{"id":"facc1309-d860-47df-9fc2-cc1908010550","arxiv_id":"2608.07133","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A general autonomization framework maps time-dependent non-unitary ODEs to time-independent linear systems, and paired with Schrödingerization, achieves log^{5/4}(1/ε) precision dependence.","lead":"The authors introduce a clock-variable framework that converts time-dependent non-unitary quantum dynamics into a time-independent linear system, so existing quantum ODE solvers can be applied directly. Combined with Schrödingerization, the method claims an improved error scaling of log^{5/4}(1/ε), beating the log^2(1/ε) of prior work.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Under the stated assumptions, the advertised log^{5/4}(1/ε) precision dependence is not established: real-analytic A(t) can grow on the expanding clock interval, and Lemma 2.1's error bound omits derivatives of B(s).","rationale":"The central construction—autonomization via a clock variable followed by Fourier spectral discretization—is coherent, and the algebraic derivations of the block-encodings are plausible. The paper's own numerics validate the formulation only on a fixed smooth two-cavity example and do not exercise the complexity claim. The reader identified analyticity as load-bearing; I partially agree, but the sharper issue is that analyticity alone is insufficient: the spectral error in Lemma 2.1 must control the full solution's derivatives, which include B and its derivatives, and the block-encoding normalization α_A is evaluated on an interval whose length grows as log^{1/2}(1/ε) for the profile used to get exponent 5/4. A dissipative real-analytic example with exponential growth makes the claimed precision exponent fail. This does not invalidate the framework, but it means the headline complexity requires an additional boundedness/derivative assumption and a proof of Lemma 2.1 that tracks B. Hence the paper should be accepted only conditionally on these corrections.","tokens_in":32487,"tokens_out":27136,"duration_ms":259718,"concrete_test":"Take A(t)=i e^t I, b(t)=0, x(0)=e_1 in Theorem 3.1 and the profile G_f. Set δ_s=ε/(4Cg_A); then R_s=2 log^{1/2}(1/δ_s)+2T and α_A=max_{s∈I_s}|e^s|=exp(2 log^{1/2}(1/δ_s)+2T). Substituting into (3.13) gives matrix-query ε-dependence exp(O(log^{1/2}(1/ε)))·log^{5/4}(1/ε), contradicting the abstract. If the intended hypothesis is bounded analyticity in a fixed strip, state it and re-derive Lemma 2.1 with an explicit bound on ||∂_s^r z(T,·)|| involving ||B^(r)||; verify that the exponent β in (2.20) is determined by B as well as G.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Lemma 2.1 uses the Fourier projection error (2.18) for z(T,s) but bounds it only by ||G^(r)||_{L2(I_s)}, leading to μ_s,max=O(log^β(1/δ_s)) in (2.20). The solution of (2.2) is z(T,s)=Texp(∫_0^T B(s-T+τ)dτ)G(s-T)u0, whose r-th s-derivative also involves B=[[A,F],[0,0]] and its derivatives along the characteristics. Thus the spectral error is controlled by the regularity of A and b on I_s, not just by G. The assumptions after (2.1) only say A,b are real analytic on R; real analytic functions need not be bounded or have uniformly controlled derivatives on an interval that grows with ε. For the headline profile G_f, R_s=2 log^{1/2}(1/δ_s); a dissipative, real-analytic example A(t)=i e^t I gives α_A=max_{s∈I_s}||A(s)||=exp(O(log^{1/2}(1/δ_s))). This ε-dependent normalization enters (2.28) and the final bound O(g_A α_A T log^{5/4}(g_A/ε)) in Theorem 3.1, so the precision dependence is e^{O(log^{1/2}(1/ε))}, not log^{5/4}(1/ε). Even with bounded A, (2.20) needs an explicit bound on ||∂_s^r z(T,·)||^{1/r}, which the paper does not provide.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an 'autonomization' framework for quantum simulation of time-dependent non-unitary linear ODEs. The non-autonomous system (1.1) is lifted via the clock-variable construction of Cao–Jin–Liu to an autonomous transport-type PDE (2.2), whose Fourier spectral discretization yields the explicit time-independent matrix Ā in (2.15). The framework is modular: Ā can be fed into any quantum ODE solver for time-independent systems. The authors combine Ā with the Schrödingerization method and with the BCOW Taylor-expansion solver, deriving query complexities in Theorems 3.1 and 4.2. The headline claim is that, under real-analyticity and dissipativity of A(t), the precision dependence of the Schrödingerization-based combination scales as log^{5/4}(1/ε), improving on the log^2(1/ε) scaling of existing approaches. Numerical experiments on a two-cavity system validate recovery of the target trajectory for both combinations.","tokens_in":32721,"tokens_out":13973,"duration_ms":127805,"significance":"The modular autonomization idea is attractive: it decouples the treatment of time dependence from the choice of the underlying time-independent solver, and the block-encoding circuits in Section 2.3 are explicit. The non-diagonalizable BCOW analysis in Section 4.2 is a useful extension of the original Berry–Childs–Ostrander–Wang result. If the complexity claims were fully established, the log^{5/4}(1/ε) precision dependence would be a noteworthy improvement over the competing log^2(1/ε) scaling. However, the two load-bearing analytical steps — the spectral error bound in Lemma 2.1 and the treatment of the clock-interval length in Theorem 3.1 — are not justified under the stated assumptions. The advertised precision dependence is therefore not currently established, although the framework and numerics remain plausible and potentially salvageable with strengthened assumptions and a corrected proof.","major_comments":[{"comment":"The proof of Lemma 2.1 is not provided; it is asserted by 'following the similar arguments in [30]'. This does not transfer to the present setting. The exact solution (2.5) is z(T,s) = Texp(∫_0^T B(s-T+τ)dτ)G(s-T)u_0, so ∂_s^r z(T,·) involves derivatives of B(s) of orders up to r, not just derivatives of G. The Fourier projection error (2.18) controls the error only through the regularity of the function being projected, which is z(T,·), not G alone. Consequently the bound ∥z(T,s_l) - z_h(T,s_l)∥ ≲ (Δs)^{r-1/2}∥G^{(r)}∥_{L^2(I_s)}∥u(0)∥ in Lemma 2.1 is not established. The assumption after (2.1) that A,b are real analytic on R does not supply uniform control of the relevant derivatives on the interval I_s, whose length grows with δ_s^{-1} for the profile G_f. A concrete counterexample to the claimed bound is A(t)=i e^t I with b=0, for which z(T,s) = exp(i(e^s - e^{s-T}))G(s-T)u_0; the r-th s-derivative grows like e^{rs} on I_s, so the spectral error is not governed by ∥G^{(r)}∥ alone. The proof must either estimate the full r-th derivative of z(T,·), including the B-dependent terms, or impose explicit regularity and boundedness assumptions on B(s) over I_s.","section":"Lemma 2.1, Section 2.2"},{"comment":"The advertised log^{5/4}(1/ε) precision dependence is not derived under the stated assumptions. The interval I_s = [-R_s, R_s+2T] has R_s = 2 log^{1/2}(1/δ_s) for the headline profile G_f in (2.9), so the grid points s_l range over an interval whose length grows with ε^{-1}. The quantity α_A in (2.22) is defined as max_{0≤l≤N_s-1}∥A(s_l)∥ and enters the final bound O(g_A α_A T log^{max{1+β/2,3β/2}}(g_A/ε)) in Theorem 3.1. Real analyticity on R does not prevent α_A from growing on this expanding interval. For the dissipative, real-analytic example A(t)=i e^t I, one has α_A = e^{R_s+2T} = e^{O(log^{1/2}(1/δ_s))}, which converts the stated complexity into e^{O(log^{1/2}(1/ε))} instead of log^{5/4}(1/ε). The paper needs either a uniform boundedness assumption on A(s) and b(s) (and suitable derivatives) over I_s with a constant independent of ε, or a more careful statement that separates the ε-dependence of α_A from the precision dependence. As written, the abstract's 'precision dependence can scale as log^{5/4}(1/ε)' is not supported.","section":"Assumptions after (2.1) and Theorem 3.1"},{"comment":"The definition of G_m is inconsistent with the stated property G_m(0)=1. With η(s)=e^{1/(s^2-1)} for |s|<1, one has η(0)=e^{-1}, hence G_m(0)=e^{e^{-1}}≠1. If the intended mollifier is η(s)=1/(s^2-1), then G_m(0)=e^{-1}≠1. The manuscript uses G(0)=1 in the recovery map (2.17), in the success-probability computation in Theorem 3.1 step (2), and in Theorem 4.2. This is a concrete error in a central definition. The profile should be normalized by dividing by G_m(0), or η should be shifted so that η(0)=0; the derivative estimates in Theorem 2.2 are unaffected up to constants, but the text and recovery formulas must be corrected.","section":"Equation (2.6), Section 2.1"}],"minor_comments":[{"comment":"The claimed improvement over [26] is stated without giving the comparison complexity of [26]. To make the 'improving upon log^2(1/ε)' assertion verifiable, the authors should state the query complexity of [26] in the same notation and identify precisely the source of the saving.","section":"Remark 3.2 and Introduction"},{"comment":"In the proof of Lemma 2.1, the phrase '∀l≥0' is unclear because the grid index l ranges over 0≤l≤N_s-1 and the pointwise estimate is needed at the recovery point s=T=s_{N_s/2}; this should be stated directly.","section":"Lemma 2.1 proof"},{"comment":"The pointwise Fourier projection estimate from [34] applies to periodic functions; for the profiles G_w and G_f, which are only approximately supported, the boundary values of G and its derivatives at s=±R_s are small but not exactly zero. The additional contribution of the boundary discontinuity to the estimate (2.18) should be addressed explicitly rather than left to the cited literature.","section":"Section 2.2, Eq. (2.18)"},{"comment":"The acknowledgements mention 'NL acknowledges funding from ...', but the author list contains no 'NL'. This appears to be a leftover from a different version and should be corrected (or the name spelled out) before publication.","section":"Acknowledgements"}],"recommendation":"major_revision","confidential_remarks":"The central error estimate of the paper (Lemma 2.1) is delegated to a citation to [30], whose author list overlaps with two of the present authors. Given the variable-coefficient structure of Eq. (2.2), the cited result does not apply as stated, and a self-contained proof is essential. The ε-dependence of α_A over the growing clock interval is a separate, potentially fatal issue for the headline log^{5/4} claim. Both can likely be repaired by strengthening the assumptions (e.g., bounded real-analytic coefficients on a fixed strip) and reworking the Fourier error bound, but the revisions are substantial. The numerical experiments are encouraging but only validate recovery of the solution, not the precision-scaling claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The autonomization construction is clean, and the modular claim is genuine: lifting time-dependent non-unitary ODEs to a time-independent transport equation via a clock variable is a real extension of [1], and the explicit block-encoding is careful. The numerical experiments do confirm that the formulation produces the right trajectory on the two-cavity example.\n\nThe soft spot is the complexity analysis. Lemma 2.1 is the load-bearing step, and it is not proved. The argument says the Fourier projection error of z(T,s) is bounded by (Δs)^{r-1/2} ||G^(r)||_{L^2(I_s)}, but z(T,s) = exp_T(∫_0^T B(s-T+τ)dτ) G(s-T) u0, and its r-th s-derivative involves derivatives of B along the characteristic. That is exactly the difference from [30], where the p-profile stays attached to the initial ψ. So the bound as written is not a small technical gap; it omits the dominant terms.\n\nThis is not just cosmetic. With G_f, the clock interval grows like 2 log^{1/2}(1/δ_s). A perfectly reasonable real-analytic, dissipative coefficient such as A(t)=i e^t I has ||A(s)|| ~ e^{O(log^{1/2}(1/δ_s))} on that interval, so α_A — which enters linearly in Theorem 3.1 — carries a factor e^{O(√log(1/ε))}. The advertised log^{5/4}(1/ε) precision dependence does not follow from the stated assumptions. To get the polylog exponent, the authors need a uniform boundedness/regularity assumption on A and b over the growing clock interval, and a proof that the r-th s-derivative of z(T,·) is bounded by a constant multiple of ||G^(r)||, or an explicit bound in terms of derivatives of B. The text flags finite regularity as future work, but the problem here is not finite regularity; it is the s-dependence of B under real-analyticity alone.\n\nSmaller items: the G_m definition in (2.6) gives G_m(0)=e^{-1}≠1 as written, inconsistent with the recovery relation; that is an easy typo to fix. The proof of Lemma 2.1 is \"similar to [30]\", but the analogy does not hold, so this is where the referee should push.\n\nWho is this for: people working on quantum differential equation solvers will find the framework useful, and the BCOW non-diagonalizable analysis is a nice cleanup, but the headline complexity claim should be treated as unproven until the error analysis is repaired. I would send it to review, with the expectation of a major revision on Lemma 2.1 and the complexity theorem. The framework deserves serious referee time.","headline":"The autonomization framework is clean and likely correct, but the advertised log^{5/4}(1/ε) precision dependence is not established: Lemma 2.1 omits the B(s)-derivatives that control the clock-direction regularity, so the main complexity claim is unsupported.","tokens_in":33356,"tokens_out":5887,"would_cite":false,"duration_ms":54005,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","65M70","65L05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that time-dependent non-unitary linear dynamics can be autonomized with a clock variable into a time-independent linear system, and that combining it with Schrödingerization yields precision dependence log^{5/4}(1/ε)…","keywords":["autonomization","clock variable","time-dependent linear differential equations","non-unitary dynamics","quantum ODE solvers","Schrödingerization","block-encoding","Fourier spectral method"],"falsifier":"Take an exactly solvable non-autonomous system whose coefficients are only $C^k$ rather than analytic, run the autonomization construction for a decreasing sequence of targets $\\delta_s$, and measure the largest Fourier mode $\\mu_{s,\\max}$ needed to keep $\\|G - P_h G\\|_{L^\\infty} \\le \\delta_s$; if $\\mu_{s,\\max}$ grows like $\\delta_s^{-1/(k-1/2)}$ rather than $O(\\log^\\beta(1/\\delta_s))$, that polynomial blow-up would falsify the $\\log^{5/4}(1/\\varepsilon)$ complexity for that regularity class.","tokens_in":32146,"feed_emoji":"⚛️","tokens_out":10930,"duration_ms":89847,"temperature":0.7,"pith_summary":"Quantum algorithms for linear differential equations have mostly targeted time-independent coefficients; time-dependent non-unitary dynamics forces a time-ordered exponential and resists direct adaptation of existing solvers. This paper builds a bridge: introduce a clock variable $s$, lift the non-autonomous system to a transport-type equation in $(t,s)$, Fourier-discretize $s$, and recover the target solution at the grid point $s=T$. The result is an explicit time-independent linear system together with a block-encoding, so any quantum ODE solver for autonomous problems can be applied as a black box. With the Schrödingerization solver, the matrix-query count depends on the error as $\\log^{5/4}(1/\\varepsilon)$, improving the $\\log^2(1/\\varepsilon)$ scaling of the best existing approaches under the assumption that $A(t)$ and $b(t)$ are analytic in time. Two numerical experiments on a time-modulated two-cavity system validate the recovery.","feed_headline":"Quantum ODE precision cost drops to log^{5/4}(1/ε)","feed_subtitle":"Clock-variable autonomization lets time-independent solvers handle non-unitary, time-dependent quantum dynamics.","key_machinery":"The carrying object is the clock-variable autonomization identity together with its Fourier spectral discretization. The paper adds a clock register $s$, freezes the coefficient matrix as $B(s)$, and solves a convection equation whose method-of-characteristics solution reproduces the original propagator; a smooth initial profile $G(s)$ with $G(0)=1$ that decays super-exponentially (one of $G_m$, $G_w$, $G_f$) makes the problem effectively periodic and controls the largest Fourier mode as $\\mu_{s,\\max} = O(\\log^\\beta(1/\\delta_s))$ with $\\beta=3,1,1/2$ respectively. This produces the explicit time-independent matrix $\\bar{A}$, its block-encoding, and the recovery map $x_h(T) = R_x w(T)$ projected at the midpoint $s=T$. The same matrix is then shifted by $\\gamma = 1/(2T)$ to restore dissipativity for Schrödingerization, or fed directly into the BCOW linear-system construction.","core_discovery":"The central claim is that the time-ordered exponential is not an obstacle for time-dependent non-unitary systems. For the enlarged homogeneous system $du/dt = B(t)u$, define $z$ by $\\partial_t z + \\partial_s z = B(s)z$, $z(0,s)=G(s)u(0)$, with $G(0)=1$; then $z(t,s=t)=G(0)u(t)$, exactly. Fourier-spectral discretization in $s$ converts this into a static matrix $\\bar{A} = -iP_s \\otimes I + \\sum_l |l\\rangle\\langle l| \\otimes B(s_l)$, which the paper block-encodes from HAM-T oracles for $A(s)$ and $F(s)$ plus a Fourier-differentiation gadget. The paper proves two instantiations: with Schrödingerization, $O(g_A \\alpha_A T \\log^{\\max\\{1+\\beta/2, 3\\beta/2\\}}(g_A/\\varepsilon))$ queries to the matrix oracles, which becomes $\\log^{5/4}(1/\\varepsilon)$ for the super-exponentially decaying profile $G_f$ ($\\beta=1/2$); with a Taylor-expansion BCOW solver, $O(C_A^2 g_A \\alpha_A T \\log^{3(\\beta+1)/2}(\\alpha_A T C_A^2 g_A/\\varepsilon))$ queries, and the analysis removes the diagonalizability assumption by bounding complexity through the growth factor $C(A) = \\sup_t \\|e^{At}\\|$.","pith_inferences":["A testable extension the paper leaves implicit: for coefficients that are only $C^k$ in time, the Fourier mesh would need to scale polynomially in $1/\\delta_s$, so the $\\log^{5/4}$ precision dependence is tied to analyticity; one could quantify the crossover by mollifying a $C^k$ coefficient and balancing regularization error against discretization error.","The modular structure suggests a menu of clock profiles: the $\\beta=1/2$ profile is optimal for the Schrödingerization query count, while a compactly supported profile ($\\beta=3$) gives simpler boundary handling and might be preferable in implementations where exact support matters.","The midpoint recovery condition ($s=T$ on an even grid) is a small numerical-engineering constraint that may generalize to other spectral discretizations, effectively giving a quantum method of characteristics for non-autonomous equations."],"forward_implications":["Any existing quantum ODE solver for time-independent non-unitary systems can be reused for time-dependent problems by consuming the autonomization matrix $\\bar{A}$ and its block-encoding; time dependence is fully pushed into the HAM-T oracles.","For the analytic clock profile $G_f$, the Schrödingerization combination uses $O(g_A \\alpha_A T \\log^{5/4}(g_A/\\varepsilon))$ queries to the matrix oracles, improving the $\\varepsilon$-dependence of earlier constructions.","The BCOW combination inherits the same framework and its new analysis extends the algorithm to non-diagonalizable matrices, with cost controlled by the growth factor $C(A)$ instead of an eigenvector condition number.","The recovery postselects on the single grid point $s=T$ and then the $x$-block; with amplitude amplification the success probability is $\\Omega(1)$, so the algorithm outputs a state $\\varepsilon$-close to $|x(T)\\rangle$ with a success flag."],"supporting_citations":[{"why":"Supplies the clock-variable autonomization technique for time-dependent Hamiltonians, which this paper extends to non-unitary and inhomogeneous systems.","marker":"[1]"},{"why":"Provides the Schrödingerization method and the smooth-initialization and error estimates whose parameters the paper reuses to derive the log^{5/4} dependence.","marker":"[30]"},{"why":"Introduces the BCOW Taylor-expansion quantum ODE solver that the second autonomization combination is built on.","marker":"[20]"},{"why":"Supplies the matrix-exponential growth factor C(A) and inverse-norm estimates that let the paper extend BCOW to non-diagonalizable matrices.","marker":"[22]"},{"why":"Gives the best existing precision dependence log²(1/ε) that the paper compares against and improves.","marker":"[26]"},{"why":"Provides the standard pointwise Fourier projection error estimate used to set the clock-variable mesh size.","marker":"[34]"}],"fun_headline_variants":["Quantum ODE solver hits log^{5/4} precision scaling","Autonomization unlocks time-dependent non-unitary dynamics","Better precision scaling for quantum ODEs via clock variables","New quantum solver improves precision to log^{5/4}(1/ε)","Time-dependent non-unitary dynamics tamed by autonomization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that $A(t)$ and $b(t)$ are real analytic in time, because only then does the Fourier spectral error in the clock variable decay fast enough to keep the number of grid points logarithmic in the required accuracy; with merely finitely smooth coefficients the mesh size would grow polynomially and the $\\log^{5/4}(1/\\varepsilon)$ dependence is not established.","fun_headline_variants_meta":{"raw":{"variants":["Quantum ODE solver hits log^{5/4} precision scaling","Autonomization unlocks time-dependent non-unitary dynamics","Better precision scaling for quantum ODEs via clock variables","New quantum solver improves precision to log^{5/4}(1/ε)","Time-dependent non-unitary dynamics tamed by autonomization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000211,"raw_usage":{"total_tokens":1510,"prompt_tokens":1138,"completion_tokens":372,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":754,"completion_tokens_details":{"reasoning_tokens":286}},"tokens_in":754,"tokens_out":372,"duration_ms":3431,"temperature":1.0,"reasoning_tokens":286,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T14:14:24.345180+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take an exactly solvable non-autonomous system whose coefficients are only $C^k$ rather than analytic, run the autonomization construction for a decreasing sequence of targets $\\delta_s$, and measure the largest Fourier mode $\\mu_{s,\\max}$ needed to keep $\\|G - P_h G\\|_{L^\\infty} \\le \\delta_s$; if $\\mu_{s,\\max}$ grows like $\\delta_s^{-1/(k-1/2)}$ rather than $O(\\log^\\beta(1/\\delta_s))$, that polynomial blow-up would falsify the $\\log^{5/4}(1/\\varepsilon)$ complexity for that regularity class.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the clock-variable autonomization technique for time-dependent Hamiltonians, which this paper extends to non-unitary and inhomogeneous systems."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the BCOW Taylor-expansion quantum ODE solver that the second autonomization combination is built on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the matrix-exponential growth factor C(A) and inverse-norm estimates that let the paper extend BCOW to non-diagonalizable matrices."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standard pointwise Fourier projection error estimate used to set the clock-variable mesh size."}],"review_version":1}