{"id":"aa86be13-f8e0-4277-8bfc-f35a86c16c83","arxiv_id":"1908.09761","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A finite sum of continuous matrix product states with different boundary operators, labeled by an ancilla, can express the continuum limit of every matrix product state, which standard continuous matrix product states cannot.","lead":"This paper introduces a generalized continuous matrix product state, a new family of quantum states, and proves it can represent the continuum limit of any matrix product state, closing a gap left open by prior work. The result matters because it provides a tensor network ansatz defined directly in the continuum, relevant for variational studies of one-dimensional quantum fields.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Transfer-matrix equality in Eq. (27) does not by itself establish state-level representability; Section 5.3 concedes many continuum states share a transfer matrix, and the ancilla-traced physical state can differ from the target pure state.","rationale":"The paper's core construction is transparent and the transfer-matrix computation in Eq. (27) is correct: orthonormality of the ancilla states kills cross terms, and the sum of Bi tensor Bi-bar is P. The worked examples are consistent with that computation, and the candid discussion in Section 5.3 shows the authors are aware of the non-uniqueness of states with a given transfer matrix. The load-bearing gap is that Theorem 4 is worded as a statement about states, not just about transfer matrices or correlation functions. Since the proof never goes beyond (Phi_R, Phi_R) = E_|R|, and since the paper itself concedes that many continuum states share E_|R|, the representability claim rests on an unstated equivalence relation. This matters physically because global coherence between sectors (as in the ferromagnetic example) is not visible in transfer-matrix correlation functions; the ansatz stores it in ancilla coefficients, so whether the original pure state has been represented depends on whether the ancilla is included and how it is read out. This is not a fatal flaw: the theorem can be read as an expressibility statement for quasi-local data, and the authors' later discussion moves in that direction. But as written, the central claim is over-stated and should be made conditional on a precise equivalence notion. I therefore keep the reader's conditional verdict, with the same weakest assumption identified.","tokens_in":18257,"tokens_out":9474,"duration_ms":101466,"concrete_test":"For a non-trivial L != 0 case (e.g. the bracket state of Eqs. (45)/(47)), construct both the continuum-limit state of the MPS by applying the p-refinement isometries of Definition 2 on finite N, and the generalized cMPS |Phi_R> of Eq. (47). Trace the ancilla out of |Phi_R> and compare the resulting reduced density matrix on H_R with the target state on a global observable such as the full-interval parity or the purity. If the ancilla-traced state is mixed while the target is pure, or if a non-local order parameter differs, transfer-matrix equality is insufficient and Theorem 4 needs an explicit ancilla projection or a quasi-local equivalence statement; if all chosen observables agree, the concern is settled.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The main result, Theorem 4, is proven only at the level of the auxiliary-space transfer matrix: Eq. (27) computes (Phi_R, Phi_R) = P e^{|R|L}, which matches the transfer matrix of the MPS continuum limit. But the statement is that the continuum limit state is 'represented' by |Phi_R>. No argument shows that a state with the same transfer matrix is the same state; Section 5.3 explicitly says 'there be many states in the continuum whose transfer matrix is E_{|R|}' and then characterizes only freedom in P, L, Kraus operators, ancilla basis, and Fock space, not a state-level identification. The issue is not merely philosophical: in the ferromagnetic example (Example 1), the coherence of |0...0> + |1...1> is carried by the ancilla superposition |v0> + |v1>, while the physical state after tracing the ancilla is the vacuum. If the ancilla is part of the physical description, the Hilbert space differs from that of the MPS and an embedding is needed; if it is not, the reduced state is a mixture rather than the original pure superposition. Thus the load-bearing premise—that transfer-matrix matching suffices for the representability claim—is imported without proof, and the exact sense of equivalence is left unspecified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a generalized continuous Matrix Product State (cMPS) ansatz that is claimed to be able to express the continuum limit of any translationally invariant MPS with a continuum limit. The ansatz is a sum of cMPS, each with a different boundary operator given by a Kraus operator of the projector quantum channel P appearing in the transfer matrix E_a = P e^{aL}, together with an ancilla label. The main result, Theorem 4, is that the transfer matrix of this generalized cMPS is exactly P e^{|R|L}, matching the transfer matrix of the MPS continuum limit. The paper also interprets the ansatz as a concatenation of a state at the closure of the set of cMPS and a standard cMPS, and illustrates the construction with several examples, including superpositions of ferromagnetic states, the completely depolarizing channel, and a 'bracket state'.","tokens_in":18350,"tokens_out":5997,"duration_ms":57918,"significance":"If the representability claim is established at the level of states, this would be a clean resolution of a limitation of cMPS identified in Ref. [19]: the missing projector P in the transfer matrix. The construction is elegant and the transfer-matrix computation in Eq. (27) is direct and correct. The paper also gives an explicit construction (Proposition 1) showing that every projector quantum channel arises as the infinite-time limit of a Markovian channel, which is a useful technical result. However, the central gap between transfer-matrix equality and state-level representability currently prevents Theorem 4 from being fully convincing.","major_comments":[{"comment":"The proof of Theorem 4 establishes only that the transfer matrix of the constructed state |Φ_R> equals P e^{|R|L}. The theorem, however, claims that the continuum limit state of |V_N(A)> is represented by |Φ_R>. Section 5.3 explicitly acknowledges that many continuum states share a given transfer matrix ('there be many states in the continuum whose transfer matrix is E_{|R|}') and then characterizes freedom in P, L, Kraus operators, the ancilla basis, and the Fock space, but it does not provide a state-level identification. To make the theorem sound, the authors must either define precisely what 'represented' means (e.g., equality of all quasi-local correlation functions, or convergence under the refining isometries of Definition 2) and prove that the generalized cMPS satisfies this definition, or restrict the theorem's statement to a transfer-matrix-level statement. Without this, the central claim is not proven.","section":"Section 5.2, Theorem 4 and Eq. (27)"},{"comment":"The ambiguity between transfer-matrix and state-level equivalence is concretely visible in Example 1. For the superposition of ferromagnetic states, the generalized cMPS is |Φ_R> = (|v0> + |v1>) ⊗ |Ω_R>. If the ancilla is traced out, the physical state is the vacuum, not a coherent global superposition; if the ancilla is kept, the state lives in C^K ⊗ H_R and no embedding of this space into the original physical space H_R is specified. The paper leaves the physical interpretation of the ancilla as an open question, which is acceptable, but the statement of Theorem 4 requires a clear specification of the sense in which the ancilla-carrying state represents the original MPS continuum limit.","section":"Section 5.3 and Example 1"},{"comment":"Theorem 4 states that 'for any N' the continuum limit state of |V_N(A)> can be represented by the generalized cMPS. The proof, however, only verifies the transfer matrix on a segment of length a, and then uses the consistency condition (E_a)^N = E_{Na} to extend to longer segments. This establishes translational consistency of the transfer matrix but does not show that the sequence of generalized cMPS states, as N varies, is related by the p-refinement and blocking operations that define the continuum limit in Definition 2. Please clarify how the N-dependence of |Φ_R> is meant to match the definition of continuum limit.","section":"Section 5.2 and 5.3, N-dependence"}],"minor_comments":[{"comment":"The notation switches between |φ_R[B_i,...]> (physical state after tracing the auxiliary space, used in Eq. (24)) and φ_R[B_i,...] (operator with open auxiliary indices, used in Eq. (25)) without an explicit reminder. Please introduce a clear notational distinction, as this is a common source of confusion in cMPS calculations.","section":"Eqs. (24) and (25)"},{"comment":"The sentence 'the first element can be seen as the thermodynamic limit of a cMPS in the thermodynamic limit' is redundant; one of the two occurrences should be removed.","section":"Section 5.4"},{"comment":"The inner product in Eq. (26) is defined on (C^K ⊗ M_D ⊗ H_R) × (C^K ⊗ M_D ⊗ H_R), but the state |Φ_R> in Eq. (24) is an element of C^K ⊗ H_R after the auxiliary trace. The proof implicitly uses the open-index version Φ_R of Eq. (25). Please state explicitly that the transfer matrix is defined using the open-index object, as is standard for cMPS.","section":"Eq. (27)"},{"comment":"The proof of Proposition 1 treats three special cases and then says that 'putting these three building blocks together' gives the general case. This is plausible, but the text would benefit from a short explicit statement of why the cases (i)-(iii) suffice to cover the general form of P in Eq. (9), especially regarding the interplay between the blocks π_k and the isometries V_k.","section":"Appendix A, proof of Proposition 1"}],"recommendation":"major_revision","confidential_remarks":"The core gap is that Theorem 4 is stated as a state-representability claim but proved only as a transfer-matrix identity. This is a correctable issue if the authors are willing to reformulate the theorem (e.g., in terms of equality of all correlation functions or a precise embedding of the ancilla space) and to address the Example 1 ambiguity. The paper is otherwise a solid contribution to the cMPS literature and extends the authors' own previous work in a natural way. I would not recommend rejection, but the revision needs to resolve the representability issue before the main claim is accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nMain thing you should know: Theorem 4 is a real result. The authors construct a generalized cMPS — a sum of ordinary equal-dynamics cMPS with Kraus-operator boundary conditions and an ancilla label — whose transfer matrix is exactly P e^{aL}, the canonical form of an infinitely divisible channel. The computation in Eq. (27) is short and sound: orthonormality of the ancilla collapses the double sum to P. This directly addresses the gap left by Ref. [19] and is not in the earlier literature. Genuinely new.\n\nWhat the paper does well: it states the characterization from [19] cleanly, gives working examples (especially the bracket state), and is honest that PL=PLP is not built into the ansatz and that the ancilla's physical meaning is open. The freedom in choosing P, L, Kraus operators, and Fock space is discussed in useful detail.\n\nWhere the soft spots are: the claim that this ansatz \"represents the continuum limit state\" goes beyond what the proof establishes. The proof only matches transfer matrices; Section 5.3 explicitly concedes that many continuum states share a transfer matrix. The ferromagnetic example shows the problem: the coherence between |0...0> and |1...1> is carried by the ancilla superposition, and tracing the ancilla leaves only the vacuum. Whether the ancilla is physical or auxiliary determines whether you get a pure superposition or a mixture. The paper never defines the equivalence relation between a generalized cMPS in C^K ⊗ H_R and the MPS continuum limit state. I think this is fixable — the framework of [19] probably supplies the right notion, namely that the continuum limit is characterized by the transfer matrix — but the paper needs to say so explicitly. As written, a careful reader can't tell what \"represented\" means.\n\nThe other issue is smaller: Appendix A's proof of Proposition 1 leaves several steps as \"immediate\" or \"easily verified,\" including the fixed-point uniqueness of a particular Liouvillian. Those steps are likely correct, but for a theorem that's supposed to hold for general P, the reader shouldn't have to fill in the work.\n\nShould you engage? Yes. This is a solid contribution to tensor networks in the continuum, and the central construction will be cited. It deserves peer review; a good referee should push for a precise statement about what is proved at the state level, and a filled-in Appendix A. I'd be interested in citing it once those points are addressed.\n\nBest,\n[Name]","headline":"A clean and useful transfer-matrix construction that plausibly closes a known representability gap, but the paper should say more precisely what it means for a generalized cMPS to 'represent' a continuum limit state.","tokens_in":19123,"tokens_out":4351,"would_cite":true,"duration_ms":45094,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45"],"pacs":[],"model":"deepseek-v4-flash","headline":"A generalized cMPS ansatz expresses the continuum limit of every MPS.","keywords":["continuous matrix product states","matrix product states","continuum limit","infinitely divisible quantum channels","projector quantum channels","Lindblad Liouvillian","tensor networks","superselection sectors"],"falsifier":"Take the MPS family $|0\\cdots0\\rangle+|1\\cdots1\\rangle$, whose transfer matrix is $P$ alone, and compute its continuum limit directly by applying the p-refinement isometries. Compare that state with the generalized cMPS states obtained from two different Kraus decompositions of $P$; Example 1 gives different states, such as $(|v_0\\rangle+|v_1\\rangle)\\otimes|\\Omega\\rangle$ and $|w_0\\rangle\\otimes|\\Omega\\rangle$. If the refined limit disagrees with one of these representations in a global or ancilla-sensitive observable, then transfer-matrix equality does not uniquely determine the continuum limit state and the state-level reading of Theorem 4 would be refuted.","tokens_in":17849,"feed_emoji":"🧮","tokens_out":12751,"duration_ms":117174,"temperature":0.7,"pith_summary":"Continuous matrix product states (cMPS) miss the continuum limits of many matrix product states (MPS), because a cMPS transfer matrix is always a Markovian channel $e^L$, whereas an MPS with a continuum limit has transfer matrix $P e^{aL}$ with $P$ a projector quantum channel. This paper claims that the missing projector can be supplied by a generalized ansatz: a finite sum of cMPS with boundary operators given by the Kraus operators of $P$, each attached to an ancilla state. The ansatz reproduces the transfer matrix $P e^{aL}$ of any MPS continuum limit, and is interpreted as a standard cMPS concatenated with an element at the closure of the set of cMPS that supplies $P$. If correct, it gives every translationally invariant MPS with a continuum limit an exact matrix-product description directly in the continuum.","feed_headline":"Generalized cMPS ansatz captures every MPS continuum limit","feed_subtitle":"Adding Kraus-operator boundary terms to cMPS restores exact continuum descriptions for all MPS with a limit.","key_machinery":"The load-bearing object is the generalized cMPS ansatz of Eq. (24): a sum over the Kraus operators $\\{B_i\\}$ of the projector quantum channel $P$, with each $B_i$ used as the boundary operator of a cMPS sharing the same $Q$ and $\\{R_\\alpha\\}$, and each term carrying a distinct ancilla state $|v_i\\rangle\\in C^K$. The argument turns on the transfer-matrix identity $(\\Phi_R,\\Phi_R)=P e^{aL}$, because the defining property of an MPS continuum limit is precisely the factorization $E_a=P e^{aL}$ with $PL=PLP$. The condition $PL=PLP$ ensures consistency $\\bigl(P e^{aL}\\bigr)^N=P e^{NaL}$, so every point of the segment is well defined. Proposition 1 supplies the physical picture: $P$ is a limit point of cMPS transfer matrices $e^{t\\tilde L}$ built from explicit jump operators, making the generalized ansatz a concatenation of a closure-of-cMPS object and a standard cMPS.","core_discovery":"Theorem 4 is the central claim. Let $\\mathcal V(A)=\\{|V_N(A)\\rangle\\}$ be a translationally invariant MPS family with a continuum limit in the sense of Definition 2, so its transfer matrix is $E_a=P e^{aL[Q,\\{R_\\alpha\\}]}$ with $P^2=P$ a projector quantum channel, $L[Q,\\{R_\\alpha\\}]$ a Lindblad Liouvillian, and $PL=PLP$. If $P=\\sum_{i=1}^K B_i\\otimes \\bar B_i$ is a Kraus decomposition of $P$, then the continuum limit state is represented by the generalized cMPS $|\\Phi_R[\\{v_i\\},\\{B_i\\},Q,\\{R_\\alpha\\}]\\rangle=\\sum_{i=1}^K |v_i\\rangle\\otimes |\\varphi_R[B_i,Q,\\{R_\\alpha\\}]\\rangle$, where $\\{|v_i\\rangle\\}$ is an orthonormal basis of the ancilla space $C^K$. The proof evaluates the transfer matrix $(\\Phi_R,\\Phi_R)=\\bigl(\\sum_i B_i\\otimes\\bar B_i\\bigr)e^{aL}=P e^{aL}$, matching the continuum limit. A further result, Proposition 1, shows that every projector channel $P$ equals $\\lim_{t\\to\\infty} e^{t\\tilde L}$ for an explicit Lindblad generator, so the projector part of the ansatz can be seen as a cMPS in the thermodynamic limit or with unbounded matrix norm.","pith_inferences":["The proof establishes equality of transfer matrices; the step from transfer-matrix equality to equality of states is imported from the continuum-limit framework of earlier work and is not re-proved here, so Theorem 4 is strictly a statement about transfer matrices and quantities determined by them unless that premise holds.","A narrower ansatz with the condition $PL=PLP$ built into the state structure may exist; the authors connect this to G-injective MPS, where superpositions with different boundary conditions reproduce the ground-state degeneracy of a parent Hamiltonian, hinting at a continuum parent Hamiltonian problem.","The physical role of the ancilla space $C^K$ is left open; identifying an observable that reads the ancilla would turn the generalized ansatz from a mathematical representation into a usable physical variational class.","One concrete stress test is to compare the continuum limit obtained by explicit p-refinement with the generalized cMPS for a case where $L=0$, such as the ferromagnetic superposition; any difference in global coherence between the two sectors would show that transfer-matrix matching alone does not fix the state."],"forward_implications":["Every translationally invariant MPS that has a continuum limit receives an exact continuum matrix-product description, closing the expressivity gap that plain cMPS left open.","The correlation-function calculus of cMPS carries over with one replacement: the boundary term $B\\otimes\\bar B$ becomes the projector $P$, so existing cMPS computational tools apply almost unchanged.","The projector's zeros, which act as superselection rules, can be imposed directly at the continuum level instead of by sending matrix norms or system sizes to infinity.","The ansatz gives a variational family for continuum theories with several superselection sectors, illustrated by superpositions of ferromagnetic states, the completely depolarizing channel, and the bracket state."],"supporting_citations":[{"why":"introduces cMPS, the continuous matrix-product ansatz that the generalized ansatz extends.","marker":"[13]"},{"why":"supplies the cMPS transfer matrix and inner product used to compute $(\\Phi_R,\\Phi_R)$, as well as the regularity conditions inherited by the generalized cMPS.","marker":"[14]"},{"why":"gives Definition 2 of the continuum limit and the characterization $E_a=P e^{aL}$ that the new ansatz is designed to match.","marker":"[19]"},{"why":"provides the characterization of infinitely divisible quantum channels underlying Theorem 2 and hence the structure of continuum-limit transfer matrices.","marker":"[25]"},{"why":"supplies the Lindblad form of Liouvillians and the fixed-point structure of projector channels used in Proposition 1 and in the discussion of $PL=PLP$.","marker":"[23]"},{"why":"justifies the assumption that the MPS transfer matrix can be taken trace-preserving without loss of generality.","marker":"[28]"}],"fun_headline_variants":["cMPS with boundary terms now reproduces all MPS continuum limits","New cMPS ansatz closes gap: includes projectors via ancilla","Generalized cMPS: any MPS limit becomes expressible","From MPS to cMPS: projector bridge found","cMPS ansatz with ancilla states matches all continuum limits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof shows the generalized cMPS has the same transfer matrix as the continuum limit, then assumes that matching transfer matrices is enough to identify the state; that identification is carried over from earlier work and is not proven in this paper.","fun_headline_variants_meta":{"raw":{"variants":["cMPS with boundary terms now reproduces all MPS continuum limits","New cMPS ansatz closes gap: includes projectors via ancilla","Generalized cMPS: any MPS limit becomes expressible","From MPS to cMPS: projector bridge found","cMPS ansatz with ancilla states matches all continuum limits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000257,"raw_usage":{"total_tokens":1610,"prompt_tokens":1006,"completion_tokens":604,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":622,"completion_tokens_details":{"reasoning_tokens":512}},"tokens_in":622,"tokens_out":604,"duration_ms":5513,"temperature":1.0,"reasoning_tokens":512,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:04:17.470370+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the MPS family $|0\\cdots0\\rangle+|1\\cdots1\\rangle$, whose transfer matrix is $P$ alone, and compute its continuum limit directly by applying the p-refinement isometries. Compare that state with the generalized cMPS states obtained from two different Kraus decompositions of $P$; Example 1 gives different states, such as $(|v_0\\rangle+|v_1\\rangle)\\otimes|\\Omega\\rangle$ and $|w_0\\rangle\\otimes|\\Omega\\rangle$. If the refined limit disagrees with one of these representations in a global or ancilla-sensitive observable, then transfer-matrix equality does not uniquely determine the continuum limit state and the state-level reading of Theorem 4 would be refuted.","supporting_citations":[{"cited_title":"Or ´us, Ann","cited_arxiv_id":null,"evidence_quote":"introduces cMPS, the continuous matrix-product ansatz that the generalized ansatz extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the Lindblad form of Liouvillians and the fixed-point structure of projector channels used in Proposition 1 and in the discussion of $PL=PLP$."},{"cited_title":"Continuous matrix product states for non-relativistic quantum fields: a lattice algorithm for inhomogeneous systems","cited_arxiv_id":"1801.02219","evidence_quote":"justifies the assumption that the MPS transfer matrix can be taken trace-preserving without loss of generality."}],"review_version":1}