{"id":"c7d11eb2-3214-4814-9db1-e43726f97981","arxiv_id":"2411.16208","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Density functional theory with extensive relaxation finds oxygen-covered gold surface structures with work function changes below 1.1 eV, resolving a long-standing 3 eV discrepancy with earlier DFT.","lead":"Oxygen on gold changes the metal's work function by less than one volt, and new computer simulations explain why earlier calculations wrongly predicted three volts. The key is that oxygen atoms drag gold atoms into new arrangements and form molecule-like pairs, keeping the work function low and matching experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Δφ comparison for Au(110)/(100) uses unreconstructed clean surfaces even though the experimental reference is Au-(1×2); missing-row reconstructions cannot emerge from ideal-slab relaxations, so the 'most stable' assignments are incomplete.","rationale":"The reader's conditional verdict identifies the finite hand-selected initial-structure set as the weakest assumption. My stress test agrees that the global-minimum claim is the load-bearing point, but sharpens it: the paper excludes experimentally known clean-surface reconstructions, so the Δφ reference itself is suspect for Au(110)/(100). This is a concrete omission rather than a general worry about global optimization, and it is directly testable. A positive result of the proposed test (missing-row surface lower in energy and different φ) would reinforce the need for revision; a null result would support the paper. The paper deserves credit for symmetric thick slabs, 30 Å vacuum, force convergence to 1 meV/Å, and HSEsol cross-checks, and the Au(111) resolution of the previous Stampfl discrepancy is plausible. However, the word 'most stable' is used more strongly than the evidence supports: local minima within 0.2 eV and structure sensitivity to initial positions are reported in Sec. IIID2, and no global structure search is performed. My concern does not change the reader's verdict; both point to the same conditional status. Therefore the verdict remains CONDITIONAL.","tokens_in":19864,"tokens_out":8168,"duration_ms":83467,"concrete_test":"Compute clean Au(110) in a p(1×2) missing-row cell with the paper's PBEsol/O_s settings, 30 Å vacuum, full relaxation to <1 meV/Å, and compare surface formation energy and φ with the unreconstructed (1×1) slab. If the missing-row surface is more stable and its φ differs by ≳0.2 eV, re-relax the 0.5 and 1 ML O-covered Au(110) structures from missing-row-like starting geometries and recompute Δφ; if Δφ leaves the <1.1 eV/experimental window, the all-surface claim needs revision. As an independent check of the Au(111) branch, run a small unbiased search (e.g., ~100 random/AIRSS 2×2 cells with 4 O per cell) and compare the lowest E_Au-O and Δφ with Fig. 7(a).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the DFT most-stable O-covered Au(100), Au(110), and Au(111) structures reproduce the experimental Δφ < 1.1 eV at ≤1 ML. This rests on identifying true ground states by fully relaxing hand-selected initial geometries (Sec. IIA2), with no global optimization. A concrete failure mode is visible in Sec. IIIB: for defect-free Au surfaces, 'surface reconstructions of Au surfaces were not observed when defects or O-atoms were absent.' For Au(110), however, the experimental comparison [23] is explicitly on the reconstructed Au-(1×2) surface, and a missing-row reconstruction cannot be generated from a bulk-terminated slab by relaxation because it requires removing half of the first-layer rows. The same holds for the quasihexagonal reconstruction of Au(100). The theoretical Δφ baseline is therefore not the experimentally relevant clean surface, and the 0.5–1 ML O-covered ground states may also be missed because no missing-row-like initial structures were included. Since the paper itself reports many local minima within 0.2 eV (Sec. IIID2), the finite structure set is not demonstrably complete; the reconstructed-clean-surface omission is the sharpest concrete instance. If the reconstructed clean reference has a different φ, or O adsorption couples to lifting the reconstruction, the claimed agreement for (110)/(100) is not established. The Au(111) resolution of the ~3 eV discrepancy additionally assumes that no lower-energy 1 ML structure with larger Δφ was overlooked by the hand-built search.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper combines Kelvin-probe measurements on polycrystalline Au films with DFT (PBEsol and HSEsol) calculations for O-covered Au(100), Au(110), and Au(111) surfaces. It reports that the most stable calculated structures give Δφ < 1.1 eV at ≤ 1 ML and an almost constant Δφ above 1 ML, in agreement with experiments by Saliba et al. and Gottfried et al., and it argues that this resolves a discrepancy with Stampfl's earlier DFT value of about 3 eV on Au(111). The authors attribute the new agreement to O-induced displacements of Au atoms and to 'molecule-like' O arrangements at high coverage, and they argue that subsurface-O structures are numerous and disordered, explaining the loss of long-range order at room temperature.","tokens_in":20183,"tokens_out":7873,"duration_ms":121863,"significance":"If the structural assignments are correct, the paper resolves a long-standing quantitative discrepancy between DFT and experimental work functions for oxygen-covered Au, and it provides a physically plausible mechanism for high-coverage oxygen chemisorption. The computational protocol is careful in several respects: force convergence below 1 meV/Å, thick symmetric slabs, vacuum-width tests (20 vs 30 Å), two different PAW potentials for oxygen, and a hybrid-functional check. The experimental part provides an independent Δφ measurement on polycrystalline films. The main limitation is that the claim of 'most stable structures' is based on a finite set of hand-selected initial geometries, and the hybrid-functional check is performed on smaller, higher-symmetry cells than those of the proposed ground states. These issues make the central conclusion plausible but not yet fully established.","major_comments":[{"comment":"The central claim that the calculated 'most stable structures' reproduce the experimental Δφ rests on a finite, hand-selected set of initial geometries, and the paper itself documents a dense set of local minima (e.g., Sec. IIID2, with structures within 0.2 eV of the most stable). For Au(110) and Au(100), the clean-surface reference is an unreconstructed bulk-terminated slab; Sec. IIIB states that surface reconstructions were not observed in the defect-free calculations. The experimental comparison [23] for Au(110) is, however, on the missing-row reconstructed Au(110)-(1×2) surface, which cannot be generated from a bulk-terminated slab by relaxation, and a similar issue exists for the quasihexagonal reconstruction of Au(100). Consequently, the Δφ baseline for these faces, and any O-covered ground states that involve lifting or modifying the reconstruction, are not explored. This is a load-bearing gap for the (110) and (100) parts of the central claim.","section":"Secs. IIA2, IIIB, IIIC1"},{"comment":"The hybrid-functional confirmation is performed on 1×1×10 supercells containing a single O atom per surface cell, i.e., high-symmetry configurations. The most stable structures identified in the larger 2×2, 3×3, and 4×4 cells involve pairs of O atoms with molecule-like O–O distances (Figs. 9(a)–9(c)), which cannot be represented in a 1×1 cell. Therefore the abstract's statement that the agreement is 'confirmed using hybrid functional' is not supported for the actual ground-state structures; the HSEsol check applies to a different structural family. The authors should either perform HSEsol on the relevant low-symmetry ground-state structures or qualify the claim accordingly.","section":"Sec. IIIC2 and Fig. 8"},{"comment":"The inference that the present experiments correspond to approximately 1 ML coverage is partly circular. The O2-to-O conversion efficiency is assumed to be 0.1%, yielding an integrated flux of 6.5×10^17 cm^-2 (which would correspond to 540 ML if all atoms stuck), and the coverage is then 'considered close to 1 ML' because the measured Δφ of 0.75 eV and the annealed φ of 4.95 eV match literature values. This inferred coverage is later used as a supporting experimental anchor for the DFT comparison, so the argument is not independent. A sensitivity analysis of the assumed conversion efficiency, or an independent coverage calibration, would strengthen the experimental claim.","section":"Secs. IIA1 and IIIA"}],"minor_comments":[{"comment":"The manuscript uses 'f' and 'df' as substitutes for φ and Δφ in several places; the notation should be made consistent throughout.","section":"Abstract and throughout"},{"comment":"Table 1 contains a formatting artifact in the (100) row ('4.94.91') and the column headers PBEa, PBEb, etc. are difficult to map to the footnoted references; please reformat the table for clarity.","section":"Table 1"},{"comment":"Fig. 7(a) is very dense, and the correspondence between data points, structures, and the sidebar labels ('a'–'g') is hard to follow; consider numbering the panels or enlarging the structure insets.","section":"Fig. 7"},{"comment":"The phrase 'the former is considered chemisorption' in the abstract is ambiguous because the antecedent could be the molecule-like O arrangements; please clarify the distinction between surface and subsurface O atoms.","section":"Abstract and Sec. IIID2"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely to be of interest to the surface-science and electrocatalysis communities. The main risk is overclaiming ground-state status from a finite structure search; I would encourage the editor to ask for a more systematic search (or at least an explicit statement of the limitation and a test of reconstructed clean surfaces) before publication. The manuscript's scope is appropriate for a physical chemistry / surface science journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper gives the most convincing explanation I've seen for the ~3 eV discrepancy between DFT and experiment for the work-function change of oxygen-covered Au(111). The key move is that at high coverage the most stable structures are not high-symmetry sites; O atoms form pairs and Au atoms are displaced, bringing Δφ below 1.1 eV. The authors show that if you restrict to the high-symmetry structures used by Shi and Stampfl, you reproduce their ~3 eV result, which supports the interpretation. The calculations are careful: force convergence to 1 meV/Å, 30 Å vacuum, two O PAW potentials, PBEsol and HSEsol checks, and layer-thickness tests. That is real work.\n\nThe main soft spot, and it is genuine, is the claim that these are the 'most stable' structures. The search is hand-built rather than a global optimization, and the authors report many local minima within 0.2 eV. For Au(111) this may not matter much because the discrepancy with the old DFT is so large and the new structures are physically reasonable. But for Au(110) and Au(100) there is a sharper problem the stress-test note rightly identifies: the experimental comparisons are to reconstructed surfaces (Au(110)-(1×2), and implicitly the quasihexagonal Au(100)), while the DFT uses unreconstructed slabs. A missing-row reconstruction cannot be generated by relaxation from an ideal slab, and the authors explicitly state that no reconstruction appeared in defect-free calculations. That means both the clean-surface baseline and possibly the O-covered ground states are missing a class of structures. If the reconstructed clean surface has a different work function, the claimed agreement for (110)/(100) is not established. The authors should include missing-row-type initial structures or soften the comparison for those faces.\n\nA lesser concern is their own experimental coverage estimate in KPFM; it relies on an assumed 0.1% O2-to-O conversion efficiency and gives a ~500 ML upper bound, not a reliable 1 ML anchor. But this is not central because the primary comparison is to the published Saliba and Gottfried data.\n\nOverall, the Au(111) resolution is likely right and important. The paper deserves serious peer review, but the 'most stable' language should be tempered and the reconstructed-surface issue for (110)/(100) should be addressed before the central claim is taken as definitive. I would send it to review with a request for a more systematic structure search or at least missing-row initial geometries.","headline":"Plausible resolution of the Au(111) work-function discrepancy, but 'most stable' overreaches the search; the reconstructed-surface problem for (110)/(100) is a real gap.","tokens_in":20757,"tokens_out":5007,"would_cite":true,"duration_ms":137025,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["73.30.+y","71.15.Mb","68.43.-h"],"model":"deepseek-v4-flash","headline":"Oxygen-covered gold surfaces shift the work function by about 1 eV, not 3 eV.","keywords":["work function","oxygen adsorption","gold surfaces","density functional theory","surface relaxation","Au(111)","chemisorption","hybrid functional"],"falsifier":"An unbiased structure search (for example, random or evolutionary generation of many more starting oxygen positions, including deeper subsurface sites) that finds a 1 ML Au(111) structure with formation energy lower than the paper's most stable one and a computed work-function shift near 3 eV would refute the central claim, since the paper's method would have missed the true ground state.","tokens_in":19700,"feed_emoji":"⚛️","tokens_out":11646,"duration_ms":92273,"temperature":0.7,"pith_summary":"This paper argues that oxygen-covered gold surfaces change the work function—the energy needed to pull an electron out of the surface—by less than about 1 eV up to one monolayer of oxygen, and that the earlier density functional theory value of about 3 eV came from relaxing only high-symmetry structures. When all atomic positions are relaxed from many starting arrangements, the stable low-symmetry structures keep the work-function shift $\\Delta\\phi$ below 1.1 eV at $\\le$1 ML and roughly constant above 1 ML. These structures show oxygen-induced displacements of gold atoms and molecule-like oxygen pairs at high coverage, and the computed shifts match experiments on the (100), (110), and (111) surfaces, including hybrid-functional checks. This matters because the work function is a reference quantity for electronics and catalysis, and the result reconciles theory with the observed stability of oxygen chemisorption on gold.","feed_headline":"Oxygen on gold shifts the work function by ~1 eV, not ~3 eV","feed_subtitle":"Fully relaxed surface structures match experiment and overturn a long-standing density functional theory prediction.","key_machinery":"The load-bearing mechanism is the structure search: initial supercells that include oxygen at low-symmetry and subsurface sites, followed by unrestricted relaxation of all atomic positions until every force is below 1 meV/Å, in slabs thick enough to let even fourth-layer gold atoms move. The work function is used as the decisive observable because it is sensitive to surface atomic arrangement and can be compared with experiment even without long-range order. The newly found stable structures show oxygen-induced gold displacements, 2×1 and 2×2 reconstructions, and molecule-like O–O pairs that are chemisorbed at 1 ML but nearly molecular-adsorbed at 2 ML.","core_discovery":"The central claim is that the experimentally observed work-function change ($\\Delta\\phi$) below 1 eV for oxygen-covered Au(100), Au(110), and Au(111) surfaces at up to one monolayer is the true ground-state behavior, while the previous density functional theory result $\\Delta\\phi \\approx 3$ eV at 1 ML on Au(111) is an artifact of metastable high-symmetry structures. Full relaxation of all ions, starting from many arrangements including oxygen at unstable sites and subsurface positions, yields surface reconstructions in which oxygen induces gold-atom displacements and, at high coverage, forms molecule-like oxygen aggregates. These reconstructed structures have $\\Delta\\phi < 1.1$ eV at $\\le$1 ML and a nearly coverage-independent $\\Delta\\phi$ above 1 ML, in agreement with reported and new Kelvin-probe experiments; hybrid-functional calculations confirm the GGA values. The paper also finds that on Au(111), many quasi-stable structures with some oxygen atoms below the surface lie close in energy to the most stable structure, which it argues explains the experimentally observed loss of long-range order at high coverage.","pith_inferences":["The same approach—using the work function as a structural fingerprint and relaxing from low-symmetry initial states—could resolve similar discrepancies on other oxidized noble-metal surfaces, where computed work-function shifts often overestimate experiment.","The molecule-like oxygen pairs at high coverage suggest O–O interactions, not just O–Au binding, control the energetics; a testable extension is to compare the vibrational frequency of the O–O pair with that of adsorbed O2.","The hybrid-functional result that Au(100) and Au(110) surfaces develop small bandgaps at 1 ML implies the oxidized surfaces may behave as poor metals or narrow-gap semiconductors, which would affect charge-transfer models in gold/oxide devices."],"forward_implications":["The earlier ~3 eV work-function shift for 1 ML oxygen on Au(111) should be abandoned as an artifact of metastable high-symmetry structures.","High-coverage oxygen can remain chemisorbed on gold even when supplied as O2, because the calculated formation energy per oxygen atom stays above the O–O bond energy, matching thermal-desorption experiments.","The ~1 eV work-function shift becomes a practical reference for estimating oxygen coverage on gold surfaces when long-range order is absent.","At 2 ML, oxygen on all three surfaces is close to molecular adsorption rather than atomic chemisorption, with the work-function shift nearly saturated.","Mixing of near-degenerate structures with subsurface oxygen destroys long-range order on Au(111) near room temperature, consistent with the absence of ordered overlayers in experiments."],"supporting_citations":[{"why":"Supplies the previous DFT result of a ~3 eV work-function shift at 1 ML on Au(111) that this paper identifies as an artifact of metastable high-symmetry structures.","marker":"[24]"},{"why":"Provides the experimental work-function shift below 1 eV on Au(111) up to about 1.3 ML used as the validity benchmark.","marker":"[15]"},{"why":"Provides the experimental work-function shift around 1 eV on Au(110) used to validate the calculations.","marker":"[23]"},{"why":"Supplies low-coverage oxygen binding-site preferences and the linear O–Au–O motif that the structure search builds on.","marker":"[17]"},{"why":"Supplies site-dependent binding energies of oxygen on gold surfaces that calibrate the low-coverage energetics.","marker":"[21]"},{"why":"Supplies the solid-tuned GGA functional used for the full relaxations and work-function calculations.","marker":"[36]"},{"why":"Supplies the hybrid functional used to confirm the work-function values and the density of states.","marker":"[37]"}],"fun_headline_variants":["Gold work function shift: 1 eV not 3 eV","Oxygen on gold: work function shifts only ~1 eV, not 3","Surface reconstruction explains gold's ~1 eV work function change","DFT overestimated oxygen's effect on gold work function","Corrected structures: O on Au gives <1 eV work function shift"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusion rests on the assumption that the hand-selected set of starting arrangements, though broad, contains the true lowest-energy structure at each oxygen coverage; if a lower-energy arrangement with a larger work-function shift exists, the match with experiment would be coincidental.","fun_headline_variants_meta":{"raw":{"variants":["Gold work function shift: 1 eV not 3 eV","Oxygen on gold: work function shifts only ~1 eV, not 3","Surface reconstruction explains gold's ~1 eV work function change","DFT overestimated oxygen's effect on gold work function","Corrected structures: O on Au gives <1 eV work function shift"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000562,"raw_usage":{"total_tokens":2745,"prompt_tokens":1102,"completion_tokens":1643,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":718,"completion_tokens_details":{"reasoning_tokens":1553}},"tokens_in":718,"tokens_out":1643,"duration_ms":12867,"temperature":1.0,"reasoning_tokens":1553,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:22:14.281608+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An unbiased structure search (for example, random or evolutionary generation of many more starting oxygen positions, including deeper subsurface sites) that finds a 1 ML Au(111) structure with formation energy lower than the paper's most stable one and a computed work-function shift near 3 eV would refute the central claim, since the paper's method would have missed the true ground state.","supporting_citations":[{"cited_title":"Shi and C","cited_arxiv_id":null,"evidence_quote":"Supplies the previous DFT result of a ~3 eV work-function shift at 1 ML on Au(111) that this paper identifies as an artifact of metastable high-symmetry structures."},{"cited_title":"Saliba, D","cited_arxiv_id":null,"evidence_quote":"Provides the experimental work-function shift below 1 eV on Au(111) up to about 1.3 ML used as the validity benchmark."},{"cited_title":"Gottfried, K.J","cited_arxiv_id":null,"evidence_quote":"Provides the experimental work-function shift around 1 eV on Au(110) used to validate the calculations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies low-coverage oxygen binding-site preferences and the linear O–Au–O motif that the structure search builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies site-dependent binding energies of oxygen on gold surfaces that calibrate the low-coverage energetics."}],"review_version":1}