{"id":"7393eac6-69fe-4f22-93fb-aecbf3d3475d","arxiv_id":"1908.10361","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Two-particle Cluster Perturbation Theory, built from a frequency-independent vertex approximation, reproduces the spin susceptibility of the 1D Hubbard model qualitatively at half filling, with quantitative deviations away from half filling.","lead":"This paper extends Cluster Perturbation Theory, a cheap computational method for single-particle properties, to compute the transverse spin susceptibility of Hubbard models by approximately solving the Bethe-Salpeter equation. The authors benchmark the method on the one-dimensional Hubbard model and find reasonable agreement with established results, suggesting it could be a low-cost tool for neutron scattering and cold-atom experiments.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The uncontrolled vertex approximation in Eq. (8) is the load-bearing concern: it assumes frequency independence and cluster-to-lattice transfer of the four-point vertex, while the singular χ_c makes the extraction of Γ_c non-unique, yet validation is only qualitative.","rationale":"The reader's weakest assumption identifies Eq. (8) and the pseudoinverse, which matches the central concern here. The vertex approximation is the philosophical core of the method: if the four-point vertex has significant internal frequency dependence, Eq. (9) does not follow from the exact Bethe-Salpeter equation, and all benchmarks become empirical curve-matching rather than a test of the derived formula. The paper's own Fig. C9 shows that the effective two-point vertex Γ(ω) deviates from RPA at U=4, hinting that the full vertex may carry more structure, but that figure is derived from the approximate χ_CPT and thus cannot validate Eq. (8). The pseudoinverse adds a second, independent ambiguity: because the uniform mode is a null vector of χ_c for all ω, Eq. (D1) selects the minimum-norm Γ_c, setting that component to zero. The paper does not check whether this choice affects physical results, particularly the uniform susceptibility which one might expect to be sensitive to that component. The proposed test, comparing the full cluster four-point vertex to the frequency-independent two-point vertex, is a direct and feasible check of the core assumption. If the vertex is indeed flat over the relevant frequency range, the method is on firmer ground; if not, the conditional acceptance is the right call, requiring either a justification of the frequency range or a reformulation. The verdict is therefore unchanged: CONDITIONAL. The paper's benchmarks and honest discussion of numerical artifacts justify publication with conditions, but not full acceptance as a production tool until the vertex approximation is directly tested and the role of the pseudoinverse is quantified.","tokens_in":102824,"tokens_out":12983,"duration_ms":131138,"concrete_test":"Compute the full four-point cluster vertex Γ_c(iν,iν',iω) on a small periodic cluster (e.g., L=4) by exact diagonalization, and compare with the frequency-independent two-point vertex Γ_c(iω) extracted from Eq. (D1). Quantify Δ = max_{iν,iν' in the range where the bubble χ0,c has spectral weight} |Γ_c(iν,iν',iω) − Γ_c(iω)| / |Γ_c(iω)|. If Δ exceeds roughly 20%, the frequency-independence assumption (Eq. 6) is invalid, directly undermining the derivation of Eq. (9).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eq. (8) is the central premise: the four-point vertex is taken to be independent of the internal fermionic frequencies (Eq. 6) and identical to the cluster vertex. This is stronger than the analogous one-particle CPT self-energy approximation because the magnetic particle-hole vertex is known to have significant frequency and momentum structure at intermediate and strong coupling, e.g., the spin-exchange vertex is non-local. The benchmarks are visual and qualitative, and the paper concedes discrepancies at U=2–8 and away from half filling. Moreover, χ_c is singular for all ω due to SU(2) symmetry (the uniform vector is a null vector), so Eq. (D1) defines Γ_c through a pseudoinverse that sets the null-space component to zero. This choice is not validated and can directly corrupt the q=0 uniform susceptibility, for which no comparison to exact results is shown. Thus the central claim rests on an ansatz that is neither derived nor quantitatively justified, and its numerical implementation contains an arbitrary component.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends cluster perturbation theory (CPT) to two-particle correlation functions, focusing on the transverse spin susceptibility of the Hubbard model. The central result is Eq. (9), χ_CPT^{-1}(q,iω) = χ_{0,CPT}^{-1}(q,iω) + χ_c^{-1}(iω) − χ_{0,c}^{-1}(iω), derived under the assumption Eq. (8) that the four-point vertex is frequency-independent and equal to the cluster vertex, Γ(iν,iν',iω) ≈ Γ(iω) ≈ Γ_c(iω). The method is benchmarked against RPA-CPT for weak coupling, DMRG for intermediate coupling, and the Müller ansatz for strong coupling in the 1D Hubbard model at half filling, and against DMRG away from half filling. The appendices provide a Q-matrix representation for cluster correlation functions, a Lehmann representation for the CPT Green's function, the evaluation of the bubble diagram, the pseudoinverse construction of Γ_c in Eq. (D1), and an admission that the method violates causality for small broadening parameters.","tokens_in":103063,"tokens_out":3737,"duration_ms":44863,"significance":"If the central approximation is reliable, the paper offers an economically attractive route to momentum-resolved two-particle correlation functions, complementing single-particle CPT and enabling direct contact with neutron scattering and cold-atom experiments. The derivation of Eq. (9) from the Bethe-Salpeter equation is clean and the method is exact in the U=0 and t=0 limits and systematically improvable with cluster size. The manuscript is commendably transparent: it benchmarks against independent RPA, DMRG, and Müller results, discusses finite-size scaling, and explicitly states the causality and pole-mismatch limitations in Appendix E. The main weaknesses are that the vertex approximation is asserted rather than validated, the pseudoinverse choice for the singular cluster vertex is not tested against q=0 uniform susceptibility, and the benchmark comparisons are largely qualitative, so the abstract's claim of 'excellent approximations' is not quantitatively established.","major_comments":[{"comment":"The four-point vertex approximation Γ(iν,iν',iω) ≈ Γ(iω) ≈ Γ_c(iω) is the load-bearing assumption of the entire method, but it is asserted rather than derived or quantitatively validated. The reference to Ref. 12 concerns a VCA calculation in a different context, and the paper does not provide any direct numerical evidence that the internal fermionic-frequency dependence of the magnetic particle-hole vertex is negligible for the 1D Hubbard model at the parameters tested. Since Eq. (9) follows exactly from this ansatz, the central claim rests on an uncontrolled approximation. I ask the authors to provide a concrete test, such as comparing the full frequency-dependent BSE result on a small cluster with the approximate vertex result, or at least plotting the four-point vertex's dependence on iν and iν' in the relevant regime.","section":"Section III, Eq. (8)"},{"comment":"Because χ_c(iω) is singular for all ω due to SU(2) symmetry, Eq. (D1) defines Γ_c through pseudoinverses, which set the null-space component of χ_c to zero. The paper does not justify that this null-space component is irrelevant for the lattice susceptibility; in fact it can directly affect the q=0 uniform susceptibility, for which no benchmark is shown anywhere in the manuscript. The figures only compare regions near q=π and q=π/2. I request a comparison of χ(q=0,ω) against DMRG or another accurate method, or an explicit argument why the null-space choice cannot bias the q=0 result.","section":"Appendix D, Eq. (D1)"},{"comment":"The manuscript admits that Eq. (9) violates causality for small broadening and that the numerical artifacts are alleviated by choosing a sufficiently large η (0.2 or 0.5 in the main figures). Since η is not a physical parameter, the apparent agreement with benchmarks is partly controlled by the choice of broadening. This is a load-bearing issue for the claim that the method provides quantitative spin susceptibilities. Please quantify the sensitivity of the benchmark comparisons to η, and either implement a causal Lehmann representation as suggested in the text or state clearly that the method is only reliable for broad features above a scale set by η.","section":"Appendix E"},{"comment":"The comparisons with RPA-CPT, DMRG, and the Müller ansatz are visual and qualitative, and the abstract's claim of 'excellent approximations' is not supported by quantitative error measures. In particular, the Müller comparison in Fig. 5 uses an intensity cutoff that normalizes the maximum of the Müller result to the CPT maximum, which weakens the benchmark. I ask for quantitative metrics such as integrated spectral weights, peak positions, peak widths, or a meaningful normalized difference, to substantiate the stated conclusions.","section":"Section IV, Figs. 3-5"}],"minor_comments":[{"comment":"There is a typo: 'one-eigth' should be 'one-eighth' in the first paragraph of Section V.","section":"Section V"},{"comment":"The name 'Lehman' is used in several places, but the standard spelling is 'Lehmann'; please make the spelling consistent.","section":"Appendices A-B"},{"comment":"The sentence 'χ(q,ω)>0 for ω>0 and χ(q,ω)<0 for ω<0' should refer to the imaginary part of the retarded susceptibility, not the full complex function; please clarify the sign convention.","section":"Appendix E"},{"comment":"The caption states 'The white dots appear where ω<0', which is unclear; please describe what the white dots represent (likely numerical artifacts) and how the reader should distinguish them from physical spectral weight.","section":"Fig. B8 caption"},{"comment":"The paper says 'Following Ref. 12' for the two-particle CPT construction, but Ref. 12 is a VCA paper; please clarify the relationship between the present CPT derivation and the VCA approach of Brehm et al., and specify which steps are new here.","section":"Introduction, Ref. 12"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is honest and clearly written, and the core idea is publishable in principle. However, the central vertex approximation is uncontrolled, the q=0 susceptibility is not benchmarked, and the known causality violation for small broadening means the method is not yet a quantitative tool as presented. I believe these issues can be addressed with additional analysis and benchmarking rather than requiring rejection. I would also encourage the authors to provide the numerical implementation or at least more detailed reproducibility information, since the method is computational in nature."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is my take. The genuinely useful part is that the paper makes two-particle CPT concrete: it derives Eq. (9), implements the Q-matrix form, and benchmarks the transverse spin susceptibility on the 1D Hubbard model at half-filling against RPA, DMRG, and the Muller ansatz. The low-U and large-U limits come out qualitatively right, and the paper is unusually candid about its failures. It admits the vertex approximation is the central assumption, shows the causality violations for small broadening, and acknowledges that the q=pi gap does not extrapolate to zero cleanly. That honesty is real and should count in the paper's favor.\n\nThe soft spot is the one the reader and stress test both point at: Eq. (8), Gamma(iNu,iNu',iOmega) ~ Gamma(iOmega) ~ Gamma_c(iOmega), is asserted. The frequency-independence is borrowed from Ref. 12 and the benchmarks give it indirect support at weak coupling, where the vertex is close to U, and at strong coupling where the spin-exchange picture of the Heisenberg limit helps. But there is no argument for intermediate U, and the magnetic particle-hole vertex is known to carry momentum and frequency structure there. The pseudoinverse in Eq. (D1) is the second load-bearing choice: because Chi_c has zero modes for all Omega, the extracted Gamma_c depends on which null-space component you drop. The paper never shows a q=0 uniform susceptibility comparison, which is exactly the quantity most sensitive to that ambiguity. I would not call that fatal for a methods paper, because the method is presented and benchmarked as approximate, but it is a genuine gap in the validation.\n\nThe validation is qualitative. That is acceptable for a first methods paper with this many benchmarks, but it means I would not yet treat two-particle CPT as a production tool for intermediate U without additional checks. The away-from-half-filling results show significant disagreement with DMRG, which the paper frames as expected; that is fair, but it also means the method's range of practical reliability is narrower than the abstract might suggest.\n\nWho is this for? Cluster-method practitioners and people computing spin response in Hubbard models for neutron or cold-atom comparisons. I would bring it to a group meeting as a methods talk. It deserves a serious referee. The natural revision asks for a q=0 comparison, a clearer statement of when the pseudoinverse is trustworthy, and ideally a suggestion for restoring causality beyond broadening. My recommendation: send to peer review. This is not a desk reject; the implementation is new, the benchmark is useful, and the limitations are openly stated.","headline":"New two-particle CPT implementation with honest benchmarks; the uncontrolled vertex ansatz is the real cost, but the paper earns referee time.","tokens_in":103553,"tokens_out":2136,"would_cite":true,"duration_ms":28140,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.10.Fd","75.40.Gb"],"model":"deepseek-v4-flash","headline":"Cluster perturbation theory is extended to spin susceptibilities: with a frequency-independent vertex approximation, small-cluster data yield a closed formula matching 1D Hubbard benchmarks at weak and strong coupling.","keywords":["cluster perturbation theory","spin susceptibility","Bethe-Salpeter equation","Hubbard model","one-dimensional Hubbard model","two-particle correlation functions","exact diagonalization","density matrix renormalization group"],"falsifier":"Compute the full four-point vertex at two different internal fermionic frequencies for a fixed low transfer frequency $\\omega$ on the half-filled one-dimensional Hubbard model at the largest cluster size available; if $\\Gamma(i\\nu,i\\nu',i\\omega)$ varies substantially with $\\nu-\\nu'$ inside the frequency window where $\\mathrm{Im}\\,\\chi(q,\\omega)$ carries spectral weight, the frequency-independent vertex assumption behind Eq. (9) is falsified precisely in the regime where the paper reports agreement.","tokens_in":102612,"feed_emoji":"🧲","tokens_out":16383,"duration_ms":149138,"temperature":0.7,"pith_summary":"Spin susceptibilities are measured by neutron scattering and ultracold-atom probes, but cluster perturbation theory (CPT), the standard economical method for computing momentum-resolved single-particle spectra from small clusters, has lacked a two-particle counterpart. This paper supplies one: it solves the Bethe-Salpeter equation approximately by assuming the four-point scattering vertex is frequency-independent and equal to the vertex computed on a small cluster, $\\Gamma(i\\nu,i\\nu',i\\omega) \\approx \\Gamma(i\\omega) \\approx \\Gamma_c(i\\omega)$. That assumption turns a costly tensor equation into the closed CPT identity $\\chi^{-1}_{\\mathrm{CPT}} = \\chi^{-1}_{0,\\mathrm{CPT}} + \\chi^{-1}_c - \\chi^{-1}_{0,c}$, in which every piece comes from exact diagonalization of a small cluster. The paper benchmarks the resulting transverse spin susceptibility against the one-dimensional Hubbard model at half filling, reporting close agreement with established results in the weak- and strong-coupling limits and reasonable agreement at intermediate interaction strengths. A sympathetic reader would care because, if correct, the method converts a cheap single-particle calculation into a momentum-resolved two-particle observable, opening a practical route to spin responses in models where more expensive methods struggle.","feed_headline":"Cluster method now computes spin susceptibilities","feed_subtitle":"A frequency-independent vertex lets cluster perturbation theory capture spin responses, matching known Hubbard limits.","key_machinery":"Two devices carry the argument. The first is the frequency-independent vertex approximation: the four-point vertex, the sum of two-particle scattering processes irreducible in the particle-hole channel, is replaced by a two-point vertex that depends only on the transfer frequency and is taken from the cluster, $\\Gamma(i\\nu,i\\nu',i\\omega) \\approx \\Gamma(i\\omega) \\approx \\Gamma_c(i\\omega)$. This collapses the rank-four Bethe-Salpeter equation into a Dyson-like two-leg equation and yields Eq. (9), the susceptibility analogue of the one-particle CPT equation. The second is the pseudoinverse construction of Appendix D, $\\Gamma_c = \\chi^+_{0,c}(\\chi_c - \\chi_{0,c})\\chi^+_c$, which extracts the cluster vertex even though the cluster susceptibility $\\chi_c$ has a zero eigenvalue at every frequency because the paramagnetic cluster preserves SU(2) symmetry. Together these convert a small-cluster exact diagonalization into a momentum-resolved spin response continuous in $q$.","core_discovery":"The paper claims that the one-particle CPT logic transfers directly to two-particle correlation functions. Starting from the Bethe-Salpeter equation for the generalized four-point susceptibility and neglecting the internal frequency dependence of the four-point vertex, $\\Gamma(i\\nu,i\\nu',i\\omega) \\approx \\Gamma(i\\omega) \\approx \\Gamma_c(i\\omega)$, the authors derive the central identity of the paper, Eq. (9): $\\chi^{-1}_{\\mathrm{CPT}}(\\tilde{q},i\\omega) = \\chi^{-1}_{0,\\mathrm{CPT}}(\\tilde{q},i\\omega) + \\chi^{-1}_c(i\\omega) - \\chi^{-1}_{0,c}(i\\omega)$, where $\\chi_{0,\\mathrm{CPT}}$ is the bare bubble built from the CPT Green's function and $\\chi_c$, $\\chi_{0,c}$ are exact cluster quantities. This is the two-particle analogue of the standard CPT Green's function equation $G^{-1}_{\\mathrm{CPT}} = G^{-1}_0 + G^{-1}_c - G^{-1}_{0,c}$, and it reduces to the exact non-interacting and atomic-limit results just as the one-particle version does. Benchmarked at half filling on the one-dimensional Hubbard model with 16-site clusters against DMRG results for 64-site chains, the method is claimed to give close agreement with known results in the weak- and strong-coupling limits and reasonable agreement at intermediate $U$, while away from half filling the comparison deteriorates as the authors expect from enhanced charge fluctuations.","pith_inferences":["My inference: the decisive unbenchmarked step is the vertex approximation itself, so a direct computation of the internal-frequency dependence of $\\Gamma$ in the magnetic particle-hole channel on the largest accessible cluster would reveal where Eq. (9) can be trusted more cheaply than full benchmarks.","Because the pseudoinverse of a singular $\\chi_c$ is not unique, different null-space choices could shift magnetic spectral weight; comparing Eq. (D1) with the paper's rejected $A(A+B)^{-1}B$ inversion route would quantify this sensitivity without invoking new physics.","A natural testable extension is the two-dimensional Hubbard model at strong coupling, where the paper's own entanglement caveat suggests the frequency-independent vertex will break down first, and a controlled large-scale comparison there would bound the method's domain more sharply than the one-dimensional benchmarks do.","The causality violations caused by pole mismatch point to a Lehmann representation of $\\chi_{\\mathrm{CPT}}$, which the paper flags as future work; that would remove the need for large broadening $\\eta$ and make the method usable for sharper real-frequency features."],"forward_implications":["A single small-cluster exact diagonalization yields a momentum- and frequency-resolved spin susceptibility, not just a single-particle spectrum, so parameter sweeps over $U$ and filling become cheap for Hubbard-type models.","The computed $\\chi(q,\\omega)$ is directly comparable with experiment: inelastic neutron scattering on magnetic materials and optical analogues in ultracold-atom lattice systems measure essentially this response.","The approximation is systematically improvable with cluster size, approaching the exact result as $L \\to \\infty$, and the residual gap at $q = \\pi$ is shown to shrink with $1/L$.","The method inherits CPT's domain of validity: it is expected to work in perturbative limits, while the paper cautions that regimes with strong entanglement, such as the two-dimensional Hubbard model at intermediate $U$, will need larger clusters or will be poorly described.","The same scheme carries over to other two-particle correlation functions, provided each channel's leading correlation structure is reflected in the form factors of the vertex approximation."],"supporting_citations":[{"why":"Supplies the one-particle CPT equation whose structure the two-particle result mirrors.","marker":"[2]"},{"why":"Earlier spin-susceptibility computation on a periodized symmetry-broken cluster that neglected inter-cluster terms, the baseline this method improves on.","marker":"[11]"},{"why":"Provides the Bethe-Salpeter cluster approach and the argument that a frequency-independent vertex works in the magnetic particle-hole channel.","marker":"[12]"},{"why":"Gives the phenomenological two-spinon estimate used as the strong-coupling benchmark.","marker":"[17]"},{"why":"Supplies the DMRG Krylov correction-vector method used to benchmark the susceptibility.","marker":"[19]"},{"why":"Provides DMRG magnetic excitation spectra of quasi-one-dimensional systems used for the large-$U$ comparison.","marker":"[20]"},{"why":"DMRG results for the Mott-insulating chain used to validate the CPT Green's function and the intermediate-coupling spin response.","marker":"[21]"},{"why":"Establishes the SU(2)-symmetry zero eigenvalue of the cluster susceptibility, which forces the pseudoinverse inversion.","marker":"[22]"}],"fun_headline_variants":["CPT computes spin susceptibilities in Hubbard model","Two-particle CPT: spin response from clusters","Cluster perturbation theory for spin correlations","Spin susceptibilities via cluster perturbation theory","Extending CPT to two-particle spin dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that the effective scattering interaction between two particles is the same at every energy and can be borrowed from a small cluster; if that interaction actually depends strongly on energy, the susceptibility formula has no controlled error.","fun_headline_variants_meta":{"raw":{"variants":["CPT computes spin susceptibilities in Hubbard model","Two-particle CPT: spin response from clusters","Cluster perturbation theory for spin correlations","Spin susceptibilities via cluster perturbation theory","Extending CPT to two-particle spin dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000208,"raw_usage":{"total_tokens":1417,"prompt_tokens":972,"completion_tokens":445,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":588,"completion_tokens_details":{"reasoning_tokens":381}},"tokens_in":588,"tokens_out":445,"duration_ms":4442,"temperature":1.0,"reasoning_tokens":381,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T10:45:02.379428+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the full four-point vertex at two different internal fermionic frequencies for a fixed low transfer frequency $\\omega$ on the half-filled one-dimensional Hubbard model at the largest cluster size available; if $\\Gamma(i\\nu,i\\nu',i\\omega)$ varies substantially with $\\nu-\\nu'$ inside the frequency window where $\\mathrm{Im}\\,\\chi(q,\\omega)$ carries spectral weight, the frequency-independent vertex assumption behind Eq. (9) is falsified precisely in the regime where the paper reports agreement.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier spin-susceptibility computation on a periodized symmetry-broken cluster that neglected inter-cluster terms, the baseline this method improves on."},{"cited_title":"Brehm , author E","cited_arxiv_id":null,"evidence_quote":"Provides the Bethe-Salpeter cluster approach and the argument that a frequency-independent vertex works in the magnetic particle-hole channel."},{"cited_title":"Nocera , author N","cited_arxiv_id":null,"evidence_quote":"Provides DMRG magnetic excitation spectra of quasi-one-dimensional systems used for the large-$U$ comparison."},{"cited_title":"Nocera , author F","cited_arxiv_id":null,"evidence_quote":"DMRG results for the Mott-insulating chain used to validate the CPT Green's function and the intermediate-coupling spin response."},{"cited_title":"Filor \\ and\\ author T","cited_arxiv_id":null,"evidence_quote":"Establishes the SU(2)-symmetry zero eigenvalue of the cluster susceptibility, which forces the pseudoinverse inversion."}],"review_version":1}