{"id":"851a9b83-924d-47e7-9490-39eaf078f4b3","arxiv_id":"1909.00816","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A connected-vertex strong-coupling diagrammatic Monte Carlo treats local interactions exactly and matches NLCE results for the infinite-U Hubbard model up to hopping order nine.","lead":"This paper derives a diagrammatic Monte Carlo method for strongly correlated electrons and spins, built from connected local vertices that are summed exactly in the atomic limit. A specialist would read it because it claims to reach higher expansion orders than prior strong-coupling methods and matches independent cluster calculations for the infinite-U Hubbard model.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The fermionic sign rules in 'Diagrammatic rules' are justified only by two examples, not proven generally; any sign error would alter Monte Carlo weights, so this is the most load-bearing unverified element.","rationale":"The central claim is that the recursive vertex construction (18) and the expansion (22) in connected vertices yield an exact, sign-correct strong-coupling diagrammatic Monte Carlo. The vertices themselves are computed exactly from local expectation values, and the recursion appears algebraically sound. The weakest point in the argument is therefore the sign convention for assembling vertices into diagrams, because every Monte Carlo weight inherits these signs. The paper's derivation of the diagrammatic rules rests on two illustrative examples and does not rigorously prove that the parity of swaps/commutes reproduces Wick's theorem signs for arbitrary topologies, especially when vertices overlap in time and operators from different sites must be regrouped. The NLCE benchmark provides empirical support for low-order contributions to the density, but it is a narrow test and cannot rule out a sign error in an untested diagram class. Other concerns, such as the narrow benchmark coverage, limited vertex-size truncation checks, and lack of released code, are secondary because they affect the breadth of evidence rather than the core correctness of the construction. Thus the sign rules are the single most load-bearing concern, and the appropriate verdict remains CONDITIONAL as the reader concluded.","tokens_in":13796,"tokens_out":29119,"duration_ms":326281,"concrete_test":"Implement an automated enumeration of all connected diagrams up to N_t=4 for the two-site Hubbard model at U=infinity. For each diagram topology, compute the sign by the paper's two rules (parity of swaps/commutes from the reference contraction) and independently compute the sign from a direct Wick contraction of the original expansion (2) at fixed external time arguments. Compare the two sign assignments for every topology; any disagreement demonstrates the rules are incomplete or incorrect. Agreement for all topologies through N_t=4 would validate the rules at low order and show the concern is not fatal.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the section 'Diagrammatic rules', the paper claims that swapping the connectivity of two fermionic external lines and commuting fermionic operators within a vertex are odd operations, while operations on bosonic lines carry no sign (Figs. 1 and 2). This rule is justified only by two examples; no general proof is given that the sign of an arbitrary diagram in the expansion (22) equals the parity of such operations relative to a global reference. In particular, when vertices overlap in imaginary time, the global time-ordered product interleaves operators from different sites, and the anticommutation signs from regrouping these operators by site are not explicitly tracked. The recursion (18) computes exact vertices, so errors could only enter through the diagram sign. Since every Monte Carlo weight in the strong-coupling expansion inherits these signs, an incorrect parity for any higher-order topology would change observable values even though the vertices themselves are exact. The NLCE benchmark gives empirical support, but it tests only the carrier density at U=infinity, mu/t=2, and T/t>=1/4, which may not exercise all relevant topologies or sign-sensitive regimes.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents a strong-coupling diagrammatic Monte Carlo method for correlated fermions and frustrated spin systems. The Hamiltonian is split into a local part (chemical potential and contact interaction) and an intersite part (hopping and boson-mediated interactions). The author defines connected strong-coupling vertices by a recursive subtraction of disconnected local contractions, Eq. (18), and argues that these vertices can be computed exactly from atomic-limit expectation values, Eq. (20), including a second-fermionization construction that restores convergence in the Hubbard case. The resulting expansion is then reorganized as connected topologies of connected vertices, Eq. (22), with diagrammatic rules stated in terms of a reference contraction and two update moves (swapping external lines and commuting operators within a vertex). Monte Carlo sampling is performed with a worm protocol, and the method is benchmarked on the U=infinity Hubbard model at mu/t=2 for temperatures 1/4 <= T/t <= 1, reaching hopping order N_t=9. The carrier density agrees with SCT-DMC and with numerical linked cluster expansion (NLCE) results.","tokens_in":14049,"tokens_out":4990,"duration_ms":161101,"significance":"If the construction is correct, this is a genuinely useful and original scheme: the local vertices are non-perturbative in the interaction, contain no fitted parameters, and are computed exactly from the local Hamiltonian, while the expansion parameter is only the nonlocal part of the Hamiltonian. The recursive definition of connected vertices and the exact local summation are the main technical contributions. The independent NLCE benchmark gives credible empirical support, and the order-by-order comparison with SCT-DMC is a useful cross-check, though it is not an independent validation because both methods are bold expansions in the hopping. The main risk is the fermionic sign convention, which is justified only by two examples and is load-bearing for every Monte Carlo weight in the expansion. The paper is therefore promising but requires a more rigorous treatment of the diagrammatic sign rules and, ideally, benchmarks that exercise sign-sensitive observables.","major_comments":[{"comment":"The assignment of the sign factor eta(alpha_1...alpha_N) in Eq. (22) is never defined algorithmically. The text justifies the two basic updates -- swapping the connectivity of two fermionic external lines and commuting operators within a vertex are odd operations, while bosonic operations carry no sign -- only through the examples in Figs. 1 and 2. This is load-bearing because every Monte Carlo weight inherits these signs, and the recursion (18) computes only the vertices, not the diagram parity. In particular, for vertices that overlap in imaginary time, the global T_tau product interleaves operators from different sites, and the anticommutation signs from regrouping them by site are not tracked. I ask for a general combinatorial proof, or an explicit parity function relative to a fixed reference contraction, and a validation on a topology with at least two interacting vertices at order N_t >= 4.","section":"Diagrammatic rules, Eq. (22)"},{"comment":"The recursion (18) is coherent only if the sign xi_{bar O,A} from Eqs. (15)-(16) is well-defined for arbitrary subsets A. However, the reordering T_tau bar O -> T_tau A x T_tau(bar O \\ A) presupposes that A can be made contiguous in the time-ordered product; when the time arguments of the two groups are interleaved, the number of fermionic commutations is not specified. The same gap affects Eq. (21), where the decomposition into products of connected vertices is asserted without a proof of the ordering convention. Please state the convention for handling interleaved time arguments and prove that the recursion yields the connected vertex for all N.","section":"Strong-coupling vertices, Eq. (18)"},{"comment":"The numerical evidence is limited to the carrier density at U=infinity, mu/t=2, and T/t >= 1/4. The expansion is claimed to be applicable to finite U, frustrated spins, and observables other than the density, but no benchmark exercises spin correlations, off-site propagators, or a regime where the diagram sign is decisive. Since the sign-rule gap in the diagrammatic rules is not covered by this benchmark, an additional comparison -- for example a spin-spin correlation function in the Heisenberg limit or a finite-U density benchmark -- would materially support the central claim.","section":"Benchmarks for the Hubbard model, Figs. 5-6"}],"minor_comments":[{"comment":"The caption of Fig. 5 is extremely dense, and the panel labels (a)-(f) are not all referenced in the main text; please restructure the caption and refer to each panel explicitly.","section":"Fig. 5 caption"},{"comment":"The statement that the number of vertices scales as 8^N and that N=10 gives about 10^9 vertices should clarify whether spin and particle-conservation symmetries have already been applied; this affects the practical storage estimate.","section":"Analytic structure of the connected vertices"},{"comment":"The relation G = 1/(Pi^{-1} - t(k)) is central to how observables are extracted, but it is presented without derivation; a brief derivation or a reference would help the reader verify the sign and normalization conventions.","section":"Eq. (29)"},{"comment":"The text should state more explicitly that the order-by-order comparison with SCT-DMC is expected to agree by construction because both methods are bold expansions in the hopping, and that the independent benchmark is the NLCE comparison.","section":"Order-by-order comparison with SCT-DMC"},{"comment":"There is a duplicated article in 'the the Swedish Research Council'; in addition, the reference list contains a typographical artifact in Ref. [21] ('Springer-V erlag').","section":"Acknowledgments"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the central construction is original and appears sound in its local-exactness part, and the NLCE benchmark provides a genuine external check. The main technical risk is the unproven fermionic sign convention, which affects every Monte Carlo weight; the SCT-DMC comparison should not be presented as independent validation because it is a cross-check with the same author's earlier method. The benchmark set is narrow, so I would ask for either a general sign-rule proof or additional sign-sensitive benchmarks before accepting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThe thing to know is that this paper offers a genuinely different construction for strong-coupling diagrammatic Monte Carlo. Instead of the usual Grassmannian Hubbard-Stratonovich transformation, Carlström builds connected vertices from a recursion over Wick contractions in the atomic limit. That is a real departure, and it buys something concrete: the t-expansion runs to order Nt=9, where his own SCT-DMC stopped at Nt=4. The benchmark against NLCE for the U=∞ Hubbard model at μ/t=2 is independent, and it holds within error bars.\n\nWhat the paper does well: the recursion in Eq. (18) is coherent, and the exact summation of local physics in the atomic limit is plausible. Appendix A gives a genuine convergence argument for the contact-interaction sum via second fermionization. The diagrammatic rules are clearly stated, and the order-by-order comparison with SCT-DMC is exactly the right check, since both expansions share the same skeleton. The discussion of vertex-size truncation is honest—truncation error is shown to be small relative to statistical noise.\n\nThe soft spot is the fermionic sign rule. The paper justifies the odd parity of swapping fermionic lines and commuting fermionic operators with two examples (Figs. 1 and 2), not a general proof. Every Monte Carlo weight inherits these signs, so an error in a higher topology would change observables even though the vertices themselves are exact. The NLCE benchmark is useful but narrow—it tests carrier density at a single chemical potential and T/t≥1/4, which may not exercise all relevant topologies. No code or data are released, so the claim of storing vertices up to 16 legs—despite the naive 8^N scaling—is hard to check. These are fixable: a general sign argument and a second benchmark at a different μ would address the main worry.\n\nNone of this undermines the core construction. The recursion is a real step forward for the field. I would send this to a serious referee; the sign-rule gap is exactly what a referee should push on. If you work on diagrammatic Monte Carlo for correlated systems, this is worth a reading-group slot.","headline":"A genuinely different strong-coupling diagrammatic Monte Carlo construction with independent NLCE benchmarks; the main open question is the general fermionic sign rule, but the paper deserves a serious referee.","tokens_in":14506,"tokens_out":3567,"would_cite":true,"duration_ms":30211,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.10.Fd","02.70.Ss","71.10.-w"],"model":"deepseek-v4-flash","headline":"This paper constructs a diagrammatic Monte Carlo method for correlated fermions and frustrated spins in which all local interactions are summed exactly into connected vertices and only nonlocal processes are expanded.","keywords":["strong-coupling expansion","diagrammatic Monte Carlo","Hubbard model","frustrated spin models","connected vertices","Wick's theorem","fermionic sign problem","infinite U"],"falsifier":"A decisive check is to extend the series for the infinite-$U$ Hubbard model at $\\mu/t=2$ and $T/t=1/2$ to order $N_t=10$ or $11$ with vertices up to 22 legs; if the parity rule fails for any newly appearing topology, the extracted density will depart from the converged cluster-expansion value by more than the statistical error, while the rule predicts continued agreement.","tokens_in":13598,"feed_emoji":"🎲","tokens_out":12925,"duration_ms":109201,"temperature":0.7,"pith_summary":"The paper aims to make diagrammatic Monte Carlo work in the strongly correlated regime by changing what is expanded. It splits the Hamiltonian into a local part (chemical potential, contact repulsion) and a nonlocal part (hopping, exchange), then sums all local contractions exactly into connected vertices via a recursion; only the nonlocal part is treated as a series. This removes the large expansion parameter that blocks weak-coupling methods at strong repulsion, and the vertices are exact, not approximate. For the infinite-$U$ Hubbard model, the resulting series reaches ninth order in the hopping and reproduces the density obtained from independent cluster expansions, opening a practical route into the doped-Mott regime where other unbiased techniques struggle.","feed_headline":"Exact local vertices take Hubbard series to ninth order","feed_subtitle":"A new diagrammatic Monte Carlo expands only nonlocal hops; density at infinite U matches cluster expansion.","key_machinery":"The central object is the connected strong-coupling vertex $V[\\bar O]$, a local sum over all contractions on one lattice site that connect a prescribed set $\\bar O$ of external operators coming from nonlocal processes. The recursion $V[\\bar O]=\\sum_n\\langle \\Gamma^n U_i^n \\bar O\\rangle_{\\hat\\mu,e}-\\sum_{A\\subsetneq\\bar O,\\ \\hat O_1\\in A}\\xi_{\\bar O,A}\\sum_{n,m}\\langle \\Gamma^n U^n A\\rangle_{\\hat\\mu,c}\\langle \\Gamma^m U^m(\\bar O\\setminus A)\\rangle_{\\hat\\mu,e}$ subtracts every disconnected grouping from the full local expectation value, so the vertex is nonperturbative in the contact interaction $U$ and, because the local trace factorizes, exactly computable from the atomic limit. The argument then expands directly in connected topologies of these vertices joined by the nonlocal $t$ or $J$ lines; diagrammatic rules assign the fermionic sign from a reference contraction, with swapping or commuting fermionic lines counting as odd and bosonic operations carrying no sign. This construction is what removes any large expansion parameter and makes the series accessible to stochastic sampling.","core_discovery":"The paper's central claim is that a strong-coupling expansion can be reorganized so that every local interaction is treated exactly: the recursive relation groups all contractions on a lattice site that connect external operators into a connected vertex, and these vertices contain all nonperturbative local physics. Because the unperturbed theory is local, each vertex is an expectation value in the atomic limit and can be computed to machine precision; the original series then becomes a sum over connected topologies of connected vertices, expanded only in the nonlocal hopping and exchange terms. The paper demonstrates the construction on the infinite-$U$ Hubbard model at $\\mu/t=2$, where the series for the carrier density reaches ninth order in the hopping, the truncation error from the vertex size is below statistical noise, and the results agree within error bars with independent cluster-expansion and spin-charge-transformed diagrammatic Monte Carlo data.","pith_inferences":["If the parity rule is correct, the sign problem should be tied to nonlocal topology rather than to the strength of the local repulsion; a direct measurement of the average sign versus $U$ and $N_t$ would confirm this and is not reported.","The same recursion should apply at finite $U$ through the second-fermionized local Hamiltonian; benchmarking at $U/t\\approx 12$, the regime where other methods struggle most, is a natural next test.","For frustrated spin models the vertex space is only four operators, so the method could deliver low-temperature unbiased correlations where conventional quantum Monte Carlo has a sign problem; the paper does not yet provide a spin benchmark.","The recursive partition of local contractions could also serve as a pre-computation step for other diagrammatic schemes, supplying exact local building blocks for determinant or bold diagrammatic methods."],"forward_implications":["For any Hamiltonian of the form (1), the local interaction can be taken to infinity and the expansion remains a finite-order series in the nonlocal terms, with vertices computed exactly to machine precision.","At the same expansion order, the method gives results identical to other bold $t$-expansion schemes, but reaches order $N_t=9$ where a comparable method reaches only $N_t=4$, so higher accuracy is available despite factorial computational scaling.","Observables are extracted as polarizations of the $t$- or $J$-lines, so two-point Green's functions and spin correlations come directly from the sampled diagrams, and many-point correlators follow from multiple measuring lines.","Truncating the vertex size at 16 legs produces errors below statistical noise at the orders reached, so the practical bottleneck is expansion order rather than vertex storage.","The method extends to spin models with fewer operators (4 versus 8 for the Hubbard model), allowing vertices up to 15 legs within the same memory budget, which is promising for frustrated magnets."],"supporting_citations":[{"why":"Supplies Wick's theorem and the Feynman-diagram sign conventions from which the strong-coupling expansion and its reference-contraction rules are derived.","marker":"[40]"},{"why":"Introduces second fermionization, which removes the divergence in the local interaction sum and makes the connected vertices exactly computable.","marker":"[32]"},{"why":"Provides the d/h operator basis whose trivial time evolution under the local Hamiltonian lets each vertex be stored as an operator-order constant.","marker":"[43]"},{"why":"Gives the numerical linked-cluster equation-of-state data for the infinite-U Hubbard model used as the benchmark.","marker":"[26]"},{"why":"Gives the order-by-order spin-charge-transformed diagrammatic Monte Carlo series that the new expansion is compared with.","marker":"[35]"},{"why":"Supplies the worm-sampling protocol used to evaluate the connected-vertex expansion stochastically.","marker":"[31]"}],"fun_headline_variants":["Exact local vertices push Hubbard expansion to ninth order","Strong-coupling expansion: local physics exact, only hops expanded","Infinite-U Hubbard: atomic vertices, ninth-order hopping series","New diagrammatic MC: nonperturbative locals, perturbative hops","Local vertices made exact: Hubbard series to ninth order"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the parity rule for the fermionic signs: the paper justifies it by only two examples, and if any higher-order topology requires a different sign, every Monte Carlo weight inherits that mistake even though the connected vertices themselves are exact.","fun_headline_variants_meta":{"raw":{"variants":["Exact local vertices push Hubbard expansion to ninth order","Strong-coupling expansion: local physics exact, only hops expanded","Infinite-U Hubbard: atomic vertices, ninth-order hopping series","New diagrammatic MC: nonperturbative locals, perturbative hops","Local vertices made exact: Hubbard series to ninth order"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1266,"prompt_tokens":806,"completion_tokens":460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":422,"completion_tokens_details":{"reasoning_tokens":376}},"tokens_in":422,"tokens_out":460,"duration_ms":201954,"temperature":1.0,"reasoning_tokens":376,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:35:28.368843+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive check is to extend the series for the infinite-$U$ Hubbard model at $\\mu/t=2$ and $T/t=1/2$ to order $N_t=10$ or $11$ with vertices up to 22 legs; if the parity rule fails for any newly appearing topology, the extracted density will depart from the converged cluster-expansion value by more than the statistical error, while the rule predicts continued agreement.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Wick's theorem and the Feynman-diagram sign conventions from which the strong-coupling expansion and its reference-contraction rules are derived."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces second fermionization, which removes the divergence in the local interaction sum and makes the connected vertices exactly computable."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the d/h operator basis whose trivial time evolution under the local Hamiltonian lets each vertex be stored as an operator-order constant."},{"cited_title":"Khatami, E","cited_arxiv_id":null,"evidence_quote":"Gives the numerical linked-cluster equation-of-state data for the infinite-U Hubbard model used as the benchmark."},{"cited_title":"Carlström, Phys","cited_arxiv_id":null,"evidence_quote":"Gives the order-by-order spin-charge-transformed diagrammatic Monte Carlo series that the new expansion is compared with."},{"cited_title":"Van Houcke, E","cited_arxiv_id":null,"evidence_quote":"Supplies the worm-sampling protocol used to evaluate the connected-vertex expansion stochastically."}],"review_version":1}