is based on measured overhead, state that explicitly and specify whether it was achieved in this experiment or in a separate test.","section":"Results, data acquisition speed paragraph"}],"minor_comments":[{"comment":"The abstract says sub-100 nm resolution, but the Results state that the 62 nm half-pitch resolution is slightly too large to resolve individual AuCs with side length approximately 100 nm. Please clarify that resolution refers to a half-pitch FSC metric and does not imply that the 100 nm cubes are individually resolved.","section":"Abstract and Results, first paragraph"},{"comment":"The caption says Fourier shell correlations but does not identify which curve corresponds to RPI-vs-ptychography and which to the split-RPI comparison. Please label the curves directly in the figure or in the caption.","section":"Fig. 1(f) caption"},{"comment":"The sentence The RPI scan was assembled by selecting one diffraction pattern from each ptychography projection should specify how the single pattern was chosen from the raster grid, for example center position, brightest position, or random. This matters for the single diffraction pattern per projection claim and for reproducibility.","section":"Methods, RPI data selection"},{"comment":"Minor formatting: '10x8 um2' should be written as '10 um x 8 um' or '10 x 8 um squared'.","section":"Methods, beam scan description"},{"comment":"Reference [22] contains the typo 'reconstuction' in the title; it should be 'reconstruction'.","section":"References"},{"comment":"The data are not publicly available. Given that the paper makes quantitative FSC claims, a public release of the diffraction patterns and reconstructions would strengthen reproducibility. If this is not possible, state the reason.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the first application of RPI to tomography is a reasonable contribution. The decisive issue is the resolution metrology: I am not convinced that the 62 nm half-pitch 3D resolution is free from missing-wedge FSC inflation. The authors can address this with directional FSC, wedge-masked FSC, or a missing-wedge phantom study, and by softening the abstract if needed. I would not require a full fast RPI acquisition if the speed claim is relabeled as a projection, but a proof-of-principle fast scan would considerably strengthen the paper. No concerns about citation practices or overlap with prior work beyond what is disclosed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this is the first experimental demonstration of randomized probe imaging (RPI) feeding a tomography workflow, and it mostly works. The abstract's sub-100 nm claim, though, is not backed by the evidence as it stands. The 62 nm half-pitch number comes from FSC analyses that share a 70-degree missing wedge and, in one case, the ptychography volume as reference. That combination can inflate correlation. The split-RPI FSC is not independent either: both halves use the same probe model and the same missing wedge, so common artifacts are correlated as strongly as signal. The paper itself notes that the 3D FSC resolution is no worse than the 2D FRC despite the missing wedge. That is a red flag; it is more likely an artifact of the shared missing wedge than a real tomographic gain. No directional FSC is shown, and the supplementary admits limited depth resolution. So the resolution claim should be treated as provisional.\n\nWhere the paper earns credit: the workflow is simple and the comparison to ptychography on the same dataset is a reasonable sanity check. The larger features in the RPI volume match the ptychography volume well. The speed extrapolation is honest — they took one diffraction pattern per projection from a four-hour dataset and estimated what a real RPI scan would take. That estimate is plausible. The paper also does not overclaim the method's readiness; it flags the missing wedge and the need for further development. The reader's worry about probe drift is less compelling: a single calibrated probe worked across the four-hour dataset, which is evidence in favor of stability rather than against it.\n\nThe soft spots beyond the FSC issue: the data are not publicly available, so no one can re-run the analysis. Also, the only external comparison is ptychography on the same sample; there is no independent resolution target. That said, the central feasibility claim — RPI projections can be reconstructed and assembled into a 3D volume — holds up. The resolution number is the weak point, not the method itself.\n\nI'd send this to peer review. The authors should be asked to provide directional FSC, a masked or independent resolution check, and to soften the sub-100 nm claim until that is done. The result is useful for people working on fast X-ray nano-tomography, even if the resolution is more like 100-150 nm than 62 nm. Would I cite it? Yes, as the first RPI-tomography demonstration, with a caveat on resolution.","headline":"First RPI-tomography demonstration works as a feasibility proof, but the 62 nm half-pitch resolution claim is likely inflated by shared missing-wedge artifacts in the FSC analyses.","tokens_in":6783,"tokens_out":4328,"would_cite":true,"duration_ms":41916,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["42.30.Rx","07.85.Qe","42.30.Wb"],"model":"deepseek-v4-flash","headline":"Tomographic phase imaging of gold nanoparticles from one diffraction pattern per projection, at 62 nm resolution.","keywords":["randomized probe imaging","tomographic phase imaging","single-shot coherent diffraction","randomized zone plate","nanoscale X-ray tomography","ptychography comparison","gold nanoparticles","time-resolved imaging"],"falsifier":"Run an actual fast RPI-tomography scan on the same type of gold nanoparticle sample, with the probe calibrated once before the scan and the full 73-second or 7-second acquisition, then compare the resulting volume to a ptychography-tomography volume of the same sample: if the 3D resolution degrades by more than the expected noise level, or if individual cubes appear in one volume but not the other in a way inconsistent with the reported 62 nm half-pitch, the fixed-probe/stability assumption is violated. A second check: simulate RPI projections from a known 3D phantom with random probe drift be","tokens_in":5918,"feed_emoji":"🧪","tokens_out":1607,"duration_ms":22187,"temperature":0.7,"pith_summary":"The paper demonstrates that a technique called randomized probe imaging (RPI) can supply the two-dimensional phase projections needed for X-ray nanotomography, using just one far-field coherent diffraction pattern per viewing angle instead of the many overlapping scans required by ptychography. The authors reconstruct a three-dimensional volume of a cluster of cubic gold nanoparticles from 220 single-shot diffraction patterns, achieving a half-pitch resolution of about 62 nm. They compare this volume with a higher-resolution ptychography-based reconstruction of the same sample and find the larger structural features match well. The core motivation is speed: with one pattern per projection, a tomographic scan that took roughly four hours with ptychography could, in principle, be collected in about 73 seconds with a step scan, or about 7 seconds with a fly scan. If this holds, RPI-tomography becomes a practical path to time-resolved nanoscale phase imaging for in-situ studies of crystal formation, fuel-cell operation, and similar dynamic processes.","feed_headline":"One diffraction pattern per angle gives 62-nm X-ray phase tomography","feed_subtitle":"Randomized-probe imaging cuts a four-hour ptychographic scan to an estimated 73 seconds without losing sub-100-nm structure.","key_machinery":"Randomized probe imaging (RPI), with a randomized zone plate (RZP) as the central optical element. The RZP focuses X-rays into a random speckle pattern with features of roughly 50 nm, so that each diffraction pattern is a scrambled encoding of the sample over a wide angular spread of illumination directions. The reconstruction algorithm uses a one-time ptychographic calibration of the complex probe (measured on a Siemens star) and then, for each single diffraction pattern, applies an iterative phase-retrieval loop with a bandwidth-limiting constraint that enlarges the effective reconstructed pixel size and stabilizes the inversion. The bandwidth-limiting constraint is what turns a single far","core_discovery":"The central claim is that tomographic phase imaging at sub-100 nm resolution does not require a scanned probe: a single far-field coherent diffraction pattern per projection, illuminated by a randomized zone plate, contains enough information to reconstruct the complex-valued projection if the probe is already known from a one-time calibration. The paper validates this by selecting one diffraction pattern from each angle of a four-hour ptychography-tomography dataset, reconstructing each projection with the RPI algorithm, feeding the phase images into a standard SART/FBP tomographic pipeline, and obtaining a 3D volume whose half-pitch resolution is 62 nm by the half-bit Fourier shell correla","pith_inferences":["The paper's two-step workflow (per-projection RPI, then tomographic reconstruction) discards information: the same diffraction patterns could in principle be fed into a joint 3D phase-retrieval algorithm that shares the known probe across all projections, likely improving resolution and robustness beyond the demonstrated 62 nm; the authors themselves hint at this possibility.","Because the probe is calibrated once and held fixed, the method's true failure point in a fast scan is mechanical and thermal stability of the sample and RZP over the full 73-second or 7-second acquisition; the paper's alignment data (1-2 μm total drift over four hours) is encouraging, but the single-shot reconstruction sensitivity to sub-pixel drift was not directly tested with genuinely fast-acq","The demonstrated resolution is in the 50-100 nm band, which is the same regime as the recently proposed coded-aperture single-distance nanotomography method; a direct experimental comparison of speed, dose, field of view, and reconstruction robustness between the two approaches would clarify which method scales better for in-situ studies.","The observation that tomographic resolution matches single-projection resolution despite a 70-degree missing wedge suggests that the missing-wedge penalty may be partially offset by the angular diversity of the single-shot reconstructions; this is a claim worth testing with simulated data before relying on RPI-tomography for anisotropic samples."],"forward_implications":["A tomographic phase scan can be collected in roughly 73 seconds with an angular step scan at 3 Hz, and potentially in 7.3 seconds with a fly scan at the detector's 30 Hz limit, opening a path toward time-resolved nanoscale phase tomography of dynamic samples.","RPI-tomography works with an ordinary far-field setup, requiring no coded-aperture motion or synchronized sample-aperture mechanics, so it can be retrofitted to existing coherent X-ray beamlines with a randomized zone plate.","Because each projection is reconstructed independently, standard tomographic alignment and reconstruction software apply unchanged, and any future algorithmic improvement to single-frame RPI phase retrieval directly improves the tomographic volume.","The consistent resolution between 2D projections and the 3D volume (62 nm half-pitch by FSC, 76 nm by FRC) suggests that tomographic integration itself does not degrade RPI reconstructions, and may in fact compensate for single-frame noise.","The method's resolution is bounded by the RZP outer zone width (here 50 nm) and by the bandwidth-limiting constraint, so fabricating RZPs with finer outer zones would directly push the achievable tomographic resolution toward the 30 nm Crowther limit of the current geometry."],"fun_headline_variants":["Single-shot X-ray phase tomography at 62 nm","Randomized probe: 3D X-ray phase from a single shot","One pattern per angle: sub-100-nm X-ray phase tomography","4-hour scan to 73 seconds: single-shot X-ray phase tomography"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire pipeline depends on the assumption that the RZP probe, calibrated once before the tomography scan, remains a perfectly fixed and faithful illumination model for every projection, and that each single-frame RPI reconstruction is a clean coherent far-field intensity of a static sample — in this data, that assumption was never tested under genuinely fast scanning conditions, because the 'RPI' frames were actually extracted from a slow four-hour ptychography dataset.","fun_headline_variants_meta":{"raw":{"variants":["Single-shot X-ray phase tomography at 62 nm","Randomized probe: 3D X-ray phase from a single shot","One pattern per angle: sub-100-nm X-ray phase tomography","4-hour scan to 73 seconds: single-shot X-ray phase tomography"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001272,"raw_usage":{"total_tokens":4997,"prompt_tokens":660,"completion_tokens":4337,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":404,"completion_tokens_details":{"reasoning_tokens":4274}},"tokens_in":404,"tokens_out":4337,"duration_ms":29806,"temperature":1.0,"reasoning_tokens":4274,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T06:26:05.008810+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an actual fast RPI-tomography scan on the same type of gold nanoparticle sample, with the probe calibrated once before the scan and the full 73-second or 7-second acquisition, then compare the resulting volume to a ptychography-tomography volume of the same sample: if the 3D resolution degrades by more than the expected noise level, or if individual cubes appear in one volume but not the other in a way inconsistent with the reported 62 nm half-pitch, the fixed-probe/stability assumption is violated. A second check: simulate RPI projections from a known 3D phantom with random probe drift be","supporting_citations":[],"review_version":1}