{"id":"2e634079-e476-4dc5-a778-2ce84ffabe78","arxiv_id":"hep-ph/9711266","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The authors derive an algorithm and explicit formulae for computing three-loop beta functions and anomalous dimensions in any local quantum field theory with canonical kinetic terms.","lead":"This paper gives a general recipe for computing how coupling constants and operator strengths change with energy scale in quantum field theories, up to three loops of perturbation theory. It supplies the explicit mathematical building blocks needed to automate these calculations, which are essential for precision predictions at particle colliders.","discovery_kind":"unclear","skeptic_critique":{"model":"selfhosted/deepseek-v4.1-flash","headline":"Eqn (25)'s cancellation is asserted, not shown, but appendix C supplies every input needed to check it in one line; the deeper gap is that the main formula (36) has no independent cross-check beyond I(3)_111111 and the authors' own applications.","rationale":"The reader's weakest_assumption is the right target, but the severity is overstated. Eqn (25) is not an unbacked assertion: appendix C gives A and B in closed form, and the cancellation is a one-line check for each (k1,k2,r) that enters (36). I verified the two leading cases explicitly and they work. So the gap is presentational -- the paper should display the check -- rather than a correctness risk that warrants treating the main formula as unvalidated. The reader's second point, that the Mathematica code is not shipped, is a 1997 artifact and not a substantive concern; the formulae are the deliverable. What the reader underweights is that eqn (36) itself has no independent cross-check in the text: the single anchor I(3)_111111=2zeta(3)/epsilon is case A and does not exercise the n!=0, higher-index sums, and [12,13] are the authors' own applications. That is the more load-bearing residual gap for a methodology paper. The m^2 counterterm argument in section 2 is acknowledged as a plausibility argument, so it is not a new finding. Overall the paper's central claim holds up under scrutiny; CONDITIONAL is defensible, and my refinement does not overturn it, so I recommend UNCHANGED rather than ACCEPT or REJECT. The concrete test above would settle whether the general formula (36) is correct beyond the single case-A value, which is the one thing that would actually move the verdict.","tokens_in":18673,"tokens_out":10645,"duration_ms":171844,"concrete_test":"Take a representative nontrivial case, e.g. n6=-1, n1=n2=n4=n5=1, n3=2, and evaluate eqn (36) using the A and B of appendix C; independently evaluate the same I(3) by the integration-by-parts recursion of [8,9] (or against a known three-loop tadpole value). If the two agree at O(1/epsilon), eqn (25) and (36) are validated for that case; if they differ, the concern lands.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader correctly identifies eqn (25), aren(k1,k2,r)+b(k1,k2,r)=O(1), as the condition that keeps the curly bracket of eqn (24) free of spurious 1/epsilon poles when n3-2+r1+r2+k is nonzero. I checked it against appendix C: for (k1,k2,r)=(1,1,0), A=0 and B=Gamma(1-epsilon)^2 Gamma(epsilon)/Gamma(2-2epsilon)=1/epsilon+O(1), so aren=-1/epsilon and aren+b=O(1); for (1,1,1), A=2Gamma(-1+epsilon)=-2/epsilon+O(1) and B=2/epsilon+O(1), again O(1). The cancellation is real and follows directly from the explicit A and B in eqns (64)-(65). So the reader's concern is a self-containedness gap, not a correctness risk: the paper says 'a short calculation' and supplies all inputs. The more load-bearing residual issue is that the main deliverable, eqn (36), is not independently validated in the text. The only in-text anchor is I(3)_111111=2zeta(3)/epsilon (eqn 16), which is case A and does not exercise the n!=0, higher-index machinery of (36); the applications [12,13] are the authors' own and are not independent checks. The m^2 counterterm argument of section 2 is explicitly a plausibility argument ('might be more convincing'). None of this is fatal -- the formulae are explicit and checkable -- but for a methodology paper whose value is the formula, an independent numerical or analytic cross-check of (36) would materially strengthen the claim.","agreement_with_reader":"partial"},"referee_report":{"model":"selfhosted/deepseek-v4.1-flash","summary":"The manuscript presents an algorithm for computing three-loop β-functions and anomalous dimensions in dimensional regularization and the MS/MS-bar scheme, for local quantum field theories with canonical kinetic terms. The infrared rearrangement is performed by adding a single common artificial mass m to every propagator denominator, so that after expansion in the physical masses and external momenta the UV counterterms reduce to pole parts of completely massive tadpoles. The paper supplies explicit formulae for all scalar integrals needed at one, two and three loops: eq. (9) for I(1)_n, the recursion (10) with boundary value (11) for I(2), and a case analysis (A, B, C) for I(3) culminating in eq. (36), which gives the pole parts of all nontrivial three-loop integrals. Four appendices cover tensor reduction, reducible integrals, the large-q^2 expansion coefficients A and B, and traceless-tensor identities. Section 5 relates the renormalization constants to β and γ in and beyond the MS scheme.","tokens_in":19095,"tokens_out":23523,"duration_ms":410185,"significance":"If correct, this is a genuinely useful methodology paper: it turns the evaluation of three-loop UV counterterms into a mechanical procedure and provides the explicit three-loop tadpole formulae that the community lacked. The work underpins the authors' own three-loop anomalous dimensions for B → X_s γ [12,13], and the same IR rearrangement was subsequently used in the four-loop QCD β-function computation of [16]. Strengths worth naming: the coefficients A and B in appendix C are given in closed form for arbitrary ω, so the main formula is checkable by the reader; the algebra of section 5 is verifiable (I checked eqs. (39) and (40) against the RG equation and they are consistent); the comparison with [16] in section 1 is candid about what is shared and what is new; and the domain of applicability (canonical kinetic terms, bounded operator dimension) is stated rather than glossed over.","major_comments":[{"comment":"The abstract claims an algorithm valid in 'any local quantum field theory with canonical kinetic terms', but the m^2-counterterm bookkeeping that makes the argument work is carried out only for QCD. Section 2 concedes that these counterterms 'may not preserve symmetries of the theory' and that their number is 'usually rather small', and then gives only the gluon mass counterterm Z_x of eq. (4). For a general theory with elementary scalars and several mass parameters, the relevant dimension-2 operators (φ^2 for each scalar, A_μ^a A^{aμ} for each gauge field) and the demonstration that the m^2 subtractions do not feed back into the pole parts are not written down. Since this is exactly where the generality claim could fail, either give the general prescription explicitly (enumerate the dimension-2 operators allowed by the field content and show the m^2 counterterm can be fixed at one loop) or restrict the abstract's claim to theories where this enumeration has been performed. The related restriction stated in §2, that the operator dimension be bounded from above, also qualifies 'any local quantum field theory' and should appear in the abstract or introduction.","section":"Abstract (abstract:s0) and §2, eqs. (3)–(4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The main deliverable, eq. (36), is not independently validated in the text. The only in-text anchor is eq. (16), I(3)_111111 = 2ζ(3)/ε, which is case A and does not exercise the n ≠ 0 machinery, the higher-index cases B.2/B.3, or the tensor decomposition of case C.2; the applications [12,13] are the authors' own and therefore not an independent check. A short table comparing a handful of entries of (36) — for instance n = 0 with n3 = 1, 2 and several choices of n1, n2, n4, n5, plus one or two n = 1, 2 entries — against the recursion algorithms of [8,9] or against a direct ε-expansion would convert the formula from 'explicit and checkable' to 'checked'. This is a correctness-risk concern rather than a claim of error, and the requested check is local.","section":"§4, eq. (36)","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The cancellation a^ren(k1,k2,r) + b(k1,k2,r) = O(1) is asserted with 'Verifying this requires a short calculation' but the calculation is not displayed. It is in fact one line from eqs. (64)–(65): for (k1,k2,r) = (1,1,0), A = 0 and B = Γ(1−ε)^2Γ(ε)/Γ(2−2ε) = 1/ε + O(1), so a^ren = −1/ε and the sum is O(1); for (1,1,1), A = 2Γ(−1+ε) = −2/ε + O(1) and B = 2/ε + O(1), again O(1). Displaying this, or pointing to eqs. (64)–(65) explicitly, would close the self-containedness gap at negligible cost.","section":"§4, eq. (25)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The truncation of the r1, r2 sums at 2−n3 (and at 2+n−n3 in eq. (36)) is not explained. A sentence stating why terms with larger r cannot contribute to the pole — namely that the denominators n3−2+r1+r2+k are then bounded away from zero and the bracket is O(1) by eq. (25) — would let the reader follow the derivation without reconstructing it.","section":"§4, eqs. (26) and (36)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The double-pole subtraction term proportional to Γ(n3−n−2+ε)/Γ(n3) needs a stated condition on n3 for it to be well defined; as written, the reader cannot tell whether the formula is intended for all n3 ≥ 1 or only for n3 > n+1.","section":"§4, eq. (36), last term","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The split G2 = (1/ε)δ_{1k1}δ_{1k2} + G2^ren is written as if the 1/ε were a UV pole of G2, but the one-loop bubble with two propagators is UV finite in D = 4−2ε. It would help to say explicitly that the pole is that of the large-q^2 coefficient b(1,1,0), not of G2 itself, since this is the point where a reader is most likely to stumble.","section":"§3, case B.3, eq. (20)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The reduction formula for the reducible three-loop integral is stated without derivation. A sentence on its origin — the decomposition of the negative-index propagators in the spirit of eq. (2), followed by the tensor reduction of appendix A — would make the appendix usable without reverse-engineering.","section":"Appendix B, eq. (60)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The relations among the coefficients a^ij and the anomalous-dimension matrix are given without derivation. Since eqs. (39) and (40) check out, a brief indication of how (43) and (45) follow from the same RG equation, or a check against a known two-loop anomalous dimension, would strengthen this section.","section":"§5, eqs. (43) and (45)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The statement that 'no explicit formulae for three-loop massive tadpoles have been published so far' is a priority claim that is difficult for a reader to verify. Consider softening it or citing the closest existing results from [8,9] explicitly, so the claim is anchored.","section":"§1, literature claim","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The symbols G2, G2^ren, a, b, A, B, a^ren and A^ren carry different arguments in different equations (k1, k2, r versus k1, k2+n, 2+n, r), and the reader must consult eq. (18) repeatedly. A small table collecting the definitions and argument lists would improve readability.","section":"Notation, §3–§4","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision","confidential_remarks":"The paper is a 1997 methodology preprint; the technical content holds up well on inspection, and the section 5 algebra in particular is correct. The two major comments are claim-scope and validation requests rather than indications of error, and both are fixable locally. The priority discussion relative to [16] in section 1 is handled fairly and does not need editorial intervention."},"author_rebuttal":null,"desk_editor":{"model":"selfhosted/deepseek-v4.1-flash","letter":"Bottom line: this is the reference for pole parts of three-loop massive tadpoles, and it holds up. The formula you would actually use is eqn (36), and it is explicit enough to check by hand at small indices. If you do three-loop renormalization in QCD or an effective theory with canonical kinetic terms, this is where you look up the integrals, and it has earned its citation count over the last quarter century by being correct and usable rather than by being elegant. The exact propagator decomposition in eqn (2) is clean, the reduction of tensor to scalar integrals via symmetric traceless tensors in appendix D is tidy, and the two-loop recursion (10) is standard but assembled in a form you can program in an afternoon. The authors are also honest about scope: canonical kinetic terms only, no HQET, and they say so up front. On the reader's main worry, eqn (25): I checked it against appendix C and it is fine. For (k1,k2,r)=(1,1,0), A=0 and B=1/epsilon+O(1), so aren+b=O(1); for (1,1,1), A=-2/epsilon and B=+2/epsilon. The cancellation is real and follows in one line from eqns (64)-(65). So that is a self-containedness nit, not a correctness risk, and the reader over-weighted it. The genuine soft spot is different: eqn (36) has no independent in-text validation. The only anchor is eqn (16), which is case A and does not exercise the n!=0, higher-index machinery. The applications [12,13] are the authors' own and are not independent checks. A numerical cross-check at, say, n3=3, n=1 against a direct Feynman-parameter evaluation would have sealed it, and its absence is the one thing I would ask for. Secondary: the Mathematica code is described but not shipped, which costs reproducibility, though the formulae are explicit enough to reimplement; and the one-loop Zx in eqn (4) is quoted without derivation, which is minor. Section 2's m^2 counterterm argument is a plausibility argument, but the authors flag it as such, so it is acknowledged rather than hidden. Who should read it: anyone computing three-loop anomalous dimensions or beta functions with massive tadpole reduction. Not for HQET or SCET practitioners. Recommendation: send it out. It is a tool paper and the tool is real; the referee should ask for the eqn (36) cross-check and the code, not for a rewrite.","headline":"A solid, still-useful tool paper: the three-loop tadpole formulae in eqn (36) are the deliverable and they hold up, with the only real gap being the absence of an independent in-text check of the general case.","tokens_in":19659,"tokens_out":1475,"would_cite":true,"duration_ms":24439,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":[],"pacs":["11.10.Gh","12.38.Bx","11.15.Bt"],"model":"selfhosted/deepseek-v4.1-flash","headline":"The paper derives an algorithm that reduces the ultraviolet divergences of any local quantum field theory with canonical kinetic terms to sums of completely massive tadpole integrals, and supplies explicit formulas for every scalar integral","keywords":["perturbative quantum field theory","beta function","anomalous dimension","dimensional regularization","MS-bar scheme","infrared rearrangement","massive tadpole integrals","three-loop Feynman diagrams"],"falsifier":"Recompute $a^{mathrm{ren}}(k_1,k_2,r)+b(k_1,k_2,r)$ to order $epsilon^0$ from the appendix C expressions for a few index triples and confirm the cancellation, then evaluate one nontrivial integral, such as $I^{(3)}_{211100}$ or $I^{(3)}_{11111-1}$, by an independent route and compare with eqn (36); the paper's own suggestion of keeping the cutoff $Lambda$ arbitrary and checking that the pole is $Lambda$-independent provides a second, self-contained test.","tokens_in":20023,"feed_emoji":"⚛️","tokens_out":3992,"duration_ms":64885,"temperature":0.7,"pith_summary":"Beta functions and anomalous dimensions are the quantities that tell you how a quantum field theory's couplings and operators change with the energy scale at which you probe them, and getting them right at three loops is what modern precision phenomenology demands. The obstacle is that the standard trick for isolating ultraviolet divergences — expand in external momenta and masses before integrating — creates fake infrared divergences that have to be regulated away, and doing so by hand is the hard part of every such calculation. The paper's proposal is to put one and the same artificial mass $m$ into every propagator denominator in every diagram, which by construction blocks any infrared singularity, and then to show that the leftover integrals are all massive tadpoles with no external momenta at all. The technical payload is a closed set of formulas, culminating in eqn (36), that gives the pole parts of all the nontrivial three-loop scalar integrals in terms of a large-momentum expansion of their two-loop subintegrals. If the construction holds, three-loop renormalization constants in any theory with standard kinetic terms become a mechanical computer-algebra exercise rather than a bespoke ordeal.","feed_headline":"Three-loop beta functions reduced to massive tadpoles","feed_subtitle":"One artificial mass tames infrared divergences, turning any three-loop counterterm into a sum of scalar tadpoles.","key_machinery":"The load-bearing device is the propagator decomposition of eqn (1) and eqn (2): each propagator is split into terms whose denominators involve only the loop momentum and the common mass $m$, ordered so that successive terms lower the overall degree of divergence by one unit. Because each term separately obeys the criteria used in Weinberg's theorem, the degree-of-divergence bookkeeping survives term by term, and the final remainder can be discarded once the theory's operator dimension is bounded. The second device is the evaluation of the surviving three-loop tadpoles: a two-loop subintegral is isolated, its behaviour at large external momentum $q^2$ is expanded as a sum of powers $(m^2/q^2)^r$ and $(m^2/q^2)^{r+epsilon}$ with coefficients $a$ and $b$ (appendix C), and the $q^2$ integration then converts that expansion into the pole part. Where the subintegral is itself divergent, its pole is split off and handled with lower-loop integrals, and where negative indices produce tensor numerators, symmetric traceless tensors are traded for scalar integrals in $4+2n-2epsilon$ dimensions via the relation of appendix D.","core_discovery":"The central claim is that the ultraviolet counterterms of a general local field theory, through three loops, can be computed from a single-parameter family of massive tadpole integrals, and that those integrals are all explicitly known. The route is an exact rewriting of each propagator, $1/[(q+p)^2-M^2]$, as a sum of simple terms depending only on the loop momentum and $m$, plus a remainder whose contribution to the degree of divergence is arbitrarily negative and which can therefore be dropped without touching the pole part of the Green's function after subdivergences are subtracted. Because every denominator carries the same $m$, no spurious infrared divergence can appear, and after expanding in the physical masses and external momenta only vacuum integrals with a common mass survive. The paper then evaluates them: the one-loop integral exactly, the two-loop integrals by an integration-by-parts recursion plus the constant $I^{(2)}_{111}$ involving $S_2 = -4/(9sqrt 3) int_0^{pi/3} dx ln(2 sin(x/2)) approx 0.2604341$, and the three-loop integrals by expanding a two-loop subdiagram at large external momentum. Among integrals with six positive indices only $I^{(3)}_{111111}$ is ultraviolet divergent, and it evaluates to $2zeta(3)/epsilon$ plus finite terms. The general result, eqn (36), covers the cases with vanishing and with negative indices and reduces the whole three-loop problem to the expansion coefficients $A$ and $B$ listed in appendix C.","pith_inferences":["The real economy of the scheme is that it converts a problem in diagram topology into a problem in series expansion: once the coefficients $A$ and $B$ of appendix C are tabulated, adding a new theory costs only the propagator expansion, not new integral technology.","The exclusion of theories with non-canonical kinetic terms, named in the text for the heavy quark effective theory, marks the natural boundary of the method; any extension would need a decomposition adapted to the different denominator structure rather than a new integral formula.","Because the artificial mass is unphysical, its independence of the final pole terms is a free consistency check, and the paper itself suggests keeping the cutoff $Lambda$ arbitrary to test the appendix C expressions against misprints.","The same tadpole machinery plausibly transfers to any setting where three-loop operator mixing must be tracked, so the practical bottleneck shifts from integration to the combinatorics of enumerating diagrams and their symmetry factors."],"forward_implications":["Any three-loop renormalization constant in a theory with canonical kinetic terms can be obtained by expanding propagators, reducing tensor integrals to scalars, and evaluating massive tadpoles, with no diagram-by-diagram infrared subtraction required.","The explicit three-loop scalar-integral formulas make the calculation programmable: the paper reports a Mathematica implementation performing the tensor reduction up to three loops for an arbitrary number of free Lorentz indices.","Because the artificial mass can be kept arbitrary throughout, the $m^2$-dependent pieces need not be computed at all; they are absorbed into a small number of local counterterms, such as the gluon-mass-like term of eqn (3) with coefficient $Z_x = -g^2/(16pi^2 epsilon)(N+2f)$ at one loop.","The paper supplies general relations, eqns (40) and (45), between renormalization constants and the beta function or anomalous-dimension matrix that remain valid when the finite parts of the constants are kept, which matters for effective theories where evanescent operators appear.","The authors note the method should extend to four loops, where the analogous task becomes expanding a three-loop propagator-type integral at large momentum, for which computer-algebra tools already exist."],"supporting_citations":[{"why":"Supplies the fact that in the MS scheme ultraviolet counterterms are polynomial in momenta and masses, which licenses expanding the integrand before integrating.","marker":"[1]"},{"why":"Provides the large-$q^2$ expansion of the two-loop wave-function integral that yields the $6zeta(3)$ coefficient and hence the $2zeta(3)/epsilon$ pole of $I^{(3)}_{111111}$.","marker":"[3]"},{"why":"Gives the general two-loop integral with one massless and two massive lines, the basis of the reducible-case formula in appendix B.","marker":"[7]"},{"why":"Introduces the integration-by-parts method from which the two-loop recursion relation of eqn (10) is derived.","marker":"[10]"},{"why":"Supplies the relations between powers of scalar products and symmetric traceless tensors used to handle negative indices in the three-loop case C.2.","marker":"[11]"},{"why":"Source of the two-loop recursion relation and of the value of $I^{(2)}_{111}$ in terms of the constant $S_2$.","marker":"[21]"},{"why":"Its eqns (18) and (A.1) are the starting point for the explicit large-momentum coefficients $A$ and $B$ in appendix C.","marker":"[22]"},{"why":"Gives the relation between one-loop tensor integrals and scalar integrals in shifted dimensions that underpins appendix D and the general formula (36).","marker":"[24]"}],"fun_headline_variants":["Three-loop beta functions from one-family massive tadpoles","Infrared rearrangement by a common mass, up to three loops","Every three-loop scalar integral you need, written out","UV counterterms of any local theory through three loops","Tadpole recursion takes beta functions to three loops"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The master formula (36) is free of spurious $1/epsilon$ poles only if the large-momentum expansion coefficients satisfy $a^{mathrm{ren}}(k_1,k_2,r)+b(k_1,k_2,r)=O(1)$ for every index triple, a cancellation the paper says 'requires a short calculation' but does not display.","fun_headline_variants_meta":{"raw":{"variants":["Three-loop beta functions from one-family massive tadpoles","Infrared rearrangement by a common mass, up to three loops","Every three-loop scalar integral you need, written out","UV counterterms of any local theory through three loops","Tadpole recursion takes beta functions to three loops"]},"model":"selfhosted/deepseek-v4.1-flash","effort":"low","attempts":1,"cost_usd":0.0,"raw_usage":{"total_tokens":955,"prompt_tokens":889,"completion_tokens":66,"prompt_tokens_details":null,"completion_tokens_details":{"reasoning_tokens":0}},"tokens_in":889,"tokens_out":66,"duration_ms":1138,"image_pages":[],"temperature":1.0,"reasoning_budget":50,"reasoning_tokens":null,"transport_provider":"selfhosted","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-10-29T18:57:03.749409+00:00","model_set":{"reader":"selfhosted/deepseek-v4.1-flash"},"falsifier":"Recompute $a^{mathrm{ren}}(k_1,k_2,r)+b(k_1,k_2,r)$ to order $epsilon^0$ from the appendix C expressions for a few index triples and confirm the cancellation, then evaluate one nontrivial integral, such as $I^{(3)}_{211100}$ or $I^{(3)}_{11111-1}$, by an independent route and compare with eqn (36); the paper's own suggestion of keeping the cutoff $Lambda$ arbitrary and checking that the pole is $Lambda$-independent provides a second, self-contained test.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Its eqns (18) and (A.1) are the starting point for the explicit large-momentum coefficients $A$ and $B$ in appendix C."},{"cited_title":"Collins, Nucl","cited_arxiv_id":null,"evidence_quote":"Supplies the fact that in the MS scheme ultraviolet counterterms are polynomial in momenta and masses, which licenses expanding the integrand before integrating."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the large-$q^2$ expansion of the two-loop wave-function integral that yields the $6zeta(3)$ coefficient and hence the $2zeta(3)/epsilon$ pole of $I^{(3)}_{111111}$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the general two-loop integral with one massless and two massive lines, the basis of the reducible-case formula in appendix B."},{"cited_title":"Tkachov, Phys","cited_arxiv_id":null,"evidence_quote":"Introduces the integration-by-parts method from which the two-loop recursion relation of eqn (10) is derived."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the relations between powers of scalar products and symmetric traceless tensors used to handle negative indices in the three-loop case C.2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source of the two-loop recursion relation and of the value of $I^{(2)}_{111}$ in terms of the constant $S_2$."},{"cited_title":"Davydychev, Phys","cited_arxiv_id":null,"evidence_quote":"Gives the relation between one-loop tensor integrals and scalar integrals in shifted dimensions that underpins appendix D and the general formula (36)."}],"review_version":1}