{"id":"9e499fd6-60be-4683-bfa4-e79403693769","arxiv_id":"2607.22273","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"Auxiliary Walking reweights DMC walkers by the descendant count of a displaced auxiliary walker to remove trial-wavefunction bias from non-local operators such as the 1RDM.","lead":"This paper introduces Auxiliary Walking, a reweighting scheme meant to make Diffusion Monte Carlo estimates of non-local quantities like the one-body density matrix unbiased. It matters because DMC produces gold-standard energies while non-local observables have remained biased by the trial wavefunction.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper gives no algorithm for obtaining d(R'_i) in a continuous system: a walker exactly at R'_i exists with probability zero and no auxiliary branching construction is described, so Eq. (8) is not yet a realizable method.","rationale":"Good faith reading: the paper's contribution is the observation that non-local accumulator bias in FWD can be cancelled by weighting with descendants of a one-particle-transposed configuration. Appendix A's derivation (linear-Δ version) plus the Hubbard dimer exact result, where AUX is implemented exactly, give independent support for the formal identity. The Hubbard dimer alone is not enough: it uses known ψ0 to sample mixed distribution and analytic descendants, and has no time-step error. The helium demonstration, the only continuous test, is one AUX point at one time step and Fig. 3 shows it still deviates from exact; the paper attributes residual error to slow distribution convergence/ψT cancellation (Appendix B), not to construction error. So the load-bearing gap is not the algebra but the missing bridge from ideal weights to an operational continuous-space estimator. This is exactly the reader's weakest assumption. Recommendation: keep CONDITIONAL; the paper should either provide a concrete algorithm for obtaining d(R'_i) (spawned auxiliary walkers, or an unbiased approximation with controlled error), show a τ→0 extrapolation of AUX for helium, and quantify population-control bias in d(R'_i). Without that, Eq. (8) is a theorem about a process that has not been specified.","tokens_in":12125,"tokens_out":8622,"duration_ms":76375,"concrete_test":"Use the authors' He_atom_QMC code: implement an explicit auxiliary-walker algorithm, e.g. for each walker R_i at a DMC step, initialize an auxiliary walker at R'_i = {r', r2,...,rN}, propagate it by τ_FW with branching to measure its normalized descendant number d(R'_i), then evaluate Eq. (8) for the 1RDM cut used in Fig. 3 at τ=0.005 and τ=0.0025 (keeping τ_FW=5 Ha). If the RMSE of the AUX 1RDM does not extrapolate toward zero, or if the routine cannot be run because the auxiliary descendants are not defined, the central helium claim fails; a successful implementation with documented pseudo-code would remove the objection.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central identity Eq. (8) is formally correct under ideal assumptions: if a walker R'_i with one electron moved to r' had descendant weight exactly proportional to ψ0(R'_i)/ψT(R'_i), then reweighting the mixed estimate O_M(R_i) by d(R'_i) cancels both the numerator and denominator trial-function factors and yields the pure expectation. The problem is that the paper never specifies how R'_i is obtained in a continuous system. The text says such a walker 'could be found (or created)' (Sec. 2, after Eq. (7)), but in a continuous 3N-electron configuration space the probability that another population walker lies exactly at R'_i is zero, and no construction is described. To estimate d(R'_i) for a configuration that is not in the walker population one would need to spawn an auxiliary branching process at R'_i for every i (or at least for every needed r' slice), and the paper gives no such algorithm, cost analysis, or bias control. Appendix C documents AUX only for the Hubbard dimer, where the four discrete configurations make exact matching trivial. The only continuous-space AUX result is a single helium time step (Fig. 3) with no implementation detail. Consequently Eq. (8) is a formal identity with no demonstrated evaluation procedure for the system in which the method is claimed to be useful; the helium 'will give an unbiased 1RDM in the limits' assertion is not yet supported by a realizable construction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes 'Auxiliary Walking' (AUX), a reweighting scheme intended to remove the trial-wavefunction (accumulator) bias from Diffusion Monte Carlo estimates of non-local operators such as the one-body reduced density matrix. The central formula, Eq. (8), reweights each DMC walker R_i by the descendant number d(R'_i) of an auxiliary configuration R'_i that differs from R_i only in one electron coordinate. The authors claim that this reweighting yields the pure (ground-state) expectation value exactly. They benchmark AUX on a symmetric Hubbard dimer, where they report exact 1RDMs, and on the helium atom, where a single-time-step AUX result improves on forward walking but is not exact. The paper also argues that forward walking and the SUB/DIV corrections cannot remove the accumulator bias for non-local operators.","tokens_in":12482,"tokens_out":30467,"duration_ms":264794,"significance":"If Eq. (8) were correct, the paper would address a recognized limitation of DMC: obtaining unbiased, pure estimates of non-local observables without leaving the DMC framework. The problem is well motivated, and the algebraic structure of the proposed correction is appealing. The paper also ships reproducible code and a clean analytic dimer test, which are strengths. However, the central identity contains a normalization error in the Appendix A derivation, and the method as stated is not even formally exact. The additional lack of a construction for the auxiliary walker in continuous systems would already be a serious obstacle; the algebraic error makes the main claim untenable.","major_comments":[{"comment":"The step from the mixed-distribution average to the ψ_T^2 average omits the Radon-Nikodym normalization. For normalized averages, ⟨f⟩_{ψ_T ψ_0} = ⟨f(ψ_0/ψ_T)⟩_{ψ_T ψ_T} / ⟨ψ_0/ψ_T⟩_{ψ_T ψ_T}. Applying this to f = [ψ_T(R')/ψ_T(R)] δ d(R'_i), one obtains an extra denominator ⟨(1+Δ(R'))(1+Δ(R))⟩_{ψ_T^2}, whereas the exact ground-state 1RDM has denominator ⟨(1+Δ(R))^2⟩_{ψ_T^2}. These denominators are not equal in general, so Eq. (A.11) is not the exact ground-state 1RDM. A direct counterexample is N=1: R'_i is a fixed point r', d(R'_i) is constant, and Eq. (8) reduces to the mixed estimator, which is not O_0. The claimed exactness is therefore unsupported.","section":"Appendix A, Eqs. (A.10)-(A.11)"},{"comment":"Even setting the normalization issue aside, the paper does not specify how R'_i is found or created in a continuous system. In a continuum, the probability that a population walker lies exactly at R'_i is zero, and no auxiliary branching or interpolation algorithm is given. The Hubbard dimer avoids the issue because the configuration space is discrete; the helium AUX result is a single time step with no implementation details, and Appendix C documents only the dimer. Thus Eq. (8) is not a realizable method for the systems the paper targets.","section":"Sec. 2, Eq. (8); Sec. 3; Appendix C"},{"comment":"The Hubbard dimer benchmark is constructed using the exact ψ_0 to sample the mixed distribution and to compute d(R)=ψ_0(R)/ψ_T(R) analytically. This tests an algebraic identity, not a stochastic DMC procedure for estimating descendant weights. It therefore cannot validate the practical AUX algorithm, and in light of the normalization error it does not constitute a proof of exactness. The statement that 'only AUX yields the exact 1RDM for all values of correlation' requires an implementation and a derivation that are not supplied.","section":"Sec. 3, Fig. 4; Appendix C"}],"minor_comments":[{"comment":"There are several typographical errors: 'derivate' (p. 8), 'pannel' (Fig. 1 caption), 'exptrapolation' (p. 11), 'roughtly' (p. 5), 'ammount' (p. 11), 'wihich' (p. 11).","section":"Throughout"},{"comment":"The AUX result in Fig. 3 appears to be a single time step. The caption and text should state explicitly how this AUX calculation was performed for helium, including how the auxiliary walkers were obtained and how d(R'_i) was estimated.","section":"Sec. 3, Fig. 3"},{"comment":"The definition of γ_QMC is written as an unnormalized expectation; the subsequent normalization by γ(r';r') is mentioned in the text but should be stated in or near the equation for clarity.","section":"Eq. (7)"}],"recommendation":"reject","confidential_remarks":"The stress-test concern about the missing construction of R'_i is real but secondary. The deeper problem is the normalization error in Appendix A: Eq. (8) is not exact even under the ideal assumptions of the paper. This is a load-bearing algebraic error, not a presentation issue, and it cannot be fixed within the manuscript's current scope. That is why I recommend rejection rather than major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. The paper introduces Auxiliary Walking (AUX), a reweighting scheme that, if it worked, would give pure estimates of non-local operators like the 1RDM within DMC. The idea is genuinely new: instead of weighting by the descendant count of the walker itself, you weight by the descendant count of an auxiliary walker with one electron displaced. The algebra in Appendix A is clean and makes the 'accumulator bias' of Forward Walking explicit. For that alone, the paper is worth reading.\n\nThe Hubbard dimer result is exact, but it's a consistency check: they construct the mixed distribution and the descendant weights analytically from the known exact ground state. That shows Eq. (8) is algebraically correct, which is valuable, but it doesn't tell you how to compute d(R'_i) in a real simulation.\n\nThe helium calculation is the only continuous-space test, and it's a single time step. The abstract claims unbiasedness 'in the limits,' but the paper does not show the extrapolation; the AUX point in Fig. 3 is at τ=0.005 only, and it still deviates from the exact 1RDM. The authors attribute the residual error to slow convergence of the ψT cancellation, which they study in Appendix B, and they are honest that AUX is not exact there yet.\n\nThe load-bearing problem is the one the stress-test note flags: the paper never specifies how R'_i is found or created in a continuous system. The probability of another walker being exactly at R'_i is zero, and no auxiliary branching construction is given. The phrase 'could be found (or created)' is not a method. Without that, Eq. (8) cannot be evaluated for helium or any non-discrete system. This is not a minor gap; it's the central step of the algorithm.\n\nI also note the paper does not ship code or data with a commit hash; a GitHub link is given, but no version or reproducibility details. Given the high stakes of the claim, that matters.\n\nFor peer review: yes, send it out. The idea is novel and the formal derivation is worth referee attention. But a serious referee should either get a concrete algorithm for generating R'_i in continuous space, or the authors should state clearly that the method is currently limited to lattice models where configurations are discrete. As written, the title overpromises.","headline":"New idea for fixing DMC's 1RDM bias, but the key step — how to build the auxiliary walker in continuous space — is missing, so the method is not yet realizable.","tokens_in":13018,"tokens_out":2766,"would_cite":false,"duration_ms":26211,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.15.-m","02.70.Ss"],"model":"deepseek-v4-flash","headline":"Auxiliary Walking removes the trial-wavefunction bias from Diffusion Monte Carlo estimates of non-local operators.","keywords":["Diffusion Monte Carlo","non-local operators","one-body reduced density matrix","Forward Walking","Auxiliary Walking","mixed estimator bias","trial wavefunction bias","Hubbard dimer"],"falsifier":"In a system with a known exact ground state, sample walkers from ψ_Tψ_0 and construct auxiliary walkers by moving one electron to a position drawn from ψ_0(R')/ψ_T(R'), computing their descendant counts analytically. If Σ d(R'_i) O_M(R_i) does not converge to O_0 as the number of walkers grows, the identity in Eq. (8) is false; the Hubbard dimer test is a discrete-state instance of this check, and the same test in a continuous system would settle the general claim.","tokens_in":11961,"feed_emoji":"⚛️","tokens_out":1835,"duration_ms":18300,"temperature":0.7,"pith_summary":"Diffusion Monte Carlo (DMC) computes ground-state expectations from a mixed distribution, and the standard Forward Walking correction fixes the distribution but leaves a separate trial-wavefunction bias inside the estimator for non-local quantities like the one-body reduced density matrix (1RDM). The paper introduces Auxiliary Walking (AUX), which reweights each DMC walker by the descendant count of an auxiliary walker that differs only in one electron position. The paper shows that this extra weight exactly cancels the non-local part of the accumulator bias, so AUX yields a pure 1RDM in the limits of zero time step and infinite walkers. The claim is demonstrated exactly on the symmetric Hubbard dimer and is shown to outperform Forward Walking on the helium atom.","feed_headline":"New DMC reweighting removes bias from non-local operators","feed_subtitle":"Auxiliary Walking yields unbiased one-body density matrices where Forward Walking cannot, shown exactly on the Hubbard dimer.","key_machinery":"The central object is the auxiliary walker R'_i = {r', r2, ..., rN}, obtained from a walker R_i = {r1, r2, ..., rN} by changing one electron coordinate. Its asymptotic descendant count d(R'_i) is proportional to ψ0(R'_i)/ψT(R'_i), the same ratio that appears in the non-local part of the 1RDM accumulator. Reweighting each walker's mixed estimator O_M(R_i) by d(R'_i) produces the pure estimate O_0. The method is an extension of the Forward Walking principle, differing only in which walker's descendant count is used as the weight.","core_discovery":"The central claim is that the equality O_A = Σ_i d(R'_i) O_M(R_i) = O_0 holds: reweighting each DMC walker R_i by the descendant number d(R'_i) of an auxiliary walker R'_i that differs from R_i only in one electron coordinate yields an unbiased pure expectation of a non-local operator. Forward Walking uses d(R_i) to correct only the sampling distribution, but the 1RDM estimator still contains ratios of trial wavefunctions in its accumulator. The auxiliary weight d(R'_i) is proportional to ψ0(R'_i)/ψT(R'_i), which is exactly the factor needed to remove the numerator bias, while the inverse weight d(R_i) cancels the denominator bias. Thus AUX removes both the mixed-distribution bias and the ac","pith_inferences":["If the auxiliary-walker construction can be made practical in continuous space, AUX could replace Forward Walking as the standard DMC estimator for 1RDM-dependent quantities, such as natural orbitals and occupation numbers used in density-functional corrections.","The slow time-step convergence of the ψ_T cancellation observed in the helium atom suggests that the method's practical accuracy in realistic systems will hinge on reducing time-step errors or on using improved sampling schemes, not on the reweighting identity itself.","A possible testable extension is to apply AUX to systems with stronger correlation or with nodes, to see whether the fixed-node approximation interacts differently with the auxiliary-walker reweighting than it does with Forward Walking."],"forward_implications":["AUX provides, for the first time within DMC, a path to unbiased 1RDMs in the zero-time-step and infinite-walker limit, removing a bias that Forward Walking cannot address.","The same reweighting scheme extends directly to other non-local observables, such as the momentum distribution and the two-body reduced density matrix, whose estimators also contain explicit wavefunction ratios.","In the Hubbard dimer, AUX recovers the exact pure 1RDM for all correlation strengths, while Forward Walking fails on the off-diagonal element.","In the helium atom, AUX outperforms Forward Walking at the same time step, showing that the accumulator bias is the dominant error for non-local operators once the distribution is corrected."],"fun_headline_variants":["Auxiliary Walking corrects DMC bias for non-local operators","Unbiased 1RDMs from DMC via Auxiliary Walking","Beyond Forward Walking: unbiased DMC density matrices","DMC's non-local bias fixed by Auxiliary Walking"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"For every walker, an auxiliary walker differing in exactly one electron coordinate must exist among the sampled walker population so its descendant count can be measured; the paper asserts such a walker 'could be found or created' but provides no construction algorithm for continuous systems.","fun_headline_variants_meta":{"raw":{"variants":["Auxiliary Walking corrects DMC bias for non-local operators","Unbiased 1RDMs from DMC via Auxiliary Walking","Beyond Forward Walking: unbiased DMC density matrices","DMC's non-local bias fixed by Auxiliary Walking"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000382,"raw_usage":{"total_tokens":1811,"prompt_tokens":642,"completion_tokens":1169,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":386,"completion_tokens_details":{"reasoning_tokens":1100}},"tokens_in":386,"tokens_out":1169,"duration_ms":10696,"temperature":1.0,"reasoning_tokens":1100,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T05:15:52.957220+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In a system with a known exact ground state, sample walkers from ψ_Tψ_0 and construct auxiliary walkers by moving one electron to a position drawn from ψ_0(R')/ψ_T(R'), computing their descendant counts analytically. If Σ d(R'_i) O_M(R_i) does not converge to O_0 as the number of walkers grows, the identity in Eq. (8) is false; the Hubbard dimer test is a discrete-state instance of this check, and the same test in a continuous system would settle the general claim.","supporting_citations":[],"review_version":1}