{"id":"0d43b001-0d44-4db7-83a0-caa1a2d5f8c0","arxiv_id":"2411.08751","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A modified coupled cluster parametrization removes the bifurcations and geometric phase failures that blocked ground-state conical intersections.","lead":"The paper develops a generalized coupled cluster method that can describe conical intersections between the ground and first excited electronic states, a case where standard coupled cluster theory fails. This matters because it removes a key obstacle to using highly accurate coupled cluster models in simulations of molecules relaxing back to the ground state after absorbing light.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim that projecting out the lowest Jacobian eigenvector makes the effective Jacobian positive definite is asserted but not proven; if a second eigenvalue approaches zero, bifurcations would persist and the central claim would fail.","rationale":"The paper presents compelling numerical evidence: continuous GCCSD surfaces around S0/S1 intersections, a sign change of the reduced-matrix eigenvectors after a 2π loop demonstrating the geometric phase, and size-extensivity tests. The central mechanism, however, is the claim that removing the lowest Jacobian eigenvector component makes the effective Jacobian positive definite and the amplitude equations convex. This is the load-bearing assumption because every subsequent claim of 'no bifurcations' depends on it, yet it is justified only by numerical observation. The reader's weakest assumption identified the same point, and the Supporting Information's dependence on the number of projected states reinforces that the framework lacks a principled rule. This is a correctness risk rather than an internal inconsistency: the method may work for the presented systems, but its generality is not established. The concern is concrete and testable by examining the spectrum of the effective Jacobian along the continuation paths. Because the numerical support is strong and the gap is addressable in principle, the reader's CONDITIONAL verdict is appropriate and no change is needed from this stress-test pass.","tokens_in":23857,"tokens_out":5339,"duration_ms":53233,"concrete_test":"Instrument the implementation to compute, at every converged geometry along the circular scan in Fig. 4 and the 2D scans in Figs. 2 and 6, the eigenvalues of the Jacobian of the full residual equations used in the DIIS solver (the projected amplitude equations together with the eigenvalue conditions defining r1, l1, and omega1). If the smallest eigenvalue or singular value crosses zero or becomes comparable to the convergence tolerance along any path, the solution branch is singular and the 'convex problem without bifurcation' claim is contradicted. As a second check, modify the ethylene S0/S1 system so that a second state is deliberately near-degenerate with S1 (e.g., by changing the geometry or basis), solve with only one projected state, and test whether the projected Jacobian becomes singular; if it does, the choice of how many states to project is load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After Eq. (15), the paper states that 'the effective Jacobian that enters the amplitude equations becomes positive definite and we obtain a convex problem without a bifurcation.' This is the theoretical foundation for the central claim that GCCSD eliminates bifurcations at ground-state conical intersections, but it is based on numerical observation and not derived. The effective Jacobian of the projected amplitude equations is not simply the restriction of A to the complement of the lowest eigenvector: because the projection operator depends on the cluster amplitudes through r1 and l1, its derivative contributes additional terms to the Jacobian of the coupled amplitude/eigenvalue system. Even setting that coupling aside, the argument removes only one eigenvector. If the second-lowest eigenvalue of A becomes small or vanishes along the scan (e.g., at a three-state intersection, or when another state approaches degeneracy), the projected Jacobian remains nearly singular and bifurcations would persist. The number of projected states is a user choice (1 for ethylene, 2 for thymine, 3 for LiF), and Table S3 shows that excitation energies shift by up to ~0.02 eV when this number is varied, so the framework is not fully specified without a criterion for choosing it. None of the finite scans rules out a geometry at which the projected Jacobian is singular, because the examples are chosen so that only one or two states are near-degenerate. A proof of positive-definiteness under stated conditions, or an adaptive procedure that projects all eigenvectors with near-zero eigenvalues, is needed for the central claim to hold generally.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a generalized coupled cluster (GCC) framework in which the cluster amplitudes are constrained to have no components along one or several selected eigenvectors of the coupled-cluster Jacobian. The amplitudes and the selected eigenvectors are determined from coupled equations, and the final ground and excited states are obtained by diagonalizing a similarity-transformed Hamiltonian in a reduced or full space that reintroduces the projected components. The authors claim that this construction removes the bifurcations of the standard CCSD amplitude equations near ground-state conical intersections and restores the correct geometric phase, while preserving size extensivity and size intensivity. The method is applied to LiF, ethylene, thymine, and 2,4-cyclohexadien-1-ylamine, with numerical demonstrations of continuous potential energy surfaces, the expected sign change of the eigenvectors of the reduced matrix after a 2π loop, and the invariance of excitation energies when non-interacting subsystems are added.","tokens_in":24114,"tokens_out":3160,"duration_ms":33768,"significance":"If the central claim holds, this is a significant step: it would provide a single-reference coupled cluster description of S0/S1 conical intersections of the same symmetry, where standard CCSD is known to give divergent, multi-valued, or non-converging solutions. The numerical results are encouraging and include several favorable features: no fitted physical parameters; continuous GCCSD surfaces for LiF and ethylene; a correct geometric phase sign change in the reduced-space eigenvectors; and explicit size-extensivity/size-intensivity tests with non-interacting molecules. The paper also outlines extensions to CC2 and to Hartree-Fock/DFT, which broaden its potential impact. However, the theoretical foundation for the key claim—that the projected amplitude equations are convex and free of bifurcations—is asserted rather than proven, and the numerical evidence does not cover all configurations. The method also depends on a user-chosen number of projected states, and no criterion is given for that choice.","major_comments":[{"comment":"The statement after Eq. (15) that 'the effective Jacobian that enters the amplitude equations becomes positive definite and we obtain a convex problem without a bifurcation' is not proven. The argument removes only the component along the lowest right/left eigenvector pair, but the projected amplitude equations (19) are solved together with the eigenvector equations (20)-(21), and the projection operator depends on the cluster amplitudes through r1 and l1. The Jacobian of this coupled system therefore contains additional terms arising from derivatives of the projection operator, not just the restriction of A to the complement of the lowest eigenvector. Even ignoring that coupling, removing one eigenvector does not guarantee positive definiteness if the second-lowest eigenvalue becomes small or vanishes, which could occur at three-state intersections or when another state approaches degeneracy. Since the central claim of the paper rests on this assertion, a proof or a precise set of conditions under which it holds is required.","section":"Generalized coupled cluster theory"},{"comment":"The number of projected states, n_proj, is a user-chosen parameter, and the paper provides no criterion for selecting it. Table S3 shows that in thymine, changing n_proj from 1 to 5 shifts excitation energies by up to about 0.02 eV and changes the ground state energy by several microhartree, while the paper itself notes that simultaneous projection of eigenvectors from different subsystems breaks size extensivity. Without a specification of how n_proj should be chosen, the framework is incompletely defined, and the size-extensivity statement in the main text applies only under conditions that may not be enforceable in practice.","section":"Applications / Supporting Information Table S3"},{"comment":"The numerical demonstrations are finite scans over selected branching-plane and circular coordinates, and they do not establish that the projected amplitude equations are free of bifurcations for all geometries in a neighborhood of the intersection or in the full configuration space. The paper's abstract and introduction claim that GCC 'avoids bifurcations of the solutions to the ground state equations' in general. To support that claim, the authors need either a general proof of the convexity/positive-definiteness assertion or an explicit characterization of the region in which the GCC equations have a unique solution. The current evidence, while suggestive, is not sufficient for the global statement.","section":"Applications (ethylene, thymine, cyclohexadienylamine scans)"}],"minor_comments":[{"comment":"There are several typographical issues, including 'T able' in the captions of Tables 1 and 2, and 'enegies' in the Supporting Information. These should be corrected in a revised version.","section":"Throughout"},{"comment":"The g vector for ethylene contains a misplaced bracket: the last component is written as '-0.10184678772]' instead of a clean number. This is a formatting error that should be fixed.","section":"Supporting Information, Table S8"},{"comment":"The phrase 'in an (N − 1) dimensional configuration space' should be hyphenated as '(N − 1)-dimensional' for clarity, and the intended meaning (intermediate normalization renders the full CC wave function undefined on a measure-zero set) could be stated more explicitly.","section":"Conclusions"},{"comment":"The small region of complex energies near the intersection is acknowledged in the text as expected from Ref. 7. It would be helpful to state explicitly that this defect is distinct from the bifurcation problem addressed by GCC and that it can be removed by the similarity-constrained transformation mentioned in the paper, to avoid confusion for readers.","section":"Fig. 2b"}],"recommendation":"major_revision","confidential_remarks":"The paper's numerical results are strong and the idea is timely, but the main theoretical claim is not yet established. The authors should either provide a rigorous proof of the positive-definiteness/convexity assertion, or substantially weaken the claim and present GCC as an empirically robust regularized scheme. The n_proj dependence is a real specification gap. The small complex-energy defect near the intersection is not fatal, since it is explicitly tied to the non-Hermitian eigenvalue problem and is treated in prior work, but it should be discussed more carefully. The overall fit with the journal is good."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Fred, two things you should know. First, the paper actually does something new: it removes Jacobian eigenvector components from the cluster amplitudes and then diagonalizes the full-space similarity-transformed Hamiltonian. That construction is not in the cited literature, and it appears to fix the S0/S1 same-symmetry intersection problem for the tested molecules. Second, the theoretical justification for why this works—the claim after Eq. (15) that the effective Jacobian becomes positive definite and the problem convex—is an assertion based on numerical observation, not a derivation. Given how much rests on that claim, it is the main thing to scrutinize in review.\n\nThe numerics are convincing: continuous surfaces in LiF, ethylene, thymine, and the amine; the geometric phase sign change across a 4π loop; size-extensivity and size-intensivity tests with non-interacting molecules. The size-extensivity argument is clear and the block-structure analysis in Eqs. (30)-(36) is correct. The implementation in eT is described, though no public code is shipped, which limits immediate reproducibility.\n\nNow the soft spots, in proportion. The stress-test note about the projection operator depending on the cluster amplitudes is a real point. The Jacobian of the coupled amplitude/eigenvalue system is not simply the restriction of A to the complement of the lowest eigenvector; there are derivative terms from the projector. The paper does not address that, so the positivity claim is even less obvious than it appears. Removing one eigenvector also does not ensure the second-smallest eigenvalue cannot cross zero; a three-state degeneracy or a different scan could break it. The choice of n_proj is a user parameter, and Table S3 shows ~0.02 eV shifts in thymine excitation energies between n_proj = 1 and 5. That is small but not negligible, meaning the method is not fully specified without a selection criterion. The complex-energy defect at the ethylene intersection is small and acknowledged, but it is present.\n\nNone of this kills the paper. The central idea is good and the evidence is strong enough to merit a serious referee. The authors could reasonably be asked to either prove positivity under stated conditions or soften the claim to what is actually shown, and to discuss the n_proj sensitivity.\n\nMy recommendation: send it to peer review, not to the desk reject pile. I would also bring it to a reading group and cite the construction in my own work. The theory gap is the thing to push on.","headline":"GCCSD is a real step forward for ground-state intersections in coupled cluster theory, but the central convexity claim is asserted rather than proven; worth refereeing with a request for proof or a softening of the claim.","tokens_in":24638,"tokens_out":1985,"would_cite":true,"duration_ms":19570,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81V55","81Q70"],"pacs":["31.15.ve","31.50.Df","03.65.Vf"],"model":"deepseek-v4-flash","headline":"The paper claims that projecting the lowest Jacobian eigenvector out of the cluster amplitudes makes the coupled cluster ground state well-behaved at conical intersections, restoring the geometric phase and eliminating bifurcations.","keywords":["coupled cluster theory","conical intersections","geometric phase","bifurcation","ground state intersections","nonadiabatic dynamics","generalized coupled cluster","size-extensivity"],"falsifier":"At a S0/S1 conical intersection in ethylene, scan the branching plane and compute the lowest eigenvalue of the effective Jacobian after the projection; any geometry where the projected eigenvalue is zero or negative would contradict the convexity claim and should show a bifurcation (two real GCCSD solutions or a non-convergence region) in a continuation run.","tokens_in":23651,"feed_emoji":"⚛️","tokens_out":6281,"duration_ms":50973,"temperature":0.7,"pith_summary":"This paper presents a modified coupled-cluster method, GCC, that keeps the ground-state equations well behaved at conical intersections between the ground and first excited states, a situation where standard coupled-cluster theory fails. By removing the component of the cluster amplitudes along the lowest eigenvector of the Jacobian, the effective Jacobian becomes positive definite and the amplitude equations become convex, eliminating the bifurcations that give standard CCSD multiple, partially divergent solutions. The removed component is reintroduced by diagonalizing a small non-Hermitian matrix, which restores the conical topology and the geometric phase while keeping the amplitudes and Jacobian eigenvectors single-valued. Demonstrations on lithium fluoride, ethylene, thymine, and 2,4-cyclohexadien-1-ylamine show continuous potential energy surfaces around ground-state intersections, and the same construction is extended to CC2, Hartree-Fock, and DFT.","feed_headline":"GCCSD restores the geometric phase at ground-state conical intersections","feed_subtitle":"Standard CCSD diverges or bifurcates at S0/S1 crossings; one projected eigenvector fixes both problems.","key_machinery":"The machinery is the projector $\\hat{P}_1 = |R_1\\rangle\\langle L_1|$ built from the biorthonormal left and right eigenvectors of the coupled-cluster Jacobian $A$ associated with the lowest eigenvalue $\\omega_1$. Writing the cluster amplitudes as $|t\\rangle = |t'\\rangle - |R_1\\rangle\\langle L_1|t'\\rangle$ strips out the diverging component, and the amplitude equations are solved in the modified manifold $\\langle\\tilde{\\mu}| = \\langle\\mu| - \\langle\\mu|R_1\\rangle\\langle L_1|$ so that the effective Jacobian is positive definite. The projected component is returned through the reduced $2\\times 2$ matrix $H_{\\mathrm{RS}}$ whose eigenstates carry the geometric phase; the full space matrix $H_{\\mathrm{FS}}$, with the block $\\langle\\tilde{\\mu}|\\bar{H}|\\tilde{\\nu}\\rangle$, gives the exact limit. The same biorthogonal projectors are used to keep the phase of the eigenvectors continuous when scanning geometries, by matching the sign of the overlap with the previous geometry.","core_discovery":"In standard coupled cluster theory the amplitude equations are solved by an exponential ansatz $\\exp(T)|\\mathrm{HF}\\rangle$ whose amplitudes are required to contain no component along the state that becomes degenerate with the ground state. Near a ground-state conical intersection, the component of the amplitude vector along the lowest left eigenvector of the Jacobian diverges, and the near-zero Jacobian eigenvalue produces a bifurcation point so that several real solutions coexist or none exist. GCC solves this by parametrizing the wave function with amplitudes from which the lowest Jacobian-eigenvector component has been projected out; after this projection the effective Jacobian is positive definite, the amplitude equations are convex and have a single solution, and the wave function is well behaved throughout the branching plane. The missing component is reintroduced by diagonalizing the similarity-transformed Hamiltonian in a space that includes the projected state, which gives the geometric phase and a correct conical intersection while keeping cluster amplitudes and Jacobian eigenvectors without a phase, and therefore single-valued.","pith_inferences":["The positive-definiteness of the projected Jacobian is asserted from numerical evidence, not proven; a spectral analysis of the projected Jacobian for minimal models would settle whether the convexity guarantee is general or geometry-dependent.","Because the lowest Jacobian eigenvalue can change character as geometries move, an automatic criterion for choosing which eigenvector (or subspace) to project could make GCC a drop-in solver for nonadiabatic dynamics; this is not implemented in the paper.","The same projection strategy could plausibly cure the analogous failures in algebraic-diagrammatic-construction methods, which the paper identifies as sharing the ground-state intersection problem, though GCC-ADC is not formulated."],"forward_implications":["GCCSD yields continuous single-valued potential energy surfaces for S0 and S1 across the full branching plane of a ground-state conical intersection, including the region where standard CCSD has a phase-effect mismatch, a flipped solution with negative excitation energy, or no convergent solution at all.","Traversing a loop around the intersection, the GCCSD amplitudes and Jacobian eigenvectors return to their starting values after $2\\pi$ while the eigenstates of the reduced/full space Hamiltonian change sign, reproducing the geometric phase without phase-carrying amplitudes.","The two-state reduced matrix reproduces the full-space eigenvalues to about $10^{-8}$ Hartree in the tested cases, so the method's cost stays close to CCSD (wall time factor about 1.7 for the ethylene example).","The method preserves size-extensivity of energies and size-intensivity of excitation energies when the projected state is localized in a single non-interacting subsystem, and projects states into the same subsystem in multi-system calculations."],"supporting_citations":[{"why":"Establishes that the ground state coupled cluster wave function fails to account for the geometric phase, producing divergences and multi-valued surfaces that GCCSD is designed to fix.","marker":"[13]"},{"why":"Characterizes crossing conditions and defects in the non-Hermitian eigenvalue problem, used to understand the defective area remaining in GCCSD spectra.","marker":"[7]"},{"why":"Shows that eigenstates of the non-Hermitian eigenvalue problem carry the correct geometric phase effect, grounding the claim for the reduced/full space diagonalization step.","marker":"[8]"},{"why":"Supplies the epsilon-MECI algorithm that provides the starting geometries for the 2D scans around the intersection seams.","marker":"[18]"},{"why":"The derivative coupling algorithm used to compute the g and h vectors defining the branching planes.","marker":"[40]"},{"why":"Defines the CC2 model, the basis for the GCC2 formulation proposed in the paper.","marker":"[30]"}],"fun_headline_variants":["Generalized CC resolves ground-state conical intersections","Projected eigenvector fixes CC ground-state crossings","GCC captures geometric phase without bifurcations","Single-valued amplitudes for ground-state intersections"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that eliminating the lowest Jacobian-eigenvector component makes the effective Jacobian positive definite everywhere the method is used; the paper asserts this from numerics, not from a proof, and if it fails at some geometries the bifurcations would reappear.","fun_headline_variants_meta":{"raw":{"variants":["Generalized CC resolves ground-state conical intersections","Projected eigenvector fixes CC ground-state crossings","GCC captures geometric phase without bifurcations","Single-valued amplitudes for ground-state intersections"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000209,"raw_usage":{"total_tokens":1363,"prompt_tokens":858,"completion_tokens":505,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":474,"completion_tokens_details":{"reasoning_tokens":449}},"tokens_in":474,"tokens_out":505,"duration_ms":5404,"temperature":1.0,"reasoning_tokens":449,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:23:24.619048+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"At a S0/S1 conical intersection in ethylene, scan the branching plane and compute the lowest eigenvalue of the effective Jacobian after the projection; any geometry where the projected eigenvalue is zero or negative would contradict the convexity claim and should show a bifurcation (two real GCCSD solutions or a non-convergence region) in a continuation run.","supporting_citations":[{"cited_title":"F.; Myhre, R","cited_arxiv_id":null,"evidence_quote":"Characterizes crossing conditions and defects in the non-Hermitian eigenvalue problem, used to understand the defective area remaining in GCCSD spectra."},{"cited_title":"M.; Kj nstad, E","cited_arxiv_id":null,"evidence_quote":"Shows that eigenstates of the non-Hermitian eigenvalue problem carry the correct geometric phase effect, grounding the claim for the reduced/full space diagonalization step."},{"cited_title":"Determining minimum energy conical intersections by enveloping the seam: exploring ground and excited state intersections in coupled cluster theory","cited_arxiv_id":"2411.08207","evidence_quote":"Supplies the epsilon-MECI algorithm that provides the starting geometries for the 2D scans around the intersection seams."},{"cited_title":"F.; Koch, H","cited_arxiv_id":null,"evidence_quote":"The derivative coupling algorithm used to compute the g and h vectors defining the branching planes."},{"cited_title":"The second-order approximate coupled cluster singles and doubles model CC2","cited_arxiv_id":null,"evidence_quote":"Defines the CC2 model, the basis for the GCC2 formulation proposed in the paper."}],"review_version":1}