Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Annals of Mathematics is collaborating with JSTOR to digitize, preserve and extend access to Annals of
Mathematics.
http://www.jstor.org
1. Introduction
> ~kk
*E or(dk) r*k+1
C*Ek?
Y(dk+l)
(d+) r*k+2
7C*E
is exact over the complement of the zero section in TX. These data we then
call an "elliptic complex" 7(E) over X.
We also definethe homologyHk{Ip(E)} of such a complex in the customary
manner by
(2.6) Hk{F(E)} = Ker dk/Image
dkl.
The de Rham complex is of course real but to fit in with our general formulation
we consider the complexification.
($(Ai) denoting the sheaf of germs of C- sections of Ai) from which our
interpretationof Hk(f *) follows readily. In short, L(f *) is just the classical
Lefschetz number of the map f.
Let us now compute the multiplicityv(P) of a fixed point for this com-
plex. According to (2.9) we have
v(P) = E (-1)k trace (Xkdfp*)/l
det (1 -dfp) I
On the other hand if A is any endomorphism of a finite dimensional vector
space and xkA and A* denote its kthexterior power and transpose respectively,
then it is well known that
-)k trace (XkA) = det(1 - A)
and that
det (A*) = det (A) .
It follows that our multiplicityreduces to
det(1 - dfp)
-(P) -
where T'X is spannedby the dzk and T"X by the d~k of a local holomorphic
coordinatesystem. The bundle T'X thereforehas a natural holomorphic
structureand T"X a naturalantiholomorphicstructure.
Correspondingto the decomposition(4.1) the bundlesof the complexified
de Rham complexdecomposecanonicallyintothe tensorproduct
X*(C OR TX) = X T'X 0c a* T"X
so that, in particular,
Ak =
Ep+q=k+ q=
with
(4.2) APsq = XP(T'X) 0c Xq (TPX)
so that Hq(X; /9(F)) _ Hoq(X; F). Using (4.2) it is then also clear that
Hq(X; O(F 0 AP,')) -HPq(X; F), so that the d" complex of a holomorphic
bundle is the basic concept.
Consider now a holomorphic map f: X X. The natural lifting of f to
A* is then compatible with d" and therefore induces endomorphismsf P * in
each of the complexes F(AP *). The correspondingendomorphismin homology
will be denoted by HP lq(f ) so that the Lefschetz numbers of f P * are given by
(4.4) L(fP*) = (-_ )q traceHPq(f)
We note that in view of (4.2) and (4.3), L(f Pa*) may also be interpretedas
the Lefschetz number of the endomorphism induced by f in the sheaf co-
homology H*(X; O(AP'?)).
Let us now compute the multiplicityof a transversal fixed point P of f
relative to the complex AP *.
Because f is holomorphicthe complexificationof df preserves the decom-
position (4.1) so that
1 0 dfp = d'fp e dep
with d'fp and d"fp endomorphismsof TpX and Tp'X respectively. It follows
that
Xk{1 0 dfP} = Ep+q=k X\{dfpf} 0 X {dffp} .
Hence the multiplicityv(P) in question is given by:
I det (1 - dfp) I v(P)
. = Eq (-)q trace XP(d'fp*) trace Xq(dffp*) .
This now yields2
. detc (1 -
(4.5) v(P) = Tracec (-XPd'fp*) dfp*)
IdetR( - dfP) I
Finally observe that under the bar operation in C OR TpX, TpJ'Xis taken
into Tp'X. Furthermore 1 0 dfp clearly commutes with this operation. It
follows that
detc (1 - d'ff) = detc (1 - duf )
whence in particular
(4.6) detR (1 - dfp) = detc (1-1 0 dfp) = detc (1 - d'fp) detc (1 - dfff)
is positive. The absolute value sign in (4.5) is therefore redundant and one
obtains the formula:
In this case then only half of the determinantal factor in the denominator
of (2.9) is cancelled out by the numerator. Note also that v(P) can be ex-
pressed without introducing the complexificationof TX. Indeed, over R,
TpM is isomorphic to TpM and so inherits a complex structure. Hence if A
is any complex linear endomorphism of TpM we can speak of its complex
tracec A, and determinant, detc A. At a fixed point P of the holomorphic
map f, df, is of course such an endomorphismand considered as a C-endomor-
phism agrees with d'f,. Hence in terms of these notions one has
(4.8) v(P) = tracec XP(dfp)/detc(1- dfp) .
To recapitulate: for a transversal endomorphismf of a complex analytic
manifold X, our fixed point formula specializes to
(4.9) L(f P *) = Ef (P)P tracec (xPdfp)/detc(1 -dfp)
where
(4.10) L(f*) = (-_)q traceHPq(f)
It is this formula which Shimura conjectured during a conference at
Woods Hole in 1965, and which furnished the impetus for this work. For
curves (4.9) had already been established by Eichler in [11]. Shimura and
Eichler were of course thinking in the framework of algebraic geometry.
There it turned out that the full duality theory of Serre and Grothendieck
yields this result even over arbitrary characteristic.
The formula (4.9) now has an easy extension to the d"-complex of an
arbitrary holomorphicbundle F. To lift a map f to this complex, one only
needs a holomorphic bundle homomorphism
A9:f *F >F .
In terms of it
-
0 Xk(d"f )*: f *(F O Aok) F? AO,k
then serves to definethe k1h lifting of f. We write T(f, p) for the induced
endomorphismof F(F 0 A',*) and Hq(f, P) for the induced homomorphismin
Hq(X; F). We also set
(4.11) L(f, A) = A (-1)q trace Hq(f, q)
for its Lefschetz number. It is then clear that the multiplicityof P relative
to this complex is given by
v(P) = tracec wp/detc(1 - dfp)
Thus the holomorphiccase of our general Lefschetz theorem takes the form
THEOREM4.12. Let X be a compact complex manifold, F a holomorphic
(4.15) =Res,
i -f'(P) - f(u)
The right hand side of (4.15) makes sense also for non-transversalmaps and
turns out to be the correct multiplicityof a general fixed point in the sense
that the formula
(4.18) 1ji==-fl
detazp
Assumenow thatthe fixedpointsof f are all non-degenerate, i.e., thatf
is transversal. On C" this is alreadythe case by our firstassumptionand on
P' it can always be arrangedbyan arbitrarilysmall deformation of the terms
of degree d in the ga, so that this case easily impliesthe generalone. With
this understood,let I' be the fixedpoints of f on P'. Applying(4.9) with
p = 0 and using the fact that HO?q(P) = 0 for q > 0, H0'0(P) = C, we have the
formula:
(4.23) 1= Spe2 1/detc (1 - dfp) + Spea' 1/detc (1 - dfp)
Next let f' be the restrictionof f to P'. Applying(4.9) to this endomor-
phismyields
(4.24) 1 = Epsz, 1/detc (1 - dfp)
Finally, it is easily checked that if d > 1, then detc (1 - df') = detc (1 - dfp)
and so (4.20) followsfrom(4.23) and (4.24).
7r a
G/H - c*F
each x e G determinesa linearisomorphism
Furthermore,
jo: F -) (C*F)(X)
by setting j((f) = u(x, f).
We come now to the induced action of G on P(c*F). Let Lg and tgdenote
left translation in G and G/H respectively. Clearly
Lg x 1:G x F - G x F
preservesthe fibersof a and henceinducesa map
Lg x1:C*F )CFF
which maps the fiberover 1gil*x linearlyinto the fiberover x. Hence 9g =
Lg x
, 1 may be interpreted as a lifting of the map 19-1. The resulting
endomorphism r(qpg)1*- of r(c*F) is denotedby Tg, and g o Tgis the desired
.
inducedrepresentation:
Tg.
C*(P(g)=
(where (G/H)x denotes the tangent space to G/H at x). The relation
Lg1 o Rh-1 o Lx = Lx o Lh ? Rh-i then implies that under (5.6) dlg-i goes over into
AdG(h), and so establishes(5.4). To see (5.5) considerjx:F -(4*F)x. We have
ji(f) = cr(xf). Hence9g ?jx(f) = c(gxf) = c(xx-lgxf) = PF(h-)j((f). q.e.d.
The expressions(5.4) and (5.5) occurin manybranchesof representation
theory. As an examplelet us show how the "HermannWeyl formula"fits
into the framework of TheoremA and Proposition(5.3).
We assume then,that G is compact,and that H = T is a maximaltorus
of G. In this case g/t breaksintoa directsum of 2-planes,ek, k = 1, ..., m,
on each of whichAdG acts by rotations
m
(5.7) g/i = ekE
Apart from some minor rewriting (5.11) is precisely the Hermann Weyl
formula. Indeed to bring(5.11) intoa morefamiliarform,assume that G is
simply connected.Then the productof the positiverootsturnsout to have
a square roota, whencethe functionA: T V C given by
A = a-fl(1 - /3k)
Trace Q = 5Trace
KQ(x, x)dx.
Tu- Tgt(g).
J G
- 50C(g)
G
Trace' Tg
(6.11) Da = -aD,
and has the multiplicative property that under the natural isomorphism
X*TX0X*TY--X*T(Xx Y)
(6.12) axxy = 0 ay .
Finally consider the action of a 0 jn on the complexificationof X*TX.
Under this action the total space breaks up into a direct sum
(6.13) X*TX0C = X+TX(X-TX
and we call the resulting set of angles {Ok} a coherent system for dfp. This
understood the multiplicityformula we are seeking is given by
(6.25) 2(P) = X f1 cot (Ok!2)
where {Ok} is a system of coherent angles for df,.
To prove (6.25) consider first the two-dimensionalcase. Then *e1 = e,
*e? - -e1 whence X+T, is spanned by 1 + ie1 A e' and e1 + ie', while - T, is
spanned by 1 - ie1 A e' and e1- ie?. Furthermore df,(el + ie?) = e`18(el+ iee)
and dfp(e, - ieD) = ei0(e1- ie') while the other elements remain fixed under
x*dfp. It follows that trace (fp*oa (0 i) = e-0 -e+iO, so that
of Atiyah-Segal) and is described in [4], [5], and [6]. Thus this method of
attack is considerably less elementary than the one under discussion here.
Furthermore it is based essentially on the fact that the group of isometries
of a Riemann structure is compact, and hence does not extend to cover other
Lefschetz problems, even with only transversal fixed point sets.
7. Two applications
Theorem (6.27) imposes strong number-theoretic restrictions on the
angles of an isometry at its fixed points. We give two illustrations of this
fact in this section. As our firstand simplest example we have
THEOREM7.1. Let X be a compactconnectedand orientedmanifold(of
positivedimension),and let f: X X be an automorphismof X, of prime
-
(7.11) r [ n I
Hence 9(x) = JJ
(x- i), Xe Z*; and so in particular
(7.12) H(2'n)=1 (1
-
) =P
It is clear that 1 - divides 1 - d in Z[fl. But for any X E Z,* E' is again
primitiveso that ($)r = $ forsomer, and so 1 - d divides1 - d. It follows
that
(7.13) (1- = (1 - d)-unit
In particular,(7.12) impliesthat
(7.14) (1 - p.)n(11/P)= puunit, n= p1
Let us now set p = 1 - d. Applying (7.13) in the formulafor ((p) we find
U(P) =-u
q*a(p) c Z[$] qu c ~mZ[f]. Then (7.14) implies (by an argument used in (7.1))
that q mustbe divisibleby pr+l where
r = n/) q.e.d.
So far in these applicationswe have onlyused the fact that Sign (f, X)
is an algebraicinteger. Its cohomologicalinterpretationwas not used, and
so we did not exploitthe fullforceof our Lefschetzformula. A moreinter-
estingapplicationwhichreallyuses the fullLefschetzformulais the follow-
ing theoremwhichconfirms an old standingconjectureof P. A. Smith.
THEOREM7.15. Let p be an odd prime and consider a smooth action of
Zp on a homology sphere which has precisely two fixed points. Then the
induced representations of Zp on the tangent spaces of the two fixed points
are isomorphic.
PROOF. Let f generatethe actionand choosea riemannianstructurefor
S2non whichf acts as an isometry.Also let P and Q be the fixedpoints of
f. We have to show that dfpand dfQhave the same set of eigenvalues. Be-
cause (dfp)P = 1 these eigenvalueswill all be pth rootsof unity. Hence if we
set e e2ri/p, and let aP denote the number of eigenvalues of dfp which are
-
2)(P) ( = I 1 + Ik
1 -k
where ek range over all the eigenvaluesof df,. In termsof our multiplicity
function a' we thereforehave
(7.18) | V(P) 12= II {(1 + $,)/(1 - e )i
(7.19) llPP)=i{ } -
(7.21) EaA = 0
and,
(7.22) fl(1 - i) = 1,
thena,_ 0.
Fromthistheoremone deducesours in the followingfashion. The condi-
tions(7.20) and (7.21) are triviallymet by our an. Using the relation
(7.23) (1 + $)/(1 - $) = (1 - V2)/(1- ?
into
the condition(7.19) is transformed
(7.24) IH (1 _ ~22)all (1 _ hi)-2a; 1
(7.30) a, = 0
and if for everydivisor d of n, d + n,
(7.31) a| (1-idi)a, = 1
thena, _ 0.
Consider now a generator f for our group Zn and apply Theorem 6.27 to
the powers f d, where d ranges over the divisors of n not equal to n, or n/2.
By assumption all these fd act freely on X - {P U Q}, whence, as none of
them has order 2, the multiplicities vp and vQrelative to fd will be non-zero.
Thus, if aiP and aQ are the multiplicityfunctions of dfp and dfQrespectively
and a, is their difference,our earlier argument leads to the condition
II ( +Ad)
(7 .32) G ,
Now for every d' I m, d' + m, we may apply (7.32) with d = 2d', to obtain
II {(1 + <22d')/(l _ <22d')}a2 =1 X Z2c
G
On the other hand here the terms X and X + m may be lumped together
whence one obtains
JJ{(1 + $22d')/(J _ <22d')}a2+a2+m 1 , ZM .
_ fi {1 -id}-2a .
Composedwiththe covariantderivative
(8.9) V+: F(A?F) > F(TX? AFF)
whichthese bundlesinheritfromthe canonicalconnectionon F, P therefore
gives rise to operators
(8.10) 6+: F(A+F) > F(AT)
and these are by definition the Dirac operatorswe were seeking. The opera-
tors are ellipticforthe followingreason. The pairing[+ inducesa map
It
(8.11) TX > Hom (A+F, A-F),
which is non-singular in the sense that ft,() is an isomorphism for e + 0.
Indeedone has
(8.12) Pt = Identity(, $)
On the otherhand observethat the symbolof the covariantderivative
V+ is inducedby the identitymap
(8.13) TX?OA+ - TX ? A+,
so that the symbolof 3+ is simply given by [(e(). Thus (8.12) implies that 3+
is non-singular. Similary for 8-. The pertinent facts concerning spinors we
have used here can be found in [3], however, for completeness, we will now
very brieflyreview the constructionof some of these objects.
If E is a real vector space with positive definiteinner product, cE shall
denote the Cliffordalgebra of E. Thus cE is definedas the quotient of the
full tensor algebra over E modulo the ideal I generated by elements of the
form e 0 e + (e, e) 1. Alternatively cE is the free algebra generated over R
by a unit 1, and an orthogonal base {e1, ..., en} for E, subject to the defining
relations
(8.14) e~,= -1, ejek + ekej =0 j k.
As a module over R, cE is then seen to be spanned by the products ej1 ... eJk,
il < i2 . . . < Jk, 0< k < n so that, qua R module, cEis isomorphicto the exterior
algebra X*E. Multiplicatively this is not so, but cE does inherit the structure
of a Z2-graded module c(E) = c+E + c&E where c+E and c&E are additively
generated by the even and odd products respectively. Furthermore E is
naturally included in cE.
The group Spin (n) exists naturally as a subgroup of the group of invert-
ible elements c*E of cE. Indeed let x + x be the anti-automorphism which,
actingon ourR-basis,sendsej1... eJkinto(-l)ke k ... ej1. ThenSpin(n) is the
subgroup of c*E characterized by
(8.15) x c+E
(8.17) ~.X = 1.
If follows from (8.15) and (8.16) that for all x e Spin (n), the transforma-
tion a(x): E E defined by e i- xex-' is an orientation preserving isometry
of E. Hence a maps Spin(n) into SO(n) and it is not difficultto show that a
is onto. In fact if' n = 2m and (f1,f/,* *, fim fm) is an orthonormalframe for
.
and
(8.22) ax = -xa for all x C c-E
while,
(8.23) ax = xa for x ccE.
This model of Spin (2m) c cE furnishes us with a natural representation
(8.24) c+: Spin (2m) - Aut (c+E),
given by the left action of c+E on c+E. This representation is far from
irreducible. Rather its complexificationbreaks up into 2m-1copies of A+ and
i\-
pairing (8.7), that this pairing satisfies the condition (8.12) etc. Note finally
that the element a = el *
...e2 is not canonically definedin c+E, but depends
e2m
on the orientation in which the frame e1,e2,...* e2m is taken. The distinction
between A+ and A- in the last analysis, therefore depends on the orientation
of E.
We turn now to the Lefschetz formula of the Dirac operator
(8.31) 6+: F(A+F) - F(&-F) .
Assume then that X admits a Spin-structure F, and let f: X > X be an
isometry. The differentialof f then induces a lifting
(8.32) df*: fF )F
and it is clear that df* lifts to a bundle isomorphism
(8.33) f: f *F - F
(qua bundles over X) if and only if df * preserves the characteristic class of
F in H'(F;Z2). There are then two possible choices of f over each component
of X and they can be distinguished by their values at a given point. In
particular, if X is 2-connected, then X has a unique Spin-structure F, and
every isometryf has precisely two liftingsf: f*F F.
A liftingf now induces liftings A+(f ): f*A+(F) A+(F) and so induces
a geometric automorphismf* of the Dirac complex
fM(y) = y x(P;
y), x(P; y) C Spin (2m)
and the conjugacy class of x(P, y) is independentof y C F,. Now just as in
the homogeneouscase (see ? 5) one concludesthat traceqP and traceq4 are
determinedby the elementx accordingto
(8.39) trace PO= traceA+x(P; y) y C Fp
(8.40) traceqp = trace &-x(P; y) y C Fp .
On the other hand ax(P, y) C SO(2m) clearly representsthe matrixof dfp
relativeto the framedeterminedby y at P. It followsthat in this frame
x(P, y) has the form +x(01/2,.., Om/2)as given by (8.18) forsomesystem
(6019 *, of coherentangles fordfp. Applying(8.30) to (8.39) and (8.40),
O9m)
the expression(8.38) goes over into
(8.41) v(P) = +im2m 1ll sin (Oj/2)/1
det (1 - dfp)I
Howeverthe denominatoris given by
(1
MJJI - eiaj)(1 - e-0j)
correspondingly.
We will call two fixedpointsP and Q, f-equivalentif they have equal
multiplicity,that is if s(P, Q;f) = 1. Then we have the followingcriterion
cencerningequivalenceof fixedpoints.
PROPOSITION8.44. Let t - s(t) be a curve in FX starting at y e Fp and
ending in FQ. Let df*: FX FX denote the map induced by the differential
and consider the curve r =- df *s. This curve then has the same endpoints
ass so that the composition r-'*s is a well defined loop c: S -F, with c(l) =
Pand c(-1) = Q. Then
P and Q are f-equivalent if and only if c = 0 in w1(F; y).
PROOF. By assumptionw0(X) = wr1(X)= 0. Hence w1(F,y) Z2. Hence
the pullbackc*F of F to S' is the non-trivialor trivialdoublecoveringof the
circle, depending on whether c is non-trivialor trivial in w1(F, y). Further-
morenote that, by construction, the compositionf oa of f withrightmulti-
plicationby a inducesan automorphism of c*F over the reflectionmap z + z
of SI = {z e C I Iz I= 1}. The resultnow followsby inspection:if c*F is the
trivialcoveringand f -a is the identityover z = 1, it will also have to be the
identityover z = -1; on the other hand,if c*F is the non-trivialcovering
et- e25 and f* a is the identity over 1 then f* a is given by 0 0-8 and so is
minusthe identityat 8 = wr/2, that is at z = -1. q.e.d.
9. Exotic involutions
Considerthe hypersurfaceV(a) in C' given by the equation
(9.(9.1)
1) ~~~~~Zal
I~+.+a~
+ ***+ Zar = o
Clearly T preserves Y and has no fixed points there. The resulting invo-
lutions occur in the work of Bredon [8] and Hirsch-Milnor [14]. In fact for
m = 2, k = 3 one obtains precisely the exotic involution of [8].
Our aim is the following generalization of this example.
THEOREM9.8. If the actions of T on the topological spheres X = (ak)
S(zF) V
~~zn e~ Ii j
maps M into itself, commutes with T, and cyclically permutes the fixedpoints
of T on M. Since S has odd order k its action on M can be lifted to a trans-
formationS of order k on FM which commutes with T. This implies that, if
P is any one of the fixedpoints of T in M,
s(P, T) = s(SP, T) .
Since S permutes these fixedpoints cyclically it follows that the signs ' (P, T)
for P e M are all the same. A similar result holds for the fixed points of
T I M' and so (9.12) takes the form
*2m-i
Since Spin (T, X) e Z[il, this implies that k _ 1 mod 22m1. q.e.d.
OXFORD UNIVERSITY
HARVARD UNIVERSITY
BIBLIOGRAPHY
[1] M. ATIYAH and R. BOTT, A Lefschetz fixed point formula for elliptic differential
operators, Bull. Amer. Math. Soc. 72 (1966), 245-250.
[2] , A Lefschetz fixed point formula for elliptic complexes: I. Ann. of Math. 86
(1967), 374-407.
[3] and A. SHAPIRO, Clifford modules, Topology 3 (1964), 3-38.
[4] M. ATIYAH and G. SEGAL, The index of elliptic operators: II, Ann. of Math. 87 (1968),
531-545.
[5] and I. M. SINGER, The index of elliptic operators: I, Ann. of Math. 87 (1968),
484-530.
[6] , The index of elliptic operators: III Ann. of Math. 87 (1968), 546-604.
[7] A. BOREL and A. WEIL, Representations lineaires et espaces homogenes Kahlerians
des groupes de Lie compactes, Seminaire Bourbaki, May 1954 (Expose by J.P.
Serre).
[8] G. E. BREDON, Examples of differentiable group actions, Topology 3 (1965), 115-122.
[9] E. BRIESKORN, Beispiele zur Differentialtopologie von Singularitaten. Invent. Math. 2
(1966), 1-14.
[10] P. E. CONNER and E. E. FLOYD, Maps of odd period, Ann. of Math. 84 (1966), 132-156.
[11] M. EICHLER, Eine Verallgemeinerung der Abelschen Integrate, Math. Z. 67 (1957), 267-
298.
[12] W. FRANZ, Uber die Torsion einer Uberdeckung, J. Reine Angew. Math. 173 (1935),
245-254.
[13] R. HARTSHORNE, Residues and duality, Lecture notes in Mathematics, Springer-Verlag
1966.
[14] M. W. HIRSCH and J. MILNOR, Some curious involutions of spheres, Bull. Amer. Math.
Soc. 70 (1964), 372-377.
[15] J. MILNOR, Whitehead torsion, Bull. Amer. Math. Soc. 3 (1966), 358-426.
[16] A. SELBERG, Harmonic analysis and discontinuous groups in weakly symmetric spaces
with applications to Dirichlet series, J. Indian Math. Soc. 20 (1956), 47-48.