{"id":"14c2762f-86a2-459c-8082-4dcd31d47155","arxiv_id":"2607.20885","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In SrTiO3−xHx, hydride pairs are stabilized by ~200 meV/H when they sit on adjacent octahedron corners (cis), a preference prior ATiO3−xHx models overlooked.","lead":"This paper maps how hydrogen ions arrange themselves in the perovskite oxyhydride SrTiO3−xHx using hybrid-DFT calculations and cluster expansion. It finds that two hydrides prefer to occupy neighboring corners of the same octahedron ('cis'), with a stabilization of about 200 meV per hydrogen—so isolated-H supercell models miss a key ordering effect.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"HSE06 is the unvalidated reference; the ~200 meV/H cis binding energy and the proxy-U fitting both inherit any hybrid-functional bias, so the central cis-preference claim could be a functional artifact.","rationale":"The reader's weakest assumption—lack of validation for HSE06 as the reference for ordering energetics—is the single most load-bearing concern. Every supporting line of evidence for the cis preference ultimately traces back to HSE06 energies or to a proxy-U fitted to HSE06. The authors acknowledge this limitation explicitly, which is intellectually honest, but it does not reduce its impact: if HSE06 is biased in this system, the quantitative ~200 meV/H binding energy and the qualitative 'strong preference' could both be artifacts. Other concerns (e.g., CE2 CV error of 28.1 meV/anion, incomplete verification of newly predicted ground states, finite-size effects in the binding-energy supercells) are real but secondary, since the direct binding energy is independent of the CE fit and the qualitative trend appears across multiple analyses. The proposed test—comparing with an independent electronic-structure method or varying the hybrid mixing fraction—directly targets the functional dependence that is the crux of the concern. Because the reader's verdict is already CONDITIONAL, my analysis does not change the verdict; it reinforces it.","tokens_in":17015,"tokens_out":9024,"duration_ms":97844,"concrete_test":"Compute the isolated-H, cis-pair, and trans-pair binding energies at x=0.125 in identical supercells using an independent method (e.g., RPA or PBE0) and/or with HSE06 exact-exchange mixing α varied from 0.2 to 0.3. If the cis stabilization does not remain ≥100 meV/H across these choices, the HSE06-based cis-preference claim is not robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim rests on HSE06 as the reference for ordering energetics, but the authors explicitly state in the Introduction: 'there is currently no direct evidence that hybrid functionals necessarily yield the most accurate ordering stability in ATiO3−xHx.' The proxy-U (U=3.3 eV) is fitted to HSE06 and used to select the HSE06 CE training set, so the entire pipeline—direct cis/trans binding energies (Figure 7c), ground-state orderings (U, V, W), and MC sampling—is conditioned on HSE06's accuracy. If HSE06 over-localizes electrons in reduced Ti oxyhydrides, it could artificially stabilize cis O4H2 configurations that favor polaron localization; the fact that plain PBE shows a 'less pronounced' cis stabilization (Figure 7c) indicates strong functional sensitivity. The ~200 meV/H quantitative claim and the qualitative 'strong preference' both hang on HSE06 being physically correct, not merely internally consistent. Without independent validation (experimental SRO or higher-level theory), the central result remains a functional artifact hypothesis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a cluster expansion (CE) framework for anion ordering in SrTiO3−xHx, using HSE06 hybrid-DFT energetics as the reference. The authors first show that ordering stability and ground states are highly sensitive to the choice of DFT+U on Ti, with variations on the order of 100 meV/anion. They fit U = 3.3 eV as an HSE06 proxy using MAD/MAPD metrics, use this proxy to perform active-learning structure selection, and then compute HSE06 energies for 68 key configurations to build a final CE (CE2). On this basis they report a strong preference for the O4H2 cis configuration: ground-state orderings U, V, and W all contain cis-type motifs; dilute-limit calculations at Sr8Ti8O23H give roughly 200 meV/H stabilization of a cis H2 pair over isolated H; and Monte Carlo sampling of the CE models shows short-range order favoring cis octahedra at 350 K. The paper explicitly acknowledges that HSE06 has not been independently validated for this ordering problem and that CE2 has limited fit quality, but the Abstract and Conclusion nonetheless present the cis preference as a robust finding.","tokens_in":17338,"tokens_out":6871,"duration_ms":73522,"significance":"If correct, the central result is important: it would overturn the common isolated-H picture for modeling ATiO3−xHx and imply that anion ordering must be included in studies of hydride migration, polaron formation, and other properties. The paper has clear strengths: the direct dilute-limit cis/trans binding-energy comparison is a clean, selection-independent DFT test; the DFT+U sensitivity analysis is systematic and useful to the community; and the proposed hybrid-DFT CE construction workflow, combining a proxy functional with active learning, is a practical methodological contribution. The authors also deserve credit for explicitly flagging the main caveat about HSE06 reliability. However, the quantitative claim of ~200 meV/H and the qualitative claim of a 'strong preference' are conditioned on the accuracy of HSE06 ordering energetics, which is not independently established. In addition, the HSE06 training set is selected using a proxy fitted to HSE06, and the final CE has a 5-fold CV error of 28.1 meV/anion—comparable to kT at the Monte Carlo temperature used for the short-range-order predictions. These issues make the paper suitable for publication only after substantial revis","major_comments":[{"comment":"The central quantitative result (~200 meV/H cis stabilization) and the broad 'strong preference' claim rest on HSE06 as the reference, but the authors state in the Introduction that 'there is currently no direct evidence that hybrid functionals necessarily yield the most accurate ordering stability in ATiO3−xHx.' Plain PBE shows a 'less pronounced' cis stabilization, so the magnitude and possibly the existence of the preference are functional-dependent. This is a load-bearing point, not a minor caveat. Please add an independent check—for example, RPA or DMC calculations for a few anion orderings, or HSE calculations with a different exact-exchange mixing fraction—or explicitly reframe the Abstract and Conclusion so that the cis preference is presented as conditional on hybrid-functional accuracy rather than as an established physical tendency.","section":"Introduction; Fig. 7c"},{"comment":"The HSE06 training set was selected using the proxy-U CE1, and U = 3.3 eV was itself fitted to HSE06 ordering stability (Eqs. 3–4). The agreement between proxy-U and HSE06 in Fig. 6a,b (MAD 7.3 meV/anion) is therefore not an independent validation of the proxy: the 68 structures are biased toward configurations that proxy-U already identifies as important. The direct dilute-limit binding calculation escapes this circularity, but the ground-state hull (U, V, W) and the MC sampling do not. Please quantify the selection bias—for example, by computing HSE06 energies for a D-optimal-only or random subset and showing that the proxy-U selection does not change the hull/CE—or explicitly restrict the proxy-U reliability claim to the dilute-limit binding energy.","section":"HSE06 and Proxy-U Cluster Expansion; Fig. 6a,b"},{"comment":"CE2 has a 5-fold CV error of 28.1 meV/anion (Fig. 6c), and the text notes that CE2 predicts additional nearly degenerate ground states that 'deepen the hull slightly' but were not recalculated with HSE06. At the MC temperature of 350 K, kT is about 30 meV/anion, so the CE2 fitting error is comparable to the thermal energy governing the disordered-state octahedral populations reported in Fig. 8. The qualitative agreement between CE1 and CE2 is reassuring, but the quantitative SRO values (e.g., the cis population relative to O5H1) may lie within the model uncertainty. Please report the magnitude of the missed hull deepening, provide error estimates on the MC SRO quantities from ECI resampling or an equivalent uncertainty propagation, and state explicitly whether the 'strong preference' conclusion is robust to these uncertainties.","section":"HSE06 CE fitting, Fig. 6c and Fig. 8"},{"comment":"The choice of U = 3.3 eV as the HSE06 proxy is based on only two compositions, Sr2Ti2O5H and SrTiO2H (Fig. 4e,f). The later comparison on 68 structures is broader, but that comparison is affected by the selection bias described above. Since the proxy-U is used to select the HSE06 training set and to guide the MC simulations, its domain of validity should be tested on a composition-diverse set not involved in the fitting—for example, by leave-one-composition-out cross-validation. This would strengthen the claim that proxy-U is a reliable structure-selection tool across the full composition range x = 0 to 1.","section":"DFT+U Sensitivity, Eqs. (3)–(4)"}],"minor_comments":[{"comment":"There is a typo in Eq. (2): 'tans c' should be 'trans c'. Also, please clarify in the caption or text how the rows of the C-matrix correspond to the cluster labels in Fig. 3, since the current notation is dense.","section":"Eq. (2) and Fig. 3"},{"comment":"The caption lists panels (d) and (e) as 'Sr2Ti2O5H and SrTiO2H differences...' while the main text refers to panels (d), (e), and (f) with overlapping descriptions. Please harmonize the panel numbering and in-text references.","section":"Fig. 4 caption"},{"comment":"The binding-energy definition is described verbally but not given algebraically. Please state explicitly how E_b for the cis and trans H2 pairs is computed from the total energies of the 2×2×2 and 2×2×4 supercells, so the reader can verify the -200 meV/H value.","section":"Fig. 7c"},{"comment":"The Abstract says '200 meV per hydride' while the body says 'nearly 200 meV/H under proxy-U and exceeds 200 meV/H with HSE06.' Please make the quantitative statement consistent and indicate which functional the Abstract value refers to.","section":"Abstract and Conclusion"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is honest about its main limitation, but the headline claim is broader than the evidence currently supports. The direct dilute-limit cis/trans comparison is a strong, selection-independent result and should be highlighted. The main risk is the self-consistent loop: a proxy fitted to HSE06 is used to select the HSE06 training set, and the resulting agreement is then cited as validation. I would ask the authors to break this loop at least partially—by adding an independent functional/method check or a selection-bias test—and to quantify the CE2 uncertainty in the MC short-range-order predictions. If these additions are made, the paper could be a solid contribution; I would not reject it, because the core methodology is sound and the cis preference is plausibly correct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First, this paper does something genuinely useful: it quantifies how much anion-ordering energetics in SrTiO3-xHx depend on the Hubbard U, showing 100 meV/anion variations, and then builds a practical workflow where U=3.3 eV acts as a low-cost proxy for HSE06 ordering energies. That workflow is a contribution in its own right. Second, the central physical claim — O4H2 cis configurations are strongly stabilized, by ~200 meV/H at dilute concentrations — is credible and worth taking seriously. The authors support it from several angles: ground-state hulls, octahedral fraction statistics, a direct dilute-limit binding calculation, and MC sampling. The direct calculation in Sr8Ti8O23H comparing isolated H with cis and trans pairs is the cleanest piece of evidence. It does not depend on the CE fit, and it shows cis binding under both HSE06 and proxy-U, while trans is unstable. That is a solid result. Now the soft spots, in proportion. The paper explicitly states there is no direct evidence that HSE06 gives the most accurate ordering stability in ATiO3-xHx. Everything is conditioned on that assumption, including the proxy-U, which is fitted to HSE06 and then used to select the HSE06 training set. So the hull agreement in Fig. 6a is partly self-consistent. If HSE06 over-localizes, the cis preference could be exaggerated. The fact that plain PBE shows a smaller effect is a warning. The authors are honest about this, but it does mean the 200 meV/H number is a functional-benchmark value, not a measured or higher-level verified quantity. The CE2 CV error of 28 meV/anion is high, but the main claim does not rest on CE2 for its dilute-binding result. The MC probabilities would be stronger with error bars. Overall, this is a competent, well-written, honest paper. The authors caveat their own assumptions more than most. The cis preference is a new and plausible result that will influence how the community models oxyhydrides. A serious referee should engage with it. It is not a desk reject, and it does not need a hostile referee — it needs one who will check the HSE06 sensitivity question and push for error bars on MC and possibly a direct higher-level check on the cis binding. I would bring it to reading group and would cite it.","headline":"Worth a serious referee: the cis-pair stabilization is a real, well-argued effect, but the unvalidated HSE06 reference and the proxy-U fitting circularity keep the quantitative ~200 meV/H claim from being closed.","tokens_in":17833,"tokens_out":2831,"would_cite":true,"duration_ms":30462,"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":"In SrTiO3−xHx oxyhydrides, hydride ions strongly prefer to sit on adjacent anion sites, a cis O4H2 ordering that prior simulation studies overlooked.","keywords":["perovskite oxyhydrides","anion ordering","cluster expansion","HSE06","DFT+U","O4H2 cis configuration","short-range order","SrTiO3-xHx"],"falsifier":"Recalculate the relative energies of the cis pair, trans pair, and isolated hydride at the dilute composition (e.g., Sr8Ti8O23H) with a method independent of semilocal and hybrid DFT self-interaction corrections, such as diffusion Monte Carlo or an embedded wavefunction approach. If the cis-trans energy difference drops below about 100 meV/H, or if the cis pair is not the lowest-energy local arrangement, the central ordering claim is undermined. On the experimental side, neutron pair distribution analysis of a sample near x = 0.6 that shows no excess of cis-like O4H2 octahedra would contradict","tokens_in":16886,"feed_emoji":"⚛️","tokens_out":4792,"duration_ms":47616,"temperature":0.7,"pith_summary":"This paper claims that hydride ions in the perovskite oxyhydride SrTiO3−xHx do not spread out randomly or stay isolated; instead, pairs of hydrides preferentially occupy first-nearest-neighbor anion sites around a titanium center, the O4H2 cis configuration. The preference is large—roughly 200 meV per hydride—and it persists even at dilute hydrogen content and in disordered high-temperature states. The authors also show that ordering energies vary by about 100 meV/anion depending on the DFT+U correction, and they identify U = 3.3 eV as a low-cost proxy that reproduces HSE06 hybrid-functional ordering energetics well enough to train a cluster expansion. If the cis preference is correct, models that assume isolated hydrides in supercells misrepresent the locally relevant configurations for diffusion and other properties.","feed_headline":"Hydride pairs prefer adjacent sites in SrTiO3–xHx","feed_subtitle":"Hybrid DFT finds a ~200 meV per hydride stabilization for the O4H2 cis motif, reshaping diffusion models.","key_machinery":"The key machinery is a cluster expansion (CE), a lattice Hamiltonian that writes the configurational energy of the O/H sublattice as a sum of interaction coefficients over clusters of anion sites, trained on HSE06 hybrid-functional energies. Because HSE06 is computationally expensive, the authors first demonstrate that PBE+U ordering energies swing by about 100 meV/anion as U varies, then select U = 3.3 eV as a proxy that best reproduces HSE06 ordering stability according to both absolute and pairwise deviation metrics. This proxy-U is used for active-learning structure selection, and a representative subset of 68 configurations is recalculated with HSE06 to train the final CE. The resulting","core_discovery":"The central claim is that SrTiO3−xHx shows a consistent, composition-wide tendency for two hydride ions to occupy first-nearest-neighbor anion sites around the same Ti—the O4H2 cis octahedral configuration—rather than the trans arrangement or isolated hydrides. Evidence comes from three independent routes: the ground-state ordered structures all share a corner-sharing 'cis-stair' pattern; a tercile analysis shows low-energy configurations are dominated by high fractions of O4H2 cis octahedra; and Monte Carlo sampling of the cluster-expansion Hamiltonians at 350 K shows short-range order in disordered states, with O4H2 cis populations well above the random limit. At the dilute composition Sr8","pith_inferences":["The same cis-pair mechanism plausibly operates in BaTiO3−xHx and CaTiO3−xHx, whose prior simulation studies also assumed isolated hydrides; this is an extension beyond the paper's explicit claim.","The proxy-U strategy—fitting a cheap DFT+U to hybrid-functional ordering energetics for structure selection before a smaller set of expensive hybrid calculations—could transfer to other systems where electron over-delocalization corrupts configurational energetics, though the fitted U is a numerical proxy rather than a physically universal Hubbard parameter.","A concrete experimental test would be neutron total scattering or pair distribution function analysis of SrTiO2.4H0.6 looking for an excess of corner-sharing O4H2 cis octahedra relative to a random anion model; such data could corroborate or refute the predicted short-range order.","Because the cis preference rests on HSE06 as the reference, a future recalculation of the 68 key configurations with a method beyond hybrid DFT would quantify how much of the preference is functional-dependent."],"forward_implications":["Dilute-limit supercells containing a single isolated hydride miss a stabilization on the order of 200 meV/H from forming nearest-neighbor cis pairs, so computed energetics for hydride-related processes should be re-examined.","Hydride migration may not be describable as a single ion hopping between equivalent sites: moving a hydride out of a stable cis pair changes the local ordering energy, potentially making diffusion cooperative or correlated.","Short-range order persists in the disordered state at 350 K, so even nominally disordered SrTiO3−xHx is not a random anion mixture; local octahedral statistics deviate clearly from the random limit.","DFT+U studies of anion ordering in reduced titanium perovskites should report U sensitivity, since the identity and stability ordering of ground states shift by more than 100 meV/anion across U = 0 to 6 eV."],"fun_headline_variants":["Cis hydride pairing dominates SrTiO3−xHx ordering","Hydride pairs stabilize SrTiO3−xHx by 200 meV","Short-range hydride order in SrTiO3−xHx"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The paper's load-bearing premise is that HSE06 hybrid-DFT correctly captures the ordering energetics of SrTiO3−xHx; the authors explicitly state there is currently no direct evidence that hybrid functionals yield the most accurate ordering stability in these oxyhydrides, and both the proxy-U calibration and the headline cis-preference claim lean on HSE06 as the reference.","fun_headline_variants_meta":{"raw":{"variants":["Cis hydride pairing dominates SrTiO3−xHx ordering","Hydride pairs stabilize SrTiO3−xHx by 200 meV","Short-range hydride order in SrTiO3−xHx"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000512,"raw_usage":{"total_tokens":2387,"prompt_tokens":867,"completion_tokens":1520,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":611,"completion_tokens_details":{"reasoning_tokens":1457}},"tokens_in":611,"tokens_out":1520,"duration_ms":12489,"temperature":1.0,"reasoning_tokens":1457,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T09:05:51.288202+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recalculate the relative energies of the cis pair, trans pair, and isolated hydride at the dilute composition (e.g., Sr8Ti8O23H) with a method independent of semilocal and hybrid DFT self-interaction corrections, such as diffusion Monte Carlo or an embedded wavefunction approach. If the cis-trans energy difference drops below about 100 meV/H, or if the cis pair is not the lowest-energy local arrangement, the central ordering claim is undermined. On the experimental side, neutron pair distribution analysis of a sample near x = 0.6 that shows no excess of cis-like O4H2 octahedra would contradict","supporting_citations":[],"review_version":1}