Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

THE STRUCTURE OF THE GAUGE THEORY VACUUM*

C.G. CALLAN, Jr.

Joseph Henry Laboratories, Princeton University. Prinecton, New Jersey 08540, USA

R.F. DASHEN *
Institute for Advanced Study. Princeton, New Jersey 08540, USA

and

D.J. GROSS
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08540, USA

Received 20 May 1976

The finite action Euclidean solutions of gauge theories are shown to indicate the existence of tunneling between
topologically distinct vacuum configurations. Diagonalization of the Hamiltonian then leads to a continuum of vacua.
The construction and properties of these vacua are analyzed. In non-abelian theories of the strong interactions one
finds spontaneous symmetry breaking of axial baryon number without the generation of a Goldstone boson, a mech-
anism for chiral SU(N) symmetry breaking and a possible source of T violation.

Polyakov [1 ] has recently pointed out that the We have, in effect, shown that the vacuum "seizes" as
Euclidean classical equations o f motion o f gauge theo- suggested by Kogut and Susskind [2], and identified
ries have soliton-like solutions and has suggested that the mechanism b y which it does so. Our primary aim
when properly included in the Euclidean functional in this letter will be to give arguments for the existence
integral they may have a bearing on the dynamics o f o f the new vacuum structure and to present the correct
confinement. The physical interpretation o f these solu- form o f the functional integral appropriate to studying
tions has, however, been obscure since they are local- the properties o f a particular vacuum. In the spirit o f
ized in time as well as space. In this letter we shall show displaying qualitative consequence o f the new vacuum
that Euclidean gauge solitons describe events in which structure we shall also briefly summarize results obtained
topologically distinct realizations o f the gauge vacuum from rather crude approximations to the functional
tunnel into one another and that this process radically integral.
changes the nature o f the vacuum state. In fact, we To explore the structure o f the vacuum we study
find a continuum o f vacua, each one o f which is a the Euclidean functional integral
superposition o f the vacua with difinite topology and
stable under the tunnelling process. The new vacua are
the ground states o f independent, and in general, in-
(0lexp(-Ht)10) ~
t--~
f[DAuD¢]
equivalent worlds (most striking, P and T are spontan-
eously violated in some o f them!). When massless x exp -fddx[ .e(A., ÷ -~gf] ) (1)
fermions are present, the vacuum tunnelling process
forces a redefinition o f the fermion vacuum as well and where d is the dimension of space time, £ is the
leads directly to spontaneous breakdown o f chiral in- Langrange density of the theory, ~Ogfis a gauge-fixing
variance without generating a " n i n t h " Goldstone boson. term and the integration is to be done over all fields
that approach vacuum values ~Fuv = 0) at infinity. Now
since Fur = 0 implies A~ = g - " (x)aug(x) takes on
* Research supported by the National Science Foundation
under Grant Number MPS 75-22514. values in the gauge group, G, any gauge field included in
* Research sponsored in part by the ERDA under Grant No. the functional integration defines a map o f the sphere
E(11-1)-2220. at Euclidean infinity into G. As pointed out b y

334
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

Belavin et al. [3] these maps fall into h o m o t o p y functional integrals corresponding to different v-classes,
classes corresponding to elements of the homotopy is then straightforward:
group, Ild_l(G). For most non-Abelian groups in four
dimensional space time and U(1) in two dimensional
spacetime, this h o m o p o t y group is Z. In these theories
(nlexp(-Ht)lm) tZ~ f [DAu...] (n-m)
(4)
(the only ones we consider) the gauge fields integrated
X exp(--fddx[.~(Au) + d~gf] } .
over in eq. (1) fall into discrete classes indexed by an
integer v running from - oo to + oo (we shall use the
notation [DAu] v to denote functional integration over The functional integral over h o m o t o p y class v describes
the vth class). Thus there is actually a discrete infinity a vacuum-to-vacuum transition in which the vacuum
of funtional integrals and one must ask which, if any, winding number changes by v! Now the minimum action
is the "right" one. for v = 0 is zero (corresponding to A u = 0) so that in the
One can clearly see what is going on by working in WKB sense the In) ~ In) amplitude is O(1). In the v 4 : 0
the gauge A 0 = 0 and requiring Fur to vanish outside sectors the minimum action is in general non-zero - for
a large, but finite, spacetime volume, V (this boundary v = 1 in four dimensions it corresponds to the Belavin
condition is, of course, gauge invariant). The dynamical et al. instanton [3], whose action is 8rr2/g 2. Thus in the
variables are now just the space components, A i, for the same WKB sense the In) ~ In + 1) amplitude is
vector potential and at large negative and positive times O(exp(_8n2/g2). This is a typical "tunnelling" amplitude,
they must take on time independent vacuum values, vanishing exponentially for small coupling and unseen
Ai(x ) = g - 1 (x)aig(x). The topological quantum number, by standard perturbation theory. Indeed, perturbation
v, associated with any particular Euclidean gauge field treatments of gauge theories expand about Au = 0 and
time history may be written as a gauge invariant volume pretend that the vacuum in = 0) is true vacuum. Because
integral of vacuum tunnelling, this is completely wrong and
causes perturbation theory to miss qualitatively sig-
v= 1 fd4x tr (Fu~puv), d =4 nificant effects.
87r2 What then is the true vacuum? A convenient way of
(2)
constructing it is to consider the generators o f time
-~rfd2xeu~F u~ , d=2.
independent gauge transformations characterized by a
gauge function xa(x):
In both cases, the integrand is a total divergence and
A 0 = 0 gauge v may be rewritten as a surface integral, Qx = f d d - l x [ F ~ i D i Xa +g ./ vl
v = n(t = + oo) _ n(t = - oo), where
where D i is the covariant derivative and g~ is the gauge
n = _ _1 e i i k f d 3 x tr (A i A / A k ) , d = 4 source of fields other than the gauge field itself. In
67r2 order to satisfy Gauss' law, DiF~i = gJ~, it is sufficient
(3)
to restrict the state space by Qxt~) = 0 for all gauge
=~-~fdxA 1 , d= 2. functions ~, which vanish at infinity. In particular, all
our vacuum states In) are annihilated by such local
With no loss of generality (we have the freedom o f gauge transformations. There also exist gauge functions
making time independent gauge transformations) we which do not vanish at infinity and generate gauge
may choose n(t = - o0) to be an integer. Then since transformations, T, which change the vacuum topology.
v is integral the gauge vacuum configuration at t = - oo One can easily construct a unitary T effecting such a
must also have integral winding number n(+ oo) = non-local gauge transformation: T = exp(iG=.), with
n(--**) + ~. Go* (21r/g)[E(oo) + E ( - oo)] for the two-dimensional
Therefore we must admit the existence of a dis- abelian theory or Go. = (2n/g) f d2S.tEa~ a for the four-
crete infinity of vacuum states, In), labelled by a wind- dimensional non-Abelian theory, and T satisfies Tin) =
ing number taking on integral values from - oo to + ~. In + 1).
The interpretation of the multiplicity of Euclidean Since T is a gauge transformation, the hamiltonian

335
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

commutes with it and energy eigenstates must be T zero field of flux +2n/g, radius/a -1 (/.t is the heavy
eigenstates. Since T is unitary, its eigenvalues are e i° , photon mass), arbitrary location and total action, SO,
0 ~< 0 ~< 27r, and the eigenstates are [0) = ~,ein° [n). proportional to la2/g 2 . We shall construct the sectors
This diagonalization of H is obviously unaffected by with topological quantum number v by superposing
including in Z? sources coupled to gauge invariant den- n+ v = + 1 vortices and n_ v = - 1 vortices with
sities. Thus, each [0) vacuum is the ground state of an n+ - n_ = v, neglecting any interactions between vor-
independent and in general physically inequivalent tices (since fields decrease exponentially this is not too
sector within which we may study the propagation of bad for low vortex density, which turns out to mean
gauge invariant disturbances. Since the different 0- small g). In this "dilute gas" approximation, the func-
worlds do not communicate with each other, there is tional integral is
no a-priori way of deciding which world is the right
one. It is gratifying that this multiplicity of worlds is
(O'[exp(-Ht)[O) ~ 5(0% O) ~ exp ( - ( n + + n_)Si~
known to exist in the Schwinger model, correspond- n+,n_=0
ing there to different values of background electric
field [4]. xexp(iO(,+-,_)) ( V__V_]"÷+n-
Finally, we must express the functional integral, n+ ! n_ ! \ Vo! (6)
eq. (4), in 0 basis:
where the factors of V come from integrating over
(0'1 exp(-Ht)lO) .... ~ 6(0 - 0')1(0) vortex locations and V0 is a normalization factor which
t..~ ~
can be calculated from the quantum corrections to this
basically semiclassical approximation. The sum is
I(O) = ~ exp ( - i v 0 )
1) trivial and yields exp [2(V/Vo)e -S° cos 0). We have
normalized the energy so that the naive perturbation
×f[DA u ...j~ exp ( - f d d x [ ~?(A u ...) + -~gft ) theory vacuum energy is zero. By contrast, the 0 vacua
have an energy per unit volume equal to - 2 V ~ -1 cos 0
= f [DAu'"J exp{- f ddx[&?gf + ~?o]) (5) e -S° . Because S O cc 1]g2, this energy difference is a
non-perturbative effect (a tunnelling effect) but poten-
where ./2 o = (i0/8rr 2) tr (Fur ffuV)for d = 4, tially important nonetheless. Mthough the 0 = 0
~o 0 = (i0/47r) eu~Fuv for d = 2 and in the second ex- vacuum has lowest energy (and no parity violation) we
pression for I(O) all gauge field topologies are summed can't conclude that it is the vacuum since the other
over. I(O) contains all possible information about 0-vacua, though higher in energy are stable to gauge
physics in the 0-world and requires no further modifi- invariant perturbations. Having constructed a vacuum
cation. The second form for I(0) makes manifest one one can then calculate Green' functions of gauge in-
of the peculiar ways in which the 0-worlds differ from variant operators perturbatively. In the path integral
one another. In four dimensional pure Yang-Mills this corresponds to performing ordinary perturbation
theory, re-expressed in Minkowski coordinates, the theory about the appropriate classical solution for
effective Lagrangian is tr [FuuFU~ + (O/8n2F,,vFUV]. each topologically distinct sector and summing.
This clearly breaks P and T invariance (except for If we try the above sort of approximation on the
0 = 0) and we must in general expect spontaneous non-abelian theory in four dimensions, there is a pro-
breaking o f space-time symmetries in all but a few blem. The classical theory is scale invariant and the basic
special 0-worlds! v = 1 solution (instanton) has an arbitrary scale param-
As a concrete illustration of these general remarks eter, ~, as well as an arbitrary position. The integration
we should like to present an approximate evaluation over ~. need not diverge since scale invariance is broken
of I ( 0 ) in two-dimensional charged scalar electro- by quantum corrections. Indeed, the renormalization
dynamics. In the sector with v = -+ 1 the field config- group should tell us whether the integral converges at
uration with minimum action is just the Nielson-Olesen the short distance end. In the dilute gas approxima-
vortex [5] in which there is a localized region of non- tion one finds for the vacuum energy density

336
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

In fact, for massless quarks, (n le-Ht lm) o: fnm !

o k g(X/u)z (7; The reason for this is that, because of the anomaly,
the conserved axial charge
where g(),]~) is the usual effective coupling, normaliz-
ed so that g (h =/a) = g, ~ is arbitrary, and ~summari-
zing the effect of loop corrections, can be computed
perturbatively. If the theory is asymptotically free 5_
and there are not too many quark multiplets, g can flavor,color
vanish rapidly enough for the integral to converge in
the limit of large )~ (small instanton size). This condi- _tr{g2NeuvxoAv(bkAa+{AxAo)}32n
2- (8)
tion is met for any pure SU(N) gauge theory and for
SU(3) with no more than ten flavors of quark. while invariant under local gauge transformations, is
On the other hand, in the limit of large instanton not invariant under global gauge transformations. In
size, one is driven to large coupling (unless ~ has a particular, one has TQ5 T -1 = Q5 - 2N where T is
small infrared fixed point) and the dilute gas approxi- the global gauge transformation, introduced earlier,
mation breaks down (instanton overlap and have long which changes gauge field winding number by one
range interactions). Thus the attempt to construct the unit and N is the number of flavors. If the vacuum
vacuum may run into an essential strong coupling pro- states of different topology are defined by In) = T n 10),
blem because the quantum corrections to vacuum with Q510) = 0 one finds that Q5 In) = 2N. nln). How-
tunnelling will be large for large instanton size. In ever, Q5 is conserved, so that it must be true that
fact, there may not be a sensible way of perturbatively (nle-Htlm) ~" fin m" In general, we must find
calculating even Green's functions of gauge invariant (nl e-l-KDvlm) ~ '6n_m, v where D v is any operator of
operators, no matter how small one makes g. This chirality 2Nv (Dv may stand for multiple insertions of
phenomenon is typical of a theory with no inherent local operators at different times - all that matters is
mass scale which produces masses dynamically. If one net chirality). Therefore, we may replace eq. (4) by
sets the renormalization mass scale, ~, equal to some (nl e-Htl m)
physical mass (e.g. 4 x/~-0 ), then g is determined (di-
mensional transmutati6n) and typically of order 1.
These problems should not, however, affect the
-+fnm f [DAu.'.]Oexp{- f dax[~(A~,...) +~f] )
standard applications of asymptotic freedom which and
rely on one's ability to compute operator product ex-
pansion coefficients at short distances. Precisely be- (nle-HtDulm) -~ fn+v,m f [DA~,...]~O,
cause of asymptotic freedom, vacuum tunnelling is
suppressed at arbitrarily small scales and leading short × exp (-fdax [~(X....) + ~gfl )
distance behavior will agree with conventional cal-
(10)
culations. There will, however, be calculated non-
leading terms suppressed by powers of momentum, with the same meaning still attached to D v. The restric-
which reflect the mass scale set non-perturbatively by tion on the topology of the gauge field histories would
the tunnelling phenomenon. actually have emerged directly from a mindless appli-
The arguments presented above require some modi- cation of eq. (4): Doing the fermion integrations for
fication when massless fermions are present. We again fixed A u yields [det(~ - ~ ) ] +1. This determinant
confine non-zero Fur to a large but finite space-time vanishes whenever (~f - 4/) has a zero eigenvalue.
volume, V, and again encounter a discrete infinity, 't Hooft [6] has noted that ifA u is taken equal to the
(In)}, of vacuum states characterized by a vacuum v = +1 or v = -1. instanton there is a zero eigenvalue,
gauge field with winding number n and a standard and our argument is just telling us that whenever Au
fermi vacuum with all negative energy states filled. In belongs to a v 4:0 class, (j~ - 4~) has a zero eigenvalue,
principle we must allow for transitions between eliminating the v 4:0 sectors from the integration.
vacuum, and evaluate (n[ e-Ittlm) for general n and m. Though vacuum tunnelling is now suppressed, the

337
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

In) vacua are not acceptable because they violate "vacuum seizing" mechanism postulated by Kogut
cluster decomposition for operators of non-zero and Susskind [2] while the non-vanishing of F F in
chirality. Consider an operator D of chirality 2N. The instanton solutions as a possible escape from the U(I)
arguments of the preceeding paragraph show that problem was noted by G. 't Hooft [7].
(nlD ln) = O, (n + llDln)4: 0. Then (n lD+(x)DO,)ln) Although the presence of zero mass fermions sup-
will not vanish for large Ix - y I as required by cluster presses vacuum tunnelling in the strict asymptotic sense,
decomposition and the vanishing of the "vacuum" ex- tunnelling does have a profound effect on the vacuum
pectation (n ID In): it obviously approached energy and other physically relevant quantities. When
(n ID+ln + 1)(n + 1 ID In). The solution to this pro- the vacuum tunnels, fermion pairs are produced. Al-
blem is obvious (it was solved in the Schwinger model though the pair must ultimately be absorbed by an
years ago!): The proper vacuum states are the 10) anti-tunnelling, since the fermions are massless the
vacua, in which basis the functional integrals have the pair may live for a long time and tunnelling occurs
form freely in intermediate states. To get some notion of
what goes on it is instructive to attempt a crude cal-
(O'[e-Ht[o) -, ~(o'- o) f [DA,...](o) culation of the basic functional integrals of eq. (11)
in the case of a single flavor.
X exp{-fdax[2(A,...) + 2gf] } We shall assume that the integral over Au is domi-
nated by configurations of widely separated instantons
(O'[e-HtDvlO) -~ 8 ( 0 ' - o)f [DA....](~) (n+ in number) and anti-instantons (n_ in number). To
compute the vacuum energy we must set n+ = n_ (con-
X exp{- fdax [fl(Au...) + ~gf] }D~. (11) figurations with v = 0), sum over n+ and integrate over
instanton locations. We will ignore the integration over
The cluster problem is resolved by the non-vanishing instanton sizes. For a given gauge field configuration,
vacuum expectation value of D in the true vacuum the integration over fermi fields yields det(~ - j ) . This
state. The fact that only one topological class of gauge determinant must also be approximated.
field history contributes to each functional integral Now, as 't Hooft has pointed out [7], individual
makes physical quantities have a trivial dependence on instantons have a zero-energy eigenfunction C0 ±(x, x±)
0: The vacuum energy density, while non-zero, is in- (x± is the instanton location and the +- label distinguishes
dependent of 0. The variation of the vacuum energy instanton from anti-instanton). Since the interesting
with respect to 0 is just (tr FuvFUV), the quantity physical effects arise precisely from these zero energy
whose non-zero value is the signal for P and T viola- solutions, we shall compute the determinant of (~ - 41)
tion. In the massless fermion case, P and T appear not in the subspace spanned by the 2n+ functions
to be spontaneously violated and, indeed, all the 10) C0 +(x, xi+), CO (x, xt:-). In the widely separated in-
vacua are physically equivalent. Finally, the axial stanton approximation, these functions are orthonor-
baryon number invariance of the original Lagrangian real and one has to compute the determinant of the
is violated by a vacuum expectation value of operators matrix
with non-zero chirality. It should be said that at this
M,j = (c~,;I 0 - ~)1%, j).
stage we have in the N-flavor case, only broken axial
U(I) and not axial SU(N). Actually, since the axial In this approximation A~ differs from a gauge trans-
charge rotates 0, a discrete subgroup of order 2N of Formation only in the neighborhood of each instanton
U(1) is left unbroken, consisting of those elements and Mii can be approximated by
which rotate 0 by a multiple of 27r. There is no as-
M / / ~ ( ~+0, i1(- ~)-ll~b~,/)
sociated Goldstone boson because the conserved, but
gauge-variant, U(1) charge takes one out of a given where q~0 = ~C0. The 75 structure of the C0 forbids
10) sector (e i~Q s[0)= l0 + 2Na)) while tr(F_~), the i and j to be both instantons or both anti-instantons.
divergence of the gauge invariant axial current, has The C0 fall off at large distances.exactly like the free
non-vanishing matrix elements. That Q5 causes transi- fermi propagator, which is why we introduce the S0 's
tions between different vacua is characteristic of the which are localized as well as the instanton itself. The

338
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

cycle expansion of the determinant then gives a sum of different helicity generated by the instanton pro-
of terms with a graphical interpretation in terms of vide new ways of identifying sums of graphs which
closed fermion loops. For instance, if n+ = n_ = 1 we can lead to the desired symmetry breakdown. We have
get constructed a simple Hartree-Fock type argument for
N = 2 which has a chance of being correct in a weak
Det ~ (~b~](~)-1 [q5~)(O~ t (~) -1 [q5~),
coupling theory and which seems, on superificial ex-
which has the obvious interpretation of massless amination, to generate quark masses. The inevitable
quark propagators connecting non-local vertices, Goldstone bosons arise in this case from iterated bubble
V + * P graphs generated by the four-fermion interactions, IT_+.
= ¢0+(x)¢0±(x ), associated with the instantions.
The functional integral weights each vertex with a fac- We do not wish to make too much of these crude
tor proportional to e -$cl, where Scl is the instanton arguments other than to suggest that the new inter-
classical action. The 75 structure of ¢~ is such that actions generated by vacuum tunnelling are likely to
V ± is like ~(1 + 75)~ in its Dirac matrix structure, play a key role in generation of quark masses and
which is to say that V ± looks like a non.local, or Goldstone bosons.
momentum-dependent mass term. Summing over num- In terms of the picture presented here Polyakov's
bers and locations of instantons simply completes the ideas about confinement appear as follows. For an
vacuum fermion loop analogy by providing all pos- isolated quark located at x the tunnelling amplitude
sible insertions of the pseudo mass terms, V +, on mass- In, x) --* In + 1, x) will be reduced relative to the vacuum
less quark loops. Then calculations of physical quanti- to vacuum amplitude. A quark state will then have
ties proceed in a perfectly conventional way so long as more energy than the vacuum, as it should. In the
we remember to add the mass term ~ e-Scl(V+ + V_) dilute gas approximation the energy difference is pro-
to the massless quark propagator. Anything which portional to the integral over all instantons which over-
directly depends on the mass term, such as the vacuum lap the quark. Integrating over instanton locations and
expectation of t~k, will be proportional to e -scl and then over the scale size X-1 leads to an integral which
will have only the by now familiar dependence on g tends to diverge at small ),. For the large instantons
characteristic of a tunnelling process. In the "scale (small X), however, the dilute gas approximation is
invariant" four dimensional theory {t~ff) q: 0 implies not valid, and one is again confronted with a strong
spontaneous generation of mass and dimensional trans- coupling problem.
mutation as before. Obviously, much remains to be done to fully ex-
If the fermion is given a bare mass tunnelling is ploit the phenomena we have found. The major dif-
allowed and one is driven directly to a 0 vacuum ficulty, of course, is that the theory we are really in-
(whose energy depends on 0). The limit of zero mass terested in, quantum chromodynamics, is basically a
is smooth. If the bare mass is small compared to the strong coupling theory and reliable calculations are
spontaneously generated mass, it acts as a small per- difficult, if not impossible. However, one may hope that
turbation on the m 0 = 0 theory. a new understanding of the qualitative physics will
If there are N flavors, the above discussion is modi- suggest new methods of calculation. We are especially
fied in an important way: the effective instanton- encouraged by the appearance, already in semiclassical
quark interaction is no longer billinear, but 2N-linear. approximations, of a vacuum that breaks chiral sym-
Indeed, the instanton (anti-instanton) vertex has the metry and sets a dynamical mass scale. We are also
structure V+ = HNI~i(1 + 75)1~i (V_ = I I N I ~ i ( I -- 3'5) intrigued by the natural appearance of spontaneous
~bi). As a result, summing over instantons in the dilute violation of P and T invariance but have so far not
gas approximation will not just produce a mass term in seen how to understand why these effects are small in
the quark propagator, but does something more com- the real world or how to exploit them to explain obser-
plicated. To produce a quark mass term, one must ved violations of these symmetries. Perhaps super-uni-
break the global chiral SU(N), while we have argued fied theories will shed some light on these questions.
that the vacuum tunnelling phenomenon is only guar-
anteed to violate the chiral U(1) symmetry. On the One of us (D.J. Gross) would like to acknowledge
other hand, the effective interactions between quarks V.N. Gribov for stimulating conversations and in par-

339
Volume 63B, number 3 PHYSICS LETTERS 2 August 1976

ticular for the suggestion that the Euclidean solitons [3] A.A. Belavin, A.M. Polyakov, A.S. Schwartz and Yu. S.
might be relevant to the structure of the vacuum. Tyupkin, Phys. Lett. 59B (1975) 85.
[4] J. Lowenstein and A. Swieca, Ann. Phys. 68 (1971) 172.
S. Coleman, Harvard preprint (1975).
[51 H. Nielsen and P. Olesen, Nucl. Phys. B61 (1973) 45.
References [6] T.S. Bell and R. Jackiw, Nuovo Cimento 60 (1969) 47,
S. Adler, Phys. Rev. 177 (1969) 2426.
[1] A.M. Polyakov, Phys. Lett. 59B (1975) 82. [7] G. 't Hooft, Harvard preprint (1976).
[2] J. Kogut and L. Susskind, Phys. Rev. D l l (1975) 3594.

340