{"id":"e9890fd8-3656-422a-8e05-a880881718b5","arxiv_id":"2608.13477","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Time-sliced SSE with boundary-based operators and loop crossings yields unbiased off-diagonal imaginary-time correlators, benchmarked on Ising and XXZ chains.","lead":"Imaginary-time correlation functions of off-diagonal operators are hard to measure in stochastic series expansion quantum Monte Carlo. This paper introduces time-sliced estimators that make them efficient and unbiased, enabling easier extraction of real-frequency spectra.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. V.A's exclusion of type-(b) loop closures is asserted without a detailed-balance proof; if biased, the O∉H estimator is uncontrolled.","rationale":"The reader's weakest assumption is correct and is the most load-bearing gap. The O∉H construction is the only part of the paper whose unbiasedness rests on a non-standard modification of the directed-loop update, and Eqs. (47)-(51) provide no proof that the Markov chain on C' samples W(C') when type-(b) closures are rejected. The TFIM O∈H estimator and diagonal estimator follow from standard reindexing and are not at issue. The conditional verdict is appropriate: the method is plausible and the XXZ benchmarks support it, but the missing detailed-balance argument should be supplied or the closure policy validated independently.","tokens_in":24598,"tokens_out":19656,"duration_ms":192452,"concrete_test":"On the L=10 XXZ chain at Δ=1, h=0.5, β=6L, Δτ=0.6, implement the Sec. V.A importance-sampling estimator twice: once excluding type-(b) closures by continuing the loop, and once accepting type-(b) closures and appending the tail as noted in Fig. 5. Compare G±(0,τ) from both variants with exact diagonalization over τ∈[0,β/2]. Agreement between the two variants and ED to within statistical error would support the no-bias claim; a systematic discrepancy would confirm the concern.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The O∉H estimator hinges on the claim in Sec. V.A (Fig. 5) that directed-loop closures of type (b) can be excluded entirely with no bias by continuing the loop until a type-(a) closure occurs. This is not derived. The directed-loop probabilities of Ref. [10] are tuned for the closed configuration space C; the extended space C' of Eq. (47) has weights that depend on the S+/S- insertion positions, and no proof is given that the standard vertex probabilities satisfy detailed balance in C' or that the modified closure rule preserves the stationary distribution. The importance-sampling weightings in Eqs. (51a)-(51b), with the factor n_x(l0) n_legs(C), are justified only by a compensation argument for the starting-leg probability; the path probabilities themselves are not shown to cancel. If this assumption fails, the histogram-based G± is biased in a way that small-chain ED benchmarks may or may not expose, and the central claim of efficient unbiased off-diagonal correlators for O∉H breaks down.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a time-sliced variant of stochastic series expansion (SSE) in which the imaginary-time axis is divided into m slices of width Δτ, and constructs estimators for three classes of imaginary-time correlation functions: diagonal operators (Sec. III), off-diagonal operators that appear in the Hamiltonian, O∈H (Sec. IV), and off-diagonal operators not in the Hamiltonian, O∉H (Sec. V). For O∈H, the boundary-valued estimator counts operators adjacent to slice boundaries (Eq. 33) with no discretization error, and an alternative time-slice-averaged estimator (Eqs. 36-39) trades a derived O(Δτ²) bias (Eq. 43) for an order-of-magnitude variance reduction. For O∉H, the paper proposes sampling an extended configuration space C' during directed-loop updates and accumulating histogram contributions from loop crossings of slice boundaries (Sec. V.A, Eqs. 50-51). The method is benchmarked against exact diagonalization for the 1D TFIM and XXZ chain, with additional larger-scale results extracting scaling dimensions and a Luttinger parameter.","tokens_in":24770,"tokens_out":27855,"duration_ms":269504,"significance":"If the method is unbiased as claimed, it removes a long-standing restriction of SSE-based dynamical calculations by making transverse off-diagonal imaginary-time correlators accessible with only modest overhead beyond standard loop updates. The paper's strengths include explicit, parameter-free estimator formulas (the only free parameter is the slice width Δτ), a clean derivation of the O(Δτ²) bias for the time-slice-averaged O∈H estimator, careful ED benchmarks that agree to few percent or better, and large-scale demonstrations of efficiency. The main weakness is that the central claim for the O∉H estimator rests on an assertion about excluding type-(b) loop closures without a detailed-balance proof, and the O∈H algorithm as written contains a potential double-counting ambiguity. These issues are fixable but are load-bearing for the paper's advertised generality.","major_comments":[{"comment":"The claim that loop closures of type (b) can be excluded entirely with no bias is asserted without proof. The directed-loop transition probabilities of Ref. [10] were derived for the closed configuration space C, whereas the estimator samples the extended space C' whose weights in Eq. (47) depend on the S+/S− insertion points. The compensation factors in Eqs. (51a)-(51b) account only for the probability of selecting the starting leg and boundary; they do not establish that the path probabilities in C' satisfy detailed balance or that continuing the loop until a type-(a) closure occurs preserves the stationary distribution. Because this assumption is load-bearing for the unbiasedness of G±(r,τ) for O∉H, please provide a derivation (or a rigorous argument) that the standard vertex probabilities, together with the modified closure rule, sample C' with the correct weights. Small-system ED benchmarks are not by themselves sufficient to rule out a systematic bias that could grow with system size or parameter range.","section":"Sec. V.A (Fig. 5, Eq. (47))"},{"comment":"The boundary-valued estimator is implemented with an 'if ... else if ...' rule that records at most one adjacent operator per time-slice boundary. However, if both the last non-identity operator of slice k−1 and the first non-identity operator of slice k are the same operator O(r), the insertion argument leading to Eq. (33) admits two distinct contributions (O inserted as H_{k−1,n_{k−1}} or as H_{k,1}), so the correct estimator should sum n_{k−1} and n_k rather than select one of them. The comment that this 'avoids double counting' appears to conflate two distinct insertion positions. Please clarify whether the implementation sums both cases; if the pseudocode is literal, the O∈H estimator is biased and the ED agreement in Fig. 6 would need to be re-examined.","section":"Sec. IV (Eq. (33), pseudocode after Fig. 3)"}],"minor_comments":[{"comment":"The derivation of ⟨H_k⟩ carries a minus sign from (−β)^n that is inconsistent with the convention H=−∑_b H_b with positive H_b adopted in Eq. (4). Equation (29) and the later statement that H_k=hσ^x (requiring division by h²) indicate the positive convention; please make the signs consistent throughout.","section":"Sec. IV, Eqs. (25)-(28)"},{"comment":"The statement that n_legs(C) is proportional to n_H, followed by a suggestion to normalize by 4mM as 'the total number of legs including identities', is confusing; please define n_legs(C) unambiguously and clarify whether the directed-loop graph includes identity vertices.","section":"Sec. V.A, Eq. (51)"},{"comment":"The extracted Luttinger parameter K=0.722 is 4% below the Bethe-ansatz value 3/4; please report the fit range, the statistical error of K, and whether the deviation is expected from finite-size or finite-Δτ effects.","section":"Fig. 10 (left)"},{"comment":"The abstract's statement that the method provides correlation functions 'with no discretization error' is qualified for the time-slice-averaged O∈H estimator in Sec. IV.A, which has an O(Δτ²) bias; consider adding a sentence noting that the no-discretization-error property applies to the boundary-valued estimators.","section":"Abstract and Sec. I"}],"recommendation":"major_revision","confidential_remarks":"The main risk to the paper is the unproved unbiasedness of the O∉H estimator in Sec. V.A. If the authors cannot supply a detailed-balance proof, the central claim for O∉H would need to be weakened or supported by additional systematic tests (e.g., convergence with Δτ and system size, or a direct comparison against an independent worm-type implementation). The O∈H pseudocode issue in Sec. IV is also worth resolving before publication. The paper is otherwise a solid and potentially useful contribution, and I would not reject it at this stage."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is a practical way to get imaginary-time correlators in SSE for off-diagonal operators, not just diagonal ones. The operator-counting estimator for O∈H (Sec. IV) is straightforward, well derived, and matches ED to statistical accuracy on the TFIM. The time-slice-averaged variant with its O(Δτ²) bias is a nice bias–variance tradeoff, and the variance reduction they show is real. That part of the paper is solid and likely to be widely used.\n\nThe O∉H estimator (Sec. V) is more delicate and is where I have the main reservation. The histogram-based directed-loop construction is clever, and the XXZ benchmarks against ED at the 0.1% level are genuinely impressive. But the unbiasedness rests on two claims that are asserted rather than proven: first, that type-(b) loop closures can be excluded with no bias, and second, that the compensating factors n_x(l0)n_legs(C) fully account for the configuration-dependent starting-leg selection. The first claim in particular looks like it could break detailed balance if the excluded closures are part of the standard directed-loop transition matrix. The authors should either provide a detailed-balance argument or demonstrate the equivalence more carefully. Small-chain ED agreement is good evidence, but it is not conclusive, especially for a methods paper.\n\nOther soft spots are minor: no code or data is released, which hinders exact reproduction; the Luttinger parameter extraction is 4% off the Bethe value, which is fine but worth a comment; and the paper leans on prior time-slicing work for the diagonal case, as it should. The citation pattern is appropriate and the positioning against worm-type approaches is fair.\n\nOverall, this is a useful contribution that removes a real bottleneck for SSE-based analytic continuation. It deserves a serious referee, but not a desk reject. The referee should ask for a proof or more extensive validation of the O∉H sampling assumption—perhaps tests on larger chains or in regimes with frequent bounces where a bias would be more visible. If that is resolved, the paper is publishable as is.","headline":"Solid SSE methods paper: the O∈H estimator is clean and benchmarked, but the O∉H importance-sampling estimator has an unproven technical assumption that deserves scrutiny before publication.","tokens_in":25347,"tokens_out":9374,"would_cite":true,"duration_ms":94119,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["05.10.Ln","75.40.Mg"],"model":"deepseek-v4-flash","headline":"Time-sliced stochastic series expansion computes imaginary-time correlation functions of diagonal and off-diagonal operators at well-defined imaginary-time points, with no discretization error and only minimal overhead on top of existing…","keywords":["quantum Monte Carlo","stochastic series expansion","imaginary-time correlation functions","directed-loop updates","time-sliced SSE","analytic continuation","transverse-field Ising model"],"falsifier":"Run the importance-sampling $O \\notin H$ estimator on a small XXZ chain in a field ($L=10$, $h=0.5$, as in Fig. 9) side by side with the boundary-start estimator and exact diagonalization, and compare the normalized $\\tau$-resolved histograms at high statistics: a systematic deviation between the two estimators or from exact diagonalization would refute the claim. A more direct check is to verify that the standard directed-loop transition probabilities leave the extended-state weights $W(C')$ invariant; if they do not, the trial probabilities in the open-defect sector are wrong and the histogram is biased.","tokens_in":2138,"feed_emoji":"🎲","tokens_out":10917,"duration_ms":256678,"temperature":0.7,"pith_summary":"Time-sliced stochastic series expansion (SSE) splits the quantum operator string into $m$ slices of width $\\Delta\\tau = \\beta/m$, turning each slice boundary into a well-defined imaginary-time coordinate. On this grid the paper derives exact estimators for all three operator classes: diagonal correlators from propagated states at the boundaries; correlators of operators that appear in the Hamiltonian from counting the operators adjacent to a boundary, with estimator $n_l n_{l'}/\\Delta\\tau^2$; and correlators of operators not in the Hamiltonian from a histogram of slice-boundary crossings accumulated while directed loops propagate. The estimators require no change to the loop or cluster updates themselves and add only minimal computational overhead. Benchmarks on the one-dimensional transverse-field Ising chain and the XXZ chain agree with exact diagonalization within statistical error, and $L=128$ runs extract scaling dimensions and a Luttinger parameter close to known values. If correct, the method removes the long-standing restriction of SSE dynamics to diagonal structure factors or SU(2)-symmetric models, opening transverse correlators to analytic continuation.","feed_headline":"Time-sliced SSE yields exact off-diagonal correlators","feed_subtitle":"Counting boundary operators and loop crossings reads out transverse spin dynamics at minimal extra cost.","key_machinery":"The central object is the sliced operator string: writing $Z = \\mathrm{Tr}[(e^{-\\Delta\\tau H})^m]$ and expanding each of the $m$ slices independently converts every slice boundary into an exact imaginary-time point $\\tau_k = k\\Delta\\tau$. Two identities carry the argument. An operator-absorption reindexing, applied within a single slice, turns an insertion of $O \\in H$ at a boundary into a factor $n_l/\\Delta\\tau$, the number of non-identity operators in slice $l$ divided by the slice width, which yields the counting estimator $f(C) = n_l n_{l'}/\\Delta\\tau^2$ for a pair of boundary-adjacent operators. For $O \\notin H$, the workhorse is the correspondence between directed-loop propagation and sampling of the extended two-defect configuration space $C'$: the loop's head and tail are the $S^+$ and $S^-$ insertions, and the correlator histogram is accumulated from the head's crossings of slice boundaries during the update, with the factor $n_x(l_0)\\,n_{\\mathrm{legs}}(C)$ compensating for the importance-sampling choice of the starting leg.","core_discovery":"The paper claims that an explicit discretization of the SSE operator string into $m$ independent slices, with the partition function written as $\\mathrm{Tr}[(e^{-\\Delta\\tau H})^m]$, makes imaginary-time correlation functions measurable on the exact grid $\\tau_k = k\\Delta\\tau$ with no discretization error. For operators $O \\in H$, it derives a counting estimator: absorbing $O$ at a slice boundary reindexes that slice's expansion and leaves a factor $n_l/\\Delta\\tau$, where $n_l$ is the number of non-identity operators in slice $l$, so the two-point correlator at nonzero separation is $f(C) = n_l n_{l'}/\\Delta\\tau^2$ (Eq. 33), with a separate expression at equal time (Eq. 34). For operators $O \\notin H$, it treats the directed loop as sampling an extended configuration space $C'$ in which $S^+$ and $S^-$ defects sit at slice boundaries, and measures the correlator as a histogram of the loop's boundary crossings, weighted by $n_x(l_0)\\,n_{\\mathrm{legs}}(C)$ in the importance-sampling variant to undo the bias of the starting-leg choice. Both estimator families are validated against exact diagonalization on small transverse-field Ising and XXZ systems and demonstrated at $L=128$, where the extracted exponents agree with expected values and the SU(2)-symmetric relation $2G_{zz} = G_{\\pm}$ holds within error bars.","pith_inferences":["The unbiasedness claim for the $O \\notin H$ estimator rests on the type-(b) closure exclusion, which the paper asserts without proof; a direct high-statistics comparison of the importance-sampling and boundary-start estimators on a small XXZ system would settle this empirically, since each is currently validated only against exact diagonalization.","Because the boundary-valued and time-slice averaged $O \\in H$ estimators differ by a known $O(\\Delta\\tau^2)$ term, their difference at fixed $\\Delta\\tau$ is a ready-made convergence diagnostic for deciding when a data set is clean enough to feed analytic continuation.","Read as a worm-type algorithm on the slice-boundary grid, the $O \\notin H$ construction should carry over to sign-problem-free boson and fermion Hamiltonians, giving two-point Green's functions $\\langle c^\\dagger(r,\\tau)c(0,0)\\rangle$ without the single-hole specialization of recent $t$-$J$-type studies.","The histogram is unnormalized and its absolute scale is fixed by pinning a known value such as $G_{\\pm}(0,0)=1/2$; applications where the equal-time value is not independently known would need a separate normalization scheme, which the paper leaves open."],"forward_implications":["Transverse ($S^x$, $S^\\pm$-type) imaginary-time correlators become accessible in SSE quantum Monte Carlo for models without SU(2) symmetry, such as anisotropic XXZ chains with external fields, lifting a restriction that previously limited SSE dynamics largely to diagonal structure factors.","The $\\tau$-resolved correlators feed directly into numerical analytic continuation and thus yield real-frequency transverse spectral functions from the same simulation ensemble that produces static observables.","The overhead stays near the cost of the simulation itself: with the FFT the diagonal and $O \\in H$ estimators scale as $O(mL^d \\log mL^d)$, while the $O \\notin H$ estimator scales as $O(mL^d)$, matching the SSE updates.","A time-slice averaged variant of the $O \\in H$ estimator trades a controlled $O(\\Delta\\tau^2)$ bias (larger, $O(\\Delta\\tau)$, at equal time) for an order-of-magnitude variance reduction, which the benchmarks show is advantageous for extracting decay exponents.","The construction transfers to any sign-problem-free model with similar operator structure, including bosonic correlators such as $\\langle c^\\dagger(r,\\tau)c(0,0)\\rangle$."],"supporting_citations":[{"why":"Supplies the directed-loop update whose slice-boundary crossings form the estimator for operators not in the Hamiltonian.","marker":"[10]"},{"why":"Provides the earlier SSE formulation of off-diagonal dynamics and the two-defect extended-space construction this paper adapts to the time-sliced grid.","marker":"[62]"},{"why":"Established the finite-temperature time-sliced SSE measurement of diagonal correlators at slice boundaries.","marker":"[60]"},{"why":"Introduced the time-sliced SSE (projector) whose slice boundaries define the imaginary-time grid used throughout.","marker":"[67]"},{"why":"Supplies the operator-absorption reindexing argument used for the in-Hamiltonian estimator.","marker":"[4]"},{"why":"Provides the cluster update used in the transverse-field Ising benchmarks of the in-Hamiltonian estimator.","marker":"[16]"},{"why":"The worm-algorithm construction motivating sampling an extended configuration space with two defects.","marker":"[61]"}],"fun_headline_variants":["Time-sliced SSE achieves exact correlators","Exact imaginary-time correlators via SSE slicing","Sliced SSE: no discretization error in correlators","Off-diagonal correlators exact with time-sliced SSE","New SSE slicing gives exact dynamic correlations"],"cache_read_input_tokens":27520,"weakest_assumption_plain":"The load-bearing premise is that, in the extended two-defect configurations sampled for operators outside the Hamiltonian, the standard directed-loop transition probabilities remain valid and loop closures of type (b) — where the loop rejoins its starting vertex through an exit leg — can be discarded from the measurement histogram without bias; the paper asserts this but gives no proof that the sampling weights stay correct.","fun_headline_variants_meta":{"raw":{"variants":["Time-sliced SSE achieves exact correlators","Exact imaginary-time correlators via SSE slicing","Sliced SSE: no discretization error in correlators","Off-diagonal correlators exact with time-sliced SSE","New SSE slicing gives exact dynamic correlations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000908,"raw_usage":{"total_tokens":3955,"prompt_tokens":1051,"completion_tokens":2904,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":667,"completion_tokens_details":{"reasoning_tokens":2829}},"tokens_in":667,"tokens_out":2904,"duration_ms":19394,"temperature":1.0,"reasoning_tokens":2829,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:09:57.849518+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the importance-sampling $O \\notin H$ estimator on a small XXZ chain in a field ($L=10$, $h=0.5$, as in Fig. 9) side by side with the boundary-start estimator and exact diagonalization, and compare the normalized $\\tau$-resolved histograms at high statistics: a systematic deviation between the two estimators or from exact diagonalization would refute the claim. A more direct check is to verify that the standard directed-loop transition probabilities leave the extended-state weights $W(C')$ invariant; if they do not, the trial probabilities in the open-defect sector are wrong and the histogram is biased.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the directed-loop update whose slice-boundary crossings form the estimator for operators not in the Hamiltonian."},{"cited_title":"Dorneich and M","cited_arxiv_id":null,"evidence_quote":"Provides the earlier SSE formulation of off-diagonal dynamics and the two-defect extended-space construction this paper adapts to the time-sliced grid."},{"cited_title":"Sandvik, Stochastic series expansion meth- ods, inMany-Body Methods for Real Materials, Vol","cited_arxiv_id":null,"evidence_quote":"Established the finite-temperature time-sliced SSE measurement of diagonal correlators at slice boundaries."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduced the time-sliced SSE (projector) whose slice boundaries define the imaginary-time grid used throughout."},{"cited_title":"Sandvik, A generalization of Handscomb’s quantum Monte Carlo scheme — Application to the 1-D Hubbard model, Journal of Physics A: Mathematical and General25, 3667 (1992)","cited_arxiv_id":null,"evidence_quote":"Supplies the operator-absorption reindexing argument used for the in-Hamiltonian estimator."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the cluster update used in the transverse-field Ising benchmarks of the in-Hamiltonian estimator."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The worm-algorithm construction motivating sampling an extended configuration space with two defects."}],"review_version":1}