{"id":"62868dcf-da1a-484d-8cfb-06acf40d9143","arxiv_id":"2607.28753","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"GA-BMPS unify Gaussian states and finite-dimensional MPS in one ansatz with closed-form expectation values and exact parent Hamiltonians.","lead":"This paper defines a new family of bosonic many-body states, GA-BMPS, that contains both Gaussian states and finite-dimensional matrix-product states, and shows how to compute their properties and construct exact parent Hamiltonians. The result is a truncation-free variational framework for 1D bosonic systems and lattice scalar field theory.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Propositions 1–3 omit bra-side complex conjugation: the transfer matrix E = Σ A_n ⊗ A_n computes Σ Tr(...)^2, not the quantum norm Σ |Tr(...)|^2; Eq. (34) fails for complex V,K,L, e.g. D=1, K=0, ℓ=1+i.","rationale":"The reader's weakest assumption targeted the parent-Hamiltonian kernel of Lemma 3 (a rigor gap in the global ODE argument). While that gap is real and worth addressing, my reading identifies a more fundamental and concrete flaw: the transfer-matrix formulas that underpin claim (ii) omit the complex conjugation on the bra side. This is not a matter of domain subtlety—it is a mismatch with the definition of the quantum inner product, demonstrable in the D=1 coherent-state limit. Since 'expectation values can be efficiently computed' is one of the three headline features of the ansatz, and the numerical tests rely on these formulas, the manuscript in its present form does not support a central claim. The flaw is likely fixable by replacing the left-side V,K,L with their entrywise conjugates in the definition of the transfer matrix (i.e., using ar A_n ⊗ A_n), and the numerics may already have done so implicitly, but the derivations as written are incorrect. This raises the correctness risk substantially, so the appropriate verdict moves from CONDITIONAL to REJECT, pending correction of the transfer-matrix calculus.","tokens_in":30015,"tokens_out":43301,"duration_ms":394510,"concrete_test":"Evaluate Proposition 1 for D=1, V=1, K=0, L=ℓ=1+i. Equation (34) yields E = exp(ℓ²) = exp(2i), whereas the exact norm of the BMPS state |ψ⟩ = e^{ℓ a†}|0⟩ is ⟨ψ|ψ⟩ = exp(|ℓ|²) = exp(2). If the transfer matrix were computing the quantum norm, these must agree. An even more decisive test: take D=2, K=0, L=diag(ℓ₁,ℓ₂) with complex ℓ₁≠ℓ₂, and V with complex off-diagonal entries; compute ⟨Ψ|Ψ⟩ from Eq. (34) and compare with the exact Fock-space contraction Σ_{n₁,...,n_N} |Tr(B A_{n₁}…A_{n_N})|². Any discrepancy confirms the missing conjugation.","verdict_should_be":"REJECT","load_bearing_attack":"The central computational claim (ii) rests on the transfer-matrix calculus of Sec. III C. Equation (33) defines E = Σ_n A_n ⊗ A_n, and Proposition 1 evaluates it as V⊗V E(L⊗1, K⊗1, 1⊗K, 1⊗L), where E(P,Q,R,S) = ⟨0| e^{P a} e^{Q a^2} e^{R(a†)^2} e^{S a†}|0⟩. For a quantum state |Ψ⟩ = Σ Tr(B A_n)|n⟩, the norm is Σ |Tr(B A_n)|^2, which requires the bra tensor to be the entrywise conjugate: ar A_n, i.e., a transfer matrix Σ ar A_n ⊗ A_n. The paper instead uses the same A_n on both sides, which computes the bilinear form Σ Tr(BA_n) Tr(BA_n). This is not the quantum inner product when the tensors are complex; the matrices V,K,L are allowed to be complex (M_D(C), Def. 1). The discrepancy appears already for D=1: the exact norm of e^{κ(a†)^2+ℓ a†}|0⟩ is (1−4|κ|^2)^{-1/2} exp((|ℓ|^2+κℓ̄^2+κ̄ℓ^2)/(1−4|κ|^2)), while Prop. 1 gives (1−4κ^2)^{-1/2} exp((ℓ^2+2κℓ^2)/(1−4κ^2)). For κ=0, ℓ=1+i these differ (e^{|ℓ|^2}=e^2 vs. e^{ℓ^2}=e^{2i}). Propositions 2 and 3 inherit the same issue because the O-transfer and Gaussian-unitary insertions are built on the same un-conjugated E(P,Q,R,S). Thus the derived expectation values are not quantum expectation values for general complex parameters. The numerics in Sec. V (e.g., Example 5 identifying a complex V) appear to contradict this, suggesting the implemented code used the correct conjugated formulas, but the text as written does not support claim (ii).","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a variational family for one-dimensional bosonic systems, the Gaussian-augmented bosonic matrix-product states (GA-BMPS). A GA-BMPS is built from a finite-dimensional auxiliary space and site operators of the form V e^{K\\otimes(a^\\dagger)^2} e^{L\\otimes a^\\dagger} applied to the Fock vacuum, optionally followed by a passive Gaussian unitary. The authors claim three properties: (i) the family contains all pure Gaussian states and finite-dimensional MPS; (ii) expectation values can be computed efficiently through a closed-form transfer-matrix calculus (Propositions 1-3); and (iii) exact parent Hamiltonians can be constructed as simple functions of the canonical operators (Section IV). They also present variational tests for the lattice phi^4 model and a parent-Hamiltonian recovery example (Section V). The paper is clearly written, the algebraic framework is appealing, and the parent-Hamiltonian construction is systematic and independent of the transfer-matrix calculus.","tokens_in":30570,"tokens_out":12859,"duration_ms":129643,"significance":"If correct, the GA-BMPS ansatz would be a useful truncation-free variational family that unifies Gaussian states and finite-dimensional MPS, with a rare combination of closed-form contractions and exact parent Hamiltonians. The parent-Hamiltonian construction via ambient annihilators and check operators is particularly interesting and goes beyond the standard finite-dimensional projector construction. However, the central computational claim (ii) is not supported as written because the transfer-matrix formulas omit the complex conjugation on the bra side. This error affects Propositions 1-3 and therefore the numerical section, although the Appendix C algebra is repairable by conjugating the left tensor factor. The claimed variational results in Section V appear to require the corrected (conjugated) transfer matrices, which are not stated in the text. The contribution is potentially valuable, but the manuscript needs a substantial, focused revision of the transfer-matrix calculus before the main claims can be accepted.","major_comments":[{"comment":"Equation (33) defines E = \\sum_n A_n \\otimes A_n and Eq. (34) evaluates it as V \\otimes V E(L\\otimes 1, K\\otimes 1, 1\\otimes K, 1\\otimes L). For complex tensors, this is the bilinear transfer matrix, not the quantum one: the bra tensor must be the entrywise conjugate \\overline{A_n}. For |\\Psi\\rangle = \\sum_n Tr(BA_n)|n\\rangle, the norm is \\sum_n |Tr(BA_n)|^2, whereas the paper computes \\sum_n Tr(BA_n)^2. The failure is already visible for D=1, K=0, L=1+i: Eq. (34) gives e^{\\ell^2}=e^{2i}, while the correct norm is e^{|\\ell|^2}=e^2, as stated in Eq. (9). Propositions 2 and 3 inherit the same defect. The Appendix C calculus can be repaired by setting P=\\overline{L}\\otimes 1, Q=\\overline{K}\\otimes 1, R=1\\otimes K, S=1\\otimes L, with conjugate sources; nevertheless, claim (ii) is unsupported as written.","section":"Section III.C, Eqs. (33)-(34), Propositions 1-3"},{"comment":"The numerical results are described as being obtained from 'the dominant fixed points of the transfer operator' defined in Section III.C. If that operator is the unconjugated one of Eq. (33), the minimized functional is not the physical energy; it can be complex and is not bounded below, so the reported variational energies and parameter recovery do not follow from the paper's formulas. Either the implementation used the corrected conjugated transfer matrices, in which case those formulas must be given in the main text, or the numerics are not evidence for the physical variational claim. The text should also reconcile this with Eq. (9), which explicitly gives the correct conjugated norm for the D=1 case.","section":"Section V, Fig. 1 and Table I"},{"comment":"The finite-MPS transfer-matrix review also writes E = \\sum_i A^i \\otimes A^i without conjugation. Unless the tensors are assumed real, the standard expression is E = \\sum_i \\overline{A^i} \\otimes A^i. This notational omission is the root cause of the error in Propositions 1-3. Please correct the convention throughout the manuscript and state it explicitly, since Definition 1 and Example 5 use complex V,K,L.","section":"Section II.B and Eq. (17)"}],"minor_comments":[{"comment":"The step 'Consequently, dim ker Q \\le n' deserves one clarifying sentence: an entire Bargmann-space function is determined by its germ at z_0 by the identity theorem, so local ODE uniqueness at a regular point bounds the global kernel even when the Wronskian has zeros. The claim itself appears sound.","section":"Appendix E, Lemma 3"},{"comment":"The O-transfer matrix should also be written with a conjugated bra: E_O = \\sum_{i,j} \\langle i|O|j\\rangle \\overline{A_i} \\otimes A_j, and the source-derivative identities in Proposition 2 should be updated accordingly.","section":"Section III.C, Eq. (35)"},{"comment":"For the multimode factorization E = E_1 \\cdots E_M, please specify that after the conjugation fix each E_\\nu is built from \\overline{V}, \\overline{K}_\\nu, \\overline{L}_\\nu on the bra side. The current notation is ambiguous about the ordering of bra and ket factors.","section":"Section III.F, Eq. (52)"},{"comment":"The target V in Example 5 is complex, but L-BFGS-B is a real-variable solver. Please state how the complex parameters were encoded (e.g., real and imaginary parts) and how automatic differentiation through the fixed-point calculation handles complex variables, so the numerical procedure is reproducible.","section":"Section V"}],"recommendation":"major_revision","confidential_remarks":"The central computational claim of the paper is currently undermined by the missing conjugation in the transfer-matrix calculus. The error is specific and fixable, and the parent-Hamiltonian part is independent and appears sound, so I recommend major revision rather than rejection. Please ask the authors to provide corrected conjugated transfer-matrix formulas for Propositions 1-3, to state explicitly which formulas were used in the numerics, and to check that the reported phi^4 and recovery results remain unchanged with the corrected calculus."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the ansatz is a good idea, and the parent-Hamiltonian construction is a real contribution. But the transfer-matrix section, which supports the paper's main computational claim, has a load-bearing bug as written.\n\nThe paper defines the transfer matrix as E = Σ_n A_n ⊗ A_n (Eq. 33) and builds Props. 1–3 on it. For a quantum state in the Fock basis with complex amplitudes, the norm requires the bra tensor to be entrywise conjugated: the correct object is Σ_n \\bar{A}_n ⊗ A_n (up to index placement). Using the same A_n on both sides computes the bilinear form, not the inner product. Since V, K, L are allowed to be complex (Definition 1), this is not a harmless convention. The counterexample is immediate at D=1, K=0, ℓ=1+i: Prop. 1 gives e^{ℓ²} = e^{2i} for the single-site norm, while the true norm is e^{|ℓ|²} = e². Eq. (9) in the paper itself contains the correct Gaussian norm, confirming that the issue is missing conjugation in the transfer matrix, not a different normalization.\n\nConsequently, claim (ii) in the abstract—that expectation values can be efficiently computed—is not supported by the text for the full family. The numerics in Sec. V appear to use the correct conjugated formulas (the complex V in Example 5 could not have worked otherwise), so the executable code may be fine, but the paper as written does not match the code.\n\nWhat the paper does well: the ansatz itself, the inclusion of Gaussian states and finite-dimensional MPS, the equivalence with MPS of photon-added Gaussian states (Prop. 4), and especially the parent-Hamiltonian construction in Sec. IV. The Q and F_j machinery in Appendices D–E is careful and, as far as I can tell, correct. The kernel argument in Lemma 3 is sketched regarding global properties of the Wronskian, but that looks fillable.\n\nVerdict: this deserves peer review, but it needs a serious revision. Fix the conjugation, rederive Props. 1–3, re-run the numerics, and the paper could be solid. As it stands, the central computational claim fails for complex parameters.","headline":"The GA-BMPS ansatz and parent Hamiltonians are worth taking seriously, but the transfer-matrix calculus misses complex conjugation and thus fails for the complex parameters the paper allows.","tokens_in":31024,"tokens_out":4749,"would_cite":false,"duration_ms":45858,"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":"The paper introduces Gaussian-augmented bosonic matrix-product states, a single ansatz that contains all pure Gaussian states and all finite-dimensional matrix-product states, with closed-form transfer matrices and exact parent Hamiltonians","keywords":["bosonic matrix-product states","Gaussian states","parent Hamiltonian","transfer matrix","photonic Fock space","variational methods","tensor networks","photon-added squeezed states"],"falsifier":"Pick a single-mode family of photon-added squeezed coherent states with degenerate squeezing parameters so that the Wronskian Δ(z) has zeros, construct Q by Lemma 3, and compute dim ker Q on the Fock-space domain. If for any parameter choice dim ker Q exceeds the number n of local states, ker(h0+hR) is larger than the target subspace W and the parent-Hamiltonian construction is not exact.","tokens_in":29936,"feed_emoji":"⚛️","tokens_out":6127,"duration_ms":59547,"temperature":0.7,"pith_summary":"The paper defines a family of bosonic many-body states—Gaussian-augmented bosonic matrix-product states—in which the virtual matrices of an MPS are interleaved with exponential factors that act like matrix-valued squeezing and displacement. It claims three things: the family contains every pure Gaussian state and every finite-dimensional matrix-product state as special cases; expectation values can be computed from closed-form transfer matrices, so variational calculations need no local Fock-space cutoff; and every state in the family is the exact ground state of a frustration-free parent Hamiltonian written as simple functions of the bosonic ladder operators. If these claims hold, this is a practical and conceptual unification of two of the main variational frameworks for one-dimensional quantum systems.","feed_headline":"Bosonic ansatz unifies Gaussian states and matrix-product states","feed_subtitle":"Expectation values and exact parent Hamiltonians are closed-form, opening truncation-free bosonic calculations.","key_machinery":"The GA-BMPS ansatz (Definition 1): a matrix-product state over the infinite Fock space whose auxiliary-space matrices sandwich exponentials K⊗(a†)^2 and L⊗a†, with an optional passive linear-optical unitary across all modes. Its two workhorses are the closed-form transfer matrix E, which turns contraction of an infinite-dimensional tensor network into a finite matrix calculation, and the Wronskian-based single-mode operators Q and F_j: Q annihilates exactly the local basis {|e_j⟩}, and F_j sends |e_k⟩ to δ_{jk}|0⟩. Together these construct the parent Hamiltonian h0+hR with kernel equal to the desired local MPS subspace.","core_discovery":"The central claim is that a bosonic MPS generated by exponentials of matrix-valued quadratic and linear terms—V e^{K⊗(a†)^2} e^{L⊗a†}|0⟩, followed by a passive linear-optical unitary—forms a self-contained framework. For commuting K and L, the transfer matrix collapses to a closed-form Gaussian integral, E = (V⊗V) Δ^{1/2} exp(Δ(K⊗L^2 + L⊗L + L^2⊗K)) with Δ = (1−4K⊗K)^{-1}, so norms and local observables are evaluated exactly without truncating Fock space. The same family admits exact parent Hamiltonians: Wronskian-built annihilation-operator polynomials Q and interpolation operators F_j give a positive local term whose kernel is exactly the local subspace of the state. Numerical tests show t","pith_inferences":["The closed-form transfer matrix should make the ansatz compatible with gradient-based time-evolution methods, such as a time-dependent variational principle, giving a truncation-free route to dynamics that the paper leaves as future work.","The linear-independence and kernel assumptions behind the Wronskian construction could be monitored numerically as a conditioning diagnostic; singular Wronskians would signal parameter regions where the parent Hamiltonian may acquire extra ground states.","The coexistence of Gaussian and MPS subfamilies suggests the ansatz could support a bosonic analogue of injectivity-based phase classification, where the overlap matrix of photon-added squeezed states plays the role of the physical inner product.","A concrete testable extension is to benchmark GA-BMPS against standard truncated-tensor-network calculations for Bose-Hubbard chains at high filling, where truncation artifacts are most severe and the advantage of exact expectation values should be largest."],"forward_implications":["Every pure Gaussian state and every finite-dimensional MPS is a subfamily, so the ansatz interpolates between the two standard variational tools for bosons.","Norm, local polynomial observables, and Gaussian-unitary operators have closed-form transfer matrices, allowing variational optimization in the full Fock space without a hard occupation cutoff.","Every BMPS—and every GA-BMPS obtained by a passive unitary—is the exact ground state of a frustration-free parent Hamiltonian whose terms are simple functions of a and a†; locality survives when the passive unitary is a finite-depth local Gaussian circuit.","Numerical results show on-site squeezing lowers the variational energy in the lattice φ^4_2 model and that the parent-Hamiltonian construction is practical, recovering a target D=2 MPS to numerical precision.","The same tensor-network structure extends to multiple boson species per site, mixed spin-boson chains, and—formally—higher-dimensional bosonic tensor networks."],"fun_headline_variants":["One bosonic ansatz for Gaussian and matrix-product states","Efficient bosonic states unify Gaussian and MPS classes","Exact parent Hamiltonians for a unified bosonic state family","Gaussian-augmented bosonic MPS with closed-form expectations"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The exact parent-Hamiltonian construction relies on the chosen local photon-added squeezed states being linearly independent and on the Wronskian-built operator Q having kernel exactly their span on the whole relevant Fock-space domain; if that kernel is larger, the parent Hamiltonian acquires extra ground states.","fun_headline_variants_meta":{"raw":{"variants":["One bosonic ansatz for Gaussian and matrix-product states","Efficient bosonic states unify Gaussian and MPS classes","Exact parent Hamiltonians for a unified bosonic state family","Gaussian-augmented bosonic MPS with closed-form expectations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000318,"raw_usage":{"total_tokens":1573,"prompt_tokens":625,"completion_tokens":948,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":369,"completion_tokens_details":{"reasoning_tokens":877}},"tokens_in":369,"tokens_out":948,"duration_ms":7510,"temperature":1.0,"reasoning_tokens":877,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T00:31:06.680662+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Pick a single-mode family of photon-added squeezed coherent states with degenerate squeezing parameters so that the Wronskian Δ(z) has zeros, construct Q by Lemma 3, and compute dim ker Q on the Fock-space domain. If for any parameter choice dim ker Q exceeds the number n of local states, ker(h0+hR) is larger than the target subspace W and the parent-Hamiltonian construction is not exact.","supporting_citations":[],"review_version":1}