{"id":"b2bffed3-818d-4f55-87c4-9c1316a982cc","arxiv_id":"2608.08821","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Finite-depth measurement-feedback preparation of 1D matrix product states is classified by pushable virtual-bond defects and their pushing relations, yielding explicit circuits and links to non-invertible symmetries.","lead":"This paper develops a framework for preparing one-dimensional quantum states with measurement and feedback circuits, classifying states by the virtual-bond defects that can be pushed through their matrix product tensors. It connects these classes to non-invertible symmetries such as Kramers-Wannier and Kennedy-Tasaki dualities.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Rank-1 FDLU criteria for associated states are algebraically wrong (Secs. V B 1, V C 1, V C 2), so the 1-round vs 2-round classification boundaries are unreliable.","rationale":"The abstract's central claim is that pushable defects and their pushing relations classify finite-depth MF-preparable states and dictate their preparation circuits. That classification is made quantitative by Lemma 1 and Lemma 2: an associated state is FDLU-preparable exactly when its transfer matrix has rank 1, and this rank condition decides whether the target state requires one or two MF rounds. The rank-1 algebra errors in Secs. V B 1 and V C 2 are therefore load-bearing: they misclassify specific families and produce asserted preparation circuits that are not supported by the stated lemmas. This is an internal inconsistency, not a mere disagreement with an external consensus. The reader's stated weakest assumption, the injectivity condition after Eq. (19), is a reasonable place to scrutinize, but the more decisive and unambiguous problem is the rank test itself; even granting the fundamental-theorem step, the classification boundaries fail. The reader did flag rank-1 conditions in the verdict rationale, though not as the named weakest assumption, so my agreement is partial. Because the framework may still be correct after correcting these rank conditions and re-deriving the affected examples, the appropriate verdict remains CONDITIONAL rather than REJECT.","tokens_in":43740,"tokens_out":17884,"duration_ms":204768,"concrete_test":"Recompute the eigenvalues of T_B = λ0 I + λ1 Z1 + λ2 Z2 + λ3 Z1Z2 for the numerical point λ0=2, λ1=λ2=1, λ3=1/2, and verify that λ0λ3=λ1λ2 holds while rank(T_B)=4; this refutes the Sec. V B 1 rank-1 criterion. Then diagonalize T_B = c I + c' Z⊗Z from Eq. (96) for c'=0 and for c'=c, confirming that the rank is 4 in the first case and 1 in the second; this settles the Sec. V C 2 boundary and determines whether the associated states in those examples are actually FDLU-preparable.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central classification is operationalized by Lemmas 1 and 2: |A> is 1-round MF-preparable iff the associated transfer matrix T_B is rank 1. The paper twice applies incorrect rank diagnostics. In Sec. V B 1 and V C 1, T_B = λ0 I + λ1 Z1 + λ2 Z2 + λ3 Z1Z2, and rank 1 is asserted from λ0λ3 = λ1λ2. In the computational basis T_B is diagonal with eigenvalues λ0 ± λ1 ± λ2 ± λ3; rank 1 requires three of these four eigenvalues to vanish. The single condition λ0λ3 = λ1λ2 does not imply that. For example, λ0=2, λ1=λ2=1, λ3=1/2 satisfies λ0λ3=λ1λ2 but yields eigenvalues 4.5, 1.5, 0.5, and 0.5, so T_B has full rank 4. Separately, Sec. V C 2 states that c' = <ψ_r|Z⊗I|ψ_r> = 0 makes T_B = c I + c' Z⊗Z rank 1; actually c'=0 gives T_B ∝ I, which has full rank, while rank 1 occurs when c' = ±c (up to normalization), not when c'=0. These errors misidentify the RI–FDLU and RZ–FDLU boundaries and the λ3=0 boundary of RZ–RIRX, directly changing which states are claimed to be FDLU- or 1-round-MF-preparable and invalidating the claimed 1-round circuits for those boundary states.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a tensor-network framework for preparing one-dimensional matrix product states with constant-depth circuits that combine measurements and unitary feedback. The central objects are pushable virtual defects; the paper defines associated states |B>, shows that a 1-round MF circuit relating |B> to |A> exists when T_B is FDLU-preparable, classifies the possible pushing relations (RI, RZ, RIRX, RZRX, RILI, etc.), and claims completeness for open-boundary MPS under left-conditioned feedback (Theorem 1). It further connects the resulting classes to non-invertible dualities, including Tambara-Yamagami, Kennedy-Tasaki, and generalized cosine symmetries, with examples such as AKLT and symmetry-broken/cluster states.","tokens_in":44098,"tokens_out":23194,"duration_ms":249445,"significance":"If correct, the paper would supply a systematic, constructive classification of finite-depth MF-preparable states in one dimension and a concrete circuit-synthesis recipe, going beyond the fusion-measurement protocols of Refs. [19,20,28,29]. The derivations are analytical and self-contained rather than fitted; Theorem 1 is an attractive completeness statement; and the connection to non-invertible symmetries, together with explicit examples (AKLT via KT, Rep(D8) symmetry breaking, non-onsite symmetry breaking), makes the paper potentially important. However, several rank diagnostics used to decide FDLU-preparability of the associated states are algebraically incorrect, and these diagnostics are load-bearing for the 1-round versus 2-round classification.","major_comments":[{"comment":"The condition λ0λ3=λ1λ2 is asserted to imply that T_B=λ0I+λ1Z1+λ2Z2+λ3Z1Z2 is rank 1. This is false. In the eigenbasis of Z1,Z2 the eigenvalues are λ0+λ1+λ2+λ3, λ0+λ1−λ2−λ3, λ0−λ1+λ2−λ3, and λ0−λ1−λ2+λ3; rank 1 requires three of these to vanish. The product condition is only necessary, not sufficient. For example, (λ0,λ1,λ2,λ3)=(2,1,1,1/2) satisfies λ0λ3=λ1λ2 but gives eigenvalues 9/2, 3/2, 1/2, 1/2, so T_B has full rank. Consequently the RI–FDLU boundary includes states whose associated B is not FDLU-preparable, and the claimed 1-round circuit for those states is not justified.","section":"Sec. V B 1, Eq. (84)"},{"comment":"The same incorrect rank diagnostic is used for the RZ–FDLU class: the text states that when λ0λ3=λ1λ2 the transfer matrix T_B is of rank 1, but T_B in Eq. (91) has the same form λ0I+λ1Z1+λ2Z2+λ3Z1Z2 as in the RI case, so the same counterexample applies. The RZ–FDLU class is therefore misidentified: states with, e.g., (λ0,λ1,λ2,λ3)=(2,1,1,1/2) do not have an FDLU-preparable associated B and are not established to be 1-round MF-preparable by this argument.","section":"Sec. V C 1, Eq. (91)"},{"comment":"The claim that ⟨ψ_r|Z⊗I|ψ_r⟩=0 (equivalently λ3=0 in Eq. (91)) makes T_B=cI+c'Z⊗Z rank 1 is incorrect. The eigenvalues of cI+c'Z⊗Z are c+c' and c−c', each with multiplicity two, so the matrix has rank 1 only when c'=±c. When c'=0, T_B is proportional to the identity and has full rank. Thus the λ3=0 line is not the boundary at which B becomes FDLU-preparable; the correct boundary is λ3=±λ0. This shifts the division between RZ–FDLU (1 round) and RZ–RIRX (2 rounds), and the examples in Sec. VII C built on this boundary need to be reexamined.","section":"Sec. V C 2, Eq. (96)"}],"minor_comments":[{"comment":"The step from equality of transfer matrices to a unitary relation between tensors uses the fundamental theorem for injective MPS; the required injectivity after site blocking should be stated explicitly at this point, since not every tensor in the paper is assumed to be in a canonical form.","section":"Sec. III A, Eq. (19)"},{"comment":"Rank statements are sometimes made before site blocking, while Lemma 1 defines FDLU-preparability after finite blocking. Blocking does not remove nonzero eigenvalues of T_B, so this is not fatal, but the text should consistently indicate whether a rank statement is meant before or after blocking.","section":"Secs. II and V"},{"comment":"The sentence 'we suspect that this state does not admit any pushable defect' is presented as a conjecture. If this is intended to support a no-go statement about finite-round MF preparation, it should be proved or explicitly labelled as an open problem.","section":"Sec. V D"},{"comment":"The notation λ0λ3=λ1λ2 appears in Secs. V B 1 and V C 1 without an accompanying statement of which λ parameters are independent and which are constrained by complete positivity of the transfer matrix; adding the explicit four-eigenvalue form would make the error-corrected conditions much easier to verify.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The rank-condition errors are concentrated in Sec. V but they affect a central classification boundary (1-round versus 2-round MF preparability). I recommend major revision rather than rejection because the framework is otherwise coherent and the errors are locally fixable by replacing the rank conditions with the correct eigenvalue conditions. No concerns about attribution or scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know: this paper has a genuinely new organizing idea for measurement-feedback (MF) state preparation in 1D, but two of the rank-1 checks that drive the classification are algebraically wrong. The reader's report and the stress-test are both right on this.\n\nThe main contribution is the pushable-defect calculus. Instead of requiring all Pauli defects to be pushable, the authors allow incomplete sets, construct an associated state |B> whose transfer matrix is T_A times the defect projector, and iterate. The classification of pushing relations (RI, RZ, RIRX, etc.) and the connections to Tambara-Yamagami dualities, KW, KT, and cosine symmetries are new and instructive. The completeness theorem for open-boundary states with left-conditioned feedback (Theorem 1) goes beyond earlier fusion-measurement results, and the paper is constructive: circuits are explicit and the examples are nontrivial.\n\nThe soft spots are serious but localized. In Sec. V B 1 and V C 1, T_B = λ0 I + λ1 Z1 + λ2 Z2 + λ3 Z1Z2 is diagonal in the computational basis with eigenvalues λ0 ± λ1 ± λ2 ± λ3. Rank 1 requires three of those four to vanish. The paper's condition λ0λ3 = λ1λ2 does not imply that; the stress-test's counterexample (λ0=2, λ1=λ2=1, λ3=1/2) gives eigenvalues 4.5, 1.5, 0.5, 0.5, i.e., rank 4. In Sec. V C 2, T_B = c I + c' Z⊗Z has rank 1 when c = ±c', not when c' = 0. The text says c'=0 makes it rank 1, which is wrong; it leaves a full-rank identity. These errors misidentify the RI–FDLU and RZ–FDLU boundaries and change which states are claimed to be FDLU- or 1-round-MF-preparable. The affected examples in Sec. VII need re-derivation.\n\nThe central framework may survive the fixes; the completeness theorem and the qualitative structure are plausible. But the classification table as written is not reliable. This is fixable, but it needs a serious referee and a corrected revision.\n\nWho is this for? Anyone working on adaptive circuits, measurement-based state preparation, or non-invertible dualities in 1D. It deserves peer review. I would send it out: the defects are concrete and the upside is high.","headline":"A novel and plausible framework for 1D measurement-feedback state preparation, but the rank-1 checks that set the classification boundaries are algebraically wrong and need correction.","tokens_in":44623,"tokens_out":2984,"would_cite":false,"duration_ms":29401,"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":"The paper claims that the set of pushable virtual-bond defects, together with their pushing relations, classifies finite-depth measurement-feedback preparable states in one dimension and dictates their preparation circuits.","keywords":["measurement and feedback","state preparation","matrix product states","pushable defects","finite-depth circuits","non-invertible symmetries","fusion measurement","Kramers-Wannier duality"],"falsifier":"A concrete falsifier: take the $Z\\leftrightarrow X$ non-example tensor of Sec.~V D and check whether any non-identity virtual-bond operator satisfies the transfer-matrix identities (17) or (18); finding one would disprove the paper's expectation that this class has no pushable defects. Alternatively, exhibit an open-boundary MPS that is preparable for all $N$ by left-conditioned single-site measurement and feedback but admits no right-pushable defects; that would refute Theorem 1.","tokens_in":43553,"feed_emoji":"⚛️","tokens_out":8686,"duration_ms":87193,"temperature":0.7,"pith_summary":"Measurement-and-feedback circuits can prepare long-range entangled states in constant depth, but finding the circuit for a target state has been an open search problem. This paper proposes that, for one-dimensional matrix product states, the answer is written in the target's \"pushable defects\": virtual-bond operators that can be moved through the MPS tensor at the cost of a physical feedback unitary. It shows that the set of pushable defects together with their pushing relations classifies finite-depth MF-preparable states and determines the preparation circuit, and for open-boundary states with left-conditioned feedback the construction is complete. The classification also reveals that many such states are product states up to non-invertible duality transformations such as Kramers-Wannier, Kennedy-Tasaki, and generalized cosine symmetries. This matters because it converts state preparation from a numerical search into a tensor classification with explicit circuits.","feed_headline":"Pushable defects classify 1D measurement-feedback states","feed_subtitle":"For open-boundary matrix-product states the defect calculus is complete, and each class maps to a non-invertible symmetry circuit.","key_machinery":"The load-bearing object is the pushable defect: a virtual-bond operator that satisfies Eq. (15) or (16), which by the MPS fundamental theorem is equivalent to an identity on the transfer matrix. From the target tensor $A$ and a set of pushable defects $\\{V_t\\}$, the paper builds an associated tensor $B$ whose $t$-th physical outcome is $AV_t$; $|A\\rangle$ is one MF round from $|B\\rangle$, so the whole preparation circuit is a chain $|A\\rangle \\leftarrow |B\\rangle \\leftarrow |C\\rangle \\leftarrow \\cdots$ ending at an FDLU-preparable state. The classification of pushing relations (which defects go to identity, to $Z$, to $X$, or mix under pushing) is what selects the unitary gates, the number of rounds, and the associated duality symmetry.","core_discovery":"For a translation-invariant MPS tensor $A$, a virtual-bond defect $V_t$ is right-pushable when $AV_t$ can be transformed into $AV'_t$ by a unitary on the physical leg, and the paper shows this is exactly a transfer-matrix identity. The paper proves (Lemma 3 and Theorem 1) that an open-boundary MPS is preparable from an associated state $|B\\rangle$ by one round of single-site measurement and left-conditioned feedback if and only if $|B\\rangle$ is built from $|A\\rangle$ through right-pushable defects; iterating this construction gives the full preparation circuit whenever the target is OLMF-preparable. When all Pauli defects are pushable, the pushing relations decompose into classes $RIRI$, $RIRX$, $RZRX$, and $RILI$ (after blocking and gauge fixing), and the target is a product state up to a Tambara-Yamagami-type duality (KW, KT, or their composition) and a transversal unitary. When only Pauli-$Z$ defects push, classes such as $RI$--$RIRX$ and $RZ$--$RIRX$ arise and require one or two rounds of MF, with the two-round class connected to generalized cosine symmetries. The central assertion is that these defect sets and relations are not bookkeeping: they classify the states and dictate the circuits.","pith_inferences":["The completeness theorem suggests a concrete decision procedure for 1D OLMF preparability: enumerate Pauli defects, test the transfer-matrix identities, form the associated state, and check whether repeated blocking eventually makes its transfer matrix rank-1.","If the paper's conjecture that the $Z\\leftrightarrow X$ class has no pushable defects is correct, those states would be natural candidates for separating OLMF-preparable from more general OMF-preparable states, resolving the paper's open question about whether left-conditioned feedback can always be assumed.","The equivalence noted in the outlook between $r$-round MF circuits and $r$ rounds of fan-out gates suggests a complexity-theoretic reading: the number of MF rounds in the defect-chain construction measures the adaptive depth needed to prepare a state, tying the classification to the magic hierarchy of shallow circuits.","A higher-dimensional analogue would require a global notion of pushability through a tensor-network environment rather than a single transfer matrix; the 1D classes here provide a testbed for whether fusion-category data still classifies preparability in $d>1$."],"forward_implications":["Every OLMF-preparable open-boundary 1D state has a preparation circuit explicitly constructed from its right-pushable defects; finding the circuit reduces to testing transfer-matrix identities, not searching over circuits.","Fusion-measurement preparable states are precisely those whose MPS tensor carries a complete set of pushable Pauli defects, and their pushing relations identify the non-invertible duality (Kramers-Wannier, Kennedy-Tasaki, or a general Tambara-Yamagami duality) that maps a product state to them.","Allowing defects to be pushed left as well as right does not enlarge the class of preparable states; the possible left/right relations collapse after blocking to the same four classes up to spatial inversion.","States in the $RI$--$RIRX$ and $RZ$--$RILI$ classes are preparable by one round of measurement and feedback with non-transversal corrections, even though no complete set of transversally pushable Pauli defects exists.","States in the $RZ$--$RIRX$ class can require two rounds of measurement and feedback and are related to product states by generalized cosine symmetries with fusion rule $L_{\\alpha}L_{\\alpha'}=L_{\\alpha+\\alpha'}+L_{\\alpha-\\alpha'}$."],"supporting_citations":[{"why":"Supplies the characterization of single-round fusion-measurement preparable MPSs and the closure conditions this paper generalizes to incomplete defect sets.","marker":"[20]"},{"why":"Establishes the sequential Clifford-circuit structure for states with all Pauli defects pushable, which the paper refines into the RIRI/RIRX/RZRX classes.","marker":"[29]"},{"why":"Introduces constant-depth adaptive preparation of MPSs and the defect-pushing picture underlying the associated-state construction.","marker":"[19]"},{"why":"Provides the fundamental theorem of matrix product states that makes transfer-matrix identities equivalent to pushability.","marker":"[55]"},{"why":"Gives the lattice realizations of KW, KT, and cosine dualities used to identify the non-invertible symmetry content of each pushing class.","marker":"[34]"}],"fun_headline_variants":["Pushable defects classify MF-preparable 1D states","Defect pushing dictates measurement-feedback circuits","Non-invertible symmetry from pushing relations","Pushing defects unify 1D state preparation schemes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction stands on the matrix-product-state fundamental theorem—after finite site blocking, equal transfer matrices force a unitary on the physical leg—together with the completeness theorem's reliance on open boundary conditions, invertible boundary tensors, and left-conditioned feedback; if any of these fails for a target state, the pushable-defect classification may not apply.","fun_headline_variants_meta":{"raw":{"variants":["Pushable defects classify MF-preparable 1D states","Defect pushing dictates measurement-feedback circuits","Non-invertible symmetry from pushing relations","Pushing defects unify 1D state preparation schemes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000179,"raw_usage":{"total_tokens":1420,"prompt_tokens":1182,"completion_tokens":238,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":798,"completion_tokens_details":{"reasoning_tokens":177}},"tokens_in":798,"tokens_out":238,"duration_ms":3133,"temperature":1.0,"reasoning_tokens":177,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:25:05.270385+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete falsifier: take the $Z\\leftrightarrow X$ non-example tensor of Sec.~V D and check whether any non-identity virtual-bond operator satisfies the transfer-matrix identities (17) or (18); finding one would disprove the paper's expectation that this class has no pushable defects. Alternatively, exhibit an open-boundary MPS that is preparable for all $N$ by left-conditioned single-site measurement and feedback but admits no right-pushable defects; that would refute Theorem 1.","supporting_citations":[],"review_version":1}