Multiloop Amplitudes and Resummation

This map compares workflows for perturbative amplitudes and infrared-safe observables when tree-level information is inadequate: multiloop integrands, master integrals, real-emission cancellation, factorization, and resummation. No method wins in the abstract. The right chain depends on particle masses, multiplicity, color structure, kinematics, desired analytic form, observable inclusiveness, and target accuracy.

Evidence cutoff. 11 August 2026.

Required background. IBP identities and master integrals supplies integral reduction, differential equations supplies one solution strategy, and generalized unitarity supplies on-shell integrand construction.

Helpful background. Fixed-order matching and resummation defines accuracy bookkeeping, Sudakov resummation supplies the logarithmic mechanism, and Landau equations constrains singularities and continuation.

Match amplitude methods to the bottleneck

Stage	Approach	Required inputs and favorable domain	Choices that control the answer
amplitude generation	Feynman diagrams	Lagrangian rules; moderate multiplicity; transparent gauge/renormalization bookkeeping	gauge, diagram basis, tensor reduction, counterterm convention
amplitude generation	generalized unitarity/integrand reduction	tree amplitudes and cut-complete ansatz; especially effective with many external states	state sums, dimension, surface terms, rational pieces, color representation
integral reduction	IBP/Laporta and finite-field reconstruction	declared integral families and dimensional regulator	family completion, ordering, master basis, sampling primes and reconstruction bounds
integration	differential equations	master basis, kinematic variables, boundary data	canonical or noncanonical basis, path and sheet, special-function representation
integration	sector decomposition/direct numerics	numerical kinematic points, including difficult masses/scales	contour deformation, sector strategy, precision, Monte Carlo/quasi-Monte Carlo
real-radiation cancellation	local subtraction	universal unresolved limits and mapped phase space	subtraction scheme, momentum map, integrated counterterms
real-radiation cancellation	slicing	factorization below a resolution variable	slicing variable, power-correction fit, extrapolation range
large logarithms	direct-QCD or EFT resummation	a proven factorization theorem and scale hierarchy	logarithmic counting, profile scales, matching, nonperturbative model

Diagrammatic and unitarity constructions should agree after integration and renormalization when they cover the same theory and regulator. They are not fully independent if both feed the same reduction and integral library. Generalized unitarity reconstructs terms visible to the selected cuts; completeness requires control of contact/surface terms and the dimensional information needed for rational contributions Bern et al. 1994, foundational method. For integral reduction, Laporta’s ordering-and-elimination strategy turns integration-by-parts relations into a finite algorithm for a declared family Laporta 2000, foundational reduction.

Amplitude and resummation error propagation

A realistic prediction error has at least four layers:

$$\delta O=\delta_{\mathrm{analytic\ construction}} +\delta_{\mathrm{numerical\ evaluation}} +\delta_{\mathrm{perturbative/factorization}} +\delta_{\mathrm{external\ inputs}}.$$

The first includes incomplete integral families, cut ansätze, IBP reconstruction, boundary constants, and analytic continuation. The second includes floating-point cancellation, integration variance, contour deformation, interpolation, and phase-space integration. The third includes missing fixed orders, missing logarithmic orders, matching ambiguity, power corrections, and possible factorization violation. The fourth includes couplings, masses, parton distributions, fragmentation, and experimental definitions.

Scale variation samples a designed set of perturbative deformations; it is neither a probability distribution nor a bound. Correlated hard, factorization, and resummation scales must be varied in a way that preserves hierarchy while exposing missing terms. For slicing, the limit should be fit over multiple small cut values with an expected power/log structure; a plateau at the available precision can hide a common bias. Finite-field reconstruction needs enough independent samples and verification primes to make a wrong rational reconstruction improbable and detectable.

Checks and known failure modes

Amplitude level. Check gauge-parameter or reference-vector independence, ultraviolet poles and counterterms, universal soft/collinear limits, factorization channels, crossing, known helicity limits, and high-precision numerical points. A compact analytic expression can still lie on the wrong branch.

Integral level. Verify IBP identities on held-out points, compare master counts across algorithms, solve differential equations around more than one boundary, track the $i0$ prescription, and compare analytic values with sector-decomposition numerics. Canonical differential equations simplify polylogarithmic sectors Henn 2013, method, but elliptic or more general periods need broader function spaces; forcing them into a polylogarithmic ansatz is a failure mode.

Observable level. Check local cancellation in every unresolved limit, pole cancellation after integration, cutoff independence, reproduction of the fixed-order expansion of the resummed result, normalization/unitarity constraints, and agreement between distinct subtraction or slicing schemes. Catani–Seymour dipoles provide a benchmark construction whose integrated counterterms cancel next-to-leading-order infrared poles Catani and Seymour 1997, subtraction benchmark. For recursively infrared-and-collinear-safe global event shapes, automated resummation supplies a benchmark for logarithmic bookkeeping Banfi, Salam, and Zanderighi 2005, resummation method. Non-global observables can generate logarithms not captured by a simple global factorization. Glauber exchange can obstruct standard factorization in specified hadronic measurements.

Benchmarks and independence

Good held-out tests include massless four- and five-point amplitudes with known analytic results; massive sunrise and elliptic sectors; Drell–Yan/Higgs inclusive and transverse-momentum spectra; event shapes; and a fiducial process computed with two independent real-radiation methods. Phase-space mappings should reproduce analytic volumes and integrable singular limits, while continuation checks should track contour deformations, branch cuts, and discontinuities against known one-loop and sunrise examples.

Agreement is strongest when workflows differ in construction, reduction, integration, and infrared treatment. Two public codes can share master integrals, subtraction kernels, PDF evolution, or original matrix elements; document these lineages before calling the result independent. A numerical benchmark should publish parameter points, precision goals, regulator/cut values, versions, and enough digits to discriminate implementations.

What the methods cannot establish

A high-loop amplitude is not automatically an observable, and a small fixed-order scale band does not prove convergence. Resummation cannot determine power corrections it factorizes away. Unitarity and analytic structure constrain but do not alone select a UV completion. Numerical sector decomposition can validate points without yielding global analytic continuation. Agreement in Euclidean or unphysical kinematics does not certify the physical sheet.

Decision aid

Need	Prefer	Require before trusting
many external states with accessible tree building blocks	unitarity/integrand methods	cut completeness and dimensional rational terms
transparent renormalization in a moderate process	diagrams plus algebraic reduction	gauge and counterterm checks
many-scale analytic families	IBP plus differential equations	boundary constants and sheet tracking
isolated physical points or nonpolylogarithmic sectors	numerical integration, possibly with differential transport	contour and independent high-precision validation
fully differential infrared cancellation	local subtraction	pointwise unresolved-limit tests
high-order inclusive result with a clean resolution variable	slicing	controlled zero-cut extrapolation
logarithmically enhanced region	proved factorization plus resummation and matching	expansion back to fixed order and profile stability

Source selection and related pages

The finite search covered INSPIRE, arXiv, journal/DOI records, software papers, and citation chaining through 11 August 2026, including targeted searches for factorization and analytic-continuation failures. Reassess when a method demonstrates a new mass/multiplicity/function-class domain with public benchmarks, when an independent reproduction fails, or when a subtraction/factorization revision changes a precision result.

See the amplitudes field guide, infrared-complete observables, color–kinematics scope, and the scattering phenomenology pathway.

References

• A. Banfi, G. P. Salam, and G. Zanderighi, “Principles of General Final-State Resummation and Automated Implementation,” JHEP 03 (2005) 073. arXiv.
• Z. Bern, L. Dixon, D. C. Dunbar, and D. A. Kosower, “One-Loop $n$-Point Gauge Theory Amplitudes, Unitarity and Collinear Limits,” Nuclear Physics B 425 (1994) 217–260. DOI.
• S. Catani and M. H. Seymour, “A General Algorithm for Calculating Jet Cross Sections in NLO QCD,” Nuclear Physics B 485 (1997) 291–419; erratum 510 (1998) 503. DOI.
• J. M. Henn, “Multiloop Integrals in Dimensional Regularization Made Simple,” Physical Review Letters 110 (2013) 251601. DOI.
• S. Laporta, “High-Precision Calculation of Multiloop Feynman Integrals by Difference Equations,” International Journal of Modern Physics A 15 (2000) 5087–5159. DOI.