{"id":"11338d28-96ef-4b28-8b84-2061bf41171a","arxiv_id":"2507.18486","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A biorthogonal alpha-deformed Fubini-Study geometry is constructed, giving dual connections and a four-way classification of metric and Berry-curvature tensors for non-Hermitian quantum systems.","lead":"This paper constructs a generalized geometry for quantum states using a biorthogonal inner product, producing alpha-deformed metrics and dual connections. It classifies four geometric tensors for non-Hermitian systems and links them to natural gradient optimization.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The α-deformed 'quantum metric' in Eq. (35) is not positive for |α|>1 because the phase covariance enters with coefficient (1−α)/(1+α); the paper's claim that the phase contribution is necessarily non-negative is only valid in an unstated restricted range.","rationale":"I read the paper as constructing an α-deformed, biorthogonal Fubini-Study geometry and then applying it to non-Hermitian systems. The reader's weakest-assumption point about pointwise biorthogonal normalisation is real, but the paper itself acknowledges that this normalisation must be imposed and is not guaranteed by non-Hermitian evolution; if one works with associated left/right states or pointwise renormalisation, the construction goes through. My concern is different and more easily checkable: the α-QMT in Eq. (35) has an explicit α-dependent prefactor on the phase covariance, so the claimed 'metric' can cease to be positive definite for |α|>1. The paper states that the phase contribution is necessarily non-negative, but that statement is only true inside (−1,1). Since classical α-connections are defined for all real α and the paper never imposes or proves the needed restriction, the one-parameter family of metrics is not well-founded over the claimed domain. This does not overturn the construction inside |α|<1, nor does it invalidate the reader's conditional verdict; it sharpens the condition under which the central claim holds. A single closed-form evaluation settles whether the issue is substantive.","tokens_in":31950,"tokens_out":16811,"duration_ms":179301,"concrete_test":"Take Ψ_θ(x)=L^{-1/2} e^{iθ x} for x∈[0,L], so P(x)=L^{-1} and φ(x)=θ x. Insert this into Eq. (35) with α=2: the amplitude term vanishes and g(2)_θθ = ((1−2)/(1+2))(E[x²]−E[x]²) = −L²/36, which is negative. If the intended domain is meant to exclude |α|≥1, repeat the same computation at α=0.9 and α=−0.9 to verify positivity, and locate the explicit restriction in the manuscript. If the computation instead yields a non-negative result, the concern is vacuous.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing spot I see is the unrestricted α-domain on which the resulting object is claimed to be a quantum metric. In Eq. (35), g(α)_ij = (1/4)E[∂_i l_θ ∂_j l_θ] + ((1−α)/(1+α))(E[∂_i φ ∂_j φ] − E[∂_i φ]E[∂_j φ]). The phase covariance tensor is positive semidefinite, but the prefactor (1−α)/(1+α) is negative whenever |α|>1. The text immediately after Eq. (35) states that, like the α=0 case, the contribution from the phase is necessarily non-negative; this is only true for −1<α<1. No such restriction is stated: the paper only excludes α=1 in the definition of l_α, and the classical α-connections are introduced for arbitrary real α. Thus the 'α-QMT' can be indefinite or negative in a regime the paper appears to include. This matters directly for the central claim: the real-valued sector of the FS tensor is supposed to support both the dual-connection geometry and the natural-gradient flow in Eqs. (57) and (66). An indefinite 'metric' does not define a steepest-descent direction, and the classification of 'metric' tensors is then not a classification of metrics. The issue is concrete: for a pure-phase state, Eq. (35) at α=2 gives a negative g(2)_θθ, so the claimed non-negativity of the phase sector fails if the α-domain is truly unrestricted. If the intended domain is |α|<1, that restriction must be stated and proved; if it is not imposed, the central metric claim is false outside the restricted interval.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a one-parameter (α) deformation of the Fubini–Study tensor for pure quantum states, using two biorthogonal families l1(α) and l2(−α) with the normalization ⟨l1(α)|l2(−α)⟩ = 1/(1−α²). It derives a four-component decomposition — a real symmetric α-QMT, an imaginary antisymmetric α-Berry curvature, and two 'flipped' tensors — and constructs gauge-invariant connections Γ1 and Γ2 that are claimed to satisfy a generalized ±α duality (Eqs. (48), (51), (54)). The same construction is then transferred to non-Hermitian systems, where LR/RL/LL/RR Fubini–Study tensors are classified and quantum natural-gradient optimization for complex-valued cost functions is discussed. The α=0 limit and the trivial-phase limit are checked against the standard QMT and the classical α-connections.","tokens_in":32341,"tokens_out":7337,"duration_ms":78313,"significance":"If the technical issues identified below are resolved, the framework would be a useful phenomenological extension of quantum information geometry: it gives explicit formulas for a deformed metric, Berry curvature, and connections in a unified biorthogonal language, and the four-way tensor classification usefully organizes several inequivalent definitions that exist in the non-Hermitian literature. The manuscript is candid about the lack of gauge invariance of the bare connections and about the role of normalization, and the limiting checks are correct as far as they go. However, the load-bearing question of whether the α-deformed object is actually a metric is not settled for the α domain advertised, and the biorthogonal normalization assumption is stated but not justified as a dynamically preserved condition. The paper would also be materially strengthened by a concrete physical example showing which of the four tensor components are nonvanishing.","major_comments":[{"comment":"Eq. (35) defines g^{(α)}_{ij} = (1/4) E[∂_i l_θ ∂_j l_θ] + ((1−α)/(1+α))(E[∂_i φ ∂_j φ] − E[∂_i φ] E[∂_j φ]). The phase covariance is positive semidefinite, but its coefficient is negative for |α|>1, so the statement immediately after Eq. (35) that the phase contribution is 'necessarily non-negative' is false outside the interval −1<α<1. For a pure-phase state at α=2, g^{(2)}_{θθ} is negative, so the object is not a metric, and the steepest-descent reading of Eqs. (57) and (66) is not defined. Please either restrict α to the interval where positivity holds and prove that restriction, or explicitly present the sign-indefinite case as a pseudo-Riemannian/indefinite structure rather than a quantum metric.","section":"Section III B 1, Eq. (35)"},{"comment":"The LR (and RL) construction, the four-component decomposition, and the natural-gradient equations (66) all depend on the pointwise biorthogonal normalization ⟨Ψ_L|Ψ_R⟩ = 1, together with the analogous α-normalization ⟨l1(α)|l2(−α)⟩ = 1/(1−α²). The text itself acknowledges immediately after Eq. (59) that this normalization is not guaranteed for evolution generated by a non-Hermitian Hamiltonian, and Appendix A supplies constraints only for a single-parameter family with commuting generators. The classification should therefore be stated as conditional on the state trajectory remaining in the normalization class, or the paper should prove invariance of that class for the physical evolutions considered. Without this, the claimed systematic classification of non-Hermitian QGTs is narrower than the presentation suggests.","section":"Section IV A, Eq. (59) and Section III B"},{"comment":"The 'derivation' of the α-FS tensor from the α-QFI is definitional, because ρ^{(α)} = |l2(−α)⟩⟨l1(α)| and the trace normalization were chosen so that (1−α²)² Tr[ρ^{(α)} ∂_i ρ^{(α)} ∂_j ρ^{(α)}] reduces to the overlap expression (31). The construction is internally consistent, but the claim that the object is obtained from a Fisher-information-like quantity requires at least one property of a genuine quantum Fisher information (for example, monotonicity under stochastic maps or a Cramér–Rao-type bound), or an explicit statement that this is only a formal analogue.","section":"Section III B 4, Eq. (42)"}],"minor_comments":[{"comment":"In the paragraph after Eq. (39), 'the moralisation is maintained throughout the evolution' should read 'the normalisation is maintained'; the typo makes the sentence difficult to parse.","section":"Section III B 3"},{"comment":"The lengthy expressions (27)–(30) are central objects of the first construction but are presented without derivation or reference to an appendix. Please provide the intermediate steps, or at least describe the structure of the calculation, so that the formulas can be checked without rederiving the whole expansion.","section":"Section III A, Eqs. (27)–(30)"},{"comment":"There are numerous typographical spacing errors ('di fferent', 'a ffect', 'e ffect') and inconsistent notation for symmetrization ('sym' versus explicit (ij) indices). A careful proofread is needed before publication.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a single-author formal construction with many intermediate algebraic steps left to the reader. The main risk is interpretive rather than purely algebraic: the object called the α-QMT is not always positive definite for the advertised α domain, and the central classification depends on a normalization condition that is not preserved by generic non-Hermitian dynamics. I would ask the authors to sharpen the positivity/domain statement and to make the scope of the normalization assumption explicit in the abstract or introduction. A small worked example (for instance, a two-level PT-symmetric model) would substantially help readers see which of the four tensor components are physically relevant."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this one. First, the biorthogonal α-deformed Fubini-Study construction is genuinely new, at least in the form presented here, and it reduces correctly in the α=0 and trivial-phase limits. Second, the central metric claim in Eq. (35) is wrong for |α|>1, because the phase covariance enters with prefactor (1−α)/(1+α), which is negative outside that interval. The paper says the phase contribution is necessarily non-negative; that is only true for −1<α<1. For a pure-phase state at α=2, g(2)_θθ is negative, so the \"α-QMT\" is indefinite. Since the natural-gradient optimization and the whole duality picture lean on this tensor being a metric, the unrestricted α-domain is a load-bearing flaw.\n\nWhat's good: the construction is explicit and coherent. The flipped tensors in (35)-(37), the gauge-invariant connections (49)-(50), and the four-way LR/RL/LL/RR classification are all clearly laid out. The author is also honest: he flags that the bare duality (48) is not physically meaningful, that affine-coordinate interpretation is open, and that the biorthogonal normalization <ΨL|ΨR>=1 is not guaranteed by non-Hermitian dynamics. That self-awareness is real credit.\n\nSoft spots, in proportion. The positivity issue is the big one; it needs either a stated and proved restriction to |α|<1 or a different argument. The biorthogonal normalization is indeed load-bearing, and the author acknowledges it but doesn't resolve it; for a paper that claims a geometry of non-Hermitian state spaces, this is a real limitation, not a footnote. Several long formulas, (27)-(30) and (52)-(53), are stated without derivation; that's fine for a first pass but should be filled in. Finally, the α-FS tensor is derived from an α-QFI defined precisely so the duality (51) holds by construction, so the depth of that result is modest.\n\nWho is this for? People working on quantum information geometry, natural gradient methods, or non-Hermitian QGT. It's a useful starting point, but it needs a careful referee. I'd send it to review, with the expectation of significant revision. My own verdict is conditional, not accept.","headline":"A genuinely new biorthogonal α-deformed geometry, but the central metric claim breaks for |α|>1 and the normalization assumption is load-bearing.","tokens_in":32862,"tokens_out":3489,"would_cite":false,"duration_ms":32351,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53B05","81Q70","53Z05"],"pacs":["03.65.-w","03.65.Vz","02.40.Ky"],"model":"deepseek-v4-flash","headline":"Biorthogonal states let non-Hermitian quantum systems carry a four-part Fubini–Study geometry.","keywords":["quantum information geometry","Fubini–Study metric","non-Hermitian quantum mechanics","biorthogonal formalism","Berry curvature","quantum geometric tensor","alpha-connections","quantum natural gradient"],"falsifier":"Take a two-level non-Hermitian Hamiltonian whose parameters trace a closed loop around an exceptional point, start from a biorthogonal pair with $\\langle\\Psi_L|\\Psi_R\\rangle=1$, and check whether the overlap remains 1 along the path; if it drifts from unity, the left–right Fubini–Study tensor, its four-piece decomposition, and the associated natural-gradient equations are not defined for that evolution, and the classification would need a different normalisation prescription.","tokens_in":31699,"feed_emoji":"🌀","tokens_out":8835,"duration_ms":83690,"temperature":0.7,"pith_summary":"The paper sets out to transfer the two pillars of classical information geometry — a metric on probability distributions and a dual family of ±α connections — to the space of pure quantum states, and claims this is possible once the inner product is made biorthogonal rather than Hermitian. It constructs a one-parameter family of Fubini–Study tensors whose real symmetric part plays the role of a quantum metric and whose gauge-invariant connections satisfy a generalised ±α duality relative to that metric, with the phase of the wavefunction entering explicitly. For non-Hermitian Hamiltonians, the same machinery classifies the quantum geometric tensor into four irreducible pieces: real symmetric, imaginary antisymmetric, imaginary symmetric, and real antisymmetric, the last two being new 'flipped' tensors that vanish in the Hermitian limit. A reader should care because this gives a single geometric language for non-Hermitian state spaces and shows which part of the geometry drives natural-gradient optimisation when the cost function is complex-valued.","feed_headline":"Four tensor types emerge when quantum geometry goes non-Hermitian","feed_subtitle":"A biorthogonal construction yields dual connections, a complex metric, and Berry curvature in one framework.","key_machinery":"The central object is the biorthogonal Fubini–Study tensor, built from a left state and a right state paired by the normalisation $\\langle\\Psi_L|\\Psi_R\\rangle=1$ rather than by complex conjugation. Its $\\alpha$-deformed version uses the two asymmetric functions $l_1^{(\\alpha)}=P^{(1-\\alpha)/2}e^{i(1-\\alpha)\\phi}/(1-\\alpha)$ and $l_2^{(-\\alpha)}$, whose mutual inner product is fixed by normalisation. Expanding the biorthogonal overlap integral supplies the metric at second order and two connections at third order; subtracting gauge-dependent terms yields invariant connections. The decomposition of the tensor into four rank-two pieces — symmetric/antisymmetric times real/imaginary — carries the classification, and the identity $\\mathrm{Re}[\\Gamma_1^{(\\alpha)}+\\Gamma_2^{(-\\alpha)}]=2\\Gamma^{(c)}$ carries the duality claim. The same machinery produces the optimisation statement: the real symmetric part of the tensor sets the natural-gradient flow for the real part of a complex cost, and the imaginary symmetric part sets the flow for its imaginary part.","core_discovery":"The central claim is that the Fubini–Study tensor can be consistently generalised by replacing the Hermitian inner product with a biorthogonal pairing. With the pointwise normalisation $\\langle\\Psi_L|\\Psi_R\\rangle=1$, the left–right tensor $FS^{LR}_{ij}=\\langle\\partial_i\\Psi_L|\\partial_j\\Psi_R\\rangle-\\langle\\partial_i\\Psi_L|\\Psi_R\\rangle\\langle\\Psi_L|\\partial_j\\Psi_R\\rangle$ decomposes into four gauge-invariant pieces: the real symmetric quantum metric $g^{LR}_{ij}$, the imaginary antisymmetric Berry curvature $\\omega^{LR}_{ij}$, and two 'flipped' tensors $\\tilde g^{LR}_{ij}$ and $\\tilde\\omega^{LR}_{ij}$ that are nonzero only because the left and right states are not conjugate. The same construction with $\\alpha$-deformed states $l_1^{(\\alpha)}$ and $l_2^{(-\\alpha)}$, normalised so that $\\langle l_1^{(\\alpha)}|l_2^{(-\\alpha)}\\rangle=(1-\\alpha^2)^{-1}$, yields an $\\alpha$-quantum metric and an $\\alpha$-Berry curvature whose real symmetric and imaginary symmetric parts govern separate natural-gradient flows. For the gauge-invariant connections built from the overlap expansion, the paper proves the duality $\\mathrm{Re}[\\Gamma_1^{(\\alpha)}+\\Gamma_2^{(-\\alpha)}]=2\\Gamma^{(c)}$, an exact quantum analogue of the classical $\\pm\\alpha$ duality, and shows that in the non-Hermitian setting the curvature of the complex Berry connection is the sum of the imaginary antisymmetric part and the flipped real antisymmetric part.","pith_inferences":["If the classification holds, it provides a principled way to resolve the existing ambiguity in definitions of the quantum metric tensor for non-Hermitian systems: choose the inner product and normalisation, and the fourfold decomposition fixes the rest.","The two incompatible natural-gradient flows suggest a least-squares or Pareto-style compromise for optimising complex costs in non-Hermitian variational algorithms, a direction the paper flags as open.","Because the flipped tensors vanish for Hermitian dynamics, their magnitude could serve as a diagnostic for proximity to exceptional points or the breakdown of unitarity along a parameter trajectory.","The $\\alpha$-deformed overlap construction may be the seed of a bona fide quantum $\\alpha$-divergence defined from biorthogonal overlaps rather than from density-matrix monotones, which would connect this geometry to parameter-estimation bounds."],"forward_implications":["For any non-Hermitian Hamiltonian with a biorthogonal pair of states, the quantum geometric tensor splits into four tensor pieces; the two 'flipped' pieces vanish identically in the Hermitian limit and can therefore serve as signatures of genuinely non-Hermitian geometry.","The gauge-invariant connections $\\Gamma_1^{(\\alpha)}$ and $\\Gamma_2^{(-\\alpha)}$ satisfy $\\mathrm{Re}[\\Gamma_1^{(\\alpha)}+\\Gamma_2^{(-\\alpha)}]=2\\Gamma^{(c)}$, recovering the classical $\\pm\\alpha$ duality with explicit phase contributions included.","Quantum natural-gradient descent for a complex cost splits into two generally incompatible flows, one governed by the real symmetric part of the non-Hermitian Fubini–Study tensor and one by the imaginary symmetric part.","For an associated left–right pair, imaginary-time evolution reduces to natural-gradient descent only when the generator is biorthogonally Hermitian; otherwise the two optimisation directions disagree.","The left–right, right–left, left–left, and right–right tensors are inequivalent, so fixing the inner product and normalisation determines which metric and curvature are physically relevant."],"supporting_citations":[{"why":"Supplies the classical α-connection duality, affine-coordinate picture, and the statistical manifold framework that the paper generalises.","marker":"[1]"},{"why":"Supplies the overlap-expansion method that defines the Fubini–Study tensor from nearby-state inner products.","marker":"[18]"},{"why":"Gives the explicit coordinate form of the Fubini–Study metric and its phase-modified Fisher information structure, the α=0 baseline.","marker":"[25]"},{"why":"Provides the biorthogonal inner product and associated-state construction on which the left–right Fubini–Study tensor is built.","marker":"[54]"},{"why":"Defines distinct left–right and right–left Berry curvatures in non-Hermitian systems that the classification here encompasses and systematises.","marker":"[57]"},{"why":"Gives the gauge-invariant connections for the Hermitian α=0 case whose non-Hermitian extension the paper constructs.","marker":"[59]"},{"why":"Establishes quantum natural-gradient descent for Hermitian systems, the baseline the paper extends to complex-valued costs.","marker":"[61]"},{"why":"Documents the inequivalent definitions of quantum metric tensor in non-Hermitian systems, motivating the fourfold classification.","marker":"[68]"}],"fun_headline_variants":["Non-Hermitian geometry yields four tensor types","Biorthogonal pairing yields dual quantum connections","Four tensor types from a generalised Fubini-Study metric","Non-Hermitian quantum geometry: dual connections and Berry curvature","Four tensor types from non-Hermitian state space geometry"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the biorthogonal overlap $\\langle\\Psi_L(\\theta)|\\Psi_R(\\theta)\\rangle$ stays exactly equal to 1 at every point of the parameter manifold, a normalisation that evolution by a non-Hermitian Hamiltonian does not automatically preserve.","fun_headline_variants_meta":{"raw":{"variants":["Non-Hermitian geometry yields four tensor types","Biorthogonal pairing yields dual quantum connections","Four tensor types from a generalised Fubini-Study metric","Non-Hermitian quantum geometry: dual connections and Berry curvature","Four tensor types from non-Hermitian state space geometry"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000641,"raw_usage":{"total_tokens":3056,"prompt_tokens":1156,"completion_tokens":1900,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":772,"completion_tokens_details":{"reasoning_tokens":1820}},"tokens_in":772,"tokens_out":1900,"duration_ms":13487,"temperature":1.0,"reasoning_tokens":1820,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:12:18.814645+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a two-level non-Hermitian Hamiltonian whose parameters trace a closed loop around an exceptional point, start from a biorthogonal pair with $\\langle\\Psi_L|\\Psi_R\\rangle=1$, and check whether the overlap remains 1 along the path; if it drifts from unity, the left–right Fubini–Study tensor, its four-piece decomposition, and the associated natural-gradient equations are not defined for that evolution, and the classification would need a different normalisation prescription.","supporting_citations":[{"cited_title":"Nagaoka, Methods of information geometry, V ol","cited_arxiv_id":null,"evidence_quote":"Supplies the overlap-expansion method that defines the Fubini–Study tensor from nearby-state inner products."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the explicit coordinate form of the Fubini–Study metric and its phase-modified Fisher information structure, the α=0 baseline."},{"cited_title":"Hasegawa, Non-commutative extension of the informa- tion geometry, inQuantum Communications and Measurement, edited by V","cited_arxiv_id":null,"evidence_quote":"Gives the gauge-invariant connections for the Hermitian α=0 case whose non-Hermitian extension the paper constructs."},{"cited_title":"Jen ˇcov´a, Reports on Mathematical Physics 47, 121 (2001)","cited_arxiv_id":null,"evidence_quote":"Establishes quantum natural-gradient descent for Hermitian systems, the baseline the paper extends to complex-valued costs."},{"cited_title":"Scalar Curvature of the Quantum Exponential Family for the Transverse-Field Ising Model and the Quantum Phase Transition","cited_arxiv_id":"2212.12919","evidence_quote":"Documents the inequivalent definitions of quantum metric tensor in non-Hermitian systems, motivating the fourfold classification."}],"review_version":2}