{"id":"2bcb7ef6-1bff-43e6-a89d-c95b112494c9","arxiv_id":"2505.11124","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Molecular quantum dynamics can be written so that all effects beyond the Born-Oppenheimer approximation are carried by the overlap between electronic states at different nuclear geometries.","lead":"This paper introduces a way to simulate how molecules move and react by tracking how the electron cloud changes shape as nuclei move, not just how the energy changes. It claims that one quantity, the electronic overlap between different geometries, automatically includes quantum interference effects that ordinary simulations miss.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Ab initio results use a path-product substitute for the direct overlap A_mn, one that can path-dependently miss quantum-metric amplitude and does not converge to the exact overlap; the numerical evidence for the exactness claim is uncontrolled.","rationale":"The strongest claim has two parts: the algebraic framework of Eq. (8) and its numerical validation. I find the algebraic part plausible: with an orthonormal DVR nuclear basis, the product basis is orthonormal, the kinetic matrix element naturally acquires the electronic overlap factor, and the potential matrix is diagonal by DVR quadrature; the continuum limit in Eq. (12) resembles a regularized gauge-covariant equation. I do not see a fatal internal inconsistency there, so I would not reject the paper on that basis. The load-bearing problem is in the ab initio implementations. The direct overlap defined in Eq. (9) is replaced by a path-ordered product of nearest-neighbor links, and for smooth real electronic states this product tends to 1 in the Delta -> 0 limit for fixed endpoint distance, so it cannot reproduce the quantum-metric amplitude of the exact overlap; it is also path-dependent and would violate closed-loop return to 1. Ref. [73] provides no error bound. Consequently, the H3+ quantum-metric and nonadiabatic demonstrations, which are the only evidence for the 'numerically exact' claim on realistic systems, are not controlled. The reader's weakest assumption already flagged this linked-product approximation; my stress-test sharpens it by showing that the approximation has the wrong continuum limit, not merely an unquantified error. I do not think this changes the overall verdict: CONDITIONAL remains appropriate because the framework could still be correct and the model-system tests in Sec. V.A are encouraging, but the central numerical claim requires direct-overlap verification or a rigorous error analysis. The paper should also complete Appendix B, whose derivation currently ends with an unresolved equation label.","tokens_in":19853,"tokens_out":20890,"duration_ms":241105,"concrete_test":"Compute the direct global overlap A_mn = <psi(R_m)|psi(R_n)> for all pairs on a small (e.g., 21x21) subgrid of the H3+ grid using the same FCI/cc-pVTZ wavefunctions, and compare it with the linked-product matrix of Ref. [73] along both the x-first and y-first paths. Measure the maximum |A_direct - A_path|, the path dependence, and the closed-loop product minus 1. Then rerun the Sec. V.B adiabatic wave-packet dynamics using the direct A and compare the (Y) trajectory and the 24.2 fs nodal pattern with the linked-product result. If the two dynamics differ materially, or if the path-product error grows as the grid spacing shrinks, the reported ab initio evidence does not support the exactness claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that Eq. (8) with A_mn = <psi(R_m)|psi(R_n)> is a numerically exact discretization of the molecular time-dependent Schrodinger equation. The ab initio demonstrations in Secs. V.B and V.C do not use this A_mn: following Ref. [73], they replace non-nearest-neighbor overlaps by a path-ordered product of nearest-neighbor links. That product is not equal to the direct overlap. For a smooth real electronic state with quantum metric component g along a path, each nearest-neighbor overlap has magnitude roughly sqrt(1 - g Delta^2), and the product over N = L/Delta steps is approximately exp(-g L Delta / 2), which tends to 1 as Delta -> 0, while the exact overlap <psi(R)|psi(R+L)> can be arbitrarily small. Thus the linked-product approximation misses exactly the amplitude (quantum metric) information that Sec. V.B claims to demonstrate, and in the dense-grid limit it collapses toward A=1, the Born-Oppenheimer limit. The approximation is also path-dependent, and no closed-loop consistency check or error estimate is reported. Because the H3+ quantum-metric and geometric-phase signatures are produced with this substituted A, those simulations do not validate the 'numerically exact' claim. The framework may still be correct, but the central numerical evidence is currently uncontrolled.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a 'topological' formulation of molecular quantum dynamics based on a discrete local trivialization ansatz Ψ = Σ_n C_n(t) ψ(r;R_n) χ_n(R), where χ_n are DVR nuclear basis functions and ψ(r;R_n) are adiabatic electronic states at grid geometries. Inserting this ansatz into the time-dependent Schrödinger equation gives i Ċ = (T A + V) C, with A_{mn} = ⟨ψ(R_m)|ψ(R_n)⟩_r. The authors argue that this single overlap matrix encodes all beyond-Born-Oppenheimer effects (diagonal BO corrections, geometric phase, and nonadiabatic transitions), is invariant under the random phases/signs produced by electronic structure codes, and is divergence-free. Numerical illustrations are given for a two-mode vibronic model, a phenol photodissociation model, and ab initio H3+ (FCI/cc-pVTZ) in both the adiabatic and nonadiabatic regimes.","tokens_in":20113,"tokens_out":12912,"duration_ms":138036,"significance":"The central identification—that the local-diabatic kinetic coupling is dressed by the electronic overlap matrix—is simple, explicit, and attractive. If the method is exact in the convergent-grid limit, it would provide a practical route to nonadiabatic and geometric-phase quantum dynamics without derivative couplings, vector potentials, or conical-intersection seam information, and the random-phase gauge-invariance argument is a genuine practical advantage. The model-system comparisons (vibronic and phenol) show the expected nodal-line signatures and agree well with full nonadiabatic references, which is encouraging. However, the paper's strongest claim—'numerically exact ab initio' dynamics—is not supported by the numerical evidence, because the H3+ calculations replace the exact overlap matrix by a linked-product approximation and no convergence tests are reported.","major_comments":[{"comment":"The ab initio calculations do not use the global overlap matrix A_{mn} of Eq. (9); following Ref. [73], they replace non-nearest-neighbor entries by a path-ordered product of nearest-neighbor links. This is not an exact representation of A_{mn}. For smooth real electronic states, |⟨ψ(R)|ψ(R+Δ)⟩| ≃ 1 − (1/2) g Δ², so the product over N = L/Δ links has magnitude ≃ exp(−g L Δ / 2), which tends to 1 as Δ → 0, while the direct overlap |⟨ψ(R)|ψ(R+L)⟩| can remain arbitrarily small. The approximation therefore suppresses exactly the quantum-metric amplitude that Section V.B claims to demonstrate, and it is not guaranteed to preserve the sign/phase of the direct overlap (path dependence in the presence of Berry curvature). No path-independence check or error quantification is provided. Consequently, the H3+ results in Figs. 3 and 4 do not validate the 'numerically exact' claim for ab initio dynamics.","section":"Section V.B, first paragraph; Section V.C"},{"comment":"The method is repeatedly called 'numerically exact', but the numerical demonstrations do not include convergence studies with respect to grid spacing, grid extent, or the number of electronic states retained. For the H3+ adiabatic example, no comparison to a direct-overlap calculation or to an independent reference is given; for the H3+ nonadiabatic example, the number of electronic states retained in the multi-state overlap matrix is not specified. The phenol simulation is explicitly not fully adiabatic (about 10% population transfer to S2), so its single-state result is approximate. The exactness claim should be either demonstrated with convergence/basis-set checks or reformulated as 'convergent in the complete-basis and dense-grid limit'.","section":"Abstract; Section VI"}],"minor_comments":[{"comment":"The continuum-limit statement C_n(t) χ_n(R) → χ(R_n,t) is schematic: C_n are DVR expansion coefficients, not pointwise values in coordinate space, so the limiting relation should involve the DVR transformation and quadrature weights. Please state the precise definition of the continuum limit.","section":"Eq. (12)"},{"comment":"The non-Abelian electronic quantum geometric tensor is said to be gauge-invariant under the stated local transformations; it is actually gauge-covariant (it transforms as U† Q U). Please correct the wording or restrict the invariance statement to the Abelian case.","section":"Eq. (26)"},{"comment":"There are minor language errors: 'potenatial' in Section V.A.2, 'developped' in Section IV, and Eq. (??) in Appendix B should be given a proper equation number.","section":"Section V.A.2 and Section IV"},{"comment":"The color scale and normalization of the overlap-matrix plot are not described; please specify the plotted quantity and explicitly explain how negative entries arise from the random sign gauge of the many-electron wavefunctions.","section":"Fig. 3b"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the formal framework is attractive and the model-system demonstrations are persuasive, but the ab initio validation is not yet at the level claimed. The load-bearing use of an unpublished companion paper (Ref. [73]) for the overlap matrix is a concern; I would require either direct-overlap benchmarks or a clear validity statement for the linked-product approximation. The novelty relative to Refs. [30,50,51] should also be clarified, though the geometric framing and the ab initio applications appear to go beyond those earlier works."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nThe paper is worth a careful read. Its central idea—that the global electronic overlap matrix A_mn contains Berry phase, diagonal BO corrections, quantum metric, and nonadiabatic transitions, with no vector potential and no global gauge fixing—is coherent and clearly articulated. The derivation of Eq. (8) from the local-diabatic DVR ansatz is algebraically sound, and the random-phase robustness argument is correct. The explicit connection to the electronic quantum geometric tensor is a genuine conceptual step beyond the earlier local-diabatic representations (Refs. 30, 50, 51), which the authors properly credit.\n\nWhat the paper does well: the vibronic model and phenol tests, where direct overlaps can be computed, demonstrate that the framework reproduces geometric phase effects using only a single adiabatic surface. That is a nice result. The nonadiabatic generalization in Eq. (21) is natural, and the promise of using random-gauge electronic structure output without post-processing is appealing.\n\nThe soft spot is in the ab initio H3+ sections. The paper calls the method \"numerically exact,\" but Secs. V.B and V.C replace the global overlap matrix by a path-ordered product of nearest-neighbor links (the linked product approximation of Ref. 73). That product is not equal to the direct overlap. For a smooth real electronic state with quantum metric g, each link has magnitude approximately sqrt(1 - g Delta^2), and the product over a path of length L scales as exp(-g L Delta / 2), which tends to 1 as the grid spacing Delta goes to zero, while the exact overlap <psi(R)|psi(R+L)> can be arbitrarily small. So the linked product systematically discards the quantum-metric amplitude information that Sec. V.B claims to demonstrate. The H3+ \"quantum metric\" effect is therefore likely an artifact of the finite grid spacing, and no convergence check or error estimate is reported. The Berry phase information is more robust to the product approximation, but the \"numerically exact\" claim cannot rest on these examples.\n\nMinor issues: Appendix B is unfinished (the target equation is literally \"Eq. (??)\"), and no code or data are provided, so the ab initio results cannot be independently checked.\n\nOverall, the framework is plausible and the model calculations are good evidence, but the numerical validation of the exactness claim is currently uncontrolled. This deserves a serious referee, but the authors should be asked to compute direct overlaps or quantify the linked-product error, and to soften \"numerically exact\" to something like \"formally exact with controlled approximations.\" I would not yet rely on the ab initio results.\n\nBest,\n\n[Your name]","headline":"A clean formal unification of beyond-Born-Oppenheimer effects in one overlap matrix, but the ab initio evidence for 'numerically exact' is undercut by an uncontrolled linked-product approximation.","tokens_in":20642,"tokens_out":4308,"would_cite":false,"duration_ms":45630,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that a single global electronic overlap matrix encodes all effects beyond the Born-Oppenheimer approximation, and it builds a numerically exact, divergence-free molecular quantum dynamics on that object.","keywords":["Born-Oppenheimer approximation","geometric phase","conical intersection","nonadiabatic dynamics","quantum geometric tensor","electronic overlap matrix","fiber bundle","quantum metric"],"falsifier":"Compute the same H3+ adiabatic dynamics twice, once with the linked-product global overlap matrix and once with direct many-electron overlaps at every grid pair; if the wave-packet trajectories or populations differ beyond convergence tolerance, the linked-product assumption fails. Alternatively, repeat the vibronic model on grids of increasing density; the nodal-line signature of the geometric phase should remain unchanged as the grid spacing shrinks, otherwise the discrete local trivialization limit is not reached.","tokens_in":19627,"feed_emoji":"⚛️","tokens_out":9894,"duration_ms":85509,"temperature":0.7,"pith_summary":"This paper argues that the standard Born-Oppenheimer picture of molecular dynamics leaves out exactly one object: the overlap between electronic wave functions evaluated at different nuclear geometries, and that this single overlap matrix accounts for every effect beyond the approximation, including geometric phase, diagonal Born-Oppenheimer corrections, and nonadiabatic transitions. It constructs a discrete local trivialization of the molecular wavefunction in which the nuclear kinetic energy is dressed by this overlap matrix, yielding a divergence-free equation of motion that requires no gauge fixing, no vector potential, and no knowledge of conical intersection locations. Numerical tests on a model conical intersection, phenol photodissociation, and ab initio H3+ dynamics show that single-surface simulations reproduce geometric-phase interference and that multi-surface simulations reproduce internal conversion. If correct, the result unifies adiabatic and nonadiabatic quantum dynamics into one quantum-geometric framework and makes ab initio wave-packet dynamics robust to random electronic phases.","feed_headline":"One overlap matrix captures every effect beyond Born-Oppenheimer","feed_subtitle":"Nuclear dynamics—adiabatic or not—reduces to one gauge-free, divergence-free equation from electronic overlaps.","key_machinery":"The load-bearing mechanism is the discrete local trivialization ansatz, which covers a globally nontrivial electron-nuclear fiber bundle by a finite set of topologically trivial product states $\\psi(r;R_n)\\chi_n(R)$ with electronic eigenstates taken at DVR grid centers. The identity that carries the argument is the equation of motion $i\\dot C=(TA+V)C$, together with the cumulant expansion $\\ln A_{mn}\\approx\\sum_\\mu F_\\mu\\Delta_\\mu-\\frac12\\sum_{\\mu\\nu}Q_{\\mu\\nu}\\Delta_\\mu\\Delta_\\nu$, where $F_\\mu$ is the gauge connection and $Q_{\\mu\\nu}$ is the electronic quantum geometric tensor. Because the overlap matrix is bounded within $[-1,1]$ and does not require smoothness or a global gauge, the divergences of derivative couplings and vector potentials are bypassed entirely. The same construction extends to arbitrary fiber bundles by replacing the electronic states with eigenstates of a fiber Hamiltonian, including non-Hermitian cases.","core_discovery":"The central claim is that the global electronic overlap matrix $A_{mn}=\\langle\\psi(R_m)|\\psi(R_n)\\rangle$, the fidelity between many-electron states at two nuclear geometries, fully encodes the quantum geometry of the adiabatic electronic manifold. In the discrete local trivialization ansatz $\\Psi=\\sum_n C_n(t)\\psi(r;R_n)\\chi_n(R)$, the Schrödinger equation becomes $i\\dot C=(TA+V)C$, where $T$ is the nuclear kinetic-energy matrix and $V$ the potential-energy matrix; the only difference from Born-Oppenheimer dynamics is the dressing of $T$ by $A$. Expanding $\\ln A_{mn}$ for nearby geometries shows that the overlap matrix's amplitude encodes the quantum metric and its phase encodes the Berry connection, so the singular quantum geometric tensor is replaced by a bounded object. The nonadiabatic generalization uses the multi-state overlap matrix $\\langle\\phi_\\beta(R_m)|\\phi_\\alpha(R_n)\\rangle$, unifying first- and second-derivative couplings, diagonal corrections, and geometric-phase vector potentials. The paper demonstrates that geometric-phase nodal lines appear in single-surface adiabatic dynamics near an energetically inaccessible conical intersection, that quantum-metric effects alter adiabatic dynamics even without degeneracies, and that ab initio nonadiabatic internal conversion in H3+ is reproduced without phase smoothing.","pith_inferences":["A practical corollary not pursued in the paper is that the computational bottleneck shifts to computing accurate many-electron overlaps between nearby nuclear geometries; improving overlap algorithms would directly determine how far the method scales.","The linked-product approximation used for the H3+ overlaps introduces potential path and gauge dependence whose error is not quantified; a direct comparison with exact overlaps would test the practical reliability of the ab initio demonstrations.","The framework suggests that Born-Oppenheimer dynamics may fail more often than assumed, since even energetically inaccessible conical intersections can alter reaction rates through geometric-phase interference.","One testable extension would be to apply the same overlap-dressed equation to exceptional points in non-Hermitian settings, where the quantum geometric tensor is ill-defined but the overlap matrix remains bounded."],"forward_implications":["Geometric phase effects in ground-state adiabatic reactions can be captured using only single-surface information, so simulations no longer need to locate conical intersections or construct vector potentials.","Nonadiabatic dynamics can be run directly from ab initio electronic states carrying random phases or signs, eliminating the diabatization bottleneck.","Quantum-metric corrections are predicted to affect adiabatic dynamics even in systems with no electronic degeneracy, as shown by the H3+ example.","The same equation of motion governs adiabatic and nonadiabatic dynamics, with the Abelian overlap matrix upgraded to a non-Abelian multi-state matrix.","Because the formalism applies to arbitrary fiber bundles, the same discrete local trivialization could be used for vibrational-rotational dynamics, band-structure problems, and non-Hermitian dynamics."],"supporting_citations":[{"why":"Supplies the linked-product path-ordered approximation used to compute the global electronic overlap matrix in the H3+ ab initio examples.","marker":"[73]"},{"why":"Provides the discrete-variable local diabatic representation and the splitting propagation scheme used to evolve the overlap-dressed equation.","marker":"[30]"},{"why":"Introduces the discrete-variable local diabatic representation for conical intersection dynamics that underpins the ansatz.","marker":"[50]"},{"why":"Shows that randomly gauged ab initio electronic states can be used directly, a property the overlap-matrix method exploits.","marker":"[51]"},{"why":"Documents the divergence and non-integrability of the diagonal Born-Oppenheimer correction at conical intersections, one of the singularities the new method removes.","marker":"[5]"},{"why":"Provides the geometric-phase concept that the phase of the overlap matrix encodes.","marker":"[18]"},{"why":"Gives the vanishing-curvature condition for exact diabatization, showing the topological obstruction that the overlap approach bypasses.","marker":"[24]"},{"why":"Provides the phenol photodissociation model and the geometric-phase reference dynamics that the single-surface simulation reproduces.","marker":"[62]"},{"why":"Supplies the two-state vibronic coupling model used for the first geometric-phase test.","marker":"[60]"},{"why":"Supplies the FCI/cc-pVTZ electronic states and energies used in the ab initio H3+ adiabatic and nonadiabatic simulations.","marker":"[72]"}],"fun_headline_variants":["One overlap matrix drives all quantum nuclear dynamics","Quantum geometry encoded in a single overlap matrix","Divergence-free quantum dynamics from one overlap","Beyond Born-Oppenheimer via a single overlap matrix"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that, on a sufficiently dense nuclear grid, the exact molecular wavefunction is completely spanned by product states formed from the electronic eigenstates at the grid points; the H3+ demonstrations further assume that a path-ordered product of nearest-neighbor electronic overlaps reproduces the exact overlap between distant nuclear geometries.","fun_headline_variants_meta":{"raw":{"variants":["One overlap matrix drives all quantum nuclear dynamics","Quantum geometry encoded in a single overlap matrix","Divergence-free quantum dynamics from one overlap","Beyond Born-Oppenheimer via a single overlap matrix"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00023,"raw_usage":{"total_tokens":1482,"prompt_tokens":944,"completion_tokens":538,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":560,"completion_tokens_details":{"reasoning_tokens":480}},"tokens_in":560,"tokens_out":538,"duration_ms":5292,"temperature":1.0,"reasoning_tokens":480,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:58:41.771747+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the same H3+ adiabatic dynamics twice, once with the linked-product global overlap matrix and once with direct many-electron overlaps at every grid pair; if the wave-packet trajectories or populations differ beyond convergence tolerance, the linked-product assumption fails. Alternatively, repeat the vibronic model on grids of increasing density; the nodal-line signature of the geometric phase should remain unchanged as the grid spacing shrinks, otherwise the discrete local trivialization limit is not reached.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the linked-product path-ordered approximation used to compute the global electronic overlap matrix in the H3+ ab initio examples."},{"cited_title":"Martinazzo and I","cited_arxiv_id":null,"evidence_quote":"Provides the discrete-variable local diabatic representation and the splitting propagation scheme used to evolve the overlap-dressed equation."},{"cited_title":"Matselyukh, V","cited_arxiv_id":null,"evidence_quote":"Introduces the discrete-variable local diabatic representation for conical intersection dynamics that underpins the ansatz."},{"cited_title":"Halpin, P","cited_arxiv_id":null,"evidence_quote":"Shows that randomly gauged ab initio electronic states can be used directly, a property the overlap-matrix method exploits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the vanishing-curvature condition for exact diabatization, showing the topological obstruction that the overlap approach bypasses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the phenol photodissociation model and the geometric-phase reference dynamics that the single-surface simulation reproduces."},{"cited_title":"Kolodrubetz, D","cited_arxiv_id":null,"evidence_quote":"Supplies the two-state vibronic coupling model used for the first geometric-phase test."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the FCI/cc-pVTZ electronic states and energies used in the ab initio H3+ adiabatic and nonadiabatic simulations."}],"review_version":1}