Loop Integrals and Reduction

Loop calculations turn the compact statement “integrate over every virtual momentum” into a controlled map from a Feynman graph to an analytic function. The essential choices are the integral family, the boundary prescription, the regulator, the reduction basis, and the continuation path. This chapter develops those choices in that order and keeps ultraviolet subtraction, infrared cancellation, and phenomenological assembly at their proper downstream interfaces.

Enter the loop calculation

A loop integral is not specified by its topology alone. Its propagator powers, numerator, loop-momentum routing, masses, external invariants, measure normalization, and $i0$ prescription are part of its definition. Dimensional regularization then supplies a translation-invariant setting in which algebraic reduction is reliable and singular regions appear as Laurent poles. Parameter representations expose denominator geometry; integration-by-parts identities and differential equations turn large integral families into finite boundary-value problems.

The progression from regulated definitions through reduction and differential equations is developed systematically in Abreu, Britto, and Duhr 2022, §§1–3, pp. 3–34 and Weinzierl 2022, §§2.4.2–2.5.3, pp. 28–55; §6.1, pp. 157–162.

Choose an entry according to the calculation in front of you:

Goal	Start here	What you should leave with
Decode or define an integral	Anatomy of a Loop Integral	A complete integral specification and a first UV, IR, and threshold diagnosis
Combine denominators	Feynman and Schwinger Parameters	A parameter-domain representation with its branch prescription intact
Dimensionally regulate an amplitude	Dimensional Regularization as an Amplitude Tool	Consistent $d=4-2\epsilon$ normalization and a Laurent expansion whose regions remain identifiable, together with warnings about scaleless and rapidity-sensitive cases
Prepare a result for renormalization	UV/IR Poles and the Renormalized-Amplitude Interface	A precise separation of bare, regulated, UV, IR, and finite information
Evaluate a one-loop graph analytically	One-Loop Integral Families and Analytic Functions	Tadpole-to-box organization, thresholds, logarithms, and dilogarithms
Remove loop momenta from numerators	Tensor Reduction	Lorentz form factors expressed through scalar families, with Gram-sensitive cases identified
Expand a hierarchy of scales	Expansion by Regions	A scaling declaration, region contributions, overlap control, and an independent check
Reduce a multiloop family	Integration-by-Parts Identities and Master Integrals	Linear relations and a finite master basis
Determine masters from kinematics	Differential Equations for Master Integrals	A matrix differential system, boundary data, and controlled analytic continuation
Produce a trustworthy number	Numerical Evaluation and Validation of Loop Integrals	Error estimates and independent analytic, representation, and limit checks

The progression follows a useful dependency chain:

$$\text{defined integral} \longrightarrow \text{regulated family} \longrightarrow \text{reduced masters} \longrightarrow \text{boundary data and continuation} \longrightarrow \text{validated value}.$$

Skipping a link is sometimes efficient, but never erase its data. For example, a numerical sector-decomposition value is not reproducible unless the measure, scale convention, kinematic point, prescription, requested Laurent coefficients, and error criterion are all known.

What is proved here—and what is handed onward

The chapter treats loop integrals and regulated amplitudes. It explains how UV and IR singularities arise and how their poles are labeled, but it does not choose counterterms, a subtraction scheme, running parameters, operator mixing, or matching coefficients; those belong to renormalization and effective-field-theory analysis. It identifies soft, collinear, potential, and threshold scalings, but a complete factorization theorem and zero-bin or rapidity-subtraction machinery require additional physical input.

Likewise, finding a branch point is not yet computing its physical discontinuity. Singularities, Cuts, and Integrand Reconstruction separates candidate contour pinches, physical cuts, generalized on-shell constraints, and maximal residues. Observable-level cancellation and phase-space integration occur later, after virtual and real contributions have been assembled consistently.

A representative exit test

Take a massive scalar bubble. A successful chapter-level calculation can:

1. state both denominators and the $i0$ prescription;
2. derive its Feynman-parameter polynomial;
3. continue it to $d=4-2\epsilon$ with an explicit scale convention;
4. locate its threshold at $s=(m_1+m_2)^2$;
5. reduce any numerator insertion to scalar integrals;
6. obtain the $\epsilon^0$ Laurent coefficient in the declared normalization, either analytically or from a differential equation; and
7. reproduce a sample value through an independent numerical representation with a stated uncertainty.

The point is not to privilege one technique. Agreement between methods tests different failure modes: algebraic reduction, branch choice, boundary constants, and numerical convergence.

For the equal-mass case, one compact thread runs through the chapter. Parameterization gives $\Delta(x)=m^2-sx(1-x)$; dimensional continuation isolates the momentum-independent UV pole; IBP reduces a doubled propagator to the bubble and tadpole masters; the master differential equation has singular points at $s=0$, $4m^2$, and infinity; regularity fixes the $s=0$ boundary; and continuation from the Euclidean region gives $\operatorname{Im}F(s)=\pi\sqrt{1-4m^2/s}$ above threshold. A numerical implementation should reproduce both the differential residual and this phase-space sign. Each step checks a different piece of the same integral definition.

Review the chapter

Use the questions below to identify topics worth revisiting.

• Which pieces of a loop integral change under a loop-momentum rerouting, and why is the final answer routing-independent only in an appropriate regulator?
• How would you decide whether a $1/\epsilon$ pole is ultraviolet or infrared when the same regulator controls both?
• Why can a canonical differential equation simplify integration without determining its boundary constants?
• What extra information must accompany a numerical value above a physical threshold?

Concise answers should mention translation invariance, region or off-shell diagnostics, boundary data, and the $i0$-selected sheet, respectively.

References

• Abreu, Samuel, Ruth Britto, and Claude Duhr. “The SAGEX Review on Scattering Amplitudes, Chapter 3: Mathematical Structures in Feynman Integrals.” Journal of Physics A: Mathematical and Theoretical 55 (2022): 443004. doi:10.1088/1751-8121/ac87de.
• Weinzierl, Stefan. Feynman Integrals. Cham: Springer, 2022. doi:10.1007/978-3-030-99558-4. Open PDF.