{"id":"65af85d8-c205-411f-9c21-4ec8489f53e8","arxiv_id":"2412.01848","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Ten exact three-particle Bethe-Salpeter ladder equations are derived, and an approximate ladder built from two-particle vertices matches exact non-linear response only qualitatively at weak coupling.","lead":"This paper derives exact ladder equations for the three-particle vertex, the next level beyond the standard two-particle Bethe-Salpeter equations, and tests a practical approximation built from two-particle vertices. The approximation matches an exact impurity-model solution only qualitatively and only for weak interactions, so the paper mainly contributes a formal framework and a cautionary benchmark.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 1/9 channel average in Eq. (53) is uncontrolled: the claim that high-order ladder terms are channel-exclusive is a stated conjecture, so the numerical benchmark cannot cleanly separate failure of the M-approximation from failure of the crossing-symmetry repair.","rationale":"The reader's weakest_assumption correctly identifies the 1/9 channel average and the channel-exclusivity conjecture as the least secure part of the numerical argument. My stress-test agrees: the formal ladder equations are plausible and are supported by the detailed reducibility classification in the Supplemental Material, and the authors honestly state that the average is a conjecture. However, the numerical conclusion of the paper, namely that the approximate ladder is only qualitatively correct at weak coupling, depends on this conjecture for every point in the comparison. If the conjecture fails, the discrepancy could be caused by the symmetrization procedure rather than by the replacement of the irreducible three-particle vertex with two-particle vertices, which would change the interpretation of the benchmark. The proposed test, an order-by-order Feynman-diagram multiplicity count, is concrete and feasible, and it would settle whether the averaging is the dominant source of error. I do not find a strong reason to reject the paper: the exact equations are the main contribution, the approximation is openly labeled as approximate, and the released code provides reproducibility. The verdict should remain conditional, as the reader already set it, pending the suggested verification.","tokens_in":36928,"tokens_out":19456,"duration_ms":179420,"concrete_test":"Perform an order-by-order diagrammatic enumeration for the pph ladders: generate all distinct three-particle diagrams through fourth order in the two-particle vertex F2, count how many of the nine pph channels generate each diagram, and verify whether any diagram with three or more vertices appears in more than one channel. This directly tests the Section VC conjecture. If the multiplicity exceeds one, recompute χ_vertex with the exact inclusion-exclusion subtraction up to that order and compare with the ED result; if the corrected curve matches ED, the 1/9 average is the dominant error, whereas if it still differs, the replacement M ≈ M12+M23+M13 itself is the limiting approximation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equations (32)-(33) are exact by construction if the irreducible building block Γ(1,⃗2,3)PI,r is known, and that formal part is not where I would direct the main objection. The load-bearing weak point is the step from these exact equations to the numerical comparison: Eq. (53) replaces the unknown irreducible vertex by M12+M23+M13 and then averages the resulting ladder plus the 1PR diagram over all nine pph channels with weight 1/9. The paper shows this average is correct through second order, but for all higher orders it rests on the conjecture, stated in Section VC, that ladder diagrams of sufficiently high order are exclusive to a single channel. Nothing in the manuscript tests this conjecture. If it is false, the 1/9 factor suppresses higher-order contributions irregularly, and the observed factor-two-to-ten discrepancies and sign changes at U=2 could be an artifact of the symmetrization procedure rather than evidence about the quality of the M ≈ M12+M23+M13 replacement. Because the same averaging is used for every data point in Figs. 21-22, the paper's central numerical conclusion is conditional on an untested diagrammatic-multiplicity assumption. A secondary but related issue is that the 'exact' reference is a six-site ED bath while the ladder is built from parquet/QMC vertices for the continuous-bath AIM, so the comparison mixes two approximations unless the six-site bath is verified to be converged for three-particle quantities.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper derives a set of three-particle ladder equations, generalizing the two-particle Bethe-Salpeter equations to ten three-particle channels. The central formal statement is that, if the three-particle vertex irreducible in one of the ten channels, Γ(1,⃗2,3)PI,r, is known, then Eqs. (32) and (33) yield the exact three-particle vertex, provided the one-particle-reducible diagram (F2GF2)r is added in the pph channels. Since Γ(1,⃗2,3)PI,r is not available, the authors approximate it by the sum of two-particle irreducible vertices M12+M23+M13, build the resulting ladder, and restore crossing symmetries approximately by averaging over all nine pph channels with a factor 1/9, Eq. (53). The approximation is then used to compute the vertex contribution to the second-order spin response function of an Anderson impurity model at DMFT self-consistency, and the results are compared with exact diagonalization for U = 0.5, 1, 2, 3, 4. The authors conclude that the approximation is only qualitatively correct for weak U and quantitatively off by factors of two to ten, with wrong signs for some components at larger U.","tokens_in":37220,"tokens_out":4262,"duration_ms":39707,"significance":"If the formal ladder equations are correct, they constitute a useful framework for organizing three-particle diagrammatics, and the careful treatment of one-, two-, and three-particle reducibility, including the one-particle-reducible diagram and the double-counting issues, is a genuine contribution. The paper also ships an open-source implementation and does not fit any target quantity: the input two-particle vertices are obtained independently from parquet/QMC and the comparison is against an ED solution of the same impurity model. The honest negative conclusion about the M12+M23+M13 approximation is valuable. However, the numerical benchmark—which is the main evidence for that negative conclusion—is conditional on an uncontrolled 1/9 channel average and on the use of a six-site ED reference without demonstrated convergence for three-particle quantities. The formal part is strong; the quantitative part needs substantial additional support.","major_comments":[{"comment":"The average over the nine pph channels with prefactor 1/9 is justified in the manuscript only for second-order diagrams. For all higher orders, the argument relies on the conjecture, stated in Section VC, that ladder terms of sufficiently high order are exclusive to a single channel. This conjecture is not tested anywhere in the manuscript. Since every data point in Figs. 21 and 22 is obtained through this averaging, the observed factor-of-two-to-ten discrepancies and the sign changes at U = 2 and U = 3 cannot be unambiguously attributed to the M12+M23+M13 replacement; they could partly be artifacts of the symmetrization procedure. I recommend a concrete check: for a fixed low-order set of ladder diagrams, explicitly enumerate the pph channels in which each topology appears and verify the 1/9 counting, or compare the averaged result with an alternative symmetrization that respects all six crossing symmetries at higher orders.","section":"Section VC and Eq. (53)"},{"comment":"The reference solution is an exact-diagonalization calculation with six bath sites obtained by pole fitting, while the ladder is evaluated using two-particle vertices from parquet and continuous-time QMC calculations for the continuous-bath AIM under DMFT self-consistency. The manuscript does not demonstrate that the six-site bath is converged for three-particle quantities such as χ(2)_vertex. If the six-site bath is not converged, part of the discrepancy between the ladder and the ED reference could come from the reference itself. At minimum, the authors should provide a convergence check (e.g., comparing ED one- and two-particle vertices to the parquet/QMC input for the same parameters, or increasing the number of bath sites), and error bars in Figs. 21 and 22.","section":"Section VA and Figs. 21–22"},{"comment":"The eigenvalue analysis in Fig. 20 shows that the largest eigenvalue of M is close to unity at U = 3 and exceeds unity at U = 4, so the geometric series defining the ladder does not converge in the ordinary sense for these interaction strengths. The manuscript nevertheless uses the closed-form expression for the geometric series. For a non-convergent series, this closed form is an analytic continuation whose connection to the original ladder diagrams is not established. Consequently, the U = 4 entries in Table IV and Fig. 22 may reflect the continuation procedure rather than the quality of the ladder approximation. This should be stated explicitly, or the U = 4 data should be excluded or reinterpreted.","section":"Section VB and Fig. 20"}],"minor_comments":[{"comment":"The phrase \"even than only qualitatively\" should be \"even then only qualitatively\".","section":"Abstract"},{"comment":"The sentence \"we express the three-particle vertex as a a geometric ladder series\" contains a duplicated article \"a\".","section":"Section IVA"},{"comment":"The sentence \"the imaginary part of χvertex,↑↑↑ and χvertex,↑↑↓ must vanish\" should use the plural \"imaginary parts\".","section":"Section IVC, after Eq. (55)"},{"comment":"The ED and ladder panels use different color scales, which makes the factor-of-two-to-ten discrepancy difficult to judge by eye; using a common color scale or explicitly plotting the ratio would improve comparability.","section":"Figs. 21 and 22"}],"recommendation":"major_revision","confidential_remarks":"The formal ladder equations and the reducibility analysis are the strongest part of the manuscript and are likely publishable even if the numerical approximation performs poorly. The main risk is that the quantitative benchmark is not yet conclusive because of the untested 1/9 channel average and the six-site ED reference. If the authors can supply a convergence check for the ED reference and a test of the channel-exclusivity conjecture, the numerical claim would be much stronger; otherwise, the numerical section should be reframed as an illustrative application rather than a quantitative benchmark."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: the exact ten-channel ladders are the real contribution, and the approximate ladder's failure is honestly reported, not a flaw. This paper deserves a serious referee, mainly for the formal reducibility framework.\n\nThe new piece is the systematic classification of one-, two-, and three-particle reducibility for the three-particle vertex, and the ten Bethe-Salpeter-like equations, Eqs. (32)-(33), which are exact conditional on knowing the appropriate irreducible vertex. The derivation of the approximate ladder from two-particle vertices is careful: double counting from the pp vertex, the permutation symmetrization, the disconnected term subtraction, and the geometric series are all handled explicitly. The code is open source, and the Supplemental Material gives a substantial algorithm for isolating the fully irreducible three-particle vertex. That is real and useful.\n\nThe soft spots are where the paper itself is honest about them. The 1/9 averaging over pph channels in Eq. (53) is only proven correct through second order; for higher orders it rests on the stated conjecture that high-order ladder diagrams are channel-exclusive. That conjecture is not tested. If it fails, the factor 1/9 suppresses higher-order contributions irregularly, so the observed quantitative failure at moderate U could be partly an artifact of the symmetrization, not purely the M ≈ M12+M23+M13 replacement. The paper cannot separate those two sources of error. This does not invalidate the formal part, but it means the numerical benchmark is less conclusive than the abstract implies.\n\nSecondary issue: the 'exact' reference is a six-site ED bath, while the ladder builds on two-particle vertices from parquet/QMC for the DMFT bath. No error bars or convergence check for three-particle quantities is given. The comparison is between two approximations unless the six-site bath is verified for three-particle response.\n\nWho this is for: people working on diagrammatic extensions of DMFT or non-linear response who need three-particle vertices. The formal equations and the supplemental reducibility algorithm are the main value; the approximate ladder is shown to be quantitatively unreliable, which is itself a useful data point.\n\nRecommendation: send to peer review. A good referee should push on the averaging conjecture and the ED convergence, but the formal derivation deserves publication. I would accept the paper for review with those revisions in mind.","headline":"The exact ten-channel ladder equations are a genuine formal contribution, and the paper's honest numerical failure is a useful data point; the main soft spot is the untested 1/9 averaging conjecture that underpins the quantitative comparison.","tokens_in":37730,"tokens_out":2791,"would_cite":true,"duration_ms":25729,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper derives exact ladder equations for the three-particle vertex and tests the natural first approximation, finding that for the Anderson impurity model it is only qualitatively correct at weak coupling and misses quantitatively by…","keywords":["three-particle vertex","Bethe-Salpeter ladder","Anderson impurity model","nonlinear response function","crossing symmetry","parquet decomposition","irreducible vertex","diagrammatic many-body theory"],"falsifier":"Compute the fourth-order (two-vertex) ladder contributions in two different pph channels, for example channels 41 and 25, and check whether any diagram appears in both; the paper's Section VC conjecture says high-order ladder terms are exclusive to one channel, so finding overlap would show that the $1/9$ averaging suppresses high orders by an uncontrolled factor. A second check is to repeat the exact-diagonalization-versus-ladder comparison at $U=0.5$ with a different exact solver, such as continuous-time quantum Monte Carlo, and see whether the factor-of-two-to-ten discrepancy persists; if it disappears, the claim that the approximation is only qualitatively correct at weak coupling would be contradicted.","tokens_in":36687,"feed_emoji":"⚛️","tokens_out":13778,"duration_ms":105683,"temperature":0.7,"pith_summary":"The paper generalizes the two-particle Bethe-Salpeter equations to ten three-particle ladders and derives exact equations that would give the full three-particle vertex if the three-particle vertex irreducible in one of ten channels were known. Since that irreducible input is not available in practice, the paper approximates it by the sum of three two-particle irreducible vertices, each connecting two of the three fermionic lines, and repairs the resulting loss of crossing symmetry by averaging the ladder over the nine particle-particle-hole channels. The approximation is tested on the Anderson impurity model with a bath fixed by dynamical mean-field theory, where the exact three-particle vertex can be obtained by exact diagonalization. The comparison shows that the approximate ladder is only qualitatively correct for weak interaction, misses the exact vertex contribution to the second-order response by factors of two to ten, and produces wrong signs for some spin components at stronger coupling. The exact ladder framework itself is the paper's main positive result; the tested approximation is a first step that the authors themselves find insufficient.","feed_headline":"Three-particle ladder is exact: first test misses by 10x","feed_subtitle":"First practical approximation of the three-particle vertex is only qualitatively right at weak coupling.","key_machinery":"The central object is the three-particle Bethe-Salpeter-like ladder, written as $\\Gamma_{1\\mathrm{PI},r} = \\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r} + \\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r} \\cdot \\Gamma_{1\\mathrm{PI},r}$, with the dot connecting three Green's function lines, and the companion equation $F_3 = \\Gamma_{1\\mathrm{PI},r} + (F_2 G F_2)_r$ that adds the single one-particle-reducible diagram of the pph channel. The arrow in $\\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r}$ marks a vertex that is one- and three-particle irreducible in channel $r$ and two-particle irreducible on the right, a distinction needed because three-particle diagrams can be cut in non-unique ways. In the approximation, this irreducible input is replaced by $M = M_{12} + M_{23} + M_{13}$, where each $M_{ij}$ connects two of the three lines through a two-particle irreducible vertex ($\\Gamma_{\\mathrm{ph}}$ or $\\Gamma_{\\mathrm{pp}}$ with the required $1/2$ factor), and the ladder is summed in closed form as a geometric series. Since the resulting ladder has only two of the six crossing symmetries of the full vertex, the final approximate vertex averages $L$ and the one-particle-reducible term over all nine pph channels with weight $1/9$.","core_discovery":"The central claim is that the full three-particle vertex $F_3$ obeys an exact ladder structure: $\\Gamma_{1\\mathrm{PI},r} = \\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r} + \\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r} \\cdot \\Gamma_{1\\mathrm{PI},r}$ with $F_3 = \\Gamma_{1\\mathrm{PI},r} + (F_2 G F_2)_r$, where $r$ labels one of ten channels and $(F_2 G F_2)_r$ is the unique one-particle-reducible diagram of that pph channel. If the vertex $\\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r}$ that is 1PI and 3PI in channel $r$ and 2PI on the right were known, iterating the first equation and adding the second term would produce the exact full three-particle vertex. The paper then implements the simplest available replacement — $\\Gamma_{(1,\\vec{2},3)\\mathrm{PI},r}$ approximated by $M_{12} + M_{23} + M_{13}$, built from irreducible two-particle vertices — and evaluates the resulting geometric ladder for the Anderson impurity model. For the vertex contribution to the second-order spin response $\\chi_{\\mathrm{vertex}}^{(2)}$, the ladder is only qualitatively correct for weak coupling and quantitatively off by factors of two to ten, with sign errors in some spin components at $U = 2$ and above.","pith_inferences":["A natural next step, not taken in the paper, is to feed a numerically estimated local three-particle irreducible vertex into the exact ladder equations instead of the two-particle-vertex sum; the paper's exact equations would then become a systematic route to three-particle vertices rather than a single uncontrolled approximation.","The sign change between $U=1$ and $U=2$ in the $\\uparrow\\uparrow\\uparrow$ component suggests that the $1/9$ averaging changes the topology of the result, not just its magnitude; tracking individual channel contributions at intermediate $U$ could show whether one channel dominates and whether the averaging hides a simpler structure.","The paper's conjecture that high-order ladder terms are exclusive to one channel is directly testable with the same code: comparing fourth-order contributions in channels 41 and 25 would reveal whether the $1/9$ factor suppresses them by an uncontrolled amount.","For physical second-order responses such as $\\chi_{nnn}$, $\\chi_{nzz}$, and $\\chi_{xyz}$, which are linear combinations of the raw spin components, the factor-of-two-to-ten errors could partially cancel or accumulate; the paper deliberately avoids this masking, but practical calculations will need to test the combined quantities directly."],"forward_implications":["If the irreducible three-particle vertex in any one of the ten channels were supplied exactly, Eqs. (32) and (33) would yield the exact full three-particle vertex, including the one-particle-reducible diagrams that are absent from ppp-type ladders.","The $1/9$ channel average makes the approximate vertex correct through second order in the two-particle vertex, but it cannot control the higher-order errors that cause the documented quantitative failures.","For the tested Anderson impurity model at $\\beta = 10$, the ladder is qualitatively reliable only up to $U \\approx 1$ for the $\\uparrow\\uparrow\\uparrow$ spin component and $U \\approx 3$ for $\\uparrow\\uparrow\\downarrow$, with quantitative errors of factors of two to ten.","At stronger coupling the approximate vertex contribution can have the wrong sign, so extracting nonlinear response coefficients from this two-particle-vertex-only ladder would not be reliable.","The ppp channel requires no separate numerical treatment because its ladder diagrams are particle-hole-symmetry related to pph ladders."],"supporting_citations":[{"why":"It defines the ten three-particle frequency notations and the decomposition of the second-order response into bubble, first-order, and vertex terms that the comparison uses.","marker":"[18]"},{"why":"It supplies the parquet decomposition and the $1/2$ factor in the pp channel that the two-particle irreducible building blocks inherit.","marker":"[33]"},{"why":"It gives the two-particle Bethe-Salpeter equation that the three-particle ladder generalizes.","marker":"[34]"},{"why":"It provides the parquet solver used to compute noise-free irreducible two-particle vertices for the weak-coupling cases.","marker":"[40, 41]"},{"why":"It supplies the exact diagonalization code that computes the exact three-particle vertex and response functions used as reference.","marker":"[42]"},{"why":"It defines the dynamical mean-field self-consistency that fixes the bath of the Anderson impurity model used in the numerical comparison.","marker":"[4]"}],"fun_headline_variants":["Exact three-particle ladder, but approximation only for weak coupling","Three-particle vertex ladder exact, approximation weak-coupling-only","Exact ladder for three-particle vertex: approximation qualitatively right only","Three-particle ladder exact, approximation misses by up to 10x"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the unknown irreducible three-particle vertex can be replaced by the sum of three two-particle irreducible vertices, $M_{12} + M_{23} + M_{13}$, and that averaging the resulting ladder over the nine pph channels with a $1/9$ factor restores the missing crossing symmetries without suppressing higher-order contributions in an uncontrolled way; the paper itself flags the second part as a conjecture in Section VC.","fun_headline_variants_meta":{"raw":{"variants":["Exact three-particle ladder, but approximation only for weak coupling","Three-particle vertex ladder exact, approximation weak-coupling-only","Exact ladder for three-particle vertex: approximation qualitatively right only","Three-particle ladder exact, approximation misses by up to 10x"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001304,"raw_usage":{"total_tokens":5318,"prompt_tokens":947,"completion_tokens":4371,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":563,"completion_tokens_details":{"reasoning_tokens":4300}},"tokens_in":563,"tokens_out":4371,"duration_ms":28018,"temperature":1.0,"reasoning_tokens":4300,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:13:52.468835+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the fourth-order (two-vertex) ladder contributions in two different pph channels, for example channels 41 and 25, and check whether any diagram appears in both; the paper's Section VC conjecture says high-order ladder terms are exclusive to one channel, so finding overlap would show that the $1/9$ averaging suppresses high orders by an uncontrolled factor. A second check is to repeat the exact-diagonalization-versus-ladder comparison at $U=0.5$ with a different exact solver, such as continuous-time quantum Monte Carlo, and see whether the factor-of-two-to-ten discrepancy persists; if it disappears, the claim that the approximation is only qualitatively correct at weak coupling would be contradicted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It supplies the parquet decomposition and the $1/2$ factor in the pp channel that the two-particle irreducible building blocks inherit."},{"cited_title":"Rohringer, A","cited_arxiv_id":null,"evidence_quote":"It gives the two-particle Bethe-Salpeter equation that the three-particle ladder generalizes."},{"cited_title":"Krien and A","cited_arxiv_id":null,"evidence_quote":"It supplies the exact diagonalization code that computes the exact three-particle vertex and response functions used as reference."},{"cited_title":"Georges and G","cited_arxiv_id":null,"evidence_quote":"It defines the dynamical mean-field self-consistency that fixes the bath of the Anderson impurity model used in the numerical comparison."}],"review_version":1}