{"id":"a13cc52c-7bac-4a80-af37-b2709e74298b","arxiv_id":"2505.18620","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The authors identify 23 candidate PT-symmetric odd-parity antiferromagnetic (AFM1) materials by combining a graph neural network screen with DFT energy comparisons.","lead":"Using an existing graph neural network and density functional theory, the authors screened more than 45,000 Materials Project compounds and propose 23 materials that may host parity-time-reversal-symmetric (PT-symmetric) antiferromagnetic order. The list includes 3 already verified, 10 synthesized but uncharacterized, and 10 as-yet-unsynthesized candidates for effects like magnetopiezoelectricity and spin-current generation.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'all 23 materials' stability claim is contradicted by the paper's own Table S1 for ErGe3, and the same no-SOC PBE protocol that fails there is used to rank the rare-earth candidates.","rationale":"The reader's weakest assumption correctly identifies the DFT level as the main risk. I partially agree, and the paper itself provides a sharper internal test: ErGe3, an experimentally verified AFM1 material, is predicted by the same no-SOC PBE protocol to favor FM over AFM1, and the Methods acknowledge this while retaining the material. That makes the 'all 23' sentence in the Results literally false and demonstrates that the validation metric can fail for exactly the rare-earth family that dominates the new candidate list. The GNN test metrics are also omitted, but the final list is DFT-filtered, so the DFT ranking is the load-bearing evidence. Because the contradiction and the SOC limitation are disclosed in the paper, and because this is a screening candidate list rather than a claim of definitive ground states, the appropriate outcome remains CONDITIONAL rather than REJECT. The concrete test above would settle whether the rare-earth candidates survive a more reliable treatment; until then, the central stability claim is overstrong as written.","tokens_in":19376,"tokens_out":4355,"duration_ms":40454,"concrete_test":"Independently recompute the magnetic configuration energies in Table S1 with spin-orbit coupling and a Hubbard U (or hybrid functional) for all non-verified candidates, starting with the rare-earth entries (ErGe3, GdSn2, GdSnGe, PmB3, Ho4Zr3O12, SmB4, NdB4) and the borderline ΔE entries (NaNiPO4, GdSn2, SrMn2Ge2, TaMnO4), using the same structures and including noncollinear initializations. If AFM1/AFM1a is no longer the lowest-energy state for any candidate, or if ErGe3 remains FM-favored, then the 'all 23' stability claim and the new-candidate list need revision. A minimal version: rerun ErGe3 with SOC+U; if the ranking flips to AFM1, the rare-earth candidates need the same treatment; if it does not flip, the PBE ranking is demonstrably unreliable for the rare-earth subset.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in 'The 23 AFM1 candidate materials' states: 'For all 23 materials, the AFM1/AFM1a configuration was found to be the most stable magnetic state among the four magnetic configurations calculated, with a minimum energy difference of at least 1 meV/atom.' This is false for ErGe3 by the authors' own numbers: Table S1 row 2 gives AFM1 = 0 and FM = -0.09 meV/atom, and the Methods explicitly say FM is 0.09 meV/atom lower but the material was retained because neutron diffraction confirms AFM1. The claim should be qualified or the table corrected. More importantly, this is not a harmless bookkeeping error. The validation metric for the 20 newly listed candidates is PBE DFT without SOC or Hubbard U, comparing only FM and three collinear AFM configurations. The Discussion concedes SOC was omitted and 'may affect the relative energy stability.' ErGe3 is an in-family counterexample: for a rare-earth 4f magnet, the same protocol predicts FM over the experimentally confirmed AFM1 state. Other listed rare-earth systems (GdSn2, GdSnGe, PmB3, Ho4Zr3O12, NdB4, SmB4) are exactly the cases where PBE without SOC/U is least reliable, and two new candidates have ΔE barely above 1 meV/atom (GdSn2 1.07, NaNiPO4 1.04). The load-bearing support for the new predictions is therefore the DFT energy ranking, and that ranking is not established at the level claimed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an AI-assisted workflow to identify parity-time-reversal-symmetric odd-parity antiferromagnetic (AFM1) materials. It applies a graph neural network originally developed for altermagnet screening, incorporates AFM1-specific symmetry constraints during dataset preparation, screens Materials Project compounds, and then ranks candidate magnetic configurations with PBE DFT. The authors report 23 AFM1 candidates, tabulate their magnetic point groups and symmetry-imposed odd-parity dispersion terms, and present a tight-binding model for TaFeO4 that illustrates asymmetric bands and the associated magnetopiezoelectric, nonreciprocal-conductivity, and SOC-free nonlinear spin Hall responses. The paper's central claim is that for all 23 candidates the AFM1/AFM1a configuration is the most stable among the four tested magnetic configurations by at least 1 meV/atom.","tokens_in":19776,"tokens_out":6996,"duration_ms":61459,"significance":"If the candidate list were validated, the paper would expand the known AFM1 family from the 123 symmetry-classified materials to include new synthesized and not-yet-synthesized compounds, and the tables of magnetic point groups and odd-parity dispersion forms would be a useful resource for experimental searches. The paper gives credit to the public code base it adapts, provides Materials Project IDs, and includes convergence data in the supplementary material, which is commendable for reproducibility. The tight-binding demonstration of ε_odd(k) and its observable consequences is a clear pedagogical contribution. However, the significance is conditional: the screening model has no held-out validation, the DFT ranking omits spin-orbit coupling and Hubbard U for rare-earth magnets, and one of the 23 entries directly contradicts the central energy-stability claim. With those issues fixed or honestly qualified, the candidate list and symmetry tables would be a useful step for the field.","major_comments":[{"comment":"The sentence 'For all 23 materials, the AFM1/AFM1a configuration was found to be the most stable magnetic state among the four magnetic configurations calculated, with a minimum energy difference of at least 1 meV/atom' is contradicted by the authors' own Table 1 and Table S1. Table 1 row 2 lists ErGe3 with ΔE = -0.09 meV/atom, and Table S1 reports FM at -0.09 meV/atom relative to the AFM1 reference. The Methods section explicitly states that FM is 0.09 meV/atom lower than AFM1 for this material and that it was retained only because neutron diffraction confirms AFM1 ordering. The Abstract repeats the same claim ('DFT calculations show that AFM1 has the lowest energy among the tested magnetic configurations in 23 candidate materials'). This is a load-bearing inconsistency, not a typo, because the paper's headline result is the 23-candidate AFM1-stability claim. Please either separate ErGe3 from the DFT-validated set, or restate the claim as covering 22 DFT-validated materials plus one experimentally confirmed AFM1 material whose computed ground state is FM, and remove the 'at least 1 meV/atom' wording for ErGe3.","section":"Results, 'The 23 AFM1 candidate materials'; Table 1, row 2; Table S1; Methods, 'DFT Calculations'"},{"comment":"The GNN screening is not independently validated. The 14 positive samples initially reserved for testing were later added to a second fine-tuning round, and the text states that 'results from this test set will not be further discussed.' This means there is no held-out test set on which model accuracy, precision, or recall is reported. Moreover, the second and third fine-tuning rounds use the previous rounds' DFT-validated candidates as positive labels, so later predictions are trained partly on earlier model output, making the screening statistics circular. The 'AI-predicted' claim in the title and abstract therefore rests on an unbenchmarked model. Please report classification metrics on a test set that was never used for training or fine-tuning, or explicitly describe the screening as exploratory rather than validated.","section":"Methods, 'AI-Based Screening'"},{"comment":"The DFT energy-ranking protocol is not reliable at the claimed level for the rare-earth candidates. The Discussion concedes that SOC is omitted, and the Methods do not mention Hubbard U corrections, yet Table 1 includes several 4f systems (ErGe3, NdB4, SmB4, GdSn2, GdSnGe, PmB3, Ho4Zr3O12). ErGe3 is a direct in-family counterexample: the same PBE-no-SOC protocol predicts FM over the experimentally confirmed AFM1 state. Several tabulated ΔE values are also small (GdSn2 at 1.07 meV/atom, NaNiPO4 at 1.04 meV/atom), which is below the expected accuracy of PBE for such localized-moment systems. Please re-rank the rare-earth and near-threshold candidates with spin-orbit coupling and, where appropriate, Hubbard U, or downgrade them to 'unconfirmed candidates' rather than DFT-validated predictions.","section":"Discussion (limitations); Methods, 'DFT Calculations'; Table 1"},{"comment":"It is not clear that the code-generated AFM1/AFM1a configurations correspond to the physically relevant experimental magnetic structures. For ErGe3 the experimental order is described as lying within the ab-plane, and for NdB4 it is described as noncollinear, yet the DFT comparisons are performed on collinear AFM1/AFM1a configurations. Table S1 shows NdB4 AFM1a at -2.00 meV/atom relative to AFM1, so it is unclear whether the ΔE = 2.01 meV/atom in Table 1 refers to AFM1a and whether that state represents the experimentally verified noncollinear order. Please specify how AFM1 and AFM1a are constructed for each verified material, and for the remaining candidates state explicitly that the AFM1 label is an assumption about the magnetic structure rather than a confirmed experimental arrangement.","section":"Methods, 'DFT Calculations'; Table S1, rows 2–3"}],"minor_comments":[{"comment":"Add a footnote to Table 1 for ErGe3 noting that the negative ΔE is an exception to the column definition and that the material is retained on the basis of experimental neutron-diffraction evidence.","section":"Table 1 and Table S1"},{"comment":"Clarify whether the reported ΔE of 2.01 meV/atom is the energy difference between AFM1a and the lowest non-AFM1 configuration or between AFM1 and the lowest non-AFM1 configuration, and whether AFM1a corresponds to the noncollinear experimental magnetic structure.","section":"Table S1, row 3 (NdB4)"},{"comment":"The caption contains an unbalanced parenthesis: 'Centering-translation-related Mn pairs: (Mn2, Mn4) and Mn3, Mn5)' should read '(Mn2, Mn4) and (Mn3, Mn5)'.","section":"Fig. S7 caption"},{"comment":"The Methods text states that a kinetic energy cutoff of 100 Ry and a 10x10x10 k-point grid were used, while Table S2 lists final converged parameters ranging from 90 to 220 Ry and finer k-point grids for most materials; clarify that the Methods sentence describes the initial screening parameters and that Table S2 lists the converged parameters used for the final energy rankings.","section":"Methods, 'DFT Calculations' and Table S2"},{"comment":"The text refers to 'Yanase et al.' while the cited references use 'Watanabe and Yanase'; use a consistent citation style for this important source.","section":"Methods, 'AI-Based Screening' and References [2,3]"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a candidate-screening report, and its claims are stated more strongly than the evidence supports. The ErGe3 contradiction is fixable, but the circular fine-tuning procedure and the absence of a truly held-out GNN test set are more fundamental concerns that should be addressed before publication. The lack of SOC/Hubbard U for rare-earth magnets is a further correctness risk that needs either additional calculations or an explicit downgrade of the affected candidates. I would recommend accepting a revision that adds honest model validation, reclassifies ErGe3, and either strengthens or carefully qualifies the DFT ranking for the rare-earth entries."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful core here is the candidate table: 23 PT-symmetric odd-parity antiferromagnets (AFM1), with magnetic point groups, symmetry-imposed odd-parity dispersion forms, and energy differences against three competing collinear states. The AFM1-specific negative-sample construction and the iterative DFT-feedback fine-tuning are a legitimate new application of the Gao GNN, and the paper is honest about what was adopted unchanged. Experimentalists working on magnetopiezoelectricity or nonreciprocal transport will want this shortlist.\n\nThe soft spots are real, and the reader's stress-test holds up. The abstract and Table 1 claim all 23 candidates have AFM1 lowest by at least 1 meV/atom, but Table S1 gives ErGe3 FM lower by 0.09 meV/atom. The Methods explain the retention on experimental grounds, which is defensible, but the claim as written is false. That matters beyond bookkeeping: ErGe3 is a rare-earth 4f magnet, and the same no-SOC, no-Hubbard-U PBE protocol that ranks FM above the experimentally confirmed AFM1 state is used to rank six other rare-earth candidates (GdSn2, GdSnGe, PmB3, Ho4Zr3O12, NdB4, SmB4). Two new candidates (GdSn2, NaNiPO4) sit just above the 1 meV/atom cutoff. So the load-bearing validation for the genuinely new predictions is exactly where the method is least reliable.\n\nThe active-learning loop is a separate concern. Each fine-tuning round adds previously DFT-validated candidates to the positive set, and the reserved test set is explicitly not discussed. That means there is no independent accuracy benchmark for the model after round one. It is not fatal — screening pipelines often do this — but it should be stated as a limitation and, ideally, supplemented with test-set numbers.\n\nWhat is solid: the symmetry analysis and MPG tables are careful, the literature verification steps are described, and the authors disclose the SOC omission and the ErGe3 exception rather than hiding them. The paper is clear on its own terms and cites the relevant prior work.\n\nWho gets value: experimentalists seeking a concrete candidate list, and method developers comparing AI-plus-DFT screening workflows. It deserves a serious referee — this is not a desk reject — but referee time should be spent on whether the corrected claim, qualified DFT protocol, and missing test metrics are adequately addressed. I would ask for revision before acceptance.","headline":"A genuinely useful candidate shortlist undermined by an internal inconsistency in the headline stability claim and a DFT protocol too light for several rare-earth entries.","tokens_in":20265,"tokens_out":1461,"would_cite":true,"duration_ms":14168,"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":"An AI-plus-DFT pipeline identifies 23 candidate PT-symmetric antiferromagnets whose symmetry-forced band asymmetry enables magnetopiezoelectric, nonreciprocal, and spin-orbit-free spin Hall responses.","keywords":["PT-symmetric magnets","AFM1 antiferromagnets","odd-parity band dispersion","graph neural network screening","density functional theory","magnetopiezoelectric effect","nonlinear spin Hall effect","materials discovery"],"falsifier":"Neutron diffraction or resonant X-ray scattering on any of the ten synthesized but magnetically uncharacterized candidates, such as SmB4, BaMn2Sn2, or TaFeO4, showing a non-AFM1 magnetic order would remove that candidate from the list, and a DFT calculation that adds spin-orbit coupling and a Hubbard U correction and flips the AFM1 state above another configuration would reopen the whole list.","tokens_in":19155,"feed_emoji":"🧲","tokens_out":14526,"duration_ms":95507,"temperature":0.7,"pith_summary":"The paper claims that a graph neural network trained on known examples, followed by density functional theory energy comparisons, can predict new parity-time-reversal-symmetric odd-parity antiferromagnets (AFM1 materials), and delivers a list of 23 candidates. These magnets break space inversion and time reversal separately while preserving their product PT, which forces an odd-parity term $\\varepsilon_{\\mathrm{odd}}(\\mathbf{k})$ into the band dispersion and makes the bands asymmetric between $\\mathbf{k}$ and $-\\mathbf{k}$. That asymmetry is the engine behind the magnetopiezoelectric effect, nonreciprocal conductivity, photocurrent generation, and a nonlinear spin Hall effect that needs no spin-orbit coupling. Of the 23 candidates, three are already verified by neutron diffraction, ten are synthesized compounds whose magnetic order remains unknown, and ten are unsynthesized predictions — together adding twenty new or previously uncharacterized entries to the known family of 123 such materials. If the predictions hold, experimentalists have a concrete target list of compounds in which to hunt for these effects.","feed_headline":"AI screening uncovers 23 PT-symmetric magnet candidates","feed_subtitle":"Twenty new candidates for magnetopiezoelectric and spin-orbit-free spin Hall effects.","key_machinery":"The load-bearing object is the AFM1 configuration itself: a parity-time-reversal-symmetric odd-parity antiferromagnet in which the nonmagnetic crystal's inversion centers sit between, rather than on, the magnetic atoms, so that inversion P and time reversal T are each broken while their product PT is preserved. Its physical consequence is the odd-parity energy term $\\varepsilon_{\\mathrm{odd}}(\\mathbf{k})$, which the paper derives for a two-sublattice tight-binding model of TaFeO4 as $\\varepsilon_{\\mathrm{odd}}(\\mathbf{k}) = M_y \\lambda_y \\sin k_y / \\sqrt{t_{x,\\mathbf{k}}^2 + t_{y,\\mathbf{k}}^2 + \\lambda_{\\mathbf{k}}^2}$, showing that the AFM1 exchange field $M_y$ acting with spin-orbit coupling tilts the dispersion into $\\varepsilon(\\mathbf{k}) \\neq \\varepsilon(-\\mathbf{k})$. Carrying the discovery pipeline are a graph neural network pretrained to reconstruct crystal graphs and fine-tuned on known AFM1 examples against 21,600 negative samples, a symmetry filter that restricts candidates to centrosymmetric cells with two or four magnetic atoms not sitting on inversion centers, and spin-polarized DFT that compares one ferromagnetic and three collinear antiferromagnetic configurations against a 1 meV/atom stability threshold.","core_discovery":"For a fair reader, the paper's claim is that AI screening plus density functional theory can find new members of a rare symmetry class, and specifically that 23 materials host the AFM1 state: a parity-time-reversal-symmetric odd-parity antiferromagnetic order that breaks inversion and time reversal separately while preserving their product PT. The central output is a table listing the 23 candidates with their magnetic element, space group, conductivity type, energy difference, and synthesis status, alongside a companion table that gives each candidate's magnetic point group and the symmetry-imposed form of the odd-parity energy contribution $\\varepsilon_{\\mathrm{odd}}(\\mathbf{k})$ that makes the bands asymmetric. The paper states that for all 23 materials the AFM1 arrangement is the most stable of the four collinear configurations tested; its own convergence section records one exception, ErGe3, whose ferromagnetic state is 0.09 meV/atom lower but which was retained because neutron diffraction had already established its AFM1 order. Of the 23, three are experimentally confirmed AFM1 materials, ten are synthesized compounds without known magnetic structure, and ten are purely computational predictions.","pith_inferences":["Read against the paper's own convergence table, the 1 meV/atom stability criterion is not applied as a hard rule: ErGe3 entered the final list with its ferromagnetic state 0.09 meV/atom lower, so the list mixes DFT-stable candidates with experiment-backed ones and the small energy-difference entries should be treated as borderline.","The DFT comparison covers only one ferromagnetic and three collinear antiferromagnetic orders; noncollinear or incommensurate states, the kind that actually win in the verified candidate NdB4, are untested competitors for the other twenty candidates, so the AFM1 label there is a plausible prediction rather than an established ground state.","A natural next step the paper does not take is to compute the actual response tensors for the energetically strongest candidates, such as the 18.36 meV/atom prediction FeGe3, which would rank the list by expected effect size rather than by energy stability alone.","The pipeline's lower-confidence band, with classification scores between 0.5 and 0.9, already yielded one additional candidate, which suggests the search is not saturated and that more DFT verification capacity would likely add further members to the AFM1 family."],"forward_implications":["The list gives experimentalists twenty new or previously uncharacterized targets in which to search for AFM1 order, and the ten synthesized candidates can be probed directly by neutron diffraction or muon spin rotation.","Because the companion table fixes the symmetry-allowed form of $\\varepsilon_{\\mathrm{odd}}(\\mathbf{k})$ for each candidate, materials with a linear term such as $\\alpha_z k_z$ are predicted to show the magnetopiezoelectric effect, nonreciprocal conductivity, and photocurrent generation, while those with cubic terms such as $\\alpha k_x k_y k_z$ show these responses only at higher order.","Candidates whose magnetic point group meets the published symmetry criteria are predicted to exhibit a nonlinear spin Hall effect without spin-orbit coupling, making light-element compounds viable spin-current sources.","The screening loop is iterative: the 23 candidates can be folded back into the training set, so the same pipeline can search for further AFM1 materials or be retargeted at other symmetry-enriched magnetic orders.","Because breaking structural inversion turns the PT-protected degeneracy into momentum-dependent spin splitting, the predicted AFM1 materials double as candidate precursors for altermagnetism under strain or substrate engineering."],"supporting_citations":[{"why":"It supplies the graph neural network architecture, pretraining scheme, and publicly available code that the screening pipeline adopts without modification.","marker":"[7]"},{"why":"It defines the AFM1 class and the symmetry analysis from which the odd-parity energy contributions and their responses follow.","marker":"[2]"},{"why":"It provides the group-theoretical classification of multipole order that yields the 123 known AFM1 materials and the magnetopiezoelectric response.","marker":"[3]"},{"why":"It derives the nonlinear spin Hall effect in PT-symmetric collinear magnets, the spin-orbit-free response the candidates are screened for.","marker":"[4]"},{"why":"It provides the crystallographic database whose structure files form the positive, negative, and candidate datasets for training and screening.","marker":"[8]"},{"why":"It supplies the experimentally determined magnetic structures used to confirm the three verified AFM1 candidates.","marker":"[31]"},{"why":"It is the neutron diffraction study that established AFM1 order in BaMn2Ge2, the first verified candidate.","marker":"[16]"},{"why":"It is the neutron diffraction study that established AFM1 order in ErGe3, the candidate retained even though its ferromagnetic state lies 0.09 meV/atom lower.","marker":"[17]"},{"why":"It is the polarized neutron study of NdB4, the noncollinear verified candidate.","marker":"[18]"}],"fun_headline_variants":["AI finds 23 PT-symmetric odd-parity magnets","23 PT-symmetric magnets predicted by AI and DFT","AI screen reveals 23 magnets with PT-symmetric order","23 odd-parity magnets with PT symmetry found via AI","AI-assisted search yields 23 PT-symmetric magnet candidates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The candidate list stands on the assumption that the collinear spin arrangements the code labels AFM1 match the magnetic order these materials would actually form, and that PBE-level density functional theory without spin-orbit coupling or Hubbard corrections, comparing only one ferromagnetic and three collinear antiferromagnetic states, ranks the true ground states correctly.","fun_headline_variants_meta":{"raw":{"variants":["AI finds 23 PT-symmetric odd-parity magnets","23 PT-symmetric magnets predicted by AI and DFT","AI screen reveals 23 magnets with PT-symmetric order","23 odd-parity magnets with PT symmetry found via AI","AI-assisted search yields 23 PT-symmetric magnet candidates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000659,"raw_usage":{"total_tokens":3010,"prompt_tokens":939,"completion_tokens":2071,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":555,"completion_tokens_details":{"reasoning_tokens":1991}},"tokens_in":555,"tokens_out":2071,"duration_ms":12485,"temperature":1.0,"reasoning_tokens":1991,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:29:06.506210+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Neutron diffraction or resonant X-ray scattering on any of the ten synthesized but magnetically uncharacterized candidates, such as SmB4, BaMn2Sn2, or TaFeO4, showing a non-AFM1 magnetic order would remove that candidate from the list, and a DFT calculation that adds spin-orbit coupling and a Hubbard U correction and flips the AFM1 state above another configuration would reopen the whole list.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the graph neural network architecture, pretraining scheme, and publicly available code that the screening pipeline adopts without modification."},{"cited_title":"Watanabe and Y","cited_arxiv_id":null,"evidence_quote":"It provides the group-theoretical classification of multipole order that yields the 123 known AFM1 materials and the magnetopiezoelectric response."},{"cited_title":"Hayami, M","cited_arxiv_id":null,"evidence_quote":"It derives the nonlinear spin Hall effect in PT-symmetric collinear magnets, the spin-orbit-free response the candidates are screened for."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It provides the crystallographic database whose structure files form the positive, negative, and candidate datasets for training and screening."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the experimentally determined magnetic structures used to confirm the three verified AFM1 candidates."},{"cited_title":"Malaman, G","cited_arxiv_id":null,"evidence_quote":"It is the neutron diffraction study that established AFM1 order in BaMn2Ge2, the first verified candidate."},{"cited_title":"Schobinger-Papamantellos, G","cited_arxiv_id":null,"evidence_quote":"It is the neutron diffraction study that established AFM1 order in ErGe3, the candidate retained even though its ferromagnetic state lies 0.09 meV/atom lower."},{"cited_title":"Metoki, H","cited_arxiv_id":null,"evidence_quote":"It is the polarized neutron study of NdB4, the noncollinear verified candidate."}],"review_version":1}