TOPOLOGICAL QUANTUM FIELD THEORIES

by MICHAV.L ATI YAH

To Ren~ Thorn on his 68th birthday.
1. Introducti on
In recent years there has been a remarkable renaissance in the relation between
Geometry and Physics. This relation involves the most advanced and sophisticated
ideas on each side and appears to be extremely deep. The traditional links between
the two subjects, as embodied for example in Einstein's Theory of General Relativity
or in Maxwell' s Equations for Electro-Magnetism are concerned essentially with classical
fields of force, governed by differential equations, and their geometrical interpretation.
The new feature of present developments is that links are being established between
quantum physics and topology. It is no longer the purely local aspects that are involved
but their global counterparts. I n a very general sense this should not be too surprising.
Both quant um theory and topology are characterized by discrete phenomena emerging
from a continuous background. However, the realization that this vague philosophical
view-point could be translated into reasonably precise and significant mathematical
statements is mainly due to the efforts of Edward Witten who, in a variety of directions,
has shown the insight that can be derived by examining the topological aspects of quant um
field theories.
The best starting point is undoubtedly Witten' s paper [11] where he explained
the geometric meaning of super-symmetry. It is well-known that the quant um Hamil-
tonian corresponding to a classical particle moving on a Ri emanni an manifold is just
the Laplace-Beltrami operator. Witten pointed out that, for super-symmetric quant um
mechanics, the Hamiltonian is just the Hodge-Laplacian. In this super-symmetric theory
differential forms are bosons or fermions depending on the parity of their degrees. Witten
went on to introduce a modified Hodge-Laplacian, dependi ng on a real-valued function f.
He was then able to derive the Morse theory (relating critical points of f to the Betti
numbers of the manifold) by using the standard limiting procedures relating the quant um
and classical theories.
176 MICHAEL ATIYAH
Wi t h this model of super - symmet r i c qua nt um mechani cs rigorously underst ood,
Wi t t en t hen went on to out l i ne t he correspondi ng ideas for super - symmet r i c quantum
field theories. Essentially such qua nt um field theories shoul d be vi ewed as t he di fferent i al
geomet r y of cer t ai n i nfi ni t e-di mensi onal manifolds, i ncl udi ng t he associated analysis
(e.g. Hodge t heory) and t opol ogy (e.g. Betti number s) .
Gr eat caut i on has of course to be used in such i nfi ni t e-di mensi onal situations
but, t aki ng one' s cue f r om physics on t he one hand and t opol ogy on t he ot her hand,
it is possible to make i nt el l i gent guesses and conj ect ures. Ther e is now ampl e evi dence
in favour of ma ny of these conj ect ures, a numbe r of whi ch have been ri gorousl y esta-
blished by al t ernat i ve met hods. Thi s applies for exampl e to results concer ni ng " elliptic
cohomol ogy " [17] and to t he topic I shall discuss in det ai l in this paper.
Perhaps a few f ur t her comment s shoul d be made to reassure t he sceptical r eader .
The qua nt um field theories of interest ar e i nher ent l y non-linear, but t he non-l i neari t i es
have a nat ur al origin, e.g. comi ng f r om non- abel i an Lie groups. Mor eover t here is
usually some scaling or coupl i ng par amet er in t he t heor y whi ch in t he l i mi t relates to
t he classical t heory. Fundament al topological aspects of such a qua nt um field t heor y
shoul d be i ndependent of t he par amet er s and it is t herefore reasonabl e to expect t hem
to be comput abl e (in some sense) by exami ni ng t he classical limit. This means t hat
such topological i nf or mat i on is essentially robust and shoul d be i ndependent of t he
fine anal yt i cal details ( and difficulties) of t he full qua nt um t heory. Tha t is why it is
not unr easonabl e to expect to under s t and these topological aspects before t he qua nt um
field theories have been shown to exist as rigorous mat hemat i cal structures. I n fact, it
ma y well be t hat such topological under st andi ng is a necessary pre-requi si t e to bui l di ng
t he anal yt i cal appar at us of t he qua nt um t heory.
My comment s so far have been of a convent i onal ki nd, i ndi cat i ng t hat t her e ma y
be i nt erest i ng topological aspects of qua nt um field theories and t hat these shoul d be
i mpor t ant for t he r el evant physics. However, we can reverse t he pr ocedur e and use
these qua nt um field theories as a concept ual tool to suggest new mat hemat i cal results.
I t is r emar kabl e t hat tbSs reverse process appears to be ext r emel y successful and has
l ed to spect acul ar progress in our under st andi ng o f geometry in low dimensions. I t is pr obabl y
no acci dent t hat t he usual qua nt um field theories can onl y be r enor mal i zed in (space-
t i me) di mensi ons ~< 4, and tiffs is precisely t he r ange in whi ch difficult phenomena arise
l eadi ng to deep and beaut i ful theories (e.g. t he works of Thur st on in 3 di mensi ons and
Donal dson in 4 di mensi ons).
I t now seems cl ear t hat t he way to investigate t he subtleties of l ow-di mensi onal
mani fol ds is to associate to t hem suitable i nfi ni t e-di mensi onal mani fol ds (e.g. spaces
of connect i ons) and to st udy these by st andar d l i near met hods (homol ogy, etc. ). I n
ot her words we use qua nt um field t heor y as a refi ned tool to st udy l ow-di mensi onal
manifolds.
Now qua nt um field theories have, because of t he difficulties i nvol ved in cons-
t r uct i ng t hem, oft en been descri bed axi omat i cal l y. Thi s identifies t hei r essential st ruct ural
TOPOLOGI CAL QUANTUM FIELD THEORI ES 177
feat ures and postpones t he quest i on of t hei r existence. We can appl y t he same appr oach,
at t he topological level, and this leads us to f or mul at e axi oms for topological quantum
f i e l d theories. These axioms ar e consi derabl y si mpl er t han for a full bl own t heor y and
t hey have a cert ai n nat ur al i t y whi ch makes t hem plausible objects of interest, i ndependent
of any physi cal i nt er pr et at i on.
I n t he next section I will t herefore present a set of such axioms. Al t hough I will
make a few comment s on t he physi cal backgr ound and not at i on, these can be i gnored
and t he axi oms t r eat ed as a basis for a ri gorous mat hemat i cal t heory.
I n t he t hi rd section I will enumer at e t he exampl es of theories, satisfying such
axioms, whi ch are now known to exist. Much, t hough not all, of this has been rigorously
established by one met hod or anot her . The hi st ory of t hese di fferent theories is qui t e
var i ed so it is cer t ai nl y helpful to see t hem all as fitting into a c ommon axi omat i c
framework.
I t will be cl ear how much this whol e subj ect rests on t he ideas of Wi t t en. I n
f or mul at i ng t he axi omat i c f r amewor k in w 2, I have also been following Gr aeme Segal
who pr oduced a ver y similar appr oach to conf or mal field theories [10]. Fi nal l y it seems
appr opr i at e to poi nt out t he maj or role t hat cobordism plays in these theories. Thus Ren6
Thom' s most cel ebr at ed cont r i but i on to geomet r y has now a new and deeper rel evance.
2. Ax i o ms f o r To po l o g i c a l Qu a n t u m Fi e l d Th e o r i e s
Before embar ki ng on t he axi oms it ma y be hel pful to make a few compari sons
wi t h st andar d homol ogy theories. We can descri be such a t hcor y as a f unct or F f r om
t he cat egor y of topological spaces (or of pairs of spaces) to t he cat egor y say of A-modul es,
wher e A is some fi xed gr ound ri ng ( commut at i ve, wi t h 1, e.g. A = Z, lit or C). Thi s
f unct or satisfies vari ous axioms i ncl udi ng
(i) a homot opy axi om, descri bed geomet r i cal l y by using " cyl i nders " X I,
(ii) an additive axi om asserting t hat , for disjoint sums, F( X 1 u X~) = F(Xx) 9 F(X2).
Not e t hat (ii) implies, for t he empt y set ~,
(ii)' F(z) = O.
The theories we shall descri be will be somewhat similar, but wi t h t he fol l owi ng
significant differences:
a ) t hey will be defi ned onl y for manifolds of a f i x e d dimension,
b ) t h e homot opy axi om is strengthened by repl aci ng cyl i nders wi t h general cobordisms,
c ) t he addi t i ve axi om is r epl aced by a multiplicative axi om, and cor r espondi ngl y t he
cmpt y set has val ue A r at her t i t an O.
Physi cal l y b ) i s rel at ed to relativistic i nvar i ance while c ) is i ndi cat i ve of t he quantum
nat ur e of t he t heory.
93
178 MICHAEL ATIYAH
We come now to the promised axioms. A topological quant um field t heory (Q.FT),
in dimension d defined over a ground ring A, consists of the following data:
(A) A finitely generated A-modul e Z(Z) associated to each oriented closed smooth
d-dimensional manifold Z,
(B) An element Z(M) ~ Z(0M) associated to each oriented smooth (d + 1)-dimensional
manifold (with boundary) M.
These dat a are subject to the following axioms, which we state briefly and expand upon
below:
(1) Z is functorial with respect to orientation preserving diffeomorphisms of Z and M,
(2) Z is involutory, i.e. Z(E' ) = Z(E) ~ where E" is E with opposite orientation and Z(E) ~
denotes the dual modul e (see below),
(3) Z is multiplicative.
We now elaborate on the precise meani ng of the axioms. (I) means first
t hat an orientation preserving diffeomorphism f : Z- + Z' induces an isomorphism
Z( f ) : Z(Z) -+ Z(Z' ) and t hat Z(gf ) = Z(g) Z( f ) for g : Z' ~ Z". Also i f f extends
to an orientation preserving diffeomorphism M -+ M' , with 0M = E, OM' = Z' , then
Z( f ) takes Z(M) to Z(M' ).
The meani ng of (2) is clear when A is afield in which case Z(Z) and Z(Z)" are
dual vector spaces. This is the most i mport ant case and, for physical examples A = Cl
(or perhaps R). However, there are interesting examples (see w 3) with A = Z. In this
case the relation between Z(Z) and Z(Z' ) is like that between integer homol ogy and
cohomology. The duality can be formalized by considering free chain complexes but
we shall not pursue this in detail. Instead we shall take A to be afield and the case A = Z
can essentially be replaced by the fields Q., zp.
The multiplicative axiom (3) asserts first that, for disjoint unions,
(3a) Z(Z 1 ~3 ~) = Z(Z1) | Z(Z, ).
Moreover if 0Ml =Zxt 3Zs , aM z =E~UZ~ and M=M I Uz . M~ is the manifold
obt ai ned by glueing together the common Z3-component (see figure)
M
Z3 ! . . . . . . . . . .
TOPOLOGICAL QUANTUM FIELD THEORIES 179
t hen we requi re:
(3b) Z( M) = ( Z ( M, ) , Z( M, ) )
wher e ( , ) denot es t he nat ur al pai ri ng
Z(X, ) | Z(Z~) | Z(X, )" | Z(X, ) -~ Z(X, ) | Z( X, ) .
Not e t hat when X 3 = o so t hat M is the disjoint uni on of M t and M l t hen (3b)
reduces to t he obvi ous mul t i pl i cat i ve r equi r ement
(3c) Z( M) = Z( M, ) | Z( Mz) .
Our mul t i pl i cat i ve axi om, involving (3b), is t herefore a ver y st rong axi om. I t asserts
t hat Z( M) can be comput ed (in many di fferent ways) by " cut t i ng M in h a l f " al ong
any X 3.
An equi val ent way of f or mul at i ng (3b) is to decompose the boundar y M i nt o
t wo component s (possibly empt y) so t hat
OM = Z t u X0;
t hen Z( M) e Z( Zo) ' | Z( ZI ) ----- Hom( Z( Xo) , Z(Xl) ). We can t herefore vi ew any cobor-
dism M bet ween X 0 and Z~ as i nduci ng a l i near t ransformat i on
Z( M) : Z(X0) -+ Z( Z, ) .
Axi om (3b) asserts t hat this is transitive when we compose bordisms.
Not e t hat t he mul t i pl i cat i ve axi om (3a) shows t hat when X = ~, t he vect or
space Z(X) is i dempot ent . I t is t herefore zero or canoni cal l y i somorphi c to t he gr ound
field A. To excl ude t hc trivial case we shoul d t hen add a non-t ri vi al i t y axi om
(4a) Z(~) = A (for ~ the empt y #di mens i onal mani fol d).
Si mi l arl y when M = ~ axi om (3b) shows t hat Z(~) e A is an i dempot ent . To excl ude
t he trivial case Z(o) = 0 we i mpose
(4b) Z(~) = 1 (for e the empt y d + 1-dimensional mani fol d).
Agai n t he muhi pl i cat i ve axi om in its cobor di sm guise cl earl y shows t hat , for a
cyl i nder X I, t he el ement
Z( E I) e End(Z(Y. ))
is an i dempot ent ~r and mor e general l y it acts as the i dent i t y on t he subspace of Z(Y.)
spanned by all el ement s Z( M) wi t h 0M = X. I f we r epl ace Z( E) by its i mage under
t he i dempot ent ~r it is easy to see t hat t he axi oms are still satisfied. Mor eover , this new
t heory cont ai ns essentially MI t he i nt erest i ng ins of the ol d one since el ement s
in t he kernel of o pl ay no real role. Thus it is r easonabl e to assume o = 1, i.e. to i mpose
a furt her non-t ri vi al i t y axi om:
(4c) Z(Y. I) is the i dent i t y.
180 MICHAEL ATIYAH
Let us now deduce a few el ement ar y consequences of our axioms. Appl y axi om (1)
wi t h M ---- M' = Y. x I and
F : M~ M'
a h o mo t o p y f t of maps Y. --+ Y. We deduce t he
homotopy invariance of Z ( f ) : Z(Y 0 ~ Z(Y.).
Thi s implies t herefore tb.at t he gr oup F( Z) of component s of t he or i ent at i on preservi ng
di f f eomor phi sms of Z acts on Z( E) .
Next let us not e t hat when M is a closed (d + 1)-di mensi onal mani f ol d so t hat
a M -= r t hen
Z( M) ~ Z(o) = A
is a const ant (el ement of t he gr ound field). Thus t he t heor y pr oduces in par t i cul ar
invariants of cl osed (d + l )-mani fol ds. Mor eover i f we cut M al ong Z into t wo part s M1, M2
so t hat we get t wo vect ors
Z(M1) z(y,),
Z( M, ) e Z( Z) ~
t hen, as a special case of t he mul t i pl i cat i ve axi om (3b), we get
Z( M) = ( Z( M1 ) , Z( M, ) ).
Thus t he i nvar i ant for a cl osed mani f ol d can be c omput e d in terms of such a decom-
posi t i on.
I f we vi ew Z( M) , for cl osed M, as a numer i cal i nvar i ant of M, t hen for a mani fol d
wi t h bounda r y we shoul d t hi nk of Z( M) E Z( aM) as a " rel at i ve " i nvari ant . The whol e
t heor y is t hen concer ned wi t h these i nvari ant s and t hei r formal propert i es.
I f we f or m t he pr oduct mani t bl d Z X S x by i dent i fyi ng opposi t e ends of t he cyl i nder
Z x I t hen our axi oms i mpl y t hat
Z( Z S 1) = Tr a c e ( I d[ Z( Y. I))
= di m Z( Z I).
Mor e general l y let f : E E be an or i ent at i on preservi ng di ffeomorphi sm, and i dent i fy
opposi t e ends of Z x I by f l Thi s gives a mani fol d Z, I and our axi oms i mpl y
Z(E~,) = Tr ace Z ( f )
wher e Z ( f ) is t he i nduced aut or nor phi sm of Z( Z) .
The obser vant r eader will have not i ced t hat our i nvol ut or y axi om refers onl y
t o reversi ng t he or i ent at i on of t he d-di mensi onal mani fol ds Z. Not hi ng has been sai d
so far about t he effect of or i ent at i on reversal on t he (d 1)-di mensi onal mani fol ds M.
I n par t i cul ar our axi oms gi ve no rel at i on as yet bet ween the i nvari ant s Z( M) and Z( M ~
for cl osed manifolds. Such a rel at i on may or ma y not exist dependi ng on t he t heor y
TOPOLOGICAL Q,UAN'IXJM FIELD THEORIES 181
(as we shall see in w 3). However , in ma ny i nt erest i ng cases t her e is such a rel at i on. Thi s
depends on t he additional assumption t hat our vect or spaces Z( Z) have a non- degener at e
hermitian st r uct ur e (rel at i ve to a conj ugat i on on A). Thi s gives a nat ur al i somorphi sm
Z ( S ) " ~ Z(X)
wher e V denot es V wi t h t he conj ugat e act i on of A. We can now consi der t he ext r a
hermitian axiom
(5) Z( M' ) = Z( M) .
I f 0M = Y~0 u Y~I so t hat Z( M) can be vi ewed as a linear t r ansf or mat i on bet ween
her mi t i an vect or spaces:
Z ( M ) : Z(Xo) Z ( X l ) ,
t hen axi om (5) is equi val ent to
(5' ) Z( M' ) is t he adj oi nt of Z( M) .
I n par t i cul ar for a closed mani f ol d M (5) asserts t hat t he numer i cal i nvar i ant Z( M)
changes to its conj ugat e under or i ent at i on reversal. Unl ess all val ues are real (fixed by
conj ugat i on) t hen these i nvari ant s can " d e t e c t " ori ent at i on.
Not e t hat for a mani f ol d IV[ wi t h boundar y ~ we can al ways form t he double
M u s M" whi ch is a cl osed mani fol d. Axi om (5) shows t hat
Z( M w z M ~ = I Z( M) I '
wher e on t he ri ght we comput e t he nor m in t he her mi t i an (possibly indefinite) met ri c.
These axi oms can be modi fi ed in vari ous i mpor t ant ways and this is necessary
for many of t he i nt erest i ng exampl es. Let me briefly i ndi cat e t he ki nd of modi fi cat i on
t hat can be i ncor por at ed.
(1) the vect or spaces Z( Z) may be mod 2 gr aded wi t h appr opr i at e signs t hen inserted,
(2) t he mani fol ds E, M may car r y mor e st ruct ure, e.g. a spin st ruct ure, a f r ami ng or
some di st i ngui shed homol ogy classes,
(3) one can consi der a " rel at i ve " t heor y for a pai r Xd_ , C Zn consisting of E and a
submani f ol d X. Thi s will have to coupl e t oget her t opol ogi cal Q FTs in di mensi ons d
and d - - r,
(4) we mi ght want to al l ow Z n to have a boundar y: this is closely r el at ed to (3) above
wi t h r = 1.
A mor e serious modi fi cat i on is
(5) al l ow Z(Y.) to be i nfi ni t e-di mensi onal .
The axi oms in this case cert ai nl y need significant changes. For exampl e cert ai n quant i t i es
become infinite (e.g. Z(Y~ S 1) ~ di m Z( E) ) .
182 MI C HAE L AT I YAH
Thi s list is not meant to be exhaust i ve par t i cul ar l y i f we move away f r om r equi r i ng
our QF T to be strictly t opol ogi cal . For exampl e conf or mal field theories are cl earl y
rel at ed to t he ideas here (see w 3).
So far in this sect i on we have del i ber at el y refrai ned from at t empt i ng to descri be
t he physi cal i nt er pr et at i on or t er mi nol ogy (except for t he acr onym OFT) . Thi s was
meant to emphasi ze t he mat hemat i cal nat ur e of t he pr esent at i on in or der to encour age
mat hemat i ci ans to t ake these ideas seriously. However , we shoul d now rect i fy t he si t uat i on
by briefly i ndi cat i ng t he physi cal backgr ound.
Z is meant to i ndi cat e t he physi cal space (e.g. d = 3 for st andar d physics) and
t he ext r a di mensi on in Z x I is " i magi nar y " t i me. The space Z(Y.) is t he Hi l ber t
space of t he qua nt um t heor y and a physi cal t heory, wi t h a Hami l t oni an H, will have
an evol ut i on oper at or e " a or an " i magi nar y t i me " evol ut i on oper at or e- ~a. The mai n
f eat ur e of t opol ogi c al Q FTs is t hat H = 0, whi ch implies t hat t here is no real dynami cs
or pr opagat i on, al ong t he cyl i nder Y~ I. However , t here can be non-t ri vi al " pr opa-
gat i on " (or t unnel i ng ampl i t udes) from E0 to Y'l t hr ough an i nt erveni ng mani f ol d M
wi t h 0M = Y'0 u Zx: this reflects t he t opol ogy of M.
I f OM = Z, t hen the di st i ngui shed vect or Z( M) in t he Hi l ber t space Z( Z) is
t hought of as t he v a c u u m s t at e defi ned by M. For a cl osed mani f ol d M t he numbe r Z( M)
is t he va c uum- va c uum expect at i on val ue. I n anal ogy wi t h statistical mechani cs it is
also cal l ed t he par t i t i on funct i on.
The r eader may wonder how a t heor y wi t h zero Hami l t oni an can be sensibly
f or mul at ed. The answer lies in t he Feynman pat h- i nt egr al appr oach to QFT. Thi s
i ncorporat es relativistic i nvari ance (whi ch cat ers for general (d + 1)-di mensi onal " space-
times ") and t he t heor y is formal l y defi ned by wri t i ng down a sui t abl e La gr a ngi a n- - a
funct i onal of t he classical fields of the t heory. A Lagr angi an whi ch involves onl y first
deri vat i ves in t i me formal l y leads to a zero Hami l t oni an, but t he Lagr angi an itself may
have non-t ri vi al feat ures whi ch rel at e it to t he t opol ogy.
For a fuller under st andi ng of t opol ogi cal Q FTs from t he physical vi ew-poi nt t he
r eader shoul d consul t t he paper s of Wi t t en on t he subj ect .
3. Examples
We shall list a number of i nt erest i ng exampl es of t opol ogi cal QFTs in di mensi ons
d~< 3. The descri pt i on will i nevi t abl y be br i ef and it shoul d be emphasi zed t hat t here
are many poi nt s t hat have yet to be fully i nvest i gat ed in some of these theories. Some
part s exist in a fully ri gorous mat hemat i cal sense whi l e ot her part s have as yet onl y
been t r eat ed formal l y. Nevert hel ess t he general pi ct ur e is ver y convi nci ng and t here
seems little doubt t hat the essential features are correct .
Nat ur al l y t he theories in general increase in difficulty wi t h t he di mensi on d. We
shall begi n wi t h the appar ent l y trivial case of d = 0 and progress up to d = 3.
TOPOLOGICAL QUANTUM FIELD THEORIES 183
( 3 0 ) = 0
Space Z now consists of finitely many (say n) points. To a single poi nt we must
associate a vect or space V ~-- Z (poi nt ), and to n poi nt s we associate t he n-fold t ensor
pr oduct : V | = V | V | . . . | V. The symmet r i c gr oup S, (di ffeomorphi sms of n points)
t hen acts on V | Thus t he classical t heor y of represent at i ons of S, appear s as a basic
i ngrcdi ent in theories for d = 0.
The quest i on now arises as to t he origin of t he vect or space V, t he Hi l ber t space
o f t he qua nt um t heory. A st andar d way to get t he qua nt um Hi l ber t spacc is first t o
gi ve a classical sympl ect i c mani f ol d (or phase space) and t hen to quant i ze this. I n par -
t i cul ar an i nt erest i ng class of exampl es arise from compact Lie groups G and t hei r homo-
geneous sympl ect i c mani fol ds; thesc arc co- adj oi nt orbits, generi cal l y copies of t he flag
mani fol d. I f we t ake " i nt e gr a l " orbi t s for whi ch t he sympl ect i c st ruct ure comes from
a l i ne-bundl e, t hen quant i zi ng leads to t hc i rreduci bl e represent at i ons V of G. Thi s
is t he physi cal i nt er pr et at i on of t he Borcl -Wci l t heor em whi ch is usual l y f or mul at ed
i n al gebr ai c- geomet r i c l anguage. The Lagr angi an of these theories is t he classical act i on
( hol onomy of t he l i ne-bundl e).
Thus t opol ogi cal QFTs wi t h d = 0 arise nat ur al l y in rel at i on to classical repre-
sent at i on t heor y of Li e groups and t he symmet r i c gr oup. I n this l ow di mensi on t her e
is no i nt erest i ng t opol ogy, onl y qua nt um symmet ri es.
( 3 . 1 ) d = 1
Ther e are t wo r at her di fferent t ypes of t heor y in tiffs di mensi on, bot h of whi ch
ar e l i nked to t he Li e gr oup theories in di mensi on zero. We descri be these in turn.
a) Floer/Gromov theory
Her e t he classical phase space consists of pat hs, in a compact sympl ect i c mani fol d X,
wi t h appr opr i at e bounda r y condi t i ons. To fit t he formal f r amewor k we set up in w 2
we shoul d consi der peri odi c boundar y condi t i ons gi ven by closed loops in X. Hol onomy
r ound such loops, used in ( 3. 0) as a Lagr angi an, is now used to modi f y t he Hami l t oni an
as i n Wi t t en I11]. For a cl osed surface M t he i nvar i ant Z( M) of t he t heor y is t he numbe r
o f pseudo- kol omor phi c maps M -~ X in the sense of Gr omov [5] (e.g. t hey are or di nar y
hol omor phi c maps i f X is a Kfi2der mani fol d). I f this numbe r is infinite, i.e. i f t here ar e
" modul i ", t hen we must fix f ur t her da t a on M. Thi s can be done by pi cki ng some
poi nt s P~ and t hen l ooki ng at hol omor phi c maps f : M --)- X wi t h f(P~) const r ai ned to
lie on a fixed hyper pl ane.
Wi t t en [14] has wr i t t en down t he r el evant Lagr angi an for this t heory. Fl oer [-3]
has gi ven a ri gorous t r eat ment , based on Wi t t en' s Mor se t heor y ideas, for t he case when
t he boundar y condi t i ons are not per i odi c but i nst ead r equi r e t he initial and end-poi nt s
o f pat hs to lie on t wo fixed Lagr angi an sub-mani fol ds. Thi s is a case when Z is an i nt erval ,
r at her t han a circle, and is a modi fi cat i on of t he t ype (4) above. The Fl oer t heor y is
nat ur al l y Mod 2 gr aded and is defi ned over t he integers.
184 MI CHAEL ATIYAH
b) Itolomorphic Conformal Field Theories
These are not strictly topological QFTs in our sense since they depend on a complex
structure, and the Hilbert spaces are infinite-dimensional. However, t hey are closely
related to topological QFTs. They have been axiomatized by G. B. Segal [10] and,
as ment i oned in w 1, his axioms motivated our version.
There are conformal field theories related to compact Lie groups G in which
the classical phase space consists of co-adjoint orbits of (a central extension of) the loop
group LG. Q uantizing these produces the Hilbert spaces of the t heory as irreducible
(projective) representations of LG. The whole theory is very similar to t hat in (3.0).
The group Diff+(S 1) now substitutes for the symmetric group and plays an i mport ant
role (see [9] for details). The partition function in such theories depends on complex
structures: it is not purely topological.
( s . 2) d = Z
There are a number of similar theories in tiffs dimension. We begin with perhaps
the most interesting and well-developed theory.
a) ]ones/Witten theory [12]
Here the classical phase space, associated to a closed surface ~, is the moduli space
of flat G-bundles over Y~. The Lagrangi an is an integer multiple of the Chern-Simons
function of a G-connection on a 3-manifold (which has to be " framed "). The integer
multiple k, called the level, is a paramet er of the t heory and k ~ oo gives the classical limit.
This theory can be naturally coupled with the d = 0 theory to produce a " relative "
t heory of the type indicated at the end of w 2. The details have been developed by Witten
who shows that the partition function for a (framed) link in the 3-sphere is just the value
of the Jones polynomials [8] for a suitable root of unity. The theory can be defined over
the relevant cyclotomic field.
Instead of coupling the d = 2 t heory to d = 0 we can couple it to the d = I
conformal theory in b) above, by considering Ri emann surfaces with. boundary.
b) Casson theory
Here th.e " Hi l ber t " spaces of the theory are essentially the homology of the moduli
spaces of flat G-bundles over Y~ (whereas in the Jones/ Wi t t en t heory one takes a holo-
morphic analogue). The theory is rood 2 graded and the hermi t i an forms are indefinite
(being given by Poincar~ duality). Witten has recently written down a Lagrangi an
for this Casson theory. The invariant Z, for homol ogy 3-spheres is the original Casson
invariant. The theory is defined over the integers. As yet there are details, related to
the singularities of the moduli-spaces, whi ch have not been fully worked out.
c) Johnson theory
Recently D. Johnson [7] has developed a t heory imitating Casson, but using
Reidemeister torsion as an essential ingredient. Witten [16] has shown t hat Johnson' s
TOPOLOGICAL QUANTUM FIELD THEORI ES 185
t heor y can be expl ai ned as a topological Q FT for a Cher n- Si mons Lagr angi an. However,
t he gr oup i nvol ved is not a compact Li e gr oup G but t he semi -di rect pr oduct I G of G
wi t h its Li e al gebr a (viewed si mpl y as abel i an group). The classical phase space is t he
modul i space of flat I G- bundl es and is not compact , so t hat t he Hi l ber t spaces of
this t heor y ar e not fi ni t e-di mensi onal . Thus t he par t i t i on f unct i on may somet i mes be
infinite.
d ) " Thurst on " theory
Wi t t en has r ecent l y consi dered [15] a Cher n- Si mons t heor y for t he non- compact
gr oup SL(2, C) in rel at i on wi t h gravi t at i onal theories in (2 + 1)-dimensions, and hence
wi t h hyperbol i c 3-manifolds. Thi s shoul d make cont act wi t h Thur st on' s wor k in due
course. The classical phase space is agai n t he ( non- compact ) modul i space of flat
SL(2, C) bundl es over a surface 2;. The Hi l ber t spaces are agai n i nfi ni t e-di mensi onal .
Remark. m Modul i spaces of flat SL(2, C) bundl es ( and t hei r general i zat i ons) have
been consi dered in a r emar kabl e paper by N. J. Hi t chi n [6]. Hi t chi n shows t hat this
modul i space is nat ur al l y fi bered by Abel i an varieties and t hat t he modul i space of flat
SU( 2) - bundl es appears as a component of t he most degener at e fibre. On t he basis of
Hi t chi n' s paper it seems likely t hat all t he theories descri bed in this sub-section can,
i n a sense, be r educed to t he abel i an case. Thi s is now bei ng i nvest i gat ed and shoul d
have maj or consequences.
( 3. 3) d -~ 3 (Fl oer/ Donal dson theory)
Donal dson [2] has defi ned i nt eger i nvari ant s of smoot h 4-mani fol ds by using
modul i spaces of SU(2)-i nst ant ons. These i nvari ant s ar e pol ynomi al s on t he second
homol ogy. Thus we shoul d consi der 4-mani fol ds wi t h ext ra dat a (consisting of an el ement
of t he symmet r i c al gebr a of H, ) . Wi t t en [13] has pr oduced a super - symmet r i c Lagr angi an
whi ch f or mal l y r epr oduces t he Donal dson t heory. Wi t t en' s f or mul a can be under st ood
as an i nfi ni t e-di mensi onal anal ogue of t he Gauss-Bonnet t heor em (as I shall expl ai n
elsewhere).
The Hami l t oni an version of t he t heor y has been devel oped by Fl oer [4] in t erms
of t he space of connect i ons on a 3-manifold. Fl oer uses t he Chern-Si mons funct i on (the
Lagr angi an of t he Jones/ Wi t t en t heory) to modi fy t he Hami l t oni an (see t he r emar ks
i n (3. 1) a ) ) . The abel i an groups he defines have a rood 2 gradi ng.
I have descri bed this Fl oer / Donal dson t heor y at gr eat er l engt h in [1]. Her e I
j ust wish to emphasi ze t hat it f or mal l y fits some version of t he axioms of w 2 and t hat
t her e is a Lagr angi an formul at i on.
Thi s t heor y is defi ned over t he integers. I t d o e s n o t satisfy any axi om of t ype (5)
of w 2. The Donal dson i nvari ant s of a 4-mani fol d wi t h its two di fferent ori ent at i ons
have no obvious r el at i on to each ot her. Inst ant ons and ant i -i nst ant ons ar e qui t e di fferent ,
par t i cul ar l y when t he si gnat ure of t he 4-mani fol d is non-zero.
94
186 MI CHAEL ATI YAH
Wi t t en [14] has also shown how one can coupl e t he d = 3 and d = 1 theories
together: this is qui t e anal ogous to t he coupl i ng bet ween d = 2 and d = 0 in t he J ones
t heory.
C o n c l u s i o n . - - These exampl es, whi ch have nat ur al geomet r i c origins, and cover
many of t he most interesting topics in l ow-di mensi onal geomet r y show t hat t opol ogi cal
QFTs have real rel evance to geomet r y. Ther e are still many t echni cal pr obl ems to
be sol ved (e.g. t he Casson i nvar i ant has so far onl y been t r eat ed for G = SU( 2) ) and
t her e are many i nt ri gui ng quest i ons. For exampl e t he Cher n- Si mons funct i on appear s
in bot h d = 2 and d ---- 3 theories, but pl ayi ng qui t e a di fferent rol e each t i me. Wha t
is t he significance of this?
RI' ;FERENCES
[1] M. F. ATIyAH, New invariants of three and four di memi onal manifolds, i n The Mathematical Heritage of Herman
Weyl, Proc. Symp. Pure Math., 48, American Mat h. Soc. (1988), 285-299.
[2] S. K. DOnAI~SON, Polynomial invariants for smooth four-manifolds, to appear i n Topo/ogy.
[3] A. FLOZR, Morse theory for fixed points of symplectic diffeomorphisms, Bull. A. M. S. , 16 (1987), 279-281.
[4] A. FLOER, An imtanton invariant for three manifolds, Courant Institute preprint, to appear.
[5] M. GROMOV, Pseudo-holomorphic curves i n symplectic manifolds, Irwent. Math., 82 (1985), 307-347.
[6] N. J . HITCHIN, The self-duality equations on a Ri emann surface, Pror London Math. Soe. (3), 85 (1987), 59-126.
[7] D. Jom~son, A geomarit form of Casson's invariant and its conntxtion with Reidaneister torsion, unpublished lecture
notes.
[8] V. F. R. JonEs, Hecke algebra representations of braid groups and l i nk polynomials, Ann. of Math., 19-8 (1987),
335-388.
[9] A. PRESSnZY and G. B. SgOAL, Loop Groups, Oxford Univemity Press (1988).
[10] G. B. SEOAL, The definition ofconformalfield theory (to appear).
[I 1] E. Wla'amN, Super-symmetry and Mome theory, J . D/ft. Geom., 1' / (4) (1982), 661-692.
[12] E. WXT'r~.N, Qua nt um field theory and the Jones polynomial, Comm. Math. Phys. (to appear).
[13] E. WXTa'ZN, Topological quant um field theory, Comm. Math. Phys., 117 (1988), 353-386.
[14] E. Wn~rEN, Topological sigma models, Comm. Math. Phys., 118 (1988), 411-449.
[15] E. WXTTEN, 2 + 1 dimensional gravity an an exactly soluble system, Nucl. Phys. B, 811 (1988/89), 46-78.
[16] E. WtTrEN, Topology changing amplitudes in 2 + 1 dimensional gravity, NuM. Phys. B (to appear).
[17] E. WrrrEN, Elliptic genera and quant um field theory, Comm. Math. Phys., 109 (1987), 525-536.
The Mat hemat i cal I nst i t ut e
Uni ver si t y of Oxf or d
24-29 St Giles
Oxf or d OX1 3LB
UK
Manuscrit re~u le 9 fi vri er 1989.