PLB34338S03702693(18)30988210.1016/j.physletb.2018.12.053The Author(s)PhenomenologyFig. 1Lmπ plotted against the pion mass for our ensembles listed in Table 1. The color coding for the ensembles is used throughout this work.Fig. 1Fig. 2Lattice data of the ground state nucleon energy, normalized to the continuum expectation (16). The color coding follows Fig. 1.Fig. 2Fig. 3Quark mass obtained from the ratio (see Eq. (20)) utilizing the PCAC relation (19) for the ensembles VI (diamonds) and VIII (circles). The result is given for various initial momenta p→ in units of 2π/L, while the final momentum is always fixed to p→′=0→ in our kinematics. Indeed, the PCAC relation is valid on the threepoint function level, up to O(a2) effects.Fig. 3Fig. 4The continuum extrapolations of the PCAC mass determined from pion twopoint functions (orange band) and nucleon threepoint functions (green band) are consistent within the statistical errors. Both determinations show the leading quadratic behaviour characteristic for O(a)improved currents. Dividing the quark mass by the pion mass squared accounts for the leading pion mass dependence given by the GellMann–Oakes–Renner relation.Fig. 4Fig. 5Ratio of correlation functions (cf. Eq. (15)) corresponding to the l.h.s. of Eq. (22) on ensemble VIII, showing the dominant excited state effects.Fig. 5Fig. 6Correlation function ratios (cf. Eq. (15)) for the axialvector (left, middle) and pseudoscalar currents (right) with three different source sink separations t/a = 9 (blue),12 (green),15 (red) for ensemble VIII with mπ ≈ 150 MeV. In both cases we compare results from the standard current (open symbols) and our modified current (filled symbols). The shown results correspond to an average over all relevant combinations of polarizations and momenta with p→=2π/L and p→′=0→ (i.e., Q2 = 0.073 GeV2). For A0 the problem is clearly visible. The signal for A0⊥ has a significantly smaller statistical error and only shows mild excited state contributions (middle, zoomed), which are resolvable with the multiexponential ansatz given in Eqs. (13) and (14). In contrast, the extent of the excited state contaminations to the data for the pseudoscalar current P is not so obvious. However, subtracting the same excited states causing the problem in the axialvector channel (by using P⊥), one finds that the true ground state plateau lies much higher. The yellow bands indicate the ground state contributions extracted from the fits for O∈{A0⊥,P,P⊥}.Fig. 6Fig. 7Violation of PCACFF. The plot shows the ratio defined in Eq. (25). The filled (open) points are obtained with (without) the excited state subtraction described in Sect. 3.2. The color coding follows Fig. 1.Fig. 7Fig. 8Violation of the pion pole dominance ansatz (26) for the induced pseudoscalar form factor. The filled and open data points are obtained with and without the excited state subtraction described in Sect. 3.2, respectively, and agree within errors. The color coding follows Fig. 1.Fig. 8Fig. 9Results for the form factors. Note that GP is scaledependent; here it is plotted for the MS‾ scale 2 GeV. In the upper panels the results using the excited state subtraction explained in Section 3.2 (filled symbols) are compared to those obtained without this method (open symbols). The largest effect is found in the pseudoscalar form factor, while the others are almost not affected. In the lower panels the yellow band shows the final result for the form factors extrapolated to the physical point using the zexpansion described in Section 4.2. In these plots the transparent points show the subtracted data itself, while the solid ones are parallel transported to the physical point in order to give an intuitive grasp on how good these fits actually describe the data points. The color coding follows Fig. 1.Fig. 9Table 1Details of the ensembles used in the analysis, including the inverse lattice coupling β, the hopping parameter κ, the lattice geometry, the pion mass mπ, and the spatial lattice extent L = aNs in units of mπ−1. The finite volume pion masses were determined in Ref. [52] and the errors include an estimate of both the systematic and statistical uncertainty. The lattice spacings and renormalization factors are listed in Table 2.Table 1Ens.βκNs3×Ntmπ [GeV]Lmπ
I5.200.13596323 × 640.2795(18)3.69

II5.290.13620243 × 480.4264(20)3.71
III323 × 640.4222(13)4.90
IV0.13632323 × 640.2946(14)3.42
V403 × 640.2888(11)4.19
VI643 × 640.2895(07)6.71
VIII0.13640643 × 640.1497(13)3.47

IX5.400.13640323 × 640.4897(17)4.81
X0.13647323 × 640.4262(20)4.18
XI0.13660483 × 640.2595(09)3.82
Table 2Lattice spacings and renormalization factors [32,57], the latter of which include the conversion to the MS‾ scheme at μ = 2 GeV.Table 2βa [fm]ZP(μ)ZA
5.200.0810.464(12)0.7532(16)
5.290.0710.476(13)0.76487(64)
5.400.0600.498(09)0.77756(33)
Table 3Parameters for the zexpansions with t0 = 0 (cf. Eqs. (27) to (29)) representing our mean results for the nucleon form factors at the physical point. For completeness, we also give the values for n = 0,1,2,3, which are fixed via Eq. (31).Table 3nanAanP˜[GeV2]anP[GeV2]
01.251.175.10
1−2.7816.2078.54
211.59−58.27−231.62
3−43.5847.6139.03
470.2327.18410.70
5−49.69−52.15−435.32
612.9818.26133.57
☆RQCD Collaboration.Solving the PCAC puzzle for nucleon axial and pseudoscalar form factorsG.S.BaliabS.CollinsaM.GruberaA.SchäferaP.WeinaT.Wurma⁎thomas.wurm@ur.deaInstitut für Theoretische Physik, Universität Regensburg, 93040 Regensburg, GermanyInstitut für Theoretische PhysikUniversität RegensburgRegensburg93040GermanybDepartment of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, IndiaDepartment of Theoretical PhysicsTata Institute of Fundamental ResearchHomi Bhabha RoadMumbai400005India⁎Corresponding author.Editor: B. GrinsteinAbstractIt has been observed in multiple lattice determinations of isovector axial and pseudoscalar nucleon form factors, that, despite the fact that the partial conservation of the axialvector current is fulfilled on the level of correlation functions, the corresponding relation for form factors (sometimes called the generalized Goldberger–Treiman relation in the literature) is broken rather badly. In this work we trace this difference back to excited state contributions and propose a new projection method that resolves this problem. We demonstrate the efficacy of this method by computing the axial and pseudoscalar form factors as well as related quantities on ensembles with two flavors of improved Wilson fermions using pion masses down to 150 MeV. To this end, we perform the zexpansion with analytically enforced asymptotic behaviour and extrapolate to the physical point.KeywordsLattice QCDNucleon structureForm factorsPartial conservation of the axialvector current1IntroductionThe axial nucleon structure is central for the description of weak interactions and plays a prominent role in longbaseline neutrino experiments, where it is important for a precise determination of the neutrino flux and the cross section for nuclear targets [1–3]. The nonperturbative information encoded in the corresponding form factors is manifold. For example, the isovector axial form factor is linked to the flavor asymmetry in the difference between helicity aligned and antialigned quark densities in impact parameter space [4]. While the axial coupling gA is measured quite precisely in βdecay, the nucleon form factors that encode the spatial structure are much less well known. The axial form factor, GA(Q2), and the induced pseudoscalar form factor, G˜P(Q2), enter the description of (quasi)elastic neutrinonucleon scattering [5–8] and exclusive pion electroproduction [9–12] (e.g., e−p→π−pν). They can also be measured in muon capture [13–16]. For reviews see, e.g., Refs. [16,17].Apart from experimental measurements, there are various tools available that can constrain form factors from the theory side. At small virtualities chiral perturbation theory can be used to obtain valuable constraints (see, e.g., Refs. [12,17,18]). At intermediate and large Q2 the form factors can be calculated (up to some systematic uncertainty of ∼15%) using lightcone sum rules [19,20]. Alternatively, one can use functional renormalization group methods [21].However, the cleanest method for the determination of hadron form factors is lattice QCD. Various determinations of the nucleon couplings and form factors using a wide variety of lattice actions and analysis methods can, e.g., be found in Refs. [22–51]. Recent calculations that have precisely determined the axial, the pseudoscalar, and the induced pseudoscalar form factors separately from lattice data yield an unexpected result: The relation between these form factors inferred from the partial conservation of the axialvector current (denoted the PCACFF relation in the following) is broken rather badly [47–51].11Note, that in Refs. [48,50] the PCAC relation is claimed to be satisfied, however the quark mass used differs by a factor of three from that extracted from the pion twopoint functions. Adding to the confusion, one should note that the PCAC relation itself is still fulfilled quite well on the level of the correlation functions, which leads to the conclusion that either discretization effects or excited state effects are responsible for the observed discrepancy. However, the former have been ruled out as (the sole) explanation in [47], while it was found in [49] that even a 3state fit cannot resolve the issue. Let us note in passing that simply enforcing the PCAC relation on the form factor level might lead to uncontrollable systematic effects.In this work we demonstrate that the largest part of the deviation from the PCACFF relation is indeed due to excited states in the temporal axialvector (A0) and pseudoscalar (P) channels. These excited states are, however, so strongly enhanced relative to the ground state that the usual multistate fit ansatz is bound to fail for any feasible time distances between the source, the sink, and the current insertion. While this is directly visible in A0 correlation functions, which are therefore usually omitted in the analysis when extracting GA(Q2) and G˜P(Q2), up to now the problem has been overlooked in the P channel (which yields GP(Q2)). However, for axialvector threepoint functions, these dominant excited state contributions violate a simple relation, derived from the equation of motion, and can be removed by a straightforward projection. The PCAC relation then suggests that a similar replacement should also be implemented for pseudoscalar threepoint correlation functions.If the reader is now eager to learn the details of the method, he or she should skip the usual description of simulation parameters and analysis methods provided in Section 2, and directly jump to Section 3 where the problem and its resolution will be explained in detail. The results are then presented in Section 4, before we conclude.2Simulation and analysis details2.1Lattice setupIn this work we analyse ensembles with two flavors of nonperturbatively improved clover fermions (also known as Sheikholeslami–Wohlert fermions [53]) and the Wilson gauge action at three different β values corresponding to lattice spacings in the range of 0.060 fm to 0.081 fm. The spacing was set using the Sommer parameter r0=0.5fm determined in Ref. [54] (see also Ref. [55]). The pion masses range from 490 MeV down to an almost physical value of 150 MeV. The ensemble details are provided in Table 1 and Fig. 1 visualizes the landscape of available pion masses and volumes. The multiple volumes for a pion mass of ∼290 MeV, ranging from Lmπ=3.4 to 6.7, enable an investigation of finite volume effects.The isovector form factors can be extracted from connected threepoint functions. The latter have been computed as part of a previous study of the nucleon isovector charges [32] using the traditional sequential source method [56], where the insertion time tins of the local current is varied, while the sink and source times, tsnk and tsrc, are fixed. To increase statistics two measurements of the threepoint functions are performed at different source positions per ensemble. Autocorrelations between configurations are taken into account by binning with a binsize of 10.In order to minimize excited state contributions to the threepoint functions (and to the twopoint functions that are also required), spatially extended source and sink interpolating operators are constructed using Wuppertal smearing [58] with APEsmeared [59] gauge links. In Ref. [32] the smearing was optimized such that ground state dominance was observed in the nucleon twopoint functions at around the same physical time, t=tsnk−tsrc≈0.8fm, for different pion masses and lattice spacings. For key ensembles, labelled III, IV, and VIII, corresponding to pion masses of 420, 290, and 150 MeV, respectively, at the lattice spacing a=0.071fm, multiple sourcesink separations for the threepoint functions were generated to enable an investigation of remaining excited state contamination. The separations correspond to t≈(1.1,1.2)fm for ensemble III, t≈(0.5,0.6,0.8,0.9,1.1,1.2)fm for ensemble IV and t≈(0.6,0.9,1.1)fm for ensemble VIII. The analysis of the isovector charges indicated that, for the smearing applied, the ground state contribution could be reliably determined for t≳1fm and this (single) sourcesink separation was employed for the remaining ensembles.The simultaneous fits to the two and threepoint functions for the present analysis, taking into account the leading excited state contribution, are discussed in Section 2.3.2.2Correlation functions and form factor decompositionsWe analyse the two and threepoint functions(1)C2ptp→(t)=a3∑x→e−ip→x→P+αβ〈ONβ(x→,t)O¯Nα(0→,0)〉,(2)C3pt,Γp→′,p→,O(t,τ)=a6∑x→,y→e−ip→′x→+i(p→′−p→)y→×Γαβ〈ONβ(x→,t)O(y→,τ)O¯Nα(0→,0)〉, where t=tsnk−tsrc, τ=tins−tsrc, ON=(uTCγ5d)u is the Wuppertalsmeared interpolating current for the nucleon with the charge conjugation matrix C, and P+=(1+γ0)/2 projects onto positive parity for zero momentum. We choose Γ to be P+i=P+iγiγ5, i=1,2,3, in our analysis. In our actual simulations, we restrict the kinematics to p→′=0→.Inserting complete sets of states, the correlation functions in Euclidean time can be expanded in terms of hadronic matrix elements. For the twopoint function this yields(3)C2ptp→(t)=∑σP+αβ〈0ONβNσp→〉〈Nσp→O¯Nα0〉e−Ep→t2Ep→+…=Zp→Ep→+mNEp→e−Ep→t+…, where only the ground state contribution is given and mN denotes the nucleon mass. The excited state corrections are discussed in Sect. 2.3. The normalization Zp→ is smearingdependent and encodes the overlap of the ground state with the interpolating operators at the source and the sink,(4)〈0ONβNσp→〉=Zp→up→,σβ, where up→,σβ is a nucleon spinor.An analogous spectral decomposition of the threepoint functions with an operator insertion O∈{P,Aμ} corresponding to the isovector pseudoscalar and axialvector currents,(5)P=u¯γ5u−d¯γ5d,Aμ=u¯γμγ5u−d¯γμγ5d, leads to(6)C3pt,Γp→′,p→,O(t,τ)=Zp→′Zp→BΓ,Op→′,p→e−Ep→′(t−τ)e−Ep→τ+… with(7)BΓ,Op→′,p→=14Ep→′Ep→Tr{Γ(p̸′+mN)J[O](p̸+mN)}. J[O] is defined by the form factor decomposition(8)〈Nσ′p→′ONσp→〉=u¯p→′,σ′J[O]up→,σ. For the different channels it reads(9)J[P]=γ5GP(Q2),(10)J[Aμ]=γμγ5GA(Q2)+qμ2mNγ5G˜P(Q2), where q=p′−p and Q2=−q2.The axial Ward identity yields a partial conservation of the axialvector current, ∂μAμ=2imqP, known as the PCAC relation. On the lattice this relation can be broken by discretization effects. For the nucleon matrix elements it implies:(11)2imq〈Nσ′p→′PNσp→〉=〈Nσ′p→′∂μAμNσp→〉+O(a2). Using the definitions (9) and (10) together with the equations of motion one can deduce the corresponding relation for the form factors (PCACFF):(12)mqGP(Q2)=mNGA(Q2)−Q24mNG˜P(Q2)+O(a2). Eqs. (11) and (12) should be satisfied to a similar degree once the ground state matrix elements have been extracted reliably.2.3Excited states analysisIn threepoint functions the signaltonoise ratio decreases exponentially with the time distance between the source and the sink. Hence, within typical separations t≲1.5fm a sufficient suppression of excited states may not be achieved. Including the leading excited state contributions to Eqs. (3) and (6) gives:(13)C2ptp→(t)=Zp→Ep→+mNEp→e−Ep→t(1+Z˜e−ΔEp→t),(14)C3pt,Γp→′,p→,O(t,τ)=Zp→′Zp→BΓ,Op→′,p→e−Ep→′(t−τ)e−Ep→τ×(1+B10e−ΔEp→′(t−τ)+B01e−ΔEp→τ+B11e−ΔEp→′(t−τ)e−ΔEp→τ),where ΔEp→ denotes the energy gap between the first excited state and the ground state. The excited state coefficient Z̃ depends on the nucleon interpolator, its smearing, and the momenta, while B10, B01, and B11 also depend on the current O and on the projector Γ.For illustrative purposes we define the ratio(15)RΓ,Op→′,p→(t,τ)=C3pt,Γp→′,p→,O(t,τ)C2ptp→′(t)C2ptp→′(τ)C2ptp→′(t)C2ptp→(t−τ)C2ptp→(τ)C2ptp→(t)C2ptp→′(t−τ)⟶t≫τ≫0Ep→′Ep→(Ep→′+mN)(Ep→+mN)BΓ,Op→′,p→, which eliminates the leading order time dependence as well as the overlap factors. Our ground state energies are well described by the continuum dispersion relation(16)Ep→=mN2+p→2, as shown in Fig. 2 and we assume this functional form in our fitting analysis. Excited states will in general contain more than one hadron and hence we make no such assumption for ΔEp→. We remark that the spectrum includes Nπ, Nππ, and higher states and that this spectrum becomes more dense as the pion mass decreases. At zero momentum the lowest multiparticle excited states are Pwave Nπ and Swave Nππ.For a given momentum transfer q2, a simultaneous fit of the form of Eqs. (13) and (14) is performed to the relevant two and threepoint functions, including all available momentum directions, hadron polarizations, and sourcesink separations. The form factors enter the fit directly as parameters, substituting the amplitudes BΓ,Op→′,p→ utilising Eqs. (7)–(10) and the dispersion relation. For ensembles with only one value of t, the parameter B11 is set to zero. The fit range is chosen to be 2a≤τ≤t−2a resulting in reasonable values of χ2/d.o.f.2.4Renormalization and O(a) improvementThe renormalization factors have been calculated nonperturbatively in an RI′MOM scheme using the Rome–Southampton method [60] and then converted into the MS‾ scheme using threeloop continuum perturbation theory. A detailed discussion can be found in [57].The isovector currents are multiplicatively renormalized using(17)OXren=ZX(β)[1+amqbX(β)]OXimp, where the relevant ZX, listed in Table 2, contain both the nonperturbative renormalization and the conversion to the MS‾ scheme at a scale of μ=2GeV. The oneloop improvement coefficients bX have been calculated perturbatively in Refs. [61,62] and are close to unity. The numeric values for our lattices are provided in Ref. [32]. Within the Symanzik improvement program [63,64] also the currents themselves have to be O(a)improved. For the axialvector current this yields(18)Aμimp=Aμ+cAa∂μP, where we used the improvement coefficient cA, nonperturbatively determined in Ref. [65], and ∂μ denotes the symmetrically discretized derivative. Note that for the pseudoscalar current Pimp=P.3PCAC on the form factor level3.1Quark mass from nucleon correlatorsSince the partial conservation of the axialvector current,(19)∂μAμ=2imqP, is an operator relation, it has to hold on the correlation function level such that, using the threepoint functions defined by Eq. (2), the PCAC quark mass can be determined as(20)mq=∂μC3pt,Γp→′,p→,Aμ2iC3pt,Γp→′,p→,P, independent of any spectral analysis. As no significant O(a2Q2) effects are observed within the statistical error, we determine the quark mass by fitting to several Q2 simultaneously (see Fig. 3). Note that the spatial derivatives can be calculated using the formula(21)〈p→′∂iOp→〉=−isin(aqi)a〈p→′Op→〉, as long as one considers only states p→〉 for which the external threemomenta have been fixed via an appropriate Fourier transform in the correlation function (2). The time derivative has to be calculated explicitly, since the energy in the correlation function is not fixed.Usually the PCAC mass is obtained from ratios of pion twopoint functions, where one can achieve very small statistical errors. Up to discretization effects, one would expect these values to agree with those obtained from ratios of baryonic threepoint functions via Eq. (20). This is indeed the case, as depicted in Fig. 4 using ensembles I, IV, and XI (which have different lattice spacings, but very similar pion masses and volumes). Note that despite the unphysical pion mass the extrapolated ratio compares reasonably well with the Nf=2 value in the physical limit: mq/mπ2=0.198(11)GeV−1 [66].3.2Uncovering the ground state contributionIn a number of lattice simulations it has been observed that, even though the PCAC relation is fulfilled on the correlation function level up to discretization effects (in our case of O(a2), cf. Figs. 3 and 4), the equivalent equation for the nucleon form factors, Eq. (12), seems to be broken rather badly. Note that this problem cannot be solved just by using the PCAC mass obtained from the ratio of nucleon threepoint functions discussed above.In the following we will demonstrate that the breaking of the PCAC relation on the form factor level is a consequence of very large excited state effects that cannot be resolved by the standard excited state analysis described in Section 2.3. Using Eq. (10) together with (p̸−mN)up→,σ=0 one finds for the nucleon ground state:(22)p‾μ〈Nσ′p→′AμNσp→〉=0, where p‾=12(p′+p). From Fig. 5 one can easily verify that this equation is violated by the data, demonstrating the presence of large excited state contaminations. This is mostly due to the A0 correlation function, which has an almost linear dependence on the insertion time, cf. the left panel of Fig. 6.One possible (and up to date the most widelyused) workaround to this problem is the exclusion of A0 from the analysis.22For comparison we will show results obtained using this method in Figs. 7–9. A more satisfactory approach is to consider the current(23)Aμ⊥=(gμν−p‾μp‾νp‾2)Aν, which fulfills p‾μAμ⊥=0 by construction. Due to Eq. (22) one can be sure that the subtraction only removes excited state contributions, while leaving the ground state contribution unchanged. For the new current the threepoint function has the expected behaviour, as shown in Fig. 6.The connection between the axial and pseudoscalar channels via the PCAC relation suggests that similar excited state contributions are present in the pseudoscalar channel and it is advantageous to construct the combination(24)P⊥=P−12imqp‾μp‾νp‾2∂μAν, such that ∂μAμ⊥=2imqP⊥. Again, Eq. (22) ensures that the ground state matrix element is not affected by this subtraction. We use the PCAC mass mq obtained from baryon threepoint functions, cf. the discussion in Section 3.1, which we found leads to smaller discretization effects compared to employing the PCAC mass from pion twopoint functions. The effect of Eq. (24) is illustrated in the right panel of Fig. 6: While the original data looked less conspicuous than for the A0 channel, the resulting shift is very significant.To conclude this section, we remark that the subtraction constructed above should not be viewed as an operator improvement, as it depends on the external momentum. Instead, one should interpret it as a method to systematically construct combinations of correlation functions that suffer less from excited states. In principle, this method can be used wherever an exact relation for the ground state matrix elements exists (in this case Eq. (22)). However, the analogous constraint for the vector current, qμ〈Np→′σ′VμNσp→〉=0, is fulfilled almost exactly by the data such that the method does not lead to an improvement.4Results4.1Restoration of PCAC on the form factor levelWe define the ratio (cf. also Ref. [47])(25)rPCAC=mqGP(Q2)+Q24mNG˜P(Q2)mNGA(Q2), where deviations from rPCAC=1 quantify the violation of the PCACFF relation (12). Fig. 7 demonstrates that using the method described in Sect. 3.2 all ensembles, in particular the ones with small pion mass that previously exhibited the largest deviations, now fulfill the PCACFF relation reasonably well. Even for large Q2≈1GeV2 we see a significant improvement, although small deviations of ∼5% remain. This residual violation can be attributed to O(a2) discretization effects of Eq. (19).In absence of better information, the induced pseudoscalar form factor is often estimated by(26)G˜P≈?4mN2GAmπ2+Q2⇒rPPD=(mπ2+Q2)G˜P(Q2)4mN2GA(Q2)=?1, usually called the pion pole dominance (PPD) assumption. Fig. 8 demonstrates that this approximation does not describe the data, especially not at small Q2. This is true for both, the original and the improved data (which reaffirms the findings of Refs. [32,47,50]). Hence, the observed disagreement with the PPD ansatz is certainly not caused by the same excited state effects that have been responsible for the PCACFF violation. Since all data for rPPD collapse onto an almost universal function of mπ2+Q2 it seems highly unlikely that the deviation is due to discretization, volume, or quark mass effects.4.2Parametrization of the form factorsWe parametrize the form factors using the zexpansion [67,68], which automatically imposes analyticity constraints. This corresponds to an expansion of the form factors in the variable(27)z=tcut+Q2−tcut−t0tcut+Q2+tcut−t0, where tcut=9mπ2 is the particle production threshold and t0 is a tunable parameter.33Varying t0 between 0 and tcut/2 has no significant impact on the result. Therefore we have simply set it to zero in our analysis. Isolating the pion pole in the pseudoscalar channels (cf. Ref. [44]) one obtains(28)GA=∑n=0NanAzn,G˜P=1mπ2+Q2∑n=0NanP˜zn,(29)GP=1mπ2+Q2∑n=0NaPnzn. To enforce the correct scaling in the asymptotic limit, GA∝1/Q4, G˜P∝1/Q6, and GP∝1/Q6 [69], one has to implement the four constraints(30)0=∑n=0NnkanX, for k=0,1,2,3, which can be incorporated by fixing the first four coefficients according to(31)akX=(−1)kk!(3−k)!∑n=4Nn!(n−4)!(n−k)anX for k=0,1,2,3, such that one is left with N−3 free coefficients. In the following we will show the fit results with 3 free coefficients (called a z3+4 fit in the literature), since the z2+4 fits failed to describe the data at low momentum transfer, while z4+4 fits did not yield further improvement. In order to extrapolate to the physical point (mπ→mπphys, a→0, L→∞), we use the parametrization(32)anX=bnX+cnXa2+dnXmπ2+enXmπ4+fnXmπ2e−mπLmπL. This allows us to perform a combined fit to all ensembles with 15 fit parameters for each form factor.4.3Form factors, charges, and radiiOur results for the form factors are shown in Fig. 9, where the band shows the value extrapolated to the physical point using the combined fit to all ensembles described in the previous section. Since we know that the PCACFF relation is fulfilled, we use GP(0)=GA(0)mN/mq to obtain a data point for GP in the forward limit (for each ensemble) in order to stabilize the continuum extrapolation of this form factor. In Table 3 we list the zexpansion coefficients corresponding to our central values at the physical point.From the form factors we obtain the charges and the corresponding mean squared radii r2=−6G′(0)/G(0) as(33)gA=GA(0)=1.25(4),rA2=0.79(15)fm2,G˜P(0)=64(23),rP˜2=7.9(4.4)fm2,gP=GP(0)=280(27),rP2=7.3(1.0)fm2. The value of gA is in perfect agreement with the experimental result gA/gV=1.2724(23) [70]. Our relatively large result for the squared axial radius still agrees within errors with zexpansion fits to experimental νd [8] and muon capture [16] data, but lies higher than other lattice results in the range 0.2–0.4 fm2 [44–47], cf. Fig. 7 in Ref. [16]. We remark that our data are also well described by dipole fits, which result in significantly smaller radii.For the induced pseudoscalar coupling defined at the muon capture point [17] we obtain(34)g˜P=mμ2mNG˜P(0.88mμ2)=2.8(7), where mμ=105.6MeV is the muon mass. Note that the small value obtained for g˜P is consistent with the strong violation of the pion pole dominance assumption at small momentum transfer shown in Fig. 8. The experimental value g˜P=8.06(48)(28) from muon capture [15] is consistent with PPD, but not with our data.5SummaryWe have presented a method to identify (and subtract) excited state contributions that spoil the PCAC relation on the form factor level. This mainly affects correlation functions involving the P and A0 currents, which have much larger coupling to the pion at small momentun transfer than Ai. After our subtraction, PCACFF is fulfilled up to small deviations at large momentum transfer, which can be interpreted as lattice artifacts. In spite of this improvement, we find that the pion pole dominance assumption that relates the axial and the induced pseudoscalar form factor is still strongly broken at small momentum transfer, which confirms the findings of Refs. [32,47,50]. A recent calculation within chiral perturbation theory [71] (along the lines of Refs. [72,73] using interpolating currents from Ref. [74], cf. also Refs. [75,76]) indicates that this deviation at small momentum transfer might be due to additional large excited state contributions in the induced pseudoscalar form factor, which is almost unaffected by the subtraction method described in Section 3.2. However, our data do not indicate any presence of these excited states.We parametrize the form factors using the zexpansion and extrapolate them to the physical point by a combined fit to all ensembles at hand. We find that these fits provide a very reasonable description of the data and we use them to extract the values of the charges gA and gP. The result for the axial form factor exhibits a rather steep slope, which is not related to our improvement, but a consequence of the applied z3+4 parametrization. It corresponds to an axial dipole mass MA=12/rA=0.77(8)GeV that is in rough agreement with phenomenological values around 1 GeV. The disagreement of our result for the induced pseudoscalar charge g˜P with experiment could point to persistent problems in the form factor G˜P and warrants further investigation. All our physical results (extrapolated to physical quark masses, to infinite volume, and to the continuum) currently come with relatively large errors that are mainly caused by the narrow range of available lattice spacings. Reducing the systematic error at the physical point (e.g., by the analysis of CLS ensembles with finer lattice spacings [77,78]) will be of high priority in the future.AcknowledgementsMG, PW, and TW are grateful to F. Hutzler, while GB and SC enjoyed fruitful discussions with R. Gupta and K.F. Liu. Many of the ensembles used in this work have been generated on the QPACE supercomputer [79,80], which was built as part of the Deutsche Forschungsgemeinschaft (DFG) Collaborative Research Centre/Transregio 55 (SFB/TRR55). The analyses were performed on the iDataCool cluster in Regensburg and the SuperMUC system of the Leibniz Supercomputing Centre in Munich.References[1]A.A.AguilarArevaloFirst measurement of the muon neutrino charged current quasielastic double differential cross sectionPhys. Rev. D81201009200510.1103/PhysRevD.81.092005arXiv:1002.2680[2]C.L.McGivernCross sections for νμ and ν¯μ induced pion production on hydrocarbon in the fewGeV region using MINERvAPhys. Rev. D94201605200510.1103/PhysRevD.94.052005arXiv:1606.07127[3]T.KatoriM.MartiniNeutrinonucleus cross sections for oscillation experimentsJ. Phys. G45201801300110.1088/13616471/aa8bf7arXiv:1611.07770[4]M.BurkardtImpact parameter space interpretation for generalized parton distributionsInt. J. Mod. Phys. A18200317310.1142/S0217751X03012370arXiv:hepph/0207047[5]L.A.AhrensA study of the axialvector form factor and secondclass currents in antineutrino quasielastic scatteringPhys. Lett. B202198828410.1016/03702693(88)900263[6]T.KitagakiStudy of νd→μ−pps and νd→μ−Δ++(1232)ns using the BNL 7foot deuteriumfilled bubble chamberPhys. Rev. D421990133110.1103/PhysRevD.42.1331[7]A.BodekS.AvvakumovR.BradfordH.BuddExtraction of the axial nucleon form factor from neutrino experiments on deuteriumJ. Phys. Conf. Ser.110200808200410.1088/17426596/110/8/082004arXiv:0709.3538[8]A.S.MeyerM.BetancourtR.GranR.J.HillDeuterium target data for precision neutrinonucleus cross sectionsPhys. Rev. D93201611301510.1103/PhysRevD.93.113015arXiv:1603.03048[9]S.ChoiAxial and pseudoscalar nucleon form factors from low energy pion electroproductionPhys. Rev. Lett.711993392710.1103/PhysRevLett.71.3927[10]V.BernardU.G.MeißnerN.KaiserComment on ‘Axial and pseudoscalar nucleon form factors from low energy pion electroproduction’Phys. Rev. Lett.721994281010.1103/PhysRevLett.72.2810[11]A.LiesenfeldA measurement of the axial form factor of the nucleon by the p(e,e′π+)n reaction at W=1125MeVPhys. Lett. B46819992010.1016/S03702693(99)012046arXiv:nuclex/9911003[12]T.FuchsS.SchererPion electroproduction, partially conserved axialvector current, chiral Ward identities, and the axial form factor revisitedPhys. Rev. C68200305550110.1103/PhysRevC.68.055501arXiv:nuclth/0303002[13]D.H.WrightMeasurement of the induced pseudoscalar coupling using radiative muon capture on hydrogenPhys. Rev. C57199837310.1103/PhysRevC.57.373[14]P.WinterMuon capture on the protonAIP Conf. Proc.1441201253710.1063/1.3700609arXiv:1110.5090[15]V.A.AndreevMeasurement of muon capture on the proton to 1% precision and determination of the pseudoscalar coupling gPPhys. Rev. Lett.110201301250410.1103/PhysRevLett.110.012504arXiv:1210.6545[16]R.J.HillP.KammelW.J.MarcianoA.SirlinNucleon axial radius and muonic hydrogen — a new analysis and reviewRep. Prog. Phys.81201809630110.1088/13616633/aac190arXiv:1708.08462[17]V.BernardL.ElouadrhiriU.G.MeißnerAxial structure of the nucleonJ. Phys. G282002R110.1088/09543899/28/1/201arXiv:hepph/0107088[18]M.R.SchindlerT.FuchsJ.GegeliaS.SchererAxial, induced pseudoscalar, and pionnucleon form factors in manifestly Lorentzinvariant chiral perturbation theoryPhys. Rev. C75200702520210.1103/PhysRevC.75.025202arXiv:nuclth/0611083[19]V.M.BraunA.LenzM.WittmannNucleon form factors in QCDPhys. Rev. D73200609401910.1103/PhysRevD.73.094019arXiv:hepph/0604050[20]I.V.AnikinV.M.BraunN.OffenAxial form factor of the nucleon at large momentum transfersPhys. Rev. D94201603401110.1103/PhysRevD.94.034011arXiv:1607.01504[21]G.EichmannC.S.FischerNucleon axial and pseudoscalar form factors from the covariant Faddeev equationEur. Phys. J. A482012910.1140/epja/i2012120096arXiv:1111.2614[22]G.MartinelliC.T.SachrajdaA lattice study of nucleon structureNucl. Phys. B316198935510.1016/05503213(89)900357[23]T.YamazakiY.AokiT.BlumH.W.LinM.F.LinS.OhtaS.SasakiR.J.TweedieJ.M.ZanottiNucleon axial charge in (2+1)flavor dynamicallattice QCD with domainwall fermionsPhys. Rev. Lett.100200817160210.1103/PhysRevLett.100.171602arXiv:0801.4016[24]T.YamazakiY.AokiT.BlumH.W.LinS.OhtaS.SasakiR.TweedieJ.ZanottiNucleon form factors with 2+1 flavor dynamical domainwall fermionsPhys. Rev. D79200911450510.1103/PhysRevD.79.114505arXiv:0904.2039[25]J.D.BrattNucleon structure from mixed action calculations using 2+1 flavors of asqtad sea and domain wall valence fermionsPhys. Rev. D82201009450210.1103/PhysRevD.82.094502arXiv:1001.3620[26]C.AlexandrouM.BrinetJ.CarbonellM.ConstantinouP.A.HarraudP.GuichonK.JansenT.KorzecM.PapinuttoAxial nucleon form factors from lattice QCDPhys. Rev. D83201104501010.1103/PhysRevD.83.045010arXiv:1012.0857[27]S.CapitaniM.Della MorteG.von HippelB.JägerA.JüttnerB.KnippschildH.B.MeyerH.WittigNucleon axial charge from lattice QCD with controlled errorsPhys. Rev. D86201207450210.1103/PhysRevD.86.074502arXiv:1205.0180[28]J.R.GreenM.EngelhardtS.KriegJ.W.NegeleA.V.PochinskyS.N.SyritsynNucleon structure from lattice QCD using a nearly physical pion massPhys. Lett. B734201429010.1016/j.physletb.2014.05.075arXiv:1209.1687[29]R.HorsleyY.NakamuraA.NobileP.E.L.RakowG.SchierholzJ.M.ZanottiNucleon axial charge and pion decay constant from twoflavor lattice QCDPhys. Lett. B73220144110.1016/j.physletb.2014.03.002arXiv:1302.2233[30]T.BhattacharyaS.D.CohenR.GuptaA.JosephH.W.LinB.YoonNucleon charges and electromagnetic form factors from 2+1+1flavor lattice QCDPhys. Rev. D89201409450210.1103/PhysRevD.89.094502arXiv:1306.5435[31]A.J.ChambersFeynman–Hellmann approach to the spin structure of hadronsPhys. Rev. D90201401451010.1103/PhysRevD.90.014510arXiv:1405.3019[32]G.S.BaliS.CollinsB.GläßleM.GöckelerJ.NajjarR.H.RödlA.SchäferR.W.SchielW.SöldnerA.SternbeckNucleon isovector couplings from Nf=2 lattice QCDPhys. Rev. D91201505450110.1103/PhysRevD.91.054501arXiv:1412.7336[33]G.von HippelT.D.RaeE.ShintaniH.WittigNucleon matrix elements from lattice QCD with allmodeaveraging and a domaindecomposed solver: an exploratory studyNucl. Phys. B914201713810.1016/j.nuclphysb.2016.11.003arXiv:1605.00564[34]T.BhattacharyaV.CiriglianoS.D.CohenR.GuptaH.W.LinB.YoonAxial, scalar, and tensor charges of the nucleon from 2+1+1flavor Lattice QCDPhys. Rev. D94201605450810.1103/PhysRevD.94.054508arXiv:1606.07049[35]A.S.MeyerR.J.HillA.S.KronfeldR.LiJ.N.SimoneCalculation of the nucleon axial form factor using staggered lattice QCDPoSLATTICE20162017179arXiv:1610.04593[36]B.YoonIsovector charges of the nucleon from 2+1flavor QCD with clover fermionsPhys. Rev. D95201707450810.1103/PhysRevD.95.074508arXiv:1611.07452[37]J.LiangY.B.YangK.F.LiuA.AlexandruT.DraperR.S.SufianLattice calculation of nucleon isovector axial charge with improved currentsPhys. Rev. D96201703451910.1103/PhysRevD.96.034519arXiv:1612.04388[38]C.BouchardC.C.ChangT.KurthK.OrginosA.WalkerLoudOn the Feynman–Hellmann theorem in quantum field theory and the calculation of matrix elementsPhys. Rev. D96201701450410.1103/PhysRevD.96.014504arXiv:1612.06963[39]C.AlexandrouM.ConstantinouK.HadjiyiannakouK.JansenC.KallidonisG.KoutsouK.OttnadA.VaqueroNucleon electromagnetic and axial form factors with Nf=2 twisted mass fermions at the physical pointPoSLATTICE2016201715410.22323/1.256.0154arXiv:1702.00984[40]E.BerkowitzAn accurate calculation of the nucleon axial charge with lattice QCDarXiv:1704.01114[41]D.L.YaoL.AlvarezRusoM.J.VicenteVacasExtraction of nucleon axial charge and radius from lattice QCD results using baryon chiral perturbation theoryPhys. Rev. D96201711602210.1103/PhysRevD.96.116022arXiv:1708.08776[42]C.C.ChangA percentlevel determination of the nucleon axial coupling from quantum chromodynamicsNature55820189110.1038/s4158601801618arXiv:1805.12130[43]C.AlexandrouM.ConstantinouK.HadjiyiannakouK.JansenC.KallidonisG.KoutsouA.Vaquero AvilésCascoConnected and disconnected contributions to nucleon axial form factors using Nf=2 twisted mass fermions at the physical pointEPJ Web Conf.17520180600310.1051/epjconf/201817506003arXiv:1807.11203[44]J.GreenN.HasanS.MeinelM.EngelhardtS.KriegJ.LaeuchliJ.NegeleK.OrginosA.PochinskyS.SyritsynUp, down, and strange nucleon axial form factors from lattice QCDPhys. Rev. D95201711450210.1103/PhysRevD.95.114502arXiv:1703.06703[45]C.AlexandrouM.ConstantinouK.HadjiyiannakouK.JansenC.KallidonisG.KoutsouA.Vaquero AvilesCascoNucleon axial form factors using Nf=2 twisted mass fermions with a physical value of the pion massPhys. Rev. D96201705450710.1103/PhysRevD.96.054507arXiv:1705.03399[46]S.CapitaniM.Della MorteD.DjukanovicG.M.von HippelJ.HuaB.JägerP.M.JunnarkarH.B.MeyerT.D.RaeH.WittigIsovector axial form factors of the nucleon in twoflavour lattice QCDarXiv:1705.06186[47]R.GuptaY.C.JangH.W.LinB.YoonT.BhattacharyaAxialvector form factors of the nucleon from lattice QCDPhys. Rev. D96201711450310.1103/PhysRevD.96.114503arXiv:1705.06834[48]N.TsukamotoK.I.IshikawaY.KuramashiS.SasakiT.YamazakiNucleon structure from 2+1 flavor lattice QCD near the physical pointEPJ Web Conf.17520180600710.1051/epjconf/201817506007arXiv:1710.10782[49]Y.C.JangT.BhattacharyaR.GuptaH.W.LinB.YoonNucleon axial and electromagnetic form factorsEPJ Web Conf.17520180603310.1051/epjconf/201817506033arXiv:1801.01635[50]K.I.IshikawaY.KuramashiS.SasakiN.TsukamotoA.UkawaT.YamazakiNucleon form factors on a large volume lattice near the physical point in 2+1 flavor QCDarXiv:1807.03974[51]J.LiangY.B.YangT.DraperM.GongK.F.LiuQuark spins and anomalous Ward identityPhys. Rev. D98201807450510.1103/PhysRevD.98.074505arXiv:1806.08366[52]G.S.BaliS.CollinsB.GläßleM.GöckelerJ.NajjarR.H.RödlA.SchäferR.W.SchielA.SternbeckW.SöldnerThe moment 〈x〉u−d of the nucleon from Nf=2 lattice QCD down to nearly physical quark massesPhys. Rev. D90201407451010.1103/PhysRevD.90.074510arXiv:1408.6850[53]B.SheikholeslamiR.WohlertImproved continuum limit lattice action for QCD with Wilson fermionsNucl. Phys. B259198557210.1016/05503213(85)900021[54]G.S.BaliNucleon mass and sigma term from lattice QCD with two light fermion flavorsNucl. Phys. B8662013110.1016/j.nuclphysb.2012.08.009arXiv:1206.7034[55]R.SommerA new way to set the energy scale in lattice gauge theories and its applications to the static force and αs in SU(2) Yang–Mills theoryNucl. Phys. B411199483910.1016/05503213(94)904731arXiv:heplat/9310022[56]L.MaianiG.MartinelliM.L.PacielloB.TaglientiScalar densities and baryon mass differences in lattice QCD with Wilson fermionsNucl. Phys. B293198742010.1016/05503213(87)900782[57]M.GöckelerPerturbative and nonperturbative renormalization in lattice QCDPhys. Rev. D82201011451110.1103/PhysRevD.82.114511M.GöckelerPhys. Rev. D862012099903ErratumarXiv:1003.5756[58]S.GüskenA study of smearing techniques for hadron correlation functionsNucl. Phys. B, Proc. Suppl.17199036110.1016/09205632(90)90273W[59]M.FalcioniM.L.PacielloG.ParisiB.TaglientiAgain on SU(3) glueball massNucl. Phys. B251198562410.1016/05503213(85)902809[60]G.MartinelliC.PittoriC.T.SachrajdaM.TestaA.VladikasA general method for nonperturbative renormalization of lattice operatorsNucl. Phys. B44519958110.1016/05503213(95)00126DarXiv:heplat/9411010[61]S.SintP.WeiszFurther results on O(a) improved lattice QCD to oneloop order of perturbation theoryNucl. Phys. B502199725110.1016/S05503213(97)003726arXiv:heplat/9704001[62]Y.TaniguchiA.UkawaPerturbative calculation of improvement coefficients to O(g2a) for bilinear quark operators in lattice QCDPhys. Rev. D58199811450310.1103/PhysRevD.58.114503arXiv:heplat/9806015[63]K.SymanzikContinuum limit and improved action in lattice theories. 1. Principles and ϕ4 theoryNucl. Phys. B226198318710.1016/05503213(83)904686[64]K.SymanzikContinuum limit and improved action in lattice theories. 2. O(N) nonlinear sigma model in perturbation theoryNucl. Phys. B226198320510.1016/05503213(83)904698[65]M.Della MorteR.HoffmannR.SommerNonperturbative improvement of the axial current for dynamical Wilson fermionsJ. High Energy Phys.03200502910.1088/11266708/2005/03/029arXiv:heplat/0503003[66]S.AokiReview of lattice results concerning lowenergy particle physicsEur. Phys. J. C77201711210.1140/epjc/s1005201645097arXiv:1607.00299[67]R.J.HillG.PazModelindependent extraction of the proton charge radius from electron scatteringPhys. Rev. D82201011300510.1103/PhysRevD.82.113005arXiv:1008.4619[68]B.BhattacharyaR.J.HillG.PazModelindependent determination of the axial mass parameter in quasielastic neutrinonucleon scatteringPhys. Rev. D84201107300610.1103/PhysRevD.84.073006arXiv:1108.0423[69]C.AlabisoG.SchierholzAsymptotic behavior of form factors for two and threebody bound states. II. Spin12 constituentsPhys. Rev. D111975190510.1103/PhysRevD.11.1905[70]M.TanabashiReview of Particle Physics, Phys. Rev. D98201803000110.1103/PhysRevD.98.030001[71]O.BärNucleon–pionstate contamination in lattice calculations of the axial form factors of the nucleonarXiv:1808.08738[72]O.BärChiral perturbation theory and nucleon–pionstate contaminations in lattice QCDInt. J. Mod. Phys. A322017173001110.1142/S0217751X17300113arXiv:1705.02806[73]O.BärMultihadronstate contamination in nucleon observables from chiral perturbation theoryEPJ Web Conf.17520180100710.1051/epjconf/201817501007arXiv:1708.00380[74]P.WeinP.C.BrunsT.R.HemmertA.SchäferChiral extrapolation of nucleon wave function normalization constantsEur. Phys. J. A47201114910.1140/epja/i2011111495arXiv:1106.3440[75]B.C.TiburziChiral corrections to nucleon two and threepoint correlation functionsPhys. Rev. D91201509451010.1103/PhysRevD.91.094510arXiv:1503.06329[76]B.C.TiburziExcitedstate contamination in nucleon correlators from chiral perturbation theoryPoSCD15201608710.22323/1.253.0087arXiv:1508.00163[77]M.BrunoSimulation of QCD with Nf=2+1 flavors of nonperturbatively improved Wilson fermionsJ. High Energy Phys.02201504310.1007/JHEP02(2015)043arXiv:1411.3982[78]G.S.BaliE.E.ScholzJ.SimethW.SöldnerLattice simulations with Nf=2+1 improved Wilson fermions at a fixed strange quark massPhys. Rev. D94201607450110.1103/PhysRevD.94.074501arXiv:1606.09039[79]H.BaierQPACE – a QCD parallel computer based on Cell processorsPoSLAT2009200900110.22323/1.091.0001arXiv:0911.2174[80]Y.NakamuraA.NobileD.PleiterH.SimmaT.StreuerT.WettigF.WinterLattice QCD applications on QPACEProc. Comput. Sci.4201184110.1016/j.procs.2011.04.089arXiv:1103.1363