Journey To The Bound States

Journey to the Bound Statesa
Paul Hoyer
Department of Physics, POB 64, FIN-00014 University of Helsinki, Finland†
Guided by the observed properties of hadrons I formulate a perturbative bound state method
for QED and QCD. The expansion starts with valence Fock states (e+ e− , q q̄, qqq, gg) bound by
the instantaneous interaction of temporal gauge (A0 = 0). The method is tested on Positronium
atoms at rest and in motion, including hyperfine splitting at O α4 , electromagnetic form factors

and deep inelastic scattering. Relativistic binding is studied for QED in D = 1 + 1 dimensions,
demonstrating the frame independence of the DIS electron distribution and its sea for xBj → 0. In
QCD a homogeneous solution of Gauss’ constraint in D = 3 + 1 implies O αs0 confining potentials
for q q̄, q q̄g, qqq and gg states, whereas q q̄ q q̄ is unconfined. Meson states lie on linear Regge
trajectories and have the required frame dependence. A scalar bound state with vanishing four-
momentum causes spontaneous chiral symmetry breaking when mixed with the vacuum.
These lecture notes assume knowledge of field theory methods, but not of bound states. Brief
reviews of existing bound state methods and Dirac electron states are included. Solutions to the
exercises are given in the Appendix.
arXiv:2101.06721v1 [hep-ph] 17 Jan 2021
CONTENTS
I. Motivations and Outline 4

A. Hadrons vs. atoms 4
B. Outline 5
II. Brief survey of present QED approaches to atoms 6

A. Recall: The Hydrogen atom in introductory Quantum Mechanics 6
B. The Schrödinger equation from Feynman diagrams (rest frame) 7
1. Bound states vs. Feynman diagrams 7
2. Forming an integral equation 8
C. The Bethe-Salpeter equation 9
D. Non-relativistic QED 11
III. Dirac bound states 12

A. Weak vs. strong binding 12
B. The Dirac equation 12
C. Dirac states 14
0
D. * Dirac wave functions for central A potentials 16
E. * Coulomb potential V (r) = −α/r 18
F. * Linear potential V (r) = V 0 r 20
IV. Fock expansion of bound states in temporal (A0 = 0) gauge 21

A. Definition of the bound state method 21
1. Considerations 21
2. Choice of approach 22
a Expanded version of lectures presented at the University of Pavia in January 2020.

Slides are available at https://www.mv.helsinki.fi/home/hoyer/Talks.html
† paul.hoyer@helsinki.fi
2
B. Quantization in QED 23
1. Functional integral method 23
2. Canonical quantization 23
3. Temporal gauge in QED 24
C. Temporal gauge in QCD 25
1. Canonical quantization 25
2. Specification of temporal gauge in QCD 26
V. Applications to Positronium atoms 27

A. The |e− e+ i Fock states of Para- and Orthopositronium atoms 28
1. Definition of the Fock states 28
2. Translations 28
3. Rotations 28
4. Parity ηP 29
5. Charge conjugation ηC 29
6. Wave functions of Para - and Orthopositronium 30
B. The Schrödinger equation for Positronium at P = 0 30
C. Positronium with momentum P 31
1. Kinetic and potential energy 32
2. The transverse photon Fock state 32
3. The bound state condition 33
D. * Hyperfine splitting of Positronium at P = 0 34
− +
1. The transverse photon Fock state |e e γi contribution 35
2. Hyperfine splitting from annihilation: e− e+ → γ → e− e+ 36
E. * Electromagnetic form factor of Positronium atoms in an arbitrary frame 37
1. Parapositronium form factor 38
2. Positronium transition form factor 39
F. * Deep inelastic scattering on Parapositronium in a general frame 40
VI. QED in D = 1 + 1 dimensions 41

0
A. QED2 bound states in A = 0 gauge 42
1. Temporal gauge in D = 1 + 1 42
2. States and wave functions of QED2 43
3. Rest frame and non-relativistic limit 44
4. Solution for any M and P 44
5. Weak coupling limit 46
− +
6. Large separations between e and e 46
7. Bound state masses and duality 47
B. * Bound state form factors in QED2 48
1. Form factor definition and symmetry under parity 48
2. Gauge invariance 49
3
3. Lorentz covariance 49
C. * Deep Inelastic Scattering in D = 1+1 51
1. The Bj limit of the form factor 51
2. Numerical evaluation of the electron distribution 52
3. xBj → 0 limit of the electron distribution 52
VII. Applications to QCD bound states 54

A. The instantaneous potential of various Fock states 54
1. The q q̄ potential 55
2. The qqq potential 56
3. The gg potential 56
4. The q q̄g potential 57
5. Limiting values of the qqq and q q̄g potentials 57
6. Single quark or gluon Fock states 58
7. The potential of q q̄ q q̄ Fock states 58
B. Rest frame wave functions of q q̄ bound states 59
1. Bound state equation for the meson wave function Φαβ (x) 59
2. Separation of radial and angular variables 60
3. The 0−+ trajectory: ηP = (−1)j+1 , ηC = (−1)j 61
−+
Non-relativistic limit of the 0 trajectory wave functions 62
−− j+1 j+1
4. The 0 trajectory: ηP = (−1) , ηC = (−1) 62
Non-relativistic limit of the 0−− trajectory wave functions 63
5. The 0++ trajectory: ηP = (−1)j , ηC = (−1)j 63
++
Non-relativistic limit of the 0 trajectory wave functions 64
C. * q q̄ bound states in motion 65
1. The bound state equation 66
2. Solution of the P 6= 0 bound state equation for V (x) = 0 66
3. Boost of the state |M, P i for V (x) = 0 67
4. Solution of the P 6= 0 bound state equation at x⊥ = 0 68
D. Properties of the q q̄ bound states 69
1. String breaking and duality 69
2. Properties of the wave function at large separations r 70
3. Discrete mass spectrum 72
4. Parton picture for M V (r) 72
E. * Glueballs in the rest frame 74
F. Spontaneous breaking of chiral symmetry 76
1. M = 0 states with vanishing quark mass m = 0 76
2. Finite quark mass mu = md = m 6= 0 77
VIII. Bound state epilogue 79
Acknowledgments 80
4
A. Solutions to exercises 80
1. Order of box diagram 80
2. Contribution of the diagrams in Fig. 2(b,c) 80
3. Derivation of (3.6) 81
5. The expressions (3.24) for vacuum state 82
6. Derivation of the identities (3.39) 82
8. Gauge transformations generated by Gauss operator 83
9. Derive (5.8). 83
11. Verify (5.22). 84
− +
12. Derive the expression for |e e γ; q, si in (5.38). 85
13. Derive the expression (6.32). 85
14. Derive the expression (6.74). 86
15. Do the x-integral in (6.74) numerically for the parameters in Fig. 7, and compare. 87
16. Derive the q q̄g potential (7.22) 88
(0)
17. Derive the expression for HV |8 ⊗ 8i in (7.36) 89
18. Verify that the expression (7.60) for Φ−+ (x) satisfies the bound state equation (7.47) given the radial
equation (7.59). 90
19. Derive the coupled equations (7.100) from the bound state equation (7.98). 90
(P )
20. Derive the frame dependence (7.101) of ΦV =0 (x) using the boost generator K0z . 91
(P )
21. Show that Φ (τ ) given by (7.113) satisfies the BSE (7.99) at x⊥ = 0. 92
22. Prove the orthogonality relation (7.115) for states with wave functions satisfying the BSE (7.98). 93
23. Verify the expression (7.118) for the global norm of Φ−+ (x) in terms of F1 (r). 94
24. Verify the expressions (7.161) for radial functions H1 (r), H2 (r) and H3 (r). 94
References 95
I. MOTIVATIONS AND OUTLINE
A. Hadrons vs. atoms
Hadrons differ qualitatively from atoms due to their relativistic binding, confinement and spontaneous chiral symmetry
breaking. Deep inelastic scattering shows that hadrons have significant sea quark and gluon constituents. Nevertheless,
the hadron spectrum has atomic features, with quantum numbers determined by the valence quarks only. Nuclei are
multiquark states analogous to molecules, being comparatively loosely bound states of nucleons. The more recently
discovered heavy multi-quark (X, Y, Z) states tend to be associated with hadron thresholds, as would be expected for
weakly bound states of hadrons.
These properties should emerge in a correct description of hadrons as QCD bound states. Many aspects have indeed
been confirmed by numerical studies (see the reviews of lattice QCD by the Particle Data Group [1] and FLAG [2]).
Valuable insights have been obtained also through studies of models, especially the quark model [1], and in more
sophisticated approaches (see the summary [3]). Yet we still lack even a qualitative understanding of main aspects,
e.g., why the degrees of freedom of the non-valence constituents are not manifest in the hadron spectrum. The contrast
to the dense excitation spectrum of nuclei due to rotational and vibrational dof.’s is striking.
5
The observed, puzzling similarities between hadrons and atoms allows us to benefit from the understanding of QED
bound states, which gradually emerged since the beginnings of quantum field theory. Unfortunately, the fields of QED
and QCD bound states have grown apart [4]. Modern textbooks on the applications of QFT to particle physics hardly
mention bound states. There are seemingly solid reasons to believe that QED methods are irrelevant for hadrons.
My lectures are motivated by a concern that this conclusion might be premature. Let me briefly indicate why I do
not find some of the common arguments completely conclusive.
a. Hadrons are non-perturbative, whereas QED is perturbative.
In their review of QED bound state calculations Bodwin et al. [5] remark that “precision bound-state calculations are
essentially nonperturbative”. Perturbation theory for atoms needs to expand around an approximate bound state,
whose wave function is necessarily non-polynomial in α. This means that there is no unique perturbative expansion
for bound states, since a polynomial in α may be shifted between the initial wave function and the higher order
terms. Measurable quantities, such as the binding energy, have nevertheless unique expansions, being independent
of the initial wave function. For example, the hyperfine splitting between Orthopositronium (J P C = 1−− ) and
Parapositronium (0−+ ) is impressively known up to O α7 corrections, and is in agreement with accurate data [6–8]
7 4 8 ln 2 α5
6
∆Eb 5 6 1367 5197 2 221 2 1 53 α
= α − + − α ln α + − π + π + ln 2 − ζ(3) 2
me 12 9 2 π 24 648 3456 144 2 32 π
7α7 2 17 217 α7
ln α + O α7

− ln α + ln 2 − (1.1)
8π 3 90 π
The freedom of choice of the initial wave function was used in the evaluation of the higher order terms, by expanding
around states given by the Schrödinger equation. Analogously, hadrons may have a perturbative expansion with a
calculable initial state incorporating the relevant features, including confinement.
b. Confinement requires a scale ΛQCD , which can arise only from renormalization.
The form of the classical atomic potential, V (r) = −α/r, follows from α being dimensionless (~ = c = 1). Conversely,
a linear potential V (r) = V 0 r requires V 0 to have dimension GeV2 , and there is no such parameter in the QCD action.
Hence it seems impossible for an ab initio approach to QCD to give a confining, instantaneous classical potential.
Accordingly, the phenomenologically successful description of heavy quarkonia based on the Schrödinger equation
with the “Cornell potential” [9, 10],
4 αs
V (r) = V 0 r − with V 0 ' 0.18 GeV2 , αs ' 0.39 (1.2)
3 r
should not be viewed as a classical potential analogous to that of QED atoms. In section IV C 2 I argue that in QCD
we may impose a boundary condition on Gauss’ law which selects solutions with a spatially constant color field energy
density. The universal magnitude of the energy density defines V 0 .
c. The QCD coupling αs (Q) is large at low scales Q, excluding a perturbative expansion.
Standard perturbative determinations of αs (Q) are restricted to Q & mτ = 1.78 GeV, with αsM S (mτ ) ' 0.33 [1].
Since αs is not directly measurable its value depends on the theoretical framework. αs is scale independent when a
classical gluon field dominates loop contributions. The issue is thus whether there is such a regime for Q < Q0 , and
what the value of Q0 is. For Q0 ' 1 GeV αs freezes at value compatible with (1.2), and a perturbative expansion
may be relevant. Strong binding is then due to the confining potential, not to the Coulomb potential ∝ αs .
The above observations, together with other theoretical and phenomenological arguments [11–15] prompt me to
consider the possibility of a perturbative expansion for QCD bound states. Perturbation theory is our main analytic
tool in the Standard Model, and merits careful consideration. Bound states are interesting in their own right, providing
insights into the structure of QFT which are complementary to those of scattering. QED methods for atoms have
been developed over a long time, and are probably now close to optimal. A perturbative approach to hadrons raises
issues which so far have received less attention.
B. Outline
To help the reader navigate through these fairly extensive lectures I provide here brief characterizations of the various
chapters and sections. Those marked with a star * may be skipped in a first reading. Students are welcome to try
the exercises, whose solutions are given in Appendix A.
Chapter II is a brief survey of established QED bound state methods. The reduction of a non-relativistic two-particle
bound state to one particle in a central potential is recalled (II A). The Schrödinger equation is derived by summing
6
Feynman ladder diagrams (II B). The derivation of the Bethe-Salpeter bound state equation is sketched, and its
non-uniqueness noted (II C). The non-relativistic effective field theory method NRQED is introduced (II D).
Chapter III covers aspects of the Dirac bound states (III A). States are defined by time ordered Feynman diagrams,
with Z-diagrams giving rise to virtual pairs (III B). A Bogoliubov transformation of the free creation and annihilation
operators allows to describe the Dirac state with a single fermion (III C). The Dirac wave functions for central A0 (r)
potentials are given in terms of radial and angular functions (III D), and the explicit example of the Coulomb potential
worked out (III E). The case of a linear potential, for which the spectrum is continuous, is considered in section III F.
Chapter IV motivates and defines the approach to bound states used here: Quantization at equal time in temporal
(A0 = 0) gauge. A perturbative expansion around valence Fock states, bound by the instantaneous gauge potential
(IV A). Comparison of quantization procedures using covariant, Coulomb and temporal gauges in QED (IV B). The
vanishing of the color octet electric field E aL for color singlet states allows to include a homogeneous solution of Gauss’
constraint for each color component of the state. The boundary condition introduces a universal constant Λ (IV C).
Chapter V applies the bound state method to Positronium atoms. The states and wave functions are defined (V A).
The Schrödinger equation is derived for atoms at rest (V B) and in a general frame (V C). The O α4 Hyperfine
splitting between Ortho- and Parapositronia is evaluated in section V D. The Poincare covariance of the Positronium
form factor is demonstrated (V E), as well as that of deep inelastic scattering on Parapositronium (V F).
Chapter VI considers e− e+ bound states in D = 1 + 1 dimensions (QED2 ). The bound state equation is solved
analytically in a general frame (VI A). The gauge and Lorentz invariance of electromagnetic form factors is verified
(VI B). The electron distribution given by deep inelastic scattering is numerically evaluated in the rest frame of the
target and shown to agree with an earlier result in the Breit frame. DIS has a sea contribution for xBj → 0 (VI C).

Chapter VII applies the bound state method to QCD hadrons. The instantaneous O αs0 potential due to the
homogeneous solution of Gauss’ constraint is evaluated for several Fock states (q q̄, qqq, gg, q q̄g and q q̄ q q̄) (VII A).
The wave functions of q q̄ states in the rest frame are determined for all J P C quantum numbers (VII B). The bound
state equation and its boundary conditions for states with general momentum P 6= 0 is formulated (VII C). The states
lie on nearly linear Regge trajectories with parallel daughter trajectories. Highly excited states have a non-vanishing
overlap with multiple hadron states, with features that are consistent with the parton model, string breaking and
duality (VII D). The glueball (gg) spectrum has features similar to that of q q̄ mesons (VII E). There is a massless 0++
q q̄ state which has vanishing four-momentum in all frames. It may mix with the vacuum without violating Poincaré
invariance, giving rise to a spontaneous breaking of chiral symmetry (VII F).
Chapter VIII is a recapitulation and discussion of the principles followed in these lectues. Experienced readers may
profit from reading this chapter before the more technical parts.
II. BRIEF SURVEY OF PRESENT QED APPROACHES TO ATOMS
A. Recall: The Hydrogen atom in introductory Quantum Mechanics
In Introductory Quantum Mechanics we write the Hamiltonian for the Hydrogen atom as a sum of the electron and
proton kinetic energies, plus the potential energy. Since the atom is stationary in time the wave function may be
expressed as exp(−iEt)Φ(xe , xp ). The Schrödinger equation (including the mass contributions) is then
h ∇2e ∇2p i
me + mp − − + V (xe − xp ) Φ(xe , xp ) = EΦ(xe , xp ) (2.1)
2me 2mp
We then transform to CM and relative coordinates,

me r e + mp r p
R= r = re − rp (2.2)
me + mp
and get,
h ∇2R ∇2 i me mp
me + mp − − r + V (r) Φ(R, r) = EΦ(R, r) µ= (2.3)
2(me + mp ) 2µ me + mp
The dependence of the wave function on R and r may now be separated. Denoting the CM momentum of the bound
state by P we have Φ(R, r) = exp(iP · R)Φ(r). The total energy is given by the electron and proton masses, the
7
kinetic energy of the CM motion and an O α2 binding energy, E = me + mp + P 2 /2(me + mp ) + Eb with,

h ∇2r i
− + V (r) Φ(r) = Eb Φ(r) (2.4)
2µ
The transformation has reduced the dynamics to that of one particle in an external potential, and the bound states are
determined by the Schrödinger equation (2.4). The two-to-one particle reduction is possible only for non-relativistic
kinematics. For relativistic motion one needs to transform the times together with the positions of the electron and
proton in (2.2). The wave function ofqa Hydrogen atom with large CM momentum |P | & me + mp can nevertheless
be determined. The energy is E = P 2 + (me + mp + Eb )2 and the wave function Φ(r) depends on P (Lorentz
contraction), as we shall see in section V C.
B. The Schrödinger equation from Feynman diagrams (rest frame)
1. Bound states vs. Feynman diagrams
Bound states are (by definition) stationary in time and thus eigenstates of the Hamiltonian. The eigenstate condition
gives the Schrödinger equation (2.4). On general grounds, bound states also appear as poles in scattering amplitudes
at ECM = M − iΓ, where M is the bound state mass and Γ its width. E.g., the e− p → e− p scattering amplitude
has poles at the masses of the ground and all excited states of the Hydrogen atom. Since the binding energy Eb < 0
the poles are below the threshold for scattering, M < me + mp . QED scattering amplitudes can be calculated
perturbatively, in terms of Feynman diagrams. The expansion is defined by the perturbative S-matrix,
h Z ∞ i
Sf i = out hf, t → ∞| T exp − i dt HI (t) |i, t → −∞iin
−∞
Z
HI = e /ψ
dx ψ̄ eA (in QED) (2.5)
where the in and out states at t = ±∞ are free. Feynman diagrams of any finite order in the coupling e for the process
i → f are generated by expanding the time ordered exponential of the interaction Hamiltonian HI . The interaction
vertices are connected by free propagators.
Unfortunately the S-matrix boundary condition of free states excludes bound states, which are bound due to in-
teractions. There is no overlap between, say, an e+ e− Positronium atom, which has finite size, and a free e+ e−
state, which has infinite size. As a consequence, there are no Positronium poles in any Feynman diagram for the
e+ e− → e+ e− scattering amplitude. This is why (as I mentioned above) atoms may be called “non-perturbative”,
and their perturbative expansion differs from that of the S-matrix.
Nevertheless, it turns out that we can generate Positronium poles by (implicitly or explicitly) summing an infinite set
of Feynman diagrams. The poles then arise through the divergence of the sum. The simplest set of diagrams to sum
are the so-called “ladder diagrams” shown in Fig. 1.
FIG. 1. Ladder diagram expansion of the e− e+ → e− e+ scattering amplitude. Momenta are shown in the fermion direction,
e.g., the positron energy p02 > 0.
At first sight it seems curious that all the ladder diagrams of Fig. 1 can be of the same order in α, allowing the
series to diverge at any value of the coupling. This is indeed true only for the special kinematics of bound states. In
the restpframe all 3-momenta are of the order of the Bohrmomentum, e.g., |p1 | is of O (αm), and its kinetic energy
Ep1 = p21 + m2 ' m + p21 /2m differs from m by O α2 m . The exchanged momentum is similar, (q1 − p1 )0 ∼ α2 m,
2
|q 1 − p1 | ∼ αm. Each propagator contributes a factor of O 1/α , making the diagram with a single ladder of
8
O (1/α). In fact all ladder diagrams are of O (1/α) for bound state kinematics, whereas all non-ladder diagrams are
of higher order in α.
Exercise A 1: Convince yourself that the diagram with two ladders in Fig. 1 is of O (1/α), like the single ladder
diagram. Hint: The relevant loop momenta are commensurate with bound state kinematics.
In processes where the momenta are even lower than in bound states the propagators are further enhanced and
the ladder series in Fig. 1 diverges more strongly. This is the kinematic region where classical fields dominate, and
Feynman diagrams give non-leading contributions. Bound states are at the borderline between quantum and classical
physics.
2. Forming an integral equation
The expression for the sum of all ladder diagrams at leading O (1/α) may be formulated as an integral equation.
Bound state poles are just below threshold, 2m − M ∼ α2 m, so also the initial and final e± must be off-shell by
O α2 . Their propagators may be expressed using
p
/+m

1 X u(p, λ)ū(p, λ) v(−p, λ)v̄(−p, λ) p
2 2
= 0
+ 0
Ep = p2 + m2 (2.6)
p − m + iε 2Ep p − Ep + iε p + Ep − iε
λ
At leading order in α we need only retain the e− pole in the electron propagator and the e+ pole for the positron, e.g.,
1/(p01 − Ep1 + iε) ∝ α−2 and 1/(−p02 + Ep2 − iε) ∝ α−2 . In the following I show for conciseness only the spinors of the
external propagators, e.g., u(p1 , λ1 ) for the incoming electron. The analysis is done in the rest frame, p1 + p2 = 0.
For bound state kinematics the spinors are trivial at leading order in α,
! !
p
/+m χλ √ χλ
u(p, λ) ≡ p = 2m + O (α)
Ep + m 0 0
! !
−p/+m 0 √ 0
v(p, λ) ≡ p = 2m + O (α) χ̄λ = iσ2 χλ (2.7)
Ep + m χ̄λ χ̄λ
The relation between χ̄λ and χλ follows from charge conjugation, see (5.15) below. In the single ladder diagram of
Fig. 2(a) the Dirac structure of the electron line is ū(q 1 , λ01 )γ µ u(p1 , λ1 ) = 2mδλ1 ,λ01 δ µ,0 + O (α). The positron line

gives a similar result. In the photon propagator (q1 − p1 )2 = (q10 − p01 )2 − (q 1 − p1 )2 = −(q 1 − p1 )2 + O α4 m2 .
The amplitude for this diagram is then, at lowest order in α, denoting p ≡ p1 , q ≡ q 1 and suppressing the conserved
helicities,
−e2
A1 (p, q) = (2m)2 ≡ (2m)2 V (q − p) (2.8)
(q − p)2
where the notation indicates that V (p − q) is the single photon exchange potential in momentum space. The factor
(2m)2 is due to my normalization of the spinors in (2.7).
A similar calculation of the double ladder amplitude in Fig. 2(b) gives, with P 0 ≡ p01 + p02 the CM energy,
d3 `
Z
1
A2 (p, q) = A1 (p, `) 0 V (q − `) (2.9)
(2π)3 P − 2E` + iε
Exercise A 2: Derive the expression (2.9) for A2 (p, q). Why does diagram 2(c) contribute only at a higher
order in α?
It is now straightforward to see that the amplitude with n ladders may be expressed as
d3 `
Z
1
An (p, q) = 3
An−1 (p, `) 0 V (q − `) ≡ An−1 (p, `) S(`) V (q − `) (2.10)
(2π) P − 2E` + iε
9
FIG. 2. (a) Single ladder diagram for the e− (p1 )e+ (p2 ) → e− (q1 )e+ (q2 ) scattering amplitude. (b) Double ladder diagram. (c)
A diagram with crossed exchanges. Momenta are shown in the fermion (arrow) direction.
where a convolution over ` is understood in the last expression. Summing over all ladder diagrams we get
∞
X
A(p, q) = An (p, q) = A1 (p, q) + A(p, `) S(`) V (q − `) (2.11)
n=1
which has the form of a Dyson Schwinger equation [16]. A bound state pole in the e− e+ → e− e+ amplitude has in
the rest frame the structure
Φ† (p)Φ(q)
A(p, q) = + ... (2.12)
P0 − M
where M is the bound state mass. Φ† (p) and Φ(q) are the bound state wave functions, expressing the coupling to the
initial and final states with relative momenta p and q. Eq. (2.11) gives a bound state equation for the wave function
since A1 (p, q) has no pole. Cancelling the factor Φ† (p)/(P 0 − M ) on both sides and extracting a factor P 0 − 2Eq
from the wave function (which gives the “truncated” wave function) we have (at P 0 = M )
d3 ` −e2 d3 ` −e2
Z Z
1
Φ(q)(M − 2Eq ) = 3
Φ(`)(M − 2E` ) 2
= 3
Φ(`) (2.13)
(2π) M − 2E` + iε (q − `) (2π) (q − `)2
where I used the explicit expression of the potential from (2.8). This is the Schrödinger equation in momentum space.
We can go to coordinate space using
Z i(q−`)·x
1 3 e
= d x
(` − q)2 4π|x|
Z
Φ(q) ≡ d3 x Φ(x) e−iq·x (2.14)
Defining the binding energy Eb by M = 2m + Eb and expanding Eq ' m + q 2 /2m on the lhs. of (2.13) we have as in
(2.4),
∇2 α
Eb + Φ(x) = V (x)Φ(x) V (x) = − (2.15)
m |x|
C. The Bethe-Salpeter equation
The Bethe-Salpeter equation [16, 17] is a generalization of the integral equation (2.13), obtained by considering all
Feynman diagrams (not just the ladder ones), and without assuming non-relativistic kinematics. It is thus a formally
exact framework for bound states with explicit Poincaré covariance, and so far is the only bound state equation which
applies in any frame. Boost covariance requires that the relative time of the constituents in the wave function is frame
dependent. It is possible to project on constituents at equal time in any one frame. I give a brief summary here,
following [18]. A comprehensive review may be found in [19].
Let GT be a truncated Green function (i.e., without external propagators) for a 2 → 2 process. Denote by K a 2 → 2
truncated “kernel” and by S a 2-particle propagator. Then if K is 2-particle irreducible, i.e., does not have two parts
10
that are only connected by S, we have the Dyson-Schwinger identity
GT = K + GT S K (2.16)
By construction, any Feynman diagram on the lhs. is either contained in K or has the structure of GT S K. Eq.
(2.16) may be regarded as an exact equation for GT since it holds for the complete sum of Feynman diagrams. The
product GT S K implies a convolution over the momenta and helicities of the two particles in the propagator S. If
GT has a bound state pole it must have the form (2.12). Identifying the residues on both sides of (2.16) gives the
Bethe-Salpeter equation for the truncated wave function shown in Fig. 3(a),
d4 `
Z
Φ(P, q) = Φ(P, `) S(`) K(q − `) (2.17)
(2π)4
p
The bound state momentum P = (P 0 , P ) satisfies P 0 = P 2 + M 2 by Poincaré invariance. The wave function
FIG. 3. (a) The exact Bethe-Salpeter equation (2.17) for a bound state of momentum P . The black dots on the fermion
propagators in S represent full self-energy corrections, and the irreducible kernel K has contributions of all orders in α. (b)
The Bethe-Salpeter equation with free propagators and a single photon exchange kernel.
Φ(P, q) depends on the relative energy q 0 of the constituents or, quivalently, on the difference in time x0 of the
constituents in coordinate space,
d4 q
Z
Φ(P, x) = Φ(P, q) e−iq·x (2.18)
(2π)4
The B-S wave function for the e− e+ component can be expressed as a matrix element of electron field operators
between the bound state |P i and the vacuum,
h0|T{ψ̄β (x2 )ψα (x1 )} |P i ≡ e−iP ·(x1 +x2 )/2 Φαβ (P, x1 − x2 ) (2.19)
It thus describes a component of the bound state which has an electron at time x01 at position x1 , and a positron at
time x02 at position x2 . For x01 = x02 the B-S wave function describes an e− e+ Fock state, belonging to the Hilbert
space of states defined at equal time and expressed in the free field basis.
A Lorentz transformation ΛP = (P 0 0 , P 0 ) transforms the electron field as ψ 0 (x0 ) = S −1 (Λ)ψ(Λx), where the 4 × 4
matrix S(Λ) is the Dirac spinor representation of the transformation, familiar from the Dirac equation [16]. Hence
the B-S wave function transforms as
Φ0 (P 0 , x01 − x02 ) = S(Λ)Φ(P, x1 − x2 )S −1 (Λ) (2.20)
Poincaré covariance thus allows the constituents be taken at equal time (x01 = x02 ) in at most one frame. The
B-S equation has “abnormal” solutions [19–21], which vanish in the non-relativistic limit and seem related to the
dependence on relative time. Their physical significance is not fully understood.
11
Expanding the propagator S and the kernel K in powers of α allows to solve the B-S equation perturbatively. There are
many formally equivalent expansions. The Dyson-Schwinger equation (2.16) determines K in terms of the truncated
Green function GT and S,
1
K= GT = GT − GT S GT + . . . (2.21)
1 + GT S
The choice of S together with the standard perturbative expansion of GT fixes the expansion of the kernel. My remark
in section I A that the perturbative expansion for bound states is not unique refers to this more precise statement.
The B-S equation is difficult to solve when the kernel K(`, q) depends on (` − q)0 , which implies retarded interactions.
Already the single photon exchange kernel has denominator (`0 − q 0 )2 − (` − q)2 . The (` − q)0 dependence arises
from the propagation of transversely polarized photons, which create intermediate e− e+ γ states that affect the e− e+
B-S wave function. No analytic solution of the B-S equation is known even for single photon exchange with free
propagators, illustrated in Fig. 3(b).
Caswell and Lepage [22] noted that S may be chosen so that the kernel K(`, q) is static (independent of ` 0 and q 0 )
at lowest order. This reduced the B-S equation to a Schrödinger equation which has an analytic solution, simplifying
the calculation of higher order corrections in the rest frame.
Bound states with arbitrary CM momenta are needed for scattering processes, e.g., form factors. It is then non-
trivial to take into account the frame dependence of the wave function [23]. Model dependent assumptions are often
made, but there are few studies based on field theory. The B-S framework was used in [24] to determine the frame
dependence of equal-time Positronium wave functions. It showed the importance of the |e+ e− γi intermediate state,
and demonstrateed (apparently for the first time) that the standard Lorentz contraction of the |e+ e− i Fock component
holds at lowest order. I verify this result using a Fock state expansion in section V C.
D. Non-relativistic QED
The realization that there are many formally equivalent versions of the Bethe-Salpeter equation underlined the need
for physical judgement in the choice of perturbative expansion. The most accurate data for atoms relates to their
binding energies. These may be calculated in the rest frame, where the constituents have mean velocities of O (α). It
has been estimated that the probability for Positronium electrons to have 3-momenta |p| & m is of order α5 ∼ 10−11
[25]. It is thus well motivated to expand the QED action in powers of |p|/m. This defines the effective theory of
non-relativistic QED (NRQED) [26, 27]. The constraints of gauge and rotational invariance allow only a limited
number of terms at each order of |p|/m in the Lagrangian. The expansion begins as,
n D2 D4 e e
LN RQED = − 41 Fµν F µν + χ† i∂t − eA0 + + 3
+ c1 σ · B + c2 ∇·E
2m 8m 2m 8m2
ie o d1 d2
+ c3 σ · (D × E − E × D) χ + 2 (χ† χ)2 + 2 (χ† σχ)2 + . . .
8m2 m m
+ positron and positron-electron terms. (2.22)
The photon action −Fµν F µν /4 is as in QED, since photons are relativistic. The field χ is a two-component Pauli
spinor, representing the electron part (upper components) of the QED Dirac field. There are further terms involving
the lower (positron) components, as well as terms mixing the positron and electron fields. D = ∇ − ieA is the
covariant derivative, E and B are the electric and magnetic field operators.
The NRQED action implies a finite momentum cutoff Λ ∼ m. The contributions of momenta |p| & Λ to low energy
dynamics are described by the UV-divergent terms in LN RQED . Their coefficients ci and di are process-independent
and may thus be determined (as expansions in powers of α) by comparing the results of QED and NRQED for selected
processes, such as a scattering amplitude close to threshold. Since both theories are gauge invariant, one may use
different gauges in their calculations. Coulomb gauge ∇ · A = 0 has been found to be convenient for bound state
calculations in NRQED, while covariant gauges (e.g., Feynman gauge) is efficient for scattering amplitudes.
The expansion in powers of |p|/m shows that the Coulomb field A0 is the dominant interaction. In (2.22) the vector
potential A, although contributing at the same order in α as A0 , is suppressed by a power of m. The choice of
initial bound state approximation is then evident: The lowest order terms in (2.22) give the familiar non-relativistic
Hamiltonian of the Hydrogen atom in Quantum Mechanics. The Schrödinger equation with the A0 potential is solved
exactly, and the terms of higher orders in |p|/m are included using Rayleigh-Schrödinger perturbation theory.
12
NRQED has turned out to be an efficient calculational method for the binding energies of atoms. It has, in particular,
allowed the impressive expression (1.1) for the hyperfine splitting of Positronium. The evaluation of the higher order
corrections are discussed in [8, 25, 26, 28–31]. A corresponding effective field theory method NRQCD has been applied
to heavy quarkonia [32, 33].
The NRQED approach is limited to the rest (or non-relativistic) frames of weakly bound states. NRQCD does not
provide insight into the dynamics of color confinement.
III. DIRAC BOUND STATES
A. Weak vs. strong binding
The QED atoms discussed above were weakly coupled (α 1). We have only a limited understanding of the dynamics
of strong binding in QFT. Some features are known in D = 1 + 1 dimensions (QED2 ), where the dimensionless
parameter is e/m [34–36]. For e/m 1 the e+ e− states are weakly bound and approximately described by the
Schrödinger equation. For e/m 1 on the other hand the spectrum is that of weakly interacting bosons. This may
be qualitatively understood since the large coupling locks the fermion degrees of freedom into compact neutral bound
states. In√the limit of e/m → ∞ (the massless Schwinger model) QED2 has only a pointlike, non-interacting massive
(M = e/ π) boson. The physical hadron spectrum does not resemble the strong binding limit of QED2 .
Solving the relativistic Bethe-Salpeter equation is complicated by the dependence of the kernel on the relative time
of the constituents (II C). The time dependence is due to the exchange of transversely polarized photons. In chapter
IV I take this into account through a Fock expansion of the bound state, keeping the instantaneous (Coulomb) part
of the interaction within each Fock state.
The Dirac equation has no retardation effects since the potential Aµ is external, i.e., fixed. A space-dependent potential
Aµ (x) breaks translation invariance, so there are no eigenstates of momentum. Nevertheless, Dirac solutions with
large potentials give insights into relativistic binding. For a linear potential eA0 (x) = V 0 |x| it has long been known
[37] (but is rarely mentioned) that the Dirac spectrum is continuous. I discuss this case in section III F.
Klein’s paradox [16, 38, 39] signals an essential difference between the Schrödinger and Dirac equations. For potentials
of the order of the electron mass (i.e., relativistic binding) the Dirac wave function does not describe a single electron.
The state has e+ e− pairs which are not constituents in the usual (non-relativistic) sense. As noted in [40] the Dirac
wave function should (when possible) be normalized to unity, regardless of the number of pairs. The Dirac pairs do
not add degrees of freedom to the Dirac spectrum, which corresponds to that of a single electron. This motivates the
study of the states described by the Dirac wave functions in section III C.
B. The Dirac equation
The Dirac equation
(i∂/ − m − eA)ψ(x)
/ =0 (3.1)
should be distinguished from the operator equation of motion for the electron field, given by δSQED /δ ψ̄(x) = 0. The
c-numbered equation (3.1) studied by Dirac in 1928 [41, 42] is a relativistic version of the Schrödinger equation, where
Aµ (x) is an external, classical field. The condition (3.1) implies that propagation in the field Aµ (x) is singular for
electrons with wave function ψ(x). Scattering in the field is explicit in a perturbative expansion,
i i i i
= − /
ieA + ... (3.2)
/ /
i∂ − m − eA / /
i∂ − m i∂ − m /
i∂ − m
For time-independent potentials Aµ (x) the static solutions ψ(t, x) = exp(−itM )Ψ(x) have both positive and negative
energy eigenvalues M . The corresponding wave functions Ψ and Ψ satisfy
Ψn (x) = Mn γ 0 Ψn (x)

/
− i∇ · γ + m + eA(x) (3.3)
Ψn (x) = −M n γ 0 Ψn (x)

/
− i∇ · γ + m + eA(x) (3.4)
13
0
−itp
where Mn , M n ≥ 0. The free (Aµ = 0) solutions are given by the
p spinors (2.7) as ψ(x) = e Ψpλ (x) = u(p, λ)e−ip·x
0
and ψ(x) = eitp Ψpλ (x) = v(p, λ)eip·x with Mp = M p = p0 = p2 + m2 . The solutions with negative kinetic energy
−M p are related to positrons. For potentials Aµ & m the wave function ψ(x) has both positive and negative energy
components, due to contributions of e− e+ pairs (section III C).
The Dirac equation with a Coulomb potential can be obtained from a sum of Feynman ladder diagrams, analogously to
the Schrödinger equation [43–45]. There are some instructive differences, however. Relativistic two-particle dynamics
cannot be reduced to that of a single particle in an external field, as in (2.2). We must therefore consider a limit
where the mass of one particle goes to infinity. The recoil of the heavy particle may then be neglected. The heavy
particle gives rise to a static potential in its rest frame.
Consider again the diagrams in Fig. 2. Let the mass of the lower (antifermion) line be mT and its charge be eZ.
We take mT → ∞ keeping the electron (fermion) momenta p1 , q 1 fixed. Thep initial momentum of the antifermion
0
is p2 = (mT , 0). Since q 2 = p1 − q 1 is fixed as mT → ∞ the energy q2 = m2T + q 22 = mT + O (1/mT ). Thus
kinematics ensures that no energy is transferred from the heavy target to the electron, i.e., p01 = q10 up to O (1/mT ).
In the diagrams of Fig. 2(b,c) the loop integral converges even without the antifermion propagator. Hence the limit
mT → ∞ can be taken in the integrand. The antifermion spinors are non-relativistic so v̄(p2 , λ2 )(−ieZ)γ µ v(q 2 , λ02 ) '
−ieZ 2mT δ µ,0 δλ2 ,λ02 . The Born diagram of Fig. 2(a) is then, for large mT and relativistic electron momenta,
−ieγ 0
A1 (p1 , q 1 ) = −ieZ 2mT ū(q 1 , λ01 ) u(p1 , λ1 ) (3.5)
(q 1 − p1 )2
This corresponds to single scattering in the field of of the heavy particle with charge −eZ.
When the electron is non-relativistic the positive energy pole of its propagator (2.6) dominates. Then the diagram
of Fig. 2(c) with crossed photons is suppressed compared to the uncrossed diagram of Fig. 2(b). Now the crossed
diagram does contribute and is required to get the result
d3 ` i(/̀ + m)
Z
1 1
A2 (p1 , q 1 ) = i(eZ)2 2mT 3
ū(q 1 , λ01 )(−ieγ 0 ) (−ieγ 0 )u(p1 , λ1 ) (3.6)
(2π) (` − p1 ) ` − m + iε (q 1 − `)2
2 2 2
corresponding to double scattering in the external potential.
Exercise A 3: Derive (3.6), and convince yourself that also the exchange of three photons reduces to scattering
in an external potential. Hint: You need only consider the antifermion line, since the upper part is the same
for the uncrossed and crossed diagrams.
For ladders with n exchanges all n! diagrams with arbitrary crossings of the photons contribute. This means that
the Bethe-Salpeter equation (2.17) reduces to the Dirac equation as mT → ∞ only for kernals of infinite degree in α
(containing arbitrarily many crossed photons). The B-S equation can, however, be modified so that it does reduce to
the Dirac equation even for finite kernels [44].
In full QED a large charge eZ is screened by the creation of e+ e− pairs. The 2 → 2 ladder diagrams that give the
Dirac equation do not describe true pair production. The e+ e− pairs in the Dirac state which are implied by Klein’s
paradox [16, 38, 39] must therefore be virtual. The pairs only arise when the diagrams are time ordered, which is
required to determine a state at an instant of time. Time ordering the electron propagator (2.6) gives a positive and
negative energy part,
0
/ + m)e−ip t
dp0 (p
Z
1 Xh −itEp itEp
i
S(t, p) ≡ i 2 = θ(t) u(p, λ)ū(p, λ) e − θ(−t) v(−p, λ)v̄(−p, λ) e (3.7)
2π p − m2 + iε 2Ep
λ
In strong potentials the electron can scatter into a negative energy state which evolves backward in time. This
corresponds to an intermediate e− e+ e− state, as illustrated in Fig. 4(b). In weakly coupled bound states, described
by the Schrödinger equation, such higher Fock components are suppressed.
Multiple scattering gives rise to Fock states with any number of intermediate e+ e− pairs. Despite its apparent
one-particle nature the Dirac wave function describes many pairs in the free Fock state basis. In order to see this
more explicitly we need to define the Dirac states in terms field operators. The following study is based on [46] and
previously published in [15].
14
FIG. 4. Time ordered diagrams for the double scattering (3.6) in an external, static potential. (a) The intermediate electron has
positive energy. (b) The intermediate electron has negative energy, corresponding to the creation and subsequent annihilation
of an e+ e− pair. This is often referred to as a “Z-diagram”.
C. Dirac states
The Dirac wave functions define eigenstates of the Dirac Hamiltonian,

→
Z

/
HD (t) = dx ψ̄(t, x) − i∇ · γ + m + eA(x) ψ(t, x) (3.8)
where ψ(t, x) now is the electron field with the canonical anticommutation relation
†
ψα (t, x), ψβ (t, y) = δα,β δ 3 (x − y)

(3.9)
This field may be expanded in the standard operator basis, which creates/annihilates free e± states,
Z
dk Xh
ik·x −ik·x †
i
ψα (t = 0, x) = uα (k, λ)e b k,λ + v α (k, λ)e dk,λ (3.10)
(2π)3 2Ek
λ
n o n o
bp,λ , b†q,λ0 = dp,λ , d†q,λ0 = 2Ep (2π)3 δ 3 (p − q)δλ,λ0 (3.11)
I take the classical, c-numbered potential Aµ (x) to be time independent. There are no physical (propagating) photons.
Since the Hamiltonian is quadratic in the fermion fields it can be diagonalized [47].
The positive (3.3) and negative (3.4) energy Dirac wave functions determine e− and e+ states defined at t = 0 by
Z X
|Mn i = dx ψα† (x)Ψn,α (x) |Ωi ≡ c†n |Ωi (3.12)
α
Z X †
Ψn,α (x)ψα (x) |Ωi ≡ c̄†n |Ωi

M n = dx (3.13)
α
Charge conjugation transforms the electron field as
Cψ † (t, x)C † = iψ T (t, x)γ 2 (3.14)
Hence
Z Z
C |M i = dx ψ T (x)iγ 2 Ψ(x) |Ωi = dx ΨT (x)iγ 2 ψ(x) |Ωi (3.15)

has the form of M in (3.13), with wave function Ψ(x) = iγ 2 Ψ∗ (x). This wave function satisfies the Dirac equation
(3.3) with M → −M and eAµ → −eAµ , as expected for a positron.
The vacuum state |Ωi is an eigenstate of the Hamiltonian with eigenvalue taken to be zero,
HD |Ωi = 0 (3.16)
15
Two equivalent expressions for |Ωi are given in (3.24) below. Using
← →
HD , ψ † (x) = ψ † (x)γ 0 (i∇ · γ + m + eA) [HD , ψ(x)] = −γ 0 (−i∇ · γ + m + eA)ψ(x)

/ / (3.17)
we see that both states (3.12) and (3.13) are eigenstates of the Dirac Hamiltonian with positive eigenvalues,
HD |Mn i = Mn |Mn i Mn > 0

HD M n = M n M n Mn > 0 (3.18)
In terms of the wave functions in momentum space,

Z Z
dp dp
Ψn (x) = Ψn (p)eip·x Ψn (x) = Ψn (p)eip·x (3.19)
(2π)3 (2π)3
the eigenstate operators defined in (3.12) and (3.13) can be expressed as
Ψ†n (p) u(p, λ)bp,λ + v(−p, λ)d†−p,λ ≡ Bnp bp + Dnp d†p

X
cn = (3.20)
p Z
X dp X
≡
p
(2π)3 2Ep
X † λ
bp,λ u† (p, λ) + d−p,λ v † (−p, λ) Ψn (p) ≡ B np b†p + Dnp dp

c̄n = (3.21)
p
In the second expressions on the rhs. a sum over the repeated index p ≡ (p, λ) is implied. In the weak binding
limit (|p| m) the positive energy spinor wave function Ψn has only upper components, whereas Ψn has only lower
components. Then |Mn i is a single electron state, whereas M n is a single positron state.
The operators cn and c̄n are related to b, d via the Bogoliubov transformations (3.20) and (3.21). Using the commu-
tation relations (3.11) and the orthonormality of the Dirac wave functions we see that they obey standard anticom-
mutation relations,
Z
dp
Ψm,α (p) uα (p, λ)u†β (p, λ) + vα (−p, λ)vβ† (−p, λ) Ψn,β (p) =
X †
cm , c†n = Ψ† (p)Ψn,α (p) = δmn

p
(2π)3 m,α
c̄m , c†n = 0

Z
dp
c̄m , c̄†n = Ψ̄† (p)Ψ̄n,α (p) = δmn

(3.22)
(2π)3 m,α
Inserting the completeness condition for the Dirac wave functions into the Dirac Hamiltonian (3.8) gives,
X
Mn c†n cn + M̄n c̄†n c̄n

HD = (3.23)
n
Exercise A 4: Derive (3.23).
The expression for the vacuum state may be found using the methods in [47]. HD |Ωi = 0 when, in terms of the B
and D coefficients defined in (3.20) and (3.21),
−1
h i h i
|Ωi = N0 exp − b†q B −1 qn Dnr d†r |0i = N0 exp − d†r D )rn B nq b†q |0i

(3.24)
Sums over the repeated indices q, n, r are implied in the exponents, and N0 is a normalization constant. The pertur-
bative vacuum satisfies bp |0i = dp |0i = 0. The vacuum state |Ωi describes the distribution of the e+ e− pairs that
arise through perturbative contributions such as Fig. 4(b). It is a formal expression, involving a sum over all states
−1
n and the inverted matrices B −1 qn and D )rn . In the weak binding limit Dnr → 0, B nq → 0 and |Ωi → |0i.

The vacuum is “empty” in the bound state basis: cn |Ωi = c̄n |Ωi = 0. The pairs appear only in bases which do not
diagonalize the Hamiltonian, such as the free basis generated by the b† and d† operators.
16
Exercise A 5: (a) Show the equivalence of the two expressions for |Ωi in (3.24). Hint: Prove that Bmp B np +
Dmp Dnp = 0. (b) Prove that HD |Ωi = 0. Hint: Note that bp essentially differentiates the exponents in (3.24).
The Dirac bound states (3.12) may be expressed in terms of their electron and positron (“hole”) distributions,
Z
dp X
e− † +

|Mn i = 3 n (p, s)bps + en (p, s)d−ps |Ωi
(2π) 2Ep s
e− †
n (p, s) = u (p, s)Ψn (p) e+ †
n (p, s) = v (−p, s)Ψn (p) (3.25)
with momentum space wave functions Ψn (p) defined as in (3.19). The corresponding electron and positron densities
dΩp p2 X ∓
Z
ρn (e∓ , p) ≡ |e (p, s, )|2 (3.26)
(2π)3 2Ep s
are normalized so that

Z ∞
dp ρn (e− , p) + ρ+ +

n (e , p) = 1 (3.27)
0
D. * Dirac wave functions for central A0 potentials
The wave functions Ψ(x) of Dirac bound states in rotationally symmetric potentials eA0 (x) = V (r) with A(x) = 0
satisfy (α ≡ γ 0 γ)
(−iα · ∇ + mγ 0 )Ψ(x) = M − V (r) Ψ(x)

(3.28)
The states may be characterized by their mass M , angular momentum j, j z ≡ λ and parity ηP = ±1. The angular
momentum operator in the fermion representation is
Z
J = dx ψ † (x) J ψ(x) (3.29)
where J is the sum of the orbital L and spin S angular momenta (which are not separately conserved),
J = L + S = x × (−i∇) + 21 γ5 α (3.30)
Operating on the states |M, jλi in (3.12) we get

Z Z
J |M, jλi = dx ψ † (x) J Ψjλ (x) |Ωi J 2 |M, jλi = dx ψ † (x) J 2 Ψjλ (x) |Ωi (3.31)
The Dirac 4-spinor wave functions are thus required to satisfy
J 2 Ψjλ = j(j + 1)Ψjλ J z Ψjλ = λΨjλ (3.32)
The parity operator is defined by
Pψ † (t, x)P† = ψ † (t, −x)γ 0

Z
P |M, jλi = dx ψ † (x) γ 0 Ψjλ (−x) |Ωi = ηP |M, jλi (3.33)
Hence the Dirac wave functions of states with parity ηP should satisfy
γ 0 Ψjλ (−x) = ηP Ψjλ (x) (3.34)

17
Denoting x = r(sin θ cos ϕ, sin θ sin ϕ, cos θ) and x̂ ≡ x/r, the angular dependence of Ψjλ (x) may be expressed using
the orthonormalized 2-spinors [16],

√ 1 
λ− 2
j + λ Y 1 (θ, ϕ) 
1  j− 2
φjλ+ (θ, ϕ) = √ 
 
1

2j  √ λ+ 2 
j − λ Y 1 (θ, ϕ)
j− 2

√ 1 
λ−
j − λ + 1 Y 12 (θ, ϕ)
1  j+ 2 
φjλ− (θ, ϕ) = p  = σ · x̂ φjλ+ (θ, ϕ) (3.35)
 
1

2(j + 1)  √ λ+ 
− j + λ + 1 Y 12 (θ, ϕ)
j+ 2
The ± notation refers to j = ` ± 21 , where ` is the order of the spherical harmonic function Y`m (θ, ϕ), which becomes
the conserved orbital angular momentum in the non-relativistic limit. In the standard notation J ± = J x ± iJ y ,
σ · L = 12 σ + L− + 12 σ − L+ + σ z Lz
p
L± |`, λi = (` ∓ λ)(` ± λ + 1) |`, λ ± 1i (3.36)
it is straightforward to verify that
1
(
cj+ = j − 2
σ · L φjλ± = cj± φjλ±
cj− = −(j + 23 )
(L + 12 σ)2 φjλ± = j(j + 1)φjλ± (3.37)
1
The 4-spinor Dirac wave functions Ψjλ± (x) describing the states |M, jλ±i of (3.12) with ηP = (−1)j∓ 2 may now be
defined in terms of two radial functions,
!
φjλ± (θ, ϕ)
Ψjλ± (x) = Fj± (r) + iα · x̂ Gj± (r) (3.38)
0
Since [J , α · x̂] = 0 the angular momentum quantum numbers (3.32) are ensured by (3.37). The parity ηP = (−1)j∓1/2
follows from γ 0 α · (−x̂) = α · x̂γ 0 and φjλ± (π − θ, ϕ + π) = (−1)j∓1/2 φjλ± (θ, ϕ).
The eigenvalue condition (3.18) determines the bound state equation for Ψjλ± (x). For this we need the relations
1
−iα · ∇ = −i(α · x̂) ∂r − α · x̂ × L
r
2 1
−iα · ∇(iα · x̂) = + ∂r + γ5 α · L (3.39)
r r
Exercise A 6: Derive the identities (3.39). Hint: Use αi αj = δ ij + iγ5 ijk αk .
We may furthermore use

! ! ! !
φjλ± 0 0 φjλ±
α · x̂ × L = = −i = −cj± iα · x̂ (3.40)
0 σ · x̂ × L φjλ± (σ · x̂)(σ · L)φjλ± 0
with cj± given in (3.37), while γ5 α · L = 2S · L contributes cj± with unit Dirac matrix. The mγ 0 term in HD gives
m(Fj± − iα · x̂ Gj± ). Identifying the coefficients in HD |M i = M |M i of the two Dirac structures in (3.38) we get
j + 3/2 j − 1/2
+ ∂r Gj+ = (Mj+ − V − m)Fj+ − ∂r Fj+ = (Mj+ − V + m)Gj+ (3.41)
r r
j + 3/2 j − 1/2
− + ∂r Fj− = (Mj− − V + m)Gj− − − ∂r Gj− = (Mj− − V − m)Fj− (3.42)
r r
18
These reduce to second order equations for F and G separately. Suppressing the subscripts j±,
2 V0 h c(c + 1) cV 0 i
F 00 + + F 0 + (M − V )2 − m2 − − F =0
r M −V +m r2 r(M − V + m)
2 V0 h (c + 1)(c + 2) (c + 2)V 0 i
G00 + + G0 + (M − V )2 − m2 − + G=0 (3.43)
r M −V −m r2 r(M − V − m)
At the potentially singular points M − V ± m = 0 the solutions behave as (M − V ± m = 0)β , with β = 0 or β = 2,

and are thus locally normalizable there.
e jλ+ (x) ≡ γ5 Ψjλ+ (x) solves this equation with m → −m and the
If Ψjλ+ (x) solves the Dirac equation (3.28) then Ψ
same eigenvalue Mj+ . This shows up as a symmetry of the bound state equations. Using φjλ− = σ · x̂ φjλ+ ,
! !
0 σ · x̂ φjλ+
e jλ+ (x) = Fj+ (r) + iα · x̂ Gj+ (r)
Ψ = α · x̂ Fj+ (r) + i Gj+ (r)
φjλ+ 0
!
φjλ−
= i Gj+ (r) − iα · x̂ Fj+ (r) = Ψjλ− (x) Fj− → iGj+ , Gj− → −iFj+ (3.44)
0
Eq. (3.42) is indeed seen to transform into (3.41) when m → −m, Mj− → Mj+ and the j− radial wave functions
are replaced with the j+ functions as indicated in (3.44). This means that the solution of (3.42), if allowed by
the quantum numbers, is given by Fj− (r, m) = Gj+ (r, −m), Gj− (r, m) = −Fj+ (r, −m) with the same eigenvalue
Mj− = Mj+ . ”Squaring” the Dirac equation (3.28) equation by multiplying it with −iα · ∇ + mγ 0 gives
− ∇2 + m2 Ψ(x) = (M − V )2 + iα · (∇V ) Ψ(x)

(3.45)
Since this equation depends on m only via m2 the eigenvalue M is independent of the sign of m. The degeneracy
Mj− = Mj+ is familiar in the case of a Coulomb potential, to which I turn next.
E. * Coulomb potential V (r) = −α/r
There is a standard and elegant method [16] for finding the Dirac spectrum in the case of a Coulomb potential
V (r) = −α/r. One starts from the squared Dirac equation (3.45), determining the eigenvalues of the Dirac matrix
iα · (∇V ). For V (r) = −α/r all terms in (3.45) may then be formally identified, based on their r-dependence, with
those of the Schrödinger equation,
1 α
− ∇2 − Ψ(x) = Eb Ψ(x) (3.46)
2m r
The known solution of this equation allows to determine the masses M of the Dirac states,
 !2 − 21
α
Mnj = m 1 + q  (3.47)
1
n − (j + 2) + (j + 21 )2 − α2
The principal quantum number n = 1, 2, 3, . . . and j ≤ n − 21 . There are two states for each mass, Mnj+ = Mnj− ,
except only Mnj+ for j = n − 12 . For α → 0 we recover the non-relativistic Schrödinger result which depends only on
n, Mnj = m(1 − α2 /2n2 ).
√
For r → ∞ (3.43) reduces to F 00 + (M 2 − m2 )F = 0, implying F (r → ∞) ∼ exp(−r m2 − M 2 ). Hence M < m for
normalizable solutions, as in (3.47). I illustrate using the the radial wave functions F (r) and G(r) of the states with
maximal spin j = n − 12 , and of the first radial excitation with j = n − 23 . The wave functions Ψnjλ± (x) are expressed
in terms of the radial functions Fnj± and Gnj± as in (3.38), with the angular functions φjλ± given in (3.35).
19
1
Maximal spin, j = n − 2
mγ
q
Mj+ 21 ,j,+ = γ≡ (j + 21 )2 − α2
j + 12
p αm
Fj+ 12 ,j,+ (r) = N1 rγ−1 exp(−µr) µ≡ m2 − M 2 =
j + 21
µ (2µ)1+2γ h µ2 i−1
Gj+ 21 ,j,+ (r) = N1 rγ−1 exp(−µr) N12 ≡ 1+ (3.48)
M +m Γ(1 + 2γ) (M + m)2
This state is not degenerate, i.e., there are no radial functions Fj+1/2,j,− , Gj+1/2,j,− . In momentum space (3.19) the
wave functions of the n = 1 ground state M, n = 1, j = 12 , λ, + are, with χ 12 = (1 0)T and χ− 12 = (0 1)T ,
!
χλ
Ψ1,1/2,λ,+ (p) = f (p) + α · p̂ g(p)
0
√ sin[δ(1 + γ)/2] αm + ip
f (p) = 4π N1 Γ(1 + γ) exp(iδ) ≡
p(α2 m2 + p2 )(1+γ)/2 αm − ip
√
4π α ∂ h sin(δγ/2) i
g(p) = − N1 Γ(γ) (3.49)
1+γ ∂p p(α2 m2 + p2 )γ/2
The electron and positron density distributions (3.26) are then
p2
ρ(e∓ , p) = Ep (f 2 + g 2 ) ± m(f 2 − g 2 ) ± 2p f g

2
(3.50)
4π Ep
The electron density ρ(e− , p) is strongly dominant. For α = 1/137 the contribution of the positron density to the
state normalization is a mere 3.2 · 10−12 . Even for α = 0.999 (see Fig. 5) the positron contributes only 3% to the
normalization. The j = 21 eigenvalue M1, 12 ,+ is complex for α > 1.

FIG. 5. Electron and positron densities (3.50) in the M, 1, 12 , λ, + Dirac state (3.49) with V (r) = −α/r and α = 0.999. The

positron density is multiplied by a factor 10.
3
First radial excitation, j = n − 2
m(1 + γ)
q
Mj+ 32 ,j,± = q γ≡ (j + 21 )2 − α2
(j + 12 )2 + 1 + 2γ
h α(1 + 2γ) i p αM
Fj+ 32 ,j,+ (r) = N2+ rγ−1 exp(−µr) r − µ≡ m2 − M 2 =
(M + m)(j + 21 − γ) + αµ 1+γ
h µ 2j + 1 − 2γ α(1 + 2γ) i
Gj+ 23 ,j,+ (r) = N2+ rγ−1 exp(−µr) r− 1 (3.51)
M +m 2α (M + m)(j + 2 − γ) + αµ
20
The degenerate state with the same mass and spin but opposite parity has, as argued at the end of the previous
section, radial functions with F− (m) = G+ (−m) and G− (m) = −F+ (−m) (up to the normalizations N2± ),
h µ 2j + 1 − 2γ α(1 + 2γ) i
Fj+ 23 ,j,− (r) = N2− rγ−1 exp(−µr) r− 1
M −m 2α (M − m)(j + 2 − γ) + αµ
h α(1 + 2γ) i
Gj+ 23 ,j,− (r) = −N2− rγ−1 exp(−µr) r − 1 (3.52)
(M − m)(j + 2 − γ) + αµ
Unbound states, M 2 − m2 > 0

States with masses M > m are unbound. The radial equations (3.41) imply for all |M, njλ, +i states,
p
Fnj+ (r → ∞) = N rβ exp(±iµr) µ = M 2 − m2
∓iµ β iα
Gnj+ (r → ∞) = N r exp(±iµr) β = ∓p −1 (3.53)
M +m 1 − m2 /M 2
In the absence of a normalization condition the mass spectrum is continuous. At large r (where V → 0) the solution is
a spherical wave with momentum p = ±µ, modulated by the phase factor rβ+1 . The norm r2 |Ψ|2 tends to a constant
at large r.
F. * Linear potential V (r) = V 0 r
Hadron phenomenology, and particularly the description [9, 10] of quarkonia using the Schrödinger equation with the
Cornell potential (1.2), motivates studying Dirac states with a linear potential, eA0 (x) = V (|x|) = V 0 r, A = 0. The
solutions of the Dirac equation for polynomial potentials have since the 1930’s [37] been known to be quite different
from those of the Schrödinger equation.
I first recall the solutions of the Schrödinger equation (3.46) for a linear potential. The ` = 0 wave function φ(r)
satisfies
h 1 2 2 i
− ∂r + ∂r + V 0 r φ(r) = Eb φ(r) (3.54)
2m r
The normalizable solutions are given by an Airy function,
N
Ai (2mV 0 )1/3 (r − Eb /V 0 )

φ(r) = (3.55)
r
The
discrete values of the binding energy Eb are determined by requiring φ(r = 0) to be regular, which implies
Ai − (2mV 0 )1/3 Eb /V 0 ) = 0. Since the potential grows linearly with r all states are bound (confined), and their wave
functions vanish exponentially for r → ∞.
The Dirac radial functions on the other hand are oscillatory at large r, as seen from (3.41) and (3.43),
im2
F (r → ∞) ' N rβ exp i(M − V )2 /2V 0

β=− −1
2V 0
G(r → ∞) ' i F (r → ∞) (3.56)
This result (and its complex conjugate) is independent of the quantum numbers n, j, ±. I retained some non-leading
terms in the exponent for ease of notation. The essential feature is that
F (r → ∞) ∼ −iG(r → ∞) ∼ N rβ exp iV 0 r2 /2 r2 |F (r → ∞)|2 = r2 |G(r → ∞)|2 = N 2

(3.57)
Thus the normalization integral diverges even though the potential is confining. In the absence of a normalization
constraint the mass spectrum is continuous for all M , in contrast to the discrete spectrum of the Schrödinger equation.
A Dirac electron state (3.12) has Fock components with positrons in the vacuum |Ωi (3.24). This is seen perturbatively
in time ordered Z-diagrams such as in Fig. 4(b). The distribution of the positrons is traced by the d-operator in the
state creation operator c†n (3.20), motivating the definitions of the e∓ (p, s) probabilities in (3.25).
21
A linear potential confines electrons, limiting their distribution to distances where V 0 r . M − m. The same potential
repulses positrons, pushing them to large distances with kinetic energy big enough to cancel their negative potential,
p−V 0 r ∼ M +m. The exponent exp(ir V 0 r/2) of F (r → ∞) in (3.57) implies momenta increasing with r as p ∼ V 0 r/2.
The relation between the F and G radial functions allows to verify that the e+ distribution indeed dominates at large
momenta (equivalent to large r),
e− (p, s) u† (p, s)Ψ(p)

lim = lim =0 (3.58)
|p|→∞ e (p, s) |p|→∞ v † (−p, s)Ψ(p)
+
Exercise A 7: Derive (3.58) for a state with j = 1/2 and parity ηP = +1. Hint: Calculate the momentum
space wave function (3.19) for |p| → ∞ using the stationary phase approximation.
An equivalent interpretation is that the wave function is a superposition of electrons, confined to low r, and acceler-
ating/decelerating positrons at large r, whose negative kinetic energy balances the positive potential. The spectrum
is continuous because the positron energies are continuous.
The tunneling of the e+ to r ' 0 is exponentially suppressed with growing fermion mass m. Hence if the initial
condition G(r = 0)/F (r = 0) of the radial equations (3.41) is such as to include a positron contribution (beyond
the tunneling rate) the wave function will grow rapidly with r and start oscillating with an amplitude which is
exponentially large in m. The precise values of G(r = 0)/F (r = 0) which suppress the positrons at r = 0 correspond
to the discrete bound state masses M of the normalizable solutions of the Schrödinger equation. All other values of
M give, in the m → ∞ limit, wave functions which grow exponentially with r.
These properties were confirmed quantitatively in D = 1 + 1 dimensions, using the analytic expression of the wave
function in terms of confluent hypergeometric functions [15, 48].
IV. FOCK EXPANSION OF BOUND STATES IN TEMPORAL (A0 = 0) GAUGE
A. Definition of the bound state method
1. Considerations
Perturbative expansions depend on the choice of a lowest order approximation. The perturbative S-matrix expands
around free states, which works well for scattering amplitudes. Bound states are stationary in time and thus, in a
sense, the very opposites of scattering amplitudes. QED approaches to atoms have been thoroughly considered, with
conceptual milestones such as the Bethe-Salpeter equation [17] (1951), the realization that it is not unique [22] (1978)
and NRQED [26] (1986). Even in a first approximation atoms are described by wave functions that are non-polynomial
in α. NRQED expands around states defined by the Schrödinger equation.
Poincaré symmetry can be explicitly realized only for generators that mutually commute. Equal-time bound states are
defined as eigenstates of the Hamiltonian, which in their rest frame have explicit (kinematic) symmetry under space
translations and rotations. The frame dependence of these states is defined by boosts. It is not trivial to determine
the boost generators of atoms, which are spatially extended. Alternatively, states with general CM momentum P
may be found as eigenstates of the Hamiltonian.pFull rotational invariance is lost for P 6= 0, but the requirement of
a correct P -dependence of the energy, E(P ) = P 2 + M 2 , is a strong constraint. Field theory ensures covariance,
as emphasized by Weinberg in the preface of [40]:
“The point of view of this book is that quantum field theory is the way it is because (aside from theories like string
theory that have an infinite number of particle types) it is the only way to reconcile the principles of quantum mechanics
(including the cluster decomposition property) with those of special relativity.”
The examples below will illustrate how subtly Poincaré covariance is realized for bound states. Much remains to be
understood in this regard. Using “relativistic wave equations” is not sufficient, as demonstrated in [49].
There are many formally equivalent approaches to bound states. In the following I briefly motivate and define my
choice, guided by the properties of atoms and hadrons. Some further comments are given in chapter VIII.
22
2. Choice of approach
Hamiltonian eigenstates
Bound states can be identified in two equivalent but distinct ways: As poles in Green functions or as eigenstates of the
Hamiltonian. The former involves propagation in time and space, allowing for explicit Poincaré invariance as in the
Dyson-Schwinger framework. The propagation of bound state constituents is complicated by their state-dependent,
mutual interactions. A Hamiltonian framework distinguishes time from space. The eigenstate condition involves no
propagation in time, and Poincaré invariance emerges dynamically. I shall use the method of Hamiltonian eigenstates,
akin to traditional quantum mechanics and NRQED.
Instant time quantization

Quantum states are traditionally defined at an instant of time t (IT), but relativistic states are also commonly defined
at equal Light-Front (LF) time t + z [50, 51]. The latter is natural in the description of hard collisions, where a
single probe (virtual photon or gluon) interacts with the target at a fixed LF time. LF states are described by frame-
independent wave functions, whereas IT wave functions transform dynamically under boosts. On the other hand, the
LF choice of z-direction breaks rotational invariance, making angular momentum (other than J z ) dynamic even in
the rest frame. The so called “zero modes” require special attention in LF quantization [52–54].
A perturbative approach allows to study the frame dependence of IT wave functions at each order of the expansion.
Rest frame states can be characterized by their angular momentum J 2 and J z . Quantization is simpler at equal
ordinary time. For these reasons I choose IT quantization.
Temporal (A0 = 0) gauge

Gauge theories have a local action, but the gauge may be fixed in all of space at each instant of time. The gauge
dependent fields A0 and AL then give rise to an instantaneous potential, such as the Coulomb potential V (r) = −α/r.
The potential allows to define an initial bound state without the complications of retardation. In temporal gauge the
longitudinal electric field E L = −∂t AL is given by a constraint for each physical state. This clearly separates the
instantaneous from the propagating fields.
Fock expansion
States are conventionally defined by their expansion in a complete basis of Fock states. In temporal gauge the Fock
state constituents are fermions and transversely polarized photons or gluons. The gauge constraint (Gauss’ law)
determines the longitudinal electric field within each Fock state.
For strong potentials the number of Fock constituents depends on the basis. The Dirac state (3.12) has an infinite
number of constituents (3.24) in the free state basis due to Z-diagrams, Fig. 4(b). In the basis of the cn operators
defined by the Bogoliubov transform (3.20) the same Dirac state has a single constituent c†n |0i. I shall define a fermion
Fock state as ψ † (t, x) |0i, without specifying the expansion of the field in creation and annihilation operators.
Initial state
I take the valence Fock state as the initial bound state of the perturbative expansion. For Positronium this means
|e− e+ i bound by the −α/r potential. Hadron (|q q̄i , |qqqi) quantum numbers correspond to their valence quarks.
Higher order corrections in α will involve Fock states with a correspondingly larger number of constituents, as well
as loop corrections to Fock states with fewer constituents. At each order of α the usual cancellation of collinear
singularities between states with different numbers of constituents should thus be ensured.
23
B. Quantization in QED
1. Functional integral method
Relativistic field theory is commonly defined using functional methods. Green functions are given by a functional
integral over the fields weighted by the exponent of the action, exp(iS/~). In QED the photon propagator is thus
Z
Dµν (x1 , x2 ) = D(Aρ )D(ψ̄, ψ) eiSQED /~ Aµ (x1 )Aν (x2 )
Z
d4 x − 14 Fµν F µν + ψ̄(i∂/ − m − eA)ψ

SQED = / Fµν = ∂µ Aν − ∂ν Aµ (4.1)
A gauge fixing term SGF must be added to the action SQED for the integral to be well defined. Explicit Poincaré
invariance is maintained with
Z
SGF = − 21 λ d4 x (∂µ Aµ )2 (4.2)
Expanding around free (hence Poincaré covariant) states gives the standard perturbative expansion of Green functions
in terms of Feynman diagrams.
This approach is well suited for scattering amplitudes. It is less convenient for bound states, for which free states are
a poor approximation. As a consequence, the perturbative S-matrix (2.5) lacks bound state poles at any finite order.
The poles can be generated through the divergence of an infinite sum of Feynman diagrams (or through an equivalent
integral equation), as discussed in section II B. However, it seems unlikely that confinement will be recovered in an
expansion starting with free quarks and gluons.
The covariant gauge fixing (4.2) introduces a time derivative ∂0 A0 for the A0 field, which is absent from SQED . This
makes A0 propagate in time like the transverse components AT , at the price of introducing a time-dependent kernel
in the Bethe-Salpeter equation (section II C). The ∂0 A0 term is avoided in Coulomb gauge, ∇ · A = 0. The field
equation for A0 (Gauss’ law) in Coulomb gauge,
δSQED
= −∇2 A0 (x) − eψ † ψ(x) = 0 (4.3)
δA0 (x)
defines A0 non-locally in terms of the electron field operator. For Positronium this gives the Coulomb potential
eA0 = −α/r, which allows an analytic solution of the Schrödinger equation and is the leading order interaction. The
evaluation of higher order corrections in Coulomb gauge is rather complicated, especially for QCD. See [55] for a
study of non-relativistic quarkonia based on the Bethe-Salpeter equation in Coulomb gauge.
2. Canonical quantization
The conjugate fields πα of the fields ϕα in the Lagrangian density L(ϕ, ∂ϕ) are defined by
∂L(ϕ, ∂ϕ)
πα (t, x) = (4.4)
∂[∂0 ϕα (t, x)]
Equal time (anti)commutation relations are imposed on the (fermion) boson fields,
[ϕα (t, x), πβ (t, y)]± = iδαβ δ 3 (x − y) [ϕα (t, x), ϕβ (t, y)]± = [πα (t, x), πβ (t, y)]± = 0 (4.5)
and the Hamiltonian is given by
Z hX i
H(t) = d3 x πα ∂0 ϕα (t, x) − L(ϕ, ∂ϕ) (4.6)
α
In gauge theories the conjugate field of A0 vanishes since L is independent of ∂0 A0 . The covariant gauge fixing term
(4.2) adds ∂0 A0 , giving the conjugate fields
π 0 = −λ ∂µ Aµ (4.7)
π i = −F 0i (4.8)
24
This allows to define covariant commutation relations for the gauge field, the non-vanishing ones being
[Aµ (t, x), πν (t, y)] = i gµν δ 3 (x − y) (4.9)
The unphysical (gauge) degrees of freedom are removed by constraining physical states not to involve photons with
time-like or longitudinal polarizations (Gupta-Bleuler method, see [16] for details).
Canonical quantization can be carried out also in Coulomb gauge, ∇ · A = 0. Due to the lack of a conjugate A0
field this requires constraints which modify the commutation relations, see [40] for QED. The generalization to QCD
is discussed in [56], demonstrating how terms related to Faddeev-Popov ghosts arise. The same study also addresses
temporal gauge (A0 = 0), which is an axial gauge without ghosts.
Temporal gauge simplifies canonical quantization since the absence of both A0 and its conjugate allows standard
commutation relations for the spatial gauge field components A. The gauge condition preserves rotational invariance
and, most importantly for the present application, Gauss’ law is implemented as a constraint on physical states which
determines E L , not as an operator relation like (4.3). The constraint is trivially satisfied for the vacuum (E L = 0),
whereas in Coulomb gauge A0 |0i would have an overlap with |e− e+ i. I next discuss canonical quantization in temporal
gauge for QED, and consider QCD in section IV C.
3. Temporal gauge in QED
The canonical quantization of QED in temporal gauge (A0 = 0) is described in [56–60]. The action (4.1) determines
the electric field E i = F i0 = −∂0 Ai to be conjugate (4.4) to Ai (i = 1, 2, 3), and iψ † to be conjugate to ψ. This gives
the canonical commutation relations without constraints,
i †
E (t, x), Aj (t, y) = iδ ij δ(x − y)

ψα (t, x), ψβ (t, y) = δαβ δ(x − y) (4.10)
All other (anti)commutators vanish. The Hamiltonian in temporal gauge is

Z Z
H(t) = dx E i ∂0 Ai + iψ † ∂0 ψ − L = dx 21 E i E i + 14 F ij F ij + ψ † (−iαi ∂i − eαi Ai + mγ 0 )ψ

(4.11)
Gauss’ operator is defined as usual by the derivative of the action wrt. A0 ,
δSQED
G(x) ≡ = ∂i E i (x) − eψ † ψ(x) (4.12)
δA0 (x)
but G(x) = 0 (Gauss’ law) is not an operator condition, since A0 = 0 is fixed by the gauge condition. The operator
relation ∂i E i (x) = eψ † ψ(x) would not even be compatible with the commutation relations (4.10).
The condition A0 = 0 does not completely fix the gauge, since it allows time independent gauge transformations
parametrized by Λ(x): A → A + ∇Λ(x). Gauss’ operator G(x) turns out to generate such transformations. An
infinitesimal, time independent gauge transformation δΛ(x) is represented by the unitary operator,
Z
U (t) = 1 + i dy G(t, y)δΛ(y) (4.13)
Exercise A 8: Show using the commutation relations (4.10) that U (t) (4.13) transforms the A(t, x) and ψ(t, x)
fields as required for a time-independent infinitesimal gauge transformation.
Constraining the physical states to satisfy
G(x) |physi = ∇ · E(x) − eψ † ψ(x) |physi = 0

(4.14)
ensures that they are invariant under time-independent gauge transformations. A physical state remains physical
under time evolution since, as may be verified, G(x) commutes with the Hamiltonian (4.11),
[G(t, x), H(t)] = 0 (4.15)

25
The electric field can be separated into its transverse and longitudinal parts, E = E T + E L , with ∇ · E T = 0. Gauss
constraint (4.14) then allows to solve for E L ,
Z
e
E L (t, x) |physi = −∇x dy ψ † ψ(t, y) |physi (4.16)
4π|x − y|
This seems like the instantaneous electric field −∇A0 in Coulomb gauge, ∇ · A = 0. The difference is that Gauss’
law is an operator equation in Coulomb gauge, whereas here it is a constraint on the physical states. The constraint
specifies E L (t, x) for each state |physi at all positions x at a given time t. The electric field of the physical vacuum
vanishes in temporal gauge,
Z
e
i
EL (x) |0i = −∂ix
dy ψ † ψ(t, y) |0i = 0 (4.17)
4π|x − y|
since the vacuum state has no net charge at any position.
The Hamiltonian (4.11) has an instantaneous part determined by Gauss’ constraint,
Z Z
QED
h e ih e i
HV 2
|physi ≡ 2 dx E L (x) |physi = 2 dxdydz ∂ix
1 1
ψ † ψ(y) ∂ix ψ † ψ(z) |physi
4π|x − y| 4π|x − z|
e2
Z
†
= 12 dxdy ψ ψ(x) ψ † ψ(y) |physi

(4.18)
4π|x − y|
HV |physi contributes a potential which depends only on the instantaneous positions of the electrons and positrons,
regardless of their momenta (which may be relativistic). The other terms of H determine the propagation of the
transverse photons and fermions in time, as well as the transitions between them.
This method can be applied to atoms in any frame. Given that non-valence Fock states are suppressed by powers of
α, calculations with a given degree of precision require to include a limited number of terms in the Fock expansion.
In section V I illustrate the method by considering several aspects of Positronium at rest and in motion. In section
VI I study the strongly bound states of QED in D = 1 + 1 dimensions.
C. Temporal gauge in QCD
1. Canonical quantization
The canonical quantization of QCD in temporal gauge A0a = 0 proceeds as in QED [56–60]. The QCD action is
Z
SQCD = d4 x − 14 Fµν a
Faµν + ψ̄(i∂/ − m − g A
/ a T a )ψ a
= ∂µ Aaν − ∂ν Aaµ − gfabc Abµ Acν

Fµν (4.19)
The electric field Eai = Fai0 = −∂0 Aia is conjugate to Aai = −Aia , giving the equal-time commutation relations
h i
Eai (t, x), Ajb (t, y) = iδab δ ij δ(x − y)
A†
ψα (t, x), ψβB (t, y) = δ AB δαβ δ(x − y)

(4.20)
The a, b (A, B) are color indices in the adjoint (fundamental) representation of SU(3). The Hamiltonian is
Z Z
†
HQCD = dx Eai ∂0 Aai + iψA ∂0 ψA − LQCD = dx 12 Eai Eai + 14 Faij Faij + ψ † (−iα · ∇ + mγ 0 − gα · Aa T a )ψ

(4.21)
where
Z Z
2
dx 14 Faij Faij = i
+ ∂i ∂j )Aja + gfabc (∂i Aja )Aib Ajc + 14 g 2 fabc fade Aib Ajc Aid Aje
1
dx 2 Aa (−δij ∇ (4.22)
has both longitudinal and transverse gluon fields.

Gauss’ operator
δSQCD
Ga (x) ≡ = ∂i Eai (x) + gfabc Aib Eci − gψ † T a ψ(x) (4.23)
δA0a (x)
26
generates time-independent gauge transformations similarly as in QED (4.13), which leave the gauge condition A0a = 0
invariant. The longitudinal electric field E aL is fixed by constraining physical states to be invariant under the gauge
transformations generated by Ga (x),
Ga (x) |physi = 0 (4.24)
This constraint is independent of time since Gauss’ operator commutes with the Hamiltonian, [Ga (t, x), H(t)] = 0. It
constrains the longitudinal electric field for physical states,
∇ · ELa (x) |physi = g − fabc Aib Eci + ψ † T a ψ(x) |physi

(4.25)
We may solve for E aL analogously1 as for QED in section IV B 3,
Z
g
E aL (x) |physi = −∇x dy Ea (y) |physi
4π|x − y|
Ea (y) = −fabc Aib Eci (y) + ψ † T a ψ(y) (4.26)

The contribution of the longitudinal electric field to the QCD Hamiltonian (4.21) is then
Z Z
2 αs
HVQCD |physi ≡ 12 dx E aL |physi = 12 dydz Ea (y)Ea (z) |physi (4.27)
|y − z|
2. Specification of temporal gauge in QCD
There is a relevant difference between QED and QCD which needs to be considered when determining the longitudinal
electric field from the QCD gauge constraint (4.25). To illustrate, compare the expectation value of the field in an
e− e+ Fock component of Positronium and in an analogous color singlet q q̄ component of a meson at t = 0,
− +
e e ≡ ψ̄α (x1 )ψβ (x2 ) |0i (4.28)
X
|q q̄i ≡ ψ̄αA (x1 )ψβA (x2 ) |0i (4.29)
A
The Dirac components α, β are irrelevant here and will be suppressed. Repeated color indices are summed. Note that
“color singlet” refers to global gauge transformations2 , the local temporal gauge being fixed by (4.16) and (4.26).
The expectation values of the QED (4.16) and QCD (4.26) longitudinal electric fields in these states are, using the
canonical commutation relations for the fermions and recalling that E L |0i = 0,
e e
he− e+ |E L (x) e− e+ = −∇x he− e+ |e− e+ i

− (4.30)
4π|x − x1 | 4π|x − x2 |
g g
hq q̄|E aL (x) |q q̄i = −∇x − a
hq q̄|ψ̄A (x1 )TAB ψB (x2 ) |0i ∝ Tr T a = 0 (4.31)
4π|x − x1 | 4π|x − x2 |
In QED the charges of e− and e+ give rise to the expected dipole electric field, while in QCD the expectation value of
an octet field in a singlet state vanishes everywhere. Comparing similarly3 the instantaneous potentials HVQED (4.18)
and HVQCD (4.27),
α
he− e+ |HVQED e− e+ = − he− e+ |e− e+ i

(4.32)
|x1 − x2 |
αs αs
hq q̄|HVQCD |q q̄i = − a
hq q̄|ψ̄A (x1 )TAB a
TBC ψC (x2 ) |0i = −CF hq q̄|q q̄i (4.33)
|x1 − x2 | |x1 − x2 |
These are the Coulomb potentials of QED and QCD, again as expected. The electron feels only the positron field,
and each quark of a given color interacts with its antiquark of opposite color. The sum over the potential energies of
all color-anticolor components A in (4.29) gives the Casimir CF = (Nc2 − 1)/2Nc of the fundamental representation.
1 At higher orders in g one needs to take into account the contribution of E L on the rhs. of (4.25). For large gauge fields this leads to
the issue of Gribov copies [61], but they do not appear in a perturbative expansion.
2 The global SU(3) transformations should not be regarded as a subgroup of the local ones, see Ch. 7 of [60].
3 The singular “self-energy” contributions ∝ 1/0 are independent of x1 , x2 and subtracted.
27
The solution of the QED gauge constraint (4.14), the longitudinal electric field (4.16), is determined using the physical
boundary condition that the electric field vanishes at spatial infinity. This boundary condition is no longer evident
for QCD, since the expectation value (4.31) of the color electric field in any case vanishes at all x, due to the sum
over quark colors. There seems to be no compelling reason to require that the gauge field of each color-anticolor
component A of the state (4.29) should vanish at spatial infinity.
The gauge constraint (4.25) fully determines E aL only given a boundary condition at spatial infinity. E aL may be
specified by the particular solution (4.26) and a homogeneous solution E aH which satisfies
∇ · E aH |physi = 0 (4.34)
There is apparently only one homogeneous solution which is invariant under translations and rotations,
Z
E aH (x) |physi = −κ ∇x dy x · y Ea (y) |physi (4.35)
where Ea (y) is defined in (4.26) and the normalization κ is independent of x, but may depend on the state |physi.
The complete longitudinal electric field is then
Z h g i
E aL (x) |physi = −∇x dy κ x · y + Ea (y) |physi (4.36)
4π|x − y|
and its contribution to the Hamiltonian (4.21) is
Z Z n Z on Z
i i
h g i h g i o
HV ≡ 1
2 dx Ea,L Ea,L = 1
2 dx ∂ix dy κ x · y + Ea (y) ∂ix dz κ x · z + Ea (z)
4π|x − y| 4π|x − z|
Z n h i αs o
dydz y · z 12 κ2 dx + gκ + 1
R
= 2 Ea (y)Ea (z) (4.37)
|y − z|

where the termsR of O gκ, g 2 were integrated by parts. The term of O κ2 is due to an x-independent field energy
density. It is ∝ dx but irrelevant provided it is universal, i.e., the same for all Fock components of all bound states.
This determines the normalization κ in (4.36) for each state |physi, up to a universal scale Λ.

The scale Λ is unrelated to the coupling g, so the gκ term in (4.37) may be viewed as an instantaneous O αs0 potential.
All relevant symmetries, in particular exact Poincaré invariance, must appear at each order of αs . The boost covariance
of Positronia in QED is ensured by a combination of the Coulomb potential and O (α) transverse photon exchange
(section V). The boost covariance of QCD bound states must at O αs0 be achieved by the instantaneous potential
alone, akin to QED in D = 1 + 1 (section VI). This appears to be satisfied (section VII C).
V. APPLICATIONS TO POSITRONIUM ATOMS
I now illustrate the approach to QED bound states described above with several applications to Positronium atoms.
The expansion starts with valence Fock states, here |e− e+ i, with higher Fock states included perturbatively. The first
task is then to define the valence Fock states and determine the constraints on their wave functions imposed by the
symmetries of translations and rotations, as well as parity and charge conjugation (section V A). I express the valence
Fock states using field operators (here the electron field ψ(t, x)), similarly as in the representations (3.12), (3.13) of
the Dirac bound states.

In section V B I determine the wave functions of Para- and Orthopositronium atoms at lowest O α2 in their binding
energy Eb . The rest frame wave functions then satisfy the Schrödinger equation. For atomic CM momentum P 6= 0
one needs to include the Fock state with one transverse
photon |e− e+ γi (V C). The hyperfine splitting between Ortho-
and Parapositronium at rest is calculated at O α4 in the rest frame (V D), taking into account the transverse photon
state |e− e+ γi and Orthopositronium annihilation into a virtual photon, e− e+ → γ → e− e+ . Finally I calculate the
electromagnetic form factor of Positronium (V E) and deep inelastic scattering on Positronia (V F) in a general frame.
28
The e− e+ Fock states of Para- and Orthopositronium atoms

A.
1. Definition of the Fock states
The |e− e+ i Fock states of Parapositronium (J P C = 0−+ ) and Orthopositronium (J P C = 1−− ) atoms, jointly denoted
by B, may be expressed in terms of two electron fields,
← →
Z
e e ; B, P ≡ dx1 dx2 ψ̄(x1 ) Λ (x1 )eiP ·(x1 +x2 )/2 Φ(P ) (x1 − x2 ) Λ (x2 )ψ(x2 ) |0i
− +
+ B −
(5.1)
(P )
where ΦB is a 4 × 4 wave function of the atom B with CM momentum P . The Λ± are Dirac projection operators,
← 1 h ← i ← 2
Λ + (x1 ) ≡ E1 − iα · ∇1 + γ 0 m = Λ + (x1 )
2E1
→ 1 h → i → 2
Λ − (x2 ) ≡ E2 + iα · ∇2 − γ 0 m = Λ − (x2 ) (5.2)
2E2
p
where E = −∇2 + m2 . The projectors select the b† operator in ψ̄(x1 ) and d† in ψ(x2 ), defined as in (3.10),
← →
Z Z
dk −ik·x † dk
v(k, λ) e−ik·x d†k,λ
X X
ψ̄(x) Λ + (x) = 3
ū(k, λ) e b k,λ Λ −
(x)ψ(x) = 3
(2π) 2Ek (2π) 2Ek
λ λ
(5.3)
Since b |0i = d |0i = 0 in (5.1) the projectors actually have

no effect. However, in operations on the states they allow
to use the anticommutation relations ψα† (t, x), ψβ (t, y) = δα,β δ 3 (x − y), to which the b and d operators contribute.

The projectors ensure that the coefficients of b and d in (5.1) vanish, so that no spurious contributions arise. I assume
the normalization
p
he− e+ ; B 0 , P 0 e− e+ ; B, P = 2EP (2π)3 δ(P − P 0 )δB,B0 EP = P 2 + 4m2

(5.4)

where EP is the energy of the atom at O α0 .
The Hamiltonian (4.11) is symmetric under translations, rotations, parity and charge conjugation. The states may
(P )
be classified by their transformation under those symmetries, giving constraints on the wave functions ΦB (x).
2. Translations
Under space translations x → x + ` the electron field is transformed by the operator

→
Z
U (`) = exp[−i` · P] where P = dx ψ † (x)(−i∇)ψ(x) (5.5)
The momentum operator satisfies

→ ←
[P, ψ(x)] = i∇ψ(x) P, ψ̄(x) = ψ̄(x)i∇ (5.6)
With P |0i = 0 we have P |e− e+ ; B, P i = P |e− e+ ; B, P i.
3. Rotations
Rotations are generated by the angular momentum operator J , which was defined already in (3.29) for the Dirac
equation. With α ≡ γ 0 γ,
Z
J = dx ψ † (x) J ψ(x) J = L + S = x × (−i∇) + 21 γ5 α (5.7)
29
Rest frame states are taken to be eigenstates of J 2 and J z . For a P = 0 Positronium state expressed as in (5.1),
← i→
Z h
(0)
J e− e+ ; B, P = 0 = dx1 dx2 ψ̄(x1 ) Λ + (x1 ) J , ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 ) |0i

(5.8)
For the state to have total angular momentum j and j z = λ in the rest frame,
J 2 e− e+ ; B, P = 0 = j(j + 1) e− e+ ; B, P = 0 J z e− e+ ; B, P = 0 = λ e− e+ ; B, P = 0

(5.9)
the wave function should satisfy

h h ii h i
(0) (0) (0) (0)
J i , J i , ΦB (x) = j(j + 1)ΦB (x) J z , ΦB (x) = λΦB (x) (5.10)
4. Parity ηP
The parity operator P transforms the electron field as
Pψ(t, x)P† = γ 0 ψ(t, −x) Pψ̄(t, x)P† = ψ̄(t, −x)γ 0 (5.11)
Changing the integration variables x1,2 → −x1,2 in (5.1) and noting that γ 0 Λ± (x) = Λ± (−x)γ 0 ,
← →
Z
(P )
P e e ; B, P = dx1 dx2 ψ̄(x1 ) Λ + (x1 )γ 0 e−iP ·(x1 +x2 )/2 ΦB (−x1 + x2 )γ 0 Λ − (x2 )ψ(x2 ) = ηP e− e+ ; B, −P
− +

(5.12)
if the wave function satisfies
(P ) (−P )
γ 0 ΦB (−x)γ 0 = ηP ΦB (x) (ηP = ±1) (5.13)
Note that parity reverses the CM momentum P of the state.
5. Charge conjugation ηC
The charge conjugation operator C transforms particles into antiparticles.
Cb(p, λ)C† = d(p, λ) Cd(p, λ)C† = b(p, λ) (5.14)
In the Dirac representation of the γ matrices (here T indicates transpose and α2 ≡ γ 0 γ 2 )
Cψ(t, x)C† = −iγ 2 ψ ∗ (t, x) = iα2 ψ̄ T (t, x) Cψ̄(t, x)C† = iψ T (t, x)α2 (5.15)
This implies v(k, λ) = −iγ 2 u∗ (k, λ) and thus χ̄λ = iσ2 χλ in (2.7). For a Positronium state to be an eigenstate of
charge conjugation,
C e− e+ ; B, P = ηC e− e+ ; B, P

(5.16)
its wave function should satisfy

(P ) T (P )
α2 ΦB (−x) α2 = ηC ΦB (x) (ηC = ±1) (5.17)

30
6. Wave functions of Para - and Orthopositronium
(0)
Non-relativistic Para- and Orthopositronium have zero orbital angular momentum in the rest frame, L, ΦB (x) = 0.
Hence their wave functions have no angular dependence. The radial dependence factorizes from the Dirac structure
since the spin, parity and charge conjugation constraints are independent of the radial coordinate r = |x|,
(0) (0)
ΦB (x) = NB ΓB F (0) (r) (5.18)
where ΓB is an x-independent 4 × 4 Dirac matrix. Para- and Orthopositronia have the same radial function F (r)
and the same binding energy Eb at O α2 . The energy degeneracy holds for all P , indicating that the factorization
(5.18) holds in any frame,
Z
(P ) (P )
ΦB (x) = NB ΓB F (P ) (x) dx |F (P ) (x)|2 = EP (5.19)
I verify this in section V C. The boosted radial function F (P ) (x) is angular dependent for P 6= 0 due to Lorentz
(P )
contraction in the P -direction. With its normalization fixed as in (5.19) the constants NB are determined by the
normalization (5.4) of the state.
In the following I take P = (0, 0, P ) in the z-direction, and consider Orthopositronium with j z = λ. The following
Dirac structures ΓB in (5.18) give the correct J P C quantum numbers of the Positronia:
Parapositronium: J P C = 0−+ ΓP ara = γ5 (5.20)
• Spin: [S, γ5 ] =
1
2 γ5 α, γ5 = 0, hence j = s = 0.
• Parity: γ 0 γ5 γ 0 = −γ5 , hence ηP = −1.

• Charge conjugation: α2 γ5T α2 = γ5 , hence ηC = +1.
1
Orthopositronium: J P C = 1−− ΓλOrtho = eλ · α e±1 = − √ (±1, i, 0) e0 = (0, 0, 1) (5.21)
2
• Spin: [S z , eλ · α] = λ eλ · α, hence j z = λ, and

P i i
i S , S , eλ · α = 2 eλ · α, hence j = 1.
• Parity: γ 0 eλ · α γ 0 = −eλ · α, hence ηP = −1.

• Charge conjugation: α2 eλ · αT α2 = −eλ · α, hence ηC = −1.
(P )
The constants NB of (5.19) are then determined by the state normalization (5.4) to be, at O α0 ,
(P ) (P ) EP (P )
NP ara = NOrtho (λ = 0) = NOrtho (λ = ±1) = 1 (5.22)
2m
Exercise A 11: Verify (5.22).
B. The Schrödinger equation for Positronium at P = 0
The Schrödinger equation for the rest frame wave function follows from the condition that the (Para- or Ortho-)
Positronium state (5.1) be an eigenstate of the Hamiltonian (4.11),
H e− e+ ; B, P = 0 = H0 (f ) + HV e− e+ ; B, P = 0 = (2m + Eb ) e− e+ ; B, P = 0

(5.23)

Transverse photons do not contribute to Eb at O α2 . Their coupling to electrons is proportional to the 3-momentum
of the electron, which in the rest frame is of O (αm).
31

The free fermion Hamiltonian acting on the electron fields gives (note that ψ, ψ̄ leaves a γ 0 , and αγ 0 = −γ 0 α)
→ ←
Z
H0 (f ), ψ̄(x1 ) = dx ψ † (x)(−iα · ∇ + mγ 0 ) ψ(x), ψ̄(x1 ) = ψ̄(x1 )(−iα · ∇1 + mγ 0 )

→
H0 (f ), ψ(x2 ) = −(−iα · ∇2 + mγ 0 )ψ(x2 )

(5.24)
p
Together with the projection operators Λ± in |e− e+ ; B, P = 0i these give energies E = −∇2 + m2 ,
← 1 ← 1 ← ←2 ←
(−iα · ∇1 + mγ 0 ) (E1 − iα · ∇1 + mγ 0 ) = E1 (−iα · ∇1 + mγ 0 ) − ∇1 + m2 = Λ+ (i∇1 )E1

2E1 2E1
1 → → 1 → →2 →
(E2 + iα · ∇2 − mγ 0 )(−iα · ∇2 + mγ 0 ) = E2 (iα · ∇2 − mγ 0 ) − ∇2 + m2 = E2 Λ− (i∇2 )

− (5.25)
2E2 2E2
Through a partial integration the ∇2 in the energies acts on the wave function. At O α2 we have E ' m − ∇2 /2m,

giving the kinetic contribution of the Schrödinger equation with reduced mass m/2,
∇21 (0)
2m − ΦB (x1 − x2 ) (5.26)
m
The potential energy arises from the instantaneous part (4.18) of the Hamiltonian,
e2
Z
1
HV e− e+ ; B, P = 0 =
†
ψ ψ(x) ψ † ψ(y) e− e+ ; B, P = 0

dxdy (5.27)
2 4π|x − y|
Because Gauss’ law in temporal gauge is imposed as a constraint on the physical states we have ψ † ψ |0i = 0 as in
(4.17). The effect of HV is then to multiply the wave function by the Coulomb potential,
α (0)
− Φ (x1 − x2 ) (5.28)
|x1 − x2 | B
Combining (5.26) and (5.28) the eigenstate condition (5.23) implies the Schrödinger equation for the wave function,
∇2 α (0) (0)
2m − − Φ (x) = (2m + Eb )ΦB (x) (5.29)
m |x| B
Since the Schrödinger equation is a Dirac scalar it gives the usual ` = 0 radial equation for F (0) (r) in (5.18),
1 h (0) 00 2 0
i hα i α3/2 m2
F (r) + F (0) (r) + + Eb F (0) (r) = 0 =⇒ F (0) (r) = √ exp(−αmr/2) Eb = − 14 mα2
m r r 4π
(5.30)
where the normalization of F (0) (r) is determined by (5.19) (EP =0 = 2m).
C. Positronium with momentum P
Fock states quantized at equal time transform non-covariantly under boosts since the definition of time is frame
dependent. The Poincaré invariance of the QED action nevertheless guarantees that measurables
will be Lorentz
covariant. In this section I demonstrate this for the binding energy of Positronia at O α2 . Lorentz covariance
determines the momentum dependence of the binding,
2mEb
q p
∆E(P ) ≡ P 2 + (2m + Eb )2 − P 2 + 4m2 = + O α4

∆E(P = 0) = Eb (5.31)
EP
p
whereEP = P 2 + 4m2 . In section V D I evaluate the hyperfine splitting between Ortho- and Parapositronia at
O α4 for P = 0. In sections V E and V F I consider the covariance of form factors and deep inelastic scattering.
The importance of properly taking into account the momentum dependence of bound state wave functions was
emphasized in [23]. The frame dependence of atomic wave functions is of general interest, since it shows how the
32
classical concept of Lorentz contraction is realized for quantum bound states. Surprisingly, there appears to be only
one study [24] of atoms with general CM momenta P , even at leading order. The following analysis is equivalent to
that one, but is formulated in terms of Fock states in temporal gauge.
In atoms with CM momentum
P transverse photons contribute at leading order to the binding energy Eb , since
they couple at O α0 to electrons whose momenta ∝ P . I consider first the kinetic and potential energies of the
|e− e+ i Fock state (5.1), and − +
then determine the wave function of the |e e γi state in Positronium. Only terms which
2
contribute to Eb at O α are retained.
1. Kinetic and potential energy
The above relations (5.24) and (5.25) are valid for all P . After a partial integration the derivatives in E1 and E2
(P ) → → → → → ←
operate in (5.1) on exp iP · (x1 + x2 )/2 ΦB (x1 − x2 ). Let ∇1 = 21 (∇1 + ∇2 ) + 21 (∇1 − ∇2 ), and similarly for ∇2 .

→ →
(P )
Then the first term gives iP /2 while the second, denoted ∇ ≡ 21 (∇1 − ∇2 ), gives O (α) derivatives of ΦB (x),
q q
E1 = ( 12 P − i∇)2 + m2 E2 = ( 21 P + i∇)2 + m2 (5.32)
√
At O α2 we need to keep two powers of ∇. Using 1 + x = 1 + 21 x − 18 x2 + O x3 and denoting

p EP P
EP ≡ P 2 + 4m2 γ≡ β≡ (5.33)
2m EP
we get
2 2 1 1 2 1
H0 (f ) e− e+ ; B, P : ∇⊥ − 2 ∇2k = EP − ∇⊥ − 2 ∇2k

E1 + E2 ' EP − (5.34)
EP γ mγ γ
where ⊥ and k refer to the P -direction, here taken to be along the z-axis.
The potential energy depends only on the instantaneous positions of the fermions and thus gives the same result as
in (5.28) for P = 0, the wave functions is multiplied by
Z
α dq 1 −iq·(x1 −x2 )
HV e− e+ ; B, P : = −e2

− e (5.35)
|x1 − x2 | (2π)3 q 2
This already signals the importance of transverse photon exchange, since the potential energy should be commensurate
with ∆E(P ) in (5.31), which is P -dependent.
2. The transverse photon Fock state

For P 6= 0 the transverse photon vertices ∝ P are of O α0 , so transverse and Coulomb photon exchanges contribute
at the same order in α. The transverse photon and its conjugate electric field in temporal gauge (A0 = 0) may at
t = 0 be expanded in photon creation and annihilation operators a† and a, with polarization 3-vectors εs , s = ±1,
Z
dq X h i
AT (x) = 3
εs (q) eiq·x a(q, s) + ε∗s (q) e−iq·x a† (q, s)
(2π) 2|q| s=±1
Z
dq X h i
E T (x) = i 3
εs (q) eiq·x a(q, s) − ε∗s (q) e−iq·x a† (q, s) (5.36)
2(2π) s=±1
a(q, s), a† (q 0 , s0 ) = (2π)3 2|q|δ(q − q 0 )δs,s0

X ∗ qi qj
q · εs (q) = 0 ε∗s (q) · εs0 (q) = δs,s0 εis (q)εjs (q) = δ ij −
s=±1
q2
The interaction between the electron and the transverse photon fields in the Hamiltonian (4.11) is given by
Z
Hint (AT ) = −e dx ψ † (x) α · AT (x)ψ(x) (5.37)
33
This creates |e− e+ γi states with a photon of momentum q and polarization s. At leading order,
← →
Z
(P )
Hint e e ; B, P = e dx1 dx2 ψ̄(x1 ) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 )
− +

Z Z
dq X 1 dq X
P · ε∗s (q) a† (q, s) − e−iq·x1 + e−iq·x2 |0i ≡ e− e+ γ; q, s

× (5.38)
(2π)3 2|q| s EP (2π)3 2|q| s
Exercise A 12: Derive the expression for |e− e+ γ; q, si in (5.38).
Operating with Hint a second time and retaining only the terms where the transverse photon is absorbed, giving an
|e− e+ i Fock state,
← →
Z
2 − + (P )
e e ; B, P = e2 dx1 dx2 ψ̄(x1 ) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 )

Hint
Z
dq X 1 2
P · ε∗s (q) P · εs (q) − e−iq·x1 + e−iq·x2 |0i

× 3 2 (5.39)
(2π) 2|q| s EP
− e−iq·x1 + e−iq·x2 2 = 2 − e−iq·(x1 −x2 ) − eiq·(x1 −x2 )

where (5.40)
The x1,2 -independent term in (5.40) corresponds to the absorption of the photon on the same fermion from which it
was emitted. This loop contribution gives a multiplicative renormalization of the state and does not contribute to the
eigenstate condition at lowest order. Neglecting this term and summing over the photon polarization s using (5.36)
we have
← →
Z
2 − + (P )
2
dx1 dx2 ψ̄(x1 ) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 )

Hint e e ; B, P = −e
Z 2
dq 2 q⊥
e−iq·(x1 −x2 ) + eiq·(x1 −x2 ) |0i

× 3
β 2
(5.41)
(2π) 2|q| q
where β = |P |/EP and q ⊥ is the component of q orthogonal to P (i.e., to the z-axis).
3. The bound state condition
The Positronium state, including the |e− e+ γi Fock component, can be expressed as the superposition
Z
dq X
|B, P i = e− e+ ; B, P + Cγ (q) e− e+ γ; q, s

3
(5.42)
(2π) 2|q| s
where |e− e+ γ; q, si is defined in (5.38) and I anticipated that its relative weight Cγ (q) is independent of s. Cγ (q)
should be determined so that |B, P i is an eigenstate of H,
Z
dq X
H |B, P i = (H0 (f ) + HV ) e− e+ ; B, P + Hint Cγ (q) e− e+ γ; q, s

3
(2π) 2|q| s
Z
dq Xh i
(P )
1 + H0 (f ) + H0 (A) Cγ (q) e− e+ γ; q, s = EB |B, P i

+ 3
(5.43)
(2π) 2|q| s
(P )
H0 (f ) + HV modify ΦB by the factors in (5.34) and (5.35). The Hint term is given by (5.41), adding the factor
Cγ (q) to the integrand of the q-integration. The first contribution to |e− e+ γ; q, si follows from (5.38). The action of
H0 (f ) is similar to that on |e− e+ ; B, P i in (5.34), but now the additional x1,2 dependence in exp(−iq · x1,2) leaves
an O (α) contribution. Since |e− e+ γ; q, si is already suppressed by a factor e we may here neglect the O α2 terms,

including its potential energy (HV ). Recalling that Ei = −∇2i + m2 and separating the O α0 contribution through,
q
q q
(E1 + E2 )eiP ·(x1 +x2 )/2 ' eiP ·(x1 +x2 )/2 1 2
4 EP − iP · ∇ 1 + 1 2
4 E P − iP · ∇ 2
h i i
' eiP ·(x1 +x2 )/2 EP − P · (∇1 + ∇2 ) (5.44)
EP
34
gives the sum of the fermion kinetic energies as
H0 (f ) e− e+ γ; q, s : E1 + E2 = EP − βqk + O α2

(5.45)
where β = P/EP and P · q = P qk .

dq
a† (q, s)a(q, s),
R P
The kinetic energy of the photon follows from H0 (A) = (2π)3 2|q| |q| s=±1
H0 (A) e− e+ γ; q, s :

Eγ = |q| (5.46)
(P )
Comparing the coefficients of |e− e+ γ; q, si in the eigenstate condition (5.43) gives, since EB 2

= EP + O α ,
1 |q| + βqk
1 + Cγ (q)(EP + |q| − βqk ) = Cγ (q)EP + O α2

=⇒ Cγ (q) = − =− 2 (5.47)
|q| − βqk q ⊥ + qk2 /γ 2
Including Cγ (q) in (5.41) we have in (5.43),

← →
Z Z
dq X − + (P )
dx1 dx2 ψ̄(x1 ) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 )

Hint 3
C γ (q) e e γ; q, s =
(2π) 2|q| s
q 2 |q| + βqk −iq·(x1 −x2 )

Z
dq
× e2 β2 ⊥ + eiq·(x1 −x2 ) |0i (5.48)

3 2 2 2 2
e
(2π) 2|q| q q ⊥ + qk /γ
The βqk term in the numerator does not contribute since the remaining integrand is symmetric under q → −q. We
may then use this symmetry to set exp[iq · (x1 − x2 )] → exp[−iq · (x1 − x2 )] and cancel a factor 2|q|. Combining this
transverse photon contribution with the Coulomb one (5.35) the integral over q becomes (with x = x1 − x2 ),
β 2 q⊥
2
q 2 /γ 2 i −iq·x e2 e−iq·x
Z Z
dq 1 h dq α
−e2 3 2
1 − 2 2 2
= 2 2 2
e = − 2 3 2 2 2
=− q (5.49)
(2π) q q ⊥ + qk /γ q ⊥ + qk /γ γ (2π) q ⊥ + qk /γ γ x 2 + γ 2 x2 ⊥ k
(P )
Adding the kinetic energy (5.34) the eigenstate condition (5.43) imposes the bound state condition on ΦB , with the
required eigenvalue (5.31),
h 1 2 1 α i
(P )
1 (P )
EP − ∇⊥ − 2 ∇2k − q ΦB (x) = EP + Eb ΦB (x) (5.50)
mγ γ γ x 2 + γ 2 x2 γ
⊥ k
A comparison with the P = 0 equation (5.29) shows that up to a normalization we have

(P ) (0)
ΦB (x) = ΦB (x⊥ , γxk ) (5.51)
i.e., standard Lorentz contraction. The contraction is the same for all Dirac components, justifying the factorization
(5.19). According to (5.30) the contracted radial function is
p
(P ) (0) α3/2 m2 −αmrP /2 q
2 2 EP P 2 + 4m2
F (x) = γ F (rP ) = γ √ e rP ≡ x⊥ + γ xk 2 γ= = (5.52)
4π 2m 2m
The Lorentz contraction of F (P ) (x) agrees with the classical result. Recall, however, that the |e− e+ γ; q, si Fock
component (5.42) also contributes. The kinetic energies of its constituents (5.45), (5.46) are of O (α), i.e., large
compared to the O α2 binding energy of |B, P i. By the uncertainty principle |B, P i fluctuates into |e− e+ γ; q, si
only a fraction α of the time. Equivalently, the norm of the |e− e+ γ; q, si Fock component is of O (α).
D. * Hyperfine splitting of Positronium at P = 0
Hyperfine splitting (hfs) is defined as the difference between the Ortho- and Parapositronium binding energies,
Ehf s = Eb (Ortho) − Eb (P ara). The hfs (1.1) is known to high accuracy, with current work addressing the O α7 m
contribution using the methods of NRQED. Here I shall illustrate the Fock state method in temporal gauge by eval-
uating the O α4 m contribution. At this order the hfs arises from transverse photon exchange between the e− and
e+ , as well as from the annihilation contribution e− e+ → γ → e− e+ (for Orthopositronium only). In the Fock state
approach this means considering the |e− e+ γi and |γi Fock states, respectively.
35
The transverse photon Fock state e− e+ γ contribution

1.
In section V C I evaluated the transverse photon exchange contribution to Positronium with P 6= 0 at leading order.
Here I consider transverse photon exchange for Positronium at rest (P = 0). The two electron-photon vertices are
then proportional to O (α) Bohr momenta, which makes the transverse photon contribution to be of O α4 in the rest
frame. I discuss only photon emission from the e− and absorption by the e+ . The converse contribution is identical
and is taken into account by a factor 2 in the final result. Photons both emitted and absorbed on the e− do not
contribute to the spin correlation (hfs) between the e− and e+ .
The Positronium state including the Fock state with a transverse photon is (5.42)
|Bi = e− e+ ; B + e− e+ γ; B

(5.53)
where |e− e+ ; Bi is defined in (5.1) with P = 0 and the wave function ΦB (x) = ΓB F (r) according to (5.18) and (5.22).
Its Dirac structures are ΓP ara = γ5 , ΓOrtho = eλ · α as in (5.20), (5.21) and the radial function F (r) is given in (5.30).
The transverse photon state is as in (5.42) with Cγ = −1/|q| from (5.47) with P = 0,
− +
Z
dq X −1
e− e+ γ; q, s

e e γ; B = (5.54)
3
(2π) 2|q| s |q|
The |e− e+ γ; q, si state created by Hint (AT ) (5.37) acting on |e− e+ ; Bi with e− → e− γ is given in (A.35).
When the Hamiltonian acts on its eigenstate |Bi the energy eigenvalue may be read off from the coefficient of the
|e− e+ ; Bi Fock state. Projecting on this state,
h he− e+ ; B|Hint |e− e+ γ; Bi i − +
H |Bi = 2m − 14 mα2 + O α4 + e e ; B + . . . (5.55)
he− e+ ; B|e− e+ ; Bi
The O α4 term represents higher order corrections to the eigenvalue from (H0 + HV ) |e− e+ ; Bi, e.g., due to the

Taylor expansion of the energies Ei in (5.25). They are the same for Para- and Orthopositronium and do not affect
the hfs. I keep only terms which are spin-, i.e., ΓB -dependent.
The absorption of the photon on the e+ is given by [Hint , ψ(x2 )] analogously as the emission from e− in (A.35),
← ←
Z XZ
− + 2 dq ∗ −iq·x1
Hint e e γ; B = −e
dx1 dx 2 ψ̄(x 1 ) Λ +
(x 1 )α · εs (q)e Λ + (x1 )ΓB F (r)
(2π)3 2q 2 s
→ →
× Λ − (x2 )α · εs (q)eiq·x2 Λ − (x2 )ψ(x2 ) |0i (5.56)
where r = |x1 − x2 | and the extra pair of Λ± project b† from ψ̄(x1 ) and d† from ψ(x2 ) (5.3). At O α4 two momenta

can contribute, one each from the photon vertices. The identities in (A.36) and (A.38) show that α bracketed by two
projectors becomes a derivative. This contribution reduces the Dirac structure of (5.56) so that the overlap in (5.55)
is independent of ΓB , i.e., it does not contribute to the hfs. Hence we need only consider the contributions
← 1 → 1
e−iq·x1 Λ(x1 ) → − α · q e−iq·x1 Λ(x2 )eiq·x2 → − α · q eiq·x2 (5.57)
2m 2m

At O α4 the remaining projectors can be set to lowest order, Λ± = (1 ± γ 0 )/2. The products of α-matrices may be
reduced through αi αj = δij + iεijk αk γ5 . The δij -function does not contribute here since q · εs (q) = 0,
∗ ∗
εis q j αi αj = iεijk εis q j αk γ5 q ` εm ` m
s α` αm = iε`mn q εs αn γ5 (5.58)
i∗
Since [γ5 , ΓB ] = 0 the γ5 ’s cancel, γ52 = 1. The sum over photon polarizations (5.36) gives m
→ δ im ,
P
s εs (q)εs (q)
since the −q i q m /q 2 term vanishes. Then i2 εijk εin` q j q ` = q 2 δ kn − q k q n .
Writing αk ΓB = [αk , ΓB ] + ΓB αk the second term does not give an hfs, while the commutator vanishes for ΓP ara = γ5 .
Hence is suffices to consider the Orthopositronium (j z = λ) contribution, [αk , eλ · α] = −2iεjkp ejλ αp γ5 . This is
multiplied by the αn in (5.58), giving −2iεjkp ejλ αp αn γ5 = 2εjkp εnip ejλ αi = 2(enλ αk − δ kn eλ · α). Combined with the
q-dependence found above we have
(q 2 δ kn − q k q n )2(enλ αk − δ kn eλ · α) = −2 q 2 eλ · α + (eλ · q)(α · q)

(5.59)
36
The contribution to (5.56) that is relevant for the hfs is thus

e2
Z Z
dq
Hint e− e+ γ; Ortho = dx1 dx2 ψ̄(x1 )Λ+ F (r)e−iq·(x1 −x2 ) q 2 eλ · α + (eλ · q)(α · q) Λ− ψ(x2 ) |0i

4m 2 3
(2π) q 2
(5.60)
where Λ± = (1 ± γ 0 )/2. In the matrix elements of (5.55) both electron fields are annihilated and the integral
d(x1 + x2 )/2 = (2π)3 δ(0) cancels with the norm (5.4) in the denominator up to a factor 4m. Recalling that we only
R
considered photon emission from e− and absorption on e+ and so are getting half of the hfs,
1 T he− e+ ; B|Hint |e− e+ γ; Bi

2 Ehf s =
he− e+ ; B|e− e+ ; Bi
e2
Z
dq
|F (r)|2 e−iq·x Tr 12 (1 − γ 0 )e∗λ · α 21 (1 + γ 0 ) q 2 eλ · α + eiλ αj q i q j

= dx (5.61)
16m3 (2π)3 q 2
The factor multiplying q i q j in the integrand is symmetric under q i → −q i , xi → −xi , allowing q i q j → 31 q 2 δ ij . The
trace factor becomes 32 q 2 Tr (e∗λ · α)(eλ · α) = 83 q 2 . With F (r) given by (5.30),
e2
Z Z
1 T 2 dq −iq·x 4πα 1
2 Ehf s = 6m3 dx |F (r)| e = |F (0)|2 = mα4 (5.62)
(2π)3 6m3 6
The contribution to the hfs from transverse photon exchange is thus as expected [6],
T 1
Ehf s = mα4 (5.63)
3
2. Hyperfine splitting from annihilation: e− e+ → γ → e− e+
The |e− e+ i → |γi → |e− e+ i transition is proportional to the square |ΦB (0)|2 of the P = 0 Positronium (5.1) wave
function at the origin. ΦB (x) = ΓB F (r) where ΓB for Para- and Orthopositronium are in (5.20), (5.21) and their
2 2 4
common radial function F (r) is in (5.30). Counting also the vertex couplings the transition
4
is ∝ e |F (0)| ∝ α .
Hence we may neglect the O (αm) relative (Bohr) momenta in evaluating the hfs at O α . The projectors Λ± in
the state (5.1) may then be replaced with 21 (1 ± γ 0 ). Annihilating both the e− and e+ in the state (5.1) by Hint gives
Z Z
Hint e− e+ ; B = −e dx ψ̄(x)α · A(x)ψ(x) dx1 dx2 ψ̄(x1 ) 12 (1 + γ 0 )ΦB (x1 − x2 ) 21 (1 − γ 0 )ψ(x2 ) |0i

(5.64)
Z Z
A(x)γ 0 21 (1 0
)ΓB F (0) 12 (1 0
− 21 eF (0) dx Tr [αi ΓB ] Ai (x) |0i

= −e dx Tr α · +γ − γ ) |0i =
As expected due to charge conjugation invariance, this vanishes for Parapositronium, ΓP ara = γ5 . Hence the annihi-
lation contribution to the hfs arises only from Orthopositronium, ΓλOrtho = eλ · α for states with j z = λ:
Z
Hint e− e+ ; Oλ = −2eF (0) dx eλ · A(x) |0i ≡ −2eF (0) |A, λi

(5.65)
The relevant action of the Hamiltonian (4.11) on this state is given by the canonical commutation relations (4.10),
Z Z Z
H |A, λi → dy 21 E 2 (y) dx eλ · A(x) |0i = i dx eλ · E(x) |0i ≡ i |E, λi (5.66)
H |E, λi has an overlap CO with Orthopositronium. Neglecting the other states which do not contribute here,
Z Z Z
Hint |E, λi = −e dy ψ (y) α · A(y)ψ(y) dx eλ · E(x) |0i = ie dx ψ † (x) eλ · α ψ(x) |0i = CO e− e+ ; Oλ + . . .
†

(5.67)
With the normalization (5.4) the overlap CO is
he− e+ ; Oλ |H |E, λi
Z
1
CO = − + == h0| dx1 dx2 ψ † (x2 ) e∗λ · α F ∗ (|x1 − x2 |) 12 (1 + γ 0 )γ 0 ψ(x1 )
he e ; Oλ |e− e+ ; Oλ i 4m(2π)3 δ 3 (0)
Z
ie ∗
× ie dx ψ † (x) eλ · α ψ(x) |0i = F (0) (5.68)
2m
37

The Orthopositronium state may at O α4 be considered as a superposition of the three states involved,
|Oλ i = e− e+ ; Oλ + CA |A, λi + CE |E, λi

(5.69)
and should be an eigenstate of H with eigenvalue EO . Using (5.65), (5.66) and (5.68),
H |Oλ i = (2m − 14 mα2 + CO CE ) e− e+ ; Oλ + CA i |E, λi − 2eF (0) |A, λi

h 2eF (0) iCA i

= E O e− e+ ; O λ − |A, λi + |E, λi = EO |Oλ i (5.70)
EO EO
The eigenstate constraint requires (at leading order) CA = −2eF (0)/2m and CE = iCA /2m = −2ieF (0)/4m2 . With
the value of F (0) in (5.30) this gives the hfs term in EO as
ie −2ie
CO CE = |F (0)|2 = 41 mα4 (5.71)
2m 4m2
as quoted in [6].
E. * Electromagnetic form factor of Positronium atoms in an arbitrary frame
In this section I evaluate the electromagnetic form factors of Positronium. The elastic form factor is evaluated with
leading order wave functions in an arbitrary frame, demonstrating the Lorentz covariance of the result. The transition
form factor from Para- to Orthopositronium is calculated in the rest frame only.
The electromagnetic current j µ (z) may be translated to the origin (z = 0) using the four-momentum operator P̂ ,
j µ (z) = ψ̄(z)γ µ ψ(z) = eiP̂ ·z j µ (0)e−iP̂ ·z (5.72)

µ
The EM form factor FAB is the expectation value of the current between atoms A, B of three-momenta P A , P B
whose four-momenta satisfy PA2 = MA2 , PB2 = MB2 ,
µ
FAB (z) = hB, P B |j µ (z) |A, P A i = ei(PB −PA )·z hB, P B |j µ (0) |A, P A i
Z
µ µ
FAB (q) = d4 z e−iq·z FAB (z) = (2π)4 δ 4 (Pb − Pa − q)GµAB (q) (5.73)
In the following I consider GµAB (q), keeping in mind the four-momentum constraint.
The Positronium state is defined in (5.1). With a short-hand notation ΨA the wave function of the incoming state is
Z
|A, P i = dx1 dx2 ψ̄(x1 )ΨA (x1 , x2 )ψ(x2 ) |0i
← →
(P A )
ΨA (x1 , x2 ) = Λ + (x1 )eiP A ·(x1 +x2 )/2 ΦA (x1 − x2 ) Λ − (x2 ) (5.74)
where the projectors Λ± are defined in (5.2). The same notation for the final state hB, P | gives
Z
GµAB (q) = dy 1 dy 2 dx1 dx2 h0|ψ † (y 2 )Ψ†B (y 1 , y 2 )γ 0 ψ(y 1 ) ψ̄(0)γ µ ψ(0) ψ̄(x1 )ΨA (x1 , x2 )ψ(x2 ) |0i (5.75)
The contraction of ψ(0) with ψ̄(x1 ) corresponds to j µ interacting with the e− . This sets x1 = y 1 = 0 and y 2 = x2 . I
denote this contribution GµAB (q, e− ). Interaction with e+ corresponds to ψ(0) contracting with ψ † (y 2 ) and is denoted
GµAB (q, e+ ). It has a minus sign due to anticommutation and sets x2 = y 2 = 0 and y 1 = x1 . Thus
Z
GµAB (q, e− ) = dx Tr ΨA (0, −x)Ψ†B (0, −x)γ µ γ 0

(5.76)
Z
GµAB (q, e+ ) = − dx Tr Ψ†B (−x, 0)ΨA (−x, 0)γ 0 γ µ

38
The two contributions are related by charge conjugation. Using (5.17) and recalling that α2 γ µ α2 = −(γ µ )T and
α2 α α2 = −αT we have
α2 ΨT (x2 , x1 )α2 = ηC Ψ(x1 , x2 ) (5.77)
Multiplying the argument of the Tr in GµAB (q, e− ) by α2 from the left and right and taking its transpose shows that
GµAB (q, e+ ) = −ηC

A B µ
ηC GAB (q, e− ) (5.78)
γ A B
As expected the photon (ηC = −1) requires ηC = −ηC when A and B are eigenstates of charge conjugation,
Z
GµAB (q) = (1 − ηC ηC ) dx Tr ΨA (0, −x)Ψ†B (0, −x)γ µ γ 0
A B

(5.79)
1. Parapositronium form factor
I take both A and B to be Parapositronium and consider only GµAB (q, e− ). This is relevant for states which are not
eigenstates of charge conjugation, e.g., a hypothetical µ− e+ atom where the muon and electron have the same mass.
Even for standard Positronium GµAB (q, e− ) 6= 0 and should have the form required by Lorentz covariance,
Gµ (q, e− ) = (PA + PB )µ F (q 2 ) (5.80)

After partial integrations in the state (5.74) we need consider only the projector derivatives acting on the O α0
(P )
phase exp[iP · (x1 + x2 )/2], since ∇ΦA,B (x) is of O (α). The projectors then become,
1 2
EP ∓ α · P ± M γ 0 = Λ†± (P ) = Λ± (P )

Λ± (P ) ≡ Λ+ (P )Λ− (P ) = 0 (5.81)
2EP
p
where EP = P 2 + M 2 and M = 2m at leading order.
The Parapositronium states are relativistically normalized (5.4), with the Lorentz contracted wave function Φ(P ) (x)
given in (5.19), (5.20), (5.22) and (5.52). Taking P = (0, 0, P ) = (0, 0, M sinh ξ) along the z-axis,
EP EP (0) EP α3/2 m2 −αmrP /2

Φ(P ) (x) = γ5 F (P ) (x) F (P ) (x) = F (rP ) = √ e (5.82)
M M M 4π
q
rP ≡ x2 + y 2 + z 2 cosh2 ξ (5.83)
The photon momentum q must be of O (α) for a leading order overlap between ΨA and ΨB . Hence we may set P B =
P A + q = P + O (α) in GµAB (q, e− ) (5.76). However, we need to retain the q-dependence of the ΨA (0, −x)Ψ†B (0, −x)
phase factor exp[i(P B − P A ) · x/2] = exp(iq · x/2), which reflects the photon wave function. The expression for
GµAB (q, e− ) is then
E 2 Z
P
Gµ (q, e− ) = dx |F (P ) (x)|2 eiq·x/2 Tr Λ+ (P )γ5 Λ− (P )Λ− (P )γ5 Λ+ (P )γ µ γ 0

(5.84)
M
From the definitions (5.81) of Λ± (P ) follows that
M
Λ+ (P )γ5 Λ− (P ) = Λ+ (P )γ 0 γ5 (5.85)
EP
Using this the trace in (5.84) becomes
M 2 M 2 2P µ
Tr µ = Tr Λ+ (P )γ µ γ 0 =

(5.86)
EP EP EP
so that
2P µ
Z
Gµ (q, e− ) = dx |F (P ) (x)|2 eiq·x/2 (5.87)
EP
39
Changing the integration variable to xR ≡ (x, y, z cosh ξ) gives dx = dxR / cosh ξ and rP = |xR | in (5.82). The photon
four-momentum q is constrained by kinematics,
M 2 = (P + q)2 = M 2 + 2P · q + O α2 = M 2

(5.88)
which at O (α) implies P · q = EP q 0 − P · q = 0. In the Positronium rest frame (P = 0) this means qR
0
= 0. Thus
z z z
q = qR cosh ξ, where qR is the z-component of the photon momentum in the rest frame. Hence q · x = q R · xR .
Recalling that cosh ξ = EP /M and using the expression for F (P ) (x) in (5.82) gives
2P µ EP α3 m4
Z
µ −

G (q, e ) = dxR exp (−αM |xR | + iq R · xR )/2 (5.89)
EP M 4π
The integral is as in the rest frame, i.e., it is P -independent, so the result is covariant and agrees with (5.80),
(αM )4
Gµ (q, e− ) = 2P µ (5.90)
(Q2 + α2 M 2 )2
where 2P µ = (PA + PB )µ + O (α) and Q2 = q 2R = −q 2 .
2. Positronium transition form factor
The γ ∗ (q) + Parapositronium → Orthopositronium transition electromagnetic form factor has the structure
Gµλ (q) = iεµνρσ Pν qρ eλσ F (q 2 ) (5.91)
where P is the four-momentum of one of the Positronia, eλ is the polarization vector (5.21) of the Orthopositronium
(with eλσ=0 = 0) and q is the photon momentum. Symmetries and gauge invariance force the kinematic factor to be
of O (q), i.e., of O (α). This reflects the spin flip, from S = 0 for Parapositronium to S = 1 for Orthopositronium.
In section V C I demonstrated that transverse photon exchange contributes to the binding of Positronium at leading
order for P 6= 0. The photon exchange may
at O (q) involve a spin flip at one of its vertices, whereupon the transition
to Orthopositronium proceeds at O α0 . I shall not here work out the O (α) corrections to the wave function of
Positronium in motion, but limit myself to evaluating the transition form factor in the rest frame of the target,
P A = 0.
Using the expressions (5.81) for the Λ± (P ) projectors (with P A = 0, P B = q, EB ' M ) the Positronium wave
functions given in (5.19), (5.20), (5.21) and (5.22) are,
ΨA (0, −x) = 21 (1 + γ 0 )γ5 F (r)
ΨB (0, −x) = Nλ Λ+ (q)eλ · αΛ− (q)F (r)e−iq·x/2 (5.92)

The expression for ΨB (0, −x) may be simplified using Λ+ (q)Λ− (q) = 0,
1
Λ+ (q)eλ · α Λ− (q) = {Λ+ (q), eλ · α} Λ− (q) = eλ · α − eλ · q Λ− (q) (5.93)
M
The form factor (5.79) is then in the target rest frame,
Nλ ∗
Z
Gµλ (q) = dx |F (r)|2 eiq·x/2 Tr µλ
2M
1 ∗ µ 0
Tr µλ = Tr (1 + γ 0 )γ5 (M + α · q − M γ 0 ) e∗λ · α − eλ · q γ γ = Tr γ5 α · q e∗λ · α γ µ γ 0

M
∗
= −4i δ µ,i εijk q j ekλ (5.94)
This agrees with the kinematic factor in (5.91) for P = 0. The Orthopositronium is transversely polarized because
eλ=0 k P = q gives Tr µ0 = 0. The normalization Nλ=±1 = 1 (5.22). The invariant form factor has the same integral
as in (5.89),
(αM )4
Z
2 2
F (q 2 ) = 2 dx |F (r)|2 eiq·x/2 = (5.95)
M M (Q2 + α2 M 2 )2
where Q2 = −q 2 .
I leave it as an exercise (without a worked-out solution) to show that the transition form factor agrees with (5.91) in
a general frame.
40
F. * Deep inelastic scattering on Parapositronium in a general frame
The target A vertex of Deep Inelastic Scattering (DIS), γ ∗ (q) + A(PA ) → X, is as in a transition form factor (5.73)
for each final state X, with the Parapositronium state A defined in (5.74). Now the photon is taken to have an
asymptotically large momentum q, and the squared vertex is summed over the final states X. In the absence of
radiative effects we may describe X in the basis of free e− e+ states,
|Xi = b†k1 ,λ1 d†k2 ,λ2 |0i (5.96)
constrained by momentum conservation, q + PA = k1 + k2 . In the Bjorken limit
Q2
xBj = (5.97)
2PA · q
is fixed as q → ∞, with Q2 = −q 2 > 0. The frame is defined by keeping the target 4-momentum PAµ fixed in the
Bj limit, with the three-momenta of q µ = (q 0 , 0, 0, −|q|) and PAµ = (EA , 0, 0, PA ) along the z-axis. The target mass
M = 2m + O α2 .
The amplitude for γ ∗ A → X corresponding to (5.75) is, suppressing the momentum conserving δ-function as in (5.73),
Z
µ
GAX (q) = dx1 dx2 h0|dk2 ,λ2 bk1 ,λ1 ψ̄(0)γ µ ψ(0)ψ̄(x1 )ΨA (x1 , x2 )ψ(x2 ) |0i (5.98)
I consider only scattering from the electron, hence the contractions

bk1 ,λ1 , ψ̄(0) = ū(k1 , λ1 )
{dk2 ,λ2 , ψ(x2 )} = e−ik2 ·x2 v(k2 , λ2 ) (5.99)
resulting in
Z
GµAX (q) = dx2 ū(k1 , λ1 )γ µ γ 0 ΨA (0, x2 )v(k2 , λ2 )e−ik2 ·x2 (5.100)
According to (5.3) the Λ− projector in (5.74) reduces to unity when acting on e−ik2 ·x2 v(k2 , λ2 ). After a partial
integration of the γ 0 Λ+ projector in the target state (5.74) it acts on ΨA (x1 , x2 ) and gives (P
/ A + M ) when the O (α)
contribution from differentiating the radial function is neglected. The radial function F (r) is as in (5.82) evaluated
with Lorentz contracted argument. Thus
EA
γ 0 ΨA (0, x2 ) = / + M )eiP A ·x2 /2 γ5 F (|x2P |)
(P
2M 2 A
x⊥ ⊥
2P = x2 z2P = z2 cosh ξ = z2 EA /M (5.101)
The integral over x2 in GµAX (q) can be done by a change of integration variable, x2 → x2P , giving
√
π α5/2 M 2
GµAX (q) = 2 ū(k1 , λ1 )γ µ (P
/ A + M )γ5 v(k2 , λ2 )
8 α2 M 2 /16 + P 22
P 2 ≡ − k⊥ 1 z z

2 , ( 2 PA − k2 )/ cosh ξ (5.102)
The denominator of GµAX (q) shows that P 2 , i.e., k⊥ 1 z z

2 and 2 PA − k2 must be of O (α) for a leading order contribution.
Squaring the form factor and summing over the helicities λ1 , λ2 ,
πα5 M 4
GµAX (q)GνAX † (q) =
X
µν
4 Tr
λ1 ,λ2 64 α2 M 2 /16 + P 22
Tr µν = Tr γ µ (P / A + M )γ ν (/

/ A + M )(/
k 2 + m)(P k 1 + m)
= 2M 2 Tr γ µ (P k 1 + m) = 8M 2 PAµ k1ν + PAν k1µ − g µν (PA · k1 − 21 M 2 )

/ A + M )γ ν (/

(5.103)
41
where I used k2 = 12 PA + O (α). We may check gauge invariance at leading order,
qµ Tr µν = 2M 2 Tr (/ / A + M )γ ν (k/1 + m) = 2M 2 Tr (m − 21 P / A + M )γ ν (k/1 + m) = 0 (5.104)

/ A )(P
k 1 + k/2 − P / A )(P
In the notation [62]

Z
1 dk1 dk2
Im W µν (Pa , q) = (2π)4 δ 4 (q + Pa − k1 − k2 )Gµ (q)Gν † (q)
2 (2π)6 4E1 E2
qµ qν Pa · q ν Pa · q
W µν = − g µν + 2 W1 + Paµ − q µ 2 Pa − q ν 2 W 2 (5.105)
q q q
Identifying W1 from the coefficient of −g µν in (5.103) (PA · k1 21 M 2 ) we get for the electron distribution,
α 5 M 6 PA · k 1
Z
1 dk2
fe/A (xBj ) = Im W1 = 2
δ(q 0 + EA − E1 − E2 ) 4 (5.106)
π (2π) 4E1 E2 16 α2 M 2 /16 + P 22
As q = (q 0 , 0, 0, −|q|) → ∞ at fixed xBj and PA ,

p q
E1 = (q + P A − k2 )2 + m2 ' q 2 − 2|q|(PAz − k2z ) ' |q| − (PA − k2 )z (5.107)
Defining the light-front notation by q ± ≡ q 0 ± q z we have,
Q2 = −q + q − = 2xBj PA · q ' xBj PA+ q − =⇒ q + = −xBj PA+ (5.108)
The energy constraint in (5.106) becomes
q 0 + EA − E1 − E2 = q 0 − |q| + (PA − k2 )z + EA − E2 = q + + PA+ − k2+ = PA+ (1 − xBj ) − k2+ = 0 (5.109)
Recalling that k2 = 12 PA at O α0 I denote the O (α) difference xBj − 21 as

xBj = 12 (1 + α x̃B ) k2+ = 12 PA+ (1 − αx̃B ) = m eξ (1 − αx̃B ) (5.110)

Neglecting terms of O α2 we have in P 2 (5.102),
m2 e−ξ
k2z − 21 Paz = 12 (k2+ − k2− ) − 12 Paz = 12 meξ (1 − αx̃B ) − − m sinh ξ = −mαx̃B cosh ξ (5.111)
2m(1 − αx̃B )
With the change of variables dk2z /E2 = dk2+ /k2+ the δ-function in (5.106) may be integrated using (5.109). Substituting
PA · k1 /E1 k2+ ' 2 and the result (5.111) we have the frame independent result
α5 M 6 dk⊥
Z
2 1 1 1 2
fe/A (xBj ) = 2 = x̃B = (xBj − 12 )
32 2
(2π) k⊥ + α2 M 2 /16 + α2 M 2 x̃2 /44 6π α (x̃B + 41 )3
2 α
2 B
(5.112)
We may check that there is a single e− in the bound state,

Z 1 Z ∞
dx̃B
dxBj fe/A (xBj ) = 21 α =1 (5.113)
0 −∞ 6πα(x̃2B + 41 )3
VI. QED IN D = 1 + 1 DIMENSIONS
In this section I apply the perturbative bound state method described in chapter IV to QED in D = 1 + 1 dimensions
(QED2 ), also known as the “Massive Schwinger model” [34–36]. QED (and QCD) in two dimensions is often considered
as a model for confinement, since the Coulomb potential is linear. The coupling e has dimension of mass, so the
dimensionless parameter relevant for dynamics is its ratio e/m to the electron mass. For e/m 1 the fermions
are weakly bound and their wave function satifies the Schrödinger equation. For e/m 1 the spectrum is that of
42
weakly interacting bosons: The strong coupling locks the fermion degrees of freedom into compact neutral bound
states. In the limit of the massless√Schwinger model (e/m → ∞) QED2 reduces to a free theory, with only a pointlike,
non-interacting massive (M = e/ π) boson field.
A perturbative approach requires the coupling to be small, e/m < 1. Highly excited states are nevertheless strongly
bound due to the linear potential. Several features are similar to those of the Dirac equation in a linear potential
discussed in section III F. There are also important differences, first of all because translation invariance allows to
define the bound state momentum. The peculiar feature of a constant (local) norm for wave functions at large
separations of the charges occurs here as well, but now it does not imply a continuous spectrum. Highly excited states
have features of duality similar to those observed for hadrons.
In section V we saw that transverse photon exchange contributes to the binding of Positronium atoms even at leading
order for non-vanishing atomic momentum P . In D = 1 + 1 there are no transverse photons. Boost covariance is
realized differently, and requires a linear potential. I shall verifty that form factors and deep inelastic scattering are
frame independent.
A. QED2 bound states in A0 = 0 gauge
1. Temporal gauge in D = 1 + 1
Quantization in temporal gauge proceeds as in section IV B 3, adapted to D = 1 + 1. The QED2 action is

Z
S = d2 x − 12 F10 F 10 + ψ̄(∂/ − m − eA)ψ

/ (6.1)
The electric field E 1 = F 10 = −∂0 A1 is conjugate to the photon field A1 , hence

1
E (t, x), A1 (t, y) = iδ(x − y)

(6.2)
The Hamiltonian is
Z Z
H = dx E 1 ∂0 A1 + iψ † ∂0 ψ − L = dx 21 (E 1 )2 + ψ † (−iα1 ∂1 + mγ 0 − eα1 A1 )ψ ≡ HV + H0 + Hint

(6.3)
Z Z Z
2
dx 12 E 1 (t, x) dx ψ̄(−iα1 ∂1 + mγ 0 ψ) dx ψ † α1 A1 (t, x)ψ

HV (t) = H0 (t) = Hint (t) = −e
Gauss’ operator is
δS
G(t, x) ≡ = ∂1 E 1 (t, x) − eψ † ψ(t, x) (6.4)
δA0 (t, x)
G(t, x) = 0 (Gauss’ law) is imposed as a constraint on physical states, fixing the remaining gauge degrees of freedom
and defining the value of E 1 for those states,
G(t, x) |physi = ∂1 E 1 (x) − eψ † ψ(x) |physi = 0

(6.5)
Solving for E 1 , using ∂x2 |x − y| = 2δ(x − y),

Z
E 1 (t, x) |physi = ∂x dy 21 e|x − y|ψ † ψ(t, y) |physi (6.6)
The vacuum |0i is a physical state with locally vanishing charge distribution,
E 1 (t, x) |0i = 0 (6.7)
The HV part of the Hamiltonian (6.3) generates an instantaneous linear potential,
e2
Z Z
1 1 2
dxdydz ∂x |x − y|ψ † ψ(y) ∂x |x − z|ψ † ψ(z) |physi

HV |physi = dx E (x) |physi =
2 8
e2
Z
=− dxdy ψ † ψ(x)|x − y|ψ † ψ(y) |physi (6.8)
4
43
2. States and wave functions of QED2
An e− e+ valence Fock state with CM momentum P is defined analogously to Positronium (5.1),

Z
|M, P i = dx1 dx2 ψ̄(x1 )eiP (x1 +x2 )/2 Φ(P ) (x1 − x2 )ψ(x2 ) |0i (6.9)
When bound by HV (6.8) this is taken as the lowest order contribution of a bound state expansion, where higher orders
are perturbatively generated by Hint . Hence the lowest order wave functions Φ(P ) (x1 − x2 ) and energy eigenvalues
E(P ) are determined by the eigenstate condition
(H0 + HV ) |M, P i = E(P ) |M, P i (6.10)

√
Each order of the perturbative expansion should be Poincaré covariant, which implies E(P ) = P 2 + M 2 . This will
be seen to be satisfied, and the P -dependence of the wave function Φ(P ) (x) determined. I do not here consider the
higher order corrections defined by Hint (6.3).
At large values of the linear potential generated by HV the state |M, P i has contributions from virtual e± pairs, as in
(3.20) and (3.24) for the Dirac case. In terms of Feynman diagrams these effects are due to Z-diagrams (Fig. 4(b)),
and they give rise to negative energy components of the wave function. The virtual pairs are implicitly included by
omitting from (6.9) the energy projectors Λ± (5.2) used for Positronium in (5.1).
Applying the free Hamiltonian to the state (6.9),
←
Z
H0 |M, P i = dx1 dx2 ψ̄(x1 )(−iα1 ∂ 1 + mγ 0 )eiP (x1 +x2 )/2 Φ(P ) (x1 − x2 )ψ(x2 )

→
− ψ̄(x1 )eiP (x1 +x2 )/2 Φ(P ) (x1 − x2 )(−iα1 ∂ 2 + mγ 0 )ψ(x2 ) |0i

(6.11)
Partially integrating the derivatives, so that they act on the wave function instead of on the electron fields,
→
Z
H0 |M, P i = dx1 dx2 ψ̄(x1 )eiP (x1 +x2 )/2 (iα1 ∂ 1 − 21 α1 P + mγ 0 )Φ(P ) (x1 − x2 )ψ(x2 )

←
+ ψ̄(x1 )Φ(P ) (x1 − x2 )(−iα1 ∂ 2 + 21 α1 P − mγ 0 )eiP (x1 +x2 )/2 ψ(x2 ) |0i

(6.12)
The instantaneous potential generated by HV (6.8) is seen from

Z
HV |M, P i = dx1 dx2 ψ̄(x1 )eiP (x1 +x2 )/2 21 e2 |x1 − x2 |Φ(P ) (x1 − x2 )ψ(x2 ) |0i (6.13)
In D = 1 + 1 the Dirac matrices can be represented by the 2 × 2 Pauli matrices. I shall use
γ 0 = σ3 γ 1 = iσ2 α1 = α1 = γ 0 γ 1 = σ1 (6.14)
With this notation the bound state condition (6.10) implies for the wave function
i∂x σ1 , Φ(P ) (x) − 12 P σ1 , Φ(P ) (x) + m σ3 , Φ(P ) (x) = E − V (x) Φ(P ) (x)

V (x) = 21 e2 |x| ≡ V 0 |x| = V 0 x (x ≥ 0) (6.15)
In the following I assume that x ≥ 0. The wave function for x < 0 is then determined by its parity ηP as in (5.13),
σ3 Φ(P ) (−x)σ3 = ηP Φ(−P ) (x) (ηP = ±1) (6.16)
It can be shown (see Exercise A 19 for the derivation in D = 3 + 1) that (6.15) is equivalent to the two coupled
equations
h 2 → i 2i iV 0
iσ1 ∂ x + mσ3 − 12 σ1 P − 1 Φ(P ) = − P ∂x Φ(P ) + σ1 , Φ(P )

E−V (E − V ) 2 (E − V )2
h ← 2 i 2i iV 0
Φ(P ) iσ1 ∂ x − mσ3 + 21 σ1 P (P )
σ1 , Φ(P )

−1 = 2
P ∂x Φ − 2
(6.17)
E−V (E − V ) (E − V )
44
The 2 × 2 wave function may be expanded in Pauli matrices,

Φ(P ) (x) = φ(P ) (P ) (P ) (P )
0 (x) I + φ1 (x) σ1 + φ2 (x) iσ2 + φ3 (x) σ3 (6.18)
where I stands for the unit 2 × 2 matrix. The coefficients of the Pauli matrices in the bound state equation (6.15)
give four conditions,
I: 2i∂x φ(P ) (P )
1 (x) = (E − V )φ0 (x)
σ1 : 2i∂x φ(P ) (P ) (P )
0 (x) + 2mφ2 (x) = (E − V )φ1 (x)
iσ2 : P φ(P ) (P ) (P )
3 (x) + 2mφ1 (x) = (E − V )φ2 (x)
σ3 : P φ(P ) (P )
2 (x) = (E − V )φ3 (x) (6.19)
3. Rest frame and non-relativistic limit
Consider first the rest frame, P = 0, E = M . The conditions (6.19) give

2i 2m
φ(0)
0 (x) = ∂x φ(0)
1 (x) φ(0)
2 (x) = φ(0) (x) φ(0)
3 (x) = 0
M −V M −V 1
V0
∂x2 φ(0) 2 2
1 (0)
1 (x) + φ(0)
1 (x) + 4 (M − V ) − m φ1 (x) = 0 (6.20)
M −V
In the NR limit, with e m and V (x) m, the equation for φ(0)
1 (x) reduces to the Schrödinger equation with
binding energy Eb = M − 2m,
h 1 i
− ∂x2 + V (x) φ(0) (0)
1 (x) = Eb φ1 (x) (6.21)
m
The normalizable solution is given by the Airy function,
0 2/3

φ(0)
1 (x) = N Ai m(V − Eb )/(mV ) (x ≥ 0) (6.22)
The coefficient N may be chosen to be real, with size fixed by the normalization (5.4) of the state. The energy
eigenvalues are determined by continuity at x = 0. From (6.16) and (6.18) follows φ(0) (0)
1 (−x) = −ηP φ1 (x), so that
φ(0)
1 (x = 0) = 0 (ηP = +1) ∂x φ(0)
1 (x = 0) = 0 (ηP = −1) (6.23)
The relations (6.20) reduce in the NR limit to φ(0) (0) (0) (0)
2 (x) = φ1 (x), φ0 (x) = φ3 (x) = 0. Hence the 2 × 2 wave function
has the structure
(0)
ΦN R (x) = (σ1 + iσ2 )φ(0)
1 (x) (6.24)
(0) (0)
The projectors (5.2) in the NR limit are Λ± = 21 (1 ± σ3 ). The wave function satisfies Λ+ ΦN R (x) = ΦN R (x)Λ− =
(0)
ΦN R (x), showing that it has no negative energy components. Hence there are no virtual e− e+ pairs in the NR bound
state (6.9).
4. Solution for any M and P
Consider now the bound state conditions (6.19) for arbitrary momenta P , without assuming V E. The last two
relations allow to express φ(P ) (P ) (P )
2 (x) and φ3 (x) in terms of φ1 (x),
E−V P
φ(P )
2 (x) = 2mφ(P )
1 (x) φ(P )
3 (x) = 2mφ(P )
1 (x) (6.25)
(E − V )2 − P 2 (E − V )2 − P 2
The denominators are the square of the kinetic 2-momentum Π(x) ≡ (E −V, P ). This motivates changing the variables
x into the “Lorentz invariant” τ (x), defined as
τ (x) ≡ (E − V )2 − P 2 /V 0

∂x = −2(E − V )∂τ (6.26)
45
The relation between ∂x and ∂τ is crucial in the following, and is valid only for a linear potential, V (x) = V 0 x (x ≥ 0).
When the equations (6.19) for φ(P ) (P )
0 (x) and φ1 (x) are expressed in terms of τ rather than x they turn out to be frame
independent, i.e., no factors of E or P appear [63]. With the shorthand notation φ0,1 (τ ) ≡ φ(P )
0,1 x(τ ) ,
i i 4m2
∂τ φ1 (τ ) = φ0 (τ ) ∂τ φ0 (τ ) = 1 − 0 φ1 (τ ) (6.27)
4 4 V τ
The superscript
(P ) on φ0,1 (τ ) is omitted since as functions of τ they are the same in all frames. The P -dependence
of φ(P
0,1
)
x(τ ) as functions of x arises only from the mapping x(τ ) defined by (6.26). The equivalence of this with the
P -dependence induced by actually boosting the state was verified in [64] (in A1 = 0 gauge).
The parity constraint (6.16) on Φ(P ) (x) implies, in view of the expansion (6.18), the relations
φ(P ) (−P )
0,3 (x) = ηP φ0,3 (−x) φ(P ) (−P )
1,2 (x) = −ηP φ1,2 (−x) (6.28)
Consider first φ0 and φ1 , which for x ≥ 0 are functions only of τ in (6.27). Since τ (x) is invariant under P → −P
we need not be concerned with the sign change of P under
parity. Continuity at x = 0 requires for ηP = +1 that
φ1 τ (x = 0) = 0 and for ηP = −1 that φ0 τ (x = 0) = 0. The relations (6.27) ensure that ∂τ φ1 (τ ) = 0 when
φ0 (τ ) = 0 and vice versa, as required by the opposite parities of φ0 and φ1 .
For P = 0 (6.26) gives V 0 τ (x = 0) = M 2 . Hence the condition φ1 (τ = M 2 /V 0 ) = 0 determines the masses M of
the bound states with ηP = +1. Similarly, the zeros of φ0 (τ ) determine the ηP = −1 masses. When P 6= 0 we
have V 0 τ (x = 0) = E 2 − P 2 , whereas the zeros of the functions φ0,1 (τ ) are independent of P . Satisfying the parity
2 2
2E to satisfy E2 − P
constraint for all P then requires the energies = M 2 , as expected from Lorentz covariance. This
0
allows to express τ (x) in (6.26) as τ (x) = M − 2EV + V /V .
The function φ(P ) (P ) (P )
2 (x) given by (6.25) has the same x → −x symmetry as φ1 (x) as required by (6.28). φ3 (x) is
(P )
likewise related to φ1 (x) by a coefficient which is symmetric under x → −x, but antisymmetric under P → −P .
Hence φ3(P ) (x) has opposite parity constraint compared to φ(P )
1 (x), which is again consistent with (6.28).
Defining the x-dependent “boost parameter” ζ(x) by
E−V P
cosh ζ = √ sinh ζ = √ (6.29)
V 0τ V 0τ
the full wave function (6.18) may be expressed using (6.25) as

h 2m(E − V ) 2mP i −σ1 ζ/2
2m
Φ(P ) = φ0 + φ1 σ1 + iσ 2 + σ 3 = e φ 0 + φ 1 σ 1 + √ φ 1 iσ 2 eσ1 ζ/2 (6.30)
V 0τ V 0τ V 0τ
In the latter expression the term in ( ) depends on τ only, whereas ζ depends also explicitly on P . In the weak
coupling limit (V m) ζ(x) reduces to the standard boost parameter ξ,
E P
cosh ξ = sinh ξ = (6.31)
M M
The expression (6.30) allows to determine the frame dependence of Φ(P ) (x) at constant x,
∂Φ(P ) (x)

xP (P ) E
σ1 , Φ(P )

= ∂x Φ − (6.32)
∂ξ
x E−V
ξ 2(E − V )
where the x-derivative on the rhs. is taken at constant P .
Exercise A 13: Derive the expression (6.32).

Hint: You may start from eσ1 ζ/2 Φ(P ) e−σ1 ζ/2 , which is a function only of τ .
Eliminating φ0 (τ ) in (6.27) gives
1 4m2
∂τ2 φ1 (τ ) + 1 − 0 φ1 (τ ) = 0 (6.33)
16 V τ
46
The regular, arbitrarily normalized analytic solutions for φ0,1 (τ ) are4 [48]
√
φ1 (τ ) = V 0 τ exp(−iτ /4) 1 F1 (1 − im2 /2V 0 , 2, iτ /2) = φ∗1 (τ )
√
φ0 (τ ) = −φ1 (τ ) − 4i V 0 exp(−iτ /4) 1 F1 (1 − im2 /2V 0 , 1, iτ /2) = −φ∗0 (τ ) (6.34)
5. Weak coupling limit
The Positronium states of QED4 were in (5.1) defined with the Λ± projectors (5.2), and the x-dependence of the
|e− e+ i Fock state wave function Φ(P ) (x) (5.19) was found to Lorentz contract (5.52). Here I verify that the QED2 wave
functions have analogous properties in the weak coupling limit, although there is no transverse photon contribution
and the states (6.9) are defined without projecting on the lowest Fock state.
According to (6.19) the Dirac structure (6.18) of the QED2 wave function reduces for V m and M ' 2m to
(P )
E P
ΦN R (x) = σ1 + iσ2 + σ3 φ(P )
1,N R (x) (6.35)
M M
and the variable τ of (6.26) simplifies to
E
V 0 τN R (x) = M 2 − 2EV (x) = M (M − 2V 0 x cosh ξ) cosh ξ = (6.36)
M

The dependence on x cosh ξ means that φ1 τN R (x) Lorentz contracts similarly as F (P ) (x) in (5.52). The leading
order expression (A.31) of the projectors Λ± is in D = 1 + 1
1
Λ± (P ) = (E ∓ σ1 P ± M σ3 ) (6.37)
2E
The Dirac structure of the QED2 wave function is σ1 + iσ2 for P = 0 (6.24). The analogy with QED4 (5.51) suggests
(P )
that ΦN R (x) should for general P be a 2 × 2 matrix proportional to
M (E + M ) E P
Λ+ (P )(σ1 + iσ2 )Λ− (P ) = σ1 + iσ 2 + σ3 (6.38)
2E 2 M M
which agrees with (6.35): The properties of the weakly bound states in QED2 and QED4 are analogous at all P .
The non-relativistic limit of φ1 (τ ) may be determined from its analytic expression (6.34). The scaling of the coordinate
x in the limit m → ∞ at fixed V 0 = 12 e2 is given by the Schrödinger equation (6.21) for P = 0: ∂x2 ∝ mV 0 x, i.e.,
x ∝ (mV 0 )−1/3 , and Eb = M − 2m ∝ (mV 0 )2/3 /m. In this limit [48, 65]
√ V 0 1/3 πm2 /2V 0 0 2/3
h
2 0
−2/3 i
φ(0) lim φ1 (τ ) = 4 V 0
1,N R (x) = m→∞ e Ai m(V − Eb )/(mV ) 1 + O m /V (6.39)
m2
which relates the normalization of the NR solution (6.22) to that of the general solution (6.34).
6. Large separations between e− and e+
The variable τ (x) (6.26) grows with the separation x of the fermions. For |τ | → ∞ the wave function φ1 (τ ) (6.34)
oscillates with constant amplitude, up to corrections of O (1/|τ |):
4V 0 p 2 0
h i
φ1 (|τ | → ∞) = √ exp(πm2 /V 0 ) − 1 e−θ(−τ )πm /2V cos 14 τ − (m2 /2V 0 ) log( 21 |τ |) + arg Γ(1 + im2 /2V 0 ) − π/2
πm
(6.40)
√
4 The factor V 0 in the wave function gives it the correct dimension, corresponding to a relativistically normalized state in D = 1 + 1.
Analytic solutions are given in [48] also for bound states of fermions with unequal masses.
47
where θ(x) = 1 (0) for x > 0 (x < 0) is the step function. From (6.27) we have φ0 (τ ) = −4i∂τ φ1 (τ ), so that
∗
φ1 (τ ) + φ0 (τ ) = lim φ1 (τ ) − φ0 (τ ) = N exp iτ /4 − i(m2 /2V 0 ) log(|τ |/2) + i arg Γ(1 + im2 /2V 0 ) − iπ/2

lim
|τ |→∞ |τ |→∞
4V 0 p 2 0
N=√ exp(πm2 /V 0 ) − 1 e−θ(−τ )πm /2V (6.41)
πm
has an x-independent local norm N 2 . This x-dependence of the wave function is made possible by modes with large
negative kinetic energy, which balance the linear potential to give a fixed energy eigenvalue. The asymptotic wave
function thus describes virtual e− e+ pairs, illustrated by time-ordered Z-diagrams such as in Fig. 4 b. The negative
energy components created by the b d operators in the state (6.9) dominate for x → ∞ [48, 65]. The pairs give rise to
a sea distribution for xBj → 0 in deep inelastic scattering (see section VI C and Fig. 8). The Dirac radial functions
(3.57) similarly have a constant local norm at large values of r.
7. Bound state masses and duality
The wave function φ1 (τ ) in (6.34) was chosen to satisfy φ1 (τ = 0) = 0, ensuring that φ2 (0) and φ3 (0) (6.25) are finite.
The general solution of the differential equation (6.33) has φ1 (τ = 0) 6= 0, giving singular φ2 and φ3 . The requirement
that the wave function is regular at τ = 0 implies a discrete QED2 spectrum. This criterion of local normalizability is
the relativistic generalization of the requirement of a finite global norm for Schrödinger wave functions. In fact, the
non-relativistic limit of the general solution for φ1 (τ ) (with singular φ2,3 (τ = 0)) adds an Airy Bi function to (6.39),
which increases exponentially at large x [48].
The bound state masses M of locally normalizable solutions are determined by the parity constraint (6.28) at x = 0,
which requires φ1 (τ = M 2 /V 0 ) = 0 for ηP = +1, and ∂τ φ1 (τ = M 2 /V 0 ) = 0 for ηP = −1. At high masses M we may
use the asymptotic expression (6.40) for φ1 (τ → ∞). The squared eigenvalues Mn2 are then given by integers n and
lie on asymptotically linear “Regge trajectories”,
Mn2 = n 2πV 0 + O m2 log n ηP = (−1)n

(6.42)
For small electron masses m the trajectory is linear down to low excitations [48]. In the range of x where V (x) M
the state (6.9) is dominated by positive energy b† d† contributions, as would be expected for parton-hadron duality
[48, 65]. The states have an overlap with multiple bound states generated via string breaking as shown in Fig. 10 a.
I return to these issues in the context of QCD in D = 3 + 1 dimensions (section VII D).
Consider next a ground state of mass M . With increasing x the first virtual e− e+ pair (string breaking) in the P = 0
wave function is expected when the potential has reached twice the mass of the bound state, i.e., at V (x) = 2M .
This energetically allows a (virtual) bound state pair to appear (as depicted
√ in Fig. 10 b below). I illustrate this with
a numerical example in Fig. 6(a), where e/m = 0.71 and M = 4.86 V 0 . The dynamics is nearly non-relativistic at
low x, as evidenced by the agreement of the blue (exact, (6.34)) and red dashed (Schrödinger, (6.39)) wave functions.
Both are exponentially suppressed with increasing x. In the relativistic range, V (x) & M , the exact wave function
(blue line) begins to increase, reaching a maximum at V (x) = 2M . The wave function is symmetric around V 0 x = M
because τ (x) in (6.26) satisfies τ (x) = τ (2M/V 0 −x).5 For P = M sinh ξ > 0 the virtual pair is expected at V (x) = 2E,
since each bound state has increased energy due to the boost: M cosh ξ = E. This is verified in Fig. 6(b), and is again
due to the symmetry of τ (x).
5 There is no symmetry for V 0 x > 2M since the wave function is defined by parity for x < 0.
48
√ √
FIG. 6. The wave function φ1 τ (x) of the ηP = −1 ground state for m = 2 V 0 , i.e., e/m = 1/ 2. Blue line: Relativistic
√
expression√(6.34), implying a bound state mass M = 4.86 V 0 . Dashed red line: Non-relativistic Airy
√ function (6.39), requiring
Eb = 0.81 V 0 . Both functions are normalized to unity at x = 0. (a) Rest frame, P = 0. (b) P = 5 V 0 , with the non-relativistic
(dashed red) line Lorentz contracted, x → xE/M as in (6.36).
B. * Bound state form factors in QED2
1. Form factor definition and symmetry under parity
The electromagnetic form factor is defined as for Positronium in D = 3 + 1, see section V E. The form factor for
γ ∗ + A → B is as in (5.73) and (5.76),
Z
µ
FAB (q) = d2 z e−iq·z hMB , PB |ψ̄(z)γ µ ψ(z) |MA , PA i = (2π)2 δ 2 (PB − PA − q)GµAB (q)
Z ∞
GµAB (q) = dx ei(PB −PA )x/2 Tr Φ†B (x)γ µ γ 0 ΦA (x) − Φ†B (−x)ΦA (−x)γ 0 γ µ

(6.43)
−∞
where the two contributions to GµAB (q) arise from scattering on e− and e+ , respectively. The wave function ΦA (x) ≡
(P )
ΦA A (x) determines the state as in (6.9), and γ 0 , γ 1 are defined in (6.14). The e− and e+ contributions can be related
using an analogy to charge conjugation in D = 3 + 1 (5.17). According to the relations (6.25) the parity relations
(6.28) become, when the sign of P is not reversed,
φ(P ) (P )
0 (x) = ηP φ0 (−x) φ(P ) (P )
1,2,3 (x) = −ηP φ1,2,3 (−x) (6.44)
This implies for the wave function expressed as in (6.18), in analogy to charge conjugation
T
σ2 Φ(P ) (−x) σ2 = ηP Φ(P ) (x)

(6.45)
Bracketing the trace of the second term in (6.43) with σ2 and transposing it,
T
−Tr σ2 Φ†B (−x)ΦA (−x)γ 0 γ µ σ2 = −ηPA ηPB Tr γ µ γ 0 ΦA (x)Φ†B (x)

(6.46)
where σ2 (γ 0 γ µ )T σ2 = γ µ γ 0 for µ = 0, 1. Hence the e− and e+ contributions are related similarly as in (5.78) and
Z ∞
GµAB (q) = (1 − ηPA ηPB ) dx ei(PB −PA )x/2 Tr Φ†B (x)γ µ γ 0 ΦA (x)

(6.47)
−∞
vanishes unless ηPA = −ηPB .

49
2. Gauge invariance
I follow the derivation in [48] and consider only scattering from e− , i.e., leave out the factor 1 − ηPA ηPB of (6.47).
0 1
Denoting EA,B = PA,B and PA,B = PA,B , gauge invariance requires that
Z ∞
qµ GµAB = (PB − PA )µ GµAB = dx ei(PB −PA )x/2 Tr Φ†B (EB − EA ) + (PB − PA )σ1 ΦA = 0

(6.48)
−∞
The bound state equation (6.15) for ΦA and Φ†B are, with V = V 0 x and x ≥ 0,

†
(EA − V )ΦA = i∂x {σ1 , ΦA } − 21 PA [σ1 , ΦA ] + m [σ3 , ΦA ] − ΦB ×

Φ†B (EB − V ) = −i∂x σ1 , Φ†B + 12 PB σ1 , Φ†B − m σ3 , Φ†B

× ΦA (6.49)

Multiplying the equations as indicated in the margin their sum becomes
Φ†B (EB − EA )ΦA = − iΦ†B ∂x {σ1 , ΦA } − i∂x σ1 , Φ†B ΦA + 21 PB σ1 , Φ†B ΦA + 12 PA Φ†B [σ1 , ΦA ]

h i
− m σ3 , Φ†B ΦA − mΦ†B [σ3 , ΦA ] (6.50)
When the trace is taken the terms with ∂x form a total derivative,
−iTr Φ†B (σ1 ∂x ΦA + ∂x ΦA σ1 ) + (σ1 ∂x Φ†B + ∂x Φ†B σ1 )ΦA = −iTr ∂x Φ†B σ1 ΦA + Φ†B ΦA σ1

(6.51)
Partially integrating this term in (6.48) it becomes
− 12 (PB − PA )Tr Φ†B σ1 ΦA + Φ†B ΦA σ1

(6.52)
The O (P ) terms in (6.50) may in the trace be expressed as

1
(σ1 Φ†B − Φ†B σ1 )ΦA + 12 PA Tr Φ†B (σ1 ΦA − ΦA σ1 ) = 12 (PB − PA )Tr Φ†B ΦA σ1 − Φ†B σ1 ΦA

2 PB Tr (6.53)
The O (m) term vanishes when traced,
−mTr (σ3 Φ†B − Φ†B σ3 )ΦA + Φ†B (σ3 ΦA − ΦA σ3 ) = 0

(6.54)
The sum of (6.52) and (6.53) gives
−(PB − PA )Tr Φ†B σ1 ΦA

(6.55)
This cancels the second term in (6.48), thus ensuring gauge invariance.
3. Lorentz covariance
Lorentz covariance requires that the form factor can be written, given the gauge invariance (6.48),
GµAB (q) = µν (PB − PA )ν GAB (q 2 ) 01 = −10 = 1 (6.56)
With EA,B = MA,B cosh ξ and PA,B = MA,B sinh ξ we have
δEA,B δPA,B
= PA,B = EA,B (6.57)
δξ δξ
Recalling that P0 = P 0 whereas P1 = −P 1 we should have
δG0AB (q) δG1AB (q)

= G1AB (q) = G0AB (q) (6.58)
δξ δξ
50
Subtracting the coupled bound state equations in (6.17) gives an expression for
P V0
∂x Φ(P ) = − 12 ∂x σ1 , Φ(P ) − 41 iP σ1 , Φ(P ) + 12 im σ3 , Φ(P ) + σ1 , Φ(P )

(6.59)
E−V 2(E − V )
Using this in (6.32) gives
∂Φ(P )
= − 12 ∂x x σ1 , Φ(P ) + 12 ix − 12 P σ1 , Φ(P ) + m σ3 , Φ(P )

∂ξ
∂Φ(P )†
= 12 ∂x x σ1 , Φ(P )† + 12 ix 12 P σ1 , Φ(P )† − m σ3 , Φ(P )†

(6.60)
∂ξ
According to the expression (6.47) for GµAB (without the factor 1 − ηPA ηPB ),
∂GµAB
Z ∞ ∂
dx ei(PB −PA )x/2 12 ix(EB − EA )Tr Φ†B (x)γ µ γ 0 ΦA (x) + Tr Φ†B (x)γ µ γ 0 ΦA (x)

= (6.61)
∂ξ −∞ ∂ξ
The second term has the contributions
n ∂Φ† o n o
γ γ ΦA = 12 Tr σ1 , ∂x (xΦ†B ) γ µ γ 0 ΦA + ix 12 PB σ1 , Φ†B γ µ γ 0 ΦA − m σ3 , Φ†B γ µ γ 0 ΦA
B µ 0

Tr
∂ξ
n ∂ΦA o 1 n o
Tr Φ†B γ µ γ 0 = 2 Tr − Φ†B γ µ γ 0 [σ1 , ∂x (xΦA )] + ix − 12 PA Φ†B γ µ γ 0 {σ1 , ΦA } + mΦ†B γ µ γ 0 {σ3 , ΦA } (6.62)
∂ξ
The O (P ) terms may be expressed as

1
1 † † µ 0 1 † µ 0
2 ixTr 2 PB (σ1 ΦB + ΦB σ1 )γ γ ΦA − 2 PA ΦB γ γ (σ1 ΦA + ΦA σ1 )
= 12 ixTr − 12 PB σ1 , Φ†B γ µ γ 0 ΦA − 12 PA Φ†B γ µ γ 0 [σ1 , ΦA ] + 12 ixTr (PB − PA )σ1 Φ†B γ µ γ 0 ΦA

(6.63)
The O (m) terms are

† µ 0 † †
1 µ 0
σ3 , Φ†B γ µ γ 0 ΦA + Φ†B γ µ γ 0 σ3 , ΦA
1
2 ixmTr ΦB γ γ (σ3 ΦA + ΦA σ3 ) − (σ3 ΦB + ΦB σ3 )γ γ ΦA = 2 ixmTr
(6.64)
In (6.61) write EB − EA = (EB − V ) − (EA − V ) and add the O (P ) (6.63) and O (m) (6.64)contributions to the
respective terms, so as to be able to make use of the bound state equations (6.49) for ΦA and Φ†B ,
∂GµAB
Z ∞
dx ei(PB −PA )x/2 12 Tr σ1 , ∂x (xΦ†B ) γ µ γ 0 ΦA − Φ†B γ µ γ 0 [σ1 , ∂x (xΦA )]

=
∂ξ −∞
+ 21 ixTr Φ†B (EB − V ) − 12 PB σ1 , Φ†B + m σ3 , Φ†B γ µ γ 0 ΦA

− 12 ixTr Φ†B γ µ γ 0 (EA − V )ΦA + 21 PA σ1 , ΦA − m σ3 , ΦA

+ 21 ixTr (PB − PA )σ1 Φ†B γ µ γ 0 ΦA

(6.65)
According to the BSE (6.49) the expression in ( ) on the second line equals −i∂x σ1 , Φ†B and that on the third line

equals i∂x {σ1 , ΦA }. Combined with the expression on the first line, noting also the ∂x x = 1 contribution we have
∂GµAB
Z ∞
dx ei(PB −PA )x/2 12 Tr σ1 Φ†B − Φ†B σ1 γ µ γ 0 ΦA − Φ†B γ µ γ 0 σ1 ΦA − ΦA σ1

=
∂ξ −∞

+ x∂x Tr σ1 Φ†B γ µ γ 0 ΦA + 12 ix(PB − PA )Tr σ1 Φ†B γ µ γ 0 ΦA

Z ∞
dx ei(PB −PA )x/2 Tr Φ†B σ1 γ µ γ 0 ΦA

=− (6.66)
−∞

I partially integrated the first term on the second line and noted that γ µ γ 0 , σ1 = 0. Comparing with the expression
(6.47) for GµAB verifies the Lorentz covariance condition (6.58).
51
C. * Deep Inelastic Scattering in D = 1+1
1. The Bj limit of the form factor
I considered Deep Inelastic Scattering (DIS) e− + A → e− + X on Positronium atoms A of QED4 in section V F,

demonstrating its frame invariance. In that case the final state was taken to be a free e− e+ pair. In QED2 there are
no free electrons, and bound states can have arbitrarily high mass. The target vertex is then described by the form
factor γ ∗ + A → B, where B is a bound state whose mass is ∝ Q in the Bj limit (5.97). In D = 1 + 1 the mass selects
a unique B, which represents the inclusive system X.
The present approach to DIS in QED2 was previously considered in the Breit frame [48], where q 0 = EB − EA = 0
while q 1 = −Q → −∞ in the Bj limit. This is a standard frame for QCD4 , which allows the target to be described in
terms of a parton distribution. It is instructive to repeat the QED2 calculation in a frame where the target momentum
is kept fixed in the Bj limit, and to verify that the parton distribution is indeed boost invariant.
The parton distribution was in Eq. (A22) of [48] defined in terms of the invariant form factor GAB (q 2 ) (6.56) as
1 1
f (xBj ) = |Q2 GAB (q 2 )|2 (6.67)
16πV 0 m2 xBj
I consider the Bj limit where the photon momentum q 1 → −∞ at fixed xBj , in a frame where the target 2-momentum
PA is fixed. Since q = PB − PA ,
Q2 (PB − PA )2 2PA · PB − MB2 MB2 MB2

xBj = =− ' =1− '1− (6.68)
2PA · q 2PA · (PB − PA ) 2PA · PB 2PA · PB 2PA+ EB
where I used 2PA · PB ' 2(EA + PA1 )EB ≡ 2PA+ EB , neglecting the finite difference between PB0 ≡ EB and −PB1 ,
MB2
q
EB = MB2 + (PB1 )2 ' |PB1 | + = −PB1 + PA+ (1 − xBj ) i.e. PB+ = PA+ (1 − xBj ) (6.69)
2|PB1 |
Defining γ ± = γ 0 ± γ 1 = σ3 ± iσ2 and with V 0 τ = [(E − V ) + P 1 ][(E − V ) − P 1 ] the expression (6.30) for the bound
state wave functions becomes
h 2m(E − V ) 2mP 1 i h mγ + mγ − i
Φ = φ0 + φ1 σ 1 + iσ 2 + σ3 = φ 0 + φ 1 σ 1 + − (6.70)
V 0τ V 0τ E − V − P1 E − V + P1
For PB1 → −∞ this gives using (6.69)

h mγ − i
ΦB = φB0 + φB1 σ1 − (6.71)
PA+ (1 − xBj ) − V
The invariant form factor GAB (q 2 ) (6.56) may be expressed in terms of G0AB (q) (6.47). Using γ + γ + = 0 and
γ + γ − = 2(1 − σ1 ) as well as (6.45) we get for bound states of opposite parities ηA ηB = −1,
Z ∞ Z ∞
1 0 2 1 4i
GAB (q 2 ) = − dx eiq x/2
Tr Φ†B (x)ΦA (x) = 1 1
Tr Φ†B (x)ΦA (x)

1
GAB (q) ' dx sin 2q x
q EB −∞ EB 0
† h 2m2 i
1
ΦB (x)ΦA (x) = φ∗B0 (τB )φA0 (τA ) + φ∗B1 (τB )φA1 (τA ) 1 + +

2 Tr (6.72)
+
PA (1 − xBj ) − V (PA − V )
The arguments of the wave functions are, with V = V 0 x:
V 0 τA = MA2 − 2EA V + V 2
M2
V 0 τB = 2EB B
+ V 2 = 2EB PA+ (1 − xBj ) − V + V 2

−V (6.73)
2EB
52
At fixed x, τB → ±∞ for PA+ (1 − xBj ) >< V . We may thus use the asymptotic expressions (6.40) and (6.41) for the
φB wave functions, see also (A.45). For ηB = +1 and with n an integer,
0 p Z ∞
n 16iV
2 2 /V 0
dx exp − θ(−τB )πm2 /2V 0

GAB (q ) =(−1) √ e πm −1
πmEB 0
n h 2m2 io
× i sin ϕB (x)φA0 (τA ) + cos ϕB (x)φA1 (τA ) 1 + + +
PA (1 − xBj ) − V (PA − V )
+ m2 V 0x
1 0 2
1
PA (1 − xBj ) − PA1 x +

ϕB (x) ≡ log − − 4V x (6.74)

2 2V 0 PA+ (1 − xBj )
1
Exercise A 14: Derive the expression (6.74).
So far I arbitrarily normalized the wave functions by adopting the solutions (6.34). Since MB ∝ Q we need to know
the relative normalization of the wave functions ΦB (τB ) in the Bj limit. This may be determined using duality, as
shown in [48]. At large MB and for V (x) MB the bound state wave functions have the form of free e− e+ states.
The normalization of ΦB (x = 0) can thus be chosen to agree with that of free e− e+ (“partonic”) states created by
a pointlike current. The result is that the normalization factor is independent of MB and a function of the electron
mass m only, see Eq. (4.16) of [48]. Hence the Bj limit and the xBj -dependence of f (xBj ) at fixed m are not affected
by this normalization of ΦB . I do not here consider the normalization of ΦA , which affects the magnitude of f (xBj ).
Using Q2 = 2xBj PA · PB from (6.68) the electron distribution (6.67) becomes
(PA · PB )2 xBj
f (xBj ) = |GAB (q 2 )|2 (6.75)
4πV 0 m2
From the boost invariance of GAB (q 2 ) shown in section VI B 3 follows that also f (xBj ) is invariant. The expression
(6.75) is finite in the Bj limit, given that 2PA · PB ' PA+ PB− ' 2PA+ EB and that EB GAB (q 2 ) (6.74) is finite. I next
discuss the numerical evaluation of the x-integral in (6.74).
2. Numerical evaluation of the electron distribution
The integrand of GAB (q 2 ) in (6.74) is regular at PA+ − V 0 x = 0 since φA1 (τA = 0) = 0. Similarly φB1 (τB = 0) = 0.
In the Bj limit τB = 0 implies PA+ (1 − xBj ) − V 0 x = 0 (6.73), i.e., x = x0 with
x0 = PA+ (1 − xBj )/V 0 (6.76)
The τB → ∞ limit of ΦB assumed in (6.74) fails at x = x0 . In the asymptotic expression for ΦB the 1/(x − x0 )
singularity is regulated by the phase ϕB (x) ∝ log |x − x0 | in cos[ϕB (x)]. This requires attention in a numerical
evaluation of the integral. A Principal Value prescription cannot be used due to the step function θ(−τB ) ' θ(x − x0 ).
In the following exercise I outline a method for the numerical evaluation of the x-integral.
Exercise A 15: Do the x-integral in (6.74) numerically for the parameters in Fig. 7, and compare the results.
A check of the boost invariance of f (xBj ) (6.75) is shown in Fig. 7. The electron distributions obtained in the rest
frame and in a frame with boost parameter ξ = 1 closely agree.
The electron distribution (6.75) in the rest frame (PA1 = 0) is compared with the one in the Breit frame (PA1 =
Q/2xBj → ∞ in the Bj limit) in Fig. 8. The agreement shows the equivalence of the Breit and target rest frames.
3. xBj → 0 limit of the electron distribution
The electron distribution in Fig. 8 increases for xBj → 0, analogously to the sea quark distribution for hadrons. In
Eq. (6.17) of [48] the leading xBj -dependence was found to be (with scale e2 = 2V 0 ),
xBj f (xBj ) ∼ cos2 m2 log xBj + 21 MA2 /e2

(6.77)
53
√
7. QED2 electron distributions (6.75) for m = 0.5 V 0 and .05 < xBj < .95 of target A ground state (ηA = −1, MA =
FIG. √
2.674 V 0 ) evaluated in the rest frame (ξA = 0, blue) and at ξA = 1 (EA = MA cosh ξA , thick red dashed line). The two curves
agree at O 10−5 , which indicates the accuracy of the numerical evaluation. The overall normalization is arbitrary.

√ √
FIG. 8. QED2 electron distribution (6.75) for m = 0.14 V 0 (MA = 2.52 √V 0 ) in the rest frame (PA = 0, red curves) compared
with the Breit frame result in Fig. 8 of [48], blue dots (m = 0.1 e = 0.14 V 0 , see (6.15)). The normalization of the red curve
was treated as a free parameter.
States of the form (6.9) appear to have just a single e− e+ pair, created by the electron fields ψ̄ and ψ. However, a
strong electric field creates virtual e− e+ pairs. In time-ordered Feynman diagrams they shows up in “Z”-diagrams like
Fig. 4(b), where an electron scatters into a negative energy state, creating an intermediate state with an additional
e− e+ pair. The mixing of the b† and d operators is explicit for the Dirac states created by the c†n operator (3.20), and
the Dirac ground state Ω (3.24) has an indefinite number of pairs in the free state basis.
The constant norm of the bound state wave functions at large x (6.41) apparently reflects the virtual pairs created
by the linear potential V 0 x. The dominant contribution to the form factor (6.74) for xBj → 0 comes from the large x
part of the integrand, namely from I1 in (A.50). More precisely, the leading behavior is due to I1c (A.52), for which
the angle ϕC (A.54) depends on x mainly through 12 PA+ xBj x. This allows the integration over u = x − x1 in I1c
(A.55) to contribute over a range ∝ 1/xBj . To leading order in the xBj → 0 limit we may set x1 = 0.
The logarithmic terms in ϕC are at leading order in the x → ∞ limit,
h x τ i
0 A
= log( 12 PA+ x) 1 + O x−1

log (6.78)
2(x − x0 )
Hence
2 0 Z i∞
4V 0 e−πm /4V p h im2 MA2 im2 i
I1c '− √ 2 0
2 sinh(πm /2V ) Im dx exp 12 iPA+ xBj x + log( 1 +
2 P A x) − i − i arg Γ 1 +
πm 0 2V 0 4V 0 2V 0
(6.79)
54
Defining v = − 12 iPA+ xBj x we have dx = 2idv/(PA+ xBj ), Im → Re and log( 12 Pa+ x) = log v − log xBj + 12 iπ. The
v-integral becomes
Z ∞ √
2 0 πm 2 0
dvv im /2V e−v = Γ(1 + im2 /2V 0 ) = p ei arg Γ(1+im /2V ) (6.80)
0 2
2V sinh(πm /2V )0
0
whose modulus and phase cancel the corresponding terms in (6.79). This leaves
8V 0 −πm2 /2V 0
cos (m2 log xBj ) + 21 Ma2 )/2V 0

I1 ' − + e for xBj → 0 (6.81)
xBj Pa
Since I2 and I3 give non-leading contributions in the xBj → 0 limit this defines, via (A.50), the frame independent
parton distribution (6.75). It agrees with the analytic result (6.77) [48], which was evaluated in the Breit frame.
The analytic approximation given by (6.81) for f (xBj → 0) is compared with the numerical evaluation in Fig. 9 for
xBj ≤ 0.1.
√
FIG. 9. Red line: Numerical evaluation of parton distribution (6.75) using (A.50) (Pa = 0 and m = 0.14 V 0 ).
Dashed blue line: Analytic approximation for xBj → 0 given by (6.81).
VII. APPLICATIONS TO QCD BOUND STATES
I consider color singlet QCD bound states in temporal gauge A0a = 0, as described in section IV C. The scale Λ
required for confinement is introduced via a boundary condition on the solutions of Gauss’ constraint (4.24), in terms
of the homogeneous solution (4.35). This affects the longitudinal electric field E aL for each color component of a Fock
state, whereas the full color singlet state does not generate a color octet field. The gauge invariance condition of
electromagnetc form factors is satisfied, and the q q̄ bound state energies have the correct frame dependence.
A. The instantaneous potential of various Fock states
The longitudinal electric field (4.36) determines the field energy HV (4.37), which defines an instantaneous potential
for each Fock state,
Z h αs i (0) (1)
HV (t = 0) = dydz y · z 12 κ2 dx + gκ + 12
R
Ea (y)Ea (z) ≡ HV + HV
|y − z|
† a
Ea (y) = −fabc Aib Eci (y) + ψA TAB ψB (y) (7.1)
(0)
where a sum over repeated indices is understood. HV is due to the homogeneous solution (4.35) of Gauss’ constraint
(1)
and generates an O αs0 potential, while HV gives the standard O (αs ) Coulomb potential. Recall that Ea (x) |0i = 0
55
(4.35) since Gauss’ constraint (4.24) is not an operator condition in temporal gauge. The potentials are independent
of the quark Dirac index α and of the gluon Lorentz index i = 1, 2, 3. I consider color singlet q q̄ (meson), qqq (baryon),
gg (glueball), q q̄g (higher Fock state of a meson) and q q̄ q q̄ (molecular or tetraquark) Fock states.
1. The q q̄ potential
A q q̄ Fock state with quarks at x1 , x2 , summed over the colors A, is invariant under global color transformations,
α β
|q q̄i = ψ̄A (x1 )ψA (x2 ) |0i ≡ ψ̄A (x1 )ψA (x2 ) |0i (7.2)
I suppress the irrelevant Dirac indices and t = 0 is understood. The canonical commutation relations (4.20) of the
fields in Ea (x) (7.1) give
Ea (x), ψ̄A (x1 ) = ψ̄A0 (x1 )TAa0 A δ(x − x1 )

a

Ea (x), ψA (x2 ) = −TAA 0 ψA0 (x2 )δ(x − x2 ) (7.3)
where T a is the SU(3) generator in the fundamental representation. I shall make use of the following relations for the
SU(Nc ) generators6
a b Nc2 − 1
T , T = ifabc T c Tr T a T b = 21 δ ab T a T a = CF I =

I
2Nc
a a 1 1 1
TAB TCD = δAD δBC − δAB δCD T aT bT a = − Tb
2 Nc 2Nc
fabc fabd = Nc δcd fabd T a T b = 21 iNc T d (7.4)

(0)
The weight y · z of HV is x21 , x1 · x2 or x22 depending on the quark field on which Ea (y) and Ea (z) act,
Z
(0)
HV |q q̄i = 21 κ2 dx + gκ
R
dydz y · z Ea (y)Ea (z)ψ̄A (x1 )ψA (x2 ) |0i
1 2
dx + gκ (x21 − 2x1 · x2 + x22 )ψ̄A (x1 ) TAB
a a
R
= 2κ TBC ψC (x2 ) |0i
= 21 κ2 dx + gκ CF (x1 − x2 )2 |q q̄i
R
(7.5)
The term 12 κ2 dx is the contribution of an x-independent field energy density EΛ . Its integral is proportional to
R
the volume of space and irrelevant only if universal, i.e., EΛ must be the same for all Fock states. In particular, EΛ
cannot depend on x1 or x2 . This requires to choose the normalization κ of the homogeneous solution for this Fock
state as
Λ2 1
κqq̄ = (7.6)
gCF |x1 − x2 |
which also serves to define the universal constant Λ. The field energy density is then
Λ4
EΛ = (7.7)
2g 2 CF
This value of EΛ must be imposed on all types of Fock states, e.g., |qqqi, and in each case will determine the
normalization of the corresponding homogeneous solution.
R
Subtracting EΛ dx in (7.5) the remaining gκ term gives
(0) (0) (0)
HV |q q̄i ≡ Vqq̄ |q q̄i Vqq̄ (x1 , x2 ) = gκqq̄ CF (x1 − x2 )2 = Λ2 |x1 − x2 | (7.8)
6 Useful properties of the SU(Nc ) generators may be found in [66].

56
(1)
The gluon exchange potential due to HV is similarly obtained using (7.3). The commutators of Ea (y) and Ea (z) with
the same quark now gives an infinite, ∼ 1/0 contribution. This “self-energy” is independent of x1 and x2 and can be
subtracted. Altogether,
(0) (1) (1) αs
HV |q q̄i = Vqq̄ + Vqq̄ |q q̄i Vqq̄ (x1 , x2 ) = −CF (7.9)
|x1 − x2 |
(0) (1)
Vqq̄ + Vqq̄ agrees with the Cornell potential (1.2) [9, 10]. The first term of the Fock expansion thus gives a good
approximation for heavy quarkonia.
2. The qqq potential
An SU(3) color singlet qqq Fock state has the form (suppressing the Dirac indices)
† † †
|qqqi = ABC ψA (x1 )ψB (x2 )ψC (x3 ) |0i (7.10)
where ABC is the fully antisymmetric tensor with 123 = 1. Note that this state is a color singlet of SU(Nc )
† † † † † †
only for Nc = 3. In a global transformation ψA (x) → ψA 0 (x)UA0 A the state is invariant: ABC UA0 A UB 0 B UC 0 C =
A0 B 0 C 0 det(U ) = A0 B 0 C 0 provided U is a 3 × 3 matrix with unit determinant.

(0)
When HV (7.1) operates on |qqqi the factor y · z is xi · xj for the commutator (7.3) of Ea (y) with ψ † (xi ) and of
Ea (z) with ψ † (xj ), or vice versa. It suffices to consider the two generic cases,
† † † 4
x21 : ABC ψA a a
00 (x1 )ψB (x2 )ψC (x3 ) |0i TA00 A0 TA0 A = CF |qqqi = |qqqi
3
† † † a a 1
† † †
x1 · x2 : 2ABC ψA 0 (x1 )ψB 0 (x2 )ψC (x3 ) |0i TA0 A TB 0 B = B 0 A0 C −
Nc A B C
0 0 ψA 0 (x1 )ψB 0 (x2 )ψC (x3 ) |0i
1 4
=− 1+ |qqqi = − |qqqi (7.11)
Nc 3
where I used (7.4). The two eigenvalues are equal and opposite for Nc = 3, which ensures translation invariance,
(0) 1 2
R 4 2
HV |qqqi = 2κ dx + gκ dqqq (x1 , x2 , x3 ) |qqqi
3
1 p
dqqq (x1 , x2 , x3 ) ≡ √ (x1 − x2 )2 + (x2 − x3 )2 + (x3 − x1 )2 (7.12)
2
To ensure the universal value of EΛ in (7.7), i.e., the universality of the spatially constant energy density, the
homogeneous solution in (4.36) should be normalized for the |qqqi Fock state (7.10) as
Λ2 1
κqqq = (7.13)
gCF dqqq (x1 , x2 , x3 )

This gives the O αs0 potential
(0)
2
4
dqqq (x1 , x2 , x3 ) = Λ2 dqqq (x1 , x2 , x3 )

Vqqq (x1 , x2 , x3 ) = gκqqq 3 (7.14)
(1)
The O (αs ) gluon exchange potential given by HV in (7.1) is determined by the eigenvalue of the x1 · x2 term in
(7.11),
(1) 2 1 1 1
Vqqq (x1 , x2 , x3 ) = − αs + + (7.15)
3 |x1 − x2 | |x2 − x3 | |x3 − x1 |
3. The gg potential
A gg Fock state which is invariant under global color SU(Nc ) transformations is expressed in terms of the gluon field
in temporal gauge as
|ggi = Aia (x1 )Aja (x2 ) |0i (7.16)
57
To find the action of HV (7.1) on this state we may use the canonical commutator in (4.20),
Ea (y), Aib (x1 ) = −facd Akc (y)Edk (y), Aib (x1 ) = ifabc Aic (x1 )δ(y − x1 )

h i
Ea (y), Aib (x1 )Ajb (x2 ) = ifabc Aib (x1 )Ajc (x2 ) − δ(y − x1 ) + δ(y − x2 )

h h ii
Ea (y), Ea (z), Aib (x1 )Ajb (x2 ) = Nc Aia (x1 )Aja (x2 ) δ(y − x1 ) − δ(y − x2 ) δ(z − x1 ) − δ(z − x2 )

(7.17)
Hence
(0) 1 2
dx + gκ Nc (x1 − x2 )2 |ggi
R
HV |ggi = 2κ (7.18)
and to ensure the universal value of EΛ in (7.7),
Λ2 1
κgg = √ (7.19)
g CF Nc |x1 − x2 |
This gives
r N αs
(0) (1) c
Λ2 |x1 − x2 | − Nc

HV |ggi = Vgg + Vgg |ggi = |ggi (7.20)
CF |x1 − x2 |
4. The q q̄g potential
The Hamiltonian (4.21) creates O (g) color singlet |q q̄gi Fock states from |q q̄i (7.2). I consider the instantaneous
potential generated by HV (7.1) for states of the form
|q q̄gi ≡ ψ̄A (x1 ) Aib (xg )TAB
b
ψB (x2 ) |0i (7.21)
Proceeding similarly as above gives the potential
Λ2 h 1 1 1 1 i
Vqq̄g (x1 , x2 , xg ) = √ dqq̄g (x1 , x2 , xg ) + 12 αs − Nc +
CF Nc |x1 − x2 | |x1 − xg | |x2 − xg |
q
dqq̄g (x1 , x2 , xg ) ≡ 14 (Nc − N2C )(x1 − x2 )2 + Nc (xg − 12 x1 − 21 x2 )2 (7.22)
Vqq̄g is a confining potential, as it restricts both |x1 − x2 | and the distance of xg from the average of x1 + x2 .
Exercise A 16: Derive the q q̄g potential (7.22).
5. Limiting values of the qqq and q q̄g potentials
When two of the quarks in the baryon |qqqi Fock state are close to each other the potential should (for Nc = 3) reduce
to that of the color 3 ⊗ 3̄ meson potential (7.8), (7.9). Setting x2 = x3 in Vqqq (7.14) and (7.15) (and subtracting the
infinite Coulomb energy) indeed gives
4 αs
Vqqq (x1 , x2 , x2 ) = Λ2 |x1 − x2 | − = Vqq̄ (x1 , x2 ) (7.23)
3 |x1 − x2 |
Similarly the potential of the q q̄g Fock state should reduce to the 3 ⊗ 3̄ meson potential when a quark and gluon
coincide. Setting xg = x2 in Vqq̄g (7.22) gives
αs
Vqq̄g (x1 , x2 , x2 ) = Λ2 |x1 − x2 | − CF = Vqq̄ (x1 , x2 ) (7.24)
|x1 − x2 |
On the other hand, when the quarks in q q̄g coincide the potential (7.22) should become the 8 ⊗ 8 glueball potential
(7.20). This is also fulfilled,
√
Λ2 Nc αs
Vqq̄g (x1 , xg , x1 ) = √ |x1 − xg | − Nc = Vgg (x1 , xg ) (7.25)
CF |x1 − xg |
58
6. Single quark or gluon Fock states
The above examples indicate that only color singlet Fock states have translation invariant potentials. For the record
I confirm this for single quark and gluon states.
In (7.5) we already saw that
(0) (0) 1 2
dx + gκ CF x2q |qi
R
HV |qi ≡ HV ψ̄A (xq ) |0i = 2κ (7.26)

For the O κ2 term to give EΛ (7.7) we need
Λ2 1
κq = (7.27)
gCF |xq |
This gives the potential
Vq(0) = Λ2 |xq | (7.28)
which is not invariant under translations.

The case of a single gluon is similar. Using our previous result (7.18),
(0) (0)
HV |gi ≡ HV Aia (xg ) |0i = 1 2
dx + gκ Nc x2g |gi
R
2κ (7.29)
This requires
Λ2 1
κg = √ (7.30)
g CF Nc |xg |
so that
√
Λ 2 Nc
Vg(0) = √ |xg | (7.31)
CF
I conclude, without further proof, that only color singlet Fock states are compatible with Poincaré invariance.
7. The potential of q q̄ q q̄ Fock states
As the number of quarks and gluons in a Fock state increases a subset of them may form a color singlet, and thus
not be confined. In this final example I show that this is the case for color singlet states |q q̄ q q̄i, with two quarks and
two antiquarks.
There are two ways to combine the four quarks into a color singlet. In a q q̄ basis we can have
|1 ⊗ 1i ≡ |(q1 q̄2 )1 (q3 q̄4 )1 i and |8 ⊗ 8i ≡ |(q1 q̄2 )8 (q3 q̄4 )8 i (7.32)
where qi ≡ q(xi ). The two independent configurations in a diquark basis |(q1 q3 )3̄ (q̄2 q̄4 )3 i and |(q1 q3 )6 (q̄2 q̄4 )6̄ i can be
expressed in terms of these. I shall use the q q̄ basis (7.32) here. Then
|1 ⊗ 1i = ψ̄A (x1 )ψA (x2 ) ψ̄B (x3 )ψB (x4 ) |0i (7.33)
a a
|8 ⊗ 8i = ψ̄A (x1 )TAB ψB (x2 ) ψ̄C (x3 )TCD ψD (x4 ) |0i
= 21 ψ̄A (x1 )ψB (x2 ) ψ̄B (x3 )ψA (x4 ) |0i − 1

2Nc |1 ⊗ 1i
(0) 1 2
R
The coefficients of y · z in HV |1 ⊗ 1i are, apart from the common factor 2κ dx + gκ :
(x1 − x2 )2 , (x3 − x4 )2 : CF |1 ⊗ 1i
a a
x1 · x3 , −x1 · x4 , −x2 · x3 , x2 · x4 : 2ψ̄A (x1 )TAB ψB (x2 )ψ̄C (x3 )TCD ψD (x4 ) = 2 |8 ⊗ 8i (7.34)
59
In terms of the relative separations

d12 = x1 − x2 d34 = x3 − x4 d = 12 (x1 + x2 − x3 − x4 ) (7.35)
(0)
and with a similar analysis for HV |8 ⊗ 8i,
(0)
HV |1 ⊗ 1i = 12 κ2 dx + gκ CF (x1 − x2 )2 + CF (x3 − x4 )2 |1 ⊗ 1i + 2(x1 − x2 ) · (x3 − x4 ) |8 ⊗ 8i
R
= 21 κ2 dx + gκ CF (d212 + d234 ) |1 ⊗ 1i + 2d12 · d34 |8 ⊗ 8i

R
(0) Nc2 −2 Nc2 −4

1 2
dx + gκ Nc d2 + 2
+ d234 ) + CF
R
HV |8 ⊗ 8i = 2κ 4Nc (d12 2Nc d12 · d34 |8 ⊗ 8i + Nc d12 · d34 |1 ⊗ 1i (7.36)
(0)
Exercise A 17: Derive the expression for HV |8 ⊗ 8i in (7.36).
(0) (1) (1)

The expression (7.1) for HV shows that the color structure of HV and HV is the same. Hence HV is given by the
coefficients of xi · xj , multiplied by 21 αs /|xi − xj | (for i 6= j).
(0)
The coefficients of the |1 ⊗ 1i and |8 ⊗ 8i states on the rhs. of (7.36) allow to determine the eigenstates of HV and
thus the normalizations κ of the corresponding homogeneous solutions. The coefficients, and thus the eigenstates,
depend on the separations d, d12 and d34 . At large separations of the q1 q̄2 and q3 q̄4 pairs, i.e., for |d| |d12 |, |d34 |,
HV |8 ⊗ 8i ∼ Nc d2 |8 ⊗ 8i. The other eigenstate then approaches |1 ⊗ 1i with the smaller eigenvalue ∼ CF (d212 + d234 ).
Since the separation of color octet charges gives a large potential energy it may be expected that eigenstates of the
full Hamiltonian are dominated by unconfined |1 ⊗ 1i color configurations at large separations.
B. Rest frame wave functions of q q̄ bound states

In this section I determine the O αs0 meson eigenstates of the QCD Hamiltonian (4.21) in the rest frame. The
valence quark state of the meson is (at t = 0) defined by its wave function Φαβ (x1 − x2 ),
1 X Z
|M i = √ dx1 dx2 ψ̄αA (x1 )δ AB Φαβ (x1 − x2 )ψβB (x2 ) |0i (7.37)
Nc A,B;α,β
where Nc = 3 in QCD and A, B are quark color indices. This is a singlet state under global color SU(Nc ) transfor-
mations. Contributions of higher orders in αs and are neglected here7 .
1. Bound state equation for the meson wave function Φαβ (x)

The bound state condition for the meson state (7.37) of mass M is at O αs0 ,
(q) (0)
H0 + HV |M i = M |M i (7.38)
The quark kinetic energy operator in the Hamiltonian (4.21) is
→
Z
(q) †
H0 = dx ψA (x)(−iα · ∇ + mγ 0 )ψA (x) (7.39)
h i ← h i →
(q) (q)
H0 , ψ̄(x1 ) = ψ̄(x1 )(−iα · ∇1 + mγ 0 ) H0 , ψ(x2 ) = (iα · ∇2 − mγ 0 )ψ(x2 ) (7.40)
(0)
The meson is bound by the instantaneous potential generated by HV (7.1). Its action on color singlet q q̄ Fock states
is given in (7.8),
h i
(0)
HV , ψ̄αA (x1 )ψβA (x2 ) = V 0 |x1 − x2 | ψ̄αA (x1 )ψβA (x2 ) V (x) = V 0 |x| = Λ2 |x| (7.41)
7 The hyperfine splitting of Positronium (section V D) gives an example of how higher Fock states are taken into account.
60
where a sum over the quark colors A is implied. I neglect the O (αs ) Coulomb energy (7.9).
Shifting the derivatives in (7.40) from the fields onto the wave function by partial integration in (7.38) the coefficients
of each Fock component gives the bound state equation for Φ(x) (x = x1 − x2 ),
→ ←
iα · ∇ + mγ 0 Φ(x) + Φ(x) iα · ∇ − mγ 0 = M − V (x) Φ(x)

(7.42)
where V (x) = V 0 |x|. Equivalent forms of this BSE are
i∇ · {α, Φ(x)} + m γ 0 , Φ(x) = M − V (x) Φ(x)

(7.43)
h 2 → i h ← 2 i
(iα · ∇ + mγ 0 ) − 1 Φ(x) + Φ(x) (iα · ∇ − mγ 0 ) −1 =0 (7.44)
M −V M −V
Introducing the notation
→ 2 → ← ← 2
h± ≡ (iα · ∇ + mγ 0 ) ± 1 h ± ≡ (iα · ∇ − mγ 0 ) ±1 (7.45)
M −V M −V
which satisfy (r = |x|)
→ → 4 →2
2 4iV 0 →
h− h+ = (−∇ + m ) − 1 + α · x (iα · ∇ + mγ 0 )
(M − V )2 r(M − V )3
← ← ←2 4 ← 4iV 0
h + h − = (−∇ + m2 ) − 1 + (iα · ∇ − mγ 0
) α · x (7.46)
(M − V )2 r(M − V )3
the bound state equation (7.44) is
→ ←
h − Φ(x) + Φ(x) h − = 0 (7.47)
The meson states (7.37) are relativistic (strongly bound) when m . Λ, and for high excitations at any m due to
the linear potential. The example of Dirac states (3.12) shows that for strong fields the vacuum (3.24) has b† d†
pairs in a free fermion basis, and fermion eigenstates are created by a superposition of the b† and d operators (3.20).
Analogously, the meson state (7.37) is a two-particle Fock state only in a Bogoliubov rotated operator basis. In the
free basis we may view the pairs as arising from the Z-diagrams (Fig. 4b) of a time-ordered perturbative expansion.
For DIS in QED2 the pairs give rise to a sea-like distribution of electrons ∝ 1/xBj , as discussed in section VI C 3 and
shown in Fig. 8. I return to this issue in section VII D 2.
2. Separation of radial and angular variables
(i)
The 4 × 4 wave function Φαβ (x) may be expressed as a sum of terms with distinct Dirac structures Γαβ (x), radial
functions Fi (r) and angular dependence given by the spherical harmonics Yjλ (x̂):
X (i)
Φjλ
αβ (x) = Γαβ Fi (r)Yjλ (x̂) (7.48)
i
where r = |x| and x̂ = x/r. The 16 independent Dirac structures Γ(i) should be rotationally invariant to allow a
simple classification of the states according to their angular momentum j and j z = λ. As discussed in section V A for
Positronium the generator of rotations for the quark fields is
Z
†
J = dx ψA (x) J ψA (x) J = L + S = x × (−i∇) + 12 γ5 α
Z
J |M i = dx1 dx2 ψ̄A (x1 ) [J , Φ(x1 − x2 )] ψA (x2 ) |0i (7.49)
For example,
the rotationally invariant Dirac structure α · ∇ satisfies [J , α · ∇] = 0 as shown in (A.25) and (A.26).
When J , Γ(i) (x) = 0 we get, since [L, Fi (r)] = [S, Fi (r)] = [S, Yjλ (x̂)] = 0,
h i
J , Γ(i) Fi (r)Yjλ (x̂) = Γ(i) Fi (r) [L, Yjλ (x̂)] (7.50)
61
Because Yjλ (x̂) is an eigenfunction of L2 and Lz this ensures that
J 2 |M i = j(j + 1) |M i J z |M i = λ |M i (7.51)
The Γ(i) (x) need contain at most one power of the Dirac vector α = γ 0 γ since higher powers may be reduced using
αi αj = δij + iijk αk γ5 (7.52)
Rotational invariance requires that α be dotted into a vector. I shall use the three orthogonal vectors x, L = x×(−i∇)
and x × L. Each of the four Dirac structures 1, α · x, α · L and α · x × L can be multiplied by the rotationally
invariant Dirac matrices γ 0 and/or γ5 . This gives altogether 4 × 2 × 2 = 16 possible Γ(i) (x). Other invariants may be
expressed in terms of these, e.g.,
1 1
i α · ∇ = (α · x) i ∂r + 2 α · x × L
r r
(α · ∇)(α · x) = 3 + r∂r + γ5 α · L (7.53)
The Γ(i) (x) may be grouped according to the parity ηP (5.13) and charge conjugation ηC (5.17) quantum numbers
that they imply for the wave function,
T
γ 0 Φ(−x)γ 0 = ηP Φ(x)

α2 Φ(−x) α2 = ηC Φ(x) (7.54)
Since Yjλ (−x̂) = (−1)j Yjλ (x̂) states of spin j can belong to one of four “trajectories”, here denoted by the parity
and charge conjugation quantum numbers of their j = 0 member:
0−+ trajectory [s = 0, ` = j] : −ηP = ηC = (−1)j γ5 , γ 0 γ5 , γ5 α · x, γ5 α · x × L
0−− trajectory [s = 1, ` = j] : ηP = ηC = −(−1)j γ 0 γ5 α · x, γ 0 γ5 α · x × L, α · L, γ 0 α · L
0++ trajectory [s = 1, ` = j ± 1] : ηP = ηC = +(−1)j 1, α · x, γ 0 α · x, α · x × L, γ 0 α · x × L, γ 0 γ5 α · L
0+− trajectory [exotic] : ηP = −ηC = (−1)j γ 0 , γ5 α · L (7.55)
The non-relativistic spin s and orbital angular momentum ` are indicated in brackets. Relativistic effects mix the
` = j ± 1 states on the 0++ trajectory, resulting in a pair of coupled radial equations. The j = 0 state on the 0−−
trajectory and the entire 0+− trajectory are incompatible with the s, ` assignments and thus exotic in the quark
model. They turn out to be missing also in the relativistic case. The bound state equation (7.43) has no solutions for
states on the 0+− trajectory (Γ(i) = γ 0 or γ5 α · L) since
i∇ · α, γ 0 = i∇ · {α, γ5 α · L} = m γ 0 , γ 0 = m γ 0 , γ5 α · L = 0

(7.56)
3. The 0−+ trajectory: ηP = (−1)j+1 , ηC = (−1)j
According to the classification (7.55) we expand the wave function Φ−+ (x) of the 0−+ trajectory states as
h i
Φ−+ (x) = F1 (r) + i α · x F2 (r) + α · x × L F3 (r) + γ 0 F4 (r) γ5 Yjλ (x̂) (7.57)
Using this in the bound state equation (7.42), noting that i∇ · x × L = L2 and collecting terms with the same Dirac
structure we get the conditions:
γ5 : − (3 + r∂r )F2 + j(j + 1)F3 + mF4 = 21 (M − V )F1

1
γ5 α · x : ∂r F1 = 12 (M − V )F2
r
1
γ5 α · x × L : F1 = 21 (M − V )F3
r2
γ 0 γ5 : mF1 = 12 (M − V )F4 (7.58)
Expressing F2 , F3 and F4 in terms of F1 we find the radial equation (denoting F10 ≡ ∂r F1 )

2 V 0 0 h1 j(j + 1) i
F100 + + F1 + 4 (M − V )2 − m2 − F1 = 0 (7.59)
r M −V r2
62
in agreement with the corresponding result in Eq. (2.24) of [67]. The wave function (7.57) may be expressed as
h 2 → i h ← 2 i
Φ−+ (x) = (iα · ∇ + mγ 0 ) + 1 γ5 F1 (r)Yjλ (x̂) = F1 (r)Yjλ (x̂) γ5 (iα · ∇ − mγ 0 ) +1
M −V M −V
→ ←
= h + γ5 F1 (r)Yjλ (x̂) = F1 (r)Yjλ (x̂) γ5 h + (7.60)
Exercise A 18: Verify that the expression (7.60) for Φ−+ (x) satisfies the bound state equation (7.47) given the
radial equation (7.59).
Hint: The identities (7.46) are useful.
Both the quark and antiquark contributions to the BSE have a spin-dependent (S = 12 γ5 α) interaction which cancels
in their sum. The contribution from the quark term is, taking into account the radial equation,
→ 8V 0 →
h − Φ−+ (x) = 3
S · (L γ5 − im x γ 0 ) F1 (r)Yjλ (x̂) (7.61)
r(M − V )
Non-relativistic limit of the 0−+ trajectory wave functions
The non-relativistic (NR) limit in the rest frame is defined by

V ∂ 1 √
→0 ∼ ∼ mV (7.62)
m ∂r r
The binding energy Eb ∼ V is defined by M = 2m + Eb . In the radial equation (7.59) we have
V0 V 1
= 1
4 (M − V )2 − m2 ' m(Eb − V ) (7.63)
M −V r(M − V ) r
so it becomes the radial Schrödinger equation in the NR limit,
00 2 0 h j(j + 1) i
F1,N R + F1,N R + m(Eb − V ) − F1,N R = 0 (7.64)
r r2
In the wave function (7.60) we have at leading order
→ 2
h+ = (iα · ∇ + mγ 0 ) + 1 ' 1 + γ 0 (7.65)
M −V
giving
ΦN R 0
−+ = (1 + γ )γ5 F1,N R (r)Yjλ (Ω) (7.66)
4. The 0−− trajectory: ηP = (−1)j+1 , ηC = (−1)j+1
According to the classification (7.55) we expand the wave function Φ−− (x) of the 0−− trajectory states as
h i
Φ−− (x) = γ 0 α · L G1 (r) + i γ 0 γ5 α · x G2 (r) + γ 0 γ5 α · x × L G3 (r) + mα · L G4 (r) Yjλ (x̂) (7.67)
Collecting terms with distinct Dirac structures in the bound state equation (7.43),
γ0 α · L : G2 − (2 + r∂r )G3 + m2 G4 = 21 (M − V )G1
j(j + 1)
γ 0 γ5 α · x : G1 = 21 (M − V )G2
r2
1
γ 0 γ5 α · x × L : (1 + r∂r )G1 = 12 (M − V )G3
r2
mα · L : G1 = 21 (M − V )G4 (7.68)
63
Expressing G2 , G3 and G4 in terms of G1 we find the radial equation for the 0−− trajectory,
2 V 0 0 h1 j(j + 1) V0 i
G001 + + G1 + 4 (M − V )2 − m2 − 2
+ G1 = 0 (7.69)
r M −V r r(M − V )
in agreement with the corresponding result in Eq. (2.38) of [67]. The 0−− radial equation differs from the 0−+ one
(7.59) only by the term ∝ V 0 /r(M − V ). Using
iL2 1
iα · ∇ γ · L = γ 0 γ5 α · x + γ 0 γ5 α · x × L 2 (1 + r∂r ) (7.70)
r2 r
allows the wave function to be expressed in terms of the projector h+ of (7.45) as,
→ → ←←
Φ−− (x) = h + γ · L G1 (r) Yjλ (x̂) = G1 (r) Yjλ (x̂) γ · L h + (7.71)
←i ←
where L = −i ∂ k xj εijk . The j = 0 state on the 0−− trajectory is missing since L Y00 (x̂) = 0. The quark contribution
to the bound state equation (7.47) is, with S = 21 γ5 α,
→ 4V 0 →2 0 →
h − Φ−− (x) = 3
L γ γ5 − 2m S · x × L G1 (r) Yjλ (x̂) (7.72)
r(M − V )
Non-relativistic limit of the 0−− trajectory wave functions
The NR limit of the radial equation (7.69) reduces as in the 0−+ case to
2 h j(j + 1) i
G001,N R + G01,N R + m(Eb − V ) − G1,N R = 0 (7.73)
r r2
The equality of the 0−+ and 0−− eigenvalues reflects the spin s independence of the NR limit, since ` = j for both.
The wave function is
ΦN R 0
−− = (1 + γ )α · L G1,N R (r)Yjλ (Ω) (7.74)
5. The 0++ trajectory: ηP = (−1)j , ηC = (−1)j
According to the classification (7.55) we expand the wave function Φ++ (x) of the 0++ trajectory states in terms of
six Dirac structures8 ,
nh i h io
Φ++ (x) = F1 (r) + i α · x F2 (r) + α · x × L F3 (r) + γ 0 γ5 α · L G1 (r) + i α · x G2 (r) + α · x × L G3 (r) Yjλ (x̂)
(7.75)
Collecting terms with distinct Dirac structures in the bound state equation (7.43),
1: − (3 + r∂r )F2 + j(j + 1)F3 = 21 (M − V )F1
1
α·x: ∂r F1 + mG2 = 12 (M − V )F2
r
1
α·x×L: F1 + mG3 = 21 (M − V )F3
r2
γ 0 γ5 α · L : G2 − (2 + r∂r )G3 = 12 (M − V )G1
1
γ0 α · x : j(j + 1)G1 + mF2 = 21 (M − V )G2
r2
1
γ0 α · x × L : (1 + r∂r )G1 + mF3 = 21 (M − V )G3 (7.76)
r2
8 The radial functions Fi and Gi are unrelated to those in sections VII B 3 and VII B 4.
64
It turns out to be convenient to express the above radial functions in terms of two new ones, H1 (r) and H2 (r):
2 4m
(M − V )2 − m2 H1 −
1
F1 = − ∂r (rH2 )
(M − V )2 4 M −V
1
F2 =− ∂r H1 + 2mH2
r(M − V )
1
F3 =− 2 H1
r (M − V )
G1 = 2H2
2 h m 2 i
G2 = ∂r − 2
H1 + ∂r (rH2 ) + (M − V )H2
r (M − V ) M −V
2h m 2 i
G3 = 2 − H 1 + ∂ r (rH2 ) (7.77)
r (M − V )2 M −V
The bound state conditions (7.76) are satisfied provided H1,2 satisfy the coupled radial equations,
V 0 0 h1
2 j(j + 1) i
H100 + H1 + 4 (M − V )2 − m2 −
+ H1 = 4m(M − V )H2 (7.78)
r M −V r2
2 V 0 0 h1 j(j + 1) V0 i mV 0
H200 + + H2 + 4 (M − V )2 − m2 − + H2 = H1 (7.79)
r M −V r2 r(M − V ) r(M − V )2
These agree with Eqs. (2.48) and (2.49) for F2GS and GGS GS
1 of [67], when H1 = (M −V )F2 and H2 = −i GGS
1 /(M −V ).
The wave function Φ++ (x) (7.75) can be expressed in terms of the H1,2 (r) radial functions and the h+ operators
(7.45) as
→ → m → 0
Φ++ (x) = h + − 21 H1 + 2 γ · L γ5 H2 + 2im α · x H2 Yjλ (x̂) +

h + γ H1 + 8H2 Yjλ (x̂) (7.80)
M −V
← ← ← m
= Yjλ (x̂) − 12 H1 − 2H2 γ5 γ · L + 2imH2 α · x h + − Yjλ (x̂) H1 γ 0 h + − 8H2

M −V
The quark contribution to the bound state equation (7.47) is, with S = 12 γ5 α,
→ 4V 0 h → m 0
i 8V 0 →2
L + m2 r2 γ 0 H2 (r)Yjλ

h − Φ++ (x) = − 3
S · L + γ r∂r H1 (r)Yjλ (x̂) + 3
(7.81)
r(M − V ) M −V r(M − V )
When m = 0 chiral symmetry implies that Φ(x) and γ5 Φ(x) define bound states with the same mass M , as is apparent
from the bound state equation (7.42). The radial equations (7.78) and (7.79) in fact decouple and coincide with the
radial equations of the 0−+ (7.59) and 0−− (7.69) trajectories, respectively. The Φ++ wave functions correspondingly
reduce to γ5 Φ−+ and γ5 Φ−− . I discuss the case of spontaneously broken chiral symmetry in section VII F.
Non-relativistic limit of the 0++ trajectory wave functions
States with the same j on the 0−+ and 0−− trajectories are degerate in the NR limit since both have ` = j. States
with the same j on the 0++ trajectory have ` = j ± 1 and thus unequal binding energies. The radial 0++ functions
H1 (7.78) and H2 (7.79) remain coupled in the NR limit,
00 2 0 h j(j + 1) i
H1,N R + H1,N R + m(Eb − V ) − H1,N R = 8m2 H2,N R (7.82)
r r2
00 2 0 h j(j + 1) i V
H2,N R + H2,N R + m(Eb − V ) − H2,N R = H1,N R (7.83)
r r2 4mr2
The lhs. of both equations scale as 1/r2 ∼ mV , implying the ratio
H2,N R V
∼ (7.84)
H1,N R m
65
0
In the expression for Φ++ the leading contribution ∝ H1 vanishes for h+ ' 1 + γ (7.65). This requires to
p (7.80)
retain the O V /m term in h+ ,
i
h+ ' 1 + γ 0 + α·∇ (7.85)
m
p p p
Then the contribution ∼ V /m H1 ∼ m/V H2 matches the leading H2 contribution mα · x H2 ∼ m/V H2 . The
2γ · L γ5 H2 term is subdominant, as are the O (V /m) corrections in h+ . This gives
i
ΦN R
(1 + γ 0 ) − α · ∇H1,N R (r) + 4m2 α · xH2,N R (r) Yjλ (Ω)

++ = (7.86)
2m
Orbital angular momentum is conserved in the NR limit, implying

2 N R
L , Φ++ = `(` + 1)ΦN ++
R
`=j±1 (7.87)
Using
→2
L , x = 2 − ∇ r2 + x r∂r + 3x

(7.88)
→2
L , ∇ = 2 x ∇2 − ∇ r∂r

(7.89)
gives
2 N R 1 0
i
L , Φ++ = i(1 + γ 0 ) α · ∇ 2rH1,N 2 2
R − j(j + 1)H1,N R − 8m r H2,N R Yjλ
2m
h 1 i
0
+ i(1 + γ 0 ) α · x 2m (Eb − V )H1,N R + 2rH2,N R + 2H 2,N R + j(j + 1)H2,N R Yjλ (7.90)
2m
Comparing with the Dirac structures in (7.86) and (7.87) gives two conditions,
2 0 1 2 0 1
8m2 H2,N R =

H1,N R + 2 `(` + 1) − j(j + 1) H1,N R = H1,N R + 2 ± (2j + 1) + 1 H1,N R (7.91)
r r r r
0
= −4m2 rH2,N 2

m(Eb − V )H1,N R R + ± (2j + 1) − 1 2m H2,N R for ` = j ± 1 (7.92)
Using the expression (7.91) for 8m2 H2,N R in the radial equation (7.82) gives the expected NR radial equation,
00
h `(` + 1) i
H1,N R + m(Eb − V ) − H1,N R = 0 (7.93)
r2
0
To check the self-consistency of (7.91) with (7.92) we may use (7.91) to express H2,N R and H2,N R in terms of
0 00
H1,N R , H1,N R and H1,N R and use this in (7.92). The result agrees with (7.93).
Using the expression (7.91) for H2,N R in the wave function (7.86) we have
i n h1
0 1 io
ΦN R
(1 + γ 0 ) α · ∇H1,N R (r) − α · x H1,N

++ = − R+ ± (2j + 1) + 1 H 1,N R Yjλ (7.94)
2m r 2r2
Separating ∇ into its radial and angular derivatives,
1 1
α · ∇ = (α · x) ∂r − i 2 α · x × L (7.95)
r r
the radial derivative of H1 cancels, so that the ` = j ± 1 NR wave functions are,
i n o
ΦN R 0 1

++ = (1 + γ ) 2 α · x ± (2j + 1) + 1 + iα · x × L H1,N R Yjλ (7.96)
2mr2
C. * q q̄ bound states in motion
A perturbative expansion for bound states should at each order respect Poincaré invariance. The spatial extent and
mutual interactions of the constituents make this non-trivial. The bound state energy needs to have the correct
66
p
dependence on the momentum, E(P ) = M 2 + P 2 , and scattering amplitudes (form factors) should transform
covariantly under rotations and boosts.
For Positronium this requires a frame dependent combination of Coulomb and transverse photon exchange, as discussed
in section V C. However, the O αs0 instantaneous potential arising from the homogeneous solution of Gauss’ law in
QCD (section VII A) must ensure Poincaré invariance on its own, without assistance from O (αs ) gluon exchange.
This is analogous to e+ e− bound states in QED2 , due to the absence of transverse photons in D = 1 + 1 dimensions.
In that case it is essential that the potential is linear (section VI A 4). The correct frame dependence of the energy
E(P ) turns out to be similarly ensured for q q̄ states, due to the linearity of the potential (7.8). I have not considered
qqq states, which have a different potential (7.14).
1. The bound state equation
In a general frame the P = 0 state (7.37) becomes, at t = 0,

Z
1 X
|M, P i = √ dx1 dx2 ψ̄ A (x1 )eiP ·(x1 +x2 )/2 δ AB Φ(P ) (x1 − x2 )ψ B (x2 ) |0i (7.97)
Nc A,B
which is an eigenstate of the momentum operator P (5.5) with eigenvalue P . In the following I take P = (0, 0, P )
(q)
along the z-axis. The derivatives in H0 (7.40) act after the partial integration also on exp[iP · (x1 + x2 )/2], giving
rise to a new term in the bound state equation (7.43),
i∇ · α, Φ(P ) (x) − 12 P · α, Φ(P ) (x) + m γ 0 , Φ(P ) (x) = E − V (x) Φ(P ) (x)

(7.98)
The potential V (x) = V 0 |x| is independent of the bound state momentum P , being determined by the instantaneous
positions x1,2 of the quarks. An alternative form of this BSE is
h → i h ← i
i∇ · α − 21 E − V + P α3 + mγ 0 Φ(P ) (x) + Φ(P ) (x) i∇ · α − 21 E − V − P α3 − mγ 0 = 0

(7.99)
It is possible to express the BSE equivalently as two coupled equations,

h 2 i 2i V0 h i
iα · ∇ + mγ 0 − 12 α · P − 1 Φ(P ) = − P · ∇Φ(P ) + iα · x, Φ(P )

E−V (E − V )2 r(E − V ) 2
h ← 2 i 2i V0 h i
Φ(P ) iα · ∇ − mγ 0 + 12 α · P (P ) (P )

−1 = P · ∇Φ − iα · x, Φ (7.100)
E−V (E − V )2 r(E − V )2
Exercise A 19: Derive the coupled equations (7.100) from the bound state equation (7.98).
Note: Can you find a simpler derivation than the one presented in A 19?
Since P breaks rotational symmetry in (7.98) (except for rotations around the z-axis) the radial and angular variables
do not separate as in (7.48). This makes a solution of the BSE more challenging. In QED2 Φ(P ) (x) can be expressed
(6.30) in terms of the rest frame wave function evaluated at a “boost invariant” variable τ (x) (6.26). A similar relation
works here as well, but only at x⊥ = (x, y) = 0. Φ(P ) (0, 0, z) then serves as a boundary condition on the BSE (7.98).
Before turning to the expression for Φ(P ) (0, 0, z) I consider the case of a vanishing potential. The exact solution can
be found for V = 0, at any P and x. This provides a boundary condition for the V 6= 0 BSE in the limit r → 0, in
which E − V 0 r → E on the rhs. of the BSE (7.98). Solving the partial differential equation with boundary conditions
at x⊥ = 0 and r → 0 should determine the wave function for all x (but this remains to be demonstrated).
2. Solution of the P 6= 0 bound state equation for V (x) = 0
The free solution of (7.98) is, for P = (0, 0, P ),

(P ) (0)
ΦV =0 (x) = exp(− 21 ξα3 ) ΦV =0 (xR ) exp( 12 ξα3 ) (7.101)
xR = (x, y, z cosh ξ) E = M cosh ξ P = M sinh ξ

67
(0) (P )
where ΦV =0 (x) is the solution of the BSE with V = 0 in the rest frame. Its relation to ΦV =0 (x) corresponds to
standard Lorentz contraction, with the j = 1/2 boost representations exp(± 12 ξα3 ) familiar from the Dirac equation.
(P )
I denote by B(x) the lhs. of the BSE (7.99) with V = 0 and Φ(P ) (x) = ΦV =0 (x) given by (7.101). Thus B = 0 is
required for (7.101) to be a solution. Multiplying by eξα3 /2 from the left and e−ξα3 /2 from the right,
→ (0)
eξα3 /2 B(x)e−ξα3 /2 = eξα3 /2 i∇ · α − 1
E + P α3 + mγ 0 e−ξα3 /2 ΦV =0 (xR )

2
(0) ←
+ ΦV =0 (xR )eξα3 /2 i∇ · α − 1
E − P α3 − mγ 0 e−ξα3 /2

2 (7.102)
Since zR = z cosh ξ we have ∂z = cosh ξ ∂zR , and

→ → →
iα3 ∂ z = eξα3 α3 i ∂ zR − sinh ξ i ∂ zR
← ← ←
iα3 ∂ z = i ∂ zR α3 e−ξα3 + i ∂ zR sinh ξ (7.103)
The terms ∝ sinh ξ give Dirac scalar contributions to (7.102), and cancel each other. Using E ± P α3 = M exp(±ξα3 ),
i∇⊥ · α⊥ exp(−ξα3 /2) = exp(ξα3 /2)i∇⊥ · α⊥ and similarly for the mγ 0 terms we get,
→ (0) (0) ←
eξα3 /2 B(x)e−ξα3 /2 = eξα3 i∇R · α − 21 M + mγ 0 ΦV =0 (xR ) + ΦV =0 (xR ) i∇R · α − 21 M − mγ 0 e−ξα3

(7.104)
(0)
Expressing exp(±ξα3 ) = cosh ξ ± α3 sinh ξ the coefficent of cosh ξ is the rest frame BSE at x = xR , which ΦV =0
satisfies by definition. The BSE allows to relate the coefficients of α3 sinh ξ, leaving the anticommutator with α3 ,
→ (0) →
(0)
eξα3 /2 B(x)e−ξα3 /2 = sinh ξ α3 , i∇R · α − 12 M + mγ 0 ΦV =0 (xR ) = 12 M sinh ξ α3 , h − ΦV =0

(7.105)
→ →
where h − is defined in (7.45) and evaluated at V = 0. The explicit expressions for h − Φ(x) in (7.61), (7.72) and
(P )
(7.81) are all ∝ V 0 and thus vanish for V = 0. Hence B(x) = 0 and ΦV =0 (x) of (7.101) solves the BSE for all P .
3. Boost of the state |M, P i for V (x) = 0
Instead of solving the BSE at a finite momentum P we may boost the rest frame state. This is feasible for V = 0
using the boost generator of free quarks. Suppressing the irrelevant color indices,
Z
K0 (t) = tP + dx ψ † (x) − xH0 + 21 iα ψ(x)

(7.106)
The expressions for the generators of translations P (5.5) in space and H0 (7.39) in time (the free Hamiltonian) are,
Z
P = dx ψ † (x)(−i∇)ψ(x)
Z Z
H0 = dx ψ † (x)H0 ψ(x) ≡ dx ψ † (x)(−iα · ∇ + mγ 0 )ψ(x) (7.107)
These operators satisfy the Lie algebra

R of the Poincaré group (I do not here consider rotations, and set t = 0). The
commutators of local operators O = dx ψ † (x)O(x)ψ(x) satisfy
Z
[Oi , Oj ] = dx ψ † (x) [Oi , Oj ] ψ(x) (7.108)
This allows to verify the Lie algebra in terms of the structures Oi (here P i = −i∂i ):
i j
P ,P = 0 since ∂i ∂j = ∂j ∂i
h i
P i , K0j = −i∂i , −xj (−iα · ∇ + mγ 0 ) + 21 iαj = iδ ij H0

H, K0i = −iα · ∇ + mγ 0 , −xi H0 + 12 iαi = iαi H0 + 21 i [H0 , αi ] = 12 i {αi , H0 } = iP i

(7.109)
68
An infinitesimal boost in the z-direction of the (non-interacting) state |M, P i with P = (0, 0, P ) is generated by the
operator 1 − idξK0z , as verified by the eigenvalues,
P z (1 − idξK0z ) |M, P i = (1 − idξK0z )P z |M, P i − idξ [P z , K0z ] |M, P i = (P + dξE)(1 − idξK0z ) |M, P i
H0 (1 − idξK0z ) |M, P i = (1 − idξK0z )H0 |M, P i − idξ [H0 , K0z ] |M, P i = (E + dξP )(1 − idξK0z ) |M, P i (7.110)
The expression (7.97) for |M, P i in terms of the wave function Φ(P ) allows to determine the wave function for
(1 − idξK0z ) |M, P i, and thus to deduce its frame dependence (7.101).
(P )
Exercise A 20: Derive the frame dependence (7.101) of ΦV =0 (x) using the boost generator K0z .
Hint: Use the bound state equation in the form of (7.100) (with V = 0).
The boost demonstrates that the relative normalizations of wave functions with different momenta P is correctly
given by (7.101). This applies also to the interacting case (V 6= 0) considered next, since the P -dependence of the
component Φ(P ) (x = 0) is given by V = 0.
4. Solution of the P 6= 0 bound state equation at x⊥ = 0
√
Apart from x⊥ = 0 the following requires E = M 2 + P 2 and a linear potential V = V 0 z with z > 0. The wave
function for z < 0 may be determined using parity or charge conjugation (7.54). As in the D = 1 + 1 case (section
VI A 4) the coordinate z is transformed into the variable τ (z)
τ (z) ≡ (E − V )2 − P 2 /V 0 = (M 2 − 2EV + V 2 )/V 0

(x⊥ = 0) (7.111)
Since τ (z) depends on E the transformation z → τ is different for the rest frame wave function Φ(0) (0, 0, z) compared
to that for Φ(P ) (0, 0, z). These two wave functions will be related at the same value of τ , and therefore at different
values of z. For V E (weak binding) τ (z) ' M 2 /V 0 − 2M z cosh ξ and the transformation is equivalent to z → zR
as in (7.101) (standard Lorentz contraction). I shall somewhat sloppily denote the wave functions expressed in terms
of τ using the same symbols, Φ(0) (τ ) and Φ(P ) (τ ). It should be kept in mind that these are related to the original
wave functions at x = (0, 0, z) through the P -dependent transformation (7.111).
The variable ζ(z) takes the place of the boost parameter ξ,
r
E−V P2 P
cosh ζ = √ = 1+ 0 sinh ζ = √ (7.112)
V 0τ V τ V 0τ
ζ(z) depends on P as well as τ . The definition (7.111) shows that V 0 τ ≥ −P 2 for real values of z. Hence when P 6= 0
there is a range of z for which V 0 τ < 0. To avoid considering complex values of ζ I shall assume values of z and P
such that τ > 0. I discuss below how to determine the x⊥ = 0 wave function in the range where τ < 0.
The solution of the BSE (7.99) at x⊥ = 0 is related to the rest frame wave function through
Φ(P ) (τ ) = exp(− 21 ζα3 ) Φ(0) (τ ) exp( 12 ζα3 ) (7.113)
The same relation holds also for ∇⊥ Φ(P ) (τ ). This requires ∇⊥ ζ = 0, which follows from ∇⊥ V (x) = 0 at x⊥ = 0.
By construction, Φ(0) (τ ) depends only on τ , whereas Φ(P ) (τ ) has an explicit P -dependence through ζ (7.112).
Exercise A 21: Show that Φ(P ) (τ ) given by (7.113) satisfies the BSE (7.99) at x⊥ = 0.
Hint: Follow the proof of section VII C 2 for the V = 0 case. Pay attention to derivatives of ζ.
As seen in section VII B the wave functions of all rest frame q q̄ states are found by solving radial equations, which
are ordinary differential equations in r. The relation (7.113) then determines Φ(P ) (0, 0, z) and ∇⊥ Φ(P ) (0, 0, z) in all
frames, when the q q̄ pairs are aligned with P . This boundary condition on the BSE (together with the one for r → 0
based on (7.101)) should allow to determine Φ(P ) (x) at all x by solving the partial differential equation (7.98) in
(x⊥ , z). This remains to be demonstrated.
For a rest frame wave function τ = (M − V 0 z)2 /V 0 ≥ 0 (7.111), whereas in general τ ≥ −P 2 . This leaves a gap
−P 2 ≤ τ < 0 in the boundary condition (7.113). In the D = 1 + 1 case the analytic functions (6.34) determine the
69
solution for all τ . Here we may use the analog of the expression (6.32),
∂Φ(P ) (x)

zP
∂z Φ (x) − 21 α3 , Φ(P ) (x)
(P )

(E − V ) = at x = (0, 0, z) (7.114)
∂P
z E P
which is a consequence of (7.113). On the lhs. the |z indicates that the P -derivative is to be taken at fixed z, while
on the rhs. the z-derivative is at fixed P . The derivation is the same as in the D = 1 + 1 case, see Exercise A 13. This
equation determines Φ(P ) (0, 0, z) for all P and z, with the rest frame wave function Φ(0) (0, 0, z) serving as boundary
condition at P = 0. In particular, the solution covers the gap −P 2 ≤ τ < 0.
D. Properties of the q q̄ bound states
The q q̄ wave functions have novel features at large values of the linear potential (7.8). There is little ab initio knowledge
of strongly bound states since they are usually associated
with large values of the coupling, i.e., non-perturbative
dynamics. Here the confining potential is of O αs0 , so relativistic binding is compatible with a perturbative expansion
in αs . It is essential that the lowest order bound states, determined by the relativistic solutions Φ(P ) (x) of the BSE
(7.98), provide a reasonable approximation of the true states.
1. String breaking and duality
A first issue is why we should trust the linear potential V (r) = V 0 r at large r. “String breaking” will prevent the
potential from reaching large values. I touched upon this already in section VI A 7, for QED in D = 1 + 1 dimensions.
The key appears to be quark-hadron duality, which is a pervasive feature of hadron data and poorly understood
theoretically, see the review [68] and conference9 .
Bound state solutions of the BSE with different eigenvalues EA 6= EB are orthogonal,
hMB , P B |MA , P A i ∝ δ(P A − P B )δA,B (7.115)
Exercise A 22: Prove (7.115) for states with wave functions satisfying the BSE (7.98).
Hint: The proof is analogous to the standard one for non-relativistic systems.
However, the q q̄ states are not orthogonal to q q̄ q q̄ states. Contracting the quark fields as in Fig. 10(a) gives
Z h Y
1 i
dx1k dx2k exp i 12 (x1A + x2A ) · P A − (x1B + x2B ) · P B − (x1C + x2C ) · P C

hB, C|Ai = √
NC k=A,B,C
× h0| ψ (x2B )Φ†B γ 0 ψ(x1B ) ψ † (x2C )Φ†C γ 0 ψ(x1C ) ψ̄(x1A )ΦA ψ(x2A ) |0i
†
(7.116)
(2π)3 3
Z
δ (P A − P B − P C ) dδ 1 dδ 2 eiδ1 ·P C /2−iδ2 ·P B /2 Tr γ 0 Φ†B (δ 1 )ΦA (δ 1 + δ 2 )Φ†C (δ 2 )

= −√
NC
where δ 1 = x1B − x2B and δ 2 = x1C − x2C . Note that the process A → B + C is not mediated by a Hamiltonian
interaction. It expresses that state A has an overlap with B + C 10 . For example, Parapositronium has no overlap
with |γγi, but decays into this state do occur through the action of Hint . A bound state can overlap two bound states
only for relativistic binding, so this phenomenon has no precedent for atoms.
Quark-hadron duality in e+ e− → hadrons means that the final state is described, in an average sense, by the q q̄ state
first created by the virtual photon. This holds also for individual resonances in the direct channel, and is consistent
with an overlap between the q q̄ and final hadron states. I show below (VII D 4) how the wave functions of highly
excited bound states indeed reduce to those of free q q̄ states.
9 “First Workshop on Quark-Hadron Duality and the Transition to pQCD”, http://www.lnf.infn.it/conference/duality05/ .

10 I thank Yiannis Makris for a helpful comment concerning this.
70
Another example of duality is the observation that the inclusive momentum distribution of hadrons produced in hard
processes agrees with the perturbatively calculated gluon distribution [14]. This “Local Parton Hadron Duality”
works down to low momenta. It shows that the transition from parton to hadron occurs with a minimal change of
momentum, in the spirit of an overlap between states.
I shall assume that highly excited bound states give a gross description of the final multi-hadron states, similarly as
high energy quarks and gluons describe hadron jets. This can and should be quantified by evaluating matrix elements
like (7.116).
FIG. 10. (a) The overlap hB, C|Ai arises through the tunneling of a q q̄ pair in the instantaneous field, “string breaking”. Since
the potential is linear the field energy is nearly the same before and after the split. (b) Hadron loop correction through the
creation and annihilation of a q q̄ pair, which may be important for unitarity.
2. Properties of the wave function at large separations r
Non-relativistic (Schrödinger) wave functions

R describe the distribution of a fixed number of bound state constituents.
The single particle probability constraint dx |Φ|2 = 1 (global norm) determines the energy eigenvalues. The Dirac
equation also describes a single electron, but in a strong potential (V & m) the wave function has negative energy
components. In time-ordered dynamics these E < 0 components correspond to e+ e− pairs in the wave function. This
is related to the Klein paradox [38] and illustrated by the perturbative Z-diagram of Fig. 4(b). For a linear potential
the local norm of the Dirac wave function approaches a constant at large r (3.57) [37], where it is dominated by the
E < 0 components (3.58). This may be interpreted as positrons, which due to their positive charge are repelled by
the potential, and accelerated to high momenta at large r.
In section II B we saw that the crossed ladder diagram of Fig. 2(c) does not contribute to atomic bound states at
lowest order in α, i.e., to non-relativistic dynamics. When time ordered this Feynman diagram is the same as the
pair-producing Z-diagram of Fig. 4(b) (after the addition of the antifermion line). In QCD the uncrossed and crossed
diagrams are distinguished also by their dependence on the number Nc of colors. This is illustrated in Fig. 11 for
q q̄ → q q̄ with single and double gluon exchange. The initial
√ and final states are taken to be SU(Nc ) singlets, implying
a sum over the colors A and B, each normalized by 1/ Nc .
FIG. 11. Color structure of QCD Feynman diagrams. (a) Single gluon exchange. Planar (b) and non-planar (c) two-gluon
exchange. The initial and final states are color singlets, implying a sum over the quark colors A and B.
71
The color factors are thus, including also their behavior for Nc → ∞:
g2 XX
a a g2 a a
2
2 Nc − 1
C(a) = √ T T = Tr (T T ) = g ∼ 1
g 2 Nc
( Nc )2 A,B a BA AB Nc 2Nc 2
g4 N 2 − 1 2
c
C(b) = Tr (T b T a T a T b ) = g 4 ∼ 1
4 g 4 Nc2
Nc 2Nc
g4 g4 2
4 Nc − 1
C(c) = Tr (T b T a T b T a ) = − Tr (T a a
T ) = −g ∼ − 14 g 4 (7.117)
Nc 2Nc2 4Nc2
The suppression of C(c) compared to C(b) for Nc → ∞ is a general feature of non-planar diagrams [69–71]. Due to
the relation of diagram (c) in Fig. 11 with the Z-diagram of Fig. 4(b) there are no such pair contributions to q q̄ states
in the Nc → ∞ limit, despite the relativistic binding. In particular, there are no sea quarks in the ’t Hooft model of
QCD2 [72], where Nc → ∞. The connection between the sea quark distribution at low xBj and the behavior of the
wave function at large quark separations (the virtual pairs) is apparent for QED2 (section VI C 3).
The normalizing integral of the 0−+ trajectory wave functions (7.60) can for all angular momenta j be expressed as,
Z ∞
2V 0
Z h i h i
†
dx Tr Φ−+ (x)Φ−+ (x) = 8 dr r2 F1∗ (r) 1 − ∂r F1 (r) (7.118)
0 (M − V )3
Exercise A 23: Verify the expression (7.118) for the global norm of Φ−+ (x) in terms of F1 (r).
For r → ∞ (at fixed j) the asymptotic radial wave function F1 (r),

1
F1 (r) ' exp i(M − V )2 /4V 0 and c.c. (r → ∞)

(7.119)
r

satisfies the radial wave equation (7.59) up to terms of O r0 . The integrand (local norm) in (7.118) tends to
r2 |F1 (r)|2 for large r, and is thus independent of r. This feature is common to states of all quantum numbers. The
probability density similarly tends to a constant also in lower spatial dimensions (D = 1 + 1 (6.41) and D = 2 + 1),
as well as in the Dirac equation for a linear potential (3.57).
At large r the radial derivative dominates in the expression (7.60) for Φ−+ (x): α · ∇ ' α · x̂ ∂r where x̂ = x/r. Using
also F10 (r → ∞) ' 21 iV 0 r F1 (r → ∞),
h 2 i
Φ−+ (r → ∞) ' − 0 (iα · x̂ ∂r + mγ 0 ) + 1 γ5 F1 (r)Yjλ (Ω) = (1 + α · x̂)γ5 F1 (r)Yjλ (x̂) (7.120)
V r
The projector Λ+ which selects the b† operator in ψ̄(x) is as in (5.3),
← ← ←
Z
dk 1 h i
ū(k, λ) e−ik·x b†k,λ
X
ψ̄(x) Λ + (x) = 3
Λ + (x) ≡ E − iα · ∇ + γ 0 m (7.121)
(2π) 2Ek 2E
λ
A partial integration in the b†k,λ component of the state (7.37) makes Λ+ operate on the wave function. Noting that
p
∇2 Φ−+ (r → ∞) ' ∂r2 Φ−+ ' − 14 (V 0 r)2 Φ−+ E Φ−+ ≡ −∇2 + m2 Φ−+ ' 12 V 0 r Φ−+ (7.122)
→
we see that Λ + annihilates the asymptotic wave function,
→ 1 h → i 1h iα · x̂ ∂r i
Λ + Φ−+ (r → ∞) = E + iα · ∇ + γ 0 m Φ−+ ' 1 + Φ−+
2E 2 E
' 12 (1 − α · x̂)(1 + α · x̂)γ5 F1 (r → ∞)Yjλ (x̂) = 0 (7.123)
Consequently the state has no b† contribution at large r. Similarly d† does not contribute, leaving only the negative
energy component d b. This is characteristic of the pair (sea quark) contributions from Z-diagrams. The negative
kinetic energy is cancelled by the positive potential energy at large r, leaving a finite eigenvalue M .
My tentative interpretation of this is as follows. The delicate cancellation between large negative kinetic and posi-
tive potential energies means that bound state components with V (r) M will not affect physical processes with
resolution below those energies. Hadron loop corrections like in Fig. 10 b may change the wave function consider-
ably. Nevertheless, since such corrections are due to overlaps the original wave function still approximately describes
physical processes. Deep inelastic processes reveal the xBj → 0 sea quark distribution as far as its resolution permits.
72
3. Discrete mass spectrum
Due to the infinite sea one cannot impose a global normalization condition on the bound state wave functions. There
is, however, a new local normalization condition. The solutions of the q q̄ bound state equation (7.42) are generally
singular at M − V (r) = 0, as indicated by the coefficients ∝ 1/(M − V ) in the radial equations (7.59), (7.69) and
(7.78)-(7.79). A physical wave function should be locally normalizable at M − V = 0, in line with the standard
requirement of local normalizability at r = 0.
The radial equation (7.59) of the 0−+ trajectory allows F1 (r) ∼ (M − V )γ with γ = 0 and γ = 2 as M − V (r) → 0.
The integrand (local norm) in (7.118) is finite at M −V = 0 only if γ = 2. For r → 0 we have as usual F1 (r) ∼ rβ , with
β = j or β = −j − 1. Only β = j makes the integrand in (7.118) finite at r = 0. The two constraints, at M − V (r) = 0
and r = 0, determine the bound state masses, but leave the magnitude of the wave function unconstrained. This is a
general feature, valid for states of all quantum numbers.
The vanishing of the radial wave functions at M − V (r) = 0 generalizes the vanishing for r → ∞ of non-relativistic
wave functions, which are defined only for V (r) M . In the limit of non-relativistic dynamics (m → ∞) the wave
functions which are regular at M − V = 0 become globally normalizable [48].
The Dirac equation may be viewed as the limit of a two-particle equation where the mass m2 of one particle tends to
infinity, turning it into a static source. The point V (r) = M (where M includes m2 ) recedes to r = ∞ as m2 → ∞.
Hence there is no condition on the Dirac wave function at M − V = 0, and the spectrum is continuous [37].
The radial equation (7.59) can readily be solved numerically, subject to the boundary conditions F1 (r → 0) ∼ rj and
F1 (r → M/V 0 ) ∼ (M − V )2 . As seen in Fig. 12, for the linear potential V (r) = V 0 r and quark mass m = 0 the
states lie on linear Regge trajectories and their parallel daughter trajectories. The mass spectra of the 0−− and 0++
trajectories are similar [15].
√
FIG. 12. (a) Masses M of the mesons on the 0−+ trajectory for m = 0, in units of V 0 . (b) Plot of the spin j vs. M 2 /V 0 for
the states listed in (a). Figure taken from [15].
4. Parton picture for M V (r)
Duality in e+ e− → hadrons implies that the perturbative process e+ e− → q q̄ gives an average of the direct channel
resonances [68]. More generally, PQCD describes inclusive cross sections and hadron distributions (jets) at high
energies in terms of unconfined quarks and gluons. Confinement (hadronization) is expected to become important as
the color charges separate beyond the hadronic scale.
Duality implies that the wave functions of highly excited bound states (M 2m) should agree with the parton
model, i.e., be compatible with the perturbative quark and gluon distribution for V = V 0 r M . This is already
indicated by the form of the bound state equation (7.42), where M and V appear only in the combination M − V .
When V M this is a free particle equation, with no negative energy (d, b) contributions from Z-diagrams. It is
instructive to see how this limit emerges from the wave functions.
I shall again use the 0−+ trajectory states to illustrate the emergence of the parton picture. The asymptotic expression
73

(7.119) for F1 (r) at large r satisfies the radial equation also when M → ∞ at finite r, up to terms of O M 0 .11 In
the expression (7.60) for the wave function the radial derivative then dominates, F10 ' −i 21 M F1 , so
h 2 i
Φ−+ (M → ∞) ' (iα · x̂ ∂r + mγ 0 ) + 1 γ5 F1 (r)Yjλ (Ω) ' (1 + α · x̂)γ5 F1 Yjλ (7.124)
M
Consider again the projector Λ+ (7.121) which projects out the b† term in the ψ̄(x1 ) field of the state (7.37). After a
partial integration it operates on Φ−+ (x), with the quark energy now giving
p
E Φ−+ (M → ∞) ≡ −∇2 + m2 Φ−+ ' 12 M Φ−+ (7.125)
→
The dominance of the radial derivative in Λ + implies,
→ 1 iα · x̂ ∂r + γ 0 m
Λ + Φ−+ (M → ∞) ' 1+ Φ−+ ' 12 (1 + α · x̂)(1 + α · x̂)γ5 F1 (r)Yjλ = Φ−+ (7.126)
2 E
The result is opposite to that of (7.123): Now only the valence quark operators b† , d† contribute to the state,
Z Z
dk1 dk2
e−i(k1 ·x1 +k2 ·x2 ) F1 Yjλ ū(k1 , λ1 )γ5 v(k2 , λ2 ) b†k1 ,λ1 d†k2 ,λ2 |0i
X
|M iV M ' dx1 dx2 6
(7.127)
(2π) 4E1 E2
λ1 ,λ2
Since F1 Yjλ depends only on x1 − x2 , and
k1 · x1 + k2 · x2 = 12 (k1 + k2 )(x1 + x2 ) + 21 (k1 − k2 )(x1 − x2 ) (7.128)
the integral over (x1 + x2 )/2 sets k1 = −k2 ≡ k. Denoting x ≡ x1 − x2 ,

Z Z
dk
e−ik·x ū(k, λ1 )γ5 v(−k, λ2 ) F1 (r)Yjλ (x̂) b†k,λ1 d†−k,λ2 |0i
X
|M iV M ' dx 3 2 (7.129)
(2π) 4Ek
λ1 ,λ2
For j = 0 the angular integral over x̂ gives, with x · k = kr cos θ,

Z 1 Z 2π √
−ikr cos θ 1 i π ikr
e − e−ikr

dcosθ dϕ e √ =− (7.130)
−1 0 4π kr
At large M V the phase φ(r) of F1 (r) (7.119) changes quickly with r, allowing to estimate the r-integral using the
method of stationary phase,
φ(r) = ±kr + (M − V )2 /4V 0 ∂r φ(rs ) = 0 V 0 rs = M − 2k φ(rs ) = k(M − k)/V 0 (7.131)
The fact that only the exp(+ikr) term in (7.130) contributes sets cos θ = −1, i.e., x and k are antiparallel (with my
choice of asymptotic solution in (7.119)). The stationary phase method may again be used in the k-integration, since
φ(rs ) depends sensitively on k,
∂k φ(rs ) = 0 ks = 21 M (7.132)
Thus the quark and antiquark momenta are each half the resonance mass, as expected by energy conservation. With
k = M/2 in (7.131) we have V 0 rs M , i.e., the stationary value rs is compatible with the limit assumed in (7.124).
A similar result will be obtained for any (fixed) angular momentum j, since each term in the angular integral
corresponding to (7.130) will be ∝ exp(±kr), multiplied by powers of kr which do not affect the stationary phase
approximation. Thus the wave functions of highly excited states on the 0−+ trajectory (7.55) take the form
Z
ū(k, λ1 )γ5 v(−k, λ2 ) Yjλ (k̂) b†k,λ1 d†−k,λ2 |0i
X
|k| = 21 M

|M iV M ∝ dk̂ (7.133)
λ1 ,λ2
The bound state wave function thus agrees with that of free q q̄ states perturbatively produced by a current with corre-
sponding J P C quantum numbers, as expected by duality. This may be used to determine the absolute normalization
of highly excited bound states, as discussed in section IV of [48].
11 In D = 1 + 1 the x → ∞ and M → ∞ limits are similarly related. The QED2 wave functions φ0 and φ1 (6.34) depend on x and M
only through the variable τ (6.26), and τ → ∞ in both limits.
74
E. * Glueballs in the rest frame
(0)
I consider states of two transversely polarized gluons |ggi, bound by the instantaneous linear potential Vgg (7.20),
r
N 2 3
Vgg (r) = Λ r = Λ2 r ≡ Vg0 r (7.134)
CF 2
(1)
The O (αs ) instantaneous gluon exchange Vgg in (7.20) as well as higher Fock components (|gggi, |ggq q̄i . . .) are
ignored. Hence the Hamiltonian (4.21) is approximated as H = H0 + HV , where HV (4.37) generates the linear
potential and the kinetic term in (4.21),
Z
+ 12 Aia,T (−∇2 )Aia,T
i i
H0 = dx 12 Ea,T

Ea,T (7.135)
involves only transversely polarized gluons Aia,T , which satisfy ∇ · Aa,T = 0, and their conjugate electric fields −Ea,T
i
.
The canonical commutation relations (4.20) imply
H0 , Aia,T (x) = iEa,T

i i
(x) = i∇2 Aia,T (x)

(x) H0 , Ea,T (7.136)
Consequently the bound state condition
(H0 + HV ) |ggi = M |ggi (7.137)
requires |ggi to have both A and E components. In terms of the wave functions Φij (x1 − x2 ),
Z
|ggi ≡ dx1 dx2 Aia,T (x1 )Aja,T (x2 )Φij i j ij i j ij i j ij

AA (x1 − x2 ) + Aa,T Ea,T ΦAE + Ea,T Aa,T ΦEA + Ea,T Ea,T ΦEE |0i (7.138)
where sums over the color a and 3-vector indices i, j are understood (here A is not a color index!). The constituent
A and E fields are assumed to be normal ordered (their mutual commutators are subtracted).
As shown in section VII A 3 the action of HV on Aia,T (x1 )Aja,T (x2 ) |0i gives the potential (7.134). Since Ea (y) (4.26)
has similar commutators with the A and E fields,
Ea (y), Aid (x) = −i fabd Aib (x)δ(x − y)

Ea (y), Edi (x) = −i fabd Ebi (x)δ(x − y)

(7.139)
the same potential (7.134) is obtained for all four components of |ggi in (7.138),
Z

HV |ggi = dx1 dx2 Vgg (|x1 − x2 |) Aa (x1 )Aa (x2 )ΦAA (x1 − x2 ) + Aa Ea ΦAE + Ea Aa ΦEA + Ea Ea ΦEE |0i
(7.140)
where I suppressed the 3-vector indices i, j and the label T of the transverse fields, which are unaffected by H0 and
HV . Using the commutation relations (7.136),
Z n
H0 |ggi = i dx1 dx2 Ea (x1 )Aa (x2 ) + Aa (x1 )Ea (x2 ) ΦAA (x1 − x2 ) + Ea Ea + Aa Aa ∇2 ΦAE

o
+ Aa Aa ∇2 + Ea Ea ΦEA + Aa Ea + Ea Aa ∇2 ΦEE |0i

(7.141)
where ∇ differentiates Φ(x1 − x2 ) wrt. x1 − x2 .

The stationarity condition (7.137) implies the following relations between the wave functions Φ(x):
∇2 (ΦAE + ΦEA ) = −i(M − V )ΦAA
ΦAA + ∇2 ΦEE = −i(M − V )ΦAE
ΦAA + ∇2 ΦEE = −i(M − V )ΦEA
ΦAE + ΦEA = −i(M − V )ΦEE (7.142)

75
where V = Vg0 |x| = Vg0 r as in (7.134). This implies
ΦAE = ΦEA = − 12 i(M − V )ΦEE

1
∇2 (M − V )ΦEE

ΦAA =
M −V
1
∇2 (M − V )ΦEE + ∇2 ΦEE = − 21 (M − V )2 ΦEE

(7.143)
M −V
The last equation is
Vg0 Vg0
∇2 ΦEE (x) − ∂r ΦEE (x) − ΦEE (x) + 14 (M − V )2 ΦEE (x) = 0 (7.144)
M −V r(M − V )
Separating the radial and angular dependence according to
ΦEE (x) = F (r)Y`λ (Ω) (7.145)
where Y`λ is the standard spherical harmonic function, the radial equation becomes
2 Vg0 0 h Vg0 `(` + 1) i
F 00 (r) + − F (r) + 14 (M − V )2 − − F (r) = 0 (7.146)
r M −V r(M − V ) r2
There is a single dimensionful parameter Vg0 . Scaling r = R/ Vg0 and M = M Vg0 the bound state equation in terms
p p
of the dimensionless variables R, M becomes
2 1 h 1 `(` + 1) i
2
∂R F (R) + − ∂R F (R) + 14 (M − R)2 − − F (R) = 0 (7.147)
R M−R R(M − R) R2
FIG. 13. (a) Glueball spectrum: Orbital angular momentum ` versus M 2 /Vg0 . (b) Glueball masses M = M/ Vg0 of the radial
p
equation (7.147).
76
For R → 0 we have the standard behaviors F ∼ Rα , with α = ` or α = −` − 1. Since ΦAA ∼ ∂R 2

ΦEE only the α = `
β
solution gives a locally finite norm at R = 0. For M − R → 0 with F ∼ (M − R) we have β = 0, and a second
solution F ∼ log(M − R). Only the β = 0 solution gives a locally finite norm at M − R = 0. These constraints on the
solutions of (7.147) at R = 0 and R = M determine the allowed
masses M. The glueball states lie on approximately
linear Regge and daughter trajectories (Fig. 13). At O αs0 the spectrum is independent of the vector indices i, j of
the wave function in (7.138).
F. Spontaneous breaking of chiral symmetry
Let us now return to the bound state equation (7.43) of a meson (q q̄) state in the rest frame,
i∇ · {α, Φ(x)} + m γ 0 , Φ(x) = M − V (x) Φ(x)

(7.148)
where the O αs0 potential is linear V = V 0 |x| (7.8). In section VII D 3 we saw that the bound state mass spectrum

is determined by the requirement that |Φ(x)|2 is integrable at r = |x| = 0 and at M − V 0 r = 0. The latter condition
is the generalization of the standard condition that the solutions of the Schrödinger equation are normalizable.
So far I did not discuss the special case of M = 0, in which case the constraints at r = 0 and M − V 0 r = 0 coincide
[48, 73]. Such solutions have vanishing four-momentum in all frames. This allows the J P C = 0++ state to mix
with the ground state (vacuum) without violating Poincaré invariance. The chiral symmetry of the QCD action for
massless quarks is then not manifest in the states, as in a spontaneous breakdown of chiral invariance.
In this preliminary study I consider two degenerate flavors, mu = md = m. Suppressing color and Dirac indices the
states (7.37) are
!
Z u
|M, ii = dx1 dx2 q̄(x1 )Φ(x1 − x2 )τ i q(x2 ) |0i q= (7.149)
d
where the τ i are Pauli matrices for isospin I = 1 states and τ i → 1 for I = 0.
1. M = 0 states with vanishing quark mass m = 0
The J P C = 0++ , I = 0 “σ” wave function may be expressed in the general form (7.48), with three Dirac structures12
allowed by (7.55) for j = 0,
Φσ (x) = H1 (r) + i α · x̂ H2 (r) + i γ 0 α · x̂ H3 (r) (7.150)
where x̂ = x/r. For M = m = 0 the BSE (7.148) is i∇ · {α, Φσ } + V 0 r Φσ = 0, and requires13

2
iα · x̂H10 − H2 − H20 + 21 V 0 r Φσ = 0 (7.151)
r
The coefficients of the three Dirac structures impose H2 = −(2/V 0 r)H10 , H3 = 0 and
1
H100 (r) + H10 (r) + 41 (V 0 r)2 H1 (r) = 0 (7.152)
r
This differential equation has an analytic solution, giving the wave function
Φσ (x) = N J0 ( 14 V 0 r2 ) + i α · x̂ J1 ( 41 V 0 r2 )

(m = M = 0) (7.153)
where N is a normalization constant and J0 , J1 are Bessel functions. The σ state (7.149),
Z
|σi = σ̂ |0i σ̂ = dx1 dx2 q̄(x1 )Φσ (x1 − x2 )q(x2 ) (7.154)
12 The present radial functions Hi (i = 1, 2, 3) are distinct from the functions H1 , H2 in (7.78) and (7.79).
13 A more detailed derivation is given in Exercise A 24 for m 6= 0.
77
has M = P = 0, i.e., vanishing 4-momentum in all frames. When mixed with the vacuum it causes a spontaneous
breaking of chiral symmetry. The I = 1 (non-anomalous) chiral transformations are generated by
Z
Q5i = dxq † (x)γ5 12 τ i q(x) (7.155)
which transform the σ state into a Goldstone boson (pion)

Z
i [Q5i , σ̂] = π̂i π̂i = dx1 dx2 q̄(x1 )Φπ (x1 − x2 )τ i q(x2 ) (7.156)
i [Q5i , π̂j ] = −δij σ̂ Φπ (x) = −iΦσ (x)γ5
where I (somewhat
arbitrarily) fixed the relative normalization of Φσ and Φπ . Like σ̂ |0i, the state π̂i |0i is an eigenstate
of the O αs0 Hamiltonian with vanishing 4-momentum in all frames. Chiral transformations thus transform a vacuum
state with a σ condensate into another one with a mixture of zero-mass pions.
It is common to describe the pion in terms of a local field ϕπ,i , which is a good approximation at low momentum
transfers. Then, recalling that |πj i = π̂j |0i is time independent and normalized as in (7.153) and (7.156),
h0|ϕπ,i (x) |πj i = δij

i
ϕπ,i (x) = − q̄(x)γ5 21 τ i q(x) (7.157)
4N
Spontaneous chiral symmetry breaking implies that the Goldstone bosons |πj i are annihilated by the axial vector
µ µ
current j5i (x) = q̄(x)γ µ γ5 12 τ i q(x) and by its divergence ∂µ j5i (x) = 2im q̄(x)γ5 12 τ i q(x). With fπ ' 93 MeV,
h0|q̄(x)γ µ γ5 12 τ i q(x) |πj , P i = i δij P µ fπ e−iP ·x (7.158)
Mπ2
h0|q̄(x)γ5 12 τ i q(x) |πj , P i = −i δij fπ e−iP ·x (7.159)
2m
Since the Goldstone pion has P µ = 0 in all reference frames the rhs. of (7.158) vanishes. The lhs. also vanishes:
−iN Tr (γ µ γ5 21 τ i γ 0 J0 (0)γ5 τ j γ 0 ) = iN δ ij Tr (γ µ ) = 0. The rhs. of (7.159) is finite in the m → 0 limit, since Mπ2 ∝ m
[74]. We have then
M2
Z
h0|q̄(x)γ5 12 τ i q(x) dx1 dx2 q̄(x1 )Φπ (x1 − x2 )τ j q(x2 ) |0i = −iN Tr (γ5 21 τ i γ 0 γ5 τ j γ 0 ) = 4iN δ ij = −i π fπ δ ij
2m
This determines the normalization of the pion wave function,
Mπ2
fπ J0 ( 14 V 0 r2 ) + i α · x̂ J1 ( 14 V 0 r2 ) γ5

Φπ (x) = i (7.160)
8m
2. Finite quark mass mu = md = m 6= 0
The pion becomes massive (Mπ > 0) through the explicit breaking of chiral symmetry by a non-vanishing quark
mass m 6= 0. The 0++ σ must remain massless (Mσ = 0) to ensure that its mixing with the vacuum does not break
Poincaré invariance. The wave function Φσ (x) which solves the BSE (7.148) with M = 0 but m 6= 0 has the structure
of (7.150) with
0 2
nh 2m2 i 2m2 o
H1 (r) = N e−iV r /4 1 0 2 1 0 2
= N J0 ( 41 r2 ) + O m2

1− L 2 1 2 iV r + L 2 1 2 iV r
(V 0 r)2 im
2V 0 − 2 (V 0 r)2 im
2V 0 + 2
0 2
nh 2 i 2 o
H2 (r) = N e−iV r /4 1 0 2 1 0 2
= N J1 ( 41 r2 ) + O m2

− i L im2 1 2 iV r − L 2 1 2 iV r
V 0 r2 2V 0 − 2 V 0 r2 im
2V 0 + 2
2m
H3 (r) = − H2 (r) (7.161)
V 0r
where the Ln (x) are Laguerre functions.
78
Exercise A 24: Verify the expressions (7.161) for radial functions H1 (r), H2 (r) and H3 (r).
The pion state in the rest frame is

Z
|π, ii = π̂i |0i = dx1 dx2 q̄(x1 )Φπ (x1 − x2 )τ i q(x2 ) |0i (7.162)
The pion wave function has the form given in (7.57), where F3 = 0 for j = 0. With the notational change F2 → F2 /r
this means
h i
Φπ (x) = F1 (r) + i α · x̂ F2 (r) + γ 0 F4 (r) γ5 (7.163)
Using this in the bound state equation (7.148) and collecting terms with the same Dirac structure we get the conditions:
2
γ5 : − F2 − F20 + mF4 = 12 (Mπ − V )F1
r
i α · x̂ γ5 : F10 = 21 (Mπ − V )F2
γ 0 γ5 : mF1 = 12 (Mπ − V )F4 (7.164)
Eliminating F2 and F4 gives the radial equation
2 V 0 0 1
F100 + F1 + 4 (Mπ − V )2 − m2 F1 = 0

+ (7.165)
r Mπ − V

For the regular solution F1 (r → 0) ∼ r0 [1 + O r2 ], F2 (r → 0) ∼ r and F4 (0) = F1 (0) 2m/Mπ . Thus
2m 0
Φ(0)
π (x = 0) = F1 (0) 1 + γ γ5 (7.166)
Mπ
The superscript on Φπ reminds that this is the rest frame wave function. In a frame with P 6= 0 the pion state takes
the form (7.97) at t = 0. For a general time, suppressing color and Dirac indices,
Z
−iP 0 t
|Mπ , i, P , ti = e dx1 dx2 q̄(t, x1 )eiP ·(x1 +x2 )/2 Φ(P ) i
π (x1 − x2 )τ q(t, x2 ) |0i (7.167)
q
where P 0 = P 2 + Mπ2 . The axial current identities (7.158) and (7.159) probe the pion state at t = x0 and
(P )
x1 = x2 = x, involving Φπ (0). Since V 0 r = 0 at r = 0 the frame dependence of the wave function at the origin is
that of a non-interacting state, given by (7.101). The wave function of a pion with momentum P = ξ̂ Mπ sinh |ξ| is
then, at the origin,
−ξ·α/2 (0)
2m 0 ξ·α h 2m i
Φ(P )
π (x = 0) = e Φπ (x = 0)eξ·α/2 = F1 (0) 1 + γ e γ5 = F1 (0) 1 + 2 (γ 0 P 0 + P · γ) γ5
Mπ Mπ
h 2m i
γ 0 Φπ(P ) (0)γ 0 = −F1 (0) 1 + 2 P/ γ5 (7.168)
Mπ
The CSB matrix elements become

2m
h0|q̄(x)γ µ γ5 12 τ i q(x) Mπ , j, P , x0 = δij e−iP ·x Tr γ µ γ5 γ 0 Φ(P ) 0
= δij e−iP ·x F1 (0) 2 4P µ

π (0)γ
Mπ
h0|q̄(x)γ5 12 τ i q(x) Mπ , j, P , x0 = δij e−iP ·x Tr γ5 γ 0 Φ(P ) 0

= δij e−iP ·x (−4)F1 (0)

π (0)γ (7.169)
Comparing with the rhs. of relations (7.158) and (7.159) gives in both cases
Mπ2
F1 (0) = i
fπ (7.170)
8m
in agreement with our previous result (7.160) for m = 0. I leave to the future a more comprehensive study of
spontaneous chiral symmetry breaking in the present context.
79
VIII. BOUND STATE EPILOGUE
I conclude with several remarks on bound states. The more subjective ones may serve to stimulate further discussion
on these topics.
Confinement is an essential aspect of hadrons, and has been demonstrated in lattice QCD. Analytic approaches to
confinement are often formulated in terms of quark and gluon Green functions, aiming to show that colored states
are unphysical. This deals directly with the fundamental fields of the theory, and benefits from the accumulated
experience with (non-)perturbative methods for local fields. A drawback is that one needs to prove that something
does not exist, with little guidance from data.
Here I try to approach confinement from the opposite direction, in terms of the color singlet bound states that are
the asymptotic states of QCD. A main challenge is that bound states are extended objects, and more difficult to deal
with than the pointlike quarks and gluons. The experience gained from QED atoms is valuable, even though it does
not address confinement. Experts regard atoms as “non-perturbative” [5], yet use PQED for precise evaluations of
binding energies. This cautions not to rush to judgement based on the non-perturbative nature of hadrons.
The hadron spectrum has surprising “atomic” features, despite their large binding energies. Why can hadrons be
classified by their valence quarks and J P C quantum numbers only [1]? Their gluon and sea quark constituents do not
feature in the spectrum as clearly as they do in deep inelastic scattering. Why do hadron decays obey the OZI rule
[75–77], e.g., favor φ(1020) → K K̄ over φ(1020) → πππ? What causes quark-hadron duality, which in various guises
pervades hadron dynamics? These features are pictured by dual diagrams ([78, 79] and Fig. 10 a) which show only
valence quarks, no gluons. We lack understanding based on QCD.
Simple features are precious: experience shows that they often have correspondingly simple explanations. Taking
the data at face value limits the options in choosing the approach. An explanation based on QCD needs to use
perturbation theory, which is our only analytic, general method. It is successful for QED atoms as well as for hard
scattering. Addressing strongly bound states in motion is conceivable (if at all) only through an expansion in αs .
Imposing the restrictive boundary condition of an explicit yet formally exact method may reveal the QCD answer, or
its incompatibility with such conditions.
Quarks and gluons do not move faster than light. This implies retarded interactions between bound state constituents.
In a Fock state expansion the retardation is described by higher Fock states, with gluons “on their way” between
valence quarks. For the non-relativistic constituents of atoms photon exchange is almost instantaneous, so retardation
is a higher order, relativistic correction. For light, relativistic quarks retarded interactions would be expected to be
prominent and higher Fock states significant. Yet data suggests that most hadrons may be viewed as q q̄ and qqq
states. Is it conceivable that the valence quark Fock states dominate also for light hadrons?
Gauge theories can have an instantaneous interaction due to the choice of gauge. The absence of the ∂t A0 and ∇ · AL
terms in the action means that A0 and the longitudinal AL fields do not propagate in time and space. Their values are
determined by the gauge, which can be fixed over all space at an instant of time. This is illustrated by a comparison
of the A0 propagator in the Coulomb and Landau gauges,
00 0 1
DC (q , q) = i Coulomb gauge: ∇ · A = 0
q2
00 0 1 − (q 0 )2 /q 2
DL (q , q) = −i Landau gauge: ∂µ Aµ = 0 (8.1)
q 2 + iε
The Coulomb gauge propagator is independent of q 0 and thus ∝ δ(t) after a Fourier transform q 0 → t. The gauge
fixing term ∝ (∂µ Aµ )2 of Landau gauge adds the missing derivatives ∂t A0 and ∇ · AL to the action, allowing all
components of Aµ to propagate. The free field boundary condition of the perturbative S-matrix at t = ±∞ is
also covariant, resulting in the explicitly Poincaré invariant expansion of scattering amplitudes in terms of Feynman
diagrams.
Bound states defined at an instant of time retain explicit symmetry only under space translations and rotations (in
the rest frame). Coulomb gauge maintains these symmetries, sets AL = 0 and determines A0 through Gauss’ law,
δSQED
= −∇2 A0 (t, x) − eψ † ψ(t, x) = 0 (8.2)
δA0 (x)
The positions of the charges at any time t determine A0 instantaneously at all positions x. This gives the Coulomb
potential V (r) = −α/r for Positronium, which is the dominant interaction in atomic rest frames. The e− e+ Fock
state wave function is determined by the Schrödinger equation, and other Fock states are suppressed by powers of α.
These features make Coulomb gauge a common choice for bound state calculations.
80
Quantization in Coulomb gauge is complicated by the absence of a conjugate field for A0 , see [40, 55, 56]. Gauss’
law (8.2) is an operator relation which defines A0 in terms of ψ † ψ as a non-local quantum field. Yet the nature of
A0 appears to be more like a classical field, whose value gets fixed by the gauge. This aspect is better realized in
temporal gauge, A0 = 0. Quantization is straightforward since AL does have a conjugate field, E L . Gauss’ law is
no longer an operator equation of motion since A0 is fixed. Physical Fock states are restricted to be invariant under
time independent gauge transformations, which preserve A0 = 0. This determines the value of E L for each physical
state to be such that Gauss’ law is satisfied ([56–60] and section IV).
The semi-classical gauge field E L provides the instantaneous Coulomb potential of Positronium, as well as the anal-
ogous gluon exchange potential for quarks. The binding is weak, being proportional to the perturbative couplings
(α, αs ). There is no parameter like ΛQCD ∼ 1 fm−1 , which is required for confinement. However, in determining
E aL from Gauss’ law one needs to specify the boundary condition. In section IV C 2 I consider adding a homogeneous
solution (4.35) to the standard Coulomb solution in QCD. This gives rise to a spatially constant gluon field energy
density for each color component of a Fock state. The color octet field necessarily cancels in the sum over the color
components of singlet Fock states, giving no external effects.

The homogeneous solution gives rise to a confining potential
of O αs0 . In section VII I determined the potential
for various Fock states and made first checks of the O αs0 dynamics. Essential and non-trivial features include the
gauge invariance of electromagnetic form factors and the correct dependence of the bound state energies on the CM
momentum, E(P ). String breaking and duality seem to arise through an overlap between single and multiple bound
states as suggested by dual diagrams (Fig. 10 a). There is a J P C = 0++ solution with P µ = 0 in all frames whose
mixing with the vacuum allows solutions with spontaneously broken chiral symmetry.
A non-vanishing field energy density would explain why confinement is not seen in Feynman diagrams, which represent
an expansion around free states. The homogeneous solution (4.35) represents a major departure from previous
experience. It requires much further study, both of theoretical consistency and phenomenological relevance.
ACKNOWLEDGMENTS
During my work on the topics presented here I have benefited from the collaboration and advice of many colleagues.
Particular thanks are due to Jean-Paul Blaizot, Stan Brodsky, Dennis D. Dietrich, Matti Järvinen, Stephane Peigné
and Johan Rathsman. I am grateful for the hospitality of the University of Pavia, and the constructive response to
my lectures there in early 2020. During earlier stages of this project I have enjoyed visits of a month or more to
ECT* (Trento), CERN-TH (Geneva), CP3 (Odense), NIKHEF (Amsterdam), IPhT (Saclay) and GSI (Darmstadt).
I have the privileges of Professor Emeritus at the Physics Department of Helsinki University. Travel grants from the
Magnus Ehrnrooth Foundation have allowed me to maintain contacts and present my research to colleagues.
Appendix A: Solutions to exercises
1. Order of box diagram
contribution comes from the range of the loop integral d4 ` where `0 ∼ α2 m and |`| ∼ αm. Thus
R
The
R 4 leading
d ` ∼ α5 . The fermions are off-shell similarly to Ep1 − m ∼ α2 , as are the photons, q 2 = (q 0 )2 − q 2 ' −q 2 ∼ α2 .
Considering also the factor e4 ∼ α2 from the vertices the power of α is altogether 5 + 4 × (−2) + 2 = −1.
2. Contribution of the diagrams in Fig. 2(b,c)
As in Ex. A 1 the leading contribution comes from the range of the loop integral d4 ` where ` is of the order of the
R
bound state momenta. Then the Dirac structures simplify as in A1 of (2.8), giving (2m)4 . The two photon propagators
reduce to the `0 -independent Coulomb exchanges contained in A1 (p, `) and V (`−q) of (2.9). The fermion propagators
with momenta ` and ` − p1 − p2 ≡ ` − P may be expressed as in (2.6), keeping only the terms with the electron and
positron poles. The relevant part of the `0 integral is then
d`0
Z
1 1 i
0 0 0
= 0 (A.1)
2π ` − E` + iε ` − P + E` − iε P − 2E` + iε
81
Noting that the factor of 1/2E` in each fermion propagator (2.6) gives 1/(2m)2 we arrive at (2.9).
In the crossed diagram (c) of Fig. 2 the second factor in (A.1) is 1/(q10 − p02 − `0 + E − iε). The pole in `0 now has
Im `0 < 0 as in the first factor of (A.1). This allows closing the contour in the Im `0 > 0 hemisphere giving no leading
order contribution. In fact only the ladder diagrams shown in Fig. 1 contribute at leading O (1/α). They generate
the classical field of the bound state.
3. Derivation of (3.6)
For any (finite) momentum exchange the antifermion energy in the loops of Fig. 2(b,c) is Ē = mT + O (1/mT ). Since
p02 = mT we need only retain the negative energy pole term in the antifermion propagators (cf. (2.6)), in which
−p02 + Ē is of O (1/mT ). For the electron lines to go on-shell requires non-vanishing energy transfer `0 − p01 6= 0,
which causes the antifermion propagator to be off-shell by O (mT ). The photon propagators are thus independent of
`0 , e.g., D00 (` − p1 ) = i/(` − p1 )2 . The only relevant poles of the `0 integration are in the antifermion propagators,
d`0 h d`0
Z Z
i i i
+ =i 2πiδ(`0 − p01 ) = −1 (A.2)
2π `0 − p01 − iε q10 − `0 − iε 2π
The factor −i at the antifermion vertices cancels with the i of the Coulomb photon propagators. The standard rules
for the electron line then gives (3.6).
In the diagram with three uncrossed photon exchanges with momenta `1 − p1 , `2 − `1 and q1 − `2 the two antifermion
propagators similarly give
d`01 d`02 i2
Z
(A.3)
(2π) (`1 − p1 − iε)(`02 − q10 − iε)
2 0 0
The five other diagrams with crossed photons ensure the convergence of the integrals similarly as in (A.2), and do not
contribute when the integration contours of `01 and `02 are closed in the upper half plane. The full result is then due to
(A.3), which equals 1 (given the convergence). If the contours are chosen differently then the same result will come
from diagrams with crossed photons. The factors associated with the electron line are again given by the standard
rules, making the three photon contribution equal to scattering from an external potential.
Inserting the completeness condition for the Dirac wave functions,
†
Ψn,α (x)Ψ†n,β (y) + Ψn,α (x)Ψn,β (y) = δαβ δ 3 (x − y)
X
(A.4)
n
into the Dirac Hamiltonian (3.8) gives, recalling that the wave functions satisfy (3.3) and (3.4),
Z
†
Ψn,α (x)Ψ†n,β (y) + Ψn,α (x)Ψn,β (y) ψβ (y)
X
HD = /
dx dy ψ̄α0 (x) − i∇x · γ + m + eA(x) α0 α
n
XZ †
dx dy ψα† (x) Mn Ψn,α (x)Ψ†n,β (y) − M̄n Ψn,α (x)Ψn,β (y) ψβ (y)

=
n
X X
Mn c†n cn − M̄n c̄n c̄†n → Mn c†n cn + M̄n c̄†n c̄n

= (A.5)
n n
In the last step I normal-ordered the operators, neglecting the zero-point energies according to (3.16).
82
5. The expressions (3.24) for vacuum state
(a) The B and D coefficients defined in (3.20) and (3.21) satisfy

Z
dp
Ψ†m,α (p) uα (p, λ)u†β (p, λ) + vα (−p, λ)vβ† (−p, λ) Ψn,β (p) =
X
Ψ† (p)Ψn,α (p) = 0

Bmp B np + Dmp Dnp =
p
(2π)3 m,α
(A.6)
−1 −1
Multiplying by (B )qm (D )rn and summing over m, n gives
−1
− (B −1 )qm Dmr = (D )rn B nq (A.7)
Using also b†q d†r = −d†r b†q shows the equivalence of the two expressions for the vacuum state in (3.24).
(b) In order to verify that cn |Ωi = 0 we note that since bp essentially differentiates the exponent in |Ωi,
Bnq bq |Ωi = −Bnq B −1 qm Dmr d†r |Ωi = −Dnr d†r |Ωi

(A.8)
This cancels the contribution of the second term in the definition (3.20) of cn . The demonstration that c̄n annihilates
the vacuum is similar. Thus
cn |Ωi = c̄n |Ωi = HD |Ωi = 0 (A.9)
6. Derivation of the identities (3.39)
We may start by evaluating (here I make no difference between lower and upper indices)
(x̂ × L)i = ijk x̂j kln xl (−i∂n ) = (δil δjn − δin δjl )x̂j xl (−i∂n ) = −ix̂i r∂r + ir∂i (A.10)
Multiplying by αi /r we have the first relation,

1
α · x̂ × L = −iα · x̂ ∂r + iα · ∇ (A.11)
r
The second identity:
xj δ ij xi xj xj 2 1
−i(α · ∇)i(α · x̂) = αi αj ∂i = (δij + iγ5 ijk αk ) − 3 + ∂i = + ∂r + γ5 α · L (A.12)
r r r r r r
1
The momentum space wave function (3.19) for j = 2 and positive parity is
!
Z ∞ Z 1 χλ
2 −ipr cos θ

Ψ1/2,λ,+ = 2π dr d cos θ r e F (r) + iG(r) α · p̂ cos θ (A.13)
0 −1 0
I chose the z-axis of the x-integration along p. The integration over the azimuthal angle ϕ leaves only the αz
component of α · x. Expressing the factor cos θ as a derivative of exp(−ipr cos θ) and using G = iF (which holds at
large r, where the stationary point is located for large p) we have
Z ∞ !
2 i i −ipr ipr
χλ
Ψ1/2,λ,+ = 2π dr r F (r) 1 − α · p̂ ∂p e −e (A.14)
0 r pr 0
For p → ∞ the phase of e±ipr is rapidly oscillating, so the leading contribution comes from the region of the r-
integration where the phase of the integrand is stationary. The stationary phase approximation is
Z
00
dr f (r)eiϕ(r) ' eε(ϕ (rs ))iπ/4 f (rs ) eiϕ(rs ) (A.15)
83
The function f (r) is assumed to be varying slowly compared to the phase exp[iϕ(r)]. The phase is stationary at
ϕ0 (rs ) = 0 and ε(x) = 1 (−1) for x > 0 (x < 0). According to (3.57) we have in the integral of (A.14) ϕ(r) =
V 0 r2 /2 ∓ pr. There is a stationary phase for r > 0 only with the exp(−ipr) term, giving rs = p/V 0 , ϕ(rs ) = −p2 /2V 0
and ϕ00 (rs ) > 0. From (3.57) the contribution proportional to the unit Dirac matrix in (A.14) is then
!
2πiN p β+1 −ip2 /2V 0 χλ
Ψ1/2,λ,+ F = e (A.16)
p V0 0
The leading term in the contribution ∝ α · p̂ has ∂p → −ip/V 0 = −irs , giving the full result
!
2πiN p β+1 2 0 χ λ
Ψ1/2,λ,+ = (1 − α · p̂) e−ip /2V (A.17)
p V0 0
Consider now the expressions for the u and v spinors for |p| → ∞,
! !
p
/+m χλ p χλ
u(p, λ) ≡ p ' |p| (1 + α · p̂) (|p| → ∞)
Ep + m 0 0
! !
−p/+m 0 p 0
v(p, λ) ≡ p ' |p| (1 + α · p̂) (|p| → ∞) (A.18)
Ep + m χ̄λ χ̄λ
Consequently, omitting a common factor in the distributions e± (p, s) (3.25),
!
− †
χλ
χ†s

e (p, s) = u (p, s)Ψ1/2,λ,+ ∼ 0 (1 + α · p̂)(1 − α · p̂) =0
0
!
†
χλ
+
χ̄†s = −2χ̄†s σ · p̂ χλ

e (p, s) = v (−p, s)Ψ1/2,λ,+ ∼ 0 (1 − α · p̂)(1 − α · p̂) (A.19)
0
which establishes (3.58).
8. Gauge transformations generated by Gauss operator
The unitary operator defined in (4.13),

Z Z
U (t) = 1 + i dy G(t, y)δΛ(y) = 1 + i dy ∂i E i (t, y) − eψ † ψ(t, y) δΛ(y)

(A.20)
transforms Aj (t, x) as,

Z
δAj (t, x) ≡ U (t)Aj (t, x)U −1 (t) − Aj (t, x) = i dy ∂i E i (t, y) − eψ † ψ(t, y) δΛ(y), Aj (t, x)

Z
= −i dy E i
(t, y)∂iy δΛ(y), Aj (t, x) = ∂j δΛ(x)
(A.21)
Similarly for the electron field,
Z
δψ(t, x) ≡ U (t)ψ(t, x)U −1 (t) − ψ(t, x) = −ie dy ψ † ψ(t, y)δΛ(y), ψ(t, x) = ie δΛ(x)ψ(t, x) (A.22)
9. Derive (5.8).
The anticommutation relation of the electron fields gives

Z h ← → ← → i
− + (0) (0)
J e e ; B, P = 0 = dx1 dx2 ψ̄(x1 ) J Λ + (x1 )ΦB (x1 − x2 ) Λ − (x2 ) − Λ + (x1 )ΦB (x1 − x2 ) Λ − (x2 )J ψ(x2 ) |0i

(A.23)
84
J commutes with the Λ± projectors (5.2), which have a rotationally invariant form. It is instructive to see how this
→ →
works out explicitly. The commutator Λ − , L + S gets contributions from the derivatives in Λ − differentiating x in
→
L = −ix × ∇, and from the commutators of the Dirac matrices in Λ − with S = 12 γ5 α.
p →
The ∇2 in E = −∇2 + m2 of Λ − (x) commutes with Li ,
∂j ∂j , εik` xk ∂` = ∂j εij` ∂` + εij` ∂` ∂j = 0

(A.24)
since ∂j ∂` = ∂` ∂j , whereas εij` = −εi`j . The commutator of iα · ∇ with Li (in my notation αj = αj ),
iαj ∂j , −iεik` xk ∂` = εij` αj ∂`

(A.25)
is cancelled by its commutator with S i . Using αi αj = δij + iεijk αk γ5 ,
iαj ∂j , 12 γ5 αi = −εjik αk ∂j

(A.26)
→
Finally, γ 0 , S = 0. We see that the commutator Λ − , J = 0 requires contributions from both L and S. Similarly
J commutes with other rotationally invariant structures. Bringing J through the Λ± projectors gives (5.8).
The commutator with the orbital angular momentum L in (5.8) arises from
→ ← →
(0) (0) (0)
x1 × (−i∇1 )ΦB (x1 − x2 ) − ΦB (x1 − x2 )x2 × (i∇2 ) = (x1 − x2 ) × (−i∇1 )ΦB (x1 − x2 ) (A.27)
h i
(0) (0) (0)
Hence in L, ΦB (x) = x × [−i∇ΦB (x)] the x-derivatives of L apply only to ΦB (x).
The transformation (5.15) of the fields under charge conjugation gives

← →
Z
(P )
C e− e+ ; B, P = − dx1 dx2 ψ(x1 )T α2 Λ + (x1 )ΦB (x1 − x2 ) Λ − (x2 )α2 ψ̄ T (x2 ) |0i

(A.28)
Take the transpose on the rhs., recalling that the anticommutation of the fields gives a minus sign,
←
Z
− + 1 (P ) T
C e e ; B, P = dx1 dx2 ψ̄(x2 )α2 T (E2 + iαT · ∇2 − γ 0 m)ΦB (x1 − x2 )

2E2
1 →
× (E1 − iαT · ∇1 + γ 0 m)α2 T ψ(x1 ) |0i (A.29)
2E1
Recalling that α2 αT α2 = −α and changing integration variables x1 ↔ x2 we get

← →
Z
− + (P ) T
C e e ; B, P = dx1 dx2 ψ̄(x1 ) Λ + (x1 )α2 ΦB (x2 − x1 )α2 Λ − (x2 )ψ(x2 ) |0i
(A.30)
Comparing with the definition (5.1) of |e− e+ ; B, P i we see that (5.16) implies (5.17).
11. Verify (5.22).
At lowest order in α the projectors Λ± in the definition (5.1) of the Positronium state may be expressed in terms of
(P )
the CM momentum P , by partial integration and ignoring the O (α) contributions from differentiating ΦB (x1 − x2 ),
1 2
EP ∓ α · P ± 2mγ 0 = Λ†± (P ) = Λ± (P )

Λ± (P ) = Λ+ (P )Λ− (P ) = 0 (A.31)
2EP
85
Anticommutating the fields in he− e+ ; B 0 , P 0 | with those in |e− e+ ; B, P i gives using (5.19),
Z
(P 0 )∗ (P ) 0 0 ∗
− + 0 0 − +
dx1 dx2 ei(P −P )·(x1 +x2 )/2 F (P ) (x1 − x2 )F (P ) (x1 − x2 )Tr B,B0

he e ; B , P e e ; B, P = NB0 NB
Z
(P ) 2
= NB (2π)3 δ(P − P 0 ) dx |F (P ) (x)|2 Tr B,B0 (A.32)
Tr B,B0 = Tr Γ†B0 Λ+ (P )ΓB Λ− (P )

Commuting Λ− (P ) through ΓB gives for ΓB = γ5 (Parapositronium) and ΓB = α3 (Orthopositronium with λ = 0),

2m † 8m2
Tr B,B0 = Tr ΓB0 Λ+ (P )γ 0 ΓB = 2 δB,B0 (A.33)
EP EP
while for ΓB = e± · α (Orthopositronium with λ = ±1),
Tr B,B0 = Tr Γ†B0 Λ+ (P )ΓB = 2δB,B0

(A.34)
Using these expressions for Tr B,B0 and the normalization (5.19) of F (P ) (x) in (A.32) the normalization (5.4) of the
state implies (5.22).
Derive the expression for e− e+ γ; q, s in (5.38).

12.
The contribution to |e− e+ γ; q, si from Hint , ψ̄(x1 ) is

← ← →
Z
(P )
e dx1 dx2 ψ̄(x1 ) Λ + (x1 )α · ε∗s (q)e−iq·x1 a† (q, s) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )ψ(x2 ) |0i (A.35)
←
where I inserted Λ + (x1 ) to select the b† contribution in ψ̄(x1 ) as in (5.3). We have then
←
←
j
← ← 1 h ←
0 j
j ← i ← −2i ∂ 1,j
Λ + (x1 )α Λ + (x1 ) = Λ + (x1 ) (E1 + iα · ∇1 − mγ )α + α , −iα · ∇1 = Λ + (x1 )
2E1 2E1
← −P j
→ Λ + (x1 ) (A.36)
EP
←
The first term in the square bracket vanishes when multiplied by Λ + (x1 ). The final result follows after partial
→
integration, with the leading order term due to i ∂ 1,j exp[iP · (x1 + x2 )/2] in (A.35).
The contribution to |e− e+ γ; q, si from [Hint , ψ(x2 )] is similarly
← → →
Z
(P )
e dx1 dx2 ψ̄(x1 ) Λ + (x1 )eiP ·(x1 +x2 )/2 ΦB (x1 − x2 ) Λ − (x2 )α · ε∗s (q)e−q·x2 a† (q, s) Λ − (x2 )ψ(x2 ) |0i (A.37)
where now
→
→ → 1 → → → 2i ∂ 2,j →
j j
(E2 − iα · ∇2 + mγ 0 ) + αj , iα · ∇2 Λ − (x2 ) =

Λ − (x2 )α Λ − (x2 ) = α Λ (x2 )
2E2 2E2 −
Pj →
→ Λ (x2 ) (A.38)
EP −
Using (A.36) in (A.35) and (A.38) in (A.37) and adding the two contributions gives (5.38).
13. Derive the expression (6.32).
The frame dependence of functions like φ0 (τ ) and φ1 (τ ) that do not explicitly depend on P or E arises only due to
the P -dependence of τ (x).

∂τ ∂ 2 2xP
M − 2EV + V 2 /V 0 = −

= (A.39)
∂P x ∂P E
86
Recalling also that for functions that depend on x only via τ ,

∂ ∂τ ∂ ∂
= = −2(E − V ) (A.40)
∂x P ∂x P ∂τ ∂τ
we have

∂φ0,1 ∂φ0,1 xp
=E = ∂x φ0,1 (A.41)
∂ξ x ∂P x E−V P
Applying this to eσ1 ζ/2 Φ(P ) e−σ1 ζ/2 gives

∂Φ(P )

∂ h σ1 ζ/2 (P ) −σ1 ζ/2 i σ1 ζ/2 1 ∂ζ
(P ) e−σ1 ζ/2
e Φ e = e σ 1 , Φ + (A.42)
∂ξ
x 2 ∂ξ
x ∂ξ
x
∂Φ(P )

xp h i
σ1 ζ/2 (P ) −σ1 ζ/2 xp σ1 ζ/2 1 ∂ζ
(P ) e−σ1 ζ/2
= ∂x e Φ e = e σ1 , Φ +
E−V ξ E−V 2 ∂x ξ ∂x ξ
which implies
∂Φ(P ) xp ∂Φ(P ) 1 xp ∂ζ ∂ζ
σ1 , Φ(P )

= + − (A.43)
∂ξ x E − V ∂x ξ 2 E − V ∂x ξ ∂ξ x

It remains to work out the derivatives of ζ. From its definition (6.29),

V 0 P (E − V ) ∂ζ P
∂x (sinh ζ)ξ = ∂x ζ ξ cosh ζ = =⇒ =
(V 0 τ )3/2 ∂x ξ τ
(E − V )(M 2 − EV ) ∂ζ M 2 − EV
∂ξ (sinh ζ)x = ∂ξ ζ x cosh ζ = =⇒ = (A.44)
(V 0 τ )3/2 ∂ξ x V 0τ
Using these in (A.43) gives (6.32).
14. Derive the expression (6.74).
The expressions (6.40) for φ1 (τ ) and φ0 (τ ) at large τ are,

4V 0 p πm2 /V 0 2 0
− 1 e−θ(−τ )πm /2V sin 41 τ − 12 m2 /V 0 log( 21 |τ |) + arg Γ(1 + im2 /2V 0 )

φ1 (|τ | → ∞) = √ e
πm
4V 0 p πm2 /V 0 2 0
− 1 e−θ(−τ )πm /2V cos 41 τ − 12 m2 /V 0 log( 12 |τ |) + arg Γ(1 + im2 /2V 0 )

φ0 (|τ | → ∞) = −i √ e (A.45)
πm
Since state B has positive parity φB1 [τB (x = 0)] = φB1 (τB = MB2 /V 0 ) = 0 according to (6.28). This determines the
masses MBn in the Bj limit,
0
1 2
4 MBn /V − 12 m2 /V 0 log( 21 MBn
2
/V 0 ) + arg Γ(1 + im2 /2V 0 ) = n · π (A.46)
where n is a large positive integer. Subtracting the lhs. from the arguments of the sin and cos functions in (A.45)
gives a sign (−1)n to φB0 and φB1 . Using also −2EB x = 2PB1 x − 2PA+ (1 − xBj )x from (6.69) gives
4V 0 (−1)n p πm2 /V 0
− 1 exp − θ(−τB )πm2 /2V 0

φB1 (|τB | → ∞) = √ e
πm
h m2 V 0x i
× sin 1 1
2 PB x − 12 PA+ (1 − xBj )x + 41 V 0 x2 − log −

+
2V 0 PA (1 − xBj )
1
4V 0 (−1)n p πm2 /V 0
− 1 exp − θ(−τB )πm2 /2V 0

φB0 (|τB | → ∞) = − i √ e
πm
h m2 V 0x i
× cos 1 1
2 PB x − 12 PA+ (1 − xBj )x + 41 V 0 x2 − log − (A.47)

+
2V 0 PA (1 − xBj )
1
87
The asymptotically large phase 12 PB1 x must cancel the one in the factor sin 12 q 1 x = sin 12 (PB1 − PA1 )x of (6.72) to

give a non-vanishing result. Using
sin α sin β = 12 cos(α − β) − cos(α + β)

sin α cos β = 21 sin(α + β) + sin(α − β)

(A.48)
and defining the angle

+ m2 V 0x
1 0 2
1
PA (1 − xBj ) − PA1 x +

ϕB (x) ≡ log 1 − + − 4V x (A.49)

2 2V 0
PA (1 − xBj )
the expression (6.72) for GAB gives (6.74) when terms with asymptotically large phases are neglected,
15. Do the x-integral in (6.74) numerically for the parameters in Fig. 7, and compare.
Separate the integral in the form factor (6.74) into three parts,
16iV 0 p πm2 /V 0
EB G(q 2 ) = (−1)n √ e − 1 (I1 + I2 + I3 ) (A.50)
πm
Z ∞
2 0
dx i sin ϕB φA0 (τA ) + cos ϕB φA1 (τA ) e−θ(x−x0 )πm /2V

I1 =
0
x0
2m2
Z
I2 = dx cos ϕB φA1 (τA )
0 V 0 (x0 − x)(PA+ − V 0 x)
∞
2m2
Z
2 0
I3 = − dx cos ϕB φA1 (τA ) 0 (x + 0
e−πm /2V (A.51)
x0 V − x0 )(PA − V x)
The I1 integrand oscillates with constant amplitude at large x, where the approximation (A.45) for φA (τA → ∞)
applies. I1 is further divided into three parts. In I1a the range is 0 < x < x1 , where x1 > x0 and τA (x1 ) > 0, ensuring
that θ(−τA ) = 0, θ(−τB ) = 1 for x > x1 . I1b integrates over x1 ≤ x ≤ ∞ with φA0 and φA1 replaced by the difference
with their large x approximations, φA − φas as
A . Finally I1c integrates φA over x1 ≤ x ≤ ∞:
Z x1
2 0
dx i sin ϕB φA0 (τA ) + cos ϕB φA1 (τA ) e−θ(x−x0 )πm /2V

I1a =
0
Z ∞ −πm2 /2V 0
dx i sin ϕB [φA0 (τA ) − φas as

I1b = A0 (τA )] + cos ϕB [φA1 (τA ) − φA1 (τA )] e (A.52)
x1
Z ∞ −πm2 /2V 0
dx i sin ϕB φas as

I1c = A0 (τA ) + cos ϕB φA1 (τA ) e
x1
The oscillations in I1b at large x are damped, allowing a numerical integration. The phase in φas
A (A.45) is
ϕA = 14 τA − 12 m2 /V 0 log( 21 |τA |) + arg Γ(1 + im2 /2V 0 ) (A.53)
The I1c integral reduces to

∞
4V 0 p
Z
I1c = −√ 1 − e−πm2 /V 0 dx sin ϕC
πm x1
m2
ϕC ≡ −(ϕA + ϕB ) = 21 PA+ xBj x − 14 MA2 /V 0 − arg Γ(1 + im2 /2V 0 ) + log( 21 τA ) − log(x/x0 − 1)

0
(A.54)
2V
This integral is evaluated by rotating the contour in u ≡ x − x1 by π/2 to ensure exponential convergence,
Z ∞ Z ∞ Z i∞
iϕC (x)
dx sin ϕC = Im dx e = Im du eiϕC (u+x1 ) (A.55)
x1 x1 0
88
In I2 the range 0 < x < x0 is transformed into 0 < y < ∞ through

dx
y = − log(1 − x/x0 ) x = x0 (1 − e−y ) = dy (A.56)
x0 − x
The y-integration is further split into 0 < y < y2 and y2 < y < ∞. The first path is finite and readily integrated
numerically. The latter path is rotated by π/2, giving exponential convergence in y2 < y < y2 + i∞ when using
cos ϕB = Re exp(−iϕB ), due to the term ym2 /2V 0 in −ϕB (6.74). x ' x0 is constant on the complex path when y2
is large, allowing the integral to be be evaluated analytically. The result for I2 should be independent of y2 .
In I3 the integration contour is split into x0 < x < 2x0 and 2x0 < x < ∞. In the first range the integration variable
is changed to
dx
y = − log(x/x0 − 1) x = x0 (1 + e−y ) = −dy (A.57)
x0 − x
The y-range 0 < y < ∞ is further split into 0 < y < y3 and y3 < y < ∞. The first is integrated numerically, and in
the second the path is rotated by π/2, ranging over y3 < y < y3 + i∞. For large y3 the value of x ' x0 is constant
on the complex contour, allowing an analytic integration. The integration over 2x0 < x < ∞ is numerically stable, as
the oscillations at large x are damped.
16. Derive the q q̄g potential (7.22)
Using the commutators in (7.3) and (7.17) the operation of Ea (y) (7.1) on the |q q̄gi state (7.21) gives
Ea (y) |q q̄gi = ψ̄A0 (x1 )TAa0 A Aib (xg )TAB
b

ψB (x2 )δ(y − x1 )
+ ψ̄A (x1 )ifabc Aic (xg )TAB

b
ψB (x2 )δ(y − xg )
− ψ̄A (x1 )Aib (xg )TAB

b a

TBB 0 ψB 0 (x2 )δ(y − x2 ) |0i (A.58)
(0)
When Ea (y) and Ea (z) in HV (4.37) act on the same (quark or gluon) constituent we may use the previous results
(7.5) and (7.18) showing that the coefficients of x21 and x22 are CF while that of x2g is Nc , multiplied by the common
factor 12 κ2 dx + gκ . The new contributions are
R
1
x1 · x2 : − 2 ψ̄A0 (x1 )TAa0 A Aib (xg )TAB
b a
TBB 0 ψB 0 (x2 ) |0i = |q q̄gi
Nc
x1 · xg : 2 ψ̄A0 (x1 )TAa0 A ifabc Aic (xg ) TAB

b
ψB (x2 ) |0i = −Nc |qg q̄i
x2 · xg : − 2ψ̄A (x1 ) ifabc Aic (xg ) TAB

b a
TBB 0 ψB 0 (x2 ) |0i = −Nc |qg q̄i (A.59)
Altogether,
(0) 1 2
dx + gκ CF (x21 + x22 ) + Nc x2g − Nc (x1 + x2 ) · xg + 1
R
HV |q q̄gi = 2κ Nc x 1 · x2 |q q̄gi
1 2
R 2
= 2κ dx + gκ dqq̄g (x1 , x2 , xg ) |q q̄gi
q
dqq̄g (x1 , x2 , xg ) ≡ 14 (Nc − N2c )(x1 − x2 )2 + Nc (xg − 12 x1 − 12 x2 )2 (A.60)

For the O κ2 term to give the universal energy EΛ (7.7) we need to choose the normalization of the homogeneous
solution as
Λ2 1
κqq̄g = √ (A.61)
g CF dqq̄g (x1 , xg , x2 )
The O (gκ) contribution to HV gives the potential,
(0) 2 Λ2
Vqq̄g (x1 , x2 , xg ) = gκqq̄g dqq̄g (x1 , x2 , xg ) = √ dqq̄g (x1 , x2 , xg ) (A.62)
CF
(1)
When the self-energies are subtracted HV has contributions only from the three terms in (A.59),
h 1 1 1 1 i
(1)
Vqq̄g (x1 , x2 , xg ) = 12 αs − Nc + (A.63)
Nc |x1 − x2 | |x1 − xg | |x2 − xg |
89
(0)
17. Derive the expression for HV |8 ⊗ 8i in (7.36)
Recall from (7.33),

b b
|8 ⊗ 8i = ψ̄A (x1 )TAB ψB (x2 ) ψ̄C (x3 )TCD ψD (x4 ) |0i
1
ψ̄A (x1 )ψB (x2 ) ψ̄B (x3 )ψA (x4 ) |0i = 2 |8 ⊗ 8i + Nc |1 ⊗ 1i (A.64)
1 2
R (0)
In the following I leave out the common factor 2κ dx + gκ in HV (7.1),
Z
(0) 1 2
R
HV = 2κ dx + gκ dy dz y · z Ea (y)Ea (z) (A.65)
and make use of the commutation relations (7.3),
Ea (x), ψ̄A (x1 ) = ψ̄A0 (x1 )TAa0 A δ(x − x1 ) a

Ea (x), ψA (x2 ) = −TAA 0 ψA0 (x2 )δ(x − x2 ) (A.66)
and of the SU(Nc ) generator relations (7.4).

The commutators of Ea (y) and Ea (z) with the same quark at xi (i = 1, . . . 4) in |8 ⊗ 8i gives y · z = x2i , color factor
T a T a = CF I and state |8 ⊗ 8i. The commutators with ψ̄(x1 ) and ψ(x2 ) gives
x1 · x2 : − 2ψ̄A0 (x1 )TAa0 A TAB

b a
TBB b
0 ψB 0 (x2 )ψ̄C (x3 )TCD ψD (x4 ) |0i =
1
Nc |8 ⊗ 8i (A.67)
The commutators with ψ̄(x1 ) and ψ̄(x3 ) give
x1 · x3 : 2ψ̄A0 (x1 )TAa0 A TAB

b
ψB (x2 )ψ̄C 0 (x3 )TCa 0 C TCD
b
ψD (x4 ) |0i (A.68)
The color factors

1 1 b b b b
2 (δA C δAC
0 0 − Nc δA A δC C )TAB TCD
0 0 = 14 δA0 C δAC 0 (δAD δBC − 1
Nc δAB δCD ) − 1
2Nc δA A δC C TAB TCD
0 0 (A.69)
give for the coefficient of x1 · x3 in (A.68),

CF
1 1
1 2
2 ψ̄B (x1 )ψB (x2 )ψ̄D (x3 )ψD (x4 ) |0i − 2Nc ψ̄D (x1 )ψB (x2 )ψ̄B (x3 )ψD (x4 ) |0i − Nc |8 ⊗ 8i = Nc |1 ⊗ 1i − Nc |8 ⊗ 8i
(A.70)
The commutators with ψ̄(x1 ) and ψ̄(x4 ) give
x1 · x4 : − 2ψ̄A0 (x1 )TAa0 A TAB

b b
ψB (x2 )ψ̄C (x3 )TCD a
TDD 0 ψD 0 (x4 ) |0i (A.71)
Now the color factors

1 1 b b b b
2 (δA D δAD
0 0 − Nc δA A δD D )TAB TCD
0 0 = 14 δA0 D0 δAD (δAD δBC − 1
Nc δAB δCD ) − 1
2Nc δA A δDD TAB TCD
0 0 (A.72)
give for the coefficient of x1 · x4 in (A.71),
Nc Nc2 −2
1 1
|8 ⊗ 8i = − C

− 2 + 2Nc ψ̄A0 (x1 )ψB (x2 )ψ̄B (x3 )ψA0 (x4 ) |0i + Nc Nc |1 ⊗ 1i −
F
Nc |8 ⊗ 8i (A.73)
The coefficients of x2 · x3 , x2 · x4 and x3 · x4 are the same as those of x1 · x4 , x1 · x3 and x1 · x2 , respectively.

Altogether,
(0) 1 2
n 1 2 2
1 1
− x2 )2
R
HV |8 ⊗ 8i = 2κ dx + gκ 2 Nc (x1 − x4 ) + 2 Nc (x2 − x3 ) − 2Nc (x1
o
1
− x4 )2 − 2 CF

− 2Nc (x3 Nc (x1 − x2 ) · (x3 − x4 ) |8 ⊗ 8i + Nc (x1 − x2 ) · (x3 − x4 ) |1 ⊗ 1i (A.74)
When expressed in terms of the separations (7.35) this gives (7.36).

90
18. Verify that the expression (7.60) for Φ−+ (x) satisfies the bound state equation (7.47) given the radial
equation (7.59).
The BSE as in (7.47) applied to Φ−+ (x) in the alternative forms of (7.60),
→ ←
h − Φ(x) + Φ(x) h − = 0
→ ←
Φ−+ (x) = h + γ5 F1 (r)Yjλ (x̂) = F1 (r)Yjλ (x̂) γ5 h + (A.75)
allows the use of

→ → 4 →2
2 4iV 0 →
h− h+ = (−∇ + m ) − 1 + α · x (iα · ∇ + mγ 0 )
(M − V )2 r(M − V )3
← ← ←2 4 ← 4iV 0
h + h − = (−∇ + m2 ) − 1 + (iα · ∇ − mγ 0
) α · x (A.76)
(M − V )2 r(M − V )3
Moving the γ5 to the right in the BSE,

→ → ← ←
h − h + γ5 F1 (r)Yjλ (x̂) + F1 (r)Yjλ (x̂)γ5 h + h −
h 8 2 2 4iV 0 xj →
0
i
= (−∇ + m ) − 2 + αj , i ∂ k α k + mγ F1 (r)Yjλ (x̂)γ5 (A.77)
(M − V )2 r(M − V )3
→ →
Using {αj , αk } = 2δjk , αj , γ 0 = 0, xj ∂ j = r ∂ r and ∇2 = (1/r2 )∂r (r2 ∂r ) − L2 /r2 with L2 Yjλ (x̂) = j(j + 1)Yjλ (x̂)

gives the radial equation (7.59).
19. Derive the coupled equations (7.100) from the bound state equation (7.98).
I make use of commutator identities such as,
[A, BC] = [A, B] C + B [A, C] (A.78)
{A, BC} = [A, B] C + B {A, C} = {A, B} C − B [A, C] (A.79)

{A, {B, C}} = − [B, [A, C]] when A, B = 0 (A.80)

{A, [B, C]} = − {B, [A, C]} when A, B = 0 (A.81)

[A, {B, C}] = − [B, {A, C}] when A, B = 0 (A.82)
{A, [A, C]} = [A, {A, C}] = A2 , C

(A.83)
{A, {A, C}} = 2A {A, C} when A2 = 1 (A.84)
[A, [A, C]] = 2A [A, C] when A2 = 1 (A.85)
Taking the commutator i∇ · [α, BSE] of the bound state equation (7.98) gives
h i h i h i h i
i∇ · α, (E − V )Φ(P ) = i∇ · α, i∇ · α, Φ(P ) − 1
i∇ · α, P · α, Φ(P ) + m i∇ · α, γ 0 , Φ(P )

2 (A.86)
The first term on the rhs. vanishes due to the commutator identity (A.83), when we recall that ∇ in the BSE always
operates on Φ(P ) . The identity (A.80) implies for the third term on the rhs. of (A.86),
h i n o
m i∇ · α, γ 0 , Φ(P ) = −m γ 0 , i∇ · α, Φ(P )

(A.87)
91
Using the original BSE (7.98) on the rhs. of (A.87) we get

h i n h io n h io n o
m i∇ · α, γ 0 , Φ(P ) = −m γ 0 , 12 P · α, Φ(P ) + m2 γ 0 , γ 0 , Φ(P ) − m γ 0 , (E − V )Φ(P )

n h io n o
= 21 m P · α, γ 0 , Φ(P ) − m(E − V ) γ 0 , Φ(P )
n o n o
1
P · α, − i∇ · α, Φ(P ) + 21 P · α, Φ(P ) + (E − V )Φ(P ) − m(E − V ) γ 0 , Φ(P )

= 2 (A.88)

where I used (A.81), (A.83) and in the last step expressed m γ 0 , ΦP using the BSE (7.98). The second term on the
rhs. of (A.88) vanishes according to (A.83). Inserting this result in (A.86) we have
h i
i∇ · α, (E − V )Φ(P ) =
h h ii n n oo h i n o
− 1
2 i∇ · α, P · α, Φ(P ) − 1
2 P · α, i∇ · α, Φ(P ) + 12 (E − V ) P · α, Φ(P ) − m(E − V ) γ 0 , Φ(P ) (A.89)
The sum of the first two terms on the rhs. simplifies. With ∇ · α = αi ∂i and P · α = P j αj ,
h h ii
αi , αj , ∂i Φ(P ) = αi (αj ∂i Φ(P ) − ∂i Φ(P ) αj ) − (αj ∂i Φ(P ) − ∂i Φ(P ) αj )αi
n n oo
αj , αi , ∂i Φ(P ) = αj (αi ∂i Φ(P ) + ∂i Φ(P ) αi ) + (αi ∂i Φ(P ) + ∂i Φ(P ) αi )αj (A.90)
so that
h h ii n n oo
αi , αj , ∂i Φ(P ) + αj , αi , ∂i Φ(P ) = (αi αj + αj αi )∂i Φ(P ) + ∂i Φ(P ) (αj αi + αi αj ) = 4∂j Φ(P ) (A.91)
Using this in (A.89) and dividing by E − V gives

1 h i n o n o 2i
i∇ · α, (E − V )Φ(P ) − 12 P · α, Φ(P ) + m γ 0 , Φ(P ) = − P · ∇Φ(P ) (A.92)
E−V E−V
For a linear potential i∇ · α V 0 |x| = iV 0 α · x/r, where r = |x|. Bringing this derivative to the rhs. in (A.92),
h i n o n o 1 V0 h i
i∇ · α, Φ(P ) − 1
2 P · α, Φ(P ) + m γ 0 , Φ(P ) = − 2iP · ∇Φ(P ) + iα · x, Φ(P ) (A.93)
E−V r
The lhs. is now the same as in the original BSE (7.98), with commutators and anticommutators interchanged. Adding
and subtracting the two equations and dividing by E − V we get equations (7.100).
(P )
20. Derive the frame dependence (7.101) of ΦV =0 (x) using the boost generator K0z .
The action of K0z (t = 0) (7.106) on the state (7.97) with P = (0, 0, P ) and V = 0,
Z
(P )
|M, P i0 = dx1 dx2 ψ̄(x1 )eiP (z1 +z2 /2 ΦV =0 (x1 − x2 )ψ(x2 ) |0i (A.94)
is determined by
← ←
K0 , ψ̄(x1 ) = ψ † (x1 ) z1 (−iα · ∇1 − mγ 0 ) + 12 iα3 γ 0 = ψ̄(x1 ) z1 (iα · ∇1 − mγ 0 ) − 21 iα3
z
→
[K0z , ψ(x2 )] = − z2 (iα · ∇2 − mγ 0 ) + 12 iα3 ψ(x2 )

(A.95)
Making the derivatives act on the wave function through partial integration,
→
Z n
(P )
K0 |M, P i0 = dx1 dx2 ψ̄(x1 ) z1 (−iα · ∇1 − mγ 0 ) − 12 iα3 eiP (z1 +z2 /2 ΦV =0 (x1 − x2 )
z

← o
(P )
− eiP (z1 +z2 /2 ΦV =0 (x1 − x2 ) z2 (−iα · ∇2 − mγ 0 ) + 12 iα3 ψ(x2 ) |0i

(A.96)
92
← ←
Using z2 (−iα · ∇2 ) + 12 iα3 = (−iα · ∇2 )z2 − 12 iα3 and then expressing z1,2 = 12 (z1 + z2 ) ± 12 (z1 − z2 ), the BSE (7.99)
(P )
satisfied by ΦV =0 ,
→ (P ) ←
(P ) (P )
iα · ∇1 − 12 P α3 + mγ 0 ΦV =0 (x1 − x2 ) + ΦV =0 (x1 − x2 ) − iα · ∇2 + 21 P α3 − mγ 0 = E ΦV =0 (x1 − x2 ) (A.97)

√
reduces the coefficient of 21 (z1 + z2 ) to −E, with E = M cosh ξ = P 2 + M 2 . We have then
Z n
(P ) (P )
K0z |M, P i0 = dx1 dx2 ψ̄(x1 )eiP (z1 +z2 /2 − 21 (z1 + z2 )E ΦV =0 (x1 − x2 ) − 12 i α3 , ΦV =0 (x1 − x2 )

(A.98)
→ ← o
(P ) (P )
− 21 (z1 − z2 ) (iα · ∇1 − 12 P α3 + mγ 0 )ΦV =0 (x1 − x2 ) + ΦV =0 (x1 − x2 )(iα · ∇2 − 12 P α3 + mγ 0 ) ψ(x2 ) |0i

→ ←
(P )
where ∇1 and ∇2 only differentiate ΦV =0 (x1 − x2 ). Subtracting the two BSE equations (7.100) gives
→ (P ) (P )
← P → (P )
iα · ∇1 − 21 P α3 + mγ 0 ΦV =0 (x1 − x2 ) + ΦV =0 (x1 − x2 ) iα · ∇2 − 12 P α3 + mγ 0 = −2i ∂ z1 ΦV =0 (x1 − x2 )

E
(A.99)
Thus
Z n
(P ) (P )
−idξK0z |M, P i0 = dx1 dx2 ψ̄(x1 )eiP (z1 +z2 /2 1
E(z1 + z2 )ΦV =0 (x1 − x2 ) − 12 dξ α3 , ΦV =0 (x1 − x2 )

2 idξ
P (P )
o
+ dξ(z1 − z2 ) ∂z1 ΦV =0 (x1 − x2 ) ψ(x2 ) |0i (A.100)
E
Let us now assume that the frame dependence (7.101) holds, and show that it agrees with the change in the wave
function (A.100) caused by the infinitesimal boost. According to (7.101),
(P +dP ) (P =0)
ΦV =0 (x) = e−(ξ+dξ)α3 /2 ΦV =0 (xR )e(ξ+dξ)α3 /2
xR = (x, y, z cosh(ξ + dξ)) = (x, y, z cosh ξ) + (0, 0, dξz sinh ξ) (A.101)
The first term within the { } of (A.100) reflects the change in the plane wave phase of |M, P i0 ,
ei(P +dξE)(z1 +z2 )/2 = ei(P (z1 +z2 )/2 1 + 21 i dξE(z1 + z2 )

(A.102)
(P +dP )
The second term is due to the exp[∓(ξ + dξ)α3 /2] factors in ΦV =0 (x),
(P =0) (P =0)
exp[−(ξ + dξ)α3 /2]ΦV =0 (x1R − x2R ) = (1 − 12 dξα3 ) exp(−ξα3 /2)ΦV =0 (x1R − x2R ) (A.103)
(P =0)
and similarly for ΦV =0 (x1R − x2R ) exp[(ξ + dξ)α3 /2]. The third term in (A.100) relates to the Lorentz contraction,
i.e., the change in xR (A.101),
∂ (P =0) sinh ξ ∂ (P )
e−ξα3 /2 dξ(z1 − z2 ) sinh ξ ΦV =0 (x1R − x2R )eξα3 /2 = dξ(z1 − z2 ) Φ (x1 − x2 ) (A.104)
∂z1R cosh ξ ∂z1 V =0
This confirms that the frame dependence of the state |M, P i0 (A.94) implied by (A.101) agrees with the transformation
of a boost.
21. Show that Φ(P ) (τ ) given by (7.113) satisfies the BSE (7.99) at x⊥ = 0.
Denoting by B the lhs. of (7.99) at x⊥ = 0 when the wave function Φ(P ) (τ ) is given by (7.113),
→
eζα3 /2 B e−ζα3 /2 = eζα3 /2 i∇ · α − 21 (E − V + P α3 ) + mγ 0 e−ζα3 /2 Φ(0) (τ )

←
+ Φ(0) (τ )eζα3 /2 i∇ · α − 21 (E − V − P α3 ) − mγ 0 e−ζα3 /2

(A.105)
93
We need to show that B = 0. For the i∂z terms,

→ → →
eζα3 /2 i ∂ z α3 e−ζα3 /2 = i ∂ z α3 − 12 i( ∂ z ζ)
← ← →
eζα3 /2 i ∂ z α3 e−ζα3 /2 = i ∂ z α3 + 12 i( ∂ z ζ) (A.106)
√
The contributions ∝ ∂z ζ cancel in (A.105). Transforming ∂z = −2(E − V )∂τ = −2 V 0 τ cosh ζ ∂τ (which requires
both a linear potential and x⊥ = 0),
→ √ → √ →
i ∂ z α3 = −2 V 0 τ eζα3 α3 i ∂ τ + 2 V 0 τ sinh ζ i ∂ τ
← ← √ ← √
i ∂ z α3 = −2i ∂ τ α3 V 0 τ e−ζα3 − 2i ∂ τ V 0 τ sinh ζ (A.107)
√
The terms ∝ sinh ζ cancel in (A.105). Expressing E − V ± P α3 = V 0 τ exp(±ζα3 ), commuting the exp(±ζα3 /2)
factors using i∇⊥ · α⊥ exp(−ζα3 /2) = exp(ζα3 /2)i∇⊥ · α⊥ (since ∇⊥ ζ = 0) and similarly for the mγ 0 terms we get,
√ → → √
eζα3 /2 B e−ζα3 /2 = eζα3 − 2 V 0 τ i ∂ τ α3 + i∇⊥ · α⊥ − 21 V 0 τ + mγ 0 Φ(0) (τ )

← √ ← √
+ Φ(0) (τ ) − 2 i ∂ τ α3 V 0 τ + i∇⊥ · α⊥ − 21 V 0 τ − mγ 0 e−ζα3

(A.108)
The terms in [ ] depend only on τ , i.e., they are as in the ζ = 0 BSE of Φ(0) (τ ). Expressing exp(±ζα3 ) = cosh ζ ±
α3 sinh ζ the coefficent of cosh ζ is the rest frame BSE, which Φ(0) (τ ) satisfies by definition. Consequently the two
terms in [ ] give equal and opposite contributions to the ζ = 0 BSE. Using this leaves an anticommutator with α3 .
→
Transforming back τ → z at ζ = 0 allows to identify the h − operator (7.45),
√ → → √
eζα3 /2 B e−ζα3 /2 = sinh ζ α3 , − 2 V 0 τ i ∂ τ α3 + i∇⊥ · α⊥ − 12 V 0 τ + mγ 0 Φ(0) (τ )

(A.109)
h → i →
1
M − V ) + mγ 0 Φ(0) (0, 0, z) = 21 (M − V ) sinh ζ α3 , h − Φ(0) (0, 0, z)

= sinh ζ α3 , i∇ · α − 2
→
The expressions in (7.61), (7.72) and (7.81) show that α3 , h − Φ(0) (0, 0, z) = 0 for all wave functions. Hence B = 0
and Φ(P ) (τ ) given by (7.113) solves the BSE for all P at x⊥ = 0.
22. Prove the orthogonality relation (7.115) for states with wave functions satisfying the BSE (7.98).
I follow the proof presented in [48]. From the expression (7.97) for the states in terms of their wave functions,
Z
(P )†
hMB , P B |MA , P A i = dx1B dx2B dx1A dx2A h0|ψ † (x2B )e−iP B ·(x1B +x2B )/2 ΦB B (x1B − x2B )γ 0 ψ(x1B )
(P A )
× ψ̄(x1A )eiP A ·(x1A +x2A )/2 ΦA (x1A − x2A )ψ(x1A ) |0i (A.110)
R R
The field contractions set x1A = x1B ≡ x1 and x2A = x2B ≡ x2 . Then dx1 dx2 = d[(x1 + x2 )/2]d(x1 − x2 ) and
the integral over x1 + x2 sets P A = P B ≡ P ,
Z h i
(P )† (P )
hMB , P B |MA , P A i = dx1 dx2 ei(P A −P B )·(x1 +x2 )/2 Tr ΦB B (x1 − x2 )ΦA A (x1 − x2 )
Z h i
3 3 (P )† (P )
= (2π) δ (P A − P B ) dx Tr ΦB (x)ΦA (x) (A.111)
(P ) (P )†
The BSE (7.98) for ΦA and ΦB are
(P ) (P ) (P ) (P )
i∇ · α, ΦA (x) − 21 P · α, ΦA (x) + m γ 0 , ΦA (x) = EA − V (x) ΦA (x)

(P )† (P )† (P )† (P )†
−i∇ · α, ΦB (x) + 12 P · α, ΦB (x) − m γ 0 , ΦB (x) = EB − V (x) ΦB (x)

(A.112)
94
(P )† (P )
Multiplying the first equation by ΦB (x) from the left, the second by ΦA (x) from the right and taking the trace
of their difference the terms ∝ 12 P , m and V cancel, giving
h (P )† i h (P )† i
(P ) (P )
2iTr α · ∇ ΦB (x), ΦA (x) = EA − EB Tr ΦB (x)ΦA (x) (A.113)
R∞
Integrating both sides −∞ dx the lhs. vanishes due to the substitution at (one component of) x = ±∞. The vanishing
of the rhs. implies (for MA 6= MB ) the orthogonality of the states according to (A.111).
23. Verify the expression (7.118) for the global norm of Φ−+ (x) in terms of F1 (r).
According to (A.111) the bound state norm is proportional to the trace on the lhs. of (7.118). The expression (7.60)
of Φ−+ (x) implies
Z
N ≡ dx Tr Φ†−+ (x)Φ−+ (x)

← →
Z n
∗
h 2 ih 2 i o
= dr r2 dΩ Tr Yjλ (Ω)F1∗ (r)γ5 (−iα · ∇ + mγ 0 ) +1 (iα · ∇ + mγ 0 ) + 1 γ5 F1 (r)Yjλ (Ω)
M −V M −V
h← → 4m2
Z
∗ ∗ 4 i
=4 dr r2 dΩ Yjλ F1 ∂ j ∂ j + + 1 F1 Yjλ
(M − V )2 (M − V )2
→2 8V 0 4m2
Z
∗ ∗
h 4 i
=4 dr r2 dΩ Yjλ F1 − ∇ − ∂r + + 1 F1 Yjλ (A.114)
(M − V )2 (M − V )3 (M − V )2
→2
Expressing ∇ in spherical coordinates and using the radial equation (7.59),
→2 h 2 j(j + 1) i h V0 i
∇ F1 (r)Yjλ (Ω) = F100 + F10 − F 1 Yjλ = − F1
0
− 1
4 (M − V )2
F1 + m2
F1 Yjλ (A.115)
r r2 M −V
Substituting this into (A.114) and using dΩ |Yjλ (Ω)|2 = 1 gives (7.118).
R
24. Verify the expressions (7.161) for radial functions H1 (r), H2 (r) and H3 (r).
The structure (7.150) of the J P C = 0++ wave function,

Φσ (x) = H1 (r) + i α · x̂ H2 (r) + i γ 0 α · x̂ H3 (r) (A.116)
follows from its parity and charge conjugation quantum numbers, as listed in (7.55). The three Dirac structures
involving the orbital angular momentum operator L do not contribute for j = 0, since L acts on Y00 in (7.48), which
has no angular dependence. The BSE (7.148) involves the (anti)commutators
1
2 {α, Φσ (x)} = αH1 (r) + ix̂H2 (r) + γ 0 x̂ × α γ5 H3 (r)
1
0 0
2 γ , Φσ (x) = iγ α · x̂H2 (r) + iα · x̂H3 (r) (A.117)
Inserting the expression (A.116) into the BSE (7.148) with M = 0 gives, using ∇f (r) = x̂f 0 (r), ∂i xj = δij and
∇ · x̂ × α f (r) = 0,
2
i α · x̂H10 − H2 − H20 + imγ 0 α · x̂H2 + imα · x̂H3 + 12 V 0 r Φσ (x) = 0 (A.118)
r
The coefficients of the Dirac structures 1, iα · x̂ and iγ 0 α · x̂ impose
2
− H2 − H20 + 21 V 0 r H1 = 0
r
H10 + mH3 + 12 V 0 r H2 = 0
mH2 + 21 V 0 r H3 = 0 (A.119)
95
The first and third equations allow to express H1 and H3 , respectively, in terms of H2 . Substituting them into the
second equation gives the differential equation
1 h1 4i
H200 + H20 + (V 0 r)2 − m2 − 2 H2 = 0 (A.120)
r 4 r
It is straightforward to verify that the expressions for Hi (r) given in (7.161) satisfy the above equations. Their
properties H1 (r → 0) ∼ r0 , H2 (r → 0) ∼ r2 and H3 (r → 0) ∼ r1 ensure that the wave function is locally normalizable
at r = 0.
[1] P. Zyla et al. (Particle Data Group), PTEP 2020, 083C01 (2020).
[2] S. Aoki et al. (Flavour Lattice Averaging Group), Eur. Phys. J. C 80, 113 (2020), arXiv:1902.08191 [hep-lat].
[3] Strong QCD from Hadron Structure Experiments: Newport News, VA, USA, November 4-8, 2019 , Vol. 29 (2020)
arXiv:2006.06802 [hep-ph].
[4] A. S. Blum, Stud. Hist. Phil. Sci. B 60, 46 (2017), arXiv:2011.05908 [physics.hist-ph].
[5] G. T. Bodwin, D. R. Yennie, and M. A. Gregorio, Rev. Mod. Phys. 57, 723 (1985).
[6] A. A. Penin, Proceedings, 12th DESY Workshop on Elementary Particle Physics: Loops and Legs in Quantum Field Theory
(LL2014): Weimar, Germany, April 27-May 2, 2014, PoS LL2014, 074 (2014).
[7] G. S. Adkins, Hyperfine Interact. 233, 59 (2015).
[8] G. Adkins, J. Phys. Conf. Ser. 1138, 012005 (2018).
[9] E. Eichten, K. Gottfried, T. Kinoshita, K. D. Lane, and T.-M. Yan, Phys. Rev. D21, 203 (1980).
[10] E. Eichten, S. Godfrey, H. Mahlke, and J. L. Rosner, Rev. Mod. Phys. 80, 1161 (2008), arXiv:hep-ph/0701208 [hep-ph].
[11] G. ’t Hooft, Nucl. Phys. B Proc. Suppl. 121, 333 (2003), arXiv:hep-th/0207179.
[12] Y. L. Dokshitzer, in 2002 European School of high-energy physics, Pylos, Greece, 25 Aug-7 Sep 2002: Proceedings (2003)
pp. 1–33, arXiv:hep-ph/0306287 [hep-ph].
[13] Y. L. Dokshitzer and D. E. Kharzeev, Ann. Rev. Nucl. Part. Sci. 54, 487 (2004), arXiv:hep-ph/0404216.
[14] Y. Dokshitzer, Proceedings, Workshop on Critical examination of RHIC paradigms (CERP 2010): Austin, USA, April
14-17, 2010, PoS CERP2010, 001 (2010).
[15] P. Hoyer (2016) arXiv:1605.01532 [hep-ph].
[16] C. Itzykson and J. Zuber, Quantum Field Theory, International Series In Pure and Applied Physics (McGraw-Hill, New
York, 1980).
[17] E. E. Salpeter and H. A. Bethe, Phys. Rev. 84, 1232 (1951).
[18] G. P. Lepage, Two-body Bound States in Quantum Electrodynamics, Ph.D. thesis, SLAC-R-0212 (1978).
[19] N. Nakanishi, Prog. Theor. Phys. Suppl. 43, 1 (1969).
[20] V. Karmanov, J. Carbonell, and H. Sazdjian, PoS LC2019, 050 (2020), arXiv:2001.00401 [hep-ph].
[21] J. Carbonell and V. K. a. H. Sazdjian, (2021), arXiv:2101.03566 [hep-ph].
[22] W. E. Caswell and G. P. Lepage, Phys. Rev. A18, 810 (1978).
[23] S. J. Brodsky and J. R. Primack, Annals Phys. 52, 315 (1969).
[24] M. Järvinen, Phys. Rev. D71, 085006 (2005), arXiv:hep-ph/0411208 [hep-ph].
[25] T. Kinoshita and G. Lepage, “Quantum electrodynamics for nonrelativistic systems and high precision determinations of
alpha,” (1990) pp. 81–91.
[26] W. Caswell and G. Lepage, Phys. Lett. B 167, 437 (1986).
[27] T. Kinoshita, in International Workshop on Hadronic Atoms and Positronium in the Standard Model Dubna, Russia, May
26-31, 1998 (1998) arXiv:hep-ph/9808351 [hep-ph].
[28] T. Kinoshita and M. Nio, Phys. Rev. D53, 4909 (1996), arXiv:hep-ph/9512327 [hep-ph].
[29] K. Pachucki, Phys. Rev. A 56, 297 (1997).
[30] A. Czarnecki, K. Melnikov, and A. Yelkhovsky, Phys. Rev. A 59, 4316 (1999), arXiv:hep-ph/9901394.
[31] M. Haidar, Z.-X. Zhong, V. Korobov, and J.-P. Karr, Phys. Rev. A 101, 022501 (2020), arXiv:1911.03235 [physics.atom-
ph].
[32] G. T. Bodwin, E. Braaten, and G. Lepage, Phys. Rev. D 51, 1125 (1995), [Erratum: Phys.Rev.D 55, 5853 (1997)],
arXiv:hep-ph/9407339.
[33] N. Brambilla et al., Eur. Phys. J. C 71, 1534 (2011), arXiv:1010.5827 [hep-ph].
[34] J. S. Schwinger, Phys. Rev. 128, 2425 (1962).
[35] S. R. Coleman, R. Jackiw, and L. Susskind, Annals Phys. 93, 267 (1975).
[36] S. R. Coleman, Annals Phys. 101, 239 (1976).
[37] M. S. Plesset, Phys. Rev. 41, 278 (1932).
[38] O. Klein, Z. Phys. 53, 157 (1929).
[39] A. Hansen and F. Ravndal, Phys. Scripta 23, 1036 (1981).
[40] S. Weinberg, The Quantum theory of fields. Vol. 1: Foundations (Cambridge University Press, 2005).
[41] P. A. Dirac, Proc. Roy. Soc. Lond. A A117, 610 (1928).
[42] P. Dirac, Proc. Roy. Soc. Lond. A A118, 351 (1928).
[43] S. J. Brodsky, Atomic physics and astrophysics. Vol.1. Brandeis University Summer Institute in Theoretical Physics , 95
(1971), (SLAC-PUB-1010).
96
[44] F. Gross, Phys. Rev. C 26, 2203 (1982).

[45] A. Neghabian and W. Gloeckle, Can. J. Phys. 61, 85 (1983).
[46] J.-P. Blaizot and P. Hoyer, unpublished (2014).
[47] J.-P. Blaizot and G. Ripka, Quantum theory of finite systems (The MIT Press, 1985).
[48] D. D. Dietrich, P. Hoyer, and M. Järvinen, Phys. Rev. D87, 065021 (2013), arXiv:1212.4747 [hep-ph].
[49] X. Artru, Phys. Rev. D 29, 1279 (1984).
[50] M. Burkardt, Adv. Nucl. Phys. 23, 1 (1996), arXiv:hep-ph/9505259.
[51] S. J. Brodsky, H.-C. Pauli, and S. S. Pinsky, Phys. Rept. 301, 299 (1998), arXiv:hep-ph/9705477.
[52] J. Collins, (2018), arXiv:1801.03960 [hep-ph].
[53] X. Ji, Nucl. Phys. B, 115181 (2020), arXiv:2003.04478 [hep-ph].
[54] P. D. Mannheim, P. Lowdon, and S. J. Brodsky, (2020), arXiv:2005.00109 [hep-ph].
[55] F. L. Feinberg, Phys. Rev. D17, 2659 (1978).
[56] N. H. Christ and T. D. Lee, Phys. Rev. D22, 939 (1980).
[57] J. F. Willemsen, Phys. Rev. D17, 574 (1978).
[58] J. D. Bjorken, in Quantum chromodynamics: Proceedings, 7th SLAC Summer Institute on Particle Physics (SSI 79),
Stanford, Calif., 9-20 Jul 1979 (1979) p. 219.
[59] G. Leibbrandt, Rev. Mod. Phys. 59, 1067 (1987).
[60] F. Strocchi, Int. Ser. Monogr. Phys. 158, 1 (2013).
[61] V. N. Gribov, Nucl. Phys. B139, 1 (1978).
[62] M. E. Peskin and D. V. Schroeder, An Introduction to quantum field theory (Addison-Wesley, Reading, USA, 1995).
[63] P. Hoyer, Phys. Lett. B172, 101 (1986).
[64] D. D. Dietrich, P. Hoyer, and M. Järvinen, Phys. Rev. D85, 105016 (2012), arXiv:1202.0826 [hep-ph].
[65] P. Hoyer (2014) arXiv:1402.5005 [hep-ph].
[66] H. E. Haber, (2019), arXiv:1912.13302 [math-ph].
[67] D. A. Geffen and H. Suura, Phys. Rev. D16, 3305 (1977).
[68] W. Melnitchouk, R. Ent, and C. Keppel, Phys. Rept. 406, 127 (2005), arXiv:hep-ph/0501217.
[69] G. ’t Hooft, Nucl. Phys. B 72, 461 (1974).
[70] E. Witten, NATO Sci. Ser. B 59, 403 (1980).
[71] S. R. Coleman, in 17th International School of Subnuclear Physics: Pointlike Structures Inside and Outside Hadrons (1980)
p. 0011.
[72] G. ’t Hooft, Nucl. Phys. B 75, 461 (1974).
[73] P. Hoyer, (2018), arXiv:1807.05598v2 [hep-ph].
[74] M. Gell-Mann, R. Oakes, and B. Renner, Phys. Rev. 175, 2195 (1968).
[75] S. Okubo, Phys. Lett. 5, 165 (1963).
[76] G. Zweig, “An SU(3) model for strong interaction symmetry and its breaking. Version 2,” in DEVELOPMENTS IN THE
QUARK THEORY OF HADRONS. VOL. 1. 1964 - 1978, edited by D. Lichtenberg and S. P. Rosen (1964) pp. 22–101.
[77] J. Iizuka, Prog. Theor. Phys. Suppl. 37, 21 (1966).
[78] H. Harari, Phys. Rev. Lett. 22, 562 (1969).
[79] G. Zweig, Int. J. Mod. Phys. A 30, 1430073 (2015).

Journey To The Bound States

Uploaded by

Copyright:

Available Formats

Journey To The Bound States

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Journey To The Bound States

Uploaded by

Copyright:

Available Formats

Journey to the Bound Statesa

I. Motivations and Outline 4

II. Brief survey of present QED approaches to atoms 6

III. Dirac bound states 12

IV. Fock expansion of bound states in temporal (A0 = 0) gauge 21

a Expanded version of lectures presented at the University of Pavia in January 2020.

V. Applications to Positronium atoms 27

VI. QED in D = 1 + 1 dimensions 41

VII. Applications to QCD bound states 54

VIII. Bound state epilogue 79

I. MOTIVATIONS AND OUTLINE

A. Hadrons vs. atoms

II. BRIEF SURVEY OF PRESENT QED APPROACHES TO ATOMS

A. Recall: The Hydrogen atom in introductory Quantum Mechanics

We then transform to CM and relative coordinates,

kinetic energy of the CM motion and an O α2 binding energy, E = me + mp + P 2 /2(me + mp ) + Eb with,

B. The Schrödinger equation from Feynman diagrams (rest frame)

1. Bound states vs. Feynman diagrams

2. Forming an integral equation

C. The Bethe-Salpeter equation

that are only connected by S, we have the Dyson-Schwinger identity

Φ0 (P 0 , x01 − x02 ) = S(Λ)Φ(P, x1 − x2 )S −1 (Λ) (2.20)

III. DIRAC BOUND STATES

A. Weak vs. strong binding

B. The Dirac equation

The Dirac equation

corresponding to double scattering in the external potential.

The Dirac wave functions define eigenstates of the Dirac Hamiltonian,

Charge conjugation transforms the electron field as

Cψ † (t, x)C † = iψ T (t, x)γ 2 (3.14)

HD |Mn i = Mn |Mn i Mn > 0

In terms of the wave functions in momentum space,

Ψ†n (p) u(p, λ)bp,λ + v(−p, λ)d†−p,λ ≡ Bnp bp + Dnp d†p

Exercise A 4: Derive (3.23).

are normalized so that

D. * Dirac wave functions for central A0 potentials

(−iα · ∇ + mγ 0 )Ψ(x) = M − V (r) Ψ(x)

Operating on the states |M, jλi in (3.12) we get

The Dirac 4-spinor wave functions are thus required to satisfy

J 2 Ψjλ = j(j + 1)Ψjλ J z Ψjλ = λΨjλ (3.32)

The parity operator is defined by

Pψ † (t, x)P† = ψ † (t, −x)γ 0

γ 0 Ψjλ (−x) = ηP Ψjλ (x) (3.34)

(L + 12 σ)2 φjλ± = j(j + 1)φjλ± (3.37)

Exercise A 6: Derive the identities (3.39). Hint: Use αi αj = δ ij + iγ5 ijk αk .

We may furthermore use

At the potentially singular points M − V ± m = 0 the solutions behave as (M − V ± m = 0)β , with β = 0 or β = 2,

− ∇2 + m2 Ψ(x) = (M − V )2 + iα · (∇V ) Ψ(x)

E. * Coulomb potential V (r) = −α/r

Unbound states, M 2 − m2 > 0

F. * Linear potential V (r) = V 0 r

F (r → ∞) ∼ −iG(r → ∞) ∼ N rβ exp iV 0 r2 /2 r2 |F (r → ∞)|2 = r2 |G(r → ∞)|2 = N 2

e− (p, s) u† (p, s)Ψ(p)

IV. FOCK EXPANSION OF BOUND STATES IN TEMPORAL (A0 = 0) GAUGE

A. Definition of the bound state method

Instant time quantization

Temporal (A0 = 0) gauge

1. Functional integral method

[Aµ (t, x), πν (t, y)] = i gµν δ 3 (x − y) (4.9)

Exercise A 6: Derive the identities (3.39). Hint: Use αi αj = δ ij + iγ5 ijk αk .