Generalized Transverse-Momentum Distributions

• Generalized TMDs are off-forward hadronic matrix elements depending on x, ξ, kₜ, and Δₜ, serving as the 'mother' distributions for GPDs, TMDs, and Wigner functions.
• They are defined via light-front correlators with essential soft factor subtraction to cancel rapidity divergences and ensure proper renormalization and evolution.
• Studies of GTMDs reveal critical spin-orbit dynamics and orbital angular momentum correlations that are vital for a multidimensional understanding of hadron structure.

Generalized transverse-momentum-dependent distributions (GTMDs) are off-forward hadronic matrix elements that depend simultaneously on the longitudinal momentum fraction $x$, the skewness $\xi$, the transverse parton momentum $\boldsymbol{k}_T$, and the transverse momentum transfer $\boldsymbol{\Delta}_T$. They are the most general leading-twist parton correlators, interpolate to generalized parton distributions (GPDs) after integration over transverse momentum, reduce to transverse-momentum-dependent distributions (TMDs) in the forward limit, and become Wigner distributions after Fourier transformation in transverse momentum transfer, thereby encoding simultaneous transverse-position and transverse-momentum information (Echevarria et al., 2016, Pasquini et al., 2013, Engelhardt et al., 2021).

1. Definition, kinematics, and the distribution hierarchy

A standard quark GTMD correlator is written as

$$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$

with $P=(p+p')/2$, $\Delta=p'-p$, and

$\xi = -\frac{\Delta^+}{2P^+}.$

In the spin-0 case, the same kinematics are often summarized by

$$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$

with the active variables $x$, $\xi$0, $\xi$1, and $\xi$2 tracked explicitly (Bhattacharya et al., 2017, Luo et al., 2020).

The GTMD framework is naturally obtained from the fully unintegrated off-forward generalized parton correlation function after integration over the light-cone energy component. In the 3-quark light-cone and light-front constituent approaches this construction yields a unified description from which TMDs, GPDs, PDFs, form factors, and charges are recovered by specific limits or projections (Lorcé et al., 2011, Pasquini et al., 2013). The same hierarchical role is emphasized in later proton, pion, kaon, and gluon studies, where GTMDs are repeatedly described as the “mother distributions” of both TMDs and GPDs (Luo et al., 2020, Zhang, 2024, Chakrabarti et al., 17 Sep 2025).

At leading twist, the number of independent GTMDs depends on the target and parton species. For quarks on a spin-$\xi$3 target, the complete leading-twist parametrization contains 16 independent complex GTMDs, or equivalently 32 real functions once naive time-reversal properties are separated (Kanazawa et al., 2014). For leading-twist gluon correlators, 16 gluon GTMDs also appear in the general decomposition (Tan et al., 2024, Chakrabarti et al., 17 Sep 2025). For a spin-0 hadron such as the pion, the twist-two correlator is parameterized by four complex GTMDs,

$\xi$4

$\xi$5

These decompositions make explicit that GTMDs are generally complex-valued and carry both polarization and nonforward kinematic information (Luo et al., 2020).

The relation to reduced distributions is structurally simple but technically delicate. Integrating over $\xi$6 yields GPD correlators, while the forward limit $\xi$7 yields TMD correlators (Luo et al., 2020, Tan et al., 2024). However, one factorization study stresses that the often-quoted direct relation “GTMD $\xi$8 GPD by transverse integration” is not straightforward and should not be oversimplified (2208.00021).

2. Soft factors, proper operator definition, and evolution

A central development in GTMD theory is the recognition that unsubtracted definitions are not genuinely well-defined QCD objects because they contain uncanceled rapidity divergences. In a formulation directly modeled on TMD factorization, the properly defined quark GTMD is

$\xi$9

where the soft factor is

$\boldsymbol{k}_T$0

The same construction extends directly to gluon GTMDs (Echevarria et al., 2016).

The role of $\boldsymbol{k}_T$1 is to remove the soft overlap responsible for spurious rapidity singularities. Without it, the unsubtracted correlator has no proper renormalization-scale evolution in $\boldsymbol{k}_T$2, no proper rapidity evolution in the auxiliary scale $\boldsymbol{k}_T$3, and even its OPE into GPDs breaks down where one would otherwise expect it to work (Echevarria et al., 2016). At one loop, the naive unsubtracted collinear GTMD contains mixed UV/rapidity divergences, the soft function also contains rapidity divergences, and only after subtracting the full soft factor and adding back half of it through $\boldsymbol{k}_T$4 do the mixed divergences cancel (Echevarria et al., 2016).

Once soft-subtracted, GTMDs obey the same evolution structure as TMDs. In impact-parameter space,

$\boldsymbol{k}_T$5

with

$\boldsymbol{k}_T$6

and Collins–Soper evolution

$\boldsymbol{k}_T$7

The full kernel is

$\boldsymbol{k}_T$8

A key result is that this evolution kernel is spin independent: the same kernel applies to polarized and unpolarized quark GTMDs, to all 16 leading-twist quark GTMDs, and analogously to gluon GTMDs (Echevarria et al., 2016).

This evolution structure underlies logarithmic resummation. With currently known perturbative ingredients, large logarithms can be resummed up to next-to-next-to-leading-logarithmic accuracy, requiring $\boldsymbol{k}_T$9 at three loops, $\boldsymbol{\Delta}_T$0 at two loops, and $\boldsymbol{\Delta}_T$1 at two loops (Echevarria et al., 2016). In the exclusive double Drell–Yan factorization theorem, the same Collins–Soper kernel $\boldsymbol{\Delta}_T$2 governs both GTMDs and light-cone wave functions after soft contamination is removed by zero-bin subtraction (2208.00021).

3. Leading-twist parametrization and the $\boldsymbol{\Delta}_T$3–$\boldsymbol{\Delta}_T$4 sector

For quark GTMDs at twist two, the chiral-even structures are commonly written as

$\boldsymbol{\Delta}_T$5

and

$\boldsymbol{\Delta}_T$6

In this basis, $\boldsymbol{\Delta}_T$7 describes unpolarized quarks in an unpolarized nucleon, $\boldsymbol{\Delta}_T$8 describes longitudinally polarized quarks in a longitudinally polarized nucleon, and $\boldsymbol{\Delta}_T$9 and $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$0 encode the spin-orbit sector (Bhattacharya et al., 2017).

Among the leading-twist GTMDs, $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$1 and $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$2 are structurally distinctive because they do not reduce to ordinary GPDs or TMDs and are directly linked to orbital and spin-orbit dynamics. For $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$3, one convenient helicity representation is

$$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$4

The corresponding model-independent relation to canonical quark orbital angular momentum is

$$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$5

while $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$6 measures the spin-orbit correlation of a parton in an unpolarized target (Kanazawa et al., 2014).

The status of these functions has been the subject of an explicit controversy. One analysis argued that the GTMDs $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$7 and $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$8 do not survive as leading-twist observables, are parity constrained in the relevant helicity channel, and that orbital angular momentum information belongs instead to twist-three observables, notably through the GPD $$W_{\lambda,\lambda'}^{q\,[\Gamma]}(P,\Delta,x,\vec{k}_\perp)= \int \frac{dz^- \, d^2\vec{z}_\perp}{2 (2\pi)^3} \, e^{i k \cdot z} \, \langle p', \lambda' | \, \bar{q}\!\left(- \tfrac{z}{2}\right) \, \Gamma \, {\cal W}\!\left(- \tfrac{z}{2}, \tfrac{z}{2}\right) \, q\!\left(\tfrac{z}{2}\right) \, | p, \lambda \rangle \Big|_{z^+ = 0} \,,$$9 in deeply virtual Compton scattering (Liuti et al., 2013). A subsequent study argued that this claim does not hold, that parity does not forbid the relevant Lorentz structures, and that $P=(p+p')/2$0 and $P=(p+p')/2$1 are genuine, nonzero twist-2 GTMDs. That conclusion was supported by scalar diquark and quark-target model calculations and by a large-$P=(p+p')/2$2 perturbative QCD analysis showing that both functions are nonzero for quarks and gluons (Kanazawa et al., 2014).

Lattice work has further tied this sector to observable proton structure. In particular, the longitudinal quark spin-orbit correlation $P=(p+p')/2$3 was identified with the GTMD $P=(p+p')/2$4, through a proton matrix element of a bilocal quark operator with a derivative in both the quark separation $P=(p+p')/2$5 and the momentum transfer $P=(p+p')/2$6 (Engelhardt et al., 2021).

4. Perturbative matching, polarization mixing, and nonforward radiative structure

At small transverse separation $P=(p+p')/2$7, GTMDs admit an operator product expansion onto GPDs. In one formulation,

$P=(p+p')/2$8

with the one-loop coefficient obtained from the difference between the one-loop GTMD and GPD parton-in-parton matrix elements after soft subtraction (Bertone, 2022). This first complete off-forward one-loop calculation established the matching coefficients that connect perturbative GTMD behavior at small $P=(p+p')/2$9 to GPDs and confirmed that the anomalous dimensions coincide with the TMD ones, so GTMD evolution follows TMD evolution up to NNLL accuracy (Bertone, 2022).

A later treatment generalized this program to the full set of leading-twist quark and gluon GTMDs in a helicity basis. The matching takes the schematic form

$\Delta=p'-p$0

and the notable novelty is that the matching is not diagonal in polarization space: for a given GTMD polarization $\Delta=p'-p$1, several GPD polarizations $\Delta=p'-p$2 may contribute (Bertone et al., 11 Feb 2025). This is a genuinely nonforward effect and has no direct forward analogue.

The same one-loop study also identified complex phases in the ERBL region. The GTMD evolution equations retain a Collins–Soper form, but in the ERBL region $\Delta=p'-p$3 they acquire a term $\Delta=p'-p$4, so that time-reversal even and odd components mix under evolution. Writing

$\Delta=p'-p$5

the evolution rotates the $\Delta=p'-p$6 and $\Delta=p'-p$7 components into one another for $\Delta=p'-p$8, while in the DGLAP region $\Delta=p'-p$9 they evolve independently (Bertone et al., 11 Feb 2025). The same paper emphasizes that matching itself also induces T-even/T-odd mixing in the ERBL region because some gluon matching coefficients become complex there (Bertone et al., 11 Feb 2025).

In the forward limit $\xi = -\frac{\Delta^+}{2P^+}.$0, $\xi = -\frac{\Delta^+}{2P^+}.$1, the off-forward coefficients reduce to the known TMD matching coefficients (Bertone, 2022, Bertone et al., 11 Feb 2025). One phenomenological implementation for the unpolarized GTMD $\xi = -\frac{\Delta^+}{2P^+}.$2 combined one-loop matching onto GPDs with GTMD/TMD Sudakov evolution and nonperturbative transverse information from modern TMD fits, and highlighted a new singularity at $\xi = -\frac{\Delta^+}{2P^+}.$3 for gluon-induced channels in the off-forward kernels (Bertone, 2022). A plausible implication is that the $\xi = -\frac{\Delta^+}{2P^+}.$4 region requires special treatment beyond naive forward intuition.

5. Factorized access, lattice matrix elements, and quasi-distributions

The first explicit proposal for direct access to quark GTMDs through a physical process was the exclusive pion–nucleon double Drell–Yan reaction

$\xi = -\frac{\Delta^+}{2P^+}.$5

or equivalently the production of two virtual photons,

$\xi = -\frac{\Delta^+}{2P^+}.$6

In this framework, polarization observables were proposed to isolate specific GTMDs, especially the OAM-sensitive $\xi = -\frac{\Delta^+}{2P^+}.$7 and $\xi = -\frac{\Delta^+}{2P^+}.$8, either directly or through interference with larger amplitudes (Bhattacharya et al., 2017).

A decisive later step was the first proof of factorization for an exclusive GTMD-sensitive process. Using SCET, the differential cross section for exclusive double Drell–Yan was shown to factorize at leading power in the regime

$\xi = -\frac{\Delta^+}{2P^+}.$9

into a perturbatively calculable hard factor, two GTMDs, and two light-cone wave functions, with soft contamination removed by zero-bin subtraction (2208.00021). The hard factor was found to be

$$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$0

identical to the single inclusive Drell–Yan hard factor, so logarithms can be resummed with the standard Drell–Yan machinery (2208.00021). This result upgraded GTMDs from formal correlators to factorization objects for a concrete process.

A complementary access route is lattice QCD. In a physical-pion-mass calculation using domain wall fermions, the relevant GTMD matrix element for $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$1 was evaluated with a staple-shaped gauge link, and the momentum-transfer derivative needed for the transverse-position weighting was obtained using a direct derivative method (Engelhardt et al., 2021). The gauge-link geometry interpolates between Ji and Jaffe–Manohar definitions: $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$2 corresponds to a straight Wilson line, while $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$3 corresponds to a staple extending to infinity (Engelhardt et al., 2021). The extracted spin-orbit correlation was negative, large in magnitude, and enhanced by about 50% when going from the Ji definition to the Jaffe–Manohar definition (Engelhardt et al., 2021).

Quasi-distributions provide a third route. In a scalar spectator model for the pion, quasi-TMDs and quasi-GPDs were defined through equal-time spatial correlation functions with spacelike Wilson lines, designed for Euclidean lattice QCD and large-momentum effective theory. In that setup, the quasi-TMD and quasi-GPD reduce to their light-cone counterparts in the limit $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$4, up to matching and power corrections (Luo et al., 2020). This establishes a direct conceptual bridge between GTMD-related physics and the LaMET program.

6. Nonperturbative structure: nucleon, pion, kaon, and gluon studies

Model and continuum studies have used GTMDs as the organizing framework for multidimensional hadron tomography across different hadrons and twist sectors. In a 3-quark light-cone picture of the nucleon, GTMDs were constructed as overlap representations of light-front wave functions in both the light-front constituent quark model and the chiral quark-soliton model, providing a unified description of TMDs, GPDs, PDFs, form factors, and Wigner functions (Lorcé et al., 2011). A subsequent light-front constituent quark study emphasized that Wigner distributions derived from GTMDs display dipole distortions for unpolarized quarks in a longitudinally polarized proton and connected these distortions to quark orbital motion (Pasquini et al., 2013).

Sub-leading twist structure has been worked out systematically in the light-front quark-diquark model. One twist-4 proton study derived 16 GTMDs,

$$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$5

at $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$6, and showed that their $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$7 limit reproduces previously published twist-4 T-even TMDs (Sharma et al., 2023). A later sub-leading-twist proton analysis derived 32 twist-3 GTMDs $$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$8, identified model-specific zeros such as

$$P=\frac{1}{2}(p+p'), \qquad \Delta=p'-p, \qquad t=\Delta^2=-\frac{1}{1-\xi^2}(4\xi^2 M^2+\vec \Delta_T^2),$$9

and found that nearly all transverse-momentum-dependent form-factor amplitudes fall to zero once $x$0 reaches or exceeds about $x$1 GeV (Sharma et al., 2024).

For the pion, both T-even and T-odd sectors have been explored. In a scalar spectator model, all four twist-two T-odd GTMDs of a spin-0 target,

$x$2

were shown to arise from one-loop final-state-interaction diagrams, vanish outside the DGLAP region, and satisfy $x$3 at $x$4 (Luo et al., 2020). In a later light-cone quark model at $x$5, 12 of the 16 possible pion GTMDs were found to be nonzero, specifically

$x$6

while

$x$7

That same study reported an elastic charge radius

$x$8

and a negative spin-orbit correlator

$x$9

indicating anti-alignment of quark spin and orbital angular momentum in the pion valence sector (Puhan et al., 21 Apr 2025).

For the kaon, a Dyson–Schwinger-equation calculation with symmetry-preserving contact interaction computed twist-two, twist-three, and twist-four GTMDs, derived GPDs, TMDs, and Wigner distributions from them, and extracted flavor-asymmetric spin-orbit correlations,

$\xi$00

In that framework, the valence $\xi$01-quark OAM is anti-aligned with its spin in the kaon, whereas the $\xi$02-quark OAM is aligned with its spin (Zhang, 2024).

Gluon GTMDs have been developed in parallel. A nonzero-skewness light-front gluon–triquark model computed the leading-twist distributions $\xi$03 and $\xi$04, derived the corresponding gluon GPDs and impact-parameter distributions, and found that the model concentrates at low $\xi$05 in the DGLAP region $\xi$06. In that model, positive $\xi$07 implies negative canonical gluon orbital angular momentum, while negative $\xi$08 implies antialignment between gluon spin and OAM (Tan et al., 2024). A later spectator-model calculation extended this to Wigner distributions for unpolarized, longitudinally polarized, and transversely polarized proton states and reported

$\xi$09

again indicating anti-alignment of gluon orbital motion with proton spin and a strongly negative gluon spin-orbit correlation (Chakrabarti et al., 17 Sep 2025).

7. Small-$\xi$10, strong coupling, and the expanding theoretical scope

At small $\xi$11 and vanishing skewness, the complete set of leading-twist gluon and sea-quark GTMDs simplifies dramatically. In the eikonal approximation, all gluon GTMDs can be expressed in terms of a basic gluon dipole operator,

$\xi$12

which leads to universal relations between otherwise distinct GTMDs (Benić et al., 6 Mar 2026). The dipole GTMD naturally separates into Pomeron and Odderon components; real parts correspond to Pomerons and imaginary parts to Odderons (Benić et al., 6 Mar 2026). In the same strict small-$\xi$13 limit, sea-quark helicity-flip and transversity GTMDs vanish, while the surviving sea-quark GTMDs are written as convolutions of the same gluon dipole with hard kernels (Benić et al., 6 Mar 2026). Their perturbative large-$\xi$14 tails are governed by small-$\xi$15 gluon GPDs (Benić et al., 6 Mar 2026).

A very different extension of the subject is the strong-coupling, holographic treatment of gluon GTMD conformal moments. At fixed even conformal spin $\xi$16, the unpolarized gluon GTMD moment splits into a local boundary sector at $\xi$17 and a finite-separation worldsheet sector at $\xi$18,

$\xi$19

with the finite-$\xi$20 sector factorizing into a universal staple-worldsheet soft factor and a stripped spin-$\xi$21 Witten amplitude (Mamo et al., 18 Jun 2026). In that framework, the cusp of the renormalized minimal area generates the Collins–Soper rapidity-logarithmic structure, and the large-$\xi$22 behavior depends on the infrared completion: soft-wall, gap-matched hard-wall, and repulsive-wall backgrounds give algebraic, exponential, and Gaussian falloffs, respectively (Mamo et al., 18 Jun 2026). Analytic continuation in $\xi$23 yields a low-$\xi$24 Regge regime governed by the holographic Pomeron spectral curve with intercept

$\xi$25

This suggests a unified strong-coupling description of hadron tomography, rapidity evolution, and Reggeization for GTMD moments (Mamo et al., 18 Jun 2026).

Across these developments, a consistent picture emerges. GTMDs are the most complete two-parton correlation functions presently used for hadron tomography; their operator definition requires soft subtraction; their evolution follows TMD logic but acquires distinctive nonforward features; their perturbative matching onto GPDs is richer than forward TMD matching because of polarization mixing and ERBL-region phases; and their phenomenology spans exclusive processes, lattice matrix elements, continuum approaches, spectator models, small-$\xi$26 limits, and strong-coupling constructions (Echevarria et al., 2016, 2208.00021, Bertone et al., 11 Feb 2025).