{"id":"5d97c7c7-0033-4918-95a4-ad01bce6f6c0","arxiv_id":"2502.00762","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A 40% overlap ratio between adjacent probe positions is sufficient for stable, high-quality reconstructions in defocused electron ptychography under a phase object approximation.","lead":"This paper studies how much adjacent illumination spots must overlap in defocused electron ptychography to get good image reconstructions. Using simulations, it finds that a 40% overlap ratio is enough for stable, high-quality results with a constrained PIE algorithm, though the exact threshold depends on the object.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 40% threshold is a best-case phase-object result: CPIE enforces phase-only in Algorithm 1, and the rice image itself needs 70% overlap to match 95%; the abstract's unconditional claim needs an amplitude-contrast test.","rationale":"The reader's weakest-assumption identification is correct and matches my own: the phase-object approximation is the softest load-bearing point for the central threshold claim. The paper's geometric redundancy analysis in Section III-B is genuinely algorithm- and object-independent and is a useful contribution; the simulation framework is reproducible and the comparison between CPIE and PIE is informative. However, the abstract generalizes a 'best-case' simulation result. The paper's own rice result, requiring 70% overlap to match 95% quality, already shows that 'comparable to 60%' at 40% is not the same as 'stable, high-quality.' Because this concern hinges on whether the 40% threshold transfers to amplitude-bearing specimens, the proposed computational test on complex-valued objects with amplitude modulation would settle it. The reader's CONDITIONAL verdict is appropriate: accept if the authors revise the claim to explicitly scope it to phase-object specimens and add the suggested robustness check; do not reject the paper's core geometric analysis. Thus the verdict remains UNCHANGED relative to the reader's assessment.","tokens_in":6965,"tokens_out":6339,"duration_ms":62580,"concrete_test":"Run the same CPIE overlap sweep (rho = 0, 5, ..., 95%) on simulated objects with non-unity amplitude, for example by multiplying the USAF and rice phase images by a smooth amplitude mask with 20-50% modulation, using the same probe, detector, Poisson noise levels, and 100 iterations. Also run an amplitude-unconstrained ePIE on the same datasets for comparison. Report NRMSE at rho = 40%, 60%, 70%, and 95% for both images and both algorithms at mSNR = 20 dB. If CPIE's NRMSE at 40% is more than 3 dB worse than its 95% NRMSE, or if 40% is no longer comparable to 60%, then the abstract's unconditional 40% claim does not generalize beyond the phase-object approximation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is the abstract's statement that 'a 40% or greater overlap ratio yields stable, high-quality reconstructions.' For this to hold as stated, the threshold must be robust when the phase-object approximation is relaxed and when 'high-quality' is treated as an absolute criterion. Section II explicitly assumes 'the specimen is a phase object, meaning the amplitude of the object is small and close to one,' and Algorithm 1 line 10 enforces this by projecting the updated object estimate to unit amplitude. All simulated objects in Section IV are phase-only by construction, so the 40% result is demonstrated only for that restricted class. Additionally, Section IV reports that the rice image requires 70% overlap to achieve quality comparable to the 95% case; saying that 40% is 'comparable to 60%' does not establish that 40% yields 'stable, high-quality' reconstructions for that image. The abstract omits the 'within the parameter space considered' qualifier that appears in the conclusion. The load-bearing assumption is therefore that the phase-only CPIE result transfers to real specimens with amplitude contrast; if real specimens have significant amplitude modulation, the unit-amplitude constraint imposed in Algorithm 1 is incorrect and the 40% threshold may shift or disappear.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies how the overlap ratio between adjacent illuminated areas affects data redundancy and reconstruction quality in defocused-probe electron ptychography. It proposes a closed-form approximation to the inverse of the overlap-ratio function, defines two geometry-only quantities D(ρ) and C(ρ) that are independent of the object and of the reconstruction algorithm, and introduces a phase-object-constrained variant of PIE (CPIE). The numerical study uses two simulated phase objects (USAF-1951 chart and a rice image), three noise levels (noiseless and Poisson noise at 20 and 26 dB mSNR), and overlap ratios from 0% to 95%. The abstract claims that a 40% or greater overlap ratio yields stable, high-quality reconstructions, while the conclusion adds the qualifier 'within the parameter space considered'.","tokens_in":7215,"tokens_out":5081,"duration_ms":49967,"significance":"The geometric analysis in Sec. III is self-contained and provides a useful, algorithm-independent way to reason about overlap-induced redundancy; the error bound for the approximate inverse, |ρ − R(R^{-1}(ρ))| < 0.008, is a concrete and checkable quantitative claim. The paper also makes a practical algorithmic contribution by adding a phase-object projection to PIE, and it tests this algorithm across several noise levels. If the 40% threshold were robust, it would offer practical dose-saving guidance for 4D STEM acquisition. However, the current evidence is narrower than the central claim: only two pure phase objects are tested, the rice object requires 70% overlap to match the 95% case, and no repeat runs or error bars are provided despite random probe ordering and Poisson noise. The significance of the paper is therefore real but contingent on a careful restatement of the conditions under which the threshold holds.","major_comments":[{"comment":"The abstract's central sentence, 'a 40% or greater overlap ratio yields stable, high-quality reconstructions,' is not supported by the paper's own numerical results. Section IV states that the rice phase object requires a minimum overlap of 70% to achieve phase retrieval quality comparable to that obtained with 95% overlap, while the USAF object reaches that level at 40%. Thus, for one of the two test objects, 40% overlap does not yield the same quality as the high-overlap reference. The conclusion already limits the claim to 'within the parameter space considered,' but the abstract omits this qualifier. Please revise the abstract to state the object-dependent nature of the threshold, or redefine 'high-quality' operationally (for example, as 'comparable to 60% overlap') and apply that definition consistently.","section":"Abstract and §IV (Fig. 4)"},{"comment":"The numerical validation is performed entirely under the phase-object approximation. Section II assumes 'the specimen is a phase object, meaning the amplitude of the object is small and close to one,' and Algorithm 1 line 10 enforces this assumption by replacing the updated object amplitude with unit amplitude. All simulated objects in Sec. IV are constructed as phase-only objects. This makes the 40% threshold a best-case result for pure phase objects. For real specimens with appreciable amplitude contrast, the constraint in line 10 is not valid, and the threshold may shift or disappear. Please either add simulations with amplitude-modulated and mixed amplitude/phase objects, or explicitly restrict the central claim to the phase-object case and state that extension to amplitude contrast is untested.","section":"§II, Algorithm 1 (line 10), and §III-C"},{"comment":"The reported NRMSE curves are single realizations with no error bars. Algorithm 1 processes probe locations in an order that the text describes as random, and the Poisson noise realizations are random as well, so the exact positions of the 40% and 70% crossover points are not established. The term 'stable' in the central claim therefore lacks statistical support. Please report the mean and standard deviation over multiple independent runs, especially at the threshold overlaps and noise levels, specify the random seed protocol, and state the NRMSE threshold used to classify a reconstruction as 'high-quality.'","section":"§IV (Fig. 4) and Algorithm 1"}],"minor_comments":[{"comment":"The sentence beginning 'The authors in [17–20] have established with applications in ptychography' is incomplete and should be rewritten or removed.","section":"§I (Introduction)"},{"comment":"The statement 'NRMSE(o, co) = ∞' appears to be a sign error. With the given ν minimizing ||o − ν ohat||², the residual for ohat = co is zero, so the dB value is −∞, not +∞. Please correct the notation and define the normalization convention explicitly.","section":"§III-C (Quality measure)"},{"comment":"The hypothesis that critical overlap ratios correspond to jump discontinuities in D, mC, or MC is stated but not tested. The paper would be strengthened by a direct comparison of the predicted critical ratios {18.2, 39, 45}% with the observed NRMSE transitions in Fig. 4.","section":"§III-B (Fig. 2)"},{"comment":"The phrase 'the behaviour of the NRMSE of CPIE correlates with the σC(ρ)/µC(ρ) curve' is qualitative. Since σC/µC is a candidate predictor, please report a numerical correlation measure such as Spearman's ρ over the stated overlap range, or weaken the claim.","section":"§IV (Fig. 4)"},{"comment":"There are minor typographical issues, e.g., 'complex objects structures' in the introduction should read 'complex object structures.' A careful proofread would improve readability.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a practically relevant question and the geometric part is sound, but the abstract overstates the universality of the 40% threshold relative to the simulations shown, especially given the rice-image result and the phase-only restriction. The lack of repeated trials is also notable for a claim about 'stable' reconstructions. I believe these issues are fixable within the manuscript's scope; hence major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the geometric redundancy quantities are a genuine addition, but the 40% rule in the abstract is a phase-object best case, not the universal statement it reads as.\n\nThe paper's real contribution is the recovery-agnostic analysis: D(rho) and C(rho) depend only on scan geometry, and the approximate inverse R^{-1}(rho) is a practical tool for converting desired overlap into scan step. These are cleanly derived, self-contained, and useful for planning defocused 4D STEM acquisitions. The CPIE algorithm, PIE with a unit-amplitude projection, is a small but sensible modification for phase objects, and the simulation framework (known probe, Poisson noise, NRMSE with global phase removal) is careful and reproducible. The observation that 40% overlap performs comparably to 60% for CPIE on these two test images is a concrete, potentially useful refinement of Bunk et al.'s 60% recommendation.\n\nThe soft spots are in the packaging. The conclusion properly says 'within the parameter space considered,' but the abstract drops that qualifier. The rice image itself requires 70% overlap to match the 95% quality level, so 'stable, high-quality reconstructions' for all images at 40% is not established. The phase-object assumption is built into the algorithm (line 10 of Algorithm 1) and into all simulated objects; if a real specimen has amplitude contrast, the unit-amplitude constraint is wrong and the threshold could shift or vanish. I'd also like to see error bars or repeated runs with different probe orders, since the algorithm randomizes probe selection. The authors note their detector-plane correction (line 7) is only suitable for noiseless settings, which is a bit awkward for the noisy experiments, though the results still look reasonable.\n\nThe citation pattern is fine: Bunk et al. is credited, and the handful of self-citations are to relevant prior work on low-dose ptychography. No circularity problem; the D/C quantities are derived from geometry, not fit to the reconstructions.\n\nWho's this for? Anyone planning defocused 4D STEM experiments who wants a quick way to think about redundancy, and anyone working on phase-object ptychographic algorithms. It deserves serious peer review; the referee should push for amplitude-contrast simulations, error bars, and an abstract that matches the conclusion.","headline":"Useful geometric redundancy analysis, but the 40% overlap claim is a phase-object best case, not the universal guideline the abstract implies.","tokens_in":7759,"tokens_out":2245,"would_cite":true,"duration_ms":20927,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a 40% overlap ratio is enough for stable, high-quality reconstructions in defocused electron ptychography.","keywords":["4D scanning transmission electron microscopy","defocused electron ptychography","overlap ratio","phase retrieval","constrained PIE","phase object approximation","Poisson noise","data redundancy"],"falsifier":"Simulate a 4D STEM dataset with a specimen that has known amplitude modulation (for example, amplitude varying by ±20% around unity) and run CPIE at 40% overlap. If the NRMSE gap versus 60% overlap widens relative to the pure-phase case, the 40% threshold depends critically on the phase-object assumption rather than on overlap alone.","tokens_in":6768,"feed_emoji":"🔬","tokens_out":5344,"duration_ms":54196,"temperature":0.7,"pith_summary":"The paper sets out to establish the minimum overlap ratio between adjacent illuminated probe positions needed for reliable defocused electron ptychography. It introduces two scan-geometry quantities, D(ρ) and C(ρ), that measure data redundancy independently of the specimen and the reconstruction algorithm. Using these quantities and a constrained PIE algorithm on simulated 4D STEM data with Poisson noise, the paper concludes that an overlap ratio of 40% or greater yields stable, high-quality phase reconstructions. This matters because lower overlap means larger fields of view and less beam damage for the same number of probe positions.","feed_headline":"40% overlap is enough for stable electron ptychography","feed_subtitle":"Simulations show 40% probe overlap matches 60% quality under the phase-object approximation, cutting dose.","key_machinery":"The argument rests on a geometric overlap model plus a modified iterative algorithm. The function $\\rho = R(\\gamma)$ converts the ratio of scan step to probe diameter into an overlap fraction, and a piecewise Taylor-based inverse $R^{-1}(\\rho)$ turns a desired overlap into a scan step. The quantities $D(\\rho)$ (number of neighbouring illuminations overlapping a primary illumination) and $C(\\rho)$ (number of times each pixel is illuminated) then measure data redundancy purely from scan geometry. On the algorithmic side, Constrained PIE (CPIE) is PIE with an added phase-object constraint that resets the object amplitude toward unity after each update, and it is this constraint that makes low-overlap reconstructions stable in the paper's simulations.","core_discovery":"For a phase object and a known probe, an overlap ratio of 40% between adjacent illuminated areas is sufficient for stable, high-quality electron ptychography reconstructions. The paper demonstrates this by defining geometry-only redundancy measures and by benchmarking a Constrained PIE (CPIE) algorithm on simulated 4D STEM datasets for two phase images under noiseless and Poisson-noise conditions. In both test images, 40% overlap gave reconstruction quality comparable to 60% overlap; one of the images required 70% overlap to match the quality of 95% overlap. The paper also finds that the unconstrained PIE algorithm needs substantially larger overlaps, showing that the phase-object constraint is central to the 40% result.","pith_inferences":["If the phase-object approximation is relaxed to allow amplitude contrast, the 40% threshold likely shifts; a natural test is to repeat the paper's simulations with mixed amplitude-and-phase objects.","The same D(ρ) and C(ρ) redundancy measures could be applied to non-raster scan patterns, such as Fermat spirals, to predict whether the critical overlap changes with scan trajectory.","Because CPIE enforces unit amplitude, real specimens with absorption or strong scattering may require a modified constraint; the 40% finding should be read as a best-case scenario for weakly scattering phase objects."],"forward_implications":["If the 40% threshold holds, experimentalists can use larger scan steps in defocused-probe ptychography, covering larger fields of view or reducing dose for a given field of view.","The geometry-only quantities D(ρ) and C(ρ) allow users to estimate critical overlap values from the scan grid before running an experiment, without knowing the specimen or choosing a specific reconstruction algorithm.","The phase-object constraint in CPIE is what makes 40% overlap viable; unconstrained PIE needs larger overlaps, so the constraint is a load-bearing part of the recipe.","The 40% threshold is object-dependent in detail: the rice test image needed 70% overlap to match 95%-overlap quality, suggesting users check their specific specimen contrast before fixing an overlap."],"supporting_citations":[{"why":"supplies the earlier 60% overlap suggestion and the linear ρ = 1 − γ approximation that this paper extends.","marker":"[15]"},{"why":"defines the PIE update that the paper's Constrained PIE algorithm modifies.","marker":"[2]"},{"why":"supplies the ambiguity-free NRMSE quality measure used to compare reconstructions.","marker":"[3]"},{"why":"provides experimental evidence that defocused electron ptychography quality depends on overlap ratio and acquisition time.","marker":"[12]"},{"why":"supplies the probe wave function model and Fraunhofer propagation used to generate simulated 4D STEM data.","marker":"[21]"},{"why":"supports the assumption that the probe is known from the aberration parameters of the probe-forming system.","marker":"[22]"}],"fun_headline_variants":["40% overlap: the sweet spot for electron ptychography","Ptychography needs only 40% probe overlap","40% overlap yields stable, high-quality ptychography","Why 40% overlap is enough for ptychography","Overlap ratio 40%: sufficient for stable ptychography"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on treating the specimen as a pure phase object whose transmission amplitude is close to one; if a real sample has strong amplitude contrast, the 40% overlap finding may not hold.","fun_headline_variants_meta":{"raw":{"variants":["40% overlap: the sweet spot for electron ptychography","Ptychography needs only 40% probe overlap","40% overlap yields stable, high-quality ptychography","Why 40% overlap is enough for ptychography","Overlap ratio 40%: sufficient for stable ptychography"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000186,"raw_usage":{"total_tokens":1254,"prompt_tokens":805,"completion_tokens":449,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":421,"completion_tokens_details":{"reasoning_tokens":366}},"tokens_in":421,"tokens_out":449,"duration_ms":4827,"temperature":1.0,"reasoning_tokens":366,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T17:46:46.435179+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a 4D STEM dataset with a specimen that has known amplitude modulation (for example, amplitude varying by ±20% around unity) and run CPIE at 40% overlap. If the NRMSE gap versus 60% overlap widens relative to the pure-phase case, the 40% threshold depends critically on the phase-object assumption rather than on overlap alone.","supporting_citations":[{"cited_title":"Influence of the overlap parameter on the convergence of the ptychographical iterative engine,","cited_arxiv_id":null,"evidence_quote":"supplies the earlier 60% overlap suggestion and the linear ρ = 1 − γ approximation that this paper extends."},{"cited_title":"A phase retrieval algorithm for shifting illumination,","cited_arxiv_id":null,"evidence_quote":"defines the PIE update that the paper's Constrained PIE algorithm modifies."},{"cited_title":"An improved ptychographical phase retrieval algorithm for diffractive imaging,","cited_arxiv_id":null,"evidence_quote":"supplies the ambiguity-free NRMSE quality measure used to compare reconstructions."},{"cited_title":"Atomic resolution defocused electron ptychography at low dose with a fast, direct electron detector,","cited_arxiv_id":null,"evidence_quote":"provides experimental evidence that defocused electron ptychography quality depends on overlap ratio and acquisition time."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the probe wave function model and Fraunhofer propagation used to generate simulated 4D STEM data."},{"cited_title":"Aberration correction in electron microscopy,","cited_arxiv_id":null,"evidence_quote":"supports the assumption that the probe is known from the aberration parameters of the probe-forming system."}],"review_version":1}