You are on page 1of 16

Nuclear Physics B236 (1984) 381-396

© North-Holland Pubhshlng Company

COMPLETELY SOLVABLE GAUGE-FIELD EQUATIONS IN


DIMENSION GREATER THAN FOUR

R S WARD
Department of Mathematlcal Sctemes, Durham Unwersltv, Durham, England

Received 29 July 1983


(Revised 24 November 1983)

This paper deals with analogues, m higher damensaons, of the self-dual Yang-Malls equations
m dim4 Requanng that they be the lntegrablhty condmons for a linear system leads to a
procedure for constructing thexr solutions A partial classlfieaaon of such equations ~s gaven, and
three examples m dimension eight are studied in detail

1. Introduction

The self-dual Yang-Mills equations 2'~B--l~~ccB = F~ in four dimensions have many


remarkable properties, not least of which being that they are, in some sense,
"completely solvable". One possible definition of this property will be described in
sect. 2; for the moment, let us note the following pieces of evidence.
(i) Many explicit solutions, for example of instanton or monopole type, are
known; there are constructions winch ymld all solutions of these types [1].
(n) There is a sequence of ans~itze which generate solutions of the (non-linear)
self-duality equations in terms of arbitrary solutions of linear equations [2].
(iii) There exist non-local symmetry transformations which generate loop algebrae
[31.
The self-duality equations imply the Yang-Mllls equations, which, it is widely
believed, are n o t completely solvable [4].
The purpose of this paper is to describe a way of constructing higher-dimensional
completely-solvable analogues of the self-duality equations. Each of these takes the
form of a linear algebraic condition on F,,, and therefore amounts to a system of
first-order semi-linear equations for the gauge potential A,.
In recent years, there has been interest in gauge theories in dimensions greater
than four, particularly because of the way in which Higgs fields and supersymmetry
can be understood through dimensional reduction from d > 4 dimensions down to
d = 4 [5]. Another area in which higher dimensions are relevant is supergravlty and
spontaneous compactlfication [6]. The ideas of this paper can be extended to
higher-dimensional curved space-times; in particular, there are "completely solvable"
381
382 R.S Ward / Completely solvable gauge-field equations

analogues of the self-dual Einstein equations [7] m arbitrarily high dimension. This
latter subject will not be discussed here, but will appear separately.
A recent paper [8] has studied a similar problem to that of this paper, namely the
finding of linear algebraic con&tions on F,, such that the Yang-Mills equations
follow automatically (as is the case for the self-duahty equations m d = 4). The
authors gave a partial classification of the cases that occur in dimensions d ~ 8, but
did not ~dentify any completely solvable equations among the higher-dimensional
examples that they discovered.
The plan of this paper is as follows. In sect. 2 we shall study overdetermlned
systems of linear equations, involving one or more complex parameters. The integra-
bihty condition for such a system is a hnear algebraic condmon on the gauge field
(curvature) F,~, of the type we are interested m. This approach leads to a construc-
tion for solutions of the non-linear equations, generahzmg the "twister theory"
construction for solutions of the self-duality equations [9,1, 2]. Some of the details of
this are relegated to an appendix.
Several classes of completely solvable equations in higher dlmensmns are listed.
Sects. 3 and 4 examine in detail three particular examples occurring in this hst, all in
d~mensIon d = 8. Finally, some concluding remarks are presented in sect. 5.

2. Linear systems and complete solvability


To begin with, let us work in the complexified picture, and consider G L ( n , C )
gauge fields on a d-dimensional complex space C J Later we shall impose reality
conditions, restricting to (say) SU(n) gauge fields on NJ. The analysis does not
depend on what the gauge group is, and works for any group.
Let x ~ ( # = 1,2 .... , d ) be complex coordinates on C ~, and let A u be a gauge
potential on C J (represented as an n × n matrix). We introduce m complex parame-
ters, thinking of an m-tuple of parameters as a point of complex projective m-space
C P m. Let 7rP ( P = 0 , 1 . . . . . m) be homogeneous coordinates on C P m, so the m
parameters are the ratios between the ~rv We want to set up a system of hnear
equations revolving the gauge potential and the parameters. To this end, let V2(~r P)
(a = 1 . . . . . k) be a set of polynomials in ¢rP, each homogeneous of degree q. The
hnear system is
V : ( ~r ) D J / ( x , Tr ) = O , (2.1)

where q, = ~k(x", ~re) IS a column n-vector, and D~ = 0, + A, is the gauge-covanant


derivative.
If k > 1, eqs. (2.1) for qJ are overdeternuned, since there are k times as many
equations as there are unknown functions. In order for (2.1) to have n hnearly
independent solutions, it is necessary and sufficient that the commutator [V,~D,, VIDe]
should vanish, which clearly gives
VCaV;IF~,~=O, (2.2)
R S Ward / Completelysolvable gauge-field equatzons 383

where F~ = 2 O[~A~I + [Au, A,] is the gauge field. (Square brackets enclosing mdaces
denotes skew-symmetrization.) Recall that the V~ are polynommls m ~re, whereas F,~
does not depend on ~re, so eq. (2.2) is a polynomial m ~rP, and equating each
coefficient of this polynomml to zero gives a set of linear equations on F~,, as
desired. The number of equations is clearly

½k( k - 1 ) ( 2q + m )

The appendix describes a construcuon which demonstrates that these equations,


considered as a system of first-order equations for Au, are "completely solvable",
provided only that the following techmcal requirement on Vfl holds. The polynomi-
als V~"can be written as V~e i~r p .. 7rR, where the V~p R are complex constants, and
are totally symmetric in the q ind~ces P . . . R. We require that Vflp R, considered as a
map from the (aP...R) space to the /~-space, be non-singular. In particular, this
gwes the constraint that these two spaces have the same d~mens~on, namely

For example, if d = 4, k = 2, m = 1 and q = 1, then the reqmrement on V~p is

[ V,o
det V~I

v?,
V1]
V?o Vl"0]
V1] 1/141]

v;lj
4:0.

Thas (simplest) case leads to the self-duality equations in d = 4, as we shall see in


sect. 3.
We can use the constraint (2.3) to obtain a classification scheme. For d ~< 11 there
are a total of 11 cases, and these are set out in table 1. (The table lists 13 cases, but
the three cases A2, B I and C 1 coincide, and correspond to the case of self-dual
Yang-Mills m d = 4. The purpose of this triplication is to illustrate the fact that each
of the classes A, B and C generahzes the self-duality equations.) In each case, the
appropriate values of k, m, q and d are given, together with a quantity which
measures the overdeterminacy of the equauons, and is defined as follows.
The equations on F~ form a gauge-invarlant set of equations for A~. Therefore
there are, in effect, ( d - 1)n 2 unknown functions, since A~ has dn 2 components, and
a gauge condition such as (say) A I = 0 reduces this by n 2. Thus the number of
equations minus the number of unknowns is

q u a n t i t y - ½k( k - 1)( 2qm+ m ) _ ( d - 1)


384 R S Ward / Completelysoh~ablegauge-fwld equattons
TABLE 1

k m q d Quantity Range

A~ k 1 1 2k l(k 2)(3k-1) k=2,3,4,5


Bq 2 1 q 2 ( q + 1) 0 q= 1,2,3,4
Cm 2 m 1 2(m + 1) 12m(m- I) m = 1,2,3,4
D 3 2 1 9 10

times n 2. We see from table 1 that only in the c a s e s Bq (and A 2 : C 1 : B2) are there
as many equations as unknowns; in the other cases we get an overdetermined set of
equations for A~. Do not confuse the overdeterminacy here with that of the linear
system (2.1).
We have seen m this section how a system of hnear equatmns, depending on a
"spectral parameter" in C P % has IntegrabIlity conditions which may be regarded as
completely solvable (in the sense of the geometric construction described in the
appendix). The d ~< 11 hst in table 1 can clearly be extended to arbitrarily high
dimensions. But tbas is not the most general situation. For one thing, we could allow
the spectral parameter to range over some more general complex manifold than
C P " (for example, over an algebraic variety). For another, we could allow V~ to
depend on x" and well as 7rP; in effect, this takes us from a fiat to a curved
space-time background. Each of these possibilities leads to more general
completely-solvable equations than those studied in this paper. In addition, there are
examples which involve not just pure gauge fields, but also (say) Higgs fields.
However, it may be the case that all of these arise from pure gauge field equations
by dimensional reduction, as happens in the case of the Bogomolny equations for
monopoles [1].
It seems reasonable to propose the following definition: that gauge field equations
are completely solvable if and only if they are the integrabllity condiuons for a linear
system of the most general type. It remains to make this definition more precise, and
to classify all the cases that can occur. The basic idea of expressing a non-linear
equation as the compatibility condition for a linear system, especially with regard to
the "inverse scattering" technique, has been around for some time [10]. But until
now, it does not seem to have been applied In a systematic way to gauge and other
theories in arbitrary dimension.

3. T h e c l a s s A k

In the class A k the polynomials V~ have the form V~(~r) = V~eTrP, where P ranges
over 0,1. We assume that V~v is non-singular as a map from the ~-space to the aP
space. So we might just as well label the coordinates on C d as x ap, with x ~ being
R S Ward / Completely solvable gauge-fieldequattons 385

defined in terms of these by x ~ V~pXaP. The linear system then takes the simple
=

form ~eDap ~ = O.
At this stage, the symmetry group acting on our equations Is the product of
G L ( k , C) (acting on the index a) and GL(2, C) (acting on the index P). Suppose we
want to reduce this symmetry group in order to preserve a euchdean metric. If k is
even, the most economical way to do this is to reduce G L ( k , C ) to Sp(½k), and
GL(2,C) to Sp(1). (The symplectic group Sp(r) is the group of r × r quatermonlc
matrices M satisfying M*M = 1.) This works as follows. Define symplectic forms Gb
and e'pQby
El2 -~- --821 = E 3 4 = --E43 . . . . . . Ek( k 1) = 1 ,

with the other components equal to zero. Impose on x ~e the reality condihon

X aP = EabEtpQX bQ.

Then the x ~e are, in effect, coordinates on R d = R ek, and the line-element

ds 2 = eabe'pQdxaedxbQ

is the euclidean metric o n R 2 k The orthogonal group SO(2k) is reduced to the


subgroup [Sp(1) × Sp(½ k ) ] / Z 2.
We shall now examine in detail the two cases k = 2 and k = 4, 1.e. dimensions
d = 4 and d = 8. But it is worth emphasizing that each one of the A~ (k >/2) gwes a
completely solvable set of equations. If k is odd, then finding a "euclidean slice" for
the equations to hve on, is not quite as natural as it is for k even.
In the case d = 4, the orthogonal group SO(4) is not reduced at all, since SO(4) is
isomorphic to [Sp(1)× Sp(1)]/7/2- The gauge field (2-form) F,p decomposes into two
irreducible pieces:

FapbQ= eabGrQ+ e'pQHab,

6 = 3 + 3, (3.1)

where GpQ ~---~ FaPbQ and Haa = ½e'PQFaehQ. The integraNhty conditions for
l_ablz"
"B'POap~ = 0 are GpQ = O, since

[ ~ P D a p , qrQDI, Q] = ¢rPTrQFaPbQ = ~abqrPq'rQGpQ.

If we write
X 1° = y = X 1 -[- lX 2 ' X 20 ~ Z ~ - - X 3-~- lX 4,

X 21 = ~ X 1 -- lX 2, X 11 = --~=X 3 q- lX 4, (3.2)
386 R S Ward / Completelysob,able gauge-field equattons

then the equations GpQ = 0 become

Fv.=Fy:=O, F~,v + F:. = 0, (3.1)'

or, equivalently,

El2 = F34, El4 = F23, 6 3 = --F24" (3.1)"

These are, of course, just the self-dual Yang-Mdls equations In d = 4. The G term
and the H term in (3.1) are, respectively, the anti-self-dual and the self-dual parts of
F~,,.
The following feature of the self-duality equations Is worth recalling here, since we
shall see presently that the other members of the class A k, for k even, also possess it
to some extent. Tins is that the action density can be written as

tr F,~F ~" = ½tr e"""~F,~F,a + 4 tr GpQG eQ . (3.3)

The first term on the right-hand side of (3.3) is a "topological charge density"; its
integral o v e r R 4 classifies the field topologically. If we look at finite-action fields
belonging to a particular topological class, then their action is bounded below by the
topological charge, and this lower bound is attained precisely when GpQ vamshes.
Thus, in particular, the self-duahty equations Geo = 0 imply the Yang-Mdls equa-
tions D"F,~ = 0, since a rmnimum of the action is necessarily also a critical point.
We move on now to consider the case k = 4, d = 8. Here SO(8) is reduced to
[Sp(1) × Sp(2)]/Z 2, a maximal subgroup. With respect to this reduced symmetry, a
2-form F~ decomposes as follows:

fapbQ = EabapQ + Kabe Q + e'pQH, h,

28=3+ 1 5 + 10, (3.4)

where

apQ = ~1_~b=r~ebp, (symmetric in PQ )

KabPQ = 7FaPba
1 + 7FaQb~,l -- e,bGeQ (symmetric in PQ, skew tracefree m ab)

11, b = ±oP
, Qz.,2.~ " ~PbQ, (symmetricin ab )

The integrabdlty condition for ~reD~pq, = 0 Is ~rP~rQF~PhQ = 0, which xs equivalent to

GpQ = O, KabPQ = 0, (3.5)


R.S Ward / Completely sob,able gauge-field equatlom 387

a total of 18 equations. Since there are only 7 unknowns (the Au with a gauge
con&tion), the system (3.5) is somewhat overdetern~ned, in contrast to the self-dual-
ity equations in d = 4. Despite this, (3.5) has many solutions, as we shall see below.
If to (3.2) we add

X30=t=X5+lX 6' X40=W = --X7+IX 8'

X41~i~_X5_lX 6' X31= --~=X7+lX 8'

then eqs. (3.5) become


Fv~=F,,t=F,v=Fwx=Ft~=Fw~=O,

G = Vw~ = G = F ~ = G = F~: = 0 ,

G-C~=C,+C:=o,

F,:- F~w= C,+ Gz= 0,

C~+C:=C++Ci=o; (3.5)'

or, eqmvalently,

FI~ = C., F13 .= _ F24 , E l 4 = F23 ,

Ce=C,, C~= -g~, &~= ~ ,


C5 = C 6 = C 7 = & ~ , C7 = - & 8 = -C5= &~,
~6 = --~5=&8 = --57' ~8=&7 = - - & 6 = --55" (3.5)"

In dimension d larger than four, there is no SO(d)-invariant 4-form e~/~ that one
can use to define "self-duality" as in dimension four. However, if one reduces from
SO(d) to a subgroup, then there may be a 4-form lnvanant under that subgroup.
This problem was studied in [8], and the maximal subgroups of SO(d), for d~< 8,
were analysed in this connection. In particular, there is a 4-form O ~ t ~= O i ~ l in
dimension 8, lnvarlant under [Sp(1) × Sp(2)]/Z 2 (unfortunately, this case was missed
in [8]). It can be defined as follows: given four vectors W~, X ~, Y" and ZC let
O~,v~I~W~'X~Y'~Z~ be the scalar quantity 6W,,pX~Y~Z be skew-symmetrlzed over W,
X, Y and Z. (In this expression, indices are lowered with the symplectlc forms e,h
and e~,O: for example, X~' = X a % ~ and Y~ = Y~%'eQ.) Now O ~ t ~ defines a hnear
map on the space of gauge field configurations, namely F~,, ~ !2,--,v~B--
ta F ~ . , so we can
look for its elgenvalues a:
388 R S Ward / Completelysolvable gauge-field equattons

In fact, the eigenspaces of this map are exactly the three irreducible subspaces
described in eq. (3.4): the G part, K part and /4 part of F,, correspond to the
eigenvalues ~ = - 5 , - 1 and 3, respectively. (To check this is straightforward
algebra.) Our set of 18 equaUons (3.5) is therefore eqmvalent to (3.6) with ~ = 3.
Hence (3.5) xmplies the Yang-Mills equations D"F~,= 0, as a consequence of the
Bianchi identities DI~F~~= 0 (cf. [8]).
The analogue of (3.3) is the identity

tr F,~F "~= ~tr O~,~F~"F~# + 4tr K~hpoKa~eQ+ ~-trGpQGeO. (3.7)

The first term on the fight-hand side of (3.7) is a total divergence and therefore a
candidate for a "topological charge density". But the formula (3,7) is not as easily
related to a topological classification and minimum action solutions as (3.3) is,
because the dimensions are wrong. Indeed, there are no finite-action solutions of the
Yang-Mllls equations on R 8 (except F ~ = 0 ) [11]. In order to get finite-action
solutions, one would have to "compactify" some of the dimensions.
We end this section by giving an analogue of the Corrigan-Falrlie-'t Hooft-
Wllczek ansatz,

A, = ~,~ 0~log ¢, (3.8)

winch "converts" an arbitrary solution q, of the Laplace equation on R 4, into a


self-dual SU(2) gauge field A,. Tins is the first of a sequence of ans~ttze which
generate solutions of the self-duality equations from solutions of linear equations [2].
These ansatze generalize to our equations (3.5) in dimension eight, and the first
ansatz again has the form (3.8). It is obtained by letting the "patching matrix" F
discussed in the appendix have the form

F =
1
rro/~.1 ,

where/" = F(~oa, ~re) is an arbitrary scalar function of six complex variables, homo-
geneous of degree zero. So F is, in effect, a free function of five variables. The
construction described in the appendix, for obtaining the gauge potential A,, is
easily carried out (see [2] for details of the analogous calculation m d = 4), and what
one ends up with is the following.
Starting with the arbitrary function F, define a scalar field ~ = ~(x) by the
Cauchy integral

f ,e,,pd,e
R S Ward / Completely solvable gauge-field equattons 389

the contour being I % / % t = 1. Put

Aae = ~1 pQ OaQlOgq~,

where T° = -7/~ = o 3, ~/1 = o l _jr_/02, and ~/o = o' - ,02, or' being the Pauli matrices.
Tlus gauge potential A~ is then a solution of (3.5), depending, in effect, on a free
function of five variables. If 0 is real-valued (and it is possible to choose F so that it
is), then Az will be trace-free and antl-herrmtian, 1.e. will be an SU(2) gauge
potential.
For example, if

/~ = I "1- q'f07/'l(0) 1 -- 093)-1(0) 2 -- 0)4) - 1 ,

then ~ = 1 + ( X - y ) - 2 , where X ~ = ( x 1, x 2, x 3, x 4) and Ya = (x 5, X 6, X 7, X 8) The


corresponding A, may be thought of as the 1-1nstanton solution in the X-space, with
the position of the instanton given by the vector Y~; or vice-versa.

4. The classes Bqand C,.


In the case of class Bq we may express our coordinates on C d as x ~p R, where
a= 1,2 and P , Q , R .... = 0 , 1 ; x ~p R has q capital Roman indices and is totally
symmetric in these, so d = 2(q + 1). The linear system (2.1) becomes

rrL..rrROap R~ = O. (4.1)

These equations have as symmetry group the product of two copies of GL(2, C),
acting on the lower-case and capital indices respectively. Suppose, as m sect. 3, that
we want to reduce this to a subgroup of SO(d). If q is odd, there is a natural way of
doing so, which reduces the symmetry to

[Sp(1) × Sp(1)]/7/2 ~ SO(4) _c S O ( d ) .

This is achieved by requiring that the two symplectic forms eab and e~,Q (defined as
m sect. 3) be preserved, and imposing the reality condmon

xaP R __
__ E a b EtP K . . . E t R M x b K M

The d-dimensional euclidean metric is

ds2=dxaP RdXap R, (4.2)

where radices are lowered by means of e and e'. Only if q is odd is (4.2) a metric
line-element: if q is even, then (4.2) is skew-symmetric rather than symmetric.
390 R S Ward / Completely solvable gauge-field equateons

T h e case q = 1 is j u s t the self-dual Y a n g - M i l l s in d = 4 again: clearly it coincides


with the case A 2. So let us exarmne the case q = 3, d = 8. H e r e a 2 - f o r m F~,
F ~ e o r b s r u , b e l o n g i n g to a 28 of SO(8), d e c o m p o s e s as 28 = 3 + 15 + 7 + 3. T h e
i n t e g r a b d i t y c o n d i t i o n for (4.1) is the vanishing of the 7 p a r t of F,~, i.e.

a b _
e~hF~*'OR s r u ) - O, (4.3)

the parentheses denoting symmetnzation. If we define

xlOOO = y = x l q- l X 2 ' X1011 ~ t = x 5 + lX 6 ,

X 1001 ~ W = - - X 7 q- l X 8 ~ X 1111 -~ Z ~ - - X 3-~- l X 4 ,

then eqs. (4.3) b e c o m e

F y , = 0 = F~.,

~ , , , - Fw. = 0 = F y , - Fw..

Fv~ + F , : - 3&~ = 0 = F~w + F,: - 3F.,,

F~-Fz: +9Ft,-9Fw~=O; (4.3)'

or, equivalently,

El2 + F34 + 9F56 + 9Fv8 = 0,

El3 -- F24 = 0,

F~4 + F23 = 0,

F15 + F26 + F37 + F48 = 0,

F16 -- F25 + F38 - F47 = 0,

11717- F28 - F35 -f- F46 q-- 3F57 - 3/768 = 0,

F18 + F2v - F36 - F45 - 3F58 - 3F67 = 0. (4.3)"

This system of e q u a t i o n s is like (3.5) in that it a d m i t s ansatze of the sort


m e n t i o n e d at the end of sect. 3. But it is unlike (3.5) in not being o v e r d e t e r m i n e d
(here, there are as m a n y equations, n a m e l y seven, as there are unknowns), a n d in
that it does not i m p l y the Y a n g - M i l l s equations.
R.S Ward / Completeh' solvable gauge-field equattons 391

Let us move on, finally, to consider the class C,,. This case 1S, in m a n y ways,
c o m p l e m e n t a r y to class A k. As coordinates we m a y use x ~p, where a is a 2-dimen-
sional index and P is an (m + 1)-dimensional index. If m is odd, we can define a
euclidean structure exactly as was dome in sect. 3 (the definitions of the symplectic
forms eah and e'pQ being adjusted in an obvious way to take account of the changed
ranges of the Indices). The dimension of the space is d = 2 ( m + 1), and the
orthogonal group S O ( d ) is broken down to [Sp(1)× Sp(~m + ½)]/Z 2.
The case m = 1 is that of the self-duahty equations in d = 4. The next interesting
case is that of m = 3, d = 8. Here, the decomposition of the field tensor F~,, is just like
(3 4), but with capital and lower-case indices swapped, namely

FapbQ=epQGab+KabpQ+eahHeQ,

28 = 3 + 15 + 10. (4.4)

The hnear system is ~reDap6= 0 and its lntegrablhty condition is

HpQ = 0, (4.5)

a set of 10 equations. If we set

X 10 = X 21 = y = X 1 4- IX 2,

X 11 = -- X 20 = Z = --X 3 4- IX 4,

X 12 = X 23 = i = X 5 + lX6~

X 13 = -- X 22 = W = - - X 7 4- IX 8,

then eqs. (4.5) become

C~=o = F~,

£,~ = o = & ~ ,

F , z + C ~ = o = ~z + &...

C,-Fw~=O=C,-F~:,
F.,-Cz = 0 = F,~- C ~ ; (4.5)'
392 R.S Ward / Completel; soh~able gaugefield equattons

or, equivalently,

~2 = --~4' ~3 = ~4' ~4 = --~3"


66= -~8, 67=~8, ~8= -~,

Ft5 + F26 + F37 + F4s = 0 ,

Fi6 - F25 + F38 - F47 = 0,

F17 - F28 - F35 + F.6 = 0 ,

F]8 + F27 - F36 - F45 = 0. (4.5)"

Note that these equations precisely complement eqs. (3.5), and that they imply the
set of seven equations (4.3).
In this case, as in the case A n, there is an invariant 4-form O , ~ ; it is unique up to
scale, and is defined as in sect. 3 but with capital and lower-case indices inter-
changed. However, eqs. (4.5) are not elgenvalue equations of the form (3.6); indeed,
(4.5) does not Imply the Yang-Mills equations. But it lmphes equations which are
obtained from a slightly modified lagranglan. For one has the identity

tr F ~ F "~ + 16 tr G, bG"h= - ½ t r O , ~ F ~ F ~B + 8trHeQH eQ (4.6)

Since the O term in (4.6) is a total divergence, the vanishing of HpQ, i.e. the system
(4.5), Implies the Euler-Lagrange equations obtained from the lagrangian on the
left-hand side of (4.6). This "modified" action density is gauge-lnvariant, positive-
definite and [Sp(1)× Sp(2)]/Z 2 invarlant, but not SO(8) invariant.
To summarize: the equations in classes Bq and Cm, although completely solvable,
do not imply the Yang-Mills equations (except for q = m = 1). But they may be of
interest in the context of lagrangians other than the "standard" Yang-Mills one.

5. Concluding remarks
We have seen that there are gauge-field equations in dimension greater than four
which share some of the remarkable properties of the self-duality equations in 4d.
However, the case of dimension four is rather special, for reasons arising from the
choice of t r F 2 as the action functional.
If higher dimensions are to be relevant, then some process of dimensional
reduction, such as spontaneous compactification, has to be involved. To this end,
one should extend the considerations of the present paper (which deals only with flat
d-dimensional space) to curved spaces. This can be done, and will be described
elsewhere.
R S Ward / Completelysob,ablegauge-fieldequauons 393

Completely solvable equations are few and far between, in the sense that almost
every non-linear equation is, it would seem, not completely solvable. But despite
being rather special, completely-solvable equations (in gauge theories, or in other
cases such as CP" models) are worth studying, for what they can tell us about
non-linear and non-perturbatlve phenomena in general.

The author would like to thank C. Devchand and D.B. Fairhe for useful
conversations.

Appendix

This appendix describes a way in which the solutions of the lntegrablhty condi-
tions arising from a linear system V,"(~re ) D , } = 0 can be constructed. It justifies our
calling such integrabllity equations "completely solvable". The construction is best
described in the language of vector bundles and geometry, but the discussion here
will largely avoid such sophistication.
Because of the assumption that V~t, s be non-singular, we may denote the
coordinates on C a as x"*" s (symmetric in P . . . S ) . The linear system now reads

7rP...TrSo, p s~k = O,

the lntegrablhty conditions for which are

skewed over ab ]
F,p sbr w symmetrlzed over P .. W / = 0. (A.1)

For the sake of simplicity let us take m = 1, so that the capital indices P, Q . . . . range
over 0,1. A later remark will describe what happens when m > 1. Solutions of (A.1)
may be constructed as follows.
Let F = F(~o "p R, ~ro) be a non-singular n × n matrix of complex-analytic func-
tions of kq + 2 variables, homogeneous of degree zero. The symbol ~0"~ R has q - 1
capital indices, and is totally symmetric in these. Thus (taking the homogeneity into
account) F 1s defined on a region of C P kq+I. We reqmre this region to be such that

G ( x , ~r)= F ( x "e Rs~rs, ~rr) (A.2)

IS analytic in a neighbourhood of [~rl/rro] = 1, for each fixed x. In other words, G is


analytic on an annular region in the ~ri/rr0 Argand plane. It follows from (A.2) that

,c(x, = 0. (A.3)

(The Indices P, Q . . . . are raised and lowered using the symplectic form e' as in sect.
3.)
394 R S Ward / Completely solvable gauge-field equanons

Now "split" G by finding non-singular n × n matrices H(x, w) and /:/(x, w),


where H is analytic for IWl/%l ~< 1 and/2/is analytic for 1~/%1 >/1 (including ~ ) ,
such that

G =/~/H t. (A.4)

From (A.3) we see that

H lwe...wSO~e sH=I21-17rP...~rSO,e sl;t. (A.5)

Now the left-hand side of (A.5) is analytic for IWl/%l ~< 1, and the nght-hand side
for ITh/%l >/1, whence each side is analytic for all w e, and so (by a form of
Liouville's theorem) must be a polynoImal in w e of degree q:

H IwP...wsO**e sH=~re...~rSA,e s, (A.6)

where the A,p s are functions only of x (not of we). This defines the G L ( n , C )
gauge potential A,. The field equations (A.1) are automatically satisfied, as one can
see by operating on (A.6) with wr...crWObv w and skewing over ab. So the
construction generates solutions out of (essentially arbitrary) matrix functions F.
Furthermore, it produces all solutions of (A.1), this is not obvious, but can be
proved without too much difficulty.

REMARKS

(i) The splitting (A.4) is not, in general, possible, we have to choose F in such a
way that it ts possible, for all x. (For example, if n = 1 and F is a 1 × 1 matrix, then
the condition is that log G should be single-valued on the region of analytlcity
specified for G.) Those values of x where the splitting (A.4) is impossible will be
singular points of A,.
(ii) The splitting is not unique, but clearly admits the freedom H ~ HA, I2I ~ I~IA,
where A = A(x) is a non-singular n × n matrix of functions of x only. From eq.
(A.6) we see that this reduces on A~ the transformation

A~ ~ A-1A~A + A-IO, A.

So the sphtting freedom corresponds to gauge freedom in Au.


(iu) The reason for requlnng Vfp s to be non-singular is as follows. If we did not
impose this restriction, we could still carry out the construction to obtain an object
A.p s- But in order to be able to interpret this as a gauge potential A., we have to
be able to invert V~p s.
(iv) The geometnc picture may be summarized as follows. For each w P, the k
vectors Vl~(~r). . . . . Vk~(Ir) span a complex k-plane Z in C d. The linear equations
V~D,~ = 0 say that ~k is covariantly constant over each of these k-planes, the
R.S Ward / Completely solvable gauge-field equatwns 395

consistency c o n d i t i o n for which is that the c u r v a t u r e F ~ should vanish when


restricted to a n y of the k-planes. This is w h a t eq. (A.1) says. T h e space of all
k - p l a n e s g e n e r a t e d b y the Vf is an o p e n subset 6~ of C P kq+l, a n d the gauge field
d e t e r m i n e s a vector b u n d l e E over ~ : the fibre E z over a p o i n t Z in ~, is defined to
be the n - d i m e n s i o n a l vector space

E z = ( ~ on ZI V [ D J / = O on Z ) .

T h e basic t h e o r e m says that solutions of (A.1) in C d c o r r e s p o n d to vector b u n d l e s E


over ~ , satisfying the c o n d i t i o n that E restricted to x (defined below) is trivial. This
x is a C P 1 in C P kq+l which c o r r e s p o n d s to x ~ ~ ; in terms of the c o o r d i n a t e s
(c0ae R, ~T) on C P kq+l, x is given b y

¢daP R : xaP RST.[S '

in terms of x aP s. T h i n k of the ~re, for fLxed x, as h o m o g e n e o u s c o o r d i n a t e s on x.


The m a t r i x F is a " p a t c h i n g m a t r i x " winch d e t e r m i n e s the vector b u n d l e E. T h e idea
ts that one cuts ~ into two pieces, c o r r e s p o n d i n g to I~q/~r0l ~< 1 a n d I~rl/~r0] >/1,
a n d on the o v e r l a p one specifies the p a t c h i n g m a t r i x F. The c o n d i t i o n that E
restricted to x be trivial j u s t says that F is " s p l i t t a b l e " for that p a r t i c u l a r x. W e see,
therefore, that solutions of the n o n - l i n e a r e q u a t i o n s (A.1) c o r r e s p o n d to vector
b u n d l e s E which are required to satisfy a regularity condition, b u t no differential
equation.
(v) All ttns was for m = 1. It r e a d i l y extends to m > 1, a n d one again has a
t h e o r e m like that in the previous remark. But n o w one needs m o r e than two patches,
in fact m + 1 in general. So one has to start with several matrices F, one for each
overlap b e t w e e n pairs of patches. Things b e c o m e s o m e w h a t m o r e c o m p l i c a t e d , b u t
the s p l i t t i n g - t y p e c o n s t r u c t i o n works as it d i d for m = 1.

References
[1] M F Atlyah, N J I-htctun, V G Dnnfeld and Yu I Mamn, Phys Lett 65A (1978) 185,
E Corrlgan and P Goddard, Commun Math Phys 80 (1981)575,
N J Hltctun, Commun Math Phys 83 (1982) 579, 89 (1983) 145
[2] M F Atlyah and R S Ward, Commun Math Phys 55 (1977)117,
E F Corngan, D B Fmrhe, R G Yates and P Goddard, Commun Math Phys 58 (1978) 223,
R S Ward, Commun Math Phys 80 (1981)563
[3] L -L Chau, Ge M -L and Wu Y -S, Phys Rev D25 (1982) 1086,
L Dolan, Phys Lett 113B (1982) 387
[4] S G Matmyan, G K Savvldy and N G Ter-Arutyunyan-Savvady. JETP (Sov Phys ) 53 (1981) 421,
E S Nlkolaevskn and L N Shur, JETP Lett 36 (1982) 218
[51 P Forgb.cs and N S Manton, Commun Math Phys 72 (1980) 15,
E Cremmer, in Supergrawty "81, eds S Ferrara and J G Taylor, (Cambridge Umverslty Press, 1982)
pp 313-368
396 R S. Ward / Completely solvable gauge-field equations

[6] P G 0 Freund and MA Rubm, Phys Lett 97B (1980) 233,


F Englert, Phys Lett 119B (1982) 339
[7] R Penrose, Gen Rel and Grav 7 (1976) 31
[8] E Corngan, C Devchand, D B Fawhe and J Nuyts, Nucl Phys B214 (1983) 452
[9] R S Ward, Phys Lett 61A (1977) 81
IO] A A Belavm and V E Zakharov, Phys Lett 73B (1978) 53
ill] A Jaffe and C Taubes, Vortices and monopoles (Blrkhauser, Boston, 1980) p 32

You might also like