Dyson-Schwinger Equations in QCD
Craig D. Rob erts
Physics Division, 203, Argonne National Lab oratory,
Argonne, Illinois 60439-4843, USA
Abstract. These three lectures describ e the nonp erturbative derivation of Dyson-
Schwinger equations in quantum eld theory, their renormalisati on and, using the
exp edient of a simple one-parameter mo del, their application to the calculation of
hadronic observables and the prop erties of QCD at nite temp erature.
1 The Dyson-Schwinger Equations
The Dyson-Schwinger equations [DSEs] provide a nonp erturbative, Poincar e in-
variant approach to solving quantum eld theory, in which the fundamental
elements are the Schwinger functions. The Schwinger functions are moments of
the measure. For example, in a Euclidean quantum eld theory describing a self-
interacting scalar eld, x, sp eci ed by a measure d[], which will involve the
classical Euclidean action for the theory and, p erhaps, gauge xing terms, etc.,
then
Z
hx :::x i d[] x :::x ; 1
1 n 1 n
R
1
where d[] represents a functional integral, is an n-p ointSchwinger function.
In a manner analogous to that when a probability measure is involved, a quantum
eld theory is completely sp eci ed if all of the moments of its measure are known.
The fo cus of lattice eld theory is a numerical estimation of these Schwinger
functions. The DSEs provide a continuum framework for their calculation.
The DSEs are an in nite tower of coupled equations, with the equation for
a given n-p oint function involving at least one m>n-p oint function. A tractable
problem is only obtained if one truncates the system. Truncations that preserve
the global symmetries of a theory; for example, chiral symmetry in QCD, are
relatively simple to e ect [Bender, et al. 1996]. It is more dicult to preserve
lo cal gauge symmetries, although much progress in this direction has b een made
in Ab elian gauge theories Bashir and Pennington, 1994.
One systematic means of truncating the system is a weak coupling expansion.
In this way one readily nds that the DSEs contain p erturbation theory in the
sense that, for a given theory, the weak coupling expansion of the equations
generates all of the diagrams obtainable in p erturbation theory. In this way,
at the very least, the DSEs can b e used as a generating-to ol for p erturbation
1
An intro duction to the functional integral formulation of quantum eld theory can
b e found in Itzykson and Zub er 1980 and Rivers 1987.
Dyson-Schwinger Equations in QCD 213
Σ D = γ
S Γ
Fig. 1. A diagrammatic representation of the DSE for a fermion self-energy, p
2 2
i p[Ap 1] + B p : S p=1=[i p + p] is the connected fermion 2-p oint
function; D k is the connected gauge b oson 2-p oint function; and p; k is the
1-particle irreducible fermion{gauge-b oson 3-p oint function. Both D k and p; k
satisfy their own DSEs and this illustrates the coupling of the equation for a given
n-p oint function to those involving m>n-p oint functions.
theory. This can also b e used as a constraint on alternative truncation schemes;
i.e., they must b e such as to preserve the feature that p erturbative results are
recovered in the weak coupling limit.
There are many familiar examples of DSEs. For example, the gap equation
that describ es Co op er pairing in ordinary sup erconductivity is simply a trun-
cated DSE for a 2-p oint electron Schwinger function; Bethe-Salp eter equations
[BSEs], which describ e relativistic two-b o dy b ound states, are DSEs for 4-p oint
functions; and covariantFadde'ev equations, which describ e relativistic three-
b o dy b ound states, are DSEs for 6-p oint functions.
The statement that a theory is solved if all of its Schwinger functions are
known can b e lo osely re-expressed as the statement that all observable S -matrix
amplitudes can b e expressed in terms of the Schwinger functions of the elemen-
tary elds in the theory. This entails that one can connect observables to the
fundamental parameters of the theory via these Schwinger functions.
The 2-p oint functions in a given theory contain imp ortant information. For
example: in a gauge theory, the form and analytic prop erties of the gauge-b oson
2-p oint function can provide information ab out whether, due to interactions,
the gauge b oson acquires a gauge invariant mass that screens the interaction
Schwinger-mass generation or a strong enhancement at small momenta that
can b e a signal of con nement. Either of these prop erties will in uence the
propagation characteristics of other mo des in the theory and hence physical
observables. In all gauge theories the fermions act as a source of the gauge eld
and their propagation characteristics are strongly a ected by their interaction
with their self-generated gauge eld. This is describ ed by the DSE for the fermion
2-p oint function, illustrated in Fig. 1, in which the gauge-b oson 2-p oint function
app ears as a driving term.
For example, in QCD, whether the gauge-b oson 2-p oint function is nite
or strongly enhanced at small momenta determines whether chiral symmetry is
214 Craig D. Rob erts
dynamically broken and/or whether quarks are con ned.
Euclidean Metric. Herein I employ a Euclidean metric formulation of eld
theory; i.e., a non-negative metric for real vectors:
4
X
a b a b a b 2
i i
i=1
2
where is the Kronecker-delta. In this case, Q is a spacelikevector if Q > 0.
My Dirac matrices satisfy
y
3
[ ] = ; f ; g =2
and . One realisation of this algebra is provided by
5 4 1 2 3
E 0 E j
; i ;j=1;2;3 4
4 i
j
where are the usual contravariant Dirac matrices in Minkowski space.
I adopt the p oint of view that the Euclidean formulation is primary; i.e.,
that a eld theory should b e de ned in Euclidean space, which is the p ersp ective
employed in constructive eld theory and, usually as a pragmatic arti ce, in the
lattice formulation and numerical simulation of eld theories. The Schwinger
functions can then b e calculated and the question of the existence of the Wight-
man functions, and hence the Minkowski space propagators, addressed subse-
quently. [A fuller discussion of these p oints can b e found in Sec. 2.3 of Rob erts
and Williams 1994.]
This is imp ortant b ecause the analytic structure of a nonp erturbatively
dressed Schwinger function is not necessarily the same as that of its p erturba-
tive seed. Given this, one cannot know a priori the singularities in the integrand
of the integral equation. Hence the true consequences of rotating the momen-
tum space integration contour, as one do es in a Wick rotation, are unclear;
i.e., the correct form of the \Wick rotated" equation mayinvolve contributions
from p oles, branch cuts, etc., that cannot b e anticipated based on the p ertur-
bative form of the Schwinger functions involved. To elucidate this, the following
Euclidean , Minkowski Transcription Rules are valid at each order in p erturba-
tion theory:
Momentum Space Con guration Space
Z Z Z Z
M E
M E
4 M 4 E 4 M 4 E
d k ! i d k 1. d x ! i dx 1.
E E
E E
2. =k ! i k 2. =@ ! i @
E E
E E
3. k q ! k q
3. A= ! i A
E E E E
4. k x ! k x , 4. A B ! A B
by which I mean that the correct Minkowski space integral for a given diagram
in p erturbation theory is obtained by applying these transcription rules to the
Euclidean integral. However, for skeleton diagrams; i.e., those in which each line
and vertex represents a fully dressed n-p oint function, this cannot b e guaranteed.
Dyson-Schwinger Equations in QCD 215
1.1 Quantum Electro dynamics in d-Dimensions, QED
d
I will use QED to illustrate the nonp erturbative derivation and form of the
d
DSEs. The generating functional or partition function for QED is
d
E E E E
Z [ ; ;J ]= 5
Z Z
E E E E d fE fE fE fE E E
d ; ;A exp ; d x + + A J
E E E E fE
where there is an implicit normalisation Z [ =0; =0;J =0]=1; ,
fE E
, J are auxiliary source elds;
E E E E
d ; ;A 6
Y Y Y