Lectures on LHC Physics

Tilman Plehn

arXiv:0910.4182v2 [hep-ph] 11 Oct 2010

Institut f¨ r Theoretische Physik u Universit¨ t Heidelberg a

arXiv:0910.4182, updated October 11, 2010

Abstract When we try to advance from a solid knowledge of field theory to LHC physics we usually encounter a frustrating problem: Higgs physics, QCD at the LHC, and phenomenological strategies appear as a impenetrable block of technical know-how, common lores, and historically grown intuition what works and what does not. These lectures should put you into a position to digest advanced writeups as well as some first research papers on the topic.

1

2

CONTENTS

Contents
1 1 Higgs physics 1.1 Electroweak symmetry breaking . . . . 1.1.1 Massive photon . . . . . . . . . 1.1.2 Standard Model doublets . . . . 1.1.3 Sigma model . . . . . . . . . . 1.1.4 Higgs boson . . . . . . . . . . 1.1.5 Custodial symmetry . . . . . . 1.2 The Standard Model . . . . . . . . . . 1.2.1 Higgs potential to dimension six 1.2.2 Unitarity . . . . . . . . . . . . 1.2.3 Renormalization group analysis 1.3 Higgs decays . . . . . . . . . . . . . . 1.4 Higgs production in gluon fusion . . . . 1.4.1 Effective gluon-Higgs coupling 1.4.2 Low-energy theorem . . . . . . 1.4.3 Signatures . . . . . . . . . . . . 1.5 Higgs production in weak boson fusion 1.5.1 Production kinematics . . . . . 1.5.2 Central jet veto . . . . . . . . . 1.5.3 Decay kinematics and signatures 1.6 Associated Higgs production . . . . . . 1.7 Beyond Higgs discovery . . . . . . . . 1.7.1 CP properties . . . . . . . . . . 1.7.2 Higgs self coupling . . . . . . . 1.8 Further reading . . . . . . . . . . . . . 5 5 5 6 9 13 15 18 18 21 23 26 27 28 31 32 36 36 38 39 40 41 41 42 44 46 46 46 49 53 54 56 59 59 63 66 68 68 70 77 82 84 87 90 90 92 96 102

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . .

2

QCD 2.1 Drell-Yan process . . . . . . . . . . . . . 2.1.1 Gauge boson production . . . . . 2.1.2 Massive intermediate states . . . 2.1.3 Parton densities . . . . . . . . . . 2.1.4 Hadron collider kinematics . . . . 2.1.5 Phase space integration . . . . . . 2.2 Ultraviolet divergences . . . . . . . . . . 2.2.1 Counter terms . . . . . . . . . . . 2.2.2 Running strong coupling . . . . . 2.2.3 Re-summing scaling logarithms . 2.3 Infrared divergences . . . . . . . . . . . . 2.3.1 Single jet radiation . . . . . . . . 2.3.2 Parton splitting . . . . . . . . . . 2.3.3 DGLAP equation . . . . . . . . . 2.3.4 Parton densities . . . . . . . . . . 2.3.5 Re-summing collinear logarithms 2.4 Scales in LHC processes . . . . . . . . . 2.5 Parton shower . . . . . . . . . . . . . . . 2.5.1 Sudakov form factor . . . . . . . 2.5.2 Soft gluon emission . . . . . . . 2.5.3 CKKW and MLM schemes . . . 2.6 NLO and parton shower . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . .

CONTENTS

3

2.7 3

2.6.1 Next-to-leading order in QCD 2.6.2 MC@NLO method . . . . . . 2.6.3 POWHEG method . . . . . . Further reading . . . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

102 105 107 110 111 111 112 113 115 118 119 120 121 128 129

LHC Phenomenology 3.1 Jets and fat jets . . . . . . . . . . . . . . . . 3.1.1 Jet algorithms . . . . . . . . . . . . . 3.1.2 Fat jets . . . . . . . . . . . . . . . . 3.2 Helicity amplitudes . . . . . . . . . . . . . . 3.3 Missing transverse energy . . . . . . . . . . 3.3.1 Measuring missing energy . . . . . . 3.3.2 Missing energy in the Standard Model 3.3.3 Missing energy and new physics . . . 3.4 Uncertainties . . . . . . . . . . . . . . . . . 3.5 Further reading . . . . . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

we are taking at least one step back and study the theory which describes Higgs production and everything else at the LHC. At the end of this part what everybody should understand is the usual set of ATLAS or CMS graphs shown in Figure 6. or missing transverse energy we cover in the third part. Many QCD issues affecting Higgs searches we will skip in the Higgs part and postpone to the. Throughout the QCD discussion we avoid the more historically interesting deep inelastic scattering process and instead rely on the Drell–Yan process or its inverted R ratio jet production for motivation and illustration. . we derive the DGLAP equation by constructing the splitting kernels. So if you think you read them very carefully and you did not email me at least a handful typos and complaints about poor explanations. To those familiar with the German system it will be obvious that the target audience of the lecture are students who know field theory and are starting their diploma or master thesis.. the literature listed at the end of each part is not meant to cite original or relevant research papers. Second. This leads us to the parton shower and to the physical interpretation of resumming different logarithms in the QCD perturbation series.QCD part. Some of them. I really do think that going through some of these mostly introductory papers will be instructive and fun once the basics have been covered by these lecture notes.. – . carefully studying these notes should put you into a position to start actual research in LHC physics. So when reading these notes. which we introduce at least on the level of simplified models. you did not read them carefully enough. One thing to keep in mind when reading is that the lectures consist of four hours a week without exercises. take a break here and there. get a coffee and think about the physics behind the calculation you just followed. like old fashioned and fat jets. so for the structure of such models and their signatures I instead refer to our review article Ref. where many colored lines represent different search channels and their discovery potential. The text is divided into three distinct parts: – In the first part we focus on Higgs physics and collider searches. To understand what we are looking for we start with the most minimalistic and not renormalizable models describing massive gauge bosons and slowly advance to the usual fundamental Higgs scalar we are really searching for. This means I calculate everything in detail on the blackboard. I collected a set of review papers or advanced lecture notes supplementing these lecture notes in different directions. Instead. there are many aspects of LHC physics we need to take into account simulating LHC events.4 CONTENTS Users manual These notes are based on a lecture at Heidelberg University in the Summer term 2009 and the Fall term 2010. Here. but not least. This is why the notes look the way they look. As the author of these notes I am confident that they are far from bug free.[1] and in particular to its second chapter. What is almost entirely missing in this review is an introduction to searches for new physics completing the Standard Model of particle physics beyond the weak scale. written up in coffee shops around the world.. Last. helicity amplitudes.. Two core discussions shape this part: first. Covering this topic appropriately would at least double the length of these notes. This part will expand over the coming years while I will keep these lecture notes up to date with actual discussions of LHC data. – Finally. there are two modern approaches combining parton shower and matrix element descriptions of jet radiation. Because this is not a text book there is less text in the notes than actual talk during the lecture.

1) 1 1 1 = − Fµν F µν + e2 f 2 Aµ − ∂µ φ 4 2 ef . The way around is also adding an innocent looking real (uncharged) scalar field without a mass and without a coupling to the photon. Only this last step will lead us to the Standard Model and the Higgs potential. so it will be gauge invariant just as it was before. backgrounds.2) under which the Lagrangian is indeed invariant: the kinetic term for the photon we left untouched. and show how we can embed this mechanism in a renormalizable fundamental gauge theory.1) . but at this stage it is a dimensionless number without any specific relevance in this interaction-less Lagrangian. 1. the Higgs boson. and the symmetries related to its implementation. we prefer to follow an effective theory approach. Even for the SU (2) gauge theory of the electroweak Standard Model we might get away without a fundamental Higgs boson. We can define a simultaneous gauge transformation of both fields in the Lagrangian Aµ −→ Aµ + 1 ∂χ ef φ −→ φ + χ . let us sketch the Standard Model Lagrangian and what we need to do to accommodate mass terms for gauge bosons and fermions.1. as we will show in Section 1. we start by breaking electrodynamics. This means we do not start by writing down the Higgs potential and deriving the measured gauge boson and fermion masses. 1.5 1 Higgs physics Understanding the nature of electroweak symmetry breaking — or slightly more specifically deciphering the Higgs mechanism — is the main goal of the ATLAS and CMS experiments at the LHC. we step by step include gauge boson and fermion masses in our gauge theories. To its kinetic F · F term we would like to add a photon mass m2 A2 /2. Therefore. The additional factor e will become the usual electric charge. where f is a common mass scale for the photon mass and the mixing term ensuring all terms in the Lagrangian have mass dimension four — remembering that bosonic fields like Aµ and φ have mass dimension one. in a first step in Section 1.1 Massive photon Even though this is not the physical problem we are interested in.1. The result is called the Boulware-Gilbert model 1 1 2 L = − Fµν F µν + (∂µ φ) + 4 2 1 1 2 = − Fµν F µν + (∂µ φ) + 4 2 1 2 2 2 e f Aµ − ef Aµ ∂ µ φ 2 1 2 2 1 e f Aµ − ∂µ φ 2 ef 2 2 − 1 2 (∂µ φ) 2 (1. (1.(1. The simultaneous gauge invariance is then defined to keep the second term in Eq.1. Only when we worry about quantum fluctuations of the relevant degrees of freedom we are led to the usual picture of the Higgs potential.1 Electroweak symmetry breaking As a starting point. before we discuss the details of Higgs signatures. giving a mass to the (massless) photon of our usual locally U (1)Q -symmetric Lagrangian. Observing some kind of Higgs boson and studying its properties involves many experimental and theoretical issues focused around understanding hadron collider data and QCD predictions to unprecedented precision.1 we will try to make our gauge bosons massive without introducing a physical Higgs field. The latter will be the main topic of the second half of this lecture. but with a scalar-photon mixing term and a non-trivial gauge transformation. Instead. Higgs physics at the LHC might well be much more than just finding a light fundamental Higgs boson as predicted by the Standard Model or particle physics. which we know is forbidden by the gauge symmetry. and related QCD aspects we should start with a discussion of electroweak symmetry breaking. As a matter of fact. see what this means for the field content of our theory. i.2. On the other hand. Because of all the dimension-4 terms our theory should be renormalizable.e.

. The only thing we have to make sure is that we combine the left. not to spoil gauge invariance. so fermion masses will be included as LD3 . (1. in spite of gauge invariance.6) 2 2 . What it illustrates is only how by introducing a neutral scalar particle without an interaction but with a mixing term we make gauge bosons heavy.(1. like a mass or a self coupling of φ. which has absorbed the real scalar φ as its additional longitudinal component.2) we immediately see that any additional purely scalar term in the Lagrangian. like a scalar potential V (φ). Dirac mass terms simply link SU (2) doublet fields for leptons and quarks with right handed singlets and give mass to all fermions in the Standard Model. This is because a massless gauge boson Aµ has only two on-shell degrees of freedom. 1.1) as 1 1 1 1 2 2 L = − Fµν F µν + e2 f 2 Bµ = − Fµν F µν + m2 Bµ 4 2 4 2 B (1. If the spontaneously broken symmetry is a gauge symmetry and its breaking induces gauge boson masses. so we know how mass terms in the renormalizable dimension-4 Lagrangian would look. An interaction term φAA would not be possible.5) We label the different parts of the Lagrangian by their mass dimension. while the massive Bµ has an additional longitudinal polarization degree of freedom. Only derivative interactions proportional to ∂φ coupling to conserved currents are fine. its broken generators correspond to physical Goldstone modes. (1. either. needs to be symmetric under the linear shift φ → φ + χ. Without any fundamental Higgs boson appearing. namely a left-handed and a right handed polarization. Such massless physical states appear in many areas of physics and are described by Goldstone’s theorem: If a global symmetry group is spontaneously broken into a group of lower rank.. What kind of properties does this field φ need to have.6 1 HIGGS PHYSICS invariant. This means that we cannot write down polynomial terms φn . This helicity flip we can easily see by introducing left-handed and right handed projectors: 1 − γ5 1 1 + γ5 1 ψL = ψ ≡ PL ψ ψR = ψ ≡ PR ψ . because the non-linear transformation only allows derivative terms in the Lagrangian. This means that our toy model is not going to be well suited to provide the three degrees of freedom needed to make SU (2) gauge bosons massive.1. so that we can use it to provide a photon mass? From the combined gauge transformation in Eq.3) and then rewrite the Lagrangian of Eq. These fields transform non-linearly under the larger and linearly under the smaller group.(1. which is why there are no degrees of freedom left after the photon gets its mass. The gauge boson mass is given by the vacuum expectation value (VEV) breaking the larger symmetry.2 Standard Model doublets One of the complications of the Standard Model is its SU (2) doublet structure. the massive photon has ‘eaten’ the real scalar field φ. LL ) and singlets (QR . these massive degrees of freedom are made out of ‘eaten’ Goldstone modes. If the smaller symmetry is also broken. In that case we can absorb the shift by χ into a total derivative in the Lagrangian. the NGBs become pseudo-NGBs and acquire a mass of the size of this hard-breaking term. This example illustrates a few vital properties of Nambu–Goldstone bosons (NGB). Fermion fields have mass dimension 3/2. LR ) properly: LD3 ∼ −QL mQ QR − LL mL LR + . They are massless and cannot form a potential.and right handed doublets (QL . In the last section we have chosen not to introduce a charged SU (2) doublet. If we now re-define the photon field as Bµ = Aµ − ∂µ φ/(ef ) we need to compare the new and the old kinetic terms Fµν B = ∂µ Bν − ∂ν Bµ = ∂µ Aν − = ∂µ Aν − ∂ν Aµ = Fµν A 1 1 ∂ν φ − ∂ν Aµ − ∂µ φ ef ef (1.4) This Lagrangian effectively describes a massive photon field Bµ .

1ψ 1ψ while the kinetic term stays diagonal ψ ∂ψ = ψ i∂ P2 + P2 ψ / / L R = ψ (PR ∂PL + PL ∂PR ) ψ / / / / = (PL ψ)∂(PL ψ) + (PR ψ)∂(PR ψ) = ψ L ∂ψL + ψ R ∂ψR . At this stage we do not need the explicit form of the gamma matrices which we will use in Eq.8) (1.R L.1. The main problem with the fermion mass terms in Eq.11) . / / (1. so we have to distinguish DL and DR ..(1. τj ] = 2i ijk τk . For our discussion of electroweak symmetry breaking it is sufficient to study one fermion generation at a time. so we will omit this complication here. Products of the Pauli matrices τ123 satisfy the relation τi τj = δij + i ijk τk or the commutator relation [τi .102). Flavor physics dealing with such 3 × 3 mass matrices is its own field of physics with its own review and lecture notes.1 Electroweak symmetry breaking 7 where ψ is a generic Dirac spinor and PL.R In general. P† = PL.(2. except that the weak interaction knows about the chirality of the fermion fields.e.R are projectors in this 4 × 4 Dirac space.j δ ij + i ijk τk = δ ij + i i=j ijk τk = δ ij + i i<j ijk + jik τk = δ ij (1. The mass term for a Dirac fermion reads ψ 1 = ψ (PL + PR ) ψ 1ψ = ψ P2 + P2 ψ L R = ψ † γ0 P2 + P2 ψ L R = ψ † PR γ 0 PL + PL γ 0 PR ψ = (PR ψ)† γ 0 (PL ψ) + (PL ψ)† γ 0 (PR ψ) = (PR ψ)1 L ψ) + (PL ψ)1 R ψ) 1(P 1(P = ψR 1 L + ψL 1 R . 4 (1. Summing over indices we see that τi τj = i. which implies that we might have to rotate the fermion fields from an interaction basis into the mass basis where these mass matrices are diagonal. The same form works for the weak SU (2) interactions. γµ } = 0 † γ5 = γ 5 . The terms 1 / / / / LD4 ∼ QL iDQL + QR iDQR + LL iDLL + LR iDLR − Fµν F µν .9) describe electromagnetic interactions introducing a covariant derivative Dµ = ∂µ + ieqAµ with the photon field also appearing in the field-strength tensor Fµν = ∂µ Aν − ∂ν Aµ .j i.10) We write the covariant derivative already in the mass eigenstates.7) with with with ψ = ψ† γ 0 {γ5 . / / The covariant derivatives in terms of the SU (2) basis matrices read DLµ = ∂µ + ieqAµ + igZ −qs2 + w DRµ = DLµ τ =0 τ3 g + − Zµ + i √ τ+ Wµ + τ− Wµ 2 2 τ+ = τ1 = 0 0 0 1 1 0 1 0 τ− = τ2 = 0 1 0 i 0 0 −i 0 τ3 = 1 0 0 −1 (1. i. the photon Aµ and the Z boson. these mass terms can be matrices in generation space..5) is that they are not gauge invariant. The interaction of fermions with gauge bosons is most easily written in terms of covariant derivatives.

Of course. To see what we need to add let us first look at the local U (1) transformations involved.µ Wµ + Z Z µ Zµ = m2 W +.5) invariant under this left handed SU (2) gauge transformation. making it more obvious how they mix with the neutral component of SU (2) to give the electric charge.g. to write a gauge-invariant Lagrangian for massive fermions we have to add something else to our minimal Standard Model Lagrangian. appropriately chosen transformation. the second set of generators reflects the fact that for SU (2) as for any SU (N ) group we can find a set of real generators of the adjoint representation.8 1 HIGGS PHYSICS The basis of three Pauli matrices we can write in terms of τ1. kinetic terms of all kinds. We start with a slightly complicated-looking way of writing the abelian hypercharge U (1)Y and electric charge U (1)Q transformations. Acting on a field with an SU (2) charge this it not a usual U (1) transformation. dependent on how picky we are with the properties of our new Lagrangian beyond the issue of gauge invariance. When we switch between the two bases we only have to make sure √ we get the factors 2 right √ + − 2 τ+ Wµ + τ− Wµ = √ 2 ⇔ + Wµ + √ 0 Wµ + 2 0 0 1 1 2 = √ Wµ − iWµ 2 0 − Wµ 0 0 1 2 = τ1 Wµ + τ2 Wµ = ! 0 1 Wµ 1 Wµ 0 + 0 2 iWµ 2 −iWµ 0 1 − 1 2 Wµ = √ Wµ + iWµ 2 (1.−. Note that this addition does not have to be a fundamental scalar Higgs field.17) β = exp i y 1 1 2 . U QL mQ QR → QL U −1 mQ QR = QL mQ QR (1. where one of the fermion field picks up a factor U and the other is unchanged.12) To track these numbers of two in the definitions of the weak gauge field we have a close look at is the dimension-2 mass term for charged and neutral gauge bosons m2 m2 m2 1 2 − W W 1.3 as well as in terms of τ+. The latter form of the generators corresponds to the two charged and one neutral vector bosons.16) In analogy to the massive photon case.µ Wµ + W 2. LD2 ∼ To understand the issue with fermion masses. but they do not affect our discussion of U (1)Y and SU (2)L gauge invariance. This electric charge transformation q is proportional to the unit matrix and hence commutes with all other matrices i exp(iβq) V † = exp(iβq) exp − βτ3 2 = exp iβ y 1 + τ3 1 2 i exp − βτ3 2 with q≡ y 1 + τ3 1 2 yQ = 1 3 yL = −1 (1. we first look at the local weak SU (2)L transformation U (x) = exp −iαa (x) τa 2 ≡ e−i(α·τ )/2 (1.2. in the complete Standard Model Lagrangian there are many additional terms.14) which only transforms the left handed fermion fields and leaves the right handed fields untouched LL → U LL LR → LR U U QL → U QL QR → QR U U (1. While the usual basis is written in terms of complex numbers. We know already that the problems with gauge invariance lies in these dimension-2 and dimension-3 mass terms.13) W 2 2 2 √ The relative factor two in front of the W mass corresponds to the factors 1/ 2 in the SU (2) generators τ± . What we can do is combine it with another.µ Wµ + Z Z µ Zµ .3 . Let us start with the neutral component of the SU (2) transformations V = exp (iβτ3 /2). e.15) It is obvious that there is no way we can make left-right mixing fermion mass terms as shown in Eq.(1. (1.

1 Electroweak symmetry breaking 9 Acting on a given field the factor τ3 in this formula is replaced by its eigenvalue ±1 for up-type and down-type fermions. without saying anything about possible representations of Σ for example in terms of physical fields LD3 ∼ −QL ΣmQ QR − LL ΣmL LR + h.22) In the second step.18) Under a combined SU (2)L and U (1)Y transformation the left handed fermions see the hypercharge. (1. As required by the above argument. The bottom line is that including a Σ field in the fermion mass terms gives a U (1)Y and SU (2)L -invariant Lagrangian. properly combined with the isospin they give the correct electric charges qQ. (1. Since τ3 and the unit matrix commute with each other the combined exponentials have not additional factor a la Baker–Campbell–Hausdorff.19) If the result should be a dimension-4 Lagrangian the mass dimension of Σ has to be m0 = 1.1. (1. Left and right handed quark and lepton fields transform under this U (1)Y symmetry as β 1 LL → exp (iβqL ) V † LL = exp i yL 1 LL 2 LR → exp (iβqL ) LR β QL → exp (iβqQ ) V † QL = exp i yQ 1 QL 1 2 QR → exp (iβqQ ) QR (1.21) For any Σ with this property the LD3 part of the Lagrangian has the required U (1)Y × SU (2)L symmetry.23) . 2 2 2 2 (1.. Its physical properties will become clear from its appearance in the Lagrangian and the requirement of gauge invariance. Following the last section. we first introduce Σ into the fermion mass term to see what it takes to turn this mass term gauge invariant under the weak SU (2) transformation QL ΣmQ QR → QL U −1 Σ(U ) mQ QR = QL ΣmQ QR U ! ⇔ Σ → Σ(U ) = U Σ . The equation of motion for the Σ field will then also have to follow from the way we introduce it in the Lagrangian. starting with the left handed covariant derivative DLµ = ∂µ + ig q− τ3 y a τa a τa Bµ + igWµ = ∂µ + ig Bµ + igWµ . This field will in some way play the role of the real scalar field we used for the photon mass generation. The U (1)Y charges or quantum numbers y are the quark and lepton hypercharges of the Standard Model.c. Just as for the SU (2) transformation U we do not even have to compute anything to see that such different transformations of the left handed and right handed fermion fields do not allow for a Dirac mass term.(1. independent of what this field really means.3 Sigma model One way of solving this problem which at this point really looks like a cheap trick is to introduce an additional field Σ(x).1. The same we can do for the U (1)Y transformation V β V 1 QL ΣmQ QR → QL exp −i y 1 Σ(V ) mQ exp (iβq) QR 2 β =QL Σ(V ) exp −i y 1 exp (iβq) mQ QR 1 2 =QL Σ(V ) V mQ QR =QL Σ mQ QR ! β exp i y 1 always commuting 1 2 ⇔ Σ → Σ(V ) = ΣV † (1. while the right handed fermions only see the electric charge.. + . we worry about the gauge boson masses. From the way it transforms Σ is a 2 × 2 matrix with mass dimension zero.L .19) gives us the required transformation property Σ → U ΣV † . 1.20) which together with Eq.

29) This choice is called unitary gauge.24) acting on a constant field we can compute the auxiliary field Vµ in unitary gauge a Vµ = Σ(Dµ Σ)† = −igWµ τa τ3 + ig Bµ 2 2 τ3 + τ+ − τ− 3 τ3 = −igWµ √ − igWµ √ − igWµ + ig Bµ 2 2 2 2 g τ3 + − = −i √ Wµ τ+ + Wµ τ− − igZ Zµ 2 2 (1. Certainly. v. First. Eq.27) with properly chosen prefactors µ.20). (1. Clearly. adding a circular symmetric trace Tr(Σ† Σ) → Tr(V Σ† ΣV † ) = Tr(Σ† Σ) gives a gauge invariant combination.V (1.10 1 HIGGS PHYSICS Instead of deriving the gauge transformation of Σ let us start with a well-chosen ansatz and work backwards step by step. as to check that we indeed arrive at the correct masses. the prefactors will just be µ and λ. Using the covariant derivative from Eq. but we can always work in a specific gauge and only check the final result for gauge invariance. This fourth term finalizes our construction of the relevant weak Lagrangian L = LD2 + LD3 + LD4 + LΣ . meaning terms with no derivatives LΣ = − µ2 v 2 λv 4 Tr(Σ† Σ) − Tr(Σ† Σ) 4 16 2 + ··· . The definitely simplest way to achieve this is to assume Σ(x) = 1 1 (1. However. while λ has mass dimension zero. The parameter µ and the factor v appearing with every power of Σ have mass dimension one. On the other hand.26) is forbidden by gauge invariance according to the SU (2) transformation of the sigma field. What we need to show next is that this form is gauge invariant and gives the correct masses. which allows for the additional potential terms. It looks like a joke to introduce Σ(x) = 1 and ask for gauge invariance.24) If we introduce the abbreviations Vµ ≡ Σ(Dµ Σ)† and T ≡ Στ3 Σ† we claim we can write the gauge boson mass terms as v2 v2 LD2 = − Tr[Vµ V µ ] − β Tr[T Vµ ] Tr[T V µ ] .25) 4 8 The trace acts on the 2 × 2 SU (2) matrices. Σ = 1 gives the correct fermion 1 masses in LD3 and makes the potential LΣ an irrelevant constant.(1. (1. To check for the correct masses in the Standard Model we start with Tr(Σ† Σ) and assume it acquires a finite (expectation) value after we properly deal with Σ. we postulate that the covariant derivative acting on the Σ field in the gauge-symmetric Lagrangian has to be Dµ Σ = ∂µ Σ − ig ΣBµ τ3 a τa + igWµ Σ . A first question is what additional terms of mass dimension four we can write down using the dimensionless field Σ and which are gauge invariant. λ. 2 2 (1. a 1 constant does not exhibit the correct transformation property under the U and V symmetries. Higher dimensional terms in a dimension-4 Lagrangian are possible as long as we limit ourselves to Tr(Σ† Σ). As rule of thumb we will later notice that once we express this potential in terms of the usual Higgs doublet |φ|2 .(1.28) organized by mass dimension. Our first attempt of a building block Σ† Σ → (U ΣV † )† (U ΣV † ) = V Σ† U † U ΣV † = V Σ† ΣV † = Σ† Σ U. What we need to check is LD2 which is supposed to reproduce the correct gauge boson masses. (1.30) . they lead to higher powers in v which we will see makes them higher-dimensional operators in our complete quantum theory.

Tr τ± τ = 1. but it leads to the most compact Lagrangian.7) from experiment.1. 2 (1. The second mass term in LD2 proportional to β is similarly simple in unitary gauge T = Στ3 Σ† = τ3 ⇒ Tr(T Vµ ) = Tr −igZ Zµ 2 τ3 2 = −igZ Zµ Tr(1 1) = −igZ Zµ 2 (1. (1.35) The role of a possible additional and unwanted Z-mass contribution β we will discuss in Section 1. In the canonical normalization we write 1 Tr(Σ† (x)Σ(x)) = 1 2 ∀x . Combining both terms yields the complete gauge boson masses LD2 = − g2 v2 v2 + 2 −g 2 Wµ W −µ − Z Zµ Z µ − β −gZ Zµ Z µ 4 2 8 2 2 v 2 g 2 + −µ v 2 gZ v 2 gZ = Wµ W + Zµ Z µ + β Zµ Z µ 4 8 8 2 v 2 g 2 + −µ v 2 gZ Wµ W + (1 + β ) Zµ Z µ .(1. What is not obvious from this argument is that we can actually write the ratio g /g in terms of a rotation angle. so they are all physically equivalent — as they should be as gauge choices. (1.31) The result in Eq. and Tr(τ3 ) = 2.13) and assuming universality of neutral and charged current interactions (β = 0) we find mW = gv 2 mZ = 1+β gZ v β =0 gZ v = 2 2 (1. Let us just recapitulate what we did until now — using this Σ field and its finite constant value Σ = 1 we have made 1 the fermions and electroweak gauge boson massive.(1. which implicitly assumes that we can rotate the B and W 3 fields into the physical mass-eigenstate photon and Z fields Aµ Zµ = cw −sw sw cw Bµ 3 Wµ .34) Identifying the masses with the form given in Eq. Given that we know the heavy gauge boson masses (mW ∼ 80 GeV) and the weak coupling (g ∼ 0.1.36) .5 on custodial symmetry. Every parameterization of Σ with this property will lead to the same massive gauge bosons. The actual minimum assumption to obtain finite gauge boson masses is Tr(Σ† (x)Σ(x)) = 0 in the vacuum. We should for example be able to see the additional degrees of freedom of the longitudinal gauge boson modes if we weaken our unitary gauge assumption. Choosing this constant finite field value for Σ is not the only and not the minimal assumption needed to make the gauge bosons heavy.32) 2 2 using τ± = 0. Tr(τ3 τ± ) = 0.33) 2 ⇒ Tr(T Vµ ) Tr(T V µ ) = −gZ Zµ Z µ . From the photon mass example.30) gives us the first of the two terms in the gauge boson mass Lagrangian LD2 Tr[Vµ V µ ] = −2 g2 g2 + − µ 2 Wµ W Tr(τ+ τ− ) − Z Zµ Z µ Tr(τ3 ) 2 4 g2 µ + = −g 2 Wµ W − − z Zµ Z µ . these relations experimentally tell us v ∼ 246 GeV. = 4 8 (1. however.1 Electroweak symmetry breaking 11 3 with Zµ = cw Wµ − sw Bµ and the two coupling constants gZ = g/cw and g = gsw /cw . we know that there must be more to this mechanism.

Because the relation between Σ and w is not linear. Even though this solution still forbids fluctuations which in the original condition on the expectation value only vanish on average. (1. The normalization scale v fixes the relevant energy scale of our Lagrangian. these gauge choices are physically equivalent. for example. If we break SU (2)L × U (1)Y → U (1)Q we expect three Goldstone bosons which become part of the weak gauge bosons and promote those from massless gauge bosons (with two degrees of freedom each) to massive gauge bosons (with three degrees of freedom each). These fields are a set of scalar Nambu-Goldstone modes. (1. (1.38) v where w(x) has mass dimension one which is absorbed by the mass scale v.40) The different ways of writing the Σ field in terms of the Pauli matrices cannot have any impact on the physics. w . 1 − 2v v 6v (1. in contrast to Σ(x) = 1 it allows a non-trivial x dependence.. which we will use later. Before we move on and introduce a physical Higgs boson we briefly discuss different gauge choices and the appearance of Goldstone modes. From the photon mass example for Goldstone’s theorem we know that these fields will become the missing degrees of freedom for the three now massive gauge bosons W ± and Z. = 1 − 2 wa wa ± . Using the commutation properties of the Pauli matrices we can expand Σ as i 1 (−1) 1 i Σ=1 − w+ 1 wa τa wb τb + wa τa wb τb wc τc v 2 v2 6 v3 1 i i 1 = 1 − w − 2 wa wa 1 + 3 wa wa w 1 v 2v 6v i 1 1 1 1 − 2 wa wa ± ..37) This means that Σ(x) is a unitary matrix which like any 2 × 2 unitary matrix can be expressed in terms of the Pauli matrices.. in which we never see Goldstone modes. something has to compensate. However. The unitary matrix can be parameterized as i Σ(x) = exp − w(x) with w(x) = wa (x)τa .. In the general (R − ξ) gauge we can actually see these Goldstone modes appear in the gauge boson propagators ∆µν (q) = VV −i qµ qν g µν + (ξ − 1) 2 2 +i − mV q − ξm2 V −i qµ qν g µν − 2 q 2 − m2 + i mV V −i g µν q 2 − m2 + i V −i qµ qν g µν − 2 2 +i q 2 − mV q q2                unitary gauge ξ → ∞ Feynman gauge ξ = 1 Landau gauge ξ = 0 (1. this is referred to as a non-linear representation of the Σ field.12 1 HIGGS PHYSICS which instead of our previous Σ(x) = 1 we can also fulfill through 1 Σ† (x)Σ(x) = 1 1 ∀x . This is the point of view of the unitary gauge. for the fact that in Feynman gauge the whole Goldstone term vanishes and the polarization sum looks like a massless . Yet another way of parameterizing the unitary field Σ in terms of the Pauli matrices is the properly normalized Σ(x) = 1 wa wa 1+ v2 i 1 − w(x) v .39) From this expression we can for example read off Feynman rules for the longitudinal gauge fields w.41) = Obviously.

additional propagating scalar degrees of freedom −i .43) as long as this physical field H has a vanishing vacuum expectation value and therefore 1 Tr(Σ† Σ) = 2 1+ H v 2 =1 (1. In other words.e.2.. From the form of the Goldstone propagators we can guess that they will indeed cancel the second term of the gauge boson propagators. The key is the Goldstone propagator. We will have a look at them in Section 1. In the static limit H = 0 we recover the vacuum condition Tr(Σ† (x)Σ(x))/2 = 1. so for a given problem one or the other might be the most efficient to use in computations or proofs. H = 0. (1.1. Tait is the guy who was hired by Edinburgh when James Clerk Maxwell was also applying for positions and the University of Edinburgh decided to not actually hire their arguably most famous son now watching over George Street. ∆V V (q 2 ) = q2 1. Loop calculations might be more efficient in Feynman gauge. but they are not realized at tree level in the Standard Model either. because of the simplified propagator structure.27) LΣ = − µ2 v 2 2 1+ H v 2 − λv 4 4 1+ H v 4 + . even a different particle content. following the original form Eq. These different gauges have different Feynman rules and Green’s functions.. W + ). i. By the way. which means this field spontaneously breaks the electroweak symmetry not by explicit terms in the Lagrangian but via the vacuum and relying on some origin of this finite VEV.44) The factor in front of the fluctuation term H/v is not fixed until we have properly defined the physical Higgs field and made sure that its kinetic term does not come with an unexpected prefactor different from unity. If we are interested in physic at or above the weak energy scale some kind of ultraviolet completion of the Σ model should tell us what the Σ field’s properties as a quantum object are.45) The dots stand for higher-dimensional terms which might or might not be there. we need to allow for quantum fluctuations of Tr(Σ† Σ) around the vacuum value Tr Σ† Σ = 2. The potential terms LΣ define a potential for this new Higgs field H. The Goldstone mass ξmV depends on the gauge: in unitary gauge the infinitely heavy Goldstones do not propagate (∆V V (q 2 ) → 0). (1. while in unitary gauge we can see the effect of these modes directly.42) − ξm2 + i V √ for both heavy gauge bosons (V = Z. the Higgs field should be added to Σ such that it matches this Goldstone. the . Some of them are not forbidden by any symmetry. etc. (1.1 Electroweak symmetry breaking 13 gauge bosons. the proof of renormalizability was first formulated in unitary gauge. If we consider our Σ model itself the fundamental theory and promote the Σ field to a quantum field like all other Standard Model fields. while in Feynman gauge and in Landau gauge we have to include them as particles. as we will see later in this section and then in more detail in Section 1. Interpreting the fluctuations around the non-trivial vacuum as a physical Higgs field is the usual Higgs mechanism. Tree-level helicity amplitudes are usually computed in unitary gauge.1. We can parameterize these new degrees of freedom as a real scalar field Σ → 1+ H v Σ .1.4 Higgs boson At this stage we have defined a perfectly fine electroweak theory with massive gauge bosons. On the other hand. All we need is a finite vacuum expectation value for Σ. For example. named after one of the University of Edinburgh’s most famous sons. it should read Tait Institute.. if we assume that the neutral Goldstone mode w3 has the correct normalization.2. in his famous paper Peter Higgs’ address is wrongly spelled Trait Institute.(1. The non-dynamical limit of this Higgs ansatz is indeed our sigma model in unitary gauge Σ† Σ = 1 equivalent to 1. while many QCD processes benefit from an explicit projection on the physical external gluons..

e. For example.49) (1. But type-II two-Higgs doublet models where one Higgs doublet gives mass to up-type and another one to down-type fermions are much more general than that. Such a non-fundamental Higgs field is the basis for so-called strongly interacting light Higgs models where the Higgs field is a light composite field with a different canonical normalization as compared to a fundamental scalar. implying −µ2 = λv 2 . (1. Counting degrees of freedom we can try to write Σ as a complex doublet with four degrees of freedom. . Any properties of the Σ field as a quantum field are left to the ultraviolet completion. In the gauge boson mass terms they appear as 1 LD3 ∼ −QL mQu Σu 1 + τ3 1 (1.52) This two-Higgs doublet model is for example the minimal choice in supersymmetric extensions of the Standard Model. The over-all factors 1/ 2 in the definition of the doublets are purely conventional and sometimes lead to confusion when for some people v = 246 GeV while for others v = 174 GeV. Apart from asking for higher precision and quantum corrections. The latter choice is a little less ˜ common but has the numerological advantage of v ∼ mt . This way. three of which are eaten Goldstones and one is the fundamental Higgs scalar.47) w2 − iw1 v + H + iw3 2 2 In this formula the normalization of the two real scalars w3 and H is indeed the same. d φj = φa τa . i.2. In the Standard Model the two Higgs doublets φ and φ give mass to up-type and down-type fermions.. which has to decide for example if Σ is a fundamental or composite field. the effective sigma model clearly breaks down at large enough energies which can excite the fluctuations of the sigma field and for example produce a Higgs boson. All we need is the finite vacuum expectation value for Σ to spontaneously break electroweak symmetry. But as mentioned already.51) which means that the longitudinal vector bosons are w = cos β φu + sin β φd . we do not distinguish between the invariant Tr Σ† Σ and its expectation value.46) = Σ→ 1+ w2 − iw1 v + H + iw3 v v v v The last step is just another way to write the 2 × 2 matrix in terms of the two SU (2) doublets containing the physical Higgs field and the Goldstone modes for the massive vector bosons W and Z 1 1 v + H − iw3 −w2 − iw1 ˜ φ= √ φ= √ . we will have the entire Section 1..14 1 HIGGS PHYSICS minimum of our potential occurs at H = 0. Remember one last time how we got to the Higgs mechanism from a static gauge invariant theory. 2 2 Each of the two Σ fields we can express in the usual representation 1 i Σj = 1 + φ 0 − φ j 1 j = u. we can write √ i 1 2 ˜ H v + H − iw3 −w2 − iw1 1 − w= 1 (φφ) . A last side remark — instead of deriving both relevant doublets for up and down-type fermion masses from one ˜ physical Higgs field φ and φ we can include two sigma fields in the fermion-mass terms 1 − τ3 1 QR − QL mQd Σd QR + . 2 2 with the isospin projectors (1 ± τ3 )/2.1 on the Higgs potential.48) 2 vu v2 Tr (Dµ Σu )† Dµ Σu + d Tr (Dµ Σd )† Dµ Σd . j vj j vj LD2 = (1. This is the job of the LHC. the Higgs field could just be one step in a ladder built out of effective theories. For this symmetry breaking we do not care about quantum fluctuations of the Σ field. From an effective field theory point of view we managed to introduce the Goldstone modes and with them gauge boson masses without introducing a fundamental Higgs scalar. so their kinetic terms will be of √ the same form. which is designed and built to take us into an energy range where we can observe the structure of electroweak symmetry breaking beyond the effective theory and the Goldstone modes. the Σ model. (1. (1. (1.50) From the gauge-boson masses we know that 2 2 v u + vd = v 2 ⇔ vu = v sin β vd = v cos β .

35) we will see that not only higher energies.1. while the Z propagator includes pure bottom and top loops. w m2 gZ Z In general. The mass values coming from Tr[Vµ V µ ] give mW and mZ proportional to g ≡ gW and gZ . The reason is that the self-energy loops in the W propagator involves a mixture of the bottom and top quark. together with the assumption that GF or g universally govern charged current (W ± ) and neutral-current (W 3 ) interactions. the general renormalizable Lagrangian for the gauge boson masses involves two terms.25. What we can hope for is that this symmetry is at least a good symmetry in the SU (2)L gauge sector and slightly broken elsewhere. but also higher precision leads to a breakdown of the effective sigma model. therefore break the symmetry . i. To bring our Lagrangian into agreement with measurements we better find a reason to constrain β to zero. and the SU (2) × U (1) gauge symmetry unfortunately does not do the job.1. the vacuum expectation value for Σ breaks SU (2)L × SU (2)R to the diagonal subgroup SU (2)L+R .3 we know that electroweak symmetry breaking by a sigma field or Higgs doublet links the couplings of neutral and charged currents firmly to the masses of the W and Z bosons. One possibility along those lines is to replace the SU (2)L × U (1)Y symmetry with a larger SU (2)L × SU (2)R symmetry.56) The last step.55) Tr(Σ Σ) → Tr V Σ U U ΣV = Tr(Σ Σ) This symmetry allows all terms in the Higgs potential LΣ . On the other hand. Giving the Σ field a finite vacuum expectation value changes the picture: in the minimal (non-Higgs) version and in the unitary gauge the Σ field now reduces to 1 which 1. mb = mt .(1. the Higgs boson.1. we can introduce a free parameter ρ which breaks this relation 2 2 gZ → gZ ρ (1.54) and which from measurements is very strongly constrained to be unity.1. Looking ahead. i. At some point we start seeing how the relative size of the W and Z masses are affected by quantum fluctuations of the sigma field. This means we are looking for an approximate symmetry of the Standard Model. both symmetric under SU (2) × U (1) and hence allowed in the electroweak Standard Model. As mentioned above. Different fermions masses in SU (2) doublets. this additional SU (2) symmetry is broken at the one-loop level.1 Electroweak symmetry breaking 15 1. we will find that in the Standard Model ρ = 1 is actually violated at the one-loop level. In LD2 the Z-mass term proportional to β precisely predicts such a deviation ρ = 1 + β = 1.e. The technical term is precisely defined this way — the two SU (2) symmetries reduce to one remaining symmetry which can be written as U = V . At tree level this relation is g2 m2 W = 2 = c2 . From the construction in Section 1. while under the diagonal symmetry group only we can accommodate a general ρ. (1. which would have to act like Σ → U ΣV † † † U ∈ SU (2)L † † † V ∈ SU (2)R because of circular trace (1. In the extended symmetry group the ρ parameter is indeed protected to be ρ = 1.e.5 Custodial symmetry Analyzing the appearance of β in Eqs. the symmetry requirement for the Lagrangian can only be satisfied if we require U = V . In other words. Their relative size can be expressed in terms of the weak mixing angle θw .53) mZ → mZ √ ρ. for the combined SU (2) transformations means Σ → U ΣV † = U 1 † = U V † = 1 1V 1 ! (1.

It is part of an effective theory parameterization of deviations from the tree-level relations between gauge boson masses. Including additional fields in a model can even turn around this prediction and prefer a heavy Higgs state. mixing angles.16 1 HIGGS PHYSICS protecting mW /mZ . Even though the Higgs contributions depend on the Higgs mass only logarithmically. Skipping the loop calculation we quote their different contributions to the gauge boson masses as m2 3GF m2 m2 ∆ρ ∼ √ m2 + m2 − 2 2 t b 2 log t t b mt − mb m2 8 2π 2 b 3GF 2 2 21 + δ ∼ √ 2mb + mb δ − 2mb log (1 + δ) m2 = m2 (1 + δ) t b δ 8 2π 2 1 δ2 δ3 3GF 2m2 + m2 δ − 2m2 +1 δ− + + O(δ 4 ) = √ b b b δ 2 3 8 2π 2 2 3GF m2 2 + δ − 2 − 2δ + δ + δ 2 − δ 2 + O(δ 3 ) = √ b 3 8 2π 2 1 2 3GF m2 = √ δ + O(δ 3 ) b 3 8 2π 2 GF 2 m2 − m2 + · · · = √ t b 8 2π 2 m2 b (1. As a matter of fact. for example the contribution from the different bottom and top masses can be translated into a contribution α∆T ∼ −β .1. First. centered around T = 0. where we first saw Goldstone’s theorem at work. For any other model it might point to something similar to a light Higgs field. A second contribution to β or the ρ parameter therefore comes from Higgs loops themselves because of g = 0 ∆ρ ∼ − 11GF m2 s2 m2 √ Z w log H . and neutral and charged current couplings. Even more importantly. One last question is: how do physical modes which we introduce as Σ(x) = exp(−iw/v) transform under the two SU (2)LR symmetries which make up the custodial symmetry? This goes back to the example of the photon mass. δρ is experimentally very strongly constrained by electroweak precision measurements. we can then derive an upper bound on the Higgs mass of the order of O(200 GeV). T. for example the mass ratio mW /mZ depends on all three parameters S. In addition. as shown in Figure 1. namely the T parameter. but does not have to be fundamental. from the covariant derivative Dµ Σ including a single τ3 we can also guess that the complete SU (2)R symmetry will not allow B-field interactions which are proportional to sw ∼ 1/4. U . which means that alternative models for electroweak symmetry breaking usually include the same kind of approximate custodial symmetry by construction. This strongly suggests that if we are faced (mostly) with the Standard Model at the weak scale the Tevatron and at the LHC will be looking for a fairly light Higgs boson — or something that very much looks like a light fundamental Higgs boson However.(1. this constraint is responsible for the almost death of (technicolor) models which describe the Higgs boson as a bound state under a new QCD-like interaction. Typical experimental constraints form an ellipse in the S vs T plane along the diagonal. In general. However. in the Standard Model we can measure the symmetry violations from the heavy quarks and from the Higgs sector shown in Eqs.58) in electroweak precision measurements.57. m2 24 2π 2 Z (1.57) In the Taylor series above the assumption of δ being small is of course not realistic. we should firmly keep in mind that the ρ parameter only points to a light fundamental Higgs boson if we assume the Standard Model Higgs mechanism. There are two reasons to discuss these loop contributions breaking the custodial symmetry in the Standard Model. . but the result is nevertheless instructive: the shift vanishes very rapidly for mt ∼ mb .58) There is even a third parameterization of the same effects.

(w · τ )] = 2iτ (α × w) . The crucial conclusion is the same. A linear transformation for example of a scalar means that we can write a potential for this particle which is symmetric under SU (2)L+R transformations. The same way we can evaluate the transformation of these physical modes under the custodial symmetry group SU (2)L+R with Σ → U ΣU † and instead find the linear transformation wa → wa = wa − εabc αb wc . so for general transformations we might end up with higher terms in α. Note that we have derived this shift-symmetry operation only for infinitesimal transformations.1.1 Electroweak symmetry breaking 17 6 5 4 July 2010 mLimit = 158 GeV Theory uncertainty ∆α(5) had = 0. As discussed in Section 1. gauge interactions.(1.60) Eq.59) In the second line we have used the Baker-Campbell-Hausdorff formula eA eB = eA+B e[A. τj ] = 2iεijk τk ⇒ ⇒ (α · τ )(w · τ ) = α · w + iτ (α × w) [(α · τ ). low Q2 data ∆χ2 3 2 1 0 Excluded 30 Preliminary 100 300 mH [GeV] Figure 1: Allowed range of Higgs masses in the Standard Model after taking into account electroweak precision data. Figure from the LEP electroweak working group. (1. The transformation properties of the Goldstone modes w follow from U Σ = e−i(α·τ )/2 e−i(w·τ )/v = e−i(α·τ )/2−i(w·τ )/v e−i[(α·τ )/2. and quadratic potential terms for these modes in Σ.00035 0.58). .02749±0.1 only derivative terms like the kinetic term and derivative couplings are allowed under the SU (2) symmetry.61) In other words.02758±0. Yukawa couplings. (1. Under the usual SU (2)L symmetry we know the transformation Σ → U Σ with U = exp(−iα · τ /2). Eq.(w·τ )/v]/2 = e−i(α·τ )/2−i(w·τ )/v eτ ·(α×w)/(2v) ∼ e−i((αv/2+w)·τ )/v = e−i(w ·τ )/v ! implying v w→w =w+ α . while the approximate symmetry SU (2)L+R leads to a linear transformation.1.59) is a non-linear transformation. 2 (1. since wa is not proportional to wa .00012 incl. their transformation is not linear.B]/2 which for the Pauli matrices includes the commutator [τi . This corresponds to our initial observation that LΣ is indeed symmetric under custodial symmetry. though: the Goldstone modes shift under the SU (2) transformation.(1. most notably the ρ parameter contribution from the Higgs itself. when we transform the physical modes corresponding to the symmetry generators in Σ by the good symmetry SU (2)L we find a non-linear transformation. When we construct a symmetric Lagrangian this non-linear symmetry transformation forbids mass terms.

If we expand the possible mass dimensions and the operator basis.1 we include higher dimensional operators on top of the usual renormalizable (dimension-4) operators in the Higgs potential. i. but it violates the custodial symmetry and leads to a very large contribution to the electroweak precision parameter ∆S = −f3 v 2 /(2Λ2 ). there are exactly two gauge-invariant operators of dimension six we can write down in terms of the Higgs doublet |φ|2 . but their effects are often small or assumed to be small. parameterized as the free mass scale Λ 2 LD6 = i=1 fi Oi . even though from the previous discussion it is clear that higher dimensional terms — stemming from higher powers of Tr(Σ† Σ) — can and should exist.64) The prefactors in the Lagrangian are conventional. as long as the typical energy scale E in the numerator in our matrix element is small (E Λ).2 The Standard Model Before we discuss all the ways we can look for a Higgs Boson at the LHC we should briefly review the Higgs mechanism in the Standard Model. Starting from the Higgs doublets introduced in Eq. but electroweak precision data tells us whatever it is the Higgs should look very similar to a light fundamental scalar.(1. we need to take the Standard Model seriously and in turn find constraints on the structure and the validity of our Standard Model with a fundamental Higgs boson.2. the corrections from the additional operators are small as well.18 1 HIGGS PHYSICS 1.47) and for now ignoring the Goldstone modes the simplified doublet 1 1 −w2 − iw1 0 ∼√ (1. before electroweak symmetry breaking O1 = 1 ∂µ (φ† φ) ∂ µ (φ† φ) 2 1 O2 = − (φ† φ)3 .1 Higgs potential to dimension six In the renormalizable Standard Model all terms in the Lagrangian are defined to be of mass dimension four. Before we compute the Higgs potential including O2 we look at the effects of the dimension-6 operator O1 . 3 (1. (1. Such large contributions to S are firmly ruled out by electroweak precision data. unless we see some seriously new states and effects around the weak scale. like ¯ ¯ mf ψψ or ψ∂µ ψ or Fµν F µν . in Section 1.2 and 1. The same renormalizability assumption we usually make for the Higgs potential.2. 1. In the last sections we have seen that there does not really need to be such a fundamental scalar. so we ignore this operator in our analysis.63) To emphasize that renormalizability is a strong and not necessarily very justified theoretical assumption in LHC Higgs physics.2. To make it a little more interesting and since we are already in the mood of not taking the Standard Model Higgs sector too literally. There is in principle one more possible operator (Dµ φ)† (φ† φ)(Dµ φ). because to construct a Lagrangian we have to multiply these operators with general coefficients of mass dimension minus two. now written in terms of the Higgs doublet and with µ2 and λ as prefactors VSM = µ2 |φ|2 + λ|φ|4 + const .65) The scale Λ suppresses these additional operators.2.e. In other words.3. It . like we will do in Sections 1.62) φ= √ v + H + iw3 v+H 2 2 leaves us with only two possible potential terms. Such operators generally occur in effective theories based on ultraviolet completions of out Standard Model. This mass dimension is easy to read off if we remember that for example scalar fields or vector-boson fields contribute mass dimension one while fermion spinors carry mass dimension 3/2. Once we want to analyze the behavior of the Higgs sector over a wide range of energy scales. we allow for more operators in the Higgs potential. Λ2 (1.

68) this gives us the Standard Model Higgs mass m2 H SM 2 = −2µ2 = 2λv0 .70) defining the minimum position |φ|2 = v 2 /2.66) ˜ We use the symbol H for the Higgs field as part of φ.1.(1. before or after symmetry breaking O1 = 1 ∂µ (φ† φ) ∂ µ (φ† φ) 2 ˜ 1 (H + v)2 ∂µ = ∂µ 2 2 = 1 ˜ ˜ ˜ (H + v)2 ∂µ H ∂ µ H . f2 f2 (1. (1.2 The Standard Model 19 contributes to the kinetic term the Higgs field in the Lagrangian. because from the formula above it is clear that there will be a ˜ difference between H and the physical Higgs field H at the end of the day.72) . The mass of the Higgs boson will be given by all quadratic terms from the potential.69) The positive sign in the last term of the potential V corresponds to a negative sign in the Lagrangian and ensures that for f2 > 0 the potential is bounded from below for large field values φ. The regular dimension-4 ˜ contributions in terms of the shifted Higgs field H are VSM = µ2 ˜ ˜ (H + v)2 (H + v)4 µ2 ˜ 2 λ ˜ +λ = H ··· + · · · 6H 2 v 2 · · · 2 4 2 4 ∼ µ2 3λv 2 + 2 2 ˜ H2 (1. The contribution from O1 leaves us with a combined kinetic term Lkin ∼ f1 v 2 1 ˜ µ˜ ∂µ H∂ H 1 + 2 2 Λ = ! 1 ∂µ H ∂ µ H 2 ⇔ H= 1+ f1 v 2 ˜ ˜ H≡N H . 2 ˜ (H + v)2 2 (1.71) 2 The first solution we have expanded around the Standard Model minimum. The two solutions of the quadratic equation for v 2 /2 are v2 λΛ2 =− ± 2 f2 λΛ2 f2 2 µ2 Λ2 − f2 1 2 = λΛ2 −1 ± f2 1− µ2 f2 Λ2 λ2 λΛ2 f 2 µ4 f2 µ2 −1 ± 1 − 2 2 − 24 4 + O(Λ−6 ) f2 2λ Λ 8λ Λ  2 4 f2 µ µ2 f2 µ2  µ  − 3 2 + O(Λ−4 ) = − 1 + 2 2 + O(Λ−4 )  − 2λ 8λ Λ 2λ 4λ Λ =  2λΛ2   − 2 + O(Λ0 ) f2 ≡ 2 v0 2 1+ 2 f2 v0 + O(Λ−4 ) 4λΛ2 (1. Λ2 (1. 3Λ2 (1. Including the additional dimension-6 operator O2 we can write the Higgs potential as V = µ2 |φ|2 + λ|φ|4 + f2 |φ|6 . v0 = −µ2 /λ. v and λ from the minimization condition of the Higgs potential.68) The term in the parentheses can be simplified once we know the relation between µ. but it also means we have to make ˜ sure we replace H with H in the entire Higgs sector. The non-trivial minimum at φ = 0 is given by ∂V 3f2 4 ! = µ2 + 2λ|φ|2 + |φ| = 0 2 ∂ |φ| 3Λ2 ⇔ |φ|4 + 2λΛ2 2 µ2 Λ2 ! |φ| + =0.67) This is a simple rescaling to define the canonical kinetic term in the Lagrangian. Inserted into Eq.

(1.71) ˜ replacing H using Eq.(1.68) gives us the complete mass term to dimension six Lmass = − µ2 ˜ 2 3 2 ˜ 2 f2 15 4 ˜ 2 H − λv H − 2 v H 2 2 Λ 24 1 5 f2 v 4 ˜ 2 =− H µ2 + 3λv 2 + 2 4 Λ2 1 f2 v 2 5 f2 v 4 =− −λv 2 1 + + 3λv 2 + 2 2 4λΛ 4 Λ2 =− 1 2 1 =− 2 2λv 2 − ˜ H2 −1 replacing µ2 using Eq. sensitive to the higher-dimensional operators.(1.20 1 HIGGS PHYSICS The second. Once we measure the Higgs mass at the LHC. in complete analogy to the Higgs mass above.75) This gives the Feynman rules ¡ ¢ 4f1 v 2 2 H ∂µ H∂ µ H Λ2 m2 H  3m2  f1 v 2 2f2 v 4 2f1 v 2 −i H 1− + 2 + 3Λ2 m2 v 2Λ2 3Λ2 mH H 3  (pj pk ) (1. 2 Λ 2Λ2 λ (1. ˜ Including this additional potential operator in terms of H reads 1 O2 = − (φ† φ)3 3 ˜ 1 (H + v)6 =− 3 8 1 ˜6 ˜ ˜ ˜ ˜ ˜ H + 6H 5 v + 15H 4 v 2 + 20H 3 v 3 + 15H 2 v 4 + 6Hv 5 + v 6 . we can compute the trilinear and quadrilinear Higgs self couplings by collecting the right powers of H in the Higgs potential. high-scale solution is not the vacuum relevant for our Standard Model.(1.(1. Note that from the W.77) j<k . Z masses we know that v = 246 GeV so v is really our first observable in the Higgs sector.67) ⇔ m2 H 5 f2 v 4 f2 v 4 f1 v 2 + 1+ 2 H2 4Λ2 4 Λ2 Λ f2 v 4 f1 v 2 2λv 2 + 2 1 − 2 + O(Λ−4 ) H 2 Λ Λ f1 v 2 f2 v 2 1 − 2 + O(Λ−4 ) H 2 = −λv 2 1 + 2λΛ2 Λ f2 v 2 f1 v 2 m2 ! + O(Λ−4 ) H 2 = − H H 2 = −λv 2 1 − 2 + Λ 2Λ2 λ 2 2 2 λv f1 v f2 v 2 =− 1− 2 + .76) j<k  3m2  f1 v 2 4f2 v 4 2f1 v 2 − i 2H 1 − 2 + 2 2 + v Λ Λ mH 3Λ2 m2 H 4  (pj pk ) (1. We find Lself = − − m2 H 2v m2 H 8v 2 1− 1− f1 v 2 2f2 v 4 + 2 2Λ 3Λ2 m2 H f1 v 2 4f2 v 4 + 2 2 2 Λ Λ mH H3 − H4 − 2f1 v 2 H ∂µ H ∂ µ H Λ2 m2 H . (1. the Higgs mass and the factor in front of the |φ|4 term in the potential has changed.74) Including dimension-6 operators the relation Eq.72) between the vacuum expectation value.73) =− 24 which after combining with the dimension-4 contributions quadratic in the Higgs field shown in Eq.

In the V rest frame we can write the three polarization vectors of the gauge bosons as       0 0 0 µ µ µ = 1  = 0  = 0  . another theory constraint on the Higgs sector.82) . for example the Goldstone couplings to the Higgs boson and the four-Goldstone couplings. w± . This comes in handy for example when we next talk about unitarity. now including the Goldstone modes 1 φ= √ 2 −w2 − iw1 v + H + iw3 ⇒ φ† φ = φ φ † 2 1 2 2 2 w + w2 + w3 + (v + H)2 2 1 2 2 wi i 2 2 wi i 1 = 4 = 1 4 + 1 2 (v + H) 2 i 1 2 wi + (v + H)4 4 (1. in the Standard Model we use only the dimension-4 operators which appear in the renormalizable Lagrangian and which give us for the Higgs mass and self couplings m2 m2 H (1.2. As mentioned before. |p|).1 = T.1 T. m2 = 2λv 2 = −2µ2 H and Lself = − 1. this L longitudinal degree of freedom is precisely the Goldstone boson.1. not the existence but the form of such relations between masses and couplings will change.78) H3 − H H4 2v 8v 2 but we should keep in mind that when the Higgs sector becomes more complicated. This relation between Goldstones and gauge bosons at very high energies is called the equivalence theorem. (1. If we write down the Higgs doublet.80) T. W ± as scalar Goldstone bosons w0 .79) T.2 The Standard Model 21 From this discussion we see that in the Higgs sector the Higgs self couplings as well as the Higgs mass can be computed from the Higgs potential and depend on the operators we take into account. However. 0. the polarization vectors become         0 0 |p| |p| 1  0  E mV 1  0  1 µ µ µ µ  1   0  −→ ≡ p . To compute cross sections for Goldstone boson scattering using the equivalence theorem we need some basic Feynman rules.2 Unitarity If we want to compute transition amplitudes at very high energies the Goldstone modes become very useful.2 = L = 0 1 0 0 mV mV mV 0 0 E |p| Very relativistic gauge bosons are dominated by their longitudinal polarization | µ | ∼ E/mV 1. giving it a four-momentum pµ = (E. 0.81) +vH i 2 wi + · · · we find for the potential — only keeping the relevant terms at dimension four V = µ2 |φ|2 + λ|φ|4 m2 ⊃ λ|φ|4 = H |φ|4 2  2v m2 1 2 = H wi 2v 2 4 i = m2 H 8v 2 2 2 wi i 2  + vH i 2 wi + · · ·  + m2 H H 2v 2 wi + · · · i (1. (1.2 L 0 1 0 0 0 1 If we boost V into the z direction. This means that at high energies we can approximate the complicated vector bosons Z.

(1.22 1 HIGGS PHYSICS Focussing on the scattering of charged Goldstones w± w± → w± w± we use the corresponding fields √ 2 2 w± = (w1 ± iw2 )/ 2. which implies an additional combinatorial factor four in the Feynman rule. there are also two choices to identify the two negatively charged Goldstones. 2s s (1. H H In terms of scalar Goldstones bosons we can compute the amplitude for W + W − → W + W − scattering at very high energies. so we find the terms V = m2 m2 m2 m2 H 4 (w+ w− )2 + H H 2 w+ w− + · · · = H w+ w− w+ w− + H Hw+ w− + · · · 8v 2 2v 2v 2 v (1. we find in the high-energy limit or for massless external particles −i j|A − A∗ T |j = −i j|A − A∗ |j = 2 ImA(θ = 0) ⇒ σ≡ 1 † 1 A A = ImA(θ = 0) . The scattering cross section including all prefactors is then given by σ= = dΩ |A|2 64π 2 s 1 (16π)2 2π 64π 2 s d cos θ −1 2 l l (2l + 1)(2l + 1) al a∗ Pl (cos θ)Pl (cos θ) l (2l + 1) |al |2 . (1.86) Assuming that our Lagrangian is hermitian this imaginary part corresponds only to absorptive terms in the scattering amplitude.84) For this process we want to test the unitarity of the S matrix.85) If we sandwich (A − A† ) between identical asymptotically free fields. which we write in terms of a transition amplitude S = 1 + iA.88) . Linking the Lagrangian to a Feynman rule involves one complication: for each positively charged Goldstone in the vertex there are two ways we can identify them with the Lagrangian fields. which obey an orthogonality condition. 2l + 1 (1. The S matrix should be unitary to conserve probability 1 1 = S † S = (1 − iA† )(1 + iA) = 1 + i(A − A† ) + A† A 1 1 1 1 ! ⇔ A† A = −i(A − A† ) . which means that we are looking at forward scattering. E mW ). Three diagrams contribute to this processes: a four-point vertex.83) which fix the two Feynman rules we need. This is the usual formulation of the optical theorem.87) ordered by the orbital angular momentum l. Pl are the Legendre polynomials of the scattering angle θ. To include the dependence on the scattering angle θ we decompose the transition amplitude into partial waves ∞ 1 A = 16π l=0 (2l + 1) Pl (cos θ) al with −1 dx Pl (x)Pl (x) = 2 δll . l 8π = s l 16π 2(2l + 1) |al | = s (1. Including a common factor (−i) the two Feynman rules then become −2im2 /v 2 and −im2 /v. In addition. They appear in the above expression as w+ w− = (w1 + w2 )/2. the s-channel Higgs exchange and the t-channel Higgs exchange: The amplitude is given by A= −2im2 H + v2 −im2 H v £ ¤ ¥ + + 2 i + s − m2 H −im2 H v 2 i im2 m2 m2 H H H =− 2 2+ + t − m2 v s − m2 t − m2 H H H .

2.90) once we recognize that the condition on Im al and on Re al is a unit cycle around al = (0. and we can even compute a maximum scale smax which leaves √ W W → W W perturbatively unitary for fixed mH : for mH v this typically becomes smax ∼ 1.2 TeV.91) In the high-energy limit s m2 this translates into an upper limit on the Higgs mass which in Eq. Or in other words. The most important result of the unitarity test is probably not the upper bound on the Higgs mass. if we limit s to a finite value this bound changes. This condition we can rewrite as (Re al )2 + Im al − 1 2 2 ⇔ |al |2 = Im al .89) = 1 4 ⇒ |Re al | < 1 . We will discuss this running of couplings in more detail in Section 2. Looking at processes like W W → f f or W W → W W H or W W → HHH we can exactly the same way fix all Higgs couplings in the Standard Model. 2 (1. 1. If something should go wrong with the Standard Model Higgs sector we can expect something else to cure unitarity around the TeV scale. In the case of a running quartic Higgs coupling λ the relevant s. gHHH .2. but the underlying assumption that the unitarity test only works in the presence of one Higgs boson if all Higgs couplings look exactly as predicted by the Standard Model.2 The Standard Model 23 Applied to the perturbative series in partial waves the optical theorem requires 16π ! 1 (2l + 1) |al |2 = 16π(2l + 1) Im al s s using Pl (cos θ = 1) = 1. This number is one of the motivations to build the LHC as a high energy collider with a partonic center-of-mass energy in the few-TeV range. t and u-channel diagrams only depending on λ itself are ¦ § ¨ © + + + . it also uniquely fixes ¯ gW W H . while perturbative unitarity only gives us an upper limit on mH . To apply this perturbative unitary constraint to Goldstone scattering we need to compute the first term in the partial wave expansion for this specific process a0 = 1 16πs 0 dt |A| = − −s 1 16πs 0 dt −s m2 m2 m2 H H H 2+ 2 + t − m2 v2 s − mH H m2 m2 s m2 H = − H2 2 + − H log 1 + 2 2 16πv s − mH s mH = − m2 H 2+O 16πv 2 m2 H s (1.1 for the running QCD coupling αs . ! (1.3 Renormalization group analysis Two additional theoretical constraints we can derive from the renormalization group equation of the Higgs potential. gHHHH .92) This is the maximum value of mH maintaining perturbative unitarity for W W → W W scattering.84) entered as H the Goldstone coupling in the numerator m2 1 H < 2 8πv 2 ⇔ m2 < 4πv 2 ≤ (870 GeV)2 H (1. including gHf f .1. specifically from the renormalization scale dependence of the self coupling λ(Q2 ). Such a scale dependence arises automatically when we encounter ultraviolet divergences and absorb the 1/ poles into a minimal counter term. One last but very important comment: this unitarity argument only works if the W W H coupling is exactly what it should be. 1/2) with radius 1/2. Of course.(1.

Any finite coupling will hit a Landau pole at some scale. The approximate renormalization group equation we can solve by replacing λ = g −1 d 1 1 dg 3 1 dλ ! = =− 2 = d log Q2 d log Q2 g g d log Q2 4π 2 g 2 dg 3 3 ⇔ =− 2 ⇔ g(Q2 ) = − 2 log Q2 + C . = 2 2 d log Q 2Q d Q 16π 4π 2 (1. This formula will be the basis of the discussion in this section.24 1 HIGGS PHYSICS Skipping the calculation we quote the complete renormalization group equation including diagrams with the Higgs boson.93) and ask a different question: how long will λ > 0 ensure that our Higgs potential is bounded from below? . Obviously. After looking at the ultraviolet regime we can go back to the full renormalization group equation Eq.(1.95) ⇒ ⇔ = λ0 1 − Q2 3 λ log 2 2 0 4π v −1 .96) Starting from scales Q ∼ v where the expression in brackets is close to one and moving towards larger scales the denominator becomes smaller until λ actually hits a pole at the critical value Qpole 1− Q2 3 pole ! λ0 log 2 = 0 4π 2 v ⇔ ⇔ ⇔ Q2 3 pole λ0 log 2 = 1 4π 2 v Q2 4π 2 pole log 2 = v 3λ0 Qpole = v exp 2π 2 4π 2 v 2 = v exp 3λ0 3m2 H (1. the top quark and the weak gauge bosons inside the loops 1 3 3 dλ 2 2 4 2 2 = 12λ2 + 6λλ2 − 3λ4 − λ 3g2 + g1 + 2g2 + (g2 + g1 )2 t t d log Q2 16π 2 2 16 with λt = √ 2mt /v.93) approximately reads dλ 1 dλ 3 2 1 = 12λ2 + · · · = λ + ··· . the coupling λ can only be zero everywhere. this divergence should not happen in a physical model and will give us a constraint on the maximum value of λ allowed.(1. which at first glance is precisely not what this theory is — trivial or non-interacting.93) The first regime we study is where this coupling λ becomes strong and the running in Eq. Such a theory with zero interaction is called trivial. .94) Because of the positive sign on the right hand side the quartic coupling will become stronger and stronger and eventually diverge for larger scales Q2 . For a given value of Qpole it translates into a maximum value of mH for which the theory is well defined.97) Such a pole is called a Landau pole and gives us a maximum scale beyond which we cannot rely on our perturbative theory to work. d log Q2 4π 4π A boundary condition λ(Q2 = v 2 ) = λ0 fixes the integration constant C 1 3 3 = − 2 log v 2 + C ⇔ C = g0 + 2 log v 2 λ0 4π 4π 3 3 3 Q2 g(Q2 ) = − 2 log Q2 + g0 + 2 log v 2 = − 2 log 2 + g0 4π 4π 4π v g0 = λ(Q2 ) = − 3 Q2 1 log 2 + 2 4π v λ0 −1 (1. This limit is often referred to as the triviality bound. The name originates from the fact that if we want our Higgs potential to be perturbative at all scales. (1. (1.

v2 (1. electroweak precision data points to the mass range mH 200 GeV.2 The Standard Model 25 Figure 2: Triviality and stability bounds for the Standard Model Higgs boson in the mH − Q plane. The second (smaller) contribution from the weak gauge coupling ameliorates this behavior λ(v 2 ) = ⇔ ⇔ m2 1 12m4 3 Q2 4 2 2 2 H =− − 4t + 2g2 + g2 + g1 log stable 2v 2 16π 2 v 16 v2 4 2 2 1 12mt 3 mH Q 4 2 2 2 = − 2g2 + g2 + g1 log stable 2 2 4 v 8π v 16 v2 mH = 70 GeV for Qstable = 103 GeV 130 GeV for Qstable = 1016 GeV .93) there are two terms with a negative sign which in principle drive λ through zero.98) The usual boundary condition at λ(v 2 ) = m2 /(2v 2 ) is the starting point from which the top Yukawa coupling drives H λ(Q2 ) to zero at another critical scale Qstable which depends on the Higgs mass mH . From the renormalization group we have two pieces of information on the Higgs mass. In the small-λ regime we therefore encounter two finite competing terms dλ 1 4m4 3 4 2 2 2 ∼ −3 4 t + 2g2 + g2 + g1 2 2 d log Q 16π v 16 12m4 3 1 4 2 2 − 4t + 2g2 + g2 + g1 ⇔ λ(Q2 ) ∼ λ(v 2 ) + 2 16π v 16 2 log Q2 . .99) Only for energy scales below Qstable (mH ) the Higgs potential is bounded from below and our vacuum stable. (1. again in the renormalizable Standard model: the Landau pole or triviality bound gives an upper limit on mH as a function of the cutoff scale. One of them vanishes for small λ ∼ 0.(1. so we can neglect it. On the right hand side of Eq. Vacuum stability gives a lower bound on mH as a function of the cutoff scale. Qstable → 1019 GeV.1. Running both cutoff scales towards the Planck mass Qpole . Summarizing what we know about the Higgs mass in the Standard Model we already have indirect experimental as well as theory constraints on this otherwise undetermined parameter. Figure slightly modified compared to the original work [2]. we see in Figure 2 that only Higgs mass values around mH = 180 · · · 190 GeV are allowed for a fundamental Higgs theory. This means at the LHC we are either looking for a light Higgs boson or we should expect some drastic modifications to our Standard Model which could alter this picture significantly and would be hard to miss and even more exciting to observe. This bound is called the stability bound. For a given maximum validity scale Qstable this stability bound translates into a minimum Higgs mass for which our theory is well defined. Strictly in the Standard Model.

interfering destructively. except for muons. The rule for different Higgs decays is simple. The LHC detectors.[3. Since the Higgs boson is a scalar there are no correlations between production and decay process. while for example hadronically . and in particular CMS are designed to measure the photon momentum and energy particularly well. This behavior we see this in Figure 3. 4]. At tree level all decay rates are determined by the Higgs coupling to Standard Model particles. The dominant decays to bottom pairs and W pairs become equal for Higgs masses around 130 GeV. one-loop induced couplings can almost compete with it. Above the top threshold the tt decay becomes sizeable. Both sides of the production and decay can contribute to unique kinematic features which we can use to extract signal from background events.26 1 _ bb BR(H) -1 1 HIGGS PHYSICS WW ZZ 10 ττ _ cc gg + − tt 10 -2 γγ Zγ 10 -3 50 100 200 MH [GeV] 500 1000 Figure 3: Branching ratios of the Standard-Model Higgs boson as a function of its mass.4. Moreover.e. Because of the small mass difference between the W and Z bosons the decay to ZZ is not as dominant. The reason for considering this decay channel are the LHC detectors. The larger W contribution fixes the sign of the loop-induced coupling.3 Higgs decays Signatures for new particles at colliders consist of a production process and a decay pattern. Figure found for example in Refs. so we can use all photon events in the Higgs search. compared to the W W decay which has two degrees of freedom (W + W − and W − W + ) in the final state. evaluated in the appropriate renormalization scheme at the right scale. mb /mW 1/30. which simplifies the calculation and simulation of Higgs signatures. photons do not decay. i. The actual new particle is then described by a Breit-Wigner propagator for unstable particles which we will discuss in detail in Section 2. Because the bottom Yukawa coupling is so small. To extract a Higgs signal from the backgrounds we usually try to measure the four-momenta of the Higgs decay products and reconstruct their invariant mass.2. Starting at low masses this first applies to decays to τ τ and b¯ The relative size b. The resolution on mγγ will at least be a factor of 10 better than for any other decay channel. For large Higgs masses the ¯ ratio of H → W W and H → ZZ decays is fixed by the relative factor of two. 1. In particular the loop-induced decay to two photons which plays an important role in LHC phenomenology. but never really dominant. In particular in the region where the W decays first becomes on-shell we see a drop of in the still off-shell Z decays.1. Off-shell effects in the decays to W W and ZZ are taken into account. they very soon dominate. The structure of the γγH coupling is similar to the production via the loop-induced ggH coupling which we will discuss in Section 1. because by definition the Higgs field couples to all particles (including itself) proportional to their masses it will preferably decay to the heaviest states allowed by phase space. Once the (off-shell) decays to W W open. Unlike the production processes the Higgs decay pattern is as simple as it can be. of their branching ratios around 10% : 90% is given by their masses or Yukawa couplings. The signal should then peak around mH while the backgrounds we expect to be more or less flat. It proceeds via a top and a W triangle which enter with opposite signs in the amplitude. which are fixed by unitarity.

e.Z ? – are these couplings dimension-3 operators? – are fermion Yukawa couplings proportional to mf ? – is there a Higgs self coupling. On the other hand. so the question is if we can find and use a loop-induced coupling of two gluons to the Higgs. Figure from Ref. as a function of the Higgs mass. This is because the Higgs boson couples to all Standard Model particles proportional to their mass — this is the same operator they get their mass from. 1. i.4 Higgs production in gluon fusion 27 Higgs boson searches at the LHC: cross section predictions 10 2 10 1 10 10 10 10 -1 σ(pp→H+X) [pb] √s = 14 TeV gg→H Mt = 174 GeV CTEQ6M 10 2 10 1 gg → H (NNLO) σ(pp → H + X) [pb] √s = 14 TeV NLO / NNLO qq’→HW _ qq→Hqq 10 _ _ -1 qq' → HW _ qq → Hqq gg/qq → ttH (NLO) _ _ -2 gg. a remnant of the Higgs potential? – do λHHH and λHHHH show signs of higher-dimensional operators? – are there any other unexpected effects. in spite of its small branching ratio around 2 · 10−3 . the protons at the LHC contain a lot of gluons. as shown in Figure 4.qq→Htt -3 10 10 -2 qq → HZ -3 _ -4 qq→HZ 0 200 400 MH [GeV] 600 800 1000 _ MRST 10 -4 100 200 300 400 500 600 700 800 900 1000 MH [GeV] ¨ Michael Kramer Page 14 ¨ Universitat Bie Figure 4: Left: production cross section for a Standard-Model Higgs boson at the LHC. .1.4 Higgs production in gluon fusion Looking for the Higgs boson at hadron colliders starts with bad news: at tree level the Higgs hardly couples to light-flavor quarks and has no coupling to gluons.. In spite of the expected suppression of the corresponding cross section by a one-loop factor (g 2 /(16π 2 ))2 we would hope to arrive at an observable production cross section pp → H. [3]. decaying W/Z → 2 jets are not particularly useful at the LHC. again something we will talk about in more detail in Section 2. Because an observed Higgs sector can deviate from the minimal Standard Model assumptions in many ways the LHC will study the different Higgs decays and. These enhancement factors make the Higgs decay to two photons a promising signature. as a function of mH . it will turn out that the production of Higgs bosons in gluon fusion is actually the dominant process at the LHC.. Numerically. like a Higgs decay to invisible particles? But before we decay the Higgs we need to produce it. Because the SU (3)C symmetry of QCD is not broken there is no coupling to gluons at all. Right: updated version including higher order corrections. answer the questions – are gauge-boson couplings proportional to mW.

(1. which do this job for us. Using it we find 4mt PT µν T µν = √ 2 −m2 + 3m2 − H t 8 (k1 q)(k2 q) − 2(k1 q) + q 2 m2 H . So we need to find another operator mediating such a coupling.101) shows that the gauge invariant operator has to be proportional to the transverse tensor 1 µν PT = √ 2 g µν − ν µ k1 k2 (k1 k2 ) with µν PT PT µν = 1 µν µν PT k1µ = 0 = PT k2ν . Switching from position space and its momentum operator into momentum space ∂ → ik Gµν Gµν = − k1µ A1ν − k1ν A1µ k2µ A2ν − k2ν A2µ ν µ k1 k2 k1 k2 = − 2 [(k1 k2 )(A1 A2 ) − (k1 A2 )(k2 A1 )] = − 2(k1 k2 ) A1µ A2ν g µν − = − m2 A1µ A2ν g µν − H ν µ k1 k2 k1 k2 (1. The Higgs can also couple to the field strength in the invariant form HGµν Gµν with Gµν ≡ ∂µ Aν − ∂ν Aµ . One thing to admit at this stage is that nobody in the world really computes Dirac traces by hand anymore. / / / / / / (1. there is a better way to compute this trace.100) µ The problem is the tensor structure of this trace. There are powerful programs.28 1 HIGGS PHYSICS 1.104) . However. (1. which would correspond HGµ Gµ with mass dimension three in the Lagrangian. using all incoming momenta with k1 = k2 = 0 and p2 = m2 . The corresponding Feynman diagram is 2 2 We start with the Dirac trace occurring the top loop. This operator has mass dimension five and arises from the dimension-6 gauge-invariant object φ† φ Gµν Gµν after breaking SU (2)L .102) Using this known tensor structure of T µν we can project out the scalar form factor F using µν T µν = F PT µν PT µν T µν = PT µν PT F = F . In a µ ν so-called axial gauge we could also get rid of the remaining terms proportional to k1 and k2 . because they would not survive the multiplication with the transverse gluon polarization (k · = 0). mt /v ∼ 0. We know that there is no tree-level Higgs coupling of the Higgs to two gluons.7. keeping in mind that it is loop induced and can therefore include a mass suppression by powers of the top mass.1 Effective gluon-Higgs coupling If an effective ggH coupling should be mediated by a closed Standard Model particle loop the top is the perfect candidate: on the one hand it has a strong coupling to gluons. like Form. so we need to compute  T µν = Tr [(q + mt ) γ µ (q + k1 + mt ) γ ν (q + k1 + k2 + mt )] .4. (1. Because of gauge invariance we can neglect terms proportional to k1 ν and k2 .103) This way we project out the relevant gluon tensor structure or the relevant degrees of freedom of the two gluons contributing to this effective coupling. Before that it pays to briefly look at the operator itself. ν and the loop momentum is q. and on the other hand it has the largest of all Standard Model couplings to the Higgs boson. The factor in front of this term is the effective coupling we are going to compute in this section. H The Dirac indices of the two gluons are µ.

which need to be removed by first regularizing the integral and then renormalizing the parameters.(1.105) The non-trivial q µ dependence of the numerator observed in Eq. mt .1. 0.e.(1.(1.2. we still do not know how to remove the (k1 q)(k2 q) terms in Eq.][. However. k2 . a leading order process which receives quantum corrections. The method of choice is a Passarino-Veltman reduction which turns tensor integrals into scalar integrals.. mt . we only compute this one-loop amplitude because there is no tree-level vertex.104) we can take care of using a few tricks.(1.108) but we usually obtain shorter analytical expressions using the Cuskosky cutting rule which links the imaginary part of a diagram or a scalar integral to the sum of all cut Feynman graphs.] 2 (1. which means no ultraviolet divergences. where a scalar integral does not have powers of the loop momentum in the numerator.106) Combining our Dirac trace in Eq. knowing that the final result for t t t this integral will be finite. i. The effective ggH vertex is of the form T µν d4 q 2 ][(q + k )2 − m2 ][(q + k + k )2 − m2 ] . For example. mt ) ≡ H d4 q 1 . which we now have to consider as part of the loop integration...][. t H H iπ 2 [. we use the relation q 2 /(q 2 − m2 ) = 1 + m2 /(q 2 − m2 ) and then shift q. For example. This gives us a very useful expression for the imaginary part of the amplitude −i (A − A∗ ) = 2 ImA = 16π 2 cut graphs ! A. (1.. Cutting diagrams means replacing all internal propagators by 1/(q 2 − m2 ) → 2π θ(q0 ) δ(q 2 − m2 ).104) with the tensor integral in Eq.104). m2 .107) The first term not proportional to any scalar integral in this expression has a little curious origin.4 Higgs production in gluon fusion 29 Inside this trace there appears the loop momentum q. While these tricks help a little. Limiting ourselves to scalar amplitudes/integrals.2. the scalar three-point function is given by 2 2 C(k1 . (x1 A1 + x2 A2 + · · · + xn An )n (1.105) and the applying a reduction algorithm gives us d4 q PT µν T µν 4mt = √ 2 + 4m2 − m2 C(0. The reason why we do not see any divergences in this process is that for a renormalization we would need something to renormalize. mt .109) The factor 16π 2 corresponds to the integral measure in the scalar integrals which can be arranged such that d4 a/(iπ 2 ) gives a typical prefactor of one. mt ) . 2 [q 2 − m2 ][(q + k )2 − m2 ][(q + k + k )2 − m2 ] iπ 1 1 2 t t t (1. including the two external cuts which correspond to the untouched amplitude A and its complex conjugate A∗ . iπ 2 [q 2 − mt 1 1 2 t t (1. because most loop calculations lead to ultraviolet divergences. This is non-trivial piece of information. m2 . so nothing to renormalize. It comes from a combination of O( ) terms from the Dirac trace in n = 4 − 2 dimensions and a two-point function which for the integration measure 1/iπ 2 always includes the ultraviolet divergence 1/ . From this imaginary part we compute the complete amplitude or scalar ... mt . the cut rule tells us that the sum of all cut one-loop or squared scalar diagrams has to vanish. Scalar integrals we can for example calculate using a Feynman parameterization 1 = A1 A2 · · · An 1 dx1 · · · dxn δ 0 xi − 1 (n − 1)! . The cut rule is directly related to the unitarity of the S matrix and the optical theorem discussed in Section 1.

mt ) = H dx log x m2 x(1 − x) − m2 t H (−m2 ) t 1 = 2 mH = dx m2 log 1 − x(1 − x) H x m2 t 1 − 4m2 /m2 t H 1 − 4m2 /m2 t H 4m2 t ≡τ <1.. For example the low-energy limit of the scalar integral. i. If we know the pole or cut structure of the amplitude. the expression in Eq.][. mt .112) is valid for arbitrary Higgs and top masses. As mentioned above. 0. 2 π (m2 +m2 ) q − q2 1 2 dq 2 (m2 +m2 ) 1 2 (1. mt . In contrast to what many people who use such effective vertices seem to believe. Expressing our Dirac trace and loop integral in terms of this function f (τ ) we find for our effective coupling in Eq. The finite three point function which appears in our effective coupling Eq.. 2 (1. this method of computing for example scalar integrals is known to produce the most compact results. m2 ..e. will be given by τ 1 which means mH < 2mt . mt .] m2 2 H 4mt = √ (2 − 2 (τ − 1) f (τ )) 2 8mt = √ (1 + (1 − τ )f (τ )) .112) including imaginary or absorptive terms for τ < 1. mt . the limit in which the top loop becomes heavy and cannot be resolved by the external energy given by the Higgs mass..30 1 HIGGS PHYSICS integral.107) 4mt d4 q PT µν T µν 2f (τ ) = √ 2 − 4m2 − m2 t H iπ 2 [.111) 1+ 1 log2 − 2m2 1− H For general top and Higgs masses it reads      1 τ 2 C(0. mt ) = − H 2f (τ ) m2 H sin −1 τ >1 2 with f (τ ) =  1   − log  4 √ 1+ 1−τ √ 1− 1−τ .110) This step assumes a sufficiently fast convergence on the integration contour for large momenta..107) has the form 1 m2 H 1 0 1 0 C(0. m2 . The dimensionless variable τ is the appropriate parameter to describe the behavior of this scalar integral.113) . we can solve the dispersion or Cauchy integral 1 2πi 1 = 2πi 2 2 ImA(q ) 2 q − q2 A(q 2 ) = dq ∞ 1 2 discA(q ) q 2 − q2 ∞ 1 1 2 2 2 = dq A(q + i ) − A(q − i ) 2 2 2πi (m2 +m2 ) q −q 1 2 1 ∞ 1 2 2 ≡ dq Im+ A(q ) . not just in the heavy-top limit.][. m2 H (1..(1.(1. 0. τ <1 − iπ (1.(1.

the top Yukawa coupling. v 8π v (1. implying gggH = −i .114) where the numerical factors originate from the closed fermion loop. This set of n-gluon couplings to the Higgs boson is again not an approximate result in the top mass.116) and combine it with all other τ -dependent terms in the full expression τ [1 + (1 − τ )f (τ )] =τ 1 + (1 − τ ) 1 1 + 2 + O(τ −3 ) τ 3τ 1 1 + O(τ −2 ) =τ 1 + − 1 − τ 3τ 2 =τ + O(τ −2 ) 3τ 2 αs = + O(τ −1 ) . and finally the result computed in Eq. the two top-gluon couplings. 4 2π v (1.115) dropping the δ ab . It is important to notice that the necessary factor in front of the dimension-5 operator is 1/v and not 1/mt .117) . which does not decouple for heavy top quarks because we have included the top Yukawa coupling in the numerator. 3 12π (1. the heavy top limit τ → ∞ of the expression would be zero.113). Of course. Without this Yukawa coupling. so for light Higgs bosons we can write it in a more compact form. We start with a Taylor series for f (τ ) in the heavy-top limit τ 1 f (τ ) = sin−1 1 τ 1/2 2 = 1 τ 1/2 + 1 + O(τ −5/2 ) 6τ 3/2 2 = 1 1 τ →∞ + 2 + O(τ −3 ) −→ 0 .101) which including prefactors gives us F PT A1µ A2ν = −F Gµν Gµν /( 2m2 ) we can H include this form factor in the effective Lagrangian.(1. This finally defines the Feynman rule we are interested in LggH ⊃ 1 gggH H Gµν Gµν v with 1 αs 1 gggH = −i τ [1 + (1 − τ )f (τ )] . √ µν Going back to Eq.(1.4. 1. the three top propagators. τ 3τ (1. just like we have three-gluon and four-gluon couplings in QCD we can compute the gggH and the ggggH couplings from the ggH coupling simply using gauge invariance. This is a particular feature of this coupling.4 Higgs production in gluon fusion 31 Using this result we can finally collect all factors from our Feynman diagram and compute its contribution to the effective ggH coupling F = − i3 (−igs )2 imt iπ 2 8mt √ (1 + (1 − τ )f (τ )) Tr(T a T b ) v 16π 4 2 2 gs mt δ ab i 8mt √ (1 + (1 − τ )f (τ )) = v 2 16π 2 2 2 gs δ ab i 8 m2 τ H √ = (1 + (1 − τ )f (τ )) v 2 16π 2 2 4 m2 1 2 H τ (1 + (1 − τ )f (τ )) = igs δ ab √ 16 2π 2 v 1 m2 H = iαs δ ab √ τ (1 + (1 − τ )f (τ )) .1.2 Low-energy theorem The general expression for gggH is not particularly handy. the unmatched loop integration measure. as we will see in a minute. the color trace. Gauge invariance completely fixes the n-gluon coupling to the Higgs via one exact dimension-5 operator in the Lagrangian.

12v 24v 2 = αs H Gµν Gµν log 1 + 12π v .32 1 HIGGS PHYSICS In this low-energy or heavy-top limit where we have decoupled the top quark from the set of propagating Standard Model particles. Attaching an external Higgs leg to the gluon self-energy diagram simply means replacing one of the two top propagators with two top propagators and adding a Yukawa coupling √ 1 1 2mt 1 → .3 Signatures To discuss the Higgs production for example in gluon fusion we would normally need to know how to deal with gluons inside the incoming protons. for example for gg+jets production its results only hold in the limit that all jet momenta are much smaller than mt .120) The first two effective couplings with one or two external Higgs legs are the triangular form factor gggH = −iαs /(12π) and the box form factor gggHH = +iαs /(12π). It also becomes problematic for example in the gg → HH process close to threshold. 1. throughout the following sections on Higgs production at the LHC we will limit ourselves to some very basic phenomenological features and postpone any discussion on how to compute these features. All of this we will piece by piece introduce in Section 2. In this low-energy limit we can easily add more Higgs bosons to the loop. as long as the Higgs mass does not exceed roughly twice the top mass. (1. Such a closed form of the Lagrangian is very convenient for simple calculations and gives surprisingly exact results for the gg → H production rate at the LHC. where the momenta of slow-moving Higgs bosons lead to an additional scale in the process. (1.118) q − mt / q − mt / v q − mt / This we can compare to a differentiation with respect to mt √ √ √ 2mt ∂ 2mt 2mt 1 −1 1 1 ∼ (−1) ∼ .1. Without knowing any theoretical particle physics we can discuss the main feature or problem of hadron collider physics: there is no such thing as a signal without a background.119) This treatment including a gamma matrix is. To obtain the correct mass dimension each external Higgs field appears as H/v. (1. v ∂mt q − mt / v (q − mt )(q − mt ) / / q − mt / v q − mt / (1. This series can be resummed to LggH = Gµν Gµν with log(1 + x) = − αs π H H2 − + . it gives us an idea how we can in the limit of a heavy top derive the ggH n couplings from the gluon self energy or the ggH coupling gggH n+1 = m2 t ∂ ∂mt 1 gggH n mt . In contrast. For n neutral Higgs fields the Lagrangian has to include a factor of 1/n which is the denominator of the logarithm’s Taylor series. how to parameterize the phase space of the Higgs decay products.121) n n=1 (−x) /n. nonsense.2. We will come to this process later in a little more detail. However. strictly speaking. and how to distinguish signal events from background events.2 we noted that there should be a relative factor between the Lagrangian and the Feynman rule accounting for more than one way to identify the external legs of the Feynman rule with the fields in the Lagrangian. it is not necessary to numerically compute processes involving ggH coupling.. there is no kinematic configuration which is unique to a signal event and cannot appear by an unlucky combination of uninteresting Standard Model or . We emphasize again that while this low-energy approximation is very compact to analytically write down the effective ggH coupling. In Section 1. More precisely.4. The ggH coupling does not depend on mt anymore and gives a finite result.. However. For this result to be finite we have to introduce the top Yukawa coupling in the numerator.

which means QCD is the theory of the evil. This number of events is the product of the proton–proton LHC luminosity measured in inverse femtobarns.8 × 10−7 (Gaussian limit) (fluctuation probability) . because when I was a graduate student they still belonged to the second class of processes. (1. People usually have good personal reasons to advertize such effects. a predicted event event rate it is split into a collider-specific number describing the initial state. The latter is the easiest number to deal with: over the sensitive region of the detector. or a wish to get the Stockholm call. Because of the sheer mass of data at the LHC. like a tenure case.4 Higgs production in gluon fusion 33 QCD processes and detector effect. As a matter of fact. the rate for a pair of 500 GeV supersymmetric gluinos is 4 × 104 fb. All other particles range somewhere in between. (1. but for us inverse femtobarns work better. so QCD as we will discuss it in Section 2 is not entirely evil. a process-specific number describing the physical process. because everyone who has been around for a few years has seen a great number of those go away. and a detector-specific efficiency for each particle in the final state. On the other hand. somewhere between 90% and 100%. the total production cross section measured in femtobarns. The first involves all particles which we know and love. Figure 5 shows that at the LHC the production cross section for a pair of bottom quarks is larger than 105 nb or 1011 fb. at least those background events which populate the signal region in phase space. In other words. and for the rest of the community the corresponding processes instantly turn into backgrounds. and the detection efficiency measured in per-cent. So all we need to do is wait for the significance to hit five sigma. This implies that any LHC measurement will always be a statistics exercise in some kind of event counting combined with a probability estimate on the signal nature of a given event. Because signals are things we had not seen before collider signals are much more rare that background events. Top quarks have an interesting history. shakes hands with the king of Sweden. The first quantity we can compute and analyze at colliders is the total number of events expected from a certain production process in a given time interval. the fiducial volume. For theorists luminosity is simply a conversion number between cross sections which we compute for a living and event numbers. we have to decide very fast if an event has the potential of being interesting in the light of the physics . the typical production rate for W or Z bosons ranges around 200 nb or 2 × 108 fb. Finally talking about cross sections and how to compute them we need to remember that at the LHC there exist two kinds of processes. These either involve particles we have not seen before or particles we want to know something about. in the Gaussian limit √ the statistical significance improves with L. and less than 1/3 for tau leptons. Instead. we could not even write every top pair event on tape. The key numbers and their orders of magnitude for typical signals are Nevents = σtot · L L = 10 · · · 300 fb−1 σtot = 1 · · · 104 fb . a promotion. we will not be able to write every LHC event on tape. People who build colliders use units involving seconds and square meters. Typical numbers are: a typical year of LHC running at design luminosity could deliver around 10 inverse femtobarns per year in the first few years and three to ten times that later.122) Different cross sections for Tevatron and LHC processes are shown in Figure 5. Digging out signal events from a large number of background events is what the remaining Higgs section will be about. They are described by QCD. If we actually see new particles someone gets a call from Stockholm. and the Higgs rate can be as big as 2 × 105 fb. To extract such signals from huge backgrounds we need to describe the backgrounds with an incredible precision.123) Never trust anybody who wants to sell you a three sigma evidence as a discovery. This number is very good for muons. the signals. The high energy community has agreed that we call a five sigma excess over the known backgrounds a signal S √ = Nσ > 5 B Pfluct < 5.1. the detection efficiency is a set of numbers depending on the nature of the detected particle and its energy. All of these processes we call backgrounds and find annoying. One last aspect we have to at least mention is the trigger. Such new particle are usually produced via QCD effects as well. but all they are really saying is that their errors do not allow them to make a conclusive statement. like old-fashioned electrons or slightly more modern W and Z bosons or most recently top quarks.

questions we are asking at the LHC.34 1 HIGGS PHYSICS proton . What makes things worse. and that very soft final states for example from bottom decays are best studied in the LHCb experiment instead of ATLAS and CMS.e. With this minimal background of collider phenomenology we can look at Higgs production in gluon fusion. this channel will as it b 33 3 3 events / sec for L = 10 cm s -2 -1 5 5 . Before there is a mis-understanding: while experimentalists are reluctant to change triggers they are not carved in stone.1 1 10 √s (TeV) Figure 5: Production rates for signal and background processes at hadron colliders. which of the decays can be distinguished from the Standard Model background statistically. Figure from Ref. are shown in Figure 6. The total Higgs production cross section through the loop induced ggH coupling we show in Figure 4. which for relevant luminosities starting around 30 fb−1 means a whopping 106 events. to Figure 5 exceeds the signal by roughly eight orders of magnitude. combined with different Higgs decays. For a reasonably light Higgs boson the cross section ranges around at least 30 pb. [6]. The two colliders correspond to the x–axis values of 2 TeV and 14 TeV. This trigger menu reflects the general attitude that the LHC is not built to study QCD. plus as much as we can events with missing energy (i. For now we can safely assume that above an energy threshold we will keep all event with leptons or photons. Only these events we will keep. The question is: after multiplying with the relevant branching ratio and detection efficiencies. The results of experimental simulations. for example by the ATLAS collaboration. is: gg → H → b¯ is hopeless. because of sheer size of the QCD continuum background gg → b¯ which according b b. ordered by decreasing branching ratio according to Figure 3. The complete list of possible Higgs decays.(anti)proton cross sections 10 10 10 10 10 10 10 10 9 10 9 8 σtot 10 8 7 Tevatron LHC 10 10 10 7 6 6 4 σb 10 10 4 2 σjet(ET jet > √s/20) σW σZ 10 10 10 10 10 2 σ (nb) 10 10 10 10 10 10 10 10 10 1 1 0 0 -1 σjet(ET jet > 100 GeV) -1 -2 -2 -3 σt σjet(ET jet 10 10 10 10 -3 -4 > √s/4) -4 -5 σHiggs(MH=120 GeV) -5 -6 200 GeV WJS2009 -6 500 GeV 10 -7 -7 0. There is little to cut on except for the invariant mass of the b¯ pair with an O(10%) mass resolution. neutrinos in the Standard Model and dark matter particles in new-physics models) and jets with very high energy coming from resonance decays. so as a functioning high energy physics community we will not miss great new physics just because we forgot to include it in the trigger menu. The discontinuity is due to the Tevatron being a proton–antiproton collider while the LHC is a proton–proton collider. Since gluon fusion really only produces the Higgs boson the production process does not help much with the background suppression.

gg → H → W + W − has a large rate. Five standard deviations over the backgrounds are required for discovery. but with a smaller rate and a further reduced branching ratio of Z → µ+ µ− or Z → e+ e− . All we can do is reconstruct a transverse mass variable. but once one of the W bosons decays leptonically the Higgs mass is hard to reconstruct.e. which we discuss in Section 3. Figure borrowed from ATLAS. because of the fully reconstructed m4 ∼ mZZ ∼ mH . so it is not likely to be seen.1. Instead of combining the advantages of H → ZZ and H → γγ this channel combines more of the disadvantages.2. theory input will play no role in this analysis. . because of the low velocity of the Higgs which makes the reconstruction of mτ τ ∼ mH hard. very promising. Of all Higgs channels it requires the least understanding of the LHC detectors.3. Because there is not much more to cut on the significance drops sharply with the decreasing branching ratio once we require off-shell W decays. The main problem is backgrounds for example from pions mistaken for photons. so we should not give up immediately. as we will discuss in Section 3. because the W coupling to fermions is purely left handed this implies that the two leptons prefer to move into the same direction as opposed to back-to-back. Its limitation is the sharp drop in the off-shell branching ratio towards smaller Higgs masses. Because mγγ can be reconstructed to O(1%) this observable has incredibly precise side bins to the left and the to right of the Higgs peak. as for any channel seeing it would give us more information on the Higgs boson. This is what the electromagnetic calorimeter for example of CMS has been designed for. gg → H → ZZ works great for ZZ to four leptons. It is widely assumed that Higgs production in gluon fusion is too close to threshold to see the decay to tau leptons.1. stands not be triggered on.5.4 Higgs production in gluon fusion 35 Signal significance ∫ L dt = 30 fb (no K-factors) 10 2 -1 ATLAS H → γγ ttH (H → bb) H → ZZ(*) → 4 l H → WW(*) → lνlν qqH → qq WW(*) → lνlν qqH → qq ττ qqH → qqZZ → llνν qqH → qqWW → lνjj Total significance 10 5σ 1 10 2 10 3 2 mH (GeV/c ) Figure 6: Simulated statistical significance for different Higgs production and decay channels for an integrated luminosity of 30 fb−1 . in spite of the small rate. Of course. On the other hand. the peak in the invariant mass of the photons is great to measure the Higgs mass: once we see a Gaussian peak there we can √ determine its central value with a precision of Γresolution / S (in a signal dominated sample with S signal events). gg → H → γγ is. but this might still change. The semi-leptonic decay of the Z pair might work once we ask for a boosted hadronically decaying Z. the backgrounds are electroweak and therefore small. at the per-mille level. Moreover. gg → H → Zγ is a little like γγ.3. in particular muons. We will discuss this decay and its reconstruction in detail in Section 1. gg → H → τ + τ − is problematic. i. by ATLAS. The most promising analysis relies on angular correlations — if the two gauge bosons come from a spin-zero resonance they have to have opposite polarization. Therefore it is referred to as ‘golden channel’.

where the u radiates a W + and the d radiates a W − . 1. The Feynman diagram for this process is If the Higgs were a Z boson. They induce a particular shape of the kinematic distributions of the final state jet which requires us to first quote from the exact calculation that in the matrix element squared we will usually find one power of pT in the numerator.2 this implies an enhancement of collinear (forward) jets. one from each of the protons. If we neglect the cutoff by mW the pT. ∼ 30 GeV. so instead of based on the top Yukawa coupling we can produce it via its gauge boson coupling. and the absolute value of the two-dimensional transverse momentum pT 0 = ! pT ∂ ∂pT E 2 − p2 − p2 − m2 3 T W 1 (−1) = 2 + pT (−2pT ) E − p2 − p2 − m2 (E 2 − p2 − p2 − m2 )2 3 3 T W T W = E 2 − p2 − p2 − m2 + 2p2 3 W T T (E 2 − p2 − p2 − m2 )2 3 T W Cm2 − p2 − m2 + 2p2 W T W T (E 2 − p2 − p2 − m2 )2 3 T W p2 − (1 − C)m2 T W (E 2 − p2 − p2 − m2 )2 3 W T p2 = (1 − C)m2 T W (1.j = pT. and ‘absolutely nothing’ is not a good signature for the trigger. In reality.3. In the Standard Model the Higgs has sizeable couplings only to the W and Z bosons and to the top quark. Because the LHC is a pp collider and because the proton mostly contains the valence quarks (uud) and low-x gluons it is important that this process can proceed as ud → duH.5. As we will show in Section 2.124) with E 2 ∼ p2 3 m2 W or E 2 − p2 = C m2 3 W ∼ = ⇔ for some number C < 1.j . but for Higgs production at colliders we safely assume mq = 0 for the first two generation fermions.j distributions will even diverge for small pT. This admittedly hand-waving argument shows that in weak-boson-fusion Higgs production the transverse momenta of the outgoing jets peak at values below the W mass. This coupling induces two channels.j like 1/pT. 1. This transverse momentum .5.36 1 HIGGS PHYSICS gg → H → invisible is not predicted in Standard Model. it could also bremsstrahlung off the incoming or outgoing quarks. it is obviously hopeless if only the Higgs is produced. The double deep inelastic scattering approximation is also a good way to compute corrections to the weak boson fusion production rate. weak boson fusion looks like double deep inelastic scattering. at least if we neglect kinematic distributions.1 Production kinematics  In the Feynman diagrams for weak-boson fusion Higgs production we encounter intermediate massive gauge boson propagators. because we would be trying to extract a signal of missing energy and nothing else. there are other promising channels to study. the larger of which is weak boson fusion qq → qqH: two incoming quarks each radiate a W or Z boson which merge and form a Higgs.W spectrum as a function of the momentum in the beam direction. That is at least unless we discuss a muon collider as a specific way to produce Higgs bosons. With this information we can look for the maximum in the pT. the peak occurs around pT.2 we will use for background suppression via the central jet veto. p3 .5 Higgs production in weak boson fusion Going back to Figure 4 we see that while gluon fusion gives the largest Higgs production rate at the LHC. In a way. This is one of the key observations which in Section 1.

j 2 σWBF ∝ log ∼ log mH mW 2 . After integrating over pT. The one-loop amplitude for gg → H is proportional to αs yt /(4π) ∼ (1/10) (2/3) (1/12) = 1/180.1. √ with the two parton momentum fractions x1. The two reasons for this behavior we mentioned already in this section: first of all.1 + pj.3.j like 1/pT. in contrast to the jets the Higgs and its decay products are expected to hit the central detector.j .125) For mH = O(TeV) this logarithm gives us an enhancement by factors of up to 10. For the production cross section this means a factor of (1/180)2 ∼ 1/40000. The difference in rate which we see in Figure 4 arises from the quark and gluon luminosities. Compared to the Higgs production in gluon fusion we buy this distinctive signature and with it the efficient extraction of the Higgs signal out of the background at the expense of the rate. for larger x values the rate for gg → H decreases steeply with the gluon density. which makes weak boson fusion the dominant Higgs production process.125) or compute such a probability for W bosons in the proton. the transverse momenta of the W boson and the forward jet are identical. Because the incoming quark inside the proton has negligible transverse momentum.j distributions will diverge for small pT. This number is a function of the partonic momentum fraction x and can be evaluated as a function of the transverse momentum pT .2 ). the gluon parton density grows very steeply for x 10−2 . In the effective W approximation we can resum the logarithms appearing in Eq. while in weak boson fusion the already huge partonic center of mass energy due to the tagging jets ensures that an increase in mH makes no difference anymore. So we are looking for two forward jets and for example two τ leptons or two W bosons in the central detector. the Higgs is produced with a finite pT H which is largely determined by the kinematic and the acceptance cuts on the forward jets and their typical transverse momentum scale pT ∼ mW .5 Higgs production in weak boson fusion 37 scale we need to compare to the longitudinal momentum given by the energy scale of valence quarks at the LHC. These two forward jets are referred to as tagging jets. Moving on to the Higgs kinematics. these numbers mean little once we battle an 800 pb tt background relying on kinematic cuts either on forward jets or on Higgs decay products.j increasing mH this logarithm becomes pmax T. (1. Producing a single Higgs in gluon fusion probes the large gluon parton density at typical parton momentum fractions √ x ∼ mH / s ∼ 10−3 . In weak boson fusion the two forward jets always combine to a large partonic center-of-mass energy x1 x2 s > (pj. several hundreds of GeV.j this yields a log pmax /pmin dependence of the total rate. but with two additional jets in the final state.e. If we neglect mW in the weak boson fusion process the pT.2 )2 = 2(pj. i. their invariant mass mjj will easily exceed a TeV. For any kind of QCD background this will not be the case. Compared to Higgs production in gluon fusion the tagging jets are an example how features of the production process which have little or nothing to do with the actual Higgs kinematics can help reduce backgrounds — the largest production rate does not automatically yield the best discovery prospects. Looking at typical LHC energies. we will talk about partons inside the proton and their probability distributions for given momenta in Section 2. Including the additional phase-space for two jets this very roughly means (2/3)6 1/(16π)2 = (64/729) (1/2500) ∼ 1/25000. These .1 pj. This means that gluon fusion wins: for a 150 GeV Higgs the gluon fusion rate of ∼ 30 pb clearly exceeds the weak boson fusion rate of ¯ ∼ 4 pb. They offer a very efficient cut against QCD backgrounds: because of their back-to-back geometry and their very large longitudinal momentum. On the other hand.(1. Even more importantly.3.j T. In Figure 6 we see that for large Higgs mass the weak boson fusion rate approaches the gluon fusion rate. Last but not least.2 and the hadronic center of mass energy s = 14 TeV. Motivated by such logarithms. This means that each of the two production processes with their specific incoming partons probes its most favorable parton momentum fraction: low-x for gluon fusion and high-x for valence quark scattering. there appear large logarithms because the low-pT enhancement of the quark-W splitting. These two numbers governing the main LHC production cross sections roughly balance each other.j pmin T. With the W mass cutoff and a typical hard scale given by the T. The cross section for weak boson fusion is proportional to g 6 .

38

1

HIGGS PHYSICS

transverse momentum distributions in pT,W = pT,j look different for transverse and longitudinal gauge bosons PT (x, pT ) ∼ PL (x, pT ) ∼
2 2 gV + gA 1 + (1 − x)2 p2 T 4π 2 2x (p2 + (1 − x) m2 )2 T W 2 2 gV + gA (1 − x)2 m2 W 2 + (1 − x) m2 )2 2 4π x (pT W

→ →

2 2 gV + gA 1 + (1 − x)2 1 4π 2 2x p2 T 2 2 gV + gA (1 − x)2 m2 W . 2 4π x p4 T

(1.126)

The couplings gA,V describe the gauge coupling of the W bosons to the incoming quarks. Looking at large transverse momenta pT (1 − x)mW at the upper end of the pT spectra the radiation of longitudinal W bosons falls off sharper than the radiation of transverse W bosons. This different behavior of transverse and longitudinal W bosons is interesting, because it allows us to gain information on the centrally produced particle and which modes it couples to just from the transverse momentum spectrum of the forward jets and without looking at the actual central particle. This problem with the effective W approximation is that it is a poor description of Higgs production at the LHC. The simple reason is that the Higgs mass is of the order of the W mass, as are the transverse momenta of the W and the final state jets, and none of them are very small. Neglecting for example the transverse momentum of the W bosons or the final-state jets will not give us useful predictions for the kinematic distributions, neither for the tagging jets nor for the Higgs. For the SSC, the competing design to the LHC in Texas which unfortunately was never built, this might have been a different story, but at the LHC we should not assume W bosons (or for that matter top quarks) are essentially massless partons inside the proton. 1.5.2 Central jet veto

From the Feynman diagram for weak boson fusion we see that the diagram describing a gluon exchange between the two quark lines multiplied with the Born diagram is proportional to the color factor Tr T a Tr T b δ ab = 0. The only way to avoid this suppression is the interference of two identical final-state quarks, for example in ZZ fusion, which first of all does not involve only valence quarks and secondly assumes a phase space configuration where one of the two supposedly forward jets actually turns around and goes backwards, so the interfering diagrams contribute in the same phase spare region. Which means that virtual gluon exchange in weak boson fusion is practically absent. In Section 2 we will see that virtual gluon exchange and real gluon emission are very closely related. Radiating a gluon off any of the quarks in the weak boson fusion process will lead to a double infrared divergence, one because the gluon can be radiated at small angles and one because the gluon can be radiated with vanishing energy. The divergence at small angles is removed by redefining the quark parton densities in the proton. The soft divergence has to cancel between real gluon emission and virtual gluon exchange. However, if virtual gluon exchange does not appear non-collinear gluon radiation cannot appear either. This means that additional QCD jet activity as part of the weak boson fusion process is limited to collinear radiation, i.e. radiation along the beam line or at least in the same direction as the far forward tagging jets. Gluon radiation into the central detector is suppressed by the color structure. While it is not immediately clear how to quantify such a statement it is a very useful feature, for example looking at the top pair backgrounds. The W W b¯ final state as a background to qqH, H → W W searches includes two bottom b jets which can mimic the signal’s tagging jets. However, it turns out that it is much more likely that we will produce ¯ another jet pp → tt+jet so only one of the two bottom jets from the top decays needs to be forward. In that case the way to isolate the Higgs signal is to look at additional central jets. As described above, for the signal additional jet activity is limited to small-angle radiation off the initial-state and final-state quarks. For background like top pairs this is not the case, which means we can reduce all kinds of background by vetoing jets in the central region above pT j 30 GeV. This strategy is referred to as central jet veto or mini-jet veto. Note that it has nothing to do with rapidity gaps at HERA or pomeron exchange, its QCD features are completely described by standard perturbative QCD. From QCD we then need to compute the probability of not observing additional central jets for different signal and background processes. Postponing the discussion of QCD parton splitting to Section 2.3.2 we already know that for small transverse momenta the pT,j spectra for massless states will diverge, as shown in Eq.(1.125). Looking at some kind of n-particle final state and an additional jet radiation we can implicitly define a reference point pcrit at which the T

1.5

Higgs production in weak boson fusion

39

diverging rate for one jet radiation σn+1 starts to exceed the original rate σn , whatever the relevant process might be

σn+1 (pcrit ) = T

dpT,j
pcrit T

dσn+1 ! = σn . dpT,j

(1.127)

This condition defines a point in pT below which our perturbation theory in αs , i.e. in counting the number of external partons, breaks down. For weak boson fusion Higgs production we find pcrit ∼ 10 GeV, while for QCD T ¯ processes like tt production it becomes pcrit = 40 GeV. In other words, jets down to pT =10 GeV are perturbatively T well defined for Higgs signatures, while for the QCD backgrounds jets below 40 GeV are much more frequent than they should be looking at the perturbative series in αs . This fixes the pT range where a central jet veto will be helpful to reject backgrounds (tag (tag (1.128) pT,j > 30 GeV and ηj 1) < ηj < ηj 2) . The second condition reminds us of the fact that only central jet radiation will be rare in weak boson fusion Higgs production. The smaller the pT can become the more efficient the central jet veto becomes, but at some point experimental problems as well as non-perturbative QCD effects will force us to stay above 20 or 30 or even 40 GeV. If we naively assume that the radiation of additional jets proceeds with a constant probability, independent of the number of jets already radiated, we can even compute the survival probability. Certainly for values pmin > pcrit we can T T postulate an n-independent constant describing the additional radiation of one jet Prad (pmin ) = T σ3 (pmin ) σn+1 (pmin ) T T = ··· = σ2 (pmin ) σn (pmin ) T T (1.129)

Using Poisson statistics the probability of observing n jets given the jet radiation probability Prad is f (n; Prad ) =
n Prad e−Prad n!

f (0; Prad ) = e−Prad

(1.130)

which means the probability of surviving the veto is exp(−Prad ). This comes out roughly as 88% for the weak boson ¯ fusion signal and as 24% for the tt background. Note that this number does not incorporate experimental effects or the non-perturbative underlying event, i.e. additional energy-dependent but process-independent jet activity in the detectors from many not entirely understood sources. This simple exponentiation model is also not guaranteed to give us the correct numbers, and by the time we get to Section 2.5.3 we will be able to properly describe the central jet veto in QCD. 1.5.3 Decay kinematics and signatures

The sizeable transverse momentum of the Higgs in the weak boson fusion process allows us to reconstruct the invariant mass of a τ τ system in the collinear approximation: if we assume that a τ with momentum p decays into a lepton with the momentum xp and a neutrino, both moving into the same direction as the tau, we can write the two-dimensional transverse momentum of the two taus from the Higgs decay in two ways p1 + p 2 ≡ k2 ! k1 + = k1 + k2 + k . / x1 x2 (1.131)

The missing transverse momentum k is the measured vector sum of the two neutrino momenta. This equation is useful / because we can measure the missing energy vector at the LHC in the transverse plane, i.e. as two components, which means the vector equation is really two equations for the two unknowns x1 and x2 . Skipping the actual calculation how to solve these two equations for x1 x2 we quote the result for the invariant τ τ mass which for the signal has to correspond to the Higgs mass (k1 k2 ) mτ τ = 2 (p1 p2 ) = 2 . (1.132) x1 x2 From the formula above it is obvious that this approximation does not only require a sizeable pi mτ , but also that back-to-back taus will not work — the vector k then largely cancels and in the Higgs center-of-mass frame and the /

40

1

HIGGS PHYSICS

computation fails. This is what happens for the production channel gg → H → τ τ , where the Higgs boson is essentially produced at rest. Again, we can make a list of signatures which work more or less well in connection to weak boson fusion production. These channels are also included in the summary plot by ATLAS, shown in Figure 6. qqH, H → b¯ is problematic because of large QCD backgrounds and because of the trigger in ATLAS. The b most worrisome background is actually overlapping events, one producing the two tagging jets and the other one two bottom jets. This overlapping scattering gives a non-trivial structure to the background events, so a brute force side-bin analysis will not work. qqH, H → τ τ can be a discovery channel for a light Higgs boson with mH 130 GeV. The approximate mass reconstruction might be good to ∼ 5 GeV. This channel is particularly useful in scenarios beyond the Standard Model, like its minimal supersymmetric extension. It has been shown to guarantee the discovery of one Higgs boson over the entire supersymmetric parameter space without a dedicated SUSY search. qqH, H → γγ should be compatible with gg → H → γγ with its smaller rate but improved background suppression. It is included in inclusive H → γγ analyses, and for neural net analyses zooming in on large Higgs transverse momenta it will dominate the inclusive analysis. qqH, H → Zγ is difficult due to simply too small event rate and no apparent experimental advantages. qqH, H → W W is the discovery channel for mH 135 GeV. In contrast to gg → H → W W it also works for off-shell W decays, i.e. for Higgs masses below 150 GeV. This is where using a multitude of background rejection cuts and generically better signal-to-background ratios win over the large rate in gluon fusion. qqH, H → ZZ is likely to work in spite of the smaller rate compared to gluon fusion. It might even be possible with one hadronic Z decay, but there are not many detailed studies available. qqH, H → invisible is the only discovery channel for an invisible Higgs which really works at the LHC. It relies on the pure tagging-jet signature, which means it is seriously hard and requires a very good understanding of the detector and of QCD effects. Just a side remark for younger LHC physicists: weak boson fusion was essentially unknown as a production mode for light Higgses until around 1998 and started off as a very successful PhD project. This meant that for example the Higgs chapter in the ATLAS TDR had to be re-written. While it sometime might appear that way there is no such thing as a completely understood field of LHC physics. Every aspect of LHC physics continuously moves around and your impact only depends on how good your ideas and their technical realization are.

1.6

Associated Higgs production

In Figure 4 there appears another class or processes at smaller rates: associated Higgs production with heavy Standard ¯ Model particles, like W or Z bosons or tt pairs. Until the summer of 2008 the Higgs community at the LHC was convinced that (a) we would not be able to make use of W H and ZH production at the LHC and (b) we would not be able to see most of the light Higgs bosons produced at the LHC, because H → b¯ does not give a promising signature b in gluon fusion or weak boson fusion production. One key to argument (a) are the two different initial states for signal and background: at the Tevatron the processes q q → ZH and q q → W H come from valence quarks. At the LHC with its proton-proton beam this is not possible, ¯ ¯ so the signal rate will relatively suffer when we go from Tevatron to LHC. The QCD background at leading order is q q → Zg ∗ → Zb¯ production, with an off-shell gluon splitting into a low mass bottom quark pair. At next-to-leading ¯ b order, we also have to consider the t-channel process q q → Z ¯ and its flipped counter part qg → Z ¯ This ¯ bbg bbq. background becomes more dangerous for larger initial state gluon densities. This means that moving from Tevatron to LHC the signal will decrease while the background increases — not a very promising starting point. Then came Ref. [7] and the whole picture changed. We will discuss this search strategy in detail in Section 3.1.2 when we know about jets and jet algorithms at the LHC. It turns out that searches for boosted Higgs bosons are not only

we still need to measure its CP properties. which altogether gives us the three coupling tensors g µν g µν − pµ pν 1 2 p1 p2 µνρσ pρ pσ . However. as discussed in Sections 1. For the three gauge-invariant coupling structures shown in Eq. i. In a way this section ties in with the effective theory picture we use to introduce the Higgs mechanism: the obvious requirements to include massive gauge bosons in an effective electroweak gauge theory leads us towards the Standard Model Higgs boson only step by step. we can also study many of its properties. These new channels are b b not yet included in the ATLAS list of processes shown in Figure 6 because the simulations are still at an early stage. can we measure at least some of these properties? In this section we will briefly touch on two interesting questions relevant to LHC Higgs physics and to determining if what we might see at the LHC is really a Higgs boson. This is another example of what is possible and what not in LHC phenomenology: it all really depends only on how creative you are. i.7 Beyond Higgs discovery The prime goal of the LHC will of course be to discover a new light scalar particle which we would suspect to be the Higgs boson. but might also resurrect the ttH.(1. Going back to the transverse momentum spectra for the tagging jets shown in Eq. so we could in principle replace it with any linear combination with g µν . Once we know that we are talking about a scalar field. But we can expect them to play a role in LHC searches for light Higgs bosons. However. The fundamental CP-even Standard-Model Higgs bosons couples to the W and Z bosons proportional to g µν .133) These are the only gauge invariant dimension-5 couplings of two gauge bosons to a (pseudo-) scalar field. For general CP-even and CP-odd Higgs bosons there are two more gauge-invariant ways to couple to W bosons. One of them gave us the effective one-loop gluon-gluon-Higgs coupling in Eqs.1. The decrease towards larger angular separation is an effect of kinematic cuts against the Z background which prefers large angular separation. the signal is depleted in the background-rich region.121). 1. these two coupling structures nicely correspond to the gauge-invariant forms W µν Wµν and W µν Wµν in the electroweak Lagrangian. the CP-even dimension-6 operator is proportional to the transverse tensor.7 Beyond Higgs discovery 41 ¯ promising in the V H.4 and 1.115. 1 2 (1. The tree-level coupling should in principle show no dependence of ∆φ at all.126) we already know one way of telling apart these couplings: by construction.1 CP properties One question for the LHC is: once we see a Higgs boson at the LHC. all its properties except for its exact mass are fixed in the minimal one-doublet Higgs sector of the Standard Model. A spin-one Higgs boson would not couple to two photons. so we leave them like this. A more useful observable is the azimuthal angle between the two tagging jets in weak boson fusion.133) we find three distinctly different normalized distributions of the azimuthal angle ∆φ. H → b¯ channel.e. it couples the longitudinal Goldstone modes in the W boson to the Higgs. In contrast.1. Therefore. how do we test its quantum numbers? One such quantum number is the spin of the Higgs boson. and that even applies to a field like Standard Model Higgs searches. The CP-odd and therefore antisymmetric µνρσ coupling vanishes .5. as a matter of fact.(1.(1. so we would need to look at angular distributions for example in weak boson fusion to distinguish spin zero from spin two.e. This is where we have seen the most progress in LHC Higgs physics over recent years: not only will we be able to see a light Higgs boson simultaneously in different production and decay channels. 1. The question is.7. the dimension-3 W W H coupling proportional to g µν comes out of the Higgs potential as a side product of the W mass term. and at any of these steps we can stop and postulate some alternative ultraviolet completion of our theory. This is a little surprising since there are no Higgs momenta involved in this measurement. which is supposedly studied to death. shown in Figure 7. The first tensor as written above is not orthogonal to g µν . the Standard Model makes many predictions concerning its properties. the angle separating the two jets in the transverse plane. H → b¯ channel. which means it couples the transverse W polarizations to the Higgs and therefore produces a harder pT spectrum of the tagging jets.

In that case the Higgs production matrix element becomes proportional to the scalar product (ptag 1 ptag 2 ) which vanishes for a relative angle ∆φ = π/2. This φ distribution can be written as dσ ∝ 1 + a cos(φ) + b cos(2φ) .7. where the two transverse jet momenta are no more linearly independent.004 SM 0. dφ (1.134) For the CP-odd Higgs coupling to W µν Wµν we find a = 0 and b = 1/4. once two momenta contracted with the Levi–Civita tensor are equal.2 Higgs self coupling   −→ If we should ever observe a scalar particle at the LHC. This trilinear coupling we can probe at the LHC studying . the maybe most important question will be if this is indeed our Higgs boson arising from electroweak symmetry breaking. To explain the opposite shape of the CP-even Higgs we go into the longitudinal limit of the tagging jets. This explains the zeros of the φ distribution at φ = 0. In other words. The signal and the leading Z background are simulated on the parton level.42 1 HIGGS PHYSICS 0. 1. Figure from Ref. π.002 CP even Z→ττ 0 0 50 100 150 ∆Φjj Figure 7: Distribution for the azimuthal angle between the two forward jets in WBF production with a subsequent decay H → τ τ . On the other hand. This method only works if we observe the decay H → ZZ → 4 with a good signal-background ratio S/B. as we can see from their Feynman diagrams: starting from the Feynman diagram for H → ZZ → 4 we can switch two fermion legs from the final state into the initial state and read the diagram right-to-left. All signal rates are normalized to the Standard-Model value. If the angle between these two planes in the Higgs rest frame is φ. The same information which we can find in the angular correlation of the tagging jets is encoded in the decay-plane correlation. T T Another promising way of telling apart the g µν structure and the CP-odd coupling is looking at the H → ZZ decay kinematics. while for the CP-even Higgs coupling g µν we find a > 1/4 depending on mH . [8].006 dσ/d∆Φjj (H→ττ) [fb] mH=120 GeV CP odd 0. Each of the decaying Z bosons defines a plane opened by the two lepton momenta. The only phenomenological advantage of the first is that a production-side correlation in weak boson fusion is independent of the Higgs decay and can be used for all Higgs masses and decay signatures. Both methods to determine the Higgs coupling structure are equivalent. does this observed scalar field have a potential with a minimum at a non-vanishing vacuum expectation value? Of course we could argue that a dimension-3 Higgs coupling to massive W bosons really is a Higgs-Goldstone self coupling. so we see it by finding a Higgs in weak boson fusion. it would be far more convincing to actually measure the self coupling of the actual Higgs field.

(k1 k2 ) (1. the Higgs self coupling sticks out of the Higgs parameters. so it only appears in the continuum (box) diagram and turns out to be numerically sub-leading over most of the relevant phase space.4.1. which means PT = 1 P2 = 1 and PT · P2 = 0.137) so the triangle diagram and the box diagram cancel. This way a shape analysis of the threshold behavior will allow us to experimentally exclude the case of λHHH = 0 which predicts an unobserved large enhancement of the production cross section at threshold.137) we can look at something like the s distribution or the mHH distribution and measure the threshold behavior. for integrated luminosities around 30 fb−1 the accuracy on the extracted couplings ranges from 30% to 70% error bands. for example in gluon fusion via the usual top quark loop  + 2 with kT = 2 Following exactly the same argument as presented in Section 1. From Section 1. a numerical fit of all couplings to all Higgs observables will be possible. both multiplying the transverse tensor µ ν g µν − k1 k2 /(k1 k2 ) for the incoming gluons gggH = −gggHH = −i αs +O 12π m2 H 4m2 t .(1. Contrasting more specific hypotheses. like a strongly interacting light Higgs.7 Beyond Higgs discovery 43 HH production. As you can find in the original paper.1 we can derive that the two tensor structures contributing to Higgs pair production for transverse gluons are √ √ µν 2PT = g µν − µν 2P2 = g µν ν µ k1 k2 (k1 k2 ) µ ν ν µ k2 kν kµ 2(k2 k3 )k1 k3 2(k1 k3 )k2 k3 kν kµ + 23 1 2 − − + 323 2 2 kT (k1 k2 ) kT (k1 k2 ) kT (k1 k2 ) kT (k1 k3 )(k2 k3 ) 2 − k3 . In the absence of any self coupling this threshold cancellation of the self coupling contribution with the continuum should be completely absent. 2 1 tensor structure is missing in gluon fusion Higgs production. If the Higgs is light and if all the of gluon fusion and weak boson fusion channels are indeed observed with sufficient luminosity. with the Standard Model will reduce these errors bars. Of course. which means the threshold contribution to Higgs pair production should be huge. The second 1. To see deviations from this self coupling in the first term of Eq.136) Close to the production threshold s ∼ (2mH )2 the leading contribution to the loop-induced production cross section for gg → HH is then proportional to 3m2 H gggH + gggHH s − m2 H 2 = gggH 3m2 H 1 −1 s − m2 H 2 ∼ gggH 3m2 H 1 −1 3m2 H 2 →0. (1. for example using the SFitter reconstruction tool [9]. At this point it is still under study if such a measurement will work at the LHC or what kind of LHC luminosities would be required. In this argument we need to use that the Higgs self coupling in the Standard Model is proportional to mH . (1. .135) 2 In this representation the two tensors are orthonormal. possibly to error bars of O(10%).2 on the effective Higgs-gluon coupling we know that in the low-energy limit we can compute the leading form factors associated with the triangle and box diagrams. but all Higgs couplings would be interesting to measure.4.

Michael and his collaborators also published a set of lecture notes on Higgs physics [17] – As always. Other papers for example by Guralnik. 12] or Brout and Englert [13]. The analysis is based on the decay HH → W W . – Abdelhak Djouadi’s compilation on ‘absolutely everything we know about Higgs phenomenology’ is indeed exhaustive. in case you need to catch up on this [20] – For some information on electroweak precision data and the ρ parameter. 5] – For more experimental details you might want to have a look at Karl Jakobs’ and Volker B¨ scher’s review of u LHC Higgs physics [16]. there is Michael Spira’s classic [3]. From the discussions it should be clear that at this stage we have to move on to QCD. there is a TASI lecture on the topic. Before moving on we mention a few papers where you can learn more about Higgs phenomenology at the LHC. – For a theory view on LHC Higgs physics. this topic is an example where you see the otherwise great particle data group fail — for one reason or another the section on electroweak precision data and new physics wrongly claims that a fourth chiral generation is ruled out. but they are rather short and not really on top of the phenomenological aspects of the topic. Hagen. [10]. . Luckily. there are James Wells’ TASI lectures [21]. Kibble [14] tend to be harder to read for phenomenologically interested students. This is where you can find more information on the low-energy theorem. – Wolfgang Kilian’s book [15] on the effective field theory approach to the Higgs mechanism is what the corresponding sections in these lecture notes are largely based on it. It has two volumes. 1. one on the Standard Model and another on its minimal supersymmetric extension [4. Figure from Ref. Dave Rainwater did not only write his thesis on Higgs searches in weak boson fusion [18]. for Higgs searches there are several very complete and useful reviews available: – You can look at the original articles by Higgs [11. he also left us all he knows about Higgs phenomenology at the LHC in his TASI notes [19].8 Further reading At this stage we are done with the brief review of the Higgs sector in the Standard Model and of contemporary Higgs phenomenology. in disagreement with the correct analysis [22]. – Tao Han wrote a very comprehensible set of TASI lecture notes on basic LHC phenomenology.44 1 HIGGS PHYSICS Figure 8: The (parton-level) sensitivity limits for Higgs pair production and a measurement of the Higgs self coupling. TASI lecture notes are generally what you should look for when interested in an area of high energy physics. to understand the strengths and weaknesses of these searches and what distinguishes a good from a bad search channel. mostly focused on gluon fusion production and its QCD aspects. By the way.

but I am not sure where you would get it from.8 Further reading 45 – For Higgs production in association with a W or Z boson and the way to observe boosted H → b¯ decays you b ¯ would need to read the original paper [7]. The same is true for the ttH analysis.1. – for cut rules and scalar integrals the best writeup I know is Wim Beenakker’s PhD thesis. .

46 2 QCD 2 QCD Just as Section 1 was not meant to be a complete introduction into electroweak symmetry breaking but was aimed at introducing the aspects of Higgs physics most relevant to the LHC this section cannot cover the entire field of QCD. These are processes with the HERA initial state e± p. This channel will lead us through all of the introduction into QCD. (2. This last part is fairly close to current research with the technical details changing rapidly. What we would like to understand instead are processes of the kind pp → W +jets. so if we assume the proton to consist of quarks and gluons we simply compute the process q q → γ. What is the production rate for this final state at the LHC. Z ¯ (2. the Drell-Yan process or lepton pair production. From a QCD perspective such processes are very complex.4. or the production of new particles with or without jets. 2. but the first two generation quarks and all leptons are usually assumed to be massless in LHC simulations. pp → H+jets. Modulo the SU (2) and U (1) charges which describe the Zf f and γ f f couplings in the Feynman rules −iγ µ ( PL + rPR ) with = e T3 − Qs2 w sw cw = r = Qe r= T3 =0 ¯ (Zf f ) ¯ (γf f ) (2. how do we account for quarks inside the protons. The general amplitude for massless fermions is M = −i¯(k2 )γ µ ( PL + rPR ) u(k1 ) v µ . e+ e− → γ. Z → + − . so we need to step back a little and start with a simple question: we know how to compute the production rate and distributions for photon and Z production for example at LEP. After discussing in detail infrared divergences in Sections 2. to get an idea how far the treatment of ultraviolet and infrared divergences work the same way. Instead. we do not have to care about factors of (−i) since we are only interested in the absolute .3. and what are the best-suited kinematic variables to use at a hadron collider? 2. so we only have to consider valence quark ¯ ¯ vs sea anti-quark scattering in the initial state. Z.3) Massless fermions are a good approximation at the LHC for all particles except for the top quark. Therefore. For the bottom quark we need to be careful with some aspects of this approximation. Luckily.2) will be the same as the corresponding matrix element squared for e+ e− → γ.1 Gauge boson production How do we compute the production of a weak gauge boson at the LHC? This process is referred to as Drell–Yan production.5 we will spend some time on modern approaches on combining QCD matrix element calculation at leading and next-to-leading order in perturbative QCD with parton showers. The problem with this approach is that for the LHC era such processes are not very useful.1 we will introduce the most important process at the LHC. we will focus on QCD as it impacts LHC physics and searches for new physics at the LHC.1 Drell-Yan process Most text books on QCD start from a very simple class of QCD processes. Squaring the matrix element in Eq. just walking through it in the opposite direction. Gluons do not couple to electroweak gauge bosons. In Section 2. ¯ pp → tt+jets. with an additional color factor.3) means adding the same structure once again.1) the matrix element and the squared matrix element for the partonic process q q → γ. Once we will arrive at infrared divergences in LHC cross sections we will specifically discuss ways of regulating them without introducing masses.1. namely deep inelastic scattering. like for example the Higgs searches discussed in the first part of the lecture. In our first attempts we will explicitly not care about additional jets. we will rely on toy models to illustrate different approaches. These kind of signal and background processes and their relevance in an LHC analysis we already mentioned in Section 1. Ultraviolet divergences and renormalization we will only mention in passing. Z under the assumption that the quarks are ¯ partons inside protons.(2.3 to 2.

Since we do not observe color in the initial state. 2 4Nc (2. the reduced polar angle y = (1 − cos θ)/2 = 0 · · · 1. γ5 } = 0 symmetric polarization sum 1 21 2 + r2 1 1 2 k2 γ ν / −gµν + qµ qν m2 Z + r2 Tr [k1 γ µ k2 γ ν ] / / −gµν + = 2Nc = 2Nc = 2Nc = 2Nc = 2Nc 2 µ ν ν µ + r2 [k1 k2 + k1 k2 − (k1 k2 )g µν ] −gµν + qµ qν m2 Z with (qk1 ) = −(k1 k2 ) with q 2 = (k1 + k2 )2 2 + r2 + r2 −2(k1 k2 ) + 4(k1 k2 ) + 2 2(k1 k2 ) + q4 q4 2 − 2m2 2mZ Z (k1 k2 )q 2 (−k1 k2 )2 − 2 mZ m2 Z 2 2 2 + r2 q2 + r2 m2 Z (2. For incoming electrons as well as incoming quarks this factor Kij includes 1/4 for the spins.1 Drell-Yan process 47 T value squared. so for the production of a massive Z boson we obtain |M| = spin. This reflects the fact that the Goldstone modes do not couple to massless fermions.(2. The second option has at least once lead to considerable confusion and later amusement at the Tevatron. Because the chiral projectors PL. . The color factor Nc accounts for the number of SU (3) states which can be combined to form a color singlet like the Z.2.4). just like the Higgs boson. we also average over the color. unless you want to compute gravitational effects on Higgs production at ATLAS and CMS..(2.5) In spite of our specific case in Eq. This means another factor 1/Nc in the averaged matrix element.4) looks that way matrix elements we compute from our Feynman rules are not automatically numbers with mass unit zero. An interesting aspect coming out of our calculation is that the 1/mZ -dependent terms in the polarization sum do not contribute — as far as the matrix element squared is concerned the Z boson could as well be transverse.color 2 u(k1 ) ( PL + rPR ) γ µ v(k2 ) v (k2 ) ( PL + rPR ) γ ν u(k1 ) ¯ ¯ −gµν + −gµν qµ qν m2 Z qµ qν m2 Z qµ qν + 2 mZ ∗ µ ν = Nc Tr [k1 ( PL + rPR ) γ µ k2 ( PL + rPR ) γ ν ] / / = Nc Tr [k1 γ µ ( PR + rPL ) ( PR + rPL ) k2 γ ν ] / / = Nc Tr k1 γ µ / = Nc 2 2 in unitary gauge with {γµ . when people noticed that the behavior of gauge bosons was dominated by gravity.4) The momenta are k1 and k2 for the massless incoming (anti-) fermions and q = −k1 − k2 for the outgoing gauge boson. If for the partonic invariant mass of the two quarks we introduce the Mandelstam variable s = (k2 + k2 )2 = 2(k1 k2 ) momentum conservation implies s = q 2 = m2 . What is still missing is an averaging factor for initial-state spins and colors. namely gauge bosons only went up or down.pol. and the color structure of the incoming q q pair has no impact on the Z–production matrix ¯ 2 element. The azimuthal angle φ plays no role at colliders.R = (1 γ5 )/2 are real and γ5 = γ5 is symmetric the left and right 1 handed gauge boson couplings do not change. This simple one-particle phase space has only one remaining free Z parameter. Any LHC Monte Carlo will either random-generate a reference angle φ for the partonic process or pick one and keep it fixed.. only the sum is included in Eq. This also means that the corresponding matrix element squared for the photon production simply means not replacing q 2 → m2 in the last Z step. which altogether becomes (in four space-time dimensions) Kij = 1 .

To make the results a little more compact we compute this process for a massless photon instead of the Z boson. In four space-time dimensions (this detail will become important later) we can compute a total cross section from the matrix element squared. where the Z might be or might not be on its mass shell.(2. i. (2. . s s = 2→1 y= τ= (2.4). we can define all momenta as incoming. for example as given in Eq. The assumption of a massless photon simplifies the couplings ( 2 + r2 ) = 2Q2 e2 as well as the polarization sums to −gµν : |M| = 16Nc (2Q2 e2 ) (2Q e2 ) = 64Nc Q2 Q e4 2 2 2 1 ν µ [k µ k ν + k1 k2 − (k1 k2 )g µν ] (−gµρ ) [pρ pσ + pσ pρ − (p1 p2 )g ρσ ] (−gνσ ) 1 2 1 2 q4 1 2 1 ν ν ν ρ k1ρ k2 + k1 k2ρ − (k1 k2 )gρ [pρ p2ν + p1ν pρ − (p1 p2 )gν ] 1 2 q4 1 2 = 64Nc Q2 Q e4 4 [2(k1 p1 )(k2 p2 ) + 2(k1 p2 )(k2 p1 ) − 2(k1 k2 )(p1 p2 ) − 2(k1 k2 )(p1 p2 ) + 4(k1 k2 )(p1 p2 )] q 1 2 = 128Nc Q2 Q e4 4 [(k1 p1 )(k2 p2 ) + (k1 p2 )(k2 p1 )] q 1 t2 u2 2 = 128Nc Q2 Q e4 2 + s 4 4 2 4 1 = 32Nc Q2 Q e 2 s2 + 2st + 2t2 s t t2 2 = 32Nc Q2 Q e4 1 + 2 + 2 2 (2.7) The coupling to the gauge bosons are and r for the incoming quarks and and r for the outgoing leptons.8) s s We can briefly check if this number is indeed positive. It would be replaced to include mW or the Higgs mass or the mass of a Kaluza-Klein graviton if needed. If the incoming trace includes the indices µ and ν and the outgoing trace the indices ρ and σ.1. using the definition of the Mandelstam variable t for massless external particles in terms of the polar angle t = s(−1 + cos θ)/2 = −s · · · 0: the upper phase space boundary t = 0 inserted into the brackets in Eq. The corresponding picture is two fermion currents interacting with each other through a gauge boson.(2. incoming anti-fermion k2 . while for the lower boundary t = −s we also find [· · · ] = 1 − 2 + 2 = 1. The additional Mandelstam variables we need to describe this (2 → 2) process are t = (k1 + p1 )2 < 0 and t = (k1 + p2 )2 < 0. outgoing fermion p1 and outgoing anti-fermions p2 .5.6) The mass of the final state appears in τ . so momentum conservation means k1 + k2 + p1 + p2 = 0. From LEP we know that such a heavy gauge boson we do not actually observe at colliders.e.5 = 0.1). The matrix element for the two sets of incoming and outgoing fermions becomes M = (−i)2 v (k2 )γ µ ( PL + rPR ) u(k1 ) u(p1 )γ ν ( PL + r PR ) v(p2 ) ¯ ¯ i q 2 − m2 Z −gµν + qµ qν m2 Z . For the central value t = −s/2 the minimum value of the brackets is [· · · ] = 1 − 1 + 0. To make it easy to switch particles between initial and final states. All we have to do is combine the traces properly. for the physical scenario where the initial-state fermions do not have enough energy to excite the intermediate Z boson. as usually with s + t + u = 0 for massless final state particles. What we should really calculate is the production for example of a pair of fermions through an s-channel Z and γ.48 2 QCD In terms of the two parameters y and φ we can describe the phase space for a one-particle final state. The matrix element for this process we can derive from the same Feynman rules in Eq. now for an incoming fermion k1 .8) gives [· · · ] = 1. with τ = 0 for a massless photon. When we combine the spinors correctly the matrix element squared factorizes into twice the trace we have computed before. the Z bosons link µ and ρ as well as ν and σ. as dσ dy π 2 Kij (1 − τ ) |M| (4π)2 1 − cos θ 2 m2 Z . The specific features of an intermediate massive Z boson we postpone to Section 2.(2.2.

For the photon the on-shell pole is not a problem.11). Everything else stays the same as for the Drell–Yan process: σ(e e → hadrons) = σ(e+ e− → µ+ µ− ) + − R≡ 4πα2 Nc 2 2 Qe Qq quarks 1 4 3s = Nc 3 + 2 2 9 9 4πα Q2 Q2 e µ 3s = 11Nc 9 (2.2.2 Massive intermediate states Before we move on to describing incoming quarks inside protons we should briefly consider the second Feynman diagram contributing to the Drell-Yan production rate in Eq.9) For our Drell–Yan process we then find the differential cross section in four space-time dimensions 1 dσ 1 π t t2 2 = 2 8 Q2 Q (4πα)2 1 + 2 + 2 2 dt s (4π)2 4Nc s s 1 2πα2 2 2 t t2 = 2 Q Q 1+2 +2 2 . but sum over all possible color combinations.1. 2 (2. R as a function of the collider energy is a beautiful measurement of the number and charges of the quarks in QCD. which in Eq. Strictly speaking we never hit it.12) for example for five quark flavors where the top quark is too heavy to be produced at the given e+ e− collider energy.11) As a side remark — in the history of QCD. namely the production rate of quarks in e+ e− scattering. which means that we hit the pole 1/q 2 in the matrix element squared only in the limit of zero incoming energy. s Nc s s which we can integrate over the polar angle or the Mandelstam variable t to compute the total cross section σ= = = = 1 2πα2 2 2 Q Q s2 Nc 2 0 (2.10) dt −s t t2 1+2 +2 2 s s 0 −s 2t3 1 2πα t2 2 Q2 Q t + + 2 s2 Nc s 3s 1 2πα 2t s 2 + 2 Q2 Q s− s2 Nc s 3s 1 2πα2 2 2 2 Q Q s Nc 3 ⇒ 2 2 s σ(q q → µ+ µ− ) ¯ QED = 4πα2 2 2 Q Q 3Nc s µ q (2. For small enough energies we can neglect the Z exchange contribution. . the on-shell or off-shell Z boson |M|2 = |Mγ + MZ |2 = |Mγ |2 + |MZ |2 + 2 Re MZ Mγ .(2. At leading order we can then compute the corresponding production cross sections for muon pairs and for quark pairs in e+ e− collisions. Moving the quarks into the final state means that we do not average of the color in the initial state.(2. This way. (2.13) where interference occurs in phase space regions where for both intermediate states the invariant masses of the muon pair are the same. It has zero mass. the same process but read right-to-left played a crucial role. 2.1 Drell-Yan process 49 The two-particle phase space integration for massless particles gives us s2 dσ dt = 2→2 π 2 Kij |M| (4π)2 t= s (−1 + cos θ) .9) gives us an averaging factor Kij = 1/4.

50 2 QCD because the energy of the incoming particles has to be large enough to produce the final state particles with their tiny by finite masses and with some kind of momentum driving them through the detector. The contour C1 through the upper half plane includes the pole at q0 ∼ −ω while the contour C2 includes the pole at q0 ∼ ω. In the Feynman i prescription the sign is crucial to correctly complete the q0 integration of the Fourier transform in the complex plane ∞ dq0 −∞ e−iq0 x0 = (θ(x0 ) + θ(−x0 )) q 2 − m2 + i = (θ(x0 ) + θ(−x0 )) ∞ dq0 −∞ ∞ e−iq0 x0 2 q0 − (ω 2 − i ) e−iq0 x0 √ (q0 − − i )(q0 + ω 2 − i ) e−iq0 x0 dq0 (q0 − ω(1 − i ))(q0 + ω(1 − i )) √ ω2 with ω 2 = q 2 + m2 dq0 −∞ = θ(x0 ) C2 +θ(−x0 ) C1 with = 2ω 2 (2. we have to make sure the exponent −ix0 iIm q0 = x0 Im q0 is negative.17) . This argument forces C1 to close for positive and C2 for negative imaginary parts in the complex q0 plane. or equivalently deforming our integration contours appropriately. Fourier transforming this result into momentum space we find G(x − x ) = + m2 G(x − x ) = = = d4 q −iq·(x−x ) ˜ e G(q) (2π)4 d4 q ˜ − m2 e−iq·(x−x ) G(q) (2π)4 d4 q ˜ (iq)2 + m2 e−iq·(x−x ) G(q) (2π)4 d4 q −iq·(x−x ) ! ˜ e −q 2 + m2 G(q) = δ 4 (x − x ) = (2π)4 ⇔ ˜ (−q 2 + m2 ) G(q) = 1 ⇔ ˜ G(q) ∼ − (2. In that case. There. keeping in mind that the integrand has two poles at q0 = ±ω(1 − i ). Z Before we can ask what such a pole means for LHC simulations we have to recall how we deal with it in field theory.14) d4 q −iq·(x−x ) e (2π)4 (2. Because the integrand has to vanish for large q0 . The sign of this infinitesimal shift we need to understand because it will become relevant for phenomenology when we introduce an actual finite decay width of intermediate states. the propagator contributes as 2 |M| ∝ 1/(s − m2 )2 which diverges on-shell Z bosons. all assuming ω > 0: ∞ dq0 −∞ q2 (−1)e−iωx0 eiωx0 e−iq0 x0 = 2πi θ(x0 ) + θ(−x0 ) 2+i −m ω + ω(1 − i ) −ω − ω(1 − i ) π = −i θ(x0 )e−iωx0 + θ(−x0 )eiωx0 . The Green function for a field obeying this equation is the inverse of the Klein–Gordon operator + m2 G(x − x ) = δ 4 (x − x ) . The contour integrals we can solve using Cauchy’s formula.15) 1 . q 2 − m2 The problem with the Green function in momentum space is that as an inverse it is not defined for q 2 = m2 . A problem arises when we consider the intermediate Z boson. We usually avoid this problem by slightly shifting this pole following the Feynman i prescription to m2 → m2 − i . ω (2.16) In the last step we have closed the integration contour along the real q0 axis in the complex q0 plane. They lie in the upper (lower) half plane for negative (positive) real parts of q0 . we encounter the same issue when we solve for example the Klein–Gordon equation.

Using this result we can complete the four–dimensional Fourier transform using d4 q q2 e−i(q·x) = −iπ − m2 + i = −iπ = −iπ = −iπ = −2πi = −2πi d3 q eiqx d4 q eiqx 1 ω 1 ω θ(x0 )e−iωx0 + θ(−x0 )eiωx0 θ(x0 )e−iq0 x0 δ(q0 − ω) + θ(−x0 )e−iq0 x0 δ(qo + ω) 1 [θ(x0 )δ(ω − q0 ) + θ(−x0 )δ(ω + q0 )] ω 1 2 2 2ω θ(x0 )δ(ω 2 − q0 ) + θ(−x0 )δ(ω 2 − q0 ) ω with δ(x) = δ(−x) d4 q e−i(q·x) d4 q e−i(q·x) 2 d4 q e−i(q·x) [θ(x0 ) + θ(−x0 )] δ(q0 − ω 2 ) d4 q e−i(q·x) [θ(x0 ) + θ(−x0 )] δ(q 2 − m2 ) 2 with q0 − ω 2 = q 2 − m2 . The solution with Re q0 > 0 would sit in the quadrant with Im q0 > 0 and the second pole at a negative real and imaginary part. From the same couplings governing the Z decay the Z propagator receives corrections.2. +) x0 < 0 θ(−x0 ) Im q0 > 0 +1 + 1 2ω (−.18) This is exactly the usual decomposition of the propagator function ∆F (x) = θ(x0 )∆+ (x) + θ(−x0 )∆− (x) into positive and negative energy contributions. the two poles would give the same result as for the correct sign of i . All steps leading to the propagator function in Eq.(2.18) we summarize in Table 1. except with a wrong over-all sign. For the wrong sign of i the two poles in the complex q0 plane would be mirrored by the real axis. (2. The residue in the Cauchy integral would now include a factor +1/(2ω). while going from negative to positive Req0 the contour C2 is defined clockwise. When we are interested in the kinematic distributions of on-shell massive states the situation is a little different.1 Drell-Yan process 1 q 2 − m2 + i q0 = ω(1 − i ) q0 = −ω(1 − i ) (+. which means we cannot simply apply the Feynman i prescription as if we were dealing with an asymptotic stable state. At the end. −) x0 > 0 θ(x0 ) Im q0 < 0 −1 + 1 2ω (−. +) x0 < 0 θ(−x0 ) Im q0 > 0 +1 − 1 2ω 1 q 2 − m2 − i q0 = ω(1 + i ) q0 = −ω(1 + i ) (+.18) for both signs of i . The factor (−1) in the C2 integration arises because Cauchy’s integration formula requires us to integrate anti-clockwise.(2. Let us briefly recapitulate what would have happened if we instead had chosen the Feynman parameter < 0. To be able to close the integration path in the upper half plane in the mathematically positive direction the real pole would have to be matched up with θ(−x0 ). Measurements of differential distributions for example at LEP include information on the physical width of the decaying particle. −) x0 > 0 θ(x0 ) Im q0 < 0 −1 − 1 2ω 51 pole complex quadrant convergence: x0 Imq0 < 0 part of real axis closed contour direction of contour residue Fourier exponent all combined e−iωx0 − e−iωx0 θ(x0 ) 2ω e+iωx0 − e+iωx0 θ(−x0 ) 2ω e+iωx0 + e+iωx0 θ(−x0 ) 2ω e−iωx0 + e−iωx0 θ(x0 ) 2ω Table 1: Contributions to the propagator function Eq. for example including fermion loops: .

52 Such one-particle irreducible diagrams can occur in the same propagator repeatedly.4.1 tells us schematically written Im M 2 ∼ M 2 |cut ≡ Γ because cutting the one-loop bubble diagram at the one possible place is nothing but squaring the two tree-level matrix element for the decay Z → µ+ µ− .1 the one-loop correction M 2 depends on the momentum flowing through the propagator. (s − m2 )2 + m2 Γ2 Z (2. For example for a massive scalar or gauge boson with a real correction M 2 this reads i iZ = 2 q 2 − m2 − M 2 q − m2 0 for q 2 ∼ m2 . Going back to the Drell-Yan process q q → µ+ µ− we now know that for massive unstable particles the Feynman ¯ epsilon which we need to define the Green function for internal states acquires a finite value. m (2. apart from the additional factor m due to dimensional analysis. The link between the self-energy matrix element squared M 2 and the partial width we can illustrate remembering one way to compute scalar integrals or one-loop amplitudes by gluing them together using tree-level amplitudes. − m2 − M 2 0 (2. is the sign of the imΓ term which just like the i prescription is fixed by causality.20) including a renormalized mass m and a wave function renormalization constant Z. (2. One thing that is important to remember.21) except we are not meant to go into the limit Γ → 0. Schematically written as a scalar they are of the form q2 i i i i i i 2 2 2 2 + q 2 − m2 (−iM ) q 2 − m2 + q 2 − m2 (−iM ) q 2 − m2 (−iM ) q 2 − m2 + · · · − m0 0 0 0 0 0 = i q 2 − m2 0 n=0  1 M2 1− 2 q − m2 0 2 QCD M2 2 − m2 q 0 n i = 2 q − m2 0 = summing the geometric series q2 i . The Cuskosky cutting rule discussed in Section 1. Above a certain threshold it can develop an imaginary part because the momentum flowing through the diagram is large enough to produce on-shell states in the loop. Just as for the ggH coupling such absorptive parts appear when a real decay like Z → µ+ µ− becomes kinematically allowed. The imaginary 0 part looks like the Feynman i term discussed before iZ i = 2 q 2 − m2 − ReM 2 − iImM 2 q − m2 − iZImM 2 0 iZ ⇔ ≡ 2 q − m2 + imΓ ! Γ=− Z Im M 2 (q 2 = m2 ) . proportional to the total width of the unstable particle.19) Requiring that the residue of the propagator be one at the pole we renormalize the wave function and the mass in the corresponding process. The resulting shape of the propagator ¯ squared is a Breit–Wigner propagator σ(q q → Z → µ+ µ− ) ∝ ¯ 1 s − m2 + imΓ Z 2 = 1 .4. The important step in our argument is that in analogy to the effective ggH coupling discussed in Section 1.22) . After splitting M 2 (q 2 ) into its real and imaginary parts we know what to do with the real part: the solution to q 2 − m2 − ReM 2 (q 2 ) = 0 defines the renormalized particle mass q 2 = m2 . This definition of a propagator of an unstable particle in the s channel is what we need for the second contribution to the Drell-Yan process: q q → Z ∗ → µ+ µ− .

c. while the gluon pdf is small at x ∼ 1 and grows very rapidly towards small x. Eq. While we cannot actually compute parton distribution functions fi (x) as a function of the momentum fraction x there are a few predictions we can make based on symmetries and properties of the hadrons. In contrast to so-called structure functions a pdf is not an observable. g and depending on the details b. Different parton densities have very different behavior — for the valence quarks (uud) they peak somewhere around x 1/3. It does not keep information about the structure of the matrix element. e. for our purposes u. They have the properties – for small energies the γ contribution dominates and can be linked to R. This steep gluon distribution was initially not expected and means that for small enough x LHC processes will dominantly be gluon fusion processes. defining a Breit-Wigner propagator. In the narrow-width approximation. e = lim √ e−x /σ = lim σ→0 σ π Γ→0 π x2 + Γ2 2π (2. the situation is very different. – in the tails of the Breit-Wigner distribution we expect Z − γ interference. For this momentum fraction x = 0 · · · 1 the parton density function (pdf) is denoted as fi (x). the matrix element factorizes into the production process times the branching ratio for Z → µµ.1 the discussion of different energy regimes for R experimentally makes sense — at an e+ e− collider we can tune the energy of the initial state.1. where i are the different partons in the proton.24) The second distribution is a Gaussian and the third one we would refer to as a Breit–Wigner shape while most other people call it a Cauchy distribution. At hadron colliders. If for now we assume that quarks move collinear with the surrounding proton. the two amplitudes interfere.g. the (2 → 2) production cross section also includes the matrix element for the decay Z → µ+ µ− in the numerator. when a non-trivial structure of the numerator gives us the spin and angular correlations between the production and decay processes. s. simply by definition of the Breit-Wigner distribution Γ→0 lim π Γµµ = Γµµ δ(s − m2 ) ≡ π BR(Z → µµ) δ(s − m2 ) . and that is the definition of the one-dimensional Dirac delta distribution in three ways and including all factors of 2 and π δ(x) = 2 2 dq −ixq Γ 1 1 . For some typical part of the relevant parameter space (x = 10−3 · · · 10−1 ) it roughly scales like fg (x) ∝ x−2 .23) uses a mathematical relation we might want to remember for life. All incoming partons we assume to be massless.2.e. Such arguments for example lead to sum rules: . We immediately see that this narrow width approximation is only exact for scalar particles. – in the limit of vanishing width the Z contribution factorizes into σ · BR. – quantum effects allow unstable particles like the Z to decay off-shell. i. which means it has to produce reasonable results when integrated over as an integration kernel. – for large energies we are again dominated by the photon channel. It is a distribution in the mathematical sense. Z exchange and the γ − Z interference terms. The energy distribution of incoming quarks as parts of the colliding protons has to be taken into account. Strictly speaking. that at the LHC incoming partons have zero pT .3 Parton densities At the end of Section 2. Z Z (s − m2 )2 + m2 Γ2 Γtot tot Z (2. – on the Z pole the rate is regularized by the Z width and Z contribution dominates over the photon.1 Drell-Yan process 53 When taking everything into account. towards smaller x values it becomes even steeper.1.23) The additional factor π will be absorbed in the different one-particle and two-particle phase space definitions. By now we have all the necessary background to compute all Feynman diagrams contributing to muon pair production at a hadron collider. 2. so we end up with three distinct contributions: γ exchange. d.(2. we can define a probability distribution for finding a parton just depending on the respective fraction of the proton’s momentum.

i. not three dimensional vectors. final state (2. the total momentum of the proton has to consist of sum of all parton momenta. Half of the proton momentum is carried by the gluons. the xi integrations and the phase space integrations can of course be interchanged. 2.1. ¯ (2.26) Similarly. Such a gauge boson not recoiling against anything else cannot develop a finite transverse momentum. but Jacobians will make your life hell when you attempt to get them right. they have zero transverse momentum. We can write this as the expectation value of xi 1 xi = 0 dx x q fq (x) + q ¯ fq (x) + fg (x) ¯ =1 (2.28) where i. In Section 2. which is an exact symmetry of QCD. Of course.e.g.11).29) This is a relation between two-dimensional.27) What makes this prediction interesting is that we can compute the same sum only taking into account the measured quark and antiquark parton densities and find that the momentum sum rule only comes to 1/2. for all partons we know p fq¯(x) = fq (x) ¯ p fq¯(x) = fq (x) ¯ p ¯ fg (x) = fg (x) (2. Given the correct definition and normalization of the pdf we can now compute the hadronic cross section from its partonic counterpart. like the QED result in Eq. The partonic energy of the scattering process is √ s = x1 x2 S with the LHC proton energy of eventually S = 14 TeV. The partonic cross section σ corresponds to the ˆ cross sections σ computed for example in Eq. e. two such differences between three particles in the final state. once we decay this gauge boson. j are the incoming partons with the momentum factions xi. While differences of azimuthal angles. each muon will have transverse momentum.(2.j .(2. When we express a general n–particle cross section σ including the phase space ˆ integration. for example into a pair of muons. are observables the over-all angle is a symmetry of the detector as well as of our physics. For example. It has to include all the necessary θ and δ functions for energy–momentum conservation. only their vector sum will be zero: pT.11). If the proton consists of three valence quarks uud.54 2 QCD The parton distributions inside the anti-proton are linked to those inside a proton through CP symmetry. The expectation values for the signed numbers of up and down quarks inside a proton have to fulfill 1 1 Nu = 0 dx (fu (x) − fu (x)) = 2 ¯ Nd = 0 dx (fd (x) − fd (x)) = 1 . plus quantum fluctuations from the vacuum which can either involve gluons or quark–antiquark pairs. This reflects the fact that the incoming quarks are collinear with the protons. For several particles in the final state we can define an azimuthal angle in the plane transverse to the beam direction.5 we will discuss an easier way to compute kinematic distributions instead of the fully integrated total rate in Eq. in the production of a single electroweak gauge boson we see that this produced particle does not have any momentum transverse to the beam direction. the contribution from the sea quarks has to be symmetric in quarks and antiquarks.1. Therefore.28). as 1 1 σtot = 0 dx1 0 dx2 ij fi (x1 ) fj (x2 ) σij (x1 x2 S) .25) for all values of x.j = 0 .4 Hadron collider kinematics Hadron colliders have a particular kinematic which is reflected in the choice of kinematic variables.(2. . ˆ (2.

(2.30) we can re-write in a way which allows us to compute it from the four-momentum for example in the LHC lab frame E + pL cosh y + sinh y 1 ey 1 1 log = log −y = y .30) E pL = exp y = 0 1 1 0 E pL = 1 cosh y + 1 . π] we can define a distance measure inside the detector as ∆Rij = (∆yij )2 + (∆φij )2 (2.(2.1 Drell-Yan process 55 In the beam direction or longitudinal direction the partonic initial state is not well defined.1. To replace both. = log 2 E − pL 2 cosh y − sinh y 2 e (2. For such massless particles we can simplify the formula for the rapidity Eq. instead expressing the rapidity in terms of the polar angle θ.35) . We use the fact that (only) for massless particles E = |p| y= 1 E + pL 1 |p| + pL 1 1 + cos θ 1 log = log = log = log 2 E − pL 2 |p| − pL 2 1 − cos θ 2 1 θ tan2 2 = − log tan θ ≡η 2 (2. They do not have a rest frame. which means we can only boost them from one finite-momentum frame to the other. following its extraction method from Eq.32) E cosh y + pL sinh y pL cosh y + E sinh y which gives for the combined rapidity. Because this is about the same range as the range of the azimuthal angle [0. A longitudinal boost for example from the rest frame of a massive particle reads E pL = exp y 0 1 1 0 m 0 m 0 y3 0 1 y2 0 1 1 ··· = 1 +y 1 + 1+ 1 0 2 6 1 0   yj yj  m 0 1 + = 1 1 0 1 0 j! j! j even j odd = 1 cosh y + 1 0 1 1 0 sinh y m 0 =m cosh y sinh y . x1 and x2 with more physical variables we need a longitudinal measure which is preferably well behaved under longitudinal boosts.(2.31) We can explicitly check that indeed the rapidity is indeed additive.30) In the first line we already see that the combination of two such exponentials involving y1 and y2 is additive y = y1 + y2 .28) we integrate over the two momentum fractions x1. 2 E − pL 2 (E − pL )(cosh y − sinh y ) 2 E − pL (2. pL ) in Eq.31).33) This combination of several longitudinal boosts is important in the case of massless particles. The rapidity y as defined in Eq. corresponding to pseudo-rapidities of 2. but coincides with the actual rapidity only for massless particles.34) This pseudo-rapidity η is more handy. To get an idea about the experimental setup at the LHC — in CMS and ATLAS we can observe all particles very well to polar angles of 5 to 10 degrees.2.(2. (2. by applying a second longitudinal boost to (E. 0 1 1 0 sinh y E pL (2.(2. In Eq.31) E + pL 1 (E + pL )(cosh y + sinh y ) 1 E + pL 1 log = log = log +y =y+y .2 and can at best determine their product x1 x2 = s/S from the final state kinematics.4 to 3.

(2. 0. The Jacobian for this change of variables reads ∂(q 2 .41) In contrast to the original form the integration over the hadronic phase space this form reflects the kinematic observables. the invariant mass of the final-state particle q 2 and its rapidity. The two incoming and approximately massless protons have the momenta p1 = (E. by leaving out the double integration in Eq. 2. 0. On the other hand.41) dσ(pp → µ+ µ− ) dq 2 dy = QED 4πα2 Q2 µ 3Nc q 4 Q2 x1 fj (x1 ) x2 f¯ (x2 ) .1. 0.36) which for the momentum of the final-state gauge boson in terms of the parton momentum fractions means       cosh y cosh y x1 + x2    0   0  ! √ 2 0     q = x1 p1 + x2 p2 = E   0  = q  0  = 2E x1 x2  0  sinh y sinh y x1 − x2 ⇔ x1 + x2 1 cosh y = √ = 2 x1 x2 2 x1 + x2 x2 x1 ⇔ ey = x1 .39) instead in terms of the hadronic phase space variables x1. −E) S = (2E)2 . j j j (2. (2.39) gives us σ(pp → µ+ µ− ) QED (2. y) q2 x2 S x1 S =s= = 1/(2x1 ) −1/(2x2 ) ∂(x1 . x2 (2. the rapidity integral still suffers from the fact that at hadron colliders we do not know the longitudinal kinematics of the initial state and therefore have to integrate over it.40) = = 4πα2 Q2 µ 3Nc 4πα 3Nc 2 dq 2 dy dq 2 dy x1 x2 1 q2 q2 1 q4 Q2 fj (x1 ) f¯ (x2 ) j j j Q2 µ Q2 x1 fj (x1 ) x2 f¯ (x2 ) .(2. The transverse momentum we already know is zero. 0. E) p2 = (E.42) . S (2. The transverse momentum of the two leptons is therefore fixed by momentum conservation. x2 ) x1 x2 which inserted into Eq.38) These relations allow us to for example compute the hadronic total cross section for lepton pair production in QED σ(pp → µ+ µ− ) QED = 4πα2 Q2 µ 3Nc 1 dx1 dx2 0 j Q2 fj (x1 ) f¯ (x2 ) j j 1 . j j j (2.2 in terms of the kinematic final state observables q 2 and y. q2 (2.5 Phase space integration In the previous example we have computed the simple two dimensional distribution.37) This result can be combined with x1 x2 = q 2 /S to obtain x1 = q2 y e S x2 = q 2 −y e .56 2 QCD Still studying the production of a single gauge boson at the LHC we can express the final state kinematics in terms of two parameters. Remember that in σ we have already integrated over the (symmetric) azimuthal angle φ and the polar angle ˆ Mandelstam variable t.

The tool which translates the vector of integration variables y into the external momenta is called a phase space generator. but imagine we want to compute a distribution dσ/dpT .1 Drell-Yan process 57 This expression we can numerically evaluate and compare to experiment.45). If we think of the integration as an 2 integration over the unit cube.43) the physical y1 distribution is then given by σ= dσ dy1 = 0 y1 dy1 · · · dyN f (y) = 0 dy2 · · · dyN f (y1 ) = dy1 dσ dy1 (2. The integration variable y we can divide into a discrete set of points yj . for each y1 value of the complete y1 · · · yN integration we decide where the value y1 goes in this array and add f (y) to the corresponding column. In terms of Eq. so we will have to rely on numerical integration tools no matter what we are doing. Numerically we can compute this distribution in two ways: one way corresponds to the first line in Eq.(2.46) numerically and read off the pT distribution as a side effect of the calculation of the total rate. Then. this weight needs to be combined with the matrix element squared |M| . Moreover. Because the phase space is not uniquely defined in terms of the integration variables. 1]. The procedure outlined above has an interesting interpretation. the phase space generator also returns the Jacobian JPS . like for example the distribution of the transverse momentum of a muon in the Drell–Yan process. the rapidity y and the momentum transfer q 2 are by no means the only distribution we would like to look at. for example equi-distant on the y axis or by choosing a chain of random numbers yj [0. we define an array with the size given by the number of bins in the y1 integration.2. Histograms mean that computing a total cross section numerically we can trivially extract all distributions in the same process.43) with an appropriate function FPS .(2.45) and means numerically evaluating the y2 · · · yN integrations and leave out the y1 integration. over the parton densities f (x) we will have to integrate numerically. because of energy–momentum conservation. but before we discuss how to best evaluate such an integral numerically. In a minute. First. with the appropriate Jacobian. This set of columns is referred to as a histogram and can be produced using canned routines. we will discuss how such a random number can be chosen in more or less smart ways. (2.(2.45) 0 dy1 · · · dyN f (y) δ(y1 − y1 ) . These momenta are not all independent. Just like computing the interesting observables we can compute the momenta of all external particles. However. we print these columns as a function of y1 to see the distribution. In the second step we have re-written the phase space integral as an integral over the four-dimensional unit cube. This histogram approach does not sound like much. Looking at a simple (2 → 2) process we can write the total cross section as σtot = dφ d cos θ dx1 dx2 FPS |M| = 0 2 1 dy1 · · · dy4 JPS (y) |M| . Imagine we do the entire phase space integrations numerically. Once we . where pT (y) is a complicated function of the integration variables. The second and much more efficient option corresponds to the second line of Eq. let us first illustrate that this integral is much more useful than just providing the total cross section. If we are interested in the distribution of an observable. Finally. but this can be taken care of. with the delta distribution defined for discretized y1 . We then simply compute dσ = dpT dy1 · · · dyN f (y) δ pT (y) − p0 T (2.44) Without any loss of generality we assume that the integration boundaries are 0 · · · 1. In the latter case we need to keep track of the bin widths (∆y)j . 2 (2. The result will be a function 0 of y1 which we then evaluate at different points y1 . Like any integral we can numerically evaluate this phase space integral by binning the variable we integrate over 1 dy f (y) 0 −→ j (∆y)j f (yj ) ∼ ∆y j f (yj ) . we need to compute dσ/dpT as a function of pT . called the phase space weight.

If you want to extract distributions via histograms you have to therefore add the total weight 2 W = WMC JPS |M| to the columns. This is not trivial once we go beyond leading order. First. and there is no guarantee for each of them to be positive. 1]. the momenta of all external particles and the weight W . If we divide each of these directions in 100 bins. This means that our distributions will if anything become more ragged. the combined 2 weight W = JPS |M| is the probability that this configuration will appear at the LHC.(2. we can for example store in a big file. and finally keep the event only if Wj /Wmax > r. p1 · · · pM )j corresponding to the vector yj . you will find that VEGAS will call the function which computes the weight W = JPS |M| for a number of phase space points and average over these points.58 2 QCD compute the unique phase space configuration (k1 . In this approach we define a distribution pY (y) such that for a one-dimensional integral we can replace the binned discretized phase space integral with a discretized version based on a set of random numbers Yj over the y integration space 1 g(Y ) = 0 dy pY (y) g(y) −→ 1 N g(Yj ) .47) All we have to make sure is that the probability of returning Yj is given by pY (y) for y < Yj < y + dy. In the ideal case where we exactly know the form of the integrand and can map it into our random numbers. From this point of view negative values for parton densities f (x) < 0 are in principle not problematic. p1 · · · pM )j can be measured. we need to add several contributions to produce a physical event. There. To achieve this we normalize all our event weights to the maximum weight Wmax . The numerical phase space integration for many particles faces two problems. The only complication is that the probability of a certain configuration is not only given by the frequency with which it appears. as long as we always keep a positive hadronic rate dσpp→X > 0. This task is called importance sampling and you will have to find some documentation for example on the standard implementation VEGAS to look at the details. the number of phase space points we need to evaluate for a 2 → 4 process is 10015 = 1030 . I should add that experimentalists have good reasons to want such unweighted events which feed best through their detector simulations.(2. which is exactly what we want — the larger the weight the more likely the event stays.47) has the advantage that we can naively generalize it to any number of N dimensions. pY (Yj ) (2. . like for example different n–particle final states. The challenge in this translation is that we will always lose events. All we have to do to compute the integral is average over N phase space values of f /pY . We have to ensure that after adding up all contributions and after integrating over any kind of unphysical degree of freedom we might have introduced. the error of the numerical integration will be zero. i. k2 . compute the ratio Wj /Wmax [0. just by organizing the random numbers Yj in one large vector instead of an N –dimensional array. the partonic phase space for n on-shell particles in the final state has 3(n + 2) − 3 dimensions. k2 . This means we have to unweight the events by translating the event weight into frequency. So what we have to find is a way to encode f (Yj ) into pY (Yj ).43) we can re-write in the same manner 1 1 dN y f (y) = 0 0 dN y f (y) pY (y) = pY (y) f (Y ) pY (Y ) → 1 N j f (Yj ) . This guarantees that each event j survives with a probability Wj /Wmax .48) Writing the phase space integral in this discretized form naively does not involve the number of dimensions anymore. This simulation is not quite what experimentalists want — they want to represent the probability of a certain configuration appearing only by its frequency. The last comment is that if the phase space configuration (k1 . This means we do not only integrate over the phase space.e. Eq. which is not realistic. the probability of a physics configuration is positive. Our N –dimensional phase space integral shown in Eq. To integrate over a large number of dimensions we use Monte Carlo integration. but also by the additional explicit weight. 1]. So if it was not for the experimentalists we would never use unweighted events. These events. j (2. we really simulate LHC events. i. So when we run our numerical integration through the phase space generator and histogram all the distributions we are interested in we really generate weighted events. then generate a flat random number r [0. 2 Technically. its weight Wj better be positive. but including another weight factor WMC representing the importance sampling.e.

50) P (s. Take for example the integration over the partonic momentum fraction. In Section 2. Of course. m) = (s − m2 )2 + m2 Γ2 For example. 2.51) This the most useful phase space mapping I know. 2. The aspect of renormalization which will guide us through this section is the appearance of the renormalization scale.49) and improve our integration significantly. 16π 2 q 2 − m2 (q + p)2 − m2 (2.2. (2. which means once we know all parameters including their counter terms our theory becomes predictive. scales automatically arise from the regularization of infrared or ultraviolet divergences.3 we will interpret the use of running parameters physically and see that in perturbation theory they re-sum classes of logarithms to all orders in perturbation theory. but before dealing with them it is very instructive to see what happens to the much better known ultraviolet divergences. They arise from the infrared momentum regime. but a well-chosen set of integration parameters will speed up simulations very significantly.1 we will review how such ultraviolet divergences arise and how they are removed.3 we will see that in QCD processes we also encounter another kind of divergences.2. Infrared divergences is what this lecture is really going be about. in Section 2. any adaptive Monte Carlo will eventually converge on such an integrand.2. m. In that situation we can substitute dx δ C = x d log x log δ d log x dx −1 C = x d log x C log δ (2. In perturbation theory. mΓ mΓ dz −1 (2. i. This computation will lead to ultraviolet divergences which can be absorbed in counter terms for any parameter in the Lagrangian.2.2 Ultraviolet divergences 59 The second numerical challenge is that the matrix elements for interesting processes are by no means flat. a Standard-Model Higgs boson with a mass of 120 GeV has a width around 0. i. is one of the most important things to understand about field theories. m) ≡ d4 q 1 1 .1 Counter terms Renormalization.2 we will review how running parameters appear in this procedure.3 we will follow exactly the same steps for infrared divergences and develop some crucial features of hadron collider physics. which means √ that the integration over the invariant mass of the Higgs decay products s requires a relative resolution of 10−5 .2 Ultraviolet divergences From general field theory we know that when we are interested for example in cross section prediction with higher precision we need to compute further terms in its perturbative series in αs . Since this is unlikely to be achievable.e. In Sections 2. where the integrand is usually falling off at least as 1/x.e. Later in Section 2. What we should really do is find a substitution which produces the inverse Breit–Wigner as a Jacobian and leads to a flat integrand — et voil´ a ds C = 2 )2 + m2 Γ2 (s − m dz C 2 )2 + m2 Γ2 ds (s − m (s − m2 )2 + m2 Γ2 C = dz 2 )2 + m2 Γ2 mΓ (s − m 1 s − m2 = dz C with tan z = .005 GeV. In Section 2. Finally. corresponding to two virtual scalars with mass m and with the external momentum p flowing through the diagram B(p2 .52) . how scale dependence is linked to the appearance of divergences. The crucial feature is that for a renormalizable theory like our Standard Model including a Higgs boson the number of counter terms is finite. Let us move on to a more relevant example: particularly painful are intermediate particles with Breit–Wigner propagators squared which we need to integrate over the momentum s = p2 flowing through it 1 . We would therefore like to help our adaptive or importance sampling Monte Carlo by defining the integration variables such that the integrand becomes as flat as possible.2. the proper treatment of ultraviolet divergences. We can see this writing down a simple scalar loop integral. more detailed discussions you can find in any book on advanced field theory.

(2. They cancel the poles in the combined leading order and virtual one-loop prediction |MLO (g) + Mvirt | = |MLO (g)| + 2 Re MLO (g)Mvirt + · · · → |MLO (g + δg)| + 2 Re MLO (g)Mvirt + · · · with g → g bare = g + δg and δg ∝ αs / . so its divergence has to be proportional to log Λ/µ2 with a dimensionless prefactor and some scale µ2 which is an artifact R R of the regularization of such a Feynman diagram.53) The constants Ci in the series in 1/ε depend on the loop integral we are considering. Counting powers of q in Eq. As we can see in Eq. Because the ultraviolet behavior of the integrand or integral cannot depend on any parameter living at a small energy scales.52) we see that the integral is logarithmically divergent in the ultraviolet. Let us have a closer look. The scalar two-point function has mass dimension zero.(2. for example absorbing the δg corrections in the leading order and virtual interference. no question asked about convergence radii µ2 R C−1 + C0 + O( ) = e2 log µR C−1 + C0 + O( ) C−1 + C0 + O( ) = 1 + 2 log µR + O( 2 ) = → C−1 C−1 + C0 + C−1 log µ2 + O( ) R + C0 + C−1 log µ2 R + O( ) . We can keep track of the renormalization scale best expanding the prefactor of the regularized but not yet renormalized integral in Eq. while the R poles 1/ cancel just fine.(2. Counter terms we include by shifting the renormalized parameter in the leading order matrix element. with massless scalars for ghosts. The poles in 1/ will cancel with the universal counter terms once we renormalize the theory.55) In the last step we have by hand corrected for the fact that log µ2 with a mass dimension inside the logarithm cannot R appear in our calculations.60 2 QCD Such two-point functions appear for example in the gluon self energy. so we have to regularize it. (2.(2. One regularization scheme is to introduce a cutoff into the momentum integral Λ. To regularize the ultraviolet divergence we assume > 0 and find mathematically well-defined poles 1/ . M2 (2. It is designed not to break gauge invariance and naively seems to not introduce a mass scale µR . A more elegant regularization scheme is dimensional regularization. and with massive scalars for supersymmetric scalar quarks. for example through the so-called Pauli–Villars regularization. (2. with a Dirac trace in the numerator for quarks.53) in a Taylor series in . From somewhere else in our calculation the scale dependence logarithm will be matched .54) the counter terms do not come with a factor µ2 in front.54) 2 2 2 where the dots indicate higher orders in αs . The integration measure 1/(16π 2 ) is dictated by the usual Feynman rules for the integration over loop momenta. This is the reason we usually find factors 1/(4π)2 = π 2 /(2π)4 in front of the loop integrals. the scale factor µ2 will not be matched between the actual ultraviolet divergence and the R counter term. the parameterization of the ultraviolet divergence in Eq. Regularizing means expressing the divergence in a well-defined manner or scheme allowing us to get rid of it by renormalization.52) cannot involve the mass m or the external momentum p2 . Defining scalar integrals with the integration measure 1/(iπ 2 ) will make for example C−1 come out as of the order O(1). When we shift the momentum integration from 4 to 4 − 2 dimensions and use analytic continuation in the number of space-time dimensions to renormalize the theory a renormalization scale µR appears if we want to ensure the two-point function and with it observables like cross sections have the correct mass dimension d4 q · · · −→ µ2 R 16π 2 iµ2 d4−2 q R ··· = 16π 2 (4π)2 C−1 + C0 + C1 + O( 2 ) . Therefore. though.

since we need to introduce the arbitrary energy scale M to separate the universal logarithm of the renormalization scale and the parameter-dependent logarithm of the physical process. For QCD corrections. but as we will see in Section 3 this is not where at the LHC we expect to measure particle masses. such schemes do not automatically work universally. An example for such a physical renormalization scheme is the on-shell scheme for masses. The interesting aspect of this squark mass counter term is that it also depends on the gluino mass. There is no way of removing ultraviolet divergences without introducing the renormalization scale if we keep track of the mass dimensions of our regularized result.(2. and vice versa.54) we replace the leading-order mass with the bare mass.(2. This little argument shows that also in dimensional regularization we introduce a mass scale µR which appears as log µ2 /M 2 in the renormalized expression R for our observables. There we choose the counter term of the weak mixing angle such that an on-shell Z boson cannot oscillate into a photon. They need to include a given divergence. This opens many ways to define a counter term for example based on physical processes where counter terms do not only cancel the pole but also finite corrections in perturbation theory. On the other hand. For the computation of total cross sections at hadron colliders or the production thresholds at e+ e− colliders the pole mass is not well suited at all. Needless to say. This logarithm is the big problem in this universality argument. To finalize our discussion of process dependent mass renormalization we quote the result for a scalar supersymmetric quark. for which we then insert the expression in terms of the renormalized mass and the counter term mbare → mt + δmt t = mt + mt αs CF 4π αs CF ≡ mt + mt 4π 1 µ2 m2 3 − + γE − log(4π) − log R − 4 + 3 log t 2 M M2 3 m2 4πµ2 1 1 R − − 4 + 3 log t ⇔ µR ≡ − γE + log M 2 . (2. it is the most convenient scheme keeping all exchange symmetries of the two scalars. For the top mass this means that just like in Eq.2 Ultraviolet divergences 61 with a log M 2 where M 2 is the typical mass or energy scale in our process. (2. In the limit of vanishing gluino contribution the counter term becomes simpler and is again proportional to the squark mass itself mbare → mq + δmq = mq + mq ˜ ˜ ˜ ˜ q ˜ αs CF 4π m2 1 q ˜ − − 3 + log 2 ˜ M .54) are not uniquely defined. but we are free to add any finite contribution we want. This means that once we define a more appropriate scheme for heavy particle masses in collider production mechanisms it better be numerically close to the pole mass.58) .e.2. it describes for example the kinematic features of top pair production at hadron colliders in a stable perturbation series. on the other hand. (2. a squark. We can generalize this scheme for mixing scalars as they for example appear in supersymmetry. Counter terms as they schematically appear in Eq. to return finite observables. where we define a counter term such that external on-shell particles do not receive any corrections to their masses. q ˜ g ˜ i. so we should do fine with something very similar to the pole mass scheme.57) with r = m2 /m2 . Another example for a process dependent renormalization scheme is the mixing of γ and Z propagators. not just the squark mass itself. but it is even less gauge invariant with respect to the weak gauge symmetries of the Standard Model than the on-shell mass scheme. One theoretical problem with this on-shell renormalization scheme is that it is not gauge invariant.56) 2 ˜ M ˜ M The convenient scale dependent pole 1/˜ is defined to include the universal additional terms like the Euler gamma function and the scaling logarithm. where in the on-shell scheme we find mbare → mq + δmq ˜ ˜ q ˜ = mq + mq ˜ ˜ αs CF 4π − m2 2r 1 q ˜ 2 − 1 − 3r − (1 − 2r) log r − (1 − r) log − 1 − 2r log 2 ˜ r M . The reason why QCD counter terms tend to depend only on the renormalized quantity itself is that the gluon is massless.

0). Treating the two divergences strictly separately and dealing with them one after the other also ensures that for ultraviolet divergences we have to choose > 0 while for infrared divergences we need to require < 0. And last but not least. When we introduce counter terms in general field theory we usually choose a slightly more model independent scheme — we define a renormalization point. Of course. like Lorentz symmetry or gauge symmetries we should simply subtract the ultraviolet pole and nothing else. so the result for B(0. . Next. divergent as well as finite. there are three scales present in such a scheme. there is the renormalization scale µR . we know that like any massive two-point function it has to be ultraviolet divergent B ∼ 1/ UV because setting all internal and external mass scales to zero is nothing special from an ultraviolet point of view.62 2 QCD One common feature of all these mass counter terms listed above is δm ∝ m. there is the scale M separating the counter term from the process dependent result. In the MS scheme the counter term is then scale dependent.(2. In our case of a top self energy this is for example the top mass mt appearing in the matrix element for the process ¯ pp → tt. Looking back at δmt as defined in Eq. For this particular integral they precisely cancel. First.59) with Z = 1 + δZm as usually in perturbation theory and δZm = δm/m linking the two ways of writing the mass counter term. i. (2. As a matter of fact. The role of this scale M will become clear when we go through the example of the running strong coupling αs . carefully keeping track of the divergent contributions from the infrared and the ultraviolet regimes.e. 0) is zero. but which as we will see implies a running of the counter term. Just a side remark for completeness: the one loop integral which has no intrinsic mass scale is the two-point function with zero mass in the loop and zero momentum flowing through the integral: B(p2 = 0. there is the physical scale in the process. We can then write the entire massless two-point function as 1/ UV − 1/ IR . we would prefer to choose all three scales the same. where we use the pole position in the propagator instead of a renormalization point shows that such a renormalization point and a physical process as a definition of an observable can sometimes be translated into each other. a reference scale which is part of the definition of any counter term. Regularization and renormalization schemes which do not break symmetries of the Lagrangian are ideal.56) we easily see a way to define a completely general counter term: if dimensional regularization. On the other hand. If we remember that chiral symmetry protects a Lagrangian from acquiring fermion masses this means that on-shell renormalization does not break this symmetry. This form implies that particles with zero mass will not obtain a finite mass through mass renormalization. the introduction of 4 − 2 dimensions does not break any of the symmetries of our Lagrangian. 0. but in a complex physical process this might not always be possible. This is the energy scale at which the counter terms cancels all higher order contributions. which we can choose however we want. Based on these dimensional arguments this integral has to vanish altogether.60) The definition of the on-shell scheme. (2. The only question is: do we subtract 1/ (MS scheme) or do we subtract 1/˜ in the MS scheme. A massless theory cannot become massive by mass renormalization. but setting it to zero too early will screw up our ultraviolet and infrared finiteness tests in practical calculations. For example. any kind of massive (2 → 3) production process naturally involves several external physical scales. The best known example is the electric charge which we renormalize in relation to the Thomson limit of zero momentum transfer through the photon propagator e → ebare = e + δe . This can only work if the scalar integral also has an infrared divergence appearing in dimensional regularization. It appears for example as a self-energy correction of external quarks and gluons. which means that we actually encounter a multiplicative renormalization mbare = Zm m = (1 + δZm ) m = 1+ δm m m = m + δm . 0.

if . as propagator corrections to gauge bosons are usually labelled. and the ghost loop which removes the unphysical degrees of freedom of the gluon from the loop. quark. the same should be true for the gluon self energy. The factor b0 arises with to one-loop corrections to the gluon self energy. where the gluon splits into two quarks or gluons and re-combines before it produces the two final state bottoms. In color space the self energy will (hopefully) be diagonal. will be a scalar. Including the gluon. replacing the factor 13/6 by a factor 11/3.(2.2 Ultraviolet divergences 63 2.52) we could have guessed that the loop integrals will only give a logarithm log p2 which then combines with the logarithm log M 2 implicitly included in the definition of ˜. and ghost loops the regularized gluon self energy with 2 a momentum flow p reads 1 Π p2 µ2 R p2 = αs 4π p2 1 − + log 2 ˜ M 1 p2 − + log 2 ˜ M b0 = 1 4π 13 2 Nc − n f 6 3 b0 + O(log m2 ) t 11 2 Nc − n f 3 3 .61) + O(log m2 ) t   + ≡ αs with + The number of fermions coupling to the gluons is nf . Later on. In Minkowski space the gluon propagator in unitary gauge is proportional to the transverse tensor T µν = g µν − pν pµ /p2 .2 Running strong coupling To get an idea what these different scales which appear in the process of renormalization mean let us compute such a scale dependent parameter. we will indicate where additional higher order corrections would enter. In the second step of Eq. so we cannot use it as our toy process this time. where at some energy range we will be dominated by valence quarks: q q → b¯ The only Feynman diagram is ¯ b. Strictly speaking.e. This is related to the fact that there are actually three types of divergent virtual-gluon diagrams in the physical process q q → b¯ the ¯ b: external quark self energies with a renormalization factor Zf . The Drell-Yan process is one for the very R few relevant process at hadron colliders where the strong coupling does not appear at tree level.2. so we can ignore the color indices for now.(2. The gluon self energy correction or vacuum polarization. As mentioned for the effective gluon-Higgs coupling. Another simple process where we can study this coupling is bottom pair production at the LHC. the internal gluon self energy ZA . Let us for now assume that all quarks are massless. just like the gluon propagator itself. and the vertex corrections ZAf f . The only physical parameters we can renormalize in this process are the strong coupling and. At next-to-leading order this gluon propagator will be corrected by self energy loops. namely the running strong coupling αs (µ2 ). from diagrams which include one additional factor αs . all fermion lines close and the Dirac trace is computed inside the loop. The Feynman diagrams for the gluon self energy include a quark look.e.2. i. From the comments on the function B(p2 .61) we have sneaked in additional contributions to the renormalization of the strong coupling from the other one-loop diagrams in the process. (2. a gluon loop. this form is the first 2 term in a perturbative series in the strong coupling αs = gs /(4π). which we therefore write as Πµν ≡ Π T µν . A useful thing to remember is the simple relation ρ ρ T µν Tν = T µρ and T µν gν = T µρ . an s-channel off-shell gluon with a momentum flow p2 ≡ s. i. 0. 0) after Eq.

because the definition of the coupling now has to account for a possible shift between the original argument p2 and the scale M 2 coming out of the MS scheme.63) 2 .(2. the strong coupling diverges in the Thomson limit because QCD is confined towards large distances and weakly coupled at small distances. All of them need to have the same divergence structure ZAf f ! ZAηη ! Z3A ! Z4A = 1/2 = 3/2 = (2. ghosts as well as the triple and quartic gluon vertex. i. This makes QCD an asymptotically free theory.65) 2 αs (M 2 ) = αs (p2 ) − αs b0 log p2 M2 (2. This means that for p2 > M 2 the ultraviolet limit of the strong coupling is zero.e. This R implies αs b0 p2 bare − αs b0 log 2 (2. 1/2 ZA Z Zf Z Zη Z A A A If we had done the same calculation in QED. we would have found that the vacuum polarization diagrams for the photon do account for the entire counter term of the electric charge. this means that αs depends on the unphysical scale M .65) αs → αs (p2 ) 1 + ˜ M This formula defines a running coupling αs (p2 ). looked for a running electric charge.64 2 QCD finite.61) we know that the ultraviolet pole which needs to be cancelled by the counter term is proportional to the function b0 bare gs → Zg gs = (1 + δZg ) gs = 2 (gs )bare → (Zg gs )2 = 1+ 2 gs = δgs gs gs 2 gs = ⇒ ⇒ 1+ ! bare αs → 1+ δαs αs δgs gs  2 αs = 1 + αs µR ˜ M δgs 1+2 gs  1+ 2 δgs 2 gs 2 gs b0  αs (M 2 ) . The other two renormalization constants ZAf f and Zf cancel because of gauge invariance. Since the bare strong coupling can be expressed in terms of αs (M 2 ) as well as in terms of αs (p2 ) we can link the two scales through the finite contributions in Eq. Because the bare coupling does not depend on any scales. Wave function renormalization constants are not physical. (2. (2. the bottom mass.67) 12π colored states . The entire divergence in our q q → b¯ process which needs to be absorbed in the strong coupling Zg is given by the combination ¯ b ZAf f = Zg ZA Zf 1/2 ⇔ ZAf f ZA Zf 1/2 ≡ Zg .j .66) ⇔ dαs (p2 ) 2 2 = −αs b0 + O(αs ) d log p2 To the given loop order the argument of the strong coupling squared can be neglected — its effect is of higher order. Similar to the top mass renormalization scheme we can switch to a more physical scheme for the strong coupling as well: we can absorb also the finite contributions of Π(µ2 /p2 ) given in Eq.61) into the strong coupling by simply identifying M 2 = p2 .(2. In this first formula for the running coupling constant we see that b0 is positive in the Standard Model.62) We can check this definition of Zg by comparing all vertices in which the strong coupling gs appears. namely the gluon coupling to quarks. We can compute the function b0 in general models by simply adding all contributions of strongly interacting particles in this loop 1 b0 = − Dj TR.64) Only in the last step we have explicitly included the scale dependence of the counter term. (2. In contrast to QED.(2. Hence lacking a well-enough motivated reference point we are tempted to renormalize αs in the MS scheme. From Eq.

because particles can only contribute to the running of αs at scales below their masses. while we assume that the pole will be properly taken care off in any of the schemes we discuss αs → αs (M 2 ) 1 + αs b0 log p2 M2 −1 .71) dp2 d log p2 n=0 .70) The factor αs inside the parentheses can again be evaluated at either of the two scales. αs (M 2 ) M (2. the difference is going to be a higher order effect.2.(2. we can use this complete formula to relate the values of αs between two reference points. The color charges are TR = 1/2 for the fundamental representation of SU (3) and CA = Nc for the adjoint representation.2 Ultraviolet divergences 65 where we need to know some kind of counting factor Dj which is -11 for a vector boson (gluon). This is why the R ratio computed in Eq. The masses of the loop particles are not relevant in this approximation because we are only interested in the ultraviolet regime of QCD where all particles can be regarded as massless. We can do even better than this fixed order in perturbation theory: while the correction to αs is perturbatively suppressed by the usual factor αs /(4π) it includes a logarithm of a ratio of scales which does not need to be small. we consider it a renormalization group equation (RGE) which evolves physical parameters from one scale to another αs (M = p ) = αs (M ) ⇔ 1 1 = αs (p2 ) αs (M 2 ) 2 2 2 p2 1 + αs b0 log 2 M p2 1 + αs b0 log 2 M −1 = 1 p2 + b0 log 2 . This is the Landau pole of or p2 dαs dαs 2 n ≡ = β = −αs bn αs (2. i.12) is so interesting once we can vary the energy of the incoming electron-positron pair. using the 2 relation d/dx(1/αs ) = −1/αs dαs /dx 1 dαs d 1 2 = −αs = −αs b0 +O(αs ) αs d log p2 d log p2 αs This is the famous running of the strong coupling constant. When we really model the running of αs we need to take into account threshold effects at the masses of heavy particles.69) Just as in the case without re-summation.e. Such a series means we replace the off-shell gluon propagator by (schematically written) T µν T µν → 2 + p2 p + T µν = 2 p ∞ T T · (−T Π) · 2 2 p p T T T · (−T Π) · 2 · (−T Π) · 2 2 p p p − Π p2 j µν µν + ··· (2. +1 for a complex scalar (squark) and +1/2 for a real scalar.64) into the denominator. +2 for a Majorana fermion (gluino). (2.68) = j=0 1 T µν . +4 for a Dirac fermion (quark). It is customary to replace the renormalization point of αs with a reference scale where the denominator of the re-summed running coupling crosses zero and the running coupling itself should diverge. This re-summation of the logarithm which occurs in the next-to-leading order corrections to αs moves the finite shift in αs shown in Eq. Instead of simply including these gluon self-energy corrections at a given order in perturbation theory we can instead include chains of on-loop diagrams with Π appearing many times in the off-shell gluon propagator. p2 1 + Π/p2 σ ν ρ To avoid indices we abbreviate T µν Tν = T · T which make sense because of (T · T · T )µν = T µρ Tρ Tσ = T µν . When we differentiate αs (p2 ) with respect to the momentum transfer p2 we find.(2.

but since we are not really sure ¯ what to do with the parton densities which are included in the actual hadronic observable. namely p2 .74) These perturbative factors rn we can assume to be dimensionless — if they are not. there is a little bit of confusion between field theorists and phenomenologists: up to now we have introduced the renormalization scale µR as the energy scale given by the renormalization point. for example of the strong coupling constant. For example.2 we have introduced the running strong coupling in a fairly abstract manner. we better use an observable at an e+ e− collider. i. Including higher order corrections we can express the result in a power series in the renormalized strong coupling αs . as discussed for the Higgs self coupling in Section 1. which means that including all orders in perturbation theory it cannot depend on any artificial scale choices M . This is QCD what we will from now on call the renormalization scale in the phenomenological sense.(2. the externally fixed p2 on the one hand and the artificial M 2 on the other.2. except that with the introduction of ΛQCD we are choosing a reference point which is particularly hard to compute perturbatively. 2. This trick is called dimensional transmutation. re-summing diagrams and changing the running of αs from Eq.72) This scheme is not that different from the Thomson limit renormalization scheme of QED. but from Section 2. One thing that is interesting in the way we have introduced ΛQCD is the fact that we have introduced a scale into our theory without ever setting it.(2. Something that will work and includes αs at least in the one-loop corrections is the R parameter given in Eq.3.3 on we will shift back to µR instead of M as the argument of the running coupling. αs M2 = n=0 rn p2 M2 n αs (M 2 ) r0 = 11Nc .73) q + e− → µ+ µ− ) σ(e 9 quarks where the numerical value at leading order assumes five quarks. In terms of language.(2.71) we did not yet link to any physics discussion. All we did was renormalize a coupling which becomes strong at large energies and search for the mass scale of this strong interaction.3 Re-summing scaling logarithms 1 + αs (M 2 ) b0 log = 1 p2 − Λ2 αs (M 2 )b0 QCD In the last Section 2.66) to Eq.2. 9 (2. In the MS scheme. In what way does the re-summation of the one-loop diagrams for the s channel gluon improve our prediction of the bottom pair production rate at the LHC? To illustrate those effects we best look at a simple observable which depends on just one energy scale. In the MS scheme we subtract 1/˜(µR /M ) and in general include a scale dependence on M in the individual prefactors rn R p2 . The first observable coming to mind is again the Drell–Yan cross section σ(q q → µ+ µ− ). Throughout this section we will keep the symbol M for this renormalization scale in the MS scheme. we can scale R appropriately using p2 .e. (2. the subtraction of 1/˜ shifts the scale dependence of the strong coupling to M 2 and moves the logarithm log M 2 /Λ2 into the definition of the renormalized parameter. This implies that the rn only depend on ratios of two scales.12) σ(e+ e− → hadrons) 11Nc R= = Nc Q2 = .2. At the same time we know that R is an observable. the argument we evaluate αs at.66 2 QCD the strong coupling. Writing this dependence as a total derivative and setting it to zero we find an equation which would be called a Callan-Symanzik equation if instead of the running coupling we had . At one-loop order this reads 1 + αs b0 log 1 1 = αs (p2 ) αs (M 2 ) Λ2 QCD ! =0 M2 ⇔ log p2 M2 Λ2 1 QCD =− 2 M αs b0 1 + b0 αs (M 2 ) log ⇔ log p2 p2 1 = log 2 − 2 M ΛQCD αs b0 = b0 log p2 Λ2 QCD (2.

75): first. . (2.2 Ultraviolet divergences 67 included a running mass 0 = M2 p2 ∂αs ∂ d ∂ R . αs (M 2 ) = M 2 + R 2 dM M2 ∂M 2 ∂M 2 ∂αs ∂ ∂ p2 n = M2 +β αs rn 2 ∂M ∂αs n=0 M2 ! p2 . The non-trivial structure.(2. Next. so αs should be considered variables by itself.77) ≡ n cn αs (p2 ) . αs M2 = n=1 M2 ∂rn n n−1 α + β rn nαs ∂M 2 s n=1 with r0 = 11Nc 9 m bm αs m=0 = M2 = M2 ∂rn n n+m+1 α − nrn αs bm ∂M 2 s n=1 m=0 n=1 ∂r1 ∂r2 αs + M 2 − r1 b0 ∂M 2 ∂M 2 2 αs + M 2 2 with β = −αs ∂r3 − r1 b1 − 2r2 b0 ∂M 2 3 4 αs + O(αs ) . the kind-of Callan-Symanzik equation requires r0 = c0 = r1 = c1 M2 M2 = c2 + c1 b0 log 2 p2 p 2 M M2 r3 = c3 + (r1 b1 + 2r2 b0 ) log 2 = c3 + c1 b1 + 2 c2 + 2c1 b0 log 2 p p 2 2 M M = c3 + (c1 b1 + 2c2 b0 ) log 2 + 2c1 b2 log2 2 0 p p ··· r2 = c2 + r1 b0 log 11Nc 9 b0 log M2 p2 (2. And finally we need to include one loop in the gluon propagator b0 and two loops for example in the vertex r2 .75) In the second line we have to remember that the M dependence of αs is already included in the appearance of β. we have a two-loop diagram in the gluon propagator b1 and a one-loop vertex correction r1 .(2. we have the appropriate NNNLO corrections r3 .76) This chain of rn values looks like we should interpret the apparent fixed-order perturbative series for R in Eq. This perturbative series in αs has to vanish in each order of perturbation theory. Altogether. namely the mix of rn derivatives and the perturbative terms in the β function we can 3 read off the αs term in Eq. They can become problematic if this logarithm becomes large enough to spoil the fast convergence in terms of αs ∼ 0.(2.74) as a series which implicitly includes terms of the order logn−1 M 2 /p2 in each rn .1 by evaluating the observable R at scales far away from the scale choice for the strong coupling constant M .2. Instead of a series in rn we can use the conditions in Eq.76) to express R in terms of the cn and collect the logarithms appearing with each cn R= n rn p2 M2 n αs (M 2 ) = c0 + c1 1 + αs b0 log M2 M2 2 + 2αs b2 log2 2 + · · · 0 2 p p M2 + ··· p2   + c2   2 αs (M 2 ) + · · · αs (M 2 ) + c2 1 + 2αs b0 log αs (M 2 ) M2 1 + αs b0 log 2 p 2 αs (M 2 ) = c0 + c1   + ··· M2  1 + αs b0 log 2 p (2.

Naively. defining the running strong coupling we sum one-loop vacuum polarization graphs. we wrote down the hadronic cross sections in terms of parton distributions at leading order in Eq.4 and 2. i. 2.3. Secondly. Last. It should behave similarly as any other 2 → 2 jet radiation.3. 2. In Section 2. In this manner all scale dependence in the perturbative series for the dimensionless observable R is moved into αs .3 we will show that these divergences have a generic structure and can be absorbed into a re-definition of the parton densities.3. the Drell–Yan process.5 we will again follow the example of the ultraviolet divergences and specify what absorbing these divergences means in terms logarithms appearing in QCD calculations.1. we can define a running parameter. This means we can start for example by computing the cross section for the partonic process q q → Zg.3. In that case introducing ΛQCD leaves us with a compact form of the running coupling αs (M 2 |ΛQCD ).(2.1 Single jet radiation So let us get back to the radiation of additional partons in the Drell–Yan process.71). In Sections 2. The perturbative question we need to ask for µ+ µ− production at the LHC is: what happens if together with the two leptons we produce additional jets which for one reason or another we do not observe in the detector. absorb the scale logarithm into the MS counter term. In Section 2.(2. which phenomenologists refer to as renormalization scale dependence.2 we will see that the purpose of a jet algorithm is to take us from some kind of energy deposition in the calorimeter to the parton radiated in the hard process. In Eq. similar to an ultraviolet renormalization of a Lagrangian parameter. They lead to a dependence of our perturbative result on the artificial scale M 2 . but even the seemingly scale invariant dimensional regularization in the conformal limit of our field theory cannot avoid the introduction of a scale. we can renormalize our parameter at a reference point. Before moving on we collect the logic of the argument given in this section: when we regularize an ultraviolet divergence we automatically introduce a reference scale µR . This re-organization of the perturbation series for R can be interpreted as re-summing all logarithms of the kind log M 2 /p2 in the new organization of the perturbative series and absorbing them into the running strong coupling evaluated at the scale p2 .77) we see that this series in cn will never lead to a scale-invariant result when we include a finite order in perturbation theory.2 and 2. To make life easier and still learn about the structure of collinear infrared divergences we will look at the crossed process qg → Zq. Such jets could for example come from the radiation of a gluon from the initial state quarks. However. in Section 3. i. as long as there is only one scale affecting a given observable.68 2 QCD In the last step we use what we already know about the running coupling from Eq.1 we will study the kinematics of radiating such jets and specify the infrared divergences this leads to. which is not really correct once we include jet algorithms and hadronization.3. this could be a ultraviolet cutoff scale. we implicitly include higher order contributions. Strictly speaking.39).e. ultraviolet and infrared finite. at each order in perturbation theory the scale dependence should vanish together with the ultraviolet poles. except that it has a different incoming state than the leading-order Drell–Yan process and hence does not involve virtual corrections. Even when we compute an observable at a given loop order. This means we do not have to deal with renormalization and ultraviolet divergences and can concentrate on parton or jet radiation from the . Following the example of the R ratio we see what our definition of the running coupling means in terms of re-summing logarithms: reorganizing our perturbative series to get rid of the ultraviolet divergence αs (p2 ) now re-sums the scale logarithms log p2 /M 2 to all orders in perturbation theory. In contrast to the rn the factors cn are by definition independent of the ratio p2 /M 2 . However. Throughout this writeup I will use the terms jets and final state partons synonymously.3 Infrared divergences After this brief excursion into ultraviolet divergences and renormalization we can return to the original example. this partonic process involves ¯ renormalization or ultraviolet divergences as well as an avalanche of loop diagrams which have to be included before we can say anything reasonable. These pdfs are only functions of the collinear momentum fraction of the partons in the proton about which from a theory point of view only know a set of sum rules. There are several ways of dealing with such a scale: first.3. On the other hand.(2.e. We will need this picture once we have introduced the infrared divergences in the following section.

The actual removal of the infrared pole — corresponding to the renormalization in the ultraviolet case — is called mass factorization and works exactly the same way as renormalizing a parameter: in a well-defined scheme we simply subtract the pole from the fixed-order matrix element squared. in the non-collinear phase space C T is not a constant. Then. However.41) and using the appropriate Jacobian this integration can be written in terms of the reduced angle y. (2. we would like to follow the same strategy as for the ultraviolet divergence. to avoid confusion with massless particles which are not radiated jets when studying the infrared behavior.78) |M| ∼ − − + s t st The Mandelstam variable t for one massless final state particle can be expressed as t = −s(1 − τ )y in terms of the rescaled gluon-emission angle y = (1 − cos θ)/2 and τ = m2 /s. as long as < 0 by slightly increasing the suppression of the integrand in the infrared regime. The amplitude for this 2 → 2 process is — modulo charges and averaging factors.82) The form C/p2 for the matrix element is of course only valid in the collinear limit. Just for reference. In this limit the Z matrix element becomes s2 − 2sm2 + 2m4 1 2m2 2 Z Z Z − + O(y) . the complete formula in terms of the angular variable y reads s dσ π(4π)−2+ = dy Γ(1 − ) µ2 F m2 Z τ (1 − τ )1−2 2 |M| ∼ y (1 − y) µ2 F m2 Z |M| . The additional factor y − regularizes the integral at y → 0. corresponding to negative t → 0 with finite u = −s + m2 .80) s2 − 2sm2 + 2m4 (s − m2 ) Z Z Z + O(p0 ) . y (1 − y) 2 (2. we regularize the divergence for example using dimensional regularization. First.2. Starting from Eq. let us go back to Z production instead of a photon. . Similarly. Approximating the matrix element as C /y or C/p2 . T s2 p2 T (2. for small angles y. i. we then need to integrate T y max dy y min C = y pmax T pmin T dp2 T C = 2 p2 T pmax T dpT pT pmin T C p2 T pmax T 2C pmin T dpT pmax 1 = 2C log T pT pmin T (2.e.83) In the second step we only keep the factors we are interested in. (2.79) becomes |M| ∼ 2 (2. so as a Z sanity check we can confirm that t + u = −s(1 − τ ) = −s + m2 . we obtain u = −s(1 − τ )(1 − y).(2. We can translate this 1/y divergence for example into the transverse momentum of the gluon or Z according to sp2 = tu = s2 (1 − τ )2 y(1 − y) = (s − m2 )2 y + O(y 2 ) T Z In the collinear limit our matrix element squared in Eq. This scale arises from the infrared regularization of the phase space integral and is referred to as factorization scale. Moreover.(2. This regularization procedure is symmetric in y ↔ (1 − y).81) The matrix element for the tree level process qg → Zq has a leading divergence proportional to 1/p2 . To compute the T total cross section for this process we need to integrate the matrix element over the entire two-particle phase space. so Eq.3 Infrared divergences 69 initial state. we find a well-defined way to get rid of it.(2. Next. What is important to notice is again the appearance of a scale µ2 with the F n-dimensional integral.82) describes well this divergence arising from quark radiation at the LHC.79) |M| ∼ s(s − m2 ) y s Z This expression is divergent for collinear gluon radiation or splitting. The collinear limit when the gluon is radiated in Z the beam direction is given by y → 0. After integrating the leading collinear pole as the combined 1/y 1+ we are left with a pole 1/(− ). but including all Mandelstam variables t s 2m2 (s + t − m2 ) 2 Z Z . Dimensional regularization means writing the two-particle phase space in n = 4 − 2 dimensions. this does not save us from the collinear divergence.

pa (pa pT ) = 0 = (npT ).3.2 Parton splitting From the discussion of the process qg → Zq we can at least hope that after taking care of all other infrared and ultraviolet divergences the collinear structure of the process q q → Zg will be similar. Infrared divergences occur for massless particles in the initial or final state. shown in Figure 9. This way we can specify n such that it defines the direction of the pb –pc decay plane. and a free factor β. c b Relative to pa we can split the opening angle θ for massless partons according to Figure 9 θ = θb + θc and pT θb = θc Eb pT Ec −1 = 1−z z ⇔ θ= θb θc = 1−z z (2. for example that of a radiated gluon p2 = 0. 2. Figure from Ref. i. The kinematics of this approximately collinear process we can described in terms of the energy fractions z and 1 − z defined as z= Eb Ec =1− Ea Ea ⇔ 2 2 p2 = 2(pb pc ) = 2z(1 − z)(1 − cos θ)Ea = z(1 − z)Ea θ2 + O(θ4 ) a θ ≡ θb + θc = 1 Ea p2 a z(1 − z) (2. Because p2 > 0 we call this final state splitting configuration a time-like branching. This time we are following the direction of the mother has to have a finite positive invariant mass p2 a b c momenta pa = pb + pc . The first and at the LHC most important case is the splitting of one gluon into two. to avoid minus signs in the particle energies. so we need to go through all ways incoming or outgoing gluons and quark can split into each other.85) in terms of an arbitrary unit four-vector n. independent of the hard process which we usually choose as the Drell–Yan process. this requires us to split the (n + 1)-particle phase space alone into an n-particle phase space and the collinear splitting. In this section we will show that ¯ we can indeed write all collinear divergences in a universal form. For this configuration we can write down the so-called Sudakov decomposition of the four-momenta pa = pb + pc = (zpa + βn + pT ) + ((1 − z)pa − βn − pT ) (2. the radiation of additional partons or the splitting into additional partons will be described by universal splitting functions. The second final state will have a finite invariant mass p2 = 0.70 2 QCD Figure 9: Splitting of one gluon into two gluons.84) in terms of the opening angle θ between pb and pc . where the two daughter gluons are close to mass shell while the p2 . In the collinear limit. a component orthogonal to the mother momentum and n. First. [23].86) Using this specific phase space parameterization we can divide an (n + 1)-particle process into an n-particle process and a splitting phase space of one of quarks or gluons. In this decomposition we can set only one invariant mass to zero. p2 . The general (n + 1)-particle phase space separating off .e.

3 dz = . 2(1 − z)(npa ) 4(1 − z)Ea (2.88) pT =  T.2.3 Infrared divergences 71 the n-particle contribution dΦn+1 = · · · d3 pb d3 pc d3 pa d3 pc Ea = ··· 3 E 2(2π)3 E 3 E 2(2π)3 E E 2(2π) b 2(2π) a c c b dpc.(2.87) is best expressed in terms of the energy fraction z and the azimuthal angle φ. 2 2 θ2 2 θ2 pa z(1 − z)Ea z(1 − z)Ea (2.(2. The z direction of pc we can translate into z in a convenient reference frame for the momenta appearing in the Sudakov decomposition       0 1 1 p  0 0 (2.91) which gives us the final result for the separated collinear phase space dΦn+1 = dΦn dzdp2 dφ a 4(2π)3 (2.2  0 0 0 −1 1 This choice has the special feature that n2 = 0 which allows us to derive β from the momentum parameterization shown in Eq.87) we also replace dp2 with dp2 according to a T 2 2 2 p2 Eb θb z 2 (1 − z)2 Ea θ2 T = = = z(1 − z) .93) .3 dpT pT dφ 1 = dΦn 2(2π)3 Ec z dpc.3 by dz in Eq.(2.85) and the additional condition that p2 = 0 c p2 = ((1 − z)pa − βn − pT ) = p2 − 2β(1 − z)(npa ) = p2 − 4β(1 − z)Ea = 0 c T T ⇔ β= p2 p2 T T = .85) for the z component of pc .88) and inserting this value for β gives us d dpc.3 dp2 dφ 1 T = dΦn 4(2π)3 Ec z at fixed pa (2. Ec 1−z (2.92) Adding the transition matrix elements to this factorization of the phase space and ignoring the initial-state flux factor which is common to both processes we can now postulate for one collinear emission dσn+1 ∼ |Mn+1 |2 dΦn+1 dp2 dzdφ a 4(2π)3 dp2 dz = |Mn+1 |2 dΦn a 2 4(2π) = |Mn+1 |2 dΦn = 2 2gs ˆ dp2 dz P (z) |Mn |2 dΦn a 2 2 pa 16π spherically symmetric assuming |Mn+1 |2 = 2 2gs ˆ P (z) |Mn |2 2 pa (2.(2.89) 2 ! Starting from Eq. expressing pa and pT following Eq.1  .90) In addition to substituting dpc.3 = dz dz (1 − z)Ea + p2 T 4(1 − z)Ea = −Ea + p2 Ec T =− 2E 4(1 − z) a 1−z ⇔ dpc. n=  pa = Ea   pT.

These six combinations contribute to the splitting matrix element a c squared as a b c ±( i j )( k pi ) ( 2 2 2 i j ) ( k pi ) /pa ⊥ ⊥ ⊥ ⊥ (−1)(−z)Ea θ −(−1)(−z)(1 − z)Ea θ −(−1)(1 − z)Ea θ z 1−z z(1 − z) 1−z z ⊥ ⊥ . To compute its transition amplitude we parameterize all the gluon momenta and polarizations entering the three gluon vertex.96) Using these kinematic relations we can tackle the actual splitting amplitude. while the polarizations in the plane are only proportional to their corresponding momenta. The perpendicular polarizations are also orthogonal to all three gluon momenta.94) As the first parton splitting in QCD we study a gluon splitting into two gluons. the three parallel as well as the three perpendicular polarization vectors are aligned. ⊥ .93) holds true it means we can compute the (n + 1) particle amplitude squared from the n-particle case convoluted with the appropriate splitting kernel. (2. Using dσn ∼ |Mn |2 dΦn we can write this relation in its most common form σn+1 = σn αs ˆ dp2 a dz P (z) p2 2π a (2.95) and (2.(2. ( b c ) or ( ⊥ ⊥ ) require a .95) with in general i = j. For three gluons the splitting amplitude will be proportional to the three gluon vertex Vggg = igf abc = igf = igf abc abc α β γ a b c α β γ a b c [gαβ (pa − pb )γ + gβγ (pb − pc )α + gγα (pc − pa )β ] [gαβ (−pc − 2pb )γ + gβγ (pb − pc )α + gγα (2pc + pb )β ] +( −( b c )( a pb ) b c )( a pb ) with pa = −pb − pc with ( j pj ) = 0 with ( a pc ) = −( a pb ) . one in the plane.(2. as shown in Figure 9. . In the limit of small scattering angles. which altogether means for the three-vectors ( i j) = −1 ( ⊥ ⊥ i j ) = −1 ( ⊥ i j) =0 ( ⊥ i pj ) =0 ( j pj ) = 0 .97) [−2( abc a b )( c pb ) a b )( c pb ) −( −( b c )( a pc ) + 2( c a )( b pc )] = −2igf [( c a )( b pc )] Given the relations of the polarization vectors in Eq.72 2 QCD This last step is an assumption we will now proceed to show step by step by constructing the appropriate ˆ splitting kernels P (z) for all different quark and gluon configurations. The only finite combinations are ( a pp ) = −Eb cos ∠( a pb ) = −Eb cos π − θb = +Eb sin θb +Eb θb = z(1 − z)Ea θ 2 π + θ = −Ec sin θ −Ec θ = −(1 − z)Ea θ ( b pc ) = −Ec cos ∠( b pc ) = −Ec cos 2 π ( c pb ) = −Eb cos ∠( c pb ) = −Eb cos − θ = +Eb sin θ +Eb θ = zEa θ 2 (2. and last but not least we a b b c can combine ( a c ) as well as ( ⊥ ⊥ ) with b .96) we know which of the three terms in brackets contribute: ( a b ) as well as ( ⊥ ⊥ ) can be combined with c . With respect to the scattering plane opened by pb and pc all three gluons have two transverse polarizations. and one perpendicular to it. If Eq. (2.

coupling the gluon current to the quark and antiquark spinors.102) . All three terms in Vggg shown in Eq.(2.97) contribute to the all-parallel polarization case.93) holds for only gluons.3 Infrared divergences 73 We also see that the six cases correspond to four different polarizations of the three external gluons.(2.2 γ0 = 1 1 0 0 −1 1 γj = 0 −σ j σj 0 γ0γ0 = 1 1 γ0γj = 0 σj σj 0 (2.100) For the massless antiquark we need to replace θ → −θ and take into account the different relative spin-momentum directions (σ p).98) 2 using CA = Nc and averaging over the color (Nc − 1) and polarization Na of the mother particle a. leading to the additional sign in the lower two spinor entries. The form of the splitting kernel is symmetric when we exchange the two gluons z and (1 − z). (2. Squaring the splitting matrix element to compute the (n + 1) and n particle matrix elements squared for the unpolarized case gives us |Mn+1 |2 = = 1 2 1 p2 a 2 2 4gs f abc f abc 1 2 Nc − 1 Na 2 ±( i j )( k pi ) |Mn |2 incoherent external polarization sum with f abc f abd = Nc δ cd with Na = 2 2 2gs p2 a 2 2gs = 2 pa 2g 2 = 2s pa 2 2gs ≡ 2 pa f abc f abc 1 ( i j )2 ( k pi )2 |Mn |2 2 Nc − 1 Na p2 a 1 ( i j )2 ( k p i )2 Nc |Mn |2 Na p2 a z 1−z Nc + + z(1 − z) |Mn |2 1−z z ˆ Pg←g (z) |Mn |2 ˆ Pg←g (z) = CA z 1−z + + z(1 − z) 1−z z . The antiquark spinors then become ˆ √ v(p) = −i E χ± χ± with χ+ = χ− = 1 −θ/2 θ/2 1 χ+ = χ− = −θ/2 −1 1 −θ/2 (spin up) (spin down) (2.2.1.101) All calculations in this section we will limit to the leading terms in the small scattering angle θ. for the coupling to a gluonic current we need the Dirac matrices which we can conveniently express in terms of the Pauli matrices listed in Section 1. The factor 1/2 in the first line takes into account that for two final state gluons the (n + 1)-particle phase space is only half its usual size. In addition to the fermion spinors. The spinors of the outgoing massless quark u(pb ) and the outgoing massless antiquark v(pc ) we can write in terms of two-component spinors u(p) = √ E χ± ±χ± with χ+ = χ− = 1 θ/2 −θ/2 1 (spin up) (spin down) (2. It diverges for ˆ ˆ either of the gluons becoming soft. The same kind of splitting kernel we can compute for the splitting of a gluon into two quarks and the splitting of a quark into a quark and a gluon g(pa ) → q(pb ) + q (pc ) ¯ and q(pa ) → q(pb ) + g(pc ) .99) Both splittings include the quark-quark-gluon vertex. ⇔ (2. The notation Pi←j ∼ Pij is inspired by a matrix notation which we can use to multiply the splitting matrix from the right with the incoming parton vector to get the final parton vector. Following the logic described above with this calculation we prove that the factorized form of the (n + 1)-particle matrix element squared in Eq.

(2. 1. 2 2 u+ (pb )γ 3 v− (pc ) ∼ ¯ 1.98) we first compute the splitting kernel Pq←g . (2. −1  1 2 2 −1 1    θc /2 −i  1   θb θb θb θb i   − . 1. θb θb . For massless fermions this means that the gluon splitting into two quarks involves two quark spin cases. sandwiching the qqg vertex between an outgoing quark u± (pb ) and an outgoing antiquark v± (pa ) for all possible spin combinations. . . For transverse gluons we can compute this vertex factor also for the other diagonal spin combination   θc /2 1  θb θb θb θb 1  1    − . 1. 2 2 θc /2 1   1 θb θb θc /2 =2. . .e. 1. 2 1 1 −1    −i θc /2   1  θb θb i     −i  −θc /2=i 1. i. 2 i −1   −1 −θc /2    1 =0 θc /2   1 −θc /2    −1  = 0 .104)  u− (pb )γ 1 v+ (pc ) ∼ ¯ u− (pb )γ 2 v+ (pc ) ∼ ¯ Before collecting the prefactors for this gluon-quark splitting. 2 2  1  −θc /2   −θc /2 θb θb  1    = θb − θc 1. θc /2 u+ (pb )γ 2 v+ (pc ) ∼ ¯ 1. θb θb . The gluon current can only couple to two fermions via a spin flip. 1. . −1   −θc /2 = − 2 . We start ¯ with all four gluon polarizations.(2. 2 2 (2. 2 2 u+ (pb )γ 1 v− (pc ) ∼ ¯ 1. 1. However. . all four gamma matrices. 1. 1.105) which vanishes. 2 . between two spin-up quarks and their spinors written out in Eqs.74 2 QCD ˆ In the notation introduced in Eq. 1. 2 2   1 θc /2  θb θb 1  1       −θc /2 = 1. after adding the two unphysical degrees of freedom cancel because of the form of our metric. 1. θc /2 (2. θb θb . we also need the same-spin case  u+ (pb )γ 1 v+ (pc ) ∼ ¯ 1. 1. 2 . −1  −i 2 2 −1 i   −1 −θc /2    1  = θb − θc θc /2   1 −θc /2    −1  = −i(θb + θc ) .103) 2 2  1  θc /2  u+ (pb )γ 0 v− (pc ) ∼ ¯ 1. each of them coupling to two transverse gluon . 1. 1. 2 . 2 2 −θc /2 1   θc /2 θb θb  1    = i(θb + θc ) 1.100) and (2. 1. −1   −θc /2=i − 2 . θb θb .  1. 1.101)  1     1  −θc /2 1   = 1  1  −θc /2 1    1 1  1  −θc /2    =  1  1  −θc /2 1    −i 1   −θc /2 i     −i   1  =i i −θc /2    1 1  −1 −θc /2      1 = 1 −θc /2 −1   1 θb −θc /2 θb  =2 1. θb θb . 2 2 Somewhat surprisingly the scalar and longitudinal gluon polarizations seem to contribute to this vertex. θb θb . 1. 2 2 u+ (pb )γ 2 v− (pc ) ∼ ¯ 1. 2 .

q(pa ) → q(pb ) + g(pc ). 1. 2 . Keeping track of all the relevant factors our vertex function for the splitting g → q q becomes ¯ Vqqg = −igs T a u± (pb )γµ µ v (pc ) ≡ −igs T a ¯ a j a F (j) Eb Ec (θb − θc )2 E 2 z(1 − z)(1 − z − z)2 θ2 |F (1) |2 = = a = (1 − 2z)2 2 p2 p2 Ea z(1 − z)θ2 a a |F (2) |2 Eb Ec (θb + θc )2 E 2 z(1 − z)(1 − z + z)2 θ2 = = a =1. In complete analogy to the gluon splitting case we can factorize the (n + 1)-particle matrix element as Tr T a T a 1 2|F (1) |2 + 2|F (2) |2 |Mn |2 2 Nc − 1 Na g2 N2 − 1 (1 − 2z)2 + 1 |Mn |2 = s TR c with Tr T a T b = TR δ ab and Na = 2 2 p2 Nc − 1 a 2g 2 = 2s TR z 2 + (1 − z)2 |Mn |2 pa 2g 2 ˆ ≡ 2s Pq←g (z) |Mn |2 pa 1 p2 a 2 gs 2 |Mn+1 |2 = ⇔ ˆ Pq←g (z) = TR (z 2 + (1 − z)2 . Comparing to the situation shown in Figure 9 ∗ ∗ for the angle relative to the scattered gluon we find θb = θ while for the incoming quark θa = −θc = −zθ. 1. b 2 2   1 1 ∗ ∗ ∗  1  θa /2    = 1. 1. Inserting the spinors listed in Eq.106) with F (j) = F+ = F− for j = 1. 2 p2 p2 Ea z(1 − z)θ2 a a (j) (j) (2. ∗ θb θ∗ . θb  1  1  2 2 ∗ 1 θa /2    −i 1 ∗ ∗ ∗   θa /2 θb θb i     −i   1 =i 1. This splitting kernel is again symmetric in z and (1 − z) u a because QCD cannot distinguish between the outgoing quark and the outgoing antiquark. i. (2. −θa /2 1 (2.(2. 2 and omitting irrelevant factors i and (−1) which will drop out once we compute the absolute value squared.2.108) ∗ ∗ with the angles θb and θa relative to the final state gluon direction pc .100) and the two relevant gamma matrices gives us  u+ (pb )γ 1 u+ (pa ) ∼ ¯ 1.e. The third splitting we compute is gluon radiation off a quark. ∗ θb θ∗ . 1.107) with TR = 1/2. In the first line we implicitly assume that the internal quark propagator can be written as something like u¯/p2 and we only need to consider the denominator. As . sandwiching the qqg vertex between an outgoing quark u± (pb ) and an incoming quark u± (pa ).3 Infrared divergences 75 polarizations. From the splitting of a gluon into a ¯ quark-antiquark pair we already know that we can limit our analysis to the physical gluon polarizations and a spin flip in the quark legs. 2 ∗ i θa /2  ∗  θa /2  1  ∗ ∗  ∗  = θa + θb θa /2 1  ∗  −θa /2  1  ∗ ∗  ∗  = i(θb − θa ) . b 2 2 u+ (pb )γ 2 u+ (pa ) ∼ ¯ 1. θb .

1. −1 u− (pb )γ 2 u− (pa ) ∼ − . This space-like splitting version of b . θb . and (2. The four results we show in Eqs. This result also means that by construction we have shown that the collinear factorization Eq.112). 1.e. . −1 u− (pb )γ 1 u− (pa ) ∼ − . ∗ −θa /2 (2.98). p2 |p2 | where pb is the momentum entering the hard interaction. θb .110) again with F (j) = F+ = F− for j = 1. 1.109) The vertex function for gluon radiation off a quark then becomes Vqqg = −igs T a u± (pb )γµ µ u± (pc ) ≡ −igs T a ¯ a j (j) a F 2 2 ∗ ∗ Ea Eb (θa + θb )2 E 2 z(z − 1) θ |F (1) |2 = (1 − z) = = a2 2 2 pa pa Ea z(1 − z)θ2 ∗ ∗ Ea Eb (θb − θa )2 E 2 z(1 + z)2 θ2 (1 + z)2 |F (2) |2 = = a2 = .111) The averaging factor 1/Na = 2 now is the number of quark spins in the intermediate state. −1  ¯    θa /2  1 2 2 2 2 −1 1   ∗  −θa /2 −i ∗ ∗ ∗ ∗   1  θb θb i   ∗ =i − θb . similar to ultraviolet divergences which get removed by counter terms these splitting kernels are universal.112) ˆ This result finalizes our calculation of all QCD splitting kernels Pi←j (z) between quarks and gluons . The factorized matrix element for this channel reads |Mn+1 |2 = a a 1 1 2 Tr T T 2|F (1) |2 + 2|F (2) |2 |Mn |2 gs 2 pa Nc Na g 2 N 2 − 1 (1 + z)2 + (1 − z)2 |Mn |2 = s c p2 2Nc 1−z a 2g 2 1 + z2 = 2s CF |Mn |2 pa 1−z 2g 2 ˆ ≡ 2s Pq←g (z) |Mn |2 pa 2 ⇔ 1+z ˆ Pq←q (z) = CF 1−z 2 .94) holds at this level in perturbation theory.(2. . so we already know p2 = t < 0 from our usual Mandelstam variables argument. −1  ¯  −i   θa /2  2 2 2 2 −1 i   −1 ∗  θa /2  ∗ ∗    1  = θa + θb ∗ −θa /2   1 ∗  θa /2  ∗ ∗    −1  = i(θa − θb ) . 2 and dropping irrelevant prefactors.111). modulo a complex conjugation   ∗  −θa /2 1 ∗ ∗ ∗ ∗  θb θb 1  1    ∗  = − θb . (2. As alluded to earlier. meaning |p2 |. (2. (2. (2. The difference to a c b the final state splitting is that now we can consider the split parton momentum pb = pa − pc as a t-channel diagram. Before using this splitting property to describe QCD effects at the LHC we need to look at the splitting of partons in the initial state. the spin-down case gives the same result. they do not depend on the hard n-particle matrix element which is part of the original (n + 1)-particle process. 2 2 2 pa pa Ea z(1 − z)θ 1−z (j) (j) (2. 1. i.107).(2. Just swapping z ↔ (1 − z) we can read off the kernel for a quark splitting written in terms of the final state gluon 1 + (1 − z) ˆ Pg←q (z) = CF z 2 .76 2 QCD expected.

which means its four-momentum squared t0 is negative and its absolute value |t0 | is small. with the one exception that for splitting in the initial state the flow factor has to be evaluated at the reduced partonic energy Eb = zEa and that the energy fraction entering the parton density needs to be replaced by xb → zxb . The minus signs in front of the tj makes these numbers positive. Since each splitting means a synchronous shift in x and t the actual path in the (x − t) plane really consists of physically defined discrete .28) then depends on the final energy fraction and the virtuality f (xn . in that case each collinear gluon radiation will decrease the quark energy xj+1 < xj and through recoil increase its virtuality |tj+1 | = −tj+1 > −tj = |tj |.85) we can solve for p2 b t ≡ p2 = (zpa + βn + pT )2 b = p2 + 2zβ(pa n) T = p2 + T = p2 z T 1−z with p2 = n2 = (pa pT ) = (npT ) = 0 a using Eq.1 T =− <0. Eq. so we can describe how the parton density f (x. t0 ) is at least probabilistically given by the energy and kind of the hadron. We start with a quark inside the proton with an energy fraction x0 . The starting point for (x0 . −tn ) where the second parameter t is new as compared to the purely probabilistic picture in Eq.2. [23]. The parton density we insert into Eq.4. 1−z 1−z The calculation of the splitting kernels and matrix elements is the same as for the time-like case. Figure from Ref. The variable t we call virtuality because for the incoming partons which if on-shell has p2 = 0 it gives the distance to the real mass shell. We then interpret each branching as a step strictly downward in xj → xj+1 where the t value we assign to this step is the ever increasing virtuality |tj+1 | after the branching. t 2π (2. for example the proton.28).89) (2. we can at least use what we know about collinear splitting to describe incoming partons. We know what the successive splitting means in terms of splitting probabilities. for example in pp → Z production.3 DGLAP equation Since we still do not have anything to say about the incoming quarks and gluons at the LHC. For a given small virtuality t0 we start at some kind of x0 distribution.1.(2. namely successive gluon radiation q → qg as illustrated in Figure 10.3 Infrared divergences 77 Figure 10: Successive space-like gluon radiation off an incoming quark.113) p2 + p2 p2 T.2 T. As we will see later it corresponds exactly to the artificial renormalization scale which appears when we re-sum the scaling logarithms which appear in counter terms.(2. as it enters the hadronic phase space integral shown in Section 2. The factorized matrix element for initial state splitting then reads σn+1 = σn αs ˆ dt dz P (z) . 2.(2.3. Let us simplify our kinematic argument by assuming that there exists only one splitting. −t) evolves in the (x − t) plane depicted in Figure 11.114) How to use this property to make statements about the quark and gluon content in the proton will be the focus of the next section. Since this quark is confined inside the proton it should have a small transverse momentum.(2.

If e define a splitting as a vertical drop in x at the target value tj+1 an incoming path hitting the square at some value t can come from any x value above the square. The probability of such a splitting to occur is given by Pq←q (z) ≡ P (z) as it appears in Eq. We use the definition of a convolution 1 1 (f ⊗ g)(x) = 0 dx1 dx2 f (x1 )g(x2 ) δ(x − x1 x2 ) = 0 dx1 f (x1 )g x1 x x1 1 = 0 dx2 f x2 x x2 g(x2 ) . When we convert a partonic into a hadronic cross section numerically we need to specify the probability of the parton density f (x. −t z 2π z dz αs ˆ x P (z)f . Figure from Ref. −t) = 0 for x > 1 . (2. [23].114) Πt (z) dtdz = αs ˆ dt P (z) dz . −t z 2π z assuming f (x . which means the x axis should be labelled −t or |t|. (2.117) where δt is the size of the interval covered by the virtuality value t. the flow into the square is δfin (−t) = δt (Π ⊗ f )(x. (2. Using our (x. ˆ ˆ points. At the end of the path we will probe this evolved parton density. i.78 2 QCD Figure 11: Path of an incoming parton in the (x − t) plane.(2.116) The incoming and outgoing flow we compute from the history of the (x. −t) δ(x − x z) 2π δt = t = δt t dz αs ˆ x P (z)f . [|tj |. the way we define the path between two splittings such a path can only enter and leave the square either vertically or horizontally. −t) . where xn and tn are given by the hard scattering process and its energy-momentum conservation. xj + δx] and. In contrast to the original authors we define t as a negative number. t) plane we compute the flow into this square and out of this square which together defines the net shift in f in the sense of a differential equation δfin − δfout = δf (x. At this stage our picture becomes a little subtle.118) .115) In this picture we consider this probability is a smooth function in t and z. if this second parameter has anything to do with physics. −t) residing in an infinitesimal square [xj . |tj | + δt]. t) evolution. 2π t (2.e. −t) = δt t 1 1 dz x 1 x 1 0 x dx αs ˆ P (z)f (x .

the finite part of the dimensionally regularized integral. (2. 2π (2. This integration appears ˆ because we do not know the normalization of P (z) or Πt (z) distribution which we interpret as a probability. This dimensionally regularized integral we can write as 1 dz 0 f (z) = (1 − z)1− = 1 dz 0 1 f (z) − f (1) + f (1) (1 − z)1− 1 dz 0 1 (1 − z)1− dz 0 1 f (z) − f (1) f (1) (1 + O( )) + 1−z f (z) f (1) (1 + O( )) + (1 − z)+ f (z) (1 + O( )) . Bringing the pole to the other side of . which we know we can cure using dimensional regularization.117) and Eq. Combining Eq. −t) = δ(−t) (−t) dz αs ˆ x P (z)+ f .3 Infrared divergences 79 The outgoing flow we define the same way as the incoming flow in Eq. the dimensionally regularized integral minus the pole.119) The boundaries of the z integration are given relative to x.122) for 4 + 2 dimensions.117) 1 δfout (−t) = δt 0 dzΠ(z) f (x. −t z δt = t δt ≡ t ⇔ dz αs z 2π ˆ dy P (y) f dz αs ˆ x P (z)+ f . We can check that this plus prescription indeed acts as a regularization technique for the parton densities.120) again assuming f (x) = 0 for x > 1 and using the specifically defined plus-subtraction scheme 1 1 F (z)+ ≡ F (z) − δ(1 − z) 0 dy F (y) or 0 dz f (z) = (1 − z)+ 1 dz 0 f (1) f (z) − 1−z 1−z (2. so they cover the entire range.121) with F (z) = 1/(1 − z). (1 − z)+ (2.119) we can compute the change in the parton density of the quarks as δf (x. −t) t dz 0 αs ˆ P (z) .e. (2.(2. −t − z 2π z ˆ P (z) − δ(1 − z) 0 1 1 dy 0 αs ˆ P (y) f (x. the integral over f (z)/(1 − z) is divergent at the boundary z → 1. −t) 2π x . The special case f (z) = 1 illustrates how dimensional regularization of infrared divergences in the phase space integration Eq.(2. As a side effect to our derivation we note that in contrast to the original z integral the plus-subtracted integral is by definition finite in the limit z → 1 where for example some of the splitting kernels diverge.83) works 1 0 1 dz = (1 − z)1− 1 dz 0 1 z 1− = z 1 = 0 1 with > 0 . −t) = δt t 0 1 x 1 0 1 dz αs ˆ x P (z) f . is the same as the plus-subtracted integral modulo terms of the order . −t) 1 = δt f (x. i. Obviously. This change in sign avoids the analytic continuation of the usual value n = 4 − 2 to < 0. −t z 2π z . −t z 2π z 1 x 1 δf (x.123) = 0 1 dz 1 ⇔ 0 f (z) f (1) dz − = (1 − z)1− dz 0 Hence.(2.(2.2.

(2. (2.80 2 QCD the equation 1 dz 0 f (z) = (1 − z)1− 1 dz f (z) 0 1 δ(1 − z) + + O( ) (1 − z)+ (2. (2.126) The infinitesimal version of Eq. quarks and gluons.128) Of the three terms on the right hand side the first and the third together define the plus-subtracted splitting kernel Pq←q (z).124) shows that the difference between a dimensionally regularized splitting kernel and a plus-subtracted splitting kernel manifests itself as terms proportional to δ(1 − z). in addition to the final state gluon.e. what happens when we include the numerator of the regularized function as compared to leaving it out? The finite difference between these results is 1 + z2 1−z − (1 + z 2 ) + 1 1−z = + 1 + z2 − δ(1 − z) 1−z 1 1 dy 0 1 + y2 1 + z2 − + δ(1 − z) 1−y 1−z 1 dy 0 1 + z2 1−y = −δ(1 − z) 0 1 dy dy 2 1 + y2 − 1−y 1−y 1 = δ(1 − z) 0 y2 − 1 = δ(1 − z) y−1 dy (y + 1) = 0 3 δ(1 − z) . comparing ˆ this to the terms proportional to Pq←q they both arise from the same splitting. If the plus prescription regularizes the pole at z → 1.129) ˆ In principle. −t z 2π z ˆ with Pq←j (z) ≡ Pq←j (z)+ . namely a quark density leaving the . Therefore. This generalization involves a sum over all allowed splittings and the plus-subtracted splitting kernels for the quark density on the right hand side is dfq (x.127) Going back to Eq. The second term is a proper convolution and the only term proportional to the gluon parton density.g 1 x dz αs x Pq←j (z) fj . 2π 1 0 dz αs ˆ x Pq←g (z) fg . −t) = δt t − 0 1 0 1 dz αs ˆ x Pq←q (z) fq . Physically. i.120) we add all relevant parton indices and splittings and arrive at δfq (x.(2.111).125) 2 We can therefore write the quark’s splitting kernel in two equivalent ways Pq←q (z) = CF 1 + z2 1−z = CF + 1 + z2 3 + δ(1 − z) (1 − z)+ 2 (2.127) without adding a plus-regulator ˆ Pq←g (z) ≡ Pq←g (z) = TR z 2 + (1 − z)2 . −t) dfq (x. As we already know quarks to not only arise in q → q splitting. (2. the splitting kernel Pg←q also generates a quark. −t) . ˆ Before we move on introducing a gluon density we can slightly reformulate the splitting kernel Pq←q in Eq. However. they represent contributions to a soft-radiation phase space integral. just following the argument above.(2. we generalize Eq. −t z 2π z (2.(2. This means we have to identify it with Pq←g in Eq.120) is the Dokshitzer–Gribov–Lipatov–Altarelli–Parisi or DGLAP integro-differential equation which describes the scale dependence of the quark parton density. but also in gluon splitting. −t) = −t = d log(−t) d(−t) j=q.120) to include the full set of QCD partons.(2. −t + z 2π z dy αs ˆ Pq←q (y) fq (x.

129). −t 1−z ˆ + Pg←q (z) fq x x .(2.3 Infrared divergences 81 ˆ infinitesimal square in the (x − t) plane via the splitting q → qg.131) requires some more work. −t ¯ z z . for the gluon density there is an contribution to δfin proportional to fq or fq which is not matched by the ¯ outgoing flow. −t + fq .2. On the other hand. and Eq. as the notation g ← q already suggests. −t + fq . t) space a gluon can either split into two gluons or radiate one of nf light-quark flavors. −t) 0 Again.127).(2. The second QCD parton density we have to study is the gluon density. We could have expected this because these off-diagonal are actually finite in the soft limit z → 0.133) The second y integral for gluon radiation becomes has to consist of a finite term and a term we can use to define the .132) This ensures that the off-diagonal contribution to the gluon density is taken into account following Eq.(2. Including the additional Pg←q (y) would be double counting and should not appear. −t ¯ z z z using Pg←g (1 − z) = Pg←g (z). −t ¯ z z z (2.131) ˆ ˆ Pg←g (y) + nf Pq←g (y) fg (x.(2. Eq. Three regularized splitting kernels entering the DGLAP equation we give in Eq.(2. −t + fg z ˆ 2Pg←g (z)fg x . z (2. −t + Pg←q (z) fq . 1. −t + Pg←q (z) fq . namely by defining without a second plus subtraction ˆ Pg←q (z) ≡ Pg←q (z) = CF 1 + (1 − z)2 . The incoming contribution to the infinitesimal square is given by the sum of four splitting scenarios each leading to a gluon with virtuality −tj+1 δfin (−t) = δt t 1 0 1 0 dz αs z 2π dz αs z 2π x ˆ Pg←g (z) fg . (2. The generic structure of the DGLAP equation implies that the two off-diagonal splitting kernels do not include any ˆ plus prescription Pi←j = Pi←j . The final splitting kernel Pg←g from the diagonal relation on the right-hand side of Eq. Applying a plus prescription would only have modified the splitting kernels at the isolated (zero-measure) point y = 1 which for a finite value of the integrand does not affect the integral on the right hand side of the DGLAP equation.126). −t) = − δt t δt t 1 0 1 dz αs z 2π dy αs 2π ˆ 2Pg←g (z)fg x x x ˆ . Combining the incoming and outgoing flows we find for the DGLAP equation δfg (x. from the quark case we already know how to deal with it. The y integral not involving the gluon parton densities we can explicitly carry out for the gluon splitting into a quark pair 1 − 0 dy αs αs ˆ nf Pq←g (y) = − nf TR 2π 2π =− 1 dy 1 − 2y + 2y 2 0 1 0 αs 2y 3 nf TR y − y 2 + 2π 3 2 αs =− nf TR .132). 3 2π (2. To leave the volume element in the (x. −t + fq .130) δt = t x x x ˆ .

One simplification we can make in this illustration is to postulate eigenvalues in the parton space and solve the equation for them. defined as the difference of two parton densities fq = (fu − fu ). let us briefly recapitulate. so the finite off-diagonal quark-gluon and gluon-quark splitting is unchanged. µF ) . including in the soft limit z → 1. at least in . µF ) = d log µ2 F 2.135) and concludes our computation of all four regularized splitting functions which appear in the DGLAP equation Eq. j (2. The original collinear divergence has vanished as well. including the factor two in front. The only approximation we make in the computation of the splitting kernels is that in the y integrals we implicitly assume that the running coupling αs does not depend on the momentum fraction.127). splitting kernels are only regularized when needed.82 ˆ plus prescription for Pg←g 1 2 QCD − 0 dy αs ˆ αs Pg←g (y) = − CA 2π 2π αs =− CA 2π =− =− αs CA 2π αs CA 2π 1 dy 0 1 y 1−y + + y(1 − y) 1−y y 2y + y(1 − y) 1−y 2(y − 1) αs + y(1 − y) − CA 1−y 2π αs 2CA 2π 1 1 dy 0 1 dy 0 1 dy 0 1 2 1−y z 1−y dy −2 + y − y 2 − δ(1 − z) 0 dy 0 αs =− CA 2π 1 1 αs −2 + − − δ(1 − z) 2CA 2 3 2π 1 0 dy 0 z 1−y (2. The DGLAP equation is formulated in terms of four splitting kernels of gluons and quarks which are linked to the splitting probabilities.136) Solving the integro-differential DGLAP equation Eq. One such parton density is the NS non-singlet parton density parton density. This means the plus prescription really acts as an infrared renormalization. µF z 2π z = αs 2π (Pi←j ⊗ fj ) (x. This defines the last remaining regularized splitting kernel Pg←g (z) = 2CA z 11 2 1−z + z(1 − z) + CA δ(1 − z) + nf TR δ(1 − z) + (1 − z)+ z 6 3 (2. The universality of the splitting kernels is obvious from the way we derived them — no information on the n-particle process ever entered the derivation. we will sketch how we would approach this. However. moving universal infrared divergences into the definition of the parton densities.(2.136) for the parton densities is clearly beyond the scope of this writeup. For the full quark and gluon particle content of QCD we have derived the DGLAP equation which describes a (so-called factorization) scale dependence of the quark and gluon parton densities. Nevertheless. This gets rid of the sum over partons on the right hand side.3.134) αs 11 αs = CA − δ(1 − z) 2CA 2π 6 2π z dy .(2.4 Parton densities 1 j x dz αs x Pi←j (z) fj . With the help of a plus-subtraction the kernels Pi←j (z) are all finite. Since ¯ gluons cannot distinguish between quarks and antiquarks. but which for the DGLAP equation have to be regularized. In its standard form and in terms of the factorization scale µ2 ≡ −t the DGLAP equation reads F dfi (x. This will give us some information on the structure of its solutions which we will need to understand the physics of the DGLAP equation. 1−y The contribution proportional to δ(1 − z) is necessary to give a finite result and is absorbed into the plus prescription ˆ of the convoluted Pg←g (z). Before discussing and solving the DGLAP equation. the gluon contribution to their evolution cancels.

xn (2. namely a Mellin transform.141) NS = M[fq ](m. µF z 2π z = αs 2π NS Pq←q ⊗ fq (x. µF. because we cut it off at Q2 etc.2.(2. d log µ2 d log µ2 d log αs αs d log µ2 d log αs d log αs This additional factor of αs on the left hand side will cancel the factor αs on the right hand side of the DGLAP equation Eq.0 ) αs (µ2 ) F. Starting from a function f (x) of a real variable x we define the Mellin transform into moment space m 1 M[f ](m) ≡ 0 dx xm−1 f (x) f (x) = 1 2πi c−i∞ dn c−i∞ M[f ](m) . (2. µ) = M[fq ](m. µ2 ) Pq←q F x x (2. µF ) = M[fq ](m. µF. This allows us to replace log µ2 in the DGLAP equation by αs .0 αs 2π = NS M[fq ](m. which fixes the offset c. Our convolutions is not really a convolution either.142) Among other things in this derivation we neglect that some splitting functions have singularities and therefore the Mellin transform is not obviously well defined.0 µ2 F µ2 F.0 ) µ2 F µ2 F. µ) 1 (0) NS =− γ (m) M[fq ](m.0 M[Pq←q ](m) αs 2π γ(m) ≡ NS M[fq ](m. The corresponding DGLAP equation with leading order splitting kernels now reads NS dfq (x. 0 .0 ) .139) NS dM[fq ](m. Instead of assuming a fixed αs in the transformed DGLAP equation Eq. µF ) .139) we can include αs (µ2 ) R in the running of the DGLAP equation.138) The integration contour for the inverse transformation lies to the right of all singularities of the analytic continuation of M[f ](m). The Mellin transform of a convolution is the product of the two Mellin transforms.3 Infrared divergences 83 the massless limit. µF ) = d log µ2 F 1 x dz αs NS x Pq←q (z) fq . (2. (2. µ0 ) exp − γ (0) (m) log 2πb0 αs (µ0 ) (2. which gives us the transformed DGLAP equation 1 NS M[Pq←q ⊗ fq ](m) = M 0 dx x2 NS NS fq (x) (m) = M[Pq←q ](m) M[fq ](m. µF. including the leading order Jacobian d d log αs d 1 dαs d d = = = −αs b0 . identifying the renormalization scale µR of the strong coupling with the factorization scale µF = µR ≡ µ. µ) d log αs 2πb0 1 αs (µ) NS NS M[fq ](m.137) To solve it we need some kind of transformation which disentangles the convolution. but the final structure in Eq. since flavor SU (3) commutes with the QCD gauge group.0 ) exp αs µ2 M[Pq←q ](m) log 2F 2π µF. µ2 ) F d log µ2 2π F and its solution NS NS M[fq ](m. µF.(2.0 αs (µ2 ) γ (0) (m) 2πb0 .139) NS dM[fq ](m. µF ) αs NS = M[Pq←q ](m) M[fq ](m.142) really holds.(2. This will be true at arbitrary loop order.140) defining γ(n).

The same kind of behavior we see in Eq. The DGLAP equation therefore does not determine parton densities. µ0 ) . but we have indeed identified the factorization and renormalization scales and this way mixed the concepts of scales arising from ultraviolet and infrared poles. µ) = M[fg ](m.e. For obvious reasons we call η the anomalous dimension of GPlanck .5 Re-summing collinear logarithms Remembering how we arrived at the DGLAP equation we notice an analogy to the case of ultraviolet divergences and the running coupling. and secondly there is η. as does the electric charge in QED. The DGLAP equation plays the role of a renormalization group equation for example for the running coupling. i. and we find in complete analogy to Eq. It links parton densities evaluated at different scales µF .142) αs (µ2 ) F. it only describes their evolution from one scale µF to another. To make sense of such a structure we remind ourselves that such ratios of αs values to some power can appear as a result of a re-summed series. Parton densities cannot be formulated in terms of a fixed final state because they include effects from any number of collinearly radiated partons summed over the number of such partons. (2.5: the splitting probability we find in the exponent. Looking at the different color factors in QCD this limit can also be derived as the leading terms in Nc . Why is γ (0) is referred to as the anomalous dimension of the parton density? This is best illustrated using a running 2 coupling with a finite mass dimension.0 αs (µ2 ) γ (0) (m) 2πb0 M[fg ](m. Such a series would need to include powers of ˆ (M[P ])n summed over n which corresponds to a sum over splittings with a varying number of partons in the final state.(2. an anomalous dimension arises through renormalization involving a ultraviolet divergence and the renormalization scale. in QCD without quarks. .143) for the Mellin-transformed parton density which has no initial mass dimension.3. There is only a slight stumbling step in this analogy: usually. (2. just like a renormalization group equation in the ultraviolet. there is the mass dimension of the bare coupling n − 2. which means that its entire running arises from quantum effects.84 2 QCD Because we will need it in the next section we emphasize that the same kind of solution appears in pure Yang–Mills theory. In our case we are officially discussing an infrared divergence and the factorization scale dependence. αs only has contributions of the second kind.143) To remind ourselves that in this derivation we unify the renormalization and factorization scales we denote them just as µ. 2. a quantum effect from the coupling renormalization. In n dimensions this gives us g bare = M n−2 GPlanck g bare = M n−2 GPlanck → Zg(M 2 ) . like the gravitational coupling GPlanck ∼ 1/MPlanck .(2. This solution to the DGLAP equation is not completely determined: as a solution to a differential equation it also includes an integration constant which we express in terms of µ0 . there are two sources of running for the renormalized coupling g(M 2 ): first. In that case there exists also only one splitting kernel defining an anomalous dimension γ. This is parameterized by a finite γ (0) . The structure of Eq.144) For the dimensionless gravitational coupling we can compute the running dg(M 2 ) d = d log M d log M 1 n−2 M GPlanck Z 1 dZ 1 dM n−2 M − 2 M n−2 = GPlanck Z dM Z d log M dZ 1 = g(M ) (n − 2 + η) with η = − Z d log M (2.3.(2. In contrast.143) already shows something we will in more detail discuss in the following Section 2. We start from universal infrared divergences which we describe in terms of splitting functions which we regularize using the plus prescription.145) Hence. When we attach a renormalization constant Z to this coupling we first define a dimensionless running bare coupling g.

148) we can then apply iteratively 1 σn+1 (x.149) µ2 F × µ2 0 dp2 αs (p2 ) T.(2.n dp2 αs (p2 ) T. We start from the basic equation defining the physical picture of parton splitting in Eq.(2.3 Infrared divergences 85 In analogy to the scaling logarithms considered in Section 2. µ0 ) (2. which is then bounded by p2 . µ0 ) µ2 0 dp2 αs (p2 ) T.(2.143).2. If beyond the single parton radiation discussed in Section 2. The recursion in Eq.1 is bounded from above by the next external scale p2 .3 we now test if we can point to a type of logarithm the DGLAP equation re-sums by reorganizing our perturbative series of parton splitting. the first splitting integrated over p2 T.94).n T.2. If we remember that t = p2 < 0 we can follow Eq. t pb pT (2.n ˆ The dz in Eq.(2.n .146) we replace by the proper convolution P ⊗ σn .1 we consider a ladder of successive splittings of one gluon into two and we for a moment forget about the actual parton densities we can schematically write the hadronic cross section in the collinear limit as σn+1 (x.n p2 2π T. µ2 ) F µ2 F = µ2 0 1 dp2 αs (p2 ) T.(2. Only taking into account gluons in pure Yang–Mills theory it precisely corresponds to the starting point of our discussion leading to the DGLAP equation σn+1 = σn αs ˆ dp2 T dz Pg←g (z) . p2 2π T (2.148) = x0 dxn ˆ Pg←g xn σn (xn .n ··· p2 2π T. To identify these re-summed logarithms we build a physical model based on collinear splitting but without using the DGLAP equation.n x xn 1 x0 dxn ˆ Pg←g xn µ2 F x xn σn (xn .146) What we now need is an exact physical definition of the virtuality variable t describing initial-state splitting. p2 2π T. Similarly. We then solve it for the structure of the parton densities and compare it to the structure of the DGLAP solutions in Eq. Last but not least. µ2 ) ∼ F x0 dxn ˆ Pg←g xn x xn 1 ··· x0 µ2 F µ2 0 dx1 ˆ Pg←g x1 x2 x1 σ1 (x1 . we define our model with the transverse momentum as the variable renormalization scale in the running coupling. evaluated at the momentum fraction x.1 The two sets of integrals in this equation we will solve one by one. the transverse momentum integral is bounded by an infrared cutoff µ0 and the physical external scale µ2 . starting from the pT integrals. For the switch of the two integrals it is F crucial that µ0 is the only scale the final matrix element σ1 depends on. For the n-fold p2 T.85) we see that p2 is really the transverse T three-momentum of of the parton pair after splitting. p2 2π T.(2.2 T.n T.3 T .1 . etc. The crucial physics assumption in our multiple-splitting model splitting concerns the integration boundaries — which ˆ until now we always assume to be somehow covered by the definition of P : in the successive transverse momentum integrations we need to assume that the transverse momenta be strongly ordered.1 T.n T. Because the splitting kernel is infrared divergent we cut off the convolution integral at x0 .147) From the definition of pT in the Sudakov decomposition Eq.113) and introduce a positive transverse momentum variable p2 such T b that −t = −p2 = − b p2 p2 T ≡ T >0 1−z 1−z ⇒ dt dp2 dp2 = 2b = 2T .3. µ0 ) (2.

2 T d log log µ2 0 p2 T.2 T T p2 µ2 T.n−1 /ΛQCD log µ2 /Λ2 0 QCD p2 .1 log 2 ΛQCD (2. The chain of integrals over pT.2 log 2 ΛQCD dp2 T. µ2 ) ∼ 1 n! 1 αs (µ2 ) 0 log 2πb0 αs (µ2 ) n 1 x0 dxn ˆ Pg←g xn x xn 1 ··· x0 dx1 ˆ Pg←g x1 x2 x1 σ1 (x1 .1 T.2 2πb0 log 2 ΛQCD 1 p2 T.n log µ2 /Λ2 0 QCD log p2 /Λ2 QCD T.(2.n 2πb0 log 2 ΛQCD dp2 T.2 T µ2 0 dp2 T.1 =p2 .1 Λ2 QCD with d(ax) = d log log x (ax) log x p2 .153) .1 p2 T. To compute the convolution integrals over the momentum fractions in the same equation σn+1 (x.n p2 T.2 .n log µ2 /Λ2 0 QCD n d log µ2 0 p2 .2 log µ2 /Λ2 0 QCD p2 .2 T d log µ2 0 log p2 /Λ2 QCD T.2 F µ2 F µ2 0 µ2 F dp2 αs (p2 ) T.2 p2 .2 p2 T.n log µ2 /Λ2 0 QCD .3 log µ2 /Λ2 0 QCD = µ2 0 d log 1 1 ··· 2 n−1 = 1 n! log log log p2 /Λ2 QCD T. p2 T.2 p2 .150) as in the last section identifying the scale of the strong coupling αs with the transverse momentum scale of the splitting µR = µF = pT .n log 2 ΛQCD p2 .1 1 p2 T.n ··· p2 2π T.86 2 QCD integration this means µ2 < p2 < p2 < · · · < µ2 .1 =µ2 0 T log p2 /Λ2 QCD T.1 − log log 2 0 2 ΛQCD ΛQCD p2 .n log µ2 /Λ2 0 QCD log p2 /Λ2 QCD T.1 1 = p2 T.149) then become 0 T. provided the integrand does T.n T ··· µ2 0 d log log p2 /Λ2 QCD T.2 log µ2 /Λ2 0 QCD = µ2 0 µ2 d log d log µ2 0 2 log p2 1 T.n not have any additional pT.1 1 ··· .2 T d log log µ2 0 p2 µ2 T.152) This is the final result for the chain of transverse momentum integrals in Eq.2 T.n 1 p2 T.n p2 T.3 T µ2 0 dp2 T. for example in the last integral p2 .n log µ2 /Λ2 0 QCD p2 .3 T µ2 0 p2 .2 T µ2 0 dp2 T.1 p2 T.n T.1 = log log 2 − log log 2 0 ΛQCD ΛQCD = log p2 .1 2πb0 log 2 ΛQCD ··· = 1 (2πb0 )n µ2 F µ2 0 1 ··· p2 T.3 T d log µ2 0 p2 .1 log µ2 /Λ2 0 QCD = µ2 0 µ2 d log log p2 /Λ2 QCD T. 1 = n! log µ2 /Λ2 QCD log log µ2 /Λ2 0 QCD αs (µ2 ) 0 αs (µ2 ) n (2. The transverse momentum integrals in Eq.1 p2 T.151) In the second line we have shifted the integration variable by a constant.n dp2 T.2 p2 2π T.3 T log p2 /Λ2 QCD T.149).2 T µ2 0 = µ2 0 ··· dp2 T.n we can solve.n p2 . To simplify the notation we have kept the integration boundaries in terms of p2 .j dependence. These integrals we can solve by switching variables.n−1 /ΛQCD ··· log µ2 /Λ2 0 QCD 2 log p2 T.1 log 2 ΛQCD = p2 .1 p2 .n−1 /ΛQCD ··· log µ2 /Λ2 2 0 QCD log n−1 log p2 /Λ2 QCD T. This gives us µ2 d log µ2 0 µ 2 log p2 /Λ2 QCD T.2 T µ2 0 dp2 αs (p2 ) T. log µ2 /Λ2 0 QCD (2.n T 2 log p2 T.3 T µ2 0 dp2 αs (p2 ) T.n T d log µ2 0 p2 .2 log µ2 /Λ2 0 QCD 2 log log p2 /Λ2 QCD T.1 T.(2.2 p2 T. µ0 ) (2.2 1 p2 T.1 ··· p2 2π T.

on the right hand side we see fixed order Drell–Yan production without any additional partons.156) we have the sum over any number of additional collinear partons.156) This is the same structure as the DGLAP equation’s solution in Eq. They are universal and simply come from the phase space integration. After renormalization as well as after mass factorization we are left with a scale artifact. Scales arise as part of a the pole subtraction: together with the pole 1/ we have a choice of finite contributions which we subtract with this pole. µ0 ) (m) γ(m)n M[σ1 ](m. for example using dimensional regularization where we choosing n = 4 − 2 < 4 for ultraviolet and n > 4 for infrared divergences is linked by analytic continuation. Comparing this to the running parton densities we can draw the analogy that any process computed with a scale dependent parton density where the scale dependence is governed by the DGLAP equation includes any number of collinear partons. µ0 ) n=0 n 1 n! 1 αs (µ2 ) 0 log γ(m) 2πb0 αs (µ2 ) n =M[σ1 ](m.154) We can now sum the production cross sections for n collinear jets and obtain ∞ M[σn+1 ](m) =M[σ1 ](m. µ0 ) exp γ(m) αs (µ2 ) 0 log 2πb0 αs (µ2 ) (2. Bending the rules of pure Yang–Mills QCD we can come back to the hard process σ1 as the Drell–Yan process q q → Z. computing perturbative higher order contributions to scattering amplitudes we encounter ultraviolet and infrared divergences. For both kinds of divergences we noticed that they are universal. Because this scale can be chosen freely we only have to make sure that it does not become too large.155) This way we can write the Mellin transform of the (n + 1) particle production rate as the product of the n-particle rate times a ratio of the strong coupling at two scales ∞ M[σn+1 ](m) = M[σ1 ](m. but with an exponentiated correction factor. µ0 ) n=0 αs (µ2 ) 0 αs (µ2 ) γ(m) 2πb0 (2. Each step in n means an additional parton in the final state.82) ceases to hold. not process or observable dependent. i. It means that we should be able to understand the physics of the DGLAP equation using our model calculation of a gluon ladder emission.4 Scales in LHC processes 87 we again Mellin transform the equation into moment space M[σn+1 ](m) ∼ = = 1 n! 1 n! 1 n! 1 αs (µ2 ) 0 log 2πb0 αs (µ2 ) 1 αs (µ2 ) 0 log 2πb0 αs (µ2 ) n 1 M x0 n dxn ˆ Pg←g xn x xn 1 ··· x0 dx1 ˆ Pg←g x1 x2 x1 σ1 (x1 . so σn+1 is Z production with n collinear partons ¯ On the left hand side of Eq. In the pure Yang–Mills example we can identify the logarithms which are re-summed beyond fixed-order perturbation theory by scale dependent parton densities. This allows us to absorb ultraviolet and infrared divergences into a re-definition of the strong coupling or the parton density.(2.2 and 2. Both of them we regularize. The upper limit of the transverse momentum of this implicit additional parton radiation is the factorization scale µF . In Eq.143). µ0 ) (2.(2. because at some point the collinear approximation C constant in Eq.(2.4 Scales in LHC processes Looking back at Sections 2. µ0 ) n 1 αs (µ2 ) 0 log γ(m) 2πb0 αs (µ2 ) M[σ1 ](m.(2.149) we see that the pT integrals lead to something like αs log µ2 /µ2 from 0 F each of the transverse momentum integrals.3 we introduced the factorization and renormalization scales step by step completely in parallel: first. 2.2. . This nominally infinite shift of parameters we refer to as renormalization for example of the strong coupling or as mass factorization absorbing infrared divergences into the parton distributions.e. Identifying the factorization and renormalization scales the DGLAP equation hence re-sums logarithms of the maximal over then minimal transverse momentum size of the incoming gluon.

2 QCD source poles cancelled summation parameter evolution large scales theory background Table 2: Comparison of renormalization and factorization scales appearing in LHC cross sections. µF ) fj (x2 . but this issue has been resolved. In a process with one . µF . which again makes them more similar to the DGLAP equation. in both cases the re-definition of parameters is not based on fixed order perturbation theory.e. but in general they are sets of coupled differential equations for all relevant parameters. Logarithms of the renormalization and factorization scales will always be part of these finite terms. i. µF ) σij (x1 x2 S. µF ) DGLAP equation increase of σtot for gluons/sea quarks factorization proven all order for DIS proven order-by-order DY. either order-by-order or rearranged into a re-summation. The universality of infrared splitting kernels has not (yet) in general been proven. universal counter terms is called renormalizability and has been proven to all orders for the kind of gauge theories we are dealing with. Once we include all orders in perturbation theory it does not matter according to which combination of couplings and logarithms we order it. An LHC production rate will then not depend on arbitrarily chosen renormalization or factorization scales µ. Our renormalization group equation for αs is a single equation. µR ) . The evolution of all running parameters from one renormalization/factorization scale to another is described either by renormalization group equation in terms of a beta function in the case of renormalization and by the DGLAP equation in the case of mass factorization. these logarithms are small and do not spoil perturbation theory. This scale dependence will only cancel once we include all implicit and explicit appearances of the scales at all orders. The fact that we can actually absorb ultraviolet divergences into process-independent. What actually happens if we include all orders in perturbation theory? For example the re-summation of the self-energy bubbles simply deals with one class of diagrams which have to be included. Practically. in Eq. ˆ R (2. Lepton pair production and Higgs production in weak boson fusion are the two prominent electroweak production processes at the LHC. For a while we thought there might be a problem with factorization in supersymmetric theories using the MS scheme.88 renormalization scale µR ultraviolet divergence counter terms (renormalization) re-sum self energy bubbles running coupling αs (µ2 ) R RGE for αs decrease of σtot renormalizability proven for gauge theories factorization scale µF collinear (infrared) divergence parton densities (mass factorization) re-sum parton splittings running parton density fj (x.(2. it does not ˆ ¯ include αs with its implicit renormalization scale dependence at leading order. The hadronic production cross section for the Drell–Yan process or other LHC production channels. Instead. A summary of the properties of the two relevant scales for LHC physics we show in Table 2. µR ) = 0 dx1 0 dx2 ij fi (x1 . reads 1 1 σtot (µF . There is one formal difference between these two otherwise very similar approaches. The way we introduce factorization and renormalization scales clearly labels them as an artifact of perturbation theories with divergences..157) we evaluate the renormalized parameters and the parton densities at some scale.157) The Drell–Yan process has the particular feature that at leading order σqq only involves weak couplings. In the ideal case. now including both scales. but on the other hand we have never seen an example where is failed for sufficiently inclusive observables like production rates. Whatever scale we choose for the strong coupling or parton densities will eventually be compensated by explicit scale logarithms. it involves summing logarithms which otherwise can become large and spoil the convergence of our perturbative series in αs . Moreover.. αs (µ2 ). The only special feature of infrared divergences as compared to ultraviolet divergences is that to identify the resummed logarithms we have to unify both scales to one.

4 Scales in LHC processes 89 distinct external scale. like the Z mass. This logarithm is an artifact in another way: it does not appear if we evaluate everything at the ‘correct’ external energy scale.2. Estimating theoretical uncertainties from scale dependence therefore requires a good understanding of the individual process and the way it is affected by the two scales. there are more or less smart choices. The transverse momentum of a final state particle is one of scale choices allowed by factorization. because a cross section even to all orders does depend on the energy. there are definitely completely poor scale choices. The appropriate interval of what we consider reasonable scale choices depends on the process and the taste of the people doing this analysis. and there are definitely very poor choices. which usually lead to an unstable perturbative behavior. which according to the argument above all scale logarithms have to have the form log µ/mZ . Even others appear though background rejection cuts in a specific analysis. At the same time we know that the next-to-leading order correction to the Drell–Yan cross section is of the order of 30%. using 1000 × mZ as a typical scale in the Drell–Yan process will if nothing else lead to logarithms of the size log 1000 whenever a scale logarithm appears. Some of them will appear in the hard process. which means that the two sources of soft divergences only cancel until we multiply each of them with numerically different parton densities. evaluating the strong coupling at the transverse momentum of the radiated jet αs (p2 ).j distributions for the additional partons because in Section 2. the different scaling behavior of a hadronic cross section shown in Table 2 implies that for example gluon-induced processes at typical x values around 10−2 show a cancellation of the factorization and renormalization scale variation. The problem in particular for LHC analyses is that after analysis cuts every process will have more than one external energy scale. i. Moreover. If we set µ = mZ all scale logarithms will vanish. Others will enter through the QCD environment where at the LHC we only consider final state jets above a certain minimal transverse momentum. In reality. or the mass of heavy new-physics states in their production process. We can turn around the argument of vanishing scale dependence to all orders in perturbation theory.3.5 we have shown that it re-sums large collinear logarithms. we usually introduce a phase-space dependent renormalization scale. scale choices violating physical properties we later need. For example. often impossible. for example the partonic center-of-mass energy s = x1 x2 S. namely mZ . this implies that in addition to poor scale choices there are wrong scale choices. which far exceeds the factorization scale dependence. Using likelihood methods does not improve the situation because the phase space regions dominated by the signal will have specific energy scales which shape the background. for example in the production of two or three particles with different masses. However. For the Drell–Yan process such a scale could be mZ . All that means that in practice there is no way to define a ‘correct’ scale. To summarize this brief section: while there is no such thing as a correct scale choice. This choice gives the best kinematic T. . These logarithms eventually have to be cancelled to all orders in perturbation theory. This dependence on the external energy is not a perturbative artifact. virtual and real gluon emission generate different multiplicities in the final state. mZ . Imagine picking a factorization scale defined by the partonic initial state. When describing jet radiation. inducing unreasonably large higher order corrections. This error estimate is not at all conservative. we know that all scale logarithms should have the form log µ/mZ . This is one of the reasons why an automatic comparison of LHC events with signal or background predictions is bound to fail once is requires an estimate of the theoretical uncertainty on the background simulation. From many discussions in this section we know that this definition is not unique: for any final state it corresponds to the well defined sum of all momenta squared. for example the renormalization scale dependence of the Drell–Yan production rate or Higgs production in weak boson fusion is zero because αs only enters are next-to-leading order. For example in Drell–Yan production at leading-order there exists only one scale. Factorization or the Kinoshita-Lee-Nauenberg theorem which ensures that soft divergences cancel between real and virtual emission diagrams are such properties we should not violate — in QED the same property is called the Bloch-Nordsieck cancellation. Only scales which are uniquely defined in the final state can serve as factorization scales. LHC observables will include several different scales. On the other hand. for example when we only consider a Drell–Yan background for mµµ > 1 TeV to Kaluza-Klein graviton production. This gives us an estimate of the minimum theoretical error on a rate prediction set by the scale dependence.e. In that sense we can think of the running coupling as a proper running observable which depends on the external energy of the process. Guessing the right scale choice for a process is hard.

5. 2π  (2. Using Poisson statistics for a something expected to occur p times the probability of observing it n times is given by P(n. p) = pn e−p n! P(0.e.2 we will discuss some of the physical properties of the parton shower.5 we have already seen that there are other structures with the same structure as solutions to the DGLAP equation.2.(2. The appropriate common starting scale t0 drops out of this ratio.160) Above the absolute starting scale of the parton shower t0 we assume that no resolvable branching occurs. Following the original discussion t2 is the target virtuality at x2 . Per se it it not clear how jet radiation described by the parton shower and jet radiation described by fixed-order QCD processes are linked.5 Parton shower In LHC phenomenology we are less interested in fixed-order perturbation theory than in typically logarithm-enhanced QCD effects.158) gives us this probability distribution which we can again implicitly solve for x2 using a flat random number rx x2 /x1 0 αs ˆ dy P (y) 2π = rx αs ˆ 1 dy P (y) 0 2π [0.5. similar to our discussion of the central jet veto in Section 1.1 Sudakov form factor ˆ ˆ The splitting kernels Pi←j (z) and Pi←j (z) we have introduced as something like splitting probabilities.160) and (2. (2.161) give us two. In Section 2. In a second step we need to compute the matching energy fraction x2 or the ratio x2 /x1 describing the momentum fraction which is kept in the splitting at x2 . t1 ) in momentum–virtuality space we step by step move to the next splitting point (xj . and for time-like final state branching the virtuality is positive tj > 0 in all points j. (2. 1] with t1 > t2 > t0 > 0 . In this section we will look for other structures which obey the DGLAP equation. but we have never applied a probabilistic approach to parton splitting. 2. Using probabilistic Sudakov factors we can use a Monte Carlo (i.(2. tj ). The on-shell mass constraint fixes a third. 1] with x1 > x2 > 0 .(2. The Sudakov factor is a function of t. so all we are left is the azimuthal angle distribution.5. Starting from a point (x1 . so it gives us the probability of not seeing any branching between t1 and t2 as ∆(t1 )/∆(t2 ) < 1.90 2 QCD 2. The y integral in the Sudakov factor in Eq.158) Before we check that such a function can obey the DGLAP equation we confirm that such exponentials appear in probabilistic arguments. p) = e−p . This will describe a quark or a gluon propagating forward in time.3. leading us to Sudakov form factors and the parton shower.3 we will discuss ways to combine the two approaches in realistic LHC simulations.(2.5.158) describes some kind of non-splitting probability of the parton i into any final state j. Therefore we will not deepen our discussion of hadronic rates as shown in Eq.157) based on fixed-order partonic cross sections convoluted with parton densities obeying the DGLAP equation. bringing us very close to the contemporary research topics. In Section 2.159) If the exponent in the Sudakov describes the integrated splitting probability of a parton i this means that the Sudakov form factor in Eq. In Section 2. The basis of such an interpretation are Sudakov form factors describing the splitting of a parton i into any of the partons j  ∆i (t) = exp − j t0 t dt t 1 dy 0 αs ˆ Pj←i (y) . (2. Of the four momentum entries of the radiated parton the two equations Eqs. a Markov process without a memory of individual past steps) to compute a chain of parton splittings as depicted in Figure 11. Using a flat random number rt the t2 distribution is implicitly given by the solution to ∆(t1 ) = rt ∆(t2 ) [0. We know from symmetry .161) For splitting kernels with soft divergences at y = 0 or y = 1 we should include a numerical cutoff in the integration because the Sudakov factor and the parton shower does not involve the regularized splitting kernels.

The derivative of the Sudakov form factor Eq. i. t) fi (x. For final state radiation this describes probabilistic branching in a Monte Carlo program.t z 2π z 1− = ∆i (t.t z 2π z . .e. t) = t t0 dt t 1− j 0 x fj . t) d∆i (t) d fi (x. z 2π z ∆i (t) dt This relation suggests to consider the derivative of the fi /∆i instead of the Sudakov factor alone 1 dfi (x.t + . (2. The resulting equation is the diagonal DGLAP equation with unsubtracted splitting kernels.t z 2π z . (2.166) = ∆i (t)fi (x. solved by the ratio of parton densities and Sudakov factors d fi (x. just based on Sudakov form factors.120) dfi (x. as shown in Eq.165) We can study the structure of these solutions of the unsubtracted DGLAP equation by integrating f /∆ between appropriate points in time fi (x. t) d∆i (t)  fi (x. t) fi (x. t) = − dt ∆i (t) ∆i (t) dt ∆i (t)2 dt   1 dz αs ˆ x fi (x.t dz αs ˆ z Pi←j (z) z 2π ∆i (t ) t t0 t ∆i (t) fi (x. t) = = t dt ∆i (t) d log t ∆i (t) 1− j 0 x fj . t) 1 = dt t = 1 t 1 j 0 1 j 0 dz αs ˆ x Pi←j (z) fj . t0 ) + t0 dt ∆i (t) t ∆i (t ) t dz αs ˆ x Pi←j (z) fj .2. t0 )fi (x. t0 ) + t0 dt ∆i (t.t − z 2π z 1 dy 0 αs ˆ Pj←i (y) fi (x. we have removed the regularization of the splitting kernel at z → 1. t) d∆i (t) 1 1 Pi←j (z) fj .t + − = ∆i (t) t j 0 z 2π z ∆i (t) dt ∆i (t)2 dt = 1 1 ∆i (t) t 1− j 0 dz αs ˆ x Pi←j (z) fj . The same statement for initial state radiation including parton densities we will put on a more solid or mathematical footing.163) dz αs ˆ x fi (x.162) is precisely the second term in df (x. t)/dt for diagonal splitting. t) 2π (2. t) d fi (x.158) 1 d∆i (t) =− ∆i (t) dt 1 t 1 dy 0 j αs ˆ Pj←i (y) 2π (2.5 Parton shower 91 arguments that QCD splitting is insensitive to this angle. so we can generate it randomly between zero and 2π.(2.164) In the last step we have cancelled what corresponds to the plus prescription for diagonal splitting.t z 2π z . t ) t j 0 dz αs ˆ x Pi←j (z) fj . t0 ) + ∆i (t0 ) dt ∆i (t) t ∆i (t ) 1− j 0 1− j 0 dz αs ˆ x Pi←j (z) fj . (2.t dz αs ˆ z Pi←j (z) z 2π ∆i (t) .(2. t) d∆i (t) Pi←j (z) fj . we need to modify the upper integration boundary by a small parameter which can in principle depend on t. Therefore. t0 ) − = ∆i (t) ∆i (t0 ) fi (x.

For this final step we define a probability measure for the backwards evolution of partons Π(t1 .(2. ideally to a point where the probabilistic picture of collinear. t2 )∆i (t1 ) with |t2 | > |t1 | . This means we need to define a probability that a parton evolved backwards from a space-like t2 < 0 to t1 < 0 with |t2 | > |t1 | does not radiate or split.98) and (2.92 2 QCD where we choose t0 such that ∆(t0 ) = 1 and introduce an alternative notation ∆(t1 .94. t2 )∆i (t1 ) (2. 1] fi (x.(2. Deriving the diagonal splitting kernels in Eqs. as dF (t. This formula for the dependence of the parton density fi (x. tn ) given by the hard subprocess. t1 )∆i (t2 ) fi (x.(2. For example. x).159) the first term can be interpreted as ‘nothing happening with f between t0 and t’ because it is weighted by the Sudakov no-branching probability ∆i .168) fi (x. t2 ) = dt ∆i (t2 ) t ∆i (t) 1− j 0 1− j 0 dz αs ˆ x Pi←j (z) fj . The second term includes the ratio of Sudakov factors which just like in Eq. Integrating this factor times the actual splitting probability over t [t0 .2 Soft gluon emission To this point we have built our parton shower on collinear parton splitting or radiation and its universal properties indicated by Eq. (2. t2 ) = ∆(t1 .5.166) and replacing t → t2 and t → t we know what happens for the combination fi (x.165) . t + δt]. e. 2.(2. Numerically it would be much easier to simulate backwards evolution where we start from the known kinematics of the hard process and the corresponding point in the (x − t) plane and evolve towards the partons in the proton. t] implies at least one branching between t0 and t. The issue of this interpretation of the Sudakov form factor in conjunction with the parton densities is its practical usability in a probabilistic approach: starting from a parton density somewhere in (x − t) space we need to evolve this parton density to a fixed point (xn . t) dt ∆i (t) (2. t2 . (2. t0 )/∆(t2 . it shows how we can algorithmically generate initial state as well as final state parton radiation patterns based on the unregularized DGLAP equation and the Sudakov factors solving this equation. t2 . The measure corresponding to a Sudakov survival probability is then t2 Π(t1 . t2 )∆i (t) = t1 fi (x. x) = 1 − d fi (x.g.160) means no branching between t and t. t) on x and t has a suggestive interpretation: corresponding to Eq. non-radiating quarks and gluons in the proton holds.167) Comparing the definition of dF to the relevant terms in Eq. One remaining issue is that in our derivation of the collinear re-summation interpretation of the parton shower we had to assume some kind of ordering of the radiated parton which we will discuss in the next section. t2 ) . t2 . stable. t)∆i (t2 ) fi (x. x). t1 )∆i (t2 ) = r [0. evaluated at (t2 .(2. t2 1 = dt ∆i (t2 ) t = dt ∆i (t2 ) This means Π(t1 .(2. t2 )dF (t. x) = 1 − t1 dF (t. q q → Z with mZ giving the scale and ¯ energy fraction of the two quarks.(2. Just like the two terms in Eq. t0 ) for the ratio of two Sudakov factors in the last line.t dz αs ˆ z Pi←j (z) z 2π ∆i (t) using Eq. we can write the probability that a parton is generated by a splitting in the interval [t.t z 2π z x fj .170) While we cannot write down this procedure in a closed form. t2 ).169) and gives us a probability measure for backwards evolution: the probability of evolving back from t2 to t1 is described by a Markov process with a flat random number as fi (x.111) we have encountered an .166) it links the splitting probability to a probability of an undisturbed evolution.

(2. This current describes the universal form of soft gluon radiation off an n-particle process 2 |Mn+1 |2 ≡ −gs (J † · J) |Mn |2 with J aµ (k. The initial quark momentum splits into a hard quark q and a soft gluon k with k 2 q 2 = m2 Mn+1 = gs T a = gs T a = gs T a ∼ gs T a → gs ∗ µ (k) ∗ µ (k) ∗ µ (k) ∗ µ (k) u(q)γ µ ¯ q+k+m / / Mn (q + k)2 − m2 1 Mn 2(qk) + O(k 2 ) u(q) [−qγ µ + 2q µ + mγ µ − γ µ k] ¯ / / u(q) ¯ q µ + O(k) / Mn (qk) + O(k 2 ) with the Dirac equation u(q)(q − m) = 0 ¯ / qµ u(q) Mn ¯ (qk)   µ qj ∗ ˆ   u(q) Mn ¯ Tja µ (k) (qj k) j (2.172) with a light-like reference vector n obeying n2 = 0. Neglecting the gluon momentum altogether is the leading term of the eikonal approximation. (qj k) (2.171) we need to square. ¯ The matrix element in Eq. namely soft gluon emission. In the last step we simply add all possible sources j of gluon radiation. Tja (qj k) (2. The matrix element squared then reads kµ nν + nµ kν + (nk) ˆ Tja j µ qj   j 2 |Mn+1 |2 = gs −gµν †  µ qj ˆ   Tja (qj k)  j  ν qj ˆ  |Mn |2 Tja (qj k) †  ˆ Tja   j j   2  = gs −   2 = −gs  j †    j (qj k) ˆ qjµ  + 2  Tja (qj k) (nk)    ˆ (qj n)   |Mn |2 Tja  (qj k) ˆ Tja µ qj   (qj k) †  j ˆ qjµ  |Mn |2 .98).173) The insertion operator in the matrix element has the form of an insertion current multiplied by its hermitian conjugate.5 Parton shower 93 additional source of infrared divergences. {qj }) = j ˆ Tja qj . For a color neutral process like our favorite Drell-Yan process adding ˆ an additional soft gluon q q → Zg it returns j Tj = 0.2. We choose the general axial gauge for massless gauge bosons ∗ µ (k) ν (k) pol = −gµν + kµ nν + nµ kν kµ kν kµ nν + nµ kν − n2 .171) The conventions are similar to Eq. This defines a color operator which we insert a a into the matrix element and which assumes values of +Tij for radiation off a quark.(2. One way to study this soft divergence without a overlapping collinear pole is gluon radiation off a massive quark with momentum q + k and mass m. attached to some hard process. = −gµν + 2 (nk) (nk) (nk) (2. −Tji for radiation off an antiquark and −ifabc for radiation off a gluon. Mn includes all additional terms except for the spinor of the outgoing quark with momentum q + k. It includes the polarization sum and will therefore depend on the gauge.174) . Its radiation pattern is equally universal.

171) ensures that the insertion currents only model soft.177) Each of the two terms we need to integrate over the gluon’s phase space. {pj }) = j ˆ Tja pj n − (pj k) (nk) (2.(2. keeping in mind that the ansatz Eq.(2. We start by symmetrizing the soft radiation dipole with respect to the two hard momenta in a particular way (J † · J)ij ∼ Wij = (qi qj ) (qi k)(qj k) 1 − cos θij = in terms of opening angles θ (1 − cos θig )(1 − cos θjg ) 1 1 − cos θij 1 1 = + − + (i ↔ j) 2 (1 − cos θig )(1 − cos θjg ) 1 − cos θig 1 − cos θjg ≡ Wij + Wij . where the current becomes dependent on the gauge vector n to cancel the n dependence of the polarization sum J aµ (k.176) we can study to see how successive soft gluon radiation will be organized for example in terms of emission angles. not collinear radiation.94 2 QCD The squared current appearing in the matrix element squared.174) and (2. we can further simplify to (J † · J) = 2 qj (qi qj ) ˆ ˆ +2 Tia Tja 2 (qj k) (qi k)(qj k) i<j j   2 qj (qi qj ) ˆ ˆ ˆ ˆ = Tja − Tia  +2 Tia Tja (qj k)2 (qi k)(qj k) j i<j i=j   q2 (qi qj ) ˆ ˆ ˆ ˆ  Tia Tja j + 2 Tia Tja = − + (qj k)2 (qi k)(qj k) i<j i<j i>j ˆ ˆ Tja Tja =2 i<j ˆ ˆ Tia Tja ˆ ˆ Tia Tja i<j 2 2 qj (qi qj ) qi − − (qi k)(qj k) 2(qi k)2 2(qj k)2 =2 =2 i<j (qi qj ) (qi k)(qj k) (qi qj ) (qi k) + (qj k) for massless partons 1 1 + (qi k) (qi k) .176) This dipole radiation term in Eqs. Calling this term a dipole is a little bit of a stretch if we compare it to the dipole term in a multi-pole series.173).(2. Eq.178) 1 = 2 dφig 0 . Just as a side remark at this stage — our definition of the insertion current J aµ in Eq. (2.(2. including the azimuthal angle integration 2π 0 dφig Wij = [i] 1 2 2π dφig 0 2π 1 − cos θij 1 1 + − (1 − cos θig )(1 − cos θjg ) 1 − cos θig 1 − cos θjg 1 1 + 1 − cos θig 1 − cos θjg 1 − cos θij −1 1 − cos θig .175) ˆ ˆ Tia Tja In the last step we bring the eikonal factor into a different form which sometimes comes in handy because it separates the two divergences associated with qi and with qj At this stage we are going back to massless QCD partons. (2.174) we can generalize to colored processes. [i] [j] (2.

0) nj = (cos θij . and the soft gluon in the full three-dimensional space ni = (1. . which for the azimuthal angle integration means 2π 0 dφig Wij = [i] π 1 + 1 − cos θig 2 1 − cos θij −1 1 − cos θig 0 2π dφig 0 2π 1 1 − cos θjg .(2. with switched roles i and j. successive emission of soft gluons in our parton shower obeys an angular ordering. (2.179) From the scalar product between these four-vectors we see that only the opening angle θjg is related to φig . 0. sin θij .180) 1 cos θig − cos θij = π+ 1 − cos θig 2 The azimuthal angle integral in this expression for Wij we can solve 2π [i] dφig 1 1 − cos θjg dφig 0 1 = 1 − cos θjg = 2π dφig 0 2π 1 1 − cos θij cos θig + sin θij sin θig cos φig 1 a − b cos φig 0 1 1 = dz z + 1/z iz unit circle a−b 2 2 1 = dz i 2az − b − bz 2 dφig 2i dz b (z − z− )(z − z+ ) 1 2i 2πi = b z− − z+ 2π =√ 2 − b2 a 2π 2π = = 2 | cos θig − cos θij | (cos θig − cos θij ) = For entire integral in Eq. 0) ng = (cos θig .2.178) this gives us 2π 0 with z = eiφig . cos φig = z + 1/z 2 with z± = a ± b a2 −1 b2 z− inside contour (2. sin θig sin φig ) 1 − cos θjg ≡ (nj ng ) = 1 − cos θij cos θig + sin θij sin θig cos φig .181) dφig Wij = [i] 1 cos θig − cos θij 2π π+ 1 − cos θig 2 | cos θig − cos θij | π = [1 + sign(cos θig − cos θij )] 1 − cos θig  π  if θig < θij 1 − cos θig = 0 else (2. sin θig cos φig . the interference partner j in the (x − y) plane. The same integral over Wij gives the same result. In other words. hard parton interference partner soft gluon (2.182) The soft gluon is only radiated at angles between zero and the opening angle of the initial parton i and its hard [j] interference partner or spectator j.5 Parton shower 95 To disentangle the different angular integrations we express the three parton vectors in polar coordinates where the initial parton i propagates into the x direction. the probability of radiating a soft gluon at an angle larger than the earlier branching is zero.

because collinear jets means many jets. The natural question then becomes: what is the range of ‘soft’ or ‘collinear’ and what is ‘hard’? Applying a consistency condition we can define collinear jet radiation by the validity of the collinear approximation in Eq. If theory does not define the golden way to order parton radiation we will have to rely on Tevatron or LHC measurements to tell us which implementation works best.5 we have learned that the physical interpretation of the collinear logarithms re-summed by the parton shower is based on the transverse momentum.5 we know that strictly speaking the parton shower only fills the phase space region up to a maximum transverse momentum pT = 0 · · · µF . 2. Collinear logarithms will become numerically irrelevant and jets will be described by the regular matrix element squared without any re-summation.(2. because quarks at the Tevatron are limited in energy. picking two particles out of four candidates with its six combinations has great potential to make our analysis a candidate for the circular folder. This final state is likely to involve more than two jets: as a start. For theorists this means that at the LHC we have to reliably model collinear and hard jets.96 2 QCD We know from Eq. This way. This is not good news.1. They fill the non-collinear part of phase space which is not covered by the parton shower. are equivalent as long as we are only interested in the leading logarithm arising from the collinear phase space integration. On the other hand. Quarks and gluons produced in association with gauge bosons at the Tevatron behave like collinear jets for pT 20 GeV. we make three histogram entries out of each signal or background event and consider all three background events plus two of the three signal combinations as background. jets produced in association with tops behave like collinear jets to pT ∼ 150 GeV. of the two most widely used event generators PYTHIA uses the virtuality or the transverse momentum while HERWIG uses the angle. In this section we have learned an argument for using the angle to order parton radiation: soft radiation is ordered in the θ. i.5 we expect many more collinear and soft jets than hard jets at the LHC. which means the pT distribution of additional jets will start dropping faster than 1/pT .3 CKKW and MLM schemes The main problem with QCD at the LHC is the range of energy scales of the jets we encounter. Because of the collinear logarithmic enhancement we have discussed in Section 2. or the virtuality t.3. Collinear jets or with their small transverse momentum are well described by partons in the collinear approximation and simulated fine by a parton shower. If we have for example three jets in the event we have to decide which of the three jet-jet combinations should go into this distribution.147) that the different variables describing parton splitting. Once we re-sum the collinear logarithms the finite shifts which appear in the transformation between the ordering parameters are also re-summed. we can ¯ . for example in the inverse Drell–Yan or R-ratio process e+ e− → q q from Section 2.82). In other words. jets produced with 500 GeV gluinos behave like collinear jets to pT scales larger than 300 GeV.1. by at least a factor of three.e. and a parton shower would be well advised to match this behavior. For harder and harder jets we will at some point become limited by the partonic energy available at the LHC. in Section 2. so the underlying variable in the parton shower does matter in the comparison to data. If this is not possible we have to consider two of the three combinations as uncorrelated ‘background’ events. For simplicity. A famous victim of such ¯ combinatorics might be the formerly promising Higgs discovery channel pp → ttH with H → b¯ b.3. This way the signal-to-background ratio decreases from NS /NB to NS /(3NB + 2NS ). the √ transverse momentum pT . At the LHC. and many jets produce combinatorial backgrounds or ruin the missing momentum resolution of the detector: if we are looking for example for two jets to reconstruct an invariant mass you can simply plot all events as a function of this invariant mass and remove the backgrounds by requiring all event to sit around a peak in mjj .81) and Eq. Different implementations of the parton shower order their jet radiation by different variables: historically. This is even more true for the matching of the parton shower and the matrix element which we will discuss in the next section. From Section 2. The maximum pT of a collinear jet is the upper end of the region for which the jet radiation cross section behaves like 1/pT or the point where the distribution pT dσ/dpT leaves its plateau. namely the angle θ.(2. In contrast. in this section we will first limit our discussion to final state radiation.3. This includes the new SHERPA parton shower which is not based on the collinear splitting kernels but on the soft-collinear QCD dipoles introduced in this section.(2. hard jets with large transverse momentum are described by matrix elements which we compute using the QCD Feynman rules. They can be transformed into each other by a simple linear transformation.5.

1. The resolution of individual jets we have identified with the matrix element–shower matching scale Qmatch . combine final states with variable number of final state particles 2.1. The analogous collinearity measure with respect to the beam we refer to as yiB . interpolate smoothly and avoid double counting of events The relevance of the first point is obvious from an experimental point of view.158) t ∆q (Qmatch . if for our closest jets we find yij > ymatch . However. In terms of Qmatch we need to define a scheme which allows us to generate events which fulfill the requirements 1. Qhard ) ≡ exp − Qmatch dq [Γg (q. like the number of jets in an event will of course depend on this resolution parameter yresol . We then define two jets as collinear if yij < yresol where yresol we give to the algorithm.2. If we shift this resolution parameter around we can choose it such that it becomes a collinear vs hard matching parameter ymatch . Its main ingredient is our knowledge of the non-splitting probabilities. The parton shower will do well for jets with yij < ymatch . If we assign the hard process a typical energy or virtuality scale Qhard we can translate the matching parameter ymatch 2 into a virtuality scale Q2 match = ymatch Qhard . we know that collinear logarithms did not play a major role. For final state radiation we have replaced t with the scale Q which corresponds to the original p2 and which moving a forward in time is ordered according to Qhard > q > Qmatch . Because this additional gluon can either be ¯ collinear with and hence geometrically close to one of the quarks or not we need an experimental measure when two somewhat collinear jets are considered one or two jets. below which we do not trust the hard matrix element. Qhard ) which for quarks and gluons are implicitly defined through the Sudakov factors which we introduced in Eq. Jet algorithms we will introduce in detail in Section 3. reproduce matrix element results in high-pT and well separated phase space region 3. Depending on the scheme this measure will involve more of the distance in R space Eq. The y integration in the original definition can be carried out in the . Analytically we can compute integrated splitting probabilities Γj (q. with no additional final state jet outside a certain resolution scale. it allows us to clearly assign each hadron collider event a number of collinear jets and a number of hard jets. For specific processes at the Tevatron the second and third point on this list have actually been tackled by so-called matrix element corrections in the parton shower Monte Carlos PYTHIA and HERWIG.183) Qhard ∆g (Qmatch . In contrast. so we should use the hard matrix element. They can transform inclusive n-particle rates into exact n-particle rates. reproduce parton shower results for collinear and soft radiation 4. i. where ‘describe’ means computing the relative probability of different phase space configurations. i. Qhard ) = exp − t0 dt t 1 dy 0 αs ˆ Pq←q (y) 2π Qhard ≡ exp − Qmatch dq Γq (q. Qhard ) + Γf (q)] .(2. Such an inclusive rate is defined as the number of events in which we for example identify n jets and ignore everything else appearing in the event. Its result.35) or more of the transverse momenta. The CKKW jet combination scheme tackles this multi-jet problem.e. which gives us a q q g final state.e.(2.5 Parton shower 97 radiate a gluon off one of the quark legs. Sudakov factors. For example for the Drell-Yan process the hard scale would be something like the Z mass. our discussion of the parton shower suggests that also from a theory point of view a fixed number of final state partons is not the right framework once we include collinear jets. Qhard ) (2. Such an event with its given number of more or less hard jets we can then describe either using matrix elements or using a parton shower. The jet algorithm simply combined jet closer than yij and keeps them separate if they are further apart than yij . They decide if or if not two jets should be re-combined to one jet based on some kind if collinearity measure yij .

but not quite. gets to Qq untouched. giving us a factor ∆q (Qmatch .185) That’s all there is. Qhard ) ≡ Γq←g (q. where the probability of no splitting should go to unity. Qg ). To get a first idea how to transform inclusive into exact n-jet rates we compute the probability to see exactly two jets.[24]. this requires us to define a vetoed parton shower which we will describe in Section 2. If we label the quark virtuality at which the second quark radiates a gluon by Qq there has to appear another factor describing that the quark. Naively we would guess that this probability is given by ∆q (Qq . We therefore compute the probability of no splitting between Qhard and Qq . we can deal with the finite terms in the Sudakov factors by requiring them to be positive semi-definite. Figure from Ref. Qhard ) < 0 by zero. Γf (q) ≡ Γq←g (q) = 3π q Γq (q. So together we have ∆q (Qmatch . Looking at Figure 12 this means that none of the two quarks radiate a resolved gluon between the virtualities Qhard (given by the qqZ vertex) and Qmatch < Qhard .184) The two arguments of Γ we have switched compared to Ref. with the exception of the intermediate quark. The two-jet (three-jet) diagram implies exactly two (three) jets at the lowest scale Qmatch . Technically.6.e. namely ∆q (Qmatch . by replacing Γ(Qmatch . [23] to agree with Ref. If we the virtuality of the radiated gluon Qg we find the gluon’s survival probability ∆g (Qmatch . Terms arising from next-to-leading logarithms spoil the limit Qmatch → Qhard . below which we rely on the parton shower. Qhard ) ≡ Γq←q (q. [24]. but what we really mean is no splitting between Qhard and Qq resolved at a third scale Qmatch < Qq given by the quark leg hitting the parton shower regime. Technically. Qg ) · · · (2. The probability we have to multiply the inclusive 2 two-jet rate with is then [∆q (Qmatch . Qhard ) Γq (Qhard . giving us 2CF αs (q 2 ) π q 2CA αs (q 2 ) Γg (q. Qq ). Qhard ). Qhard ). Qq ) ∆g (Qmatch . Qhard )] . Unfortunately. starting from Qhard . Whatever happens at virtualities below Qmatch will be governed by the parton shower and does not matter anymore.3. leading logarithm approximation. Qhard )= π q nf αs (q 2 ) . Qq ) the second quark survives to Qmatch . once for each quark. After splitting at Qq with the probability Γq (Qhard .98 2 QCD x x x x x x x x x x xx x x x x x x x x x Figure 12: Vetoed showers on two-jet and three-jet contributions. again following Figure 12? We know that it contains a factor ∆q (Qmatch . It would imply no splittings resolved at the fixed lower scale Qq . Qq ) ∆q (Qmatch . Next. but under the condition that splittings from Qq down to Qmatch are explicitly allowed. The scale at the gauge boson vertex is Qhard . what is the probability that the initially two–jet final state evolves exactly into three jets. i. Qhard )= Qhard 3 − q 4 11 Qhard − log q 12 log (2. this formula is somewhat understandable from a probabilistic picture of parton splitting. As will become important later we specify that this no-radiation statement assumes a jet resolution as given by end of the quark legs. Qhard ) for one untouched quark. to avoid unnecessary approximations in the y integration more recent CKKW implementations integrate the splitting kernels numerically. The virtualities Qhard > q correspond to the incoming (mother) and outgoing (daughter) parton. For the general argument this problem with the analytic expressions for the splitting functions is irrelevant. .

Qq ) ∆g (Qmatch . statistically picking the correct process and combining them properly. Qq ) (2. i) with its probability Pn. assumed hard. At the end. Qq ) with an unknown splitting point q. The purpose ¯ ¯ ¯ of the algorithm is to assign a weight (probability. Qhard ) exp Qmatch dq Γq (q. everything else magically drops out. Qg ) . combining different hard n-jet channels into one consistent set of events. compute the transition matrix element |M| including parton shower 5. For an arbitrary number of possible splittings between Qq and Qmatch we find the sum Qq ∆q (Qmatch .j σk. keep all splitting virtualities Qj in this event. the shower history using the kT algorithm 6. Once again: the probability of nothing happening between Qhard and Qq we compute from the probability of nothing happening between Qhard and Qmatch times any number of possible splittings between Qq and Qmatch . like q q gg and q q q q for n = 4. check that this history corresponds to possible Feynman diagrams and does not violate any symmetries 7. This cutoff ensures that all jets are hard and that all corresponding cross sections σn.186) = ∆q (Qmatch . Qhard ) ∆q (Qmatch . we only integrate over the splitting point Qq . just like in the two–parton case.e. combine matrix element. assign the momenta from the phase space generator to..5 Parton shower 99 If no splitting gives us ∆q (Qmatch .i are finite. we need exactly one splitting producing a gluon. both of which lead the same final state. Qhard ). and this gluon cannot split further. Qq ) (2. and αs into a final weight 2 k. for one one splitting between Qq we add a factor Γ(q. matrix element squared.j .(2. re-weight the αs values of each splitting using the kT scale from the shower history 9. This result is pretty much what we would expected: both quarks go through untouched.2. Qhard ) Γq (Qhard . Qq ) [∆q (Qmatch . Collecting all factors from Eq. Qq ) + · · · Qq = ∆q (Qmatch . i. Sudakovs. Qhard ) 1 + Qmatch dq Γq (q. select a final state (n. The second index i describes different non-interfering parton configurations for a given number of final state jets. we compute all leading-order cross sections for n-jet production with a lower jet radiation cutoff at Qmatch . Qq ) ∆q (Qmatch . This example illustrates how we can compute these probabilities using Sudakov factors: adding a gluon corresponds to adding a splitting probability times the survival probability for this gluon. Qhard )]2 ∆g (Qmatch . This means we one by one turn inclusive n-jet events into exact n-jet events.i 3. for each jet final state (n. This discussion allows us to write down the first step of the CKKW algorithm.) to a given phase space point.186) gives us the complete probability to find exactly three partons resolved at Qmatch as part of the inclusive sample ∆q (Qmatch . Qhard ) . In addition. This point q we integrate over between the resolution point Qmatch and the original splitting point Qq . Qg ) ∆q (Qmatch .i / 2. It proceeds event by event: 1. as requested above.185) and Eq.i = σn. We can write down the slightly simplified algorithm for final state radiation. As a starting point.(2.. external particles 4. i) compute the relative probability Pn. Qq ) = The factors 1/n! in the Taylor series appear because for example radiating two jets in the same q interval can proceed ordered in two ways. for each internal and external line compute the Sudakov non-splitting probability down to Qmatch 8. ∆q (Qmatch .187) = Γq (Qhard ..

Its main difference to the CKKW scheme is that it avoids computing the corresponding survival properties using Sudakov form factors. In other words. for example implemented in ALPGEN or MadEvent. Unfortunately. This avoids double counting of events which on the one hand are generated with n hard jets from the matrix element and on the other hand appear for example as (n − 1) hard jets with an additional jet from the parton shower. This corresponds to adding the Sudakov non-splitting probabilities in the CKKW scheme. namely if the ymatch dependence drops out of the final combined probabilities. so only experiment will tell us which scheme as part of which event generator produces the most usable results to understand LHC data. All we need to make sure is that there is no double counting. The CKKW scheme is implemented in the publicly available SHERPA and MadGraph/MadEvent event generators. If there is a significant change between the original hard parton event and the showered event this event has to go. as proven in the original paper. But the devil is in the details. After applying a jet algorithm (which in the case of ALPGEN is a cone algorithm and in case of MadEvent is a kT algorithm) we compare the hard event with the showered event by identifying each reconstructed showered jet with the partons we started from.184) which can otherwise lead to a mismatch between the actual shower evolution and the analytic expressions of the Sudakov factors. after the hard matrix element and the parton shower have defined a set of complete events. Aside from such technical details all merging schemes are conceptually similar enough that we should expect them to reproduce each others’ results. We still need a backwards evolution of a generated event to know the virtuality scales which fix αs (Q2 ). As a side effect. If all jet–parton combinations match and there exist no additional resolved jets we know that the showering has not altered the hard structure of the event. The only exception to this rule is the set of events with the highest jet multiplicity for which additional jets can only come from the parton shower. In step 4 the CKKW event generation chooses an appropriate hard scale for the parton shower to start depending on the event kinematics which affects the shower history and thereby the final weight.100 2 QCD This final event weight we can use to compute distributions from weighted events or to decide if to keep or discard an event when producing unweighted events. it attempts to veto those events which CKKW removes with the Sudakov non-splitting probabilities.(2. From the discussion of Eq. Therefore. The answer for final state radiation is yes. the veto approach allows the MLM scheme to combine a set of n–parton events after they have been generated using hard matrix elements.(2. The construction ensures that the relative weight of the different n–jet rates is identical to the probabilities we initially computed.186) we know that Sudakovs describing the evolution between two scales and using a third scale as the resolution are going to be the problem. including a hypothetical next-to-leading logarithm parton shower. To summarize. both regimes consistently combine different n–jet processes. For this parton shower simulation we need to know the interval of virtualities over which for example the additional gluon in the previous two-jet example can split. to compute the αs values at each splitting point we start the parton shower at an universal hard scale. we can use the CKKW and MLM schemes to first combine n-jet events with variable n and then consistently add the parton shower. This way it avoids problems with splitting probabilities beyond the leading logarithms. Individually. Carefully distinguishing the scale of the actual splitting from the scale of jet resolution is the key. for example the finite terms appearing in Eq. There is a simple way to check this. we can for example simulate Z + n jets production at the LHC to . The CKKW scheme starts each parton shower at the point where the parton first appears. From the final resolution scale Qresol to a matching scale Qmatch we rely on the parton shower to describe jet radiation while above the matching scale jet radiation is explicitly forbidden by the Sudakov non-splitting probabilities. In the MLM scheme we veto events which are simulated the wrong way. In a second step of the CKKW scheme we match this combined hard matrix element with the parton shower. and it turns out that we can use this argument to keep the resolution regimes y > ymatch and y < ymatch separate. chosen as the hard(est) scale of the process. given the matching point Qmatch . The lower end of this interval is universally given by Qmatch . In addition. if we also know the Feynman diagrams describing an event we can check that a certain splitting with its color structure is actually possible. but the upper end we cannot extract from the event record. and they largely do. the vetoing approach causes a slight problem for the MLM scheme. A widely used variant of the CKKW scheme is the MLM scheme. After defining the proper exclusive n-jet event sets we can again use the parton shower to describe more collinear jet radiation between Qmatch and Qresol . Instead.

Using a very pragmatic approach and tune the parton shower to correctly describe LHC data even in this parameter region will most likely violate basic concepts like factorization.5 Parton shower 101 1 1/σtot dσ/dNjet (tt/GG+jets) jet radiation – 1 1/σtot dσ/dNjet (tt/GG+jets) 10 -1 – 1 1/σtot dσ/dNjet (tt/GG+jets) 10 -1 – 10 -1 pT > 30 GeV jet radiation pT > 50 GeV jet radiation pT > 100 GeV 10 -2 GG600 GeV tt – 10 -2 10 tt – -2 GG600 GeV GG300 GeV – GG300 GeV GG600 GeV GG300 GeV -3 10 -3 10 0 1 2 3 4 5 6 7 8 Njet -3 tt 10 0 1 2 3 4 5 6 7 8 Njet 0 1 2 3 4 5 6 7 8 Njet Figure 13: Number of additional jets with a transverse momentum of at least 30. where most of these jets vanish once we require transverse momenta of at least 100 GeV. arbitrarily large numbers of jets. the only parameter which matters is their mass.j + pT. limited only by computational resources and the physical argument that at some point in Q additional jet radiation will always be described by the parton shower. [25] with technical help by Johan Alwall. i. In other words. This analysis beautifully shows that the naive parton shower is not a good description of QCD background processes to the production of heavy particles. The emission of real particles is included. Before moving on to an alternative scheme we will illustrate why Higgs or exotics searches at the LHC really care about progress in QCD simulations: one way to look for heavy particles decaying into jets. An example for such heavy states are scalar gluons with a mass of 300 or 600 GeV. In Figure 13 we show the number of jets expected to be produced in association with a pair of top quarks and a pair of heavy new states at the LHC.2. 50 or 100 GeV radiated of top pair production and the production of heavy states at the LHC.2. Using the same simulational tools we also show the missing transverse energy distribution.j + j ET. The details of these heavy scalar gluons are secondary for the basic features of these distributions. leptons) (2.188) = pT + / This variable and its relatives we will discuss in detail in Section 3. We see that heavy states come with many jets radiated at pT 30 GeV.e. This figure tells us that an analysis which asks for a reconstruction of two W decay jets better give you a very good argument why it should not we swamped by combinatorics. (for massless quarks. leptons and missing energy is the variable meff = ET + / j ET. . This combination will describe all jets correctly over the entire collinear and hard phase space. pT. pair-produced in gluon fusion. the hard scale of the process which sets the factorization scale and defines the upper limit of collinearly enhanced initial-state radiation.3. just like an old-fashioned Monte Carlo. virtual corrections are not. the CKKW and MLM schemes give us all jet distributions but leave the normalization free. For gluon-induced QCD processes the effective mass should be small while the signal’s effective mass scale will be determined by the heavy masses. For the Z+jets background we show this distribution using the CKKW scheme and purely parton shower in Figure 14. Looking at the individual columns in Figure 13 there is one thing we have to keep in mind: each of the merged matrix elements combined into this sample is computed at leading order. Curing this shortcoming will be the topic of the next section. Figures from Ref.

Obviously. This is what we call a K factor: K = σ improved /σ MC = σ improved /σ LO .3. On the other hand.miss [pb/GeV] dσ/dHT [pb/GeV] 10 pT. 2.miss > 20 GeV Sherpa ME0j+PS Sherpa ME0j1j2j3j+PS 100 dσ/dpT. we consider all contributions of the order GF αs .102 2 QCD pp ->νν+X 1000 100 pp ->νν+X Sherpa ME0j+PS Sherpa ME0j1j2j3j+PS 10 pT. As a solution we compute higher order cross sections to normalize the total cross section entering the respective Monte Carlo simulation. 2. In other words. With its collinear logarithm it sums particular terms to all orders in αs .5 we have seen that the parton shower does not respect a fixed order perturbation theory.1 0.001 0 100 200 300 400 500 600 700 800 900 1000 pT. An additional set of diagrams we should not forget are the crossed ¯ channels qg → Zq and q g → Z q . so we find the ¯ ¯ . but never touches the total cross section it started from. which could originate in supersymmetric squark and gluino production. Thank you to Steffen Schumann for providing these Figures. in Section 2.6.1 0. and there are two ways to improve this traditional Monte Carlo approach. we should try to do better than that.01 0 50 100 150 200 250 300 0. the normalization of the simulated event sample will always be leading order in perturbative QCD and hence object to large theoretical uncertainties. the virtual gluon exchange for example between the incoming ¯ quarks. The reason for this shortcoming is that collinear jet radiation starts from a hard process and the corresponding production cross section and works with splitting probabilities. and the real gluon emission q q → Zg. In such a case we need to compute observables to higher order in QCD. namely the Born contribution q q → Z. for example to Drell–Yan production. Apart from the collinear approximation for jet radiation they are limited by the fact that they ‘only do shapes’. There are three obvious sets of Feynman diagrams we have to square and multiply.miss [GeV] HT [GeV] Figure 14: Transverse momentum and HT distributions for Z+jets production at the LHC. The two curves correspond to the SHERPA parton shower starting from Drell–Yan production and the fully merged sample including up to three hard jets. These distributions describe typical backgrounds for searches for jets plus missing energy.1 Next-to-leading order in QCD When we compute the next-to-leading order correction to a cross section. Only amplitudes with the same external particles can be squared.01 0. It is crucial to remember that higher order cross sections integrate over unobserved additional jets in the final state.6 NLO and parton shower As we know for example for the R ratio the precision of a leading order QCD calculation in terms of the strong coupling constant αs is not always sufficient to match the experimental accuracy. So when we normalize the Monte Carlo we assume that we can first integrate over additional jets and obtain σ improved and then just normalize the Monte Carlo which puts back these jets in the collinear approximation.jets > 50 GeV 1 1 0. So how can we on the one hand compute higher order corrections to for example the Drell–Yan cross section and distributions and in addition consistently combine them with the parton shower? Such a combination would remove one of the historic shortcomings of parton shower Monte Carlos.

(2. for numerical applications and event simulation we need to implement this cancellation differently.189) Strictly speaking we have to include counter terms. we can extract a histogram from of the integral for O in Eq. and schematically obtain 1 O ∼ 0 dx O(x) O(0) . so x < xs < 1. including Born terms. When we compute next-to-leading order corrections to a hard process. x (2. However. When we compute for example a distribution of the energy of one of the heavy particles in the process.(2. the different contributions (now neglecting crossed channels) are dσ dx = B δ(x) dσ dx = αs − B +V 2 δ(x) dσ dx = αs R(x) .191) An observable computed beyond leading order includes contributions from real gluon emission and virtual gluon exchange. From factorization which we have shown via the universality of the leading splitting kernels we know that in the collinear and soft limits the real emission has to follow the Born matrix element x→0 lim R(x) = B . However.193) and obtain a normalized distribution. This factorization scale factor we will casually drop in the following.54) are a modification of |MB |2 .(2. The virtual corrections alone diverge.192) This kind of combination has a finite limit for → 0. is given by 1 O ≡ O B + O V + O R = µ−2 F dx 0 O(x) dσ x−2 dx B + dσ dx V + dσ dx R . this is not the issue we want to discuss. We immediately see that the integral over x in the real emission rate is logarithmically divergent in the soft limit. just like in Section 2.2.(2. There exist two methods to combine the virtual and real contributions to an observable and produce a finite physics result. The coupling constant αs ignores factors 2 and π or color factors. It describes simplified particle radiation off a hard process: the energy of the system before radiation is xs and the energy of the outgoing particle (call it photon or gluon) is x.190) B V R The constant B describes the Born process and the factorizing poles in the virtual contribution.3. and the individual parts of the integrand in Eq.193) are not integrable. − x1−2 2 (2. soft and collinear divergences. The problem is that we have to numerically integrate over x. These counter terms we add to the interference of Born and virtual gluon diagrams to remove the ultraviolet divergences. The expectation value of any infrared safe observable over the entire phase space. The first way historically introduced by the Dutch loop school for example to compute QCD corrections to top . which following Eq. but the expectation value including virtual and real gluon contributions after dimensional regularization is finite. Because we are interested in infrared divergences we choose n = 4 + 2 dimensions with > 0. similar to the collinear divergences we now know and love.3.193) The same way the renormalization and factorization scales appeared dimensional regularization now yields an additional factor 1/x−2 . ¯ (2. Infrared poles arise from two sources. |MZ q |2 ∝ GF αs . virtual corrections and real emission.6 NLO and parton shower 103 matrix-element-squared contributions |MB |2 ∝ GF 2Re M∗ MB ∝ GF αs V |MZg |2 ∝ GF αs |MZq |2 . If the observable is infrared safe it will have a smooth limit towards vanishing gluon energy O(x) → O(0). To avoid the complication of overlapping collinear and soft divergences we will follow a toy model by Bryan Webber. (2.

(2.193).190) (2.195) ds4 log 2 δ(s4 ) = ds4  max  s − ∆ − s4  µ µ 0 0 4 and similarly for log2 ∆.(2.104 2 QCD pair production is the numerically highly efficient phase space slicing: we divide the divergent phase space integral into a finite part and a pole. The fundamental shortcoming of phase space slicing is that it requires an analytical integration to produce sensible observable distributions.(2.190) .194) dx ∆ R(x)O(x) + αs V O(0) + O( ) . To avoid cancellations between integrals and replace them by cancellations among integrands we use a subtraction method to define integrable functions under the x integral in Eq.(2. This representation we can integrate along with the real emission phase space. The result will be a finite value for the next-to-leading order rate in the limit ∆ → 0 and exactly = 0. Starting from the individually divergent virtual and real contributions we first subtract and then add again a smartly chosen term 1 O R + O V = 0 dx αs R(x)O(x) + O x1−2 1 V 1 =αs B O(0) 0 1 dx dx 0 1 x1−2 1 x1−2 1 + 0 dx 1 αs R(x)O(x) αs BO(0) − x1−2 x1−2 dx R(x)O(x) − BO(0) + O x1−2 V + O V V =αs B O(0) =αs =αs 0 + αs 0 B O(0) + αs 2 1 dx 0 R(x)O(x) − BO(0) + O x dx R(x)O(x) − BO(0) + αs V O(0) x using Eq. If the eikonal approximation is given in terms of a Mandelstam variable δ(s4 ) and the cut-off has mass dimension two we can write   smax log 4 2 smax smax 4 4  ∆ 1 ∆ µ  log 2 = (2.196) In the subtracted real emission integral we take the limit → 0 because the asymptotic behavior of R(x → 0) makes the numerator vanish and hence regularizes this integral without any dimensional regularization required. so we can evaluate the integral at finite but small values of ∆. We end up with a perfectly finite x integral for the . The first integral precisely cancels the divergence from the virtual correction. by introducing a small parameter ∆. which acts like 1 O R + O V = 0 dx ∆ O(x) dσ x−2 dx 1 R + O V = 0 + ∆ dx αs ∆ R(x)O(x) + O x1−2 1 x1−2 dx 1 V =αs R(0) O(0) 0 dx 1 ∆ + αs ∆ dx R(x)O(x) + O x V V =αs B O(0) =αs =αs ∆2 + 2 R(x)O(x) + O x 1 B O(0) 2 log ∆ + O( 2 ) + 2 B O(0) log ∆ + 2 1 dx ∆ R(x)O(x) + αs V O(0) x using Eq. An amusing numerical trick is to re-write the explicit log ∆ contribution into a NLO-type phase space integral. x The two sources of log ∆ dependence have to cancel in the final expression.

provided we integrate over the additional parton’s phase space. 2. Obviously. there is no numerically small parameter in the expression 1 O = O B + O V + O R = B O(0) + αs V O(0) + αs 0 1 dx B x R(x) O(x) − B O(0) x + O(x) αs R(x) x . which in this special case we can omit because of the integration boundaries. but there is really no problem with them as long as any physical prediction we make after adding all leading oder and next-to-leading order contributions gives a positive cross section. For a hard additional parton or jet the cross section looks well defined and finite.197) = 0 dx O(0) B + αs V − αs This subtraction procedure is a standard method to compute next-to-leading order corrections involving one-loop virtual contributions and the emission of one additional parton.e.199) dz LO z dz NLO z The first term describes parton shower radiation from the Born diagram at order αs .190). The transfer functions we would have to include in the general form of Eq. On the one hand. This soft and collinear subtraction scheme for next-to-leading order calculations leads us to the first method of combining or matching next-to-leading order calculations with a parton shower. Negative events trigger negative reactions with experimentalists. Because we know they describe collinear jet radiation correctly such a modification will implement the emission probabilities using Sudakov factors.(2. we have an infrared divergence appearing after integrating over small but finite x → 0 from real emission. (2. and on the other hand we cancel it with an infrared divergence in the virtual corrections proportional to a Born-type momentum configuration δ(x). the Born-type kinematics would come with an explicit factor δ(x). possibly including experimental cuts dσ = dO 1 dx I(O)LO 0 B + αs V − αs B x + I(O)NLO αs R(x) x . i. Instead of the contribution from the virtual corrections contributing at δ(x) what we would rather want is a smeared virtual corrections pole which coincides with the justified collinear approximation and cancels the real emission over the entire low-x range.193) we integrate over the entire phase space of the additional parton.198) . We can re-write the same formula in a more appropriate way to compute distributions. We know what this spectrum has to look like for the collinear and hard kinematic configurations BP (z) dσ R(z) dσ = αs = αs . the radiated photon spectrum as a function of the (external) energy scale z. counter-events.(2.(2. (2. Once we integrate over the entire phase space this modified subtraction scheme has to give the same result as the next-to-leading oder rate. while the second term is the hard real emission defined in Eq. In terms of a histogram in x we encounter the real emission divergence at small x.6. and this divergence is cancelled by a negative delta distribution right at x = 0.6 NLO and parton shower 105 sum of all three contributions. even in the limit = 0. Note that we avoid the complicated proper two–dimensional description of Eq. The formula Eq.(2.(2. this will only give us a well-behaved distribution after integrating over at least small x.158) in favor of the simpler picture just in terms of particle energy fractions as introduced in the last section.e. We can write them as a function of the energy fraction z and find dP = αs P (z)/z dz. This contribution we can view as events with a negative weight. Let us consider the perturbatively critical but otherwise perfectly fine observable. (2.2. i.198) The transfer function I(O) is defined in a way that formally does precisely what we required by hand before: at leading order we evaluate I(O) using the Born kinematics x = 0 while for the real emission kinematics it allows for general x = 0 · · · 1. Smearing the integrated soft–collinear subtraction term using the splitting probabilities entering the parton shower means that the MC@NLO subtraction scheme has to be adjusted to the parton shower we use.197) is a little tricky: usually.2 MC@NLO method For example in Eq.

now written for general observables dσ = dO 1 dx I(O. we also knew we would fall short. Now. On the other hand. because our strategy included a smeared virtual subtraction term which for finite x should cancel the real emission.(2. 1) I(z. 1) 0 1 B + αs V − αs B x B x + I(z.201) All Born terms proportional to δ(z) vanish because their contributions would be unphysical. However.199) to turn the real emission term into a finite contribution R(x) − BP (x) R(x) −→ .(2.(2.202) x x Because this is an ad hoc subtraction term we also have to add it to the Born-type contribution. we can integrate over z and calculate the total cross section σtot with a cutoff zmin for consistency.e. Following the above argument the subtraction scheme implemented in the MC@NLO Monte Carlo describes hard emission just like a next-to-leading order calculation. xM ) = αs P (z) z P (z) θ(xM (x) − z) . This leads us to a modified version of Eq.203) When we use this form to compute the z spectrum to order αs it will include an integrated subtraction term contributing to the Born-type kinematics dσ −→ dz = 0 1 dx αs 0 1 BP (z) + R(z) BP (x) 2 − αs δ(x − z) + O(αs ) z x dx αs BP (z) + R(z) − BP (z) 2 + O(αs ) z (2.200) LO NLO = δ(z − x) + αs The second term in the real radiation transfer function arises because at the next order in perturbation theory the parton shower also acts on the real emission process. So we are still missing something.198). This includes the next-to-leading order normalization of the rate . without having done anything except for including a transfer function. z < xM .(2. It requires that enough energy to radiate a photon with an energy z has to be available. 1) 0 B + αs V − αs B αs BP (x) + x x + I(O. (2.201) includes an additional term which spoils the result: the same kind of jet radiation is included twice. This subtraction is not yet included. These transfer functions we can include in Eq. xM ) αs R(x) − BP (x) x . This is precisely the double counting which we avoided in the CKKW scheme. where xM is the energy available at the respective stage of showering. i. Factorization tells us how to write such a subtraction term using the splitting function P as defined in Eq. z This is exactly the distribution we expect. Eq. (2.198) dσ = dz = 0 1 1 dx I(z. now in the collinear limit. xM ) αs R(x) x R(x) x dx αs dx αs P (z) z B + αs V − αs + (δ(x − z) + O(αs )) αs = R(z) BP (z) 2 + αs + O(αs ) z z 0 BP (z) + R(z) 2 + O(αs ) = αs z (2. This already fulfills the first requirement for our scheme. z (2.106 2 QCD to arrive at this equation are I(z. once through the matrix element and once through the shower.204) = αs R(z) 2 + O(αs ) .

The main ingredient to this structure are Sudakov factors introduced in Section 2. It considers the next-to-leading order calculation of a cross section as a combination of an m-particle and an (m + 1)-particle process and attempts to adjust the parton shower attached to each of these two contributions such that there is no double counting. It starts from a given parton shower and avoids double counting by modifying the next-to-leading corrections. The first term is already suppressed by one power of αs .3 POWHEG method As described in Section 2. This means that we do cannot assume the form R/x for the real emission.190). Most importantly. so we can add a parton shower to it without much worry.6. For example. i. precisely like a parton shower. We can combine it with appropriate soft and collinear subtraction terms C in the (m + 1)-particle phase space where for simplicity we assume that the integrated subtraction terms exactly cancel the divergences from the virtual corrections: dσ = B dφm + αs − B +V 2 dφm + αs R dφm+1 after soft-collinear subtraction αs R (1 − P) dφ+1 B (2. it simulates additional collinear particle emissions using Sudakov factors.(2.205) we have to attach a parton shower.5. we can combine all m-particle contributions in one term dσ = [B + αs V + αs (R − C)Pdφ+1 ] dφm 1 + ≡ B dφm 1 + αs R 2 (1 − P) dφ+1 + O(αs ) B αs R 2 (1 − P) dφ+1 + O(αs ) B (2. Finally. To the three terms in Eq. However. An interesting question is: can we also combine a next-to-leading order calculation by keeping the next-to-leading order structure and apply a modified parton shower.6 NLO and parton shower 107 as well as the next-to-leading order distributions for the particles produced in the hard process. . As usually. but the setup is the same. so we denote the real emission as R alone.3. there is nothing which forces this combination of virtual and subtracted real emission to be positive. it avoids double counting between the first hard emission and the collinear jets. which means we have to re-organize the terms some more.206) where the combination B can only become negative if the regularized next-to-leading contribution over-compensates the Born term which would indicate a breakdown of perturbation theory. while for the production of heavy new particles their distributions hardly change at next-to-leading order.5.(2. The distribution of the radiated parton is included at leading order. The starting point of our argument is the next-to-leading order computation of a cross section following Eq. which do not feature in the next-to-leading order calculation. 2.1 and used for the CKKW merging scheme in Section 2. as we see in Eq. which means it describes the entire pT range of jet emission for the first and hardest radiated jet consistently.2 the matching scheme for a next-to-leading order correction and the parton shower is based on an extended subtraction scheme. For example for W +jets production such corrections to the W and leading jet distributions matter. In contrast to the MC@NLO scheme the POWHEG (Positive Weight Hardest Emission Generator) scheme does not introduce counter events or subtraction terms. Those additional jets. The projector P links the nominal (m + 1)-particle phase space of the real emission to the m-particle phase space in the Born-type kinematics.205) = B dφm + αs V dφm + αs (R − CP) dφm dφ+1 = [αs V + αs (R − C)Pdφ+1 ] dφm + B dφm 1 + One change to the earlier formula is that we replace the toy phase space variable by the actual m-particle and (m + 1)-particle phase space integrations.(2. including their ratio φ+1 .6. what looked to easy in our toy example is much harder in QCD reality.204).2. in the collinear approximation.e. are added through the parton shower.

The introduction of B is nothing but a re-weighting factor for the events contributing to the m-particle configuration which we need to maintain the next-to-leading order normalization of the combined m-particle and (m + 1)-particle rates. t . From the next-to-leading order T normalization we can compute the splitting probability entering the Sudakov factor from the perturbative series as ∆(pmatch ) = exp − T dφ+1 αs R θ(pT − pmatch ) T B . The second factor αs R/B is essentially the PYTHIA or HERWIG matrix element correction used for an improved simulation for example W +jet events.166) with the modification of an explicit veto. For the (m + 1)-particle contribution we have to ensure that the hardest jet radiation is given by the matrix element R. Using 1 = θ(g) + (1 − θ(g)) we find Eq. z) can be used to for example forbid hard jet radiation. which means we need to attach a Sudakov factor to veto additional jet radiation dσ = B dφm ∆ + αs R 2 (1 − P) dφ+1 + O(αs ) .(2. t) = ∆(t. but now with the additional requirement that hard jet radiation is vetoed. z) − pmatch T 2π .210) The first Sudakov factor is not vetoed which means it is evaluated at pmatch = 0.(2. The bottom line of this formula is that we can consistently write down vetoed Sudakov probabilities and build a parton shower out of them. The proof of this formula starts from Eq. (2. B (2.209) where we again show the diagonal case to simplify the notation. pmatch ) T 2π αs ˆ P (z) θ pT (t .206) would be B and not B. B (2.209) more or less straight away.(2. The Sudakov survival probabilities have to include a veto condition in the splitting to avoid that the shower populates the hard radiation phase space with pT > pmatch . t0 .166) including the veto condition in the splitting kernel t f (x. pmatch )f (x.211) The reason is that in Eq. (2. This gives us for the combined next-to-leading order exclusive contributions dσ = B dφm ∆(0) + ∆(pmatch ) T αs R 2 dφ+1 + O(αs ) .(2.108 2 QCD From CKKW jet merging we know what we have to ensure when we combine a m-particle process with an (m + 1)-particle process: the m-particle process has to be exclusive.(2. If for the (m + 1)-particle contribution T the vetoed shower should avoid double counting pmatch has to be chosen event by event.(2.(2.158) by including a step function θ(g) which dependent on some kinematic condition for g < 0 can become zero and this way veto a splitting contributing to the survival probability t ∆(t. Such a vetoed shower is easily defined starting from the definition of T the (diagonal) Sudakov survival probability Eq. pmatch ) = exp − T t0 t dt t dt t 1 dz 0 1 αs ˆ P (z) θ g(t .206) we have to replace (1 − P) with a probability measure.212) .208) ∼ exp − t0 dz 0 In the second line we indicate how the function g(t . For these modified Sudakov factors we need to show that they obey a DGLAP equation like Eq. z. t0 ) + T t0 dt ∆(t. (2. In perturbative QCD we can show that provided the include αs R/B as the splitting probability we indeed find ∆(0) + ∆(pmatch ) T αs R dφ+1 = 1 − B dφ+1 αs R + B αs R 2 2 dφ+1 + O(αs ) = 1 + O(αs ) . This term projects the real emission on the actual (m + 1)-particle phase space. pmatch ) T t 1 0 x dz αs ˆ P (z) θ (g) f .t z 2π z . B (2.210) the term in brackets integrated over the entire phase space has to give unity. In the second term of Eq. The appearance of these modified Sudakov factors indicates that in contrast to the MC@NLO method we now will modify the structure of the parton shower which we combine with the higher order matrix elements.207) In the CKKW scheme the factor in the front of the brackets in Eq.

defining a truncated shower. namely Eq. . The first emission of the parton shower might well appear at large angles but with small energy. When for some reason we are mainly interested in one hard jet we can use MC@NLO or POWHEG and benefit from the next-to-leading order normalization.6 NLO and parton shower MC@NLO/POWHEG matching first jet correct all jets correct. The condition to treat soft emission separately we know from CKKW merging. for example in the MENLOPS package. The main features and shortcomings of the matching and merging approaches we summarize in Table 3.213) A technical detail which we have not mentioned yet is that the POWHEG scheme assumes that for our Sudakov factors can be ordered in such a way that the hardest emission always occurs first.2. For the POWHEG shower this soft radiation has to be removed or moved to a lower place in the ordering of the splittings. which means it will not be particularly hard. In the POWHEG scheme the events get re-weighted according to standard building blocks of a next-to-leading order calculation. If I were a string theorist I would definitely call this achievement a revolution or even two.2-2.6. ordered by angle. but the shower has to be adapted.(2. just like the old-fashioned parton shower Monte Carlo we need to include the normalization of the rate by hand.5. MadEvent. and it is by far the fastest method. However. In contrast to a lot of other progress in theoretical physics jet merging solves a problem which would otherwise have limited our ability to understand LHC data.5. if there is a well defined and sufficiently hard scale in the process. like 1917 but with the trombones and cannons of Tchaikovsky’s 1812.186): the scale at which we resolve a parton splitting does not have to identical with the lower boundary of simulated splittings. Following the discussion in Section 2. We can construct a parton shower taking into account such splitting kernels. but the shower stays the same.6.3 and Sections 2. First of all. the old-fashioned Monte Carlo with a tuned parton shower will be fine. tuned correct to LO plus real emission MLM–Alpgen. In cases where we really need a large number of jets correctly described we will end up with CKKW or MLM-type simulation. I am not getting tired of emphasizing that the conceptual progress in QCD describing jet radiation for all transverse momenta is absolutely crucial for LHC analyses. tuned correct to NLO CKKW merging/SHERPA all jets correct all jets correct. d∆(pmatch ) T = ∆(pmatch ) T dpmatch T d∆(pmatch ) T = ∆(pmatch ) T ∆max = log ∆min − log ∆min = dpmatch T dφ+1 dφ+1 dφ+1 dφ+1 αs R δ(pT − pmatch ) T B αs R δ(pT − pmatch ) = T B dφ+1 αs R B αs R B αs R B (2.3 we have introduced different ways to simulate jet radiation at the LHC. no matter what kind of Higgs or new physics we are looking for.... 109 hard jets collinear jets normalization variants Table 3: Comparison of the MC@NLO and CKKW schemes combining collinear and hard jets. This is the case for example when a gluon splits into two bottoms in the initial state and we are interested in the radiated bottom jet and its kinematics. In Sections 2. Or we are lucky and combined version of CKKW and POWHEG is available. At this stage it is up to the competent user to pick the scheme which describes their analysis best. This modified shower is the big difference between the MC@NLO scheme and the POWHEG scheme to combine next-to-leading order corrections with a parton shower. In the MC@NLO scheme we modify the next-to-leading order correction for a give shower.2 we expect any collinear transverse momentum ordering to be disrupted by soft radiation.

There is no point in giving a list of original references. Rick Field works it all out [29]. James Stirling. – aimed more at perturbative QCD at the LHC is the QCD primer by John Campbell. – even though they are just hand written and do not include a lot of text it might be useful to have a look at Michael Spira’s QCD lecture notes [39] to view some of the topics from a different angle. – the only review on leading order jet merging is by Michelangelo Mangano and Tim Stelzer [34]. – a little more phenomenology you can find in G¨ nther Dissertori. you will have wait for the review article which Frank Krauss and Peter Richardson promised to write for Journal of Physics G. If you want to know more. Paolo Nason. I still use his appendices for Feynman rules because I have not seen another book with as few (if not zero) typos [26]. – for those of you who are now hooked on QCD and jet physics at hadron colliders there are two comprehensive reviews by Steve Ellis [30] and by Gavin Salam [31]. namely the resummation of logarithms in QCD and the physical meaning of these logarithms. and Bryan Webber. Joey Huston. . in particular when a calculation did not affect the main topic. Ian Knowles and Michael Schmelling’s book u on QCD and phenomenology [28]. Some corners I had to cut. – if you would like to learn how to for example compute higher order cross sections to Drell–Yan production. I borrowed some of the discussions in the QCD section from this them. Different processes you can find discussed in detail in a later paper by Stefano Frixione. and James Stirling [6]. but maybe a little more modern is the Standard Model primer by Cliff Burgess and Guy Moore [27]. At the end of it you will find more literature. Both of them do not exactly use my kind of notations and are comparably formal. The original CKKW paper beautifully explains the general idea for final state radiation. and I followed their analysis [24]. For other approaches there is a very concise discussion included with the comparison of the different models [35]. – similar. there are are Dave Soper’s [32] and George Sterman’s [33] notes. – the POWHEG method is really nicely described in the original paper by Paolo Nason [37]. It includes everything you ever wanted to know about QCD [23] and more. but I will to list a few books and review articles which should come in handy if you would like to know more: – I started learning high–energy theory including QCD from Otto Nachmann’s book. Again.7 Further reading Just like the Higgs section the QCD part is something in between a text book chapter and a review of QCD as it will be relevant to LHC searches. This QCD section essentially follows its Chapter 5.110 2 QCD 2. and Carlo Oleari [38]. – to understand MC@NLO there is nothing like the original papers by Bryan Webber and Stefano Frixione [36]. – coming to the usual brilliant TASI lectures. – the best source to learn QCD at colliders is the pink book by Keith Ellis. but they are a great read if you know something about QCD already.

For example. the helicity method.2. in Section 2. Then. we turn to a very active field of current research. They are not experimental physics either.2 we have seen that the strong coupling explodes at low energy scales ΛQCD which means that something has to happen with quarks and gluons on their way through the detectors. Can we try to reconstruct the original partons? The answer is yes. to experimentally observed QCD objects. Wikipedia can of course tell you more about this. 1. phenomenology is a school based on Edmund Husserl. hopefully triggered by LHC measurements challenging theorist and their simulations. Most professional tool computing LHC cross sections or simulating LHC events use this method instead of squaring amplitudes analytically. because they go beyond understanding the direct results of the LHC detectors.3. the gluon and all quarks except for the top quark hadronize before they decay and form bunches of baryons and mesons which in turn decay in many stages.2 we introduce a particularly efficient way of computing transition amplitudes from Feynman rules. In the public arXiv version more sections will follow. It goes beyond what is usually found in theoretical physics text books and is usually referred to as ‘phenomenology’.1 we fill in some blanks from Section 1.5. in Section 2 we use the terms parton and jet synonymously.62) we apply wave function renormalization factors to them. Originally. both with a bias towards aspects relevant to the LHC. Sometimes. but as part of the LHC program we hope to find dark matter candidates that way. The third section will reflect much better what many theorists working on LHC topics need to know. We know from Section 2 that in QCD we can produce an arbitrary number of hard jets in a hard matrix element and another arbitrary number via soft or collinear 1 The term ‘phenomenology’ is borrowed from philosophy where it means exactly the opposite from what it means in physics. they lie in between the two fields and need to be well understood to allow theorists and experimentalists to interact with each other.6.. In Section 3. These particles carry a lot of energy. Indeed. Instead.4. From these lecture notes it is clear that LHC physics is field theory. in the sense that we can combine a large number of proto-jets into smaller numbers. hadronized and decayed to stable particles. Finally. Imagine we observe a large number of energy depositions in the ATLAS or CMS calorimeter which we would like to combine into jets. even though they cannot exist individually because QCD is asymptotically free. Relativistic kinematics then tells us that these baryons and mesons are strongly boosted together to form jets.1 Jets and fat jets Throughout Section 2 we pretend that quarks and gluons produced at the LHC are actually what we observe in the LHC detectors. In Section 3. Section 3. In addition. partons. physicist phenomenologists are theorists who really care about measurements.3 discussed how to reconstruct particles which interact too weakly to be actually observed in LHC detectors. where unfortunately nothing tells us what the final number of jets should be. or — not covered in these notes — physics models describing the symmetry structure of the extension of our Standard Model at the TeV scale. Consequently. in Eq. We first discuss jets and how to link the asymptotic final states of QCD amplitudes. so-called fat jets. 1 This terms indicates that these topics are not really theoretical physics in the sense that they rely on for example field theory tool box. In the Standard Model those would be neutrinos. Those jets we measure at hadron colliders and link to the partons produced in the hard interaction. both introductions really are theoretical physics. Unlike other fields of philosophy their main interest are not observations but the actual nature of things. essentially assuming that each parton at the LHC turns into a jet and that the measured jet four-momentum can be linked to the parton four-momentum. jets. The way we usually define jets are so-called recombination algorithms including for example the Cambridge–Aachen or (anti-) kT algorithms.4 we very briefly discuss LHC uncertainties from a theory point of view. either electroweak symmetry breaking. We know that they come from a small number of partons which originate in the hard QCD process and which since have undergone a sizeable number of splittings. but with few exceptions do not justify the title of the lecture notes. On the other hand. in Section 3. 3.(2. and 2. phenomenology has the reputation of not being proper theoretical physics.3. In perturbative QCD those particles are the final states. QCD. around the typical energy scales O(mH ) we are interested in at the LHC. This third chapter supplements the pure theory aspects and links them to experimental issues of the ATLAS and CMS experiments.111 3 LHC Phenomenology The two first parts of these lecture notes focus on Higgs physics and on QCD. Doing exactly the opposite. ..

go back to (1) (2c) if y min > ycut keep all proto-jets. To decide if two proto-jets have arisen from one parton leaving the hard process we have to define a collinearity measure. the jet mass. purely geometric (Cambridge–Aachen C/A) or hard constituents (anti-kT ) to form a jet.35) and on the other hand the transverse momentum of one proto-jet with respect to another or to the beam axis.j D yij = yiB = pT. which means the radiation of an additional soft parton cannot affect the global structure of the reconstructed jets. where ycut is the scale we give to the algorithm. Because of the power dependence on the transverse momenta the three algorithms start with soft constituents (kT ). All of them have in common that they link physical objects. and you can find examples in QCD lectures by Mike Seymour. 3. Moreover.j yiB = 1 yiB = p−1 .(3.j ) D ∆Rij yij = D ∆Rij yij = min p−1 . namely calorimeter towers.1) The parameter D is a measure of the angular size of the jet and can range anywhere between 0. This measure will on the one hand include the distance in R space as introduced in Eq. Therefore. The three-momentum vectors we simply add ki + kj → ki . whatever the measure for this resolution might be. The technical question is what ‘combine jets’ means in terms of the four-momentum of the new jet.1. . we can give the algorithm the minimum number of physical jets and stop there. go back to (1) (2b) if y min = yiB < ycut remove proto-jet i. yiB ) (2a) if y min = yij < ycut merge proto-jets i and j and their momenta. following a series of actual physical intermediate splittings. For other algorithms we have to ensure this property explicitly. pT. which is inspired by the massless parton we are usually looking for.1 Jet algorithms The basic idea of recombination algorithms is to ask if a given jet has a soft or collinear partner.1) we have all the freedom in the world to weight the angular and transverse momentum distances relative to each other.i .4 and 1. A transverse momentum distance between jets is therefore best suited to combine the correct proto-jets into the original parton from the hard interaction. done The result of the algorithm will of course depend on the resolution ycut . this is not clear at all for the anti-kT algorithm. As we can see from the different choices in Eq. to other more or less physical objects. In the coming section we will therefore discuss what criteria for a proto-jet recombination to an assumed physical jets exist. While for the kT and the C/A algorithms it is fairly clear that the intermediate steps have a physical interpretation. this transverse momentum measure is intrinsically infrared safe. call it beam radiation. we need to tell the jet algorithm either how many jets it should arrive at or what should be the resolution of the smallest proto-jets we consider partons.5. p−1 T. namely partons from the hard process. or the top quark. Alternatively. the Higgs boson.(2.j (3.112 3 LHC PHENOMENOLOGY radiation. weighted by the relative power of the two ingredients kT C/A anti-kT ∆Rij min (pT. For the zero component we can assume that the new physical jet have zero invariant mass. Explicit measures are. This follows from the physics arguments in Section 2 from which we know that partons produced in a hard process preferably turn into collinear pairs of partons as approximately described by the parton shower. As we will see in the next section this allows us to extend the concept of jet algorithms to massive particles like a W or Z boson. The jet algorithm is (1) for all combinations of two proto-jets in the event find the minimum y min = minij (yij . T.i T. If instead we add the four-momenta we can compute the invariant mass of the jet constituents. From Section 2 and the derivation of the collinear splitting kernels it is obvious why theorists working on perturbative QCD often prefer the kT algorithm: we know that the showering probability or the collinear splitting probability is best described in terms of virtuality or transverse momentum. We define two proto-jets as coming from one jet if yij < ycut .

A comparison of different jet algorithms can be found in the original paper on associated Higgs and gauge boson production.g. Based on the kT algorithm they look for structures in the chain of y values introduced in Eq. For such an analysis of y values it is crucial that the intermediate steps of the jet algorithm have a physics interpretation. choosing a large geometric size R = 1. Such noise is easiest understood geometrically in a probabilistic picture.1 Jets and fat jets 113 The problem of the kT algorithm arises with the pile-up or underlying event. Historically.H in terms of the transverse momentum of the boosted Higgs and the momentum fractions of the two bottom jets. each subjet ji that is kept we further decompose recursively until it reaches some minimum value. Such jets involving heavy particles and (usually) a large geometrical size are referred to as fat jets. For Higgs and top decays the situation is significantly more promising. like a heavy Higgs boson. which define the kinematics of each jet. This fat jet we then uncluster again. the question arises if we can look for other heavy objects inside a jet. the goal of such a Higgs search is a distribution of the invariant mass of the bottom quarks which gives us a signal peak and side bins to estimate the background. dependent on our physics model heavy objects like W bosons or top quarks might be boosted enough to fit into a regular jet of angular size R 0. Basically. i. a jet algorithm might include hadronic decay products which we would not trust to include in a regular mass reconstruction based on reconstructed detector objects. How much energy deposit we have to subtract from a reconstructed jet depends on the area the jet covers in the detector. (3. Making use b 2 of the scalar nature of the Higgs boson we can add an additional requirement on the balance min(p2 j1 . applying jet algorithms with very large R .1.(3. i. searching for a mass drop in jet mass indicating the decay of the massive Higgs to two essentially massless quarks. More recent fat jet algorithms looking for not too highly boosted heavy particles are based on the C/A algorithm which appears to be best suited to extract massive splittings inside the jet clustering history.1). Three main motivations lead us into the direction of fat jets: first. The C/A and anti-kT algorithms are more compact and easier to interpret experimentally. Starting with the Higgs tagger we search for jets which include two bottom quarks coming from a massive Higgs boson with mH 120 GeV. for example searching for top quarks with a transverse momentum in excess of 1 TeV is unlikely to appear in the Standard Model and will only become interesting if we encounter very heavy resonances decaying to a pair of top quarks. At the LHC the experimental situation pT /m 1 we are guaranteed to encounter for electroweak gauge bosons.7. If the mass drop criterion mj1 > 0. This way we obtain a splitting pattern which should only include massive splittings and which for the Higgs jet uniquely identifies the H → b¯ decay. the first fat jet searches were designed to look for strongly interacting W bosons.e. j2 ordered such that mj1 > mj2 . p2 j2 )∆Rj1 j2 . This is why we focus on the moderate scenario.2 Fat jets Starting from the way the experiments at the Tevatron and the LHC search for bottom jets.8 mj is fulfilled. we run the C/A algorithm over the event. including several detailed requirements on the content of such jets. Using a C/A algorithm we can search for hadronically decaying boosted Z bosons. Secondly. Higgs bosons. The more extreme case of pT m. so we subtract it from each event. we discard j2 and keep j1 . T T Of course. very soft QCD activity entering the detectors undirectionally. However. Experimentally. giving two subjets j1 . Therefore. The problem is that for those we only have one hard criterion based on which we can reject QCD backgrounds: the mass of the Z resonance.e.2) b pT. Adding a second Z boson and possibly the mass of a resonance decaying to these two Z bosons. it is a major step that even for the kT algorithm we can indeed compute an IR–safe geometric jet size. First.H z(1 − z) pT. otherwise we keep both j1 and j2 . 3. adds the necessary QCD rejection. j2 is soft enough to come from the underlying event or soft or collinear QCD emission.2 estimated to cover 1 mH 2mH Rb¯ ∼ > . and top quarks. The iterative unclustering we start by undoing the last clustering of the jet j. e. mji < 30 GeV ensuring it does not involve heavy states.3. the low-energy jet activity is constant all over the detector. even if only a fraction of the heavy particles we are searching for are sufficiently boosted such an algorithm automatically resolves signal combinatorics known to limit LHC analyses. all actual numbers in this selection are object to experimental scrutiny and can only be determined after testing the algorithm on LHC data. And finally.

e. For a Standard Model b ¯ Higgs boson the production mechanisms are pp → W H/ZH (left) and pp → ttH (center).Events / 60 40 20 114 100 120 140 160 180 20 0 0 20 40 60 80 100 120 140 160 180 20 3 Mass (GeV) LHC PHENOMENOLOGY Mass (GeV) Events / 8GeV / 30fb-1 qq V+jets VV V+Higgs 140 120 100 80 60 40 (d) S/ B = 4. Instead. so we usually add two bottom tags on the constituents which according to detector simulations can be very efficient.5 in 112-128GeV qq V+jets VV V+Higgs 1 0. Next. The invariant mass we include in the histogram is the mass of the three hardest constituents. in ttH production. This filtering significantly reduces the y-φ surface area of the relevant constituents and b thereby the ??? by the underlying event and pile-up. but we would like to know their invariant mass.4 0. which might even allows us to run a Higgs tagger over any kind of event sample and see if we find any Higgs bosons for example in new physics decays. the mass of the b¯ pair.2 dσ/dmbb [fb/5 GeV] ¯ ¯ tth ¯ ttz ¯b ttb¯ 20 100 120 140 160 180 20 0 0 20 40 60 80 100 120 140 160 180 20 0 Mass (GeV) Mass (GeV) 30 60 90 120 150 mbb [GeV] ¯ 180 210 Figure 15: Invariant mass distributions for Higgs searches using fat jets in H → b¯ decays. Such a reconstruction is possible in a reasonably clean sample without too many QCD jets. From the above discussion we see that Higgs taggers rely only on one kinematic criterion.8 0. Instead of these two distinct steps we can also apply a two-dimensional condition on the kinematics of the three top decay products which avoids assigning the two W decay jets in cases where two combinations of decay jets give similar invariant masses. The two combined add to a QCD rejection of at least 10−4 . In b terms of background rejection this is not much. When we use top taggers to look for new particles decaying to two top quarks we are not only interested in finding boosted top quarks.2 (∆R12 )4 — the original Jade distance includes is given by pT. To avoid shaping the background side bins we can include several (filtered) subjet combinations. the two bottom quarks and possibly a radiated gluon. On the simulation level both methods give equivalent results. provided the top quarks have large enough energy to boost all three decay jets into a small enough cone. i. This means reconstructing their direction and their energy. ordered in the modified Jade distance pT.3. we apply a filtering procedure looking at the same constituent with a higher resolution which can for example be Rfilt = min(0. Rb¯/2).6 0. Applying these three kinematic conditions gives a QCD rejection of a few per-cent. the entire fat jet has to reproduce the top mass. and light supersymmetric Higgs searches in the decays of squarks and gluinos.2. Examples for the invariant mass distributions for different Higgs search channels we show in Figure 15. Therefore. Starting from a C/A jet of size up to R = 1. we cannot simply use the mass of a set of constituents of the fat jet. In a busy QCD environment another problem arises: errand jets from initial state radiation or other particles in the final state enter the fat jet algorithm and give us several mass drops in the fat jet history. size makes us increasingly vulnerable to underlying event. or even regular initial state radiation as described in Section 2.3.5 we again search for mass drops. They include Standard Model Higgs ¯ searches in W H/ZH production. Looking for massive particles decaying to quarks top taggers are next in line. .1 pT.2 (∆R12 )2 . looking for other boosted heavy particles is usually motivated by new physics scenarios. [40] and [41] (left to right). we require a mass drop corresponding to the W decay and effectively constrain a second combination of two decay jets evaluating the helicity angle of the left handed W decay. In cascade decays of supersymmetric squarks and gluinos we can apply the same search for the light Higgs boson (right). Figures from Ref [7]. After appropriate filtering we apply these two mass windows conditions. If this should not be sufficient for a given analysis we can increase the rejection rate by requiring a bottom tag which as a bonus also tells us which of the three top decay jets should reconstruct the W mass. While fat jet Higgs searches are targeted mostly at the Standard Model. pile-up.1 pT. this time corresponding to the top and W masses.

1 )/ ).-1.-1.1 CALL OXXXXX(P(0.NHEL(2 ).ZERO . even event shapes without any jet algorithm can search for such signals.3 )) ). so we do not have to be careful with the different spinors u and v.AMP(1 1) . u (3. while matx is a real positive number with a fixed mass dimension depending on the number of external particles. 3. interaction vertices and propagators and squaring the amplitude analytically. NCOLOR PARAMETER (NWAVEFUNCS= 5. u The entries in the external spinors are given by the spin of the massless fermions obeying the Dirac equation.1 CALL IOVXXX(W(1. polarization vectors.NEXTERNAL=4) INTEGER NWAVEFUNCS .5 ).inc’ DATA Denom(1 DATA (CF(i. currents of all kinds.W(1.W(1. For each value of µ = 0 · · · 3 this object is a complex number.NHEL(3 ).1 we compute the cross section for Z production by writing down all external spinors.2 )) ). µ are massless fermions.1 ) / 3/ CALL IXXXXX(P(0.1 )) ).4 CALL JIOXXX(W(1.W(1. Instead. NCOLOR ).5 )) )) .GAL .ZERO .W(1. As an example.ZERO . computed from the four component of each spinor and the respective 4 × 4 gamma matrix γ µ shown in Eq.W(1. Summing over µ and ν in both currents forms the matrix element. The intermediate photon propagator has the form gµν /s. we use powerful tools like Form to 2 compute traces of Dirac matrices in the calculation of |M| . which is a real number for each value of µ = ν.NEIGEN= 1. For our toy process Eq.3 CALL OXXXXX(P(0. This approach we illustrate for muon pair production based on the MadGraph/Helas implementation. For the LHC nobody does gamma matrix traces by hand anymore.ZERO ).(3. Once the top quarks have very large transverse momenta the two-step mass drop criterion becomes less critical because the three decay jets are too close to be cleanly resolved. Instead of squaring this amplitude symbolically we can follow the steps described above and compute an array of numbers for different spin combinations numerically.ZERO . In this section we illustrate that we can do even better. we consider our usual toy process u¯ → γ ∗ → µ+ µ− .i=1 1/ .2 ).2 Helicity amplitudes In Section 2. Summing over the internal Dirac indices we compute the matrix element and summing over external fermion spin directions we compute the matrix element squared.+1. The amplitude itself inherits external indices for example from the polarization vectors.3) the MadGraph output reads REAL*8 FUNCTION UUB_MUPMUM(P.W(1.+1. MadGraph as part of MadEvent is a tool to compute matrix elements this way.W(1. just inserting the correct values for each component of each spinor.4 )) ).102). To square this matrix element we need to sum M∗ × M over all possible spin directions of the external fermions.4 ). In this situation analyses based on the kT or anti-kT algorithms can be as promising. NEXTERNAL PARAMETER (NGRAPHS= 1.NHEL(4 ).3) The structure of the amplitude M with two internal Dirac indices µ and ν involves one vector current on each side (¯f γµ uf ) where f = u. external polarization vectors.W(1.NHEL(1 ).2 CALL IXXXXX(P(0. NEIGEN.GAU . NCOLOR= 1) REAL*8 P(0:3.(2.3 JAMP( 1) = +AMP( DO I = 1. etc.NEXTERNAL) INTEGER NHEL(NEXTERNAL) INCLUDE ’coupl.NHEL) C C FUNCTION GENERATED BY MADGRAPH C RETURNS AMPLITUDE SQUARED SUMMED/AVG OVER COLORS C FOR PROCESS : u u˜ -> mu+ muC INTEGER NGRAPHS.3.2 Helicity amplitudes 115 While it appears that the C/A jet algorithm with its purely geometric selection has the best potential to search for massive particles in its jet history there exists a multitude of algorithms searching for boosted top quarks.ZERO . It automatically produces a Fortran routine which calls functions from the Helas library to compute spinors.

Its first two entries are the left-chiral part of the fermionic spinor. As two more arguments Jio includes the four-momentum flowing through the gauge boson propagator. so we numerically compute T Fo J g(1) / 1 − γ5 1 2 + g(2) 1 + γ5 1 2 Fi (3.116 3 LHC PHENOMENOLOGY DO J = 1. The fermion wave function output is a complex array F (1 : 6). Because for massless particles helicity and chirality are identical our quarks and leptons will only have finite entries F (1 : 2) for nhel = −1 and F (3 : 4) for nhel = +1. g. m. The four entries correspond to the size of one γ matrix. 3 Jio (5) = −Fi (5) + Fo (5) ∼ −pi (0) + po (0) + i (−pi (3) − po (3)) Jio (6) = −Fi (6) + Fo (6) ∼ −pi (1) + po (1) + i (−pi (2) + po (2)) . These four numbers can directly be computed from the four-momentum if we know the helicity. m. Fo . Moreover. so we can compute the trace of the chain of gamma matrices. F ) does the same for a fermion with outgoing fermion flow. Γ. and similarly for the spinor v . because by convention MadGraph defines its particles as u and µ− . Again. nsf . The left-chiral and right-chiral components now read F (1 : 2) = u(1 − γ5 )/2 and ¯ 1 F (3 : 4) = u(1 + γ5 )/2. nsf = +1 marks the incoming fermion u and nsf = −1 for the outgoing anti-fermion µ+ . Remember that helicity and chirality are identical only for massless fermions because chirality is defined as the eigenvalue of the projectors (1 ± γ5 )/2. As input it requires the four-momentum. (3. i. V ) computes the amplitude of a fermion–fermion–vector coupling using the two external fermionic spinors Fi and Fo and an incoming vector current J which in our case comes from JIOXXX. For each point in phase space and each helicity combination the MadGraph subroutine SUUB MUPMUM computes the matrix element using standard Helas routines. ImJio (6). Jio ) computes the (off-shell) current for the vector boson attached to the two external fermions Fi and Fo .6) . the mass and the helicity of this fermion. They allow us to reconstruct q µ based on the information from the last to entries of the external fermions’ wave functions q µ = (ReJio (5). (3. nsf . The entries F (3 : 4) 1 1 are the right-chiral spinor. The entries of nhel will be either +1 or −1. nhel . the coupling g(1 : 2) is a complex array. · OXXXXX(p. The last two entries of F contain the four-momentum in the direction of the fermion flow. namely F (5) = nsf (p(0) + ip(3)) and F (6) = nsf (p(1) + ip(2)). F ) computes the wave function of a fermion with incoming fermion number. F (1 : 2) = (1 − γ5 )/2 u or F (1 : 2) = (1 − γ5 )/2 v for nsf = ±1. J. The output array Jio again has six components which for the photon with momentum q are Jio (µ + 1) = − 1 − γ5 1 1 + γ5 1 i T µ F γ g(1) + g(2) q2 o 2 2 Fi µ = 0. · JIOXXX(Fi . NCOLOR ZTEMP = ZTEMP + CF(J. 2. g.4) The first four entries in Jio correspond to the index µ or the dimensionality of the Dirac matrices in this vector T current. our incoming u and ¯ our outgoing µ− . while helicity is defined as the projection of the spin onto the 1 momentum direction. i. as the left or right handedness. For a general Breit-Wigner propagator we need to know the mass m and the width Γ of the intermediate vector boson. ImJio (5)) . The last two entries are F (5) = nsf (p(0) + ip(3)) and ¯ 1 ¯ F (6) = nsf (p(1) + ip(2)). The spinor index is contracted between Fo and Fi . nhel .e. so either an incoming fermion or an outgoing anti-fermion.e.I)*JAMP(J) ENDDO UUB_MUPMUM =UUB_MUPMUM+ZTEMP*DCONJG(JAMP(I))/DENOM(I) ENDDO END The input to this function are the external four-momenta p(0 : 3. m. · IXXXXX(p.5) · IOVXXX(Fi .e. Fo . The coupling g(1 : 2) is a complex array with the interaction of the left-chiral and right-chiral fermion in the upper and lower index. 1 : 4) and the helicities of all fermions nhel (1 : 4) in the process. ReJio (6). i. 1.

MadGraph calls IOVXXX with Fi = W (∗. -1. 1 : 4). 1).IHEL=1. which in our simple case means one color structure with a color factor Nc = 3. 1. -1.NCOMB).3. not an array. -1/ 1/ -1/ 1/ -1/ 1/ -1/ 1/ -1/ . even if they appear several times in the different Feynman diagrams. (NHEL(IHEL. 5). combined with the photon current J = W (∗. Since this number gives the final amplitude. -1. -1. it u always calls wave functions for all external particles and puts them into the array W (1 : 6.4) 5). 3) correspond to Fi (u) and Fi (µ+ ). 1. Its output is the photon current Jio ≡ W (∗.ANS) C C C C C C C FUNCTION GENERATED BY MADGRAPH RETURNS AMPLITUDE SQUARED SUMMED/AVG OVER COLORS AND HELICITIES FOR THE POINT IN PHASE SPACE P(0:3. -1. They are transferred through common blocks in the file coupl. 2) JIOXXX computes the vector current for the massless photon in the s channel. Momentum conservation is not enforced by IOVXXX. Given the wave functions Fi = W (∗.4) 7).inc and computed elsewhere. 1.4) / / / / / / / / / -1. except for the arguments m and Γ in the JIOXXX call.IHEL=1. -1. -1. PARAMETER (NEXTERNAL=4.4) 9). 2) and W (∗.IC(NEXTERNAL) INTEGER IPROC. -1. -1. The last step combines this current with the two outgoing muons coupling to the photon. NCOMB. To obtain the full transition amplitude numerically we correspondingly sum over all helicity combinations of the external fermions.FALSE. u The first vertex we evaluate is the incoming quark-photon vertex. 1) and Fo = W (∗. (NHEL(IHEL. -1. 1.IHEL=1.4) 3). -1. it should return a complex number. The only remaining sum left to compute before we square JAMP is the color structure. so we have to take care of it by hand. because in the helicity amplitude approach it is easy to accommodate complicated tensors. (NHEL(IHEL.4) 2). As an added bonus MadGraph produces a file with all Feynman diagrams in which the numbering of the external particles corresponds to the second argument of W and the numbering of the Feynman diagrams corresponds to the argument of AMP. -1. First.NTRY REAL*8 T. As mentioned above. Not much changes if we instead choose a massive Z boson. The result AMP is copied into JAMP without an additional sign which can have come from the relative ordering of external fermions in different Feynman diagrams contributing to the same process. Given the list above it is easy to recapitulate how MadGraph computes the amplitude for u¯ → γ ∗ → µ+ µ− .2 Helicity amplitudes 117 We see that all spinor and Dirac indices of the three input arguments are contracted in the final result.IHEL=1. (NHEL(IHEL.4) 8). When we square M symbolically we need to sum over the spins of the outgoing states to transform a spinor product of the kind u¯ into the residue or numerator of u a fermion propagator. (NHEL(IHEL. UUB_MUPMUM INTEGER IHEL. -1.IHEL=1. -1. (NHEL(IHEL. to calculate the transition amplitude MadGraph requires all masses and couplings.IHEL=1. while W (∗. 1. 1.IHEL=1. The function UUB MUPMUM described above is not yet the full story. 4). 1) and W (∗.IDEN. The vectors W (∗.4) 4).4) 6). SUBROUTINE SUUB_MUPMUM(P1. MadGraph uses unitary gauge for massive vector bosons. (NHEL(IHEL. -1. 5). -1.JC(NEXTERNAL) LOGICAL GOODHEL(NCOMB) DATA GOODHEL/THEL*. in exchange for a large number of Feynman diagrams. NCOMB= 16) INTEGER THEL PARAMETER (THEL=NCOMB*1) REAL*8 P1(0:3. 4) mean Fo (¯) and Fo (µ− ).IHEL=1. (NHEL(IHEL. This helps us to identify intermediate results W which each are only computed once. 1. 3) and Fo = W (∗.IHEL=1.ANS INTEGER NHEL(NEXTERNAL. -1. 1.NEXTERNAL) FOR PROCESS : u u˜ -> mu+ muINTEGER NEXTERNAL./ DATA NTRY/0/ DATA DATA DATA DATA DATA DATA DATA DATA DATA (NHEL(IHEL.NEXTERNAL). in our case 24 = 16 combinations. In general. 1.

JC(1)) ANS = ANS + T IF (T .IHEL=1.IHEL=1. 1. the originally produced particles have to radiate quarks or gluons to get shed their color charge.IHEL= 1. 1.TRUE.118 3 LHC PHENOMENOLOGY DATA DATA DATA DATA DATA DATA DATA DATA DATA (NHEL(IHEL. For example. in little-Higgs models T parity. Adding all different helicity combinations means a loop over the second argument and a call of UUB MUPMUM with the respective helicity combination.4) (NHEL(IHEL.NCOMB IF (GOODHEL(IHEL. Because of all naive helicity combinations many are physically not allowed there the array GOODHEL keeps track of valid combinations.GT. otherwise MadGraph does not waste time to compute the matrix element.4) (IDEN(IHEL). 1.IHEL=1. provided they exist. NTRY . In supersymmetry this is called R parity. -1.IPROC)) THEN GOODHEL(IHEL. 16). but those depend on the details of the weakly interacting new physics sector.e.IPROC)=. From cosmological constraints we know that dark matter definitely interacts gravitationally and that it cannot carry electromagnetic or color charges. 13). It it the reason why in modern TeV-scale model building every model and it’s dog has a stable WIMP. This helicity method might not seem particularly appealing for a simple 2 → 2 process. 14).4) (NHEL(IHEL. . 4/ 36/ NTRY=NTRY+1 DO IHEL=1. .IHEL=1.IHEL).3 Missing transverse energy Some of the most interesting signatures at the LHC involve dark matter particles. for each external momentum configuration. 1). ENDIF ENDIF ENDDO ANS = ANS/DBLE(IDEN) END The important part of this subroutine is the list of possible helicity combinations stored in the array nhel (1 : 4. 11). -1. 1/ 1. a complete spin–color averaging factor is included as IDEN and in our case given by 2 2 × 2 × Nc = 36. 1/ 1.IHEL=1. 3. i. hence invisible particles in the ATLAS and CMS detectors. 1.IHEL=1. -1/ 1. and in extra dimensional models Kaluza-Klein parity. 1. 1) / / / / / / / / / 1. 1.IHEL=1. If a conserved dark matter quantum number exists we will always produce them in pairs. but it makes it possible to compute processes with four and more particles in the final state and typically up to 10000 Feynman diagrams which we could never square symbolically. -1.4) (NHEL(IHEL.LT. Weak interactions are allowed because of their limited reach and it turns out that a weakly interacting particle with a mass around the TeV scale typically give the observed relic density in the universe. 1/ 1. 1 : 16). a multiplicative quantum number for (a sector) of the newly introduced particles.4) (NHEL(IHEL. -1. From supersymmetry we know that this is not hard to achieve: all we need is a Z2 symmetry. Each of them decays to the weakly interacting sector which includes a stable dark matter agent.NEXTERNAL JC(IHEL) = +1 ENDDO DO IHEL=1.IPROC) .AND. After an initialization to all ‘false’ this array is only switched to ‘true’ if UUB MUPMUM returns a finite value. On the way. MadGraph provides us with the subroutine SUUB MUPMUM and the function UUB MUPMUM which 2 together compute |M| for each phase space point. -1. GOODHEL(IHEL. 1.NOT. 2) THEN T = UUB_MUPMUM(P1. 10). 3. If in some kind of cascade decays they also radiate leptons or photons which can be very useful to trigger on the events and reduce QCD backgrounds. -1/ 1.4) (NHEL(IHEL. 1.4) ( IC(IHEL. 2. but they do not have to be. At the LHC we typically produce strongly interacting new particles. 1/ 1. -1. 15).OR. Altogether. 0D0 . At the very end. -1/ 1. the decay steps ideally are two body decays from on-shell particle to on-shell particle. 12). no matter how many graduate students’ lives we turn into hell.NHEL(1.IHEL=1.4) (NHEL(IHEL. i. What we therefore look for is jets in association with pairs of only weakly interacting.e. This is called the WIMP miracle.

[42].3. Some sources of noise. i. We can convert it into a missing transverse energy which in the absence of any information on particle masses is defined as the absolute value of the two-dimensional missing momentum vector. It will turn out that jet merging is needed to reliably predict the missing transverse momentum distributions in Standard Model processes.(2. It nicely illustrates that by just measuring missing transverse energy.3.3 we focused on the proper simulation of W +jets and Z+jets samples.5.1 Measuring missing energy The left panel of Figure 16 is a historic missing transverse energy distribution from DZero. We can identify such bad runs by looking at Standard Model physics. LHC events including dark matter are characterized by a high jet multiplicity and large missing transverse energy. it is purely a detector effect. the vector sum of the transverse momenta of all invisible particles. we have to subtract bad runs. which are the Standard Model backgrounds to such signals. The actual observable is the missing transverse momentum. At the end of Section 2. Figure from Ref. There are several sources of missing energy which we have to understand before we get to search for new physics: – first.3 nb LCW 100 120 140 Emiss [GeV] T Figure 16: Left: missing energy distribution from the early running phase of the DZero experiment at the Tevatron. while for missing transverse energy we need all of them. From Eq. In the same spirit. The way to look for invisible particles therefore is a mis-balance of three-momentum in the transverse plane.28) and the discussion of parton densities we remember that at hadron colliders we do not know the entire kinematics of the initial state.e. like gauge bosons. To cut on a variable like missing transverse momentum we need to understand our detectors really well. there will be coherent noise in the calorimeter. Figure from Beate Heinemann. They happen if for a few hours parts of the detector might not have worked properly. Tevatron would have discovered supersymmetry based on two beautiful peaks in the distribution around 150 GeV and around 350 GeV. whereas in the transverse plane by definition the incoming partons have zero momentum. They can be cosmic rays or . Of 200000 cells we know that some of them will individually fail or produce noise. and for this level of understanding we need time and luminosity. Unless something goes wrong with the machine we should not base a new physics analysis in early LHC running on a missing energy cut. However.3 Missing transverse energy 119 Events / 1 GeV 105 104 103 102 10 1 10-1 0 20 40 60 80 ATLAS Preliminary Data 2010 | |<4. The problem of missing energy we can illustrate using a simple number: to identify and measure a lepton we need around 500 out of 200000 calorimeter cells in an experiment like ATLAS. This means that after the studies in Section 2 we are at least theoretically on safe ground. and remove them from the data sample. but not coming from the interaction point. like leaking voltage or other electronic noise can be correlated geometrically. In beam direction we only know its boost statistically. there might also be particles crossing our detector. However. It will lead to beautiful missing momentum signals. Right: corrected missing energy distribution in QCD events at ATLAS using only data from April/May 2010 at 7 TeV collider energy. this (preliminary) experimental result has nothing to do with physics.5 Data MC QCD Jets s = 7 TeV -1 Ldt=14. this is not the whole story of missing transverse momentum. – next. 3.

120

3

LHC PHENOMENOLOGY

errand beam radiation, and they will lead to unbalanced energy deposition as well. Again, the way to get rid of such noise is looking at Standard Model candles and remove sets of events where problems occur. – the third class of fake missing energy is failing calorimeter cells, like continuously hot cells or dead cells, which can be removed for good once we know the detector really well. – while not really a detector fake the main source of missing energy at hadron colliders are mis-measured QCD jets. If parts of jets point into regions with poor calorimetry, like support structures, their energy will be wrongly measured, and the corresponding QCD event will show missing transverse energy. One way to tackle this problem is to require a geometric separation of the missing momentum vector and all jets in the event. Nevertheless, as a rule of thumb, O(1%) of all hard QCD events still lead to more than 100 GeV of fake missing transverse energy. Figure 5 shows that this is not at all a negligible number. Once we understand all the source of fake missing momentum we can focus on real missing momentum. This missing transverse momentum we compute from the momenta of all tracks seen in the detector. This means that any uncertainty on these measurements, like the jet or lepton energy scale will smear the missing momentum. Moreover, we know that there is for example dead matter in the detector, so we have to compensate for this. This compensation is obviously a global correction to individual events, which means it will generally smear the missing energy distribution. The right panel of Figure 16 shows a very early missing transverse energy distribution of ATLAS after some of the corrections described above. To simulate a realistic missing transverse momentum distribution at the LHC we have to smear all jet and lepton momenta, and in addition apply a Gaussian smearing of the order ∆ET / 1 ∼ GeV 2 ET GeV 20 . (3.7)

While this sounds like a trivial piece of information I cannot count the number of papers where people forget this smearing and discover great channels to look for Higgs bosons or new physics. They fall apart when experimentalists take a careful look. The simple rule is: phenomenological studies are right or wrong based on if they can be reproduced by real experimentalists and real detectors or not. 3.3.2 Missing energy in the Standard Model

In the Standard Model there exists a particle which only interacts weakly: the neutrino. At colliders we produce them in reasonably large numbers in W decays. This means that in W + jets production we can learn how to reconstruct the mass of a leptonically decaying W , i.e. one observed and one missing particle. We construct a transverse mass in analogy to an invariant mass, but neglecting the longitudinal momenta of the decay products m2 = (ET,miss + ET, ) − (pT,miss + pT, ) T
2 2

= m2 + m2 + 2 (ET, ET,miss − pT, · pT,miss ) , miss

(3.8)

2 in terms of a transverse energy ET = p2 + m2 . Since the transverse mass is always smaller than the actual W mass T and reaches this limit for realistic phase space regions we can extract mW from the upper edge in the mT,W distribution. Obviously, we can define the transverse mass in many different reference frames. However, its value is invariant under — or better independent of — longitudinal boosts. Moreover, given that we construct it as the transverse projection of an invariant mass it is also invariant under transverse boosts. By construction we cannot analyze the transverse mass event by event, so this W mass measurement only uses the fraction of events which populate the upper end of the transverse mass distribution.

Alternatively, from single top production we know another method to conditionally reconstruct masses and momenta involving one invisible particle: from a leptonically decaying top quark we only miss the longitudinal momentum of the neutrino. On the other hand, we know at least for the signal events that the neutrino and the lepton come from an on-shell W boson, so we can use this on-shell condition to reconstruct the longitudinal neutrino momentum under the assumption that the neutrino have zero mass.

3.3

Missing transverse energy

121

From Higgs searches we know how to extend the transverse mass to two leptonic W decays, i.e. two neutrinos in the final state. The definition of this transverse mass m2 W = (ET,miss + ET, ) − (pT,miss + pT, ) T,W
2 2

= m2 + m2 + 2 (ET, ET,miss − pT, · pT,miss ) miss

(3.9)

is not unique because it is not clear how to define mmiss , which also enters the definition of ET,miss . From Monte Carlo studies it seems that identifying mmiss ≡ m , which is correct at threshold, is most strongly peaked. On the other hand, setting mmiss = 0 to define a proper bounded-from-above transverse mass variable seems to improve the Higgs mass extraction. For an unspecified number of visible and invisible particles in the final state there exist global observable we can rely on. The visible mass it is based on the assumption that we are looking for the decay of two heavy new states where the parton densities will ensure that these two particles are highly non-relativistic. We can then approximate the partonic √ ˆ energy s ∼ m1 + m2 by some kind of visible energy. Since we assume that the heavy states are produced with little energy, boost invariance is not required for this construction. Without taking into account missing energy it looks like   m2 visible =
,j

2


,j

2 p . (3.10)

E − 

Similarly, the Tevatron experiments have for a long time used an effective transverse mass scale which is usually evaluated for jets only, but can trivially be extended to leptons: HT =
,j

ET =
,j

pT ,

(3.11)

where the last step assumes that all final state particles are massless. In an alternative definition of HT we sum over a number of jets plus the missing energy and skip the hardest jet in this sum. When combining such a measure with missing transverse momentum the question arises: do we want to pair up the missing transverse momentum with the visible transverse momenta or with the complete visible momenta? For example, we can use the scalar sum of all transverse momenta in the event, now including the missing transverse momentum meff = ET = pT , (3.12)
,j,miss ,j,miss

This effective mass is known to trace well the mass of the heavy new particles decaying for example to jets and missing energy. This interpretation relies on the non-relativistic nature of the production process and our confidence that all jets included are really decay jets. 3.3.3 Missing energy and new physics

The methods described in the last section are well studied for different Standard Model processes and can be applied in new physics searches for various lengths of decay chains. However, there is need for one significant modification, namely to account for a finite unknown mass of the carrier of the missing energy. This is a problem of relativistic kinematics and at leading order does not require any knowledge of QCD or new physics models. The chain of three successive three-body decays shown in Figure 17 is the typical left handed squark cascade decay in supersymmetry. The same topology we can interpret in extra-dimensional models as the decay of a Kaluza-Klein quark excitation qL → χ0 q → ˜± ˜ ˜2 q → χ0 ˜1
+ −

q

QL → Z (1) q →

(1)

(1) ±

q → γ (1)

+ −

q.

(3.13)

In both cases the last particle in the chain, the lightest neutralino or the Kaluza-Klein photon excitation pass the detectors unobserved. The branching ratio for such decays might not be particularly large; for example in the supersymmetric parameter point SPS1a with its mass spectrum shown in Figure 21 the long squark decay ranges

122

3

LHC PHENOMENOLOGY

q

µ µ

q

~

χo ˜2

µ χo ˜1

~

Figure 17: Feynman diagram for the long decay chain shown in Eq.(3.13). around 4%. On the other hand, strongly interacting new particles should be generously produced at the LHC, so we usually assume that there will be enough events to study. The question is how we can then extract the four masses of the new particles appearing in this decay from the three observed external momenta. The proposals to solve this problem can be broadly classified into three classes. While all of them should in principle work and would then differ mostly by statistics, only for the first strategy we know how QCD and detector smearing affect them. 1. endpoint methods extract masses from lower (threshold) and upper (edge) kinematic endpoints of invariant mass distributions of visible decay products. This method is best suited for long decay chains, where the number of independent endpoint measurements in one leg at least matches the number of unknown masses in the cascade. An implicit assumption of these endpoint techniques is that the form of the matrix element populates the phase space close to the endpoint well. Otherwise, the endpoint will be soft and hard to identify on top of the continuum background. The squark decay Eq.(3.13) has a particular kinematic feature: the invariant mass distributions of the two leptons m . Looked at from the rest frame if the intermediate slepton is a current-current interaction, similar to the Drell-Yan process computed in Eq.(2.11). Because in the s channel there now appears a scalar particle there cannot be any angular correlations between the two currents, which means the m distribution will have a triangular shape. Its upper limit, called the dilepton edge, we can compute: in the rest frame of the scalar lepton the three-momenta of the incoming and outgoing pair of particles have the absolute values |p| = |m2 0 − m2 |/(2m ˜). The lepton mass we set to zero. The invariant mass of the two lepton reaches its ˜ χ1,2 ˜ maximum if the two leptons are back to back, i.e. the scattering angle is cos θ = −1 m2 = (p
+

+ p − )2
− −

= 2 (E + E < 2 (E + E =4 m2 0 χ2 ˜ − 2m ˜

− |p + ||p − | cos θ) + |p + ||p − |) m2 − m2 0 ˜ χ ˜
1

m2 ˜

2m ˜

using E 2± = p2± .

(3.14)

This kinematic statement is independent of the shape of the m distribution. For both particle assignments shown in Eq.(3.13) the kinematic endpoint are given by 0 < m2 < (m2 0 − m2 )(m2 − m2 0 ) ˜ ˜ χ ˜ χ ˜
2 1

m2 ˜

0 < m2 <

(m2 (1) − m2(1) )(m2(1) − m2 (1) ) Z γ m2(1)

.

(3.15)

An problem of realistic applications of endpoint methods is that to avoid combinatorics we need to either clearly separate the decays of two heavy new states, or we need to combine a short decay chain on one side with a long chain on the other side. In supersymmetry this is naturally the case for associated squark and gluino production. A right handed squark often decays directly to the lightest neutralino which is the dark matter

3.3

Missing transverse energy

123

1

2

3

4

8 5 6 7

Figure 18: Example for the mass relation method using three successive two-body decays on both sides of the events (left). After detector smearing we can reconstruct the masses for the supersymmetric parameter point SPS1a with the squark decay chain shown in Eq.(3.13). Figure from Ref. [43].

candidate in the model. The gluino has to radiate two quark jets to reach the weakly interacting sector of the model and can then further decay in many different ways. In other models this feature is less generic. The impressive potential of endpoint methods in the case of supersymmetry we will illustrate later in this section. When looking at long cascade decays for example with two leptons we usually cannot tell which of the two leptons is radiated at an earlier stage of the cascade. Therefore, endpoint techniques will always be plagued with combinatorics from the mapping of the particle momenta on the decay topology combinatorial background. The same applies to QCD jet radiation vs decay jets. In this situation it is useful to consider the correlation of different invariant masses and their endpoints. The endpoint method can be extended to use invariant mass distributions from both sides of the event (hidden threshold techniques), and correlations between the distributions from each leg (wedgebox techniques). Instead of looking for endpoints in all events we can also reconstruct masses by forcing the events into extreme kinematic configurations. One such region are phase space points close kinematic endpoints where particles are produced at rest. Another example is the approximate collinear Higgs mass reconstruction in a decay to boosted tau pairs described in Section 1.5.3. 2. mass relation methods generalize the single-top example in Section 3.3.2 and completely reconstructing the kinematics event by event. For each event this construction provides a set of kinematic constraints. While for one event the number of parameters can be larger than the number of measurements, adding signal events increases the number of measurements while keeping the number of unknowns constant. Eventually, the system of equations will solve, provided all events are really signal events. Implicitly, we always assume that all decaying particles are on-shell, but off-shell effects we do not expect to have a huge effect on the results at the LHC. In Figure 18 we show the general topology of a three-step cascade decay on each side of the event, like we expect it for a pair of left handed squarks following Eq.(3.13). To extract the masses of the new particles we need to solve the system of equations (p1 + p2 + p3 + p4 )2 = m2 Z (p2 + p3 + p4 )2 = m2 Y (p3 + p4 )2 = m2 X p2 = m2 4 N (3.16)

m2 .m[χ0] / GeV 1 1 Figure 19: Simulations for different mT X .miss = q1 + q2 . MT2 methods are based on a global variable mT 2 . In addition.8 2 mTX(m[χ0]) . so it can be inverted..(3. one from each leg of the event.m[χ0] 1 1 0 0.). E4 . this minimum will simply be the zero transverse momentum limit of mT on one leg. ˆ Inspired by the usual transverse mass we are interested in a mass variable with a well-defined upper edge. One strategy to solve this problem is to assume eight test masses m = (m2 . m2 . It generalizes the transverse mass known from W decays to the case of two massive invisible particles. assuming we know the invisible particle’s mass mT. (3.j as a function of the splitting of pT. Using the first equal sign we we can collect the two missing four-momenta into pmiss = (p4 .8 ≡ (s4 . we define the measured combinations s4. Naively.(3.6 0. The first three lines in Eq. Together.5.124 3 LHC PHENOMENOLOGY m[π] mT4 ee mT3 eπ mT2 ππ m[χ+] . Similarly.17) we can conveniently solve for the missing momentum −2(p1 p4 ) = m2 − m2 + 2(p1 p2 ) + 2(p1 p3 ) ≡ s1 Y Z −2(p2 p4 ) = m2 − m2 + 2(p2 p3 ) X Y −2(p3 p4 ) = m2 N − m2 X ≡ s2 ≡ s3 .(3. mmiss ). The observed missing energy in the event we can divide into two scalar fractions pT. The matrix A includes only components of measured momenta and is almost block diagonal. s8 ). E8 ) and express this vector in terms of measured quantities and masses in the cascade A · pmiss = s.17) for each event plus the two missing transverse momentum components to determine both missing four-momenta and test the consistency of this solution using the last line of Eq.17) we can then write s = B · m + c. we measure the missing transverse momentum as pT.2 0. the ˆ second argument means an external assumption.6 from the opposite chain. p8 . Figure from Ref. The mass relation method has also been successfully applied to single legs as well as in combination with kinematic endpoints.18) The result for all masses in the decay chain using 25 events per set and including all combinatorics we show in Figure 18. i. On the other . which is not very interesting. [44]. m2 ..17) for each of the two legs. so we need to use some kind of minimum of mT.4 0. Given the two fractions qj we can construct a transverse mass for each side of the event. use the three Z Y X N first equations in Eq. All of these 12 unknowns we can determine if we add a sufficiently large number of events.8 1 1. In this consistency test we combine the information from several events. 3. Following the second equal sign in Eq.4 + pT. for each side of the event. The blue mT 2 line applies to ˜1 ˜1 ˜1 ˜1 the two-body decay.e.4 1. for the decay χ+ → χ0 π or χ+ → χ0 e+ ν. mmiss is an assumed value for mmiss . (3.6 1.17) for simplicity assuming massless Standard Model decay products. For each event there are eight unknown masses and six unknown missing momentum components of which we measure two combinations as the missing transverse momentum. . this gives us pmiss = A−1 s = A−1 B m + A−1 c .(3.2 1.j (qj . where the matrix B only contains non-zero entries ±1 and the vector c consists of measured quantities.miss .

In favorable cases mT 2 may allow the measurement of both the parent particle and the LSP based on a single-step decay chain. There are a two properties we know by construction mlight + mmiss < mT 2 (mmiss ) ˆ ˆ mlight + mmiss < mT 2 (mmiss ) < mheavy (3.miss =q1 +q2 (3. Moreover. i. for the correct value of mmiss the mT 2 distribution has a sharp edge at the mass of the parent particle.(3. the procedure can be generalized to any one-step decay. so we can instead define mT 2 (mmiss ) = ˆ min max mT. so for massless Standard Model decay products there will be a global mT 2 threshold at the missing particle’s mass. This aspect we can exploit by scanning over mmiss and looking for so-called kinks. The classic example for the endpoint method is the long supersymmetric left-handed squark decay chain shown in Eq. in this case the transverse mass from the other leg reaches a maximum. The way such mass measurements can lead to proper model reconstruction we sketch for one example.3.j lie between the sum of the two decay products’ masses and the mass of the decaying particle. Similar to the more global meff variable we can generalize mT 2 to the case where we do not have a clear assignment of the two decay chains involved. Figure from Giacomo Polesello (ATLAS).13) and in Figure 17. which unfortunately is not quite as sharp as the one in mT 2 .3 Missing transverse energy 125 1500 Events/1 GeV/100 fb-1 OS-SF SUSY OS-OF SUSY SM 1000 500 0 0 20 40 60 m(ll) (GeV) 80 100 Figure 20: Invariant mass of two leptons after selection cuts for the SPS1a parameter point: SUSY signal Opposite Sign Same Flavor (OS-SF): full line. This modification MT Gen again has an upper edge. points where different events kinematics all return the same value for mT 2 . hand.20) The first line means that each of the mT. mmiss ) ˆ j pT . When we use such kinematic endpoints or other methods to extract mass parameters it is crucial to start . while the upper edge of mT 2 − mχ0 coincides with the ˜1 ˜1 dashed line for mχ+ − mχ0 .e. Similarly. For a wrong assignment of ˆ mmiss it has nothing to do with the actual kinematics and hence with any kind of invariants.19) as a function of the unknown missing particle mass. Such MT X distributions are useful as long as they have a sharp enough edge. as illustrated in Figure 19. These two aspects for the correct value mmiss = mmiss we can see in Figure 19: the ˆ lower threshold is indeed given by mT 2 − mχ0 = mπ . Standard Model background: grey. SUSY signal Opposite Sign Opposite Flavor (OS-OF): dotted line. for example a three-body decay with either one or two missing particles on each side of the event.j (qj . ˜1 ˜ 1 An interesting aspect of mT 2 is that it is boost invariant if and only if mmiss = mmiss . and house numbers are not boost invariant.

This is due to the form of the endpoint formulas which involve the difference of mass squares m2 − m2 = (m1 + m2 )(m1 − m2 ).mχ 0 ∼ l 1 mχ 0 . In complete analogy to the dilepton edge shown in Eq. we solve the system for the intermediate masses without any model assumption.mb ~ ~ m~ g m~ . [45]. a point at which we might have to start thinking about off-shell propagators and at some point even define what exactly we mean by ‘masses as appearing in cascade decays’.mχ 0 ∼ ∼ 1 1 m~ q L mg . Provided the gluino or heavy gluon is heavier than the squarks or heavy quarks we can measure its mass by extending the squark chain by one step: g → q q . which allows us to even measure the dark matter mass to O(10%).m∼ 0 b χ 1 1 R 2 1 m~ b 1 0 100 200 300 400 Sparticle masses and mass differences [GeV] 500 600 Figure 21: Masses extracted from the gluino-sbottom decay chain. where the edge is softened by neutrinos from tau decays.m∼ 0 g χ m~ . For jets.(3. including estimated errors. The supersymmetric mass spectrum is given by the SPS1a parameter point. The way around ¯ is to ask for two bottom jets from the strongly interacting decay: g → b˜∗ or G(1) → bB (1) . Nevertheless. This measurement is hard if one of the two jets from the gluino decay is not ˜ ˜ very hard. The key observation is that in long cascade decays the leptons are flavor-locked. The limiting factors will likely be our chances to observe enough endpoints in addition to mmax and the jet energy scale uncertainty. as illustrated in Figure 20. correlated jet and lepton energy scale uncertainties do not make life easier either.126 3 LHC PHENOMENOLOGY m∼ 0 χ 1 m~ m∼ 0 l χ R 2 m~ . The long squark decay in SPS1a-like parameter points with squark masses in the 500 to 600 GeV range has an important advantage: for such a large mass hierarchy we should be able to isolate the one decay jet just based on its energy. because its information will be buried by the combinatorial error due to QCD jet radiation. but with somewhat reduced elegance we can measure the threshold and edge of the + − q distribution and the edges of the two ± q combinations. which means the combination e+ e− + µ+ µ− − e− µ+ − e+ µ− becomes roughly twice µ+ µ− for the signal. while it cancels for top pairs. the common lore that kinematics only constrain mass differences is obviously not true for two body decays. The summary of all ˜ b measurement in Figure 21 shows that for example the gluino mass we can extract at the per-cent level. A generic feature or all methods relying on decay kinematics is that it is easier to constrain the differences of squared masses than the absolute mass scale. Experimentally. Figure from Ref. from a signal-rich sample to avoid combinatorics and washed out endpoints vanishing in a fluctuating or even sculptured background. In that case the invisible heavy state is produced . leptons and missing energy a major background will be top pairs.15). An interesting question is how well we will do with tau leptons. Using such a combination for the endpoint analysis means the top background is subtracted purely from data. Alternatively. Then. This is also shown in Figure 21. This combination is much more 1 2 sensitive to (m1 − m2 ) than it is to (m1 + m2 ). The faint blue lines indicate wrong solutions when inverting the endpoint-mass relations. we can use the same gluino decay to first reconstruct the intermediate neutralino or Kaluza-Klein Z momentum for lepton pairs residing near the m edge.m∼ 0 q χ L 1 m~ .

13) there exists a promising method to simultaneously determine the spin of all new particles in the chain: 1. Invariant masses are just an invariant way of writing angular correlations between outgoing particles.2 is problematic for low-pT jets. like left or right handed sbottoms. For short decay chains mT 2 is best suited to measure the masses of particles decaying directly to the dark matter agent. typical for supersymmetric theories. This second method of reconstructing an invariant mass of the intermediate state could ideally distinguish different intermediate states. so we can contrast it with another hypothesis where the spins in the decay chain follow the Standard Model assignments. Keeping in mind that endpoint analyses only use a small fraction of the events. .(3. Those depend on the spin and quantum numbers of all particles involved. approximately at rest. Kinematic endpoints can be supplemented by any other method to measure new physics masses.5. For the squark decay chain given in Eq.3. for the mass spectrum shown in Figure 21 we can measure the gluino mass to few per-cent. This means that we should be careful when extracting information for example from kinematic endpoints we do not observe. depending on the systematic errors. However. this strategy bears a serious danger. In supersymmetry. is hard in the absence of fully reconstructed events.21) If both neutralino masses (or the Kaluza-Klein photon and Z masses) are known we can extract the sbottom (Kaluza-Klein bottom) and gluino (Kaluza-Klein gluon) masses by adding the measured bottom momenta to this neutralino (Kaluza-Klein photon) momentum. where each Standard Model particle acquires a Kaluza-Klein partner from the propagation in the bulk of the additional dimensions. kinematic endpoints can for example be softened. Supersymmetry switches this fermion/boson nature compared to the corresponding Standard Model particle. which following Sections 2. [46]. this happens for right handed sleptons or right handed squarks. The spins inside the decay chain can only alternate between fermions and bosons. and the momenta are correlated as p χ0 = ˜2 1− mχ0 ˜1 m p pZ (1) = 1− mγ (1) m p (3. Instead. instead of trying to measure spins in a general parameterization we start from the observation that cascade decays radiate particles with known spins. namely those with extreme kinematics. Measuring discrete quantum numbers. Figure taken from Ref. Again. The spectrum j is assumed to be hierarchical. an obvious way to improve their precision is to include the complete shape of the invariant mass distributions. This argument we can turn around.5.3 Missing transverse energy 127 Figure 22: Asymmetry in mj /mmax for supersymmetry (dashed) and universal extra dimensions (solid). Depending on the quantum numbers and mixing angles in the new physics scenario. in particular when the final state includes missing transverse energy. we rely on angular correlation in decays. so they vanish in the background noise. This is most obvious for long gluino decays where we know that the radiated bottom quarks as well as muons are fermions. An example of such a model is Universal Extra Dimensions (UED). The issue with short decay chains is that they often require on some kind of jet veto.3 and 1. like the spin of new particles. The usual threshold behavior is not observable at hadron colliders.

In other words. A decaying squark radiates a quark while an antisquark radiates an antiquark. They are hardest to model because they are often dominated by higher-order QCD effects in fixed order perturbation theory. The only caveat is the shape of far-away tails. 3. In contrast. once we consider a statement about perturbative QCD a statement about an uncertainty we are probably only giving a wild guess. LHC physics always means extracting signals from often large backgrounds. For LHC calculations we are usually confronted with three types of errors. Some of the readers might remember a bunch of theorists mistaking a forward pion for an electron — that happened right around my TASI in 1997 and people not only discovered supersymmetry but also identified its breaking mechanism. For example the mj distribution allows us to analyze spin correlations in squark decays in a Lorentz invariant manner. Systematic errors which do not follow a Gaussian distribution we quantify in terms of nσ = Nsig /Nback which means they do not improve with increasing luminosity. which means that we can define a non-zero production-side asymmetry between mj + and mj − . 3. which often turn out to be higher than the exponentially suppressed Gaussian tails. Of course. In high energy physics five standard deviations above a known background we call a discovery. They show that the LHC can do much more than just discover some kind of particles beyond the Standard Model. like the calibration of the jet and lepton energy scales. and for large numbers they converge to the Gaussian limit.9 or 1. or the efficiencies to for instance identify a muon as a muon. Again. weak-boson-fusion signatures. and once we are systematics dominated waiting for more data does not help. If we can compute or reliably estimate some higher order terms in the perturbative QCD series we call this a prediction. Such an asymmetry we show in Figure 22. gluino decays with decay-side asymmetries. cascades including charginos. but experimentally we determine efficiencies and scaling factors largely from well understood background processes. the shape of the distribution between the endpoints is nothing but an angular correlation in some reference frame. our experimentalist CDF lecturer told us that the whole thing was a problem of identifying a particle in the detector with an efficiency which does not have to zero or one. the measurements of the luminosity. For small numbers of events these experimental errors are described by Poisson statistics. The first and easiest one are the statistical errors. like decays including gauge bosons. It means that for small number of signal and background events we need more than The second set of errors are systematic errors. Naively. To model theoretical uncertainties it is crucial to realize that higher order effects are not any more likely to give a K factor of 1.0 than 0. Provided the masses in the decay chain are not too degenerate we can indeed distinguish the two hypotheses. In that limit the number of standard √ deviations in terms of the number of signal and background events is nσ = Nsig / Nback . likelihood distributions accounting for theoretical errors do not have a .128 3 LHC PHENOMENOLOGY 2. Kinematic endpoints cannot distinguish between supersymmetry and universal extra dimensions. This means.1. etc. five standard deviations are required to claim a discovery. In other words. This basic idea has since been applied to a great many similar situations. Such counting experiments in background channels like Z → leptons do behave Gaussian. For the SPS1a spectrum their ratio ranges around 2:1. The Poisson region is about the only complication we encounter for statistical errors. three sigma is often referred to as an evidence. we would not assume that systematic errors follow a Gaussian distribution. From Section 2 we know that higher order corrections for example to LHC production rates do not follow a naive power counting in αs but are enhanced by large logarithms. The two event numbers are proportional to the integrated luminosity L which means that the statistical significance in the Gaussian limit increases √ with L. a proton–proton collider like the LHC produces considerably more squarks than antisquarks in the squark–gluino associated channel. for the SUSY and for the UED hypotheses.4 Uncertainties As we argue in the very beginning of the lecture. The third source of errors are theoretical errors. a correct error estimate is crucial. three-body decays. thresholds and the edges of all invariant masses of the radiated fermions are completely determined by the masses inside the decays chain.

and a measure in model space or error space is not defined by physics. 3. and the proper description of error bars is our smallest problem. For the center of the distribution this means we need to cut open the experimental Gaussian distribution and insert a flat theory piece. systematic and theoretical errors gives us a Gaussian distribution altogether. (3. A profile likelihood instead projects the best fitting point of the unwanted direction onto the new parameter space. If that happens we need to conclude that perturbative QCD breaks down. The downside is that any integration needs a measure. the centrally flat theory probability distribution for an LHC observable has to go to zero for very large deviations from the currently best value...5 Further reading 129 peak and are definitely not Gaussian. But this approach assumes the we know much more about QCD than we actually do which means it is not conservative at all. The same thing we can do for a Gaussian experimental error and a flat theory errors.n−1 . there exist several good review articles with more in-depth discussions of different aspects touched in this Section: – as mentioned in Section 2 two very useful reviews of jet physics are available by Steve Ellis and collaborators [30] and by Gavin Salam [31]. We therefore assume such a prior and have to test our final result by varying this prior in what we consider a reasonable range.. An intuitively better solution which gives a centrally flat likelihood distributions for the combined errors is the Rfit scheme. In other words. for each binned parameter point in the (n − 1)-dimensional space we explore the nth direction which is to be removed L(x1. we also know that theoretical errors cannot be arbitrarily large.. In the Bayesian picture we define individual likelihood distributions for the different errors and integrate over the corresponding uncertainties. Such a projection avoids defining a measure but it does not maintain for example the normalization of the likelihood distribution. . The integral returns a two-sided error distribution which still has a peak. In the center the distribution is flat. It is a frequentist of profile likelihood construction marginalizing the unknown errors. The impact of the size of the flat box we need to text.22) This distribution implies that for very large deviations there will always be tails from the experimental errors.n−1 .. which makes things numerically much easier.. so we can neglect the impact of the theoretical errors on this range. which is that folding three Gaussian shapes for statistical. Along this direction we pick the best value and it with the lower-dimensional parameter point L(x1. This is where the convolution of several Gaussians comes from. used for example by CKMfitter or SFitter.5 Further reading Again. Such a marginalization of an unwanted direction in the parameter space of our likelihood is a mathematical problem: to keep the mathematical properties of a likelihood as a probability measure we would prefer to integrate over unwanted directions... a K factor for a total LHC cross section should not be larger than something like two or three. xn ). While this is not the only way to describe all sources of errors on LHC measurements it is certainly consistent and convincing... xn ). This is what the Bayesian pictures does.i = σi   d − d − σ (theo)  i ¯i  i   (exp) σi (theo) ¯ |di − di | < σi (theo) ¯ di − di < −σi (theo) ¯ di − di > σi . reflecting our ignorance about the theory prediction.n−1 ) ≡ Lmax(n) (x1.. For the log-likelihood χ2 = −2 log L given a set of measurements d and in the presence of a general correlation matrix C this gives us χ2 = χT C −1 χd d  0    d − d + σ (theo)  i ¯i  i (exp) χd. The Rfit scheme projects away a flat theory error as well as a Gaussian experimental error. Nothing stops us from computing likelihood maps to measure the top mass. Unless there is a very good reason. On the other hand. There is a good reason to choose the Gaussian short cut. analyze the Higgs sector or identify new physics at the LHC..3.

130 3 LHC PHENOMENOLOGY – for the general phenomenology of the heaviest Standard Model particles. – the pivotal paper on determining spins in cascade decays is Jennie Smillie’s PhD thesis [46]. Chris Lester’s thesis [50] is a good point to start with. errors. On the same topic there is a nicely written review by Liantao Wang and Itay Yavin [53]. Recently. and the statistical treatment of new-physics parameter spaces you can find in the big SFitter publication [54]. The SFitter analysis of the Higgs sector [9] is very similar in structure. – if you use MadGraph/HELAS to compute helicity amplitudes there is the original documentation which describes every routine [48]. the top quark. Alan Barr and Chris Lester published a broad review on techniques to measure masses in models with dark matter particles [51]. – many useful pieces of information on mass extraction. – a lot of insight into new physics searches at the LHC and at a linear collider you can find in a huge review article collected by Georg Weiglein [52]. – a lot of experimental knowledge on new physics searches well described and theoretically sound you can find in the CMS technical design report. Chris Lester and Phil Stephens [44]. Some key analyses are described in detail while most of the documentation focuses on the physics expectations [49]. – more on the magical variable mT 2 you can find in an article by Alan Barr. have a look at Sally Dawson’s corresponding TASI lectures [47]. . – as mentioned in the introduction. there is a more advanced review on new physics at the LHC which includes an extensive chapter on LHC signatures [1]. but different in the physics application.

For this longer version I am indebted to Steffen Schumann and Thomas Binoth for helping me out on many QCD questions over the past years. guys. . I would like to thank all the people who have convinced me that theoretical physics even including QCD is fun — at least most of the time.5 Further reading 131 Acknowledgments The list of people I would like to thank is long and historically grown: starting with Peter Zerwas. Florian G¨ rtz or o Michael Spannowsky followed my appeal to give me lists of typos. and I would like to thank Tom DeGrand and Ben Allanach for their comments on the TASI notes and Sally Dawson for her encouragement to put these notes on the web. to Jan Pawlowski for uncountable cups of coffee on general field theory and QCD. Alert readers like Sebastian Bock. As a long-term collaborator I am missing Dave Rainwater who should have stayed in LHC physics. Dieter Zeppenfeld and Uli Baur. to Fabio Maltoni and Johan Alwall for teaching me jet merging. Michael Spira and Michael Kr¨ mer I would like to thank all the people who taught me theoretical o a physics and phenomenology over many years. The QCD part of this lecture is based on my 2008 TASI lectures. Sorry. This later on included Tao Han. but what you are holding in your hands is really the best I could do.3. Roland H¨ pker. and Manuela Wirschke carefully read the notes removing language mistakes — thank you very much to everyone who helped make this writeup more correct and more readable. Most importantly. Wim Beenakker. Not surprisingly this includes many US colleagues mostly from our TASI year ’97.

Rev. M. Rainwater and D.” Phys.” Phys. [11] P. “Jet substructure as a new Higgs search channel at the LHC. Rev. “Broken Symmetries and the Masses of Gauge Bosons. W. 585 (1964). Rauch. Rept. “New Physics at the LHC. L. Zeppenfeld. Davison. 242001 (2008) [arXiv:0802.” Phys. Djouadi.” Phys. Maiani.” arXiv:0912. Morrissey. C. 100. 009 (2009) [arXiv:0904. Lett. P. 132 (1964). Higgs.” Phys. 13.132 REFERENCES References [1] D. [10] U. Plehn and D. I: The Higgs boson in the standard model. [5] A. S. Prog. Campbell. Stirling. 70. The Higgs bosons in the minimal supersymmetric model. Hagen and T. P. [14] G. G. [7] J. M. Phys. “Hard Interactions of Quarks and Gluons: A Primer for LHC Physics. J. [12] P. Djouadi. B 158. 033003 (2003) [arXiv:hep-ph/0211224]. Rept. “Determining the structure of Higgs couplings at the LHC. . 1 (2008) [arXiv:hep-ph/0503172]. [6] J. Rev. 88. Tait. 051801 (2002).2470 [hep-ph]]. massless particles and gauge fields. Petronzio. [9] R. “The Anatomy of electro-weak symmetry breaking. Phys. Lett. B. W. Rev.” Fortsch. Cabibbo. Baur. “Broken symmetry and the mass of gauge vector mesons. M. “Bounds On The Fermions And Higgs Boson Masses In Grand Unified Theories. Phys.” Phys. Plehn and T. D 67. Salam. [13] F. “Determining the Higgs boson selfcoupling at hadron colliders. T.” Rept. Kibble. Plehn. 1 (2008) [arXiv:hep-ph/0503173].” Phys. W. M. Rev. [2] N. Englert and R. D. u “Measuring the Higgs Sector. “Broken symmetries. A. “Global Conservation Laws and Massless Particles. [8] T. Higgs. II. 89 (2007) [arXiv:hep-ph/0611148]. T. Rainwater. Rubin and G. Guralnik. 203 (1998) [arXiv:hep-ph/9705337]. Lett. D. 457. “The Anatomy of electro-weak symmetry breaking. 13. Brout. E. [3] M.3866 [hep-ph]]. Lafaye. Lett.” Phys.” JHEP 0908. W. L.” Phys. D¨ hrssen. 13. M.3259 [hep-ph]. 12. “QCD effects in Higgs physics. Huston and W. Parisi and R. Lett. 321 (1964). 459. J. R. Rev.” Nucl. 508 (1964). 46. T. Butterworth. Plehn. Lett. 295 (1979). [4] A. R. Spira. Zerwas and M.

77) . 8. Kilian. R. Webber. Plehn and T.REFERENCES 133 [15] W. Cosmol.” Int. 1 (1996). F. Catani. Ellis. M. Nucl. Schmelling. Phys.” Phys. Han. D.” arXiv:hep-ph/9908378. 075001 (2009) [arXiv:0810. “Four generations and Higgs physics. P. Plehn. I. “The Standard Model: A Primer” Cambridge. P. Burgess and G. Phys.” arXiv:hep-ph/0508097. A 20. “Electroweak symmetry breaking: The bottom-up approach. M. Phys. 198 (2003) 1. D.” J. [17] M. Kribs. [20] T. D 76. M. “TASI lecture notes: Introduction to precision electroweak analysis. “Applications of Perturbative QCD” Redwood City. “QCD and collider physics” Camb. [26] O.3718 [hep-ph]]. “Elementary Particla Physics: Concepts and Phenomena” Berlin. Phys. [18] D. “QCD matrix elements + parton showers” JHEP 0111. Knowles and M. [25] T.3919 [hep-ph]]. [24] S. [23] R. u “Concepts of Electroweak Symmetry Breaking and Higgs Physics. M. USA: Addison-Wesley (1989) 366 p. T. R. R. Field. [22] G. [21] J. B¨ scher and K. Pr. Dissertori. UK: Cambridge Univ. Rev. K. 075016 (2007) [arXiv:0706. G 36.” arXiv:hep-ph/0512342. 063 (2001) [arXiv:hep-ph/0109231]. Spira and P. Gomez-Bock. M¨ hlleitner. Rainwater.” arXiv:0712. M. “Searching for the Higgs boson. L. Phys. Zerwas. (2007) 542 p [28] G. G. Moore. Mondragon. u “Higgs boson searches at hadron colliders. D.” Springer Tracts Mod. Monogr. D. J. “Intermediate-mass Higgs searches in weak boson fusion.” arXiv:hep-ph/0702124. Rainwater. Kuhn and B. 2523 (2005) [arXiv:hep-ph/0504099]. “Collider phenomenology: Basic knowledge and techniques. [16] V. W. (Frontiers in physics. Krauss. Part. J. UK: Clarendon (2003) 538 p [29] R. Tait. “Seeking Sgluons. Tait. Jakobs. Stirling and B. M. Mod. Germany: Springer (1990) 559 p [27] C.2419 [hep-ph]. Spannowsky and T. Wells. M. Nachtmann. [19] D. Webber. “QCD — High Energy Experiments and Theory” Oxford. P.

E.” people. “Discovering the Higgs Boson in New Physics Events using Jet Substructure. Alwall et al. Spannowsky. Nucl.ch/spira/vorlesung/qcd [40] T.” arXiv:0912. Salam and M. Stelzer.” JHEP 0711. 040 (2004) [arXiv:hep-ph/0409146]. “QCD and jets” arXiv:hep-ph/0412013. Tonnesmann. R. Frixione and B. K. J. 124 (2009) [arXiv:0907. [41] G. “Fat Jets for a Light Higgs. 473 (2008) [arXiv:0706.psi. P. 029 (2002) [arXiv:hep-ph/0204244]. C 53. Part. A. Huston. 55. [33] G. Nason and C. Martin. Loch and M. “Mass determination in sequential particle decay chains. 111801 (2010) [arXiv:0910. P. Nucl. G. “Jets in hadron-hadron collisions. Hatakeyama. “A new method for combining NLO QCD with shower Monte Carlo algorithms. D. [38] S. P. “Performance of the Missing Transverse Energy Reconstruction and Calibration in Proton-Proton Collisions at a √ Center-of-Mass Energy of s = 7 TeV with the ATLAS Detector” ATLAS-CONF-2010-057 [43] B. [34] M. 070 (2007) [arXiv:0709. P. “Tools For The Simulation Of Hard Hadronic Collisions” Ann. Mangano and T.5472 [hep-ph]]. Sterman. Sci. [36] S. Lett. “Matching NLO QCD computations with Parton Shower simulations: the POWHEG method. “QCD. Kribs. S. [39] M. Ellis. [37] P.” JHEP 0909. J.web.” Phys. Plehn.” arXiv:hep-ph/0011256. Soper. [35] J. Salam.2569 [hep-ph]]. Webber. Rev. [32] D. Rev. 60. J.5307 [hep-ph]]. Nason. Frixione. Roy and M. D. . Phys.” arXiv:0906.4731 [hep-ph] [42] ATLAS collaboration.2447 [hep-ph]]. Phys. 104. T. “Matching NLO QCD computations and parton shower simulations” JHEP 0206..1833 [hep-ph]. “Basics of QCD perturbation theory.” Prog. [31] G. “Comparative study of various algorithms for the merging of parton showers and matrix elements in hadronic collisions” Eur. Oleari. Part. “Towards Jetography.” JHEP 0411. L. 484 (2008) [arXiv:0712. Webber. CERN-PH-TH-2005-074. Spannowsky. Spira.134 REFERENCES [30] S.2092 [hep-ph]]. 555 (2005).

Mod. “CMS technical design report.REFERENCES 135 [44] A. Webber. Stephens.2732 [hep-ph]]. Phys. Lester. “Model independent sparticle mass measurement at ATLAS. Murayama. Barr and C.” arXiv:hep-ph/0303191. www. [LHC/LC Study Group]. J. L. Dawson. Wang and I.” J. Smillie and B. G.ac. A 23. Rept. Lester. 2343 (2003) [arXiv:hep-ph/0304226]. I. K. Phys. 015 (2005) [arXiv:hep-ph/0501033]. 426. J. 4647 (2008) [arXiv:0802. and new physics.hep. Zerwas. [53] L.” CERN-THESIS-2004-003.” J. G 29. . [46] J. M. “A Review of the Mass Measurement Techniques proposed for the Large Hadron Collider. Rauch and D.3985 [hep-ph]]. “The Top quark. [45] B. Lafaye. Lester and P. C 54.2726 [hep-ph]].uk/˜lester/thesis/index. Weiglein et al. T. 617 (2008) [arXiv:0709. [47] S. 995 (2007).phy. “A Review of Spin Determination at the LHC. [48] H.” JHEP 0506. Barr.” Eur.” Phys. [51] A. “Measuring Supersymmetry. 069 (2005) [arXiv:hep-ph/0507170].html. T. volume II: Physics performance.” J. G in print [arXiv:1004.cam. Hagiwara. [49] G. [52] G. 47 (2006) [arXiv:hep-ph/0410364]. Miller and P. [50] C. R. Phys. “HELAS: HELicity amplitude subroutines for Feynman diagram evaluations” KEK-91-11. Watanabe and K. QCD. Phys. C. Osland. M. J. “Physics interplay of the LHC and the ILC.” Int. Phys. Plehn. “Distinguishing Spins in Supersymmetric and Universal Extra Dimension Models at the Large Hadron Collider. Gjelsten. . [CMS Collaboration]. D. [54] R.” JHEP 0510. G 34. Bayatian et al. “Measurement of the gluino mass via cascade decays for SPS 1a. J. Yavin. “m(T2) : The Truth behind the glamour.

94 Dirac delta distribution. 34 total 2 → 1. 50 collinear divergence. 15. 119 derivative interaction. 103 jet. 108 parton shower. 96 combinatorial background. 102 kinematic endpoint. 16 stability bound. 125. 78 covariant derivative. 113 convolution. 8 electroweak precision measurements ρ parameter. 129 statistical error. 113 jet mass. 17 non-linear transformation. 71. 53 Cauchy integral. 96. 93 azimuthal angle. 17 Feynman rules. 33. 96. 28 self coupling. 21 error RFit scheme. 6 helicity amplitudes. 20. 43 Higgs field. 49 Cuskosky cut rules. 108 unweighted events. 18 histogram. HELAS. 95. 25 triviality bound. 28 loop induced. 96. 68. 79 dimensional transmutation.absorptive integral. 89. 122 Klein–Gordon equation. 68 collinear limit. 12 Goldstone’s theorem. 52 angular ordering. 108 MadEvent. 109 anomalous dimension. 112 top tagger. 13 Higgs mass. 58 weighted events. 105 re-weighting. 120 DGLAP equation. 60. 119. 100. 100. 115. 8 infrared safety. 41 form factor. 16 equivalence theorem. 40. 97. 48 total 2 → 2. 39 LHC cross sections. 22 non-linear representation. 7 cross section hadronic. 87 collinear radiation. 117 PYTHIA. 111 Cambridge-Aachen algorithm. 16 Higgs mass. 54 inclusive rate. 30. 24 Higgs potential. 102. 10. 55. 17 Gaussian distribution. 37 eikonal approximation. 6 detector smearing. 84 asymptotic freedom. 94 Callan-Symanzik equation. 57 hypercharge. 41. 58 event generators ALPGEN. 97 Tevatron and LHC processes. 19 dimension-6 operators. 66 dipole radiation. 113 Higgs tagger. 13 quantum fluctuations. 108 SHERPA. 66 cascade decay. 100 HERWIG. 73 effective W approximation. 50 136 . 69. 101. 97. 53 Dirac matrices. 19 experimental upper bound. 26 collinear decay. 80–82. 122. 93 electric charge. 12. 82 dimensional regularization. 127 Cauchy distribution. 64 axial gauge. 109 gauge transformation linear transformation. 53 Goldstone boson. 112 fat jet. 34. 128 systematic error. 128 event generation negative weights. 115–117 Higgs boson branching ratios. 27 Higgs coupling. 96. 6. 37. 29 dark matter. 128 theoretical error. 23 dimension-6 CP basis. 91 solutions. 114 K factor. WIMP miracle.

Dirac mass. 87 renormalization scale. 104 phase space subtraction. 33. 14. 85. 6. 49. 87 scales factorization scale. 60. 8. 97. 22 particle width. 62 mass. 69. 28. 62 top quark mass. 50 gauge boson. 80. 6 Mellin transform. 109 vetoed shower. 81 subtracted Pq←g . 24 strong coupling. 79 propagator Breit–Wigner propagator. 70 space-like branching. 13 residue. 68 running coupling. 41 trigger. 72. 38 Monte Carlo importance sampling. 51 Breit–Wigner propagator. 73 ˆ unsubtracted Pg←q . 57 phase space mapping. 90 phase space. 18 renormalization MS scheme. 90. 12 gluon. 83 mini-jet veto. 104 Sudakov decomposition. 105 transverse momentum ordering. 55 renormalizable operators. 98. 90 Monte Carlo generator. 48 mass fermion mass. 52 spinor ???. 75 ˆ unsubtracted Pq←q . 53 parton shower backwards evolution. 61 strong coupling. 82 subtracted Pg←q . 69 massive photon. 52 narrow width approximation. 38. 78. 122. 7 perturbative unitarity. 33. 85 scaling logarithms. 66 rapidity.137 Laundau pole. 59 phase space slicing. 107 optical theorem. 61 wave function. 63. 9 MT2 construction. 105 supersymmetry. 24 likelihood. 82. 124 transverse mass. 37. 90 Monte Carlo integration. 53 parton densities. 52 renormalization group equation Higgs self coupling. 9 gauge boson mass. 47. 120 visible mass. 127 tagging jet. 83. 75 pseudo-rapidity. 103. 87 transverse tensor. 70 splitting kernel. 63 Goldstone boson. 63 QCD sum rules. 40. 76 spontaneous symmetry breaking. 42 transfer function. 129 luminosity. 23. 119. 76 ˆ unsubtracted Pq←g . 108 Pauli matrices. 13 Sudakov factor. 58 next-to-leading order corrections. 23. 33. 119 two Higgs doublet model. 14 . 52 squark mass. 60 Thomson limit. 80 ˆ unsubtracted Pg←g . 76 time-like branching. 83 soft gluon emission. 105. 70 plus subtraction. 65 resummation collinear logarithms. 121 mass factorization. 28 QCD perturbative series. 53 R ratio. 53 cutting. 53. 108 subtracted Pg←g . 128 Mandelstam variable. 89 scalar-photon mixing. 90. 58 Markov process. 80 subtracted Pq←q . 85 transverse momentum size. 26 phase space phase space generator. 29 Feynman i . 102. 60. 41. 5 scale artifact. 55 QCD field strength. 70. 92 truncated shower. 92 splitting no-splitting probability.

15. 11 neutral current. 12. 119 charged current. 11 .138 unitary gauge. 77 weak interaction. 15. 7. 10. 117 virtuality.

Sign up to vote on this title
UsefulNot useful