{"id":"a7e4da11-ef30-4d94-bd07-e5aa67a26b9c","arxiv_id":"2502.07571","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper presents QSVT-based algorithms that estimate Renyi and von Neumann entropies from copies of a quantum state by measuring ancillas, with improved sample complexity over prior copy-based methods.","lead":"This paper designs quantum algorithms that estimate the entropy of a quantum state from repeated copies of the state, by building a block encoding of the state and measuring small ancilla registers. The algorithms cut the number of copies needed compared with earlier copy-based entropy estimation methods.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 1 is unproven and may be false: density matrix exponentiation yields a channel, not a unitary block encoding of πρ/4; all copy-complexity claims depend on it.","rationale":"I agree with the reader's conditional verdict, but for a more fundamental reason. The ρmin assumption is real and affects the domain of the polynomial approximations, but equality to zero only excludes rank-deficient states; the rest of the analysis could be adapted or stated for full-rank states. Lemma 1, however, is needed for every state. The paper repeatedly asserts it but never proves it, and the cited DME technique provides a channel, not a unitary block encoding. Without Lemma 1, the entire QSVT machinery—constructing powers of ρ, computing logarithms, and the measurement protocol—does not get off the ground. The paper contains no machine-checked proof, no reproducible code, and no independent verification, so the unproven lemma is the main risk to correctness. I keep CONDITIONAL because the issue might be salvageable with an explicit construction, but the condition should be a complete proof of Lemma 1, not just an assumption about ρmin.","tokens_in":980,"tokens_out":3956,"duration_ms":257139,"concrete_test":"Provide an explicit construction and error analysis for Lemma 1. For a one-qubit state ρ = diag(p, 1−p) (e.g., p = 1/3), implement the DME circuit for exp(−iρ/2) using n copies, apply the algorithm of Corollary 71, and test whether the resulting unitary U satisfies <0|U|0> → πρ/4 in spectral norm as n scales like O(1/Δ log(1/Δ)). Also verify that U can be controlled and inverted without access to a purification of ρ. If the convergence fails or a purification is required, Lemma 1 is false.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim—the sample complexity in Tables I and II—rests on Lemma 1, which states that from O(1/Δ log(1/Δ)) copies of ρ one can construct a Δ-approximated block encoding of πρ/4. The paper's justification is to use density matrix exponentiation (DME) to simulate exp(−iρ/2) and then apply Corollary 71 of [23]. But DME is a quantum channel, not a unitary oracle: it acts by consuming fresh copies of ρ and approximates the map σ ↦ e^{−iρt} σ e^{iρt}. Corollary 71 requires a controlled unitary and its inverse, and the paper's own Appendix A Lemma 6 requires a purification unitary, which is not assumed. The paper provides no proof that the DME channel can be dilated to a fixed unitary block encoding with the claimed copy count. Consequently, the derivation of Eq. (7) and the measurement formula p0 = Tr(AρA†) are unjustified. If Lemma 1 fails, the QSVT power-manipulation steps and all sample-complexity improvements collapse, even for full-rank states.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes quantum algorithms for estimating the Rényi entropy S_α = (1/(1−α)) log Tr(ρ^α) and the von Neumann entropy S_v = −Tr(ρ log ρ) from copies of an unknown state ρ. The central idea is to convert copies of ρ into an approximate unitary block encoding of πρ/4 via density-matrix exponentiation and a corollary of quantum singular value transformation, then use block-encoding arithmetic to implement powers of ρ, measure an ancilla, and convert the measurement probability into an estimate of the desired entropy. The paper reports sample-complexity bounds in Tables I and II, claiming improvements over the prior works of Wang et al. (Ref. [29]) and Acharya et al. (Ref. [31]), especially in low-rank regimes.","tokens_in":18425,"tokens_out":5930,"duration_ms":56914,"significance":"If the central construction were valid, the claimed results would be significant: they would give the first copy-based entropy-estimation algorithms with sample complexity polynomial in rank and 1/ϵ with power improvements over prior art, and they would demonstrate a new application of QSVT. The paper is mostly self-contained in its use of external lemmas, and the asymptotic claims are concrete and falsifiable. However, the entire edifice rests on Lemma 1, which is asserted without a proof and, as stated, appears to be false. The manuscript also contains a clear algebraic error in the negative-c case of Section III B. Because these issues affect the central derivation rather than presentation, the significance cannot be assessed until they are resolved.","major_comments":[{"comment":"Lemma 1 is unproven and the justification given is invalid. The text claims that density-matrix exponentiation (Ref. [39]) simulates exp(−iρ/2), and then Corollary 71 of Ref. [23] converts this into an approximate block encoding of πρ/4. But density-matrix exponentiation implements a quantum channel, not a fixed unitary oracle: it consumes fresh copies of ρ and approximates the map σ ↦ e^{−iρt}σe^{iρt}. Corollary 71, on the other hand, requires a controlled unitary U = exp(−iH) and its inverse. The Appendix's own Lemma 6 likewise requires a purification unitary, which is not assumed in this paper. No argument is given that the density-matrix-exponentiation channel can be dilated to a Δ-approximated block-encoding unitary with O((1/Δ)log(1/Δ)) copies. Since Eq. (7), the probability expressions p0, and all sample complexities in Tables I and II depend on this block encoding, the central claim of the paper is not established.","section":"Section III B, after Eq. (31)"},{"comment":"The identity after Eq. (31) is algebraically wrong. The paper defines α = 2k+1+c, but then states that 2k+1 = α and concludes that 1/4 (π/4)^{2k} (1/ρ_min^c) Tr(ρ^α) = (1/ρ_min^c)(1/π) Tr((πρ/4)^α). This equality holds only when c=0. The correct relation has an extra factor (π/4)^c: p0 = (1/ρ_min^c)(1/π)(π/4)^c Tr((πρ/4)^α). The subsequent rescaling δ → δ/(4ρ_min^c) and the derived sample complexity for negative c in Eq. (47) are therefore not justified.","section":"Section III B, after Eq. (31)"},{"comment":"The algorithms require a positive lower bound on the smallest nonzero eigenvalue ρ_min of ρ, and the sample-complexity claims in Tables I and II diverge as ρ_min → 0. Lemmas 12, 13, and 5 all require a spectral lower bound I/κ ≤ A; for a rank-deficient state, ρ_min = 0 and these lemmas do not apply. The manuscript calls ρ_min the 'non-zero minimum eigenvalue' but does not analyze rank-deficient states, nor does it propose truncating small eigenvalues. This is a substantive scope limitation on the main claim, not a technical footnote.","section":"Sections III and IV, Lemmas 3–5"}],"minor_comments":[{"comment":"Equation (7) is written as an exact equality for an approximate block encoding. For an approximate block encoding, the off-block terms are not exactly orthogonal to |0⟩⟨0| ⊗ AρA†, so the measurement probability differs from Tr(AρA†) by terms that must be bounded using the approximation error. The paper later says errors add linearly, but this is not derived.","section":"Section II, Eq. (7)"},{"comment":"The text says 'density matrix exponentiation method in [39]' after citing Ref. [38] for the block-encoding recipe, but Ref. [39] is the supervised/unsupervised machine-learning paper, while density-matrix exponentiation is introduced in Ref. [38] (quantum principal component analysis). The citations appear to be swapped.","section":"References [38] and [39]"},{"comment":"Table I is difficult to read because the O-arguments are not formatted clearly; for example, the 0 < α < 1 entry contains a large log^5 expression whose arguments are ambiguous. The table would be easier to verify if the asymptotic expressions were typeset more carefully.","section":"Table I"},{"comment":"The symbol δ is used both for the additive error in estimating Tr(ρ^α) and for the block-encoding approximation error, sometimes in the same paragraph. Please use distinct notation for these two error parameters.","section":"Section III A and III B"}],"recommendation":"reject","confidential_remarks":"The core problem is Lemma 1: no valid proof is supplied, and the suggested route via density-matrix exponentiation plus Corollary 71 of [23] is conceptually incompatible with the block-encoding definition, because density-matrix exponentiation is a channel simulation rather than a fixed unitary oracle. The negative-c factor error in Eq. (31) is fixable, but the Lemma 1 issue is not a local repair within the current manuscript. I recommend rejection, while noting that the paper could be reconsidered if the authors provide a rigorous and correct construction of a block encoding of ρ from copies without a purification oracle."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Thanks for the report. The paper gives concrete sample-complexity bounds for estimating Rényi and von Neumann entropies from copies of ρ alone, using QSVT on a block encoding of πρ/4. If that block encoding existed, the argument is mostly clean. The authors do a good job assembling standard lemmas (Gilyén et al., Rall, van Enk–Beenakker) and being explicit about error propagation. The comparison to Wang et al. is careful, and the polynomial-approximation route for von Neumann entropy is a nice addition. No circularity: the self-citations are only for eigenvalue estimation. The load-bearing step is Lemma 1, and I think the stress-test note is right. The paper justifies Lemma 1 by saying DME simulates exp(−iρ/2) and then applying Corollary 71 of [23]. That doesn't work. DME is a channel, not a unitary oracle; Corollary 71 requires controlled-U and its inverse. The paper never explains how to extract a fixed unitary block encoding from the DME channel with O(1/Δ log(1/Δ)) copies. Without Lemma 1, the measurement formula (7) and all subsequent complexity claims are unsupported. This is not a minor gap; it is the foundation. Smaller issues: the negative-c identity after Eq. (31) is algebraically wrong—the paper uses 2k+1=α while α=2k+1+c, missing a factor (π/4)^c. That is fixable but should be corrected. The ρmin>0 assumption is never discussed; rank-deficient states are common, and the bounds blow up as ρmin→0. The paper should either extend the analysis or state the limitation. Overall: the framework is coherent given Lemma 1, and the derived bounds are a useful starting point. But as written, the central claim is not established. The issue is localized; a rigorous construction of a block encoding of ρ from copies (or a proof that it can't be done efficiently) is what the paper needs. I would not cite it yet, but I would send it to a referee who knows QSVT—this is exactly the kind of step that peer review should catch. Conditional reject with a clear path to revision, not a desk reject.","headline":"A coherent QSVT framework for copy-based entropy estimation, but the central Lemma 1 is unproven and likely wrong: DME gives a channel, not a unitary block encoding, and all the claimed improvements rest on that step.","tokens_in":719,"tokens_out":1937,"would_cite":false,"duration_ms":69728,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P68"],"pacs":["03.67.-a","03.67.Lx"],"model":"deepseek-v4-flash","headline":"This paper gives a copy-only quantum algorithm that estimates Rényi and von Neumann entropies by building a block encoding of the state and measuring ancillas, with sample complexity improved over prior copy-based methods in both rank and…","keywords":["quantum entropy estimation","Rényi entropy","von Neumann entropy","block encoding","quantum singular value transformation","sample complexity","ancilla measurement","density matrix exponentiation"],"falsifier":"For a fixed small quantum state, count the number of copies the protocol actually consumes to estimate S_α to error ε, and compare with the Table I bound; since the derivation hinges on Lemma 1's copy-to-error rate and Lemma 5's polynomial degree, a disagreement beyond the stated poly-log factors would show one of those lemmas fails at the assumed parameter ranges.","tokens_in":17995,"feed_emoji":"⚛️","tokens_out":12999,"duration_ms":100954,"temperature":0.7,"pith_summary":"This paper claims that the Rényi and von Neumann entropies of an unknown quantum state ρ can be estimated using only copies of ρ, without any unitary that prepares a purification of the state. The core construction turns copies of ρ into a block encoding of πρ/4, then uses quantum-singular-value-transformation tools to raise that encoding to arbitrary powers and reads out a probability proportional to Tr(ρ^α), or to a constant plus the von Neumann entropy, by measuring ancilla qubits. For non-integer Rényi order α>1, the claimed sample complexity is roughly $r^{{3(α-1)}}$/$ε^{3}$ up to logarithmic factors, an almost power-of-two improvement in rank and error over the previous copy-based approach of Ref. [29]. A sympathetic reader would care because entropy estimation is a central subroutine in quantum information and because the method removes the need for purification access while improving on existing copy-only protocols.","feed_headline":"Estimate quantum entropies from copies alone with fewer samples","feed_subtitle":"Block-encoding copies of the state yields entropy estimates with improved rank and error scaling.","key_machinery":"The central object is a block encoding of the operator πρ/4: a unitary matrix whose upper-left block equals πρ/4, constructed from copies of ρ via density-matrix exponentiation and quantum singular value transformation (Lemma 1). The argument then rides on QSVT lemmas that convert this encoding into block encodings of (πρ/4)^k for integer k, of (πρ/4)^c for fractional c>0, and of (ρ/ρ_min)^c for fractional c<0, each with a controlled approximation error; these transformed operators are applied to ρ (or to the maximally mixed state) with an ancilla, and the ancilla measurement probability is the fundamental quantity whose sample complexity drives the bounds.","core_discovery":"The central claim is that nonlinear functions of a density matrix—specifically Tr(ρ^α) and Tr(ρ log ρ)—can be extracted from copies of ρ by constructing a block encoding of ρ, applying quantum-singular-value-transformation (QSVT) based transformations to obtain block encodings of arbitrary powers of ρ, and measuring an ancilla register. Concretely, the probability that the ancilla of the block-encoded operator A=(πρ/4)^k returns |0⟩ after being applied to ρ is Tr((πρ/4)^k ρ (πρ/4)^k) = (π/4)^{2k+1} Tr($ρ^{{2k+1}}$), so choosing 2k+1=α yields Tr(ρ^α) up to a known constant; for 0<α<1 the same idea with a maximally mixed input gives a factor of the dimension times Tr(ρ^α), and for the von Neumann entropy a block encoding of γ log($4ρ^{{-1}}$/π) yields a probability γ log(4/π)+γ S_v. The paper derives sample-complexity bounds for each regime and compares them with the two most relevant prior copy-based algorithms, claiming an almost power-of-two improvement in both the rank dependence and the error tolerance for non-integer α.","pith_inferences":["If the bounds are right, entropy estimation from copies becomes practical for low-rank states, so quantum certification and entanglement-quantity estimation that currently rely on full state tomography could adopt copy-only protocols; the paper does not discuss these downstream applications.","The ρ_min^{-2} factors suggest a scaling bottleneck for nearly pure states; an extension that truncates small eigenvalues or adapts to rank-deficient ρ would be needed before the method applies to, say, ground states with exponentially small spectral gaps.","The same block-encoding-plus-ancilla-measurement skeleton could be turned on other nonlinear functionals of ρ, such as non-integer purity moments Tr(ρ^k), since the machinery already constructs arbitrary real powers of ρ; this is an extension the authors do not pursue.","A numerical test on small systems—comparing the empirical copy count against Table I for a fixed ρ_min—would reveal whether the hidden constants and poly-log factors make the bound tight or loose in practice, which the paper does not address."],"forward_implications":["For non-integer Rényi order 1<α<2, the protocol's sample complexity is O(ε^{-3} ρ_min^{-2} r_ρ^3 log^5(r_ρ/(ρ_min ε))), compared with O(ε^{-5} ρ_min^{-2} r_ρ^5) for the previous copy-based method—an almost power-of-two saving in both rank and error.","For non-integer α>2, the complexity scales as O(ε^{-3} |1-α|^{-3} r_ρ^{3(α-1)}...) (with an extra ρ_min^{-3c} factor when the floor of α is even), again improving the rank and error exponents of the prior bound.","For 0<α<1, using a maximally mixed register, the cost scales with (dim ρ)^2 rather than (dim ρ)^{2/α} of the prior dimension-dependent method, a power-of-two improvement in dimension for fixed error.","For von Neumann entropy, the polynomial-approximation variant achieves O(ε^{-2} ρ_min^{-2} log^4(1/ρ_min) log^2(1/ε)) copies, removing the dimension dependence and improving the error dependence of the QSVT-based approach."],"supporting_citations":[{"why":"Supplies the QSVT framework, the logarithmic-unitary lemma (its Corollary 71), and the positive/negative power-exponent lemmas that carry the block-encoding transformations.","marker":"[23]"},{"why":"The prior copy-based algorithm for estimating Rényi and von Neumann entropies whose sample complexity this work improves; also supplies the error-propagation relation used to convert Tr(ρ^α) error into entropy error.","marker":"[29]"},{"why":"The prior dimension-dependent algorithm for quantum entropy estimation that the protocol avoids, providing the O(n^2/ε^2) baseline compared in Tables I and II.","marker":"[31]"},{"why":"Introduced the recipe, later refined in [23], for constructing a block encoding of ρ from multiple copies, the starting point of Lemma 1.","marker":"[38]"},{"why":"Provides density-matrix exponentiation, the subroutine used to simulate exp(-iρ t) and hence to build the block encoding of πρ/4.","marker":"[39]"},{"why":"Gives the random single-copy measurement method for estimating Tr(ρ^n) for integer n, used for integer Rényi orders and for the polynomial-approximation approach to von Neumann entropy.","marker":"[40]"},{"why":"Provides the positive-power-exponent lemma (Lemma 12 of the appendix) used to construct block encodings of (πρ/4)^c for 0