Transverse-Spin and Transverse-Momentum Effects in High-Energy Processes
aa r X i v : . [ h e p - ph ] N ov Transverse-Spin and Transverse-Momentum Effectsin High-Energy Processes
Vincenzo Barone , , Franco Bradamante , , Anna Martin , Di.S.T.A., Universit`a del Piemonte Orientale, 15121 Alessandria, Italy INFN, Gruppo Collegato di Alessandria, 15121 Alessandria, Italy Dipartimento di Fisica, Universit`a degli Studi di Trieste, 34127 Trieste, Italy INFN, Sezione di Trieste, 34127 Trieste, ItalyOctober 26, 2018
Abstract
The state of the art concerning transverse-spin and transverse-momentum phenomena in hardhadronic reactions is reviewed. An account is given of single-spin and azimuthal asymmetries insemiinclusive deep inelastic scattering, e + e − annihilation, Drell-Yan production, and hadroproduc-tion. The ongoing experiments and the main theoretical frameworks are described in the first partof the paper. The second part is devoted to the experimental findings and their phenomenologicalinterpretations. A brief discussion of the perspectives of future measurements is finally presented. Contents T -odd couple: Sivers and Boer-Mulders distributions . . . . . . . . . . . . . 193.4 Higher-twist distributions and quark-gluon correlators . . . . . . . . . . . . . . . . . . . 213.5 Generalised parton distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223.6 Distribution functions in the impact-parameter space . . . . . . . . . . . . . . . . . . . 233.7 Model calculations of TMD distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 251.8 Fragmentation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263.8.1 The Collins function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 / e + e − annihilation . . . . . . . . . . . . . . . . . 394.3 Drell-Yan production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.3.1 Kinematics and observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.3.2 DY asymmetries in the TMD approach . . . . . . . . . . . . . . . . . . . . . . . 434.3.3 DY double transverse asymmetries . . . . . . . . . . . . . . . . . . . . . . . . . 454.3.4 DY azimuthal and spin asymmetries in QCD . . . . . . . . . . . . . . . . . . . . 464.4 Inclusive hadroproduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464.4.1 Hadroproduction in the extended parton model . . . . . . . . . . . . . . . . . . 474.4.2 Single-spin asymmetries at twist three . . . . . . . . . . . . . . . . . . . . . . . 484.4.3 Λ production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494.5 Other processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 e + e − annihilation . . . . . . . . . . . . . . . . . . . . . . . . . . 555.2.3 Phenomenology of the Collins effect and determination of transversity . . . . . . 575.2.4 Two-hadron asymmetries in SIDIS and e + e − annihilation . . . . . . . . . . . . . 595.2.5 Λ polarisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625.3 Accessing TMD distributions: T-odd leading twist functions . . . . . . . . . . . . . . . 635.3.1 Sivers effect in SIDIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635.3.2 Boer Mulders effect in SIDIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655.3.3 Boer-Mulders effect in DY production . . . . . . . . . . . . . . . . . . . . . . . . 695.4 Accessing the TMD distributions: leading-twist T -even functions and higher-twist functions 705.5 Inclusive hadroproduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.5.1 SSA’s in inclusive hadroproduction . . . . . . . . . . . . . . . . . . . . . . . . . 725.5.2 Spin-averaged hadroproduction cross sections . . . . . . . . . . . . . . . . . . . 75 Introduction
For quite a long time the common lore in the hadron physics community has been that transversepolarisation effects are negligibly small in hard processes. In the last two decades a growing theoreticaland experimental evidence has shown that this is not the case and that transverse-spin phenomena are,on the contrary, rather relevant in various high-energy hadronic reactions.The prehistory of the subject started in the mid-70s, when substantial single-spin asymmetries(SSA’s) were found in inclusive pion hadroproduction at the center-of-mass energies of the Argonnesynchrotron (few GeV) [1, 2, 3]. At the same time, at Fermilab Λ hyperons produced in unpolarised pN collisions at √ s ≃
24 GeV and moderate transverse momenta P T (below 1.5 GeV) were found topossess a large transverse polarisation [4], a result subsequently confirmed at slightly higher √ s and P T [5]. These findings stimulated both experimental and theoretical work. An experimental programmeto investigate both longitudinal and transverse spin effects in high energy pp and ¯ pp scattering wasproposed in 1978 at FNAL [6] and carried out more than 10 yeas later by the E704 Collaboration.Measuring inclusive pion production in collisions of transversely polarised proton and antiproton beamswith an hydrogen target at the center-of-mass energy √ s = 19 . P T up to 2 GeV, theE704 Collaboration found single-spin asymmetries as large as 40 % in the forward region [7, 8, 9, 10].More recently, the experimental collaborations STAR, PHENIX and BRAHMS, working at RHIC,have confirmed the early Fermilab findings on single-spin asymmetries in hadroproduction, pushing thefrontier of the c.m. energy to √ s = 200 GeV and covering wider kinematical ranges in P T and in theFeynman variable x F [11, 12, 13, 14, 15].On the theoretical side, soon after the first experimental findings of transverse spin effects, Kane,Pumplin and Repko proved in a famous paper [16] that in collinear perturbative QCD (applicable tohigh P T ) SSA’s are of the order of α s ( m q / √ s ) (where m q is the quark mass) and therefore vanish in themassless limit. Also, in other important theoretical works [17, 18, 19] it was shown that non-vanishingtransverse single-spin asymmetries may arise in QCD only if one consider higher-twist contributions(na¨ıvely expected to behave as a power of ( M/P T ), where M is a hadronic scale.It took a while to realise that theory allows for transverse polarisation effects which are in somecases unsuppressed (for a review see f.i. Ref. [20]). In the early 90s various authors [21, 22, 23, 24]rediscovered the distribution of transversely polarised quarks in a transversely polarised nucleon firstintroduced by Ralston and Soper in 1979 [25]. This “transversity” distribution, usually denoted by h ( x ) or by ∆ T q ( x ), is a leading-twist quantity that contributes dominantly to the double transverseasymmetry in Drell-Yan (DY) production. Due to its chiral-odd nature h is not measurable in inclusiveDeep Inelastic Scattering (DIS), where transverse SSA’s are prohibited by time-reversal invariance atlowest order in α em [26]. This argument, however, does not hold in semi-inclusive DIS (SIDIS), whereat least one hadron in the final state is detected on top of the scattered lepton. In SIDIS processesno first principles forbid SSA’s. Various theoretical proposals were soon put forward to measure h [27, 28, 29, 30, 31, 32, 33]. In particular, Collins proposed a mechanism, based on a spin asymmetryin the fragmentation of transversely polarised quarks into an unpolarised hadron (the “Collins effect”),which involves a transverse-momentum dependent (TMD) fragmentation function, H ⊥ . This mechanismwas originally proposed as a “quark polarimeter”, and could be conveniently exploited to measure thetransversity function h ( x ) in SIDIS.In a different approach, one year before the publication of the E704 results, Sivers had suggestedthat single-spin asymmetries could originate, at leading twist, from the intrinsic motion of quarks inthe colliding hadrons [34, 35]. The idea, in particular, was that there exists an azimuthal asymmetryof unpolarised quarks in a transversely polarised hadron (the so-called “Sivers effect”), and a new T -odd TMD distribution function, now commonly called Sivers function and usually denoted by f ⊥ T , wasproposed to describe the partons in a transversely polarised nucleon.3riginally this mechanism seemed to violate time-reversal ( T ) invariance [29] and it was demon-strated that f ⊥ T had to be zero. Brodsky, Hwang and Schmidt [36, 37] proved however by an explicitcalculation that final-state interactions in SIDIS, arising from gluon exchange between the struck quarkand the nucleon remnant, or initial-state interactions in DY, produce a non-zero Sivers asymmetry.The situation was further clarified by Collins [38] who pointed out that, taking correctly into accountthe gauge links in the TMD distributions, time-reversal invariance does not imply a vanishing f ⊥ T , butrather a sign difference between the Sivers distribution measured in SIDIS and the same distributionmeasured in DY.The phenomenological analysis of the E704 results [39, 40], however, showed that both the Collinsand the Sivers effects are at work to generate the observed asymmetries, but a satisfactory theoreticaldescription of the data is still missing today.New experimental opportunities came out in the 90’s, when it was realised that high energy SIDISexperiments were needed to investigate the helicity structure of the nucleons. In fact a major eventhad focused on the nucleon spin the attention of the high energy physics community. In 1988 the EMCCollaboration at CERN, scattering a high energy polarised muon beam on a transversely polarisedproton target obtained a totally unexpected result, namely that the fraction of the nucleon spin carriedby the quarks was small, even compatible with zero, within the accuracy of the measurement. From theinclusive cross-section difference for parallel and antiparallel spins one could extract a linear combinationof the quark helicity distributions ∆ q (or g ), defined as the difference of the quark densities for quarkspin parallel or antiparallel to the longitudinal nucleon spin. Using complementary information on thequark helicities derived from the weak decays of the hyperons it was possible to add up the quarkhelicities, thus obtaining ∆Σ, the overall quark contribution to the nucleon spin. The result, whichcame to be known as the “spin crisis” was at variance with the current paradigm, i.e. the quark modeland the beautifully simple explanation of the baryon magnetic moments. More than one thousandtheoretical papers were written on the subject, and many experiment (SMC at CERN, E142, E143,E154 and E155 at SLAC) were proposed and executed to confirm the effect, to extend the result to theneutron, and to improve the accuracy of the measurement. The confirmation of this finding led to agrowing attention to the other contributions to the proton spin, namely the gluon polarisation and theorbital angular momentum of both quarks and gluons, as well as to a deeper look at the QCD descriptionof the nucleon and to the relation between h and g . Thus, new generation experiments, well suitedto investigate SIDIS with both longitudinally and transversely polarised targets, like COMPASS andHERMES, were proposed and started their operations about 10 years ago.In 2004 HERMES [41] and COMPASS [42, 43] presented the first data collected with transverselypolarised proton and deuteron targets, which showed clear evidence of transverse SSA’s on proton. Oneof the main advantages of SIDIS is that the Collins and Sivers effects, as well as the other TMD effects,are not mixed, as in hadroproduction, but generate different azimuthal asymmetries, which can beseparately explored. Thus, the Collins and Sivers asymmetries could be extracted analysing the samedata.Another major step in the understanding of the Collins effect occurred from the Belle Collaborationstudies of the azimuthal correlation between the hadrons in the two jets created in e + e − annihilations[44]. In the process e + e − → q ¯ q the transverse polarisations of the q ¯ q pair are correlated, thus theCollins effect is expected to cause correlated azimuthal modulations of the hadrons into which the q and the ¯ q fragment. The high precision of the Belle data provided very accurate measurements of suchmodulations, and a combined analysis has allowed a first extraction of both the Collins function and ofthe transversity distribution [45]. The Sivers and Collins effects are by now theoretically well establishedand the overall picture is essentially in agreement with the still limited set of results produced by theSIDIS experiments.All this work on transverse polarisation effects eventually opened up the Pandora box of thetransverse-momentum structure of hadrons. The importance of the intrinsic transverse momentum4f quarks in hadrons has been acknowledged since many years. The transverse momentum of thequarks is responsible for the large azimuthal asymmetries of the hadrons produced in SIDIS processeson unpolarised nucleons (the so-called Cahn effect). In a similar way it is largely responsible for theazimuthal asymmetries observed in DY processes, namely in the production of a lepton pair in hadron-hadron scattering at high energy. When the intrinsic transverse momentum of the quarks in the nucleonis taken into account, several new functions are needed to describe the transverse spin structure of thenucleon. Transverse spin, in fact, couples naturally to the intrinsic transverse momentum of quarks,and the resulting correlations are expressed by various transverse-momentum dependent distributionand fragmentation functions, that give rise to a large number of possible single-spin and azimuthalasymmetries [34, 35, 29, 46, 47, 48, 49]. Of particular interest are the correlations between the quarktransverse momentum and the nucleon spin, the quark spin and the transverse momentum of the frag-menting hadron, and the quark transverse spin and its transverse momentum in an unpolarised nucleon,which give rise to the Sivers function, the Collins function and the so-called Boer-Mulders function re-spectively. All these three functions are (na¨ıvely) T -odd, and all three are responsible for transversespin asymmetries in SIDIS. In particular the Boer-Mulders function [50], measures the transverse-spinasymmetries of quarks inside an unpolarised hadron, and contributes to the cos φ and cos 2 φ azimuthalmodulations in the cross sections of unpolarised SIDIS and DY processes which have been observedsince many years and are presently been accurately measured.The TMD description of hard processes has been put on a firm basis by the proof of a non-collinearfactorisation theorem for SIDIS and DY, in the low transverse momentum regime [51, 52]. On the otherhand, it is known that twist-3 collinear effects, expressed by quark-gluon correlation functions, can alsoproduce single-spin and azimuthal asymmetries [53, 54, 55]. This mechanism works at high transversemomenta, P T ≫ M . Thus, there is an overlap region where both the collinear twist-3 factorisationand the non-collinear factorisation should be both valid. The relation between these two pictures, thatis, between the T -odd TMD functions on one side and the multiparton correlators on the other side,has been clarified in a series of recent papers [56, 57, 58]. These works have opened the way to thederivation of the evolution equations for the TMD functions, a longstanding problem in transverse-spinand transverse-momentum physics [59, 60, 61, 62].Before concluding these introductory remarks, a caveat is in order. This review is far from beingexhaustive. Transverse spin physics is in fact developing so fast that it is nearly impossible to cover allthe results and the ongoing work. The following pages necessarily reflect the specific competence and thepreference of the authors. Among the various processes involving the transverse spin and the transversemomentum structure of hadrons, we chose to focus on SIDIS. From a theoretical viewpoint these are thecleanest and best understood reactions. A related important process, that we will also treat, is hadronpair production in e + e − annihilation, which probes transverse-spin fragmentation functions. As wementioned, a great wealth of data on transverse-spin phenomena come from inclusive hadroproduction.We will pay less attention to these processes, because a recent review [63] is largely dedicated to theirphenomenology. Generalised Parton Distributions will also be only mentioned since they are nicelycovered in a very recent and comprehensive review [64].A comment on the notation is in order. The proliferation of distribution and fragmentation functionsinvolved in transverse-spin and transverse-momentum phenomena makes the issue of notation andterminology a very problematic one. Throughout this paper we adopt for the distribution functionsthe Jaffe-Ji nomenclature [23], extended to transverse momentum dependent distributions by Muldersand collaborators [48, 50] illustrated in detail in Section 3.3.. Thus, f ( x ), g ( x ), and h ( x ) are theunpolarised, the helicity and the transversity distribution functions, respectively, with the subscript1 denoting leading-twist quantities. The main disadvantage of this nomenclature is the use of g todenote a quark distribution function whereas the same notation is universally adopted for one of thetwo structure functions of polarised deep inelastic scattering. Other common names in the literatureare q ( x ) for the unpolarised distribution, ∆ q ( x ) for the helicity distribution, ∆ T q ( x ) for the transversity5istribution (which is also called sometimes δq : here we reserve this name to the tensor charge). Thefragmentation functions are denoted by capital letters: D (unpolarised), G (longitudinally polarised), H (transversely polarised). Thus, D is be the usual leading-twist unpolarised fragmentation function, G the fragmentation function of longitudinally polarised quarks, H the fragmentation function oftransversely polarised quarks. Note that capital letters are also used for the gluon distribution functions[65] and for the generalised parton distributions [66] and the reader should be aware of this possiblesource of confusion.The outline of this paper is the following. In Section 2 a brief account of the technical features of themain ongoing experiments is presented. Section 3 is devoted to the formal aspects of the transverse-spinand transverse-momentum structure of hadrons. In Section 4 we introduce the relevant processes andobservables, and to the theoretical frameworks that describe them. Section 5 illustrates the experimentalfindings about single-spin and azimuthal asymmetries, and their phenomenological interpretations. InSection 6 we discuss the short- and mid-term perspectives of planned and proposed measurements.Finally, Section 7 contains some concluding remarks. Deep inelastic scattering as a tool to unveil the nucleon structure was invented in the late 60’s atSLAC, when for the first time a high energy (1 GeV) electron accelerator became available, a wealthof e-N scattering data were collected, and eventually it became clear that scattering at large transversemomentum could be interpreted as elastic scattering off the nucleon constituents, the “partons”. Fromthe dependence of the cross-section on the energy and the momentum transfered to the nucleon it hasbeen possible to identify the charged partons with the quarks, and assess the existence of the gluons,as carriers of half of the proton momentum. In these experiments, only the scattered electrons weredetected with suitable magnetic spectrometers, and no coincidence experiments were possible, due tothe small duty cycle of the intense electron beam. In the subsequent years the Linear Acceleratorenergy was gradually increased, to reach eventually 50 GeV in the most recent experiments, allowingto measure at larger and larger Q , and at smaller and smaller x . Higher energy experiments couldbe performed at CERN and at FNAL by constructing muon beams from π and K decays: all thesedata and the neutrino-Nucleon data eventually led to the extraction of the parton distribution functions(PDF’s) and of their Q dependence.In a second generation of experiments polarised lepton beams and polarised targets have been used.Sophisticated techniques have been developed to polarise the electron beam at SLAC: in the latestexperiments (for instance E155), 85% beam polarization was typically achieved by using as sourcethe photoelectrons emitted by a gallium arsenide surface. The high energy muon beams on the otherhand were naturally polarised in the weak decay process: by suitably choosing the ratio between muonand pion momentum (0.94 in the EMC experiment at the CERN SPS) muon polarisation of ∼ g and g , and to verify the Bjorken sum-rule. The beam was longitudinally polarised, while the targets werelongitudinally polarised to measure g and transversely polarised to measure g . A variety of solid targetshave been used, butanol, ammonia or LiD, which are kept at very low temperatures ( < . ∼ /σ stat , is proportional to f P tgt , where f is the ratio between the polarisable nucleons and the total number of nucleons in thetarget, and P tgt is the free protons (deuterons) polarisation. These experiments allowed to discover the“spin crisis”, to extract the helicity quark distributions g , to provide first measurements of the quarkcontribution to the nucleon spin ∆Σ, and to verify the validity of the Bjorken sum-rule.A third generation of experiments aiming at the study of the nucleon spin structure started in thepast decade. They still use polarised beams and polarised targets, but complement the detection of thescattered lepton with the reconstruction and identification of the hadrons produced in the fragmentationof the struck quark, the so-called current jet. A suitable trigger system still allows these experimentsto record DIS events, but the main objective of the measurement is SIDIS events. These coincidenceexperiments require the disposal of a continuous beam and of a large acceptance spectrometer with fullparticle identification. At HERA the HERMES experiment was designed to utilise the circulating electron or positron beam.At the experiment, the stored beam (27.5 GeV and 40 mA) passes through a cell, a tube 60 cm long,coaxial with the beam, in which polarised atoms of hydrogen or deuterium are pumped in from an AtomicBeam Source. After diffusing in the storage cell, the atoms are pumped away by a huge pumping systembefore they diffuse into the electron beam pipe. Polarisation is achieved in the Atomic Beam Source byStern-Gerlach filtering followed by radio-frequency transitions to the selected spin state. A schematicview of the target system is given in Fig. 1. This target system is particularly attractive when comparedto the solid polarised targets because there is no dilution of the target polarisation due to the presenceof the unpolarised nucleons bound in the other nuclei present in the material. Of course, the targetthickness cannot be increased at will, not to destroy the circulating electron beam, but densities of 10 nucleons/cm were regularly achieved. After the target, a large acceptance magnetic spectrometer basedon a 1.3 Tm dipole magnet analysed all charged particles up to 170 mrad in the horizontal plane andbetween 40 and 149 mrad in the vertical plane. The reduced acceptance in the vertical plane was dueto the fact that since both the electron and the proton beam pipes of the HERA collider go through themiddle plane of the magnet, its gap was divided into two identical sections by a horizontal septum plate7igure 2: Side view of the COMPASS polarised target. One can see the three target cells inside themixing chamber, the microwave cavity, the solenoid coil, the correction coils, and the end compensationcoil.that shields the electron and the proton beams from the dipole magnetic field. Charge particle trackingis provided by several micro-strip counters, multiwire proportional chambers, and drift chambers locatedbefore, inside and behind the magnet. Charged particle identification is provided by a RICH Cherenkovcounter, while electron-hadron discrimination is achieved with a lead-glass calorimeter with a pre-showerhodoscope in front, and by a Transition Radiation detector. At the end of the spectrometer, a muonhodoscope located after an iron absorber helps the muon identification. The experiment took data withpolarised targets until 2005. After an upgrade to implement the spectrometer with a recoil detector toinvestigate exclusive channels, it took data on unpolarised targets from 2008 to July 2009, when HERAceased operation, and it has afterwards been dismantled.At CERN the COMPASS experiment has been assembled in the Hall 888, where the EMC andafterwards the SMC experiments were installed. Its physics program includes not only the investigationof the spin structure of the nucleon, but also the search of exotic light-quark hadronic states, likeglueballs and hybrids. For spectroscopy the experiment uses hadronic beams (mostly pions and protons),and data have been collected with a liquid hydrogen target and various nuclear targets in 2008 and2009. In the following only the configuration which has been used for the study of the spin structure ofthe nucleon will be described, which uses the high energy muon beam at the CERN SPS. COMPASShas been designed to deal with a beam intensity five times larger than that of the previous EMC orSMC experiments. Typical intensities are 2 · muons per spill (about 5 s every 18 s) at 160 GeV,the momentum at which most of the data have been collected. Since there is no problem of beamradiation in the target, the target length has been chosen as long as possible, within the boundariesof the complexity and cost of the cryogenic system. The target materials which have been used inso far are LiD ( f ≃ .
4) as a deuteron target, and NH ( f ≃ .
15) as a proton target. The targetsystem uses a solenoidal superconducting magnet, providing a highly homogeneous field of 2.5 T overa length of 130 cm along the axis, and is schematically shown in Fig. 2. About one kg of material iscontained in a 4 cm diameter cylinder, coaxial with the beam, over a length of 120 cm, distributed either8igure 3: An artistic view of the polarised He gas target used in Hall A of Jlab for several experiments.For the E06-010 neutron transversity experiment important upgrades have been made, including theisntallation of a third pair of Helmoltz coil, enabling to orient the target spins in any direction.over two cells (in 2002, 2003, and 2004), or over three cells (since 2006). Nearby cells are oppositelypolarised, so that scattering data on the two orientation of the target are taken simultaneously tominimise systematic effects. A cryogenic system allows to keep the target at temperatures of about 0.5K, and to polarise it with the DNP method. Once high enough polarisation values are reached, the R-Fis switched off, the temperature drops to less than 50 mK, the spins get frozen, and data taking canstart. A set of two saddle coils allows to get a transverse field of up to 0.6 T which can be either used toadiabatically rotate the target polarisation from parallel to antiparallel to the beam, or to set it in thetransverse mode, orthogonal to the beam direction. At regular intervals, the polarisation orientationof the target cells are reversed by changing the frequency of the microwaves, so that possible effectsdue to the different acceptances of the different cells can be cancelled in the analysis. The experimentis still running, and in 2010 and 2011 will take more data with a NH polarised proton target, 50%of the time in the transverse polarisation mode and 50% of the time in the longitudinal mode. Largeangular and momentum acceptance is guaranteed by a two-stage magnetic spectrometer, 60 m long,centred around two dipole magnets with 1 Tm and 4.4 Tm bending power respectively. A variety oftracking detectors ensures charged particles tracking from zero to ∼
200 mrad scattering angle, andcharged particle identification is provided by a RICH Cherenkov counter. Two hadronic calorimeters,two electromagnetic calorimeters and two muon filters complement the particle identification and allowthe reconstruction of neutral pions.At Jefferson Lab (JLab) many measurements of electron-nucleon scattering and in particular of DIShave been performed over the past 10 years using the electron beam of CEBAF (Continuous ElectronBeam Accelerator Facility), energies from 0.8 to 6 GeV, and polarised targets. In Hall A, the focushas been on measurements on a neutron target. Thanks to the very high beam current, a polarised He gas target could be used as a neutron target. The advantage of this target is that to first orderone can think that its spin is carried entirely by the neutron, since the two protons have their spinanti-aligned. The He gas fills a pressurised glass vessel (10 atm, typically) and is mixed with Rubidiumvapour whose electrons can be polarised via optical pumping with circularly polarised laser light. The He nuclei get polarised through spin exchange collisions with the Rubidium atoms. A schematic viewof the target system is shown in Fig. 3. These targets can stand a much larger beam intensity than asolid target, without suffering of radiation damage. With a 15 µ A electron beam on a 40 cm target at90 atm a luminosity of 10 cm − s − is achieved. A series of high precision experiments have providedinvaluable information on g and g for the neutron particularly at large x values. The measurementshave utilised a pair of high resolution magnetic spectrometers to measure the scattered electron, andby changing the angular settings and the momentum settings the structure functions could be preciselymeasured over a broad ( Q − W ) plane. On the other hand, no SIDIS measurements were possible, dueto the small angular acceptance of the spectrometers.Complementary measurements, using solid polarised targets (both Ammonia, and Deuterated am-monia) have been carried on in Hall B, using the CEBAF Large Angle Spectrometer (CLAS). Thelarge acceptance of this spectrometer has allowed to detect also hadrons to study SIDIS and exclusiveevents, but the target geometry did not allow to put the polarisation orthogonal to the beam, so thatno transversity measurements have been possible in so far.Recently, in 2009, the first transversity measurements have been performed at JLab, by the E06-010collaboration, in Hall A. Another important feature of the He target is that the field necessary tohold the polarisation is low, so that it is easy with a set of three pairs of Helmoltz coils to rotate thepolarisation in any direction, in particular to a direction orthogonal to the beam and measure SSA’s ona transversely polarised target. The experiment has used the High Resolution Spectrometer to detectthe hadrons in the SIDIS reaction ( e → e ′ π ± ), and a new large acceptance (64 msr) spectrometer,the BigBite spectrometer, to measure the electrons. Thus the first ever measurement of SIDIS ona transversely polarised neutron target should soon be available, to complement the HERMES andCOMPASS proton and deuteron data.The HERMES, COMPASS and JLab experiments have given an important contribution to preciselymeasuring ∆Σ, the quark contribution to the nucleon helicity (0 . ± . ± . G/G , the gluon contribution to the nucleon spin, from the spin asymmetry of the cross sectionof pairs of hadrons. An important result obtained by the CLAS collaboration is the first evidence fora non-zero beam-spin azimuthal asymmetry in the semi-inclusive production of positive pions in theDIS kinematical regime. This effect, observed also by HERMES, is not leading twist, and should giveinformation on quark-gluon correlation. Most relevant to this report, HERMES and COMPASS havebeen the first experiments to measure SIDIS events on transversely polarised targets, and the datathey have collected are still the only ones in this field. HERMES took SIDIS data from 2002 to 2005and new results from those data are coming and will still come. COMPASS took SIDIS data withthe transversely polarised deuteron target in 2002, 2003 and 2004, and with the transversely polarisedproton target in 2007, and more data will be collected in 2010. JLab has carried on measurements withthe transversely polarised He3 target in 2009, but no data exist yet at the time of this report.
A novel attack to transverse spin phenomena is provided by the Relativistic Heavy Ion Collider RHIC,at BNL, which has been designed to accelerate and store not only ion beams, but also polarised protonbeams. As a high-energy polarised proton collider, RHIC by now is the world flagship facility for spineffects in hadron physics. The polarised proton beams are accelerated at the AGS to an energy of about25 GeV, then transferred into the RHIC rings and accelerated up to the desired energy, typically in therange 100- 250 GeV. An overview of the accelerator complex is shown in Fig. 4. The polarisation inRHIC is maintained by two Siberian snakes, namely two sets of four helical dipole magnets which rotatethe proton spin by 180 o . The snakes are placed at two opposite points along the rings, so that the beamdeflection between the two snakes is exactly 180 o and the spin tune is thus equal to a half-integer andenergy independent, thus cancelling the effect of both imperfection and intrinsic resonances. In thisway polarisation values of about 70% have been obtained. Two more sets of the same helical dipolemagnets before and after the two major collider experiments (PHENIX and STAR), for a total of 810igure 4: The Brookhaven hadron facility complex, which includes the AGS Booster, the AGS, andRHIC. Two snakes per ring and four spin rotators per each of the two large experiments (STAR andPHENIX) are also shown.spin “rotators”, allow to set the spin direction from vertical to parallel to the beam and then back tovertical, to allow the experiments to use either longitudinally polarised protons or transversely polarisedprotons. Polarisation-averaged cross-sections for pion production at √ s = 200 GeV have already beenmeasured at RHIC both at mid-rapidity and in the forward region and found to be well described bynext-to-leading order (NLO) pQCD calculations.The major players at RHIC are the three experiments BRAHMS, PHENIX and STAR. Since no spinrotators are installed before and after the BRAHMS experiment, this experiment always took data withtransversely polarised proton beams. The experimental set-up consisted of two movable spectrometerarms to measure charged hadrons over a large rapidity and transverse momentum range. The forwardspectrometer consists of 4 dipole magnets, providing a total bending power of 9.2 Tm, it can be rotatedbetween 2.3 and 15 degrees, and has particle identification capability thanks to a RICH Cerenkovcounter. The second spectrometer uses a single dipole magnet (1.2 Tm bending power), can be rotatedfrom 34 to 90 degrees relative to the beam, and covers a solid angle of approximately 5 msr. Theexperiment completed data taking in 2006. The PHENIX experiment is one of the two large ongoingexperiments investigating the proton spin structure at RHIC. The detector consists of two spectrometerarms at mid-rapidity ( η < .
35) and two larger-rapidity spectrometer arms at 1 . < η < .
4. Themid-rapidity spectrometers identify and track charged particles, and are equipped with electromagneticcalorimeters. Two more electromagnetic calorimeters cover the large rapidity region, 3 . < η < . p T , to probe thegluon polarisation ∆ G/G , and first results have already been published, which confirm the smallness of∆
G/G observed in the SIDIS processes. In the transverse spin mode the large single spin asymmetriesin meson (pion) production in proton-proton scattering already observed in the late 1970’s at lowerenergy at Argonne and at CERN, and in the 1990’s by E704 at FNAL, have been confirmed to persistat the RHIC energies. For inclusive pion SSA’s the overall accumulated statistics by now is so largethat data can be binned in P T , in rapidity and in x F . The behaviour of the SSA’s as a function of thevarious kinematical variables is essential to constrain the parameters of the phenomenological modelsand assess the physical origin of the observed asymmetries. Experiments at electron-positron colliders play a special role in the extraction of the transversity PDF’s.It has been seen in the Introduction that the Collins conjecture, namely that in the hadronisation ofa transversely polarised quark the hadrons of the jet might exhibit a left-right asymmetry relativeto the plane defined by the quark momentum and the quark spin, might conveniently be exploitedto measure the transversity distribution. To unfold the measured Collins asymmetry and extract thetransversity distribution a knowledge of the Collins function is mandatory, but QCD tools are notcapable to calculate either the quark distribution functions, or the fragmentation functions, and theCollins function is no exception. However, such an effect might be detected in high energy e − e + annihilations into two jets, and it was unambiguously observed by the Belle Collaboration [44], analysingdata collected at the asymmetric e − e + KEKB storage rings, as will be explained in Section.4.2. The Belledetector is a large-solid-angle magnetic spectrometer, based on a superconducting solenoidal magnetand many different tracking detectors, calorimeters and Cherenkov counters, which provide excellentparticle identification. Particle identification is particularly important because the Collins effect whichis needed for a global analysis with the SIDIS data is the effect occurring in the fragmentation of thelight quarks. The fragmentation of charm quark or b -quark is expected to dilute the effect, becausehelicity is only conserved for nearly massless quarks, thus removal from the data sample of all the eventsassociated with c - or b - quarks is very important. The measurement of the Drell-Yan cross-sections is not straightforward due to the smallness of the cross-section. Up to now only unpolarised hadrons have been used as either beams or targets, and the quarkPDF’s have been extracted from the coefficients of various angular modulation in the cross-section.The NA10 experiment [67, 68] impinged a high intensity ( ∼ · particles per burst, 95% π − ,corresponding to a mean intensity of 2 · /sec and a duty cycle of 30%) negative beam from theCERN SPS unto a nuclear target, detecting muon pairs in a spectrometer whose analysing magnet hada hexagonal symmetry and produced a toroidal field. The experiment was run at 140 GeV, 194 GeV,and 286 GeV beam momenta, and different targets were used: a tungsten target (either 5.6 or 12 cmlong) at all three beam momenta, and a liquid deuterium target, 120 cm long, placed 2 m upstreamof the tungsten target, for the 286 GeV runs. To reduce the flux in the spectrometer, the target wasfollowed by a beam dump/hadron absorber, placed between the target and the core of the dump, ata distance (120 cm) such that there was no contamination in the data from muon-pairs created in thedump. After the absorber, the trajectories of the two muons were detected with two sets of multiwireproportional chambers, one upstream and one downstream of the spectrometer magnet. In the analysis,12 / Ψ and Υ events were eliminated from the muon-pair sample by suitable cuts in the pair invariantmass, so that the final sample of ∼ . · events contained only DY events in the continuum. Verylarge modulations in cos 2 φ , where φ is the azimuthal angle between the hadron plane and the leptonplane, have been measured, in strong disagreement with the prediction of collinear QCD.Similar results have been obtained at Fermilab by the experiment E615 [69]. The experiment wascarried out in the Proton-West High intensity Area at Fermilab, and the measured reaction again was π − N → µ − µ + X . The negative hadrons (93% π − ) beam momentum was set at 252 GeV, and had anintensity of 2 × particles/sec, with a duty cycle of 33%. The experimental lay-out is similar tothat of the CERN experiment, a nuclear target (a 20 cm long tungsten cylinder) was followed by along (7.3 m) absorber, acting both as beam stopper and hadron absorber. The absorber was made oflight material (beryllium-oxide bricks first, and graphite afterword), inserted between the pole faces ofa dipole magnet, in order to sweep away from the spectrometer the low energy muons (correspondingto low-mass muon-pairs) and at the same time to focus the high mass pairs into the central part of thespectrometer. Low-Z material was chosen for the absorber in order to minimise the multiple scatteringof the muons. In order to extend the measurements to very small muon angles with respect to thebeam, the absorber was uniform in the plane transverse to the beam, and had no central plug of high-Zmaterial, like the uranium/tungsten core of the CERN experiment. After the absorber, a momentumanalysing spectrometer consisting of a system of wire chambers upstream and downstream of a seconddipole magnet was used to measure the muon-pair trajectories. The angular distributions have beenmeasured in the invariant mass region 4 . < M < .
55 GeV, which is free from resonances.Recently the E866 collaboration at Fermilab has carried out measurements of the angular distribu-tions of DY muon pairs produced by scattering 800 GeV protons on a deuterium and a proton target[70]. The Collaboration uses the upgraded Meson-East magnetic pair spectrometer at Fermilab. Theprimary proton beam, with an intensity of ∼ × protons/spill impinges over one of three identical50.8 cm long target flasks containing either liquid hydrogen, liquid deuterium, or vacuum. A copperbeam dump located inside the second dipole magnet absorbs the protons that passed through the target.Very much as in the other experiments, downstream of the beam dump an absorber wall removes allhadrons produced in the target and in the beam dump. The muon trajectories are detected by fourtracking stations (drift chambers) and a momentum analysing dipole magnet. Extrapolating the tracksto a vertex in the target the parameters of each muon track are optimised, the invariant mass of themuon pair is evaluated, and the events from the J/ Ψ and Υ region are rejected from the DY final eventssample. This experiment probes the DF’s of the sea antiquarks, so it provides information which iscomplementary to that of the π beam experiments.To summarise, the study of the DY process is very promising and several new experiments are beingproposed. At variance with all the past experiments, the new experiments will all use polarised beamsand/or polarised targets. Parton distribution functions (PDF’s) and fragmentation functions (FF’s) incorporate a large part ofthe information on the internal structure of hadrons that can be probed in hard processes, that is,in strong-interaction processes characterised by at least one large momentum scale Q . We will startdiscussing the nature and the formalism of PDF’s, and then extend our presentation to the FF’s.Right after the first DIS experiments at SLAC, Feynman proposed the concept of parton distributionsas the probability densities of finding a parton with a certain momentum fraction inside a nucleon [71].In its original formulation, Feynman’s parton model was based on the observation that the time scale ofthe interaction between the virtual photon and the partons is ∼ /Q , hence much smaller than the time13cale of the binding interactions of partons, which is ∼ /M ( M is the nucleon mass) in the target restframe and gets dilated in a reference system where the nucleon moves with a very large momentum (the“infinite-momentum frame”). Thus, we can approximately assume that in DIS the electron interactselastically with free partons, and define the PDF’s as the single-particle momentum distributions of thenucleon’s constituents.Focusing for definiteness on quarks, f ( x ) denotes the number density of quarks carrying a fraction x of the longitudinal momentum of the nucleon. It is not difficult to introduce polarisation into thissimple picture. Consider first a longitudinally polarised nucleon. The helicity distribution function g ( x ) is defined as the helicity asymmetry of quarks in a longitudinally polarised nucleon, that is, thenumber density f + ( x ) of quarks with momentum fraction x and polarisation parallel to that of thenucleon minus the number density f − ( x ) of quarks with the same momentum fraction but antiparallelpolarisation: g ( x ) = f + ( x ) − f − ( x ). In terms of f ± the unpolarised distribution f is simply the sum ofthe two probability densities: f ( x ) = f + ( x ) + f − ( x ). The case of transverse polarisation can be treatedin a similar way: for a transversely polarised nucleon the transversity distribution h ( x ) is defined asthe number density of quarks with momentum fraction x and polarisation parallel to that of the hadron,minus the number density of quarks with the same momentum fraction and antiparallel polarisation,that is, denoting transverse polarisations by arrows, h ( x ) = f ↑ ( x ) − f ↓ ( x ). In a basis of transversepolarisation states, h too has a probabilistic interpretation. In the helicity basis, in contrast, it has nosimple meaning, being related to an off-diagonal quark-hadron amplitude.Moving from this intuitive approach to quantum field theory, the PDF’s admit a rigorous definition interms of correlation functions of parton fields taken at two space-time points with a light-like separation[72, 73] (a modern treatment is given in Ref. [74]). Since DIS probes the parton dynamics on the light-cone (see, e.g., Ref. [75]), it is convenient to introduce here some notions concerning light-cone geometry.The light-cone components of a four vector a µ are defined as a ± = ( a ± a ) / √
2, and grouped in tripletsof the form a µ = ( a + , a − , a T ), where the transverse bi-vector is a T = ( a , a ). The norm of a µ isgiven by a = 2 a + a − − a T . It is customary to define two light-like vectors n + = (1 , , T ) and n − =(0 , , T ), sometimes called “Sudakov vectors”, which identify the longitudinal direction and are suchthat n + · n − = 1. Any vector a µ can be written as a µ = a + n µ + + a − n µ − + a µT , where a µT = (0 , , a T ). Thisis the four-dimensional generalisation of the familiar decomposition of a three-vector into longitudinaland transverse components with respect to a given direction. The reference frame of DIS (or SIDIS)is chosen so that the nucleon’s momentum is purely longitudinal: P µ ≃ P + n µ + , where the approximateequality means that we are neglecting the nucleon mass (a legitimate approximation in the deep inelasticlimit). The infinite momentum frame corresponds to P + → ∞ . Dominant contributions to DIS are O ( P + ), whereas subleading corrections are suppressed by inverse powers of P + , or equivalently, in termsof the momentum transfer, by inverse powers of Q .The field-theoretical expression of the quark number density f ( x ) is (we postpone the QCD sub-tleties) f ( x ) ∼ Z d ξ − e i xP + ξ − h N | ψ † (+) (0) ψ † (+) ( ξ ) | N i , (1)with ξ + = 0 , ξ T = T . The peculiarity of eq. (1) is the appearance of the so-called “good” components ψ (+) of the quark fields ψ . These admit the general decomposition ψ = ψ (+) + ψ ( − ) , with ψ ( ± ) = γ ∓ γ ± ψ . The good components ψ (+) are the dominant ones in the infinite momentum frame, whereasthe “bad” components ψ ( − ) are not dynamically independent: using the equations of motion, they canbe eliminated in favour of “good” components and terms containing quark masses and gluon fields. Dueto the structure of eq. (1), one can insert between the quark fields a complete set of intermediate states | X i , obtaining a modulus squared: f ( x ) ∼ X X δ (cid:0) P + − xP + − P + X (cid:1) |h N | ψ (+) (0) | X i| . (2)14 j Φ Figure 5: The quark correlation matrix Φ.This expression confirms in a field-theoretical form the probabilistic interpretation of PDF’s: f ( x ) is theprobability to extract from the nucleon N a quark with longitudinal momentum xP + , leaving an inter-mediate state X of longitudinal momentum P + X . A similar reasoning applies to polarised distributionsand yields g ( x ) ∼ X X δ (cid:0) P + − xP + − P + X (cid:1) (cid:8) |h N |Q + ψ (+) (0) | X i| − |h N |Q − ψ (+) (0) | X i| (cid:9) , (3) h ( x ) ∼ X X δ (cid:0) P + − xP + − P + X (cid:1) (cid:8) |h N |Q ↑ ψ (+) (0) | X i| − |h N |Q ↓ ψ (+) (0) | X i| (cid:9) , (4)where Q ± and Q ↑↓ are the helicity and transversity projectors, respectively.If the quarks are perfectly collinear with the parent hadron, the three distribution functions we havementioned so far, f ( x ), g ( x ), h ( x ), exhaust the information on the internal dynamics of hadronsat leading twist, i.e., at zeroth order in 1 /Q (for an operational definition of twist, see Ref. [75]). Ifinstead we admit a non negligible quark transverse momentum, the number of distribution functionsconsiderably increases. At leading twist, there are eight of them. In order to understand their originand meaning, it is necessary to adopt a more systematic approach. Quark distribution functions are contained in the correlation matrix Φ (Fig. 5), defined asΦ ij ( k, P, S ) = Z d ξ (2 π ) e i k · ξ h P, S | ¯ ψ j (0) W [0 , ξ ] ψ i ( ξ ) | P, S i . (5)where | P, S i is the nucleon state of momentum P µ and polarisation vector S µ , i and j are Dirac indicesand a summation over colour is implicit. The Wilson line W , which guarantees the gauge invariance ofthe correlator, is a path-ordered exponential of the gluon field (see below) arising from multigluon finalstate interactions between the struck quark and the target spectators. The presence of this gauge linkintroduces in principle a path-dependence in Φ, which in some cases turns out to be highly non trivial(Section 3.3).Integrating Φ( k, P, S ) over the quark momentum, with the condition x = k + /P + that defines x asthe fraction of the longitudinal momentum of the nucleon carried by the quark, yieldsΦ( x ) = Z d k Φ( k, P, S ) δ ( k + − xP + ) = Z d ξ − π e i xP + ξ − h P S | ψ (0) W − [0 , ξ ] ψ ( ξ ) | P, S i| ξ + =0 , ξ T = , (6)where the Wilson line W − [0 , ξ ] connects (0 , , T ) to (0 , ξ − , T ) along the n − direction and reads ( P denotes path ordering) W − [0 , ξ ] = P exp − i g Z ξ − d z − A + (0 , z − , T ) ! . (7)15n the light-cone gauge, A + = 0, the Wilson link reduces to unity and can be omitted. The situation ismore complicated in the case of transverse-momentum distributions, which are defined in terms of fieldseparations of the type (0 , ξ − , ξ T ): we shall return to this issue in Section 3.3.Φ( x ) contains the collinear (i.e., x -dependent, k T -integrated) quark distribution functions. In-troducing the longitudinal and transverse components of the the polarisation vector of the nucleon, S µ = ( S L /M ) P µ + S µT , the expression of Φ( x ) at leading twist, that is at leading order in P + , isΦ( x ) = 12 ( f ( x ) /n + + S L g ( x ) γ /n + + h ( x ) γ [ /S T , /n + ]2 ) . (8)Here one sees the three distributions already introduced: the number density f ( x ), the helicity dis-tribution g ( x ) and the transversity distribution h ( x ), first identified by Ralston and Soper [25]. Thequark distributions can be extracted from (8) by tracing Φ with some Dirac matrix Γ. We will usethe notation Φ [Γ] ( x ) ≡ Tr [Φ( x ) Γ]. The explicit expressions of the leading-twist distributions are (thetransverse Dirac matrix γ T is either γ or γ ): f ( x ) = Φ [ γ + ] ( x ) = Z d ξ − π e i xP + ξ − h P, S | ¯ ψ (0) W − [0 , ξ ] γ + ψ ( ξ ) | P, S i| ξ + =0 , ξ T = T , (9) g ( x ) = Φ [ γ + γ ] ( x ) = Z d ξ − π e i xP + ξ − h P, S | ¯ ψ (0) W − [0 , ξ ] γ + γ ψ ( ξ ) | P, S i| ξ + =0 , ξ T = T , (10) h ( x ) = Φ [ γ + γ T γ ] ( x ) = Z d ξ − π e i xP + ξ − h P, S | ¯ ψ (0) W − [0 , ξ ] γ + γ T γ ψ ( ξ ) | P, S i| ξ + =0 , ξ T = T . (11)In QCD the operators appearing in (9-11) are ultraviolet divergent, so they have to be renormalised.This introduces a scale dependence into the distribution functions, f ( x ) → f ( x, µ ), etc., which isgoverned by the renormalisation group equations, the well known DGLAP equations [76, 77, 78]. The main properties of the “third” parton density, the transversity distribution h , eq. (11), are: i) it is chirally-odd and therefore does not appear in the handbag diagram of inclusive DIS, which cannotflip the chirality; in order to measure h , the chirality must be flipped twice, so one always needs twohadrons, both in the initial state, or one in the initial state and one in the final state, and at least oneof them must be transversely polarised ; ii) there is no gluon transversity distribution: this would imply a helicity-flip gluon-nucleon amplitude,which does not exist since gluons have helicity ± h have been worked out at leading order [21], and years later at next-to-leading order [79, 80, 81]. There are two noteworthy features of the evolution of h : first of all,since there is no gluon transversity distribution, h does not mix with gluons and evolves as a non-singlet density [21]; second, at low x , h is suppressed by the evolution with respect to g [82]. Thishas important consequences for those observables that involve h at low x and large Q , such as theDrell-Yan double transverse asymmetry at collider energies [83].The transversity distribution satisfies a bound discovered by Soffer [84]: | h ( x ) | ≤
12 [ f ( x ) + g ( x )] . (12)This inequality, which is derived in the context of the parton model from the expressions of the distribu-tion functions in terms of quark-nucleon forward amplitudes, is strictly preserved in leading-order QCD1682, 85]. At next-to-leading order, parton densities are not univoquely defined, but a regularisationscheme can be chosen such that the Soffer inequality is still valid [81].The integral of Φ( x ) over x gives the local matrix element h P, S | ¯ ψ (0) ψ (0) | P, S i , which can beparametrised in terms of the vector, axial and tensor charge of the nucleon. In particular, the tensorcharge (that we call δq , for the flavour q ) is given by the matrix element of the operator ¯ ψ i σ µν γ ψ , h P, S | ¯ ψ q (0)i σ µν γ ψ q (0) | P, S i = 2 δq ( S µ P ν − S ν P µ ) , (13)and is related to the transversity distributions as follows Z d x [ h q ( x ) − ¯ h q ( x )] = δq . (14)Note that, due to the charge-conjugation properties of ¯ ψ i σ µν γ ψ , which is a C -odd operator, the tensorcharge is the first moment of a flavour non-singlet combination (quarks minus antiquarks).An important distinction between transverse spin and transverse polarisation [22] is in order. Thetransverse spin operator, i.e. the generator of rotations, for a quark is Σ T = γ γ γ T , and does notcommute with the free quark Hamiltonian H = α z p z . Thus, there are no common eigenstates of Σ T and H : said otherwise, in a transversely polarised nucleon quarks cannot be in a definite transverse spinstate. The distribution related to Σ T , called g T ( x ), is a twist-three quantity that reflects a complicatedquark-gluon dynamics with no partonic interpretation. On the other hand, the transversity distribution h carries information about the transverse polarisation of quarks inside a transversely polarised nucleon.The transverse polarisation operator is Π T = γ Σ T , and commutes with H , owing to the presence ofan extra γ . Therefore, in a transversely polarised nucleon, quarks may exist in a definite transversepolarisation state, and a simple partonic picture applies to h .The argument above shows that the integral R d x ( h q + ¯ h q ) does not represent the quark + antiquarkcontribution to the transverse spin of the nucleon. A transverse spin sum rule containing the firstmoment of h + ¯ h has been derived in Ref. [86] within the parton model, but, in the light of what wehave just said and of other general considerations, is subject to some controversy (see the discussion inRef. [64]). A sum rule for the total angular momentum of transversely polarised quarks in an unpolarisedhadron [87, 88], involving the generalised parton distributions, will be introduced in Section 3.5. The intrinsic transverse motion of quarks, is a source of azimuthal and spin asymmetries in hadronicprocesses. Taking into account its transverse component, the quark momentum is given by k µ = xP + n µ + + k µT . As we will see later in this Section, at leading twist there are eight TMD distributions:three of them, once integrated over k T , yield f , g , h ; the remaining five are new and vanish upon k T integration. Integrating Φ( k, P, S ) over k + and k − only, one obtains the k T -dependent correlationmatrix Φ( x, k T ) = Z d k + Z d k − Φ( k, P, S ) δ ( k + − xP + ) , (15)which contains the TMD distribution functions. The field-theoretical expression of Φ( x, k T ) [89] turnsout to be quite complicated due to the structure of the gauge link, which now connects two space-timepoints with a transverse separation. One has [90, 91]Φ( x, k T ) = Z d ξ − π Z d ξ T (2 π ) e i xP + ξ − e − i k T · ξ T × h P, S | ¯ ψ (0) W − [0 , ∞ ] W T [0 T , ∞ T ] W T [ ∞ T , ξ T ] W − [ ∞ , ξ ] ψ ( ξ ) | P, S i| ξ + =0 , (16)17 (0;0) (1;1)(1;0) (1;1)(1;(cid:24))((cid:24)(cid:0);(cid:24)) PP Figure 6: The gauge-link structure of TMD distributions in SIDIS.with two longitudinal Wilson lines directed along n − , from (0 , , T ) to (0 , ∞ , T ) and from (0 , ∞ , ξ T ) to(0 , ξ − , ξ T ), and two Wilson lines W T at ξ − = ∞ containing the transverse gluon field A µT (Fig. 6). Thislink structure, with the longitudinal Wilson lines W − running to ξ − = + ∞ , applies to semi-inclusivedeep inelastic scattering. In Drell-Yan processes, the Wilson line runs to −∞ and this may change thesign of the distributions, as we will discuss later (Section 3.3.1).It is important to stress in eq. (16) the presence of the transverse links, which survive in thelight-cone gauge A + = 0, enforcing gauge invariance under residual gauge transformations. Thesetransverse links are responsible for the final-state or initial-state interactions that generate some TMDdistributions otherwise forbidden by time-reversal invariance (the so-called T -odd distributions). Innon singular gauges, on the contrary, the gauge potential vanishes at infinity and one is left with thelongitudinal links. It is known that in this case there are light-cone logarithmic divergences arising in thelimit z + → ζ = (2 P · v ) /v ( v is a vector slightly off the light-cone), acting as a rapidity cutoff. The light-conedivergences now appear as large logarithms of ζ , which are resummed by the so-called Collins-Soperequation [89, 92]. A lucid presentation of this subject is contained in Ref. [74]Coming back to the quark correlator, at leading twist Φ( x, k T ) has the following structure [48, 50] Φ( x, k T ) = 12 ( f /n + − f ⊥ T ǫ ijT k T i S T j
M /n + + (cid:18) S L g L + k T · S T M g T (cid:19) γ /n + + h T [ /S T , /n + ] γ (cid:18) S L h ⊥ L + k T · S T M h ⊥ T (cid:19) [ /k T , /n + ] γ M + i h ⊥ [ /k T , /n + ]2 M ) , (17)where ǫ ijT is the two-dimensional antisymmetric Levi-Civita tensor, with ǫ T = 1. By tracing Φ( x, k T ) In the “Amsterdam classification” of TMD distributions [48] which we follow in this review the letters f , g , h refer tounpolarised, longitudinally polarised, and transversely polarised distributions, respectively (as first proposed by Jaffe andJi [22, 23]). The subscript 1 labels the leading twist. Subscripts L and T indicate that the parent hadron is longitudinallyor transversely polarised. A superscript ⊥ signals the presence of k iT factors in the quark correlation function. [Γ] ≡ Tr (ΓΦ), one getsΦ [ γ + ] = f ( x, k T ) − ǫ ijT k T i S T j
M f ⊥ T ( x, k T ) , (18)Φ [ γ + γ ] = S L g L ( x, k T ) + k T · S T M g T ( x, k T ) , (19)Φ [i σ i + γ ] = S iT h ( x, k T ) + S L k iT M h ⊥ L − k iT k jT + k T g ijT M S T j h ⊥ T ( x, k T ) − ǫ ijT k T j
M h ⊥ ( x, k T ) , i = 1 , . (20)where h ≡ h T +( k T / M ) h ⊥ T . The three quantities Φ [ γ + ] , Φ [ γ + γ ] and Φ [i σ i + γ ] represent the probabilitiesof finding an unpolarised, a longitudinally polarised and a transversely polarised quark, respectively,with momentum fraction x and transverse momentum k T . In eqs. (18-20) eight independent TMDdistributions are present: f , f ⊥ T , g , g T , h , h ⊥ L , h ⊥ T , h ⊥ . Upon integration over k T , only three of these, f ( x, k T ) , g ( x, k T ) , h ( x, k T ), survive, yielding the x -dependent leading-twist distributions f ( x ), g ( x ), h ( x ).From eq. (20) one sees that the spin asymmetry of transversely polarised quarks inside a transverselypolarised nucleon is given not only by the unintegrated transversity h ( x, k T ), but also by the TMDdistribution h ⊥ T ( x, k T ), which has been given the name of “pretzelosity”, as it is somehow related tothe non-sphericity of the nucleon shape [93] (for a review of the properties of h ⊥ T , see Ref. [94]). Notethat, due to the intrinsic transverse motion, quarks can also be transversely polarised in a longitudinallypolarised nucleon ( h ⊥ L ), and longitudinally polarised in a transversely polarised nucleon ( g T ). T -odd couple: Sivers and Boer-Mulders distributions From eq. (18) the probability of finding an unpolarised quark with longitudinal momentum fraction x and transverse momentum k T inside a transversely polarised nucleon is f q/N ↑ ( x, k T ) = f ( x, k T ) − ( ˆ P × k T ) · S T M f ⊥ T ( x, k T ) , (21)where ˆ P ≡ P / | P | . Thus the azimuthal asymmetry is f q/N ↑ ( x, k T ) − f q/N ↑ ( x, − k T ) = − P × k T ) · S T M f ⊥ T ( x, k T ) , (22)which is proportional to the so-called Sivers function f ⊥ T [34, 35]. A non vanishing f ⊥ T signals thatunpolarised quarks in a transversely polarised nucleon have a preferential motion direction: in particular, f ⊥ T > z with transverse polarisation in the +ˆ y direction,unpolarised quarks tend to move to the right, i.e. towards − ˆ x .Specularly, the distribution of transversely polarised quarks inside an unpolarised nucleon is [95] f q ↑ /N ( x, k T ) = 12 " f ( x, k T ) − ( ˆ P × k T ) · S qT M h ⊥ ( x, k T ) , (23)and from this we get a spin asymmetry of the form f q ↑ /N ( x, k T ) − f q ↓ /N ( x, k T ) = − ( ˆ P × k T ) · S qT M h ⊥ ( x, k T ) , (24)19hich is proportional to h ⊥ , the Boer–Mulders distribution [50]. Positivity bounds for f ⊥ T and h ⊥ were derived in Ref. [96]. Note that in the literature (see Ref. [63] and bibliography therein) one alsoencounters the notation ∆ N f q/p ↑ ≡ − | k T | M f ⊥ q T , ∆ N f q ↑ /p ≡ − | k T | M h ⊥ q . (25)The Sivers and Boer-Mulders functions are associated with the time-reversal ( T ) odd correlations ( ˆ P × k T ) · S T and ( ˆ P × k T ) · S qT , hence the name of “ T -odd distributions”. To see the implications oftime-reversal invariance one has to recall the operator definition of these distributions which, in thecase of the Sivers function, is: f ⊥ T ( x, k T ) ∼ Z d ξ − Z d ξ T e i xP + ξ − − i k T · ξ T h P, S T | ψ (0) γ + W [0 , ξ ] ψ ( ξ ) | P, S T i| ξ + =0 (26)If the overall Wilson link W is na¨ıvely set to unity, the matrix element in (26) changes sign under timereversal, and the Sivers function must therefore be zero [29]. On the other hand, a direct calculation [36]in a spectator model shows that f ⊥ T is non vanishing: gluon exchange between the struck quark and thetarget remnant generates a non-zero Sivers asymmetry (the presence of a quark transverse momentumsmaller than Q ensures that this asymmetry is proportional to M/k T , rather than to M/Q , and thereforeis a leading-twist observable). The puzzle is solved by carefully considering the Wilson line in eq. (26)[38]. In fact W [0 , ξ ] includes transverse links at infinity that do not reduce to unity in the light-conegauge [91]. Since time reversal changes a future-pointing Wilson line into a past-pointing Wilson line, T -invariance, rather than constraining f ⊥ T to zero, gives a relation between processes that probe Wilsonlines pointing in opposite time directions. In particular, since in SIDIS the Sivers asymmetry arisesfrom the interaction between the spectator and the outgoing quark, whereas in Drell-Yan production itarises from the interaction between the spectator and an incoming quark, one gets f ⊥ T ( x, k T ) SIDIS = − f ⊥ T ( x, k T ) DY . (27)A similar relation holds for the Boer-Mulders function h ⊥ . Eq, (27) is an example of the “time-reversalmodified universality” of distribution functions in SIDIS, DY production and e + e − annihilation studiedin Ref. [97]. The relation (27) is a direct consequence of the gauge structure of parton distributionfunctions, and its experimental check would be extremely important.Gauge link patterns of hadroproduction processes are more complicated and do not result in a simplesign flip of the distributions [98, 99, 100, 101, 102, 103]. For these processes the authors of Refs. [99, 100]suggested that a factorisation scheme should hold with k T distributions containing process-dependentWilson lines. This “generalised TMD factorisation” evidently differs from the standard TMD factori-sation, wherein the k T distributions are fully universal quantities. A recent study [104], however, hasshown that even the generalised factorisation scheme is violated in hadroproduction of nearly back-to-back jets or hadrons, a process investigated experimentally by the STAR collaboration at RHIC[105].The quark Sivers function has an exact gluonic counterpart, f ⊥ g T , which represents the distributionof unpolarised gluons in a transversely polarised hadron. This function is called G T in Ref. [65], wherea complete classification of leading-twist k T -dependent gluon distributions is presented. There is nogluonic equivalent of the Boer-Mulders function, but a somehow similar quantity is the distribution oflinearly polarised gluons in an unpolarised hadron.A sum rule for the Sivers function was derived in QCD by Burkardt [106, 107], who showed that thesum of all contributions to the average transverse momentum of unpolarised partons in a transverselypolarised target (that is, the average transverse momentum induced by the Sivers effect), must vanish: X a = q, ¯ q,g h k aT i| Sivers = 0 . (28)20 g = x ′ − xP Px x ′ Figure 7: The quark–gluon correlation matrix.In terms of the Sivers function, the condition (28) becomes [108] X a = q, ¯ q,g Z d x f ⊥ (1) a T ( x ) = 0 , (29)where the first k T -moment of f ⊥ T is given by f ⊥ (1)1 T ( x ) ≡ Z d k T k T M f ⊥ T ( x, k T ) . (30)Although some QCD aspects, such as ultraviolet divergences and light-cone singularities, were notconsidered in the original derivation, eq. (30) is likely to be valid in general. In Ref. [109] it has beenshown that the Burkardt sum rule is fulfilled for a quark target in perturbative QCD at one-loop order.From a phenomenological point of view, the importance of eq. (29) is that one can infer the sizeof the gluon Sivers function from fits to SIDIS observables involving the quark and antiquark Siversfunctions [110, 111]. At twist three, suppressed by 1 /Q , i.e., by 1 /P + in the infinite momentum frame, with respect toleading twist, the quark correlator Φ( x ) admits the general decomposition [22, 23]Φ( x ) | twist 3 = M P + (cid:26) e ( x ) + g T ( x ) γ /S T + S L h L ( x ) [ /n + , /n − ] γ (cid:27) , (31)displaying the three distribution functions e ( x ) , g T ( x ) , h L ( x ). In particular, g T ( x ) contributes to thepolarised DIS structure function g ( x, Q ) (see, e.g., Ref. [112]). Higher-twist distributions do not havea probabilistic interpretation. They involve in fact both good and bad components of the quark fields,so the procedure leading to expressions such as eqs. (2-4) cannot be applied.Higher-twist effects are a manifestation of quark-gluon correlations inside hadrons [113, 114]. Attwist three there are four quark-gluon correlators, which depend on two momentum fractions, x and x ′ (see Fig. 7): G F ( x, x ′ ), ˜ G F ( x, x ′ ), H F ( x, x ′ ), E F ( x, x ′ ). In the literature [53, 55, 56, 60, 115], thesecorrelators are also called T F , ˜ T F , ˜ T ( σ ) F , T ( σ ) F , respectively, with a normalisation varying from one paperto another. In QCD the quark-gluon correlation functions acquire a dependence on a scale µ . Theequations governing the evolution in µ have been recently written down and solved [59, 60].Using the QCD equations of motion and integrating over x ′ one can show that [48, 116] g T ( x ) = g (1)1 T ( x ) x + ˜ g T ( x ) , h L ( x ) = − h ⊥ (1)1 L ( x ) x + ˜ h L ( x ) , e ( x ) = ˜ e ( x ) , (32)21here g (1)1 T , h ⊥ (1)1 L are the first transverse moments of g T and h ⊥ L , defined as in eq. (30), and the tildefunctions are genuinely twist-three distributions related to the quark-gluon correlators.Ignoring the contributions of tilde functions and of quark mass terms one gets the generalisedWandzura-Wilczek (WW) approximation, so called in analogy with the original WW relation [117]between the polarised DIS structure functions g ( x, Q ) and g ( x, Q ) [118, 112, 119]. The generalisedWW approximation relates twist-three distributions to twist-two distributions. It has been investigatedby various authors [120, 48, 121, 122] and also applied in phenomenological analyses [123].Coming to the transverse motion of quarks, the structure of the k T -dependent quark correlatorΦ( x, k T ) at twist three has also been studied by various authors [48, 50, 124, 125]. It is now knownthat there are 16 twist-three TMD distributions: e , e ⊥ T , e L , e T , f T , f ⊥ L , f ⊥ T , f ⊥ , g T , g ⊥ L , g ⊥ T , g ⊥ , h L , h T , h , h ⊥ T in the classification of Ref. [126]). Some of these functions, namely g ⊥ , e ⊥ T , f T , f ⊥ T , not identifiedin earlier studies, exist because the Wilson line in the quark correlator provides an extra independentvector ( n − ) for the Lorentz decomposition of Φ( x, k T ). If we integrate Φ( x, k T ) over k T , the only nonvanishing distributions are the three T -even functions in eq. (31).Concerning twist four, the integrated parton distributions were first identified in Refs. [22, 23, 127].More recently, the complete expression of the k T -dependent correlator Φ( x, k T ) has been given in Ref.[125], where it is shown that up to twist four there are in total 32 TMD distributions. The unintegratedcorrelation matrix Φ is also composed of 32 Lorentz-scalar structures: 12 amplitudes associated to thefour-vectors k, P, S and 20 amplitudes associated to n − . The number of distribution functions beingequal to the number of amplitudes of Φ, all the TMD distributions are independent and there are nogeneral relations among them. In earlier studies [120, 48], some “Lorentz-invariance relations” (LIR’s)were derived from an expansion of Φ that did not take into account the amplitudes associated to thegauge link vector n − . Two of these relations are: g T ( x ) = g ( x ) + dd x g ⊥ (1)1 T ( x ) , h L ( x ) = h ( x ) − dd x h ⊥ (1)1 L ( x ) . (33)The presence of the n − -dependent amplitudes invalidate the LIR’s, which are not valid in QCD [128,129]. However, they approximately hold in the generalised WW approximation [122].A general remark about higher-twist distributions is in order. While the distributions e ( x ), g T ( x ), h L ( x ) – or, to be precise, the corresponding quark-gluon correlators – enter into the collinear twist-three factorisation theorem of QCD [53, 54], the k T -dependent higher-twist distributions are employedin factorisation formulae that lack a solid QCD foundation. Thus, they should rather be interpreted asa way to model subleading effects. The generalised parton distributions (GPD’s), which are related to non-forward quark-quark (or gluon-gluon) correlators, emerge in the description of hard exclusive processes, such as deeply-virtual Comptonscattering and exclusive meson production, characterised by a non-zero momentum transfer to the targetnucleon [130, 131, 132, 66, 133, 134]. Here we will be mostly concerned with the relations existingbetween the GPD’s and the transverse spin distributions (for more details, see Ref. [135]).The kinematics of GPD’s is represented in fig. 8 (we follow the conventions of [66]). The momenta ofthe incoming and the outgoing nucleon are p = P − ∆ and p ′ = P + ∆, respectively. The momentumtransfer squared is t = ∆ . The GPD’s depend on t and on two light-cone momentum ratios: x = k + /P + and ξ = − ∆ + / P + . The variable ξ is sometimes called “skewness”, and the GPD’s are also known as“skewed parton distributions”. At leading twist, there are 8 GPD’s [136]: H ( x, ξ, t ) , E ( x, ξ, t ) , ˜ H ( x, ξ, t ) , ˜ E ( x, ξ, t ) , H T ( x, ξ, t ) , E T ( x, ξ, t ) , ˜ H ( x, ξ, t ) , ˜ E ( x, ξ, t ) . (34)22 − ∆ k + ∆ P − ∆ P + ∆Figure 8: Kinematics of GPD’sThe first four are chirally even and are related to the familiar form factors. Integrating H , E , ˜ H , ˜ E over x , one gets in fact the Dirac, Pauli, axial and pseudoscalar form factors, respectively. The quantity Z d x E q ( x, ,
0) = κ q , (35)is the contribution of the flavour q to the anomalous magnetic moment of the nucleon, that is, to thePauli form factor F at t = 0. The GPD’s H, ˜ H, H T , taken at ξ = t = 0, coincide with the integratedquark distributions f , g , h : H ( x, ,
0) = f ( x ) , ˜ H ( x, ,
0) = g ( x ) , H T ( x, ,
0) = h ( x ) . (36)The original interest in GPD’s was prompted by Ji’s sum rule relating the total angular momentum ofquarks (in a nucleon with polarisation vector S ) to the second moment of H and E [131]: h J iq i = S i Z d x x [ H ( x, ,
0) + E ( x, , . (37)A similar decomposition for the angular momentum of quarks with transverse polarisation vector S q inan unpolarised nucleon has been derived in Ref. [87, 88] and is given by: h J iq ( S q ) i = S iq Z d x x [ H T ( x, ,
0) + 2 ˜ H T ( x, ,
0) + E T ( x, , . (38)Here H T ( x, ,
0) coincides with transversity, whereas the combination 2 ˜ H T + E T appears in the impact-parameter description of the Boer-Mulders effect (see Section 3.6).Note in conclusion that there are no direct and model-independent connections between the GPD’sand the TMD distributions, as stressed in Refs. [135, 137]. GPD’s are instead directly related to thedistribution functions in the impact-parameter space. In the impact-parameter space one can get a more intuitive picture of some transverse spin and trans-verse momentum effects. To define the impact-parameter distributions (IPD’s), one first introducesnucleon states localised at a transverse position R T , by means of an inverse Peierls-Yoccoz projection: | P + , R T ; S i = N Z d P T (2 π ) e − i P T · R T | P, S i , (39)where N is a normalisation factor. The IPD’s are light-cone correlations in these transverse-positionnucleon eigenstates. For instance, the unpolarised IPD is given by q ( x, b T ) = Z d z − π e i xP + z − h P + , T ; S | ¯ ψ ( z ) W − [ z , z ] γ + ψ ( z ) | P + , T ; S i , (40)23ith z , = (0 + , ∓ z − , b T ). This is the number density of quarks with momentum fraction x andtransverse position b T inside an unpolarised hadron. The polarised IPD’s are obtained by inserting inthe matrix element of eq. (40), instead of γ + , the matrices γ + γ and i σ i + γ .IPD’s are Fourier transforms not of the TMD distributions, but of the GPD’s. The impact-parametertransform of a generic GPD X for ξ = 0 (which implies ∆ = − ∆ T ) is defined as X ( x, b T ) = Z d ∆ T (2 π ) e − i ∆ T · b T X ( x, , − ∆ T ) . (41)It is straightforward to show [138] that the unpolarised IPD q ( x, b T ) coincides with the Fourier transformof H ( x, , − ∆ T ), that is q ( x, b T ) = H q ( x, b T ) . (42)The impact-parameter density of unpolarised quarks in a transversely polarised nucleon ( N ↑ ) is [138,139, 135] q N ↑ ( x, b T ) = H q ( x, b T ) + ( ˆ P × b T ) · S T M E ′ q ( x, b T ) , with E ′ q ( x, b T ) ≡ ∂∂b T E q ( x, b T ) , (43)where E ( x, b T ) is the Fourier transform of E ( x, , − ∆ T ). Notice the formal similarity with eq. (21) andthe correspondence f ⊥ T ↔ −E ′ . . Due to the E ′ q term, which can be regarded as the b T -space analogueof the Sivers distribution, q N ↑ ( x, b T ) is not axially symmetric and describes a spatial distortion of thequark distribution in the transverse plane. Final-state interactions can translate this position-spaceasymmetry into a momentum-space asymmetry. For instance, if the nucleon moves in the +ˆ z directionsand is polarised in the +ˆ x direction, a positive E ′ q implies that quarks tend to be displaced in the − ˆ y direction, and final-state interactions, which is expected to be attractive on average, convert thistransverse distortion into a momentum asymmetry in the +ˆ y direction. This is the intuitive explanationof the Sivers effect in the impact-parameter picture [106, 140]. A measure of the space distortion isgiven by the flavour dipole moment d iq = Z d x Z d b T b iT q N ↑ ( x, b T ) = − ǫ ijT S jT M Z d x E q ( x, ,
0) = − ǫ ijT S jT M κ q , (44)where κ q is the contribution of the quark flavour q to the anomalous magnetic moment of the nucleon,see eq. (35). The argument developed so far is summarised by the following qualitative relation betweenthe Sivers function f ⊥ T and κ q [141, 142, 140] (any quantitative relation between these two quantitiesis necessarily model-dependent [135, 137]) f ⊥ q T ∼ − κ q , (45)where the minus sign is a consequence of attractive final-state interactions that transform a preferentialdirection in the b T -space into the opposite direction in k T . Eq. (45) leads to an immediate prediction:since the quark contributions to the anomalous magnetic moment of the proton κ p , extracted from theexperimental value of κ p using SU(2) flavour symmetry, are κ u ≃ . , κ d = − .
0, one expects f ⊥ u T < f ⊥ d T >
0. This prediction has been corroborated by the SIDIS experiments (see Section 5.3.1).Consider now the case of transversely polarised quarks inside an unpolarised nucleon. Their impact-parameter distribution is [139] q ↑ ( x, b T ) = 12 ( H q ( x, b T ) + ( ˆ P × b T ) · S qT M [ E ′ T q ( x, b T ) + 2 ˜ H ′ T q ( x, b T )] ) , (46)24 (cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:8)(cid:9)(cid:10) (cid:0)(cid:11)(cid:12)(cid:13)(cid:14)(cid:15)(cid:16)(cid:17)(cid:18)(cid:19)(cid:20)(cid:21)(cid:22)(cid:23)(cid:24)(cid:25)(cid:26)(cid:27)(cid:28)(cid:29)(cid:30)(cid:31) !" Figure 9: First x -moments of the densities of unpolarised quarks in a transversely polarised nucleon(left) and transversely polarised quarks in an unpolarised nucleon (right) for u (upper plots) and d (lower plots) quarks. Quark spins (inner arrows) and nucleon spins (outer arrows) are oriented in thetransverse plane as indicated. From Ref. [143].The term E ′ + 2 ˜ H ′ T is the analogue of the Boer-Mulders function in the b T -space – see eq. (23). Again,we see that transverse spin (of quarks, in this case) causes a spatial distortion of the distribution, whichis at the origin of the Boer-Mulders effect. One can repeat the same reasoning developed for the Siverseffect and introduce a transverse anomalous moment κ qT , defined by κ qT ≡ Z d x [ E qT ( x, ,
0) + 2 ˜ H qT ( x, , . (47)The Boer-Mulders function is expected to scale with this quantity, h ⊥ q ∼ − κ qT . (48)where the minus sign has the same meaning as before. Unfortunately, no data exist for κ qT . This quantity,however, and the impact-parameter distributions have been calculated in lattice QCD [143, 144] andare shown in Fig. 9). The result for κ T is: κ uT = 3 . κ dT = 1 .
9. Thus, at variance with f ⊥ T , we expectthe u and d components of h ⊥ to have the same sign, and in particular to be both negative. Moreover,assuming simple proportionality between the ratio h ⊥ /f ⊥ T and κ/κ T , the u component of h ⊥ should beapproximately twice as large as the corresponding component of f ⊥ T , while h ⊥ d and f ⊥ d T should have acomparable magnitude and opposite sign. These predictions are well supported by a phenomenologicalanalysis of SIDIS data [145] as will be shown in Section 5.3.1. Models and other non-perturbative approaches like lattice calculations play a very important rˆole whenthe experimental information about distribution functions is scarce or lacking at all. So, it is notsurprising that a considerable effort has been made to compute the TMD distributions in variousmodels of the nucleon and by lattice QCD. Here we will be not be able to give an exhaustive accountof all this work still largely in progress and we will limit ourselves to a general discussion. For a recentreview of model results see Ref. [146]. 25he first calculation of TMD distributions was performed in a quark-diquark spectator model [147].This class of models, with various quark-diquark vertex functions, has been subsequently used bymany authors. In particular, using a simple scalar spectator model with gluon exchange is was shownexplicitly [36] that the Sivers function is non vanishing. Since Wilson links, representing gluon insertions,are crucial in order to guarantee the existence of the T -odd distribution functions, these can only becomputed in models containing gluonic degrees of freedom. Following Ref. [36], more refined calculationsof the Sivers and Boer-Mulders functions were performed in spectator models with both scalar and axial-vector diquarks and various quark-diquark vertices [148, 149, 150, 109, 151, 152, 153]. Other modelsused to evaluate the T -odd functions include the MIT bag model [154, 155, 156, 157], the constituentquark model [158, 157] and a light-cone model [159]. In Ref. [155] final state interactions were assumedto be induced by instanton effects.What emerges from models is that the Sivers function, although quite variable in magnitude, isnegative for u quarks and positive for d quark. A different sign of f ⊥ d T is however found in the model ofRef. [155]. As for the Boer-Mulders function, the general prediction (with the exception of Ref. [150])is that both the u and the d distributions are negative. These signs for f ⊥ T and h ⊥ are also expected inthe impact-parameter picture [138, 141, 140, 142, 87], in the large- N c approach [160], which predicts theisoscalar component of f ⊥ T and the isovector component of h ⊥ to be suppressed, and in chiral models[161].Spectator models have been also used [162, 163] to calculate T -odd twist-3 distributions, in particular g ⊥ , which contributes to the longitudinal beam spin asymmetry in SIDIS.Models without gluonic degrees of freedom can be used to compute T -even TMD distributionsonly. These distributions have been calculated in a spectator model [152], in light-cone quark models[164, 165, 166], in a covariant parton model with orbital motion [167] and in the bag model [94, 168].In particular, Ref. [168] presents a systematic study of leading and subleading twist TMD distributionsand of the relations among them.In any quark model without gluons, the Lorentz–invariance relations, obtained by neglecting theamplitudes of the quark-quark correlator related to the gauge link (Section 3.4), must obviously bevalid. There are also a number of other relations that hold in some specific models like [94] g ( x, k T ) − h ( x, k T ) = k T M h ⊥ T ( x, k T ) . (49)According to this relation, h ⊥ T can be interpreted as a measure of the relativistic effects in the nu-cleon, which are known to be responsible for the difference between the helicity and the transversitydistributions [169]. Other model-dependent relations involving the TMD distributions are listed inRefs. [146, 168].Finally, one should keep in mind that models provide a dynamical picture of the nucleon at somefixed, very low, scale µ < [170, 171, 172, 173]. The quark distributions that one gets aretherefore valid at this unrealistic scale and must be evolved to the experimental scales. The evolutionof the TMD distributions has been unknown until very recently and is therefore usually neglected orapproximated in current phenomenological analyses. In partially inclusive processes a parton hadronises into a particle h carrying away a fraction of theparton’s momentum. In the following it is supposed that the fragmentation process is initiated by aquark, as is the case in SIDIS and e + e − annihilation at leading order. The momentum of the fragmentingquark is indicated as κ µ and z is the fraction of its longitudinal component carried by the final hadron, z = P − h /κ − . Since the hadron moves in the opposite direction with respect to the target nucleon,the minus components of momenta are the dominant ones. The fragmenting quark has a transverse26omentum κ T with respect to the final hadron. Conversely, the hadron has a transverse momentum p T = − z κ T with respect to the quark.The fragmentation analogue of Φ ij ( x, k T ) is:Ξ ij ( z, z κ T ) = 12 z X X Z d ξ + π Z d ξ T (2 π ) e i P − h ξ + /z e − i κ T · ξ T ×h |W [+ ∞ , ξ ] ψ i ( ξ ) | P h , S h ; X ih P h , S h ; X | ψ j (0) W [0 , + ∞ ] | i| ξ − =0 , (50)where each Wilson line includes a longitudinal link along n + and a transverse link at infinity [126]. Inthe case of fragmentation one has the same gauge structure in SIDIS and in e + e − annihilation, whichmeans that there is no difference between the fragmentation functions of these processes, i.e. they areuniversal quantities in a full sense [97]. The integrated fragmentation correlator is given byΞ( z ) = Z d p T Ξ( z, p T ) = 12 ( D ( z ) /n − + S L G ( z ) γ /n − + H ( z ) [ /S T , /n − ] γ ) + h . t . , (51)where “h.t.” denotes higher-twist terms. D , G and H are the integrated leading-twist fragmentationfunctions (FF’s): D is the ordinary unpolarised fragmentation function, G is the analogue of thehelicity distribution, H is the analogue of the transversity distribution (it describes the fragmentationof a transversely polarised quark into a transversely polarised hadron).To compute azimuthal asymmetries the transverse-momentum dependent FF’s are needed. Forsimplicity, we limit ourselves to listing the FF’s of main phenomenological interest. The traces of thefragmentation matrix corresponding to unpolarised and transversely polarised quarks are [48]Ξ [ γ − ] ( z, p T ) = D ( z, p T ) + ǫ T ij p iT S jhT zM h D ⊥ T ( z, p T ) , (52)Ξ [i σ i − γ ] ( z, p T ) = S ihT H ( z, p T ) + ǫ ijT p T j zM h H ⊥ ( z, p T ) + . . . . (53) D ⊥ T is analogous to the Sivers distribution function and describes the production of transversely po-larised hadrons from unpolarised quarks. For this reason it is called “polarising fragmentation function”[174]. The most noteworthy FF appearing in (53) is H ⊥ ( z, k T ), the so-called Collins function, describingthe fragmentation of a transversely polarised quark into an unpolarised hadron [29]. The resultingtransverse-momentum asymmetry of hadrons is D h/q ↑ ( z, p T ) − D h/q ↑ ( z, − p T ) = 2 ( ˆ κ × p T ) · S ′ qT zM h H ⊥ ( z, p T ) , (54)where S ′ q is the spin vector of the fragmenting quark. From the structure of the correlation ( ˆ κ × p T ) · S ′ qT one sees that a positive H ⊥ corresponds to a preference of the hadron to be emitted on the left sideof the jet if the quark spin points upwards. Through this mechanism the transverse momentum of theproduced hadron with respect to the jet direction acts as a quark polarimeter.The Collins function satisfies a sum rule arising from the conservation of the intrinsic transversemomentum during quark fragmentation. This sum rule, discovered by Sch¨afer and Teryaev [175], reads X h Z d z z H ⊥ (1) q ( z ) = 0 , with H ⊥ (1)1 ( z ) ≡ z Z d κ T κ T M h H ⊥ ( z, z κ T ) . (55)27igure 10: The fragmentation process in the recursive string model [177]. The ˆ z direction is along thestring and the fragmenting quark q is supposed to be polarised in the +ˆ y direction, out of the page.A simple qualitative explanation of the Collins effect is provided by the “recursive string model”of Ref. [28, 176, 177], which is illustrated in Fig. 10. Suppose that a quark q , polarised in the +ˆ y direction, i.e. out of the page in the figure, fragments into a pion with an antiquark ¯ q created by stringbreaking. If the q ¯ q pair is in a P state, the orbital angular momentum of the pair is L = 1, andthe pion, inheriting the transverse momentum of ¯ q , moves in the +ˆ x direction. The quark q , with thesubleading pion that contains it, moves in the opposite direction. This model predicts opposite Collinsasymmetries for π + and π − assuming u dominance, and a positive (negative) sign for the favoured(unfavoured) Collins function. “Favoured” refers to the fragmentation of a quark or an antiquarkbelonging to the valence component of the final hadron, e.g. u → π + , d → π − , ¯ d → π + , etc..The Collins function for pions has been computed in various fragmentation models [178, 179, 180,181, 182]. What is common to these approaches is that H ⊥ arises from the interference between a treelevel amplitude and loop corrections that provide the necessary imaginary parts. The differences residein the pion-quark couplings and in the nature of the virtual particles in the loops (pions or gluons). Anassessment of model calculations of the Collins function is contained in Ref. [183]. In this section we will present a general description of the processes probing the transverse-spin andtransverse-momentum structure of hadrons, and of the relevant observables: single-spin asymmetries,double-spin asymmetries and unpolarised azimuthal asymmetries. The focus will be on two classes ofreactions that have clear and well-established theoretical descriptions, namely SIDIS with the relatedprocess e + e − annihilation into hadron pairs, and DY production. Hadroproduction will also be de-scribed, but in lesser detail. The last subsection contains a sketchy presentation of other processessomewhat related to transverse spin. SIDIS is the process ℓ ( l ) + N ( P ) → ℓ ′ ( l ′ ) + h ( P h ) + X ( P X ), where ℓ ( ℓ ′ ) is the incoming (outgoing)lepton, N the nucleon target, h the detected hadron. The corresponding four-momenta are given inparentheses. In the following we will denote by S k and S ⊥ the longitudinal and transverse componentof the target spin vector, respectively, and by λ ℓ the longitudinal polarisation of the incident lepton.SIDIS is usually described in terms of the invariant variables x B = Q / P · q , y = P · q/P · l , z h = P · P h /P · q , with q = l − l ′ and Q ≡ − q . In the deep inelastic limit, Q is much larger than the28 zx had r on p l an e l e p t on p l an e l l S ? P h P h ? f h f S Figure 11: Lepton and hadron planes in semi-inclusive deep inelastic scattering. The reference frame isdefined according to the convention of Ref. [184].mass M of the nucleon and the mass M h of the final hadron. Hereafter mass corrections are neglectedunless otherwise stated.To parametrise the cross section in terms of structure functions the so-called “ γ ∗ N collinear frame”[184] is usually adopted. As shown in Fig. 11 in this reference frame the virtual photon and the targetnucleon are collinear and directed along the z axis, with the photon moving in the positive z direction,and the final hadron has a transverse momentum P h ⊥ . All azimuthal angles are referred to the leptonscattering plane: φ h is the azimuthal angle of the hadron h , φ S is the azimuthal angle of the nucleonspin vector S ⊥ . The phase space of the process contains another angle, ψ , which is the azimuthal angleof the outgoing lepton around the beam axis with respect to an arbitrary fixed direction, which is chosento be given by the target spin. Up to corrections of order M /Q one has d ψ ≃ d φ S [185].In this Section we consider the case of a spinless or unpolarised detected hadron, while leptopro-duction of transversely polarised hadrons will be treated in Section 4.1.4. The SIDIS differential crosssection in the six variables x B , y, z h , φ S , P h ⊥ ≡ | P h ⊥ | and φ h , is given byd σ d x B d y d z h d φ S d φ h d P h ⊥ = α Q yz h L µν W µν , (56)where L µν is the usual DIS leptonic tensor and W µν is the hadronic tensor W µν = 1(2 π ) X X R d P X (2 π ) E X (2 π ) δ ( P + q − P X − P h ) ×h P, S | J µ (0) | X ; P h , S h ih X ; P h , S h | J ν (0) | P, S i . (57)Neglecting for simplicity the M x /Q corrections the complete SIDIS cross section can be parametrised29n terms of 18 structure functions as follows [185, 126];d σ d x B d y d z h d φ h d P h ⊥ d φ S = α x B yQ (cid:26) (1 − y + 12 y ) F UU,T + (1 − y ) F UU,L +(2 − y ) p − y cos φ h F cos φ h UU + (1 − y ) cos 2 φ h F cos 2 φ h UU + λ ℓ y p − y sin φ h F sin φ h LU + S k h (2 − y ) p − y sin φ h F sin φ h UL + (1 − y ) sin 2 φ h F sin 2 φ h UL i + S k λ ℓ (cid:20) y (1 − y ) F LL + y p − y cos φ h F cos φ h LL (cid:21) + S ⊥ (cid:20) sin( φ h − φ S ) (cid:18) (1 − y + 12 y ) F sin( φ h − φ S ) UT,T + (1 − y ) F sin( φ h − φ S ) UT,L (cid:19) +(1 − y ) sin( φ h + φ S ) F sin( φ h + φ S ) UT + (1 − y ) sin(3 φ h − φ s ) F sin(3 φ h − φ S ) UT + (2 − y ) p − y sin φ S F sin φ S UT + (2 − y ) p − y sin(2 φ h − φ S ) F sin(2 φ h − φ S ) UT i + S ⊥ λ ℓ (cid:20) y (1 − y ) cos( φ h − φ S ) F cos( φ h − φ S ) LT + y p − y cos φ S F cos φ S LT + y p − y cos(2 φ h − φ S ) F cos(2 φ h − φ S ) LT io . (58)The structure functions F depend on x B , y, z h and P h ⊥ . Their first and second subscript denote thepolarisation of the beam and of the target, respectively ( U = unpolarised, L = longitudinally polarised, T = transversely polarised), whereas the third subscript refer to the polarisation of the virtual photon.If we integrate (58) over P h ⊥ , only 5 structure functions survive: F UU,T , F
UU,L , F LL , F cos φ S LT , and F sin φ S UT . The first two, upon a further integration in z and a sum over all outgoing hadrons, yieldthe unpolarised DIS structure functions F T ( x B , Q ) = 2 xF ( x B , Q ) and F L ( x B , Q ) = F ( x B , Q ) − x B F ( x B , Q ). The second two lead to combinations of the structure functions g ( x B , Q ) and g ( x B , Q )of longitudinally polarised DIS. The fifth one vanishes. The fact that X h Z d z h z h Z d P h ⊥ F sin φ S UT = 0 (59)is a consequence of time-reversal invariance [185] and is another way to express the Christ-Lee theorem[26], according to which there cannot be transverse spin asymmetries in inclusive DIS. On the contrary,in SIDIS no first principle forbids the existence of transverse spin asymmetries.In the literature, the spin and azimuthal asymmetries of SIDIS are defined in two different ways:- as the structure function ratios: A w ( φ h ,φ S ) XY ( x, z, P h ⊥ ) ≡ F w ( φ h ,φ S ) XY F UU , (60)where X and Y label the polarisation of the beam and of the target respectively, w ( φ h , φ S ) is atrigonometric function of its arguments, and F UU ≡ (1 − y + y ) F UU,T + (1 − y ) F UU,L ;- as the azimuthal moments of cross sections [184]: A w ( φ h ,φ S ) ( x B , y, z h , P h ⊥ ) ≡ h w ( φ h , φ S ) i ≡ R d φ h R d φ S w ( φ h , φ S ) d σ ( φ h , φ S ) R d φ h R d φ S d σ ( φ h , φ S ) , (61)where d σ is a shorthand notation for the fully differential cross section.Notice that the two definitions of asymmetries differ for y -dependent factors.30 Pq qP h P h k kκ κ ΞΦ Figure 12: Diagram contributing to semi-inclusive DIS in the parton model.
In the parton model the virtual photon strikes a quark (or an antiquark), which successively fragmentsinto a hadron h . The process is represented by the diagram in Fig. 12. We will take transverse momentaof quarks into account and refer to this description as the “extended parton model”.For the partonic description of SIDIS we work in a reference frame where the momenta of the targetnucleon and of the outgoing hadron are collinear and define the longitudinal direction. In this “ hN collinear frame”, one has P µ = P + n µ + and P µh = P − n µ − , whereas the virtual photon momentum acquiresa transverse component q T . The incoming quark momentum is k µ = xP µ + k µT , with x = k + /P + ;the fragmenting quark momentum is κ µ = P µh /z + κ µT , with z = P − h /κ − . Notice that the “transverse”quantities in the hN frame (labelled by the subscript T ) differ from the “perpendicular” quantities inthe γ ∗ N frame (labelled by the subscript ⊥ ) by terms suppressed at least as 1 /Q . In particular, q T isrelated to P h ⊥ by q T = − P h ⊥ /z h , up to 1 /Q corrections.The hadronic tensor corresponding to Fig. 12 is given by W µν = X a e a Z d k Z d κ δ ( k + q − κ ) Tr [Φ a ( k ) γ µ Ξ a ( κ ) γ ν ] , = 2 z h X a e a Z d k T Z d κ T δ ( k T + q T − κ T ) Tr [Φ a ( x B , k T ) γ µ Ξ a ( z h , κ T ) γ ν ] , (62)where the second expression has been obtained by working out the momentum-conservation delta func-tion and neglecting 1 /Q terms. In this case one has z = z h and x = x B .Inserting the expressions of Φ and Ξ into (62) and contracting W µν with L µν leads to the SIDISstructure functions. With the following notation for the transverse momenta convolutions C [ wf D ] = X a e a x Z d k T Z d κ T δ ( k T − κ T − P h ⊥ /z ) w ( k T , κ T ) f a ( x, k T ) D a ( z, κ T ) , (63)31he non vanishing structure functions at leading twist are [48, 50, 126] F UU,T = C [ f D ] , (64) F cos 2 φ h UU = C " −
2( ˆ h · k T )( ˆ h · κ T ) − k T · κ T M M h h ⊥ H ⊥ , (65) F sin 2 φ h UL = C " −
2( ˆ h · k T )( ˆ h · κ T ) − k T · κ T M M h h ⊥ L H ⊥ , (66) F LL = C [ g L D ] (67) F sin( φ h − φ S ) UT,T = C " − ˆ h · k T M f ⊥ T D , (68) F sin( φ h + φ S ) UT = C " − ˆ h · κ T M h h H ⊥ , (69) F sin(3 φ h − φ S ) UT = C "
2( ˆ h · κ T )( k T · κ T ) + k T ( ˆ h · κ T ) −
4( ˆ h · k T ) ( ˆ h · κ T )2 M M h h ⊥ T H ⊥ , (70) F cos( φ h − φ S ) LT = C " ˆ h · k T M g T D . (71)where ˆ h ≡ P h ⊥ / | P h ⊥ | . The structure function F UU,T gives the dominant contribution to the unpolarisedcross section integrated over φ h .The Collins term F sin( φ h + φ S ) UT couples the Collins function H ⊥ to the transversity distribution h ,thus representing one of the privileged ways to access this quantity. Note that in the original paper[29] the Collins angle Φ C was defined as the angle between the momentum of the produced hadron andthe spin of the fragmenting quark, i.e. Φ C ≡ φ h − φ S ′ q . In terms of the azimuthal angle of the targetspin φ S , one has Φ C = φ h + φ S − π . On the other hand, according to the conventions of Ref. [184],the Collins angle is defined as Φ ′ C ≡ φ h + φ S . Thus, one gets different signs for the Collins asymmetry,depending on which definition of the Collins angle, either Φ C or Φ ′ C , is adopted.Another leading-twist asymmetry source is the Sivers term F sin( φ h − φ S ) UT,T , which couples the Siversfunction f ⊥ T to the unintegrated unpolarised fragmentation function D . In the transversely polarisedcase, a further angular modulation, of the type sin(3 φ h − φ S ), involves the distribution function h ⊥ T .In unpolarised SIDIS, a leading-twist azimuthal asymmetry is generated by the structure function F cos 2 φ h UU , which couples the Boer-Mulders distribution h ⊥ to the Collins fragmentation function H ⊥ .Going to twist three, i.e. to order 1 /Q , it turns out that the leading-twist structure functions (64-70)do not acquire any extra contribution, but there appear other non vanishing structure functions [126].Among them, of particular phenomenological importance are those related to the cos φ h and sin φ h modulations. Ignoring, in the spirit of the parton model, interaction-dependent terms, i.e. quark-gluoncorrelations, and quark mass contributions (the generalised Wandzura–Wilczek approximation) onefinds [126] F cos φ h UU = 2 MQ C " − ( ˆ h · κ T ) k T M h M h ⊥ H ⊥ − ˆ h · k T M f D , (72) F sin φ h UL = 2 MQ C " − ( ˆ h · κ T ) k T M h M h ⊥ L H ⊥ . (73)In the same approximation one gets F sin φ h LU = 0. Thus a deviation of the beam-spin sin φ h asymmetryfrom zero might signal the relevance of interaction effects in the nucleon. One should recall however32hat at high transverse momenta F sin φ h LU is non zero in next-to-leading order QCD. The term in F cos φ h UU containing the product of the unpolarised functions f D is a purely kinematical contribution arisingfrom the intrinsic transverse motion of quarks, with no relation to spin. This contribution was discoveredlongtime ago by Cahn [186, 187], and the corresponding azimuthal asymmetry is referred to as the “Cahneffect”. A similar contribution emerges at twist four, that is at order 1 /Q , in the cos 2 φ h term: F cos 2 φ h UU,
Cahn = M Q C " (2( ˆ h · k T ) − k T ) M f D . (74)All the above parton-model results have been obtained using the most general decompositions ofthe correlation matrices Φ and Ξ, and inserting them into the SIDIS hadronic tensor W µν . There isan alternative approach, which relies on the helicity formalism and expresses the cross section as aconvolution of helicity amplitudes of elementary subprocesses with partonic distribution and fragmen-tation functions, taking fully into account non collinear kinematics [188]. Considering for simplicity anunpolarised lepton beam and a spinless or unpolarised final hadron, and adopting the γ ∗ N collinearframe, the basic factorisation formula for the SIDIS cross section in this picture is, up to order 1 /Q ,d σ = X q i X λ qi λ ′ qi Z d k ⊥ Z d p ⊥ ρ q i λ qi λ ′ qi f q i ( x, k ⊥ )dˆ σ λ qi λ ′ qi D h/q f ( z, p ⊥ ) δ ( z k ⊥ − p ⊥ − P h ⊥ ) , (75)where the λ ’s are helicity indexes, f q i ( x, k ⊥ ) is the probability of finding a quark q i with momentumfraction x and transverse momentum k ⊥ inside the target nucleon, ρ qλ qi λ ′ qi is the helicity density matrixof the quark q i , D h/q f ( z, p ⊥ ) is the fragmentation function of the struck quark q f into the hadron h withtransverse momentum p ⊥ with respect to the fragmenting quark, and dˆ σ λ qi λ ′ qi ∼ ˆ M λ ℓ ′ λ qf ; λ ℓ λ qi ˆ M ∗ λ ℓ ′ λ qf ; λ ℓ λ ′ qi is the cross section of lepton-quark scattering ℓq i → ℓ ′ q f at tree level. Note that, whereas in the collinearcase the produced hadron is constrained to have P h ⊥ = 0 and the entire process takes place in thescattering plane, the intrinsic transverse momentum of quarks introduces a non planar geometry. Theelementary scattering amplitudes ˆ M ’s take into account this non collinear and out-of-plane kinematics.Neglecting O ( k ⊥ /Q ) contributions, no Jacobian factors appear in eq. (75) and one has x = x B , z = z h .Despite their apparent dissimilarity, the two parton model approaches described so far, namely theapproach based on quark correlation matrices and eq. (62) and the approach based on the helicityformalism and eq. (75), are equivalent as far as parton interactions are ignored. In other terms, all theleading-twist asymmetries listed in eqs. (64-71) can be exactly reobtained from eq. (75) [189], whereasat twist three the results of the two approaches are identical if one neglects the “tilde” distribution andfragmentation functions arising from quark-gluon correlations. So far, we have only considered the extended parton model, i.e. the parton model incorporating intrinsictransverse momenta. One may wonder whether the results we have presented have any solid QCD foun-dation. The answer to this question is positive, at least in a particular kinematic regime. Semi-inclusiveprocesses are characterised by two scales, besides the confinement scale Λ
QCD : the momentum transfer Q and the transverse momentum of the final hadron P h ⊥ or, equivalently, the transverse momentum Q T ≡ | q T | of the virtual photon in the hN collinear frame.Extending the pioneering work on back-to-back jet production of Ref. [89], a TMD factorisationtheorem for SIDIS and DY has been proven [52, 51, 190]. The proof is valid in the low transverse-momentum region P h ⊥ ( Q T ) ≪ Q . In this framework the unpolarised SIDIS structure function is33 ′ q k ′′ qq l ′ l ′ k ′ lk k kk ′ k ′′ k ′ k ′′ l l Figure 13: Feynman diagrams of the elementary processes contributing to SIDIS at first order in α s .written as F UU,T ( x B , z h , Q , Q T ) = X a e a x Z d k T Z d κ T Z d l T δ ( k T − κ T + l T + q T ) × H ( Q ) f a ( x B , k T ) D a ( z h , k T ) U ( l T ) . (76)For simplicity the dependence of the distribution functions on ζ = (2 v · P ) /v and of the fragmentationfunction on ζ h = (2˜ v · P ) / ˜ v ( v and ˜ v are vectors off the light-cone) is omitted . The variables ζ and ζ h serve to regulate the light-cone singularities, as explained in Section 3.3. H is a perturbative hardfactor written as a series in powers of α s . The soft factor U arises from the radiation of soft gluons(of transverse momentum l T ) and is a matrix element of Wilson lines in the QCD vacuum. Also notdisplayed in eq. (76) is the dependence of all quantities on the renormalisation scale µ and on thesoft-gluon rapidity cut-off ρ = p (2 v · ˜ v ) /v ˜ v . Of course, the physical observable F does not dependon any of these regulators.The generalisation of eq. (76) to the polarised structure functions, in particular to those generatingtransverse SSA’s, has been proposed in Refs. [191, 192]. The parton model expressions of Section 4.1.1are recovered at tree level, i.e. O ( α s ), since H (0) = 1 and U (0) ( l T ) = δ ( l T ). At high transverse momenta, Q T ≫ Λ QCD , SIDIS structure functions can be described in collinearQCD. The azimuthal angular dependence of hadrons in leptoproduction was proposed longtime ago asa test of perturbative QCD [193]. In collinear factorisation, transverse momenta are generated by gluonradiation. At first order in α s the hard elementary processes shown in Fig. 13 contribute to the fourunpolarised SIDIS structure functions F UU and to the two double-longitudinal structure functions F LL .Introducing the partonic variables ˆ x and ˆ z , defined as ˆ x = Q / k · q = x B /x , ˆ z = k · k ′ /k · q = z h /z ,where k and k ′ are the four-momenta of the incident and fragmenting partons, respectively, and x and z are the usual light-cone momentum fractions, i.e. k = xP and k ′ = P h /z , one has for the F UU ’s atleading order in α s and leading twist [194, 195, 192] F UU ( x, Q ) = α s π z q X a e a x B Z x B dˆ x ˆ x Z z h dˆ z ˆ z δ (cid:18) Q T Q − (1 − ˆ x )(1 − ˆ z )ˆ x ˆ z (cid:19) × h f a (cid:16) x B ˆ x (cid:17) D a (cid:16) z h ˆ z (cid:17) C γ ∗ q → qgUU + f a (cid:16) x B ˆ x (cid:17) D g (cid:16) z h ˆ z (cid:17) C γ ∗ q → gqUU + f g (cid:16) x B ˆ x (cid:17) D a (cid:16) z h ˆ z (cid:17) C γ ∗ g → q ¯ qUU i , (77)and analogous formulae for the F LL ’s. The Wilson coefficients C represent elementary cross sectionsand are listed in Ref. [192].The structure function F sin φ h LU encountered in Section 4.1.1, which produces a beam-spin asymmetryand vanishes in the parton model, gets a non zero perturbative QCD contribution at leading twist andorder α s [196, 197]. 34n the contrary, the transversely polarised structure functions F UT , which vanish at leading twist incollinear factorisation, since there is no chirally-odd fragmentation function, emerge at twist three, as theresult of quark-gluon correlations. Following the early work of Ref.[17, 18, 198], a twist-three collinearfactorisation theorem valid at large transverse momenta was proven [53, 54, 55]. In this approach thecross section for SIDIS with a transversely polarised target has the general form [191, 199, 200]d σ ∼ G F ( x, x ′ ) ⊗ dˆ σ ⊗ D ( z ) + h ( x ) ⊗ dˆ σ ′ ⊗ ˆ E F ( z, z ′ ) , (78)where the first term contains a quark-gluon correlation function for the transversely polarised nucleonand the ordinary unpolarised fragmentation function for the final hadron, whereas the second termcombines the transversity distribution with a twist-three fragmentation function. Let us focus on thefirst contribution (twist-three effects in the initial state). The hadronic tensor can then be written as W µν ( P, q, P h ) = X a Z d zz w aµν ( P, q, P h /z ) zD a ( z ) , (79)where the partonic tensor w µν contributing to the transversely polarised structure functions is (seeFig. 14) w µν ( P, q, P h ) = Z d k Z d k ′ Tr [Φ A ( k, k ′ ) H µν ( k, k ′ , q, P h )] . (80)In this expression Φ A is the quark-gluon correlation matrixΦ A ( k, k ′ ) = Z d ξ (2 π ) Z d η (2 π ) e i k · ξ e i( k ′ − k ) · η h P, S | ¯ ψ (0) W − [0 , η ] gA + ( η ) W − [ η, ξ ] ψ ( ξ ) | P, S i , (81)and H µν represents the perturbatively calculable partonic hard scattering. By means of the collinearexpansion [54] one can get H ( k, k ′ ) = H ( x, x ′ ) + ∂H∂k α (cid:12)(cid:12)(cid:12)(cid:12) x,x ′ ( k α − xP α ) + ∂H∂k ′ α (cid:12)(cid:12)(cid:12)(cid:12) x,x ′ ( k ′ α − x ′ P α ) , (82)and finally end up with w µν = i Z d x Z d x ′ Tr (cid:20) Φ αF ( x, x ′ ) ∂H ( x, x ′ ) ∂k ′ α (cid:21) , (83)where the quark-gluon correlation matrix Φ αF , defined asΦ αF ( x, x ′ ) = Z d ξ − π Z d η − π e i xP + ξ − e i( x ′ − x ) P + η − h P, S | ¯ ψ (0) W − [0 , η ] gF + α ( η ) W − [ η, ξ ] ψ ( ξ ) | P, S i , (84)contains the multiparton distributions G F , ˜ G F , H F , E F introduced in Section 3.4. It is easy to verifythat, due to the structure of Φ αF , the hadronic tensor receives contributions only from the imaginarypart of the hard blob, arising from internal propagator poles.Considering for definiteness the Sivers contribution to the cross section, its explicit expression is[191, 199]d σ | Siv ∼ Z d xx Z d zz δ (cid:18) Q T Q − (cid:18) − xx B (cid:19) (cid:18) − zz h (cid:19)(cid:19) × X a e a (cid:20) x d G aF ( x, x )d x ˆ σ D + G aF ( x, x ) ˆ σ G + G aF ( x,
0) ˆ σ F + G aF ( x, x B ) ˆ σ H (cid:21) D a ( z ) + . . . (85)35 Pq qP h P h k k ′ σ H Φ A Figure 14: General diagram contributing to SIDIS SSA’s in the twist-three factorisation.The first two terms represent the so-called “soft-gluon pole” contribution ( x g = x ′ − x = 0), the thirdterm is the “soft-fermion pole” contribution ( x ′ = 0), the fourth term is the “hard pole” contribution( x ′ = x B ). The dots represent the contributions of ˜ G F and of the gluonic correlation functions.In the intermediate transverse-momentum region, i.e. Λ QCD ≪ Q T ( P h ⊥ ) ≪ Q , one expects thatboth the TMD and the twist-three pictures should hold. This has been explicitly verified in Refs. [56, 58].The output of these important works is a set of relations that connect the T -odd TMD distributions(Sivers and Boer-Mulders functions) on one side, with the quark-gluon correlations on the other side. / baryons The leptoproduction of spin 1 / ℓ + N ↑ → ℓ ′ + B ↑ + X , thespin transfer between the initial and the final particle occurs in collinear kinematics. The cross sectionintegrated over the transverse momentum gets in fact a double-spin term proportional to the productof the transversity distribution h and the “transversity” fragmentation function H .This doubly polarised process has the advantage of being free from the theoretical complicationsrelated to the transverse motion of quarks: the ordinary collinear QCD factorisation theorem appliesand a perturbative study is possible, since we know the Q evolution of both h (Section 3.2) and H [201]. At leading order in QCD, the transverse polarisation P BT of the produced baryon is given by P BT = P NT ˆ D NN ( y ) P a e a h a ( x B , Q ) H B/a ( z h , Q ) P a e a f a ( x B , Q ) D B/a ( z h , Q ) , (86)where P NT is the nucleon polarisation and ˆ D NN ( y ) = (1 − y ) / (1 − y + y /
2) is the spin transfer coefficientof quarks.In the case of Λ hyperons, information on the spin transfer in the fragmentation process can beobtained from the Λ polarisation extracted from the angular distribution in the weak Λ → p π − decay.The transverse polarisation of Λ’s produced in hard processes with initial transversely polarised hadronswas studied long time ago in Refs. [202, 203], and reinvestigated more recently [204, 32]. Phenomeno-logical analyses are presented in Refs. [205, 206, 207]. The transversity fragmentation function H Λ1 ismeasurable in e + e − production of transversely polarised Λ pairs.Transverse Λ polarisation can also be observed in SIDIS with an unpolarised target, by measuringasymmetries in the transverse momentum distribution of the hyperons. One contribution ( T -odd inthe final state) involves the quark density f and the polarising fragmentation function D ⊥ Λ1 T . This36echanism for Λ polarisation in SIDIS was studied in Ref. [208]. Another contribution ( T -odd inthe initial state) involves the Boer-Mulders function h ⊥ coupled to the (unintegrated) fragmentationfunction H Λ1 ( z, p T ). At high P h ⊥ , this effect has a counterpart in the twist-three approach. Theinitial-state T -odd mechanisms for producing Λ polarisation have been investigated in the context oftwist-three factorisation by Zhou, Yuan and Liang [209]. Another partially inclusive DIS reaction that can provide information on the transverse-spin structure ofhadrons, and in particular on transversity, is two-particle leptoproduction from a transversely polarisedtarget, ℓ ( l )+ N ↑ ( P ) → ℓ ′ ( l ′ )+ h ( P )+ h ( P )+ X ( P X ), with the two spinless final hadrons in the same jet.Two-hadron production in SIDIS has been proposed and studied by various authors [210, 33, 211, 212] asa process probing the transverse polarisation distribution in combination with a dihadron fragmentationfunction (DiFF). The idea is to look at an angular correlation between the spin of the fragmentingquark and the relative transverse momentum of the hadron pair, without involving the transversemomenta of quarks. Integrating over the total transverse momentum of the final hadrons, one getsan asymmetry in the azimuthal angle between the two-hadron plane and the scattering plane. Thisasymmetry is determined by a fragmentation function usually called H < ) , which does not depend on theintrinsic transverse motion of quarks and arises from the interference between different channels of thefragmentation process into the two-hadron system. Thus, all the difficulties related to non-collinearityare in this process avoided.The first authors who suggested resonance interference as a way to produce non-diagonal fragmen-tation matrices of quarks were Cea et al. [213] in their attempt to explain the observed transversepolarisation of Λ hyperons produced in pN interactions [4]. The unpolarised dihadron fragmentationfunctions appeared for the first time in the context of jet calculus [214, 215]. The extension to thepolarised case was discussed in Refs. [210, 30, 216, 33] and a complete classification of the DiFF’s wasgiven at leading twist in Ref. [211] and at twist-3 in Ref. [217].The kinematics of the process in the γ ∗ N frame is shown in Fig. 15. We introduce the totalmomentum of the hadron pair P h = P + P (with invariant mass M h = P h ), the relative momentum R = ( P − P ) /
2, and the variables z = z + z = P − /κ − + P − /κ − = P − h /κ − (the light-cone fractionof the fragmenting-quark momentum carried by the hadron pair) and ζ = 2 R − /P − h (which describeshow the total momentum of the pair is split into the two hadrons). R T is the transverse component of R with respect to P h , and φ R is the azimuthal angle of R T in the plane orthogonal to the γ ∗ N axis,measured with respect to the scattering plane. The azimuthal angle of the target spin vector is φ S .Calling κ T the transverse momentum of the fragmenting quark with respect to P h , the unpolarised andtransverse-spin projections of the two-hadron fragmentation matrix ∆ at leading twist are∆ [ γ − ] = D ( z, ζ , κ T , R T , κ T · R T ) , (87)∆ [i σ i − γ ] = 1 M + M (cid:2) ε ijT κ T j H ⊥ ( z, ζ , κ T , R T , κ T · R T ) + ε ijT R T j H < ) ( z, ζ , κ T , R T , κ T · R T ) (cid:3) . (88)These are the probabilities for an unpolarised quark and for a transversely polarised quark, respectively,to fragment into a hadron pair. Upon integration over κ T , the contribution of the Collins-type DiFF H ⊥ disappears, and the only remaining transverse-spin term is the one containing the (integrated)interference fragmentation function H < ) ( z, ζ , M h ).It is convenient to consider the centre-of-mass frame of the two hadrons (Fig. 15), where R T ≡| R T | = | R | sin θ and θ is the angle between the direction of the hadron emission and P h (in the γ ∗ N frame). For two alike hadrons of mass m , one has ζ = 2 ( | R | /M h ) cos θ and | R | = p M h − m .In terms of these variables, keeping only the unpolarised and the transversely polarised terms, the37 π − P π + P h θ P π − π + π − CMframe R T S T P π + P h φ R ⊥ P φ S qk k ′ Figure 15: Kinematics of two-hadron leptoproduction.leading-twist partonic expression for the cross section of two-hadron leptoproduction readsd σ d x B d y d z h d φ R d φ S d M h d cos θ = α πxyQ X a e a x (cid:26)(cid:18) − y + y (cid:19) f a ( x B ) D a ( z h , M h , cos θ ) − (1 − y ) S ⊥ | R | M h sin θ sin( φ R + φ S ) h a ( x B ) H < ) a ( z h , M h , cos θ ) (cid:27) . (89)Here it is z = z h ≡ P · P h /P · q , a relation valid modulo 1 /Q corrections. At twist three, thereappear extra terms in the cross section: a cos φ R unpolarised contribution and a sin φ S transverse-spincontribution [217].The first model for the two-pion DiFF H < ) ,sp was presented in Ref. [33], where the phase differencebetween s and p waves was taken from ππ phase shifts in elastic scattering. The resulting fragmentationfunction (called δ ˆ q I in Ref. [33]), changes sign around the ρ mass.In a more recent model [218] the fragmentation functions are expanded in Legendre polynomials ofcos θ keeping only the first few terms, corresponding to the lowest values of relative orbital momentum.This truncation is expected to be legitimate for not very large M h . Thus one can write D ( z h , M h , cos θ ) = D o ( z h , M h ) + D sp ( z h , M h ) cos θ + D pp ( z h , M h ) 14 (3 cos θ − , (90) H < ) ( z h , M h , cos θ ) = H < ) ,sp ( z h , M h ) + H < ) ,pp ( z h , M h ) cos θ (91)(remember that H < ) multiplies a sin θ factor in the cross section). The interpretation of these terms,signaled by their superscripts, is the following [218]: D o is a diagonal component, receiving contributionsfrom s and p waves of the dihadron system separately (the “background”); D sp and H < ) ,sp originate fromthe interference of a s wave and a p wave; D pp and H < ) ,pp arise from the interference of two p waves.The main channels contributing to the fragmentation of a quark q into a π + π − pair are: 1) incoherentfragmentation, q → π + π − X ; 2) fragmentation via a ρ resonance, q → ρX → π + π − X ; 3) fragmentationvia a ω resonance decaying into three pions, q → ωX → π + π − π X . Pions in channel 1 are expected tobe mostly produced in s wave; pions in channel 2 come from the two-body decay of a vector meson andare in a relative p wave; pions in channel 3 are prevalently in p wave, but a fraction of them may alsobe in s wave. The functions D sp and H < ) ,sp arise from the interference of channels 1-2 and 1-3 [219].A model based on a more sophisticated analysis of the fragmentation channels [219] predicts acompletely different behaviour for H < ) ,sp , with a peak at the ρ mass and a broader maximum at the ω mass. Its size is about 30 % of the unpolarised fragmentation function as shown in Fig. 16.A different definition of the relative transverse momentum is proposed by Artru [28], who uses thevector r ⊥ = ( z P ⊥ − z P ⊥ )( z + z ), which is perpendicular to the γ ∗ N axis, and its azimuthal angle38 | ~ R | M h H < ) , o t / D , oo M h (GeV) − | ~ R | M h H < ) , o t / D , oo z Figure 16: Model prediction for the ratio ( −| R | H < ) ,sp / ( M h D o ) as a function of M h (left) and z (right).The dotted lines represent the positivity bounds.Figure 17: Left: Configuration of the process e + e − → h h X in the jet frame, used for the cos( φ + φ )reconstruction of the Collins asymmetry. Right: The same process in the Gottfried-Jackson frame, usedfor the the cos 2 φ reconstruction of the Collins asymmetry. φ r . The advantage of φ r is that it is by construction invariant with respect to boosts along the γ ∗ N direction, which is not the case of φ R although the two angles are the same in the γ ∗ N frame. Usingthe recursive fragmentation string model (see Section 4.1.5) the “joint p T spectrum” of the first and thesecond rank hadrons has been calculated [177]. By suitably integrating the spectrum, expressions forboth the Collins FF and the dihadron FF in principle may be obtained. e + e − annihilation An independent source of information on the Collins fragmentation function H ⊥ is inclusive two-hadronproduction in electron–positron collisions, e + + e − → h + h + X , with the two hadrons (typicallypions) in different hemispheres. We know that the Collins function H ⊥ produces a cos φ modulation,where φ is the azimuthal angle between the plane containing the quark and the hadron momenta, andthe plane normal to S q . Considering a single jet in e + e − hadron production, the Collins modulationwould average to zero in a large event sample. Thus, in e + e − annihilation the Collins effect can onlybe observed in the combination of two fragmenting processes of a quark and an antiquark, resulting inthe product of two Collins functions with an overall modulation of the type cos( φ + φ ), where φ and φ are the azimuthal angles of the final hadrons around the quark-antiquark axis, with respect to the e + e − → q ¯ q scattering plane.Two-hadron production in e + e − collisions was studied in Refs. [220, 221, 222, 223, 224, 225]. Thetree-level differential cross section in the jet frame with respect to the quark-antiquark direction (Fig. 17,39eft) reads d σ dΩd z d z d φ d φ = 3 α s X a = q, ¯ q e a z z n (1 + cos θ ) D a [0]1 ( z ) D a [0]1 ( z )+ sin θ cos( φ + φ ) H ⊥ a [1]1 ( z ) H ⊥ a [1]1 ( z ) o , (92)where dΩ = d cos θ d φ ℓ ( θ is the angle between the lepton axis and the q ¯ q axis, in the q ¯ q centre-of-mass frame, whereas φ ℓ gives the orientation of the scattering plane around the q ¯ q axis), and we haveintroduced the one-dimensional moments F [ n ] ( z ) ≡ Z d p T (cid:18) p T M h (cid:19) n F ( z, p T ) . (93)Eq. (92) and the following results refer to the case of photon-mediated e + e − annihilation. Z productionand γ ∗ Z interference effects have been investigated in Refs. [222, 225].From an experimental point of view, the quark-antiquark direction is not directly accessible, and isapproximated by the dijet thrust axis ˆ n , defined by T = max P h | P h · ˆ n | P h | P h | , (94)where the sum is over all detected particles. The resulting cos( φ + φ ) asymmetry is given by a ( θ, z , z ) = sin θ θ P a e a (cid:16) H ⊥ a [1]1 ( z ) H ⊥ a [1]1 ( z ) (cid:17)P a e a (cid:16) D a [0]1 ( z ) D a [0]1 (cid:17) . (95)We will refer to this method of extracting the Collins asymmetry by measuring azimuthal distributionsaround the thrust axis as to the “cos( φ + φ ) method” [221, 44, 226, 225].There is another way of reconstructing the asymmetry, the so-called “cos 2 φ method”, which isbased on a different geometry and does not require the knowledge of the thrust axis. In this case onemeasures the hadron yields as a function of φ , the angle between the plane containing the momentumof hadron 2 and the leptons, and the plane defined by the two hadron momenta [222, 223, 225] (thisframe, shown in Fig. 17 (right), is similar to the Gottfried-Jackson frame in Drell-Yan processes [227]).The corresponding cross section isd σ dΩd z d z d q T = 3 α Q X a e a z z ( (1 + cos θ ) C [ D a D a ] + sin θ cos 2 φ C " h · κ T ˆ h · κ T − κ T · κ T M M H ⊥ a H ⊥ a , (96)where ˆ h ≡ P ⊥ / | P ⊥ | and κ T , κ T are the transverse momenta of the two fragmenting quarks withrespect to the hadron directions. The cos 2 φ asymmetry reads a ( θ, z , z ) = sin θ θ P a e a C h (2 ˆ h · κ T ˆ h · κ T − κ T · κ T ) H ⊥ a H ⊥ a i M M P a e a C [ D a D a ] . (97)Electron-positron scattering can also allow accessing the fragmentation function H of transverselypolarised baryons [27, 220, 228]. In the case of Λ’s, the specific process is back-to-back ΛΛ inclusive40igure 18: Geometry of two hadron-pair production in e + e − collisions.production, e + e − → Λ Λ X , with the hyperon and the anti-hyperon decaying into pπ − and ¯ pπ + , respec-tively. It was shown in Ref. [220] that the cross section of this process contains an azimuthal modulationproportional of the type sin θ cos( φ + ¯ φ ) H Λ1 ( z ) H Λ1 ( z ) ∝ H Λ1 ( z ) H Λ1 ( z ), where θ is the angle betweenthe collision axis and the q ¯ q axis in the jet frame (Fig. 17, left) and φ, ¯ φ are the azimuthal angles ofthe proton and antiproton in the same frame (with the z direction given by the q ¯ q jet and the x axis inthe plane of the beams and jets. The cos( φ + ¯ φ ) asymmetry is determined by measuring the differencebetween the number of p ¯ p pairs on the same side of the scattering plane and the number of pairs onopposite sides.Finally, the interference fragmentation function H < ) can be extracted from the production of twohadron pairs in electron-positron annihilation: e + e − → ( h h )( h ′ h ′ ) X , where the particles in bracketsbelong to two back-to-back jets [216]. The (complicated) geometry of this process is shown in Fig. 18.The observable quantity is the angular correlation of the production planes, expressed by the so-calledArtru-Collins asymmetry. The kinematics of the process is described by doubling the variables intro-duced in Section 4.1.5. If we call φ R and ¯ φ R the azimuthal angles of the transverse relative momenta R T and ¯ R T of the two hadron pairs, the Artru-Collins azimuthal asymmetry is the cos( φ R + ¯ φ R ) correlation.In e + e − annihilation, the interference fragmentation functions and the Collins function are typicallyprobed at much larger scales compared to SIDIS. However, the evolution of H < ) , differently from thatof H ⊥ , is known [229]. Therefore, a consistent combined analysis of dihadron production in e + e − annihilation and SIDIS is possible and may provide an alternative way to extract the transversitydistributions [230]. Drell-Yan (DY) dilepton production with various polarisations of the two particles in the initial stateis a very rich source of knowledge on the hadronic structure. The main advantage of this class ofreactions is that they do not involve fragmentation functions, but only parton distributions. However,unless one considers antiproton–proton scattering, or pion-proton scattering, DY processes necessarilyinvolve sea × valence products. This means that, while they provide direct information about antiquarkdistributions, which are less determined in SIDIS, their asymmetries are generally small.In principle, DY production with two transversely polarised hadrons is the cleanest reaction forstudying the transversity distribution h ( x ) and the pioneering works of Ref. [25] and [231] were indeed41 zP P h^ f lepton plane (cm) q l’ l Figure 19: The Collins-Soper frame. The z-axis bisects the angle between P and − P , the momentaof the two initial state hadrons.devoted to this process. However, in order to observe sizable double-spin asymmetries and extract h ,we probably have to wait for a new generation of experiments with polarised antiprotons [232]. On theother hand, unpolarised and singly-polarised DY processes can probe a large variety of TMD’s relatedto transverse spin, and are now attracting a wide theoretical and experimental interest. Drell-Yan lepton-pair production is the process A ( P ) + A ( P ) → ℓ + ( l ) + ℓ − ( l ′ ) + X , where A and A are hadrons and X is an undetected system. The center-of-mass energy squared of this reaction is s = ( P + P ) ≃ P · P , having neglected in the approximate equality the hadron masses M and M .The lepton pair originates from a virtual photon with four-momentum q = l + l ′ . In contrast to DIS, q is a time-like vector: Q = q >
0, and the invariant mass M of the lepton pair coincides with Q .The deep inelastic limit corresponds to Q , s → ∞ , with τ ≡ Q /s fixed and finite.The DY cross section is usually expressed in a dilepton center-of-mass frame and can be written asd σ d q dΩ = α sQ L µν W µν , (98)where L µν is the familiar leptonic tensor and W µν is the DY hadronic tensor. Among the infinitedilepton c.m. frames, related to each other by a rotation, the most often used is the Collins-Soper (CS)frame [233], characterised by a z axis that bisects the angle between P and − P as shown in Fig. 19.Another common dilepton c.m. frame is the Gottfried–Jackson frame [227], where the z axis coincideswith the direction of one of the colliding hadrons.In the unpolarised case the DY hadronic tensor contains four independent structure functions [234].Using the classification of Ref. [235] the cross-section becomesd σ UU d q dΩ = α sQ n (1 + cos θ ) W UU + sin θ W UU + sin 2 θ cos φ W cos φUU + sin θ cos 2 φ W cos 2 φUU o . (99)The double subscript refers to the polarisation states of the two colliding hadrons: U = unpolarised, L = longitudinally polarised, T = transversely polarised. In literature [234], the structure functions W UU , W UU , W cos φUU , W cos 2 φUU are also called (apart from a common factor), W T , W L , W ∆ , W ∆∆ , respec-tively. The angular distribution of leptons is often parametrised as1 N tot d N dΩ = 34 π λ + 3 (cid:16) λ cos θ + µ sin 2 θ cos 2 φ + ν θ cos 2 φ (cid:17) . (100)The three quantities λ, µ , and ν are related to W UU , W UU , W cos φUU , and W cos 2 φUU by λ = W UU − W UU W UU + W UU , µ = W cos φUU W UU + W UU , ν = 2 W cos 2 φUU W UU + W UU . (101)42he so-called Lam-Tung relation [234, 236, 237] λ + 2 ν = 1 , (102)which corresponds to W UU = 2 W cos 2 φUU , is valid at order α s in collinear QCD [238] (see below) and isslightly violated at order α s [239].In the polarised case, a complete analysis of the DY hadronic tensor is contained in Ref. [235], whereit is shown that in single-polarised DY, first studied in [231], the number of independent structurefunctions is 16, whereas the double-polarised hadronic tensor contains 28 structure functions, so thataltogether the number of independent structure functions in DY is 48. The full cross section can be foundin Ref. [235]. Limiting ourselves to unpolarised, single-transverse and double-transverse contributions,the cross section readsd σ d q dΩ = α sQ nh (1 + cos θ ) W UU + sin θ W UU + sin 2 θ cos φ W cos φUU + sin θ cos 2 φ W cos 2 φUU i + S T h sin φ S (cid:16) (1 + cos θ ) W T U + sin θ W T U + sin 2 θ cos φ W cos φT U + sin θ cos 2 φ W cos 2 φT U (cid:17) + cos φ S (sin 2 θ sin φ W sin φT U + sin θ sin 2 φ W sin 2 φT U ) i + (1 ↔ , T ↔ U )+ S T S T (cid:2) cos( φ S + φ S ) (cid:0) (1 + cos θ ) W T T + sin θ W T T + sin 2 θ cos φ W cos φT T + sin θ cos 2 φ W cos 2 φT T (cid:17) + cos( φ S − φ S ) (cid:16) (1 + cos θ ) W T T + sin θ W T T + sin 2 θ cos φ W cos φT T + sin θ cos 2 φ W cos 2 φT T (cid:17) + sin( φ S + φ S ) (sin 2 θ sin φ W sin φT T + sin θ sin 2 φ W sin 2 φT T )+ sin( φ S − φ S ) (sin 2 θ sin φ W sin φT T + sin θ sin 2 φ W sin 2 φT T ) i + . . . o . (103) φ S and φ S are the azimuthal angles of the spin vectors of hadrons A and B , respectively. This angularstructure is valid in any dilepton c.m. frame, but the numerical values of the structure functions areframe-dependent.In the parton model and at leading order in QCD the two invariants x = Q / P · q and x = Q / P · q can be interpreted as the fractions of the longitudinal momenta of the hadrons A and B carried by the quark and the antiquark that annihilate into the virtual photon. In the c.m. frameof the two colliding hadrons, which is the most convenient frame to study the partonic structure ofthe hadronic tensor, the photon momentum q µ can be parametrised as q µ = ( x P +1 , x P − , q T ) andacquires a transverse component q T . Neglecting terms of order 1 /Q , one has Q /x x s = 1, that is τ ≡ Q /s = x x . The structure functions in eq. (103) can be expressed in terms of the four variables x , x , Q T ≡ | q T | , Q .Other variables customarily used are the rapidity of the virtual photon, y ≡ ln( q + /q − ) = ln( x /x )and the Feynman variable x F = 2 q L / √ s = x − x . In a dilepton c.m. frame, y = (1 + cos θ ). Therelation between ( x , x ) and ( τ, y ) is x = √ τ e y , x = √ τ e − y . The DY cross-section can be variouslyreexpressed in terms of these variables:d σ d q = 2 s d σ d x d x d q T = 2 d σ d y d Q d q T = 2 ( x + x ) d σ d x F d Q d q T . (104) In the parton model, calling k and k the momenta of the quark (or antiquark) coming from hadron A and A respectively, the hadronic tensor shown in Fig. 20 is W µν = 13 X a e a Z d k Z d k δ ( k + k − q ) Tr [Φ( k ) γ µ ¯Φ( k ) γ ν ] . (105)43 P P P q qk ′ k ′ k k Φ Φ Figure 20: The parton-model diagram for the DY hadronic tensor.Here Φ is the quark correlation matrix for hadron A , Φ is the antiquark correlation matrix for hadron A , and the factor 1 / k ) replaced by Φ( k ) and ¯Φ( k ) replaced by ¯Φ( k ), which accounts for the casewhere a quark is extracted from A and an antiquark is extracted from A . In the following formulaewe shall denote this term symbolically by [1 ↔ ξ = k +1 /P +1 and ξ = k − /P − , and working outthe delta function of four-momentum conservation, one finds ξ = x , ξ = x , and the hadronic tensorbecomes W µν = 13 X a e a Z d k T Z d k T δ ( k T + k T − q T )Tr [Φ( x , k T ) γ µ Φ( x , k T ) γ ν ] + [1 ↔ . (106)Inserting here the explicit partonic expressions of the quark correlators, it is not difficult to get theparton model expressions of the DY structure functions [47, 240, 235]. Only 24 of these 48 structurefunctions are non vanishing at leading twist. The most relevant ones are, in the Collins-Soper frame(but the Gottfried-Jackson expressions differ from these only by subleading terms O ( Q T /Q )) W UU = 13 C [ f ¯ f ] , (107) W cos 2 φUU = 13 C "
2( ˆ h · k T )( ˆ h · k T ) − k T · k T M M h ⊥ ¯ h ⊥ , (108) W T U = − C " ˆ h · k T M f ⊥ T ¯ f , W UT = 13 C " ˆ h · k T M f ¯ f ⊥ T , (109) W sin(2 φ − φ S ) T U = 13 C " ˆ h · k T M h ¯ h ⊥ , W sin(2 φ − φ S ) UT = − C " ˆ h · k T M h ⊥ ¯ h , (110) W cos(2 φ − φ S − φ S ) T T = 13 C [ h ¯ h ] , (111)where ˆ h ≡ q T /Q T and C denotes the transverse-momentum convolution of eq. (63) with the addition ofthe [1 ↔
2] term. In eq. (110) we defined the combinations W sin(2 φ − φ S S ) T U ( UT ) ≡ − ( W cos 2 φT U ( UT ) − W sin 2 φT U ( UT ) )and W cos(2 φ − φ S − φ S ) T T ≡ ( W cos 2 φT T + W sin 2 φT T ), which correspond to the angular modulations indicated bytheir superscripts. 44qs. (107-111) contain a series of interesting results. First of all, the Boer-Mulders function h ⊥ generates a cos 2 φ asymmetry in unpolarised DY. At leading twist, the ν parameter is given by ν = 2 W cos 2 φUU W UU = 2 C h (2( ˆ h · k T )( ˆ h · k T ) − k T · k T ) h ⊥ ¯ h ⊥ i M M C [ f ¯ f ] . (112)Since ν = 0 and λ = 1, the Lam-Tung relation is violated, and this is one of the remarkable consequencesof the intrinsic transverse motion of quarks.Concerning singly-polarised Drell-Yan processes, the Boer-Mulders function combines with thetransversity distribution in the sin(2 φ − φ S ), or sin(2 φ − φ S ), asymmetry, whereas the Sivers functionis probed via the F T U (or F UT ) structure function associated with the sin φ S (or sin φ S ) asymmetry.Note that upon integration over the lepton angles only the Sivers asymmetry is non vanishing. With two transversely polarised colliding hadrons, the cos(2 φ − φ S − φ S ) term provides a direct accessto transversity. Inserting eqs. (107, 111) into eq. (103) and integrating the cross section over q T , onegets d σ d x d x dΩ = α Q X a e a (cid:2) (1 + cos θ ) f a ( x ) ¯ f a ( x )+ S T S T sin θ cos(2 φ − φ S − φ S ) h a ( x )¯ h a ( x ) (cid:3) + [1 ↔ . (113)This parton-model expression can be generalised to QCD by resorting to the collinear factorisationtheorem, which for the polarised DY process reads [241]d σ = X a X λ λ ′ λ λ ′ Z d ξ Z d ξ ρ (1) λ ′ λ f a ( ξ , µ ) ρ (2) λ ′ λ ¯ f a ( ξ , µ ) dˆ σ λ λ ′ λ λ ′ ( Q , µ , α s ( µ )) , (114)where ξ and ξ are the momentum fractions of the quark (from hadron A ) and antiquark (from A ), ρ (1) and ρ (2) are the quark and antiquark spin density matrices, dˆ σ λ λ ′ λ λ ′ is the cross-section matrixof the elementary subprocesses in the quark and antiquark helicity space, µ is the factorisation scale.At leading order, i.e. O ( α s ), the only contributing subprocess is q ¯ q → ℓ + ℓ − and ξ = x , ξ = x . Inthe transversely polarised case, one reobtains eq. (113), except that all distribution functions acquire a Q -dependence. Thus the LO double transverse asymmetry is A DYT T = a T T P a e a h a ( x , Q )¯ h a ( x , Q ) + [1 ↔ P a e a f a ( x , Q ) ¯ f a ( x , Q ) + [1 ↔ , (115)where a T T = sin θ θ cos(2 φ − φ S − φ S ) , (116)is the elementary double-spin asymmetry for q ¯ q → ℓ + ℓ − . We see that a measurement of A DYT T woulddirectly provide the product of quark and antiquark transversity distributions, with no mixing with otherunknown quantities. At next-to-leading order (NLO) the DY transverse cross section gets contributionsfrom virtual-gluon (vertex and self-energy) corrections and real-gluon emission, which were calculatedby several authors with different methods [242, 243, 244, 81]. The NLO double transverse asymmetrywas investigated in Refs. [245, 246, 247, 248]. 45 .3.4 DY azimuthal and spin asymmetries in QCD
As in the case of SIDIS, perturbative gluon radiation can generate a non-zero transverse momentum Q T .For instance, the contribution of the quark-antiquark annihilation process q ¯ q → γ ∗ g to the unpolarisedangular distribution in the Collins-Soper frame is [238]1 N tot d N dΩ = 316 π (cid:18) Q + Q T Q + Q T + Q − Q T Q + Q T cos θ + 12 Q T Q + Q T sin θ cos 2 φ + . . . (cid:19) , (117)where we have omitted the sin 2 θ cos φ term which is the only one depending on the quark and antiquarkdistributions. From (117) one gets λ = Q − Q T Q + Q T , ν = Q T Q + Q T , (118)and the Lam-Tung relation is fulfilled. The contribution of the qg → γ ∗ q is more complicated, but alsosatisfies the Lam-Tung relation, which holds for the complete leading-order cross section.The perturbative QCD approach to the DY angular distribution sketched above holds for Q T ∼ Q .At small Q T , large logarithms of the form ln( Q /Q T ) appear, which must be resummed. This is donein the space conjugate to q T and gives rise to a Sudakov form factor, according to the Collins-Soperprocedure [89]. The Sudakov resummation for the structure functions of unpolarised DY production,including those related to azimuthal asymmetries, has been studied in Refs. [249, 250] and the Lam-Tungrelation is found to be unaffected by the resummation.The perturbatively generated cos 2 φ asymmetry is suppressed as Q T /Q at small Q T . A furthercontribution to this asymmetry can arise in the twist-three approach from the product of two quark-gluon correlation functions E F , one associated with the quark from hadron A , the other with theantiquark from hadron A [251].In the singly-polarised DY case the situation is again analogous to SIDIS. At large Q T ∼ Q ≫ Λ QCD ,a DY single-spin asymmetry is generated by the G F quark-gluon correlator of the polarised hadron. Asmooth transition from this twist-three mechanism to the Sivers effect occurs in the intermediate regionΛ QCD ≪ Q T ≪ Q , where both the higher-twist and the TMD factorisations apply [57]. Another SSA’sarises from the chirally-odd quark-gluon correlation function E F of the unpolarised hadron coupledto the transversity distribution of the transversely polarised hadron, but in the low Q T limit thiscontribution vanishes (after integration over the lepton angles only the Sivers asymmetry survives). We finally discuss a third class of reactions that probe the transverse-spin and transverse-momentumstructure of hadrons: inclusive hadroproduction with one transversely polarised hadron in the initialstate, that is A ↑ + B → h + X , where an unpolarised (or spinless) hadron h is produced with atransverse momentum P T with respect to the collision axis. An interesting variation of this process isthe production of a transversely polarised hadron, i.e., a Λ hyperon, from unpolarised hadron-hadronscattering, that is A + B → h ↑ + X (Section 4.4.3). We will limit ourselves to a brief description ofthese reactions, referring the reader for more detail to some reviews [63, 252] and to the original papers.We first consider hadroproduction with a transversely polarised colliding particle. The measuredquantity is the single-spin asymmetry A N = d σ ↑ − d σ ↓ d σ ↑ + d σ ↓ , (119)with the cross sections usually expressed as functions of P T and of the Feynman variable x F = 2 P L / √ s ( P L being the longitudinal momentum of the produced hadron). In terms of the scattering angle θ ,46eynman’s x can be written as x F = 2 P T / √ s tan θ . Another often used variable is the pseudorapidity η = − ln tan( θ/ P T can be formally written asd σ = X abc X λ a λ ′ a λ c λ ′ c ρ aλ a λ ′ a f a ( x a ) ⊗ f b ( x b ) ⊗ dˆ σ λ a λ ′ a λ c λ ′ c ⊗ D h/cλ c λ ′ c ( z ) . (120)Here f a ( f b ) is the distribution of parton a ( b ) inside the hadron A ( B ), ρ aλ a λ ′ a is the spin density matrixof parton a , D h/cλ c λ ′ c is the fragmentation matrix of parton c into hadron h , and dˆ σ is the (perturbativelycalculable) cross-section of the elementary process a + b → c + . . . (a two-body scattering, a + b → c + d ,at lowest order).If the produced hadron is unpolarised, or spinless, only the diagonal elements of D h/cλ c λ ′ c are non zero,i.e. D h/cλ c λ ′ c ∼ δ λ c λ ′ c D h/c , where D h/c is the unpolarised fragmentation function. Together with helicityconservation in the partonic subprocess, this implies λ a = λ ′ a . Therefore, the cross section (120) carriesno dependence on the spin of hadron A and all single-spin asymmetries vanish [16]. In order to escapesuch a conclusion one must consider either the intrinsic transverse motion of quarks [34, 29, 30], orhigher-twist effects [17, 18, 198, 19, 53, 54]. In the former case, one can probe a number of distributionand fragmentation functions, including the transversity distribution (transversely polarised quarks inhadron A ↑ ), the Sivers function (unpolarised quarks in hadron A ↑ ), the Boer-Mulders function (trans-versely polarised quarks in hadron B ), the Collins function (transversely polarised quarks fragmentinginto hadron h ). The twist-three single-spin asymmetries involve various quark-gluon correlators, eitherin the initial state (distribution functions), or in the final state (fragmentation functions). The mainproblem is that all TMD or twist-three contributions mix up in a single observable, A N , which makesthe physical interpretation of the results quite unclear. When the intrinsic transverse motion of quarks is taken into account, the QCD factorisation theoremfor inclusive hadroproduction is not proven, and actually is known to be explicitly violated in somecases [254, 104]. Nevertheless, one can write a non-collinear factorisation formula in the context of theextended parton model, with a tree-level elementary kernel. One must obviously recall that: i) the gen-eralisation of this kernel to higher order in α s is not a legitimate procedure; ii) the transverse-momentumdependent distribution and fragmentation functions appearing in the hadroproduction factorisation for-mula are not guaranteed to be universal quantities, i.e., to be the same functions as in other processes.The extended-parton model formula generalising eq. (120) is (we consider the production of a spinlesshadron) [188] d σ = X abcd X λ a λ ′ a λ c λ ′ c ρ aλ a λ ′ a f a ( x a , k T a ) ⊗ f b ( x b , k T b ) ⊗ dˆ σ λ a λ ′ a λ c λ ′ c ⊗ D h/cλ c λ ′ c ( z, p T ) . (121)where the convolutions ⊗ are now not only on the longitudinal momentum fractions x a , x b , z , but alsoon the transverse momenta k T a , k T b , p T . Note that even though h is unpolarised, its p T -dependent frag-mentation matrix D h/cλ c λ ′ c is non diagonal. The elementary cross sections have the structure dˆ σ λ a λ ′ a λ c λ ′ c ∼ P λ b λ d ˆ M λ c λ d ,λ a λ b ˆ M ∗ λ ′ c λ d ,λ ′ a λ b . The amplitudes ˆ M refer to the elementary subprocess a b → c d (rememberthat we are considering the tree level only). A natural reference frame is the center-of-mass frameof the colliding hadrons. The collision axis forms with the direction of the produced hadron a plane,that we call the hadronic plane. The tricky point about eq. (121) is that, due to intrinsic transversemomenta, the partonic scattering does not take place in the hadronic plane. This non-planar geometry47ives rise to some non-trivial phases in the distribution and fragmentation matrices. Also, the ampli-tudes ˆ M appearing in eq. (121) must be Lorentz transformed to the canonical amplitudes ˆ M definedin the partonic center-of-mass frame, an operation which introduces further phases. This complicatedstructure has been fully worked out in Ref. [188], where all the details concerning the kinematics andthe scattering amplitudes can be found. Here we limit ourselves to quoting some general results. Thecontribution to the transverse single-spin asymmetry from the qq → qq subprocess schematically readsd∆ σ qq → qq ∼ f ⊥ a T ⊗ f b ⊗ d∆ˆ σ ′ ⊗ D c + h a ⊗ f b ⊗ d∆ˆ σ ′′ ⊗ H ⊥ c + h a ⊗ h ⊥ b ⊗ d∆ˆ σ ′′′ ⊗ D c + f ⊥ a T ⊗ h ⊥ b ⊗ d∆ˆ σ ′′′′ ⊗ H ⊥ c . (122)One recognises the Sivers effect (first term), the Collins effect (second term), the Boer-Mulders effect(third term) and a mixed effect (fourth term). The other contributions, q ¯ q → gg , qg → qg , qg → gq , gq → gq , gq → qg , gg → q ¯ q , gg → gg , are explicitly given in Ref. [188]. They contain, besides thedistributions and fragmentation functions of linearly polarised gluons, the gluon Sivers function f ⊥ g T .For instance, the gq → gq and gg → ¯ qq contributions ared∆ σ gq → gq ∼ f ⊥ g T ⊗ f b ⊗ d∆ˆ σ I ⊗ D c + . . . , d∆ σ gg → ¯ qq ∼ f ⊥ g T ⊗ f g ⊗ d∆ˆ σ II ⊗ D c + . . . . (123)Processes that select these terms and allow accessing the gluon Sivers function are D meson production(which is dominated by the gg → ¯ cc channel) [255] and pion production at midrapidity (which probesin the RHIC kinematics the small x a region and thus proceeds predominantly via gluonic channels)[256]. Other reactions probing the Sivers functions of quarks and/or gluons without contributions fromthe fragmentation sector are prompt-photon production A ↑ + B → γ + X [257, 258, 259], photon–jetproduction A ↑ + B → γ + jet + X [258, 260], back-to-back dijet production A ↑ + B → jet + jet + X [261].On the contrary, the Collins effect can be singled out by studying asymmetric azimuthal correlationof hadrons inside a jet, that is A ↑ + B → jet + X → h + X [262]. As pointed out in Ref. [17, 198], non-vanishing single-spin asymmetries can be obtained in perturbativeQCD at higher-twist level. A twist-three factorisation theorem was proven for direct photon production[53, 54] and hadron production [55]. This work has been extended to cover the chirally-odd contributions[263, 264, 265]. Here we limit ourselves to quoting the main general results of these works. The twist-three phenomenological studies of data is treated in Section 5.5.At twist three the hadroproduction cross section is formally given byd σ = X abc n G aF ( x a , x ′ a ) ⊗ f b ( x b ) ⊗ dˆ σ ′ ⊗ D h/c ( z ) + h a ( x a ) ⊗ E bF ( x b , x ′ b ) ⊗ dˆ σ ′′ ⊗ D h/c ( z )+ h a ( x a ) ⊗ f b ( x b ) ⊗ dˆ σ ′′′ ⊗ ˆ E cF ( z, z ′ ) o , (124)where G F ( x a , x ′ a ) and E F ( x a , x ′ a ) are the quark–gluon correlation functions introduced in Section 3.4,ˆ E cF ( z, z ′ ) is a quark–gluon correlator in the fragmentation process and dˆ σ ′ , dˆ σ ′′ and dˆ σ ′′′ are cross-sections of hard partonic subprocesses. The first term in (124) corresponds to the chirally-even mecha-nism considered by Qiu and Sterman [55]. The second term is the initial-state chirally-odd contributionanalysed in Ref. [263]. The third term is the final-state contribution studied in Ref. [266]. The detailsand the elementary cross-sections can be found in the original papers.48 .4.3 Λ production The origin of the large transverse polarisation of hyperons measured since the 70’s [4, 5] in high-energyunpolarised hadron-hadron scattering is a longstanding problem (for reviews see Refs. [267, 268]). Atransverse-momentum mechanism able to produce sizable asymmetries in A + B → Λ ↑ + X involvesthe polarising fragmentation function D ⊥ T introduced in Section 3.8, which describes the fragmentationof an unpolarised quark in a transversely polarised hadron [174]. In the twist-three approach theΛ polarisation is generated by the chirally-odd spin-independent quark-gluon correlation function E F [209].While in A + B → Λ ↑ + X the Λ polarisation must vanish at pseudorapidity η = 0 for symmetryreasons. in Λ+jet production no such constraint exists. The process A + B → jet+jet+ X → Λ+jet+ X has been proposed as an alternative way of probing D ⊥ T through the correlation between the transversemomentum and spin of the Λ with respect to the dijet axis [269]. We conclude our discussion of the transverse-spin effects in hard processes by listing a series of reactionsthat have been proposed as sources of information on transversity and TMD’s.- Hadron production in transversely polarised lepton-proton scattering: ℓ + p ↑ → h + X . Note thatthe final lepton is not detected, so this process is similar to p + p ↑ → h + X . Its transverse SSAhas been calculated in the twist-three factorisation approach [265] and in the extended partonmodel [270].- Dilepton photoproduction: γ + N ↑ → ℓ + + ℓ − + X . It has been shown [271] that the transverseSSA of this reaction involves the transversity distribution multiplied by the chiral-odd distributionamplitude of the photon.- Exclusive π electroproduction: e + p ↑ → e ′ + π + p ′ . Using a model for the GPD’s and relating H T ( x, ξ, t ) to the tensor charge, the authors of Ref. [272] show that the transverse-spin asymmetryof this process can provide information on the tensor charge δu .- Photo- and electroproduction of two vector mesons: γ ( ∗ ) + N ↑ → ρ + ρ + N ′ . If one of themesons is transversely polarised this reaction probes the transversity GPD H T ( x, ξ, t ) [273]. In the last decade, many transverse-spin effects have been measured in SIDIS on transversely polarisedtargets mainly by the HERMES and COMPASS Collaborations, in hadron-hadron scattering by theRHIC spin experiments, and in unpolarised Drell-Yan processes at Fermilab.In this section, we review some of the recent experimental findings together with their phenomenolog-ical interpretation. The selected data are organised according to the physics information they provide.We start with the measurements aiming to access the transversity distribution, including the relatedmeasurements which are being performed in e + e − collisions. Then we describe the experimental re-sults related to the T -odd TMD’s (Sivers and Boer-Mulders function) and we conclude this part witha brief discussion of some measurements involving the T -even TMD’s and higher-twist PDF’s. TheRHIC hadroproduction results cannot easily be fitted in this scheme, and are presented in a separatesubsection.Since most of the results currently used to access transversity and TMD’s come from the SIDISexperiments, we feel useful to give in section 5.1 some details on the kinematical ranges of the present49IDIS experiments and on the analyses these experiments are doing, which are essentially common toall the SSA’s extraction.Note: Although one should in principle distinguish between the longitudinal momentum fraction x = k + /P + and the Bjorken variable x B = Q / P · q , we have seen that they coincide as far as 1 /Q corrections are neglected. For the sake of simplicity, in this section we ignore this distinction andwrite the distributions as functions of x . Analogously, we take z = P − h /k − to be the the argument offragmentation functions. In the data plots, obviously x and x B always stay for x B and z and z h for z h = P · P h /P · q .Other notations: k T ≡ | k T | is the transverse momentum of the initial quark and p T ≡ | p T | is thetransverse momentum of the produced hadron with respect to the fragmenting quark. In SIDIS P h ⊥ ≡| P h ⊥ | is the transverse momentum of the final hadron with respect to the γ ∗ N axis. In hadroproduction P T ≡ | P T | is the transverse momentum of the final hadron with respect to the collision axis. The SIDIS events are usually identified with standard cuts, with some differences for the differentchannels and for the various experiments due to the different beam energies and thus to the differentkinematical domain. The DIS events are selected requiring the photon virtuality Q to be larger than1 GeV . The fractional energy y transfered from the beam lepton to the virtual photon has to belarger than 0.1, to remove events affected by poor energy resolution, and smaller than 0.9 (or 0.95), toavoid the region most affected by radiative corrections. A minimum value of invariant mass of the finalhadronic state W ≃ W larger than 4, 10 and 25 GeV are required in the data analyses of the JLab, HERMES and COMPASSexperiments respectively.The variables x , Q and W , for the selected events, cover ranges which strongly depend on thelepton beam energy. Fig. 21 shows the regions of the ( x , Q ) plane kinematically accessible with leptonbeams of 160 GeV, 27.5 GeV and 6 GeV momenta, corresponding to the COMPASS, HERMES andJLab experiments respectively. In the COMPASS experiment the x range is between 0.004 and 0.3,where the upper limit is given by the low luminosity; for the HERMES experiment 0 . < x < . x > .
1, in the valence region.The average Q values are also different and there is a strong x − Q correlation. At x ≃ . Q value is 6.4 GeV at COMPASS and about 2.5 GeV at HERMES, while at x ≃ . and 6.2 GeV respectively. In the overlap region the ratio of the Q mean valuesmeasured in the two experiments goes from 2 to 3 with increasing x , in spite of the similar mean valueswhen integrating over the whole x range. The W values are between 25 and 200 GeV for COMPASS,between 10 and 50 GeV for HERMES, and below 10 GeV for the present JLab experiments. Thedifferences in the covered kinematical regions make the experiments complementary, and, all together,they guarantee a very good coverage of the phase space.In addition to the requirements on the inclusive DIS variables, in the data analysis cuts on theenergy final state hadrons are applied, which also depend on the experiment and on the physics channelunder consideration. Particle identification implies momenta above the RICH thresholds, which dependon the detector used in the experiment. In the single hadron analyses, the relative energy z of eachhadron has to be between 0.2 and 0.8. The upper limit, usually not required in the COMPASS analyses,is chosen to reject exclusively produced hadrons. The lower limit is used to select hadrons from thecurrent fragmentation region. To do that, a selection based on the hadron rapidity should be applied,or, equivalently, the so-called Berger criterion should be fulfilled. This criterion [274] has been tunedon unpolarised SIDIS data and allows to relax the request based on the W value alone by asking foreach hadron a large enough z . Thus, if W > . x − Q DIS regions with lepton beams of 160 GeV, 27.5 GeV and 6 GeV momenta (leftto right). At large x , the region is limited by W >
25 GeV or y > . W >
10 GeV and W > for the three beam momenta respectively, while at low x it is limited by the requirement y < . Q − x correlations for W equal to 100, 25 and 4 GeV .Going down to W ≃ z > . W ≃ z > . A f (Φ) XY = F f (Φ) XY /F UU , (125)where f (Φ) is a trigonometric function of a linear combination Φ of φ H and φ S , and XY refer to thebeam and target polarisations (U, L, or T). The asymmetries can be extracted from the measureddistribution of the final state hadrons in the relevant azimuthal angle Φ. In the following, the methodsused to evaluate the transverse SSA’s in the HERMES and in the COMPASS experiments will brieflydescribed.In principle, since all the trigonometric functions appearing in the cross-section are orthogonal,the amplitudes of the azimuthal modulations (the so-called “raw asymmetry” a ) can be obtained astwice the mean value of f (Φ), or by fitting the azimuthal distribution with a function of the type F (Φ) = const · [1 + a · f (Φ)]. In practice this procedure requires to correct the azimuthal distributionfor possible acceptance effects by means of Monte Carlo simulations. This can be avoided in the caseof the SSA’s by collecting data with two opposite spin orientations, indicated with “+” and “-” in thefollowing. If the acceptance and the detector efficiencies are the same for the two sets of data, no MonteCarlo correction is needed when fitting the function F (Φ) on the quantities A (Φ) = N + (Φ) − rN − (Φ) N + (Φ) + rN − (Φ) . (126)Here N ± are the numbers of events in a given Φ bin, r is the normalisation factor between the two setsof data, and Φ is always measured assuming the same orientation of the target polarisation for both sets51f data. In the case of the HERMES experiment, the target spin orientation is flipped every second, sothat acceptance and overall efficiencies can safely be assumed to be the same for the two sets of data. Inthe case of the COMPASS experiment, in which the target polarisation can be reversed typically onlyafter a few days of data taking, the target is divided in cells with opposite polarisation directions. Thisallows to minimise the possible systematic effects due to acceptance variations by using, instead of A (Φ)the so-called “ratio product” quantities [43] which combine the number of events from the different cellsand with the different target polarisation orientation. These quantities do not dependent on the beamflux and on the acceptances, under the assumption that the relative variations are the same for all thetarget cells, and they have a very simple expression in terms of the azimuthal modulation one wants toextract.The methods described above are simple and direct, still some systematic effect can be relevant.In particular, the apparatus acceptance can introduce correlations between the physical asymmetries.For this reason different and more elaborated methods have been developed, which include the binningof the data in the ( φ h , φ S ) plane and the fit with a function which includes all the modulations whichappear in the cross-section. In the most recent analyses, both HERMES [275, 276] and COMPASS [277,278, 279] have introduced “unbinned” maximum-likelihood methods based on maximum-likelihood fitswith the data unbinned in φ h and φ S . The probability distributions include all the expected azimuthalmodulations, both for the spin independent and the transverse target spin dependent parts of thecross-section. The spin independent part turned out not to influence the results for the SSA’s. In theCOMPASS case, the same is true for the acceptance of the apparatus, assumed to have the same relativevariations for all the target cells [278], and the SSA’s obtained with the “unbinned” maximum-likelihoodmethod are in very good agreement with those extracted from the simpler methods used previously.In order to obtain the final results, here called SSA’s, the raw asymmetries obtained with such fitshave to be divided by the target (and beam) polarisation. The result are the HERMES “moments” or“amplitudes”, usually indicated with 2 < f (Φ) > . In the COMPASS data, the raw asymmetries aredivided by the target polarisation, by its dilution factor f , and by the kinematical y -dependent factors,usually dependent on the experimental acceptance in y . As an example, the Collins SSA’s published byCOMPASS are obtained by dividing the raw asymmetries a by the target polarisation, by the dilutionfactor f , and by the mean value of the transverse polarisation transfer from the initial to the finalquark in the elementary lepton-quark scattering D NN = (1 − y ) / (1 − y + y / x , z and P h ⊥ . Since in the experiments there is a strong x − Q correlation, veryfew attempts have been done to measure the Q dependence of the asymmetries in the various x bins,which would be better studied by comparing the results of the different experiments. Usually the SSA’sare measured binning the data alternatively in x , z or P h ⊥ , and integrating on the other two variables.This extraction introduces some correlation between the data, which should be taken into account whenfitting all the results in a global analysis. To avoid this problem, the HERMES Collaborations is doingmulti-dimensional analysis, which are not yet possible in COMPASS, due to the limited statistics. Today, the most direct information on transversity is coming from SIDIS measurements with trans-versely polarised targets, which are complementary to the DY experiments and have the advantage of52llowing a flavor separation by identification of the final state hadrons.Among the various SIDIS observables related to transversity, the measurements performed so farhave provided data on three of them: the Collins asymmetry, the two-hadron asymmetry, and the Λpolarisation. They will be presented in the following subsections, after a a brief description of the eventand hadron selection.
The main source of information on the transversity PDF’s is at present the Collins asymmetry, whichcouples h to the Collins fragmentation function H ⊥ . The Collins asymmetry has been measured bythe HERMES [41, 275] and by the COMPASS [42, 43, 280, 279] Collaborations.Before describing these results, it has to be mentioned that the asymmetries measured by the twoexperiments differ for the already mentioned correction by the D NN factor, applied by COMPASS only,and for the sign because of the different definition of the Collins angle Φ C . In HERMES following theso-called “Trento convention” [184] it is defined as Φ C = φ h + φ S , while in COMPASS the originaldefinition [29] Φ C = φ h + φ S − π is used, as mentioned in Section 4.1.1.The first signal for a non-zero Collins asymmetry came from HERMES in 2005 [41], when the resultson the data collected with the transversely polarised target in 2002 were published. The asymmetryhad values clearly different from zero in the valence region and of opposite sign for positive and negativepions, and this was the first evidence that both the transversity and the Collins FF had to be differentfrom zero. An interesting feature of the HERMES results is that the size of the asymmetry turned outto be roughly the same for positive and negative pions. As suggested in Ref. [41], this result implied thatthe favoured ( u → π + , d → π − ) and the unfavoured ( u → π − , d → π + ) Collins fragmentation functions H ⊥ , fav1 and H ⊥ , unf1 should be of the same size. In fact, neglecting the sea contribution (the asymmetryis different from zero only in the valence region) the flavour structure of the Collins asymmetry for aproton target can be written as A p,π + Coll ∼ e u h u H ⊥ , fav1 + e d h d H ⊥ , unf1 , A p,π − Coll ∼ e u h u H ⊥ , unf1 + e d h d H ⊥ , fav1 . (127)Due to the weight factor given by the quark charge, the measured asymmetries are not sensitive to h d , thus the result | A p,π + Coll | ≃ | A p,π − Coll | implies that H ⊥ , fav1 ≃ − H ⊥ , unf1 . This finding, unexpected at thetime, can be understood [41] in the framework of the string model of fragmentation which in its mostrecent version [177] is described in Section 3.8.1. If a favoured pion forms at the string end createdby the first break, a disfavoured pion from the next break will be opposite in charge and will inherittransverse momentum from the first break in opposite direction from that acquired by the first pion.Also, assuming that the u and d quark contributions add up in the asymmetries, in the same model itis expected that their transversity distributions have opposite sign.The COMPASS experiment started data taking with the transversely polarised deuteron target, andthe first results [42] were published almost at the same time as the HERMES results. The measuredCollins asymmetries were all compatible with zero, both for positive hadrons and for negative hadrons.This result is compatible with the HERMES finding. Limiting again the analysis to the valence region,the deuteron asymmetries can be written as A d,π + Coll ∼ ( h u + h d )( e u H ⊥ , fav1 + e d H ⊥ , unf1 ) , A d,π − Coll ∼ ( h u + h d )( e u H ⊥ , fav1 + e d H ⊥ , unf1 ) . (128)The straightforward conclusion from the COMPASS deuteron measurements is that h u and h d musthave roughly the same size and opposite sign, very much as in the case of the helicity quark distributions.Both HERMES and COMPASS have continued the measurements with the proton and the deuterontargets respectively, and have produced results with considerably better statistics, which have confirmedthe first measurements. 53igure 22: HERMES results for the Collins asymmetry from the 2002-2005 data collected with thetransversely polarised proton target [281]. The asymmetries are shown as function of x , z and P h ⊥ .The left plots show the asymmetries for pions, the right plots the asymmetries for charged kaons.The HERMES results based on the whole data collected from 2002 to 2005 [281] are shown in Fig. 22for pions (left) and charged kaons (right).The applied cuts are Q > , 0 . < y < . W >
10 GeV and all the hadrons with 2 < P h <
15 GeV, 0.2 < z < π + and π − haveopposite sign, increase from very small values at x ≃ .
03 to about 5% at the highest x values, andhave a similar magnitude. The π asymmetries are compatible with zero.The COMPASS results for the Collins asymmetry from all the data collected from 2002 to 2004with the deuteron target are shown in Fig. 23 for charged positive and negative hadrons [43]. The errorbars are statistical only. The systematic errors have been estimated to be negligible with respect tothe statistical precision, and the overall scale uncertainty is 7.3% including the uncertainties on thetarget polarisation and on its dilution factors. Here the DIS events are selected requiring Q > ,0 . < y < . W >
25 GeV . The hadrons used in the analysis have P h ⊥ > z > . z hadron with z > .
25. Also in this case, the asymmetriesturned out to be compatible with zero.Again compatible with zero are the asymmetries measured by COMPASS on deuteron for chargedpions and for kaons. The final results are shown in Fig. 24 as functions of x , z and P h ⊥ for chargedpions (top), charged kaons (middle) and neutral kaons (bottom). They have been obtained using allthe 2002-2004 data for K o and the 2003 and 2004 data for the charged hadrons, since in 2002 the RICHwas not working during the transverse target polarisation data taking. The kinematical cuts are thesame as for the unidentified charged hadrons, plus the requirement to have charged pion and kaonmomenta above 3.1 GeV and 10 GeV respectively, and below 50 GeV. The lower limit is due to the54igure 23: COMPASS results for the Collins asymmetry from the 2002, 2003, and 2004 data collectedwith the transversely polarised deuteron target [43]. The asymmetries are shown as functions of x , z and P h ⊥ for all positive (full circles) and all negative hadrons (open circles). In the plots the opencircles are slightly shifted horizontally with respect to the measured value.RICH threshold and the upper correspond to 1.5 σ mass separation between the two mass hypotheses.The charged pion and kaon asymmetries have been corrected for the purity of the particle identi-fication, which, anyhow, is quite good [280]. The overall systematic errors have been estimated to benegligible with respect to the statistical errors.The quantitative interpretation of SIDIS data on Collins asymmetries and the extraction of thetransversity distributions require external information on the other unknown quantity of the process, theCollins fragmentation function. This has been recently obtained from the inclusive hadron productionin e + e − annihilation, described in the next section. e + e − annihilation The first indication of the Collins effect in e + e − annihilation came from a study of the DELPHI data oncharged hadron production at the Z pole [283], which gave an estimate of about 10 % for the analysingpower h H ⊥ i / h D i , with a considerably uncertainty.More recently, data on azimuthal asymmetries in inclusive production of back-to-back hadrons from e + e − annihilation at s ≃
110 GeV have been presented by the Belle Collaboration [44, 226]. In theiranalysis they use both reconstruction methods described in Section 4.2. The measured quantities are R ≡ N ( φ + φ ) h N i , R ≡ N ( φ ) h N i , (129)where N ( φ + φ ) and N ( φ ) are the numbers of hadron pairs with cos( φ + φ ) and cos 2 φ modulation,respectively, and h N i , h N i are the total average number of pairs. In terms of the asymmetries a and a defined in eqs. 95 and 97, R and R are given by R = 1 + a cos( φ + φ ) , R = 1 + a cos 2 φ . (130)In order to eliminate the contribution of gluon radiation which is insensitive to the charge of the hadronsand the acceptance effects, the ratios of the normalised distributions for unlike-sign (U) hadron pairsover like-sign (L) hadron pairs are taken, R U12 /R L12 and R U0 /R L0 . Focusing on the cos( φ + φ ) modulation,one finds R U12 R L12 ≃ φ + φ ) A UL12 ( z , z ) , (131)55 o ll d A -0.100.10.2 + p - p C o ll d A -0.200.20.4 + K - K x -2 -1 C o ll d A -0.200.2 K z (GeV/c) hT p Figure 24: COMPASS results for the Collins asymmetry for charged pions and kaons, and for neutralkaons from all the data collected with the transversely polarised deuteron target [280]. The full andopen circles refer to positive and negative hadrons respectively. The asymmetries are shown as functionof x , z and P h ⊥ .with the asymmetry parameter A UL12 given by A UL12 = sin θ θ P a e a (cid:16) H ⊥ , fav [1]1 ¯ H ⊥ , fav [1]1 + H ⊥ , unf [1]1 ¯ H ⊥ , unf [1]1 (cid:17)P a e a (cid:16) D ⊥ , fav [0]1 ¯ D ⊥ , fav [0]1 + D ⊥ , unf [0]1 ¯ D ⊥ , unf [0]1 (cid:17) − P a e a H ⊥ , fav [1]1 ¯ H ⊥ , unf [1]1 P a e a D ⊥ , fav [0]1 ¯ D ⊥ , unf [0]1 , (132)where the superscripts “fav” and “unf” denote, as usual, the favored and the unfavored FF’s, respec-tively. Another independent combination of these functions, given by the ratio R U12 /R C12 of unlike-signpairs over all charged (C) pairs, is also determined, following a suggestion of Ref. [284]. This quantitycan be written as in eq. (131) with an asymmetry parameter A UC12 .An analysis similar to the one we have just sketched is performed with the cos 2 φ method, leadingto the ratios R U0 /R L0 and R U0 /R C0 , and to the asymmetry parameters A UL0 and A UC0 . Notice that thecos( φ + φ ) and the cos 2 φ analyses of the same events are not independent, and thus cannot beincluded together in a fit.The Belle results are presented in Fig. 25. A clear rising behaviour of the asymmetries with z and z is visible, suggesting a similar trend for the ratio H ⊥ /D (recall that the FF’s are probed by Belle atthe scale Q = s ≃
110 GeV ). Due to the quadratic nature of the asymmetries in terms of the Collinsfunction, the difference between H ⊥ , fav1 and H ⊥ , unf1 is poorly determined. However, combining the Belleconstraint on the product H ⊥ , fav1 · H ⊥ , unf1 with the SIDIS measurements of the Collins asymmetry, thetwo Collins functions can be separately determined [45, 285, 284, 286].56 A < z < UL A UC < z < z A < z < z < z < A < z < UL A UC < z < z A < z < z < z < Figure 25: A (left panel) and A (right panel) as a function of z for some z bins [226]. The UL dataare represented by triangles and their systematic uncertainty is given by the upper error band. The UCdata are represented by the squares and their systematic uncertainty is given by the lower error band. The first phenomenological analysis of the SIDIS experimental results on the Collins asymmetry wasperformed by Vogelsang and Yuan [287], who assumed Soffer saturation of transversity, i.e., | h | =1 / f + g ), to extract from the first HERMES measurement [41] the favoured and unfavoured Collinsfunctions. Given the poor statistics of those data, the uncertainties on H ⊥ , fav1 and H ⊥ , unf1 were large.Efremov et al.[284, 286] analysed the same data with the transversity distributions taken from thechiral quark-soliton model [288]. The resulting H ⊥ was shown to reproduce satisfactorily also theCOMPASS deuteron data [42, 43], and to be compatible with the Collins function determined from theBelle data [44]. The main finding about H ⊥ supports the HERMES interpretation of their data, namelythat the favoured and unfavoured Collins functions are opposite in sign, and that H ⊥ , unf1 is surprisinglylarge, being comparable in magnitude to H ⊥ , fav1 at the average Q scale (few GeV ) of the HERMESexperiment. Moreover, it explains why the π asymmetry is nearly zero.A combined analysis of the first SIDIS data from HERMES and COMPASS, and of the e + e − Belledata, was performed by Anselmino et al. [45] and led to the first extraction of the u and d -quarkstransversity distributions. This analysis has been updated in Ref. [285] using the preliminary HERMESdata [275], and the COMPASS [280] and Belle [226] published data. The fit does not include K ± and π data, nor the preliminary COMPASS results with the proton target. The TMD’s are writtenas factorised functions of x and k T , and their transverse-momentum dependence is assumed to havea Gaussian form. These two simplifying assumptions are supported by recent lattice studies [289,290]. Fragmentation functions are parametrised in a similar way. Thus the unintegrated unpolarisedquantities f ( x, k T ) and D ( z, p T ) are expressed as f ( x, k T ) = f ( x ) e − k T / h k T i π h k T i , D ( z, p T ) = D ( z ) e − p T / h p T i π h p T i . (133)The resulting average transverse momentum of the hadron is h P h ⊥ ( z ) i = √ π q z h k T i + h p T i . (134)57 d ( x ) T D x u ( x ) T D x ) d ( x , k T D x ) u ( x , k T D x x (GeV)k −0.100.10.20.30.4 −0.100.10.20.30.4 x = 0.1 x = 0.1 Figure 26: The transversity distributions x ∆ T u ≡ xh u and x ∆ T d ≡ xh d at Q = 2 . from the fitof Ref. [285]. The shaded bands represent the uncertainty of the fit. The solid line is the Soffer bound.Also shown are the helicity distributions (dashed curves).The widths h k T i and h p T i are those obtained in [291], namely: h k T i = 0 .
25 GeV and h p T i = 0 .
20 GeV .The parametrisation of the u- and d- transversity distribution and of the Collins function adopted inRefs.[45, 285] is h q ( x, k T ) = N q x a (1 − x ) b [ f q ( x ) + g q ( x )] e − k T / h k T i π h k T i , H ⊥ q ( z, p T ) = N Cq z c (1 − z ) d D ( z ) e − p T /µ C , (135)where N u , N d , N Cfav , N
Cunf , a, b, c, d, µ C are free parameters. Antiquark contributions are ignored. Allthe data used as input of the combined analysis, i.e. the COMPASS deuteron, the HERMES protonand the Belle e + e − data, are very well fitted.Note that since the SIDIS and the e + e − data are taken at very different Q (up to 6 and 20 GeV forHERMES and COMPASS respectively vs. ∼ GeV ), some assumption about the scale dependenceof H ⊥ is required. The simple hypothesis adopted in all the phenomenological analyses is that H ⊥ hasthe same evolution as D , so that the ratio H ⊥ /D is the same at all scales.In Fig. 26 we show the transversity distributions extracted from the SIDIS measurements in Ref. [285].They have opposite sign, with | h d | smaller than | h u | . While the magnitude and the intermediate- x behaviour of h are reasonably well constrained, its high- x tail is not determined by the data. Byintegration of h q , the tensor charges are found to be δu = 0 . +0 . − . and δd = − . +0 . − . at the referencescale Q = 0 . . The value for u is smaller than the predictions of lattice QCD [143] and of mostmodels. However, one should recall that the model scales are very small and usually just guessed, sothe evolution from these scales to a higher Q is affected by large uncertainties [292].A general caveat about the phenomenological analyses of the Collins asymmetry is in order. Theyall ignore the soft factor appearing in the TMD factorisation formulae. At tree level, this factor is equalto 1, but as Q rises it increasingly suppresses the asymmetry [293]. This effect, which is not taken intoaccount in the present fits, leads to underestimate the Collins function extracted from Belle data andconsequently to overestimate the transversity distributions obtained by using that function.Given the relatively large Q values of the COMPASS data in the quark valence region, where theHERMES data showed the largest values of the Collins asymmetry, a comparison of the COMPASSproton data with the HERMES results was regarded as very important to establish the leading twist58 o ll p A -0.0500.050.1 0.2 0.4 0.6 0.8-0.0500.050.1 0.5 1 1.5-0.0500.050.1 + COMPASS protons hAnselmino et al, Nucl. Phys.Proc. Suppl. 191 (2009) 98 x -2 -1 C o ll p A -0.0500.050.1 z hT p - COMPASS protons hAnselmino et al, Nucl. Phys.Proc. Suppl. 191 (2009) 98
Figure 27: COMPASS results for the Collins asymmetry for positive (top) and negative (bottom)hadrons from the data collected with the transversely polarised proton target [279]. The error barsare statistical only. The curves show the calculation by Anselmino et al. based on the global fit ofRef. [285].nature of the effect and to check the robustness of the overall picture and of the phenomenologicalanalysis.In 2008 COMPASS produced the first preliminary results [294] from part of the data collected in2007 with the transversely polarised proton target. The results obtained using all the available statisticshave been published recently [279] and are shown in Fig. 27 for positive (top) and negative (bottom)hadrons. The applied kinematical cuts for DIS event and hadron selection are the same as for thedeuteron data. The systematic uncertainties have been evaluated to be about 0.5 the statistical one,and include a 5% scale uncertainty due to the target polarisation measurement. At small x , in thepreviously unmeasured region, the asymmetries are compatible with zero. At larges x a clear signaldevelops both for positive and negative hadron. The results are compatible with the HERMES resultsin the overlap region, and in very good agreement with the values expected values expected on the basisthe global fit by Anselmino et al. [285], shown by the curves in fig. 27. For the first time, the Collinsasymmetry has been measured to be different from zero at Q ∼
10 GeV, and in perspective these datashould provide information on the Q evolution of the transversity and the Collins function. e + e − annihilation The transverse spin asymmetry in the distribution of the azimuthal plane of hadron pairs in the currentjet of DIS have been measured by HERMES [295] with the proton target and by COMPASS both withthe deuteron [296, 297, 298, 299, 300] and the proton [278] target.Also for this asymmetry there are some differences between the analysis performed by the twoexperiments. The first difference concerns the azimuthal angle of the two-hadron production plane: withreference to the definitions introduced in Section 4.1.5, the angle φ R is used in the HERMES analysis,while φ R is used in the most recent COMPASS analysis (see f.i. Ref. [301]), following the suggestionsof Ref [282, 216]. The two angles, however, coincide in the γ ∗ N system. Also, HERMES measuresthe amplitude of the modulation in the angle φ R + φ S , while COMPASS measures the modulation in φ r + φ S + π , in line with the definition of the Collins angle. Thus the asymmetries measured by the two59 U ⊥ Æ x æ Æ z æ Æ M ppæ [ G e V ] Æ x æ s i n ( f R ⊥ + f S ) s i n q M pp zx . % sca l e un ce r t a i n t y r M pp [ GeV ] x z Figure 28: The two hadron asymmetries versus M h , x , and z measured by the HERMES experiment withthe transversely polarised proton target. The bottom panels show the average values of the variablesthat were integrated over. The bands represent the systematic uncertainty.experiments are expected to have opposite sign.Other differences are in the treatment of the kinematical factor D NN , as in the Collins case, andin the kinematical cuts. Finally, the HERMES extraction of the asymmetries is based on a Legendreexpansion of the dihadron fragmentation functions, as suggested in Ref. [217].In HERMES [295] the events are selected requiring Q > , 0 . < y < . W >
10 GeV and the missing mass larger than 2 GeV to avoid contributions from exclusive two pion production.Also a minimum pion momentum of 1 GeV is required for pion identification. For each event all thepossible combinations π + π − have been used, labelling as 1 the positive particle.If one adopts the partial wave expansion of Ref. [218], a convenient observable is the asymmetry A sin( φ R + φ S ) sin θUT ( x B , y, z h , M h ) ≡ R d cos θ R d φ R R d φ S sin( φ R + φ S ) d σ/ sin θ R d cos θ R d φ R R d φ S d σ = −
12 ˆ D NN ( y ) s − m M h P a e a h a ( x B ) H < ) ,sp a ( z h , M h ) P a e a f a ( x B ) D o a ( z h , M h ) , (136)which selects the interference fragmentation function H < ) ,sp . Thus, in HERMES, the spin asymmetriesdefined in eq. (126) and divided by the target polarisation, have been measured in each ( φ R , θ ′ ) bin,and fitted with the function a sin φ R sin θ ′ / [1 + b (3 cos θ ′ − θ ′ = || θ − π/ | − π/ | , a is thefree parameter and b has been varied to take into account the unknown dependence on cos θ of theunpolarised dihadron fragmentation function. where A sinφ HR sinθU ⊥ = a is the free parameter and b is variedto take into account the unknown dependence on cos θ of the unpolarised dihadron fragmentationfunction. The published asymmetries A sinφ HR sinθU ⊥ are the fitted values of a . The final results from thedata collected from 2002 to 2005 with the proton target [295] are shown versus M h , x and z = z + z .The bottom plots give the average values of the other two variables that were integrated over. Theasymmetries in bins of x and z have been evaluated requiring 0.5 < M h < . b and acceptance effects, are given by theband centred at -0.01. The scale uncertainty due to the target polarisation uncertainty was about 8%.As apparent from the figure, the asymmetries are different from zero, indicating that the spin-dependentpart of the dihadron fragmentation function is different from zero. Also, the data show clear trends ineach of the three kinematical variables, and the signal have the same sign and smaller values than thatof the Collins asymmetry. 60 R S A x z ] [GeV/c inv M COMPASS 2007 transverse proton data p r e l i m i n a r y -2 -1
10 1-0.100.1 pairs - h + h -0.100.1 -0.100.1 Figure 29: The COMPASS proton preliminary results for the two hadron asymmetries versus x , z and M h [278].In the COMPASS analysis, the event selection is similar to that described for the single hadronasymmetries, namely only events with Q > , 0 . < y < . W >
25 GeV are accepted.Only hadrons with z , > . z < . x F > . r T > θ is taken into account. This is justified even in the frameworkof Ref. [217], since in the COMPASS kinematics the sin θ distribution is strongly peaked at one ( < sin θ > = 0 .
94) and the cos θ distribution is symmetric around zero. The asymmetries are extractedfrom the azimuthal distributions using the same methods described for the Collins asymmetry. Inparticular the results from the transversely polarised target data were obtained with the “ratio productmethod”.Preliminary results using the data collected with the deuteron target have been produced looking atdifferent selections for the hadron pair. The asymmetries have been evaluated using all the combinationsof the positive and negative selected hadrons [296]; taking only the two hadrons with higher transversemomentum with different charge combinations [297]; taking only the two hadrons with the highest z i , again for the different charge combinations [297]; taking all the possible combinations of identifiedcharged pions and kaons [299]; taking only the charged pions and kaon with higher z i [300]. All thecorresponding asymmetries turned out to be compatible with zero and no clear signal could be seen.This result could have been expected on the basis of the null result on the measured Collins asymmetry.Also for this SSA there was a strong interest for the COMPASS measurement with the protontarget, since not much variation was expected going from the HERMES to the COMPASS energy. Thepreliminary results from all the data collected in 2007 [278] are shown in Fig. 29 versus x , z , and M h forall the combinations of positive and negative hadrons, selected as in the deuteron case. The systematicuncertainties have been evaluated to be not larger than one half the statistical errors. As can be seen,the asymmetries are clearly different from zero. They have the same sign of the Collins asymmetrieson proton, the behaviour in x is very similar and the absolute values at large x are even larger. Also,there is no clear indication for a structure, in particular as a function of the invariant mass.Very recently, preliminary results on the Artru-Collins asymmetries described in Section 4.2 havebeen produced by BELLE [302]. They show asymmetries different from zero, which depend on z h and M h . Both these results and the COMPASS proton results are fresh, and no attempt to perform globalanalysis has been done yet.On the contrary, The HERMES data on A sin( φ R + φ S ) sin θUT have been recently analysed [230] in termsof the model for D ( z, M h ) and H < ) ,sp developed in Ref. [219]. It turns out that, in order to get a fairdescription of the HERMES asymmetry, using for the transversity distributions the parametrisation61igure 30: COMPASS results for Λ and ¯Λ polarisation versus x from the data collected with thetransversely polarised proton target [304].of Ref. [285], the interference fragmentation function H < ) ,sp of Ref.[219] must be reduced by a factor3. The predicted M h dependence of the analysing power H < ) ,sp /D , with the typical bumps at the ω and ρ masses, is not incompatible with the data, within their large errors. However, the interferencefunction that fits HERMES data largely undershoot the asymmetry measured by COMPASS and, onceevolved to high Q by means of the DGLAP equations for dihadron fragmentation functions [229],yields values of the Artru-Collins asymmetry which are much smaller than those found by Belle [299].Moreover, the invariant mass behaviour predicted in Ref. [230] does not match the Belle findings. Areconsideration of interference fragmentation functions in the light of the recent HERMES, COMPASSand Belle measurements seems to be necessary. Λ polarisation As shown in Section 4.1.4 detecting a transversely polarised spin 1 / h ( x ) H ( z ). The typical example of such processes is Λ (or Λ) production [32]. The Λ polarisation isin fact easily measured by studying the angular distribution of the Λ → pπ decay.From the phenomenological viewpoint, the main problem is that, in order to compute the Λ po-larisation, one needs to know the fragmentation functions H Λ /q ( z ), which are completely unknown.Predictions for P Λ T have been presented by various authors [205, 206, 207] and span a wide range ofvalues. In the calculation of Ref. [205], where the transversity distributions are assumed to saturatethe Soffer bound, P Λ T lies in the interval ±
10% at x ∼ .
1. In particular, in the SU(6) non relativisticmodel, the entire spin of the Λ is carried by the strange quark and one therefore expects a polarisationclose to zero, due to the smallness of h s .The only existing results for this experimentally difficult channel are from the COMPASS exper-iment. On deuteron, the preliminary analysis gave polarisation values for Λ and ¯Λ compatible withzero within the non negligible statistical errors [303]. The same indication comes from the polarisationmeasured with the proton target. Fig. 30 shows Λ and ¯Λ polarisation versus x from part of the 2007proton data [304].As in the case of the Collins function, independent information on H can be obtained from e + e − an-nihilation [27, 220, 228]. The specific process is back-to-back ΛΛ inclusive production, e + e − → Λ Λ X ,with the hyperon and the anti-hyperon decaying into pπ − and ¯ pπ + , respectively. It was shown inRef. [220] that the unpolarised cross section of this process contains an azimuthal modulation propor-tional to H Λ /q H Λ / ¯ q = ( H Λ /q ) , which is selected by the asymmetry between the number of p ¯ p pairs onthe same side of the scattering plane and the number of pairs on opposite sides. The measurement of62uch asymmetry was attempted by the ALEPH Collaboration at LEP [305], but the scarce sensitivityof data did not allow getting any significant result. SIDIS represents, at this moment, the best source of knowledge on the two T -odd distribution functions:the Sivers function, involved in transversely polarised SIDIS, and the Boer-Mulders function, whichgenerates asymmetries in unpolarised SIDIS. Some information on the Boer-Mulders distribution comesalso from unpolarised Drell-Yan processes. DY can probe the Sivers function as well, if one of the twocolliding hadrons is transversely polarised, but this reaction so far has not been experimentally explored. As in the case of the transversity PDF’s, the only measurements which give today a clean access tothe Sivers function are the SSA’s in SIDIS on transversely polarised targets. The relevant quantity isthe so-called Sivers asymmetry A S iv = F sin( φ h − φ S ) UT /F UU , where F sin( φ h − φ S ) UT and F UU are the structurefunctions introduced in eq.(58). This SSA couples f ⊥ T to the unpolarised fragmentation function D .Till now it has been measured only by the COMPASS and the HERMES experiments. The same data asfor the measurement of the Collins asymmetry have been used, and the same kind of analysis, describedin Section 5.1, has been performed. In this case the relevant modulation is that in the azimuthal angle( φ h − φ S ), and, at variance with the Collins case, the Sivers asymmetry is defined in the same way andwith the same sign in the COMPASS and in the HERMES experiments. It has to be noted that in theHERMES papers the Sivers asymmetry is indicated as 2 < sin( φ h − φ S ) > .The first results on the Sivers asymmetry have been produced by the HERMES experiment using thedata collected in 2002 with the transversely polarised proton target [41], and showed large positive valuesfor the π + and K + while for π − and K − the asymmetries were compatible with zero. The preliminaryresults from the 2002-2005 data [275] and the final results published recently [276] confirmed with betterstatistical precision the previous measurement. Fig. 31 shows the final Sivers asymmetries versus x , z and P h ⊥ for pions and charged kaons. The error bars are statistical only. The systematic uncertaintyis shown by the bands and include possible contributions due to the longitudinal component of thetarget spin as well as acceptance and smearing effects, radiative effects, and effects due to the hadronidentification. The additional scale error due to the target polarisation uncertainty is quoted to be7.3%. As can be seen, the π + asymmetry is of the order of 5% almost over all the measured x range,compatible with zero for π − , and slightly positive for π . Slightly positive signals can be seen also for K − , while the values for K + are quite large, up to 10%. The pion results can naively be explained inthe framework of the quark model. With the assumptions used in eq. (127), they can be understoodas due to a the d -quark Sivers function of roughly twice the size of that of the u quark, and of oppositesign. The K results are more difficult to be explained, and further studies on the difference between π + and K + results are quoted in [276]. Also, the asymmetry in the difference of the distributions of π + and π − , which should be more related to the u and d quark Sivers functions in the valence region,has been measured. Finally, a study of the Q dependence has been performed, without finding cleareffects indicating sizable 1 /Q effects. It has to be reminded, however, that the mean values of Q areall in the range between 1 GeV and about 7 GeV .As in the Collins case, COMPASS has measured for the first time the Sivers asymmetry on thedeuteron. First results for charged hadrons from the 2002 data were published in 2005 [42], and later onfinal results from all the collected deuteron data have been produced for charged hadrons [43] and foridentified pions and kaons [280]. The measured Sivers asymmetries for identified hadrons are shown inFig. 32. The errors are statistical only. The quoted systematic errors are negligible with respect to thestatistical ones. All the measured values are compatible with zero within the small statistical errors.63 Æ s i n ( f - f S ) æ U T p + -0.100.1 p -0.0500.05 10 -1 x p - z P h ⊥ [GeV] Æ s i n ( f - f S ) æ U T K + -0.100.1 10 -1 xK - z P h ⊥ [GeV] Figure 31: HERMES results for the Sivers asymmetries on proton from the 2002-2005 data [276] asfunctions of x , z and P h ⊥ , for pions (left) and charged kaons (right).This result for pions can be interpreted naively in the framework of the parton model [43] as due toopposite Sivers functions for the u and d quarks.After the first phenomenological study [108] of preliminary HERMES data on the P h ⊥ –weightedSivers asymmetry [306], various theoretical groups [291, 287, 307, 308] extracted the Sivers distributions(or their moments) from the HERMES measurement of A sin( φ h − φ S ) UT [41]. The fits were then extended[291, 308] to higher-precision HERMES preliminary data [309] and to the COMPASS deuteron data[42]. A comparison of the results of these analyses [310] shows a certain qualitative agreement andsome common features: a negative f ⊥ u T and a positive f ⊥ d T , as predicted by the the impact-parameterapproach [141], with comparable magnitudes, as expected in the large- N c limit [160] or in chiral models[161].More recent fits [110, 111] take into account the new HERMES [276] and COMPASS deuteron data[280]. The surprisingly large values of the K + asymmetry and of the K + − π + difference call for acareful reconsideration of the sea. The Sivers functions are factorised in x and k T , with a Gaussiandependence on k T . Taking as an example the parametrisation of Ref. [110], the functional form of f ⊥ q T is f ⊥ q T ( x, k T ) = N q x α (1 − x ) β f q ( x ) e − k T /µ S , (137)where N q , α q , β and µ S are free parameters, the last two being taken to be the same for all flavors.It turns out that the exponent β governing the high- x tail of the distributions is not well constrainedby the data which extend up to x ∼ .
3. The overall quality of the fit is rather good and the firstmoments of the extracted Sivers functions for the u and d quarks displayed in Fig. 33. As expected, anon negligible strange sea is required to reproduce the K + data. The ¯ u and ¯ d distributions are moreuncertain, even in their sign.The two fits of Ref. [110] and Ref. [111] indicate that the Burkardt sum rule is approximately satu-rated by the quark and antiquark distributions, thus little room is left for the gluon Sivers component64 i v d A -0.100.10.2 + p - p S i v d A -0.200.20.4 + K - K x -2 -1 S i v d A -0.200.2 K z (GeV/c) hT p Figure 32: Final COMPASS results for the Sivers asymmetry on deuteron against x , z and P h ⊥ forcharged pions and kaons [280].and the orbital motion is restricted to valence quarks.In this situation, in which all the existing experimental data on the Sivers asymmetry could beexplained coherently, the COMPASS measurements with the transversely polarised proton target cameas a surprise. The preliminary results for charged hadrons from part of the 2007 data were released in2008 [294] and showed small asymmetries both for positive and negative hadrons, compatible with zerowithin the statistical errors. The analysis of the complete set of data was concluded only recently [279],and the final results are shown in Fig. 34. For positive hadrons the data indicate small positive values,up to about 3% in the valence region. These values are somewhat smaller than but still compatiblewith the ones measured by HERMES at smaller Q . The systematic errors is estimated to be 0.8 σ stat ,plus a ± .
01 systematic uncertainty in the absolute scale due to a systematic difference in the meanvalues of the asymmetry which was found between the first and the second parts of the 2007 run.Given the importance of the Sivers function the COMPASS Collaboration has decided to remeasureSSA’s on NH in 2010 with an improved spectrometer and better statistics. The existence of cos φ h and cos 2 φ h asymmetries in unpolarised SIDIS is experimentally well established.They have been investigated many years ago by the EMC and ZEUS experiments [311, 312, 313] inthe large Q region where they are dominated by perturbative QCD effects. It is indeed known thatthese asymmetries are perturbatively generated by gluon radiation [193, 194, 314, 195, 315, 316] andat high Q and high P h ⊥ they are dominated by these effects. The recent HERMES [317], COMPASS[318] and CLAS [319] results cover a kinematical region (small Q and P h ⊥ < -3 -2 -1 x f T ^ ( ) ( x ) xQ =2.4 GeV updown Figure 33: The first moments of the Sivers u and d quark distributions from the fit of Ref. [110]. x -2 -1 p S i v A -0.100.1 z positive hadronsnegative hadrons ) c (GeV/ hT p Figure 34: COMPASS results for the Sivers asymmetry on proton against x , z and P h ⊥ for positive(closed points) and negative hadrons (open points) [279].contributions.The unpolarised azimuthal asymmetries have been measured by the COMPASS and HERMESCollaborations using part of the data collected with polarised targets and combining them in order tocancel the net target polarisation.The COMPASS preliminary results [318] have been obtained from part of the data collected in 2004with the LiD target polarised both transversely and longitudinally with respect to the muon beamdirection. The data have been combining in such a way to cancel the net target polarisation. To reducethe acceptance effects, only the events with a vertex in the downstream target cell have been used, andthe statistics for each polarisation orientation chosen in such a way to have a zero net polarisation. Theevents are selected requiring the usual cuts Q > , mass of the final hadronic state W > . < y < .
9. For the final state hadrons it is required that 0 . < z < .
85 and 0 . < P h ⊥ < . x , z and P h ⊥ and in each bin the measured φ h distributionhas been corrected for the acceptance of the spectrometer, evaluated with a Monte Carlo simulation.Since the COMPASS beam is longitudinally polarised, a sin φ h modulation is also possible, so thecorrected φ h distributions have been fitted with a function containing the cos φ h , the cos 2 φ h and thesin φ h modulations. This last amplitude turned out to be always compatible with zero. The preliminaryresults for the cos 2 φ h asymmetries are shown in Fig. 35 as functions of x , z and P h ⊥ for positive andnegative hadrons. The measured asymmetries have been corrected for the corresponding y dependentkinematical factor appearing in the cross-section. The errors are statistical only. The systematic errors66 j x -2 -1 ˛ / D f c o s A h-h+ z [GeV/c] T p LiD (part) COMPASS 2004 preliminary
Figure 35: The preliminary results for the cos 2 φ azimuthal asymmetries versus x , z and P h ⊥ forpositive and negative hadrons measured by the COMPASS experiment on LiD [318].are of the order of 2% in both cases and are largely dominated by the acceptance correction. Boththe cos φ h and the cos 2 φ h asymmetries are quite large, with a strong dependence on the kinematicalvariables. Also, for the first time, the asymmetries have been produced separately for positive andnegative hadrons and the measured difference points to different contributions of the u and d quarks tothe underlying mechanisms.The HERMES experiment has produced results for the cos φ h and the cos 2 φ h asymmetries in hydro-gen and deuterium using the data collected in 2000, 2005, and 2006. The cuts applied in the event andhadron selection are: x > . < Q <
20 GeV , 10 < W <
45 GeV , 0 . < y < . x F > . . < z < .
75, and 0 . < P h ⊥ < .
75 GeV. To correct for acceptance of the spectrometer, detectorsmearing, and QED radiative effects, the data were analysed in a 5-dimensional grid in the variables x , y , z , P h ⊥ and φ h . A 10-dimensional smearing matrix was populated by Monte Carlo simulation andincorporated into the fitting procedure, which has been extensively tested with Monte Carlo data. Thereleased asymmetries are one dimensional projections of the asymmetries in which the other four vari-ables have been integrated over. The measured cos φ h asymmetries for protons and deuterons are shownin Fig. 36 for positive and negative hadrons versus x , y , z , and P h ⊥ . The error bars are statistical, x -1
10 1 UU æ ) h f c o s ( Æ -0.3-0.2-0.100.10.2 - h + hHydrogen y z [GeV] h P HERMES Preliminary x -1
10 1 UU æ ) h f c o s ( Æ -0.3-0.2-0.100.10.2 - h + hDeuterium y z [GeV] h P HERMES Preliminary
Figure 36: The preliminary results for the spin-independent cos φ h azimuthal asymmetries versus x , y , z and P h ⊥ for positive and negative hadrons measured by the HERMES experiment with the 28 GeVelectron beam [317] on p (top) and on d (bottom).67 -2 -1 A c o s f x COMPASS Deuteron p - CahnBoer-MuldersCahn+BM-0.15-0.1-0.05 0 0.05 0.1 0.15 A c o s f p + CahnBoer-MuldersCahn+BM p + CahnBoer-MuldersCahn+BM 0.3 0.4 0.5 0.6 0.7z p - CahnBoer-MuldersCahn+BM p + CahnBoer-MuldersCahn+BM 0.1 0.3 0.5 0.7 0.9P (GeV) T p - CahnBoer-MuldersCahn+BM -0.15-0.1-0.05 0 0.05 0.1 0.15 0.1 0.2 0.3 0.4 0.5 A c o s f x HERMES Deuteron p - CahnBoer-MuldersCahn+BM-0.15-0.1-0.05 0 0.05 0.1 0.15 A c o s f p + CahnBoer-MuldersCahn+BM p + CahnBoer-MuldersCahn+BM 0.3 0.4 0.5 0.6 0.7 0.8z p - CahnBoer-MuldersCahn+BM p + CahnBoer-MuldersCahn+BM 0.1 0.3 0.5 0.7P (GeV) T p - CahnBoer-MuldersCahn+BM
Figure 37: The preliminary results for the cos 2 φ spin-independent azimuthal asymmetries for deuteronfrom COMPASS (left panel) and HERMES (right) as functions of x , z and P h ⊥ compared with the fitsto the data of [145].and the bands show the systematic error. As can be seen in the figure, the proton and the deuteronasymmetries are very similar, giving a hint for alike sign for the u and d Boer Mulders functions. Thedifference between positive and negative hadrons is remarkable, as in the case of the COMPASS results.At variance with the COMPASS data shown here, the HERMES results incorporate the y -dependentkinematical factor appearing in the cross section.The CLAS results at JLab for the cos 2 φ h asymmetry for π + [319] agree with the HERMES mea-surements at large z . A striking feature of their data is the large values measured for z < .
2, whichincrease with P h ⊥ and for which there is no comparison with other experiments.Phenomenologically, the Cahn contribution to the cos φ h asymmetry was studied by Anselmino andcoworkers [291]. Using the EMC data [311, 312] in the region P h ⊥ ≤ h k T i = 0 .
25 GeV in the distribution functions, h p T i = 0 .
20 GeV in thefragmentation functions.The cos 2 φ h asymmetry has been analysed in detail in Refs.[320, 145, 321]. In Ref. [320], whichanticipated the measurements, it was predicted that A cos 2 φ h UU is of the order of at most 5%, and that the π − asymmetry should be larger than the π + asymmetry, as a consequence of the Boer-Mulders effect.These predictions have been substantially confirmed by the experimental results. A fit to the HERMESand COMPASS preliminary data has been recently presented in Ref. [145]. It assumes that A cos 2 φ h UU can be described by the leading-twist Boer-Mulders component and by the twist-4 Cahn term (whichis however only part of the full twist-4 contribution, still unknown). The available data do not allowa complete determination of the x and k T dependence of h ⊥ . Thus, the Boer-Mulders functions aresimply taken to be proportional to the Sivers functions of Ref. [110], h ⊥ q = λ q f ⊥ q T , and the parameters λ q are obtained from the fit (the Boer-Mulders sea is not constrained by the data and is taken to beequal in magnitude to the Sivers sea). The result is h ⊥ u = 2 . f ⊥ u T , h ⊥ d = − . f ⊥ d T , (138)and the comparison with the data is shown in Fig. 37 Since f ⊥ u T is negative and f ⊥ d T is positive, the u and d Boer-Mulders distributions are both negative. This is what one expects in large- N c QCD[160] and in some other models [152, 157]. The results are also consistent with the predictions of theimpact-parameter picture [87] combined with lattice calculations [143], which indicate a u componentof h ⊥ larger in magnitude than the corresponding Sivers component, and the d distributions with thesame magnitude and opposite sign. A very recent model calculation of h ⊥ [322] is in good agreementwith the findings of Ref. [145]. 68 .3.3 Boer-Mulders effect in DY production As shown is Section 4.3.1 the cross-section for the DY process on unpolarised nucleon is given by eq. 99where the quantities λ and ν are related by the Lam-Tung relation λ + 2 ν = 1.The NA10 data on π − N DY off a tungsten target show that the angular distribution for the DYevents do not show any c.m. energy dependence nor any nuclear dependence. The data show that thevalue for λ is close to 1, as expected by the naive parton model (for massless quarks and no intrinsicquark momentum the angular distribution should just be (1 + cos θ ). Also, the value for µ is close toexpectation: it is essentially consistent with zero, indicating that the annihilating partons contributeequally to the transverse momentum of the muon-pair. Both the values of λ and of µ are essentiallyindependent of any kinematical variable. But the most striking result from this experiment is the largevalue they obtain for ν and the strong dependence of ν on the dimuon transverse momentum.Similar results have been obtained at FermiLab by the experiment E615 which investigated thesame π − N → µ + µ − X DY process. As shown in Fig. 38 ν was found to be as large as 30%, and steeplyrising with Q T , an effect which is not explained by pQCD and was interpreted as a manifestation of theBoer-Mulders mechanism [240]. Also, by projecting the data points on the x π axis, the pion valencestructure function F π ( x π ) could be precisely determined in the x π range from 0.21 to 1, and found ingood agreement with the extraction of NA3 and NA10. n Q T [GeV] Figure 38: The amplitude ν of the cos 2 φ modulation in π − N Drell-Yan process [323]. The Collins-Soperframe is adopted. The curve is the prediction of Ref. [240].Recently, the E866/NuSea Collaboration at FNAL has presented data on the angular distributionsof dimuon production in pd [324] and pp [70] collisions. Much smaller ν values (less than 0.05) than in πN DY are found, as shown in Fig. 39, and the Lam-Tung relation is satisfied. These data could inprinciple give some information about the antiquark Boer-Mulders distributions [325, 326]. However,in the P T region above 1-1.5 GeV they are expected to be described by pQCD [238]. Thus the onlypoints that have likely to do with the Boer-Mulders effect are those below P T ∼ . P T in the context of twist-three factorisation has been performed in Ref. [251].69 p T (GeV/c) n p - + W at 194 GeV/c p - + W at 252 GeV/cp + d at 800 GeV/c -0.1-0.0500.050.10.150.20.250.3 0 0.5 1 1.5 2 2.5 3 3.5 4 p T (GeV/c) n p + d at 800 GeV/cp + p at 800 GeV/c Figure 39: Left: the ν coefficient in the Collins-Soper frame for three DY measurements (dots:E866/NuSea [324]; squares: E615 [69]; triangles: NA10 [323]). The curves are fits similar to thatof Ref. [240]. Right: the ν coefficient for pd [324] and pd DY [70]. The dot-dashed curve is the pertur-bative QCD contribution. The solid and dotted curves are the calculations of Ref. [325] for pp and pd ,respectively, based on the Boer-Mulders effect. T -even functions andhigher-twist functions At leading twist, besides the Collins, Sivers and Boer–Mulders terms, there are other three angularmodulations in the SIDIS cross section, which probe T -even TMD distributions:- Unpolarised beam and longitudinally polarised target: sin 2 φ h modulation, involving h ⊥ L .- Unpolarised beam and transversely polarised target: sin(3 φ h − φ S ) modulation, involving h ⊥ T .- Longitudinally polarised beam and transversely polarised target: cos( φ h − φ S ) modulation, in-volving g T .From the data collected with the deuteron target, the COMPASS experiment has produced prelim-inary results on these asymmetries for charged hadrons [327, 328]. All in all, it is hard to find a signalin any of these observables. It will be interesting to look at the corresponding results with the protontarget.The HERMES collaboration has measured the sin 2 φ h moment, both with a proton [329] and adeuteron target [330], finding it to be compatible with zero. Signals of a non vanishing sin 2 φ h asymmetryhave been recently reported by the CLAS collaboration [331].For the moment, all these data can only be confronted with model predictions. Focusing on h ⊥ T ,which has attracted some attention for its interesting physical content, a calculation of A sin(3 φ h − φ S ) UT based on positivity bounds [121] gives limits of about ± .
03 for the asymmetry on a deuteron targetand a slightly larger value for a proton target. The COMPASS deuteron data lie within these bounds.The light-cone constituent quark model of Ref. [165] predicts an asymmetry A sin(3 φ h − φ S ) UT smaller than0.01, also consistent with the COMPASS findings. All the asymmetries related to T -even TMD’s arecalculated in Ref. [165] and found to be generally close to zero, hence compatible with the COMPASSfindings. 70t subleading twist, that is at order 1 /Q , the situation is much more involved. The SIDIS structurefunctions of eq. (58) have have in fact the general form: F ∼ l . t . TMD ⊗ h . t . FF + h . t . TMD ⊗ l . t . FFwhere “l.t.” = leading twist, and “h.t” = higher twist. They contain both leading-twist and twist-threeTMD distributions and fragmentation functions. Thus, the phenomenological interpretation of theseobservables is rather intricate. In SIDIS with unpolarised (U) and/or longitudinally (L) and transversely(T) polarised beams and targets there are 8 twist-three modulations:
U U : cos φ h ; LU : sin φ h ; U L : sin φ h , sin 2 φ h ; LL : cos φ h ; U T : sin φ S ; U T : sin(2 φ h − φ S ); LT : cos φ S , cos(2 φ h − φ S ) . COMPASS has measured all these quantities on a deuteron target [327, 318, 328] and found them allto be consistent with zero. HERMES presented results on the sin 2 φ h and sin φ h modulations witha longitudinally polarised proton [329] and deuteron target [330]. While the sin 2 φ h asymmetry wasfound to vanish, the sin φ h asymmetry showed a large signal (up to 4% for proton and 2% in deuteron),incompatible with zero for positive and neutral pions, originally interpreted in terms of large transversityPDF’s.A remark about the definition of the target polarisation is now in order. Experimentally, the targetpolarisation is defined with respect to the a longitudinal (transverse) polarisation with respect to thebeam axis has a transverse (longitudinal) component with respect to the virtual photon axis, whichis kinematically suppressed by a factor 1 /Q [112]. Thus, any measured w ( φ h , φ S ) modulation with a“longitudinally” (“transversely”) polarised target is mixed with a transverse (longitudinal) modulationof the same type, suppressed by 1 /Q . This may be relevant in some cases. For instance, the sin φ h asymmetry measured with a target longitudinally polarised with respect to the beam axis gets a twist-three contribution from A sin φ h UL , but receives also contributions from the leading-twist Collins and Siversasymmetries, A sin( φ h + φ S ) UT and A sin( φ h − φ S ) UT , which are multiplied by a kinematical factor ∼ /Q . In prin-ciple, all these contributions might be equally relevant. The HERMES analysis of the sin φ h asymmetry[332] has shown that the Collins and Sivers contributions to h sin φ h i are small, and therefore the largemeasured asymmetry is a genuine subleading-twist effect.Another indication of the relevance of twist-three effects comes from the beam-spin asymmetry A sin φ h LU , which has been measured by CLAS at 4.3 GeV for positive pions [333] and by HERMES at 27.6GeV for charged and neutral pions [334]. The asymmetry for positive pions is large and positive. TheCLAS and HERMES results nicely agree if one rescales the HERMES data by the mean value of Q and the y dependent kinematical factor, as shown in Fig. 40. Sizable SSA’s in polarised inclusive hadroproduction have been reported since the 70s [1, 2, 3, 335],for center-of mass energies in the range 5-10 GeV. In the same years, Fermilab experiments discoveredthat Λ hyperons produced in unpolarised pp collisions have a large transverse polarisation with respectto the production plane [4, 5]. These findings provoked a certain theoretical interest, as it was widelyheld that large transverse polarisation effects could not be reproduced in the framework of perturbativeQCD [16]. On the other hand, the small values of √ s and P T explored by those experiments left thedoor open to interpretations based on soft physics.In 1991 the E581/E704 experiment at Fermilab extended the investigation of transversely polarisedhadroproduction to higher energies and found remarkably large transverse SSA’s in the forward region[8, 9]. These results were confirmed by the RHIC measurements, which moved the energy frontier71igure 40: Comparison of the kinematically rescaled sin φ h asymmetries between the HERMES (circles)and CLAS (triangles) measurements. The full square represents a previous HERMES measurementaveraged over the range 0 . < z < .
7. The outer error bars represent the quadratic sum of thesystematic uncertainty and the statistical uncertainty (inner error bars) [334]..one order of magnitude upwards. On the Λ polarisation front, the striking effect discovered in theearly experiments was observed also at higher P T values [336, 337, 338]. Thus, it has become clearthat there must be some hard mechanism behind the transverse polarisation phenomena observed inhadroproduction. As seen in Section 5.1, the left-right asymmetries associated to an azimuthal modulation of a cross-section, are best measured by comparing data taken with up- and down-polarised beam or target.Integrating over φ , a global left-right asymmetry A N can then be defined as A N = 1 P N L − N R N L + N R , (139)with P the beam or target polarisation.The modern era of experimental (and theoretical) work on SSA’s in hadroproduction was inauguratedby the E704 investigation of pN and ¯ pN collisions with transversely polarised secondary proton andantiproton beams at the c.m. energy √ s ≃ . . ≤ x F ≤ .
6, with P T in the range 0 . − . | x F | ≤ .
15, with P T up to 4 GeV. In pion production, large SSA’s were foundat high x F [8, 7, 9, 10]: the asymmetries are nearly zero up to x F ∼ . x F ,reaching 15 % for π and 30-40 % for π ± . In p ↑ N collisions A N is positive for π + and negative for π − ,with about the same size. Signs are reversed in ¯ p ↑ N scattering. The asymmetry for π is roughly half ofthat of charged pions and always positive. As for the P T dependence, A N is zero below P T ∼ . P T . In the central rapidity region, where larger valuesof P T are reached, the asymmetries turn out to be consistent with zero [339]. E704 has also measuredsizable SSA’s in inclusive Λ and η production [340, 341] but the rather low transverse momentum ofthe Λ’s does not allow a safe perturbative QCD analysis.72he E704 findings have been substantially confirmed by other fixed-target experiments at lowerenergies, at IHEP (Protvino) [342, 343] and at the BNL–AGS [344, 345].On the phenomenological side,the E704 pion asymmetries have been interpreted in terms of the Siverseffect [346, 39, 257] and of the Collins effect [176, 40, 347]. A recent reassessment of the situation [348]has shown that, contrary to a previous prediction of a strong suppression of the Collins effect [349], bothCollins and Sivers mechanisms may give sizable contributions to the SSA’s. The E704 results have alsobeen studied in the context of twist-3 factorisation, considering quark-gluon correlations in the initialstate [53, 55, 263, 264, 350] and in the final state [265]. All these approaches are able to reproduce atleast qualitatively the data, thus showing that many different physical mechanisms may be at work inpolarised hadroproduction. Thus, it is impossible for the moment to draw definite conclusions as to thedynamical source of single-spin transverse asymmetries.The E704 measurement might have left the doubt that transverse SSA’s would disappear at colliderenergies. Studying the p ↑ p → π X reaction at √ s = 200 GeV in the first polarised collisions at RHIC,the STAR Collaboration showed that this is not the case: the large effects found by E704 persist atan order of magnitude higher energy [11]. As shown in Fig. 41 STAR measured a large positive A N above x F ∼ . . < h P T i < . x F region has been explored, finding an asymmetry consistent with zero, and the P T dependence ofthe SSA has been determined [351]. In Fig. 42 one sees that the rise of the SSA’s at large x F is fairlywell reproduced by the Sivers mechanism in the generalised parton model with the Sivers functionextracted from the HERMES SIDIS data [257, 352] and by the twist-3 factorisation scheme [350]. Onthe contrary the P T behaviour of the data, showing a clear tendency to increase at fixed x F , contradictsthe theoretical expectations, which predict a decrease of A N with P T .Concerning the description of hadroproduction SSA’s in terms of TMD’s taken from SIDIS analysis,one should recall that RHIC asymmetries scan the parton distributions over a wide range of the Bjorkenvariable, including the large- x region, whereas the SIDIS data are limited to x < . x F [353]. -0.20.00.20.40 0.2 0.4 0.6 0.8Total energy p mesons x F A N ( A ss u m i ng A CN I = . ) CollinsInitial state twist-3Final state twist-3Sivers Æ p T æ = 1.0 1.1 1.3 1.5 1.8 2.1 2.4 GeV/c N Figure 41: The asymmetry A N ( p ↑ p → π X ) measured by STAR [11] at √ s = 200 GeV. The curves arethe predictions of Ref. [40] (solid), Ref. [39] (dotted), Ref. [55] (dot-dashed), Ref. [265] (dashed).Measurements of forward charged pions by BRAHMS at √ s = 200 GeV [354] and √ s = 62 . π + A N ( p ↑ p → π X ) measured by STAR [351] at √ s = 200 GeV, as a functionof x F for two different values of the average pseudorapidity h η i (left) and the same asymmetry as afunction of P T at fixed x F (right). The curves are the predictions of Ref. [257, 352] (dashed line) andof Ref. [350] (solid line).and π − and large absolute values of A N . The SSA’s for π + and π − at √ s = 62 . P T in Fig. 44. A clear rise with P T is visible up to a transverse momentum of about 1 GeV,where the asymmetries reach magnitudes of about 0.3. BRAHMS has also measured kaon production F x ) p ( N A -0.2-0.100.10.2 )<0.8 GeV/c p ( T + p E704 - p Figure 43: Charged pion asymmetries measured by BRAHMS at √ s = 200 GeV [354] and √ s = 62 . √ s = 19 . √ s = 62 . K + asymmetry has the same sign and approximately the same magnitudeas the K − asymmetry [355]. The results are fairly well reproduced both by the twist-3 approach andby the Sivers effect in the generalised parton modelIn the midrapidity region ( | η | < . √ s = 200 GeV [14] at √ s = 200 GeV, showed transverse SSA’s consistent withzero. This result is in agreement with the fixed-target E704 finding and extends it to higher P T , up to5 GeV. 74 N A -0.4-0.200.20.4 + p - p (a) )<0.5 GeV/c p ( T (b) )<0.6 GeV/c p ( T F x (c) )<0.8 GeV/c p ( T (d) )<1.0 GeV/c p ( T (e) )<1.2 GeV/c p ( T Figure 44: Transverse SSA of charged pions at √ s = 62 . P T , measured by BRAHMS[355]. The QCD-based descriptions of the E704 and RHIC results on hadroproduction SSA’s have been criti-cised [356] on the ground that, whereas the spin-averaged cross sections at √ s = 200 GeV [12, 357, 15]are well described by next-to-leading order perturbative QCD [358], the cross sections at lower energies(for instance at √ s = 19 . x T ≡ P T / √ s → In this section account is given of near future and more distant future SIDIS and DY experiments. Themain goal of the SIDIS experiments is to measure the SSA’s over a detailed grid of the kinematic variables x , P h ⊥ , and z , facilitating the extraction of the DF’s and of the FF’s. In addition, it should be possibleto study sub-leading twist effects by probing their 1 /Q dependence, and to explore the transition fromnon-perturbative small transverse momentum, typically lower than 1 GeV, to the transverse momentumlarge regime. In the case of the DY measurements, the main goal is to perform for the first timeexperiments with polarised nucleon targets and/or polarised beams, to test the predicted test of signof the T-odd DF’s. COMPASS at CERN:
The analysis of the 2007 transversely polarised proton data is still ongoing,and several results on SSA’s have still to be obtained. Many more data on the same target (NH ) willbe collected in the long 2010 run, so that in the near future a large amount of data is expected.For what concerns a more distant future, the COMPASS Collaboration is presently preparing a pro-posal for measurements aiming to study chiral perturbation theory, generalised parton distributions viaDeeply Virtual Compton Scattering (DVCS), and TMD parton distributions via Drell-Yan processes.75he DVCS measurements will be performed using a liquid hydrogen target and a 190 GeV muon beam.In parallel with these measurements, SIDIS data will be collected to extract with high precision theunpolarised cos φ h and cos 2 φ h azimuthal asymmetries as well as the beam helicity dependent sin φ h asymmetry. Such information cannot be extracted from the data collected with the transversely po-larised target because of the complications of using a nuclear target. JLab experiments:
In the near future also the CLAS Collaboration in Hall B will take SIDIS dataon a transversely polarised target. For transverse running, the use of a novel HD-ice target is planned,which in a frozen-spin state requires only small holding fields. The use of the HD-ice target by theE08-015 Collaboration has many advantages: being a solid target, it can be short, a few cm, and thanksto the smallness of the holding field it can be located in the centre of the detector, thus increasing theacceptance of the spectrometer. In addition, the HD target has almost no dilution, which maximisesthe figure-of-merit, and being of low atomic number, comparatively few bremsstrahlung photons will beproduced in the target. The experiment should run in the second half of 2011, and an upgraded versionof the detector has already been proposed for JLab12.In a medium term range, Jlab12 GeV upgrade could meet the requirements to study TMD’s in the va-lence region, thus covering a complementary kinematic region with respect to COMPASS. The Clas12experiment in Hall-B is designed to achieve a very broad kinematic coverage while increasing by a factor10 the luminosity with respect to the current 6 GeV setup. In particular, the forward spectrometercomprises a 2 T toroid with improved geometry to minimise the not-active azimuthal coverage anda RICH detector is under study to extend the hadron identification over the full energy range of theexperiment. The spectrometer is complemented by a central detector embedded in a 5 T solenoid.Also, an upgraded version of experiment E06-010 (PR09-018) has already been proposed and condi-tionally approved to run in Hall A. The experiment aims to measure the SSA’s of the SIDIS process e + n → e ′ hX , where h is either a π or a K . The experiment will use the large-solid-angle Super BigBiteSpectrometer as hadron arm, the BigBite Spectrometer as electron arm, and a novel polarised Hetarget that includes alkali-hybrid optical pumping and convection flow to achieve very high luminosity.Thanks to the large acceptances of the electron and hadron arms, an electron - polarised nucleon lumi-nosity at the level of 4 × cm − s − , and a target polarisation of 65%, the experiment should collectin a two-month run about 100 times more statistics than that obtained by the past experiments. e − N and e − A future colliders: In depth studies of hadron structure can be best performed ata high energy polarised electron-polarised proton collider. Large Collaborations at BNL and JLab areelaborating proposals which are well advanced and are being encouraged by the USA agencies. Aswritten in the NSAC 2007 Long Range Plan, ”the allocation of resources are recommended to developaccelerator and detector technology necessary to lay the foundation for a polarised Electron Ion Collider(EIC). The EIC would explore the new QCD frontier of strong color fields in nuclei and precisely imagethe gluons in the proton”. To carry out a rich and diversified physics program the recommended energiesfor the electrons are between 3 and 10 GeV, for the protons between 25 and 250 GeV, and for the heavyions between 25 and 100 GeV. The luminosity in the case of the e − p collider should be 10 − cm − s − , i.e. about 100 times the luminosity of the HERA collider. Recently preliminary ideas for apolarised electron-nucleon collider (ENC) at GSI, Darmstadt, have been discussed mostly amongst theGerman community.The advantage of the Collider configuration over fixed target experiments are manifold:- it provides a large range of Q , x , W and P h ⊥ .- the figure of merit for asymmetry measurements is very much better. For ammonia (NH ) f ≃ . ≃
50. Needless tosay, the comparison is done assuming the same number of collected events, so a high luminosity for theEIC is a prerequisite.- It provides access to the interaction region, so that modern vertex tracking systems can identify shortliving particles, like D produced in the interaction.76aving access to the interaction region, exclusive reactions are at reach. This opens up the whole fieldof GPDs, which to-day are the only way to quantify how the orbital motion of quarks in the nucleoncontributes to the nucleon spin. Also, it allows measurements in the target fragmentation region,which is presently poorly known due to the difficulty of measuring slowly moving hadrons in fixed-target experiments, opening a window to the study of spin-independent and spin-dependent ”fracture”functions.The US groups of RHIC and JLab are proceeding jointly to the formulation of two different proposalsfor two different colliders, eRHIC and ELIC, based in the two different laboratories. The RHIC projectclearly foresees the use of the highly polarised proton and nuclear existing beams. Two acceleratordesign options are being worked upon, both aiming at high brightness 10 GeV electron beams. ARing-Ring option, which requires a new electron storage ring for polarised electron or positron beams,is technologically more mature, and could provide a peak luminosity of 0 . × cm − s − . The secondoption is a Linac-Ring option, which offers higher luminosity (by a factor of 5) and possibly higherenergy, but requires intensive R&D for the high-current polarised electron source.The starting point of the JLab project is the availability of the 12 GeV electron beam from the upgradedCEBAF. The proton complex has to be built from scratch, so it is being designed taking full advantageof the expertise matured at RHIC and other laboratories on acceleration and storage of polarised protonbeams. The design goal for the collider luminosity is very ambitious, 3 × cm − s − for beam energiesof 10 and 250 GeV for the electron and protons respectively.Quite recently, in the summer of 2008, discussions started , about a possible low-cost realisation of anENC at GSI. The central idea is to use the 15 GeV high energy storage ring HESR, which is plannedto store an antiproton beam for the PANDA experiment (and possibly PAX) as the ring where to storethe polarised proton beam. By constructing a 3-3.5 GeV electron ring, a ”low energy” ENC could berealised. The cm energy would be 14 GeV, i.e. in between the HERMES and the COMPASS energies.To inject polarised protons in HESR a new 70 MeV p -linac will be needed. The protons would then beinjected into the existing SIS18 ring, accelerated up to 1.4 GeV, and transferred then into HESR. Newhardware for the spin manipulations will be needed in SIS 18 and in HESR, but it is the same which willbe necessary for the PAX experiment. The electron complex has to be constructed from scratch. An e -linac and an electron synchrotron will accelerate the electron beam, which will be stored into a newstorage ring of about the same length as HESR, and housed in the same tunnel. Preliminary machinestudies indicate that a luminosity of at least 10 could be achieved, as well as large polarisations ( 80%)for the two beams. To further reduce the cost of the project, it is proposed to use the PANDA detector,and to operate the collider in time sharing with the PANDA Collaboration. As described in section 4.3, the DY process in transversely polarised hadron scattering is theoreticallya very clean and safe way to access transversity. The original suggestion of measuring DY in p ↑ p ↑ scattering, which could be done at RHIC, turned out to be difficult because of the small value expectedfor the asymmetry,of the order of 1 −
2% [83, 245, 246]. The measurement will also require external inputto disentangle the quark and the antiquark distributions. It will be done when the RHIC luminositywill be increased.These problems can be circumvented by studying DY production with polarised antiprotons atmoderate energies, which is the ideal process to observe a sizable double transverse asymmetry [83, 362,363, 247], dominated by the valence distributions. Such a measurement has been proposed by the PAXCollaboration [232] at the FAIR complex to be built at GSI. Since the production rate for relativelylarge dilepton masses M ( > A DYT T , it has been Private communication from D. von Harrach. A N s i n ( fg - f S ) x F COMPASS: p p › p =160 GeV GRVPI p - p + A TT ( y ) y s=30 GeV s=45 GeV s=80 GeV s=200 GeV Figure 45: Left: the prediction of Ref. [364] for the Drell-Yan Sivers asymmetry to be measured byCOMPASS in πp collisions. Right: the double spin asymmetry integrated over M in transverselypolarised proton-antiproton DY for different values of √ s [247].proposed [362] to exploit the J/ψ peak to measure the asymmetry.Finally, we recall that the Sivers effect can also be observed in Drell-Yan processes with a transverselypolarised proton, where it gives rise to a sin( φ γ − φ S ) asymmetry. No measurement has been made sofar, but many experimental collaborations worldwide plan to investigate this class of reactions in thenear future. In the following we describe very briefly the DY proposed experiments.
COMPASS:
Among the proposed measurements for a second phase of the COMPASS experimentan important issue is the possibility to investigate for the first time a π − induced DY process on atransversely polarised proton target. The high mass of the COMPASS target (about 1 kg of NH )and the excellent performance of the COMPASS spectrometer make this measurement feasible, and thenumber of events collected in two years of running would allow to check the expected change of signof the Sivers function. Assuming for the magnitude of the Sivers function the value extracted from theHERMES measurements in SIDIS (see Fig. 45, left), the significance of the measurement is expectedto be 3 to 4 σ . PAX and PANDA at GSI:
As already mentioned, the PAX Collaboration has proposed to measureDY processes in ¯ pp scattering at FAIR. An asymmetric collider is proposed, consisting of HESR, wherepolarised protons will be stored, and of a new storage ring for the polarised antiprotons, which could bethe existing COSY Storage Ring, suitably modified. The predictions for A DYT T in ¯ p ↑ p ↑ collisions at GSI-FAIR are shown in Fig. 45 (right). One sees that asymmetries of the order of 10-15 % can be expectedat PAX. Polarisation of the stored antiproton beam will be done using the “spin filtering” technique[365]. The antiprotons beam traverses a polarised proton storage cell, and a beam polarisation buildsup by repeatedly passing through the cell as long as the cross section for parallel spin is different fromthat of antiparallel spins. The method has been proven to work for proton beams, but in the caseof antiprotons the spin dependent cross-sections are not known and corresponding measurements havebeen proposed at the Antiproton Decelerator at CERN.The PANDA collaboration also envisage the measurement of DY process in ¯ pp scattering, where bothparticles are unpolarised. The option to put a transversely polarised proton target in the PANDAdetector is very interesting, but it is technically very difficult and has presently been discarded.78 ermilab Experiment E906: This experiment is scheduled to run in 2010 for 2 years of data col-lection. It will extend DY measurements of E866 (which were done with 800 GeV protons) using anupgraded spectrometer and the 120 GeV proton beam from the main injector. The use of the lowerbeam energy gives a factor 50 improvement of luminosity with respect to E866. To cut down costs, itwill use many components from E866, and data will be taken with Hydrogen, Deuterium and NuclearTargets. The main goal of the experiment is the study of the structure of the nucleon, in particular the¯ d/ ¯ u ratio at high x .In the future, there are plans to measure SSA’s on a transversely polarised target, and check the changeof sign of the Sivers function with respect to SIDIS. J–PARC:
Two proposals for DY experiments have been submitted:- P04: measurement of high-mass dimuon production at the 50 GeV proton synchrotron;- P24: polarised proton acceleration.The advantage of an experiment at JAPRC is the high proton beam intensity, and consequently thehigh luminosity. The disadvantage is that at the same invariant mass the cross section is smaller atlower energy. The transverse polarisation program is best carried on by P24, but is clearly conditionedby the realisation of the polarised proton beam, which is not yet approved.
STAR and PHENIX at RHIC:
According to the present accelerator schedule, which foresees a longlongitudinal run for W-physics, a DY program with transverse spin at PHENIX and STAR will notstart before 2015. However, since in the intersection regions IP-2 and IP-10, where the PHOBOS andthe BRAHMS experiments were installed, there are no spin rotators, the polarisation of the beams isalways transverse and ideas are being put forward to prepare both a collider experiment and a fixedtarget experiment for DY measurements.
RHIC internal target:
Quite recently ideas have been put forward for a DY experiment at RHICscattering one beam off an internal target. With the 250 GeV beam, the kinematic range exploredwould be x = 0 . − . x = 0 . − . NICA:
To investigate the hadron structure a Nuclotron-based Ion Collider fAcility (NICA) is beingplanned at the JINR in Dubna, based on the existing proton synchrotron Nuclotron. The acceleratorcomplex will require new ion and polarised proton sources, a new linear accelerator, a new boostersynchrotron and the two new superconducting storage rings of the collider. Both polarised proton andpolarised deuteron beams should be available in the two rings. The main physics objectives will be thestudy of elastic processes and of Drell-Yan processes.
The relevant parameters for all these projects are summarised in Table 1.Theoretical predictions for COMPASS ( πp ↑ ), PAX (¯ pp ↑ ), RHIC ( p ↑ p at √ s = 200 GeV) and J–PARC ( p ↑ p at √ s ≃
10 GeV) have been presented by various authors [108, 366, 367, 368, 369, 370, 364,371, 372]. The conclusion one can draw from these analyses is that the future experiments will largelybe complementary to each other.As for p ↑ p DY, the RHIC data in the negative x F region will probe the contribution of the sea Siversfunction, while experiments at lower energies, like J–PARC (operating at √ s ≃
10 GeV) will provideinformation on the large- x behaviour of f ⊥ T . A comprehensive discussion of all future DY measurementsof Sivers asymmetries can be found in Ref. [364]. 79xperiment particles energy √ s x or x luminosityCOMPASS π ± + p ↑
160 GeV 17.4 GeV x = 0 . − . × cm − s − PAX p ↑ + ¯ p collider 14 GeV x = 0 . − . × cm − s − PANDA ¯ p + p ↑
15 GeV 5.5 GeV x = 0 . − . × cm − s − J–PARC p ↑ + p
50 GeV 10 GeV x = 0 . − . cm − s − NICA p ↑ + p collider 20 GeV x = 0 . − . cm − s − RHIC p ↑ + p collider 500 GeV x = 0 . − . × cm − s − RHIC IT phase 1 p ↑ + p
250 GeV 22 GeV x = 0 . − . × cm − s − RHIC IT phase 2 p ↑ + p
250 GeV 22 GeV x = 0 . − . × cm − s − Table 1: Compilation of the relevant parameters for the future planned DY experiments. For RHIC,IT stays for Internal Target
The original finding of the EMC collaboration, that the quark spin does not account for the total spin ofthe proton, has been a strong motivation for in-depth studies of the QCD structure of the nucleon and fora new generation of experimental investigation of hard scattering processes on polarised nucleons. Thegrowing interest in the contribution of the quark and gluon orbital angular momentum to the nucleonspin naturally led to an increased attention to transverse spin and transverse momentum phenomena.In this context, the most important experimental finding has been the discovery that there is acorrelation between the spin of a transversely polarised quarks and the P T of the hadrons created inthe quark hadronisation process. Convincing evidence for this correlation has been provided by bothSIDIS processes on transversely polarised nucleons and high energy e + e − annihilations into hadrons.Thanks to this correlation, it is now possible to measure the transversity distribution function. Globalanalysis of the existing SIDIS data and of the e + e − data have already provided first rough informationof this two new function.A second important discovery of the recent years is that there is also a non-zero correlation betweenthe spin of a transversely polarised nucleon and the intrinsic transverse momentum of the quarks.In polarised proton-proton scattering the most impressive result in transverse spin physics is theconfirmation that large SSA’s for inclusive mesons production persist at centre of mass energies which bynow are more than one order of magnitude greater than those of the previous fixed target experiments.On the theoretical side, the main achievement has been the discovery that the Wilson line structureof parton distributions, which is necessary to enforce gauge invariance, has also striking observableconsequences, allowing for single-spin asymmetries that would otherwise be forbidden by time-reversalsymmetry. At leading order the most general descriptions of SIDIS and Drell-Yan processes have beenrevisited and a number of structure functions have been introduced to take into account all possiblecorrelations among the transverse momentum and spin of the quarks and the spin of the nucleon. ManyQCD studies have been performed to understand the properties and the gauge structure of these unin-tegrated distribution functions, which have been named transverse momentum dependent distributionfunctions. Non-collinear factorisation schemes have been developed and extended to polarised pro-cesses. In an alternative approach, twist-three effects have been evaluated and compared to the TMDdescription in the intermediate P T region. A third line of attack which is vigorously being pursued isQCD computation on the lattice. Recent refined lattice QCD results shed light on the fine structureof TMD’s. This approach is particularly interesting for transversity, because the tensor charge, thefirst moment of the transversity distribution, which is an all-valence object, is believed to have beenevaluated with good accuracy on the lattice, thus a good measurement of this quantity could provide agood test of the correctness of the calculation. 80n spite of these achievements, the amount of work which is still needed is not small.On the experimental side, the available SIDIS data are only a glimpse to a new territory, and manymore data are needed to obtain the P T and Q dependence of the asymmetries in the different x-bins, aprerequisite to a model independent extraction of the TMD functions. The situation is worse howeverin the DY sector, since no polarised DY data exist at all. The existing unpolarised DY data allow onlyto access the Boer-Mulders function, but the full exploitation of the DY potential is not even at thehorizon.On the theoretical side, the present fits to the data do not make use of the Q evolution schemeswhich are already now available for the TMD’s and for the quark-gluon correlators, and which clearlymust be integrated in the calculations. Also , a better understanding of higher-twist contributions toSIDIS observables (which are significant since h Q i is rather small) requires more insight and moreeffort.In the near future more data will be collected in the SIDIS sector by the COMPASS and by theJLab experiments at quite different energies. In the proton-proton sector, the luminosity of the RHICcollider should be enough to allow for measurements of DY processes.In a more distant future, COMPASS should providing new measurements of azimuthal asymmetriesin SIDIS on a liquid hydrogen target and a the first measurement of polarised DY process in π − p ↑ scattering. Also, higher energy SIDIS data will come from JLab upgraded at 12 GeV.In an even more distant future, many projects have been proposed. The PAX experiment at FAIRaims to investigate DY pairs in polarised antiproton-polarised proton scattering, a very clean way toaddress the transversity functions, and a terribly difficult experiment. Fixed target DY experimentsscattering polarised protons on polarised protons are being planned at JPARC, in Japan, and at NICA,in Dubna. It is fair to say, however, that the future of the field will depend in a crucial way on thepending decisions to construct a polarised electron-polarised proton collider. Ambitious projects arebeing pursued at BNL and at JLab. In Europe, the electron-proton collider has a long story, which needsnot to be summarised here, we only mention the most recent proposal for a polarised collider, which istailored to the new accelerators complex FAIR presently being realised at GSI. All these projects arebased on existing laboratories and existing infrastructures so that optimism is mandatory. This review on transverse spin and transverse momentum phenomena in hard processes has been possiblethanks to the friendly exchange of ideas and information which characterizes the ”spin community”, andwe would like to thank all our colleagues for their invaluable contribution in elucidating the many facetsof this new and rapidly growing field. This work has been partly supported by the Italian Ministry forEducation, University and Research (MIUR) as a Research Project of National Interest (PRIN).
References [1] L. Dick et al. , Phys. Lett.
B57 , 93 (1975).[2] R. D. Klem et al. , Phys. Rev. Lett. , 929 (1976).[3] W. H. Dragoset et al. , Phys. Rev. D18 , 3939 (1978).[4] G. Bunce et al. , Phys. Rev. Lett. , 1113 (1976).[5] K. J. Heller et al. , Phys. Rev. Lett. , 607 (1978).816] I. P. Auer et al. , FERMILAB-PROPOSAL-0581.[7] E581, D. L. Adams et al. , Phys. Lett. B261 , 201 (1991).[8] FNAL-E704, D. L. Adams et al. , Phys. Lett.
B264 , 462 (1991).[9] E581, D. L. Adams et al. , Z. Phys.
C56 , 181 (1992).[10] Fermilab E704, A. Bravar et al. , Phys. Rev. Lett. , 2626 (1996).[11] STAR, J. Adams et al. , Phys. Rev. Lett. , 171801 (2004), hep-ex/0310058.[12] STAR, J. Adams et al. , Phys. Rev. Lett. , 152302 (2006), nucl-ex/0602011.[13] PHENIX, S. S. Adler et al. , Phys. Rev. Lett. , 241803 (2003), hep-ex/0304038.[14] PHENIX, S. S. Adler et al. , Phys. Rev. Lett. , 202001 (2005), hep-ex/0507073.[15] BRAHMS, I. Arsene et al. , Phys. Rev. Lett. , 252001 (2007), hep-ex/0701041.[16] G. L. Kane, J. Pumplin, and W. Repko, Phys. Rev. Lett. , 1689 (1978).[17] A. V. Efremov and O. V. Teryaev, Sov. J. Nucl. Phys. , 140 (1982).[18] A. V. Efremov and O. V. Teryaev, Sov. J. Nucl. Phys. , 962 (1984).[19] P. G. Ratcliffe, Nucl. Phys. B264 , 493 (1986).[20] V. Barone, A. Drago, and P. G. Ratcliffe, Phys. Rept. , 1 (2002), hep-ph/0104283.[21] X. Artru and M. Mekhfi, Z. Phys.
C45 , 669 (1990).[22] R. L. Jaffe and X.-D. Ji, Phys. Rev. Lett. , 552 (1991).[23] R. L. Jaffe and X.-D. Ji, Nucl. Phys. B375 , 527 (1992).[24] J. L. Cortes, B. Pire, and J. P. Ralston, Z. Phys.
C55 , 409 (1992).[25] J. P. Ralston and D. E. Soper, Nucl. Phys.
B152 , 109 (1979).[26] N. Christ and T. D. Lee, Phys. Rev. , 1310 (1966).[27] X. Artru, (1992), hep-ph/9301296.[28] X. Artru, (1993), hep-ph/9310323.[29] J. C. Collins, Nucl. Phys.
B396 , 161 (1993), hep-ph/9208213.[30] J. C. Collins, S. F. Heppelmann, and G. A. Ladinsky, Nucl. Phys.
B420 , 565 (1994), hep-ph/9305309.[31] X.-D. Ji, Phys. Lett.
B284 , 137 (1992).[32] R. L. Jaffe, Phys. Rev.
D54 , 6581 (1996), hep-ph/9605456.[33] R. L. Jaffe, X.-m. Jin, and J. Tang, Phys. Rev. Lett. , 1166 (1998), hep-ph/9709322.[34] D. W. Sivers, Phys. Rev. D41 , 83 (1990). 8235] D. W. Sivers, Phys. Rev.
D43 , 261 (1991).[36] S. J. Brodsky, D. S. Hwang, and I. Schmidt, Phys. Lett.
B530 , 99 (2002), hep-ph/0201296.[37] S. J. Brodsky, D. S. Hwang, and I. Schmidt, Nucl. Phys.
B642 , 344 (2002), hep-ph/0206259.[38] J. C. Collins, Phys. Lett.
B536 , 43 (2002), hep-ph/0204004.[39] M. Anselmino and F. Murgia, Phys. Lett.
B442 , 470 (1998), hep-ph/9808426.[40] M. Anselmino, M. Boglione, and F. Murgia, Phys. Rev.
D60 , 054027 (1999), hep-ph/9901442.[41] HERMES, A. Airapetian et al. , Phys. Rev. Lett. , 012002 (2005), hep-ex/0408013.[42] COMPASS, V. Y. Alexakhin et al. , Phys. Rev. Lett. , 202002 (2005), hep-ex/0503002.[43] COMPASS, E. S. Ageev et al. , Nucl. Phys. B765 , 31 (2007), hep-ex/0610068.[44] Belle, K. Abe et al. , Phys. Rev. Lett. , 232002 (2006), hep-ex/0507063.[45] M. Anselmino et al. , Phys. Rev. D75 , 054032 (2007), hep-ph/0701006.[46] A. M. Kotzinian, Nucl. Phys.
B441 , 234 (1995), hep-ph/9412283.[47] R. D. Tangerman and P. J. Mulders, Phys. Rev.
D51 , 3357 (1995), hep-ph/9403227.[48] P. J. Mulders and R. D. Tangerman, Nucl. Phys.
B461 , 197 (1996), hep-ph/9510301.[49] R. D. Tangerman and P. J. Mulders, Phys. Lett.
B352 , 129 (1995), hep-ph/9501202.[50] D. Boer and P. J. Mulders, Phys. Rev.
D57 , 5780 (1998), hep-ph/9711485.[51] X.-d. Ji, J.-p. Ma, and F. Yuan, Phys. Rev.
D71 , 034005 (2005), hep-ph/0404183.[52] X.-d. Ji, J.-P. Ma, and F. Yuan, Phys. Lett.
B597 , 299 (2004), hep-ph/0405085.[53] J.-w. Qiu and G. Sterman, Phys. Rev. Lett. , 2264 (1991).[54] J.-w. Qiu and G. Sterman, Nucl. Phys. B378 , 52 (1992).[55] J.-w. Qiu and G. Sterman, Phys. Rev.
D59 , 014004 (1999), hep-ph/9806356.[56] X. Ji, J.-W. Qiu, W. Vogelsang, and F. Yuan, Phys. Rev. Lett. , 082002 (2006), hep-ph/0602239.[57] X. Ji, J.-w. Qiu, W. Vogelsang, and F. Yuan, Phys. Rev. D73 , 094017 (2006), hep-ph/0604023.[58] Y. Koike, W. Vogelsang, and F. Yuan, Phys. Lett.
B659 , 878 (2008), arXiv:0711.0636 [hep-ph].[59] Z.-B. Kang and J.-W. Qiu, Phys. Rev.
D79 , 016003 (2009), arXiv:0811.3101 [hep-ph].[60] J. Zhou, F. Yuan, and Z.-T. Liang, Phys. Rev.
D79 , 114022 (2009), arXiv:0812.4484 [hep-ph].[61] V. M. Braun, A. N. Manashov, and B. Pirnay, Phys. Rev.
D80 , 114002 (2009), arXiv:0909.3410[hep-ph].[62] W. Vogelsang and F. Yuan, Phys. Rev.
D79 , 094010 (2009), arXiv:0904.0410 [hep-ph].[63] U. D’Alesio and F. Murgia, Prog. Part. Nucl. Phys. , 394 (2008), arXiv:0712.4328 [hep-ph].8364] M. Burkardt, A. Miller, and W. D. Nowak, Rept. Prog. Phys. , 016201 (2010), arXiv:0812.2208[hep-ph].[65] P. J. Mulders and J. Rodrigues, Phys. Rev. D63 , 094021 (2001), hep-ph/0009343.[66] M. Diehl, Phys. Rept. , 41 (2003), hep-ph/0307382.[67] NA10, S. Falciano et al. , Phys. Lett.
B158 , 92 (1985).[68] NA10, M. Guanziroli et al. , Z. Phys.
C37 , 545 (1988).[69] J. S. Conway et al. , Phys. Rev.
D39 , 92 (1989).[70] FNAL E866/NuSea, L. Y. Zhu et al. , Phys. Rev. Lett. , 182001 (2009), arXiv:0811.4589[nucl-ex].[71] R. P. Feynman,
Photon–Hadron Interactions (Benjamin, 1972).[72] D. E. Soper, Phys. Rev.
D15 , 1141 (1977).[73] D. E. Soper, Phys. Rev. Lett. , 1847 (1979).[74] J. C. Collins, Acta Phys. Polon. B34 , 3103 (2003), hep-ph/0304122.[75] R. L. Jaffe, hep-ph/9602236.[76] Y. L. Dokshitzer, Sov. Phys. JETP , 641 (1977).[77] V. N. Gribov and L. N. Lipatov, Sov. J. Nucl. Phys. , 675 (1972).[78] G. Altarelli and G. Parisi, Nucl. Phys. B126 , 298 (1977).[79] A. Hayashigaki, Y. Kanazawa, and Y. Koike, Phys. Rev.
D56 , 7350 (1997), hep-ph/9707208.[80] S. Kumano and M. Miyama, Phys. Rev.
D56 , R2504 (1997), hep-ph/9706420.[81] W. Vogelsang, Phys. Rev.
D57 , 1886 (1998), hep-ph/9706511.[82] V. Barone, Phys. Lett.
B409 , 499 (1997), hep-ph/9703343.[83] V. Barone, T. Calarco, and A. Drago, Phys. Rev.
D56 , 527 (1997), hep-ph/9702239.[84] J. Soffer, Phys. Rev. Lett. , 1292 (1995), hep-ph/9409254.[85] C. Bourrely, J. Soffer, and O. V. Teryaev, Phys. Lett. B420 , 375 (1998), hep-ph/9710224.[86] B. L. G. Bakker, E. Leader, and T. L. Trueman, Phys. Rev.
D70 , 114001 (2004), hep-ph/0406139.[87] M. Burkardt, Phys. Rev.
D72 , 094020 (2005), hep-ph/0505189.[88] M. Burkardt, Phys. Lett.
B639 , 462 (2006).[89] J. C. Collins and D. E. Soper, Nucl. Phys.
B193 , 381 (1981).[90] X.-d. Ji and F. Yuan, Phys. Lett.
B543 , 66 (2002), hep-ph/0206057.[91] A. V. Belitsky, X. Ji, and F. Yuan, Nucl. Phys.
B656 , 165 (2003), hep-ph/0208038.8492] A. Idilbi, X.-d. Ji, J.-P. Ma, and F. Yuan, Phys. Rev.
D70 , 074021 (2004), hep-ph/0406302.[93] G. A. Miller, Phys. Rev.
C76 , 065209 (2007), arxiv:0708.2297 [nucl-th].[94] H. Avakian, A. V. Efremov, P. Schweitzer, and F. Yuan, Phys. Rev.
D78 , 114024 (2008),arXiv:0805.3355 [hep-ph].[95] M. Boglione and P. J. Mulders, Phys. Rev.
D60 , 054007 (1999), hep-ph/9903354.[96] A. Bacchetta, M. Boglione, A. Henneman, and P. J. Mulders, Phys. Rev. Lett. , 712 (2000),hep-ph/9912490.[97] J. C. Collins and A. Metz, Phys. Rev. Lett. , 252001 (2004), hep-ph/0408249.[98] C. J. Bomhof, P. J. Mulders, and F. Pijlman, Phys. Lett. B596 , 277 (2004), hep-ph/0406099.[99] A. Bacchetta, C. J. Bomhof, P. J. Mulders, and F. Pijlman, Phys. Rev.
D72 , 034030 (2005),hep-ph/0505268.[100] C. J. Bomhof, P. J. Mulders, and F. Pijlman, Eur. Phys. J.
C47 , 147 (2006), hep-ph/0601171.[101] C. J. Bomhof and P. J. Mulders, JHEP , 029 (2007), hep-ph/0609206.[102] C. J. Bomhof, P. J. Mulders, W. Vogelsang, and F. Yuan, Phys. Rev. D75 , 074019 (2007),hep-ph/0701277.[103] C. J. Bomhof and P. J. Mulders, Nucl. Phys.
B795 , 409 (2008), arXiv:0709.1390 [hep-ph].[104] T. C. Rogers and P. J. Mulders, Phys. Rev.
D81 , 094006 (2010), arXiv:1001.2977 [hep-ph].[105] STAR, B. I. Abelev et al. , Phys. Rev. Lett. , 142003 (2007), arXiv:0705.4629 [hep-ex].[106] M. Burkardt, Phys. Rev. D69 , 057501 (2004), hep-ph/0311013.[107] M. Burkardt, Phys. Rev.
D69 , 091501 (2004), hep-ph/0402014.[108] A. V. Efremov, K. Goeke, S. Menzel, A. Metz, and P. Schweitzer, Phys. Lett.
B612 , 233 (2005),hep-ph/0412353.[109] K. Goeke, S. Meissner, A. Metz, and M. Schlegel, Phys. Lett.
B637 , 241 (2006), hep-ph/0601133.[110] M. Anselmino et al. , Eur. Phys. J.
A39 , 89 (2009), arXiv:0805.2677 [hep-ph].[111] S. Arnold, A. V. Efremov, K. Goeke, M. Schlegel, and P. Schweitzer, (2008), arXiv:0805.2137[hep-ph].[112] V. Barone and P. G. Ratcliffe,
Transverse spin physics (World Scientific, Singapore, 2003).[113] R. K. Ellis, W. Furmanski, and R. Petronzio, Nucl. Phys.
B207 , 1 (1982).[114] R. K. Ellis, W. Furmanski, and R. Petronzio, Nucl. Phys.
B212 , 29 (1983).[115] J. Zhou, F. Yuan, and Z.-T. Liang, (2009), arXiv;0909.2238 [hep-ph].[116] D. Boer, P. J. Mulders, and F. Pijlman, Nucl. Phys.
B667 , 201 (2003), hep-ph/0303034.[117] S. Wandzura and F. Wilczek, Phys. Lett.
B72 , 195 (1977).85118] M. Anselmino, A. Efremov, and E. Leader, Phys. Rept. , 1 (1995), hep-ph/9501369.[119] A. Accardi, A. Bacchetta, W. Melnitchouk, and M. Schlegel, JHEP , 093 (2009),arXiv:0907.2942 [hep-ph].[120] R. D. Tangerman and P. J. Mulders, (1994), hep-ph/9408305.[121] H. Avakian et al. , Phys. Rev. D77 , 014023 (2008), arXiv:0709.3253 [hep-ph].[122] A. Metz, P. Schweitzer, and T. Teckentrup, Phys. Lett.
B680 , 141 (2009), arXiv:0810.5212[hep-ph].[123] A. Kotzinian, B. Parsamyan, and A. Prokudin, Phys. Rev.
D73 , 114017 (2006), hep-ph/0603194.[124] A. Bacchetta, P. J. Mulders, and F. Pijlman, Phys. Lett.
B595 , 309 (2004), hep-ph/0405154.[125] K. Goeke, A. Metz, and M. Schlegel, Phys. Lett.
B618 , 90 (2005), hep-ph/0504130.[126] A. Bacchetta et al. , JHEP , 093 (2007), hep-ph/0611265.[127] P. Hoodbhoy and X.-D. Ji, Phys. Rev. D50 , 4429 (1994), hep-ph/9307304.[128] R. Kundu and A. Metz, Phys. Rev.
D65 , 014009 (2002), hep-ph/0107073.[129] K. Goeke, A. Metz, P. V. Pobylitsa, and M. V. Polyakov, Phys. Lett.
B567 , 27 (2003), hep-ph/0302028.[130] A. V. Radyushkin, Phys. Lett.
B380 , 417 (1996), hep-ph/9604317.[131] X.-D. Ji, Phys. Rev. Lett. , 610 (1997), hep-ph/9603249.[132] K. Goeke, M. V. Polyakov, and M. Vanderhaeghen, Prog. Part. Nucl. Phys. , 401 (2001),hep-ph/0106012.[133] A. V. Belitsky and A. V. Radyushkin, Phys. Rept. , 1 (2005), hep-ph/0504030.[134] S. Boffi and B. Pasquini, Riv. Nuovo Cim. , 387 (2007), arXiv:0711.2625 [hep-ph].[135] S. Meissner, A. Metz, and K. Goeke, Phys. Rev. D76 , 034002 (2007), hep-ph/0703176.[136] M. Diehl, Eur. Phys. J.
C19 , 485 (2001), hep-ph/0101335.[137] S. Meissner, A. Metz, and M. Schlegel, (2008), arXiv:0807.1154 [hep-ph].[138] M. Burkardt, Int. J. Mod. Phys.
A18 , 173 (2003), hep-ph/0207047.[139] M. Diehl and P. Hagler, Eur. Phys. J.
C44 , 87 (2005), hep-ph/0504175.[140] M. Burkardt and D. S. Hwang, Phys. Rev.
D69 , 074032 (2004), hep-ph/0309072.[141] M. Burkardt, Phys. Rev.
D66 , 114005 (2002), hep-ph/0209179.[142] M. Burkardt, Nucl. Phys.
A735 , 185 (2004), hep-ph/0302144.[143] QCDSF, M. G¨ockeler et al. , Phys. Rev. Lett. , 222001 (2007), hep-lat/0612032.[144] QCDSF-UKQCD, D. Br¨ommel et al. , Prog. Part. Nucl. Phys. , 73 (2008).86145] V. Barone, S. Melis, and A. Prokudin, Phys. Rev. D81 , 114026 (2010), arXiv:0912.5194 [hep-ph].[146] H. Avakian et al. , Mod. Phys. Lett.
A24 , 2995 (2009), arXiv:0910.3181 [hep-ph].[147] R. Jakob, P. J. Mulders, and J. Rodrigues, Nucl. Phys.
A626 , 937 (1997), hep-ph/9704335.[148] G. R. Goldstein and L. Gamberg, (2002), hep-ph/0209085.[149] L. P. Gamberg, G. R. Goldstein, and K. A. Oganessyan, Phys. Rev.
D67 , 071504 (2003), hep-ph/0301018.[150] A. Bacchetta, A. Schaefer, and J.-J. Yang, Phys. Lett.
B578 , 109 (2004), hep-ph/0309246.[151] L. P. Gamberg, G. R. Goldstein, and M. Schlegel, Phys. Rev.
D77 , 094016 (2008), arXiv:0708.0324[hep-ph].[152] A. Bacchetta, F. Conti, and M. Radici, Phys. Rev.
D78 , 074010 (2008), arXiv:0807.0323 [hep-ph].[153] J. R. Ellis, D. S. Hwang, and A. Kotzinian, Phys. Rev.
D80 , 074033 (2009), arXiv:0808.1567[hep-ph].[154] F. Yuan, Phys. Lett.
B575 , 45 (2003), hep-ph/0308157.[155] I. O. Cherednikov, U. D’Alesio, N. I. Kochelev, and F. Murgia, Phys. Lett.
B642 , 39 (2006),hep-ph/0606238.[156] A. Courtoy, S. Scopetta, and V. Vento, Phys. Rev.
D79 , 074001 (2009), arXiv:0811.1191 [hep-ph].[157] A. Courtoy, S. Scopetta, and V. Vento, Phys. Rev.
D80 , 074032 (2009), arXiv:0909.1404 [hep-ph].[158] A. Courtoy, F. Fratini, S. Scopetta, and V. Vento, Phys. Rev.
D78 , 034002 (2008),arXiv:0801.4347 [hep-ph].[159] Z. Lu and B.-Q. Ma, Nucl. Phys.
A741 , 200 (2004), hep-ph/0406171.[160] P. V. Pobylitsa, (2003), hep-ph/0301236.[161] A. Drago, Phys. Rev.
D71 , 057501 (2005), hep-ph/0501282.[162] A. V. Afanasev and C. E. Carlson, Phys. Rev.
D74 , 114027 (2006), hep-ph/0603269.[163] L. P. Gamberg, D. S. Hwang, A. Metz, and M. Schlegel, Phys. Lett.
B639 , 508 (2006), hep-ph/0604022.[164] B. Pasquini, S. Cazzaniga, and S. Boffi, Phys. Rev.
D78 , 034025 (2008), arXiv:0806.2298 [hep-ph].[165] S. Boffi, A. V. Efremov, B. Pasquini, and P. Schweitzer, Phys. Rev.
D79 , 094012 (2009),arXiv:0903.1271 [hep-ph].[166] J. She, J. Zhu, and B.-Q. Ma, Phys. Rev.
D79 , 054008 (2009), arXiv:0902.3718 [hep-ph].[167] A. V. Efremov, P. Schweitzer, O. V. Teryaev, and P. Zavada, Phys. Rev.
D80 , 014021 (2009),arXiv:0903.3490 [hep-ph].[168] H. Avakian, A. V. Efremov, P. Schweitzer, and F. Yuan, (2010), arXiv:1001.5467 [hep-ph].[169] V. Barone, T. Calarco, and A. Drago, Phys. Lett.
B390 , 287 (1997), hep-ph/9605434.87170] G. Parisi and R. Petronzio, Phys. Lett.
B62 , 331 (1976).[171] R. L. Jaffe and G. G. Ross, Phys. Lett.
B93 , 313 (1980).[172] A. W. Schreiber, A. I. Signal, and A. W. Thomas, Phys. Rev.
D44 , 2653 (1991).[173] V. Barone and A. Drago, Nucl. Phys.
A552 , 479 (1993).[174] M. Anselmino, D. Boer, U. D’Alesio, and F. Murgia, Phys. Rev.
D63 , 054029 (2001), hep-ph/0008186.[175] A. Sch¨afer and O. V. Teryaev, Phys. Rev.
D61 , 077903 (2000), hep-ph/9908412.[176] X. Artru, J. Czyzewski, and H. Yabuki, Z. Phys.
C73 , 527 (1997), hep-ph/9508239.[177] X. Artru, (2010), arXiv:1001.1061 [hep-ph].[178] A. Bacchetta, R. Kundu, A. Metz, and P. J. Mulders, Phys. Lett.
B506 , 155 (2001), hep-ph/0102278.[179] A. Bacchetta, R. Kundu, A. Metz, and P. J. Mulders, Phys. Rev.
D65 , 094021 (2002), hep-ph/0201091.[180] A. Bacchetta, A. Metz, and J.-J. Yang, Phys. Lett.
B574 , 225 (2003), hep-ph/0307282.[181] L. P. Gamberg, G. R. Goldstein, and K. A. Oganessyan, Phys. Rev.
D68 , 051501 (2003), hep-ph/0307139.[182] L. P. Gamberg, D. S. Hwang, and K. A. Oganessyan, Phys. Lett.
B584 , 276 (2004), hep-ph/0311221.[183] D. Amrath, A. Bacchetta, and A. Metz, Phys. Rev.
D71 , 114018 (2005), hep-ph/0504124.[184] A. Bacchetta, U. D’Alesio, M. Diehl, and C. A. Miller, Phys. Rev.
D70 , 117504 (2004), hep-ph/0410050.[185] M. Diehl and S. Sapeta, Eur. Phys. J.
C41 , 515 (2005), hep-ph/0503023.[186] R. N. Cahn, Phys. Lett.
B78 , 269 (1978).[187] R. N. Cahn, Phys. Rev.
D40 , 3107 (1989).[188] M. Anselmino et al. , Phys. Rev.
D73 , 014020 (2006), hep-ph/0509035.[189] M. Anselmino et al. , paper in preparation.[190] J. C. Collins, T. C. Rogers, and A. M. Stasto, Phys. Rev.
D77 , 085009 (2008), arXiv:0708.2833[hep-ph].[191] X. Ji, J.-W. Qiu, W. Vogelsang, and F. Yuan, Phys. Lett.
B638 , 178 (2006), hep-ph/0604128.[192] A. Bacchetta, D. Boer, M. Diehl, and P. J. Mulders, JHEP , 023 (2008), arXiv:0803.0227[hep-ph].[193] H. Georgi and H. D. Politzer, Phys. Rev. Lett. , 3 (1978).[194] A. Mendez, Nucl. Phys. B145 , 199 (1978). 88195] J.-g. Chay, S. D. Ellis, and W. J. Stirling, Phys. Rev.
D45 , 46 (1992).[196] K. Hagiwara, K.-i. Hikasa, and N. Kai, Phys. Rev.
D27 , 84 (1983).[197] M. Ahmed and T. Gehrmann, Phys. Lett.
B465 , 297 (1999), hep-ph/9906503.[198] A. V. Efremov and O. V. Teryaev, Phys. Lett.
B150 , 383 (1985).[199] H. Eguchi, Y. Koike, and K. Tanaka, Nucl. Phys.
B763 , 198 (2007), hep-ph/0610314.[200] H. Eguchi, Y. Koike, and K. Tanaka, Nucl. Phys.
B752 , 1 (2006), hep-ph/0604003.[201] M. Stratmann and W. Vogelsang, Phys. Rev.
D65 , 057502 (2002), hep-ph/0108241.[202] N. S. Craigie, F. Baldracchini, V. Roberto, and M. Socolovsky, Phys. Lett.
B96 , 381 (1980).[203] F. Baldracchini, N. S. Craigie, V. Roberto, and M. Socolovsky, Fortschr. Phys. , 505 (1981).[204] X. Artru and M. Mekhfi, Nucl. Phys. A532 , 351 (1991).[205] M. Anselmino, M. Boglione, and F. Murgia, Phys. Lett.
B481 , 253 (2000), hep-ph/0001307.[206] B.-Q. Ma, I. Schmidt, J. Soffer, and J.-J. Yang, Phys. Rev.
D64 , 014017 (2001), hep-ph/0103136.[207] J.-J. Yang, Nucl. Phys.
A699 , 562 (2002), hep-ph/0111382.[208] M. Anselmino, D. Boer, U. D’Alesio, and F. Murgia, Phys. Rev.
D65 , 114014 (2002), hep-ph/0109186.[209] J. Zhou, F. Yuan, and Z.-T. Liang, Phys. Rev.
D78 , 114008 (2008), arXiv:0808.3629 [hep-ph].[210] J. C. Collins and G. A. Ladinsky, (1994), hep-ph/9411444.[211] A. Bianconi, S. Boffi, R. Jakob, and M. Radici, Phys. Rev.
D62 , 034008 (2000), hep-ph/9907475.[212] A. Bianconi, S. Boffi, R. Jakob, and M. Radici, Phys. Rev.
D62 , 034009 (2000), hep-ph/9907488.[213] P. Cea, G. Nardulli, and P. Chiappetta, Phys. Lett.
B209 , 333 (1988).[214] K. Konishi, A. Ukawa, and G. Veneziano, Phys. Lett.
B78 , 243 (1978).[215] K. Konishi, A. Ukawa, and G. Veneziano, Nucl. Phys.
B157 , 45 (1979).[216] X. Artru and J. C. Collins, Z. Phys.
C69 , 277 (1996), hep-ph/9504220.[217] A. Bacchetta and M. Radici, Phys. Rev.
D69 , 074026 (2004), hep-ph/0311173.[218] A. Bacchetta and M. Radici, Phys. Rev.
D67 , 094002 (2003), hep-ph/0212300.[219] A. Bacchetta and M. Radici, Phys. Rev.
D74 , 114007 (2006), hep-ph/0608037.[220] K. Chen, G. R. Goldstein, R. L. Jaffe, and X.-D. Ji, Nucl. Phys.
B445 , 380 (1995), hep-ph/9410337.[221] W. Bonivento et al. , Internal Note DELPHI-95-81 PHYS 516, 1995.[222] D. Boer, R. Jakob, and P. J. Mulders, Phys. Lett.
B424 , 143 (1998), hep-ph/9711488.89223] D. Boer, R. Jakob, and P. J. Mulders, Nucl. Phys.
B504 , 345 (1997), hep-ph/9702281.[224] D. Boer, Ph. D. Thesis, Vrije Universiteit, Amsterdam, 1998.[225] D. Boer, Nucl. Phys.
B806 , 23 (2009), arXiv:0804.2408 [hep-ph].[226] Belle, R. Seidl et al. , Phys. Rev.
D78 , 032011 (2008), arXiv:0805.2975 [hep-ex].[227] K. Gottfried and J. D. Jackson, Nuovo Cim. , 309 (1964).[228] A. P. Contogouris, O. Korakianitis, Z. Merebashvili, and F. Lebessis, Phys. Lett. B344 , 370(1995).[229] F. A. Ceccopieri, M. Radici, and A. Bacchetta, Phys. Lett.
B650 , 81 (2007), hep-ph/0703265.[230] A. Bacchetta, F. A. Ceccopieri, A. Mukherjee, and M. Radici, Phys. Rev.
D79 , 034029 (2009),arXiv:0812.0611 [hep-ph].[231] B. Pire and J. P. Ralston, Phys. Rev.
D28 , 260 (1983).[232] PAX, V. Barone et al. , (2005), hep-ex/0505054.[233] J. C. Collins and D. E. Soper, Phys. Rev.
D16 , 2219 (1977).[234] C. S. Lam and W.-K. Tung, Phys. Rev.
D18 , 2447 (1978).[235] S. Arnold, A. Metz, and M. Schlegel, Phys. Rev.
D79 , 034005 (2009).[236] C. S. Lam and W.-K. Tung, Phys. Lett.
B80 , 228 (1979).[237] C. S. Lam and W.-K. Tung, Phys. Rev.
D21 , 2712 (1980).[238] J. C. Collins, Phys. Rev. Lett. , 291 (1979).[239] E. Mirkes and J. Ohnemus, Phys. Rev. D51 , 4891 (1995), hep-ph/9412289.[240] D. Boer, Phys. Rev.
D60 , 014012 (1999), hep-ph/9902255.[241] J. C. Collins, Nucl. Phys.
B394 , 169 (1993), hep-ph/9207265.[242] W. Vogelsang and A. Weber, Phys. Rev.
D48 , 2073 (1993).[243] A. P. Contogouris, B. Kamal, and Z. Merebashvili, Phys. Lett.
B337 , 169 (1994).[244] B. Kamal, Phys. Rev.
D53 , 1142 (1996), hep-ph/9511217.[245] O. Martin, A. Sch¨afer, M. Stratmann, and W. Vogelsang, Phys. Rev.
D57 , 3084 (1998), hep-ph/9710300.[246] O. Martin, A. Sch¨afer, M. Stratmann, and W. Vogelsang, Phys. Rev.
D60 , 117502 (1999), hep-ph/9902250.[247] V. Barone, A. Cafarella, C. Coriano’, M. Guzzi, and P. Ratcliffe, Phys. Lett.
B639 , 483 (2006),hep-ph/0512121.[248] P. G. Ratcliffe, Eur. Phys. J.
C41 , 319 (2005), hep-ph/0412157.90249] D. Boer and W. Vogelsang, Phys. Rev.
D74 , 014004 (2006), hep-ph/0604177.[250] E. L. Berger, J.-W. Qiu, and R. A. Rodriguez-Pedraza, Phys. Rev.
D76 , 074006 (2007),arXiv:0708.0578 [hep-ph].[251] J. Zhou, F. Yuan, and Z.-T. Liang, Phys. Lett.
B678 , 264 (2009), arXiv:0901.3601 [hep-ph].[252] Z.-t. Liang and C. Boros, Int. J. Mod. PHys.
A15 , 927 (2000), hep-ph/0001330.[253] J. C. Collins, D. E. Soper, and G. Sterman, Nucl. Phys.
B261 , 104 (1985).[254] J. Collins and J.-W. Qiu, Phys. Rev.
D75 , 114014 (2007), arXiv:0705.2141 [hep-ph].[255] M. Anselmino, M. Boglione, U. D’Alesio, E. Leader, and F. Murgia, Phys. Rev.
D70 , 074025(2004), hep-ph/0407100.[256] M. Anselmino, U. D’Alesio, S. Melis, and F. Murgia, Phys. Rev.
D74 , 094011 (2006), hep-ph/0608211.[257] U. D’Alesio and F. Murgia, Phys. Rev.
D70 , 074009 (2004), hep-ph/0408092.[258] I. Schmidt, J. Soffer, and J.-J. Yang, Phys. Lett.
B612 , 258 (2005), hep-ph/0503127.[259] U. D’Alesio and F. Murgia, AIP Conf. Proc. , 559 (2007), hep-ph/0612208.[260] A. Bacchetta, C. Bomhof, U. D’Alesio, P. J. Mulders, and F. Murgia, Phys. Rev. Lett. , 212002(2007), hep-ph/0703153.[261] D. Boer and W. Vogelsang, Phys. Rev. D69 , 094025 (2004), hep-ph/0312320.[262] F. Yuan, Phys. Rev. Lett. , 032003 (2008), arXiv:0709.3272 [hep-ph].[263] Y. Kanazawa and Y. Koike, Phys. Lett.
B478 , 121 (2000), hep-ph/0001021.[264] Y. Kanazawa and Y. Koike, Phys. Lett.
B490 , 99 (2000), hep-ph/0007272.[265] Y. Koike, AIP Conf. Proc. , 449 (2003), hep-ph/0210396.[266] Y. Koike, Nucl. Phys.
A721 , 364 (2003), hep-ph/0211400.[267] A. D. Panagiotou, Int. J. Mod. Phys. A5 , 1197 (1990).[268] J. Felix, Mod. Phys. Lett. A14 , 827 (1999).[269] D. Boer, C. J. Bomhof, D. S. Hwang, and P. J. Mulders, Phys. Lett.
B659 , 127 (2008),arXiv:0709.1087 [hep-ph].[270] M. Anselmino et al. , Phys. Rev.
D81 , 034007 (2010), arXiv:0911.1744 [hep-ph].[271] B. Pire and L. Szymanowski, Phys. Rev. Lett. , 072002 (2009), arXiv:0905.1258 [hep-ph].[272] S. Ahmad, G. R. Goldstein, and S. Liuti, Phys. Rev.
D79 , 054014 (2009), arXiv:0805.3568[hep-ph].[273] R. Enberg, B. Pire, and L. Szymanowski, Eur. Phys. J.
C47 , 87 (2006), hep-ph/0601138.[274] E. L. Berger, ANL-HEP-CP-87-45. 91275] HERMES, M. Diefenthaler, (2007), arXiv:0706.2242 [hep-ex].[276] HERMES, A. Airapetian et al. , Phys. Rev. Lett. , 152002 (2009), arXiv:0906.3918 [hep-ex].[277] COMPASS, A. Bressan, (2009), arXiv:0907.5508 [hep-ex].[278] COMPASS, H. Wollny, (2009), arXiv:0907.0961 [hep-ex].[279] COMPASS, M. G. Alekseev et al. , (2010), arXiv:1005.5609 [hep-ex].[280] COMPASS, M. Alekseev et al. , Phys. Lett.
B673 , 127 (2009), arXiv:0802.2160 [hep-ex].[281] HERMES, A. Airapetian et al. , (2010), arXiv:1006.4221 [hep-ex].[282] X. Artru, (2002), hep-ph/0207309.[283] A. V. Efremov, O. G. Smirnova, and L. G. Tkachev, Nucl. Phys. Proc. Suppl. , 49 (1999),hep-ph/9812522.[284] A. V. Efremov, K. Goeke, and P. Schweitzer, Phys. Rev. D73 , 094025 (2006), hep-ph/0603054.[285] M. Anselmino et al. , Nucl. Phys. Proc. Suppl. , 98 (2009), arXiv:0812.4366 [hep-ph].[286] A. V. Efremov, K. Goeke, and P. Schweitzer, Eur. Phys. J. ST , 1 (2008), arXiv:0801.2238[hep-ph].[287] W. Vogelsang and F. Yuan, Phys. Rev.
D72 , 054028 (2005), hep-ph/0507266.[288] P. Schweitzer et al. , Phys. Rev.
D64 , 034013 (2001), hep-ph/0101300.[289] LHPC, B. U. Musch et al. , PoS
LAT2007 , 155 (2007), arXiv:0710.4423 [hep-lat].[290] B. U. Musch et al. , PoS
LC2008 , 053 (2008), arXiv:0811.1536 [hep-lat].[291] M. Anselmino et al. , Phys. Rev.
D71 , 074006 (2005), hep-ph/0501196.[292] M. Wakamatsu, Phys. Rev.
D79 , 014033 (2009), arXiv:0811.4196 [hep-ph].[293] D. Boer, Nucl. Phys.
B603 , 195 (2001), hep-ph/0102071.[294] COMPASS, S. Levorato, (2008), arXiv:0808.0086 [hep-ex].[295] HERMES, A. Airapetian et al. , JHEP , 017 (2008), arXiv:0803.2367 [hep-ex].[296] COMPASS, F. Sozzi, Nucl. Phys. Proc. Suppl. , 47 (2007).[297] COMPASS, F. Bradamante, (2006), hep-ex/0602013.[298] COMPASS, A. Martin, Czech. J. Phys. , F33 (2006), hep-ex/0702002.[299] COMPASS, A. Vossen, (2007), arXiv:0705.2865 [hep-ex].[300] COMPASS, F. Massmann, Eur. Phys. J. ST , 85 (2008).[301] COMPASS, R. Joosten, AIP Conf. Proc. , 646 (2007).[302] A. Vossen et al. , (2009), arXiv:0912.0353 [hep-ex].92303] A. Ferrero, AIP Conf. Proc. , 436 (2007).[304] COMPASS, T. Negrini, AIP Conf. Proc. , 656 (2009).[305] ALEPH, D. Buskulic et al. , Phys. Lett. B374 , 319 (1996).[306] HERMES, R. Seidl, Prepared for 12th International Workshop on Deep Inelastic Scattering (DIS2004), Strbske Pleso, Slovakia, 14-18 Apr 2004.[307] J. C. Collins et al. , Phys. Rev.
D73 , 014021 (2006), hep-ph/0509076.[308] J. C. Collins et al. , (2005), hep-ph/0510342.[309] HERMES, M. Diefenthaler, AIP Conf. Proc. , 933 (2005), hep-ex/0507013.[310] M. Anselmino et al. , (2005), hep-ph/0511017.[311] European Muon, J. J. Aubert et al. , Phys. Lett.
B130 , 118 (1983).[312] European Muon, M. Arneodo et al. , Z. Phys.
C34 , 277 (1987).[313] ZEUS, J. Breitweg et al. , Phys. Lett.
B481 , 199 (2000), hep-ex/0003017.[314] A. Konig and P. Kroll, Z. Phys.
C16 , 89 (1982).[315] P. M. Nadolsky, D. R. Stump, and C. P. Yuan, Phys. Rev.
D61 , 014003 (2000), hep-ph/9906280.[316] P. M. Nadolsky, D. R. Stump, and C. P. Yuan, Phys. Lett.
B515 , 175 (2001), hep-ph/0012262.[317] On behalf of the HERMES, F. Giordano and R. Lamb, AIP Conf. Proc. , 423 (2009),arXiv:0901.2438 [hep-ex].[318] COMPASS, W. K¨afer, (2008), arXiv:0808.0114 [hep-ex].[319] CLAS, M. Osipenko et al. , Phys. Rev.
D80 , 032004 (2009), arXiv:0809.1153 [hep-ex].[320] V. Barone, A. Prokudin, and B.-Q. Ma, Phys. Rev.
D78 , 045022 (2008), arXiv:0804.3024 [hep-ph].[321] B. Zhang, Z. Lu, B.-Q. Ma, and I. Schmidt, Phys. Rev.
D78 , 034035 (2008), arXiv:0807.0503[hep-ph].[322] B. Pasquini and F. Yuan, (2010), arXiv:1001.5398 [hep-ph].[323] NA10, S. Falciano et al. , Z. Phys.
C31 , 513 (1986).[324] FNAL-E866/NuSea, L. Y. Zhu et al. , Phys. Rev. Lett. , 082301 (2007), hep-ex/0609005.[325] B. Zhang, Z. Lu, B.-Q. Ma, and I. Schmidt, Phys. Rev. D77 , 054011 (2008), arXiv:0803.1692[hep-ph].[326] Z. Lu and I. Schmidt, (2009), arXiv:0912.2031 [hep-ph].[327] COMPASS, A. Kotzinian, (2007), arXiv:0705.2402 [hep-ex].[328] COMPASS, I. A. Savin et al. , (2010), arXiv:1002.3052 [hep-ex].[329] HERMES, A. Airapetian et al. , Phys. Rev. Lett. , 4047 (2000), hep-ex/9910062.93330] HERMES, A. Airapetian et al. , Phys. Lett. B562 , 182 (2003), hep-ex/0212039.[331] CLAS, H. Avakian et al. , (2010), arXiv:1003.4549 [hep-ex].[332] HERMES, A. Airapetian et al. , Phys. Lett.
B622 , 14 (2005), hep-ex/0505042.[333] CLAS, H. Avakian et al. , Phys. Rev.
D69 , 112004 (2004), hep-ex/0301005.[334] HERMES, A. Airapetian et al. , Phys. Lett.
B648 , 164 (2007), hep-ex/0612059.[335] J. Antille et al. , Phys. Lett.
B94 , 523 (1980).[336] K. J. Heller et al. , Phys. Rev. Lett. , 2025 (1983).[337] B. Lundberg et al. , Phys. Rev. D40 , 3557 (1989).[338] E. J. Ramberg et al. , Phys. Lett.
B338 , 403 (1994).[339] FNAL E704, D. L. Adams et al. , Phys. Rev.
D53 , 4747 (1996).[340] E704, A. Bravar et al. , Phys. Rev. Lett. , 3073 (1995).[341] Fermilab E704, D. L. Adams et al. , Nucl. Phys. B510 , 3 (1998).[342] V. D. Apokin et al. , Phys. Lett.
B243 , 461 (1990).[343] V. V. Abramov et al. , Nucl. Phys.
B492 , 3 (1997), hep-ex/0110011.[344] K. Krueger et al. , Phys. Lett.
B459 , 412 (1999).[345] C. E. Allgower et al. , Phys. Rev.
D65 , 092008 (2002).[346] M. Anselmino, M. Boglione, and F. Murgia, Phys. Lett.
B362 , 164 (1995), hep-ph/9503290.[347] M. Boglione and E. Leader, Phys. Rev.
D61 , 114001 (2000), hep-ph/9911207.[348] M. Anselmino et al. , (2008), arXiv:0809.3743 [hep-ph].[349] M. Anselmino, M. Boglione, U. D’Alesio, E. Leader, and F. Murgia, Phys. Rev.
D71 , 014002(2005), hep-ph/0408356.[350] C. Kouvaris, J.-W. Qiu, W. Vogelsang, and F. Yuan, Phys. Rev.
D74 , 114013 (2006), hep-ph/0609238.[351] STAR, B. I. Abelev et al. , Phys. Rev. Lett. , 222001 (2008), arXiv:0801.2990 [hep-ex].[352] M. Boglione, U. D’Alesio, and F. Murgia, Phys. Rev.
D77 , 051502 (2008), arXiv:0712.4240[hep-ph].[353] M. Anselmino et al. , (2009), arXiv:0907.3999 [hep-ph].[354] BRAHMS, J. H. Lee and F. Videbaek, (2009), arXiv:0908.4551 [hep-ex].[355] BRAHMS, I. Arsene et al. , Phys. Rev. Lett. , 042001 (2008), arXiv:0801.1078 [nucl-ex].[356] C. Bourrely and J. Soffer, Eur. Phys. J.
C36 , 371 (2004), hep-ph/0311110.[357] PHENIX, A. Adare et al. , Phys. Rev.
D76 , 051106 (2007), arXiv:0704.3599 [hep-ex].94358] D. de Florian, Phys. Rev.
D67 , 054004 (2003), hep-ph/0210442.[359] D. de Florian and W. Vogelsang, Phys. Rev.
D71 , 114004 (2005), hep-ph/0501258.[360] D. de Florian, W. Vogelsang, and F. Wagner, Phys. Rev.
D76 , 094021 (2007), arXiv:0708.3060[hep-ph].[361] D. de Florian, W. Vogelsang, and F. Wagner, Phys. Rev.
D78 , 074025 (2008), arXiv:0807.4515[hep-ph].[362] M. Anselmino, V. Barone, A. Drago, and N. N. Nikolaev, Phys. Lett.
B594 , 97 (2004), hep-ph/0403114.[363] A. V. Efremov, K. Goeke, and P. Schweitzer, Eur. Phys. J.
C35 , 207 (2004), hep-ph/0403124.[364] M. Anselmino et al. , Phys. Rev.
D79 , 054010 (2009), arXiv:0901.3078 [hep-ph].[365] P. Lenisa, Int. J. Mod. Phys.
E18 , 484 (2009).[366] J. C. Collins et al. , Phys. Rev.
D73 , 094023 (2006), hep-ph/0511272.[367] A. Bianconi and M. Radici, Phys. Rev.
D72 , 074013 (2005), hep-ph/0504261.[368] A. Bianconi and M. Radici, Phys. Rev.
D73 , 034018 (2006), hep-ph/0512091.[369] A. Bianconi and M. Radici, Phys. Rev.
D73 , 114002 (2006), hep-ph/0602103.[370] M. Anselmino et al. , Phys. Rev.
D72 , 094007 (2005), hep-ph/0507181.[371] A. Sissakian, O. Shevchenko, A. Nagaytsev, O. Denisov, and O. Ivanov, Eur. Phys. J.
C46 , 147(2006), hep-ph/0512095.[372] A. Sissakian, O. Shevchenko, A. Nagaytsev, and O. Ivanov, Eur. Phys. J.