{"id":"2f4c940f-cbc2-4ad9-a277-8488aa86541c","arxiv_id":"2505.00697","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Symmetry-tailored and parallel-readout variants of adaptive quantum gradient estimation cut the state-preparation query count for fermionic k-RDM estimation, giving a quadratic speedup over prior QGE methods at fixed particle number.","lead":"This paper offers two upgraded versions of a quantum algorithm that estimates many properties of a quantum system at once, one exploiting a symmetry of the state and one reading out all answers in a single shot. If the results are correct, computing fermionic properties like 2-electron reduced density matrices would need 100 to 500 times fewer calls to the state preparation circuit in the benchmarked cases.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The k-RDM norm identity is stated without a normalization convention; under the standard Hermitianized observables it already fails for k=1, N=3, η=1, so the 100x/500x query claims are not grounded in this text.","rationale":"The reader's weakest_assumption identified the unproven k-RDM norm identity as load-bearing; I agree and sharpen it: the identity is not merely unproven but ambiguous, since its numerical value depends on whether the observables are the non-Hermitian k-RDM elements or their Hermitianized real/imaginary parts. A direct k=1 check shows that the standard Hermitianized convention gives a different value from the claimed binomial product, so the paper's query-count formulas and the 100x/500x reductions are not derivable from the text as written. The companion-paper deferral compounds the issue, but the normalization ambiguity is the more elementary and testable problem. I am not claiming the final claims are false; under a consistent convention the algorithm may be correct and the constants may shift. However, the letter does not provide enough information to verify the central quantitative claims, so the reader's CONDITIONAL verdict remains appropriate. I therefore recommend no change to the verdict.","tokens_in":30414,"tokens_out":34940,"duration_ms":365481,"concrete_test":"For k=1, N=3, eta=1, explicitly list the Hermitianized 1-RDM observables as defined in the text (diagonal n_p, off-diagonal (a^dagger_p a_q + a^dagger_q a_p)/2 and (a^dagger_p a_q - a^dagger_q a_p)/(2i)), restrict each to the eta=1 sector, compute ||sum_j (O_j^{(eta)})^2||, and compare with the claimed value 3. Repeat with the unnormalized off-diagonal convention. If neither gives 3, the identity is not valid for the Hermitianized observables used in Theorems 1 and 2, and the associated query-complexity claims require a revised normalization or a separate proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central speedup rests on the identity immediately before Fig. 1(d): sum_j [O_j^{(eta)}]^2 = C(eta,k) C(N-eta+k,k), which is fed into Theorems 1 and 2 (Eqs. 5 and 7). The identity is asserted without proof and, more importantly, without specifying which operator convention is used. For k=1, N=3, eta=1, the right-hand side is 3. If O_j are the Hermitianized 1-RDM operators in the standard convention — diagonal D_p = a^dagger_p a_p and off-diagonal X_pq = (a^dagger_p a_q + a^dagger_q a_p)/2, Y_pq = (a^dagger_p a_q - a^dagger_q a_p)/(2i) — then on the eta=1 sector the diagonal terms contribute I and the off-diagonal terms X_pq^2+Y_pq^2 = (|p><p|+|q><q|)/2, whose sum over p<q is I, so sum_j (O_j^{(eta)})^2 = 2I. If instead one drops the factor 1/2, the same sum is 5I. Neither equals 3. The value 3 matches the non-Hermitian ordered sum sum_{p,q} (a^dagger_p a_q)^dagger (a^dagger_p a_q), which acts as eta(N-eta+1)I. Thus the identity holds for non-Hermitian k-RDM elements, but Theorems 1 and 2 are stated for Hermitian observables with operator norm at most 1. The text's remark about Hermitianized operators does not specify the normalization, and the complete proofs are deferred to the companion paper [35], which cannot be checked from this letter. If the companion proves the theorem for a different observable convention, the query complexities in Eqs. (5) and (7) and the numerical 100x/500x factors are not the quantities being proved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript proposes a generalized adaptive quantum gradient estimation (QGE) framework and two variants for multiple observables estimation: Method I exploits a symmetry (direct-sum structure) of the target state, and Method II adds a single-shot parallel readout scheme. The main results, Theorem 1 (Eq. 5) and Theorem 2 (Eq. 7), bound the total number of queries to the state-preparation oracles U_psi and U_psi^dagger needed to estimate M observables with max MSE <= epsilon^2, in terms of epsilon^{-1} times sqrt(sum_j [O_j^{(lambda)}]^2) up to logarithmic factors. For fermionic k-RDM estimation under a fixed particle number eta, the paper asserts the identity sum_j [O_j^{(eta)}]^2 = C(eta,k) C(N-eta+k,k), which converts the general bound into a quadratic speedup over prior adaptive QGE for extreme fillings. Numerical comparisons in Figures 2 and 3 claim query reductions by factors of 100 for the FeMoco active space (N=152, eta=113) and 500 for a 100-site doped Hubbard model at epsilon=10^{-3}.","tokens_in":30800,"tokens_out":9318,"duration_ms":94375,"significance":"If fully established, the proposed symmetry-aware complexity bound would be a useful and nontrivial advance: it offers a quadratic improvement in the observable-dependent prefactor for particle-number-restricted fermionic partial tomography, and the parallel single-shot readout is a conceptually interesting way to remove repetition overhead. The general adaptive-QGE framework, the pseudocode in Algorithm 1, and the proof of Lemma 1 in Appendix A provide a helpful foundation. However, the two main theorems are not proved in the text (Theorem 2 has no proof at all, and Theorem 1's proof is a sketch), and the k-RDM norm identity that powers the central speedup is stated without a normalization convention or proof. Because the numerical 100x/500x reductions inherit these unverified ingredients, the paper's main quantitative claims cannot currently be checked from the submitted manuscript. The strengths are the clarity of the framework and the concrete application targets; the weakness is the deferral of essentially all load-bearing proof content to the companion paper [35].","major_comments":[{"comment":"The identity sum_j [O_j^{(eta)}]^2 = C(eta,k) C(N-eta+k,k) is asserted without proof and without specifying the normalization of the Hermitianized k-RDM operators. Under the standard convention with off-diagonal Hermitianized operators X_pq = (a^dagger_p a_q + a^dagger_q a_p)/2 and Y_pq = (a^dagger_p a_q - a^dagger_q a_p)/(2i), the k=1, N=3, eta=1 case gives sum_j [O_j^{(eta)}]^2 = 2I, not 3I; with unnormalized off-diagonal operators the sum is 5I; the value 3 matches the non-Hermitian ordered sum sum_{p,q} (a^dagger_p a_q)^dagger (a^dagger_p a_q). Since Theorems 1 and 2 are stated for Hermitian observables with spectral norm at most 1, the query complexities in Eqs. (5) and (7) and the derived 100x/500x reductions are not directly tied to the asserted identity as written. Please state the operator convention explicitly, prove the identity under that convention, and reconcile it with the Hermitian-observable assumption in the theorems, or revise the claimed complexity bounds accordingly.","section":"Application to fermionic problems, p.4 (before Fig. 1(d))"},{"comment":"The complete proofs of both main theorems are deferred to the companion paper [35]; for Theorem 2 the text contains no proof at all, only a pointer to Sec. V.C of [35]. Since the central claims of the letter are these query-complexity bounds, the manuscript is not self-contained: a reader cannot verify the logarithmic factors, the validity of uniform singular value amplification for the restricted observables O_j^{(lambda)} without access to the projector Pi_lambda, or the claimed dependence on d_lambda and M. Please include full proofs of Theorems 1 and 2, or at minimum complete proof sketches with all technical lemmas stated and proved in the appendix.","section":"Theorem 2 (Eq. 7) and Theorem 1 proof sketch"},{"comment":"The text states that the parallel scheme 'reduces the query complexity from O(epsilon^{-1} R sqrt(M) log d) to O(epsilon^{-1} sqrt(M) R log d), achieving quadratic speedup regarding R.' These two expressions are identical up to commutativity of the factors, so the claimed quadratic speedup is not reflected in the displayed formulas. Please correct the baseline and improved scaling (e.g., distinguishing the number of repetitions R from the number of observables M) and state precisely what quantity is quadratically improved.","section":"Method II paragraph, p.3"}],"minor_comments":[{"comment":"The phrase 'Hermitianized operators such as (kD_q^p + kD_q^q)/2' appears to contain a typo: the indices should likely be arranged as (kD_q^p + kD_p^q)/2 or similar. Please correct this to the intended Hermitianized 2-RDM operator.","section":"Application to fermionic problems, p.4"},{"comment":"The table in Figure 1(d) is difficult to read in the provided rendering; the asymptotic scalings are partially garbled (e.g., 'O(N^{k/2})/epsilon' versus 'O(N^k)/epsilon^2'). Please ensure the figure is typeset legibly in the final version.","section":"Figure 1(d)"},{"comment":"The input condition M >= 2 log_2 d + 24 and the confidence parameter c bound c in (0, 3/(8(1+pi)^2)] are introduced in the pseudocode but not connected to the theorem statements; please state how these conditions are used or remove them from the main algorithm description.","section":"Algorithm 1 (Appendix B)"},{"comment":"The captions do not define the exact query-count evaluation method (e.g., which constant factors are kept, what block-encoding costs are assumed for the observables, and how the 'QAE algorithm proposed in accompanying paper [35]' is implemented). Please specify these details so the numerical factors 100 and 500 are reproducible.","section":"Figures 2 and 3 captions"}],"recommendation":"major_revision","confidential_remarks":"The central proofs of Theorems 1 and 2 and the QAE-with-Heisenberg-limit baseline used in the numerical comparisons are all attributed to the companion paper arXiv:2505.00698 [35]. If that companion is not included in the review package, the present letter cannot be independently evaluated for correctness; I recommend asking the authors to supply it or to move the full proofs into the appendix. I also note that the 100x/500x query-reduction claims are presented as part of the letter's headline results but depend directly on the unproven norm identity and the companion's QAE baseline; please ensure these claims are either fully supported in this text or explicitly labeled as conditional on [35]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the core idea is real: replacing the full-space norm in adaptive QGE with a symmetry-adapted subspace norm, plus a parallel single-shot readout, is a genuine advance that could reduce query counts for symmetric states. Second, the fermionic application section is not currently checked: the key identity sum_j [O_j^(eta)]^2 = C(eta,k)C(N-eta+k,k) is asserted without proof or a normalization convention, and under the standard Hermitianized p<q convention it is false. That directly undermines the 100x/500x speedup claims.\n\nThe paper does a few things well. Theorem 1's bound is a clean generalization of [16], and the observation that you don't need access to the projector Pi_lambda is valuable. Theorem 2's parallel scheme, turning R repetitions into a single entangled readout, is a neat trick and properly reduces the cost. The QSVT lemma in Appendix A is proved. The comparison table in Fig. 1(d) is helpful.\n\nThe soft spots are substantial. First, complete proofs of Theorems 1 and 2 are in the companion paper; Theorem 2 has no proof sketch in this text beyond a few sentences. For a letter that's acceptable only if the companion is easily available and the statements are precise. Second, the norm identity. The stress-test example is worth checking: for k=1, N=3, eta=1, the standard Hermitianized real+imaginary operators give sum of squares 2, not 3. The identity matches the non-Hermitian ordered sum. If the authors intend to list every ordered pair (p,q) and include duplicate Hermitianized operators, the identity holds, but then M and log M change, and the text never says that. The claim as written is at best ambiguous and at worst wrong. Since this identity feeds directly into Eqs. (5) and (7) and the reported 100x/500x factors, those numbers are not grounded in this manuscript. Third, the baselines in Figures 2-3 are drawn from the same group's companion paper with undisclosed constants, so the head-to-head factor should be treated as provisional.\n\nIs the central idea salvageable? Probably yes. The subspace-adapted complexity bound is plausible independent of the RDM identity. But the application section needs a correct, stated normalization and a proof of the identity (or a different identity). A referee should demand that.\n\nWho this is for: quantum algorithm researchers working on fermionic tomography and expectation-value estimation. It deserves a serious referee; the novelty is enough, and the flaw is fixable. I would not cite it yet. Bring it to reading group as a cautionary example of how a missing normalization can sink an otherwise good idea.","headline":"Real algorithmic advance in adaptive QGE, but the fermionic RDM speedup rests on an unproved and convention-sensitive norm identity, so the 100x/500x query claims are not yet grounded.","tokens_in":31450,"tokens_out":13322,"would_cite":false,"duration_ms":123402,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A quantum algorithm for estimating many fermionic observables at once claims a quadratic query reduction by exploiting particle-number symmetry and a single-shot parallel readout, with numerical 100x and 500x savings for 2-RDM estimation.","keywords":["quantum gradient estimation","multiple observables estimation","fermionic partial tomography","reduced density matrices","particle-number symmetry","query complexity","quantum singular value transformation","parallel quantum readout"],"falsifier":"Work through the paper's normalization for a minimal case, say $k=1$, $\\eta=1$, $N=2$, by explicitly constructing the Hermitianized $k$-RDM operators, restricting to the one-particle subspace, and summing squared spectral norms; if the sum does not equal $\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}$, the claimed 100x and 500x reductions fail by that factor. The same check can be repeated by reproducing the FeMoco and Hubbard query counts at $\\varepsilon=10^{-3}$ from the paper's formulas.","tokens_in":30130,"feed_emoji":"⚛️","tokens_out":12554,"duration_ms":115389,"temperature":0.7,"pith_summary":"Estimating many expectation values of a fermionic many-body state is a bottleneck in quantum simulation, and this paper proposes a way to do it with fewer calls to the state-preparation oracle. The core idea is a generalized adaptive quantum gradient estimation loop with two upgrades: restrict the encoding of observables to the particle-number subspace that actually contains the target state, and replace repeated probe readouts by a single-shot parallel measurement. The paper states as Theorem 2 that when the state lives in an $\\eta$-particle subspace, the total query count scales as $\\epsilon^{-1}\\sqrt{\\sum_j [O_j^{(\\eta)}]^2}\\,\\log d_\\eta \\log M$ for root-mean-square error $\\varepsilon$. For $k$-RDM elements it asserts the squared-norm sum equals $\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}$, which yields a quadratic speedup over previous adaptive quantum gradient estimation in low- or high-filling regimes. Numerically, at $\\varepsilon=10^{-3}$, this reduces 2-RDM estimation queries by about 100x for the FeMo cofactor active space and 500x for a 100-site doped Fermi-Hubbard model.","feed_headline":"Quantum algorithm cuts fermionic estimation queries 100-500x","feed_subtitle":"Particle-number symmetry and parallel readout give adaptive gradient estimation a quadratic speedup.","key_machinery":"The machinery is an adaptive quantum gradient estimation loop in which the expectation values to be estimated are encoded as phases of a probe register; the loop's cost is set by the block-encoding normalization of the averaged observable. Method I uses the direct-sum decomposition $O_j=\\bigoplus_\\lambda O_j^{(\\lambda)}$ and applies quantum singular value transformation (QSVT), the technique for implementing polynomial functions of an operator's singular values, only on the relevant symmetry subspace via Lemma 1, bringing the normalization down to $\\sigma_\\lambda=O(\\sqrt{\\sum_j [O_j^{(\\lambda)}]^2}\\,\\log d_\\lambda)$. Method II replaces the iterated median-of-readouts step by preparing $R$ entangled copies of the probe state and reading them in parallel, which changes the repetition scaling from $O(\\varepsilon^{-1}R\\sqrt{M}\\log d)$ to $O(\\varepsilon^{-1}\\sqrt{MR}\\log d)$; with $R=O(\\log M)$ this yields the $\\sqrt{\\log M}$ factor in Theorem 2.","core_discovery":"The central claim is that the cost of estimating $M$ observables from a state preparation oracle can be made to depend on the size of the symmetry-restricted observable set, not the full Hilbert space. Concretely, Theorem 2 asserts that for a target state supported on a particle-number subspace $\\eta$, $M$ observables can be estimated with worst-case mean squared error at most $\\varepsilon^2$ using $\\varepsilon^{-1}\\cdot O(\\sqrt{\\sum_j [O_j^{(\\eta)}]^2}\\,\\log d_\\eta \\log M)$ queries to $U_\\psi$ and $U_\\psi^\\dagger$. For fermionic $k$-RDM elements the paper identifies the squared-norm sum as $\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}$, and for $\\eta=k+O(1)$ or $\\eta=N-O(1)$ this turns the bound into a quadratic improvement over the previous adaptive quantum gradient estimation algorithm. The same paper reports that, for 2-RDM estimation at $\\varepsilon=10^{-3}$, the query count is reduced by a factor of about 100 for the FeMo cofactor ($N=152$, $\\eta=113$) and about 500 for a 100-site doped Fermi-Hubbard model ($\\eta=\\lceil 7N/8\\rceil$).","pith_inferences":["If the binomial norm identity is correct, the same subspace-block-encoding trick should extend to other symmetries of fermionic states, such as spin or momentum conservation, wherever the analogue of $\\sum_j [O_j^{(\\lambda)}]^2$ can be computed; the speedup would then be available for a wider class of collective observables than $k$-RDMs.","The parallel readout buys its $R\\to\\sqrt{R}$ improvement with extra probe qubits, so on a device where qubit count is the scarcer resource the symmetry-only Method I may be the more practical choice even though it uses more state-preparation calls.","The quoted 100x and 500x factors count queries to $U_\\psi$ and its inverse only; an end-to-end resource estimate that includes the block-encoding and QSVT circuits could shrink the practical gap, so a natural next calculation is a full fault-tolerant cost for FeMoco at $\\varepsilon=10^{-3}$."],"forward_implications":["For $k$-RDM estimation at $\\eta=k+O(1)$ or $\\eta=N-O(1)$, the oracle query count scales as $\\sqrt{\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}}\\,\\log d_\\eta\\log M/\\varepsilon$, a quadratic improvement over the previous adaptive QGE baseline.","In the paper's numerical comparison at $\\varepsilon=10^{-3}$, the total queries to the state-preparation oracle drop by roughly 100x for the FeMoco active space and 500x for the 100-site doped Hubbard model when estimating 2-RDM elements.","For 2-RDM estimation at filling $\\eta=\\lceil 7N/8\\rceil$, Method II achieves the lowest query count among the compared algorithms for every system size $N$ considered.","For the FeMoco active-space model, Method II has the lowest query count among the compared algorithms for target precision $\\varepsilon \\lesssim 10^{-2}$.","For $k \\geq 3$, the asymptotic saving over competing methods becomes even larger because the query count is set by the square root of the same binomial product divided by $\\varepsilon$."],"supporting_citations":[{"why":"Earlier multiple-observables estimation algorithm used as a baseline that the new methods improve on.","marker":"[14]"},{"why":"Defines the adaptive QGE algorithm whose query complexity the two new methods reduce.","marker":"[16]"},{"why":"Provides quantum singular value transformation, the technique behind Lemma 1 and the subspace-restricted encodings.","marker":"[32]"},{"why":"Supplies uniform singular value amplification used in step (ii) of the encode phase.","marker":"[33]"},{"why":"Gives the Hamiltonian simulation result that fixes the O(t) query scaling in the encode step.","marker":"[34]"},{"why":"Companion paper to which the complete proofs of Theorems 1 and 2 are deferred.","marker":"[35]"},{"why":"Fermionic classical-shadow tomography algorithm used as the main state-of-the-art baseline in the numerical query comparisons.","marker":"[29]"}],"fun_headline_variants":["Fermionic estimation gets quadratic speedup via symmetry","Adaptive QGE improves fermionic tomography with symmetry","Parallel scheme and symmetry slash fermionic query count","100x fewer queries for FeMo cofactor 2-RDM estimation","New algorithm speeds fermionic observable estimation quadratically"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The quantitative speedups rest on an asserted but unproved identity for the sum of squared norms of $k$-RDM operators in an $\\eta$-particle subspace, $\\sum_j [O_j^{(\\eta)}]^2=\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}$, together with complete proofs of Theorems 1 and 2 that are deferred to a companion paper.","fun_headline_variants_meta":{"raw":{"variants":["Fermionic estimation gets quadratic speedup via symmetry","Adaptive QGE improves fermionic tomography with symmetry","Parallel scheme and symmetry slash fermionic query count","100x fewer queries for FeMo cofactor 2-RDM estimation","New algorithm speeds fermionic observable estimation quadratically"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000259,"raw_usage":{"total_tokens":1618,"prompt_tokens":1010,"completion_tokens":608,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":626,"completion_tokens_details":{"reasoning_tokens":529}},"tokens_in":626,"tokens_out":608,"duration_ms":5711,"temperature":1.0,"reasoning_tokens":529,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T04:38:38.142266+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Work through the paper's normalization for a minimal case, say $k=1$, $\\eta=1$, $N=2$, by explicitly constructing the Hermitianized $k$-RDM operators, restricting to the one-particle subspace, and summing squared spectral norms; if the sum does not equal $\\binom{\\eta}{k}\\binom{N-\\eta+k}{k}$, the claimed 100x and 500x reductions fail by that factor. The same check can be repeated by reproducing the FeMoco and Hubbard query counts at $\\varepsilon=10^{-3}$ from the paper's formulas.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier multiple-observables estimation algorithm used as a baseline that the new methods improve on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the adaptive QGE algorithm whose query complexity the two new methods reduce."},{"cited_title":"Gilyén, Y","cited_arxiv_id":null,"evidence_quote":"Provides quantum singular value transformation, the technique behind Lemma 1 and the subspace-restricted encodings."}],"review_version":1}