betaFeedback

WorksPhysics and Astronomy

Polarization options in inclusive DIS off tensor polarized deuteron

Abstract

In the near future, the Jefferson Lab b1b_1 experiment will provide the second measurement of tensor polarized asymmetries in inclusive DIS on the deuteron. In this asymmetry, 4 independent tensor polarized structure functions contribute. This necessitates systematic approximations in the extraction of the leading twist structure function b1b_1 from a single tensor asymmetry measurement. Contamination from higher twist structure functions and kinematic effects is discussed here. Using a deuteron convolution model, we quantify the systematic errors from these approximations for two different choices for the target polarization direction (momentum transfer, electron beam direction). For Jefferson Lab 12 GeV kinematics, the systematic error turns out to be comparable between the two polarization options, while at higher Q2Q^2 values the momentum transfer direction is preferred.

Introduction

The b1b_1 experiment is an experiment to run in Hall C of Jefferson Lab (JLab) that aims to measure the leading-twist tensor polarized structure function (SF) b1b_1 of spin-1 hadrons, in this case the deuteron [1]. The b1b_1 SF can be extracted from an inclusive DIS measurement in the reaction

e(k)+d⃗(pd)→e′(k′)+X,(1)e(k) + \vec{d}(p_d) \to e^\prime(k^\prime) + X, \tag*{(1)}

where the deuteron is polarized and no electron polarization is required.

The observable of interest is the tensor polarized asymmetry

AT=dσ(Λd=+1)+dσ(Λd=−1)−2dσ(Λd=0)dσ(Λd=+1)+dσ(Λd=0)+dσ(Λd=−1),(2)A^T = \frac{d\sigma(\Lambda_d=+1)+d\sigma(\Lambda_d=-1)-2d\sigma(\Lambda_d=0)}{d\sigma(\Lambda_d=+1)+d\sigma(\Lambda_d=0)+d\sigma(\Lambda_d=-1)}, \tag*{(2)}

[−2≤AT≤1][-2 \leq A^T \leq1]

where the individual terms denote the differential cross section for the deuteron state with polarization Λd\Lambda_d along some fixed polarization direction (discussed in more later).

As shown in Figure 1, the JLab experiment will cover DIS kinematics 0.8 GeV2<Q2<5.0 GeV20.8\ \mathrm{GeV}^2 < Q^2 < 5.0\ \mathrm{GeV}^2 and 0.16<x<0.490.16 < x < 0.49, with 0<x<20 < x < 2 the rescaled Bjorken scaling variable and Q2Q^2 the momentum transfer squared. Figure 2 depicts the reaction and specifies the momenta of the initial and final-state particles, were the exchanged virtual photon carries momentum transfer q=k−k′q = k-k^\prime.

Kinematical coverage of the $b_1$ experiment over $x$ and $Q^2$

Figure 1. Kinematical coverage of the b1b_1 experiment over xx and Q2Q^2. The blue region is the kinematical coverage of the left SHMS spectrometer in JLab Hall C, and the red area is that of the right HMS spectrometer. The grey regions are excluded by a cut of W2>1.85 GeV2W^2 > 1.85\ \mathrm{GeV}^2. The darker areas of each region indicate kinematic points with higher rates [1, 2].

Figure 2. Unpolarized electron-polarized deuteron deep-inelastic scattering. Incoming (outgoing) electron have momentum kk (k′k'), the deuteron has momentum pdp_d. NeN_e (NqN_q) denote the deuteron polarization direction considered here, along the electron beam (virtual photon direction).

The collinear partonic distribution function (usually denoted δTq\delta_Tq or f1LL=23δTqf_{1LL} = \frac{2}{3}\delta_Tq [3]) that appears in the b1b_1 SF has a probabilistic interpretation in the parton model [4]. It is linked to the difference of quark (longitudinal momentum) distributions in the deuteron with polarization Λd=+1\Lambda_d=+1 and Λd=0\Lambda_d=0:

b1∝[q+1(x)−q0(x)].(3)b_1 \propto\left[q^{+1}(x)-q^0(x)\right]. \tag*{(3)}

For the proton-neutron (pnpn) component of the deuteron, a non-zero b1b_1 SF is linked to the non-central (tensor) part of the nucleon-nucleon interaction. As such b1b_1 provides a measure of how nuclear interactions affect collinear partonic structure. In a convolution model [5, 4, 6, 7], b1b_1 is non-zero only in the presence of a deuteron D-wave (L=2L = 2). f1LLf_{1LL}, the partonic distribution function that contributes to b1b_1, enters in the quark total and orbital angular momentum sum rules for spin-1 hadrons [8].

In the past, HERMES has provided the only measurement of A1TA^T_1 in the DIS region and provided an extraction of b1b_1 [9]. The tensor asymmetries and associated b1b_1 values are typically small for inclusive DIS measurements. This is due to i) the D-wave being a small component of the deuteron wave function at internal pnpn-momenta ≤\leq few 100 MeV, ii) the inclusive measurement averaging over all initial deuteron configurations. Conventional nuclear convolution models are not able to explain the HERMES result, with calculations typically smaller than the measured values [6, 7]. This could be due to physics not included in the convolution models (non-nucleonic components in the deuteron for instance [10]), but could also be due to the systematic errors associated with approximations in the extraction procedure. The latter effect is discussed and quantified in more detail in this article.

As mentioned earlier, Jefferson Lab has an approved experiment to measure the inclusive tensor asymmetry, both in the quasi-elastic (x≥1x \geq1) and deep inelastic region. The experiment aims to extract b1b_1 over a similar kinematic region as the HERMES data. While the HERMES reported a 44% relative uncertainty for their ATA^T measurement, the proposal for the b1b_1 experiment at Jefferson Lab calls for a measurement capable of achieving 9.2% relative uncertainty on ATA^T [9, 11]. In this light, we discuss in this article some of the approximations that are needed to extract b1b_1 from the tensor asymmetry measurement, especially with regard to the freedom that one has in picking the direction in which to polarize the deuteron target. As it is not trivial to change the polarization direction once the target is operated, carrying out this study is of relevance and importance to the planned measurement.

In § 2 we discuss how the polarization for spin-1 targets is quantified through variables that enter theoretical calculations and experimental polarimetry of spin-1 ensembles. Next, in § 3, we discuss the general form of the inclusive DIS cross section and tensor asymmetry for spin-1 targets, while in § 4 we cover how the polarization direction affects the asymmetry and what approximations are needed to obtain b1b_1 from a single

tensor asymmetry. In § 5 we estimate the systematic error that follows from the used approximations in the b1b_1 extraction at JLab kinematics. Conclusions are stated in Sec. 6.

Deuteron Polarization

The b1b_1 experiment uses a solid polarized deuteron target, which will be achieved using a cryogenic target material of deuterated ammonia (ND3\mathrm{ND}_3) which is doped with free radicals as well as the process of dynamic nuclear polarization (DNP). The usage of DNP introduces a number of technical requirements for the target such as the usage of cryogenics to hold the target at a low temperature (typically about 1 K), and a magnetic holding field (typically either 2.5 T or 5 T) [12]. Due to the Zeeman splitting of nuclear spin states, at low temperature nuclear spin states will fill up preferentially at low temperature according to Boltzmann statistics. This effect can be quantified by the nuclear spin polarization nI=⟨Iz⟩/In_I = \langle I_z\rangle/I where II is the total population, and IzI_z is the population of nuclei with spin aligned along B-field axis ZZ.

For a spin-1 particle such as a deuteron, the Zeeman splitting creates three energy sublevels each corresponding to the Λd=+1,0,\Lambda_d = +1, 0, and −1-1 spin states. The “vector” polarization P\mathcal{P} is then:

P=n+−n−(4)\mathcal{P} = n_+ - n_- \tag*{(4)}

where n+n_+, n−n_-, and n0n_0 (n++n0+n−=1n_+ + n_0 + n_- = 1) are the relative populations of the spin Λd=1\Lambda_d = 1 state, the spin Λd=−1\Lambda_d = -1 state, and the spin Λd=0\Lambda_d = 0 state respectively. For deuterons in a magnetic field BB and at temperature TT the thermal equilibrium vector polarization is

P(TE)=4tanh⁡(gdμNB2kBT)3+tanh⁡2(gdμNB2kBT)(5)\mathcal{P}^{(\mathrm{TE})} = \frac{4\tanh\left(\frac{g_d\mu_N B}{2k_B T}\right)}{3+\tanh^2\left(\frac{g_d\mu_N B}{2k_B T}\right)} \tag*{(5)}

where BB is the magnetic holding field, TT is the target temperature, gdg_d is the deuteron g-factor (gd≃0.857g_d \simeq0.857) and μN\mu_N is the nuclear magneton.

At a magnetic field of B=5B = 5 T and at a temperature of T=1T = 1 K the deuteron polarization is still quite small, only about 0.1%. However, at the same field and temperature the electron polarization is quite high, over 99% polarized. The nuclear polarization can then be enhanced by driving electron spin transitions in the target, and which causes the unpaired electrons from the free radicals to transfer their spins to the nuclei. The spin transitions are driven with the application of microwaves tuned to the electron spin transition frequency (approximately 140 GHz at 5 T) during this process of microwave “enhancement” of the polarization, the nuclear spin polarization builds up according to a number of factors including the density of free radicals in the target sample, the microwave power, and the magnetic field uniformity within the target sample.

In addition to vector polarization, enhancement of deuterons also drives an increase in “tensor” polarization QQ [13] which is

Q=n++n−−2n0.(6)Q = n_+ + n_- - 2n_0. \tag*{(6)}

As discussed in § 3, the inclusive cross section with an unpolarized electron beam is only sensitive to tensor polarization QQ, not vector polarization P\mathcal{P}.

From the theory side, the deuteron ensemble with different polarization states is described by a spin-1 density matrix. Here, we use the covariant density matrix, which can be parameterized in terms of an axial 4-vector sμs^\mu and a traceless and symmetric 4-tensor tμνt^{\mu\nu}

ρμν=13(−gμν+pdμpdνMd2)+i2Mdϵμνρσpd,ρsσ−tμν,(7)\rho^{\mu\nu} = \frac{1}{3}\left(-g^{\mu\nu}+\frac{p_d^\mu p_d^\nu}{M_d^2}\right)+\frac{i}{2M_d}\epsilon^{\mu\nu\rho\sigma}p_{d,\rho}s_\sigma-t^{\mu\nu}, \tag*{(7)}

where MdM_d is the deuteron mass, pdp_d the deuteron momentum and sμ(tμν)s^\mu(t^{\mu\nu}) are the polarization vector (tensor) and ϵ0123=1\epsilon_{0123}=1. For more details on ρμν\rho^{\mu\nu}, see Ref. [14].

Pure deuteron states are characterized by a polarization direction NμN^\mu [(Npd)=0(N p_d)=0] and a polarization state Λd=±1,0\Lambda_d=\pm1,0, with total angular momentum states quantized along NN. In the deuteron rest frame NμN^\mu reduces to the spatial direction used to polarize the deuteron ensemble Nμ=(0,N)N^\mu=(0,\boldsymbol{N}). For pure states, the density matrix parameters take the following form. The axial 4-vector can be expressed as

sμ(N,Λd)=ΛdNμ,(pdN)=0,N2=−1.(8)s^\mu(N,\Lambda_d)=\Lambda_d N^\mu,\qquad(p_dN)=0,\quad N^2=-1. \tag*{(8)}

For pure states, the symmetric and traceless 4-tensor can be expressed as

tαβ(N,Λd)=16(gαβ−pdαpdβpd2+3NαNβ)×{+1Λd=±1−2Λd=0.(9)t^{\alpha\beta}(N,\Lambda_d)=\frac{1}{6}\left(g^{\alpha\beta}-\frac{p_d^\alpha p_d^\beta}{p_d^2}+3N^\alpha N^\beta\right)\times \begin{cases} +1 & \Lambda_d=\pm1\\ -2 & \Lambda_d=0 \end{cases}. \tag*{(9)}

In the deuteron rest frame the eigenvalues of ραβ\rho^{\alpha\beta} (being n+,n0,n−n_+,n_0,n_-) can be expressed entirely using the degrees of polarization PP and QQ, see Eqs. (4) and (6).

Cross section and tensor polarized asymmetry

In the inclusive DIS cross section, the tensor degrees of freedom present in the spin-1 density matrix lead to four additional tensor polarized structure functions in comparison with the spin-1/2 case. For unpolarized electron scattering6, this results in the following decomposition of the cross section [14]

dσdx dQ2=πy2α2Q4(1−ϵ)[F[UU,T]+ϵF[UU,L]+TLL(F[UTLL,T]+ϵF[UTLL,L])+2ϵ(1+ϵ)TLTcos⁡ϕTLF[UTLT]+ϵTTTcos⁡2ϕTTF[UTTT]].(10)\frac{d\sigma}{dx\,dQ^2}=\frac{\pi y^2\alpha^2}{Q^4(1-\epsilon)}\left[F_{[UU,T]}+\epsilon F_{[UU,L]}+T_{LL}\left(F_{[UT_{LL},T]}+\epsilon F_{[UT_{LL},L]}\right)+\sqrt{2\epsilon(1+\epsilon)}T_{LT}\cos\phi_{TL}F_{[UT_{LT}]}+\epsilon T_{TT}\cos2\phi_{TT}F_{[UT_{TT}]}\right]. \tag*{(10)}

Here, α\alpha is the electromagnetic fine structure constant. The kinematical invariants Bjorken xx, inelasticity yy and Rosenbluth variable ϵ\epsilon are defined as follows

xd=−q22(pdq),(0<xd<1),(11)x_d=\frac{-q^2}{2(p_dq)},\qquad(0<x_d<1), \tag*{(11)}
y=(pdq)(pdk),(0<y<1),(12)y=\frac{(p_dq)}{(p_dk)},\qquad(0<y<1), \tag*{(12)}
ϵ=1−γ2y2/41−y+y2/2+γ2y2/4.(13)\epsilon=\frac{1-\gamma^2y^2/4}{1-y+y^2/2+\gamma^2y^2/4}. \tag*{(13)}

In Figure 1 and what follows we use the rescaled

x=2xd,(0<x<2),(14)x=2x_d,\qquad(0<x<2), \tag*{(14)}

which permits more natural comparison with the Bjorken-xx variable on the nucleon (i.e. x≈1x\approx1 corresponds to quasi-elastic scattering). The variable γ\gamma is defined as

γ=MdQ(pdq)=2xdMdQ.(15)\gamma=\frac{M_dQ}{(p_dq)}=\frac{2x_dM_d}{Q}. \tag*{(15)}

In the deuteron rest frame (RF) depicted in Figure 2, the polarization parameters appearing in Eq. (10) can be identified with components of the polarization tensor of Eq. (9):

TLL=tzz[RF],(16)T_{LL}=t^{zz}[\mathrm{RF}], \tag*{(16)}
TLTcos⁡ϕTL=txz[RF],(17)T_{LT}\cos\phi_{TL}=t^{xz}[\mathrm{RF}], \tag*{(17)}
TTTcos⁡2ϕTT=txx[RF]−tyy[RF].(18)T_{TT}\cos2\phi_{TT}=t^{xx}[\mathrm{RF}]-t^{yy}[\mathrm{RF}]. \tag*{(18)}

For details on the construction and interpretation of these parameters as Lorentz invariants, see Refs. [15, 14].

We follow the notation used in [14] to identify the structure functions (SF) indexed with the different electron, deuteron, and virtual photon polarizations:

F[electron deuteron,photon](x,Q2).(19)F_{[\text{electron deuteron,photon}]}(x,Q^2). \tag*{(19)}

The dependence on xx and Q2Q^2 is usually suppressed and implicit in any of the other equations. The structure functions contain information the high-energy DIS process and the internal deuteron structure probed in the reaction.

For the unpolarized structure functions, we have the following relations with F1,F2F_1,F_2

F[UU,L]=(1+γ2)F2xd−2F1,F[UU,T]=2F1.(20)F_{[UU,L]}=(1+\gamma^2)\frac{F_2}{x_d}-2F_1,\qquad F_{[UU,T]}=2F_1. \tag*{(20)}

The tensor polarized structure functions were originally introduced in [4] as b1b_1 to b4b_4 using a different decomposition of the hadronic tensor. The linear relations with those appearing in Eq. (10) is given by

F[UTLL,T]=−[2(1+γ2)b1−γ2xd(16b2−12b3)](21)F_{[UT_{LL,T}]}=-\left[2(1+\gamma^2)b_1-\frac{\gamma^2}{x_d}\left(\frac{1}{6}b_2-\frac{1}{2}b_3\right)\right] \tag*{(21)}
F[UTLL,L]=1xd[2(1+γ2)xdb1−(1+γ2)2(13b2+b3+b4)](22)F_{[UT_{LL,L}]}=\frac{1}{x_d}\left[2(1+\gamma^2)x_db_1-(1+\gamma^2)^2\left(\frac{1}{3}b_2+b_3+b_4\right)\right] \tag*{(22)}
−(1+γ2)(13b2−b4)−(13b2−b3)(23)-\left(1+\gamma^2\right)\left(\frac{1}{3}b_2-b_4\right)-\left(\frac{1}{3}b_2-b_3\right) \tag*{(23)}
F[UTLT]=−γ2xd[(1+γ2)(13b2−b4)+(23b2−2b3)](24)F_{[UT_{LT}]}=-\frac{\gamma}{2x_d}\left[(1+\gamma^2)\left(\frac{1}{3}b_2-b_4\right)+\left(\frac{2}{3}b_2-2b_3\right)\right] \tag*{(24)}
F[UTTT]=−γ2xd(16b2−12b3).(25)F_{[UT_{TT}]}=-\frac{\gamma^2}{x_d}\left(\frac{1}{6}b_2-\frac{1}{2}b_3\right). \tag*{(25)}

The SF used here have the advantage that the geometric dependence on the different tensor polarization parameters is more clear, while the b1−b4b_1-b_4 have a more clear interpretation in the parton model. In the Bjorken limit (Q2→∞Q^2\to\infty, xx fixed), b1b_1 and b2b_2 obey a Callan-Gross-like relation

b2=2xdb1=xb1.(26)b_2=2x_db_1=xb_1. \tag*{(26)}

b1b_1 and b2b_2 are leading twist, while b3b_3 and b4b_4 are higher twist. For the SF used here, Eqs. (17) imply that F[UTLL,L]F_{[UT_{LL,L}]} and F[UTLL,T]F_{[UT_{LL,T}]} are leading twist, while F[UTLT]F_{[UT_{LT}]} and F[UTTT]F_{[UT_{TT}]} are higher twist. By using Eqs. (10) (see Ref. [15]), we can write the asymmetry as

AT=1Q2FUU,T+ϵFUU,L[TLL(FUTLL,T+ϵFUTLL,L)+TLTcos⁡ϕTL2ϵ(1+ϵ)F[UTLT]+TTTcos⁡2ϕTTϵF[UTTT]](27)A^T=\frac{1}{Q}\frac{2}{F_{UU,T}+\epsilon F_{UU,L}}\left[T_{LL}(F_{UT_{LL,T}}+\epsilon F_{UT_{LL,L}})+T_{LT}\cos\phi_{TL}\sqrt{2\epsilon(1+\epsilon)}F_{[UT_{LT}]}+T_{TT}\cos2\phi_{TT}\epsilon F_{[UT_{TT}]}\right] \tag*{(27)}

where it is understood that QQ is the combination of Eq. (6) made using the eigenvalues of the density matrix of the deuteron ensemble that is being considered.

Experiments wishing to extract b1b_1 have to do so from a tensor asymmetry measurement. One tensor asymmetry measurement is, however, not sufficient to uniquely extract one SF as four unknown structure functions enter in the numerator of the asymmetry. In order to extract the b1b_1 structure function either multiple measurements with different polarization directions NN (which results in different values for the tensor polarization parameters) and/or ϵ\epsilon are needed or an approximation has to be used to eliminate some of the unknowns. As performing measurements with different polarization directions is technically challenging and results in larger beam time needs, the HERMES measurement and the upcoming JLab experiment use a single tensor asymmetry measurement with approximations to extract b1b_1. These approximations result in additional systematic errors for the extraction. However, for a single measurement there is still the question which polarization direction NN could be the most optimal and yields the best approximation.

Polarization direction considerations

In this section, we discuss possible approximations that have or can be used to extract b1b_1 from a single ATA^T measurement. While from a theory viewpoint, we could analyse Eq. (27) for any direction NN, the experimental setup has real constraints in which directions the deuteron can be polarized.

Because of the need for a cryogenic target in a high magnetic field with microwave transmission for polarization enhancement, the maintenance of the target can get quite complex for any practical measurement of ATA^T. This complexity has the effect of practically limiting the target and detector configurations that an inclusive measurement can be designed to have. In particular, Jefferson Lab’s most recent polarized target run employed a 5 T superconducting magnet, which has an opening bore of ±35∘\pm35^\circ and is designed to produce a field direction either longitudinal or transverse to the beamline (as seen in Figure 3). In the longitudinal configuration, which is planned to be used for the b1b_1 experiment, the deuteron is polarized along the beam momentum [16]. Rotating the magnet anywhere within its ±35∘\pm35^\circ opening bore angle requires the use of a magnetic chicane to “pre-steer” the electron beam in order to correctly reproduce the longitudinal-field kinematics, which can complicate the running of the experiment. However, running with a longitudinal field also creates a nonzero angle θq≠0\theta_q\ne0 between the target polarization direction and the momentum transfer direction NqN_q, which as shown below results in a more complicated relation between the asymmetry ATA^T and b1b_1.

The 5 T superconducting magnet to be used for the polarized target field

Figure 3. A photograph of the 5 T superconducting magnet to be used for the polarized target field for the b1b_1 experiment.

The tensor asymmetry of Eq. (27) depends on the polarization direction of the deuteron through the tensor polarization parameters of Eqs. (14). Two natural polarization directions we can consider is i) polarizing along the virtual photon direction N≡NqN\equiv N_q and ii) polarizing along electron beam direction N≡NeN\equiv N_e. For a deuteron polarized along the virtual photon direction, using Eqs. (9) and (14), the tensor polarization parameters for Λd=+1\Lambda_d=+1 reduce to

TLL=Q3,(28)T_{LL}=\frac{Q}{3}, \tag*{(28)}
TLTcos⁡ϕTL=TTTcos⁡2ϕTT=0.(29)T_{LT}\cos\phi_{TL}=T_{TT}\cos2\phi_{TT}=0. \tag*{(29)}

(N=Nq)(N=N_q)

These values result in only two tensor polarized structure functions appearing in the asymmetry of Eq. (27), reducing the complexity of the asymmetry. For this choice of polarization direction we have that P\mathcal{P} and QQ introduced in § 2 are related to the tensor polarization parameters as

P=SL=sz[RF],(30)\mathcal{P}=S_L=s^z[\mathrm{RF}], \tag*{(30)}
Q=3TLL.(31)Q=3T_{LL}. \tag*{(31)}

For a deuteron polarized along the electron beam direction, the polarization parameters take the form

TLL=16(3cos⁡2θq−1)Q=3cos⁡2θq+112Q,(32)T_{LL}=\frac{1}{6}(3\cos^2\theta_q-1)Q=\frac{3\cos2\theta_q+1}{12}Q, \tag*{(32)}
TLTcos⁡ϕTL=−12cos⁡θqsin⁡θq=−sin⁡2θq4Q,(33)T_{LT}\cos\phi_{TL}=-\frac{1}{2}\cos\theta_q\sin\theta_q=-\frac{\sin2\theta_q}{4}Q, \tag*{(33)}
TTTcos⁡2ϕTT=sin⁡2θq2=1−cos⁡2θq4Q,(34)T_{TT}\cos2\phi_{TT}=\frac{\sin^2\theta_q}{2}=\frac{1-\cos2\theta_q}{4}Q, \tag*{(34)}

(N=Ne)(N=N_e)

where 0≤θq≤π0\leq\theta_q\leq\pi is the angle between virtual photon and electron beam in the target rest frame

cos⁡θq=Nq⋅Ne,(35)\cos\theta_q=\boldsymbol{N}_q\cdot\boldsymbol{N}_e, \tag*{(35)}

see Figure 2. This rest frame angle can also be expressed using invariants:

cos⁡θq=1+γ2y/21+γ2,(36)\cos\theta_q=\frac{1+\gamma^2y/2}{\sqrt{1+\gamma^2}}, \tag*{(36)}
sin⁡θq=γ1−y−γ2y2/41+γ2.(37)\sin\theta_q=\frac{\gamma\sqrt{1-y-\gamma^2y^2/4}}{\sqrt{1+\gamma^2}}. \tag*{(37)}

In the Bjorken limit, we have γ→0\gamma\to0, and Eqs. (24) reduce to cos⁡θq=1,sin⁡θq=0\cos\theta_q=1,\sin\theta_q=0. This means that in the Bjorken limit the polarization parameters for polarization along the virtual photon direction (Eqs. (20)) or the electron beam direction (Eqs. (22)) become identical. Using the scaling limit Callan-Gross relations of Eq. (26) and F2=2xdF1=xF1F_2=2x_dF_1=xF_1, Eqs. (17) reduce in the Bjorken limit to

F[UTLL,T]=−2b1,(38)F_{[UT_{LL},T]}=-2b_1, \tag*{(38)}
F[UTLL,L]=F[UTLT]=F[UTTT]=F[UU,L]=0.(39)F_{[UT_{LL},L]}=F_{[UT_{LT}]}=F_{[UT_{TT}]}=F_{[UU,L]}=0. \tag*{(39)}

Consequently, for both polarization considerd here (N=Ne,Nq)(N=N_e,N_q) the tensor asymmetry simplifies in the Bjorken limit to

AT=−23b1F1.(40)A^T=-\frac{2}{3}\frac{b_1}{F_1}. \tag*{(40)}

This is the relation that was used in the b1b_1 extraction of the HERMES result [9].

The HERMES and JLab measurements have moderate Q2≲5 GeV2Q^2\lesssim5\ \mathrm{GeV}^2 and are thus not close the Bjorken limit. For reference, for the two Q2Q^2 values we show in the plots in this paper, we have at x=0.5x=0.5 that γ=0.66\gamma=0.66 (Q2=2 GeV2Q^2=2\ \mathrm{GeV}^2) and γ=0.30\gamma=0.30 (Q2=10 GeV2Q^2=10\ \mathrm{GeV}^2). For this reason, we can attempt to relax some of the limits taken to arrive at Eq. (40). Table 1 shows a summary of the different limits considered in this analysis and the relation between ATA^T and b1b_1 they result in. Next to the Bjorken limit case of Eq. (40) (first row), we retain the finite value for γ\gamma in the other two approximations that we use (second and third row). Here, we still impose b3,b4=0b_3,b_4=0 for the higher twist SF and use the tensor Callan-Gross relation, as otherwise we still have too many unknowns to solve for b1b_1 from a single asymmetry. This approach can be considered including kinematical higher twist corrections, while still neglecting (out of necessity) the dynamical higher twist contributions.

TwistNqN_qNeN_eγ\gammaSFApproximation used for b1/(F1AT)b_1/(F_1A^T)
Leading✓\checkmark✓\checkmark0b2=xb1b_2=xb_1−32-\frac{3}{2}
Kinematical Higher✓\checkmarkFiniteb2=xb1,b_2=xb_1,<br>b3=b4=0b_3=b_4=09(1+ϵγ2)−6−5γ2−2ϵγ4\frac{9(1+\epsilon\gamma^2)}{-6-5\gamma^2-2\epsilon\gamma^4}
Kinematical Higher✓\checkmarkFiniteb2=xb1,b_2=xb_1,<br>b3=b4=0b_3=b_4=018(1+ϵγ2)[(3cos⁡2θq−1)][(−6−5γ2)−2ϵγ4]−[cos⁡θqsin⁡θq]2ϵ(1+ϵ)(9γ+3γ3)−[sin⁡2θq]ϵ(3γ2)\frac{18(1+\epsilon\gamma^2)}{[(3\cos^2\theta_q-1)][(-6-5\gamma^2)-2\epsilon\gamma^4]-[\cos\theta_q\sin\theta_q]\sqrt{2\epsilon(1+\epsilon)}(9\gamma+3\gamma^3)-[\sin^2\theta_q]\epsilon(3\gamma^2)}

Table 1. Summary of the different approximations used to relate b1b_1 to ATA^T. The first row is the Bjorken limit approximation that for both polarization along the electron or photon direction leads to Eq. (40). The second and third row keep contributions proportional to γ\gamma but put the higher twist b3=b4=0b_3=b_4=0. All approximations use the tensor Callan-Gross relation.

Comparing directions: systematic error estimation

For kinematics away from the Bjorken limit, using the approximate relations in the right column of Table 1 to extract b1b_1 from data (measured tensor asymmetries ATA^T and F1F_1 from unpolarized measurements) results in an extracted b1b_1 value that is different from the true b1b_1. The approximations lead to the introduction of an additional systematic error. This systematic error can be checked and quantified using a model calculation, where both the true and extracted b1b_1 values can be computed and compared within the same framework. Within the model, we can compute the structure functions appearing in Eq. (2) (and thus also the b1−4b_{1-4}) and the tensor asymmetry using Eq. (27). Next, we can extract b1b_1 using the approximations listed in Table 1 and compare with the original model calculation of b1b_1, referred to as the true b1b_1. Through Eq. (2), we can disentangle the different contributions to the asymmetry: the different terms, but also within each term the effect of polarization, electron kinematics and additional structure functions. We can study the dependence on model inputs (nucleon structure functions, deuteron wave function) and kinematics (Q2,xQ^2,x dependence) to investigate the robustness of the results.

We use a standard convolution model [5, 6, 7] that calculates the proton–neutron component contribution to the deuteron structure functions. In the convolution model, the unpolarized and tensor polarized deuteron SF are computed through a convolution of unpolarized inclusive nucleon SF with deuteron light-front densities, see Ref. [7] for details of the formalism. For the nucleon SF that enter the calculations, we compare both direct data-driven parametrizations of F2NF_{2N} [17] (which include resonance contributions) and parametrizations based on the leading order relation with the unpolarized partonic distribution function (pdf), using the MSTW2008 [18] and CTEQ [19] pdfs. The deuteron light-front densities are matched to non-relativistic wave functions using the procedure detailed in [14] and we compare deuteron wave functions from Argonne V18 [20], CD-Bonn [21] and Paris [22] parametrizations. We want to stress that no non-nucleonic components are included in the current analysis, for a study of these see [10].

Using the model, we can test some of the assumptions common to all approximations of Table 1. The first is the CallanGross relation b2=xb1b_2=xb_1 which is valid in the Bjorken limit. Figure 4 shows that, even at lower Q2Q^2, using the Callan-Gross relation works quite well. For Q2=2 GeV2Q^2=2\ \mathrm{GeV}^2 the difference is on average ∼20%\sim20\% in the region of interest for the JLab experiment and becomes larger at very small xx or in regions where either of the structure functions has a zero crossing. At Q2=10 GeV2Q^2=10\ \mathrm{GeV}^2 the relative difference is smaller as would be expected for this scaling limit relation and the effect of the resonances is diminished.

Figure 4. Validity of Callan-Gross relation at Q2=2 GeV2Q^2 = 2\ \mathrm{GeV}^2 (2 top panels) and 10 GeV2^2 (bottom 2 panels). Top panels for each Q2Q^2 show the absolute value of the relative difference between b2b_2 and xb1xb_1. This would be identically zero if the Callan-Gross relation is exactly valid. The convolution model calculation uses the SLAC nucleon structure functions and Paris deuteron wave function. Black arrow indicate the central xx-values for the JLab experiment, see Figure 1.

Next, in Figure 5, we show the contributions of the different tensor polarized SF to the numerator of the tensor asymmetry (Eq. (27)) for different polarization directions and Q2Q^2 values. We see that overall F[UTLL,T]F_{[UT_{LL,T}]} is larger than F[UTLL,L]F_{[UT_{LL,L}]}. The higher-twist SF F[UTLT]F_{[UT_{LT}]}, F[UTTT]F_{[UT_{TT}]} do not contribute to the asymmetry when polarization along the virtual photon direction is used, but do contribute when deuteron polarization is away from that direction, such as along the electron beam shown in the bottom panels of Figure 5. We see that F[UTTT]F_{[UT_{TT}]}, which is twist 4 is negligible down to the lowest Q2Q^2 value shown, while the twist-3 F[UTLT]F_{[UT_{LT}]} can be quite sizeable and can cause significant cancellations in the asymmetry when it has the opposite sign of F[UTLL,T]F_{[UT_{LL,T}]}.

Figure 5. Contributions to the numerator of the tensor asymmetry proportional to the different tensor SF appearing in eq. (27). Curves include the polarization parameter and ϵ\epsilon-dependent factors. Comparison between polarization direction along the virtual photon direction (NqN_q, top row) and the electron beam direction (NeN_e, bottom row). The convolution model calculation shown here uses the SLAC nucleon SF and the Paris deuteron wave function.

A second approximation used in Table 1 is neglecting the higher twist b3,b4=0b_3,b_4=0 in the kinematical higher twist approximations. The extent to which b3b_3 and b4b_4 can be neglected in the total asymmetry is shown in Figure 6 for polarization along NqN_q and NeN_e. At Q2=2 GeV2Q^2=2\ \mathrm{GeV}^2, we see that the contribution from b3,b4b_3,b_4 to the asymmetry is not negligible and especially for polarization along the photon direction is on the order or larger than that of the leading twist b1b_1 and b2b_2. For both polarization directions, as the value of Q2Q^2 increases, the contributions proportional to the higher twist structure functions b3b_3 and b4b_4 decrease as expected.

Figure 6. As Figure 5 but now the ATA^{T} numerator contributions are shown with the leading twist b1b_1, b2b_2 terms and higher twist b3b_3, b4b_4 terms.

Figure 7 illustrates a comparison between the different extraction approximations of Table 1. Note that for the leading twist approximation we still have two different curves. We start from asymmetries with different polarization directions (Nq,Ne)(N_q,N_e) but use the same approximation relation (Eq. (40)), which results in different b1b_1 curves. While qualitatively the curves are quite similar, the oscillations due to the nucleon resonance contributions and the sign change in the structure function can result in actual sizeable relative differences between an extracted b1b_1 and the “truth” input. While the kinematic region of the JLab experiment avoids the region with larger resonance effects, the difference is still sizeable. The uncertainties shown on the “truth” curve in Figure 7 were propagated from projected uncertainties of ATA^T for the experiment [23]. Here, we take the entire error on b1b_1 (σb1\sigma_{b_1}) coming from σAT\sigma_{A^T}, the error on the asymmetry and we can write

Figure 7. Comparison at Q2=2 GeV2Q^{2}=2\,\mathrm{GeV}^{2} between the original model input (yellow curve) and the b1b_1 values extracted from the calculated asymmetry. Different curves show the different polarization directions and approximations and follow the notation of Table 1. The convolution model calculation shown here uses the SLAC nucleon SF and the Paris deuteron wave function. Black arrow indicate the central xx-values for the JLab experiment, see Figure 1.

σb1=σATATb1.(41)\sigma_{b_1}=\frac{\sigma_{A^T}}{A^T}b_1. \tag*{(41)}

To quantify the systematic error between the extracted b1b_1 and model truth, we take the calculated grid from the xx-region of interest to the experiment (x∈[0.05,0.5]x \in[0.05, 0.5]) with interval Δx=0.01\Delta x = 0.01 and interpolate 10610^6 points for 1) the true b1b_1, 2) its uncertainty and 3) the extracted b1b_1 estimates that we are quantifying the error for. This provides enough statistics for the analysis. Out of the 10610^6 interpolated points, we take random samples of 10310^3 points and calculate the following estimate

h=∑iN(b1itrue−b1iestimate)2Nσb1,i2.(42)h = \sum_{i}^{N} \frac{\left(b_{1i}^{\mathrm{true}} - b_{1i}^{\mathrm{estimate}}\right)^2}{N\sigma_{b_1,i}^2}. \tag*{(42)}

This corresponds to a numerical approximation of the weighted L2L_2 norm between the true and extracted b1b_1 curve in the interval x∈[0.05,0.5]x \in[0.05, 0.5]. Here N=103N = 10^3 is the number of points in the random sample and the sum runs over the values of each variable in the sample. A value of ⟨h⟩=1\langle h \rangle= 1 means that the systematic error induced by the approximation used in the extraction is of the same size as the other combined errors in the experiment. We use 10310^3 different random samples from the 10610^6 interpolated points to generate the histogram for hh, shown in Figure 8 for one choice of kinematics. That Gaussian nature of the distribution in Figure 8 is representative for all other kinematics considered in this analysis.

Figure 8. Distribution for 10310^{3} generated hh values, using the same model inputs used in Figure 7. Here, b1b_1 was estimated using the leading twist approximation in kinematics with the polarization of the deuteron along NqN_q. The red curve shows a Gaussian fit to the histogram values.

To compare the error estimates, we analyze the means of the Gaussian distributions that were fitted to the histograms, using calculations with different model inputs, polarization directions and extraction approximations. The results are summarized in Figure 9. The interpretation of these bar charts is as follows: lower ⟨h⟩\langle h \rangle corresponds to an estimate closer to the ’true’ b1b_1 and a smaller systematic error. ⟨h⟩=1\langle h \rangle= 1 corresponds to the error of the estimate being equal to the rest of the experimental uncertainty. As depicted in the charts, changing parameters affects which one of the approximations results in the best b1b_1 estimate. On the one hand, we are interested if a certain polarization direction and/or approximation yields a consistently better result. For this we can compare the different bars sharing the same parameter inputs (groups of 4). On the other hand, we want to see trends that would be independent of the model inputs. For that we can compare bars with different parameter labels using the same approximation (bar style and color).

Figure 9. Average ⟨h⟩\langle h\rangle-values for different convolution model inputs (see horizontal axis labels) and different polarization directions and approximations (see legends and Table 1). Top panel shows results for Q2=2 GeV2Q^2 = 2\ \mathrm{GeV}^2, the bottom panel for Q2=10 GeV2Q^2 = 10\ \mathrm{GeV}^2.

By examining the two panels of Figure 9, one can observe how ⟨h⟩\langle h \rangle changes with Q2Q^2. At sufficiently high Q2Q^2, the virtual photon direction consistently provides the better estimate. This is in line with expectations as the NqN_q direction tensor asymmetry has only 2 tensor SF contributing, see § 4. The average error at Q2=10 GeV2Q^2 = 10\ \mathrm{GeV}^2 is about a factor of 10 smaller than at Q2=2 GeV2Q^2 = 2\ \mathrm{GeV}^2 for the SLAC and MSTW nucleon inputs, reflecting the fact that all our considered approximations become better closer to the Bjorken limit. At the lower Q2Q^{2} value, there is no consistent better polarization direction emerging. While one would expect the photon direction to give the better result given that the asymmetry numerator contains less terms, from

Figure 6 we see that the neglected higher twist b3,b4b_3, b_4 are much larger for polarization in the photon direction, thus worsening the used approximation again. While there is little dependence on the deuteron wave function, the results depend quite heavily on the nucleon SF input, with the SLAC parametrization giving a smaller error for polarization along NeN_e (for two of the three deuteron wave functions), while the CTEQ parametrization produces no significant difference and MSTW has smaller errors for the NqN_q direction. Keep in mind that the SLAC parametrization includes resonance contributions which should produce a more realistic estimate at the JLab kinematics. Even though a quasi-elastic nucleon W2>1.85 GeV2W^2 > 1.85\ \mathrm{GeV}^2 cut is imposed, due to the deuteron D-wave the tensor polarized SF receive the largest contributions in the convolution integral from nucleon momenta of a few 100 MeV. Consequently, resonance effects can still contribute in the kinematics of the experiment. The Q2=2 GeV2Q^2 = 2\ \mathrm{GeV}^2 results indicate that the error originating from the approximation in the extraction would be on the order of the other errors. It is worth noting that the approximation using kinematical higher twist corrections (blue bars) never produces a significantly better error than the (simpler) leading twist approximation.

Finally, for further tests of the robustness of our results we did calculations of ⟨h⟩\langle h\rangle using different xx-intervals by varying the upper limit of the interval used in the hh estimation. This is shown in Figure 10. The approach of Eq. (41) has the disadvantage that whenever b1b_1 crosses a zero, its error goes to zero. This causes a large blowup in the value of hh. This also makes regions with smaller uncertainty on b1b_1 contribute the most to the eventual ⟨h⟩\langle h\rangle values, which could bias the conclusions drawn from the results. For example, looking at Figure 7 there is a zero crossing at around x=0.43x = 0.43 and this makes the hh values for the estimates for polarization along NqN_q blowup which can be seen in the left plot in Figure 10. This might lead to the conclusion that whatever curve has had less zero crossings gives the best extraction result. This happens as well for the estimates along NeN_e as can be seen around x=0.63x = 0.63 where there is another zero crossing in Figure 7. The effect is less pronounced for the high Q2Q^2 calculation with NqN_q providing the better estimate across the considered intervals. A constant uncertainty weighs all events similarly in the summation of Eq. (42) which avoids the blowup around a zero crossing. We show results in Figure 11 using a constant absolute uncertainty for which we use a value

Figure 10. ⟨h⟩\langle h\rangle values for different x∈[0.05,xup]x \in[0.05,x^{\mathrm{up}}] intervals. Left panel shows results for Q2=2 GeV2Q^2=2\ \mathrm{GeV}^2, the right panel for Q2=10 GeV2Q^2=10\ \mathrm{GeV}^2. The model inputs for plots shown here uses the SLAC nucleon SF and the Paris deuteron wave function.

Figure 11. As in Figure 10 except the uncertainty is taken to be constant throughout the entire range, see text and Eq. (43) for details.

σb1=σATAT∣min⁡(b1)∣+∣max⁡(b1)∣2.(43)\sigma_{b_1} = \frac{\sigma_{A_T}}{A_T}\frac{|\min(b_1)|+|\max(b_1)|}{2}. \tag*{(43)}

We see as a result that the magnitude of the error estimate is significantly smaller. At high Q2Q^2 we see in Figure 11 that the virtual photon is still the better approximation and the zero crossing blowup has disappeared. For the low Q2Q^2 value, one still sees variation with the considered interval which polarization direction produces the best error. This behavior also persists for other choices of the model inputs, not shown here. Consequently, it is not conclusive which direction produces less error in the kinematical regions of the experiment. What is shown is that at high Q2Q^2 the virtual photon direction does give less error.

Conclusion

An analysis was performed using a deuteron convolution model of the systematic errors induced in the extraction of b1b_1 from a single tensor polarized asymmetry using approximations. In all scenarios, approximations that include some kinematical higher twist corrections did not perform better than the leading twist approximation. While at larger Q2∼10 GeV2Q^2 \sim10\ \mathrm{GeV}^2 values polarizing the deuteron along the virtual photon direction provides a smaller error, at the kinematics relevant to the JLab experiment at lower values of Q2Q^2, there is no clear preference between polarizing along the electron beam or virtual photon direction. Different model inputs yield different results, with the magnitude of the error depending on how the error on b1b_1 was treated. This demonstrates that i) this is an effect that should be included in a proper error analysis, ii) there is no clear preference at low Q2Q^2 in which direction to polarize the deuteron to minimize this error, with the electron beam the more practical choice in the fixed target JLab setup. While these conclusions are based on calculations within one model, it would be interesting to do a similar exercise using a global analysis framework to quantify an error estimate using closure tests. Using measurements with different tensor polarizations at the same kinematics could reduce the uncertainties and in principle allow extraction of all independent structures (See [24] for a similar approach using polarization transfer in elastic eded scattering).

Acknowledgements

We want to thank Elena Long for useful discussions. WC is supported through NSF grants PHY-2111442 and PHY-2239274, AS acknowledges support through a Bridge to Doctorate NSF fellowship at FIU. This work has been supported by EURO-LABS “EUROpean Laboratories for Accelerator Based Science” which received funding from the Horizon Europe research and innovation programme under grant agreement No. 101057511. This material is also based upon work supported by the U.S. Department of Energy, Office of Science, Office of Nuclear Physics under grant DE-FG02-88ER40410. We used Jaxodraw [25] for Figure 2.

References

  1. [1]K. Slifer, et al., Jefferson Lab PAC 40 (2013). URL https://www.jlab.org/exp_prog/proposals/13/PR12-13-011.pdf
  2. [2]E. Long, O. Rondon, P. Solvignon, Rates and statistical uncertainty calculations for a measurement of $A_{zz}$. Tech. rep., University of New Hampshire (2013)
  3. [3]S. Kumano, Eur. Phys. J. A, in press (2024)
  4. [4]P. Hoodbhoy, R. Jaffe, A. Manohar, Nucl.Phys. B312, 571 (1989). DOI 10.1016/0550-3213(89)90572-5DOI
  5. [5]L.L. Frankfurt, M.I. Strikman, Nucl. Phys. A 405, 557 (1983). DOI 10.1016/0375-9474(83)90518-3DOI
  6. [6]H. Khan, P. Hoodbhoy, Phys. Rev. C 44, 1219 (1991). DOI 10.1103/PhysRevC.44.1219DOI
  7. [7]W. Cosyn, Y.B. Dong, S. Kumano, M. Sargsian, Phys. Rev. D95(7), 074036 (2017). DOI 10.1103/PhysRevD.95.074036DOI
  8. [8]W. Cosyn, S. Cotogno, A. Freese, C. Lorcé, Eur. Phys. J. C 79(6), 476 (2019). DOI 10.1140/epjc/s10052-019-6981-3DOI
  9. [9]A. Airapetian, et al., Phys. Rev. Lett. 95, 242001 (2005). DOI 10.1103/PhysRevLett.95.242001DOI
  10. [10]G.A. Miller, Phys. Rev. C 89(4), 045203 (2014). DOI 10.1103/PhysRevC.89.045203
  11. [11]K. Slifer et al.,The deuteron tensor structure function experiment. E12-13-011. Jefferson Lab PAC51. 2023
  12. [12]D.G. Crabb, B. Meyer, Annual Review of Nuclear and Particle Science 47(Volume 47, 1997), 67 (1997). DOI https://doi.org/10.1146/annurev.nucl.47.1.67. URL https://www.annualreviews.org/content/journals/10.1146/annurev.nucl.47.1.67
  13. [13]W. Meyer, et al., Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 244(3), 574 (1986). DOI https://doi.org/10.1016/0168-9002(86)91087-9. URL https://www.sciencedirect.com/science/article/pii/0168900286910879
  14. [14]W. Cosyn, C. Weiss, Phys. Rev. C 102, 065204 (2020). DOI 10.1103/PhysRevC.102.065204DOI
  15. [15]W. Cosyn, C. Weiss, Phys. Lett. B799, 135035 (2019). DOI 10.1016/j.physletb.2019.135035DOI
  16. [16]J. Maxwell, C. Keith. Personal Communication
  17. [17]A. Bodek, et al., Phys. Rev. D 20, 1471 (1979). DOI 10.1103/PhysRevD.20.1471DOI
  18. [18]A.D. Martin, W.J. Stirling, R.S. Thorne, G. Watt, Eur. Phys. J. C 63, 189 (2009). DOI 10.1140/epjc/s10052-009-1072-5DOI
  19. [19]H.L. Lai, J. Botts, J. Huston, J.G. Morfin, J.F. Owens, J.w. Qiu, W.K. Tung, H. Weerts, Phys. Rev. D 51, 4763 (1995). DOI 10.1103/PhysRevD.51.4763DOI
  20. [20]R.B. Wiringa, V.G.J. Stoks, R. Schiavilla, Phys. Rev. C 51, 38 (1995). DOI 10.1103/PhysRevC.51.38
  21. [21]R. Machleidt, Phys. Rev. C 63, 024001 (2001). DOI 10.1103/PhysRevC.63.024001
  22. [22]M. Lacombe, B. Loiseau, J.M. Richard, R. Vinh Mau, J. Cote, P. Pires, R. De Tourreil, Phys. Rev. C 21, 861 (1980). DOI 10.1103/PhysRevC.21.861
  23. [23]E. Long, private communication
  24. [24]D. Abbott, et al., Phys. Rev. Lett. 84, 5053 (2000). DOI 10.1103/PhysRevLett.84.5053
  25. [25]D. Binosi, J. Collins, C. Kaufhold, L. Theussl, Comput. Phys. Commun. 180, 1709 (2009). DOI 10.1016/j.cpc.2009.02.020

Paper details

vlnc:W_hbjupa4g
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

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