Theorem 3.3 (Atiyah). Let
G
be a ompat Lie group. Then(i) K 0 (BG) ∼ = R(G) [ I
,(ii) K 1 (BG) = 0
.I will nowuse this theorem to determine
K ∗ (BS 1 )
andK ∗ (BC n )
.Lemma 3.4. Let
T : S 1 ֒ → C ∗
be the natural 1-dimensional representation ofS 1
, and lett = [T ] − 1 ∈ K 0 (BS 1 )
. ThenR(S 1 ) = Z[T, T − 1 ], I = h T − 1 i , K 0 (BS 1 ) ∼ = R(S \ 1 ) I = Z[[t]].
Proof. Firstnotethatarepresentation
ρ : S 1 −→ GL n (C)
anbeonjugatedto
ρ : S 1 −→ U(n)
, by hoosing an inner produt onC n
(all of whih areonjugate) whih is
S 1
-invariant. So it sues to look at representationsρ : S 1 −→ U (n)
. Nowρ(t) ∈ U (n)
(fort ∈ [0, 2π]
) is diagonizable,ρ(t) ∼
diag
(e iθ 1 , . . . , e iθ n )
. This also diagonalizesρ(kt)
,k ≥ 1
, so if we hooset
rationallyindependent of
π
, this diagonalizationworks fora dense subset ofS 1
. Soby ontinuity we an diagonalizeρ(t)
for allt
simultaneously, and soρ
isgivenbydiag(ρ 1 (t), . . . , ρ n (t))
,whereρ k : S 1 −→ S 1
isahomomorphism.This means
ρ k (z) = z m k , m k ∈ Z
. Using the natural representationT : z 7→
z
,anditsinverseT − 1 : z 7→ z − 1
,weanreformulatethisbysayingthatevery representation ofS 1
has the formP N
i= − N n i T i
,n i ≥ 0
. The Groethendiek onstrution yieldsR(S 1 ) = ( N
X
i= − N
n i T i | n i ∈ Z )
= Z[T, T − 1 ].
Now tothe augmentationideal. By denition
I = ( N
X
i= − N
n i T i | X N i= − N
n i = 0 )
.
Clearly,
T − 1 ∈ I
, and also,P N
i= − N n i T i ∈ I
is divisible byT − 1
, beausethe sum of the oeients is zero. So
I = h T − 1 i
. NowI k =
(T − 1) k
,
and
R(S 1 )/I k
hasgenerators1, T − 1, (T − 1) 2 , . . . , (T − 1) k − 1
. Consequently, puttingt = [T ] − 1
,we getK 0 (BS 1 ) ∼ = R(S \ 1 ) I = Z[[t]].
Lemma3.5. Let
n ∈ N
be anumber withprime fatorisationn = Q
p | n p i(p)
.Then
K 0 (BC n ) ∼ = Z ⊕ M
p | n
(ˆ Z p ) p i(p) − 1 ,
where
Z ˆ p
denotes thep
-adi integers.Proof. Let
W
be the natural 1-dimensional representation ofC n ⊆ C ∗
. Asin the proof of Lemma 3.4 above, we only need look at representations
ρ : C n −→ U (m)
and diagonalize, so thatρ =
diag(ρ 1 , . . . , ρ m )
. Here eahρ j : C n −→ S 1
is a group homomorphism, and so is a power ofW
, withthe relation
W n = 1
. ConsequentlyR(C n ) = Z[W ]/ h W n − 1 i
. Theaug-mentation ideal is
I = h W − 1 i
for the same reason as before, and we mustomputetheinverse limit
lim
←− k
R(C n )/I k
. This wepropose todointwosteps:First assume
n = p i
. ThenC p i
is ap
-group, and aording to [Atiyah2℄the
I
-adi andp
-adi topologiesonI = I(C p i )
are equivalent,so thatK 0 (BC p i ) ∼ = R(C \ p i ) I = Z ⊕ I \ (C p i ) I ∼ = Z ⊕ I \ (C p i ) p .
To alulate this, let
w = W − 1
, and note thatI(C p i ) = h w i
in the ringZ[w]/ D
(w + 1) p i = 1 E
,and so
I(C p i ) ∼ = Z p i − 1
. ThusI \ (C p i ) p ∼ = (ˆ Z p ) p i − 1
.Now take any
n ∈ N
. Observe thatC p i(p)
, wheren = p i(p) m
withgcd(p, m) = 1
, are exatly the Sylowp
subgroups ofC n
. Then by [Atiyah2℄Prop. 4.10, there isaninjetivemap
K 0 (BC n ) −→ M
p | n
K 0 (BC p i(p) ),
and inpartiular
I(C \ n ) I(C n ) −→ M
p | n
I(C \ p i(p) ) I(C
pi (p) )
is injetive. By using that
C n ∼ = Q
p | n C p i(p)
by the Chinese RemainderThe-orem, itis easilyseen that this map is an isomorphism,sothat
K 0 (BC n ) ∼ = Z ⊕ M
p | n
I(C \ p i(p) ) ∼ = Z ⊕ M
p | n
(ˆ Z p ) p i(p) − 1 ,
by the result for
p i
above.Withthese results, letusrst take alook atthe
K ∗ (BS 1 )
-modulestru-ture on
K ∗ (X hS 1 )
, whereX
isanS 1
-spae,as desribed in Setion1.4. Fol-lowing the notation in Lemma 3.4, we have the anonial representation
T
of
S 1
, whih by (27) gives a bundle overBS 1
, whih we also allT
. OnK
-theory,
T
denes alassinK ∗ (BS 1 )
,andK ∗ (BS 1 ) = Z[[t]]
,wheret = T − 1
.Usingthe projetion pr
1 : X hS 1 −→ BS 1
,weget lasses pr∗ 1 (T )
and pr∗ 1 (t)
inK ∗ (X hS 1 )
. We willsuppress the map pr1
from the notation, and simplyallthese lasses
T
andt
again.Weannowdeterminethe
K ∗ (BS 1 )
modulestrutureon∆(r) ≃ G(r) hS 1
:Lemma 3.6. The
K ∗ (BS 1 ) = Z[[t]]
module struture onK(∆(r))
is givenby
t 7→ (x − y)/(y + 1)
. In partiular,t 2r
ats as0
.Proof. We use the results from ohomology, where the
H ∗ (BS 1 ) = Z[u]
module struture on
H ∗ (G(r)/S 1 ) = Z[x 1 , x 2 ]/ h Q r , Q r+1 i
is given byu 7→
x 1 − x 2
,f. [Bökstedt-Ottosen℄Cor. 3.7. Reallthatx = [X] − 1
,y = [Y ] − 1
,where
x 1 = c 1 (X)
andx 2 = c 1 (Y )
aretherstChernlasses. Alsou = c 1 (T )
.The rst Chern lass gives a group isomorphism fromomplex linebundles
over
∆(r)
toH 2 (∆(r))
, sosinec 1 (T ⊗ Y ) = c 1 (T ) + c 1 (Y ) = u + x 2 = x 1 = c 1 (X ).
we get
T ⊗ Y = X
. Then we alulate(T − 1) ⊗ (Y − 1) = T ⊗ Y − Y − T + 1 = (X − 1) − (Y − 1) − (T − 1)
Isolating
T − 1
gives(T − 1) = ((X − 1) − (Y − 1)) ⊗ Y − 1 .
In
K ∗ (∆(r))
this equality givest = (x − y)(y + 1) − 1
, as desired. Sine inK(∆(r)) ∼ = Z[x, y]/ h Q r , Q r+1 i
all non-zero elements have a total degree inx, y
whih is less than2r
, we see thatt 2r = (x − y) 2r (y + 1) − 2r = 0
.Now we prove the main Theorem of this setion, but rst we introdue
a bit of notation: We write
K hS ∗ 1 (X)
forK ∗ (ES 1 × S 1 X)
, whenX
is anS 1
-spae. Reall the diagram(26)K hS ∗ 1 (G(r) (n) ) oo K ∗ (∆(r))
K ∗ (BS 1 )
OO
K ∗ (BS 1 )
B P n
oo OO
This gives amap
K ∗ (BS 1 ) (n) ⊗ K ∗ (BS 1 ) K ∗ (∆(r)) −→ K hS ∗ 1 (G(r) (n) )
where the
K ∗ (BS 1 ) (n)
denotes that the mapB P n
should be applied in thetensor produt, as the diagramindiates.
Theorem 3.7. Let
n ∈ N
. Then the mapK ∗ (BS 1 ) (n) ⊗ R(S 1 ) K ∗ (∆(r)) −→ K hS ∗ 1 (G(r) (n) )
isan isomorphism of rings. In partiular,
K hS 1 1 (G(r) (n) ) = 0
.Tox the notationand avoidlong, umbersome expressions, put
R = R(S 1 ) = Z[U, U − 1 ], R ˆ = K 0 (BS 1 ) = Z[[u]], u = U − 1.
S = R(S 1 ) = Z[T, T − 1 ], S ˆ = K 0 (BS 1 ) = Z[[t]], t = T − 1.
M = K ∗ (∆(r)) = Z[x, y]/ h Q r , Q r+1 i .
Here
S
isanR
-module by the mapU 7→ T n
, and likewiseS ˆ
is anR ˆ
-moduleby
u 7→ (t +1) n − 1
. ByLemma3.6,M
isanR ˆ
-modulebyu 7→ (x − y)/(1+y)
,and thusan
R
-module byU 7→ (x − y)/(1 + y) + 1
.The Theorem says that
S ˆ (n) ⊗ R M ∼ = K hS 1 (G(r) (n) )
. The reason forrestriting to
R
instead ofR ˆ
is given by the following lemma, whih alsoshows that for the isomorphism, this restrition does not matter.
Lemma 3.8.
S ˆ
is a atR
-module, andS ˆ ⊗ R ˆ N ∼ = ˆ S ⊗ R N.
forany nitely generated
R ˆ
-moduleN
whereu m
ats as0 onN
for somem
.In partiular this holds for the ltration modules
M j
from Remark 3.2, forM = M 2r+1
, and for the quotientsM j /M j+1
.Proof. Clearly,
S
is a freeR
-module (with basis{ 1, U, . . . , U n − 1 }
), soS
isat over
R
. SineS
isNoetherian,S ˆ
isat overS
,see [Atiyah-MaDonald℄, Prop. 10.14. By the natural isomorphism,for anyR
-moduleM
,S ˆ ⊗ R M ∼ = ˆ S ⊗ S S ⊗ R M,
wesee that
S ˆ
isat overR
.Take
N
as inthe lemma. Then the ompletion by the idealI = h u i ⊆ R ˆ
gives
N ˆ = lim ←
k
N/u k N = N.
Also by [Atiyah-MaDonald℄, Prop. 10.13, sine
R
is Noetherian andN
is nitely generated,
N ˆ ∼ = ˆ R ⊗ R N
. Combining these two fats yields theisomorphism
S ˆ ⊗ R ˆ N ∼ = ˆ S ⊗ R ˆ N ˆ ∼ = ˆ S ⊗ R ˆ ( ˆ R ⊗ R N ) ∼ = ˆ S ⊗ R N.
Now onsider the
R ˆ
-moduleM j
. Sineu
ats as(x − y)/(1 + y)
, andM j
onsistsofpolynomialsdegreeatleast
j
,u 2r+1
atsaszero. ForthequotientM j /M j+1
,u
itself ats as zero. So the requirements ofN
holds for thesemodules.
We will use the ltration
M j
ofM
to prove the Theorem, so we need aLemma whih proves the Theorem in the ase
M = Z
:Lemma 3.9. The followingmap is an isomorphism:
K ∗ (BS 1 ) ⊗ R(S 1 ) Z −→ K ∗ (BC n ).
Proof. Let
A = S ⊗ R Z
andB = ˆ S ⊗ R Z
, and letA −→ B
be the mapindued by the ompletion
S −→ S ˆ
. We now deneanother mapA −→ R(C n ) = Z[W ]/ h W n − 1 i , T 7→ W.
This is learly an isomorphism,and preserves the augmentation ideal.
Con-sider the diagram:
B
A
oo ∼ = //
R(C n )
ˆ
B oo A ˆ ∼ = // K 0 (BC n )
Herethe vertialarrowsdenoteompletionwithrespettotheaugmentation
ideals; respetively
tB
,(T − 1)A
, andh W − 1 i )
. To prove the Lemma, wemust show
B ∼ = ˆ A
. FirstnotethatA ˆ −→ B ˆ
isanisomorphism,sineforanyk
,the map given byT 7→ t + 1
, is anisomorphism:A/(T − 1) k = Z[T ]/
T n − 1, (T − 1) k
−→ Z[t]/
(t + 1) n − 1, t k
= B/t k B.
Next we show that
B −→ B ˆ
is anisomorphism. To show this, onsider the exat sequene given by multipliationbyu − 1 ∈ R
,0 // R u − 1 // R // Z // 0 .
Sine
S ˆ
is at overR
, we obtain anew exat sequene,0 // S ˆ ⊗ R R 1 ⊗ (u − 1) // S ˆ ⊗ R R // S ˆ ⊗ R Z // 0 ,
whih, afterapplying the natural isomorphism,beomes
0 // S ˆ (t+1)
n − 1
// ˆ S // ˆ S ⊗ R Z // 0 .
(28)Completing this with respet to the ideal
h t i
, whih is an exat funtor, weobtainyet another exat sequene
0 // lim
←
S/ ˆ
t k (t+1) n − 1
// lim
←
S/ ˆ t k
// lim
← ( ˆ S ⊗ R Z)/
t k
// 0 .
Reall
S ˆ = Z[[t]]
. After applying the isomorphismlim ← S/ ˆ
t k ∼ = ˆ S
, we getthe exat sequene,
0 // S ˆ (t+1)
n − 1
// ˆ S // lim ← ( ˆ S ⊗ R Z)/
t k
// 0 .
(29)Comparing(28) and (29), we see that
B ∼ = ˆ B
. Asalready noted, this meansthat
A ˆ ∼ = B
, and this provesthe result.Now we an provethe mainTheorem 3.7:
Proof of Theorem3.7. Firstwe laim that the map
K ∗ (BS 1 ) ⊗ Z K hS ∗ 1 (G(r)) −→ K hS ∗ 1 (G(r) (n) )
(30)issurjetive. Tosee this, werst note that the map
K ∗ (BC n ) −→ K ∗ (BS 1 )
issurjetive. Thisfollowsfromthefatthatthe mapof representationrings,
R(C n ) −→ R(S 1 )
is surjetive, sine any representation ofC n
an beex-tended toarepresentationof
S 1
. Nowtoprove surjetivity of (30), weuse a ltrationargumentin the spetral sequeneH ∗ (∆(r); K ∗ (BC n )) ⇒ K hS ∗ 1 (G(r) (n) ).
This ollapses, sine everything sits in even degrees. As in the proof of
Theorem 3.1, we now use Cor. 2.5 of [Atiyah-Hirzebruh℄, so let
A
denotethe image of
K ∗ (BS 1 ) ⊗ Z K hS ∗ 1 (G(r))
inK hS ∗ 1 (G(r) (n) )
. In ltration degree0 we have
K ∗ (BC n )
. As already shownK ∗ (BS 1 )
is surjetive onto this,so the lowest ltration an be hit. Anything else in
H ∗ (∆(r); K ∗ (BC n ))
isgeneratedby monomials
x i 1 x j 2
, andwe havex i y j ∈ A
withh(x i y j ) = x i 1 x j 2 +
higherterms. This shows that
A = K hS ∗ 1 (G(r) (n) )
,so (30) is surjetive.Nowweshowthatthemapisinjetive. Wewillusealtrationargument,
where we lter
M = K 0 (S(τ )/S 1 )
as in Remark 3.2. We look at the exatsequene,
0 −→ M i+1 −→ M i −→ M i /M i+1 −→ 0.
As
S ˆ
isat overR
by Lemma3.8, weget the exat sequene0 −→ S ˆ ⊗ R M i+1 −→ S ˆ ⊗ R M i −→ S ˆ ⊗ R M i /M i+1 −→ 0.
(31)Werstapplythisto
K
-theorywithF p = Z/pZ
oeients. Forthe eldF p
,wehavebytheUniversalCoeientTheorem,K ∗ (X; F p ) ∼ = K ∗ (X) ⊗ F p
.Clearly the ltration
M i ′ = M i ⊗ F p
works forF p
oeients, so we anuse the result above. But sine
F p
is a eld, the exat sequene (31) splits,so we an do a ounting argument quite easily. Observe that
M i ′ /M i+1 ′ = F p [x, y] i / h Q r , Q r − 1 i = (F p ) n i
, wheren i ∈ N
. By Lemma3.9, we knowS ˆ ⊗ R M i ′ /M i+1 ′ ∼ = (K 0 (BC n ; F p )) n i .
(32)and in addition,
K 0 (BC n ; F p )
is a nite numberof opies ofF p
, so it makessense to ount them. Also
M 2r ′ − 1 = F p
, soS ˆ ⊗ R M 2r ′ − 1 ∼ = K 0 (BC n ; F p )
. Soindutively, sine
M ⊗ F p
is a graded ring with a total ofr(r + 1)
opies ofF p
, thenS ˆ ⊗ R M ⊗ F p ∼ = (K 0 (BC n ; F p )) r(r+1) .
Weompare this with
K ∗ (G(r) (n) ; F p )
via the spetralsequene for thever-tial brationin Prop. 1.6:
E 2 = H ∗ (∆(r); K ∗ (BC n ; F p )) ⇒ K ∗ (G(r) (n) ; F p ).
Weseeeverythingsitsinevendegreesin
E 2
,sotherearenodierentials,and, workingoveraeldF p
,weansimplyountthedimensionofK 0 (G(r) (n) ; F p )
as the sum of the dimensions of
E 2 m,n
on the diagonalm + n = 0
. SineH ∗ (∆(r); F p ) ∼ = F p [x, y]/ h Q r , Q r+1 i
also has a total ofr(r + 1)
opies ofF p
,again by Lemma 3.5, we get,
K 0 (G(r) n ; F p ) ∼ = (K 0 (BC n ; F p )) r(r+1) .
Sothe map of
F p
-vetor spaesS ˆ ⊗ R M ⊗ Z p = K 0 (BS 1 ) ⊗ R(S 1 ) K 0 (S(τ)/S 1 ; Z p ) −→ K 0 (S(τ) (p) hS 1 ; Z p )
isasurjetion between spaesof the samedimension, and isthusan
isomor-phism, and this holds for every prime number
p
.Nowwe ompare
Z
- andF p
-oeients (for aprimep
)by the diagramK 0 (∆(r)) // K 0 (G(r) (n) ; F p )
S ˆ ⊗ R M //
ϕ
OO
S ˆ ⊗ R M ⊗ F p
∼ =
OO
(33)
Assume
a ∈ S ˆ ⊗ R F
isin the kernel ofϕ
. Then, by the diagram,a
reduedmod
p
iszero, soa = p · a 1
for somea 1
. But then, sineK 0 (∆(r))
is torsionfree,
a 1 ∈ ker(ϕ)
, soa 1 = p · a 2
, et. Consequently, ifa ∈ ker(ϕ)
, thena
is divisible by
p
innitely often. Reall that this holds for any primep
, andthusalso for
n
, soa
is innitelyoften divisible byn
.Now take a look atthe ltration again
0 −→ S ˆ ⊗ R M i − 1 −→ S ˆ ⊗ R M i −→ S ˆ ⊗ R M i /M i − 1 −→ 0.
(34)If
a ∈ S ˆ ⊗ R M i
isdivisible byn
innitelyoften, then the image inS ˆ ⊗ R M i /M i − 1 ∼ = Z N ⊕ M
p | n
(ˆ Z p ) N p
is zero (the isomorphism is Lemma 3.5 and Lemma 3.9). So
a
omes froma ′
inS ˆ ⊗ R M i − 1
anda ′
is also innitely oftendivisible byn
. Soindutivelya
omes froma 0 ∈ S ˆ ⊗ R F 0 ∼ = Z ⊕ L
p | n (ˆ Z p ) p i − 1
, anda 0
is divisible byn
innitelyoften, and so
a 0 = 0
, whih impliesa = 0
.This shows that the kernel of
ϕ
is zero,and thusthe mapϕ : K 0 (BS 1 ) ⊗ R(S 1 ) K 0 (S(τ )/S 1 ) −→ K 0 (S(τ) (p) hS 1 )
(35)isan isomorphism.
4 The free loop spae and Morse theory
Now we turn to study the free loop spae
L(FP r )
, where as usualF = C
orF = H
. First a denition:Denition 4.1. Let
X
be a topologialspae. The spaeLX = { f : [0, 1] −→ X | f(0) = f (1), f
isontinuous} ,
with the ompat-open topology,is alled the free loopspae of
X
.We are going to use Morse theory to study
LM
for a smooth manifoldM
, where we will takeM = FP r
. It is a fat that it does not hange thehomotopy typeof
LM
if we requireallf ∈ LM
tobe dierentiable,or even smooth, sowe do that.Now let us onsider how one ould do Morse theory on the free loop
spae
LM
as well the spae of homotopy orbitsLM hS 1
, whereM
denotesa ompat
n
-dimensional manifold. For details, I refer to [Klilngenberg1℄, and [Bökstedt-Ottosen ℄, espeially hapters 7 and 8.LM
is not anite-dimensionalmanifold,but one an makeamodelof
LM
whihisaso-alledHilbert manifold, f. [Klilngenberg1℄ 1.2, meaning there are harts on
LM
making it loally homeomorphi to a Hilbert spae. The tangent spae of
a loop
f ∈ LM
is the spaeΓ(f )
of vetor elds alongf
. Leth· , ·i
denotethe Riemannian metri on
M
. Now the tangent spaeT f LM
arries thestruture of a Hilbert spae via
h ξ, η i c = Z
S 1
h ξ(t), η(t) i + c h∇ ξ(t), ∇ η(t) i
dt,
(36)where
ξ, η ∈ T f LM
are vetor elds alongf
inLM
, and∇
denotes theovariant derivative along
f
. The onstantc ∈ R
makes the inner produtvary. This isneessary toensurethat the
n
-folditerationmap,P n
, beomesanisometry
P n ∗ = D f ( P n ) : T f LM −→ T P n f LM, P n ∗ (ξ(z)) = ξ(z n )
sine
hP n ∗ ξ, P n ∗ η i 1 = h ξ, η i n 2
,see [Bökstedt-Ottosen ℄ 7.Weare going todo Morse theory via the energy funtion
E : LM −→ R, f 7→
Z
S 1 | f ′ (t) | 2 dt.
Foreah
a ∈ R
, wesetF (a) = E − 1 (] − ∞ , a]) ⊆ LM
. The ritial points ofE
are the losed geodesis onM
. We shall assume that the ritial pointsare olleted on ompat submanifolds, eah of whih satisfy the Bott
non-degeneray ondition. This strong ondition is needed for the Morse theory
mahinery,anditissatisedfor
M = FP r
,andmoregenerallyforsymmetrispaes, [Ziller℄. Callthe ritial values
0 = λ 0 < λ 1 < . . .
, and onsider theltration
F (λ 0 ) ⊆ F (λ 1 ) ⊆ · · · ⊆ LM.
(37)This ltration is equivariant with respet to the
S 1
ation. This means itinduesa ltrationof
LM hS 1
,F (λ 0 ) hS 1 ⊆ F (λ 1 ) hS 1 ⊆ · · · ⊆ LM hS 1 .
(38)Thenon-degeneray ondition ensurethat eahritialsubmanifold
N (λ) = E − 1 (λ)
isnite-dimensional,and the tangent bundleT (LM ) | N(λ) ⊆ T (LM )
splits
S 1
-equivariantly:T (LM) | N (λ) ∼ = µ − (λ) ⊕ µ 0 (λ) ⊕ µ + (λ),
intothe bundles of negative, zero-,and positivediretions, respetively, and
the negativebundle
µ − (λ)
is nite-dimensional. To ease the notation, writeF n = F (λ n )
andµ − n = µ − (λ n )
. The main result of Morse theory in thissetting is proved by Klingenberg in [Klilngenberg1℄, 2.4: There is an
S 1
-equivariant homotopy equivalene
F n )/ F n − 1 ≃ T h(µ − n ).
(39)We want a similar result for
(LM) hS 1
, so we onsider the quotients of theltration(38):
ES 1 × S 1 F n /ES 1 × S 1 F n − 1 ∼ = ES + 1 ∧ S 1 F n / F n − 1 ,
where
ES + 1 ∧ S 1 X = (ES + 1 ∧ X)/S 1
is the smash produt modded out bythe diagonal
S 1
ation. The obvious map dened on representatives is a homeomorphism. Thus by the Morse theorem in(39),ES 1 × S 1 F n /ES 1 × S 1 F n − 1 ≃ ES + 1 ∧ S 1 T h(µ − n ).
We an use [Bökstedt-Ottosen ℄ Lemma 5.1tond that
( F n ) hS 1 /( F n − 1 ) hS 1 ≃ ES + 1 ∧ S 1 T h(µ − n ) ∼ = T h((µ − n ) hS 1 )
(40)where for an
S 1
-vetor bundleξ
given by a projetion mapp : E −→ B
wedenoteby
ξ hS 1
thebundlewithprojetionid× p : ES 1 × S 1 E −→ ES 1 × S 1 B
.This means wealso have a Morse theorem for the
S 1
-equivariantltration.4.1 The negative bundle
In [Bökstedt-Ottosen ℄ Lemma 5.1, it is shown that the negative bundle
(µ − n ) hS 1
is an oriented vetor bundle ifµ − n
is. But to use the Thomiso-morphismin
K
-theoryweneedtoknowthatthenegativebundleisomplex,ormore preisely
Proposition 4.2. The negative bundle
µ − n
for the energyltration ofLCP r
an be written as
ε ⊕ ν
, whereε
is a trivial realS 1
-line bundle, andν
is aomplex
S 1
vetor bundle. Consequently, the negative bundle(µ − n ) hS 1
for theenergyltration of
LCP r hS 1
an also be written asε ⊕ ν hS 1
.Proof. ThereisaHermitianinnerprodut
h· , ·i C
onT CP r
,andtheRieman-nian metri is
h· , ·i =
Re( h· , ·i C )
. The tangent spaeT f LCP r
is a omplexvetor spae,and itarries the struture of a Hilbert spae via
h ξ, η i = Z
S 1
( h ξ(t), η(t) i + h∇ ξ(t), ∇ η(t) i ) dt,
where
ξ, η ∈ T f LCP r
are vetor elds alongf
inLCP r
, and∇
denotes theovariant derivativealong
f
. Sineh· , ·i =
Re( h· , ·i C )
, wegeth zξ, zη i = h ξ, η i
forz ∈ S 1 .
(41)If
f
is a ritial point of the energy funtionalE
(a geodesi), then thetangent spae of
LCP r
splits asT f LCP r = Γ(Rf ′ ) ⊕ Γ(Rif ′ ) ⊕ Γ((f ′ ) ⊥ )
(42)where e.g.
Γ(Rf ′ ) ⊆ Γ(f)
denotes the vetor eldsξ
alongf
withξ(t) ∈ Rf ′ (t) ⊆ T f(t) CP r
. We an use the inner produt to represent the HessianH = D 2 E
ofE
by a linear operatorA = A f
onT f LCP r
, by requiringh Aξ 1 , ξ 2 i = H(ξ 1 , ξ 2 )
. Then we get by (41) thatzAz ¯ = A
forz ∈ S 1
, whihimplies that
A
isomplex linear.Aording to Klingenberg, [Klilngenberg1℄ Thm. 2.4.2,
A f =
id− (1 − ∇ 2 ) − 1 ◦ ( ˜ K f + 1),
where
K ˜ f (ξ)(t) = R(ξ(t), f ′ (t))f ′ (t)
= π 2 (f ′ (t) h ξ(t), f ′ (t) i − 2f ′ (t) h f ′ (t), ξ(t) i + ξ(t) h f ′ (t), f ′ (t) i )
Notethefator
π 2
;itappearsbeauseourmetrionCP r
issaledsothattheirumferene is 1, not
π
. This gives us the following eigenvalue equationAξ = λξ
:(λ − 1) ∇ 2 ξ = ( ˜ K f + λ)ξ
(43)Thenegativebundleonsistsofsolutionstothis equationwith
λ < 0
. Notiethatbytheformulafor
A
,itpreservesthedeomposition(42),sineovariant derivativeommuteswiththeomplexstrutureonT CP r
. Thusweansolve(43) inthe three spaes separately.
(i) ξ ∈ Γ(Rf ′ )
: Thenξ(t) = g(t)f ′ (t)
whereg : [0, 1] −→ R
is a smoothfuntionwith
g(0) = g(1)
. ThenK ˜ f (t) = 0
,andequation(43)beomes(λ − 1)g ′′ = λg ⇔ g ′′ = λ λ − 1 g ⇒ g = 0
sine
λ < 0
andg
mustbeperiodi. Sowehavenonon-trivialsolutions.(ii) ξ ∈ Γ((f ′ ) ⊥ )
: Sine(f ′ ) ⊥
is a omplex vetor spae, andA
is omplexlinear as noted, we see that
Aξ = λξ
impliesA(iξ) = λ(iξ)
. So thisspaeof solutions has a omplex struture.
(iii) ξ ∈ Γ(Rif ′ )
: Thenξ(t) = g (t)if ′ (t)
, whereg : [0, 1] −→ R
is asmoothfuntionwith
g(0) = g(1)
. ThenK ˜ f (t) = 4π 2 k f ′ (t) k 2 ξ(t) = 4π 2 n 2 ξ(t)
,sine
f
is ageodesi of lengthn
. The equation (43) then beomes(λ − 1)g ′′ = (4π 2 n 2 + λ)g ⇔ g ′′ = 4π 2 n 2 + λ
λ − 1 g
To get a periodi solution
g
, we must have4π 2 λ n − 2 1 +λ ≤ 0
, i.e.λ ≥
− 4π 2 n 2
. Forλ = − 4π 2 n 2
we must haveg
onstant, and this gives thetrivialreal linebundle
ε
. If− 4π 2 n 2 < λ < 0
we have the solution setspanned over
R
byg 1 K (t) = cos(K · 2πt),
andg 2 K (t) = sin(K · 2πt), t ∈ [0, 1]
where
K = s
− 4π 2 n 2 + λ
2π(λ − 1) ,
andK ∈ N,
sine the funtions must be periodi with period 1. This happens if
and onlyif
λ = 4π 2 (K 2 − n 2 ) 4π 2 K 2 + 1 ,
so for a xed
n
we get solutions withλ < 0
forK = 1, . . . , n − 1
.Thisspae ofsolutionsan be given aomplexstruture
J
by rotatingt 7→ t − 4K 1
,wheret ∈ [0, 1]
,i.e.J(g 1 K ) = g 2 K , J (g 2 K ) = − g K 1 .
and extendinglinearly. Clearly
J
satisesJ 2 = −
id.This gives the bundle
ν
, whih is learly anS 1
bundle, with theS 1
ationgiven by rotation.
Nowletussee thattheresultfor
µ − n
impliesthatfor(µ − n ) hS 1
. Thebundle(µ − n ) hS 1
isdened sothatthe pullbakof(µ − n ) hS 1
agreeswith pr∗ (µ − n )
inthefollowingdiagram,
µ − n
pr
∗ (µ − n )
//
oo (µ − n ) hS 1
G n (r) ES 1 × G n (r) //
proo ES 1 × S 1 G n (r)
where
G n (r)
denotesthespaeofn
-timesiteratedgeodesis. Sineµ − n = ε ⊕ ν
isadeompositionin