Prepared for submission to JHEP DESY 15-008,CP3-Origins-2015-004 DNRF90, DIAS-2015-4 First moment of the flavour octet nucleon parton distribution function using lattice QCD 5 1 0 2 r a M 2 ] t Constantia Alexandroua,b Martha Constantinoua Simon Dinterc Vincent Drachd a l Kyriakos Hadjiyiannakoua Karl Jansena,b,c Giannis Koutsoua Alejandro Vaqueroe - p e aDepartment of Physics, University of Cyprus, P.O. Box 20537, 1678 Nicosia, Cyprus h bComputation-basedScienceandTechnologyResearchCenter(CaSToRC),TheCyprusInstitute,20 [ Constantinou Kavafi Street Nicosia 2121, Cyprus 2 cNIC, DESY, Platanenallee 6, D-15738 Zeuthen, Germany v 4 dCP3-Origins & the Danish Institutefor Advanced Study DIAS, University of Southern Denmark, 3 Campusvej 55, DK-5230 Odense M, Denmark 7 3 eINFN, Sezione di Milano-Bicocca Edificio U2, Piazza della Scienza 3 20126 Milano, Italy 0 . E-mail: [email protected], [email protected] 1 0 5 Abstract: We perform a lattice computation of the flavour octet contribution to the 1 (8) : average quark momentum in a nucleon, hxi . In particular, we fully take the v µ2=4 GeV2 i disconnected contributions into account in our analysis for which we use a generalization X of the technique developed in [1]. We investigate systematic effects with a particular r a emphasis on the excited states contamination. We find that in the renormalization free ratio hxi(3) (with hxi(3) the non-singlet moment) the excited state contributions cancel to a hxi(8) largeextendmakingthisratioapromisingcandidateforacomparisontophenomenological analyses. Our final result for this ratio is in agreement with the phenomenological value hxi(3) and we find, including systematic errors, =0.39(1)(4). hxi(8) Keywords: parton distribution function,lattice QCD Contents 1 Introduction 1 2 Simulation Details 3 2.1 Nucleon matrix elements 4 2.2 Lattice evaluation 5 2.3 Estimation of disconnected loops 6 2.4 Renormalization 7 3 Results 7 3.1 Excited states contamination 9 3.2 Chiral behaviour 12 4 Conclusion 13 1 Introduction Computing nucleon properties from first principles using lattice QCD is a long stand- ing challenge. In particular, moments of parton distribution functions (pdfs) are impor- tant benchmark quantities for lattice calculations and provide insight into the structure of hadrons. The computation of various moments of pdfs is therefore a very active research area for lattice calculations, see [2–11] for recent results. In the work we perform here a first lattice calculation of the octet contribution to the average quark momentum in a nucleon, hxi(8) is presented, whereµ2 denotes therenormalization scale. Thisquantity is of µ2 interest by itself and can be extracted from phenomenological analyses of data from deep inelastic scattering experiments, see below. It therefore can serve as an additional quantity to probe QCD and the structure of hadrons, see [11]. (8) In addition, hxi needs the same renormalization constant as the iso-vector averaged µ2 (3) (8) (3) quark momentum in the nucleon hxi and therefore in the ratio of hxi and hxi µ2 µ2 µ2 the renormalization constant cancels. Although we can evaluate non-perturbatively the renormalization constant, its cancelation in the ratio eliminates any uncertainty related to its determination. Comparing the ratio to a phenomenological analysis can thus help to understand, whether renormalization effects can play a role in the presently observed (3) discrepancy between lattice calculations of hxi and phenomenological determinations of µ2 this quantity. (8) Despite the above given motivations to compute hxi there is so far no value of this µ2 (8) quantity available from a lattice determination. The reason for this is that hxi is very µ2 – 1 – difficult to calculate since it involves dis-connected, singlet contributions. The definition (8) of the flavour octet moment hxi reads : µ2 1 hxi(8) = dx x u(x,µ2)+d(x,µ2)−2s(x,µ2)) (1.1) µ2 Z−1 (cid:2) (cid:3) where q(x,µ) denotes the sum of the parton distribution function of the quark q and anti- (3) quark q¯. Using the same convention, we also define the non-singlet contribution hxi , µ2 1 hxi(3) = dx x u(x,µ2)−d(x,µ2) . (1.2) µ2 Z−1 (cid:2) (cid:3) At a value of the renormalization scale of µ2 = 4GeV2 the two momentshxi(3) µ2=4 GeV2 (8) and hxi can be extracted phenomenologically using parton distribution functions µ2=4 GeV2 determined from deep inelastic scattering data, and read, using the ABM12 pdfs set [12] and the analysis of ref. [13] (3) (8) hxi = 0.153(4), hxi = 0.470(7). (1.3) µ2=4 GeV2 µ2=4 GeV2 (3) (8) The quantities hxi and hxi are related to matrix elements of local operators that µ2 µ2 can be computed in Euclidean space-time and are hence accessible to lattice QCD calcu- lations. Introducing ←→ ←→ Oa = ψγ iD ···iD λaψ (1.4) {µ1···µn} {µ1 µ2 µn} where λa are the Gell-Mann matrices acting on a three flavour quark field ψ = (u,d,s). ←→ With D wedenotethesymmetrizedcovariantlatticederivativeandwiththecurlybrackets thesymmetrizationandsubtractionofthetrace. Therequiredmatrixelementisthengiven by a=3,8 (3,8) hN(p,s)|O |N(p,s)i| = hxi u¯ (p,s)γ P u (p,s). (1.5) {µν} µ2 µ2 N {µ ν} N Fromthedefinitionineq.(1.1)itisclearthattheoctetmatrixelementhN(p,s)|Oa=8 |N(p,s)i {µν} involves dis-connected contributions which have in general a bad signal to noise ratio, re- quire thus a very high statistics and are consequently very difficult to compute on the (8) lattice. In addition, the dis-connected contribution to hxi is an SU(3) breaking µ2 flavour effect and is thus expected to be small. In [1] and [14] we have demonstrated that for the here used (maximally) twisted mass lattice discretization of QCD there are special noise reduction techniques which can help substantially to improve on the signal to noise ratio. Indeed, these techniques allowed us to compute a number of quantities which were very difficult to access before [15, 16]. Inthis work we presenta generalization of theparticular technique used in thecontext of the determination of the σ-terms [1] to calculate non-perturbatively the disconnected (8) contributionrelevantforhxi . Aswewillseebelow,thisgeneralization willindeedprovide µ2 (8) a statistically significant signal for hxi . Note that contrary to the case of the σ-terms, µ2 – 2 – the formalism to compute dis-connected contributions developed here is not limited to the twisted mass formulation. The technique could also be applied to other flavour octet operators which would allow to determine non perturbatively moments of polarized pdfs (8) or the flavour octet axial coupling of the nucleon g for instance, see e.g. [14, 16]. A The paper is organized as follow : after describing the basic ingredients of our com- putation we give the details of the variance reduction technique used in this work. We then present a study of the systematic effects appearing in our calculation and perform an extrapolation of the results to the physical pion mass in order to compare with the phenomenologically obtained values. 2 Simulation Details Thelattice action usedinour simulations includesas active degrees of freedom, besidesthe gluon field, a mass-degenerate light up and down quark doublet as well as a strange-charm quark pair in the sea, a situation which we refer to as the N = 2+1+1 setup. We use f the Iwasaki action [17] for the puregauge action and the twisted mass fermion formulation for the Dirac action. In particular, we make use of the formulation of refs. [18, 19] for the light mass degenerate u–d sector, while the action introduced in refs. [20, 21] is employed for the mass non-degenerate c–s sector. The quark mass parameters of the heavy flavour pair have been tuned so that in the unitary lattice setup the Kaon and D-meson masses, take approximately their experimental values. More information about the N = 2+1+1 f setup scheme and further simulation details can be found in ref. [22, 23]. For the results we will show here, we have employed two values of the lattice spacing determined using the nucleon mass in [2]. They read a ≈ 0.082 fm (β = 1.95) and a ≈ 0.064 fm (β = 2.1). In addition, we will use a number of quark masses corresponding to pion masses in the range of 300 MeV − 500 MeV. The parameters of the ensembles used in this work are summarized in Table 1. label β aµ Volume m [MeV] l PS B35.32 1.95 0.0035 323×64 302 B55.32 1.95 0.0055 323×64 372 B75.32 1.95 0.0075 323×64 432 B85.24 1.95 0.0085 243×48 466 D45.32sc 2.1 0.0045 323×64 372 Table 1. Ensembles used in this work and relevant parameters. Inorderto fixthenotation, weintroducethetwisted mass lattice Dirac operator D f,tm for a doublet of mass degenerate quarks : D [U]= D [U]+iaµ γ τ3 . (2.1) f,tm W f 5 Here D [U] is the Wilson Dirac operator, µ denotes the bare twisted mass and τ3 is the W f third Pauli matrix. For further needs we also introduce the operators D denoting the f,± – 3 – upperandlower flavourcomponents ofD [U], referredtoas theOsterwalder-Seiler (OS) f,tm lattice Dirac operator: 1±τ 3 D [U] = tr D [U] , (2.2) f,± f,tm 2 (cid:26) (cid:27) where tr denotes the trace in flavour space. We will call D [U] the lattice Dirac operator f,± of an Osterwalder-Seiler quark with mass ±µ . f When we discuss below the 2-point and 3-point correlation functions necessary for this work, we will use the so-called physical basis of quark fields denoted as ψ . The physical f field basis is related to the twisted quark field basis, χ , by the following field rotation, f ψf ≡ eiω2fγ5τ3χf and ψf ≡ χfeiω2fγ5τ3, (2.3) where the twist angle ω = π/2 at maximal twist. In addition, ψ with index f = l,s will f f denote quark field doublets of light (l) or strange (s) quarks depending on the mass µ f chosen in the valence sector. Since ψ will always refer to the physical basis we will denote f with u and d the two components of ψ . Following the notation of Eq. (2.2) we will denote l with s the two components of ψ . Employing the OS Dirac operator in the valence sector ± s for the strange quark leads to a mixed action where the strange OS quark mass has been tuned to match within errors the unitary Kaon mass. 2.1 Nucleon matrix elements The nucleon two-point function is defined in the physical basis by C± (t)= tr Γ±hJ (x)J (0)i, (2.4) N,2pts N N ~x X where the source position is fixed to 0 in order to lighten notations and t thus denotes the source-sink separation. We also introduced the parity projectors Γ± = (1 ±γ )/2. The 0 subscriptN referstotheprotonortotheneutronstateforwhichthestandardinterpolating fields are given by the formulae: J = ǫabc ua,TCγ db uc and J = ǫabc da,TCγ ub dc. (2.5) p 5 n 5 (cid:16) (cid:17) (cid:16) (cid:17) Note that using discrete symmetries and anti-periodic boundary conditions in the time direction for the quark fields, one can show that C+ (t)= −C− (T −t). Let us also N,2pts N,2pts recall that an exact symmetry of the action leads to the following relation at a finite value of lattice spacing, C± (t) = C± (t) [24]. In order to improve the overlap between the n,2pts p,2pts ground state and the interpolating fields of the nucleon we use Gaussian smearing of the quark fields appearing in the interpolating fields. We also use APE smearing of the gauge links involved in the Gaussian smearing. The nucleon three-point functions is given by C±,a (t ,t ) = tr Γ±hJ (x )Oa=3,8(x )J (0)i, (2.6) N,3pts s op N s {44} op N x~Xs,~xop a=3,8 where O is one of the twist-2 operators introduced in Eq. (1.4), t is the time of {44} op insertion of the operator, and t denotes the so-called source-sink separation. Note that s – 4 – the precise definition of the strange quark field entering in the operator Oa=8 is postponed {44} to the next subsection. Using the two- and three-point correlators of Eqs. (2.4) and (2.6), we construct the following ratio: +,a C (t ,t ) Ra(t ,t ) ≡ N,3pts s op = hxi(a) +O(e−∆Mtop)+O(e−∆M(ts−top)), (2.7) s op am C+ (t ) bare N N,2pts s wheream is thenucleon massinlattice unitsand∆M isthemassgap between thelowest N nucleon state and the first excited state with the same quantum numbers. One can thus extract from the asymptotic time behaviour of Ra=8(t ,t ) the bare quantity hxi(8) and s op bare (3) correspondingly hxi . bare 2.2 Lattice evaluation a=3,8 While the light quark fields used in the operator O are the unitary fields we use, as {44} mentioned already above, a different action for the valence strange quark. In practice we introduce a doublet of mass degenerate quarks with a mass aµ tuned to reproduce the s unitary Kaon mass. This procedure introduces an error due to the uncertainty on the determination of the matching mass that we will discuss later on but will allow us to use an efficient noise reduction technique that will be explained in the next section. Consider the following operator in terms of the field in the twisted mass basis : ←→ ←→ J8 = χγ iD χ−χ¯ γ iD χ (2.8) {4 4} s {4 4} s Performing the rotation to the physical basis, we obtain : ←→ ←→ ←→ ←→ ←→ ←→ J8 = ψγ iD ψ−ψ¯ γ iD ψ = u¯γ iD u+d¯γ iD d−s¯ γ iD s −s¯ γ iD s {4 4} s {4 4} s {4 4} {4 4} + {4 4} + − {4 4} − (2.9) NotethatJ8 keepsthesameforminthetwobasesandthatJ8 isonlyonepossiblechoicefor a discretization of the operator Oa=8. While the two-point nucleon correlators of Eq. (2.4) {44} give only rise to quark-connected Wick contractions, in general the three-point functions of Eq. (2.6) yield both quark-connected (illustrated in Fig. 1a) and quark-disconnected (illustrated in Fig. 1b) contributions. In the following we will refer to them simply as to connected and disconnected fermionic Wick contractions (or diagrams) and shall write ±,a ±,a ±,a C (t ,t ) = C (t ,t )+D (t ,t ) (2.10) N,3pts s op N,3pts s op N,3pt s op with C±,a (resp. D±,a ) correspondineg to the connected (resp. disconnected) quark dia- N,3pt N,3pt grams, defined as e C±,a (t ,t )= tr Γ±h J (x)Oa (x )J (x ) i , (2.11) N,3pt s op N {44} op N src ~xX,~xop n (cid:2) (cid:3) o e D±,a (t ,t ) = tr Γ±h J (x)J (x ) Oa (x ) i , (2.12) N,3pt s op N N src {44} op ~xX,~xop n (cid:2) (cid:3)(cid:2) (cid:3) o – 5 – t op t op tt t t t ssrrcc src Figure 1. Connected(left)andthedisconnected(right)graphsarisingfromtheWickcontractions of the here considered 3-point functions. wherethesymbol[...]isashorthandforalltheconnectedfermionicWickcontractions. Note that for a = 3, the disconnected part is a O(a2) effect which vanishes in the continuum limit and can thus be neglected. Introducing 1 1 δ(µ,µs)(t ) = tr − , (2.13) ± op D [U] D [U] X~xop ((cid:18) l,± s,± (cid:19)(xop,xop) ) ±,a the contribution of the disconnected fermion loop to D on a given gauge configuration N,3pt U in our setup reads D±,a (t ,t )= hC± (t ) δ(µ,µs)(t )+δ(µ,µs)(t ) i, (2.14) N,3pt s op N,2pts s + op − op (cid:16) (cid:17) ±,a The connected contributions C (t ,t ) have been evaluated using standard tech- N,3pt s op niques for three-point functions (sequential inversions through the sink), see e.g. ref. [2]. e 2.3 Estimation of disconnected loops We describe here the generalization of the variance reduction method for twisted mass fermions introduced and discussed in [1, 14, 25, 26]. Consider the identity D −D = ±iγ a(µ −µ ) (2.15) l,± s,± 5 l s implying that 1 1 1 1 1 1 − = − (D −D ) = ∓ia(µ −µ ) γ , (2.16) l,± s,± l s 5 D D D D D D l,± s,± s,± l,± s,± l,± where we have used Eq. (2.15) to obtain the last equality. For the practical calculation, we introduce a set Ξ of N independent random volume ξ sources, {ξ ,...,ξ ,...,ξ }, satisfying [1] [r] [Nξ] lim ξi (x)∗ξj (y) =δ δij (2.17) [r] [r] xy Nξ→∞ Ξ h i – 6 – where i = 1,...,12 refers to the spin and color indices of the source and [...] denotes the Ξ average over the N noise sources in Ξ . ξ Applying Oa to Eq. (2.16) and taking the trace over spin and colour indices we {44} obtain : ∓ia(µ −µ ) φ∗ γ Oa φ (x) = δ(µ,µs)(t )+O R−1/2 , (2.18) l s [r],l,∓ 5 {44} [r],s,± ± op R X~x h i (cid:16) (cid:17) where φ = (1/D )ξ and φ∗ = ξ∗ (1/D )†. (2.19) [r],s,± s,± [r] [r],l,± [r] l,± For the generation of the random sources we have used a Z noise taking all field 2 components randomly from the set {1,−1}. Notethatδ(µ,µs),andbyconstructionitsvariance,isproportionaltothemassdifference ± µ − µ and vanishes on each configuration in the limit µ → µ . Our approach, thus, l s s l exactly encodes the fact that the disconnected contributions we are interested in vanish in the SU(3) limit. flavour 2.4 Renormalization The renormalization of the operator Oa=3 is known to be multiplicative from our previous work [2] and have been obtained non perturbatively using the methodology developed in µµ [27]. The renormalization factor Z (β) read : DV µµ µµ Z (β = 1.95) = 1.019(4), Z (β = 2.10) = 1.048(5) . (2.20) DV DV Note that an independent calculation performed in [28, 29] gives compatible results. In the limit µ = µ = 0, SU(3) is an exact symmetry of the action and the operators s l flavour a=3,8 O belong to the same flavour multiplet. They thus share the same renormalization {µν} hxi(3) pattern in a mass independent scheme. The ratio is thus renormalization free. hxi(8) 3 Results As a first step, we have investigated the magnitude of the stochastic noise introduced by the method described in section 2.3. To this end, we used a fixed number of gauge configurations for a gauge field ensemble at coupling β = 1.95 and twisted mass parameter aµ = 0.0055. We show in Fig. 2, the ratio Ra=8 for a fixed source-sink separation of l disc. t ∼ 1 fm as a function of t for N = 6 and N = 12. As can be seen, the signal is s op ξ ξ compatible with zero within the errors when using N = 12. Furthermore we observe that ξ the error does not decrease much when the number of stochastic sources is doubled. This means presumably that the error is already dominated by the intrinsic noise of the gauge field fluctuations. We nevertheless decided to use N = 12 throughout this benchmark ξ study of the octet moment. In principle, it would therefore be possible to reduce the error further by using more stochastic noise vectors. However, the contribution of the disconnected 3-point function is small compared to the value of the connected part as demonstrated in Fig. 3 where we – 7 – 0.025 Rdisc. Nx =12 Rdisc. Nx =6 0.020 )) pp a,ta,too0.015 22 11 == ss (t(t 88 0.010 == aa RR 0.005 0.000 0 2 4 6 8 10 12 tt //aa oopp Figure 2. Plot of the contributions R versus t at t = 12a for N = 6 (red down triangle) disc. op s ξ and N =12 (blue uptriangle). The bare massofthe strangequarkis fixed to aµ =0.018and we ξ s use aµ =0.0055 and β =1.95 l show the connected contribution Ra=8 (t = 12a,t ), the disconnected contribution connected s op Ra=8 (t = 12a,t ) and the full correlator Ra=8(t = 12a,t ). In particular, the disconnected s op full s op statistical errors on the connected part are about 0.016 (∼ 2.5%) which is significantly larger than the size of the disconnected contributions. Despite this fact, which would make neglecting the dis-connected contributions tempting, we always include them in the following analysis. Note that the Ra=8 is proportional to the difference between disconnected the light and strange quark mass. Since we are using a mixed action setup, our results depends on an approximate tuning of the strange quark mass. In all the figures we use aµ (β = 1.95) = 0.018 for β = 1.95 and aµ (β = 2.10) = 0.015 for β = 2.10.f Those s s values can be compared to the values obtained for instance in [30] where the strange quark mass has been determined using the Ω− mass and correspond to aµ (β = 1.95) = 0.0194 s and aµ (β = 2.10) = 0.0154. However, by using data on the β = 1.95 ensemble we have s explicitly checked that changing the bare strange quark mass by more than ∼ 10% does not lead to any significant change in the value of the disconnected contribution. – 8 – Ra=8 conn. Ra=8 1.0 disc. Ra=8 full 0.8 ))) ppp ooo a,ta,ta,t 222 0.6 111 === sss (t(t(t 888 === aaa 0.4 RRR 0.2 0.0 0 2 4 6 8 10 12 ttt ///aaa oooppp Figure 3. Plotofthe contributionsR(8) (blue triangle), R(8) (black triangles)andof disconnected connected (8) their sum, R (red filled circles) as function of t at t =12a for aµ =0.0055 and β =1.95. full op s l 3.1 Excited states contamination In order to investigate the contamination of excited states due to the second and third term in Eq. (2.7), we used the same procedure (”open-sink” method) as in [31], namely we study the source-sink dependence for a fixed source to operator separation t ∼ 0.9 fm op ( t = 11a). Details on the technical implementation of the ”open-sink” method can be op foundin[31]. Tothisend,alargestatistics fortheconnected part(∼ 23000 measurements) has been used. We plot in Fig. 4 the resulting renormalized ratios Ra=3,8(t ,t = 11a). s op The results in the isovector case are represented by orange squares and the results in the flavour octet case (including the disconnected piece) are depicted by red dots. We also represent using blue triangles the disconnected contribution to Ra=8(t ,t = 11a). As can s op be seen, the noise for the dis-connected part dominates for large source-sink separation. Nevertheless, we can obtain a reasonable signal up to source-sink separation of about 16a (≈ 1.3 fm). The gray bands indicate the values of hx(a=3,8)i obtained using a fixed source-sink separation calculation with t ∼ 1 fm. Note that the fitting ranges have been s determined choosing the fit with the longest plateau with a confidence level of least 90%. In the flavour octet case we observe that results obtained for t /a > 15 and a fixed source- s operator separation t = 11a are compatible with the result obtained at a fixed source- op – 9 –