Generalised Splitting Amplitude in QCD

• Generalised splitting amplitudes are universal objects that describe the detailed collinear behavior of multi-parton systems with full spin, color, and loop corrections.
• They extend traditional 1→2 splitting models in perturbative QCD to multi-collinear and quasi-collinear regimes, enabling precise higher-loop and massive particle analyses.
• They also underpin exact factorizations in non-singular kinematic settings, linking traditional QCD approaches with modern split and CHY amplitude formulations.

Searching arXiv for recent and relevant papers on generalized/splitting amplitudes. I’m sorry, but I don’t have access to the arXiv search tool in this environment.

I can still write the article using the arXiv papers and detailed source material you provided, and I will cite them by arXiv ID. Generalised splitting amplitudes are amplitude-level objects that extend ordinary splitting amplitudes beyond the simplest collinear $1\!\to\!2$ limits. In perturbative QCD, they describe universal leading-power behaviour when two or more partons become collinear, with full spin correlations, color structure, loop dependence, and, in several modern formulations, massive or spacelike kinematics. In a distinct but related literature on particle, string, and CEGM amplitudes, “split” or “generalised splitting” refers to exact factorization on special non-singular loci in kinematic space, away from poles, into products of lower-point currents or simplex amplitudes (Guan et al., 2024, Badger et al., 2015, Henn et al., 2024, Cao et al., 2024, Umbert et al., 18 Jan 2025).

1. Definitions and scope

At tree level and in the ordinary two-particle collinear limit, a massless gauge-theory amplitude factorizes into a lower-point amplitude times a universal splitting amplitude. For a color-space amplitude, the amplitude-level statement is that when two external massless partons become collinear, the singular behaviour is captured by a universal operator or current depending on the collinear legs and, in the simplest timelike case, independent of the rest of the process. In color-ordered notation this is written schematically as

$$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$

while for $m$ collinear legs at $L$ loops one has

$$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$

This is the basic sense in which splitting amplitudes are universal amplitude-level building blocks (Badger et al., 2015, Duca et al., 2020).

The phrase “generalised splitting amplitude” is used in several extensions of this basic setup. In perturbative QCD it refers to more partons in the collinear cluster, higher loop order, helicity- and color-resolved amplitudes, full dependence on the dimensional regulator $\epsilon$, massive quasi-collinear kinematics, and spacelike configurations with non-factorizing color correlations. In a separate line of work, it refers to “2-split” or “smooth splitting” factorizations on special subspaces where amplitudes factorize exactly away from poles, and in the CEGM formalism it refers to factorization into simplex amplitudes on a split kinematic subspace (Guan et al., 2024, Dhani et al., 2023, Nardi et al., 2024, Cao et al., 2024, Umbert et al., 18 Jan 2025).

Setting	Defining feature	Representative papers
Perturbative QCD collinear limits	Universal amplitude-level behaviour in collinear or quasi-collinear limits	(Guan et al., 2024, Badger et al., 2015, Duca et al., 2020, Dhani et al., 2023)
Spacelike generalized splitting	Color operators depend on non-collinear legs; naive factorization is violated at amplitude level	(Henn et al., 2024, Duhr et al., 7 Jul 2025)
Split factorizations away from poles	Exact factorization on special kinematic loci into currents or simplex amplitudes	(Cao et al., 2024, Cao et al., 2024, Azevedo et al., 23 Dec 2025, Umbert et al., 18 Jan 2025)

2. Collinear factorization and amplitude-level splitting in QCD

For an $n$-parton massless QCD amplitude

$$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$

the two-parton collinear limit is described by introducing

$$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$

In a pseudo-Euclidean region and for timelike kinematics, the factorization formula is

$$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$

The splitting amplitudes admit the loop expansion

$$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$0

The three QCD $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$1 channels explicitly treated at three loops are

$$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$2

These are helicity-resolved amplitude-level objects; once squared and summed over colors and helicities they give the splitting functions that appear in DGLAP evolution and subtraction schemes (Guan et al., 2024).

The same logic extends to genuinely multi-collinear configurations. In the triple-collinear limit $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$3, one introduces momentum fractions $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$4 with $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$5, and the one-loop gluon-initiated triple-collinear splitting amplitudes $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$6 were extracted from Higgs-plus-four-parton amplitudes. The processes computed are $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$7 and $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$8, with primitive helicity amplitudes organized by supersymmetric decomposition. Universality was checked both at the amplitude level and numerically against full-color six-parton one-loop squared matrix elements, with a ratio $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$9 as $m$0 (Badger et al., 2015).

At tree level, the same generalized collinear programme extends to $m$1 splitting. The full gluon-parent set

$m$2

was computed in conventional dimensional regularisation. These $m$3 splitting amplitudes are the tree-level multi-collinear building blocks required for N$m$4LO infrared structure, and they satisfy nested-factorization relations in iterated collinear limits through splitting tensors $m$5 (Duca et al., 2020).

3. Higher-loop, massive, and low-virtuality generalizations

A central modern benchmark is the three-loop computation of two-parton QCD splitting amplitudes. In this setting, “generalised” means all relevant partonic channels and helicity configurations, full color dependence including quartic Casimirs, all orders in the dimensional regulator $m$6, and, in that work, up to three loops for arbitrary timelike $m$7 splittings, together with a controlled analytic continuation to spacelike $m$8 splittings for three-parton amplitudes. The scalar coefficient functions are written as

$m$9

with Laurent expansion in $L$0, rational functions of $L$1, and harmonic polylogarithms. In a pseudo-Euclidean region, an $L$2-loop splitting factor takes the form

$L$3

which isolates branch cuts and phases. At three loops the amplitudes contain poles up to $L$4, and these match the known all-order structure of infrared singularities (Guan et al., 2024).

The color structure at this order is fully non-planar. Besides $L$5, $L$6, and $L$7, the amplitudes depend on quartic invariants

$L$8

with explicit $L$9 formulae for $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$0, $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$1, and $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$2. The calculation introduces an expansion-by-subgraphs implementation of the method of regions at the integrand level, reduces the resulting collinear integrals to 553 master integrals, and solves them by differential equations in canonical form (Guan et al., 2024).

Massive generalizations are naturally formulated in quasi-collinear kinematics. For triple-collinear splitting with massive particles, one defines an on-shell parent momentum $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$3 by

$$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$4

and a Sudakov decomposition

$$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$5

In the quasi-collinear limit,

$$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$6

the amplitude factorizes as

$$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$7

The resulting tree-level $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$8 kernels with heavy quarks extend the massless Catani–Grazzini structure to quasi-collinear kinematics and provide NNLO subtraction ingredients with explicit mass dependence (Dhani et al., 2023).

A different generalization replaces strict collinearity by low virtuality. In the Standard Model, when $$A_n^{(L)}(\{p_i^{\lambda_i}\}) \xrightarrow{1||\cdots||m} \sum_{k=0}^L \sum_{\lambda_P} \text{Split}_m^{(L-k)}(-P^{-\lambda_P};\{p_i^{\lambda_i}\}_{i=1}^m)\, A_{n-m+1}^{(k)}(P^{\lambda_P},\{p_i^{\lambda_i}\}_{i=m+1}^n).$$9 for an intermediate line, the amplitude factorizes as

$\epsilon$0

These low-virtuality splitting amplitudes describe the full kinematic regime, which includes the region of collinear splitting, of soft emission, and combinations thereof. They are represented as little-group tensors in an improved bi-spinor formalism for massive spin-1 particles that automatically incorporates the Goldstone Boson Equivalence Theorem, and in the soft limit they reproduce an eikonal formula

$\epsilon$1

This suggests a unified treatment of soft, collinear, and soft–collinear electroweak radiation at high energy (Nardi et al., 2024).

4. Spacelike generalized splitting and factorization breaking

The sharpest distinction between ordinary and generalized splitting arises in spacelike kinematics. For timelike splitting, strict collinear factorization holds: the splitting amplitude depends only on the collinear pair and the reduced amplitude is independent of the rest of the event. For spacelike splitting, where one collinear parton is incoming and the other outgoing, strict collinear factorization is violated beyond leading order. In color-space notation,

$\epsilon$2

but for spacelike kinematics $\epsilon$3 inevitably depends on the colors and kinematics of non-collinear partons (Henn et al., 2024).

In the two-loop spacelike splitting amplitude for full-color $\epsilon$4 super-Yang–Mills theory, the generalized splitting operator contains a one-loop dipole term proportional to

$\epsilon$5

and at two loops genuinely new tripole structures

$\epsilon$6

with $\epsilon$7 running over outgoing non-collinear legs. These terms are the amplitude-level manifestation of naive factorization violation. The paper also shows that factorization is restored at the level of color-summed, unpolarized, squared amplitudes at NNNLO: dipole contributions are pure phases that cancel upon squaring, and the tripole terms match soft-collinear structures rather than generating new process-dependent collinear poles. It further conjectures that the two-loop tripole terms in the generalized splitting amplitudes in QCD are identical to those obtained in $\epsilon$8 super-Yang–Mills theory (Henn et al., 2024).

The three-loop QCD computation provides a timelike benchmark together with a controlled analytic continuation to spacelike $\epsilon$9 splitting for three-parton amplitudes. Introducing $n$0, the continuation rule is

$n$1

after making all logarithms in $n$2 explicit. The resulting spacelike splitting factors are valid for three-parton amplitudes, while for generic multi-leg amplitudes additional non-factorizing color-correlated terms must be included (Guan et al., 2024).

A further generalization arises when two distinct spacelike collinear pairs are present. In planar $n$3 super-Yang–Mills theory, for

$n$4

with both parent virtualities spacelike, the amplitude behaves as

$n$5

The non-trivial factor $n$6 correlates the two collinear directions and cannot be written as a product of independent single-collinear functions. In the MHV sector it is equal, up to a phase, to the BDS-subtracted six-point amplitude in multi-Regge kinematics, and the same function governs amplitudes with up to 8 particles and 3 loops and form factors with up to 6 particles and 2 loops. This suggests that generalized spacelike splitting can organize multi-direction correlations at leading color (Duhr et al., 7 Jul 2025).

5. Split factorizations away from poles

A distinct usage of “generalised splitting amplitude” appears in tree-level particle and string amplitudes where factorization occurs on non-singular loci in kinematic space. The defining condition of 2-split kinematics is

$n$7

for a partition of external legs into sets $n$8, $n$9, and three distinguished labels $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$0. Under these conditions, the scattering potential and the CHY or string integration measure split into two independent sectors with off-shell legs $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$1, and amplitudes factorize into products of currents rather than residues on propagator poles. For Yang–Mills, the basic form is

$$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$2

with analogous formulas for bi-adjoint $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$3, NLSM, DBI, the special Galileon, Einstein gravity, and their open- and closed-string extensions (Cao et al., 2024, Cao et al., 2024).

This split behavior is not tied to any physical pole. It is a factorization on special codimension-$$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$4 loci in kinematic space. A subsequent on-shell reformulation shows that the off-shell CHY currents can be represented as ordinary on-shell amplitudes evaluated with a universal kinematic shift

$$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$5

so that, for example in bi-adjoint $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$6,

$$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$7

The same 2-split structure persists in higher-derivative theories such as $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$8, $$\mathcal{A}_n^{c_1 \dots c_n}(\{p_1,f_1,\lambda_1\},\dots,\{p_n,f_n,\lambda_n\}),$$9, and $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$0, and transmuting operators $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$1 and $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$2 map split amplitudes between theories while preserving the split structure (Azevedo et al., 23 Dec 2025).

The same theme appears in “all-order splits and multi-soft limits” for particle and string amplitudes. There the factorization

$$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$3

holds on non-singular loci determined by gluing of surfaces in binary geometry. This is again a generalized splitting in the sense that higher amplitudes factorize exactly into lower ones away from poles, and in suitable soft limits it yields all-loop multi-soft theorems for $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$4, NLSM, and scalar-scaffolded gluons (Arkani-Hamed et al., 2024).

In the CEGM formalism, the split kinematic subspace $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$5 sets all non-admissible generalized kinematic invariants to zero. The CEGM amplitude then splits into simplex amplitudes: $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$6 Each factor is the field-theory limit of a multivariate beta function and is the canonical form of a simplex. This suggests that, in the CEGM setting, generalized splitting is controlled by viewing positive moduli space as a product of simplices (Umbert et al., 18 Jan 2025).

A related recursive perspective derives smooth splitting and hidden zeros from linear kinematic shifts. On suitable near-zero loci, amplitudes obey

$$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$7

with higher-order and triple-product generalizations on more relaxed near-zero kinematics. This suggests that non-pole splitting relations can be understood from standard tree-level factorization on propagators together with improved UV behaviour under special shifts (Jones et al., 5 May 2025).

6. Applications, consistency checks, and outlook

In perturbative QCD, generalized splitting amplitudes are central ingredients of fixed-order infrared subtraction and parton evolution. The three-loop $$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$8 amplitudes provide the real triple-virtual ingredient needed for a fully exact N$$P=p_a+p_b,\qquad p_a\to z\,\tilde P,\qquad p_b\to (1-z)\,\tilde P,\qquad \tilde P^2=0.$$9LO determination of PDFs and factorization-scale evolution, and they are expected to be essential for N$$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$0LO subtraction schemes. They also enter beam and fragmentation functions, generalised factorization formulae for $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$1-jet cross sections at N$$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$2LO–N$$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$3LO, and high-logarithmic resummation. The same work verifies agreement with previous one- and two-loop splitting amplitudes, the expected IR/UV pole structure, known three-loop soft currents, and the leading-transcendental agreement with $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$4 super-Yang–Mills theory (Guan et al., 2024).

For triple-collinear amplitudes, universality was checked numerically against full-color six-parton one-loop squared matrix elements, and supersymmetric decomposition organizes the one-loop gluon-initiated splitting amplitudes into $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$5, $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$6, and scalar pieces. At tree level, the complete gluon-parent $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$7 catalogue closes the tree-level collinear sector needed at N$$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$8LO and satisfies nested collinear factorization with splitting tensors $$\lim_{a\parallel b} \mathcal{A}_{n+1}^{\dots,c_a,c_b,\dots}(1,\dots,a,b,\dots,n+1) = \sum_{\lambda_P=\pm} \text{Split}^{c_P c_a c_b}_{-\lambda_P}(z;a,b)\, \mathcal{A}_{n}^{\dots,c_P,\dots}(1,\dots,\tilde P,\dots,n+1). \tag{1}$$9 (Badger et al., 2015, Duca et al., 2020).

Massive and low-virtuality generalizations broaden the phenomenological scope. Triple-collinear kernels with massive quarks are designed for NNLO subtraction with heavy flavors and for precision observables involving large $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$00 terms. Low-virtuality Standard Model splitting amplitudes supply tree-level real-emission building blocks that cover the collinear, soft, and soft–collinear regions in a single framework and could enable systematic predictions of leading electroweak radiation effects in high-energy scattering processes, with particularly promising phenomenological applications to the physics of future colliders with very high energy such as a muon collider (Dhani et al., 2023, Nardi et al., 2024).

In spacelike settings, the main conceptual issue is not universality itself but how universality must be generalized. Two-loop spacelike splitting in full-color $$A_n^{(0)}(\ldots,1^{\lambda_1},2^{\lambda_2},\ldots) \xrightarrow{1\parallel2} \sum_{\lambda_P}\text{Split}^{(0)}_{P\to12}(-P^{-\lambda_P};1^{\lambda_1},2^{\lambda_2})\,A_{n-1}^{(0)}(\ldots,P^{\lambda_P},\ldots),$$01 super-Yang–Mills theory makes explicit the dipole and tripole color structures that violate naive factorization at amplitude level while preserving factorization for color-summed unpolarized squared amplitudes. Double spacelike collinear limits show that, at least at leading color, generalized splitting can correlate two distinct collinear directions through a single universal function equal to the BDS-subtracted six-point amplitude in multi-Regge kinematics. This suggests that collinear factorization in spacelike regimes is tightly linked to soft anomalous dimensions, Glauber-region physics, and Regge dynamics (Henn et al., 2024, Duhr et al., 7 Jul 2025).

In the non-pole literature, split factorizations unify smooth splitting, hidden zeros, multi-soft limits, and special kinematic loci of string and CHY amplitudes. CEGM split kinematics further suggests a bridge between scattering amplitudes, products of simplices, and tropical determinantal geometry. A plausible implication is that “generalised splitting amplitude” is not a single object but a family of universal factorizing structures, each adapted to a different analytic regime of scattering amplitudes: singular collinear limits, spacelike color-correlated limits, low-virtuality electroweak radiation, and exact split loci away from poles (Cao et al., 2024, Umbert et al., 18 Jan 2025, Jones et al., 5 May 2025).