{"id":"293c7334-0948-4b22-89b8-a3d8156a5bcf","arxiv_id":"2508.12858","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"BELT block-encodes the output of an arbitrary linear map N applied to a density matrix, using a block encoding of the partially transposed Choi matrix of N.","lead":"This paper introduces BELT, a quantum algorithm that can apply any linear transformation to a quantum state and store the result inside a larger unitary operation. It works even for maps that do not correspond to real quantum processes, with applications in detecting entanglement, inverting quantum channels, and simulating differential operators.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"BELT's sample-complexity advantage is in a stronger oracle model; with U_rho unavailable for unknown states, the separation from copy-based protocols may vanish.","rationale":"The paper's Theorem 1 is a correct oracle reduction: given a purification oracle for rho and a block encoding of the partially transposed Choi matrix, it produces a block encoding of N(rho). I checked the derivation in Appendix A, including the index structure and the error propagation, and found no internal inconsistency. The applications (reduction map for entanglement detection, inverse channel recovery, pseudo-differential operators) work under the stated assumptions. The main issue is that the advertised sample-complexity improvements are measured in oracle calls to U_rho, a resource that is stronger than copies of rho. The reader's weakest_assumption already identifies exactly this, plus the related dependence on ||Lambda_N^{T1}||_infty. I agree with that assessment. The paper is transparent about these caveats in the text after Eq. (7) and in the Table I note, but the abstract and introduction do not foreground them, which justifies the CONDITIONAL verdict. No change to the reader's verdict is needed; the concern is real but does not invalidate the central theorem.","tokens_in":14860,"tokens_out":30233,"duration_ms":308799,"concrete_test":"Fix a resource measure by counting copies of rho rather than oracle calls: for an unknown n-qubit state, estimate the number of copies needed to implement one U_rho call via quantum state tomography (standard lower bound Omega(2^n)). Re-evaluate Theorem 2's '6 oracle calls' and Table I; if the copy count is exponential, the claimed constant-vs-exponential separation over single-copy protocols vanishes when both are measured in copies.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorem (Theorem 1) is mathematically sound: the tensor-network proof in Appendix A correctly shows that the circuit block-encodes N(rho). The load-bearing concern is about the interpretation of the sample-complexity claims. BELT requires oracle access to a purification unitary U_rho and its inverse, whereas the comparison baselines (HME, single-copy protocols) operate on copies of rho. This is a strictly stronger resource: given U_rho one can prepare copies of rho, but not vice versa without exponential tomography overhead for an unknown state. Consequently, the '6 oracle calls' of Theorem 2 and the 'constant vs exponential' separation of Table I do not constitute a sample-complexity improvement over copy-based protocols in the same resource model. A second facet is the norm dependence: Eq. (7) gives success probability alpha^{-2} Tr[N(rho) sigma N(rho)^dagger]; for the transpose map highlighted in the abstract, Lambda_T^{T1} = |Phi+><Phi+| has norm 2^n, so the protocol succeeds with probability 4^{-n} when sigma is the maximally mixed state. The paper acknowledges these caveats (text after Eq. (7) and Table I note), but the abstract and introduction still advertise the transpose map and improved sample complexity without foregrounding them, so the practical reach of the central claim is narrower than advertised.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"BELT is a protocol for block-encoding the output N(rho) of an arbitrary linear map N acting on an unknown state rho. The construction uses a block encoding of the partially transposed Choi matrix Lambda_N^{T1} and an oracle U_rho that prepares a purification of rho. Theorem 1 proves that the circuit (I_m tensor U_rho^dagger tensor I_k)(U_N tensor I_r)(I_m tensor U_rho tensor I_k) is an (alpha, m+r+n, epsilon)-block encoding of N(rho). The paper applies BELT to entanglement detection via the reduction criterion, to exact inversion of known quantum channels, and to simulating pseudo-differential operators; it also proposes an alternative QETU+HME approach in Appendix D. The central derivation is self-contained and uses no fitted parameters.","tokens_in":15103,"tokens_out":11748,"duration_ms":126099,"significance":"If Theorem 1 holds, and the Appendix A proof indeed gives a clean error-propagation argument, BELT is a conceptually useful primitive: it extends block encoding from spectral functions of rho to general linear maps, including non-CP maps, and it enables coherent postprocessing of N(rho) with QSVT. The reduction-criterion entanglement protocol and the exact channel-recovery protocol are interesting, and the discussion of maps with bounded Choi-transpose norm is honest. The main caveat, acknowledged in part but under-emphasized, is that the claimed sample-complexity improvements are obtained in a stronger oracle model than the copy-based models used by the HME and single-copy baselines. The paper would be stronger if the resource-model distinction and the norm dependence were placed at the center of the exposition.","major_comments":[{"comment":"Theorem 2 states that the entanglement-detection protocol uses 6 oracle calls to U_rho or U_rho^dagger, but the proof in Appendix B sets K=2, and each run of the circuit in Fig. 2(a) contains one U_rho and one U_rho^dagger. The protocol therefore uses 4 oracle calls, not 6. Since K=2 already gives success probability at least 2/3, the theorem should be corrected to 4 calls, or changed to K=3 with 6 calls.","section":"Theorem 2 and Appendix B"},{"comment":"The constant-versus-exponential separation in sample complexity is not a comparison in the same resource model. BELT requires oracle access to U_rho and U_rho^dagger, which for an unknown state is strictly stronger than access to copies of rho; the HME and single-copy baselines use only copies. The Table I note and the text after Theorem 3 acknowledge this 'stronger oracle model', but the abstract and introduction still advertise improved sample complexity. The claims should be reframed as oracle-query complexity, with an explicit statement that no sample-complexity improvement over copy-based protocols is claimed in the identical resource model.","section":"Abstract, Section III.A, and Table I"},{"comment":"The norm dependence of the protocol is load-bearing for the advertised scope. Eq. (7) gives success probability alpha^{-2} Tr[N(rho) sigma N(rho)^dagger], and for the transpose map one has alpha = ||Lambda_T^{T1}||_infty = 2^n, so the post-selection probability is exponentially small (e.g., 4^{-n} for a real pure state with sigma = rho). The text after Eq. (7) does note that maps with bounded alpha are preferable, and the applications indeed use such maps, but the abstract's mention of the transpose map as a headline example invites an opposite reading. The norm caveat should be moved into the abstract and introduction.","section":"Eq. (7) and Abstract"}],"minor_comments":[{"comment":"The statement 'A block encoding of A exists iff ||A||_infty <= alpha' is correct for exact block encodings (epsilon=0) but not as an iff statement for approximate block encodings, since a matrix with ||A||_infty > alpha can still be approximately block encoded. The sentence should be qualified.","section":"Section II, after Definition 1"},{"comment":"There is a typo: 'In the identity on n qubits' should read 'I_n the identity on n qubits'.","section":"Section II, first paragraph"},{"comment":"The displayed formula describing the spectral map in Fig. 1(c), containing TN and Lambda_{E^{-1}}^{T1}, appears garbled in the manuscript text; the intended expression should be typeset legibly.","section":"Figure 1 and surrounding text"},{"comment":"In the paragraph after Theorem 2, the phrase 'sample complexity' should be replaced by 'oracle-query complexity' or 'number of oracle calls' to avoid conflating the two resource measures.","section":"Section III.A"}],"recommendation":"major_revision","confidential_remarks":"The central theorem is sound and the main issues are local-to-the-exposition: an incorrect oracle-count in Theorem 2 and an inadequate foregrounding of the stronger-oracle-model comparison. Once these are fixed, I would be comfortable with the paper. The self-citation to the HME paper is appropriate given that the comparison is direct and acknowledged."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The central construction is sound and worth knowing. BELT block-encodes an arbitrary linear map N acting on rho by taking a block encoding of the partially transposed Choi matrix Lambda_N^{T1} and sandwiching it with a purification oracle and its inverse. That is a natural generalization of the swap-based block encoding of rho, and it extends the reach of both QSVT and HME to maps that are not Hermitian-preserving. The tensor-network proof in Appendix A is clean, the error propagation in Eq. (A3) is correct, and the applications to entanglement detection and channel inversion are concrete. The reduction map has norm-2 Choi-transpose, so the entanglement detection protocol runs with constant post-selection overhead, and the 2-sparse structure gives an efficient gate implementation.\n\nThe paper is mostly honest about its main caveat: the success probability scales with ||Lambda_N^{T1}||_infty^{-2}, and the abstract's mention of the transpose map is not one of the efficient examples. The text after Eq. (7) explicitly says to prefer maps with bounded Choi-transpose norm, and the applications do that. So the norm dependence is not a flaw in the applications, just a limitation of the general claim.\n\nThe weaker point is the sample-complexity comparison. BELT assumes a unitary oracle U_rho that prepares a purification of rho and its inverse, which is strictly stronger than having copies of rho. The comparison to single-copy protocols is therefore not apples-to-apples: the constant-versus-exponential separation is really 'with a state-preparation oracle' versus 'without.' The paper does acknowledge this in the Table I note and in the channel inversion subsection, but the abstract and introduction still advertise 'improved sample complexity' without foregrounding that qualifier. A revision should move that caveat to the front.\n\nMinor issues: Theorem 2 says 6 oracle calls, but the proof's K=2 repetitions require 4 (one U_rho and one U_rho^dagger per run). That is a constant-count inconsistency, easy to fix. Also, the pseudo-differential operator section is a bit of a side application, since it only uses the CP case, but it is harmless.\n\nOverall, the primitive is a real contribution and the paper deserves a serious referee. The math holds up, the soft spots are presentation and constant-factor issues, not load-bearing flaws. I would take it for peer review and expect it to be accepted after a minor revision that clarifies the oracle model.","headline":"The core BELT construction is clean, correct, and genuinely new; the sample-complexity claims need to be framed more carefully around the stronger purification-oracle model.","tokens_in":15637,"tokens_out":4566,"would_cite":true,"duration_ms":47426,"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 authors prove that any linear map acting on a quantum state — even one that would make the output nonphysical — can be embedded as a block of a unitary operator by the BELT protocol.","keywords":["block encoding","linear maps on density matrices","non-completely positive maps","Choi matrix","partial transpose","entanglement detection","quantum channel inversion","quantum singular value transformation"],"falsifier":"Take a small exactly known case, such as the transpose map on a single qubit with $\\rho=|+\\rangle\\langle+|$; construct $U_\\mathcal{N}$ as a block encoding of $\\Lambda_\\mathcal{N}^{T_1}$, run the BELT circuit, and compare the top-left block of the resulting unitary with $\\rho^{T}/\\alpha$. If the operator-norm error exceeds $\\epsilon$, Theorem 1 is false; equivalently, measuring the postselection probability and checking that it equals $\\alpha^{-2}\\operatorname{Tr}[\\mathcal{N}(\\rho)\\sigma\\mathcal{N}(\\rho)^\\dagger]$ tests the identity underlying the protocol.","tokens_in":14673,"feed_emoji":"⚛️","tokens_out":7312,"duration_ms":70789,"temperature":0.7,"pith_summary":"The paper introduces BELT, a protocol that simulates an arbitrary linear map $\\mathcal{N}$ acting on an $n$-qubit density matrix $\\rho$. Because $\\mathcal{N}(\\rho)$ need not be a valid density matrix, it cannot be prepared directly; BELT instead embeds $\\mathcal{N}(\\rho)$ as a block of a larger unitary, using a purification oracle for $\\rho$ and a block encoding of the partially transposed Choi matrix of $\\mathcal{N}$. The central theorem states that a simple three-part circuit gives an $(\\alpha,\\dots)$-block encoding of $\\mathcal{N}(\\rho)$. This extends quantum simulation beyond completely positive and Hermitian-preserving maps to all linear maps, including the transpose map and channel inverses, and yields efficient protocols for entanglement detection and for recovering a state from a noisy channel's output.","feed_headline":"A three-part unitary circuit embeds any linear map's output as a block","feed_subtitle":"Nonphysical maps like transposition become runnable, speeding up entanglement tests and channel inversion.","key_machinery":"The load-bearing object is the partially transposed Choi matrix $\\Lambda_\\mathcal{N}^{T_1}=(I\\otimes\\mathcal{N})(|\\Phi^+\\rangle\\langle\\Phi^+|)^{T_1}$, which encodes the whole linear map and can be block-encoded even when $\\mathcal{N}(\\rho)$ is not a valid state. The circuit sandwiches the block-encoding unitary $U_\\mathcal{N}$ between the purification oracle $U_\\rho$ and its inverse, so the ancilla register of the purification carries the input state while the Choi-transpose matrix applies $\\mathcal{N}$; replacing the swap operator used for the identity channel with $\\Lambda_\\mathcal{N}^{T_1}$ is what lifts block encoding from states to arbitrary linear transformations.","core_discovery":"Theorem 1 is the core: for a linear map $\\mathcal{N}:\\mathcal{L}(\\mathbb{C}^{2^n})\\to\\mathcal{L}(\\mathbb{C}^{2^k})$, let $U_\\rho$ prepare a purification of $\\rho$ and let $U_\\mathcal{N}$ be an $(\\alpha,m,\\epsilon)$-block encoding of the partially transposed Choi matrix $\\Lambda_\\mathcal{N}^{T_1}$. Then the circuit $(I_m\\otimes U_\\rho^\\dagger\\otimes I_k)(U_\\mathcal{N}\\otimes I_r)(I_m\\otimes U_\\rho\\otimes I_k)$ is an $(\\alpha,m+r+n,\\epsilon)$-block encoding of $\\mathcal{N}(\\rho)$. The tensor-network proof replaces the only non-unitary ingredient, $\\Lambda_\\mathcal{N}^{T_1}$, by its block-encoding unitary; choosing $\\mathcal{N}$ as the identity recovers the standard block encoding of $\\rho$ via the swap operator, so BELT is a direct generalization that turns the nonphysical object $\\mathcal{N}(\\rho)$ into a physically realizable unitary block.","pith_inferences":["If BELT is correct, the notion of physical simulability of a linear map shifts from complete positivity or Hermitian preservation to the cost of block-encoding $\\Lambda_\\mathcal{N}^{T_1}$; maps that are hard to block-encode remain hard regardless of their positivity properties.","Because the protocol requires a purification oracle rather than copies of $\\rho$, it trades a stronger input assumption for an exponential sample-complexity gain; a natural test is whether the gain survives when $U_\\rho$ must itself be learned from copies.","The norm $\\|\\Lambda_\\mathcal{N}^{T_1}\\|_\\infty$ is the real resource: for maps where it grows exponentially, BELT's postselection probability falls exponentially, so the method's practical reach will be decided by identifying useful maps with bounded Choi-transpose norm, which the paper leaves as an open direction."],"forward_implications":["BELT can simulate maps that lie outside quantum singular value transformation: the transpose map $\\rho\\mapsto\\rho^T$ is basis-dependent and not a spectral function, yet it becomes block-encodable.","For entanglement detection on states drawn from the paper's distribution, BELT detects entanglement with six calls to the purification oracle and its inverse, while any single-copy protocol needs exponentially many copies.","For a known invertible channel $\\mathcal{E}$ acting on an unknown pure state $\\psi$, BELT prepares $\\psi$ from $\\mathcal{E}(\\psi)$ exactly upon postselection, with sample complexity polynomial in $\\|\\Lambda_{\\mathcal{E}^{-1}}^{T_1}\\|_\\infty$ and logarithmic in $1/\\delta$.","Combining BELT with QSVT yields block encodings of $f(\\alpha^{-1}\\mathcal{N}(\\rho))$ for spectral functions $f$, enabling amplitude amplification that boosts the postselection probability.","For completely positive maps with Stinespring dilation $\\mathcal{F}(\\rho)=\\operatorname{Tr}_Z(A\\rho A^\\dagger)$, BELT block-encodes $\\mathcal{F}(\\rho)$ efficiently when $\\|A\\|_\\infty$ is bounded, which applies to pseudo-differential operators through existing block encodings of $T$."],"supporting_citations":[{"why":"Supplies the definition of block encoding used in Theorem 1.","marker":"[36]"},{"why":"Provides the sparse-matrix block-encoding lemma and the QSVT framework used to process the block-encoded output.","marker":"[18]"},{"why":"Gives the standard block encoding of a density matrix via the swap operator, which BELT generalizes.","marker":"[26]"},{"why":"The Hermitian-preserving map exponentiation algorithm that BELT extends and compares against.","marker":"[7]"},{"why":"Gives the single-copy exponential lower bound that BELT's entanglement-detection protocol outperforms.","marker":"[34]"},{"why":"The reduction criterion, the positive map used in the entanglement-detection application.","marker":"[48]"},{"why":"Efficient block encodings of pseudo-differential operators used in the PDO application.","marker":"[65]"},{"why":"Robust oblivious amplitude amplification used to boost postselection in channel inversion.","marker":"[62]"}],"fun_headline_variants":["Block-encode any linear map, even nonphysical ones","Unitary block encoding tames non-CP maps like transposition","BELT: run nonphysical maps on quantum hardware","Quantum protocol embeds forbidden maps into unitary blocks"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The protocol is efficient only when the operator norm of the partially transposed Choi matrix of the map stays bounded or grows slowly with system size, because the postselection success probability is controlled by the square of that norm.","fun_headline_variants_meta":{"raw":{"variants":["Block-encode any linear map, even nonphysical ones","Unitary block encoding tames non-CP maps like transposition","BELT: run nonphysical maps on quantum hardware","Quantum protocol embeds forbidden maps into unitary blocks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000205,"raw_usage":{"total_tokens":1389,"prompt_tokens":936,"completion_tokens":453,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":387}},"tokens_in":552,"tokens_out":453,"duration_ms":4904,"temperature":1.0,"reasoning_tokens":387,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:18:48.955463+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small exactly known case, such as the transpose map on a single qubit with $\\rho=|+\\rangle\\langle+|$; construct $U_\\mathcal{N}$ as a block encoding of $\\Lambda_\\mathcal{N}^{T_1}$, run the BELT circuit, and compare the top-left block of the resulting unitary with $\\rho^{T}/\\alpha$. If the operator-norm error exceeds $\\epsilon$, Theorem 1 is false; equivalently, measuring the postselection probability and checking that it equals $\\alpha^{-2}\\operatorname{Tr}[\\mathcal{N}(\\rho)\\sigma\\mathcal{N}(\\rho)^\\dagger]$ tests the identity underlying the protocol.","supporting_citations":[{"cited_title":"Gily´ en, Y","cited_arxiv_id":null,"evidence_quote":"Provides the sparse-matrix block-encoding lemma and the QSVT framework used to process the block-encoded output."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Hermitian-preserving map exponentiation algorithm that BELT extends and compares against."},{"cited_title":"Liu and F","cited_arxiv_id":null,"evidence_quote":"Gives the single-copy exponential lower bound that BELT's entanglement-detection protocol outperforms."},{"cited_title":"entangled","cited_arxiv_id":null,"evidence_quote":"The reduction criterion, the positive map used in the entanglement-detection application."},{"cited_title":"Regula, R","cited_arxiv_id":null,"evidence_quote":"Robust oblivious amplitude amplification used to boost postselection in channel inversion."}],"review_version":1}