Double Deeply Virtual Compton Scattering

• DDVCS is an exclusive lepton–hadron process with both incoming and outgoing photons off-shell, allowing independent variation of spacelike and timelike virtualities.
• It enables a broader mapping of generalized parton distributions compared to DVCS and TCS by decoupling the dependence on momentum fractions.
• Experimental strategies leverage beam-spin and charge asymmetries alongside interference with Bethe–Heitler processes to extract precise Compton form factors.

Double deeply virtual Compton scattering (DDVCS) is the exclusive lepton–hadron process in which a spacelike virtual photon probes a hadron and the final-state Compton photon is itself virtual, subsequently decaying into a lepton pair, schematically $eN\to eN\gamma^\ast\to eN\ell^+\ell^-$. Equivalently, at the hadronic level one considers $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$ with both photons off shell. DDVCS is the doubly virtual generalization of deeply virtual Compton scattering (DVCS) and timelike Compton scattering (TCS), and it is studied as a probe of generalized parton distributions (GPDs) because the independent variation of the spacelike and timelike photon virtualities opens a larger GPD phase space than the crossover kinematics of DVCS or TCS. Although it remains unmeasured, recent phenomenology treats it as a central channel for constraining Compton form factors (CFFs), especially in the helicity-conserving sector, and for testing refined QCD, holographic, and small-$x$ descriptions of exclusive Compton amplitudes (Alvarado et al., 15 Sep 2025, Deja et al., 2023).

1. Definition and relation to DVCS and TCS

In the standard electroproduction channel, DDVCS is written as

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$

with the hard subprocess

$$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$

Its defining feature is that both photons are virtual: the incoming photon is spacelike with virtuality $Q^2=-q^2$, while the outgoing photon is timelike with virtuality $Q'^2=q'^2$. In DVCS the final photon is real, $Q'^2=0$; in TCS the initial photon is real, $Q^2=0$. DDVCS therefore interpolates continuously between the two better-known limiting cases while retaining richer control over the hard scales (Zhao, 2019, Alvarado et al., 12 May 2026).

Several phenomenological studies emphasize the dimuon channel,

$ep\to e'p'\mu^+\mu^-,$

because a muon pair avoids the antisymmetrization and identical-lepton complications associated with an $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$0 final state in electron-beam experiments. More generally, the final leptons may be either an electron or a muon pair, but the muon channel has been favored in Jefferson Lab proposals because it is cleaner for reconstruction and interpretation (Zhao, 2019, Boër et al., 2024).

The distinction from DVCS and TCS is not merely kinematic. In the DDVCS literature, the process is repeatedly presented as the only experimental channel that allows the dependence of GPDs on the average and transferred momentum fractions to be studied independently, rather than restricting observables to the special lines that define the DVCS and TCS limits. That claim is the basis for the recurrent description of DDVCS as a particularly differential probe of nucleon structure (Zhao et al., 2021).

2. Kinematics and access to off-diagonal GPD domains

The basic invariants are

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$1

A common leading-twist convention defines the generalized Bjorken variables

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$2

so that DVCS corresponds to $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$3 and TCS to $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$4. Alternative phenomenological treatments use a variable $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$5 instead of $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$6, with the central point unchanged: the hard kernel depends on a momentum-fraction variable that is no longer locked to the skewness $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$7 once both photon virtualities vary independently (Zhao, 2019, Deja et al., 2023).

This leads directly to the characteristic DDVCS CFF structure. For the chiral-even GPDs,

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$8

one may write

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$9

The real part is a principal-value convolution over the full $x$0-domain, whereas the imaginary part probes the GPD at $x$1. In the $x$2-notation used in some papers, the same point is expressed as direct leading-order access to $x$3. This is the formal origin of the claim that DDVCS enables a “complete mapping” or a decoupled measurement of GPD dependence beyond the crossover line sampled by DVCS and TCS (Alvarado et al., 2024, Deja et al., 2023).

That broader access is also the reason DDVCS is tied to several broader goals of the GPD program. Hall A studies explicitly motivate it as a way to constrain extrapolations to zero skewness, $x$4, which is the kinematic limit used for tomographic interpretations of the nucleon’s partonic structure. Other proposals stress that DVCS and TCS admit deconvolution ambiguities and even “shadow GPDs” that reproduce observables while differing in their underlying $x$5-dependence, whereas DDVCS probes the GPDs away from the $x$6 ridge (Boër et al., 2024, Alvarado et al., 12 May 2026).

The literature is not entirely uniform in how far this gain is taken. Experimental proposals often describe DDVCS as access to the full three-variable GPD dependence, but the positron-beam analysis at SoLID notes that with a timelike final photon the physical DDVCS region obeys

$x$7

so the direct imaginary-part access lies in the $x$8 region. This suggests a broader but still structured enlargement of phase space, rather than the complete disappearance of kinematic restrictions (Zhao et al., 2021).

3. Factorization, amplitudes, and CFF representation

In the leading-twist handbag mechanism for a spin-$x$9 target, DDVCS is described by the same four chiral-even GPDs that appear in DVCS and TCS,

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$0

and the resulting observables are expressed through CFFs. Recent feasibility studies explicitly frame DDVCS observables as directly interpretable in terms of those CFFs, with particular emphasis on the helicity-conserving sector (Alvarado et al., 15 Sep 2025, Alvarado et al., 4 Feb 2025).

A representative leading-order decomposition of the pure DDVCS amplitude is

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$1

where $$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$2 and $$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$3 are the vector and axial contributions. In practice, the experimentally observed exclusive lepton-pair electroproduction process is not pure DDVCS. It also contains Bethe–Heitler (BH) mechanisms, usually denoted $$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$4 and $$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$5, and interference terms between DDVCS and BH. After integration over the decay-lepton solid angle, the $$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$6 interference terms vanish in the five-fold cross section in the formulation used in early JLab feasibility studies (Deja et al., 2023, Zhao, 2019).

A substantial methodological development in recent phenomenology is the use of Kleiss–Stirling spinor techniques to write DDVCS and BH amplitudes in compact helicity form. The basic building blocks are the spinor products

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$7

together with current contractions such as

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$8

and

$$e(k)+N(p)\to e'(k')+N'(p')+\ell^+(\ell_+)+\ell^-(\ell_-),$$9

These formulations were developed precisely because they are well suited for numerical implementation in PARTONS and the EpIC Monte Carlo generator, and they were validated by taking the $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$0 and $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$1 limits and comparing with independent DVCS and TCS calculations (Deja et al., 2023, Deja et al., 2023).

The same validation studies also stress a phenomenologically important point: the pure DDVCS term is typically small compared with the full electroproduction sample. Accordingly, practical extraction strategies rely on BH–DDVCS interference rather than on isolating the DDVCS contribution directly (Deja et al., 2023).

4. Observables and separation of real and imaginary CFF components

The central DDVCS observables are chosen because different beam and target combinations isolate different linear combinations of CFFs. In Jefferson Lab feasibility studies, the two flagship observables are the beam-spin asymmetry (BSA) and the muon-charge asymmetry ($$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$2CA). The BSA is especially sensitive to the imaginary parts of the CFFs, while the charge asymmetry accesses the real parts. At the EIC, the projected observable set includes the BSA and the unpolarized cross section $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$3, extending the analysis beyond asymmetries alone (Alvarado et al., 15 Sep 2025).

A representative helicity-dependent combination is

$$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$4

so, for a proton target, the dominant sensitivity is usually to $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$5 through the $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$6 term. This is the basis for the widespread use of longitudinal beam-spin observables as the primary handle on the imaginary part of the DDVCS amplitude (Boër et al., 2024, Alvarado et al., 12 May 2026).

Unpolarized observables and cosine moments are complementary. In the PARTONS-based impact studies, the Fourier cosine moment of the unpolarized cross section,

$$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$7

is proportional to the BH–DDVCS interference term and therefore gives access to the real part of the CFFs, whereas

$$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$8

enhances sensitivity to the imaginary part and thus to GPDs sampled at $$\gamma^\star(q_1)+N(p)\to N'(p')+\gamma^\star(q_2)\to N'(p')+\ell^+\ell^-.$$9 or $Q^2=-q^2$0 (Deja et al., 2023).

Charge separation adds a further layer of control. In the SoLID positron-beam program, electron and positron data are combined into charge-odd observables,

$Q^2=-q^2$1

which isolate the interference amplitude, and charge-even combinations,

$Q^2=-q^2$2

which isolate the DDVCS signal. In that framework, the four-observable set $Q^2=-q^2$3 improves the determination of both $Q^2=-q^2$4 and $Q^2=-q^2$5 (Zhao et al., 2021).

Polarized-target studies broaden the observable basis further. In the CEBAF and EIC sensitivity analyses with polarized beams and polarized proton targets, the observables

$Q^2=-q^2$6

were compared across several GPD models. At CEBAF kinematics, $Q^2=-q^2$7 and $Q^2=-q^2$8 are dominated by GPD $Q^2=-q^2$9, $Q'^2=q'^2$0 shows stronger model dependence but a more complicated GPD mixture, and $Q'^2=q'^2$1 is dominated by $Q'^2=q'^2$2 but is too small to be measurable within the assumed beam time. At EIC kinematics, $Q'^2=q'^2$3 and $Q'^2=q'^2$4 remain promising, while $Q'^2=q'^2$5 becomes valuable because of its strong dependence on GPD $Q'^2=q'^2$6 (Alvarado et al., 2024, Alvarado et al., 4 Feb 2025).

A distinctive prediction repeated across several studies is that the beam-spin observable changes sign when crossing the line $Q'^2=q'^2$7, equivalently when $Q'^2=q'^2$8 changes sign. That sign flip is presented as a stringent test of the GPD-based description of DDVCS (Zhao, 2019, Zhao et al., 2021).

5. Experimental programs, projected yields, and detector requirements

The present experimental DDVCS program is organized around high-luminosity, muon-capable measurements at Jefferson Lab and future EIC measurements in a complementary small-$Q'^2=q'^2$9 regime. Recent feasibility studies conclude that DDVCS appears viable at both facilities, but with distinct strengths: Jefferson Lab primarily probes the valence region through asymmetries, whereas the EIC extends reach into the sea-quark region and may also enable unpolarized cross-section measurements (Alvarado et al., 15 Sep 2025).

Monte Carlo studies based on PARTONS and the EpIC event generator provide representative integrated cross sections and luminosities needed to accumulate $Q'^2=0$0 events:

Configuration	$Q'^2=0$1	Luminosity for $Q'^2=0$2 events
JLab12	$Q'^2=0$3 pb	$Q'^2=0$4
JLab20+	$Q'^2=0$5 pb	$Q'^2=0$6
EIC $Q'^2=0$7	$Q'^2=0$8 pb	$Q'^2=0$9
EIC $Q^2=0$0	$Q^2=0$1 pb	$Q^2=0$2

For these same studies, the predicted beam-spin asymmetry is roughly $Q^2=0$3–$Q^2=0$4 at JLab and $Q^2=0$5–$Q^2=0$6 at the EIC. The event-generator analysis also finds that the pure DDVCS contribution is relatively small compared with the full sample, reinforcing the strategy of targeting interference-sensitive observables rather than attempting to isolate the DDVCS term directly (Deja et al., 2023, Deja et al., 2023).

At Jefferson Lab, projected luminosities up to

$Q^2=0$7

and model-predicted large asymmetry amplitudes are taken to make precise BSA and $Q^2=0$8CA measurements realistic. The relevant instrumentation includes $Q^2=0$9CLAS12 and SoLID$ep\to e'p'\mu^+\mu^-,$0. The $ep\to e'p'\mu^+\mu^-,$1CLAS12 proposal emphasizes operation at $ep\to e'p'\mu^+\mu^-,$2, forward $ep\to e'p'\mu^+\mu^-,$3 coverage, a PbWO$ep\to e'p'\mu^+\mu^-,$4 electromagnetic calorimeter (wECal), machine-learning-based PID with a pion-pair suppression factor of at least $ep\to e'p'\mu^+\mu^-,$5, removal of more than $ep\to e'p'\mu^+\mu^-,$6 of background events by the classifier, an expected single-MIP trigger rate of about $ep\to e'p'\mu^+\mu^-,$7 kHz, and an expected DDVCS yield of more than $ep\to e'p'\mu^+\mu^-,$8 events (Alvarado et al., 12 May 2026).

Hall A studies pursue a complementary route by augmenting SoLID with a dedicated muon detector and trigger. The proposed detector consists of 3 layers of iron shielding, 3 layers of straw tubes for tracking, and 2 layers of plastic scintillators for triggering on the muon pair. The Hall A projections use an $ep\to e'p'\mu^+\mu^-,$9 GeV beam, luminosity up to

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$00

and kinematics selected to avoid the central region $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$01, where current theoretical control over factorization is weaker (Boër et al., 2024).

Sensitivity studies with polarized beams and targets sharpen the experimental priorities. Assuming an upgraded luminosity about 100 times larger than nominal CLAS12, together with $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$02 combined reconstruction/acceptance efficiency and 100 days of beam time, CEBAF analyses conclude that $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$03, $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$04, and $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$05 should be measurable, whereas $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$06 is not feasible. For the EIC, with roughly $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$07 per year and $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$08 acceptance/reconstruction efficiency, $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$09, $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$10, and $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$11 are judged measurable, while $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$12 and $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$13 are too small in the explored region (Alvarado et al., 2024).

6. Higher-order QCD, holographic structure, and nuclear extensions

Beyond leading-order phenomenology, DDVCS has become a useful arena for testing more formal structures of exclusive amplitudes. In the flavor non-singlet channel, threshold resummation has been derived near the partonic threshold $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$14, where the coefficient function develops large logarithms of $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$15. The resulting leading-logarithmic exponentiation at $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$16 takes the form

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$17

That analysis also stresses that DDVCS threshold factorization is not the same as DVCS threshold factorization, because the limit $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$18 and the $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$19 expansion do not commute. As a byproduct, the leading threshold term of the two-loop coefficient function was obtained and shown to agree with a recent explicit two-loop calculation (Schoenleber, 2024).

A distinct line of work uses DDVCS as the cleanest setting for a structural comparison between holographic Witten-diagram amplitudes and the conformal-OPE formulation of perturbative QCD. In the collinear regime

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$20

the fixed-$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$21 holographic DDVCS amplitude yields the same Gauss hypergeometric hard kernel that appears in the $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$22-basis Wilson coefficients,

$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$23

In that construction, the open-string channel matches the $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$24 eigenchannel and the closed-string channel matches the protected $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$25 eigenchannel at a single scale $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$26, with $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$27 serving as the anchor moment. The authors explicitly state that this is a fixed-$\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$28, fixed-scale structural matching statement, not a claim of all-scale equality or a global fit (Mamo, 13 Apr 2026).

DDVCS has also been studied in the color dipole picture for proton and nuclear targets. In that approach, the proton and nuclear amplitudes are written as convolutions of photon wave functions with dipole scattering amplitudes, and both quark shadowing and gluon shadowing are included. The resulting DDVCS cross section is small, of order picobarns, making it primarily relevant for future high-luminosity facilities. For nuclei, shadowing can be substantial: at $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$29, the nuclear cross section can be suppressed by about a factor of two relative to naive $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$30 scaling, with stronger suppression for heavier nuclei and a gradual return toward unity with increasing $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$31 through color transparency (Kopeliovich et al., 2010).

These developments place DDVCS at an unusual intersection of phenomenology and formal theory. It is simultaneously a candidate experimental channel for disentangling GPD kinematics, a test bed for higher-order threshold dynamics, a precise setting for conformal and holographic structural matching, and, at small $\gamma^\ast(q_1)N(p)\to \gamma^\ast(q_2)N(p')$32, a probe of coherent nuclear effects. The common thread is the doubly virtual structure itself: the final-state lepton pair makes the process experimentally difficult, but it is exactly that extra virtuality that gives DDVCS its distinctive leverage on hadron structure.