{"id":"a2d53be4-8fb7-42a1-9c3c-54ce26378def","arxiv_id":"2604.17561","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"For delta-DRX spinel materials, STEM-HAADF images along [110] classify domain boundaries into four Fourier profiles, and four of 28 boundary types are undetectable.","lead":"This paper simulates electron microscope images of a battery material that forms eight differently ordered 'spinel' domains, and shows the image patterns can look identical across some domain boundaries. It warns that some boundaries are completely invisible in standard atomic-resolution imaging, so scientists may misread the material's true structure.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'undetectable boundary' claim rests on a projected-potential-plus-Gaussian HAADF proxy; multislice channeling could break the A1–A2 depth degeneracy, so the four-profile classification needs realistic-image validation.","rationale":"The paper's geometric enumeration—eight variants, four projected motif pairs, and the Fourier-discontinuity classification—is coherent and likely correct as a statement about projected occupancies at a given imaging approximation. That part is parameter-free and does not depend on scattering physics. However, the strongest claim is phrased as a statement about what 'atomically resolved electron microscopy' can and cannot detect. The reader's weakest assumption correctly identifies the HAADF proxy in Methods 4.5 as the least secure link. My stress test sharpens this: the undetectable class is not merely a borderline case; it is the case where the two variants are identical in projection and only differ along the beam direction. A projection-only model cannot possibly reveal them, so the conclusion is an artifact of the model unless a depth-sensitive simulation or experiment shows otherwise. This is a legitimate correctness risk, not a consensus disagreement. The proposed multislice check directly tests whether the classification survives under realistic imaging physics. I do not find an internal inconsistency in the geometric argument, and the authors are appropriately cautious about interface widths and about generalization beyond {100}. Thus the appropriate verdict remains CONDITIONAL, unchanged from the reader.","tokens_in":17200,"tokens_out":3397,"duration_ms":41023,"concrete_test":"Run multislice STEM-HAADF simulations (e.g., abTEM or Prismatic) for all 28 variant-pair interfaces at 300 kV, 30 mrad convergence, a realistic HAADF detector, and including oxygen (and optionally Li), at thicknesses of 3.4, 10, 50, and 100 nm. Repeat the same Fourier-filtering procedure and compare the Table 1 discontinuity pattern, paying special attention to the four nominally undetectable pairs (A1–A2, B1–B2, C1–C2, D1–D2). If any of these four shows a fringe discontinuity at any thickness, the 'undetectable' claim must be restricted to the thin-slab projection approximation; if all remain continuous and the 24 detectable pairs remain discontinuous across thicknesses, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that certain variant interfaces are 'undetectable in atomically resolved electron microscopy' along [110]—depends on the HAADF model in Methods 4.5: the projected Mn potential is squared, convolved with a Gaussian (sigma=3 pixels), and interpreted as Z-contrast. This is a thin-slice, incoherent, projection-only proxy. It omits dynamical diffraction, channeling, finite detector geometry, and the ~100 nm lamella thickness of the experimental samples; the simulated slab is only c=34.24545 Å along [110]. The A1–A2 (and B1–B2, C1–C2, D1–D2) pairs differ only by the depth ordering of pure 16c/16d columns along the beam. In a pure projection these are identical by construction, so 'undetectable' is baked into the model rather than demonstrated. If channeling or thickness-dependent column intensities break that depth degeneracy, the boundary could become visible and the paper's central conclusion—that single-axis STEM-HAADF systematically undercounts domain boundaries—would be weakened. The paper itself acknowledges that the projected interface width is thickness-dependent (Section 2.2) but does not test whether the detectability classification itself is thickness-dependent. The Discussion's extension of Table 1 to non-{100} interfaces is likewise asserted from symmetry rather than derived or simulated, but the more load-bearing issue is the imaging proxy itself: Table 1 is presented as a statement about real STEM-HAADF, not merely about blurred projected potentials.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes single-zone-axis STEM-HAADF imaging of δ-DRX (disordered rocksalt derived spinel) materials, focusing on the eight crystallographic variants of spinel ordering that form on the rocksalt lattice. The authors construct {100} antiphase boundary models between every pair of the eight variants, simulate HAADF-like images using a projected Mn potential squared and Gaussian blurred, and then apply Fourier filtering to isolate two spinel superlattice frequencies indexed near the (111) and parent-lattice (222) positions. They find that all 28 variant pairs fall into four discontinuity profiles: continuous in both frequencies, discontinuous in both, or discontinuous in only one. Four pairs (Ai–Ai, Bi–Bi, Ci–Ci, Di–Di) are reported as producing no discontinuity, making the boundary 'undetectable' in STEM-HAADF along [110]. The paper argues that this explains apparent remnant disorder and layered-like contrast in experimental images as projection overlap from slanted domain boundaries, and warns that single-axis HAADF systematically undercounts domain boundaries in δ-DRX.","tokens_in":17510,"tokens_out":4317,"duration_ms":53073,"significance":"The crystallographic enumeration and the four-profile taxonomy are internally consistent and provide a useful conceptual framework for interpreting spinel-domain contrast in the [110] projection. The complete simulation of all 28 variant pairs and the explicit connection to Fourier filtering are valuable strengths. If the 'undetectable boundary' claim holds for real STEM-HAADF imaging, the paper identifies an important and non-obvious limitation that could affect many experimental studies of Mn-rich cathodes and analogous topotactic systems. However, the central quantitative claim about detectability in actual electron microscopy rests on a projection-only imaging proxy; the paper's own method section describes it as 'projected electrostatic potential calculations with Gaussian blurring.' This gap needs to be addressed before the abstract-level claim can be accepted as a statement about real STEM-HAADF rather than about a simplified model.","major_comments":[{"comment":"The 'undetectable' classification for A1–A2, B1–B2, C1–C2, and D1–D2 is obtained from an imaging model that squares the projected Mn potential and convolves it with a Gaussian (σ=3 pixels). For these pairs, the two variants differ only by the depth ordering of pure 16c/16d columns along the beam direction; in any strictly projected model they are mathematically identical. Thus the undetectability is built into the model rather than demonstrated for real STEM-HAADF, where dynamical diffraction, channeling, finite detector geometry, and sample thickness can break the depth degeneracy. The simulated slab is only c=34.24545 Å along [110], whereas the experimental lamella is ~100 nm. Please either (a) restrict the central claim to 'projected-potential imaging' and adjust the Abstract and Conclusions accordingly, or (b) add multislice simulations for representative boundaries—at least one 'und","section":""},{"comment":"The statement that 'the results summarized in Table 1 are, in principle, applicable to interfaces beyond the {100} family, since the Fourier discontinuity arises from the symmetry relationship between variants rather than from the specific plane along which the boundary forms' is asserted without derivation or simulation. The Fourier discontinuity depends on the relative projected offset of the two variant lattices, and for a general boundary plane (or a boundary with local relaxations) this offset need not be the same as for the specifically constructed slanted {100} boundaries. Only {100} boundaries were simulated; the abstract's unqualified 'each domain interface' is therefore broader than the evidence. Please either add a second boundary orientation (e.g., {111} or {110}) to substantiate the generality, or explicitly label this as a hypothesis and qualify the abstract.","section":""},{"comment":"The binary 'continuous/discontinuous' classification in Table 1 is based on visual inspection of filtered fringe images ('interfaces were classified as continuous or discontinuous depending on whether the filtered fringe phase and intensity remained coherent across the boundary'). No quantitative criterion (phase-shift threshold, fringe-amplitude jump, automated metric) or uncertainty estimate is provided, so the reproducibility of the four-profile taxonomy from the simulations is not established. Given that the entire paper rests on this classification, a quantitative definition would strengthen confidence; at minimum, the authors should state whether the classification was performed blinded and report the robustness of the category assignments to the Gaussian blur width and mask parameters.","section":""}],"minor_comments":[{"comment":"Typo: 'was subsequentely tiled' should be 'was subsequently tiled'.","section":""},{"comment":"The statement that a {100} boundary viewed along [110] intersects the slab at 45° is clear, but the projected-width dependence on thickness is mentioned only qualitatively. A quantitative statement about the simulated thickness effect would help the reader assess the relevance to the ~100 nm experimental lamella.","section":""},{"comment":"The suppression of the DRX-like frequency within regions of clear spinel ordering is a processing choice that could affect the apparent localization of the DRX-like fringes at the interface. The description says this is done 'to enable visualization' but does not specify the suppression amplitude or criterion; please give the exact procedure or a reference to a previous implementation.","section":""},{"comment":"Ref. 33 (the authors' own prior work) is the source of the eight-variant enumeration and the low-energy {100} boundary selection, and this is correctly acknowledged in the text. The dependence on that prior work is substantive; consider citing the relevant sections or equations of Ref. 33 more specifically in §2.1 and §2.2.","section":""}],"recommendation":"major_revision","confidential_remarks":"The paper is clearly written and the crystallographic enumeration is valuable, but the central claim about 'undetectable' boundaries in real STEM-HAADF is not yet supported: the imaging model is a projected-potential proxy, which makes the depth-ordering degeneracy an input rather than a finding. The authors have two reasonable paths: add multislice validation (the cleanest fix) or soften the abstract and conclusions to 'projected-potential imaging.' The generalization to non-{100} interfaces also needs either a simulation or an explicit qualifier. A major revision with these changes is appropriate. The self-citation to Ref. 33 is understandable because that paper provides the variant enumeration, but the present paper should be careful not to present the prior work's assumptions as established experimental fact."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is the first systematic map of what [110] STEM-HAADF plus Fourier filtering can and cannot see for all 28 spinel-variant interfaces in δ-DRX, and the invisible-boundary class is a useful warning that should change how people read filtered HAADF images of these cathodes. The classification itself is clean and parameter-free: given the eight variants from ref 33, the four Fourier profiles follow from octahedral-column bookkeeping, and Table 1 is coherent. That is real credit, not repackaging—ref 33 gave the variant framework and {100} energies, not the [110] projection signatures. The self-citation is not a problem.\n\nThe soft spots are in the imaging model, not the crystallography. The HAADF simulation is a projected Mn potential, squared, then Gaussian-blurred (Methods 4.5). That is a thin-slice projection proxy: no dynamical diffraction, no channeling, no detector geometry. For the four undetectable pairs (A1/A2 etc.), the two variants differ only by depth ordering of pure 16c/16d columns along the beam, so 'invisible' is baked into the projection by construction. The abstract's 'leaves the boundary undetectable in atomically resolved electron microscopy' overstates what a projected-potential calculation can establish, especially since the experimental lamella is ~100 nm while the simulated slab is ~3.4 nm. The authors note the projected interface width is thickness-dependent but do not test whether detectability is thickness-dependent. A multislice sensitivity check on a few representative pairs, or at least explicit hedging in the abstract, would close the gap.\n\nTwo smaller points. Table 1 is about {100} interfaces only; the extension to other planes is asserted from symmetry, and the Discussion itself adds a caveat for boundaries perpendicular to the beam. The abstract's 'each domain interface' is broader than the evidence. And the experimental check is qualitative: one image, four marked regions; it supports the scheme but does not validate it. No code or data files are shipped, which is annoying but not disqualifying.\n\nNet: the central geometric classification is sound and likely correct for ideal {100} boundaries; the load-bearing imaging claim needs strengthening before it should appear as a categorical statement. This paper deserves a serious referee. Send it out, and ask for multislice validation and a more careful abstract.","headline":"Worth reading and worth refereeing: the enumeration of invisible spinel-variant boundaries is a real contribution, but the undetectability claim outruns the projected-potential imaging model and needs a multislice check before it can stand as a categorical statement.","tokens_in":18073,"tokens_out":3535,"would_cite":true,"duration_ms":35973,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"STEM-HAADF imaging along [110] cannot detect a class of antiphase boundaries between spinel domains in δ-DRX cathodes.","keywords":["STEM-HAADF","antiphase boundaries","spinel variants","disordered rocksalt cathode","delta-DRX","Fourier filtering","projection overlap","domain detection"],"falsifier":"Perform multislice STEM-HAADF simulations of an A1–A2 boundary at experimental thicknesses (e.g., 50–100 nm) with full dynamical scattering and a realistic detector; if the (-11-1)- or (-111)-type filtered fringes show a detectable shift across the interface, the claim that this class is undetectable along [110] is falsified. Alternatively, image a sample containing a known A1/A2 boundary along [110] and see whether the boundary appears.","tokens_in":17012,"feed_emoji":"🔬","tokens_out":7101,"duration_ms":70752,"temperature":0.7,"pith_summary":"This paper asks what atomically resolved STEM-HAADF imaging can actually reveal about the nanometer-scale spinel domains that form when a disordered rocksalt cathode material (δ-DRX) transforms to a spinel-like ordered structure. Simulating all 28 pairings of the eight crystallographically distinct spinel variants, it shows that every interface falls into one of four Fourier-filtered fringe profiles: continuous in both spinel superlattice frequencies, discontinuous in both, or discontinuous in only one. One profile—interfaces between variants related by a twofold rotation, such as A1–A2—leaves the boundary completely undetectable along the standard [110] zone axis. The paper further argues that 'disordered' or 'layered' regions seen in real micrographs can be projections of ordered spinel domains overlapping across slanted {100} boundaries, not true remnant disorder. If correct, single-axis imaging systematically misses a class of domain boundaries and overestimates disorder, which matters for understanding the electrochemical performance of Mn-rich cathodes.","feed_headline":"One in seven spinel domain-boundary types is invisible to [110] STEM","feed_subtitle":"Projection overlap hides some antiphase boundaries and makes ordered domains look disordered in δ-DRX cathodes.","key_machinery":"The key machinery is the crystallographic relationship among the eight spinel variants and their projection onto the [110] zone axis. In spinel, the octahedral sublattice splits into occupied 16d and vacant 16c sites; along [110] these form pure 16d columns, pure 16c columns, and mixed columns, and the eight variants differ by rotations and translations that rearrange these columns. Fourier filtering of simulated STEM-HAADF images isolates two spinel superlattice frequencies (indexed as (-11-1)- and (-111)-type), and the continuity or discontinuity of these fringes across a slanted {100} interface defines the four boundary classes. The slanted {100} interface, intersecting [110] at 45°, is t","core_discovery":"The central claim is that, viewed along [110], the eight spinel variants collapse into four projected motifs, and the 28 possible antiphase boundaries between them produce exactly four detectable Fourier-filtered profiles. Boundaries between variants in the same rotational pair (A1/A2, B1/B2, C1/C2, D1/D2) preserve both (-11-1)- and (-111)-type fringes and are invisible in atomically resolved STEM-HAADF; all other pairings are detectable, with eight disrupting both fringes, eight only the (-11-1)-type, and eight only the (-111)-type. The paper also claims that apparent excess disorder and layered-like contrast in prior STEM-HAADF studies of δ-DRX can arise from projection overlap across low-","pith_inferences":["A direct test of the undetectability claim would be multislice simulations of the same interfaces at realistic thicknesses: the paper's Gaussian-squared-potential proxy may miss dynamical contrast that either reveals or further hides a boundary.","If variant fractions in δ-DRX are roughly equal, roughly one seventh of all antiphase boundaries are invisible; correcting published domain-size distributions for this factor could shift quantitative conclusions about ionic transport and phase-transformation suppression.","The same projection-overlap logic should apply to other topotactic transformations that preserve a parent lattice while reducing symmetry, such as perovskite-to-infinite-layer reductions; boundaries there may also be hidden when the beam direction aligns with the preserved sublattice.","Combining multi-axis imaging or ptychography with the paper's Table 1 classification could identify which variant pairs actually dominate δ-DRX, testing whether all eight variants form with roughly equal probability."],"forward_implications":["Single-zone-axis STEM-HAADF will systematically undercount domain boundaries: roughly one seventh of variant-pair types (A1/A2, B1/B2, C1/C2, D1/D2) are invisible along [110], so domain-size and morphology statistics derived from such images are biased.","Regions that look disordered or layered in atomic-resolution images of δ-DRX can be fully ordered spinel domains seen through overlapping slanted boundaries; apparent remnant disorder should not be interpreted as real DRX without complementary diffraction evidence.","Even for visible boundaries, multiple variant pairings share the same Fourier-filtered profile, so exact variant identity cannot be assigned from [110] images alone; boundaries, not domains, are the only reliable unit of characterization.","Diffraction-based methods such as SEND and XRD remain necessary to quantify remnant disorder, while STEM-HAADF is best used for domain size and morphology once its blind spots are known."],"fun_headline_variants":["STEM sees only 4 of 28 spinel boundary motifs along [110]","Projection overlap hides antiphase boundaries, faking disorder in δ-DRX","Invisible boundaries: why ordered spinel domains look layered in STEM","Spinel domain boundaries: four projection types, one escapes STEM's view"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The undetectability classification rests on the Methods 4.5 simulation of HAADF contrast as the squared projected Mn potential convolved with a Gaussian, which neglects dynamical diffraction, channeling, detector geometry, and thickness; if real scattering changes fringe continuity, the set of 'invisible' boundaries could change.","fun_headline_variants_meta":{"raw":{"variants":["STEM sees only 4 of 28 spinel boundary motifs along [110]","Projection overlap hides antiphase boundaries, faking disorder in δ-DRX","Invisible boundaries: why ordered spinel domains look layered in STEM","Spinel domain boundaries: four projection types, one escapes STEM's view"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000585,"raw_usage":{"total_tokens":2612,"prompt_tokens":791,"completion_tokens":1821,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":535,"completion_tokens_details":{"reasoning_tokens":1749}},"tokens_in":535,"tokens_out":1821,"duration_ms":14701,"temperature":1.0,"reasoning_tokens":1749,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T05:24:13.746374+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform multislice STEM-HAADF simulations of an A1–A2 boundary at experimental thicknesses (e.g., 50–100 nm) with full dynamical scattering and a realistic detector; if the (-11-1)- or (-111)-type filtered fringes show a detectable shift across the interface, the claim that this class is undetectable along [110] is falsified. Alternatively, image a sample containing a known A1/A2 boundary along [110] and see whether the boundary appears.","supporting_citations":[],"review_version":2}