{"id":"aa5d9820-b186-42e6-a24b-e531fb0a19bb","arxiv_id":"2412.15916","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"The self-induced spin glass state of elemental neodymium shows dynamic heterogeneity: coexisting fast and slow magnetic regions, plus a field-driven evolution of magnetic periodicities that thermal cycling reinitializes.","lead":"Using spin-polarized scanning tunneling microscopy, the authors directly imaged the frozen magnetic patterns in the spin glass state of elemental neodymium and found that after magnetic field cycles, some regions change while others stay stable, a coexistence of slow and fast dynamics called dynamic heterogeneity. The result suggests local length scales, not only mean-field randomness, shape how spin glasses age, linking them to structural glasses.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The by-eye slow/fast classification (S5) and the L=12 nm grid choice (S7) are not validated against a quantitative local-dynamics metric or a null model, so the real-space clustering at the core of the DH claim could be an analysis artifact.","rationale":"The reader's conditional verdict is well calibrated. I agree that the weakest assumption is the manual slow/fast classification and the box-size choice; indeed, Supplementary Text S5 explicitly says boxes are 'examined by eye,' and S7 states that L=12 nm is optimal because it maximizes the fraction of classifiable boxes. This is the load-bearing point because the paper's central claim is the real-space coexistence of slow and fast dynamics, not the S(Q) evolution, which is quantitative but does not by itself demonstrate dynamic heterogeneity. The simulations provide independent support for spatially heterogeneous dynamics, and the multi-sample S(Q) evolution and the 27-hour stability check are genuine strengths; however, the simulations are analyzed with the same kind of threshold-based subtraction and also lack a statistical null model. My proposed check would settle whether the blue/green maps are a real property of the magnetization or an artifact of sorting: an automated metric plus a permutation/noise null would either confirm the spatial clustering or show it is not significant. Since the reader already recommended a conditional verdict and this concern is testable rather than fatal, the verdict should remain unchanged: the claim is promising but not yet quantitatively established.","tokens_in":16616,"tokens_out":4612,"duration_ms":43761,"concrete_test":"Reanalyze the raw image pairs used in Fig. 4A/B, 4D/E, and S6 (and, where possible, the ASD trajectories) with an automated quantitative local-overlap metric: divide each image pair into 12 nm boxes, compute a local similarity (e.g., normalized cross-correlation of M(r) or the JS divergence of the local S(Q)) between t0 and t1 for every box, mask defect-containing boxes if needed, and threshold using an objective criterion such as Otsu's method instead of by-eye sorting. Then compute a spatial clustering statistic of the resulting slow/fast map (e.g., fraction of same-type nearest neighbors or a finite-size four-point-type correlation) and compare it to null distributions generated by (i) randomly permuting the box labels and (ii) applying the same metric to the repeated zero-field images in Fig. S1, which isolate noise-only changes.","verdict_should_be":"UNCHANGED","load_bearing_attack":"For the central claim of dynamic heterogeneity to hold, the coexistence of slow and fast regions must reflect a real, spatially correlated dynamical propensity, not the sorting procedure. The experimental evidence for that coexistence is Fig. 4C/F and Fig. S6G-I, which are produced by manual classification (Supplementary Text S5: 'Each box was examined by eye') on a grid whose box size is selected in S7 because it maximizes the fraction of boxes that can be sorted into slow or fast categories. No quantitative mobility metric is given, no measure of spatial correlation of the labels (e.g., a four-point susceptibility or a clustering statistic) is computed, and no null-hypothesis test is reported. Because the protocol compares only two frozen snapshots separated by a field cycle, any local pattern change - including noise, drift, or field-response differences - is binned as 'fast,' and a purely random arrangement of changed boxes would also produce apparent spatial clusters at the chosen box size. The simulations (Fig. 5, S10-S11) use a threshold-based subtraction method and likewise lack a comparison against a random-switching null. The paper's remark that a four-point correlation function is difficult to apply does not remove the need for a quantitative test of spatial heterogeneity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports spin-polarized scanning tunneling microscopy (SP-STM) measurements and atomistic spin dynamics (ASD) simulations of the self-induced spin glass state of elemental Nd(0001). The central claim is the observation of dynamic heterogeneity (DH): a spatial coexistence of slow and fast evolving magnetic patterns in the frozen spin glass state, reminiscent of DH in structural glasses. The paper also reports that zero-field cooling imprints a preferred set of metastable magnetic periodicities (Q-pockets) that evolve systematically under repeated magnetic-field cycling and can be thermally reinitialized; this evolution is quantified with the Jensen-Shannon divergence between successive magnetic structure factors. The DH claim is supported by qualitative slow/fast classification of real-space boxes by eye (Supplementary Text S5), by a box size chosen to maximize classifiable boxes (S7), and by threshold-based subtraction in the simulations (S10).","tokens_in":16809,"tokens_out":6397,"duration_ms":59289,"significance":"If the DH claim is established quantitatively, this would be a significant real-space observation of dynamic heterogeneity in a spin glass, bridging glassy dynamics in magnetic systems and structural glasses, and providing evidence beyond the mean-field description. The S(Q)-evolution and reinitialization part is more quantitative and rests on multiple samples, 27-hour stability checks, and a well-defined similarity metric; this portion is convincing and provides a solid basis for the metastable-periodicity claim. The simulations reproduce spatially heterogeneous dynamics and glassy autocorrelation functions, and the paper includes a data-availability statement and reproducible methodology via UppASD. The principal weakness is that the experimental DH classification is by eye and lacks a quantitative mobility metric or null-model test, so the core claim is not yet rigorously supported.","major_comments":[{"comment":"The central claim of dynamic heterogeneity rests on classifying 12-nm boxes as slow or fast by eye, as explicitly stated in S5 ('Each box was examined by eye'). S7 states that L = 12 nm is selected because it maximizes the fraction of boxes that can be sorted into slow/fast categories. No quantitative local mobility metric (e.g., an overlap function or a local autocorrelation), no null-hypothesis test (e.g., random permutation of box labels or a comparison to a Poisson process), and no repeatability/inter-rater check is provided. The spatially clustered appearance in Fig. 4C,F and Fig. S6G-I could therefore be produced by the sorting procedure itself rather than by a physical property of the spin glass. The remark that a four-point correlation function is difficult to apply does not remove the need for a quantitative demonstration that the slow/fast labels are spatially correlated beyond what a random arrangement would produce.","section":"Main text, 'Real-space imaging of DH'; Supplementary Text S5 and S7"},{"comment":"The simulated evidence for DH uses a subtraction method that is defined by multiple hand-set parameters (top 30% of extremal magnetization values, a Gaussian filter, threshold tau_M = 0.25, hole filling, and neglect of the smallest regions). The main text claims that the evolution shown in Fig. 5D is 'neither consistent with stochastic behavior (i.e. random switching) nor with the clear movement of a favorable domain,' but no statistical comparison to a stochastic null model or to a coarsening model is presented. Because the same subtraction defines both the changed regions and their spatial clustering, the simulation analysis is circular unless the observed clustering is compared with, for example, randomly shuffled spin flips or randomly placed changed regions. The ASD snapshots are valuable and show qualitative heterogeneous dynamics, but they do not yet quantitatively establish DH or the nucleation claim.","section":"Fig. 5 and Supplementary Text S10"},{"comment":"The choice of L = 12 nm is justified only by maximizing the fraction of boxes that can be classified as slow or fast, which conflates the analysis resolution with the physical length scale of the heterogeneous dynamics. A quantitative measure of the spatial correlation of the slow/fast labels, such as a spatial correlation function of the binary map, a cluster-size distribution compared to a random permutation, or a four-point susceptibility computed on the box labels, is needed to establish that the heterogeneity has a characteristic length scale. Without such a measure, the statement that the DH 'persisted' for variable box sizes (as long as L stays smaller than the pattern size) does not distinguish a real dynamical length scale from an artifact of the grid.","section":"Supplementary Text S7 and Fig. 4"}],"minor_comments":[{"comment":"The phrase 'ubiquitous behavior of dynamic heterogeneity' is unclear and overbroad: the paper studies a single material, and 'ubiquitous' is not defined. Suggest rewording to 'we demonstrate dynamic heterogeneity in the self-induced spin glass state of elemental neodymium.'","section":"Abstract"},{"comment":"The text in S7 says 'for diminishing box sizes L,' while the Fig. S7 caption says 'increasing box sizes L.' Please reconcile the direction of the series.","section":"Supplementary Text S7 and Fig. S7 caption"},{"comment":"The main text refers to the 'subtraction method in S9' when describing Fig. 5D, but the subtraction method is described in S10. Please correct the cross-reference.","section":"Main text, Fig. 5D description"},{"comment":"The conclusion ends with an orphan citation '(43-46)' that is not attached to a sentence or in-text discussion. These references are used in the Methods and Supplementary, so they should not appear as a dangling citation in the main text.","section":"Conclusion and References"},{"comment":"The threshold tau_M = 0.25 is introduced without a definition of how it is applied to the Gaussian-filtered image. Please specify the exact operation (e.g., pixels above 0.25 times the maximum intensity) so that the method is reproducible.","section":"Supplementary Text S10"}],"recommendation":"major_revision","confidential_remarks":"The paper is potentially important and the S(Q)-evolution part is solid, but the central DH claim currently depends on qualitative by-eye classification and threshold-based subtraction with no null-model validation. This issue is addressable with additional quantitative analysis, so major revision rather than rejection is appropriate. The manuscript fits the journal's scope and would be strengthened by adding a quantitative mobility metric, a spatial correlation function, and a random-switching null test for both the experimental and simulated data."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read. The main claim—that the self-induced spin glass in elemental Nd shows coexisting slow and fast magnetic regions in real space—is genuinely new and, if correct, connects spin glass aging to the DH literature in structural glasses. The paper deserves careful attention.\n\nWhat it does well: the SP-STM imaging is demanding and the authors control for the obvious troubles. They show the frozen state is stable for 27 hours (S1), they register images on defects, and the DH maps are reproduced on multiple samples, including one with lower defect density (S6). The S(Q) evolution under field cycling and the thermal reinitialization is documented quantitatively with the Jensen-Shannon divergence across four samples, and that part looks solid. The ASD simulations use the same DFT-derived Hamiltonian as in the group's earlier nature paper, with no fitting to the new experiment; the simulation shows nucleation-like heterogeneous dynamics that look like the experimental maps. The paper is also honest about the qualitative nature of the analysis and about not understanding the S(Q) evolution.\n\nThe soft spot is exactly what the stress-test says. The slow/fast assignment is done by eye, the box size L=12 nm is chosen in S7 because it maximizes the fraction of sortable boxes, and the simulation subtraction thresholds are hand-adjusted. There is no null-hypothesis test, no clustering statistic, and no quantitative mobility metric. A random arrangement of changed boxes would not necessarily look like the large patches in Fig. 4C, but the human eye is good at seeing patterns that aren't there. For a claim like DH, the field expects a four-point susceptibility or at least a simple spatial correlation of the labels. The paper's argument that the four-point function is hard to apply in a multi-Q system is fair, but that doesn't excuse providing no alternative metric.\n\nThis is not a fatal flaw in my view. The stability checks, the multiple samples, and the independent simulation support make the claim plausible. But the current analysis is not strong enough for the conclusions to be accepted at face value. The authors should be asked to provide a quantitative DH metric and a permutation test, and to release the raw images behind the by-eye maps. The S(Q) aging part is a separate, stronger result.\n\nFor whom: spin glass experimentalists and theorists, glass physics community. It deserves peer review. I would send it out, but with a clear request for quantitative analysis. If the authors can pass that bar, this becomes a high-impact paper.","headline":"A potentially important first real-space look at dynamic heterogeneity in a spin glass, but the key experimental classification is done by eye and needs a quantitative check before the claim is solid.","tokens_in":17409,"tokens_out":3303,"would_cite":true,"duration_ms":30199,"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":"Slow and fast magnetic regions coexist in neodymium's spin glass.","keywords":["spin glass","dynamic heterogeneity","neodymium","spin-polarized scanning tunneling microscopy","atomistic spin dynamics simulations","aging dynamics","magnetic structure factor","self-induced spin glass"],"falsifier":"Reanalyze the same before-and-after magnetization images with an automated, pre-registered classifier that extracts the dominant Q-states in each box without human judgment and compare the resulting slow/fast maps to the by-eye maps of Fig. 4 and Fig. S6; if the automated maps do not reproduce the coexistence of stable and changing regions across multiple samples and field cycles, the claim of dynamic heterogeneity would fail. A complementary check is to compute the four-point dynamic susceptibility from the image series and look for a growing dynamical correlation length, whose absence would contradict the dynamic-heterogeneity interpretation.","tokens_in":16378,"feed_emoji":"🧲","tokens_out":6272,"duration_ms":62667,"temperature":0.7,"pith_summary":"This paper reports that the self-induced spin glass state of elemental neodymium is dynamically heterogeneous: when the frozen magnetic patterns are nudged by magnetic-field cycles, some real-space regions keep their local order while others change it, and these two behaviors coexist side by side. The authors argue this is the same phenomenon known as dynamic heterogeneity in structural glasses, where different regions relax at very different rates. They also find that zero-field cooling imprints a reproducible set of magnetic periodicities into the glass, repeated field cycling shifts those periodicities along the high-symmetry axes until the change saturates, and warming through the ordered phase reinitializes the periodicities. If this is right, spin-glass aging cannot be captured by length-invariant mean-field descriptions alone; local length scales play an essential role.","feed_headline":"Slow and fast magnetic regions coexist in neodymium's spin glass","feed_subtitle":"Magnetic-field cycling reveals spatial regions that relax at different rates, linking spin glasses to structural glasses.","key_machinery":"The central objects are the nearly degenerate magnetic wave-vector pockets $Q_A$, $Q_B$, and $Q_C$ read from the static magnetic structure factor $S(Q)$, obtained by Fourier transforming spin-polarized scanning tunneling microscopy images of the frozen magnetization. The experimental protocol is a three-step dynamical cycle: zero-field cool through the Néel and glass transitions to 1.3 K, apply magnetic-field cycles to induce dynamics and re-image at zero field, and warm through the ordered phase to reinitialize. Slow and fast dynamics are mapped by dividing images into $12 \\times 12$ nm boxes and sorting each box by eye, using criteria in Supplementary Text S5, into stable, changing, or undetermined categories; the Jensen-Shannon divergence $\\mathcal{D}_{JS}$ quantifies the similarity between successive structure factors. The simulations use atomistic spin dynamics based on a Heisenberg Hamiltonian with density-functional-theory-derived exchange interactions, tracking the magnetization over hundreds of picoseconds, and use a subtraction-filtering method to identify regions of change.","core_discovery":"The central claim is that the self-induced spin glass state of elemental neodymium exhibits dynamic heterogeneity, both in spin-polarized scanning tunneling microscopy experiments and in atomistic spin dynamics simulations. In the frozen glass at 1.3 K, the magnetization is a patchwork of locally ordered multi-Q patterns; after dynamics are triggered by cycling the magnetic field to 2-7 T and the system is re-imaged at zero field, comparison of before-and-after images shows spatial regions whose patterns are unchanged (slow dynamics) beside regions whose patterns have changed (fast dynamics). This coexistence persists across repeated cycles, different field values, and several samples, and the simulations reproduce it, showing regions of change that nucleate at particular locations and then grow. In parallel, the magnetic structure factor $S(Q)$ evolves systematically under field cycling: the $Q_A$ pocket moves radially outward while $Q_B$ and $Q_C$ move inward, and the evolution saturates after a few cycles. Zero-field cooling reproducibly imprints a similar initial $S(Q)$, showing that the freshly cooled glass is an intermediate metastable state that ages under field cycling and can be thermally reinitialized.","pith_inferences":["A testable extension beyond this paper would be to apply the same imaging protocol to other frustrated but disorder-free magnets: if dynamic heterogeneity is generic to self-induced spin glasses, similar slow/fast coexistence should appear in those systems too.","The by-eye box classification could be replaced by an automated local-FFT classifier that assigns dominant Q-states per box; if such a classifier reproduces the slow/fast maps across samples, it would put the heterogeneity claim on a more quantitative footing and could yield a four-point-like dynamic correlation length.","The separation between reproducible periodicities and non-reproducible spatial patterns suggests a distinction between thermodynamic memory and configurational memory; partial warming excursions could reveal which Q-pockets reset first, something the present study does not test.","If field-history can reliably set and reset specific Q-pockets, repeated cycling might be used as a form of magnetic state writing, analogous to phase-change memory; the paper mentions this pathway but does not demonstrate it."],"forward_implications":["A correct account of aging in this spin glass must include local length scales, not only the global, length-invariant picture of mean-field spin-glass theory.","The systematic, sample-independent evolution of $S(Q)$ implies the zero-field-cooled state is an intermediate metastable state that ages toward a different set of periodicities under repeated field cycling.","Thermal cycling above the glass transition resets the periodicities but not the specific real-space arrangement of patterns, so reinitialization is statistical rather than a recovery of the exact configuration.","The coexistence of slow and fast magnetic regions provides a concrete experimental bridge between spin-glass aging and dynamic heterogeneity in structural glasses.","The simulations reproduce spatially heterogeneous dynamics without an accompanying change of $S(Q)$, suggesting that structural-factor aging and heterogeneous real-space dynamics are distinct aspects of the glassy behavior."],"supporting_citations":[{"why":"Establishes the self-induced spin glass state in elemental neodymium and supplies the imaging methods, sample preparation, and the Heisenberg Hamiltonian used in the simulations.","marker":"[24]"},{"why":"Provides the glass transition and multi-Q phase context plus the Jensen-Shannon divergence method used to quantify the similarity between structure factors.","marker":"[25]"},{"why":"Defines dynamic heterogeneity in glasses, the phenomenon this paper claims to observe in a spin glass.","marker":"[26]"},{"why":"Documents dynamical heterogeneities below the glass transition in simulations, the structural-glass counterpart used for comparison.","marker":"[30]"},{"why":"Introduces the four-point correlation function that motivates the real-space and time analysis and that the authors note cannot be directly applied to these multi-Q patterns.","marker":"[35]"},{"why":"Supplies the theory of nonlinear susceptibility and correlation length in glasses, used as context for quantifying dynamically correlated regions.","marker":"[36]"},{"why":"Supplies the atomistic spin dynamics code used for the simulations of neodymium magnetization dynamics.","marker":"[45]"},{"why":"Describes the algorithmic foundations of the atomistic spin dynamics simulations, including the Langevin heat-bath treatment.","marker":"[46]"}],"fun_headline_variants":["Neodymium spin glass shows mix of slow and fast dynamics","Magnetic patterns in neodymium relax at two speeds","Spin glass in neodymium reveals dynamic patchwork","Neodymium's frozen magnetic state has dual dynamics","Slow and fast magnetic regions mark neodymium's glass"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The classification of real-space regions into slow and fast dynamics is done by eye using criteria in Supplementary Text S5, and the box size of 12 nm is chosen in S7 because it maximizes the fraction of boxes that can be sorted into one of the two categories; if that visual sorting is biased or coincidental, the central observation of dynamic heterogeneity could be an artifact of the analysis.","fun_headline_variants_meta":{"raw":{"variants":["Neodymium spin glass shows mix of slow and fast dynamics","Magnetic patterns in neodymium relax at two speeds","Spin glass in neodymium reveals dynamic patchwork","Neodymium's frozen magnetic state has dual dynamics","Slow and fast magnetic regions mark neodymium's glass"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00043,"raw_usage":{"total_tokens":2205,"prompt_tokens":963,"completion_tokens":1242,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":579,"completion_tokens_details":{"reasoning_tokens":1157}},"tokens_in":579,"tokens_out":1242,"duration_ms":8531,"temperature":1.0,"reasoning_tokens":1157,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:57:43.293530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reanalyze the same before-and-after magnetization images with an automated, pre-registered classifier that extracts the dominant Q-states in each box without human judgment and compare the resulting slow/fast maps to the by-eye maps of Fig. 4 and Fig. S6; if the automated maps do not reproduce the coexistence of stable and changing regions across multiple samples and field cycles, the claim of dynamic heterogeneity would fail. A complementary check is to compute the four-point dynamic susceptibility from the image series and look for a growing dynamical correlation length, whose absence would contradict the dynamic-heterogeneity interpretation.","supporting_citations":[{"cited_title":"Kamber, A","cited_arxiv_id":null,"evidence_quote":"Establishes the self-induced spin glass state in elemental neodymium and supplies the imaging methods, sample preparation, and the Heisenberg Hamiltonian used in the simulations."},{"cited_title":"Verlhac, L","cited_arxiv_id":null,"evidence_quote":"Provides the glass transition and multi-Q phase context plus the Jensen-Shannon divergence method used to quantify the similarity between structure factors."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines dynamic heterogeneity in glasses, the phenomenon this paper claims to observe in a spin glass."},{"cited_title":"V ollmayr-Lee, W","cited_arxiv_id":null,"evidence_quote":"Documents dynamical heterogeneities below the glass transition in simulations, the structural-glass counterpart used for comparison."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the four-point correlation function that motivates the real-space and time analysis and that the authors note cannot be directly applied to these multi-Q patterns."},{"cited_title":"Donati, S","cited_arxiv_id":null,"evidence_quote":"Supplies the theory of nonlinear susceptibility and correlation length in glasses, used as context for quantifying dynamically correlated regions."},{"cited_title":"https://github.com/UppASD/UppASD, (2020)","cited_arxiv_id":null,"evidence_quote":"Supplies the atomistic spin dynamics code used for the simulations of neodymium magnetization dynamics."},{"cited_title":"What can we ‘learn’ with atoms?","cited_arxiv_id":null,"evidence_quote":"Describes the algorithmic foundations of the atomistic spin dynamics simulations, including the Langevin heat-bath treatment."}],"review_version":1}