Feynman Residues in Quantum Field Theory

• Feynman residues are multivariate localizations at singular loci in Feynman integrals, capturing on-shell cuts, higher-order poles, and spectral weight contributions.
• They facilitate extraction of master-integral coefficients and enable cancellation of problematic higher-order poles through on-shell projections and counterterm design.
• Residue techniques unite diverse formulations—from Mellin–Barnes and Grassmannian approaches to renormalization residues—revealing deep analytic, geometric, and arithmetic structures.

“Feynman residues” denotes several related constructions attached to singularities of Feynman integrals, propagators, and amplitude representations. In perturbative quantum field theory, the term can refer to residues of loop-energy poles that define cuts, residues of higher-order poles produced by raised propagators, residues of propagator poles that carry spectral weight or LSZ normalization data, and residue-type contributions appearing in Mellin–Barnes, Grassmannian, twisted-cohomology, or configuration-space formulations (Abreu et al., 2017, Baumeister et al., 2019, Hirata et al., 2023, Lewandowski, 2017). The common principle is localization on singular loci—on-shell hypersurfaces, complex poles, Landau varieties, or exceptional divisors—while the operational meaning depends on the framework.

1. Cut residues and multivariate on-shell localization

A precise residue-theoretic definition of a cut Feynman integral is given at one loop by interpreting a cut as a multivariate residue of the loop integrand on the locus where selected propagators are put on shell (Abreu et al., 2017). For a one-loop scalar integral

$$I_n^D=\int \omega_n^D,\qquad \omega_n^D= \frac{e^{\gamma_E\epsilon}}{i\pi^{D/2}} \frac{d^Dk}{D_1\cdots D_n},$$

with propagators $D_j=(k-q_j)^2-m_j^2+i0$, and a cut set $C\subseteq\{1,\dots,n\}$, the composed residue is

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$

where $S_i=\{D_i=0\}$. The associated cut integral is expressed as

$$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$

Here $\delta_C$ is the iterated Leray coboundary, and the cut contour is the tubular cycle around the on-shell variety.

This residue definition makes the geometry of the cut explicit. The cut variety $S_C=\bigcap_{j\in C}S_j$ is a sphere $S_C\cong S^{D-c}$, and the cut contour is its Leray coboundary (Abreu et al., 2017). First-type Landau singularities arise at finite loop momentum and are characterized by the vanishing of the modified Cayley determinant $Y_C$. Second-type singularities are associated with pinches at infinity and are controlled by Gram determinants in the compactified description. A notable consequence is that cuts associated with second-type singularities can be written as specific combinations of ordinary cuts, leading to linear relations among distinct cuts of the same integral.

Multidimensional residues also provide a direct method for extracting master-integral coefficients when propagators occur with generic powers (Zhang, 2011). For

$D_j=(k-q_j)^2-m_j^2+i0$0

the local residue at an isolated common zero $D_j=(k-q_j)^2-m_j^2+i0$1 is

$D_j=(k-q_j)^2-m_j^2+i0$2

When $D_j=(k-q_j)^2-m_j^2+i0$3, the residue becomes a derivative: $D_j=(k-q_j)^2-m_j^2+i0$4 At one loop this allows direct extraction of the scalar box coefficient from a quadruple cut, bypassing iterative IBP reduction (Zhang, 2011).

2. Higher-order poles from raised propagators

Starting from two loops, self-energy insertions on internal lines generate propagators raised to higher powers, so the integrand develops higher-order poles (Baumeister et al., 2019). If an internal line is represented by

$D_j=(k-q_j)^2-m_j^2+i0$5

a self-energy insertion produces factors such as $D_j=(k-q_j)^2-m_j^2+i0$6, and at higher loops even higher powers can appear. Analytically, such integrals can be reduced by IBP identities, but in numerical approaches the residue of the higher-order pole becomes a direct computational obstacle.

The standard residue formula for a pole of order $D_j=(k-q_j)^2-m_j^2+i0$7 is

$D_j=(k-q_j)^2-m_j^2+i0$8

For $D_j=(k-q_j)^2-m_j^2+i0$9, derivatives of the full integrand are required, and in the multivariate case the problem becomes still more cumbersome (Baumeister et al., 2019). This is precisely the situation encountered in loop-tree duality and numerical unitarity when a raised propagator goes on shell.

For renormalised amplitudes in the on-shell scheme, the problematic residue can be made to vanish already at the integrand level by combining the self-energy graph with a suitable counterterm graph (Baumeister et al., 2019). The key device is the on-shell projection

$C\subseteq\{1,\dots,n\}$0

together with “flattened” denominators such as

$C\subseteq\{1,\dots,n\}$1

The loop integrand and its counterterm are engineered so that their sum vanishes quadratically as the raised propagator goes on shell: $C\subseteq\{1,\dots,n\}$2 In that situation the double pole and the accompanying subleading pole cancel, and the residue of the raised propagator vanishes (Baumeister et al., 2019).

The scope of the construction is scheme dependent. The quadratic on-shell cancellation is tied to the on-shell renormalisation scheme and does not hold in $C\subseteq\{1,\dots,n\}$3, because finite scheme-changing terms remain nonzero in the on-shell limit (Baumeister et al., 2019). The same logic extends to quark and gluon self-energies in QCD, although for massive quarks and gluons some terms vanish only linearly or are proportional to $C\subseteq\{1,\dots,n\}$4 or $C\subseteq\{1,\dots,n\}$5; these terms are harmless when contracted with physical gauge-invariant quantities.

3. Nested residues, causal denominators, and Landau geometry

In the Loop-Tree Duality framework, multiloop contour integration is organized into iterated residues in the loop-energy variables (Aguilera-Verdugo et al., 2020). For a generic rational integrand with quadratic denominators,

$C\subseteq\{1,\dots,n\}$6

the primitive variables are complexified successively, not simultaneously. This produces residue chains with an important dichotomy: displaced poles cancel, while the surviving terms are the nested residues.

The paper proves that contributions from displaced poles cancel pairwise (Aguilera-Verdugo et al., 2020). Those poles are generated when the position of a pole in one variable depends on the residue already taken in another variable. After explicit Laurent expansion, the relevant iterated residues have equal magnitude and opposite sign. The surviving nested residues encode the physically relevant information and are naturally mapped onto nondisjoint on-shell states. A central structural consequence is that unphysical singularities vanish, and the final expressions can be written using only causal denominators.

This cancellation pattern is closely related to Landau geometry. At one loop, first-type singularities are tied to finite-momentum pinch configurations and second-type singularities to pinch configurations at infinity (Abreu et al., 2017). In a more algebraic formulation of integral reduction, the maximal-cut Landau locus in the Baikov representation,

$C\subseteq\{1,\dots,n\}$7

controls the critical syzygies used in reduction (Coro et al., 5 Dec 2025). The decomposition

$C\subseteq\{1,\dots,n\}$8

expresses critical syzygies as a sum over irreducible components of the Landau locus. The paper describes the associated saturation and ideal-quotient operations as residue-like, in the sense that they remove unwanted singular contributions and retain the components relevant for reduction (Coro et al., 5 Dec 2025).

A plausible implication is that several apparently distinct residue constructions—cut residues, nested LTD residues, and Landau-controlled syzygies—are different localizations of the same singular-support data. The supplied literature does not collapse these viewpoints into a single formalism, but it repeatedly identifies the singular locus as the organizing object.

4. Propagator-pole residues, spectral weights, and LSZ factors

In many-body Green’s function theory, the residue at a propagator pole is the spectral weight carried by that pole (Hirata et al., 2023). The exact one-particle Green’s function satisfies Dyson’s equation

$C\subseteq\{1,\dots,n\}$9

or equivalently

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$0

If $$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$1 is a pole, the residue is

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$2

and in the diagonal approximation

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$3

For the exact Green’s function, these residues satisfy the sum rules quoted in the paper (Hirata et al., 2023).

Finite-order Feynman–Dyson perturbation theory does not preserve this structure uniformly (Hirata et al., 2023). The second-order self-energy is described as mostly physical: principal roots have residues close to $$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$4, while satellite roots have near-zero residues. At odd perturbation orders, however, the self-energy can acquire the wrong concave or convex shape within a frequency bracket, yielding complex roots or real roots with residues outside the physical range $$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$5. At higher even orders, numerous phantom poles can appear with essentially zero residues. The paper attributes the nonconvergence to the nonanalyticity of the rational-function exact Green’s function at many frequencies, and argues that Padé approximants can largely restore the correct pole and residue structure (Hirata et al., 2023).

A distinct but related notion appears in LSZ reduction for mixed propagators. For systems with non-diagonal fermionic or scalar propagators, the pole part factorizes at each complex pole (Lewandowski, 2017). For scalars,

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$6

while for Majorana fermions

$$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$7

The matrices $$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$8 and $$\mathrm{Res}_C[\omega_n^D]=\mathrm{Res}_{S_1\cdots S_c}[\omega_n^D],$$9 are the “square-rooted” residues. The paper gives all-orders prescriptions for these factors in arbitrary renormalization schemes for Majorana fermions, Dirac fermions, generic mixing fermions, and scalars (Lewandowski, 2017). In the stable-particle case they match the usual LSZ wave-function normalization logic; for unstable particles they remain useful for defining effective couplings and resonant amplitudes.

5. Hypergeometric, Grassmannian, and Yangian residue structures

Residues also organize exact function spaces of Feynman integrals. In conformal integral bootstrap, the $S_i=\{D_i=0\}$0-dimensional box integral with generic propagator powers is fixed by Yangian symmetry to a specific linear combination of Appell $S_i=\{D_i=0\}$1 functions (Loebbert et al., 2019). In Mellin–Barnes form,

$S_i=\{D_i=0\}$2

with

$S_i=\{D_i=0\}$3

Closing the contour in compatible cones yields

$S_i=\{D_i=0\}$4

In cone $S_i=\{D_i=0\}$5, the fundamental residue reproduces the Appell $S_i=\{D_i=0\}$6 coefficient structure exactly (Loebbert et al., 2019). The same paper argues that Yangian invariance and Mellin–Barnes residue calculus are two descriptions of the same analytic structure: the Yangian PDEs determine the coefficient system, while the Mellin–Barnes contour expresses those coefficients as sums over residues.

In four-dimensional undeformed kinematics, the conformal box collapses to the Bloch–Wigner function (Loebbert et al., 2019). In this sense, the residue sum is not merely a computational tool but an analytic mechanism selecting the physically relevant branch from a larger solution space.

A different residue geometry appears in the Grassmannian formulation of scattering amplitudes. Leading singularities are obtained as residues of the Grassmannian integral over $S_i=\{D_i=0\}$7, after enough minors are set to zero to localize the integral (Zhang et al., 2010). The localization dimension is

$S_i=\{D_i=0\}$8

Composite residues arise when vanishing of one minor forces further factorization. The coordinates $S_i=\{D_i=0\}$9 introduced in the paper place composite and non-composite residues on equal footing and make residue theorems more uniform (Zhang et al., 2010).

For generalized biadjoint amplitudes $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$0, certain $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$1-dimensional residues are even larger objects: they reproduce the entire ordinary biadjoint partial amplitude $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$2 rather than a single diagram (Cachazo et al., 2022). The main theorem states

$$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$3

In the $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$4 example, $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$5 has $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$6 generalized Feynman-diagram terms, but only $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$7 survive at the chosen residue, and $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$8 is exactly the number of planar cubic diagrams in $$\mathcal{C}_C I_n = (2\pi i)^{-\lceil c/2\rceil}\int_{\Gamma_C}\omega_n^D, \qquad \Gamma_C=\delta_C S_\bot.$$9 (Cachazo et al., 2022). Here “Feynman residue” denotes a residue whose value is a full tree-level amplitude for fixed planar ordering.

6. Renormalization residues, higher residue pairings, and period invariants

In configuration-space renormalization, residues appear as Poincaré residues of the pulled-back Feynman form on a wonderful compactification of the graph configuration space (Ceyhan et al., 2010). For a local form

$\delta_C$0

the Poincaré residue along $\delta_C$1 is

$\delta_C$2

If $\delta_C$3 is the exceptional divisor associated with a divergent subgraph $\delta_C$4, regularization is achieved by replacing the singular part of the real integration cycle with a Leray coboundary, and the ambiguity is measured by

$\delta_C$5

For primitive logarithmically divergent graphs, the pulled-back form has a simple pole along the deepest exceptional divisor; with logarithmic subdivergences, iterated residues appear on intersections of exceptional divisors indexed by nests of divergent subgraphs (Ceyhan et al., 2010).

A formally different but conceptually adjacent construction is provided by higher residue pairings in the twisted-cohomology description of Feynman integrals (Mizera et al., 2019). After Schwinger parametrization, a generic $\delta_C$6-loop integral takes the form

$\delta_C$7

with twisted differential

$\delta_C$8

The relevant intersection number has an $\delta_C$9 expansion

$S_C=\bigcap_{j\in C}S_j$0

whose coefficients are Saito’s higher residue pairings. The leading term is the ordinary Grothendieck residue

$S_C=\bigcap_{j\in C}S_j$1

Because the connection matrix is polynomial in $S_C=\bigcap_{j\in C}S_j$2, finitely many higher residue pairings suffice to reconstruct the exact differential equations near four dimensions (Mizera et al., 2019).

Residue language also appears in the arithmetic side of perturbation theory. Feynman periods are the renormalization-group independent parts of logarithmically divergent, subdivergence-free graphs and are equivalently the $S_C=\bigcap_{j\in C}S_j$3 residues of those graphs in dimensional regularization (Borinsky et al., 2022). The period is defined by

$S_C=\bigcap_{j\in C}S_j$4

Using graphical functions and conformal four-point integrals, all subdivergence-free Feynman periods in $S_C=\bigcap_{j\in C}S_j$5 theory up to six loops and $S_C=\bigcap_{j\in C}S_j$6 of $S_C=\bigcap_{j\in C}S_j$7 Feynman periods at seven loops were computed (Borinsky et al., 2022). In $S_C=\bigcap_{j\in C}S_j$8 graph theory, the extended graph permanent was introduced as an infinite sequence of residues from prime order finite fields and shown to be preserved by completion/decompletion, planar duality, and the Schnetz twist (Crump, 2017). This suggests a broad residue hierarchy ranging from analytic pole coefficients to arithmetic graph invariants.

A recurring misconception is that “residue” always means a simple coefficient extracted from a one-variable pole. In the literature summarized here, residues can instead be multivariate Leray residues, higher-order derivatives at repeated poles, spectral weights of propagator roots, factorized LSZ wave-function data, Mellin–Barnes lattice sums, Poincaré residues on exceptional divisors, or $S_C=\bigcap_{j\in C}S_j$9 coefficients of renormalized amplitudes. The unifying content is not a single formula but the localization of Feynman-theoretic information on singular structures.