{"id":"d4ebdc2d-64de-4183-9c78-51a3b8454ecc","arxiv_id":"2606.02549","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Diagrammatic Monte Carlo stochastically sums the divergent virtual-positronium ladder series in positron-molecule self-energies, reproducing exact-diagonalisation binding energies for LiH.","lead":"This paper applies diagrammatic Monte Carlo to sum the infinite electron-positron ladder diagrams that describe virtual positronium formation in positron-molecule scattering. It reproduces exact binding energies for LiH while cutting memory scaling by roughly 100-1000x, potentially enabling calculations for larger molecules.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Cesàro–Riesz + exponential 1/N extrapolation for the divergent Γ series is the decisive unvalidated step; a deterministic partial-sum cross-check would settle it.","rationale":"The reader's weakest_assumption is the same one I would flag. The paper's central claim is not that low-order diagrams are sampled correctly—that is expected—but that an infinite, divergent series can be recovered by Cesàro–Riesz damping plus an ad hoc exponential fit. That is precisely the step with no independent support. The deterministic cross-check I propose is feasible because the authors already have the EXCITON+ code and the same basis; it would separate extrapolation bias from MC noise. I do not see a more load-bearing flaw: the MC update scheme, normalization via type-0 sector, and memory reduction from N^4 to N^3 are plausible and internally consistent. The single-molecule benchmark against the same group's code and the absence of released code/data are real limitations but secondary; they do not by themselves invalidate the claim. Hence the reader's CONDITIONAL verdict remains appropriate; no verdict change.","tokens_in":9079,"tokens_out":5365,"duration_ms":58601,"concrete_test":"Compute the exact order-by-order Γ-series partial sums Σ^(n) for the same LiH basis using the deterministic EXCITON+ code (or a small exactly diagonalisable model with identical integral data), then apply the paper's full extrapolation protocol—Eq. (9) with δ∈[1,3], N≥5, and the exponential fit—to these noiseless sums. If the resulting C does not reproduce the exact BSE binding energy (1197 meV at the 2+Γ level) within ~30 meV, the extrapolation is biased and the diagMC agreement is not evidence for the method. If it does reproduce it, the remaining 26 meV spread is MC noise and the central claim survives.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central new claim—that diagMC stochastically sums the divergent virtual-positronium (Γ) ladder and recovers 1207±26 meV vs 1197 meV (Table I)—depends entirely on the infinite-order extrapolation, not on the Monte Carlo sampling. Eq. (9) applies Cesàro–Riesz damping to a series whose bare partial sums diverge; for such a series the statement that 'in the limit N→∞ the resummation reproduces the original series' has no well-defined content, since the original series has no sum. The practical protocol then fits ε_b(1/N)=A(e^{B/N}-1)+C to N≥5 over δ∈[1,3]. The reported error (26 meV) is only the standard deviation of C across δ; it does not include (i) the choice of the exponential model, (ii) the choice N_min=5, (iii) the MC statistical error of each Σ^(n), or (iv) the possibility that the regularized finite-N sequence has a limit that depends on δ or does not converge to the exact BSE value. Because the Γ contribution dominates the binding energy (1207 vs 636 meV at TDHF@TDA), any bias in this extrapolation directly shifts the headline numbers, even if the low-order sampling is unbiased.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a diagrammatic Monte Carlo (diagMC) method for evaluating the positron-molecule self-energy in a Gaussian-orbital basis, including the GW@TDA, virtual-positronium (Γ) ladder, and positron-hole (Λ) ladder diagram classes. A Markov-chain Monte Carlo algorithm samples diagram orders and internal indices, with density fitting to store only three-centre integrals. The infinite-order limit of the Γ series, which grows without bound order-by-order, is handled by Cesàro–Riesz damping (Eq. 9) followed by an exponential fit in 1/N. The method is benchmarked on LiH against the authors' deterministic EXCITON+ Bethe–Salpeter exact-diagonalisation code. The reported binding energies agree with the benchmarks to within 5–30 meV at five levels of theory, and the largest memory arrays scale as O(N^3) instead of O(N^4).","tokens_in":9433,"tokens_out":6914,"duration_ms":72582,"significance":"If the extrapolation protocol is sound, this is a potentially important advance: it would be the first demonstration of stochastic, all-order summation of the virtual-positronium ladder in positron-molecule calculations, removing a major computational bottleneck and extending the reach of many-body positron theory to larger molecules. The memory reduction from N^4 to N^3 is concrete and practically valuable. The paper is also explicit about its proof-of-principle scope, benchmarks against a deterministic code, and uses physical diagram classes. The central risk is that the infinite-order Γ result depends on an unvalidated resummation whose uncertainty is understated; the reported agreement for LiH is encouraging but not conclusive. Overall, the significance is conditional on a validation of the resummation and on a more complete uncertainty propagation.","major_comments":[{"comment":"The infinite-order extrapolation of the divergent Γ series is the load-bearing step. The reported 1207±26 meV at the 2+Γ level is only the standard deviation of the fitted C across δ∈[1,3]; it omits the MC statistical error of each Σ^(n), the uncertainty in the exponential model, the choice of the N≥5 fit window, and the possibility that the Cesàro–Riesz limit depends on δ. Since the bare series diverges, the statement after Eq. (9) that 'in the limit N→∞ the resummation reproduces the original series' is not meaningful; resummation defines a generalized sum that may differ from the physical BSE solution. The authors should validate the protocol on a deterministic order-by-order series, e.g., by computing finite-order Γ contributions with EXCITON+ in a small basis and applying the same resummation to those partial sums to see if they recover the BSE eigenvalue. They should also propagate","section":"Eq. (9) and Results, Table I"},{"comment":"Per-order Monte Carlo statistical uncertainties are not reported. The text states that 10^7–10^8 steps per element are used and that the Γ channel has larger stochastic uncertainty, but no error bars are assigned to Σ^(n) or to the resummed partial sums. Without these, the reader cannot determine whether the 10–30 meV differences between diagMC and EXCITON+ are statistically significant or simply reflect undiagnosed bias in the extrapolation. The reported 26 meV error bar for 2+Γ appears to be a purely systematic spread over δ, not a Monte Carlo error. This is a key omission for a stochastic method.","section":"Results, Eq. (8) and Table I"},{"comment":"The exponential model ε_b(1/N)=A(e^{B/N}-1)+C fitted to N≥5 for δ∈[1,3] is not fully specified. The maximum order N_max is not stated, and no sensitivity analysis (e.g., varying N_min, using alternative fit forms, or showing the fitted curves for each δ) is provided. The argument that the leading correction is linear in 1/N applies to the resummed self-energy matrix elements, not directly to the binding energy after solving the Dyson equation, which is a nonlinear functional of Σ. The paper should state N_max, show the N-range and fitted curves, and discuss the stability of C with respect to the fitting window and model choice.","section":"Results, extrapolation model"}],"minor_comments":[{"comment":"The citation placeholder '[23? ?]' appears in the sentence about annihilation-rate corrections; this must be fixed.","section":"Footnote [27]"},{"comment":"Reference [43] lists two arXiv identifiers in a confusing format; check for a typo. Reference [31] also has an arXiv number that appears similar to another entry.","section":"References [43] and [31]"},{"comment":"The choice D0=Σ^(2)_if is problematic if the second-order self-energy vanishes for some (i,f) pair. The authors should describe the fallback procedure or justify that this does not occur in the systems considered.","section":"Eq. (1), type-0 sector"},{"comment":"The update set is described qualitatively. The proposal probabilities in Eq. (5)–(7) need to be stated explicitly to ensure reproducibility, especially for the 'add interaction' update where the ratio of proposal probabilities must cancel the combinatorial factors.","section":"Algorithm details"},{"comment":"The figure captions and the text do not specify the number of MC steps per element per order, the total sampled orders, or whether the plotted points include any error bars. This information is essential for evaluating the reliability of the extrapolations.","section":"Figures 2 and 3"}],"recommendation":"major_revision","confidential_remarks":"The core idea is interesting and the benchmark is favourable, but the decisive step—the resummation of a divergent series—is not yet established. The authors should be asked to provide a deterministic cross-check of the Cesàro–Riesz + exponential protocol and to report per-order MC uncertainties. If they can do that, the paper would be a strong contribution. I would also note that the benchmark code is from the same group, which is acceptable but should be complemented by an independent or self-contained validation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the first sensible application of diagrammatic Monte Carlo to positron-molecule self-energies, and the memory reduction from ~N^4 to ~N^3 is real. The paper deserves a serious referee. The central new ingredient is the stochastic summation of the divergent virtual-positronium ladder series with Cesaro-Riesz damping. That is the part I worried about, and the stress-test note is right that the extrapolation is heuristic. But the paper does something the stress-test undersells: it validates that extrapolation against the exact diagonalisation value for LiH. 1207 +/- 26 vs 1197 meV, and 1271 +/- 18 vs 1276 for the combined level. That is not an independent check - it is the same group's EXCITON+ code - but it is a deterministic, parameter-free reference, and the agreement across five levels of theory is genuine evidence.\n\nWhat is actually new: the update scheme, the type-0 normalisation trick, and the demonstration that the divergent Gamma series can be tamed. The density-fitted three-centre integrals are the largest stored objects, so the memory claim is plausible. I also like that they sample order-by-order and reconstruct Sigma^(n) separately; that makes the Monte Carlo part auditable.\n\nSoft spots, in proportion:\n1. The Cesaro-Riesz factor in Eq. (9) is not a standard resummation of a divergent series; the statement that it reproduces the original series in the N->infinity limit has no clear meaning when the original partial sums diverge. The exponential fit in 1/N is ad hoc. The reported error is only the spread in C across delta in [1,3], not the model uncertainty, the MC noise per order, or the choice N_min=5. That is a real weakness.\n2. The benchmark is a single molecule, LiH, and the reference comes from the same authors' code. One independent calculation on a second molecule would substantially raise confidence.\n3. No code, data, or per-order MC error estimates are shipped. For a method paper, that is a fixable deficiency.\n\nThe extrapolation concern is the load-bearing one, but it is not a demonstrated error. The central argument holds for LiH. If a referee pushes for a second benchmark and a more honest treatment of the extrapolation uncertainty, the paper will be much stronger.\n\nWho should read it: anyone working on positron binding, annihilation, or scalable many-body methods for molecules. I would take it to a reading group. I would not cite it yet without a second system, but I would keep it in view.\n\nRecommendation: send to peer review, conditional on substantial revision focused on the resummation and a second benchmark.","headline":"Useful proof-of-principle for diagMC in positron-molecule theory; the infinite-order resummation is the soft spot, but the LiH benchmark carries more weight than the stress-test allows.","tokens_in":9872,"tokens_out":2819,"would_cite":false,"duration_ms":26782,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Diagrammatic Monte Carlo reproduces exact positron binding energies in LiH by stochastically summing the infinite electron-positron ladder series.","keywords":["positron-molecule interactions","diagrammatic Monte Carlo","virtual positronium ladder","Bethe-Salpeter equations","Tamm-Dancoff approximation","Cesaro-Riesz resummation","positron binding energy","ladder series"],"falsifier":"Perform the same stochastic resummation on a small molecule or model where the exact diagonalisation can be extended to orders N large enough that the extrapolation becomes essentially direct, or where the true infinite-order sum is known independently, and check whether the delta-averaged extrapolated value converges to that known number as N grows; a mismatch beyond the quoted uncertainty would indicate a bias in the resummation model.","tokens_in":8974,"feed_emoji":"🎲","tokens_out":5882,"duration_ms":53211,"temperature":0.7,"pith_summary":"The paper claims that diagrammatic Monte Carlo can stochastically sum the entire infinite ladder series describing virtual positronium formation in a positron bound to a molecule, without ever diagonalizing the huge two-particle Hamiltonian that deterministic methods require. For LiH, the stochastic binding energies match exact-diagonalisation benchmarks: 1207 ± 26 meV versus 1197 meV at the second-order-plus-ladder level, and 1271 ± 18 meV versus 1276 meV when the electron-positron and positron-hole ladders are combined with the time-dependent Hartree-Fock series. The largest memory footprint drops from roughly N^4 to N^3 scaling, where N is the number of molecular orbitals. This is achieved by Cesaro-Riesz damping of a series whose terms grow with order, followed by an exponential extrapolation in 1/N.","feed_headline":"Monte Carlo sums divergent positron ladder, matches exact binding","feed_subtitle":"Diagrammatic Monte Carlo reproduces LiH binding energies to within a few tens of meV, with memory down from N^4 to N^3.","key_machinery":"The central object is the irreducible self energy of the positron, expanded in Goldstone-type diagrams in powers of the Coulomb interaction; the ladder diagrams (electron-positron and positron-hole) are sampled order by order. The mechanism that makes it work is a Markov-chain update set that adds and removes interaction vertices and modifies internal lines, together with a normalising 'type-0' sector; the infinite-order limit is then extracted by Cesaro-Riesz weighted partial sums, where the nth-order contribution is multiplied by ((N-n+1)/N)^delta, and a fit of the binding energy versus 1/N to an exponential model to take N to infinity.","core_discovery":"The central claim is that the divergent infinite-order electron-positron ladder (virtual positronium) series in the positron-molecule self energy can be summed stochastically rather than by exact diagonalisation. Each diagram order is sampled via Markov-chain Monte Carlo, with sign-accumulated estimates combined through Cesaro-Riesz weighted partial sums and an exponential 1/N extrapolation to reach the infinite-order limit. With this machinery, the method reproduces the exact diagonalisation binding energies for LiH at all tested levels, including the strongly divergent Gamma series, while storing only three-centre density-fitted integrals rather than the full two-particle Bethe-Salpeter ma","pith_inferences":["A testable consequence left implicit: the delta-scan procedure (delta from 1 to 3) plus exponential fit could be validated on a model where the exact infinite-order sum is known analytically, isolating any resummation bias from Monte Carlo sampling error.","The method's success with bare Coulomb interactions suggests the next step is to combine it with self-consistent screening (beyond the Tamm-Dancoff approximation), which may extend quantitative accuracy to a wider class of molecules.","The memory reduction opens the door to embarrassingly parallel implementations, but the scaling of the number of Monte Carlo steps required for the divergent Gamma series is not yet analysed; this will matter for larger basis sets.","If the extrapolation model proves transferable, a practical protocol for other molecules would be to compute binding energies at several delta values and report the mean and spread, as done here."],"forward_implications":["Positron binding energies can be computed in larger molecules than deterministic Bethe-Salpeter diagonalisation allows, because the dominant memory cost is the three-centre integrals (~N^3) rather than two-particle matrices (~N^4).","The virtual-positronium ladder, the main non-perturbative correlation channel, is shown to be summable stochastically despite its order-by-order divergence.","The same resummation pipeline applies to the RPA and TDHF GW series, which converge smoothly and extrapolate stably across the resummation parameter range.","Combining the electron-positron and positron-hole ladders with the TDHF series reproduces the exact diagonalisation binding energy for LiH to within the stochastic uncertainty.","The method removes the terabyte-scale memory bottleneck of deterministic two-particle diagonalisation, making positron-molecule studies feasible on modern distributed architectures."],"fun_headline_variants":["Stochastic summation tames divergent positron ladder, nails LiH binding","Diagrammatic Monte Carlo sums virtual positronium series to exact LiH binding","Positron ladder series summed stochastically, matches exact diagonalisation","Monte Carlo overcomes divergent positron ladder, reproducing LiH binding","Memory reduced as Monte Carlo sums divergent positron-molecule ladder"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The recovered infinite-order binding energy rests on the assumption that Cesaro-Riesz damping with delta between 1 and 3, together with the exponential 1/N fit, recovers the true sum of the divergent Gamma series; if that extrapolation is biased, the reported energies could be systematically wrong even though each sampled order is unbiased.","fun_headline_variants_meta":{"raw":{"variants":["Stochastic summation tames divergent positron ladder, nails LiH binding","Diagrammatic Monte Carlo sums virtual positronium series to exact LiH binding","Positron ladder series summed stochastically, matches exact diagonalisation","Monte Carlo overcomes divergent positron ladder, reproducing LiH binding","Memory reduced as Monte Carlo sums divergent positron-molecule ladder"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000679,"raw_usage":{"total_tokens":2933,"prompt_tokens":768,"completion_tokens":2165,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":512,"completion_tokens_details":{"reasoning_tokens":2070}},"tokens_in":512,"tokens_out":2165,"duration_ms":15132,"temperature":1.0,"reasoning_tokens":2070,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T12:30:38.022507+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Perform the same stochastic resummation on a small molecule or model where the exact diagonalisation can be extended to orders N large enough that the extrapolation becomes essentially direct, or where the true infinite-order sum is known independently, and check whether the delta-averaged extrapolated value converges to that known number as N grows; a mismatch beyond the quoted uncertainty would indicate a bias in the resummation model.","supporting_citations":[],"review_version":2}