{"id":"2b445aed-b1db-40e0-8cff-d3ffef18d322","arxiv_id":"2412.08126","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Simulations identify reconstructed BiVO4(010) surfaces with exposed, under-coordinated Bi atoms that spontaneously dissociate water, providing a mechanistic explanation for experimentally observed bismuth-rich reactive surfaces.","lead":"This paper uses machine-learned interatomic potentials and density functional theory to map out 494 possible surface structures of the solar water-splitting material BiVO4, then predicts that some reconstructed, bismuth-rich surfaces spontaneously split water. It offers concrete atomic models for the bismuth-rich surfaces seen in experiments and points to under-coordinated bismuth atoms as the active sites.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central dissociation claim rests on one 7.5 ps PBE0-MD run per surface started from a water slab equilibrated for the stoichiometric surface; this is too short and too initialization-dependent to establish equilibrium dissociation percentages.","rationale":"The reader's weakest-assumption analysis points to the same place I would: the water dissociation result is derived from a single 7.5 ps PBE0-MD run per surface, with the water slab initialized from a prior stoichiometric-surface study and the temperature raised to 350 K. This is the most load-bearing part of the paper because the paper's headline contribution is the 'first theoretical report' of spontaneous water dissociation on reconstructed Bi-rich BiVO4(010) and the attribution of that reactivity to bare, under-coordinated Bi sites. Everything else—the MLIP search, the Pourbaix diagram, the structural characterization—is input that selects the surfaces, but it is the MD that directly supports the mechanism. The lack of multiple independent initial conditions and the short trajectory mean the reported dissociation fractions and the site-specific mechanism are not yet shown to be converged properties of the aqueous interface. I do not regard this as a reason to reject the paper: the reconstructed surfaces contain genuinely under-coordinated Bi sites, and the observed proton-transfer events are mechanistically plausible. It is a reason to keep the verdict at CONDITIONAL rather than ACCEPT, and to require the proposed replication test before the 'significant spontaneous dissociation' claim is treated as established. I agree with the reader's choice of weakest assumption; the p(1x1) search-cell limitation is a secondary generality concern, and the lack of released training data is a reproducibility concern, not a direct threat to the dissociation mechanism.","tokens_in":13724,"tokens_out":8321,"duration_ms":84788,"concrete_test":"Replicate the t-BiO2 and t-BiVO4(010) interface simulations with two additional 10 ps PBE0 AIMD runs per surface, using water initializations generated by 5 ps classical NVT equilibration (different random seeds) of the same water slab around each reconstructed surface. Compute the dissociated-water fraction over the final 5 ps of each run; also extend one t-BiO2 run to 20 ps. If the fraction varies by more than 15 percentage points between initializations, or if t-BiO2 does not reproducibly exceed the stoichiometric t-BiVO4 fraction, the single-trajectory 'significant spontaneous dissociation' claim is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim—that reconstructed Bi-rich BiVO4(010) surfaces exhibit significant spontaneous water dissociation driven by bare, under-coordinated Bi sites—is quantified from Figure 3 using one 7.5 ps Born-Oppenheimer PBE0 MD trajectory per surface (Methods paragraph: 'The simulations were run for 7.5 ps with a time step of 0.5 fs... with the temperature set at 350 K'). The water slab is not re-equilibrated for each reconstructed surface; the same initial water configuration from a previous stoichiometric BiVO4(010)-water study is used (Ref. 39). This matters for three reasons. First, the exposed under-coordinated Bi atoms are placed in direct contact with a water arrangement optimized for a different surface, so early proton-transfer events in the first 1-2 ps may reflect transient relaxation rather than equilibrium spontaneous dissociation. Second, 7.5 ps is comparable to or shorter than water reorientation and proton-transfer timescales at 350 K, and Figure 3a shows dissociation fractions still rising or fluctuating at the end of the run, so the reported maxima (75% on Bi-rich surfaces) are not converged properties. Third, there are no independent replicates or error bars, and no sensitivity test to initial water positions. If the initial contact geometry is what triggers the first dissociation events, the mechanistic attribution to bare Bi sites is weakened even though those sites are likely reactive. This is the most direct threat to the central claim; the p(1x1) search-cell constraint is a secondary generality concern.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a computational workflow for exploring reconstructed BiVO4(010) surfaces and their aqueous interfaces. A Gaussian Approximation Potential (GAP) MLIP, iteratively trained via active learning, is combined with simulated annealing in p(1x1) cells to generate 494 unique surface configurations across 13 stoichiometries. These are screened with PBE and PBE0 (alpha=0.22) to construct a surface Pourbaix diagram, identifying Bi-rich (t-BiO2, t-BiO3) and V-rich (t-VO2, t-VO4) structures as electrochemically stable under relevant conditions. Born-Oppenheimer PBE0-MD simulations (7.5 ps, 350 K) of five aqueous interfaces show no dissociation on the stoichiometric surface but significant water dissociation on the reconstructed surfaces, with the largest fractions attributed to bare, under-coordinated Bi atoms on t-BiO2/t-BiO3. The authors claim this is the first theoretical observation of spontaneous water dissociation on BiVO4(010).","tokens_in":14153,"tokens_out":7612,"duration_ms":64086,"significance":"If the dissociation result is robust, the paper makes a valuable contribution: it proposes concrete structural models for experimentally observed Bi-rich BiVO4 surfaces, identifies a possible role for under-coordinated Bi sites in surface hydration, and demonstrates an active-learning MLIP workflow for multicomponent oxide surfaces with modest DFT cost (355 DFT calculations). The workflow is described in enough detail to be reproduced, and the training data are promised on Materials Cloud. However, the central dissociation claim currently rests on short, unreplicated MD simulations with a non-equilibrated initial water configuration, so the quantitative conclusions and the mechanistic attribution require additional validation.","major_comments":[{"comment":"The central claim of significant spontaneous water dissociation on reconstructed Bi-rich surfaces rests on a single 7.5 ps PBE0-MD trajectory per surface, with the initial water configuration taken from a previous stoichiometric BiVO4(010)-water study (Ref. 39) and not re-equilibrated for each reconstructed surface. Figure 3a shows the dissociation fraction still rising or fluctuating at the end of the 7.5 ps runs, so the reported 'maximum dissociation percentage of 75%' is a transient maximum rather than a converged steady-state value. With no independent replicates or sensitivity tests to the initial water arrangement, the mechanistic attribution of dissociation to bare, under-coordinated Bi sites is not yet quantitatively established. The authors should provide longer trajectories, multiple starting water configurations, and time-averaged dissociation fractions with error estimates.","section":"Aqueous BiVO4(010) Interfaces and their Dynamical Properties; Methods (DFT Calculations)"},{"comment":"The structure search is confined to the p(1x1) surface unit cell; the 494 unique structures are therefore only unique within this cell. As reconstructed surfaces with longer periodicity are common in multicomponent oxides, the claim that the workflow 'identified 494 unique reconstructed surface structures' should be explicitly qualified as p(1x1)-limited. The later p(2x2) re-optimization of selected structures is a useful check, but it does not recover structures that would only emerge from a genuinely p(2x2) or larger-cell global search. This limitation should be stated in the abstract and conclusions.","section":"Global Optimization for Reconstructed BiVO4(010) Sampling"}],"minor_comments":[{"comment":"The number of distinct aqueous interfaces is inconsistent: the Methods states 'five different aqueous BiVO4(010)-water interfaces', while the Results discusses six (t-BiVO4, t-BiVO6, t-BiO2, t-BiO3, t-VO2, t-VO4). Please clarify.","section":"Methods (DFT Calculations)"},{"comment":"The phrase 'spontaneous water dissociation, at the early stages of the PBE0-MD simulations' contains an awkward comma after 'dissociation'; more importantly, 'early stages' should be reconciled with the claim that the observed dissociation is an equilibrium property.","section":"Results and Discussion, Aqueous BiVO4(010) Interfaces and their Dynamical Properties"},{"comment":"The caption does not define the 'percentage of dissociated water molecules' (it is defined in the text as the number of dissociated water molecules divided by the total number of outermost metal atoms) nor the meaning of 'spikes' relative to 'steps'. Adding this information to the caption would improve readability.","section":"Figure 3 caption"},{"comment":"The claim that this is 'the first theoretical report' of spontaneous water dissociation on BiVO4(010) should be supported by a more thorough literature search or softened to avoid an unsupported novelty assertion.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The paper's main novelty—spontaneous water dissociation on reconstructed Bi-rich BiVO4(010)—is interesting but currently rests on one short trajectory per surface without equilibration. The p(1x1) limitation of the structure search should also be clearly stated. The work is worth revising rather than rejecting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nShort version: this is a serious computational paper with a genuinely new workflow and a central claim that is plausible but not fully substantiated. The active-learning GAP global optimization on BiVO4(010) turns up 494 reconstructed candidates, and the Pourbaix screening narrows them to six electrochemically stable surfaces, including Bi-rich surfaces exposing bare Bi. The first theoretical report of spontaneous water dissociation on these reconstructed surfaces is an appealing and mechanistically interesting result.\n\nWhat is genuinely new: the integration of an existing active-learning GAP framework with simulated annealing for a multicomponent oxide surface, and the specific Bi-rich/V-rich structural models that go beyond bulk-truncated or chemical-intuition surfaces. The two-step DFT screening (GGA then dielectric-dependent PBE0) for the Pourbaix diagram is sensible, and the structural analysis of O2-like dimers and distorted BiOn polyhedra is careful. The AIMD analysis, while limited, consistently points to Bi as the adsorption site and to under-coordinated Bi as the reactive center.\n\nThe main soft spot is the MD sampling. Figure 3's dissociation percentages come from one 7.5 ps Born-Oppenheimer PBE0 trajectory per surface, initialized from a water configuration equilibrated for a different (stoichiometric) surface and run at 350 K. The dissociation fractions are still rising or fluctuating at the end of the runs, so the maxima are not converged. There are no replicates or error bars. That means the quantitative claim of 75% dissociation and the 'spontaneous' label are stronger than the evidence supports; the actual point, that bare Bi sites promote water dissociation, is likely correct but needs longer or repeated simulations with properly re-equilibrated water.\n\nThe search confined to p(1x1) cells is a secondary generality concern, though they do extend selected structures to p(2x2) for re-optimization. They also promise training data on Materials Cloud, which would help, but it is not yet available.\n\nOverall: I think the central argument is credible but the quantitative claims are not fully substantiated. This deserves a serious referee and likely a major revision, not a desk reject. I would bring it to a reading group to discuss the strengths of the structure-search pipeline and the pitfalls of short AIMD trajectories.","headline":"Valuable MLIP-driven surface structure search and Pourbaix analysis; the water dissociation claim is plausible but the quantified percentages rest on one short, un-replicated MD run per surface.","tokens_in":14570,"tokens_out":3043,"would_cite":true,"duration_ms":30801,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["68.08.-p","71.15.Mb","82.65.+r"],"model":"deepseek-v4-flash","headline":"The paper argues that reconstructed Bi-rich BiVO4(010) surfaces spontaneously dissociate water, with bare under-coordinated Bi sites as the active centers.","keywords":["BiVO4","surface reconstruction","machine-learning interatomic potential","active learning","simulated annealing","Pourbaix diagram","ab initio molecular dynamics","water dissociation"],"falsifier":"Repeating the hybrid-functional molecular dynamics on the t-BiO2 surface with several independent water starting configurations and trajectories longer than 20 picoseconds would test the claim: if the dissociation fraction falls to the near-zero level of the stoichiometric surface, the spontaneous-dissociation mechanism would be falsified. On the experimental side, operando surface-specific vibrational spectroscopy on a Bi-rich BiVO4(010) electrode under anodic conditions should detect surface hydroxyl signals if the predicted dissociation occurs.","tokens_in":13524,"feed_emoji":"💧","tokens_out":8604,"duration_ms":78528,"temperature":0.7,"pith_summary":"The paper aims to show that the surface that actually does the work in a bismuth vanadate photoanode is not the clean, bulk-truncated one that most models assume. Using a machine-learned interatomic potential driven by active learning and simulated annealing, the authors searched the surface reconstruction space of BiVO4(010) and found 494 distinct structures, then used hybrid-functional thermodynamics to pick the ones stable under photoelectrochemical conditions. When they ran hybrid-functional molecular dynamics with explicit water on those reconstructed surfaces, they observed spontaneous water dissociation on the Bi-rich and V-rich surfaces—up to 75 percent of surface metal atoms carrying dissociated water on Bi-rich models—while the stoichiometric surface stayed intact. The paper argues that bare, under-coordinated bismuth sites, which only exist in the reconstructed surfaces, are the active centers for this dissociation. If true, this reframes how the aqueous BiVO4 interface should be modeled and suggests that surface reconstruction is essential for surface hydration and reactivity.","feed_headline":"Bi-rich BiVO4 surfaces split water spontaneously","feed_subtitle":"Reconstructed surfaces with bare bismuth sites dissociate water, unlike the textbook stoichiometric surface.","key_machinery":"The load-bearing element is a multi-stage computational workflow rather than a single identity. First, an active-learning-trained Gaussian Approximation Potential (a machine-learned surrogate of the density-functional potential-energy surface) drives simulated-annealing searches that rearrange surface and near-surface atoms, producing 494 unique reconstructed BiVO4(010) structures across Bi-rich, V-rich, and mixed stoichiometries. Second, a two-step screening with PBE and a dielectric-dependent PBE0 hybrid functional builds a surface Pourbaix diagram that selects electrochemically stable structures under Bi- and V-rich electrolyte conditions. Finally, Born-Oppenheimer molecular dynamics with the same hybrid functional and explicit water at 350 K for 7.5 ps probes the aqueous interface. The mechanistic heart is the bare, under-coordinated Bi site: water binds there and dissociates via an indirect pathway through a transient hydronium ion or a direct proton transfer to a bridging oxygen, forming surface hydroxyls.","core_discovery":"The paper's central claim is that reconstructed Bi-rich BiVO4(010) surfaces, and to a lesser extent V-rich ones, spontaneously dissociate water at the aqueous interface, whereas the stoichiometric surface does not. The authors report this as the first theoretical observation of spontaneous water dissociation on BiVO4. On the t-BiO2 and t-BiO3 surfaces, water adsorbed at a bare, under-coordinated Bi site transfers a proton to a neighboring water molecule or directly to a bridging oxygen, yielding surface hydroxyls; the maximum dissociation percentage reaches about 75 percent on Bi-rich surfaces. This behavior is attributed to the low-coordinated Bi sites exposed only by reconstruction, which act as the active centers for dissociation, and it is presented as a mechanistic explanation for the enhanced hydration of experimentally observed Bi-rich surfaces under photoelectrochemical anodic conditions.","pith_inferences":["The dissociation percentages come from single 7.5 ps trajectories initialized from one water configuration; longer runs or multiple starting configurations could shift quantitative fractions, though the qualitative contrast between reconstructed and stoichiometric surfaces may persist.","If the bare-Bi mechanism holds, then the anodic vanadium dissolution that creates Bi-rich overlayers would actually enhance water activation, implying that photoelectrochemical activity and surface stability are dynamically coupled rather than fixed properties.","The short O-O motifs observed on oxidized reconstructed surfaces resemble dioxygen-like species; connecting these to oxygen-evolution-reaction intermediates would be a natural next step, but the paper does not make that claim.","An applied-electrode-potential treatment (for example, grand canonical molecular dynamics) could test whether the dissociation behavior persists under explicit bias, which the present simulations at the water interface do not include."],"forward_implications":["Experimentally observed Bi-rich BiVO4 surfaces under photoelectrochemical anodic conditions likely correspond to the t-BiO2- and t-BiO3-type reconstructions identified here, which are far more hydrated than the stoichiometric surface.","Surface reconstruction, not the bulk-truncated termination, should be the starting model for aqueous BiVO4 interfaces in studies of stability and reactivity.","Bare, low-coordinated Bi sites act as the active centers for water dissociation; V sites play a secondary role on reconstructed surfaces.","The active-learning-plus-Pourbaix workflow can be transferred to other multicomponent oxides to predict stable reconstructed surfaces and their interfacial reactivity.","The predicted surface structures offer concrete targets for experimental validation with operando or ex situ surface characterization."],"supporting_citations":[{"why":"Operando infrared spectroscopy revealing that BiVO4 surfaces reconstruct during the PEC cycle; motivates searching beyond bulk-truncated models.","marker":"3"},{"why":"Experimental demonstration that surface Bi/V ratio changes interfacial energetics and photoelectrochemical properties; provides the Bi-rich surface context and earlier intuition-based models.","marker":"4"},{"why":"The data-efficient iterative active-learning protocol for training Gaussian Approximation Potentials that the authors extend to BiVO4 surface exploration.","marker":"19"},{"why":"Supplies the dielectric-dependent PBE0 hybrid functional setup (alpha = 0.22) used for stability screening and molecular dynamics.","marker":"32"},{"why":"Provides the initial water configuration used in the aqueous interface simulations and a prior theoretical model of the stoichiometric BiVO4-water interface.","marker":"39"},{"why":"Introduces the Gaussian Approximation Potential framework used as the low-cost surrogate potential-energy surface.","marker":"42"}],"fun_headline_variants":["Bi-rich BiVO4 surfaces spontaneously dissociate water","Machine learning maps BiVO4 reconstructions that split water","Reconstructed Bi-rich BiVO4 surfaces trigger water dissociation","First theory: BiVO4 Bi-rich surfaces dissociate water","Under-coordinated Bi sites make BiVO4 dissociate water"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire mechanistic conclusion rests on the assumption that a single 7.5-picosecond simulation, started from a water configuration built for the stoichiometric surface and run at 350 K, represents the equilibrium aqueous interface on each reconstructed surface.","fun_headline_variants_meta":{"raw":{"variants":["Bi-rich BiVO4 surfaces spontaneously dissociate water","Machine learning maps BiVO4 reconstructions that split water","Reconstructed Bi-rich BiVO4 surfaces trigger water dissociation","First theory: BiVO4 Bi-rich surfaces dissociate water","Under-coordinated Bi sites make BiVO4 dissociate water"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000795,"raw_usage":{"total_tokens":3516,"prompt_tokens":975,"completion_tokens":2541,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":591,"completion_tokens_details":{"reasoning_tokens":2459}},"tokens_in":591,"tokens_out":2541,"duration_ms":17892,"temperature":1.0,"reasoning_tokens":2459,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:10:43.780637+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeating the hybrid-functional molecular dynamics on the t-BiO2 surface with several independent water starting configurations and trajectories longer than 20 picoseconds would test the claim: if the dissociation fraction falls to the near-zero level of the stoichiometric surface, the spontaneous-dissociation mechanism would be falsified. On the experimental side, operando surface-specific vibrational spectroscopy on a Bi-rich BiVO4(010) electrode under anodic conditions should detect surface hydroxyl signals if the predicted dissociation occurs.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Operando infrared spectroscopy revealing that BiVO4 surfaces reconstruct during the PEC cycle; motivates searching beyond bulk-truncated models."},{"cited_title":"The Impact of Surface Composition on the Interfacial Energetics and Photoelectrochemical Properties of BiVO _4","cited_arxiv_id":null,"evidence_quote":"Experimental demonstration that surface Bi/V ratio changes interfacial energetics and photoelectrochemical properties; provides the Bi-rich surface context and earlier intuition-based models."},{"cited_title":"G.; Margraf, J","cited_arxiv_id":null,"evidence_quote":"The data-efficient iterative active-learning protocol for training Gaussian Approximation Potentials that the authors extend to BiVO4 surface exploration."},{"cited_title":"Comprehensive modeling of the band gap and absorption spectrum of BiVO _ 4","cited_arxiv_id":null,"evidence_quote":"Supplies the dielectric-dependent PBE0 hybrid functional setup (alpha = 0.22) used for stability screening and molecular dynamics."},{"cited_title":"Electron and Hole Polarons at the BiVO _ 4 --Water Interface","cited_arxiv_id":null,"evidence_quote":"Provides the initial water configuration used in the aqueous interface simulations and a prior theoretical model of the stoichiometric BiVO4-water interface."},{"cited_title":"P.; Payne, M","cited_arxiv_id":null,"evidence_quote":"Introduces the Gaussian Approximation Potential framework used as the low-cost surrogate potential-energy surface."}],"review_version":1}