Two-Photon Exchange (TPE) in Scattering

• Two-photon exchange (TPE) is a radiative correction mechanism in lepton–hadron scattering that arises from the interference between one‐photon and two‐photon exchange amplitudes.
• TPE studies use hadronic, dispersion, and partonic approaches to quantify corrections that reconcile differences between Rosenbluth separations and polarization transfer measurements of proton form factors.
• Experimental comparisons of e⁺p and e⁻p scattering, along with studies in muon and nuclear targets, validate TPE’s role in precision measurements and inform form-factor extractions.

Searching arXiv for recent and foundational papers on two-photon exchange in elastic and related scattering. Two-photon exchange (TPE) denotes the contribution to lepton–hadron scattering amplitudes arising from the interference of the leading one-photon-exchange amplitude with amplitudes containing two virtual photons. In elastic electron–proton scattering, TPE is the leading radiative correction that breaks the Born approximation and modifies both unpolarized cross sections and selected spin observables. Its phenomenological importance is tied most directly to the longstanding discrepancy between proton electromagnetic form factors extracted from Rosenbluth separations and from polarization transfer measurements, and more broadly to precision studies of elastic $e^\pm p$, $\mu^\pm p$, nuclear, and time-like hadronic processes (Moteabbed et al., 2013, Afanasev et al., 2017, Borisyuk et al., 2019).

1. Formal definition and amplitude structure

In the one-photon-exchange, or Born, approximation, elastic $ep$ scattering is described by the Rosenbluth formula

$$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$

with

$$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$

Equivalent Rosenbluth forms are also used in the review literature, for example

$$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$

or

$$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$

depending on convention (Moteabbed et al., 2013, Afanasev et al., 2017, Bernauer et al., 2021).

With TPE included, the unpolarized cross section is written as

$$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$

where the correction is defined through the interference term

$$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$

The Born amplitude is

$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$

while the two-photon contribution is the sum of box and crossed-box graphs,

$\mu^\pm p$0

with leptonic and hadronic tensors $\mu^\pm p$1 and $\mu^\pm p$2 (Afanasev et al., 2017).

Beyond Born kinematics, a general massless-electron elastic $\mu^\pm p$3 amplitude is commonly decomposed into three invariant amplitudes or generalized form factors. One representation is

$\mu^\pm p$4

with $\mu^\pm p$5 generalized form factors. Related bases use generalized electric, magnetic, and spin-flip amplitudes, often denoted $\mu^\pm p$6, $\mu^\pm p$7, and $\mu^\pm p$8, or $\mu^\pm p$9, $ep$0, and $ep$1 (Borisyuk et al., 2019, Borisyuk et al., 2012).

A central observable is the beam-charge ratio

$ep$2

which is sensitive to TPE because the interference term changes sign under $ep$3. After correcting for charge-odd bremsstrahlung interference, one has

$ep$4

or, in related conventions,

$ep$5

to first order in the small correction (Moteabbed et al., 2013, Afanasev et al., 2017, Bernauer et al., 2021).

2. Radiative corrections, infrared structure, and observables

In practical analyses, TPE does not appear in isolation. Standard radiative corrections already contain the infrared-divergent soft part of the two-photon amplitude. The finite hard contribution is therefore separated as

$ep$6

with $ep$7 the universal infrared piece included in Mo–Tsai or Maximon–Tjon prescriptions (Afanasev et al., 2017). The review literature likewise emphasizes the infrared-finite correction after subtraction of the standard Mo–Tsai soft term (Arrington et al., 2011).

Charge-even and charge-odd radiative effects must also be distinguished experimentally. In elastic $ep$8 scattering one writes

$ep$9

where $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$0 is the lepton–proton bremsstrahlung interference term and $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$1 denotes charge-even radiative corrections. After applying the calculable bremsstrahlung-interference correction, the TPE-isolation ratio is defined as

$$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$2

(Moteabbed et al., 2013).

The real and imaginary parts of the TPE amplitude affect different observables. The real part modifies unpolarized cross sections and double-spin observables and is, most likely, responsible for the discrepancy between Rosenbluth and polarization methods in proton form-factor measurements (Borisyuk et al., 2019). The imaginary part generates beam and target normal single-spin asymmetries, which vanish in one-photon exchange. These asymmetries take the form

$$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$3

and are proportional to $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$4 through unitarity relations (Borisyuk et al., 2019).

At very low $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$5, the nonrelativistic limit reduces to the McKinley–Feshbach Coulomb correction,

$$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$6

while full second-Born and dispersion-based low-$$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$7 treatments indicate that TPE shifts the extracted proton radius by $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$8 fm and $$\frac{d\sigma_{\rm Born}}{d\Omega} = \sigma_{\rm Mott}\, \Bigl[ G_E^2(Q^2)+\tau \epsilon^{-1} G_M^2(Q^2) \Bigr],$$9 fm. The same review concludes that TPE corrections at low $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$0 are too small to resolve the muonic hydrogen proton-radius puzzle (Borisyuk et al., 2019).

3. Theoretical approaches

Theoretical calculations of TPE corrections are intrinsically sensitive to proton structure. In the elastic and low-to-moderate $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$1 domain, three frameworks recur: hadronic calculations, dispersion relations, and partonic or GPD-based methods (Afanasev et al., 2017, Borisyuk et al., 2019).

In hadronic approaches, the two photons couple to a nucleon or low-lying hadronic intermediate state. One inserts either the proton intermediate state or resonances such as $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$2 into the hadronic tensor $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$3, and evaluates the resulting one-loop integrals numerically using phenomenological form-factor fits at each vertex. In this framework, TPE corrections are typically a few percent at backward angles, change sign and vanish as $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$4, and provide a positive slope in $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$5 versus $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$6 at moderate $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$7, helping restore agreement between Rosenbluth and polarization-transfer extractions of $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$8 (Afanasev et al., 2017).

Dispersion-relation methods reconstruct the real part of the TPE amplitude from its imaginary part using analyticity and unitarity. They parametrize the TPE amplitude with generalized form factors $$\sigma_{\rm Mott}=\frac{\alpha^2 E' \cos^2(\theta/2)}{4 E^3 \sin^4(\theta/2)}, \qquad \tau=\frac{Q^2}{4M_p^2}, \qquad \epsilon=\bigl[1+2(1+\tau)\tan^2(\theta/2)\bigr]^{-1}.$$9, determine $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$0 from on-shell $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$1 cuts, and then obtain $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$2 through fixed-$$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$3 dispersion relations,

$$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$4

This method enforces unitarity and allows inclusion of inelastic channels via electroproduction data (Afanasev et al., 2017).

A later analytic study of exact relations between dispersion relations and hadronic models argues that both approaches should be modified to general forms, and that after the modifications the new forms give the same results. In that formulation, seagull interaction, meson-exchange effect, contact interactions, and off-shell effect are automatically and correctly included (Cao et al., 2020). This suggests that some apparent discrepancies between frameworks arise from incomplete implementations rather than from the basic DR–HM distinction itself.

At larger $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$5, partonic methods become relevant. In the handbag GPD approach, TPE occurs on one active quark while the remaining constituents act as spectators. In pQCD factorization, the dominant TPE mechanism involves scattering off two separate quarks with one hard gluon exchange, leading to parametric $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$6 scaling of $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$7 (Afanasev et al., 2017). Soft-collinear effective theory has also been used to formulate TPE factorization at moderately large momentum transfer, separating hard-spectator and soft-spectator contributions and introducing a universal SCET form factor that also appears in wide-angle Compton scattering (Kivel et al., 2012).

A distinct low-energy effective-field-theory development is the exact heavy-baryon chiral perturbation theory analysis for MUSE kinematics. In that treatment, the TPE correction is expanded as

$$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$8

with $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ \tau G_M^2(Q^2)+\epsilon G_E^2(Q^2) \Bigr]$$9. The analysis includes recoil and proton-structure effects through $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$0, finds non-vanishing residual proton-structure effects at this order, and reports that the next-to-next-to-leading-order corrections are small, indicating reasonably good perturbative convergence (Goswami et al., 21 Jan 2026).

4. Proton form factors and the Rosenbluth–polarization discrepancy

The central phenomenological role of TPE in elastic $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$1 scattering is its connection to the proton form-factor discrepancy. Rosenbluth separations of unpolarized cross sections and polarization transfer extractions of $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$2 have long yielded inconsistent results at moderate and high $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$3. The standard interpretation is that uncorrected hard TPE introduces an $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$4-dependent shift in the reduced cross section, biasing the Rosenbluth slope used to determine $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$5, while polarization transfer is largely insensitive to $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$6 (Moteabbed et al., 2013, Bernauer et al., 2021, Borisyuk et al., 2019).

Review and model studies agree on the qualitative kinematic structure required for such a resolution. Model calculations find $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$7 at the few-percent level with pronounced $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$8-dependence: at low $$\frac{d\sigma}{d\Omega} = \frac{\sigma_{\rm Mott}}{\epsilon(1+\tau)} \Bigl[ G_E^2(Q^2)+\frac{\tau}{\epsilon}G_M^2(Q^2) \Bigr] \Bigl[1+\delta_{2\gamma}(Q^2,\epsilon)\Bigr]$$9 GeV$$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$0, $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$1–$$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$2 rising from $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$3 down to $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$4, while at higher $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$5–$$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$6 GeV$$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$7, $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$8 can reach $$\sigma=\sigma_{\rm Born}\bigl(1+\delta_{2\gamma}(Q^2,\epsilon)\bigr),$$9–$$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$0 at small $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$1 and is $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$2 at forward angles (Moteabbed et al., 2013). A related review states that the largest TPE effects occur at backward angles, $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$3, and moderate to high $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$4 in the $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$5–$$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$6 GeV$$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$7 range, where corrections can reach $$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$8–$$\delta_{2\gamma} = \frac{2\,\Re\!\bigl[{\cal M}_{1\gamma}^*{\cal M}_{2\gamma}\bigr]} {|{\cal M}_{1\gamma}|^2}.$$9 (Afanasev et al., 2017).

A phenomenological reexamination based on linear-in-$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$0 parameterizations writes

$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$1

with

$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$2

for $${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$3 GeV$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$4, implying

$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$5

That extraction concludes that applying $${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$6 to unpolarized cross sections brings LT extractions of $${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$7 into agreement with polarization-transfer results up to $${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$8 GeV$${\cal M}_{1\gamma} = -\frac{e^2}{Q^2}\,j_\mu J^\mu,$$9 (Qattan et al., 2011).

A complementary global-fit analysis asks how much TPE is needed to resolve the discrepancy and concludes that the answer depends strongly on which global fit is used for the unpolarized data. It finds that recent hard-TPE measurements can easily accommodate the hypothesis that TPE underlies the form-factor discrepancy, but that the magnitude of the discrepancy itself is not well-constrained (Schmidt, 2019). This underscores a recurrent point in the literature: the consistency of TPE as an explanation is robust, but the exact required magnitude remains sensitive to the treatment of the world data.

Theoretical work on inelastic intermediate states sharpens this picture. The $\mu^\pm p$00 contribution mainly influences the generalized electric form factor, unlike the elastic contribution, which affects the magnetic form factor, and its effect grows with $\mu^\pm p$01. For polarization transfer, the shift

$\mu^\pm p$02

was estimated to become comparable to or larger than experimental systematic uncertainties above $\mu^\pm p$03 GeV$\mu^\pm p$04 (Borisyuk et al., 2012). A dispersive calculation including resonances below $\mu^\pm p$05 GeV later found that among the resonant states, the $\mu^\pm p$06 becomes dominant for $\mu^\pm p$07 GeV$\mu^\pm p$08, with a sign opposite to the $\mu^\pm p$09 contribution, and that the combined results are in good overall agreement with recent $\mu^\pm p$10 ratio and polarization transfer measurements (Ahmed et al., 2020).

5. Experimental determinations in elastic $\mu^\pm p$11 scattering

Direct experimental access to hard TPE comes from comparing positron–proton and electron–proton elastic cross sections. Three modern experiments are repeatedly identified as the key benchmarks: VEPP-3, CLAS TPE, and OLYMPUS (Afanasev et al., 2017).

Experiment	Key configuration	Reported kinematic emphasis
VEPP-3	Monoenergetic beams at $\mu^\pm p$12, $\mu^\pm p$13 GeV; non-magnetic spectrometers	Medium and large angles
CLAS TPE	Simultaneous mixed $\mu^\pm p$14 beam, large acceptance detection	$\mu^\pm p$15 GeV$\mu^\pm p$16, $\mu^\pm p$17
OLYMPUS	$\mu^\pm p$18 GeV $\mu^\pm p$19 storage-ring beams, windowless $\mu^\pm p$20 target	$\mu^\pm p$21

At $\mu^\pm p$22 GeV$\mu^\pm p$23, VEPP-3, CLAS, and OLYMPUS all find $\mu^\pm p$24 at low $\mu^\pm p$25, rising to $\mu^\pm p$26–$\mu^\pm p$27 for $\mu^\pm p$28, while at $\mu^\pm p$29, $\mu^\pm p$30 returns to unity within $\mu^\pm p$31. Global fits combining these experiments exclude $\mu^\pm p$32 at $\mu^\pm p$33 C.L., and hadronic loop and dispersive calculations reproduce both the magnitude and the $\mu^\pm p$34-dependence of the modern data to within $\mu^\pm p$35–$\mu^\pm p$36 (Afanasev et al., 2017).

The 2013 CLAS demonstration established a new technique for making direct $\mu^\pm p$37 comparisons. A $\mu^\pm p$38 GeV, $\mu^\pm p$39–$\mu^\pm p$40 nA primary electron beam struck a thin gold radiator to produce bremsstrahlung photons, which then impinged on a downstream gold converter foil, generating $\mu^\pm p$41 pairs. A three-dipole magnetic chicane separated and recombined the lepton beams, while a photon blocker stopped the photon beam, producing a combined tertiary lepton beam with energies from $\mu^\pm p$42 to $\mu^\pm p$43 GeV delivered into CLAS (Moteabbed et al., 2013).

Elastic events were identified with over-constrained kinematics, coplanarity, transverse-momentum balance, and beam-energy reconstruction cuts. Charge-dependent acceptance effects were handled with a “swimming” algorithm, and unknown charge- and polarity-dependent efficiencies were canceled with the double-ratio method

$\mu^\pm p$44

For $\mu^\pm p$45 GeV$\mu^\pm p$46 and seven bins in $\mu^\pm p$47 between $\mu^\pm p$48 and $\mu^\pm p$49, the measured ratio before radiative corrections was

$\mu^\pm p$50

and after applying the lepton–proton bremsstrahlung correction the TPE-isolation ratio became

$\mu^\pm p$51

The data showed no significant $\mu^\pm p$52-dependence over this narrow range, as expected at low $\mu^\pm p$53, and were consistent with the Blunden–Melnitchouk–Tjon hadronic prediction $\mu^\pm p$54–$\mu^\pm p$55 in that kinematic region (Moteabbed et al., 2013).

This low-$\mu^\pm p$56 result is significant mainly as a validation of the expected forward-angle behavior: it confirms that $\mu^\pm p$57 is small, at the percent level, for forward angles. It does not by itself test the larger corrections required in the high-$\mu^\pm p$58, low-$\mu^\pm p$59 region where the form-factor discrepancy is largest (Moteabbed et al., 2013, Afanasev et al., 2017).

6. Extensions beyond elastic proton scattering

TPE is not restricted to elastic $\mu^\pm p$60 scattering. The review literature explicitly treats elastic $\mu^\pm p$61, $\mu^\pm p$62-nucleus, and $\mu^\pm p$63 scattering, and later work extends the phenomenology to deep inelastic scattering and time-like form factors (Borisyuk et al., 2019).

For trinucleon targets, elastic electron scattering from $\mu^\pm p$64 and $\mu^\pm p$65 has a richer TPE structure because photons can couple to the same nucleon or to different nucleons. In a semirelativistic calculation with Paris and CD-Bonn wave functions, all three TPE generalized form factors for electron–$\mu^\pm p$66 elastic scattering were computed. The resulting TPE corrections to $\mu^\pm p$67 and $\mu^\pm p$68 reach several percent at large $\mu^\pm p$69 and $\mu^\pm p$70 fm$\mu^\pm p$71, and the ratio $\mu^\pm p$72 departs from unity by up to $\mu^\pm p$73–$\mu^\pm p$74 at $\mu^\pm p$75 fm$\mu^\pm p$76. The same work concludes that TPE corrections in $\mu^\pm p$77-$\mu^\pm p$78 are typically $\mu^\pm p$79–$\mu^\pm p$80 times larger than in elastic $\mu^\pm p$81 scattering, with Type II diagrams, where the photons couple to different nucleons, giving the dominant share of the correction (Kobushkin et al., 2013).

In $\mu^\pm p$82 electroproduction, a hadronic calculation including only elastic nucleon intermediate states finds that TPE effects on $\mu^\pm p$83 are very small, while $\mu^\pm p$84 reaches about $\mu^\pm p$85–$\mu^\pm p$86 near $\mu^\pm p$87 GeV$\mu^\pm p$88, depending on whether MAID or SAID is used to emulate the data. For $\mu^\pm p$89, the TPE effects decrease rapidly with increasing $\mu^\pm p$90 while growing with increasing $\mu^\pm p$91, reaching $\mu^\pm p$92–$\mu^\pm p$93 with $\mu^\pm p$94 GeV$\mu^\pm p$95 at $\mu^\pm p$96. The corresponding shifts in $\mu^\pm p$97 are small, but the shifts in $\mu^\pm p$98 are comparable to or larger than current experimental uncertainties (Zhou et al., 2017).

In the time-like channel $\mu^\pm p$99, the Born cross section is symmetric under $ep$00, while TPE introduces an angular asymmetry through the interference term

$ep$01

The asymmetry

$ep$02

is therefore nonzero only beyond one-photon exchange. In a pQCD treatment including twist-2 and twist-3 pion distribution amplitudes, $ep$03–$ep$04 at $ep$05 for $ep$06–$ep$07 GeV$ep$08, falling to a few percent by $ep$09 (Chen et al., 2018).

For time-like proton form factors, BESIII reports the first unambiguous observation of OPE$ep$10TPE interference. The differential cross section for $ep$11 is decomposed as

$ep$12

where the odd $ep$13 terms are $ep$14-odd and arise from OPE$ep$15TPE interference. At $ep$16 GeV, the symmetric fit is excluded by $ep$17, corresponding to an $ep$18 effect. The integrated asymmetry at this energy is

$ep$19

and the extracted interference coefficients are

$ep$20

$ep$21

The measured asymmetry is of order $ep$22 relative to the Born term, as expected for a loop-level correction (Xia, 18 Aug 2025).

Inclusive DIS provides a contrasting case. Using HERA and SLAC $ep$23 DIS data, a recent study forms

$ep$24

for $ep$25 GeV$ep$26 at HERA and $ep$27–$ep$28 GeV$ep$29 at SLAC. It finds

$ep$30

with no statistically significant dependence on $ep$31 or $ep$32. The inferred size of $ep$33 is $ep$34 at the one-$ep$35 level across the HERA kinematic range. The study concludes that TPE effects in inclusive DIS are strongly suppressed relative to the elastic form-factor context and are negligible at the current level of PDF precision (Klest, 30 Jul 2025).

7. Open issues and future measurements

Several unresolved questions define the present TPE program. The most immediate is the lack of dedicated $ep$36 data above $ep$37 GeV$ep$38, where the form-factor puzzle is most pronounced and where hadronic and partonic descriptions are expected to diverge (Afanasev et al., 2017). The same review identifies extending direct $ep$39 measurements to $ep$40–$ep$41 GeV$ep$42 as decisive for establishing whether TPE fully resolves the Rosenbluth–polarization discrepancy (Afanasev et al., 2017).

The proposed CLAS12 program is explicitly designed for this regime. It plans alternating $ep$43 and $ep$44 beams at $ep$45, $ep$46, $ep$47, and $ep$48 GeV on a $ep$49 cm liquid-hydrogen target, with kinematic coverage from $ep$50 to $ep$51 GeV$ep$52 and $ep$53 down to $ep$54, emphasizing $ep$55 where TPE grows. Predicted statistical uncertainties on $ep$56 are $ep$57–$ep$58 up to $ep$59 GeV$ep$60, rising to $ep$61 at $ep$62 GeV$ep$63. The proposal frames this as a direct test of whether TPE transitions from a hadron- to quark-dominated regime (Bernauer et al., 2021).

A complementary proposal at DESY, the TPEX experiment, aims to determine the ratio of positron-proton to electron-proton elastic scattering with an extracted beam at $ep$64 and $ep$65 GeV. It targets luminosity $ep$66 cm$ep$67s$ep$68sr$ep$69, approximately $ep$70 times the luminosity achieved by OLYMPUS, with ten symmetric spectrometer arms covering polar angles from $ep$71 to $ep$72. The $ep$73 GeV run would extend measurements up to $ep$74 GeV$ep$75, roughly twice the range of current measurements. Simulations using phenomenological and dispersive models predict $ep$76 and $ep$77 (Alarcon et al., 2023).

At low energies, MUSE will compare $ep$78 and $ep$79 scattering at very low $ep$80, with the objective of pinning down TPE in proton-radius extractions (Afanasev et al., 2017). The exact HB$ep$81PT calculation for the MUSE kinematic regime already indicates that TPE effects enter at the few-percent level and that order $ep$82 structure effects are nonzero, even though they are small (Goswami et al., 21 Jan 2026). A plausible implication is that the dominant theoretical issue at low $ep$83 is not the existence of TPE itself, but the required precision with which recoil and structure-dependent pieces must be controlled.

Across all of these directions, the consensus of the recent literature is technically narrow but conceptually firm. In elastic $ep$84 scattering, modern $ep$85 measurements confirm the predicted size and angular dependence of TPE at the percent level for $ep$86 GeV$ep$87, theory and experiment agree at the few-per-mille level in that domain, and the remaining frontier is the high-$ep$88, low-$ep$89 region where the elastic form-factor discrepancy, resonance contributions, and hadron–parton interpolation all become most consequential (Afanasev et al., 2017, Ahmed et al., 2020).