{"id":"6360c86b-5020-4c2a-b7b8-e2e5e1dfc44f","arxiv_id":"2505.18147","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"Full-dimensional machine-learned electronic friction tensors for H2 on four Cu facets show nonadiabatic electron-hole-pair effects are subtle and dissociation is controlled by the potential energy surface.","lead":"This paper builds machine-learning models of electronic friction for hydrogen molecules colliding with four different copper surfaces, then simulates the scattering with and without those friction effects. The simulations match measured sticking and survival probabilities well and show that electron-hole-pair excitations in the metal barely change the outcomes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"ODF surrogate friction errors near the transition state could bias the weak-nonadiabaticity conclusion; the low-velocity defense rests on single example trajectories rather than a systematic analysis.","rationale":"I read the paper in good faith: it is a careful study with large trajectory statistics, public code, and a genuine methodological advance in multi-facet EFT surrogates. The central conclusion that electronic friction has only subtle effects on H2 scattering on copper is physically plausible and consistent with earlier LDFA work. However, the new element is the ODF surrogate model, and the paper's own validation shows non-negligible errors in the ODF friction near the transition state (Figs. 3 and 4). The argument that these errors are harmless because velocities are low there is supported by only one trajectory per facet (Figs. S16/S17). This is the weakest point in the chain: vibrational de-excitation, the very observable used to fingerprint nonadiabaticity, depends on Lambda_dd, which is among the components with acknowledged discrepancies. A systematic check with reference EFT for the most sensitive observable would settle whether the 'weak nonadiabaticity' conclusion is physical or an artifact of the surrogate. The reader's concern about PES circularity is valid but less load-bearing: regardless of how the PES was fitted, the BO-vs-MDEF comparison uses the same PES, so the relative nonadiabatic effect is still meaningful; the circularity mainly weakens the 'excellent agreement with experiment' validation. I therefore keep the conditional verdict, but recommend that the primary condition be the EFT surrogate accuracy in the dynamically relevant region, not the PES training history.","tokens_in":31615,"tokens_out":11720,"duration_ms":103162,"concrete_test":"For the sensitive observable H2(v=2,J=1 -> v=1,J=1), recompute the MDEF-ODF transition probability on Cu(111) and Cu(110) at collision energies 0.15 eV and 0.25 eV using the reference DFT EFT (evaluated on-the-fly or with a corrected model) for 2,000-5,000 trajectories, and compare with the ACE-friction-based result. Also compute the trajectory-weighted friction-force error (sum over Lambda(R)*v along each trajectory) between the ML and DFT EFTs to test whether the surrogate error is systematic. If the transition probabilities change by more than the ~0.1% statistical error, the weak-nonadiabaticity claim is conditional on the surrogate EFT accuracy.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim that electronic friction has negligible effects on H2/Cu scattering rests on MDEF simulations where the ODF friction tensor is evaluated by an ACE-friction surrogate model. The paper itself acknowledges (Sec. III A) that this surrogate deviates significantly from DFT reference values near the dissociation barrier, for example Lambda_dd on Cu(211) and Lambda_phiphi on Cu(110). The authors dismiss these deviations because velocities are low in the transition-state region, citing one example trajectory per facet (Figs. S16 and S17). This defense is not systematic: for over-barrier collisions at 0.4-0.5 eV the molecule crosses the barrier with finite velocity, and for vibrationally excited H2(v=2) the d-coordinate velocity is large precisely where Lambda_dd peaks. An error in Lambda_dd there directly affects the computed vibrational de-excitation probability (v=2 -> v=1), which is the key nonadiabatic fingerprint in Sec. III D. If the surrogate error is biased, the BO vs LDFA vs ODF comparison could shift, making the 'subtle differences' conclusion an artifact of the surrogate rather than a physical result. The inherited PES from Refs. 58/59 and the SRP48 functional add a secondary fitting-to-experiment confound, but the primary load-bearing issue is surrogate EFT accuracy in the dynamically relevant high-friction region.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents full-dimensional machine-learning surrogate models of the electronic friction tensor for H2 on Cu(111), Cu(100), Cu(110), and Cu(211): an isotropic LDFA model built from an ACE density surrogate, and an anisotropic orbital-dependent friction (ODF) model built with the ACE-friction framework of Sachs et al. Combined with a MACE interatomic potential for the H2/Cu PES taken from prior work by the same group, the models are used in molecular dynamics with electronic friction (MDEF) to compute state-resolved sticking, survival, and rovibrational transition probabilities with 20,000 trajectories per condition. The paper reports good agreement with absolute survival probabilities on Cu(100) and with the available Cu(110) data, finds that friction changes sticking and survival only slightly across all four facets, and concludes that electron-hole-pair effects are weak, that the PES shape and initial vibrational state control dissociation, and that LDFA and ODF friction give nearly indistinguishable results. It also reinterprets the previously reported Spiering-Meyer fingerprint of anisotropic friction on Cu(111) as an artefact of the low-temperature Lambda-zero approximation.","tokens_in":31837,"tokens_out":21269,"duration_ms":171133,"significance":"If the central conclusions hold, this is a valuable contribution: it delivers transferable, publicly available PES and EFT models for reactive hydrogen chemistry on multiple copper facets; it provides a systematic, statistically well-converged comparison of adiabatic, LDFA, and ODF dynamics against experimental state-resolved observables; and it makes falsifiable predictions (negligible effect of friction on sticking and survival; weak kinetic-energy dependence of nu=2 to nu=1 de-excitation; no experimentally resolvable nonadiabatic fingerprint on any of the four facets). The manuscript reports 20,000 trajectories per condition with bootstrap error estimates, documents DFT-level validation of the friction surrogates along representative trajectories, includes convergence studies for k-grid, broadening, and slab size, and makes code, data, and models publicly available; these are genuine strengths. The qualitative conclusion of weak nonadiabaticity is consistent with prior LDFA-based studies and is supported by both the LDFA and ODF simulations, so it is probably robust.","major_comments":[{"comment":"The manuscript documents systematic deviations of the ACE-friction surrogate near the dissociation barrier, for example Lambda_dd on a Cu(211) dissociation trajectory and Lambda_phiphi on Cu(110) scattering (Fig. 3 and Fig. S11), and states that these persist in models retrained with different settings, indicating a systematic rather than statistical error. The dismissal of their dynamical impact argues that transition-state velocities are low, citing single example scattering trajectories per facet (Figs. S16-S17). This defense is not systematic and is weakest for the component that matters most for the paper's key nonadiabatic fingerprint (Fig. 9): via Eq. (1), Lambda_dd couples to the internal stretch, whose velocity for H2(nu=2) remains large (~0.1 Angstrom/fs) at the barrier, whereas the 'low velocity' argument strictly applies to the center-of-mass Z-mode near its turning point. Moreover, the deviations are documented on a dissociation trajectory, while the velocity illustration is taken from scattering trajectories, so the two do not directly confront each other. Because the BO/LDFA/ODF differences in Fig. 9 are at the 10^-3 level in probability and the documented Lambda_dd errors near the barrier are commensurate with the friction values themselves, I request a systematic quantification, e.g., the distribution of the error in the friction force Lambda-v along ensembles of dissociating and scattering trajectories, or a sensitivity test that injects the documented surrogate errors, before the 'even more subtle' LDFA-vs-ODF conclusion is drawn. I note that the main qualitative conclusion (friction is weak) is independently supported by the well-validated LDFA model and is not overturned by this concern.","section":"Sec. III A / Figs. 3-4, S16-S17"},{"comment":"The text reports that the MACE PES database was built by adaptive sampling with retraining 'until the final dynamical observable matched the reference values,' but it does not state what those reference values were. If, as in Ref. [58], they are dynamics results computed with DFT using the SRP48 functional, then the PES inherits the semi-empirical character of SRP48, which is parameterized for this very system, and the subsequent agreement with experiment is partly inherited rather than an independent test of the PES. The manuscript should state the nature of the reference values explicitly and discuss the independence of the experimental validation; this would also sharpen the central claim that dissociative adsorption is 'dominated by the shape of the underlying potential energy surface,' which otherwise risks being partially circular. This is a provenance and clarity issue: the authors are transparent about the refinement, but the key term 'reference values' is left undefined in a way that bears on the interpretation of the headline conclusion.","section":"Sec. II C a"},{"comment":"The abstract claims that 'the predicted sticking coefficient and survival probabilities are in excellent agreement with experiment,' but no direct quantitative comparison of sticking probabilities is presented. The Anger et al. sticking data discussed in Sec. III C are rovibrationally unassigned, are compared only in terms of facet ordering, and the paper itself notes that the experimental order differs from the computed nu=2,J=1 sticking. The quantitative experimental validation in Sec. III B concerns survival and rotational-excitation probabilities on Cu(100) (absolute, 500 K) and Cu(110) (one absolute point with +/-0.13 error plus slope-matched relative data). The abstract should either be revised to specify which observables are validated quantitatively, or a direct quantitative sticking comparison should be added.","section":"Abstract / Sec. III C"}],"minor_comments":[{"comment":"The sentence 'Lambda_ij(R,z) is a component of of the 3N x 3N EFT Lambda' contains a duplicated 'of.'","section":"Sec. II A"},{"comment":"The sentence 'We evaluate the the Wigner-Seitz radius rs(rho_emb)' contains a duplicated 'the.'","section":"Sec. II C b"},{"comment":"The cross-reference '(Lambda, Eq. ??,)' is a broken reference and should point to Eq. (2) of the main text.","section":"SM Sec. SVIII"},{"comment":"The error-evaluation paragraph describes the transition 'H2(nu=2,J=1 -> nu=2,J=1),' which appears to be a typo for nu=2 -> nu=1 as in Fig. S18.","section":"SM Sec. SX"},{"comment":"The explanation that the disappearance of the Spiering-Meyer ODF/LDFA fingerprint arises from the Lambda_0-versus-Lambda approximation is plausible but is confounded by the simultaneous change of PES (static 6D versus full-dimensional moving-surface); a sentence acknowledging this confound would make the 'may be an artefact' claim properly qualified.","section":"Sec. III D / SM Fig. S15"},{"comment":"The claim that the models are 'the most accurate publicly available, full-dimensional models for H2 on copper to date' is not directly established by the benchmarks shown; recommend 'to our knowledge' or an explicit quantitative comparison with the alternative models cited in the text.","section":"Sec. IV / Abstract"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the manuscript is a strong follow-on to the authors' Refs. [58,59] and is within the scope of a chemical-physics dynamics journal. The main technical risk is the one identified in Major Comment 1: the ACE-friction surrogate's systematic deviations near the transition state are dismissed with anecdotal trajectory arguments, and the requested systematic analysis is needed before the quantitative edge claims (LDFA vs ODF subtlety; de-excitation fingerprint) are fully supported. The central qualitative conclusion is probably safe because it is supported by the validated LDFA model and by prior literature. The citation practice is fair; self-citations are substantial but appropriate given the direct lineage of the models. Please ensure the GitHub repository is archived with a persistent DOI before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read arXiv:2505.18147, H2 scattering on four copper facets with ML electronic friction surrogates. My take: genuinely useful paper, central conclusion probably right, but the authors dismiss surrogate error near the transition state too quickly, and a referee should push on that.\n\nWhat's new: the first full-dimensional ODF friction tensor surrogates for H2/Cu across four facets, trained on the temperature-dependent Maurer Λ rather than the low-temperature Λ0. That choice matters — the paper shows the Spiering-Meyer ODF-versus-LDFA fingerprint on Cu(111) mostly disappears when you use Λ, backed by MEP comparisons and 0 K de-excitation curves. That is a real resolution of a live debate. The public models, the k-grid/broadening/slab-size convergence tests, and the 20,000-trajectory statistics with bootstrap errors are also done properly. The Cu(100) survival and rotational excitation agreement with Watts is genuinely good.\n\nSoft spots, in order of softness. First: the ACE-friction surrogate has known deviations in Λdd on Cu(211) and Λφφ on Cu(110) near the barrier, and the defense is 'velocities are low in the transition-state region,' backed by one example trajectory per facet (S16/S17). That is not systematic. The stress-test concern lands: for ν=2 trajectories the d-coordinate velocity is not obviously small where Λdd peaks, and the v=2→v=1 probability is the observable most sensitive to this. The friction force is Λ·v, so unquantified Λ error times finite v is unquantified bias in the key nonadiabatic fingerprint. I don't think it breaks the conclusion — the LDFA model, which is accurate, also gives weak friction, and the BO/LDFA/ODF curves cluster everywhere — but the authors should bound Λ·v over the trajectory ensemble rather than assert it from two plots.\n\nSecond: the PES is inherited from Refs. 58/59, adaptively refined until dynamical observables matched reference values, and SRP48 is itself fitted to beam data, so part of the experimental agreement is inherited. That is standard practice and not fatal, but it makes 'most accurate publicly available models' stronger than the evidence, especially given the Cu(110) absolute survival offset. Third, minor: no commit hash and no one-command reproduction path.\n\nWho it's for: the gas-surface dynamics and MDEF-development crowd; they'll use and cite it. Send it to peer review — yes. The referee should ask for a systematic Λ·v analysis in the high-friction region and temper the 'most accurate' claim.","headline":"A solid, reproducible computational study whose core 'friction is weak' conclusion is probably right, but the ODF surrogate error near the transition state needs a systematic Λ·v bound before the LDFA/ODF indistinguishability claim is fully trusted.","tokens_in":32436,"tokens_out":6153,"would_cite":true,"duration_ms":48529,"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":"Electron-hole friction barely affects H$_2$ dissociation on copper surfaces.","keywords":["dissociative chemisorption","electronic friction","electron-hole pair excitations","machine learning interatomic potentials","atomic cluster expansion","state-to-state scattering","hydrogen on copper","orbital-dependent friction"],"falsifier":"Measure the H$_2(v=2,J=1)\\rightarrow(v=1,J=1)$ vibrational de-excitation probability from Cu(110) with state-resolved resolution better than a few tenths of a percent; the paper predicts a weak kinetic-energy dependence and only a fraction-of-a-percent nonadiabatic enhancement, so a several-fold enhancement at low collision energies would falsify the near-adiabaticity claim. A complementary check is the vibrational lifetime of a hydrogen atom chemisorbed on Cu(111) or Cu(110), where LDFA and ODF differ by more than a factor of three.","tokens_in":31365,"feed_emoji":"⚛️","tokens_out":13636,"duration_ms":121340,"temperature":0.7,"pith_summary":"The paper sets out to test whether electron-hole pair excitations, the main nonadiabatic energy-loss channel in metal-surface scattering, alter the outcome of H$_2$ dissociative chemisorption on copper. It builds full-dimensional machine-learned surrogate models of the electronic friction tensor for four copper facets (Cu(111), Cu(100), Cu(110), and Cu(211)), couples them to an existing machine-learned potential energy surface, and runs thousands of state-resolved scattering trajectories with and without friction. The central claim is that electron-hole pair influence is weak: sticking and survival probabilities are controlled by the shape of the potential energy surface and the initial vibrational state, while nonadiabatic effects only produce subtle changes in rovibrationally inelastic transition probabilities. If correct, this means the least expensive adiabatic simulations capture the essentials of hydrogen dissociation on copper, and the difference between simple isotropic friction models and full anisotropic ab initio friction models is not dynamically decisive. A sympathetic reader would care because these are the first transferable, publicly available models of their kind for hydrogen on copper.","feed_headline":"Electron-hole friction barely dents H2 dissociation on copper","feed_subtitle":"New machine-learned friction models match experiment and show that potential shape, not electron friction, decides who sticks.","key_machinery":"The central machinery is molecular dynamics with electronic friction (MDEF), a Langevin equation in which a friction tensor $\\Lambda$ and a matching stochastic force dissipate nuclear energy into electron-hole pairs. Two friction models are represented: the isotropic local density friction approximation (LDFA), where the friction coefficient is a function of the local electron density, and orbital-dependent friction (ODF), where $\\Lambda$ is built from Kohn–Sham electron–phonon coupling matrix elements at the Fermi level. The transferable full-dimensional surrogates are constructed with the atomic cluster expansion (ACE), a body-ordered basis that yields symmetric positive semi-definite friction tensors with the correct equivariance; the ODF surrogate uses a row-wise coupling ansatz that factors the friction tensor into a sum of outer products of 3×3 matrix-valued environment functions. These machine-learned friction models are combined with a machine-learned interatomic potential for the potential energy surface, enabling 20,000-trajectory quantum-state-resolved scattering simulations per state and energy on all four facets.","core_discovery":"Using molecular dynamics with electronic friction (MDEF), the authors show that for quantum-state-resolved H$_2$ scattering on Cu(111), Cu(100), Cu(110), and Cu(211), the probability of dissociative adsorption is dominated by the topology of the potential energy surface and the initial vibrational quantum state, not by electron-hole pair excitation. The nonadiabatic contribution, modelled either with isotropic local-density friction (LDFA) or with full orbital-dependent friction (ODF) computed from first-order time-dependent perturbation theory, changes sticking and survival probabilities by only small amounts across all facets and collision energies. The one observable previously proposed as a sensitive fingerprint of nonadiabaticity, vibrational de-excitation H$_2(v=2,J=1)\\rightarrow(v=1,J=1)$ on Cu(111), shows only a weak kinetic-energy dependence, and the difference between LDFA and ODF is substantially smaller than earlier low-temperature friction calculations suggested. Because the new friction models reproduce ab initio reference data along scattering and dissociation trajectories, the authors conclude that the computed near-adiabaticity is not an artefact of a crude friction estimate, and they present the combined potential and friction models as the most accurate publicly available full-dimensional models for H$_2$ on copper to date.","pith_inferences":["If nonadiabaticity is this weak across all four facets, copper is a poor benchmark for testing electronic-friction theories; surfaces such as silver, gold, or platinum, where electron-hole pair coupling is stronger or barriers are shaped differently, would make sharper discriminators, and the paper's machinery transfers to them directly.","The near-degeneracy of LDFA and ODF in these dynamics suggests the cheaper isotropic friction model is sufficient for H$_2$ on copper, but the full tensorial model may still matter for observables that weight the chemisorbed state heavily, such as hot-electron yields or kinetic isotope effects.","The ODF surrogate deviates most from reference data near the transition state, where velocities are low, so the dynamical insensitivity may hide genuine friction-tensor errors; testing on heavier adsorbates with higher arrival velocities at the barrier would reveal whether the agreement is structural or coincidental."],"forward_implications":["For thermal sticking and survival probabilities of H$_2$ on the four low-index copper facets, adiabatic Born–Oppenheimer dynamics is sufficient; the added cost and complexity of electronic friction changes the result only slightly.","The facet identity and the initial vibrational state, not electron-hole pair friction, set the reactivity ordering: Cu(110) and Cu(211) react at lower translational energies when the molecule starts in $v=2$, while Cu(111) and Cu(100) stay inert until higher energies.","The previously reported kinetic-energy fingerprint of anisotropic nonadiabatic effects in H$_2$ vibrational de-excitation on Cu(111) does not survive when the full orbital-dependent friction expression is used instead of the low-temperature approximation, suggesting that the earlier anisotropy signal was likely an artefact of that approximation.","The publicly released machine-learned potential and friction models for all four facets allow future simulations that include surface phonons and electronic dissipation without re-fitting electronic structure data."],"supporting_citations":[{"why":"Supplies the row-wise coupling ACE representation of friction tensors on which the ODF surrogate model is built.","marker":"[61]"},{"why":"Provides the DFT database and iterative refinement protocol from which the machine-learned potential energy surface is trained.","marker":"[58]"},{"why":"Provides the machine-learned interatomic potential used for all Born–Oppenheimer and MDEF dynamics across the four facets.","marker":"[59]"},{"why":"Gives the first-order time-dependent perturbation theory expression used to compute the orbital-dependent friction reference data.","marker":"[49]"},{"why":"Documents the convergence problems of the low-temperature friction expression and justifies using the full ODF formula instead of the quasi-static approximation.","marker":"[50]"},{"why":"Supplies the tabulated atom-in-jellium friction coefficients used for the LDFA surrogate model.","marker":"[67]"},{"why":"Defines the vibrational de-excitation fingerprint on Cu(111) that the paper re-evaluates and explains away.","marker":"[15]"},{"why":"Provides absolute experimental survival and rotational excitation probabilities at Cu(100) used as validation benchmarks.","marker":"[20]"},{"why":"Provides relative experimental survival probabilities at Cu(110) used as a validation benchmark.","marker":"[79]"}],"fun_headline_variants":["H2 sticking on copper: potential rules, friction whispers","Copper facets reveal weak electron-hole friction in H2 scattering","Friction barely influences H2 dissociation on copper surfaces","For H2 on copper, geometry beats electron friction","Machine-learned friction shows H2 dissociation is nearly adiabatic"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The machine-learned potential energy surface was iteratively retrained until its reaction probabilities matched reference values, so the agreement with experiment and the conclusion that potential-energy shape controls dissociation partly inherit the fitness target of the fitting procedure rather than being purely independent predictions.","fun_headline_variants_meta":{"raw":{"variants":["H2 sticking on copper: potential rules, friction whispers","Copper facets reveal weak electron-hole friction in H2 scattering","Friction barely influences H2 dissociation on copper surfaces","For H2 on copper, geometry beats electron friction","Machine-learned friction shows H2 dissociation is nearly adiabatic"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000549,"raw_usage":{"total_tokens":2679,"prompt_tokens":1057,"completion_tokens":1622,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":673,"completion_tokens_details":{"reasoning_tokens":1541}},"tokens_in":673,"tokens_out":1622,"duration_ms":12996,"temperature":1.0,"reasoning_tokens":1541,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:34:17.532185+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the H$_2(v=2,J=1)\\rightarrow(v=1,J=1)$ vibrational de-excitation probability from Cu(110) with state-resolved resolution better than a few tenths of a percent; the paper predicts a weak kinetic-energy dependence and only a fraction-of-a-percent nonadiabatic enhancement, so a several-fold enhancement at low collision energies would falsify the near-adiabaticity claim. A complementary check is the vibrational lifetime of a hydrogen atom chemisorbed on Cu(111) or Cu(110), where LDFA and ODF differ by more than a factor of three.","supporting_citations":[{"cited_title":"G.; Maurer, R","cited_arxiv_id":null,"evidence_quote":"Supplies the row-wise coupling ACE representation of friction tensors on which the ODF surrogate model is built."},{"cited_title":"G.; Westermayr, J.; Douglas-Gallardo, O","cited_arxiv_id":null,"evidence_quote":"Provides the DFT database and iterative refinement protocol from which the machine-learned potential energy surface is trained."},{"cited_title":"G.; Van Der Oord, C.; Batatia, I.; Zhang, Y.; Jiang, B.; Cs´ anyi, G.; Maurer, R","cited_arxiv_id":null,"evidence_quote":"Provides the machine-learned interatomic potential used for all Born–Oppenheimer and MDEF dynamics across the four facets."},{"cited_title":"J.; Askerka, M.; Batista, V","cited_arxiv_id":null,"evidence_quote":"Gives the first-order time-dependent perturbation theory expression used to compute the orbital-dependent friction reference data."},{"cited_title":"L.; Stark, W","cited_arxiv_id":null,"evidence_quote":"Documents the convergence problems of the low-temperature friction expression and justifies using the full ODF formula instead of the quasi-static approximation."},{"cited_title":"I.; Meyer, J","cited_arxiv_id":null,"evidence_quote":"Supplies the tabulated atom-in-jellium friction coefficients used for the LDFA surrogate model."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides absolute experimental survival and rotational excitation probabilities at Cu(100) used as validation benchmarks."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides relative experimental survival probabilities at Cu(110) used as a validation benchmark."}],"review_version":1}