kotschick.

qxp 3/17/99 10:09 AM Page 335

Gauge Theory Is
Dead!—Long Live
GaugeTheory!
Gauge Theory!
D. Kotschick

I
n late October and early November 1994 cation of smooth 4-manifolds is very different
many mathematicians received e-mail from from their classification up to homeomorphism.
colleagues trumpeting the death of gauge Combined with Freedman’s work, it produced ex-
theory. More than a decade earlier, S. K. otic differentiable struc-
Donaldson (Oxford) had found a deep but tures on Euclidean 4-
mysterious link between Yang-Mills theory from space, an anomaly that
mathematical physics on the one hand and 4-di- does not arise in other di-
mensional differential topology on the other. mensions. Given the
Since then, many topologists had become fasci-
nated by gauge theory. Last autumn they found
In gauge theory one
considers connections or
enormous impact
themselves on the receiving end of wry com- covariant derivatives A of the U(1)
ments, when rumour had it that a new set of on a principal G -bundle
equations proposed by E. Witten (IAS, Prince- over a smooth 4-manifold monopole
ton) had made gauge theory obsolete in topol- X endowed with an ori-
ogy. This was, of course, an exaggeration. entation and a Riemann- equation on
Witten’s equations, originating in his joint
work [SW] with N. Seiberg (Rutgers) in quantum
ian metric; here G is a
compact Lie group. The
4-dimensional
field theory and appearing in [W], do provide connections of interest topology, there are
shortcuts to many of the consequences of gauge are the so-called instan-
theory and quickly lead to proofs of very im- tons, the solutions of the high expectations
portant new results. The equations are them- anti-self-dual Yang-Mills
selves part of a gauge theory and shed new light equation, defined as fol-
for the other
on the Yang-Mills equations that Donaldson used. lows: Let ∗ be the Hodge equations.
Rather than making gauge theory obsolete, Wit- star operator defined by
ten’s equations make gauge theory even more in- the orientation and the
teresting and more powerful. Riemannian metric on X.
For A a G -connection, let
The Old Gauge Theory F A be its curvature . The self-dual part of the cur-
Coming on the heels of the work of M. H. Freed- vature is
man (University of California, San Diego) on topo- 1
F+A = (F A + ∗F A )
logical 4-manifolds, Donaldson’s use of gauge 2
theory showed that the differentiable classifi- and the anti-self-dual Yang-Mills equation is
∗F A = −F A ⇐⇒ F+A = 0.
D. Kotschick is professor of mathematics at the
University of Basel in Switzerland.

MARCH 1995 NOTICES OF THE AMS 335

kotschick.qxp 3/17/99 10:09 AM Page 336

Instantons are the minima

R of the Yang-Mills X
s
functional Y M(A) = X |F A |2. When G is U(1), (1) q = eQ/2 ai eKi ,
the anti-self-duality equation is linear and the in- i=1
stantons are completely described by Hodge the- where Q is the intersection form of X and the
ory. When G is SU(2), say, the equation is a non- ai ∈ Q and Ki ∈ H 2 (X, Z) are certain character-
linear PDE which is elliptic on the space of istic elements for the intersection form. The
connections modulo bundle automorphisms or “basic classes” Ki are constrained by the in-
gauge transformations. Its space of solutions, the equality 2g(S) − 2 ≥ Q(S, S) + Ki (S) for the genus
moduli space of instantons up to gauge trans- g of any smoothly embedded surface S ⊂ X with
formations, is generically a finite-dimensional normal bundle of positive degree.
smooth manifold M . This manifold is usually
noncompact, partly due to the conformal in- The New Gauge Theory
variance of the equation, and this noncompact- To write down Witten’s equations on a smooth
ness leads to many technical difficulties in the oriented Riemannian 4-manifold X, one has to
topological applications of gauge theory. choose a Spinc-structure, that is, a lift of the
In 1981–82, Donaldson had the insight that frame bundle from SO(4) to Spinc (4) =
the algebraic topology of the moduli space M Spin(4) ×±1 U(1). Associated with this structure
contains subtle information about the differen- are bundles V± of positive and negative spinors
tiable structure of X. He first proved that certain and a complex determinant line bundle
topological 4-manifolds do not support any dif- L = det(V± ). Further, there is a canonical map
ferentiable structure at all. Later he defined dif- σ : V+ × V+ → Λ2+ defined by taking the trace-
ferentiable invariants of large classes of mani- free part of an element in V+ ⊗ V+ considered as
folds which, although difficult to calculate an endomorphism of V+ .
completely, were very successful in distin- A U(1)-connection A on L , together with the
guishing nondiffeomorphic differentiable struc- Levi-Civita connection of the Riemannian metric
tures on X. induce a covariant derivative Γ (V+ ) →
This initially came as a complete surprise to Γ (V+ ⊗ T ∗ X) . Composing this with Clifford mul-
topologists. Even after gauge theory had been tiplication Γ (V+ ⊗ T ∗ X) → Γ (V− ) defines a Dirac
firmly established as a tool in topology, there was operator DA : Γ (V+ ) → Γ (V− ). Witten’s equations
no conceptual understand- for a connection A and a positive spinor
ing of how and why in- φ ∈ Γ (V+ ) are
stantons were related to the
DA φ = 0
structure of 4-manifolds.
Rather than Early on Donaldson F+A = iσ (φ, φ).
proved that in the case of
making gauge complex algebraic surfaces These equations are invariant under bundle au-
connections with anti-self- tomorphisms of L , but they are not conformally
theory obsolete, dual curvature are the same invariant. The solutions, called monopoles, are
the minima for the functional
Witten’s as stable holomorphic
bundles in the sense of geo- Z
equations make metric invariant theory.
Furthermore, the instanton X
(|F+A − iσ (φ, φ)|2 + |DA φ|2 ).

gauge theory invariants are nontrivial for

algebraic surfaces. This es- The space of monopoles modulo bundle auto-
even more tablished a strong link be- morphisms is generically a smooth manifold,
tween gauge theory and and it is always compact. Compactness follows
interesting and algebraic geometry. Four- easily from the Weitzenböck formula for the
dimensional differential Dirac operator combined with standard elliptic
more powerful. topology was seen as being theory. Further, if the scalar curvature of the
very close to complex Riemannian metric is nonnegative, all the solu-
geometry. tions of the monopole equation have φ = 0 and
Over the last four years, so are U(1) instantons. Even if the scalar curva-
P. B. Kronheimer (Oxford) and T. S. Mrowka (Cal- ture is negative somewhere, there are only finitely
ifornia Institute of Technology) and others de- many cohomology classes which are the first
veloped a structure theory for the instanton in- Chern classes of complex line bundles L admit-
variants. They found that under suitable ting nontrivial solutions to the monopole equa-
hypotheses on X all the invariants derived from tions.
different SU(2)-bundles over X could be pack- The compactness of the monopole moduli
aged into a single analytic function spaces makes them much easier to handle than
q : H2 (X, R) → R of the form the instanton moduli spaces. This is the main rea-

336 NOTICES OF THE AMS VOLUME 42, NUMBER 3

kotschick.qxp 3/17/99 10:09 AM Page 337

son why Witten’s approach is much simpler than

Donaldson’s. Although instanton and monopole
moduli spaces seem to contain very similar in-
formation, the monopoles are tied much more
closely to the Riemannian geometry of X. This
inspires hope for the development of a combi-
natorial approach to gauge theory.
One can use the monopole moduli spaces to
reprove and generalize most of Donaldson’s the-
orems about the nonexistence of differentiable
structures on certain topological 4-manifolds.
One can also define invariants of differentiable
structures, for example by counting the number
of points in zero-dimensional moduli spaces.
These invariants are trivial for manifolds ad-
mitting either a Riemannian metric of positive
scalar curvature or a smooth connected sum de-
composition in which both summands have in-
tersection forms which are not negative defi-
nite.
On the other hand, the invariants are nontrivial
for complex algebraic surfaces and, more gen-
erally, for symplectic manifolds [T]. This puts
strong restrictions on the differential topology
and geometry of symplectic 4-manifolds. It im-
plies, for example, that the connected sum of
three copies of the complex projective plane
does not admit a symplectic structure, although
all other known constraints for symplectic man-
ifolds are satisfied in this case.
Witten [W] predicted that the “basic classes”
Ki appearing in (1) should be precisely the first

Laura Pedrick/NYT Pictures

Chern classes of complex line bundles L for
which the new monopole invariants are nonzero
and that the coefficients ai in (1) should be de-
termined by the values of the monopole invari-
ants. For large classes of manifolds, this pre-
diction was quickly proved to be true, because
one can calculate both sets of invariants, for in- E. Witten
stantons and for monopoles, and compare the
answers. There are promising attempts to prove
Witten’s conjecture in a more direct way. The Physical Origins
One of the spectacular new theorems proved Yang-Mills theory is a conformally invariant clas-
using Witten’s equations concerns the problem sical field theory whose groundstates are the in-
of finding the minimal genus of a smoothly em- stantons. Experience with quantum field theory
bedded surface in a 4-manifold representing a suggests a recipe for turning this into a so-called
given homology class. While formula (1) solved topological field theory whose correlation func-
this problem in many cases, it seemed very dif- tions are Donaldson’s instanton invariants. As
ficult to generalize the argument using instan- they are invariants of the differentiable structure,
tons to cover, for example, the case when the one can vary the Riemannian metric used with-
manifold is the complex projective plane. Now out affecting the invariants. Witten [W] outlines
the nontriviality of the monopole invariants as- how studying a family of metrics gt = tg1 , where
sociated with the first Chern classes of algebraic t > 0 is a real parameter, leads naturally to the
surfaces allows one to show that algebraic curves appearance of the monopole equations.
have minimal genus among all smoothly em- For small t , the classical approximation to
bedded surfaces representing the same homol- quantum field theory coincides with Donald-
ogy class (if it has nonnegative self-intersection). son’s definition of instanton invariants. For large
This statement for the case of algebraic curves t , however, the quantum vacuum states of the
in the complex projective plane [KM] was previ- theory, parametrized by a complex variable u,
ously known as the Thom conjecture. become relevant. It turns out [SW] that the quan-

MARCH 1995 NOTICES OF THE AMS 337

kotschick.qxp 3/17/99 10:09 AM Page 338

tum theory naturally leads one to consider the manifolds there is more to the u-plane than one
family of elliptic curves sees near the special values u = ±1 .
The physical theory developed by Seiberg and
y 2 = (x2 − 1)(x − u).
Witten [SW] suggests a whole family of coupled
For generic u, the elliptic curve is smooth, but equations, of which the monopole equations are
it degenerates to a rational curve at u = ±1 . just the simplest example. If one considers
For large classes of manifolds, all the topo- SU(N) instantons (instead of SU(2)), then there
logical information of the instanton theory can is a related set of equations for G -connections
A , where G is the dual of the maximal torus of
SU(N) , and certain spinor fields φ. The equa-
tions are the Dirac equation for φ and an equa-
tion expressing F+A by a certain hyper-Kähler
moment map, generalizing σ (φ, φ) in the case
when G = U(1) . These generalized monopole
equations are related to higher-dimensional
Abelian varieties rather than elliptic curves. Given
the enormous impact of the U(1) monopole equa-
tion on 4-dimensional topology, there are high
expectations for the other equations. Far from
being dead, gauge theory is more active and ex-
citing than ever.

References
[KM] P. B. Kronheimer and T. S. Mrowka, The genus
of embedded surfaces in the projective plane, Math. Res.
Lett. 1 (1994), 797-808.
[SW] N. Seiberg and E. Witten, Monopoles, duality and
chiral symmetry breaking in N = 2 supersymmetric
QCD, Nuclear Phys. B (to appear).
[T] C. H. Taubes, The Seiberg-Witten invariants and
symplectic forms, Math. Res. Lett. 1 (1994), 809-822.
[W] E. Witten, Monopoles and four-manifolds, Math.
Res. Lett. 1 (1994), 769-796.

S.K. Donaldson be extracted by looking at appropriate elliptic

functions in infinitesimal neighbourhoods of the
points ±1 in the u-plane. The reason is that for
these special values of the parameter, certain par-
ticles in the quantum theory—the monopoles—
become massless, although they are not mass-
less in the classical theory. (The monopole
equations, like the whole quantum theory, are not
conformally invariant, although the classical the-
ory is.) The information about instantons can
then be derived from the monopole equations
alone, leading to formulae for the Donaldson in-
variants. In fact, (1) is the simplest such for-
mula, and understanding this was one of the
motivations for Witten’s approach.
The monopole equations seem to detect the
simplest part of Donaldson theory, which for
simple manifolds is all there is. However, there
are manifolds like the complex projective plane,
on which there are no monopoles at all, for which
Donaldson theory has another, more compli-
cated part to it. In the quantum field theory de-
scription, this is said to be detected by integra-
tion over the u -plane, because for these

338 NOTICES OF THE AMS VOLUME 42, NUMBER 3