betaFeedback

WorksPhysics and Astronomy

Proton internal pressure from deeply virtual Compton scattering on collider kinematics

Abstract

The unique experimental connection to the QCD energy-momentum tensor offered by generalised parton distributions has been strongly highlighted in the past few years with attempts to extract the pressure and shear forces distributions within the nucleon. If, in principle, this can be performed in a model independent way from experimental data, in practice, the current limited precision and kinematic coverage make such an extraction very challenging. Moreover, the limitation to a leading-order description in the strong coupling of the data has provided only an indirect and weakly sensitive access to gluon degrees of freedoms, solely through their mixing to quarks via evolution. In this paper we address this issue by pro viding a next-to-leading order formalism allowing a reanalysis of global fits with genuine gluonic degrees of freedom. In addition, we provide an estimate of the reduction in uncertainty that could stem from the extended kinematic range relevant for the future Electron Ion Collider. Finally, we stress the connection between the analysis of the dispersion relation in terms of generalised parton distributions and the deconvolution problem.

Introduction

More than a century after its discovery by E. Rutherford [1], the proton is still at the core of an intense research activity. Among other aspects, its mechanical properties, described through the quantumchromodynamics (QCD) energy-momentum tensor (EMT), have attracted a significant attention in the last decade yielding many theoretical (see for instance [2, 3, 4]), phenomenological (for instance [5, 6, 7]), lattice (as in [8, 9, 10]) and continuum studies (see e.g. [11, 12]). The reason for this interest is that the macroscopic properties of the proton such as its mass or its spin are expected to be emergent phenomena from the microscopic interaction between quarks and gluons. The computation of the quark and gluon contributions to the macroscopic properties of the proton, and the comparison with the experimental extraction of these contributions has become one of the main objectives of modern hadron physics.

The connection to experimental data is certainly one of the main factors explaining the recent interest for the EMT. Indeed, it was shown almost three decades ago [13] that one can build an indirect experimental access to the EMT via generalised parton distributions (GPDs) [14, 13, 15, 16, 17]. The latter enter the description of deep exclusive processes according to QCD factorisation theorems [18, 19]. One can for instance highlight deeply virtual Compton scattering (DVCS) [15], timelike Compton scattering (TCS) [20], deep virtual meson production (DVMP) [21], multiparticle production [22, 23, 24] or single diffractive hard exclusive processes [25, 26, 27]. However, the main source of experimental information remains today DVCS which was measured in several facilities in the last two decades (see e.g. [28, 29, 30, 31, 32, 33, 34, 35]) and which is currently the core of an intense experimental program at the Thomas Jefferson Laboratory in the USA and at COMPASS at CERN.

This experimental effort has triggered an important theoretical interest for DVCS, which is today the deep exclusive process with the clearest theoretical framework. Higher order corrections up to next-to-next to leading order (NNLO) have been derived [36]. Higher-twist kinematic corrections are also available [37]. However, despite its mature theoretical description, DVCS does not allow to extract unambiguously GPDs. The reason is to be found in the so-called deconvolution problem [38, 39], that is the ill-definedness of the inverse problem relating DVCS form factors to GPDs, embodied by the notion of shadow GPDs. One form factor of the EMT has been particularly studied since it can be accessed from a dispersive analysis of DVCS [40] without requiring the full extraction of the GPDs. Indeed, this form factor only depends on the Polyakov-Weiss DD-term [41], accessible directly from the real and imaginary parts of the Compton Form Factors. However, this extraction is also plagued by shadow DD-term contributions, hinted at in [7] and presented in greater detail here.

In this paper, we assess the feasibility of extracting from DVCS data independently genuine quarks and gluon contributions to the pressure and shear forces inside the proton, taking into account shadow DD-terms and evolution. After introducing our notations and conventions in § 2, we propose in § 3 a new derivation of the dispersion relations of DVCS amplitudes at any order of perturbation theory. This presentation highlights that dispersion relations can provide more information than solely the canonical one concerning the DD-term. We present the issue of the deconvolution problem and introduce the notion of shadow DD-terms in § 4. Then, we apply our formalism on an existing global fit in § 5 and present the results of the first next-to-leading order extraction. In § 8, we investigate more closely the impact of shadow DD-term on a kinematic range relevant at future facilities, in view of the future electron-ion collider (EIC).

EMT experimental access through GPDs

The proton matrix element of the local gauge-invariant QCD energy-momentum tensor (EMT) operator can be parameterised in terms of five gravitational form factors (GFFs) Aa(t)A_a(t), Ba(t)B_a(t), Ca(t)C_a(t), C‾a(t)\overline{C}_a(t) and Da(t)D_a(t) [42, 43] where t=Δ2t = \Delta^2 with Δ=p′−p\Delta= p' - p the four-momentum transfer to the proton (see Ref. [7]). The label aa denotes either the quark flavour (a=qa = q) or the gluon (a=ga = g) contribution to the EMT. These GFFs allow to define various distributions of so-called mechanical properties of the proton, like distributions of pressure and shear stress induced by the nucleon’s partonic structure [44, 2, 3]. Some of these GFFs are accessible thanks to their remarkable relation to generalised parton distributions (GPDs) introduced in [14, 13, 15, 16, 17], such as [45]:

∫−11dx x1−paHa(x,ξ,t)=Aa(t)+4ξ2Ca(t),(1)\int_{-1}^{1} \mathrm{d}x\,x^{1-p_a} H^a(x,\xi,t) = A_a(t) + 4\xi^2 C_a(t), \tag*{(1)}
∫−11dx x1−paEa(x,ξ,t)=Ba(t)−4ξ2Ca(t).(2)\int_{-1}^{1} \mathrm{d}x\,x^{1-p_a} E^a(x,\xi,t) = B_a(t) - 4\xi^2 C_a(t). \tag*{(2)}

where pq=0p_q = 0, pg=1p_g = 1 and Ha(x,ξ,t)H^a(x,\xi,t) and Ea(x,ξ,t)E^a(x,\xi,t) are leading-twist chiral-even GPDs depending on xx, the average longitudinal light-front momentum fraction of the active parton and ξ\xi, the skewness variable describing the transfer of longitudinal light-front momentum to the system. The link between GFFs and GPDs offers a unique opportunity for an experimental access to the mechanical properties of hadron matter, thanks to the sensitivity to GPDs of a wide class of exclusive experimental channels. We also highlight that the specific polynomial ξ\xi dependence of the Mellin moments of eqs. (1) and (2) is in fact a general property called polynomiality [46, 47] and is generalised to higher moments as:

∫−11dx xm−paHa(x,ξ,t)=∑j=0⌊m2⌋Am;2ja(t)(2ξ)2j+mod⁡(2,m)(2ξ)m+1Cma(t),(3)\int_{-1}^{1} \mathrm{d}x\,x^{m-p_a} H^a(x,\xi,t) = \sum_{j=0}^{\left\lfloor\frac{m}{2} \right\rfloor} A^a_{m;2j}(t)(2\xi)^{2j} + \operatorname{mod}(2,m)(2\xi)^{m+1} C^a_m(t), \tag*{(3)}
∫−11dx xm−paEa(x,ξ,t)=∑j=0⌊m2⌋Bm;2ja(t)(2ξ)2j−mod⁡(2,m)(2ξ)m+1Cma(t),(4)\int_{-1}^{1} \mathrm{d}x\,x^{m-p_a} E^a(x,\xi,t) = \sum_{j=0}^{\left\lfloor\frac{m}{2} \right\rfloor} B^a_{m;2j}(t)(2\xi)^{2j} - \operatorname{mod}(2,m)(2\xi)^{m+1} C^a_m(t), \tag*{(4)}

where ⌊⋯ ⌋\lfloor\cdots\rfloor is the floor function and mod⁡(2,m)\operatorname{mod}(2,m) is 0 for mm even and 1 for mm odd. The polynomiality property is equivalent to the so-called double distribution formalism introduced independently in [14, 17] (see also [48, 49] for a modern picture of the connection between the two). The double distributions FaF^a and KaK^a are connected to the GPDs through:

Ha(x,ξ,t)=∫Ωdβ dα[βpaFa(β,α,t)+ξ1+paDa(α,t)δ(β)]×δ(x−β−αξ),(5)H^a(x,\xi,t) = \int_{\Omega} \mathrm{d}\beta\,\mathrm{d}\alpha\left[ \beta^{p_a} F^a(\beta,\alpha,t) + \xi^{1+p_a} D^a(\alpha,t)\delta(\beta) \right]\times\delta(x-\beta-\alpha\xi), \tag*{(5)}
Ea(x,ξ,t)=∫Ωdβ dα[βpaKa(β,α,t)−ξ1+paDa(α,t)δ(β)]×δ(x−β−αξ),(6)E^a(x,\xi,t) = \int_{\Omega} \mathrm{d}\beta\,\mathrm{d}\alpha\left[ \beta^{p_a} K^a(\beta,\alpha,t) - \xi^{1+p_a} D^a(\alpha,t)\delta(\beta) \right]\times\delta(x-\beta-\alpha\xi), \tag*{(6)}

with Ω={(α,β)∣∣α∣+∣β∣<1}\Omega= \{(\alpha,\beta)\lvert|\alpha|+|\beta| < 1\} and where DaD^a is the so-called Polyakov-Weiss DD-term whose first Mellin moments yields the GFF Ca(t)C_a(t) in eq. (1):

Ca(t)=14∫−11dα α1−paDa(α,t).(7)C_a(t) = \frac{1}{4}\int_{-1}^{1}\mathrm{d}\alpha\,\alpha^{1-p_a}D^a(\alpha,t). \tag*{(7)}

However, the unambiguous model-independent extraction of GPDs from one of the most promising current channels, namely deeply virtual Compton scattering (DVCS), has already been demonstrated to be practically unfeasible in Ref. [38]. The reason is to be found in the relation between the DVCS amplitude, parametrised with Compton Form factors (H\mathcal{H}, E\mathcal{E}, ...) and GPDs:

Ha(ξ,t,Q2)=∫−11dxξTa(xξ,Q2μ2,αs)Ha(x,ξ,t,μ2)ξpa(8)\mathcal{H}^{a}(\xi,t,Q^{2})=\int_{-1}^{1}\frac{\mathrm{d}x}{\xi}T^{a}\left(\frac{x}{\xi},\frac{Q^{2}}{\mu^{2}},\alpha_{s}\right)\frac{H^{a}(x,\xi,t,\mu^{2})}{\xi^{p_{a}}} \tag*{(8)}

where TaT^{a} is the DVCS coefficient function. The convolution reveals itself not to be numerically invertible on any relevant range in Q2Q^{2}, the virtuality of the photon mediating the interaction between the lepton beam and the proton target in DVCS, yielding out of control uncertainties. This situation can partly be tamed by exploiting the theoretical constraints applied on GPDs [39] (see also [50] for a exhaustive list of these properties), but theoretical uncertainties remain significant.

In this context, the GFF Ca(t)C_{a}(t) has attracted a specific interest since it does not require a full extraction of GPDs, but is instead sensitive to the DD-term only. The latter can be probed more specifically in a dispersive formalism of DVCS [51, 40, 52]. A careful analysis of the world DVCS data using this dispersive formalism at leading order (LO) was performed in Ref. [7] using realistic uncertainties on DVCS form factors coming from a neural network analysis [53]. The dependence on Q2Q^{2} was taken into account through the use of evolution equations for the scale dependence of the DD-term, but the LO approach did not take into account any direct gluon contribution to the subtraction constant. It is our objective in this paper to propose a full next-to-leading order (NLO) treatment, whose relevance becomes stringent with the lever arm in Q2Q^{2} promised by future collider experiments, in particular those to be conducted in the electron-ion collider (EIC) [54, 55], Chinese electron-ion collider (EicC) [56, 57] or large hadron-electron collider (LHeC) [58].

DVCS dispersion relations beyond leading order

In this section, we provide a brief summary of the state-of-the-art regarding DVCS dispersion relations and provide an alternative proof beyond leading order. We also generalise the dispersion relation for an arbitrary number of subtractions, allowing a new way to connect moments of GPDs with experimental data. Note however that we focus only on ss and uu channel dispersion relations, constraining the xx and ξ\xi dependence of GPDs. tt-channel dispersion relations have been investigated in [59], and are beyond the scope of this paper.

State of the art and dispersion relations at LO and NLO

Dispersion relations at Born order were first derived in ref. [51, 40]. The authors took advantage of the explicit leading-order expression of the CFF in terms of GPDs to derive the dispersion relations. In a nutshell, at LO, the connection between GPDs and CFFs are given by:

RHq(ξ)=LOeq2∫−11Hq(x,ξ)[1ξ−x−1ξ+x]dx(9)\mathfrak{R}\mathcal{H}^{q}(\xi)\overset{\mathrm{LO}}{=}e_{q}^{2}\int_{-1}^{1}H^{q}(x,\xi)\left[\frac{1}{\xi-x}-\frac{1}{\xi+x}\right]\mathrm{d}x \tag*{(9)}
IHq(ξ)=LOπeq2[Hq(ξ,ξ)−Hq(−ξ,ξ)](10)\mathfrak{I}\mathcal{H}^{q}(\xi)\overset{\mathrm{LO}}{=}\pi e_{q}^{2}\left[H^{q}(\xi,\xi)-H^{q}(-\xi,\xi)\right] \tag*{(10)}

where f\mathfrak{f} indicates that the integrals are regularised through the Cauchy principal value prescription. Using the dispersion relation relating the real and imaginary part of the Compton Form Factor:

RHq(ξ)=1π∫01dxIHq(x)[1ξ−x−1ξ+x]+Sq(ξ)(11)\mathfrak{R}\mathcal{H}^{q}(\xi)=\frac{1}{\pi}\int_{0}^{1}\mathrm{d}x\mathfrak{I}\mathcal{H}^{q}(x)\left[\frac{1}{\xi-x}-\frac{1}{\xi+x}\right]+\mathcal{S}^{q}(\xi) \tag*{(11)}

one introduces Sq\mathcal{S}^{q}, the so-called subtraction constant associated with a flavour qq. Combining eqs. (9), (10) and (11), one gets the following expression for the subtraction constant at LO:

Sq(ξ)=LOeq2∫−11Hq(x,ξ)−Hq(x,x)ξ−xdx\mathcal{S}^{q}(\xi)\overset{\mathrm{LO}}{=}e_{q}^{2}\int_{-1}^{1}\frac{H^{q}(x,\xi)-H^{q}(x,x)}{\xi-x}\mathrm{d}x
−eq2∫−11Hq(−x,ξ)−Hq(−x,x)ξ−xdx-e_{q}^{2}\int_{-1}^{1}\frac{H^{q}(-x,\xi)-H^{q}(-x,x)}{\xi-x}\mathrm{d}x
=LO2eq2∫01[Hq(x,ξ)−Hq(x,x)][1ξ−x−1ξ+x]dx(12)\overset{\mathrm{LO}}{=}2e_{q}^{2}\int_{0}^{1}\left[H^{q}(x,\xi)-H^{q}(x,x)\right] \left[\frac{1}{\xi-x}-\frac{1}{\xi+x}\right]\mathrm{d}x \tag*{(12)}

Expanding Hq(x,ξ)H^{q}(x,\xi) as a Taylor series around x=ξx=\xi, and using the polynomiality condition (3) or equivalently the DD representation (5), one recovers the well-known leading-order relation between the subtraction constant and the DD-term:

Sq(t)=LO2eq2∫−11dzDq(z,t)1−z(13)\mathcal{S}^{q}(t)\overset{\mathrm{LO}}{=}2e_{q}^{2}\int_{-1}^{1}\mathrm{d}z\frac{D^{q}(z,t)}{1-z} \tag*{(13)}

DVCS dispersion relations beyond the Born order were considered in ref. [52] whose analysis is based on the dispersion relation of eq. (11). Then, introducing GPDs through the leading-twist all-order factorisation theorem (eq. (8)) as well as the polynomiality property, they derive the general form of the subtraction constant as:

Sq=2π∫1∞dωITq(ω,Q2μ2,αs)∫−11dαDq(α,t,μ2)ω−α(14)\mathcal{S}^{q}=\frac{2}{\pi}\int_{1}^{\infty}\mathrm{d}\omega\mathfrak{I}T^{q}\left(\omega,\frac{Q^{2}}{\mu^{2}},\alpha_{s}\right)\int_{-1}^{1}\mathrm{d}\alpha\frac{D^{q}(\alpha,t,\mu^{2})}{\omega-\alpha} \tag*{(14)}

To the best of our knowledge, this is the sole study on the topic of dispersion relations beyond Born order until now.

This integral generalises the expression of the subtraction constant at higher order in perturbation theory, at the price of introducing a second integration variable, to be integrated to infinity. In the following, we present another approach to generalise dispersion relations at higher orders, allowing us to write the subtraction constant as a single integral over the DD-term.

Analytic properties of the Compton Form Factors

In order to derive an alternative expression to the DVCS dispersion relation at higher-pQCD order, we need to recall the analytic properties of the CFF, which are a special case of the 2-particle scattering amplitude (see Figure 1). These types of amplitudes are fully characterised by the three Mandelstam variables ss, tt and uu defined as:

Figure 1. Diagram for the abstract 2-particle process considered

s=(p+q)2=(p′+q′)2(15)s=(p+q)^2=(p^\prime+q^\prime)^2 \tag*{(15)}
t=(p′−p)2=(q′−q)2(16)t=(p^\prime-p)^2=(q^\prime-q)^2 \tag*{(16)}
u=(p′−q)2=(p−q′)2(17)u=(p^\prime-q)^2=(p-q^\prime)^2 \tag*{(17)}

such that s+t+u=p2+q2+q′2+p′2s+t+u=p^2+q^2+q^{\prime2}+p^{\prime2}. In the following, we will assume working at fixed tt and fixed q2q^2 for a real outgoing photon, such that the amplitude can be fully described by either ss or uu. To distinguish between the two, we will write Fs(s)\mathcal{F}_s(s) and Fu(u)\mathcal{F}_u(u) respectively. From that point, several postulates allow us to define an analytic continuation to the complex plane. We briefly give them here but more details can be found in ref. [60].

  1. Causality allows us to extend the physical amplitude to the upper-half complex plane Ps+\mathcal{P}_s^+, such that:

Fs(s)=lim⁡ϵ→0Fs(s+iϵ).(18)\mathcal{F}_s(s)=\lim_{\epsilon\to0}\mathcal{F}_s(s+i\epsilon). \tag*{(18)}
  1. The Schwartz reflection principle allows us to extend the upper plane analytic continuation to the lower one following:

Fs(s∗)=Fs∗(s),(19)\mathcal{F}_s(s^*)=\mathcal{F}_s^*(s), \tag*{(19)}

and provided that Fs\mathcal{F}_s is real on at least a segment of the real axis. It immediately follows:

lim⁡ϵ→0Re⁡(Fs(s+iϵ))=lim⁡ϵ→0Re⁡(Fs(s−iϵ)),(20)\lim_{\epsilon\to0}\operatorname{Re}\left(\mathcal{F}_s(s+i\epsilon)\right)=\lim_{\epsilon\to0}\operatorname{Re}\left(\mathcal{F}_s(s-i\epsilon)\right), \tag*{(20)}
lim⁡ϵ→0Im⁡(Fs(s+iϵ))=−lim⁡ϵ→0Im⁡(Fs(s−iϵ)),(21)\lim_{\epsilon\to0}\operatorname{Im}\left(\mathcal{F}_s(s+i\epsilon)\right)=-\lim_{\epsilon\to0}\operatorname{Im}\left(\mathcal{F}_s(s-i\epsilon)\right), \tag*{(21)}

and thus, F\mathcal{F} is discontinuous on the real axis if the imaginary part doesn’t vanish.

  1. The imaginary part of the amplitude can be computed from the optical theorem, and corresponds to the sum of all possible on-shell intermediate states. Single stable states are responsible for poles on the real axis, while multi-particle stable states trigger cuts.

Importantly, if ss, tt and uu are all space-like, then no on-shell intermediate state is allowed and the amplitude is real and continuous on the real axis.

Until now we have discussed the analytic continuation in the complex plane of Fs(s)\mathcal{F}_s(s), but the same procedure could be applied to Fu(u)\mathcal{F}_u(u). The two variables are connected through:

s+u=2p2+q2−t=Σ(22)s+u=2p^2+q^2-t=\Sigma \tag*{(22)}

and therefore, Fs(s)\mathcal{F}_s(s) and Fu(u)\mathcal{F}_u(u) are expected to be related by the crossing symmetry and thus by analytical continuation. However, in the case of CFFs, because one of the outgoing particles is massless (a real photon), to the best of our knowledge, there is no formal proof showing how to connect Fs(s)\mathcal{F}_s(s) and Fu(u)\mathcal{F}_u(u). We therefore stick to the standard assumption stating that the analytic continuations of amplitudes Fs(s)\mathcal{F}_s(s) and Fu(u)\mathcal{F}_u(u) on their respective physical upper half-planes connect through a real interval between the two thresholds, as highlighted on Figure 2.

Representation of half-planes $\mathcal{P}_s^+$ and $\mathcal{P}_u^+$ aligned according to the crossing condition

Figure 2. Representation of half-planes Ps+\mathcal{P}_s^+ and Pu+\mathcal{P}_u^+ aligned according to the crossing condition. Singularities for the uu and ss channel can appear only on the highlighted intervals.

In the specific case of DVCS, we are able to easily identify such an interval. Indeed, since q2q^2 is negative and much larger in absolute value than the hadron mass and tt, Σ\Sigma is negative. We can thus identify an interval between (s=0,u=Σ)(s=0,u=\Sigma) and (s=Σ,u=0)(s=\Sigma,u=0) where both Fs(s)\mathcal{F}_s(s) and Fu(u)\mathcal{F}_u(u) are real, allowing us to define the analytic continuation between the amplitudes. We also highlight that in the case Σ>0\Sigma>0, the crossing structure is more complicated as the connection has to be done through the respective cuts of the amplitudes. This situation is expected to arise in the case of Time-like Compton Scattering (TCS) where the hard scale is provided by a deeply timelike outgoing virtual photon. We do not consider this case in the present analysis.

The last step to characterise the properties of the CFF is to describe their singularity structure in the Bjorken limit4. We recall that in this limit, all masses, thresholds, and ∣t∣|t| are very small compared to Q2=Q^2= −q2-q^{2}. Let us introduce the variable ν\nu (differing by a factor 2 compare to the one of [52])

ν=s−uΣ≃u−sQ2≃1ξ,(23)\nu= \frac{s-u}{\Sigma} \simeq\frac{u-s}{Q^{2}} \simeq\frac{1}{\xi}, \tag*{(23)}

such that for ν=1\nu= 1, (s=Σ,u=0)(s = \Sigma,u = 0) and for ν=−1\nu= -1, (s=0,u=Σ)(s = 0,u = \Sigma). Consequently, the CFF F(ν)\mathcal{F}(\nu) is real and continuous for ν∈]−1;1[\nu\in]-1;1[, and analytic for the entire complex plane but ν∈]−∞,−1]∪[1,∞[=P\nu\in]-\infty,-1] \cup[1,\infty[ = \mathcal{P}. This structure is simple and illustrated on Figure 3. The direct and crucial consequence is that for any ∣ν∣<1|\nu| < 1, one can write the CFFs as:

Figure 3. Analytic structure of the amplitude in the ν\nu-plane

F(ν)=∑j=0∞fjνj(24)\mathcal{F}(\nu) = \sum_{j=0}^{\infty} f_j \nu^j \tag*{(24)}

or in terms of the variable ξ\xi, for ∣ξ∣>1|\xi| > 1:

F(ξ)=∑j=0∞fj1ξj.(25)\mathcal{F}(\xi) = \sum_{j=0}^{\infty} f_j \frac{1}{\xi^j}. \tag*{(25)}

Note that because of the Schwartz principle, the fjf_j are all real and uniquely define the analytic continuation in the complex plane. This generalises the proof provided of analyticity provided in ref. [51] which was based on a Taylor expansion of the LO DVCS kernel in the unphysical region ∣ξ∣>1|\xi| > 1.

All order dispersion relation with arbitrary subtraction

The analyticity of the CFFs of eq. (25) is a major result, and our next goal is to connect the fjf_j coefficients with the associated GPDs. To derive these relations, we first go back to eq. (8), where the factorisation theorem is applied to write the CFF as a convolution of coefficient function TT and a GPD HH. The ξ\xi dependance in these formula can be simplified by using the Double Distribution representation of GPDs. Thus introducing eq. (5) into eq. (8), one obtains:

Hq=1ξ∫ΩTq(α+βξ)Fq(β,α) dβdα+h0,(26)\mathcal{H}^{q} = \frac{1}{\xi}\int_{\Omega} T^{q}\left(\alpha+\frac{\beta}{\xi}\right)F^{q}(\beta,\alpha)\,\mathrm{d}\beta\mathrm{d}\alpha+h_{0}, \tag*{(26)}

where we define:

h0=∫−11Tq(ω)Dq(ω) dω.(27)h_{0}=\int_{-1}^{1}T^{q}(\omega)D^{q}(\omega)\,\mathrm{d}\omega. \tag*{(27)}

Note that we have assumed that the Radon transform can be permutted with the convolution over xx. From eq. (26), one sees that the analytic properties of H\mathcal{H} are a direct consequence of the analytic properties of TT. And in fact these properties are the same, since both H\mathcal{H} and TT describe the scattering of two particles. T(z)T(z) is analytic for ∣z∣<1|z| < 1 and in our case for ∣ξ∣>1|\xi| > 1:

−1<α−∣βξ∣≤α+βξ≤α+∣βξ∣<1,(28)-1 < \alpha-\left|\frac{\beta}{\xi}\right| \leq\alpha+\frac{\beta}{\xi} \leq\alpha+\left|\frac{\beta}{\xi}\right| < 1, \tag*{(28)}

as the support of the DD is limited to {(α,β)∣∣α∣+∣β∣≤1}\{(\alpha,\beta)||\alpha|+|\beta|\leq1\}. Consequently, for every value of α\alpha in the DD support, T(α+β/ξ)T(\alpha+\beta/\xi) is analytic in the unphysical region (see Figure 4), and thus Taylor expanded around α\alpha into:

Figure 4. The range of the argument inside function TT

T(α+βξ)=∑n=0∞1n!∂nT(∂α)n(α)(βξ)n,for ∣ξ∣>1.(29)T\left(\alpha+\frac{\beta}{\xi}\right)=\sum_{n=0}^{\infty}\frac{1}{n!}\frac{\partial^{n}T}{(\partial\alpha)^{n}}(\alpha)\left(\frac{\beta}{\xi}\right)^{n},\qquad\text{for }|\xi|>1. \tag*{(29)}

Injecting eq. (29) into eq. (26), we can compute the coefficients introduced in eq. (25) in terms of moments of the DD ff and derivatives of the coefficient function TT:

H(ξ)=∑j=0∞hj1ξjfor ∣ξ∣>1,(30)\mathcal{H}(\xi)=\sum_{j=0}^{\infty}h_j\frac{1}{\xi^j}\quad\text{for }|\xi|>1, \tag*{(30)}
h0=∫−11Tq(ω)Dq(ω) dω(31)h_0=\int_{-1}^{1}T^{q}(\omega)D^{q}(\omega)\,\mathrm{d}\omega \tag*{(31)}
hj+1=1j!∫−11dα ∂jT(∂α)j(α)∫−1+∣α∣1−∣α∣βjFq(β,α) dβ(32)h_{j+1}=\frac{1}{j!}\int_{-1}^{1}\mathrm{d}\alpha\,\frac{\partial^{j}T}{(\partial\alpha)^{j}}(\alpha)\int_{-1+|\alpha|}^{1-|\alpha|}\beta^{j}F^{q}(\beta,\alpha)\,\mathrm{d}\beta \tag*{(32)}

This completes our first goal, but these results are restricted to the unphysical region. One needs to use the dispersion relation to derive relations between these coefficients in the physical region. To do this, we define the following In\mathcal{I}_n integrals in the complex plane:

In(ξ)=∮ΓRH(ξ′)ξ′−ξ(ξ′ξ)ndξ′=∑jhj∮ΓRξ′−jξ′−ξ(ξ′ξ)ndξ′(33)\mathcal{I}_n(\xi)=\oint_{\Gamma_R}\frac{\mathcal{H}(\xi')}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' =\sum_j h_j\oint_{\Gamma_R}\frac{\xi'^{-j}}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' \tag*{(33)}

where the contour ΓR\Gamma_R is illustrated on Figure 5 and chosen such that the CFF is analytic all along. Note that in eq. (33), ξ′\xi' is in the unphysical region, allowing us to expand the CFF H\mathcal{H}, but ξ\xi can be safely choosen in the physical region. Indeed, taking ξ∈[−1,1]\xi\in[-1,1], one gets:

Figure 5. The contours used in the proof of dispersion relations. In red, the singularities and branch cuts.

∮ΓRξ′−kξ′−ξdξ′={0if k>0,2iπξ−kif k≤0(34)\oint_{\Gamma_R}\frac{\xi'^{-k}}{\xi'-\xi}\mathrm{d}\xi'= \begin{cases} 0 & \text{if } k>0,\\ 2i\pi\xi^{-k} & \text{if } k\leq0 \end{cases} \tag*{(34)}

as in the case k>0k>0, the contributions of the two poles exactly compensate each other. We deduce:

In(ξ)=2πi∑j=0nhj1ξj,(35)\mathcal{I}_n(\xi)=2\pi i\sum_{j=0}^{n}h_j\frac{1}{\xi^j}, \tag*{(35)}

showing that the In\mathcal{I}_n truncate eq. (30) to order nn. The next step is to connect In\mathcal{I}_n with the value of the CFF within the physical region. To do so, we can deform the contour integration from ΓR\Gamma_R to Γ′\Gamma' such that:

In(ξ)=∫−11H(ξ′−iϵ)ξ′−ξ−iϵ(ξ′ξ)ndξ′−∫−11H(ξ′+iϵ)ξ′−ξ+iϵ(ξ′ξ)ndξ′=∫−11H(ξ′−iϵ)ξ′−ξ(ξ′ξ)ndξ′+iπH(ξ−iϵ)−∫−11H(ξ′+iϵ)ξ′−ξ(ξ′ξ)ndξ′+iπH(ξ+iϵ)(36)\begin{aligned} \mathcal{I}_n(\xi)&=\int_{-1}^{1}\frac{\mathcal{H}(\xi'-i\epsilon)}{\xi'-\xi-i\epsilon}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' -\int_{-1}^{1}\frac{\mathcal{H}(\xi'+i\epsilon)}{\xi'-\xi+i\epsilon}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi'\\ &=\int_{-1}^{1}\frac{\mathcal{H}(\xi'-i\epsilon)}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' +i\pi\mathcal{H}(\xi-i\epsilon) -\int_{-1}^{1}\frac{\mathcal{H}(\xi'+i\epsilon)}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' +i\pi\mathcal{H}(\xi+i\epsilon) \tag*{(36)} \end{aligned}

where we used the Sokhotski–Plemelj formula. The Schwartz reflexion principle in eq. (19) allows us to rewrite the formula in terms real and imaginary parts of the CFF such that:

In(ξ)=2iπℜH(ξ)+2i∫−11ℑH(ξ′)ξ′−ξ(ξ′ξ)ndξ′(37)\mathcal{I}_n(\xi)=2i\pi\Re\mathcal{H}(\xi)+2i\int_{-1}^{1}\frac{\Im\mathcal{H}(\xi')}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi' \tag*{(37)}

after safely taking the limit ϵ→0+\epsilon\to0^+. An additional subtlety comes from the fact that ℑ[H]\Im[\mathcal{H}] is defined as the limit coming from the upper half-plane (see eq. (18)) using Mandelstam variables, and thus ν\nu. Since ξ=1/ν\xi=1/\nu, we have:

ℑH=lim⁡ϵ→0+H(ν+iϵ)=lim⁡ϵ→0+H(ξ−iϵ),(38)\Im\mathcal{H}=\lim_{\epsilon\to0^+}\mathcal{H}(\nu+i\epsilon)=\lim_{\epsilon\to0^+}\mathcal{H}(\xi-i\epsilon), \tag*{(38)}

triggering a plus sign in front of integral.

One should realise that the result in eq. (37) is conditioned to the behaviour of H\mathcal{H} within the integration contour, and in particular for ξ→0\xi\to0. Indeed the contour deformation of eq. (36) can be performed only if the integrand remain integrable on the contour, especially for ξ→0\xi\to0 (or equivalently ν→∞\nu\to\infty). We expect the CFF to present a Regge behaviour in ξ−α\xi^{-\alpha} for ξ→0\xi\to0. Then expressions (36) and (37) for In\mathcal{I}_n are only valid for n>α−1n>\alpha-1.

Combining eqs. (35) and (37) we can deduce the general expression for the nn-times subtracted dispersion relation at any order of perturbation theory:

ℜH(ξ)+1π∫−11ℑH(ξ′)ξ′−ξ(ξ′ξ)ndξ′=∑j=0nhj1ξj,(39)\Re\mathcal{H}(\xi)+\frac{1}{\pi}\int_{-1}^{1}\frac{\Im\mathcal{H}(\xi')}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^n\mathrm{d}\xi'=\sum_{j=0}^{n}h_j\frac{1}{\xi^j}, \tag*{(39)}

with hjh_j being given in eqs. (31) and (32). It is easy to verify that h0h_0 is in fact the subtraction constant SQS^\mathcal{Q} we introduced in eq. (11).

New expression for the subtraction constant and consistency with previous results

Eq. (39) is the key result of this section allowing us to connect the real and imaginary parts of the CFF to the D-term and the Double Distribution. Yet, it can be simplified using the symmetries of DVCS and GPDs. Importantly, the CFFs are ξ\xi even and thus H(ξ)=H(−ξ)\mathcal{H}(\xi)=\mathcal{H}(-\xi). When combining this parity argument with the Schwarz reflection principle H(ξ∗)=H∗(ξ)\mathcal{H}(\xi^*)=\mathcal{H}^*(\xi) we obtain the following constraint:

H(−ξ∗)=H∗(ξ).(40)\mathcal{H}(-\xi^{*})=\mathcal{H}^{*}(\xi). \tag*{(40)}

From this we deduce that the real part of the CFF must be even in ξ\xi, while the imaginary one must be odd along the real axis, i.e. as a function of the real part of ξ\xi. This restricts our expansion in eq. (30) to

H(ξ)=∑j=0j even∞hj1ξj.(41)\mathcal{H}(\xi)=\sum_{\substack{j=0\\ j\ \mathrm{even}}}^{\infty}h_j\frac{1}{\xi^j}. \tag*{(41)}

As a consequence, restricting ourselves to ξ∈]0,1[\xi\in]0,1[, the dispersion relation (39) can be simplified into:

∑j=0j evennhjξ−j=ℜH(ξ)−1π∫01ℑH(x)(xξ)n[1ξ−x−(−1)n1ξ+x]dx(42)\begin{aligned} \sum_{\substack{j=0\\ j\ \mathrm{even}}}^{n}h_j\xi^{-j} &=\Re\mathcal{H}(\xi)\\ &-\frac{1}{\pi}\int_0^1\mathop{\Im}\mathcal{H}(x)\left(\frac{x}{\xi}\right)^n\left[\frac{1}{\xi-x}-(-1)^n\frac{1}{\xi+x}\right]\mathrm{d}x \tag*{(42)} \end{aligned}
=ℜH(ξ)−2π∫01(xξ)nxℑH(x)(ξ−x)(ξ+x)dx(43)=\Re\mathcal{H}(\xi)-\frac{2}{\pi}\int_0^1\left(\frac{x}{\xi}\right)^n\frac{x\mathop{\Im}\mathcal{H}(x)}{(\xi-x)(\xi+x)}\mathrm{d}x \tag*{(43)}

where k=2[n2]k=2\left[\frac{n}{2}\right] is the largest even number inferior or equal to nn. n=0n=0 and n=1n=1 are therefore equal, and exactly identical to the usual formula in eq. (11). For phenomenological CFFs, for which xℑH(x)x\mathop{\Im}\mathcal{H}(x) is integrable, this expression converges. Note that the terminology can be misleading. SS is usually called the subtraction constant, while we show here that it is actually extracted from an unsubtracted dispersion relation.

On top of eq. (27) for the quarks contribution, the above discussion can be generalised to the gluons contribution with:

Sg=∫−11Tg(ω)Dg(ω)dω.(44)S^g=\int_{-1}^{1}T^g(\omega)D^g(\omega)\mathrm{d}\omega. \tag*{(44)}

Because of the Schwartz principle, h0h_0 (and in fact all the hjh_j) is real, which means that only the real part of Tq,gT^{q,g} contribute to eq. (27) and (44). We are thus left with the following results:

Sa=∫−11ℜTa(ω)Da(ω)dω.(45)S^a=\int_{-1}^{1}\Re T^a(\omega)D^a(\omega)\mathrm{d}\omega. \tag*{(45)}

Eq. (45) can be recovered from the results presented in ref. [52] and recalled in (14). There, the subtraction constant is expressed as a double convolution involving the DD-term and the imaginary part of the perturbative kernel. Reshuffling the expression using the odd-parity of the DD-term:

Sq=2π∫1∞dωℑTq(ω)∫−11dαDq(α)ω−α=∫−11dαDq(α)1π∫1∞dωℑTq(ω)[1ω−α−1ω+α](46)\begin{aligned} S^q&=\frac{2}{\pi}\int_1^\infty\mathrm{d}\omega\mathop{\Im}T^q(\omega)\int_{-1}^{1}\mathrm{d}\alpha\frac{D^q(\alpha)}{\omega-\alpha}\\ &=\int_{-1}^{1}\mathrm{d}\alpha D^q(\alpha)\frac{1}{\pi}\int_1^\infty\mathrm{d}\omega\mathop{\Im}T^q(\omega)\left[\frac{1}{\omega-\alpha}-\frac{1}{\omega+\alpha}\right] \tag*{(46)} \end{aligned}

and injecting now the dispersion relation of the hard scattering kernel obtained in [52]:

ℜTq(ν)=1π∫1∞dωℑTq(ω)[1ω−ν−1ω+ν],(47)\Re T^q(\nu)=\frac{1}{\pi}\int_1^\infty\mathrm{d}\omega\mathop{\Im}T^q(\omega)\left[\frac{1}{\omega-\nu}-\frac{1}{\omega+\nu}\right], \tag*{(47)}

we recover our eq. (45).

Higher subtractions and connection with the DDs

An unexpected result of this new derivation of the higher-order connection between the subtraction constant and the DD-term, is that the dispersion relation formalism allows us to connect the Mellin moments of the Double Distribution with higher-order subtraction constants. Indeed, going back to eq. (30)-(32), one realises that we have only exploited the connection provided by eq. (31). It is possible to isolate the hjh_j for j≠0j\ne0 by subtracting two consecutive terms in eq. (42) (we recall that ξ∈]0,1[\xi\in]0,1[):

hnξ−n=1π∫01ℑH(ξ′)ξ′−ξ(ξ′ξ)n−1(ξ′ξ−1)dξ′+(−1)nπ∫01ℑH(ξ′)ξ′+ξ(ξ′ξ)n−1(ξ′ξ+1)dξ′.(48)\begin{aligned} h_n\xi^{-n} &=\frac{1}{\pi}\int_0^1\frac{\mathop{\Im}\mathcal{H}(\xi')}{\xi'-\xi}\left(\frac{\xi'}{\xi}\right)^{n-1}\left(\frac{\xi'}{\xi}-1\right)\mathrm{d}\xi'\\ &\quad+\frac{(-1)^n}{\pi}\int_0^1\frac{\mathop{\Im}\mathcal{H}(\xi')}{\xi'+\xi}\left(\frac{\xi'}{\xi}\right)^{n-1}\left(\frac{\xi'}{\xi}+1\right)\mathrm{d}\xi'. \tag*{(48)} \end{aligned}

After simplification, this can be simply written for j≥1j\ge1 as:

h2j=2π∫01ℑH(ξ′)(ξ′)2j−1dξ′.(49)h_{2j}=\frac{2}{\pi}\int_0^1\mathop{\Im}\mathcal{H}(\xi')(\xi')^{2j-1}\mathrm{d}\xi'. \tag*{(49)}

Reinjecting eq. (32), we get for ℓ\ell odd:

2π∫01ℑH(ξ′)(ξ′)ℓdξ′=1ℓ!∫−11dα∂ℓT∂(α)ℓ(α)∫−1+∣α∣1−∣α∣dβ βℓfq(β,α)(50)\frac{2}{\pi}\int_0^1\mathop{\Im}\mathcal{H}(\xi')(\xi')^{\ell}\mathrm{d}\xi' =\frac{1}{\ell!}\int_{-1}^{1}\mathrm{d}\alpha\frac{\partial^\ell T}{\partial(\alpha)^\ell}(\alpha)\int_{-1+|\alpha|}^{1-|\alpha|}\mathrm{d}\beta\,\beta^\ell f^q(\beta,\alpha) \tag*{(50)}

Rewording this equation, we note that the ℓth\ell^{\mathrm{th}} Mellin moment of the imaginary part of the CFF is connected with the β\beta-moment of the Double Distribution, convoluted then with the derivative of the hard scattering kernel. This equation highlights well the deconvolution problem of GPDs and DDs from DVCS data [38]. Indeed, while two indices are required to independently deconvolve the (α,β)(\alpha,\beta) dependence of the DDs, only a single index, ℓ\ell, appears here. The impact of such relation on the characterisation of shadow GPDs is left for future work.

Finally, let us mention that the fact of implementing dispersion relation at the level of the CFF directly may have a significant impact already at the level of CFF extraction (see Figure 2 of [62] for an illustration of the impact of the one-subtracted dispersion relation for DVCS). Assessing the impact of higher-subtracted dispersion relations on CFF extraction is also left for future studies.

Extraction of the pressure distribution on collider kinematics: an inverse problem

If we limit ourselves to the lowest subtraction, which only involves the DD-term, we have demonstrated that the CFFs give us in principle access to the subtraction constant:

Sa(t,Q2)=∫−11dω ReTa(ω,Q2μ2,αs)Da(ω,t,μ2).(51)S^{a}(t,Q^{2})=\int_{-1}^{1}\mathrm{d}\omega\,\mathop{\mathrm{Re}}\mathcal{T}^{a}\left(\omega,\frac{Q^{2}}{\mu^{2}},\alpha_{s}\right)D^{a}(\omega,t,\mu^{2}) . \tag*{(51)}

We will elaborate on the challenges related to the characterization of SaS^{a} from experimental data in the next section. For now, we are interested in the fact that the GFF Ca(t,μ2)C_{a}(t,\mu^{2}), which is related to the pressure distribution in the proton, writes as another integral of the DD-term, eq. (7) which we recall with its full variable dependence:

Ca(t,μ2)=14∫−11dα α1−paDa(α,t,μ2).(52)C_{a}(t,\mu^{2})=\frac{1}{4}\int_{-1}^{1}\mathrm{d}\alpha\,\alpha^{1-p_{a}}D^{a}(\alpha,t,\mu^{2}) . \tag*{(52)}

An obvious question is whether the knowledge of the subtraction constant Sa(t,Q2)S^{a}(t,Q^{2}) allows an unambiguous extraction of the GFF Ca(t,μ2)C_{a}(t,\mu^{2}). This question is very similar to that known as the deconvolution problem [38], which aims at determining whether the measurement of CFFs – i.e. the convolution of the perturbative coefficient function to the full GPD – allows the unambiguous reconstruction of the GPD. In fact, the problem at hand in this paper is exactly the restriction of the general deconvolution problem to the DD-term.

As we have already hinted at, the root of the deconvolution problem is that DVCS experimental data offers one less kinematic variable compared to the parton distributions we want to extract. CFFs are functions of (ξ,t,Q2)(\xi,t,Q^{2}), whereas GPDs are functions of (x,ξ,t,μ2)(x,\xi,t,\mu^{2}); the subtraction constant is a function of (t,Q2)(t,Q^{2}) whereas the DD-term of (α,t,μ2)(\alpha,t,\mu^{2}). One might argue that the GFF CaC_{a} which we are fundamentally interested in, is just a function of (t,μ2)(t,\mu^{2}) – so a similar kinematic dependence as the subtraction constant Sa(t,Q2)S^{a}(t,Q^{2}). However, one cannot write a straightforward relation between the subtraction constant Sa(t,Q2)S^{a}(t,Q^{2}) and the GFF Ca(t,μ2)C_{a}(t,\mu^{2}) which does not involve in practice the extraction of the DD-term Da(α,t,μ2)D^{a}(\alpha,t,\mu^{2}).

There is however a theoretical solution to the missing variable problem. At a given order in perturbation theory, the scale dependence of Da(α,t,μ2)D^{a}(\alpha,t,\mu^{2}) is given by renormalization group equations, removing in principle one degree of freedom. In practice, evolution equations entangle the (α,μ2)(\alpha,\mu^{2}) dependence of the DD-term (and the (x,ξ,μ2)(x,\xi,\mu^{2}) dependence of GPDs). However, although this solves the issue on paper, in practice, effect of QCD evolution are rather weak on the range of Q2Q^{2} accessible to exclusive processes. Ref. [38] offered an explicit construction of very different GPDs (with vanishing DD-terms) such that their CFFs would be indiscernible to experimental data. These shadow GPDs represent an illustration of particularly badly constrained objects to DVCS in any foreseeable data. Solutions to this issue involve, on the one hand, the introduction of more theoretical constraints to reduce the functional space accessible to GPDs [39], and on the other hand an ambitious program of global fits on a variety of exclusive processes. In particular, processes which do not show an exacerbated sensitivity to a pole x=ξx=\xi as DVCS, TCS or DVMP, but rather to a pole where xx and ξ\xi are entangled to an external kinematic variable are very desirable, like DDVCS [63, 64, 65] or two-or-three exclusive processes [23, 66, 24, 67, 25, 26, 68, 27].

In a similar fashion to the study of the deconvolution problem led in Ref. [38], there exist shadow DD-terms, which bring barely any contribution to the subtraction constant over current ranges in Q2Q^{2} and are therefore extremely hard to discern in the data. Ref. [7] gave a hint at such shadow DD-term when it highlighted a tremendous increase of uncertainty as soon as the parametrization of the DD-term was made slightly more flexible. In practice, it is common to parametrize the DD-term through an expansion in Gegenbauer moments due to their friendly LO evolution properties:

Dq(α,t,μ2)=(1−α2)∑nodddnq(t,μ2)Cn(3/2)(α),(53)D^{q}(\alpha,t,\mu^{2})=(1-\alpha^{2})\sum_{\substack{n\\\mathrm{odd}}}d_{n}^{q}(t,\mu^{2})C_{n}^{(3/2)}(\alpha), \tag*{(53)}
Dg(α,t,μ2)=32(1−α2)2∑nodddng(t,μ2)Cn−1(5/2)(α).(54)D^{g}(\alpha,t,\mu^{2})=\frac{3}{2}(1-\alpha^{2})^{2}\sum_{\substack{n\\\mathrm{odd}}}d_{n}^{g}(t,\mu^{2})C_{n-1}^{(5/2)}(\alpha). \tag*{(54)}

As Gegenbauer polynomials form a complete orthogonal family, this representation is fairly general – but comes with the drawback that fixed-order truncations are usually oscillating functions. We refer to Ref. [7] for an account of the LO scale dependence of dna(t,μ2)d_{n}^{a}(t,\mu^{2}). In this representation, the LO subtraction constant reads as:

S(t,Q2)LO=4∑qeq2∑nodddnq(t,μ2),(55)S(t,Q^{2})^{\mathrm{LO}}=4\sum_{q}e_{q}^{2}\sum_{\substack{n\\\mathrm{odd}}}d_{n}^{q}(t,\mu^{2}), \tag*{(55)}

where we will use conventionally in the following μ2≡Q2\mu^{2}\equiv Q^{2}. On the other hand, the GFF Ca(t,μ2)C_{a}(t,\mu^{2}) reads:

Ca(t,μ2)=15d1a(t,μ2).(56)C_{a}(t,\mu^{2})=\frac{1}{5}d_{1}^{a}(t,\mu^{2}) . \tag*{(56)}

The problem of relating the subtraction constant to the pressure inside the proton turns into the question of extracting d1ad_{1}^{a} from the sum of all dnad_{n}^{a} at LO (and more complicated infinite linear combinations of the dnad_{n}^{a} at higher order). A simple solution to the ill-definedness of this extraction is to assume that only a finite number of coefficients dnad_{n}^{a} actually contribute to the subtraction constant. In fact, the study of [5] used only n=1n=1, and evaluated the systematic uncertainty caused by such a rigid modelling of the DD-term by inputs from a quark soliton model. In [7], effects of a truncation at n=1n=1 and n=3n=3 were compared. It was observed that the – already large – uncertainty on d1d_1 inflated by a factor 20 when d3d_3 was allowed to be non-zero. In fact, the reason is fairly simple to understand. Since evolution effects are relatively small on the narrow range in Q2Q^2 available to the current precise DVCS data, d1(t,μ2)d_1(t,\mu^2) and d3(t,μ2)d_3(t,\mu^2) do not exhibit a significantly different behavior in μ\mu. Therefore, parasitic contributions such that

d1q(t,μ2)≈−d3q(t,μ2)(57)d_1^q(t,\mu^2) \approx-d_3^q(t,\mu^2) \tag*{(57)}

amount to almost no contribution to the LO subtraction constant of eq. (55), and are virtually unconstrained. An object which brings exactly no contribution to the subtraction constant at a given scale μ02\mu_0^2 will be called a shadow D-term, and we have already highlighted a very simple example at LO:

d1q(μ02)=λ ;d3q(μ02)=−λ ,(58)d_1^q(\mu_0^2)=\lambda\,;\quad d_3^q(\mu_0^2)=-\lambda\,, \tag*{(58)}

or equivalently

DS,LOq(α,μ02)=λ(1−α2)[C1(3/2)(α)−C3(3/2)(α)].(59)D_{S,\mathrm{LO}}^q(\alpha,\mu_0^2)=\lambda(1-\alpha^2)\left[C_1^{(3/2)}(\alpha)-C_3^{(3/2)}(\alpha)\right]. \tag*{(59)}

The space of shadow DD-terms at a fixed scale is a vector space (it is the kernel, or null-space of the integral transform and there exist shadow DD-terms of arbitrary size). Under evolution to another scale μ2≠μ02\mu^2\ne\mu_0^2, the contribution of a shadow DD-term to the subtraction constant becomes non-zero. Indeed, (58) can only be true at one scale, since the μ2\mu^2 dependence of both sides of the equation are ruled by different anomalous dimensions. Therefore, the range in scales on which DVCS is measured precisely directly constrains the maximal size of shadow DD-terms, and the uncertainty of the deconvolution procedure.

To give a simple approximate example, if there were no mixing between quarks and gluons, the evolution of dnqd_n^q would be entirely dictated by the anomalous dimension γn\gamma_n following

dnq(μ2)=Γnqq(μ2,μ02)dnq(μ02),d_n^q(\mu^2)=\Gamma_n^{qq}(\mu^2,\mu_0^2)d_n^q(\mu_0^2),

where

Γnqq(μ2,μ02)=(αs(μ2)αs(μ02))2γn/β0,(60)\Gamma_n^{qq}(\mu^2,\mu_0^2)=\left(\frac{\alpha_s(\mu^2)}{\alpha_s(\mu_0^2)}\right)^{2\gamma_n/\beta_0}, \tag*{(60)}

where β0\beta_0 is the first coefficient in the β\beta function of αs\alpha_s. Using γ1=16/9\gamma_1=16/9, γ3=157/45\gamma_3=157/45 and β0=11−2nf/3=9\beta_0=11-2n_f/3=9, we find that the contribution to the subtraction constant of the simple shadow DD-term of eq. (58) is:

SSq(Q2)=83(Γ1qq(Q2,μ02)d1q(μ02)+Γ3qq(Q2,μ02)d3q(μ02)),(61)S_S^q(Q^2)=\frac{8}{3}\left(\Gamma_1^{qq}(Q^2,\mu_0^2)d_1^q(\mu_0^2)+\Gamma_3^{qq}(Q^2,\mu_0^2)d_3^q(\mu_0^2)\right), \tag*{(61)}
≈83λ[(αs(Q2)αs(μ02))0.395−(αs(Q2)αs(μ02))0.775].(62)\approx\frac{8}{3}\lambda\left[\left(\frac{\alpha_s(Q^2)}{\alpha_s(\mu_0^2)}\right)^{0.395}-\left(\frac{\alpha_s(Q^2)}{\alpha_s(\mu_0^2)}\right)^{0.775}\right]. \tag*{(62)}

This gives of course 0 if Q2=μ02Q^2=\mu_0^2 by definition of the shadow DD-term. Linearizing the previous relation yields this approximate contribution of the shadow DD-term to the subtraction constant:

SSq(Q2)≈λ[1−αs(Q2)αs(μ02)].(63)S_S^q(Q^2)\approx\lambda\left[1-\frac{\alpha_s(Q^2)}{\alpha_s(\mu_0^2)}\right]. \tag*{(63)}

If the experimental uncertainty of the subtraction constant is characterized by a quantity ΔS\Delta S, and the measurement have been performed on a range in scales of [Qmin⁡2,Qmax⁡2][Q^2_{\min},Q^2_{\max}], our approximate approach tells that the shadow DD-term of eq. (58) will be typically bring a dispersion:

σS,d1q≈σS,d3q≈38ΔSΓ1qq(Qmax⁡2,Qmin⁡2)−Γ3qq(Qmax⁡2,Qmin⁡2).(64)\begin{aligned} \sigma_{S,d_1^q}\approx\sigma_{S,d_3^q} \\ &\approx\frac{3}{8}\frac{\Delta S}{\Gamma_1^{qq}(Q^2_{\max},Q^2_{\min})-\Gamma_3^{qq}(Q^2_{\max},Q^2_{\min})}. \tag*{(64)} \end{aligned}
≈ΔS(1−αs(Qmax⁡2)αs(Qmin⁡2)).(65)\approx\frac{\Delta S}{\left(1-\frac{\alpha_s(Q^2_{\max})}{\alpha_s(Q^2_{\min})}\right)}. \tag*{(65)}

The approximate form of the last line should only be used for scales close to the charm mass, as it is derived with the anomalous dimensions γf=3\gamma_f=3. (64) is general on the other hand, provided the true evolution operator is used. Of course, there exist many more shadow DD-terms if the parametrization in terms of dnd_n is made more flexible than solely d1qd_1^q and d3qd_3^q, in particular if explicit gluons contributions are included. The simple eq. (65) represents a typical estimate of the uncertainty of the deconvolution procedure within the parametric space which we have chosen. Despite the simplifying assumptions that we have made, this result captures the essence of the propagation of uncertainty: first a dependence on the experimental uncertainty of the data through ΔS\Delta S, and then a characterization of how different the evolution of the different parameters is with respect to the scale.

Formally, since the fit of the dnqd_n^q coefficients is linear, it is straightforward to write the exact solution of the fit. In particular, the covariance matrix of d1qd_1^q and d3qd_3^q that we are interested in for this simple example writes:

\begin{pmatrix} \sigma_{d_1^q}^2 & \operatorname{cov}[d_1^q,d_3^q]\\ \operatorname{cov}[d_1^q,d_3^q] & \sigma_{d_3^q}^2 \end{pmatrix} =(C^T\Omega^{-1}C)^{-1}, \tag*{(66)}\end{pmatrix}

where Ω\Omega is the covariance matrix of the fitted dataset and CC is the so-called design matrix which contains the values of the fitted functions at the fitted kinematics, here:

C=83(Γ1qq(Q12,μ02)Γ3qq(Q12,μ02)Γ1qq(Q22,μ02)Γ3qq(Q22,μ02)⋮⋮).(67)C=\frac{8}{3} \begin{pmatrix} \Gamma_1^{qq}(Q_1^2,\mu_0^2) & \Gamma_3^{qq}(Q_1^2,\mu_0^2)\\ \Gamma_1^{qq}(Q_2^2,\mu_0^2) & \Gamma_3^{qq}(Q_2^2,\mu_0^2)\\ \vdots& \vdots \end{pmatrix}. \tag*{(67)}

One can draw a parallel between this exact general formula and the approximate shadow DD-term uncertainty that we have derived in eq. (65). If there were only two measurements in Q2Q^2, one at Qmin⁡2Q_{\min}^2 and one at Qmax⁡2Q_{\max}^2, we chose to evaluate μ02=Qmin⁡2\mu_0^2=Q_{\min}^2, and the experimental dataset was uncorrelated with standard deviation ΔS\Delta S, we would find:

σd1q=38ΔS1+[Γ1qq(Qmax⁡2,Qmin⁡2)]2Γ1qq(Qmax⁡2,Qmin⁡2)−Γ3qq(Qmax⁡2,Qmin⁡2)(68)\sigma_{d_1q} = \frac{3}{8}\Delta S \frac{\sqrt{1+[\Gamma^{qq}_{1}(Q^2_{\max},Q^2_{\min})]^2}}{\Gamma^{qq}_{1}(Q^2_{\max},Q^2_{\min})-\Gamma^{qq}_{3}(Q^2_{\max},Q^2_{\min})} \tag*{(68)}
σd3q=38ΔS1+[Γ1qq(Qmax⁡2,Qmin⁡2)]2Γ1qq(Qmax⁡2,Qmin⁡2)−Γ3qq(Qmax⁡2,Qmin⁡2)(69)\sigma_{d_3q} = \frac{3}{8}\Delta S \frac{\sqrt{1+[\Gamma^{qq}_{1}(Q^2_{\max},Q^2_{\min})]^2}}{\Gamma^{qq}_{1}(Q^2_{\max},Q^2_{\min})-\Gamma^{qq}_{3}(Q^2_{\max},Q^2_{\min})} \tag*{(69)}

Using that Γqq(Qmax⁡2,Qmin⁡2)<1\Gamma^{qq}(Q^2_{\max},Q^2_{\min})<1, we find a very similar estimate to the one we have derived using the notion of shadow DD-term, without needing the concept at all. Shadow DD-terms are after all merely an attempt at simplifying, or making more intuitive, the analysis of the inverse linear problem by identifying obvious directions that are dominant in the uncertainty propagation. It is truly useful when it comes to making broad predictions based on general characteristics of the data without going through the process of generating pseudo-data and fitting them. We show such exercise to broadly evaluate the plausible impact of the EIC in the last section of this paper, where we will construct NLO shadow DD-terms and treat the evolution equations properly.

Let us note in passing that, with the same assumptions that were used to derive eqs. (68)–(69), we find:

corr⁡[d1q,d3q]=−1−Γ1qqΓ3qq(1+[Γ1qq]2)(1+[Γ3qq]2)(70)\operatorname{corr}[d^{q}_{1},d^{q}_{3}] = \frac{-1-\Gamma^{qq}_{1}\Gamma^{qq}_{3}}{\sqrt{(1+[\Gamma^{qq}_{1}]^2)(1+[\Gamma^{qq}_{3}]^2)}} \tag*{(70)}

where we have omitted the argument (Qmax⁡2,Qmin⁡2)(Q^2_{\max},Q^2_{\min}). If evolution is very weak, Γ1qq≈Γ3qq≈1\Gamma^{qq}_{1}\approx\Gamma^{qq}_{3}\approx1, which gives as expected corr⁡[d1q,d3q]≈−1\operatorname{corr}[d^{q}_{1},d^{q}_{3}]\approx-1, with σd1q≈σd3q≈+∞\sigma_{d_1q}\approx\sigma_{d_3q}\approx+\infty.

Results on current experimental data

We now conduct a re-analysis of the LO extraction of the GFF Ca(t)C_a(t) of Ref. [7] using the NLO DVCS coefficient function, and putting to use the understanding of the deconvolution uncertainty stemming from shadow DD-terms. A neural network analysis of the global DVCS dataset was conducted in 2019 [53], leveraging 30 observables over 2500 kinematic configurations acquired during 17 years of measurements. The real and imaginary parts of the four leading-twist CFFs were modelled independently. The result of the fit was 100 sets of CFFs which represent a sample of the functional distribution of the CFFs.

The computation of the subtraction constant from eq. (11) requires the evaluation of the imaginary part of the CFF on the full range of ξ∈]0,1[\xi\in]0,1[. Since the skewness ξ\xi is related to the plus component of the four-momentum transfer Δ\Delta, it is bound kinematically by the value of tt according to:

∣ξ∣≤−t−t+4Mp2(71)|\xi| \le\frac{\sqrt{-t}}{\sqrt{-t+4M_p^2}} \tag*{(71)}

where MpM_p is the proton mass. It means that part of the integral must be evaluated over a domain where the CFF is continued analytically and where experimental measurements are impossible. This does not represent a theoretical issue per se. For instance, the Double Distribution can be characterized from a limited range in ξ\xi as highlighted in [69], and then used to construct the CFF in the range where it is not measured. Likewise, the extraction of GPDs on the lattice, which is performed in Euclidean space using a space-like definition of the kinematic variables gives access to any ξ\xi at any tt [70]. Yet, in the context of an analysis based on experimental data, the severe kinematic limitation on the information on the CFFs represents a challenge. The flexibility of the neural network parametrization attempts to introduce as little bias as possible in the analytic continuation of the CFFs outside of their experimental determination.

From the 100 sets of CFFs stemming from the neural network analysis, we compute 100 samples of the functional distribution of the subtraction constant which we will call replicas in the following. The result at one kinematic point (t,Q2)(t,Q^2) as a function of ξ\xi is presented in Figure 6. We present the result using both the traditional sample standard deviation estimate:

Figure 6. Subtraction constant as a function of ξ\xi at a given value of (t,Q2)(t,Q^2) obtained from the dispersion relation applied to a neural network extraction of H\mathcal{H} from the world DVCS dataset in 2019. We show a subset of 50 replicas and the uncertainty computed using the ordinary sample standard deviation and the MAD robust estimator.

σsample=1N−1∑i=1N(Xi−mean⁡(X))2(72)\sigma_{\mathit{sample}} = \frac{1}{\sqrt{N-1}}\sqrt{\sum_{i=1}^{N}(X_i-\operatorname{mean}(X))^2} \tag*{(72)}

and using the outlier robust estimate of the standard deviation known as mean absolute deviation (MAD):

σMAD(X)=λmed⁡(∣X−med⁡(X)∣).(73)\sigma_{\mathit{MAD}}(X)=\lambda\operatorname{med}\left(\left|X-\operatorname{med}(X)\right|\right). \tag*{(73)}

med stands for the median, and λ=1/Φ−1(3/4)=1.4826\lambda= 1/\Phi^{-1}(3/4) = 1.4826 where Φ(x)\Phi(x) is the standard normal cumulative distribution function. This constant rescaling allows the MAD operator to coincide with the standard deviation in infinite statistics under the assumption that the distribution is Gaussian. The large difference between the two estimates in Figure 6 highlights the important contamination by outliers, that we elaborate on in the next paragraph.

Since the real and imaginary parts of the CFFs have been modelled independently without enforcing the connection between them induced by the polynomiality of GPDs, there is in principle no expectation that the subtraction constant will end up independent of ξ\xi. However, the result is indeed globally compatible with a constant. We exclude from the analysis the subtraction constant for ξ>0.5\xi> 0.5, as there seems to be a slight systematic downward shift of the subtraction constant at large ξ\xi. This may stem from the fact that Im⁡H(ξ)\operatorname{Im}\mathcal{H}(\xi) is not constrained to go to 0 as ξ→1\xi\to1, leading to slightly less behaved subtraction constant integrals in this limit. At the value of (tt, Q2Q^2) presented in Figure 6 – one of the most precise since it corresponds to a region well explored by the JLab 6 GeV data – the subtraction constant is still only one standard deviation away from 0 evaluated using outlier robust statistics. We present in Figure 7 the strength of the signal of the subtraction constant on the kinematic domain. The fact that neural network analyses of the DVCS dataset available before the JLab 12 GeV upgrade lead to subtraction constants which are compatible with 0 was also established in [6]. This results to a large extent from the poor constraints on the real part of the CFF H\mathcal{H} within the current experimental dataset. This highlights the interest of a better determination of this quantity, achievable for instance by measuring unpolarised beam charge asymmetry observables with a positron beam at JLab [71, 72].

Strength of the signal of the subtraction constant

Figure 7. Strength of the signal of the subtraction constant (robust estimates), expressed in number of standard deviations from 0. We use the most precise value of ξ\xi for each kinematic (tt, Q2Q^2), excluding ξ>0.5\xi> 0.5 (see text). Only the best kinematics allow a characterization at 1σ1\sigma from 0.

Treatment of outliers

As we have already noticed, our data suffers from noticeable outliers. Therefore, instead of the sample means, variances and covariances, we should use outlier robust estimates. In all this analysis, we will use the sample median instead of the sample mean and the MAD instead of the sample standard deviation. There exists a considerable literature devoted to the question of a robust estimate of the covariance / correlation matrix (see among the most popular suggestions [73, 74, 75]). We will use in this paper the straightforward generalization of the MAD estimator for the correlation, which we have not seen used before in the existing literature:

r~≡med⁡((X−med⁡(X))(Y−med⁡(Y)))med⁡((X−med⁡(X))2)med⁡((Y−med⁡(Y))2)(74)\widetilde{r} \equiv\frac{\operatorname{med}\left((X-\operatorname{med}(X))(Y-\operatorname{med}(Y))\right)}{\sqrt{\operatorname{med}\left((X-\operatorname{med}(X))^2\right)\operatorname{med}\left((Y-\operatorname{med}(Y))^2\right)}} \tag*{(74)}

A comparison of this estimator to the many others presented in the literature would divert us from the physics purpose of this paper, and is conducted in a separate more statistically focused paper [76]. We stick therefore to a succinct presentation. In effect, this estimator does not directly relate to the correlation r≡corr⁡[X,Y]=cov⁡[X,Y]/(σXσY)r \equiv\operatorname{corr}[X,Y] = \operatorname{cov}[X,Y]/(\sigma_X\sigma_Y): just like a factor λ\lambda was necessary to relate the MAD estimator to the standard deviation for a normal distribution, some procedure is needed to match (74) to r=corr⁡[X,Y]r = \operatorname{corr}[X,Y]. It is likely that there exist no closed-form matching formula in the case of a bivariate normal distribution. However as presented in [76] and demonstrated empirically in Figure 8, the following approximation is accurate beyond the per mille level and fully satisfactory considering the precision of this study:

Figure 8. Empirical measurements of r/rˉr/\bar{r} from extensive samplings of correlated normal distributions, and their fit by the functional form of eq. (75).

r/r~=1−0.5635ln⁡∣r∣(75)r/\widetilde{r} = 1 - 0.5635\ln|r| \tag*{(75)}

Isolating properly the correlation coefficient yields:

r=0.5635r~ W(10.47∣r~∣)(76)r = 0.5635\widetilde{r}\,W\left(\frac{10.47}{|\widetilde{r}|}\right) \tag*{(76)}

where WW is the Lambert-WW function, defined as the inverse function of f(W)=WeWf(W)=We^W. One should note that this definition does not guarantee that the resulting covariance / correlation matrix is positive definite. Only in the limit of infinite statistics of a true Gaussian distribution do we formally expect that this property is fulfilled. This may however not be as much of a drawback that it appears at first sight. The maximal size of negative eigenvalues in the spectrum provides a clear physical evidence of the magnitude of poorly estimated correlations. Instead of using the inverse of the covariance matrix to compute the χ2\chi^2, one could therefore use a singular value decomposition (SVD) with a threshold on the singular values larger than the absolute value of the negative eigenvalues.

To demonstrate practically the interest of our outlier resilient covariance matrix estimate, we deliberately create an outlier-ridden distribution, made of a mixture of a narrow normal distribution, and a fraction of samples from a wider normal distribution. The result is depicted in Figure 9 and shows that our robust estimators are less affected by the outliers. Note that in general, our robust estimators have a larger dispersion than the sample ones (in statistical terms, they are less efficient). In the example presented here, the dispersion of our robust operators is typically 20 to 30% larger than the sample ones. But although the sample operators are less dispersed, their expectation is further from the value of interest in presence of strong outliers.

High-statistics sample with 10% outliers

Figure 9. We generate a high-statistics sample with 10% of outliers. The median based estimators are less contaminated by the outliers than the ordinary sample estimators.

A model of the DD-term

As we have already noted, our data-driven neural network extraction of the subtraction constant is largely unconstrained as ∣t∣|t| and Q2Q^2 increase. We have stressed

in § 4 that the scale dependence is instrumental to perform a model independent extraction of the DD-term or the C(t)C(t) GFF from the experimental data. To proceed forward and obtain sensible results at large ∣t∣|t|, we are forced to use a model of DD-term. We choose the following usual strategy:

  • The scale dependence of the DD-term is given by the LO renormalization group equation resummed at leading-logarithmic accuracy. We use the LO running of αs\alpha_s with αs(MZ2)=0.118\alpha_s(M_Z^2) = 0.118 and threshold crossing at mc=1.27m_c = 1.27 GeV and mb=4.18m_b = 4.18 GeV.

  • We truncate the Gegenbauer expansion (54) to either n=1n = 1 or n=3n = 3, which allows us to probe a part of the shadow DD-term uncertainty as we have explained in the previous section.

  • We assume an equal contribution of the light quarks dnu=dnd=dns=dnudsd_n^u = d_n^d = d_n^s = d_n^{uds} and a purely radiatively generated charm contribution, that is dnc(μ02=mc2)=0d_n^c(\mu_0^2 = m_c^2) = 0.

  • We enforce a factorized tt-dependence under the form of a tripole Ansatz:

Da(t)=Da(t=0)(1−tM2)−3,(77)D^a(t) = D^a(t = 0) \left(1 - \frac{t}{M^2}\right)^{-3}, \tag*{(77)}

where M=0.8M = 0.8 GeV. The lack of distinctive tt-dependence in the CFF extraction makes this fully constrained Ansatz satisfactory.

Our model is therefore entirely defined by the coefficients dnuds(t=0,μ02)d_n^{uds}(t = 0,\mu_0^2) and dng(t=0,μ02)d_n^g(t = 0,\mu_0^2) at a conventionally fixed scale μ0=2\mu_0 = 2 GeV. We want to obtain the distribution of those parameters so that

∫−11dw Ta(w,αs(Q2))Σab(Q2,μ02)⊗Db(w,t,μ02)(78)\int_{-1}^{1} \mathrm{d}w\, T^a(w,\alpha_s(Q^2)) \Sigma^{ab}(Q^2,\mu_0^2) \otimes D^b(w,t,\mu_0^2) \tag*{(78)}

approximates as much as possible the distribution of the 100 replicas S(ξ,t,Q2)S(\xi,t,Q^2). Σab(Q2,μ02)\Sigma^{ab}(Q^2,\mu_0^2) represent the leading-logarithmic evolution operator of the DD-term. Implicit summation on the repeated indices is subtended.

The problem presents itself as finding the best fit of a target function by a parametrized function. A simple way to proceed is to sample the target function and perform a least-squares fit. However, the answer may depend on the choice of kinematics where the sample is performed. In order to perform a correlated fit, we need to select much fewer kinematic values than the number of replicas which are available. With 100 replicas, we will use NkinN_{\rm kin} kinematics selected because they represent the strongest signal of the subtraction constant. Precisely, we decompose the (ξ,t,Q2)(\xi,t,Q^2) phase-space in a regular grid with a logarithmic spacing in ξ\xi characterized by a multiplicative factor of 1.5, a uniform spacing in tt characterized by a pace of 0.1 GeV2^2, and a logarithmic spacing in Q2Q^2 (multiplicative factor of 1.35). Then we select the NkinN_{\rm kin} kinematics where the ratio of the median of the replicas by the MAD is the largest. This prevents from overfitting our model on a region where the neural network is left largely unconstrained.

Let’s call (ξi,ti,Qi2)(\xi_i,t_i,Q_i^2) the set of kinematics we have just described, on which the replicas of S(ξ,t,Q2)S(\xi,t,Q^2) are sampled. Once this choice is fixed, several options present themselves to determine the best parameters.

  1. The most natural option is to determine the distribution of our free parameters so as to minimize the correlated least squares:

∑i,i′(model⁡(ξi,ti,Qi2)−Sˉ(ξi,ti,Qi2))×cov⁡−1[S]i,i′×(model⁡(ξi′,ti′,Qi′2)−Sˉ(ξi′,ti′,Qi′2)).(79)\sum_{i,i'} \left(\operatorname{model}(\xi_i,t_i,Q_i^2)-\bar{S}(\xi_i,t_i,Q_i^2)\right)\times\operatorname{cov}^{-1}[S]_{i,i'}\\ \times\left(\operatorname{model}(\xi_{i'},t_{i'},Q_{i'}^2)-\bar{S}(\xi_{i'},t_{i'},Q_{i'}^2)\right). \tag*{(79)}

Sˉ\bar{S} is the sample median of the dataset and cov⁡[S]\operatorname{cov}[S] the robust covariance matrix. Since the fit is linear, the best-fit parameters are normally distributed and we do not need to use individual replicas.

  1. If we used an uncorrelated least-squares fit (only the diagonal terms of the covariance matrix), then we would not be limited in the number of kinematics where to perform the fit. However, neglecting outright the statistical information of correlation of the target function seems unjustified.

  1. The LO study in Ref. [7] used an hybrid approach: for each replica Sj(ξ,t,Q2)S_j(\xi,t,Q^2) where 1≤j≤1001\leq j\leq100 labels the replica, the best-fit value was found with an uncorrelated least squares:

∑i(model⁡(ξi,ti,Qi2)−Sj(ξi,ti,Qi2))2(ΔS(ξi,ti,Qi2))2(80)\sum_i \frac{\left(\operatorname{model}(\xi_i,t_i,Q_i^2)-S_j(\xi_i,t_i,Q_i^2)\right)^2}{\left(\Delta S(\xi_i,t_i,Q_i^2)\right)^2} \tag*{(80)}

where ΔS(ξi,ti,Qi2)\Delta S(\xi_i,t_i,Q_i^2) is the outlier robust standard deviation computed from the 100 replicas. The result is then made of the distribution of the best-fit value over each replica. This strategy can in principle be applied to an arbitrary number of kinematics, and yet encompasses part of the correlated information through the use of the distribution of replicas. In the absence of outliers distorting the distribution, if we used cov⁡−1[S]\operatorname{cov}^{-1}[S] instead of 1/(ΔS)21/(\Delta S)^2 in eq. (80), we would find exactly the first method, while if we used Sˉ\bar{S} instead of the individual SjS_j replicas, we would recover exactly the second method.

To appreciate the difference between the three methods independently from the question of the outlier suppression, we construct in Figure 10 a fictitious normal distribution of replicas (grey curves), sample it on some kinematics (blue points), and apply the different methods in the absence of outliers. The correlated fit (top plot) exhibits a smaller variance than the uncorrelated one (middle plot) as a reflection of the fact that the grey replicas exhibit a long-distance anti-correlation: the replicas that go down at small xx tend to go up at large xx and vice-versa. This has the effect of pinning down the best constant more precisely than when this information is simply neglected. The

Figure 10. Fit by a constant function of a functional distribution represented by the grey replicas. The fit is performed on the sampled kinematics represented by the blue points using 1) a fully correlated fit, 2) a fully uncorrelated fit and 3) the hybrid method of uncorrelated fit replica-by-replica.

hybrid method (bottom plot) exhibits a significantly larger uncertainty than the other two. In the following, we will present results using the correlated method. It is clear that the most reliable strategy would be to perform a full refit of the experimental data – which is outside of the scope of this study which only aims at giving a qualitative understanding of the effect of switching from a LO to a NLO analysis.

Finally, to stress once again the importance of our robust estimate of the covariance, we plot in Figure 11 a comparison between the spectrum of eigenvalues of the sample covariance matrix versus our robust estimate with Nkin=20N_{\mathrm{kin}}=20 on the subtraction constant dataset. The difference between the largest eigenvalues is mainly driven by the fact that σMAD\sigma_{MAD} is smaller than the sample standard deviation which is inflated by outliers. The eigenvalues of the robust estimator then decrease more slowly than the sample one, which means that they will produce an increased stability in a χ2\chi^2 fit. Six eigenvalues of the robust estimator are negative. The dotted line represents the largest of them in absolute value, and represents a physical criterion to discard smaller eigenvalues as unreliable. In the end, at most 7 eigenvalues of the covariance matrix are reliably estimated.

Figure 11. Comparison of the spectrum of eigenvalues of the covariance matrix for Nkin=20N_{\mathrm{kin}}=20 on the subtraction constant dataset on which this study is performed. We compare the sample covariance to our robust estimator. The dotted line represents the largest negative eigenvalue in the spectrum of the robust estimate.

The fits presented in the following section are stable when NkinN_{\mathrm{kin}} varies in the interval 10 to 30. Below, we under-sample the phase-space available to the study, resulting in larger uncertainties. Above, the quality of inference of the covariance matrix decreases sharply since NkinN_{\mathrm{kin}} becomes fairly large compared to the 100 replicas. When we use the unreliable sample covariance estimator, we do not find such an extended region of stability. We will show all results of the fits for Nkin=20N_{\mathrm{kin}}=20 and the robust estimators. Nkin=20N_{\mathrm{kin}}=20 corresponds to probing the subtraction constant in the region: ξ∈[0.1,0.4]\xi\in[0.1,0.4], −t∈[0.2,0.4] GeV2-t \in[0.2,0.4]\ \mathrm{GeV}^2 and Q2∈[1,2.5] GeV2Q^2 \in[1,2.5]\ \mathrm{GeV}^2, which corresponds to the bulk of the most constraining dataset, JLab 6 GeV.

Results of the fits at LO

LO radiative gluons and n=1n=1

At first, we consider only terms with n=1n=1 in the Gegenbauer expansion of eq. (54). Furthermore, we assume a radiative gluon generation, that is that d1g(t=0,μg2)=0d_1^g(t=0,\mu_g^2)=0 for some low-lying scale μg2\mu_g^2. Therefore, d1uds(t=0,μ02)d_1^{uds}(t=0,\mu_0^2) is really the only free parameter. Using μg=300 MeV\mu_g=300\ \mathrm{MeV}, we obtain the LO result:

d1uds(t=0,2 GeV2)=−0.6±1.1(81)d_1^{uds}(t=0,2\ \mathrm{GeV}^2)=-0.6\pm1.1 \tag*{(81)}
d1g(t=0,2 GeV2)=−0.8±1.5d_1^g(t=0,2\ \mathrm{GeV}^2)=-0.8\pm1.5
d1c(t=0,2 GeV2)=−0.003±0.005d_1^c(t=0,2\ \mathrm{GeV}^2)=-0.003\pm0.005

(LO n=1n=1 radiative gluons)

In spite of the different fitting methodology compared to Ref. [7], the results are very similar: there was determined d1uds=−0.5±1.2d_1^{uds}=-0.5\pm1.2 and d1g=−0.6±1.6d_1^g=-0.6\pm1.6. It was also noticed in Ref. [7] that the threshold μg\mu_g were gluons are introduced has barely any impact on the fitted value of d1udsd_1^{uds}. For instance, if we use μg=1 GeV\mu_g=1\ \mathrm{GeV}, we still obtain d1uds(t=0,2 GeV2)=−0.6±1.1d_1^{uds}(t=0,2\ \mathrm{GeV}^2)=-0.6\pm1.1, while on the other hand, d1g=−0.1±0.2d_1^g=-0.1\pm0.2.

In order to understand this interesting observation, we need to remind ourselves that at LO, there is no direct contribution of the gluons to the subtraction constant. The only contribution is indirect, through the radiation of quarks by gluons in the perturbative evolution. On the range of Q2Q^2 relevant for this analysis, that is Q2∈[1,2.5] GeV2Q^2\in[1,2.5]\ \mathrm{GeV}^2, the evolution operator resummed to leading logarithmic accuracy reads:

(d1uds(2.5 GeV2)d1g(2.5 GeV2)d1c(2.5 GeV2))=(0.920.0150.230.950.0010.007)(d1uds(1 GeV2)d1g(1 GeV2))(82)\begin{pmatrix} d_1^{uds}(2.5\ \mathrm{GeV}^2)\\ d_1^g(2.5\ \mathrm{GeV}^2)\\ d_1^c(2.5\ \mathrm{GeV}^2) \end{pmatrix} = \begin{pmatrix} 0.92 & 0.015\\ 0.23 & 0.95\\ 0.001 & 0.007 \end{pmatrix} \begin{pmatrix} d_1^{uds}(1\ \mathrm{GeV}^2)\\ d_1^g(1\ \mathrm{GeV}^2) \tag*{(82)} \end{pmatrix}

We label the coefficients of the evolution matrix from μ02\mu_0^2 to μ2\mu^2 as:

\begin{pmatrix} \Gamma_1^{qq}(\mu^2,\mu_0^2) & \Gamma_1^{gq}(\mu^2,\mu_0^2)\\ \Gamma_1^{qg}(\mu^2,\mu_0^2) & \Gamma_1^{gg}(\mu^2,\mu_0^2)\\ \Gamma_1^{cq}(\mu^2,\mu_0^2) & \Gamma_1^{cg}(\mu^2,\mu_0^2) \end{pmatrix}. \tag*{(83)}\end{pmatrix}

Notice how small Γ1qg\Gamma_1^{qg}, the radiation of light quarks by gluons, is in the range of scales covered by the bulk of the experimental data. This means that gluon contribution to the subtraction constant at LO is heavily suppressed. Introducing the gluon radiation threshold, we obtain that:

d1uds(μ2)=[Γ1qq(μ2,μ02)+Γ1gq(μ2,μ02)Γ1qg(μ02,μg2)Γ1qq(μ02,μg2)]d1uds(μ02).(84)d_1^{uds}(\mu^2)=\left[\Gamma_1^{qq}(\mu^2,\mu_0^2)+\Gamma_1^{gq}(\mu^2,\mu_0^2)\frac{\Gamma_1^{qg}(\mu_0^2,\mu_g^2)}{\Gamma_1^{qq}(\mu_0^2,\mu_g^2)}\right]d_1^{uds}(\mu_0^2). \tag*{(84)}

The maximal effect of gluons on the fit of d1udsd_1^{uds} is obtained when μ02\mu_0^2 and μ2\mu^2 are taken at the extreme values covered reliably by the experimental data, so here 1 and 2.5 GeV2\mathrm{GeV}^2. We find therefore that the features of the fitted data in the interval [1,2.5] GeV2[1,2.5]\ \mathrm{GeV}^2 that can be imputable to gluons are typically of the order of Γ1qg(2.5,1)/Γ1qq(2.5,1)×Γ1qg(1,0.09)/Γ1qq(1,0.09)=0.015/0.92×2.1=2.1%\Gamma_1^{qg}(2.5,1)/\Gamma_1^{qq}(2.5,1)\times\Gamma_1^{qg}(1,0.09)/\Gamma_1^{qq}(1,0.09)=0.015/0.92\times2.1=2.1\% of the flavour contribution imputable to d1udsd_1^{uds}. If μg\mu_g increases, the gluonic contribution decreases even more, but that is in any case completely imperceptible. In other words, in a LO analysis, radiative gluons might as well be equivalent to no gluons at all. Although this means that the quark contribution to the GFF ∑qCq(t)\sum_q C_q(t) is quite independent of the choice of radiative threshold, it clearly means that the overall GFF C(t)=∑qCq(t)+Cg(t)C(t)=\sum_q C_q(t)+C_g(t) is extremely unreliable.

LO radiative gluons and n=3n = 3

Still using a radiative gluon generation with μg=300\mu_g = 300 MeV, we now allow both d1uds(t=0,2 GeV2)d^{uds}_1(t = 0, 2\ \mathrm{GeV}^2) and d3uds(t=0,2 GeV2)d^{uds}_3(t = 0, 2\ \mathrm{GeV}^2) to be fitted. At LO, we find:

d1uds(t=0,2 GeV2)=−2.1±26.6d^{uds}_1(t = 0, 2\ \mathrm{GeV}^2) = -2.1 \pm26.6
d3uds(t=0,2 GeV2)=1.5±26.5(85)d^{uds}_3(t = 0, 2\ \mathrm{GeV}^2) = 1.5 \pm26.5 \tag*{(85)}
d1g(t=0,2 GeV2)=−2.9±37d^g_1(t = 0, 2\ \mathrm{GeV}^2) = -2.9 \pm37
d3g(t=0,2 GeV2)=0.2±4.1d^g_3(t = 0, 2\ \mathrm{GeV}^2) = 0.2 \pm4.1

(LO n=3n=3 radiative gluons)

d1d_1 and d3d_3 are anti-correlated in excess of 99% as identified in Ref. [7] before. One observes that, within uncertainty, d1uds≈−d3udsd^{uds}_1 \approx-d^{uds}_3. In other words, the uncertainty in the extraction is almost entirely stemming from contamination of LO shadow DD-terms. We have derived in § 4 an approximate estimator of the uncertainty linked precisely to this shadow DD-term d1uds≈−d3udsd^{uds}_1 \approx-d^{uds}_3 with a simplified evolution kernel. We found that (65):

σd1q≈σd3q≈ΔS1−αs(Qmax⁡2)αs(Qmin⁡2)(86)\sigma_{d_1^q} \approx\sigma_{d_3^q} \approx\frac{\Delta S}{1 - \frac{\alpha_s(Q^2_{\max})}{\alpha_s(Q^2_{\min})}} \tag*{(86)}

ΔS\Delta S can be obtained by noting that the fit with d1d_1 alone reads S=8/3×d1udsS = 8/3 \times d^{uds}_1, and therefore ΔS≈8/3×1.1\Delta S \approx8/3 \times1.1. This is probably an underestimation, since it takes the uncertainty of the simplest fit as a measure of the uncertainty of the full quantity. Then using Qmax⁡2=2.5 GeV2Q^2_{\max} = 2.5\ \mathrm{GeV}^2 and Qmin⁡2=1 GeV2Q^2_{\min} = 1\ \mathrm{GeV}^2, we find

σd1q≈σd3q≈16(87)\sigma_{d_1^q} \approx\sigma_{d_3^q} \approx16 \tag*{(87)}

to compare with the value of 26.5 that our fit produced. Besides the likely underestimation of ΔS\Delta S, the main drawback of this approximation is the reduction of the information contained in the scale dependence to a sole interval [Qmin⁡2,Qmax⁡2][Q^2_{\min}, Q^2_{\max}] where we assume that the data is uncorrelated and uniformly constraining. Then, the result is of course sensitive to the choice of this interval. For instance, simply raising Qmin⁡2Q^2_{\min} to 1.4 GeV2\mathrm{GeV}^2 [and therefore reducing the range in scales where we believe the data to be truly constraining] would produce an estimate of σd1q≈σd3q≈25.5\sigma_{d_1^q} \approx\sigma_{d_3^q} \approx25.5, very similar to the one truly observed. The approximate evolution used to derive our estimate (65) only represents a minor imprecision owing to the negligible effect of radiative gluons and the fact that our scales are close to the charm mass.

LO free gluons and n=1n = 1

This time, we allow d1g(t=0,μ02)d^g_1(t = 0, \mu_0^2) to also be a free parameter. We find at LO:

d1uds(t=0,2 GeV2)=−0.6±1.1d^{uds}_1(t = 0, 2\ \mathrm{GeV}^2) = -0.6 \pm1.1
d1g(t=0,2 GeV2)=−11±132(88)d^g_1(t = 0, 2\ \mathrm{GeV}^2) = -11 \pm132 \tag*{(88)}
d1g(t=0,2 GeV2)=−0.04±0.47d^g_1(t = 0, 2\ \mathrm{GeV}^2) = -0.04 \pm0.47

(LO n=1n=1 free gluons)

There again, the results are in good agreement with the results of Ref. [7]. Two interesting features are noticeable: the extraction of d1udsd^{uds}_1 has been left unchanged by the addition of free gluons, and the uncertainty of the gluon term has increased by a factor 90 compared to the radiative gluons. This factor of 90 can be understood as being related to Σqq(2.5,1)/Σgg(2.5,1)=0.92/0.015≈60\Sigma^{qq}(2.5,1)/\Sigma^{gg}(2.5,1) = 0.92/0.015 \approx60, the factor by which d1g(1 GeV2)d^g_1(1\ \mathrm{GeV}^2) must be larger than d1uds(1 GeV2)d^{uds}_1(1\ \mathrm{GeV}^2) so that both terms contribute with the same order of magnitude to the fitting of the data. As for the reason why d1udsd^{uds}_1 is unchanged by free gluons at LO, it comes from the fact that the fitted distributions of d1udsd^{uds}_1 and d1gd^g_1 are largely uncorrelated (correlation coefficient of -0.10). As we have stressed before, the large similarity of Γ1qq\Gamma^{qq}_1 and Γ3qq\Gamma^{qq}_3 is the root of the very large correlation between d1qd^q_1 and d3qd^q_3. On the other hand, Γ1qq\Gamma^{qq}_1 and Γ1?g\Gamma^{\mathrm{?}g}_1 present very different functional forms, and result therefore in far less correlated fits as can be observed on Figure 12.

Figure 12. Comparison of the functional form of the operators Γ1qq\Gamma^{qq}_1, Γ3qq\Gamma^{qq}_3, αsΓ1gq\alpha_s\Gamma^{gq}_1 and Γ1gg\Gamma^{gg}_1 as a function of the scale. The essence of the deconvolution problem is that, when the fitted functional forms are too similar to one another, the associated parameters are extremely difficult to differentiate and inflate considerably the uncertainty.

Results of the fits at NLO

At NLO, for μ2=Q2\mu^2 = Q^2 and with a truncation up to Gegenbauer moments of order n=3n = 3, the subtraction constant reads:

S=∑qeq2Sq+Sg,(89)S = \sum_q e_q^2 S^q + S^g, \tag*{(89)}
SNLOq=d1q(4−49αsCF4π)+d3q(4+14759450αsCF4π),(90)S^q_{\mathrm{NLO}} = d^q_1\left(4 - \frac{4}{9}\frac{\alpha_s C_F}{4\pi}\right) + d^q_3\left(4 + \frac{14759}{450}\frac{\alpha_s C_F}{4\pi}\right), \tag*{(90)}
SNLOg=∑qeq2αsTF4π(−1729d1g−3317150d3g),(91)S^g_{\mathrm{NLO}} = \frac{\sum_q e_q^2 \alpha_s T_F}{4\pi}\left(-\frac{172}{9}d^g_1 - \frac{3317}{150}d^g_3\right), \tag*{(91)}

where CF=4/3C_F = 4/3 and TF=1/2T_F = 1/2. We use the LO running of αs\alpha_s from APFEL [77, 78], which remains continuous at the heavy quark mass thresholds. However, a naive implementation of the gluon coefficient function is discontinuous at threshold due to the factor ∑qeq2\sum_q e_q^2. This makes no practical numerical difference for radiative gluons fits, but becomes important for the fit results with free gluons. The appropriate course of action would be to include heavy quark mass effects in the coefficient function, but this extends beyond the scope of this paper. To avoid spurious effects, we will therefore consider for the rest of the paper that the factor ∑qeq2\sum_q e_q^2 in the gluonic contribution to the subtraction constant is fixed to 10/910/9, the value it assumes naively if nf=4n_f = 4.

NLO radiative gluons and n=1n = 1

With a threshold of radiative gluon generation at 300 MeV, we find:

d1uds(t=0,2 GeV2)=−0.7±1.3(92)d_1^{uds}(t = 0, 2\ {\rm GeV}^2) = -0.7 \pm1.3 \tag*{(92)}
d1g(t=0,2 GeV2)=−0.9±1.8d_1^g(t = 0, 2\ {\rm GeV}^2) = -0.9 \pm1.8
d1c(t=0,2 GeV2)=−0.003±0.006d_1^c(t = 0, 2\ {\rm GeV}^2) = -0.003 \pm0.006

(NLO n=1 radiative gluons)

The results are almost identical to the LO results of the same fit. There again, changing the threshold for gluon production is only a little effect on d1udsd_1^{uds}, although larger than at LO. With a threshold μg2=1\mu_g^2 = 1 GeV2^2, we find d1uds=−0.6±1.1d_1^{uds} = -0.6 \pm1.1 and d1g=−0.1±0.2d_1^g = -0.1 \pm0.2.

To understand the similarity between the LO and NLO fit in the case where only d1udsd_1^{uds} is fitted, we observe that αs\alpha_s is at most αs,max⁡=0.35\alpha_{s,\max} = 0.35 in the fitted range. Then the NLO quark term reads (90):

∑qeq2(4−49αs,max⁡CF4π)≈3.98∑qeq2≈2.65,(93)\sum_q e_q^2\left(4-\frac{4}{9}\frac{\alpha_{s,\max}C_F}{4\pi}\right)\approx3.98\sum_q e_q^2\approx2.65, \tag*{(93)}

where we neglected the charm contribution to the subtraction constant, and for gluons (91):

−1729∑qeq2αs,max⁡TF4π≈−0.30,(94)-\frac{172}{9}\frac{\sum_q e_q^2\alpha_{s,\max}T_F}{4\pi}\approx-0.30, \tag*{(94)}

which is of the order of 10% of the quark contribution. Due to its negative sign, it causes a slight increase in the fitted value of d1udsd_1^{uds} compared to the situation at LO when the gluon threshold is small enough. Let us note that although gluons still play a minor role in the extraction, it is a much bigger one than at LO where we estimated it to 2% because of the smallness of Γ1qq\Gamma_1^{qq}.

NLO radiative gluons and n=3n = 3

Still with a radiative gluon threshold at μg=300\mu_g = 300 MeV, and allowing d1udsd_1^{uds} and d3udsd_3^{uds} to be fitted, we obtain:

d1uds(t=0,2 GeV2)=−1.7±21(95)d_1^{uds}(t = 0, 2\ {\rm GeV}^2) = -1.7 \pm21 \tag*{(95)}
d3uds(t=0,2 GeV2)=0.7±15d_3^{uds}(t = 0, 2\ {\rm GeV}^2) = 0.7 \pm15

(NLO n=3 radiative gluons)

d1g(t=0,2 GeV2)=−2±30d_1^g(t = 0, 2\ {\rm GeV}^2) = -2 \pm30
d3g(t=0,2 GeV2)=0.1±2.3d_3^g(t = 0, 2\ {\rm GeV}^2) = 0.1 \pm2.3

The general uncertainty is of the same order of magnitude as the one obtained at LO and the anti-correlation still in excess of 99%. However, whereas at LO, we had d1uds≈−d3udsd_1^{uds} \approx-d_3^{uds}, the situation has changed a bit. Since the subtraction constant has been modified at NLO, it does not admit exactly the same shadow DD-terms. Indeed, the d3udsd_3^{uds} term in (90) gives with αs(2 GeV2)≈0.3\alpha_s(2\ {\rm GeV}^2)\approx0.3:

∑qeq2(4+14759450αsCF4π)≈3.36.(96)\sum_q e_q^2\left(4+\frac{14759}{450}\frac{\alpha_s C_F}{4\pi}\right)\approx3.36. \tag*{(96)}

An NLO shadow DD-term must now also cancel the contribution stemming from the gluons, which we can no longer ignore even in the radiative approximation. Using the reference scale of 2 GeV2^2 and the explicit relation between dngd_n^g and dnudsd_n^{uds} offered by the radiation threshold, the gluonic contribution of n=1n = 1 reads:

−1729∑qeq2αsTF4πΓ1gg(2,0.09)Γ1qq(2,0.09)d1uds≈−0.36d1uds,(97)-\frac{172}{9}\frac{\sum_q e_q^2\alpha_s T_F}{4\pi}\frac{\Gamma_1^{gg}(2,0.09)}{\Gamma_1^{qq}(2,0.09)}d_1^{uds}\approx-0.36d_1^{uds}, \tag*{(97)}

whereas for n=3n = 3:

3317150∑qeq2αsTF4πΓ3gg(2,0.09)Γ3qq(2,0.09)d3uds≈0.05d3uds,(98)\frac{3317}{150}\frac{\sum_q e_q^2\alpha_s T_F}{4\pi}\frac{\Gamma_3^{gg}(2,0.09)}{\Gamma_3^{qq}(2,0.09)}d_3^{uds}\approx0.05d_3^{uds}, \tag*{(98)}

We note in passing that the effect of gluons at NLO on d3udsd_3^{uds} is much smaller than on d1udsd_1^{uds}. Finally, the NLO subtraction constant at 2 GeV2^2 reads approximately as:

S(2 GeV2)NLO=(2.65−0.36)d1uds+(3.36+0.05)d3uds,(99)S(2\ {\rm GeV}^2)^{\rm NLO}=(2.65-0.36)d_1^{uds}+(3.36+0.05)d_3^{uds}, \tag*{(99)}

to compare with the LO:

SLO=2.65d1uds+2.65d3uds.(100)S^{\rm LO}=2.65d_1^{uds}+2.65d_3^{uds}. \tag*{(100)}

An expectation of NLO shadow DD-term with radiative gluons at a 300 MeV threshold is therefore:

d1uds≈−3.36+0.052.65−0.36d3uds≈−1.5d3uds.(101)d_1^{uds}\approx-\frac{3.36+0.05}{2.65-0.36}d_3^{uds}\approx-1.5d_3^{uds}. \tag*{(101)}

Using the observed value of σd3q=15\sigma_{d_3q}=15, this would predict σd1q≈22.5\sigma_{d_1q}\approx22.5, which is close to the true value of 21. The estimator is made somewhat more complicated than at LO by the necessary inclusion of the gluon term and αs\alpha_s. It provides however reliable results, demonstrating that the interpretation of the uncertainty in terms of a simple shadow DD-term is a valuable tool.

NLO free gluons and n=1n = 1

If we now allow the d1gd^{g}_{1} term to be freely fitted alongside d1udsd^{uds}_{1}, we find:

d1uds(t=0,2 GeV2)=−1.1±7.7d^{uds}_{1}(t = 0, 2\ \mathrm{GeV}^{2}) = -1.1 \pm7.7
d1g(t=0,2 GeV2)=−6±78(102)d^{g}_{1}(t = 0, 2\ \mathrm{GeV}^{2}) = -6 \pm78 \tag*{(102)}
d1q(t=0,2 GeV2)=−0.02±0.27d^{q}_{1}(t = 0, 2\ \mathrm{GeV}^{2}) = -0.02 \pm0.27

(NLO n=1n=1 free gluons)

One will notice that the uncertainty on d1udsd^{uds}_{1} has increased by a large factor compared to the case where d1gd^{g}_{1} was not a free parameter. This indicates very large correlation between d1udsd^{uds}_{1} and d1gd^{g}_{1} at NLO, and the impact of an underlying shadow DD-term.

The reason why a shadow DD-term produces a large effect at NLO in the joint fit of d1udsd^{uds}_{1} and d1gd^{g}_{1} whereas it was not visible at LO is that gluons now contribute to the subtraction constant in their own right, mostly through αs(Q2)Γ1gg(Q2,μ02)\alpha_{s}(Q^{2})\Gamma^{gg}_{1}(Q^{2},\mu_{0}^{2}). At LO they could only contribute through the radiation term Γ1qg(Q2,μ02)\Gamma^{qg}_{1}(Q^{2},\mu_{0}^{2}). But Γ1gg\Gamma^{gg}_{1} is a diagonal term in the evolution matrix, whose functional dependence is very similar to that of Γ1qq\Gamma^{qq}_{1}, and fundamentally different of the off-diagonal term Γ1qg\Gamma^{qg}_{1}. The operators can be compared on Figure 12 where it will be apparent that the impact of the shadow DD-term related to d1gd^{g}_{1} at NLO remains smaller than that of d3udsd^{uds}_{3}. Therefore σd1q\sigma_{d1q} remains less affected by the inclusion of a free NLO d1gd^{g}_{1} than by a LO or NLO d3udsd^{uds}_{3}.

An EIC perspective

In § 4, we derived a simple estimate of the uncertainty of d1qd^{q}_{1} and d3qd^{q}_{3} when they are fitted jointly in a LO framework with simplified evolution over a range [Qmin⁡2,Qmax⁡2][Q^{2}_{\min}, Q^{2}_{\max}]. As we studied NLO fits in § 5, we extended the concept and started to consider the case of explicit gluonic degrees of freedom. Let us give here final general expressions and apply them on a kinematic range relevant for the EIC.

We assume that the contribution of heavy quarks remains always negligible in the subtraction constant and that the contribution of all three light flavors is the same dnudsd^{uds}_{n}. In the absence of appropriate heavy quark mass effects in the gluon coefficient function, we fix ne=∑qeq2=10/9n_{e} = \sum_{q} e_{q}^{2} = 10/9. Then, at a given value of tt, we remind that the subtraction constant at NLO truncated to the Gegenbauer moments n=3n = 3 reads:

SNLO=23d1uds(4−49αsCF4π)−1729neαsTF4πd1g+23d3uds(4+14759450αsCF4π)−3317150neαsTF4πd3g(103)S^{\mathrm{NLO}} = \frac{2}{3}d^{uds}_{1}\left(4-\frac{4}{9}\frac{\alpha_{s}C_{F}}{4\pi}\right)-\frac{172}{9}\frac{n_{e}\alpha_{s}T_{F}}{4\pi}d^{g}_{1} +\frac{2}{3}d^{uds}_{3}\left(4+\frac{14759}{450}\frac{\alpha_{s}C_{F}}{4\pi}\right)-\frac{3317}{150}\frac{n_{e}\alpha_{s}T_{F}}{4\pi}d^{g}_{3} \tag*{(103)}

where SS, dnad^{a}_{n} and αs\alpha_{s} have all an implicit dependence on Q2Q^{2}.

First let us study the impact of EIC kinematics on the free extraction of a gluon contribution with n=1n = 1 only. Following the reasoning of § 4, we cancel the subtraction constant at some reference scale:

0=a(μ02)d1uds(μ02)+b(μ02)d1g(μ02),(104)0 = a(\mu_{0}^{2})d^{uds}_{1}(\mu_{0}^{2}) + b(\mu_{0}^{2})d^{g}_{1}(\mu_{0}^{2}), \tag*{(104)}

where the coefficients aa and bb are read straightforwardly from the first line of eq. (103). We choose μ02=mc2\mu_{0}^{2} = m_{c}^{2}, which will also serve as the Qmin⁡2Q^{2}_{\min} of our data. Then at any scale, we have:

S(Q2)=a(Q2)(Γ1qq(Q2,μ02)d1uds(μ02)+Γ1qg(Q2,μ02)d1g(μ02))+b(Q2)(Γ1gq(Q2,μ02)d1uds(μ02)+Γ1gg(Q2,μ02)d1g(μ02)).(105)\begin{aligned} S(Q^{2}) ={}& a(Q^{2})\left(\Gamma^{qq}_{1}(Q^{2},\mu_{0}^{2})d^{uds}_{1}(\mu_{0}^{2})\right.\\ &\left.+\Gamma^{qg}_{1}(Q^{2},\mu_{0}^{2})d^{g}_{1}(\mu_{0}^{2})\right)\\ &+b(Q^{2})\left(\Gamma^{gq}_{1}(Q^{2},\mu_{0}^{2})d^{uds}_{1}(\mu_{0}^{2})\right.\\ &\left.+\Gamma^{gg}_{1}(Q^{2},\mu_{0}^{2})d^{g}_{1}(\mu_{0}^{2})\right). \tag*{(105)} \end{aligned}

Assuming that a quantity ΔS\Delta S represents the typical experimental uncertainty on the subtraction constant for Q2∈[μ02,Qmax⁡2]Q^{2} \in[\mu_{0}^{2},Q^{2}_{\max}] where Qmax⁡2Q^{2}_{\max} is the largest scale where the subtraction constant is extracted reliably, we find:

σd1q(μ02)=ΔS×∣a(Qmax⁡2)Γ1qg−a(Qmax⁡2)a(μ02)b(μ02)Γ1qg+b(Qmax⁡2)Γ1gg−a(μ02)b(Qmax⁡2)b(μ02)Γ1gg∣−1,(106)\sigma_{d1q}(\mu_{0}^{2})=\Delta S\times\left|a(Q^{2}_{\max})\Gamma^{qg}_{1}-\frac{a(Q^{2}_{\max})a(\mu_{0}^{2})}{b(\mu_{0}^{2})}\Gamma^{qg}_{1} +b(Q^{2}_{\max})\Gamma^{gg}_{1}-\frac{a(\mu_{0}^{2})b(Q^{2}_{\max})}{b(\mu_{0}^{2})}\Gamma^{gg}_{1}\right|^{-1}, \tag*{(106)}

where all the Γn\Gamma_{n} operators have an implicit argument (Qmax⁡2,μ02)(Q^{2}_{\max},\mu_{0}^{2}). Straightforwardly,

σd1g(μ02)=∣a(μ02)b(μ02)∣σd1q(μ02).(107)\sigma_{d1g}(\mu_{0}^{2})=\left|\frac{a(\mu_{0}^{2})}{b(\mu_{0}^{2})}\right|\sigma_{d1q}(\mu_{0}^{2}). \tag*{(107)}

We depict in Figure 13 the value of this estimate of the precision using ΔS=3\Delta S = 3, that is considering that the current precision on the subtraction constant is extended to a much larger range in Q2Q^{2}. The current estimated precision with the CFF extraction used in this study is depicted by the red star. One observes that constraining the CFFs up to Q2=20 GeV2Q^{2} = 20\ \mathrm{GeV}^{2} could reduce the uncertainty by a factor 3 to 4. Let us notice also that our estimator predicts very accurately that σd1g≈10σd1q\sigma_{d1g}\approx10\sigma_{d1q}, a relation that one can observe to be accurately verified in our actual fits as well.

Figure 13. Evolution of the uncertainty of d1udsd_1^{uds} and d1gd_1^g when the latter is a free parameter depending on the range of Q2Q^2 available for the extraction of CFFs. The red star denotes approximately the current situation.

Now we study the impact of EIC kinematics on the extraction of both d1udsd^{uds}_{1} and d3udsd^{uds}_{3}. We will assume that the gluon part of the shadow DD-term is 0 at the reference scale. Then:

0=a(μ02)d1uds(μ02)+c(μ02)d3uds(μ02),(108)0 = a(\mu_{0}^{2})d^{uds}_{1}(\mu_{0}^{2}) + c(\mu_{0}^{2})d^{uds}_{3}(\mu_{0}^{2}), \tag*{(108)}

and

S(Q2)=[a(Q2)Γ1qq(Q2,μ02)+b(Q2)Γ1gq(Q2,μ02)]×d1uds(μ02)+[c(Q2)Γ3qq(Q2,μ02)+d(Q2)Γ3gq(Q2,μ02)]×d3uds(μ02).(109)\begin{aligned} S(Q^{2}) ={}& [a(Q^{2})\Gamma^{qq}_{1}(Q^{2},\mu_{0}^{2})+b(Q^{2})\Gamma^{gq}_{1}(Q^{2},\mu_{0}^{2})]\\ &\times d^{uds}_{1}(\mu_{0}^{2})+[c(Q^{2})\Gamma^{qq}_{3}(Q^{2},\mu_{0}^{2})+d(Q^{2})\Gamma^{gq}_{3}(Q^{2},\mu_{0}^{2})]\\ &\times d^{uds}_{3}(\mu_{0}^{2}). \tag*{(109)} \end{aligned}

Combining the two expressions gives the estimator:

σd1q(μ02)=ΔS×∣a(Qmax⁡2)Γ1qq+b(Qmax⁡2)Γ1gq−c(Qmax⁡2)a(μ02)c(μ02)Γ3qq−d(Qmax⁡2)a(μ02)c(μ02)Γ3gq∣−1(110)\sigma_{d1q}(\mu_{0}^{2})=\Delta S\times\left|a(Q^{2}_{\max})\Gamma^{qq}_{1}+b(Q^{2}_{\max})\Gamma^{gq}_{1} -\frac{c(Q^{2}_{\max})a(\mu_{0}^{2})}{c(\mu_{0}^{2})}\Gamma^{qq}_{3} -\frac{d(Q^{2}_{\max})a(\mu_{0}^{2})}{c(\mu_{0}^{2})}\Gamma^{gq}_{3}\right|^{-1} \tag*{(110)}

We produce the corresponding plot in Figure 14. Constraining the CFFs up to Q2=20 GeV2Q^2 = 20\ \mathrm{GeV}^2 would now result in a decrease of uncertainty by a factor 2 to 3.

Evolution of the uncertainty of $d_1^{uds}$ and $d_3^{uds}$

Figure 14. Evolution of the uncertainty of d1udsd_1^{uds} and d3udsd_3^{uds} depending on the range of Q2Q^2 available for the extraction of CFFs, considering only the role of shadow DD-terms with no explicit gluon contribution. The red and green stars denote approximately the current situation obtained when fitting with radiative gluons.

We remind that we have only considered here shadow DD-terms with no explicit gluon contribution, which corresponds to the uncertainty of the fits we have performed earlier with radiative gluons. Adding freedom of explicit gluon contributions would result in yet a far larger increase of uncertainty, combining the effect of Figure 13 and Figure 14.

However, let us stress that our estimator is only focusing on the impact of the measured range of scales Q2Q^2. The EIC will also bring precious high-quality data in regions in ξ\xi which are poorly constrained so far, which will likely decrease the value of ΔS\Delta S. More work to estimate this impact remains to be done.

Conclusion

We have provided a re-derivation of dispersion relations for DVCS at all order in perturbation theory. Our presentation highlights that dispersion relations contain far more information than the commonly acknowledged restriction to the DD-term. In fact, eq. (42) along with eq. (30) give a very synthetic picture of the information that can possibly be extracted from an arbitrary knowledge of CFFs at a given scale. We have then stressed that the scale dependence of measurements is crucial to mitigate the issues related to the deconvolution problem. An intuitive presentation in terms of shadow DD-terms allows to construct simple estimates of the impact on the uncertainty of the range in scales on which DVCS is measured.

Using those tools, we have re-considered the dispersion relation global analysis of Ref. [7]. We have proposed a new statistical method aiming at giving a sounder account of correlated fits in the presence of outliers. Our re-analysis of the LO DD-term gives however very similar results to the previous publication. We extend the analysis to NLO. We find that the NLO results are generally fairly similar to the LO. The major modification comes when an explicit gluon DD-term is allowed to be freely fitted on the data. Then we find at NLO a very large correlation between the fitted quarks and gluons that does not exist at LO. We provide an explanation for this fact, linked to the similarity of scale dependence of the evolution operators.

Using the simple formalism of shadow DD-terms, we finally establish generic estimates of the reduction of uncertainty that one could expect from the range of scales probed at an EIC. We find that extending the region of measurements with a similar statistical accuracy as current best measurements to 20 GeV220\ \mathrm{GeV}^2 could bring a reduction of uncertainty in the deconvolution problem by a factor 2 to 5 depending on the quantity of interest. One should however keep in mind that the EIC will additionally reduce the statistical uncertainty on the subtraction constant, by measuring CFFs at values of Bjorken-xx where they are poorly constrained so far. Therefore, the potential EIC impact on our experimental knowledge of d1d_1 could be larger and remains to be fully estimated. In the meantime, we expect the knowledge of the pressure and shear forces within the nucleon to be mostly driven by lattice-QCD calculations.

Acknowledgements The authors thank V. Bertone, C. Lorcé, K. Orginos, J. Rodriguez-Quintero, E. Romero Alcalde and P. Snajder for valuable discussions. This project was supported by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 824093, and with the support of the French Government scholarships programme. This research was funded, in part, by l’Agence Nationale de la Recherche (ANR), project ANR-23-CE31-0019, supported by the GLUODYNAMICS project funded by “P2IO LabEx (ANR-10-LABX-0038)” in the framework of Investissements d’Avenir (ANR-11-IDEX-0003-01), managed by the Agence Nationale de la Recherche (ANR), France. HD was supported by U.S. DOE Grant #DE-FG02-04ER41302 and under the Laboratory Directed Research and Development Program (LDRD 2412) at the Thomas Jefferson National Accelerator Facility for the U.S. Department of Energy. For the purpose of open access, the author has applied a CC-BY public copyright licence to any Author Accepted Manuscript (AAM) version arising from this submission.

Open source resources

The neural network global fit of CFFs used in this analysis is available in the PARTONS framework [79]. The code of this framework is open source and can be found online at https://drf-gitlab.cea.fr/partons/core/partons on version 3 of the GPL (GPLv3).

References

  1. [1]E. Rutherford. Collision of α particles with light atoms. IV. An anomalous effect in nitrogen. Phil. Mag. Ser. 6, 37:581–587, 1919.
  2. [2]Maxim V. Polyakov and Peter Schweitzer. Forces inside hadrons: pressure, surface tension, mechanical radius, and all that. Int. J. Mod. Phys. A, 33(26):1830025, 2018.DOI
  3. [3]Cédric Lorcé, Hervé Moutarde, and Arkadiusz P. Trawiński. Revisiting the mechanical properties of the nucleon. Eur. Phys. J. C, 79(1):89, 2019.DOI
  4. [4]Adam Freese. Noether’s theorems and the energy-momentum tensor in quantum gauge theories. Phys. Rev. D, 106(12):125012, 2022.DOI
  5. [5]V. D. Burkert, L. Elouadrhiri, and F. X. Girod. The pressure distribution inside the proton. Nature, 557(7705):396–399, 2018.DOI
  6. [6]Krešimir Kumerički. Measurability of pressure inside the proton. Nature, 570(7759):E1–E2, 2019.
  7. [7]H. Dutrieux, C. Lorcé, H. Moutarde, P. Sznajder, A. Trawiński, and J. Wagner. Phenomenological assessment of proton mechanical properties from deeply virtual Compton scattering. Eur. Phys. J. C, 81(4):300, 2021.DOI
  8. [8]C. Alexandrou, M. Constantinou, K. Hadjiyiannakou, K. Jansen, C. Kallidonis, G. Koutsou, A. Vaquero Avilés-Casco, and C. Wiese. Nucleon Spin and Momentum Decomposition Using Lattice QCD Simulations. Phys. Rev. Lett., 119(14):142002, 2017.DOI
  9. [9]C. Alexandrou, S. Bacchio, M. Constantinou, J. Finkenrath, K. Hadjiyiannakou, K. Jansen, G. Koutsou, H. Panagopoulos, and G. Spanoudes. Complete flavor decomposition of the spin and momentum fraction of the proton using lattice QCD simulations at physical pion mass. Phys. Rev. D, 101(9):094513, 2020.DOI
  10. [10]Gen Wang, Yi-Bo Yang, Jian Liang, Terrence Draper, and Keh-Fei Liu. Proton momentum and angular momentum decompositions with overlap fermions. Phys. Rev. D, 106(1):014512, 2022.DOI
  11. [11]Sreeraj Nair, Chandan Mondal, Siqi Xu, Xingbo Zhao, Asmita Mukherjee, and James P. Vary. Gravitational form factors and mechanical properties of quarks in protons: A basis light-front quantization approach. Phys. Rev. D, 110(5):056027, 2024.arxiv.org/abs/2403.11702
  12. [12]Z. Q. Yao, Y. Z. Xu, D. Binosi, Z. F. Cui, M. Ding, K. Raya, C. D. Roberts, J. Rodríguez-Quintero, and S. M. Schmidt. Nucleon Gravitational Form Factors. 9 2024.arxiv.org/abs/2409.15547
  13. [13]Xiang-Dong Ji. Gauge-Invariant Decomposition of Nucleon Spin. Phys. Rev. Lett., 78:610–613, 1997.
  14. [14]Dieter Mueller, D. Robaschik, B. Geyer, F.M. Dittes, and J. Hořejší. Wave functions, evolution equations and evolution kernels from light ray operators of QCD. Fortsch.Phys., 42:101–141, 1994.
  15. [15]Xiang-Dong Ji. Deeply virtual Compton scattering. Phys.Rev., D55:7114–7125, 1997.DOI
  16. [16]A.V. Radyushkin. Asymmetric gluon distributions and hard diffractive electroproduction. Phys.Lett., B385:333–342, 1996.DOI
  17. [17]A.V. Radyushkin. Nonforward parton distributions. Phys.Rev., D56:5524–5557, 1997.DOI
  18. [18]John C. Collins, Leonid Frankfurt, and Mark Strikman. Factorization for hard exclusive electroproduction of mesons in QCD. Phys.Rev., D56:2982–3006, 1997.DOI
  19. [19]John C. Collins and Andreas Freund. Proof of factorization for deeply virtual Compton scattering in QCD. Phys.Rev., D59:074009, 1999.DOI
  20. [20]Edgar R. Berger, M. Diehl, and B. Pire. Time - like Compton scattering: Exclusive photoproduction of lepton pairs. Eur.Phys.J., C23:675–689, 2002.DOI
  21. [21]Dieter Müller, Tobias Lautenschlager, Kornelija Passek-Kumericki, and Andreas Schaefer. Towards a fitting procedure to deeply virtual meson production - the next-to-leading order case. Nucl. Phys., B884:438–546, 2014.DOI
  22. [22]R. Boussarie, B. Pire, L. Szymanowski, and S. Wallon. Exclusive photoproduction of a γ pair with a large invariant mass. JHEP, 02:054, 2017. [Erratum: JHEP 10, 029 (2018)].
  23. [23]G. Duplančić, K. Passek-Kumerički, B. Pire, L. Szymanowski, and S. Wallon. Probing axial quark generalized parton distributions through exclusive photoproduction of a γ π± pair with a large invariant mass. JHEP, 11:179, 2018.DOI
  24. [24]Oskar Grocholski, Bernard Pire, Pawel Sznajder, Lech Szymanowski, and Jakub Wagner. Collinear factorization of diphoton photoproduction at next to leading order. Phys. Rev. D, 104(11):114006, 2021.DOI
  25. [25]Jian-Wei Qiu and Zhite Yu. Exclusive production of a pair of high transverse momentum photons in pion-nucleon collisions for extracting generalized parton distributions. JHEP, 08:103, 2022.DOI
  26. [26]Jian-Wei Qiu and Zhite Yu. Single diffractive hard exclusive processes for the study of generalized parton distributions. Phys. Rev. D, 107(1):014007, 2023.DOI
  27. [27]Jian-Wei Qiu and Zhite Yu. Extraction of the Parton Momentum-Fraction Dependence of Generalized Parton Distributions from Exclusive Photoproduction. Phys. Rev. Lett., 131(16):161902, 2023.DOI
  28. [28]F.X. Girod et al. Measurement of Deeply virtual Compton scattering beam-spin asymmetries. Phys.Rev.Lett., 100:162002, 2008.pubmed.ncbi.nlm.nih.gov/18518188
  29. [29]A. Airapetian et al. Beam-helicity and beam-charge asymmetries associated with deeply virtual Compton scattering on the unpolarised proton. JHEP, 07:032, 2012.DOI
  30. [30]H. S. Jo et al. Cross sections for the exclusive photon electroproduction on the proton and Generalized Parton Distributions. Phys. Rev. Lett., 115(21):212003, 2015.DOI
  31. [31]M. Defurne et al. E00-110 experiment at Jefferson Lab Hall A: Deeply virtual Compton scattering off the proton at 6 GeV. Phys. Rev., C92(5):055202, 2015.DOI
  32. [32]M. Defurne et al. A glimpse of gluons through deeply virtual Compton scattering on the proton. Nature Commun., 8(1):1408, 2017.
  33. [33]R. Akhunzyanov et al. Transverse extension of partons in the proton probed in the sea-quark range by measuring the DVCS cross section. Phys. Lett. B, 793:188–194, 2019. [Erratum: Phys.Lett.B 800, 135129 (2020)].
  34. [34]F. Georges et al. Deeply Virtual Compton Scattering at High Bjorken xB. Phys. Rev. Lett., 128(25):252002, 2022.
  35. [35]G. Christiaens et al. First CLAS12 Measurement of Deeply Virtual Compton Scattering Beam-Spin Asymmetries in the Extended Valence Region. Phys. Rev. Lett., 130(21):211902, 2023.
  36. [36]V. M. Braun, A. N. Manashov, S. Moch, and J. Schoenleber. Two-loop coefficient function for DVCS: vector contributions. JHEP, 09:117, 2020.
  37. [37]V. M. Braun, A.N. Manashov, and B. Pirnay. Finite-t and target mass corrections to deeply virtual Compton scattering. Phys.Rev.Lett., 109:242001, 2012.DOI
  38. [38]V. Bertone, H. Dutrieux, C. Mezrag, H. Moutarde, and P. Sznajder. The deconvolution problem of deeply virtual Compton scattering. Phys. Rev. D, 103(11):114019, 4 2021.DOI
  39. [39]H. Dutrieux, H. Dutrieux, O. Grocholski, O. Grocholski, H. Moutarde, H. Moutarde, P. Sznajder, and P. Sznajder. Artificial neural network modelling of generalised parton distributions. Eur. Phys. J. C, 82(3):252, 2022. [Erratum: Eur.Phys.J.C 82, 389 (2022)].DOI
  40. [40]I. V. Anikin and O. V. Teryaev. Dispersion relations and subtractions in hard exclusive processes. Phys. Rev. D, 76:056007, 2007.DOI
  41. [41]Maxim V. Polyakov and C. Weiss. Skewed and double distributions in pion and nucleon. Phys.Rev., D60:114017, 1999.DOI
  42. [42]B. L. G. Bakker, E. Leader, and T. L. Trueman. A Critique of the angular momentum sum rules and a new angular momentum sum rule. Phys. Rev. D, 70:114001, 2004.DOI
  43. [43]E. Leader and C. Lorcé. The angular momentum controversy: What’s it all about and does it matter? Phys. Rept., 541(3):163–248, 2014.
  44. [44]M. V. Polyakov. Generalized parton distributions and strong forces inside nucleons and nuclei. Phys. Lett., B555:57–62, 2003.DOI
  45. [45]M. Diehl. Generalized parton distributions. Phys.Rept., 388:41–277, 2003.
  46. [46]Xiang-Dong Ji. Off forward parton distributions. J.Phys., G24:1181–1205, 1998.DOI
  47. [47]A.V. Radyushkin. Symmetries and structure of skewed and double distributions. Phys.Lett., B449:81–88, 1999.DOI
  48. [48]N. Chouika, C. Mezrag, H. Moutarde, and J. Rodríguez-Quintero. Covariant Extension of the GPD overlap representation at low Fock states. Eur. Phys. J., C77:906, 2017.DOI
  49. [49]Cédric Mezrag. An Introductory Lecture on Generalised Parton Distributions. Few Body Syst., 63(3):62, 2022.DOI
  50. [50]José Manuel Morgado Chavez, Valerio Bertone, Feliciano De Soto Borrero, Maxime Defurne, Cédric Mezrag, Hervé Moutarde, José Rodríguez-Quintero, and Jorge Segovia. Pion generalized parton distributions: A path toward phenomenology. Phys. Rev. D, 105(9):094012, 2022.
  51. [51]O. V. Teryaev. Analytic properties of hard exclusive amplitudes. In 11th International Conference on Elastic and Diffractive Scattering: Towards High Energy Frontiers: The 20th Anniversary of the Blois Workshops, 17th Rencontre de Blois, 10 2005.DOI
  52. [52]M. Diehl and D. Yu. Ivanov. Dispersion representations for hard exclusive processes: beyond the Born approximation. Eur. Phys. J. C, 52:919–932, 2007.DOI
  53. [53]H. Moutarde, P. Sznajder, and J. Wagner. Unbiased determination of DVCS Compton Form Factors. Eur. Phys. J., C79(7):614, 2019.DOI
  54. [54]A. Accardi et al. Electron Ion Collider: The Next QCD Frontier. Eur. Phys. J., A52(9):268, 2016.arxiv.org/abs/1212.1701
  55. [55]R. Abdul Khalek et al. Science Requirements and Detector Concepts for the Electron-Ion Collider: EIC Yellow Report. arXiv:2103.05419, 3 2021.DOI
  56. [56]Xurong Chen. A Plan for Electron Ion Collider in China. PoS, DIS2018:170, 2018.arxiv.org/abs/1809.00448
  57. [57]Daniele P. Anderle et al. Electron-ion collider in China. Front. Phys. (Beijing), 16(6):64701, 2021.DOI
  58. [58]J. L. Abelleira Fernandez et al. A Large Hadron Electron Collider at CERN: Report on the Physics and Design Concepts for Machine and Detector. J. Phys. G, 39:075001, 2012.
  59. [59]B. Pasquini, M. V. Polyakov, and M. Vanderhaeghen. Dispersive evaluation of the D-term form factor in deeply virtual Compton scattering. Phys. Lett. B, 739:133–138, 2014.DOI
  60. [60]H. M. Nussenzveig. Causality and dispersion relations, volume 95. Academic Press, New York, London, 1972.
  61. [61]Gary R. Goldstein and Simonetta Liuti. The Use of Dispersion Relations in Hard Exclusive Processes and the Partonic Interpretation of Deeply Virtual Compton Scattering. Phys. Rev. D, 80:075101, 2009.DOI
  62. [62]Marija Čuić, Krešimir Kumerički, and Andreas Schäfer. Separation of Quark Flavors Using Deeply Virtual Compton Scattering Data. Phys. Rev. Lett., 125(23):232005, 2020.DOI
  63. [63]M. Guidal and M. Vanderhaeghen. Double deeply virtual Compton scattering off the nucleon. Phys. Rev. Lett., 90:012001, 2003.pubmed.ncbi.nlm.nih.gov/12570602
  64. [64]Andrei V. Belitsky and Dieter Mueller. Probing generalized parton distributions with electroproduction of lepton pairs off the nucleon. Phys. Rev. D, 68:116005, 2003.DOI
  65. [65]K. Deja, V. Martinez-Fernandez, B. Pire, P. Sznajder, and J. Wagner. Phenomenology of double deeply virtual Compton scattering in the era of new experiments. Phys. Rev. D, 107(9):094035, 2023.DOI
  66. [66]A. Pedrak, B. Pire, L. Szymanowski, and J. Wagner. Electroproduction of a large invariant mass photon pair. Phys. Rev. D, 101(11):114027, 2020.DOI
  67. [67]Oskar Grocholski, Bernard Pire, Pawel Sznajder, Lech Szymanowski, and Jakub Wagner. Phenomenology of diphoton photoproduction at next-to-leading order. Phys. Rev. D, 105(9):094025, 2022.
  68. [68]Goran Duplančić, Saad Nabeebaccus, Kornelija Passek-Kumerički, Bernard Pire, Lech Szymanowski, and Samuel Wallon. Probing chiral-even and chiral-odd leading twist quark generalized parton distributions through the exclusive photoproduction of a γρ pair. Phys. Rev. D, 107(9):094023, 2023.DOI
  69. [69]P. Dall’Olio, F. De Soto, C. Mezrag, J. M. Morgado Chávez, H. Moutarde, J. Rodríguez-Quintero, P. Sznajder, and J. Segovia. Unraveling generalized parton distributions through Lorentz symmetry and partial DGLAP knowledge. Phys. Rev. D, 109(9):096013, 2024.arxiv.org/abs/2401.12013
  70. [70]Hervé Dutrieux, Robert G. Edwards, Colin Egerer, Joseph Karpie, Christopher Monahan, Kostas Orginos, Anatoly Radyushkin, David Richards, Eloy Romero, and Savvas Zafeiropoulos. Towards unpolarized GPDs from pseudo-distributions. JHEP, 08:162, 2024.DOI
  71. [71]A. Accardi et al. An experimental program with high duty-cycle polarized and unpolarized positron beams at Jefferson Lab. Eur. Phys. J. A, 57(8):261, 2021.DOI
  72. [72]H. Dutrieux, V. Bertone, H. Moutarde, and P. Sznajder. Impact of a positron beam at JLab on an unbiased determination of DVCS Compton form factors. Eur. Phys. J. A, 57(8):250, 2021.DOI
  73. [73]R. Gnanadesikan and J. R. Kettenring. Robust estimates, residuals, and outlier detection with multiresponse data. Biometrics, 28(1):81–124, 1972.DOI
  74. [74]Peter J. Rousseeuw. Least median of squares regression. Journal of the American Statistical Association, 79(388):871–880, 1984.DOI
  75. [75]Olivier Ledoit and Michael Wolf. A well-conditioned estimator for large-dimensional covariance matrices. Journal of Multivariate Analysis, 88(2):365–411, 2004.
  76. [76]Hervé Dutrieux. Robust covariance and extremely correlated fit. in preparation, 2024.
  77. [77]V. Bertone, S. Carrazza, and J. Rojo. APFEL: A PDF Evolution Library with QCD corrections. Comput. Phys. Commun., 185:1647–1668, 2014.DOI
  78. [78]Valerio Bertone. APFEL++: A new PDF evolution library in C++. PoS, DIS2017:201, 2018.arxiv.org/abs/1708.00911
  79. [79]B. Berthou, D. Binosi, N. Chouika, L. Colaneri, M. Guidal, C. Mezrag, H. Moutarde, J. Rodríguez-Quintero, F. Sabatié, P. Sznajder, and J. Wagner. PARTONS: PARtonic Tomography Of Nucleon Software. A computing framework for the phenomenology of Generalized Parton Distributions. Eur. Phys. J., C78(6):478, 2018.

Paper details

vlnc:W_7us9x239
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents