{"id":"1ad5d36e-18ac-4ea0-a882-b5f824ac81d6","arxiv_id":"2606.14873","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A variational tensor-network framework for continuum 1D models that encodes non-orthogonal finite-element basis overlaps as an MPO and solves a generalized DMRG eigenvalue problem.","lead":"This paper presents a matrix-product-state method for simulating quantum particles in a one-dimensional continuum using overlapping 'tent' basis functions, handling the non-orthogonality by encoding all overlaps in a compact tensor operator. The method keeps a strict energy upper bound and adds an exact grid-refinement step, useful for cold-atom and impurity problems.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Monotonic variational convergence across grid refinements is not guaranteed after particle-number truncation; the formal claim in Sec. VII A overstates an empirical observation.","rationale":"The reader's weakest assumption correctly identifies that the particle-number cutoff breaks the nested-subspace property under grid refinement. After careful review of the MPO construction in Sec. IV, the exact representation of N for a fixed grid is valid: the virtual index bounds are independent of L, the Cholesky factor is banded, and the contraction W†W yields the claimed bond dimension. The generalized eigenvalue problem is well-posed since N is positive definite. However, the paper's language in Sec. VII A ('strictly variational upper bound that approaches the continuum limit monotonically') and the multigrid claims in Sec. VI B imply a formal monotonicity that the construction does not guarantee. The refinement map R produces occupations up to 2n_c, and truncation back to n_c projects out of the exact image, so the fine truncated variational space is not a superset of the coarse one. Thus the monotonic convergence observed in the benchmarks is an empirical finding, not a theorem. This does not undermine the core method for a fixed grid, but it does affect the advertised multigrid and variational-convergence advantages, warranting a conditional acceptance with a request to qualify the claim and provide evidence for monotonicity across the tested regime.","tokens_in":23970,"tokens_out":39015,"duration_ms":342983,"concrete_test":"Run the FE-MPS ground-state search (n_c=2, D=15) for the Gaussian-barrier benchmark of Fig. 8 with Ω=500 for L=100,200,400,800. If E(L) is not strictly decreasing, the monotonic variational claim is refuted. If it is decreasing, the concern stands as an unproven empirical regularity; additionally, compute the maximum local occupation of R|Ψ_L> for the coarse-grid ground state to confirm it exceeds n_c, demonstrating the exact refined state lies outside the fine truncated variational space.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central MPO construction (Sec. IV) is mathematically sound for bosons: the overlap N = W†W is exactly represented with bond dimension χ^2 independent of L. The load-bearing weakness lies in the stronger claim of strict variational monotonic convergence across grid refinements (Sec. VII A). The exact refinement map R (Sec. VI B) maps coarse-grid computational states to fine-grid states with local occupations up to 2n_c (Eqs. 48-49). Since the fine-grid variational space is truncated back to n_c, the image of the coarse truncated space under R followed by truncation is not a superset of the fine truncated space; hence the fine-grid energy is not guaranteed to be below the coarse-grid energy. The paper acknowledges the truncation error ('does not introduce significant error') but does not prove nesting; the observed monotonicity in Figs. 5,6,8 is empirical. This weakens the formal variational/multigrid guarantees advertised in the abstract and Sec. VII A, though it does not invalidate the generalized eigenvalue solver for a fixed grid.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an MPS framework for one-dimensional continuum quantum many-body systems using a non-orthogonal, localized single-particle basis. The physical Fock space is mapped to an auxiliary canonical computational space via an operator W, so that the many-body overlap N = W^†W becomes the metric of a generalized eigenvalue problem H|Φ> = E N|Φ>. The central technical result is that N can be represented exactly as an MPO with bond dimension independent of the number of basis functions L when the single-particle overlap matrix is banded with bandwidth R. The authors then implement a generalized DMRG solver and apply it to the Lieb-Liniger gas in a first-order finite-element basis, including inhomogeneous potentials and a Gaussian barrier. They also introduce an exact refinement map for multigrid optimization and present benchmarks against Bethe-ansatz and LDA results.","tokens_in":24300,"tokens_out":7505,"duration_ms":87390,"significance":"The main result—an exact, compact MPO representation of the many-body overlap for non-orthogonal local basis sets—is a genuine methodological advance. It enables a variational, local, non-orthogonal discretization of continuum models without the need to orthogonalize the basis and destroy locality. The paper provides detailed derivations, reproducible code and data (Julia packages and Zenodo archive), and benchmarks against independent Bethe-ansatz/LDA results. The finite-element MPS approach also correctly captures UV observables such as Tan's contact, which is a nontrivial check. If the formal claims are appropriately qualified, this is a solid contribution with clear value for continuum DMRG applications.","major_comments":[{"comment":"The claim that FE-MPS 'provides a strictly variational upper bound that approaches the continuum limit monotonically' is not formally supported by the construction. The exact refinement map R maps coarse-grid computational states to fine-grid states with local occupations up to 2n_c (Eqs. 48–49). In practice the fine grid is truncated back to the cutoff n_c, as acknowledged in Sec. VI B ('we enforce a fixed cutoff n_c on the refined state'). Consequently, the truncated fine-grid variational space is not a superset of the image of the coarse truncated space under R, and the refined-grid energy is not guaranteed to be below the coarse-grid energy. The observed monotonicity in Figs. 5, 6, and 8 is empirical, not a formal consequence of the method. The paper's own statements that truncation 'does not introduce significant error' and that compression 'typically introduces <5% energy error' ar","section":"Sec. VI B and Sec. VII A, Eqs. (48)–(49)"},{"comment":"The fermionic sign factor ξ_i is introduced with a short counting argument but no derivation or proof that it factorizes into local contributions as in Eq. (26). Since the paper explicitly claims to generalize the framework to fermionic statistics, this is a load-bearing point for that generalization. The absence of any fermionic benchmark makes it difficult for the reader to verify correctness. Please provide a complete derivation (e.g., by induction on the number of sites and on R) or state clearly that the fermionic extension is conjectural and defer numerical validation to later work.","section":"Sec. IV, Eq. (26)"}],"minor_comments":[{"comment":"Typo: 'an upper an upper-triangular F' should read 'an upper-triangular F'.","section":"Sec. III"},{"comment":"The bond dimension of N is stated as χ^2, but the construction would benefit from an explicit diagram showing the contraction W^†W and the local physical dimensions (n_c+1)^2. This would clarify why the intermediate output dimension (R+1)n_c of W does not enter the final MPO bond dimension.","section":"Sec. IV, after Eq. (28)"},{"comment":"The horizontal axis label in panel (a) appears garbled ('k 0 500 100 0'); the intended tick labels should be cleaned up.","section":"Fig. 7"},{"comment":"The environment rescaling procedure is described qualitatively. A few details on how the extensive logarithmic factors are stored and combined when computing energies would improve reproducibility.","section":"Sec. V A"},{"comment":"The Fourier transform convention for the tent functions should be stated explicitly, as the prefactor depends on the normalization of the Fourier transform.","section":"Appendix C, Eq. (C8)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is technically strong on its central MPO construction and numerical benchmarks. The main issue is overstatement of the formal guarantees: the monotonic variational convergence across grid refinements is not actually proven after the n_c truncation, and the paper itself contains passages acknowledging the truncation. This is fixable with careful rewording and a truncation-error analysis, so I do not recommend rejection. The fermionic sign-factor derivation should also be completed or explicitly deferred."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper does what it says. It takes the overlap-MPO construction from [arXiv:2405.10285] and generalizes it from chiral fermions to bosons and arbitrary local orbitals, then uses it to run variational DMRG in a first-order finite-element basis for the Lieb-Liniger model. The central result—an exact MPO representation of the many-body overlap with bond dimension independent of L—is derived cleanly and the benchmarks hold up. I checked the algebra for the R=1 case and it works; the benchmarks against Bethe ansatz and LDA are convincing, and the UV momentum tail is a nice practical payoff. The authors also ship code and raw data on Zenodo, which makes this reproducible. The generalization from the same group's prior paper is handled honestly—they re-derive the construction rather than importing it as a black box, so the self-citation is not a circularity problem.\n\nThe main soft spot is a formal claim that outruns the construction. Section VII A states that FE-MPS converges monotonically to the continuum limit across grid refinements. That is observed in the numerics, but it is not guaranteed by the variational principle once the local occupation cutoff is imposed. The exact refinement map in Sec VI B produces fine-grid occupancies up to 2n_c (Eqs. 48-49), and the code then truncates back to n_c. After that truncation, the refined variational space is not a superset of the coarse one, so the fine-grid energy is not provably below the coarse-grid energy. The paper mentions that the truncation 'does not introduce significant error' but doesn't try to prove nesting. For a fixed grid, the generalized eigenvalue solver is on firm ground; the multigrid story is 'good practice plus strong numerics' rather than a theorem. I'd ask the authors to either prove a weaker form of monotonicity or rewrite the abstract/claim to say 'empirically monotonic'. The fermionic sign factor (Eq. 26) is derived but never tested, which is minor given the bosonic focus but worth flagging. The error analysis is also on the thin side—they report gradient norms and convergence rates but no systematic DMRG error bars.\n\nBottom line: this is a solid technique paper that deserves a serious referee. I'd send it out and I'd cite it. The formal overreach is real but contained, and the numerical evidence is strong enough to carry the paper while the authors fix the wording.","headline":"A sound and useful extension of the overlap-MPO construction to bosonic finite-element MPS; treat the claimed monotonic multigrid convergence as empirical rather than proven.","tokens_in":24715,"tokens_out":3249,"would_cite":true,"duration_ms":34096,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Non-orthogonal single-particle bases can be used directly in DMRG: the many-body overlap operator is exactly representable as a matrix product operator with bond dimension set by the basis bandwidth, not by system size.","keywords":["matrix product states","non-orthogonal basis","finite element method","generalized eigenvalue problem","DMRG","Lieb-Liniger model","continuum many-body systems","multigrid optimization"],"falsifier":"For a tridiagonal overlap matrix (R=1) with L=20 and n_c=2, build the dense many-body overlap matrix N and compare it element-by-element with the MPO constructed from the paper's tensors; any element with relative error above machine precision would disprove the claimed exact representation. Similarly, apply the refinement map R to a coarse-grid state with an occupied site, verify that R† N_fine R = N_coarse holds to machine precision, and check that truncating the refined state at n_c changes the physical norm by more than the reported few-percent level.","tokens_in":23912,"feed_emoji":"⚛️","tokens_out":9422,"duration_ms":88973,"temperature":0.7,"pith_summary":"This paper develops a matrix product state method for one-dimensional continuum quantum many-body systems that keeps a strict variational principle without forcing the basis to be orthogonal. The authors show that for any local single-particle basis whose overlap matrix is banded—such as the tent functions of a first-order finite-element discretization—the many-body overlap operator that encodes the non-orthogonality can be written exactly as a matrix product operator whose bond dimension depends only on the bandwidth and the local occupation cutoff, not on the number of basis functions. This recasts the ground-state search as a generalized eigenvalue problem solved by a generalized density matrix renormalization group sweep. Applied to the Lieb-Liniger gas in traps and barriers, the method gives energies that are strict upper bounds, converge monotonically with grid spacing, resolve sub-grid potentials, and reproduce ultraviolet properties such as the k^-4 momentum tail. The same construction yields an exact coarse-to-fine grid refinement map, enabling multigrid optimization.","feed_headline":"Overlaps compress into a fixed-size MPO for continuum DMRG","feed_subtitle":"Finite-element DMRG for the Lieb-Liniger gas becomes variational, monotonic, and multigrid-ready.","key_machinery":"The central object is the many-body overlap operator N=W†W acting on an auxiliary canonical Fock space. The construction that carries the argument is the explicit local MPO tensor for W, obtained from a Cholesky factorization N=F†F with banded F; the virtual index Q_{i,r} at each bond is the number of particles crossing that bond whose destination lies r sites to the left. This particle-flow picture turns the multinomial expansion of the basis change into nearest-neighbor tensors, with multinomial coefficients and (for fermions) a locally computable sign, so that W and hence N become matrix product operators. The bond dimension of N is the square of the number of allowed flow states per bond","core_discovery":"For a non-orthogonal single-particle basis with band-diagonal overlap matrix N (bandwidth R), the many-body overlap operator N=W†W admits an exact MPO representation with bond dimension χ=(∏_{r=1}^R(n_c r+1))^2, independent of the number of basis functions L. The proof is constructive: the basis change from the physical non-orthogonal modes to the computational canonical modes is implemented by a many-body operator W that acts locally, with Cholesky superdiagonals F of N, and whose matrix elements are organized through a particle-flow picture: virtual indices Q_{i,r} count how many particles cross the bond between sites i and i+1 on their way to a site r positions to the left. With this MPO","pith_inferences":["The same overlap-MPO machinery should transfer to quantum-chemistry tensor networks built from localized atomic orbitals, where overlap matrices are naturally banded; the generalized DMRG would then avoid Lowdin orthogonalization and its delocalizing effects.","Because the metric N is fixed and sparse-structured, the scheme could be extended to real-time evolution via a time-dependent variational principle, enabling non-equilibrium studies of impurities or quenches in non-orthogonal bases.","The exact refinement map suggests an adaptive h-refinement strategy where only selected cells are refined, with the small occupation-truncation error monitored locally; this would concentrate the O(Δx) error where the wavefunction is cusped.","The explicit bond-dimension formula makes the cost of higher-order elements predictable: going to quadratic elements (R=2) with n_c=2 squares the bond dimension relative to R=1, so the reported O(Δx) convergence of tent functions is a practical sweet spot unless the wavefunction is smooth enough to justify a larger overlap bandwidth."],"forward_implications":["For any 1D continuum Hamiltonian projected onto a finite-element basis, the ground-state energy is an upper bound to the continuum energy, and the bound becomes systematically tighter as the grid is refined.","The banded-overlap MPO construction applies to fermions as well as bosons, and to any local orbital family beyond tent functions, as long as the single-particle overlap matrix has a small bandwidth.","The coarse-to-fine refinement map is exact at the many-body level before occupation truncation, so a coarse-grid solution can be lifted to a fine grid to seed DMRG, yielding converged results at multiple resolutions in a single run.","Observables such as momentum distributions and Tan's contact are computed directly in the continuum, recovering the physical k^-4 tail that finite-difference MPS cannot capture.","The algorithm scales as O(L d χ D^3) with χ independent of L, so for a fixed per-element accuracy the cost grows only linearly with the number of basis functions."],"fun_headline_variants":["MPO for continuum overlaps: bond dimension fixed, not system size","Finite-element DMRG: overlaps become size-independent MPOs","Exact MPO for non-orthogonal overlap in continuum DMRG","Continuum DMRG: band-diagonal overlap yields fixed-size MPO","Lieb-Liniger with DMRG: overlap MPO independent of lattice size"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The central construction is exact only under two conditions that must hold simultaneously: the single-particle overlap matrix has small, system-size-independent bandwidth, and the bosonic Fock space is truncated at a finite occupation n_c; if either fails, the MPO bond dimension or the variational bound after grid refinement loses its guarantees.","fun_headline_variants_meta":{"raw":{"variants":["MPO for continuum overlaps: bond dimension fixed, not system size","Finite-element DMRG: overlaps become size-independent MPOs","Exact MPO for non-orthogonal overlap in continuum DMRG","Continuum DMRG: band-diagonal overlap yields fixed-size MPO","Lieb-Liniger with DMRG: overlap MPO independent of lattice size"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000749,"raw_usage":{"total_tokens":3149,"prompt_tokens":698,"completion_tokens":2451,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":442,"completion_tokens_details":{"reasoning_tokens":2365}},"tokens_in":442,"tokens_out":2451,"duration_ms":15641,"temperature":1.0,"reasoning_tokens":2365,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T11:24:07.524108+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a tridiagonal overlap matrix (R=1) with L=20 and n_c=2, build the dense many-body overlap matrix N and compare it element-by-element with the MPO constructed from the paper's tensors; any element with relative error above machine precision would disprove the claimed exact representation. Similarly, apply the refinement map R to a coarse-grid state with an occupied site, verify that R† N_fine R = N_coarse holds to machine precision, and check that truncating the refined state at n_c changes the physical norm by more than the reported few-percent level.","supporting_citations":[],"review_version":1}